CERN-Superconductivity For Accelerators

CERN–2014–005
17 July 2014
ORGANISATION EUROPÉENNE POUR LA RECHERCHE NUCLÉAIRE
CERN EUROPEAN ORGANIZATION FOR NUCLEAR RESEARCH
Superconductivity for Accelerators
Erice, Italy
24 April – 4 May 2013
Proceedings
Editor: R. Bailey
GENEVA
2014
ISBN 978–92–9083–405–2
ISSN 0007–8328
Copyright c CERN, 2014
Creative Commons Attribution 4.0
Knowledge transfer is an integral part of CERN’s mission.
This CERN Yellow Report is published in Open Access under the Creative Commons Attribution 4.0 license
(http://creativecommons.org/licenses/by/4.0/) in order to permit its wide dissemination and use.
The submission of a contribution to a CERN Yellow Report shall be deemed to constitute the contributor’s
agreement to this copyright and license statement. Contributors are requested to obtain any clearances that may
be necessary for this purpose.
This reports in indexed in: CERN Document Server (CDS), INSPIRE, Scopus.
This report should be cited as:

Proceedings of the CAS–CERN Accelerator School: Superconductivity for Accelerators,
Erice, Italy, 24 April – 4 May 2013, edited by R. Bailey, CERN-2014-005 (CERN, Geneva, 2014),
DOI: 10.5170/CERN–2014–005
A contribution in this report should be cited as:

[Author name(s)], in Proceedings of the CAS–CERN Accelerator School: Superconductivity for Accelerators,
Erice, Italy, 24 April – 4 May 2013, edited by R. Bailey, CERN-2014-005 (CERN, Geneva, 2014),
pp. [first page] – [lastpage], DOI: 10.5170/CERN–2014–005. [first page]
Abstract
These proceedings collate lectures given at the twenty-seventh specialized course organised by the CERN Accel-
erator School (CAS). The course was held at the Ettore Majorana Foundation and Centre for Scientific Culture
(EMFCSC) in Erice, Italy, from 24 April to 4 May 2013.
Following recapitulation lectures on basic accelerator physics and superconductivity, the course covered topics
related to the design, production and operation of superconducting RF systems and superconducting magnets for
accelerators. The participants pursued one of six case studies in order to get ’hands-on’ experience of the issues
connected with the design of superconducting systems. A series of topical seminars completed the programme.
iii
iv
Foreword
The aim of the CERN Accelerator School (CAS) is to collect, preserve and disseminate the knowledge accumu-
lated in the world’s accelerator laboratories over the years. This applies not only to general accelerator physics,
but also to related sub-systems and associated technologies, and to how these are adapted to particular require-
ments. This wider aim is achieved by means of specialized courses currently held yearly. The topic of the 2013
specialized course was Superconductivity for Accelerators and was held at the Ettore Majorana Foundation and
Centre for Scientific Culture (EMFCSC) in Erice, Italy from 24 April to 4 May 2013.
The backing of the CERN management and the guidance of the CAS Advisory and Programme Committees
enabled the course to take place, while the attention to detail by the staff at EMFCSC ensured that the school was
held under optimum conditions.
Special thanks must go to the lecturers for the preparation and presentation of the lectures, even more so to
those who have written a manuscript for these proceedings.
For the production of the proceedings we are indebted to the efforts of Barbara Strasser and to the CERN
E-publishing Service, especially Valeria Brancolini for her very positive and efficient collaboration.
Finally, the enthusiasm of the participants of 23 nationalities, from institutes in 14 countries, provides con-
vincing proof of the usefulness and success of the course.
These proceedings have been published in paper (black and white) and electronic form. The electronic version,
with full colour figures, can be found at https://cds.cern.ch/record/1507630?ln=en.
Roger Bailey
CERN Accelerator School
v
PROGRAMME FOR SUPERCONDUCTIVITY FOR ACCELERATORS
24 April – 4 May 2013, Erice, Italy

Time Wednesday Thursday Friday Saturday Sunday Monday Tuesday Wednesday Thursday Friday Saturday
24 April 25 April 26 April 27 April 28 April 29 April 30 April 1 May 2 May 3 May 4 May
09:00 Introduction AC/RF Cavity Design Superconductors Mechanical Cryostat Stability of SC Superconductor
Superconductivity & Ancillaries for Magnets Design of Design Cables Dynamics
II I SC Magnets I
I
10:00 R. Bailey G. Ciovati H. Padamsee R. Flukiger F. Toral V. Parma L. Bottura F. Gomory

10:00 Basic Material Fabrication & Principles of SC Heat Heat Protection of Vacuum D
Thermodynamics Properties at LT Materials Magnet Design Transfer & Transfer & SC Magnets Techniques for
A for SC Cooling Cooling SC Devices E
Techniques Techniques
R E I II P
11:00 P. Duthil P. Duthil W. Singer H. Ten Kate B. Baudouy B. Baudouy H. Ten Kate P. Chiggiato
R COFFEE COFFEE COFFEE X COFFEE COFFEE COFFEE COFFEE COFFEE A
11:30 Superconductivity HOMS and Limitations & Superconductors Mechanical Cryostat Superfluid He Manufacturing
I I Heating Possible C for Magnets Design of Design Technology/ and Testing R
Solutions II SC Magnets II Applications
V U II T
12:30 A D. Larbalestier B. Holzer C. Antoine R R. Flukiger F. Toral V. Parma P. Lebrun L. Rossi U
vi
LUNCH LUNCH LUNCH LUNCH LUNCH LUNCH LUNCH LUNCH
15:00 L Transverse Beam Refrigeration Measurement S Superconducting Case Study Case Study Current Leads, R
Dynamics I Techniques Cables F Work Presentations Links and Buses
I I R E
E
16:00 B. Holzer A. Alekseev D. Reschke O P. Bruzzone E A. Ballarino
16:00 D Superconductivity Refrigeration Measurement Magnetic Design Case Study Case Study Large SC
II II Techniques N of SC Magnets Work Presentations Magnet Systems D
A II A
17:00 F A
Y D. Larbalestier A. Alekseev D. Reschke E. Todesco T P. Vedrine
TEA TEA TEA TEA E TEA TEA TEA Y
17:30 Longitudinal Cavity Design & Event Seminar R Seminar Case Study Case Study
Beam Dynamics Ancillaries Creation’s NMR/MRI N HTS Power Presentations Summary
I Birthday O Applications
O
18:30 B. Holzer H. Padamsee H. Padamsee T. Havens N M. Noe P. Ferracin
18:30 Case Study Seminar Closing Talk
Introduction ITER
L. Bottura N. Mitchell
20:00 Dinner Dinner Dinner Dinner Dinner Dinner Dinner Dinner Banquet Dinner
Contents
Foreword
R. Bailey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Basic Thermodynamics
P. Duthil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Introduction to Transverse Beam Dynamics
B.J. Holzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Introduction to Longitudinal Beam Dynamics
B.J. Holzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
AC/RF Superconductivity
G. Ciovati . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Material Properties at Low Temperature
P. Duthil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Higher-order Modes and Heating
B.J. Holzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Basics of Low-temperature Refrigeration
A. Alekseev . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Design Topics for Superconducting RF Cavities and Ancillaries
H. Padamsee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
SRF Cavity Fabrication and Materials
W. Singer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
How to Achieve the Best SRF Performance: (Practical) Limitations and Possible Solutions
C. Z. Antoine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Superconductivity for Magnets
R. Flukiger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Magnetic Design of Superconducting Magnets
E. Todesco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Mechanical Design of Superconducting Accelerator Magnets
F. Toral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Heat Transfer and Cooling Techniques at Low Temperature
B. Baudouy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
Cryostat Design
V. Parma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
Cable Stability
L. Bottura . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
Cooling with Superfluid Helium
Ph. Lebrun and L. Tavian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
Superconductor Dynamics
F. Gömöry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
Vacuum Technology for Superconducting Devices
P. Chiggiato . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
vii
Manufacturing and Testing of Accelerator Superconducting Magnets
L. Rossi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
Current Leads, Links and Buses
A. Ballarino . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547
Large Superconducting Magnet Systems
P. Védrine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559
Case Studies on Superconducting Magnets for Particle Accelerators
P. Ferracin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585
List of Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 591
viii
Published by CERN in the Proceedings of the CAS-CERN Accelerator School: Superconductivity for Accelerators, Erice,
Italy, 24 April – 4 May 2013, edited by R. Bailey, CERN–2014–005 (CERN, Geneva, 2014)
Basic Thermodynamics
P. Duthil 1
Institut de Physique Nucléaire d’Orsay, IN2P3-CNRS/Université de Paris Sud, Orsay, France
Abstract
The goal of this paper is to present a general thermodynamic basis that is
useable in the context of superconductivity and particle accelerators. The
first part recalls the purpose of thermodynamics and summarizes its
important concepts. Some applications, from cryogenics to magnetic
systems, are covered. In the context of basic thermodynamics, only
thermodynamic equilibrium is considered.
Keywords: thermodynamics, expansion of gas, heat machines, energy, phase

transition.
1 Introduction
Thermodynamics provides macroscopic descriptions of the states of complex systems and of their
behaviours when they interact or are constrained under various circumstances. A thermodynamic
system is a precisely specified macroscopic region of the Universe. It is limited by boundaries of a
particular nature (real or not) with particular properties. All space in the Universe outside the
thermodynamic system is known as the surroundings, the environment, or the reservoir. By
(European) convention, the scalar quantities exchanged between the system and the surroundings are
counted negatively if they leave the system and positively in the opposite case. A thermodynamically
isolated system is completely isolated from its environment: it does not exchange energy (heat, work)
or matter with the environment. If matter is able to flow in or out of its boundaries, the system is said
to be open. Otherwise it is said to be closed (by real boundaries). Processes that affect the interior of
the region of the system are studied using general principles: the four laws of thermodynamics.
Although historically established before modern physics, most of these laws are based on the fact that
a system is governed by general statistical principles which apply to the random motion of its
particles. Hence, two approaches are possible when studying a system by consideration of its
thermodynamics: a very general and practical one is to establish balances of the transfers between the
system and its surroundings; the other is to apply statistical laws to the system. The latter approach is
suitable for simple systems, and may also be used to clear up some of the more abstract concepts
within thermodynamics.
A unique macroscopic description of a system requires thermodynamics to use a number of
macroscopic parameters: the pressure p, the volume V, the magnetization M̂ , or the applied magnetic
field Ĥ , for example. Some other fundamental macroscopic parameters are defined by the laws of
thermodynamics: the temperature T, the total internal energy U and the entropy S. Knowing
(experimentally) the minimum number of parameters required for this unique description fixes the
number of independent state variables chosen for the macroscopic description. Thermodynamics then
expresses constraints on the system by use of some relationships between these parameters.
Parameters that are proportional to the size of the system are extensive variables and are written in the
following in upper-case characters. Parameters that do not depend on the mass of the system are
intensive quantities and are written in lower-case characters. An exception is made for the
temperature, an intensive quantity denoted by T . For homogeneous systems, extensive parameters can
1
duthil@ipno.in2p3.fr
978–92–9083–405-2; 0007-8328 – c CERN, 2014. Published under the Creative Common Attribution CC BY 4.0 Licence. 1
http://dx.doi.org/10.5170/CERN-2014-005.1
P. D UTHIL
be related to intensive variables via the relation X= m ⋅ x , where m is the mass. They then refer to
bulk properties of the system. An example is the volume V (m3) and the specific volume v (m3·kg−1).
2 Equilibrium and equation of state

A thermodynamic system is in thermodynamic equilibrium when all its state variables remain constant
with time: there is no net flow of matter or energy, no phase changes, and no unbalanced potentials (or
driving forces) within the system. A system that is in thermodynamic equilibrium experiences no
change when it is isolated from its surroundings. It should be stressed that thermodynamic equilibrium
implies steady state, but that steady state does not always induce thermodynamic equilibrium (e.g.
steady heat diffusion in a support). For different combined subsystems forming one thermodynamic
system at thermodynamic equilibrium, the extensive parameters of each subsystem are summed
(volume, for example) and the intensive parameters are equal (pressure, temperature). At equilibrium,
a functional relationship between the parameters (state variables) of a system can be expressed as an
equation of state.
As an example, a gas is completely described at equilibrium by the parameters p, V, and T. The
equation of state, of the form f ( p,V , T ) = 0 , describes an equilibrium surface on which a point
represents an equilibrium state. The isotherms of a (dilute) gas may be described according to Boyle’s
law (or the Boyle–Mariotte law) by pV / n = cst , where n denotes the number of moles. An
experimental temperature scale can be defined by any function of the temperature associated with
these isotherms. The simplest function is pV = nRT , known as the ideal gas law, which defines the
isotherms as hyperbolic curves in the p − V plane—see Fig. 1(a). Using an air thermometer,
Amontons suggested in 1702 that the zero on this temperature scale would be for the zero pressure
limit. Taking into account the interaction between the molecules, the van der Waals equation
( p + a / v ² )( v − b ) =
rT is also an equation of state for a gas: it modifies the pressure term by
considering the attraction effect –a between molecules; it also corrects the volume by considering
what is really accessible by one mole of molecules: a volume b is indeed not available due to the
repulsion effect of the molecules. It is accurately applicable for a larger range of pressures. In the van
der Waals equation, there is a critical isotherm for which (∂p / ∂V )T =∂ ( 2 p / ∂V 2 )T =0 at a critical
point C having coordinates pC = a / (27b 2 ) , VC = 3nb , and TC = 8a / (27Rb) . Tending to this critical
point, the van der Waals hyperbolic isotherms experience an inflection point—see Fig. 1(b). Below
this critical point, two inflection points are revealed (furthermore, for a given volume and temperature,
the calculated pressure can be negative). This domain coincides with the liquid phase of the fluid (see
Section 15).
Fig. 1: Equilibrium surfaces given by the ideal gas (a) and the van der Waals (b) equations of state
2
BASIC T HERMODYNAMICS
3 Transformations
Thermodynamics often consider a system evolving from an initial equilibrium state to a final
equilibrium state via a thermodynamic transformation. A reversible transformation is a
thermodynamic process that can be assessed via a succession of thermodynamic equilibria by
infinitesimally modifying some external constraints. It can be reversed without changing the nature of
the external constraints, and only by reversing the parameters (such as time). Otherwise it is
irreversible. When the thermodynamic process occurs infinitely slowly, it might ensure that the
system goes through a sequence of states that are infinitesimally close to equilibrium: state variables
and equation of states are well defined. But the system is not in equilibrium with its surroundings: the
transformation is thus quasi-static. A reversible transformation is quasi-static. But a quasi-static
transformation can be irreversible (heat diffusion in a support, for example). If the final state is the
same as the initial state, the process is a thermodynamic cycle.
4 Work, heat, and the first law of thermodynamics
4.1 Work
Mechanical work δ W= F ⋅ dx is achieved when displacement dx or deformation occurs by means of a
force field. Consider a closed system, for example a gas contained in a cylinder, as shown in Fig. 2(a).
Fig. 2: Work of a closed system (a) and an open system (b)
For a reversible expansion or compression of the gas, induced by the movement of a piston of cross-
sectional area A,
δ Wepf =
− F dx = − F / A ⋅ A dx =
− p dV
and
Wepf = ∫ − p dV , (1)
where the subscript ‘epf’ refers to the external pressure forces. During expansion, the volume change
dV is positive and thus δ Wepf < 0 : work is given by the system to the surroundings. During
compression, work is received from the surroundings. For an isochoric process, dV = 0 and thus
δ Wepf = 0 .
Considering the open system shown in Fig. 2(b), a mass of gas dm is received by the cylinder
during an admission step preceding the expansion and is then rejected after compression. The flow of
matter entering the system achieves some work as a piston of fluid; the system also performs work
onto the exiting flow of matter. This total flow work is given by:
δ Wflow =pin Ain dlin − pout Aout dlout =pin dVin − pout dVout =pin vin δmin − pout vout δmout =[ pvδm]out .
in
3
P. D UTHIL
where v is the specific volume (unit: m3·kg−1) of the fluid. The work which may be performed by the
system on some mechanical device is often called the ‘shaft’ work (useful work), and is the difference
between the total work δ W and the flow work δ Wflow :
δ Wshaft =
δ W − δ Wflow = V dp . (2)
4.2 Internal energy

The internal energy U (unit: J) is the sum of the kinetic and potential energies of all the physical
=
particles of the system (and of the rest-mass energies) at some microscale: U ∑ Ec,micro + ∑ Ep,micro .
It is a state function and can thus be defined by macroscopic parameters. For example, for a non-
magnetic compressible fluid, if p and V are fixed, U = U ( p,V ) is also fixed.
4.3 First law of thermodynamics and energy balances

Between two thermodynamic equilibria, δ= = δW + δ Q ,
U δ W + δ Q ; for a reversible process, dU
where W is the exchange work (mechanical, electrical, magnetic interaction, etc.) and Q is the
exchange heat. For a cyclic process during which the system evolves from an initial state I to an
identical final state F: ΔU = U F − U I = 0 .
Between two thermodynamic equilibria, the total energy change is given by
ΔE =ΔEc,macro + ΔEp,macro + ΔU =W + Q . Ec,macro and Ep,macro are the kinetic and potential energies of
the system at a macroscopic scale, and they are null if the system does not move or is not in a force
field. This leads to:
ΔU= W + Q . (3)
If work is due to external pressure forces only, then W = Wepf = ∫ − p dV and ΔU= Wepf + Q . For an
isochoric process, the volume remains constant and ΔU = Q . This relation is the basis of calorimetry.
For an open system, the work δ Wflow achieved by the pressure forces to make the fluid circulate
in and out is distinguished from the useful work performed by other forces (mechanical, magnetic,
electrical, etc.) and denoted δ Wu . We note that for the system of Fig. 2(b) δ Wu = δ Wshaft . The first law
then yields:
ΔE = ΔEc,macro + ΔEp,macro + ΔU + δ m [u ]inlet =Wu + Wflow + Q .
outlet
This can be rewritten as follows:

ΔEc,macro + ΔEp,macro + ΔU + δ m [u + pV ]inlet =Wu + Q .
outlet
Another function of state, called enthalpy, is then defined as

H= U + pV (unit: J). (4)
Thus, h being the enthalpy per unit mass (J·kg −1), we obtain
ΔE =ΔEc,macro + ΔEp,macro + ΔU + δ m [ h ]inlet =Wu + Q .
outlet
Ec,macro Δ=
For a cyclic process ( ΔU = 0 ), if Δ= Ep,macro 0 ,
= Wu + Q .
ΔH (5)
In the case of an isobaric process (for which the pressure remains constant) and if no other type of
work is performed other than the pressure work, ΔH = Q .
4
The energy balance for an open system at steady state or undergoing a cyclic process is written
as follows:
= Wu + Q ∑ mh − inlet boundaries
∑ mh (unit: J),
outlet boundaries
(6)
= Wu + Q ∑ mh  −
outlet boundaries
∑ mh  (unit: W).
inlet boundaries
5 Entropy and second law of thermodynamics

For a system considered to be between two successive states, there is a function of state S called
entropy, unit J·K−1, such that:
ΔSsyst
= Δ= S ΔS e + S i , (7)
where ΔS e = ∫ δ Q / T relates to the heat exchange and S i is an entropy production term:

S i ΔSsyst + ΔSsurroundings . For a reversible process, S i = 0 ; for an irreversible process, S i > 0 . For an
=
adiabatic ( δ Q = 0 ) and reversible process ( S i = 0 ) the thermodynamic transformation is isentropic:
ΔS = 0 .
5.1 Entropy of an isolated system

An isolated system has no exchange with the surroundings; thus ΔS e = 0 and ΔSsyst= S i ≥ 0 . At
thermodynamic equilibrium, the isolated system experiences no change, and its state variables remain
constant with time. Hence, ΔSsyst = 0 and S i = 0 .
The statistical weight Ω denotes the number of possible microstates of a system: the position of
the atoms or molecules, or the distribution of the internal energy. The different possible microstates
correspond to (and are consistent with) the same macrostate described by the macroscopic parameters
(p, V, etc.). The probability of finding the system in one microstate is the same as that of finding it in
another microstate. Thus, the probability that the system is in a given macrostate must be proportional
to Ω . The entropy can be expressed as follows:
S = kB ln(Ω ) , (8)
where =kB 1.38 ⋅ 10−23 J·K−1 is the Boltzmann constant. Considering two systems A and B, the number
of microstates of the combined system A ∪ B is Ω A ⋅ Ω B and its entropy is = S S A + S B : the entropy is
additive. Similarly, it is proportional to the mass of the system, and is thus an extensive parameter, i.e.
S= m ⋅ s , where s is the specific entropy (unit: J·K−1·kg−1).
5.2 Principle of increase of entropy

One statement of the second law of thermodynamics is that “entropy of an isolated system tends to
increase”. The system changes in response to this tendency. Such a change is said to be ‘spontaneous’.
If one imagines two possible macrostates for an isolated system, corresponding to two values of the
number of possible microstates Ω , the system will experimentally be found in the macrostate for
which the number of microstates is the greater, thus having larger entropy. We consider, for example,
a gas in a box, and assume that the gas is on one side of the box at some initial time corresponding to
Ω = Ω1 . If the gas is free to fill up the whole box, corresponding to Ω = Ω 2 , it will ‘spontaneously’
go to that state. Indeed, Ω 2 >> Ω1 , and there is a higher probability of finding the gas throughout the
whole box than of staying on one side of the box. This principle applies to isolated systems only.
Otherwise, one would have to consider exchanges with the surroundings, and thus the change of
entropy of the surroundings. Even though not very convenient, it is always possible to transform a
system into an isolated one by extending its boundaries, keeping in mind that the whole Universe is an
isolated system.
5
P. D UTHIL
6 Thermodynamic temperature and the zeroth law of thermodynamics

Let us consider two closed systems, A at temperature TA and B at TB , each having a constant volume
and separated by a diathermal wall: they are in thermal contact, they are not isolated from each other,
but they cannot exchange work (or matter). The combined system A ∪ B is isolated and therefore
0 δ Q A + δ QB . Thus, energy δ U = δ Q (heat) can flow from A to B and from B to A
d(U A + U B ) ==
such that δ Q A = −δ QB . The net change of entropy of the isolated combined system is given by
 1 1 
δS = δ S e + δ S i =+
0 δ Si =
δ S A + δ SB = −  δ QA ≥ 0 .
 TA TB 
Indeed, if TA > TB , we expect that heat will flow spontaneously from A to B such that δ Q A < 0 and
then δ S > 0 . If, however, TA < TB , then δ QB =−δ Q A < 0 and δ S > 0 . We find a relationship between
temperature and entropy which is consistent with the second law of thermodynamics.
The thermodynamic temperature is defined as:
T= ( ∂U / ∂S )V . (9)
For the isolated combined system A ∪ B , the net change of entropy can rewritten as
A  ∂S   ∂S B   1 1 
δS =
  δU A +   δU B =
 −  δ QA .
 ∂U A VA ∂U
 B VB T
 A TB 
At thermodynamic equilibrium, maximization of the entropy requires that δ S = δ S e + S i = 0 + 0 = 0

and thus that TA = TB . Similarly, one can observe that two systems both in thermal equilibrium with a
third system are in thermal equilibrium with each other. This equilibrium can be assessed by the
equality of scalar quantities, i.e. the temperature of each system, an intensive parameter. This is the
zeroth principle of thermodynamics, or the principle of temperature existence.
7 Boltzmann factor and the third law of thermodynamics: statistical point of view
We now consider a system SYST , having energy E , in relation to its much larger surroundings or
environment, named EXT , which has energy E EXT at temperature T . We assume that the combined
= SYST ∪ EXT is isolated (system E EXT is possibly the entire Universe outside SYST ).
system TOT
Then the probability that SYST has energy E (corresponding to the number of microstates Ω ) is the
probability that EXT has energy ETOT − E corresponding to the number of microstates
Ω EXT ( ETOT − E ) . Noting that S EXT = kB ln Ω EXT , we have ln Ω EXT ( ETOT − E ) =
S EXT / k B . S EXT is thus a
function of ETOT − E = EEXT , which can be expanded around ETOT (noting that E << ETOT ):
1 E  ∂S EXT  1 E E
ln Ω EXT ( E
=TOT − E ) S EXT ( ETOT ) −  =  S EXT ( ETOT
= )− ln Ω EXT ( ETOT ) − .
kB k B  ∂ETOT  E =0 k B k BT k BT
The probability of observing the macroscopic system SYST with energy E is thus proportional to the
Boltzmann factor: exp(− E / kBT ) . The temperature limit T → 0 means that the system is in a
minimum-energy state. At the zero of temperature, any system in thermal equilibrium must exist in its
lowest possible energy state. If we consider, for example, SYST to be a molecule with energy ε in
an ideal gas of constant volume and energy E EXT , the probability of the molecule having energy ε is
exp(−ε / kBT ) . As T → 0 , the whole gas is in a state of minimum energy as no molecule can be in an
excited state. Thus, if T → 0 , the minimum-energy state is unique ( Ω → 1 ) and S = 0 . It therefore
follows that an absolute entropy can be computed and referenced from the temperature.
6
8 Relating entropy to functions of state
8.1 Entropy with internal energy

U and S are state functions. Any macrostate at equilibrium can be described by a relationship
between these functions and the independent state variables. For example, a simple compressible
system is completely characterized by its energy and its volume, and U = f ( S ,V ) . Therefore, for a
reversible process ( S i = 0 ), the change of the state functions can be written as
 ∂U   ∂U   ∂U 
dU =
  dS +   dV =+
T dS   dV .
 ∂S V  ∂V  S  ∂V  S
Identifying this equation with the expression of the first law of thermodynamics for a closed system
yields p = −(∂U / ∂V ) S and thus
dU T dS − p dV .
= (10)
The kinetic temperature of an ideal gas can be derived from the expression of its internal energy
and from the motion of its molecules. This kinetic temperature can be related to the thermodynamic
temperature using the Boltzmann distribution. It follows that the equation of state of an ideal gas can
be written as
= pV Nk =BT nN A kBT , (11)
where N is the number of molecules of the gas and= N A 6.022 ⋅1023 mol−1 is Avogadro’s number.
Stating that R = N A kB , we find again pV = nRT , and thus we relate the experimental temperature
defined in Section 2 to both the thermodynamic and kinetic temperature.
8.2 Thermodynamic potentials

We consider a non-isolated system in thermal contact with its surroundings ( EXT ) which is at
temperature TEXT , with a large heat capacity such that heat flows from the surroundings into the
system without changing TEXT . The second principle states that for the system combined with its
surroundings the total change of entropy is given by δ STOT = δ S + δ S EXT ≥ 0 with δ S EXT = −δ Q / TEXT .
For the considered system having a constant volume, δ U = δ Q . Thus, we obtain δ (U − TEXT S ) ≤ 0 :
this spontaneous change is related to a decrease of U − TEXT S .
A new thermodynamic equilibrium ( T = TEXT ) will thus be reached for the minimum of the
(Helmholtz) free energy of the system F= U − TS . Similarly, for a system in thermal contact with its
surroundings and held at constant pressure, the Gibbs free energy G =U − TS + pV tends to a
minimum at thermodynamic equilibrium. The free energy and the Gibbs free energy are
thermodynamic potentials.
Thermodynamic potentials are specific functions of some state variables: the natural variables
of the potential. When these natural variables are kept constant, a closed thermodynamic system
evolves towards equilibrium if the potentials diminish. The thermodynamic equilibrium corresponds
to a minimum of one of the thermodynamic potentials. The internal energy U = f ( S ,V ) is such a
thermodynamic potential of natural variables S and V . For a mechanical system (no heat exchange),
we saw that an increase of energy is the result of a gain of mechanical work which is the product of a
force by a small displacement. We can generalize this concept by stating that the increase in energy of
a thermodynamic system is the result of the addition of several products of generalized
thermodynamic forces which induce thermodynamic displacements from equilibrium. The
thermodynamic force is driven by an intensive variable which is conjugated to an extensive variable
representing the thermodynamic displacement: p and V are such conjugated variables, as are T and
S . If S and V are kept constant, U will tend to a minimum as the considered closed system tends to
7
P. D UTHIL
thermodynamic equilibrium. The other thermodynamic potentials tending to a minimum for a closed
system at equilibrium are:
the enthalpy, if p and the extensive parameters (such as S ) are held constant,
H = U + pV ⇒= dH T dS + V dp ; (12)
the Helmholtz free energy, if T and the extensive parameters (such as V ) are held constant,
F =−
U TS ⇒ dF = − S dT − p dV ; (13)
the Gibbs free energy if p , T , and the extensive parameters are held constant,
G =− U TS + pV ⇒ dG = − S dT + V dp . (14)
9 Maxwell relations
For a compressible system U = f ( S ,V ) , the differential form dU is exact if the two mixed second
partial derivatives are equal:
∂ 2U ∂ 2U  ∂   ∂U   ∂   ∂U 
= ⇒    =    .
∂V ∂S ∂S ∂V  ∂V  S  ∂S V  ∂S V  ∂V  S
dU T dS − p dV , we deduce an appropriate relation between the first derivatives of T and p :

As =
 ∂T   ∂p 
  = −  .
 ∂V  S  ∂S V
By the same argument, we obtain from the different potentials the useful Maxwell relations:
 ∂T   ∂p   ∂p   ∂S 
U=
U ( S ,V ) ⇒   = −  , F = F (T ,V ) ⇒   =  ,
 ∂V  S  ∂S V  ∂T V  ∂V T
(15)
 ∂T   ∂V   ∂S   ∂V 
H = H ( S , p) ⇒   =  , G=
G (T , p ) ⇒   = −  .
 ∂p  S  ∂S  p  ∂p T  ∂T  p
10 Heat capacities
The amount of heat that must be added to a system reversibly to change its temperature is the heat
capacity: C = δ Q / dT (unit: J·K −1). The conditions under which heat is supplied, at constant volume
V or constant pressure p , must be specified:
 ∂S   ∂U   ∂S   ∂H 
=CV T=
   =  , C p T=
    . (16)
 ∂T V  ∂T V  ∂T  p  ∂T  p
It turns out that

T CV
for V = cst , S (T ,V ) = ∫ dT and ΔU = ∫ CV dT ;
0 T
T Cp
for p = cst , S (T , p ) = ∫ dT and ΔH = ∫ C p dT .
0 T
Q = ∫ Ci dT is known as the ‘sensible heat’ (see the lecture ‘Materials properties at low temperature’ in
these proceedings).
8
The relationship between C p and CV is expressed by Meyer’s relation:

 ∂p   ∂V 
C p − CV =
T    , (17)
 ∂T V  ∂T  p
which yields, for an ideal gas,

R γR
C p − CV =
nR , CV = n , and C p = n , (18)
γ −1 γ −1
where γ is the ratio of the heat capacities:

−1
Cp  ∂V   ∂V 
γ
= = T =
   . (19)
CV  ∂p T  ∂p  S
11 Some thermodynamic transformations
11.1 Adiabatic expansion of a gas

For a reversible adiabatic expansion,
 ∂T   ∂T   ∂S   ∂T   ∂V  T  ∂V 
 ∂p  =−    =     =   .
 S  ∂S  p  ∂p T  ∂S  p  ∂T  p C p  ∂T  p
As C p > 0 and (∂V / ∂T ) p > 0 , (∂T / ∂p ) S > 0 . Thus, as for an expansion ∂p < 0 , we get ∂T < 0 ,
expressing a cooling. During the adiabatic expansion of a gas, work is extracted and the gas is cooled.
This is achieved in a reciprocating engine or in a turbine (turbo-expander).
11.2 Joule–Kelvin (Joule–Thomson) expansion of a gas

We consider a gas flowing and expanding through a throttling valve from a fixed high pressure p1 to a
fixed low pressure p2 . The (open) system is thermally isolated and thus the transformation is
isenthalpic:
∆U = W ⇒ U1 + p1V1 = U 2 + p2V2 ⇒ H1 = H 2 .
It follows that:
 ∂T   ∂T   ∂H  1  ∂H  1  ∂S   V
 ∂p  =
−    =
−   =−   + V  = (α T − 1) ,
 H  ∂H p  ∂p T C p  ∂p T C p  ∂p T  C p
where α is the coefficient of thermal expansion:

1  ∂V 
α=   . (20)
V  ∂T  p
For the ideal gas, α T = 1 and thus (∂T / ∂p ) H =

0 : the isenthalpic expansion does not change the
temperature of the gas.
For a real gas, if α T < 1 , (∂T / ∂p ) H > 0 , and the expanding gas is warmed; if α T > 1 ,
(∂T / ∂p ) H < 0 , and the gas is cooled. The latter case occurs when the gas temperature at the inlet of
the valve is below a certain temperature called the inversion temperature. The inversion temperature
curve of helium-4 (4He) is plotted as a function of pressure and temperature in Fig. 3. The maximum
inversion temperature (as p → 0 ) is about 45 K. In a helium liquefier (or refrigerator), the gas is
9
P. D UTHIL
usually subcooled below the inversion temperature by combining adiabatic expansion and heat
removal within turbines and heat exchangers. It is finally liquefied by use of a Joule–Thomson
expansion. Note that the maximum inversion temperatures of hydrogen, nitrogen and oxygen are
202 K, 623 K and 761 K, respectively.
Fig. 3: Inversion temperature of helium-4 as a function of p and T . Constant enthalpy lines are dashed; the
solid line is the inversion curve.
11.3 Thermodynamic transforms of an ideal gas

Table A.1 of Appendix A lists different reversible thermodynamic processes involving an ideal gas. It
provides a plot of the considered evolution in a p − V diagram, expresses the heat and work
exchanged, and gives the changes in internal energy, enthalpy and entropy.
12 Heat machines
The basic function of a heat machine is to convert heat into work or, conversely, to produce a heat
transfer by means of supplied work. This work can be electrical, mechanical, etc., but the second
principle states that its production by a heat engine requires at least two heat reservoirs at two different
temperatures: the machine is driven with some heat supplied by the hotter reservoir, produces work,
and dumps waste heat to a colder reservoir (see Fig. 4). A heat pump is driven by an amount of work
and aims to transfer heat from a colder reservoir to a hotter one. It can therefore be used to warm or to
cool. In the latter case, the heat pump is a refrigerator. A dithermal machine is a heat machine
working by way of two heat reservoirs: one hot at temperature TH and one cold at TC . Over one
thermodynamic cycle, the system in the final state recovers its initial states. Thus,
energy balance (first law): ∆U = U f − U i = 0 ⇒ W + QC + QH = 0 ;
∆S e + S i =
entropy balance (second law): ∆S = 0 ⇒ QC / TC + QH / TH + S i =
0.
It follows that:
QH = − TH / TC ⋅ QC − TH S i .
−QC − W = (21)
10
Fig. 4: Principle of a heat machine: engine (top) and heat pump or refrigerator (bottom)
From the two expressions given for the energy and entropy balances, we can plot the domain of
the existence of heat machines in a QH − QC diagram, called a Raveau diagram and shown in
Figs. 5(a) and (b).
Fig. 5: Raveau diagrams of a heat engine (a) and a heat pump (b). Domains of existence of both types of heat
machines are indicated by pale grey zones. The domains of impossible cycles (as in not respecting the second
law) are indicated in dark grey.
12.1 Heat engine

As the work is produced (given) by the engine, we have W < 0 and, if TH > TC , then QH > 0 and
QC < 0 (see Fig. 5(a)). The engine is supplied by heat coming from the hot reservoir and rejects excess
heat at the cold reservoir. The efficiency η of an engine can be defined by the (positive) ratio of the
amount of produced work to the energy that is supplied. The supply being the heat coming from the
hot source, we have
W −W QH + QC QC
η= = = = 1+ ,
QH QH QH QH
i.e., making use of equation (21),

η =− 1 TC / TH < 1 since S i ≥ 0 .
1 TC / TH − TC S i / QH ≤ ηC =− (22)
i
The maximum possible efficiency ηC is obtained in the case of no irreversibility ( S = 0 ) and is called
the Carnot efficiency, referring to the ideal Carnot cycle presented in Section 12.3.
11
P. D UTHIL
12.2 Heat pump or refrigerator

Similarly, for a heat pump or a refrigerator, work is supplied: W > 0 . Thus, if TH > TC , then QH < 0
and QC > 0 (see Fig. 5(b)); heat is pumped from the cold reservoir and heat is rejected at the hot
reservoir. The efficiency of a heat pump or a refrigerator is defined by the Coefficient Of Performance
(COP), which is the ratio of the energy of interest to the energy supplied to the machine. The energy
of interest has to be distinguished between a heat pump, which is designed to warm, and a refrigerator,
which is designed to cool.
Thus, for a heat pump,
QH −QH −QH 1
COPHeating
= = = = ,
W W −QH − QC 1 + QC / QH
and, noting that COPC Heating is the coefficient of performance of an ideal Carnot heat pump, we have
1 1
COP
= Heating ≤ COP
= C Heating . (23)
1 − TC / TH − TC S i / QH 1 − TC / TH
For a refrigerator,
QF QF 1
COPCooling
= = = ,
W −QH − QC −1 − QH / QC
and, noting that COPC Cooling is the coefficient of performance of an ideal Carnot heat pump, we have
1 1
COP
= Cooling ≤ COP
= C Cooling . (25)
TH / TC − 1 + TC S / QH
i
TH / TC − 1
12.3 The Carnot cycle
Fig. 6: Carnot cycle
The Carnot thermodynamic cycle (see Fig. 6) performs four successive reversible transforms:
• two isothermal processes for which reversibility means that the heat transfers occur under a
very small temperature difference;
• two adiabatic processes for which reversibility leads to isentropic transformations.
12
This fully reversible cycle performs the maximum efficiency obtainable with a heat machine and thus
serves as a reference for comparing efficiencies of heat machines, taking into account the temperatures
of their heat reservoirs. The relative efficiencies of heat machines are thus defined as follows:
η TC S i
engine η r = = 1− ; (26)
ηC QH (1 − TC / TH )
COPHeating 1 − TC / TH
heat pump COP
= r Heating = ; (27)
COPC Heating 1 − TC / TH − TC S i / QC
COPCooling 1 − TH / TC
refrigerator COP
= r Cooling = . (28)
COPC Cooling 1 − TH / TC − TC S i / QH
Carnot efficiencies of a heat pump and a refrigerator are plotted in Fig. 7 versus the ratio of the
temperatures of their reservoirs. As COP tends to very small values as the cold temperature TC tends
to zero, the cryogenic refrigeration community often employs the inverse of the COP and speaks in
terms of ‘watt to supply per watt of produced cooling power’.
Fig. 7: COP values of a heat pump and a refrigerator based on the ideal Carnot cycle
The vapour compression cycle (see Fig. 8) achieved by our domestic refrigerators has a
relatively good COP compared with the Carnot cycle. This intrinsically good performance comes from
the fact that (i) the vaporization of a saturated liquid and the liquefaction of a saturated vapour are two
isothermal process (heat is, however, transferred irreversibly); and (ii) the isenthalpic expansion of a
saturated liquid is sensitively closed to an isentropic expansion. However, this cycle cannot be used in
cryogenic applications; it relies on the phase change of a working fluid, and thus the temperatures of
the heat sources cannot be very different. Even when implemented in cascade, the continuity of the
process towards a low temperature cannot be insured as the existence of critical fluids (for which a
liquid–gas phase transition is possible) does not cover continuously the domain of cryogenic
temperature.
13
P. D UTHIL
Fig. 8: Vapour compression heat machine
13 Maximum available work

Considering an isothermal ( T = cst ) and reversible thermodynamic process, dU = δ W + T dS and
thus δ W =dU − T dS =dF . The work produced by a system ( δ W < 0 ) is equal to the reduction of its
free energy. For the reversible process, this work is the maximum that can be provided. Similarly, for
an open system,= dH δ Wshaft + T dS and thus δ Wshaft =dH − T dS =dG . The maximum produced
work (other than due to the external pressure forces) is equal to the reduction of the Gibbs free energy.
In these cases, all the energy is free to perform useful work because there is no entropy production due
to irreversibilities. Sources of entropy production come from heat transfers (due to temperature
difference), friction (due to the movements of solid or fluid components), dissipative structure induced
by fluid motions, matter diffusion, electric resistance, etc.
Heat and work are not equivalent: they do not have the same ‘thermodynamic grade’. Whereas
mechanical or electric work can be integrally converted into heat, integrally converting heat into work
is impossible according to the second law. Moreover, transfers are generally irreversible, which
implies that thermodynamic processes have a direction (heat flows from a hot to a cold body, mass
transfers from high to low pressures, etc.). The exergy allows us to ‘rank’ energies by involving the
concept of ‘usable’ or ‘available’ energy: it is the maximum useful work that brings a system into
equilibrium with its surroundings at TEXT (generally the room temperature). Exergy is defined as
Ex =−U U EXT + pEXT (V − VEXT ) − TEXT ( S − S EXT ) , (29)
such that it tends to zero when the energy U , the volume V , and the entropy S of the system tend to
those of the surroundings: U EXT , VEXT , and S EXT . Exergy is a state function of the system relative to
its surroundings, not of the system only.
Considering a closed system in thermal and pressure contact with its surroundings, we can
deduce from the first and second law that
Δ(U + pEXT V − TEXT S ) =
Wu − TEXT S i ,
where Wu is the useful work produced by the system (other than the work due to pressure forces).
This work is maximum for S i = 0 , and thus
Wu ≤ Δ(U + pEXT V − TEXT S ) =
ΔEx . (30)
The maximum produced work is equal to the reduction of the exergy ΔEx . If we define the Gibbs free
energy of our system by imposing T = TEXT and p = pEXT , we see that ΔEx = ΔG . We can thus
determine the exergy by use of the thermodynamic potentials:
dEx
= dH − TEXT dS ; dEx= dF − pEXT dV ; dEx = dG .
14
From the previous considerations, and in view of the energy and entropic balances, we can define the
exergetic balance of an open system, noting that mi ⋅ exi = Exi :
dEx d  TEXT 
= ) ∑ m i ⋅ exi − ∑ m i ⋅ exi + Wu + ∑ Q i 1 −
(U + pEXT V − TEXT S = i
 − TEXT S . (31)
dt dt i∈inlet i∈outlet i  Ti 
The term −TEXT S i represents a destruction of exergy.
For a stationary state and a closed system, it reduces to

 TEXT 
Wu + ∑ Q i 1 − i 0 .
 − TEXT S = (32)
i  Ti 
The exergetic balance of one of the heat machines we considered in Section 12 is thus
 TEXT    TEXT 
Wu + Q H 1 −  + QC 1 −
i 0 .
 − TEXT S =
 TH   TC 
Then, we can deduce that, for an engine, the higher the temperature of the supplied heat, the larger the
work it can potentially produce.
14 States of matter and phase changes

We have already presented two equations of state for a compressible fluid: the ideal gas law and the
van der Waals law. A pure substance can indeed be solid, liquid, or gaseous, and its state at
thermodynamic equilibrium depends on the two intensive parameters ( p, T ) , as shown in the p − T
phase diagram of Fig. 9. Two phases coexist in equilibrium along each solid line. The three curves
intercept at the triple point J, where the three phases coexist.
Fig. 9: p −T phase diagram
15
P. D UTHIL
The Gibbs phase rule gives the number of degrees of freedom D (number of independent
intensive variables) of a system of mixed substances. In the absence of a chemical reaction or external
constraint, the variance of the system depends on the number of (chemically independent) constituents
c and the number of phases φ :
D = c + 2 −φ . (33)
The variance of a monophasic (in a single phase) pure substance in thermodynamic equilibrium
is D = 2 ; a full description of the system requires both pressure and temperature to be set. At the
triple point J, where the substance is at equilibrium in three phases, D = 0 . For a pure substance, this
unique point is defined by a unique couple ( pJ , TJ ) . Triple points of different pure substances are used
as metrological references for the temperature scale and for the calibration of reference temperature
sensors. When two phases coexist at equilibrium, the variance is equal to 1. Thus, setting p or T
variable will fix the other one. This is the case at point P (placed on the saturated curve [JC] ) on
Fig. 9: the gas (or vapour) and liquid phases coexist in equilibrium. The condition for this equilibrium
is that the free Gibbs energy G tends to a minimum. Consider a small quantity of matter δ m that
transfers from the liquid to the gas phase, then the change in the total free Gibbs energy is
δ m( g gas − g liq ) , which is minimum for g gas = g liq . If we now consider a neighbouring point Q on the
saturated liquid–vapour curve, we can write, per mass unit:
 ∂g gas   ∂g gas 
for the gaseous phase g gas (Q) = g gas ( P ) +   dp +   dT = g gas ( P) + vgas dp − sgas dT ,
 ∂p T  ∂T  p
 ∂g liq   ∂g liq 
for the liquid phase g liq (Q) = g liq ( P ) +   dp +   dT = g liq ( P ) + vliq dp − sliq dT ,
 ∂p T  ∂T  p
which yields, by subtraction,
0 = (vgas − vliq )dp − ( sgas − sliq )dT ,
where vgas and vliq (unit: m3·kg −1) are the specific volumes (the reciprocals of the mass densities) of
the gas and liquid phases.
The slope of the saturated vapour tension curve is thus given by the Clausius–Clapeyron
equation:
dp ( sgas − sliq ) lliq → vap
= = , (34)
dT (vgas − vliq ) T (vgas − vliq )
where lliq=→ vap T ( sgas − sliq ) ( > 0 ; unit: J·kg−1) is the specific latent heat for the liquid → gas
transition (vaporization). It is the quantity of heat required to transform, at constant temperature and
pressure, a unit mass of liquid into vapour. We can express this specific latent heat by use of the
enthalpy difference per mass unit between the vapour and the liquid. Indeed, during the reversible
phase transition process,
vap vap vap
lliq → vap = q = ∫ [du + p dv ]= ∫ [du + d( pv)]= ∫
liq liq liq
dh = hvap − hliq . (35)
Quantification of the vaporization latent heat from a T − s diagram is shown in Fig. 10. It can be
deduced that lliq → vap tends to zero as we approach the critical point. Also, lliq → vap can be expressed as
a function of T by neglecting vliq (as vgas >> vliq ) and using an equation of state to express dp /dT .
16
Fig. 10: Evaluation of the latent heat of the liquid–gas transition in a T − s diagram
The same argument can be applied to solid–liquid and solid–gas phase transitions defining
fusion (melting) and sublimation (deposition) latent heats, respectively. These phase transitions are
characterized by a discontinuity in the first derivatives of the Gibbs free energy S = −(∂G / ∂T ) p : at
equilibrium, the entropy of the liquid phase is not the same as that of the gaseous phase. Thus, these
phase transitions are first-order transitions.
At the critical point= ( T T=C; p pC ), however, vgas = vliq and dp / dT is finite so the Clausius–
Clapeyron equation states that lliq → vap= 0= ( sgas − sliq ) . No latent heat is involved during the transition
and S = −(∂G / ∂T ) p is continuous. But second derivatives of Gibbs free energy are not continuous,
and the transition is of second order.
15 Thermodynamics of magnetic systems: type I superconductor phase transition

In a vacuum, the magnetic field B̂ (unit: T ≡ V⋅s⋅m−² ≡ N⋅A−1⋅m−1) is proportional to the excitation (or
applied) magnetic field Ĥ (unit: V·s·A−1·m−1 ≡ A⋅m−1): Β ˆ = µ0 Η
ˆ . The proportional factor is the
permeability of free space, µ= 0 4π ⋅ 10 (N⋅A ). In a material,=
−7 −2
Bˆ µ0 ( Ηˆ +M ˆ ) , where M ˆ = χΗ
ˆ
(unit: A⋅m−1) is the magnetization, which represents how strongly a region of material is magnetized
by the application of Ĥ ; χ is the magnetic susceptibility. Thus, Βˆ = µ0 (1 + χ )Η
ˆ = µ0 µ r Η
ˆ , where µr
is the relative magnetic permeability.
Fig. 11: Type I superconductor. (a) Evolution of the magnetization with the excitation magnetic field; (b) Phase
diagram; (c) Evolution of the Gibbs free energy with the excitation magnetic field below the critical temperature.
We consider a long cylinder sample of type I superconductor placed in a solenoid producing the
excitation magnetic field Ĥ= n ⋅ i . Note that in the Meissner state, i.e. below a critical temperature
17
P. D UTHIL
TCM and below a critical magnetic field Hˆ CM (T ) , type I superconductors are perfect diamagnets, with
χ = −1 , B = 0 , and M ˆ = −Η
ˆ (see Fig. 11(a)). We assume that the sample has a uniform
magnetization parallel to the field. If we increase the external field from Ĥ to Hˆ + dHˆ , it induces a
change dBˆ of B̂ and dMˆ of M̂ . The resulting reversible magnetic work produced per unit volume
(unit: J·m−3) is given by
dWTOT = Hˆ ⋅ dBˆ = µ0 ( Hˆ ⋅ dHˆ + Hˆ ⋅ dMˆ ) .
The first term is the energy of the magnetic field produced by the solenoid in the absence of the
material. The second term is the energy supplied to the sample: d= W µ0 Hˆ ⋅ dMˆ . The change of
internal energy of the material for this reversible process is then dU = T dS + µ0 Hˆ ⋅ dMˆ . Considering
Ĥ ≡ − p and M̂ ≡ V , we see that it is analogous with = dU T dS − p dV . Similar to the liquid–gas
phase transition we studied at fixed p and T , this phase transition from the superconducting to the
normal state at fixed Ĥ and T can be studied using the Gibbs free energy (per unit volume):
= Gˆ U – TS − µ0 HM ˆ ˆ . (36)
At constant sample volume V , the condition for coexistence of the two phases (normal and
superconducting) at equilibrium along the critical curve Hˆ CM (T ) is that the total free Gibbs energy
tends to a minimum, and so Gibbs free energy is equal in the two phases: Gˆ norm = Gˆ sup . Below the
critical temperature ( T < TCM ), as for point P in Fig. 11(b), equilibrium is reached for Hˆ = Hˆ CM (T ) ,
and thus Gˆ norm (T , Hˆ CM ) = Gˆ sup (T , Hˆ CM ) . For a neighbouring point Q on the critical curve,
Gˆ norm (T +dT , Hˆ CM +dHˆ CM ) =
Gˆ sup (T +dT , Hˆ CM +dHˆ CM ) , and by subtraction dGˆ norm = dGˆ sup . Taking into
account that (per unit volume)
dGˆ = dU – T dS − S dT − µ0 Hˆ ⋅ dMˆ − µ0 Mˆ ⋅ dHˆ = − S dT − µ0 Mˆ ⋅ dHˆ ,
and that the magnetization in the normal state is negligible ( Mˆ norm << 1 ), we obtain
dHˆ CM 1 dHˆ CM
2
( Snorm − Ssup ) =−V ⋅ µ0 ⋅ Hˆ CM ⋅ =−V (unit: J). (37)
dT 2 dT
Experimentally, Hˆ CM (T ) = Hˆ CM (0) − Hˆ CM (0) ⋅ (T / TCM ) 2 (see Fig. 11(b)), dHˆ CM2

/ dT ≤ 0 , and
thus Snorm ≥ Ssup ; in the Meissner state, the material has a lower entropy due to the Cooper pairing of
electrons in a single quantum state. Integrating the Gibbs free energy along an increasing Ĥ , we find
that for the phase in the superconducting state ( Mˆ = − Hˆ )
Hˆ 1
Gˆ sup ( Hˆ , T ) =
Gˆ sup (0, T ) − µ0 ∫ M dH =
Gˆ sup (0, T ) + µ0 Hˆ 2 . (38)
0 2
As Mˆ norm << 1 , the Gibbs free energy of the normal state phase is not dependent on Ĥ . The Gibbs free
energies (per unit volume) of the normal and superconducting phases are plotted in Fig. 11(c) as a
function of the excitation magnetic field. We see that for T < TCM , Gˆ sup < Gˆ norm : the superconducting
state has a lower free energy. However, as T > TCM , Gˆ sup > Gˆ norm , and transition to the normal state
occurs, the phase of minimum free enthalpy.
The entropy difference is related to the latent heat of the phase transition:
dHˆ CM T2  T2 
Lnorm →sup = −V µ0T ⋅ Hˆ CM
( S norm − Ssup )T = =2V µ0 Hˆ CM
2
(0) 2 1 − 2  (unit: J). (39)
dT TCM  TCM 
As T < TCM , Lnorm →sup > 0 ; during the transition from the superconducting to the normal state due to
the magnetic field, the material absorbs a quantity of heat which is provided by this latent heat. This is
also a first-order transition as the first derivative of the Gibbs free energy S = −(∂G / ∂T ) p is not
continuous from one phase to another. At T = TCM , Lnorm →sup = 0 and the entropy is continuous.
18
However, the heat capacity C = δ Q / dT =T dS / dT = −T (∂ 2G / ∂T 2 ) p , a second derivative of the

Gibbs energy, is not continuous:
(Cnorm − Csup ) = −4 µ0VHˆ CM
2
/ TCM . (40)
Therefore at T = TCM the superconducting → normal transition is a second-order transition.
16 Second-order phase transitions

The liquid–gas transition of a fluid and the superconducting–normal state of a type I superconductor
are first-order transitions, except when the temperature reaches the critical temperature: no latent heat
is involved and they become of second-order transitions as the second derivatives of the Gibbs free
energy are not discontinuous.
Continuity of the first derivatives of the Gibbs free energy (per unit of mass) during a second-
order transition between phases I and II implies that dsI = dsII and dvI = dvII . Using Eqs. (15), (16)
and (20) we find that
 ∂si   ∂si  c pi  ∂vi  c pi
dsi =
  dT +   dp = dT −   dT = dT − viα i dp .
 ∂T  p  ∂p T T  ∂T  p T
Similarly, using Eq. (20) and introducing the coefficient of isothermal compressibility
χT = −1 / V (δ V / δ p )T , we obtain
 ∂vi   ∂vi 
dvi =
  dT +  viα i dT − vi χTi dp .
 dp =
 ∂T  p  ∂p T
Hence, from the equality of the derivatives on the coexistence curve, and as there is no change in
volume during the transition, we obtain, for the Ehrenfest formulae,
dp c pI − c pII dp α I − α II
= and = . (41)
dT Tv(α I − α II ) dT χTI − χTII
An example of a second-order transition is the normal–superfluid transition of liquid helium-4.

The equilibrium curve of these two phases in a p − T diagram is called the lambda line. This name
originates from the shape of the heat capacity curve plotted versus temperature, which has the form of
the Greek letter lambda near the transition, as can be seen in Fig. 12(a). The lambda point is the triple
point where the normal phase, the superfluid liquid phase, and the vapour phase coexist. As T
diminishes to Tλ , the heat capacity of He-I (normal state) increases. At T = Tλ , the heat capacity is
discontinuous, as shown in Fig. 12(b). Below Tλ , the heat capacity reaches a maximum and decreases
as T diminishes.
Fig. 12: Specific heat of helium-4 as a function of temperature near the lambda point: (a) large temperature
scale; (b) very close to Tλ (from [1]).
19
P. D UTHIL
It should be noted that higher-order phase transitions can be found in nature, even though they
are less abundant than first- or second-order ones. Their order is defined as the minimum integer i
such that (∂ i G / ∂x i ) I ≠ (∂ i G / ∂xi ) II .
References
[1] J.A. Lipa et al., Phys. Rev. Lett. 76 (1995) 944.
Appendix A: Thermodynamic processes for an ideal gas
Table A.1: Summary of the properties of some reversible processes involving an ideal gas
Process Isochoric Isobaric Isothermal Adiabatic Polytropic

Fixed
V = cst p = cst T = cst δq = 0
parameter
p–v diagram
Relationship pB TB vB TB p=
A vA p=
B vB nRTA pA vAγ = pBvBγ
= = pA vAz = pBvBz
between state pA TA vA TA Laplace law
1< z < γ
variables Charles law Gay–Lussac law Mariotte law γ = c p / cV
ΔU = UB − UA
ncV (TB − TA ) ncV (TB − TA ) ncV (TB − TA ) =
0 ncV (TB − TA ) ncV (TB − TA )
(J)
ΔH = HB − HA nc p (TB − TA ) nc p (TB − TA ) nc p (TB − TA ) =
0 nc p (TB − TA ) nc p (TB − TA )
(J)
Heat QAB VB z −γ
∆U ∆H nRT ln 0 ncv (TB − TA )
(J) VA z −1
Work WAB − p (VB − VA ) VB γ −1

0 −nRTA ln ∆U ncv (TB − TA )
(J) = nR (TB − TA ) VA z −1
ΔS = SB − SA TB TB V VB γ − n VB
ncV ln c p ln = c p ln B nR ln 0 nR ln
(J·K −1) TA TA VA VA γ − 1 VA
20
Introduction to Transverse Beam Dynamics
B.J. Holzer 1
CERN, Geneva, Switzerland
Abstract
In this chapter we give an introduction to the transverse dynamics of the
particles in a synchrotron or storage ring. The emphasis is more on
qualitative understanding rather than on mathematical correctness, and a
number of simulations are used to demonstrate the physical behaviour of
the particles. Starting from the basic principles of how to design the
geometry of the ring, we review the transverse motion of the particles,
motivate the equation of motion, and show the solutions for typical
storage ring elements. Following the usual treatment in the literature, we
present a second way to describe the particle beam, using the concept of
the emittance of the particle ensemble and the beta function, which
reflects the overall focusing properties of the ring. The adiabatic
shrinking due to Liouville’s theorem is discussed as well as dispersive
effects in the simplest case.
Keywords: accelerators, transverse beam dynamics, ring geometry,

adiabatic shrinking.
1 Introduction
The transverse beam dynamics of charged particles in an accelerator describes the movement of single
particles under the influence of external transverse bending and focusing fields. It includes the detailed
arrangement of the accelerator magnets used (for example, their positions in the machine and their
strength) to obtain well-defined and predictable parameters of the stored particle beam. It also
describes methods to optimize the single-particle trajectories as well as the dimensions of the beam as
an ensemble of many particles. A detailed treatment of this field in full mathematical lucidity and
including sophisticated lattice optimizations, such as the right choice of the basic lattice cells, the
design of dispersion suppressors or chromaticity compensation schemes, is beyond of the scope of this
basic overview. For further reading and for more detailed descriptions, therefore, we would like to
refer the reader to more detailed explanations [1–4, 9, 10].
2 Geometry of the ring

Magnetic fields are used in general in circular accelerators to provide the bending force and to focus
the particle beam. In principle, the use of electrostatic fields would be possible as well, but at high
momenta (i.e., if the particle velocity is close to the speed of light), magnetic fields are much more
efficient. The force acting on the particles, the Lorentz force, is given by
F= q·( E + ( v × B ) ) .
1
bernhard.holzer@cern.ch
B.J. H OLZER
As an example let us consider a magnetic B field of 1 T, which is a conservative size. The

resulting Lorentz force acting on a high-energy particle (v ≈ c ≈ 3 × 108 m/s) is
m Vs
F ≈ q·3·108 ·1
s m2
MV
F ≈ q·300
m
This corresponds therefore to an equivalent electric field of 300 MV m−1, which is far beyond
the technical limits, which are set by field breakdown and discharges. In high-energy storage rings, or,
more precisely, as soon as the speed of the particles is large enough, magnetic fields are much more
efficient. It has to be pointed out, however, that for low-energy heavy-ion rings situations might occur
where the speed is low enough to make electrostatic fields more worthwhile.
2.1 The ideal circular orbit

We describe the transverse movement of the particles in a coordinate system rotating with the so-
called ideal particle. As indicated in Fig. 1, the vertical coordinate, describing the displacement of a
particle with respect to the ideal orbit, is called y, the corresponding horizontal one x, and the
coordinate that is pointing into the direction of the longitudinal motion and that is moving with the
particles around the ring is called s.
Fig. 1: Coordinate system used and orbit of an idealized particle
Assuming ideal conditions and a homogeneous dipole field to provide the bending force, the
condition for a circular orbit is given by the equality between the Lorentz force and the centrifugal
force. Neglecting any electrostatic field we get
mv 2
q·v·B =
ρ
In a constant transverse magnetic field B, a particle will see a constant deflecting force and the
trajectory will be part of a circle, whose bending radius ρ is determined by the particle momentum
p = mv and the external B field:
p
B·ρ =
q
The term B·ρ is called the beam rigidity. Inside a dipole magnet, therefore, the bending angle
(sketched in Fig. 2) is
α= ∫Bds . (1)
Bρ
22
I NTRODUCTION TO T RANSVERSE B EAM DYNAMICS
For the lattice designer, the integrated B field along the design orbit of the particles is one the
most important parameters, as it is the value that enters Eq. (1) and defines the field strength and the
number of such magnets that are installed. By requiring a bending angle of 2π for a full circle, we
obtain a condition for the magnetic dipole fields in the ring. Figure 3 shows a photo of a small storage
ring [5], where only eight dipole magnets are used to define the design orbit. The magnets are powered
symmetrically, and therefore each magnet corresponds to a bending angle θ of the beam of exactly
45°. The field strength B in this machine is of the order of 1 T.
Fig. 2: Magnetic B field in a storage ring dipole and, schematically, the particle orbit
Fig. 3: The TSR heavy-ion storage ring at the Max-Planck-Institut in Heidelberg [5]
The lattice and, correspondingly, the beam optics in modern storage rings are usually split into
several different characteristic parts. These include arc structures, which are used to guide the particle
beam and to establish a regular pattern of focusing elements, leading to a regular, periodic beam
dimension. These structures define the geometry of the ring and, as a function of the installed dipole
magnets, the maximum energy of the stored particle beam. The arcs are connected by so-called
insertions, long lattice sections where the optics is modified to establish the conditions needed for
particle injection, to reduce the dispersion function, or to reduce the beam dimensions in order to
increase the particle collision rate, for example, where the beam in a collider ring must be prepared for
particle collisions.
In the case of the Large Hadron Collider (LHC) at CERN, for a momentum p = 7000 GeV/c, 1232
dipole magnets are needed in the arc to define the machine geometry, each having a length of 15 m and a
B field of 8.3 T (Fig. 4). To be very precise, the LHC, as any other storage ring, is not really a ring in the
true sense of the word, but rather a polygon – or, to be even more precise, a 1232-gon.
23
B.J. H OLZER
Fig. 4: The s.c twin-aperture dipole magnets of LHC
In general, we will try to cover a maximum part of the circumference by the main bending
magnets, as they define the highest energy of the machine. For technical reasons, as a general rule
about two-thirds of the circumference of the machine can be used to install the dipoles.
Requiring an overall bending angle of 2π for the complete dipole bending strength and
approximating the integral of the B field by the sum over the magnets, we can calculate the required
B field. In the case of the LHC we get
∮Bdl ≈ N
l B=
2πp / e
2π·7000·109 eV
B≈ 8.3 T
=
8 m
1232·15m·3·10 ·e
s
A part of the LHC tunnel is shown in Fig. 5, where the shape of the ring is hardly visible due to
the large bending radius of ρ = 2.5 km.
Fig. 5: The LHC tunnel with the dipole cryostats. The normalized bending force, represented by the radius of
ρ = 2.5 km, is hardly visible.
24
3 Quadrupole magnets and equation of motion
Once the geometry and specification of the arc have been determined and the layout of the bending
magnets has been done, the next step is to worry about the focusing properties of the machine. In
general, we have to keep more than 1012 particles in the machine, distributed over a number of
bunches, and these particles have to be focused to keep their trajectories close to the design orbit.
Gradient fields generated by quadrupole lenses are used to do this job. These lenses generate a
magnetic field that increases linearly as a function of the distance from the magnet centre:
By =
− g·x , Bx =
− g· y
Here, x and y refer to the horizontal and vertical planes and the parameter g is called the
gradient of the magnetic field. It is customary to normalize the magnetic fields to the momentum of
the particles. In the case of dipole fields, we obtain from Eq. (1)
=α ∫=
Bds 1
·Leff
B·ρ ρ
where Leff is the so-called effective length of the magnet. The term 1/ρ is the normalized bending
strength of the dipole. In the same way, the field of the quadrupole lenses is normalized to Bρ. The
strength k is defined by,
g
k=
B·ρ
and as a consequence the focal length of the quadrupole is given by
1
f =
k·l
As a rule of thumb, we get the normalized gradient k as the ratio between gradient g expressed
in T m−1 and the particle momentum p in units of GeV/c and multiplying by 0.3 as an approximation
of the speed of light:
T 
g 
m
k = 0.3  
 GeV 
p 
 c 
It is worth emphasizing that the gradient g will describe the focusing property of the quadrupole
in both transverse planes. From Maxwell’s equation,
∂E
∇ × B = 0
j+ =
∂t
because at the position of the beam (i.e., inside the quadrupole bore) the density j of the magnet
current is zero and no explicit changing electric field exists. We conclude that
∂By ∂Bx
=
∂x ∂y
25
B.J. H OLZER
To derive the equation of motion of the particles under the influence of the fields described
above, we refer to the linear approximation. In the case of the fields this means that only constant
terms or terms linearly dependent on x (or y) will be taken into account. Mathematically speaking, for
a general Taylor expansion, all higher-order terms will be neglected:
B( x) 1 1 1
= + kx + mx 2 + nx 3 + ...
p e ρ 2! 3!
This approximation is more serious than it might look like. Nonlinear terms would lead to an
equation of motion that cannot be solved analytically any more. For the storage ring design, the
consequence is to optimize the magnets and suppress higher multipoles in the dipole and quadrupole
magnets as well as possible. Moreover, in general, the magnets in modern accelerators will be split
and optimized according to their job: dipoles for bending, and quadrupoles for focusing. Multipoles
like sextupole and octupole magnets will be introduced – if needed – carefully for higher-order
corrections only.
In Fig. 3 we saw already an example of such a separate function storage ring [5] with eight
dipole magnets and 4 × 5 quadrupole lenses for beam focusing.
3.1 Equation of motion
For the derivation of the equation of motion, we start with a general expression for the radial
acceleration as known from classical mechanics:
2
d 2 ρ  dθ 
ar = 2 - ρ  
dt  dt 
The first term refers to an explicit change of the bending radius, and the second one to the centrifugal
acceleration. Referring to our coordinate system, and replacing for the general case the ideal radius ρ
by ρ + x (Fig. 6), we obtain for the balance between radial force and the counter-acting Lorentz force
the relation
d2 mv 2
F m
= ( x + ρ ) − = eBy v
dt 2 x+ ρ
Fig. 6: Ideal circular orbit and real particle trajectory with its transverse coordinates x and y
On the right-hand side of the equation, we take only linear terms of the magnetic field into account,
∂By
By = B0 + x
∂x
and for convenience we replace the independent variable t by the coordinate s,
26
dx dx dt
x′
= =
ds dt ds
to obtain an expression for the particle trajectories under the influence of the focusing properties of the
quadrupole and dipole fields in the ring, described by a differential equation. This equation is derived
in its full beauty elsewhere [6], so we shall just state it here:
x″ + Kx = 0. (2)
Here, x describes the horizontal coordinate of the particle with respect to the design orbit; the
derivative is taken with respect to the orbit coordinate s, as usual in linear beam optics; and the
parameter K combines the focusing strength k of the quadrupole and the weak focusing term 1/ρ2 of
the dipole field. (Note that a negative value of k means a horizontal focusing magnet.) The value of K
is given by
K = −k + 1/ρ2.
In the vertical plane, in general, the term 1/ρ2 is missing, as in most (but not all) accelerators the
design orbit is in the horizontal plane and no vertical bending strength is present. At the same time, the
sign of the gradient changes due to the geometry of the quadrupole field lines, k → −k. So, in the
vertical plane, we have
K = k.
3.2 Single-particle trajectories
The differential equation (2) describes the transverse motion of a particle with respect to the design
orbit. This equation can be solved in a linear approximation, and the solutions for the horizontal and
vertical planes are independent of each other in the sense that the motions in the two transverse planes
are uncoupled.
If the focusing parameter K is constant, which means that we are referring to a position inside a
magnet where the field is constant along the orbit, the general solution for the position and angle of the
trajectory can be derived as a function of the initial conditions x0 and x0′. In the case of a focusing lens,
we obtain
x0′
=x( s ) x0 *cos( K *s) + *sin( K * s ),
K
x′( s ) =
− x0 * K *sin( K * s ) + x0′ *cos( K * s ),
or, written in a more convenient matrix form,

x  x 
 ′  = M · ′ 
 x s  x 0 .
Given the particle amplitude and angle in front of the lattice element, x0 and x0′, we obtain their
values after the element by a simple matrix multiplication. The matrix M depends on the properties of
the magnet, and we obtain the following expressions for three typical lattice elements. Also shown
schematically are the situations for the three cases described.
27
B.J. H OLZER
Focusing quadrupole:
 1 
 cos( K l sin( K l 
M QF = K  (3a)
 
 − K sin( K l cos( K l 
Defocusing quadrupole:
 1 
 cosh( K l sinh( K l 
M QD = K  (3b)
 
 K sinh( K l cosh( K l 
Drift space:
1 l 
M drift =   (3c)
 0 1
3.3 Transformation through a system of lattice elements
Given the matrix description for the lattice elements, it is evident that, to obtain the particle amplitude
at a certain position in the storage ring, we can either start at an initial point and transform the vector
(x, x′) step-by-step through the machine, or – mathematically equivalent, but much more convenient –
determine the product matrix between the two points of interest and apply it to the initial coordinates
(x, x′)0. As an example, we refer to Fig. 7. For simplicity, we consider a storage ring built only out of
focusing and defocusing quadrupoles and dipole magnets in between.
28
Fig. 7: Ideal storage ring consisting of focusing and defocusing quadrupoles and dipole magnets for bending.
(Courtesy of K. Wille [7].)
Starting in front of a focusing magnet, a typical part of this structure, expressed in matrix form,
would be e.g.
M total = M QF * M D * M QD * M Bend * M D*.....
The particle coordinates at point s2 can be obtained by repetitive multiplication of the

corresponding single element matrices, or by applying the product matrix above:
x  x 
=
 ′ M ( s2 , s1 ) ∗  
 x s 2  x′  s1
Fig. 8: Single-particle trajectory in a storage ring (green line). In red, the pattern of the beta function is shown,
which can be interpreted as the maximum beam size or aperture needed in the example.
Using reasonable values for 1/ρ and k, we obtain the picture shown in Fig. 8 for the particle
trajectory. The particle movement – looking zig-zag in the simplified model used here – represents in
each single element the solution of a harmonic oscillator, each having eventually a different restoring
force, still however leading to an overall focusing effect around the storage ring. The overall number
of oscillations, or better the corresponding oscillation frequency of this transverse motion, is of major
importance for beam stability reasons and represents the eigenfrequency of the motion, the so-called
‘tune’. In the example of Fig. 8 we get Q ≈ 1.3.
29
B.J. H OLZER
4 The Twiss parameters α, β and γ
Following the single-particle trajectory as shown in Fig. 8, the natural question will be how the
trajectory of the second turn will look like; and then the third one; and so on. Reproducing the
calculations that lead to the orbit of the first turn will result in a large number of single-particle
trajectories that will overlap somehow and form the beam envelope. Figure 9 shows the result for
50 turns. Clearly, as soon as we talk about many turns, or many particles, the use of the single-
trajectory approach is quite limited and we need a description of the beam as an ensemble of many
particles.
Fortunately, in the case of periodic conditions in the accelerator, there is another way to describe
the particle trajectories that, in many cases, is more convenient than the above-mentioned formalism.
It is important to note that, in a circular accelerator, the focusing elements are necessarily periodic in
the orbit coordinate s after one revolution. Furthermore, storage ring lattices have in most cases an
inner periodicity: they are often constructed, at least partly, from sequences in which identical
magnetic cells, the lattice cells, are repeated several times in the ring and lead to periodically repeated
focusing properties.
In this case, the equation of motion has to be written now in a slightly different form:
x′′( s ) − k ( s ) x( s ) =
0
Unlike the treatment above, the focusing parameters – or the restoring forces – are no longer
constant, but are functions of the coordinate s. However, they are periodic in the sense that, at least
after one full turn, they repeat each other, k(s + L) = k(s), leading to the special so-called Hill
differential equation. Following Floquet’s theorem, the solution of this equation can be written in its
general form as [2]
𝑥(𝑠) = √𝜀�𝛽(𝑠)cos(𝜓(𝑠) + 𝜙) (4)
−√𝜀
𝑥 ′ (𝑠) = sin(𝜓(𝑠) + 𝜙) + 𝛼(𝑠)cos(𝜓(𝑠) + 𝜙)
𝛽(𝑠)
The position and angle of the transverse oscillation of a particle at a point s are given by the
value of a special β-function at that location, and ε and δ are constants of the particular trajectory. The
β-function depends in a quite sophisticated manner on the overall focusing properties of the storage
ring. It cannot be calculated directly in an analytical approach, but it has to be either determined
numerically or deduced from properties of the single-element matrices described above (see e.g. [8]).
In any case, like the lattice itself, it has to fulfil the periodicity condition
β ( s + L) =
β (s)
Inserting the solution (4) into the Hill equation and rearranging slightly, we get
s
ds
ψ (s) = ∫ (5)
0
β (s)
which describes the phase advance of the oscillation. It should be emphasized that ψ depends on the
amplitude of the particle oscillation. At locations where β reaches large values, i.e., the beam has a
large transverse dimension, the corresponding phase advance will be small; and vice versa, at locations
where we create a small β in the lattice, we will obtain a large phase advance. In the context of Fig. 8
we introduced the tune as the number of oscillations per turn, which is nothing other than the overall
30
phase advance of the transverse oscillation per revolution in units of 2π. So by integrating Eq. (5)
around the ring, we get the expression
1 ds
Qy =
2π ∫ β (s)
Fig. 9: Beam as an ensemble of many single-particle trajectories. Left: transverse shape of a typical particle
beam. Right: overlapping of many particle trajectories.
The practical significance of the β-function is shown in Fig. 9. The left part shows schematically
the transverse shape of the beam inside the vacuum chamber (red) and the hyperbolic pole shoe profile
of the quadrupole lens. On the right-hand side the single-particle trajectories are shown. The envelope
of the overlapping trajectories is given by 𝑥� = �𝜀𝛽(𝑠) and is used to define the beam size in the
sense of a Gaussian density distribution.
The integration constant ε has a clearly defined physical interpretation. Given the solution of
Hill’s equation
=x( s ) ε β ( s ) cos(ψ (s) + φ ) (6)
and its derivative
ε
x′( s ) =− {α ( s) cos(ψ ( s) + φ ) + sin(ψ ( s) + φ )} (7)
β ( s)
we can transform Eq. (6) to

x( s )
cos(ψ ( s ) + φ ) =
ε β (s)
and insert the expression into Eq. (7) to get an expression for the integration constant ε:
γ ( s ) x 2 ( s ) + 2α ( s ) x( s ) x′( s ) + β ( s ) x′2 ( s )
ε= (8)
Here we follow the usual convention in the literature and introduce the two parameters
31
B.J. H OLZER
−1
α (s) = β ′( s )
2
(9)
1 + α ( s)2
γ (s) =
β (s)
We obtain for ε a parametric representation of an ellipse in x–x′ ‘phase’ space, which according
to Liouville’s theorem is a constant of motion, as long as conservative forces are considered. The
mathematical integration constant thus gains physical meaning.
Fig. 10: A single-particle trajectory …
Fig. 11: … and its phase space coordinates at a position s turn by turn
Referring again to the single-particle trajectory, discussed above (see Fig. 10), but now plotting
at a given position s in the ring the coordinates x and x′ turn-by-turn, we get the picture shown in
Fig. 11. Plotted in phase space the particle will follow the shape of an ellipse whose shape and
orientation are defined by the optical parameters at the reference position s in the ring. Each point in
Fig. 11 represents the coordinates for a certain turn at that position in the ring, and the particle will
have performed from one turn to the next a number of revolutions in phase space that corresponds to
its tune. It has to be emphasized that, as long as conservative forces are considered (i.e., no interaction
between the particles in a bunch, no collisions with remaining gas molecules, no radiation effects,
etc.), the size of the ellipse in x–x′ space is constant and can be considered as a quality factor of the
single particle. Large areas in x–x′ space (where, to be exact, the area is given as A = πε) mean large
amplitudes and angles of the transverse particle motion, and we would consider it as a bad particle
‘quality’.
32
To discuss the dependence of the shape of the ellipse on the beam optics in a bit more detail, we
replace the two parameters α and γ by their corresponding expressions in Eq. (9),
x2 α 2 x2
ε= + + 2α ·xx′ + β ·x′2
β β
solve for x′
−α ·x ± εβ − x 2
′ =
x1,2
β
and determine the maximum values of the particle angle via

dx′
=0
dx
to obtain for the maximum angle x′ and the position x at that point:
xˆ ′ = εγ , x = ±α ε
γ
For completeness and to demonstrate the conservation of phase space area, we plot in Fig. 12
the emittance invariants of four particles in phase space. For the position s this time the centre of a
focusing quadrupole has been chosen, where the orientation of the ellipses is flat, due to the fact that at
this location the beam will have its largest size, i.e., α = 0. Turn by turn the coordinates of the four
particles will lie on the corresponding ellipses and never cross each other.
Fig. 12: Phase space ellipses at the location of a focusing quadrupole for four particles
33
B.J. H OLZER
4.1 The emittance of a particle ensemble

The ensemble of many single particles, as shown in Fig. 13, forms a pattern of overlapping trajectories
that in the end we will observe as transverse intensity (or charge) distribution and that we will use to
define the beam size. Along the storage ring coordinate s, its transverse size, in the sense of the
maximum amplitude of a trajectory that will be observed at a given location, is defined by the β-
function.
Fig. 13: The overlapping trajectories of many single particles define the beam cross-section as defined by ε
and β.
As the general solution of the equation of motion is given by
=x( s ) ε β ( s )·cos(Ψ( s ) + φ )
the maximum amplitude, or beam size, is obtained from
xˆ ( s ) = ε β ( s )
and depends via the β-function on the focusing properties of the lattice and via ε on the quality of the
particle ensemble.
In many cases, the transverse particle density of the particles follows a Gaussian distribution.
Referring to a particle within this distribution that is situated at one standard deviation σ, the ε
parameter (sometimes called the Courant–Snyder invariant) of this particle can be used as
representative of the complete beam. In this sense we talk about a beam emittance and thus about a
general quality factor of the whole particle ensemble. Referring to the example of the LHC we have, in
the arc of the storage ring at flat-top energy, the following values:
β = 180 m
ε = 5·10−10 m·rad
and the beam size (1σ) is
=σ ε ·β · 180 m·5·10−10 m·rad 0.3 mm
=
Figure 14 shows the result of a measurement of the transverse beam size. The points represent
the measurement values, the curve a Gaussian fit, to obtain the sigma of the distribution as a number to
qualify the beam dimension. In general, we will define the aperture dimensions of the vacuum
chamber as a certain multiple of this beam sigma. In the case of the LHC, for example, we require a
minimum aperture of r0 = 18 σ inside the mini-beta quadrupoles.
34
Fig. 14: Measured transverse particle density in a beam and a Gaussian fit to obtain the σ of the distribution
5 Liouville’s theorem during acceleration

There is a special issue concerning the phase space area of a particle beam as soon as we start to
accelerate. It has been explained that the expression of the beam emittances
γ ( s ) x 2 ( s ) + 2α ( s ) x( s ) x′( s ) + β ( s ) x′2 (s)
ε=
represents an ellipse in the x–x′ space, which we call in a rather sloppy manner ‘phase space’.
The real phase space, however, as defined in classical mechanics, relates the position x with its
canonical conjugate momentum px. It is the area in this real phase space that is conserved according to
Liouville’s theorem. So if
∫ p dx = const
x
and rewriting for the angle

dx dx dt β x
x′
= = =
ds dt ds β
which is the ratio of the relativistic β-parameter between the transverse motion and the longitudinal
one, we get from Liouville’s theorem
with the expression in the integral being our ‘phase space’. Still, it is true that the beam emittance is
constant for a given energy, but as soon as we start to accelerate ε will shrink as
1
=ε ∫ x′dx ∝ βγ
where γ and β are the relativistic parameters and should not be confused with the beam optics
functions.
As a consequence, it turns out that a proton machine or an electron linac will need the highest
aperture at injection energy. As soon as we start to accelerate, the beam size shrinks as (βγ)−1/2 in both
transverse planes. At lowest energy the machine will have major aperture problems, and here we have
35
B.J. H OLZER
to minimize the beta function to obtain a beam envelope that fits into the vacuum chamber. At high
energy this aperture restriction is more relaxed and we can develop optics solutions that allow for
higher values of β. For example, a mini-beta concept that, as we will see, leads to extreme beta values
in the focusing magnets will only be adequate at flat top. An example is shown in Fig. 15 for the case
of the HERA proton storage ring. The particles were injected into this machine at an energy of 40 GeV
and after acceleration reached a flat-top energy of 920 GeV. For the same beam optics, the figure
compares the situation at injection energy and at flat top. Owing to the factor of 23 in energy increase,
the beam size shrinks by nearly a factor of 5 in both planes.
Fig. 15: Transverse beam envelope for design parameters of the HERA proton ring at injection and flat top. For
the same optics, the beam size shrinks considerably during the acceleration.
In the case of the LHC, the two different beam optics optimized following these needs are
shown in Fig. 16. While at injection energy of 450 GeV the maximum beta values have to be limited
to 500 m, much higher values are possible at high energy, and close to the mini-beta insertions a
maximum of β = 4.5 km is possible.
Fig. 16: LHC beam optics optimized for injection (left) and luminosity (right). The large beta functions needed
for luminosity operation can only be applied at flat-top energy where the beam emittance is considerably
reduced.
36
6 Dispersive effects
Until now we have treated the beam and the equation of motion as a mono-energetic problem.
Unfortunately, in the case of a realistic beam, we have to deal with a considerable distribution of the
particles in energy or momentum. Typical values are
Δp
≈ 1.0 10-3
p
This momentum spread will lead to several effects concerning the bending of the dipole
magnets and the focusing strength of the quadrupoles. It turns out that the equation of motion, which
was a homogeneous differential equation until now, will get a non-vanishing term on the right-hand
side:
1 Δp 1
x′′ + x( 2
− k) = ·
ρ p ρ
The general solution therefore is the sum of the solution of the homogenous equation of motion
plus a special solution of the inhomogeneous one:
x( s ) xh ( s ) + xi ( s )
=
Here xh is the solution that we have discussed until now and xi is an additional contribution that
has yet to be determined. For convenience, we usually normalize this second term and define a
function, the so-called dispersion:
xi ( s )
D( s) =
∆p
p
This describes the dependence of the additional amplitude of the transverse oscillation on the
momentum error of the particle. In other words it fulfils the condition
1 ∆p
xi′′( s ) + Κ ( s )·xi ( s ) =·
ρ p
The dispersion function is usually calculated by optics programs in the context of the
calculation of the usual optical parameters and is of equal importance. Analytically it can be
determined for single elements via
s1 s1
1 1
=D( s ) S ( s ) ∫ C ( s )ds − C ( s ) ∫ S ( s )ds
s0
ρ s0
ρ
where S(s) and C(s) correspond to the sine-like and cosine-like elements of the single-element
matrices in the lattice [6].
Typical values in the case of a high-energy storage ring are
xβ = 1–2 mm, D(s) = 1–2 m
and, for a typical momentum spread of ∆p / p ≈ 10−3 , we obtain an additional contribution to the beam
size from the dispersion function that is of the same order as the one from the betatron oscillations xβ.
An example of a high-energy beam optics including the dispersion function is shown in Fig. 17.
37
B.J. H OLZER
Fig. 17: Beam optics function β(s) (top) and dispersion function D(s) (bottom) for a part of a typical collider
ring. The dispersion must vanish at the collision point in the middle of the horizontal axis.
It should be pointed out that the dispersion describes that special orbit that an ideal particle
would have in the absence of no betatron oscillations (xβ = xβ′ = 0) for a momentum deviation of Δp/p
= 1. Still, it describes ‘just another particle orbit’ and so it is subject to the focusing forces of the
lattice elements, as seen in Fig. 17.
7 Mini-beta insertions and luminosity

The straight sections of a storage ring are often designed for the collision of two counter-rotating
beams. For a given overall cross-section σR of the particle collision (i.e., the probability for a physics
process to occur during the interaction of the particles), the event rate of the collision process is
determined by the so-called luminosity of the storage ring.
R = σ R ·L
Schematically the situation is shown in Fig. 18.
Fig. 18: Colliding bunches and luminosity
The luminosity is characterized by the storage ring parameters and depends on the stored beam
current and the transverse size of the colliding bunches at the interaction point:
1 I p1 I p 2
=L ∗
4π e f 0 nb σ xσ y
2
Here Ip1 and Ip2 are the values of the stored beam currents, f0 is the revolution frequency of the
machine, and nb is the number of stored bunches. The quantities σx and σy in the denominator are the
beam sizes in the horizontal and vertical planes at the interaction point. For a high-luminosity collider,
the stored beam currents have to be large and, at the same time, the beams have to be focused at the
interaction point to very small dimensions. The β-functions at the collision points are therefore very
38
small compared with their values in the arc cells. Typical values are more in the range of centimetres
than of metres. At the same time, a large drift space is needed where focusing elements cannot be
installed due to the particle detector that will be placed around the interaction point (Fig. 19).
As an example, the parameter list of the LHC including the design luminosity is presented:
=β x*, y 0.55
= m f 0 11.245 kHz
=ε x , y 5·10−10 m·rad
= = L 1.0·1034 cm −2s −1
nb 2808
=σ x , y 17
= µ m I p 584 mA
Fig. 19: Particle detector of the ATLAS collaboration at the LHC storage ring
Figure 20 shows the typical layout of such a mini-beta insertion. It consists in general of:
– a symmetric drift space that is large enough to house the particle detector and whose beam
waist (where α0 = 0) is centred at the interaction point (IP) of the colliding beams;
– a quadrupole doublet (or triplet) installed on each side as close as possible to the IP;
– additional quadrupole lenses to match the optical parameters of the mini-beta insertion to the
optical parameters of the lattice cell in the arc.
Fig. 20: Layout of a mini-beta insertion scheme
Following Liouville’s theorem, the request of a small beam size immediately will lead to a
corresponding increase of beam divergence. Figure 21 shows the corresponding situation in phase
space. It compares the x–x′ ellipse in a typical location of the storage ring with the situation obtained at
the IP.
39
B.J. H OLZER
Fig. 21: Phase space ellipse in the arc of a storage ring and at the interaction point
While the beam size shrinks with reduced β as 𝜎 = �𝜀𝛽, the corresponding divergence is
increased as 𝜎 ′ = �𝜀 ⁄𝛽. More quantitatively, the behaviour of the beta function in the vicinity the IP
can be calculated and we obtain within the symmetric drift space of the mini-beta insertion
12
β ( s=
) β0 + .
β0
Inevitably, the strong focusing that is needed to obtain the smallest spot size of the beam leads
to a large increase of the beam inside the first quadrupoles after the IP. It is at this location that we will
need the strongest and at the same time the largest aperture quadrupoles – this is the price that we have
to pay for the luminosity that we will get out of the design.
References and further reading

[1] P. Bryant and K. Johnsen, The Principles of Circular Accelerators and Storage Rings
(Cambridge University Press, Cambridge, 1993).
[2] F. Hinterberger, Physik der Teilchenbeschleuniger (Springer, Berlin, 1997).
[3] M. Sands, The Physics of e+e− Storage Rings, SLAC Report 121 (1970).
[4] D. Edwards and M. Syphers, An Introduction to the Physics of Particle Accelerators,
Superconducting Super Collider Laboratory Report (1990).
[5] E. Jaeschke et al., in European Particle Accelerator Conference, Rome (1988), p. 365.
[6] P. Schmüser, Basic course on accelerator optics, in CERN Accelerator School: 5th General
Accelerator Physics Course, CERN 94-01 (1994).
[7] K. Wille, Physics of Particle Accelerators (Oxford University Press, Oxford, 2005).
[8] B. Holzer, Lattice design, in CERN Accelerator School: Intermediate Accelerator Physics
Course, CERN 2006-002 (2006).
[9] A. Chao and M. Tigner, Handbook of Accelerator Physics and Engineering (World Scientific,
Singapore, 1999).
[10] M. Reiser, Theory and Design of Charged Particle Beams (Wiley-VCH, Weinheim, 2008).
40
Introduction to Longitudinal Beam Dynamics
B.J. Holzer 1
Abstract
This chapter gives an overview of the longitudinal dynamics of the particles
in an accelerator and, closely related to that, of the concept of
synchronization between the particles and the accelerating field. Beginning
with the trivial case of electrostatic accelerators, the synchronization
condition is explained for a number of driven accelerators like Alvarez
linacs, cyclotrons and finally synchrotrons and storage rings, where it plays a
crucial role. In the case of the latter, the principle of phase focusing is
motivated qualitatively as well as on a mathematically more correct level
and the problem of operation below and above the transition energy is
discussed. Throughout, the main emphasis is more on physical
understanding rather than on a mathematically rigorous treatment.
Keywords: accelerators, longitudinal beam dynamics, cyclotrons,

synchrotrons.
1 Introduction
The longitudinal movement of the particles in a storage ring – and, closely related to this, the
acceleration of the beam – is strongly related to the problem of synchronization between the particles
and the accelerating system. This synchronization might be established via the basic hardware and
design of the machine (like in a Wideroe structure), via the orbit (in a cyclotron), or in a more
sophisticated way it can be a fundamental feature of the ring, which leads in the end to the name
‘synchrotron’ for this special machine. We will treat these different aspects in more detail. But, before
we do so, we would like to start on a very fundamental ground and at the same time go back a bit in
history.
2 Electrostatic machines
The most prominent example of these machines, besides the Cockcroft–Walton generator, is the Van
de Graaff accelerator. A sketch of the principle is shown in Fig. 1.
Using a moderate d.c. high voltage, charges are sprayed on a moving belt or chain with isolated
links and transported to a kind of Faraday screen, where the charges accumulate, leading to
considerable high voltages. Basically, it is the transformation of mechanical energy into electrostatic
energy via the movement of the belt that defines the physical principle behind the generator. The high-
voltage part is usually connected to a particle source (protons or heavy ions) and the potential
difference between the high-voltage end and the ground potential will define the particle energy.
Voltages in the range of some megavolts can be achieved, where usually a discharge-suppressing gas
is needed to achieve the highest possible terminal voltages (up to ≈ 30 MV). By their design concept,
these machines deliver an excellent energy resolution, as basically each and every particle sees the
1
B.J. H OLZER
same accelerating potential (which is no longer the case as soon as we have to consider a.c.
accelerating structures).
Fig. 1: Technical principle of a Van de Graaff accelerator

An example of such a machine is shown in Fig. 2. This ‘tandem’ Van de Graaff accelerator uses
a stripper foil in the middle of the structure, thus, in the first half of the structure, accelerating negative
ions that are produced in a Cs-loaded source. After stripping the electrons, the same voltage is applied
in the second part of the structure to gain another step in energy. However, after leaving the
accelerator at the down end, the particle energy is still determined by the high voltage that can be
created, and that is finally always limited due to discharge effects of the electric field.
The kinetic energy of the particle beam is given by the integration over the electric field E, in
direction z of the particle motion and measured as usual in units of electronvolts (eV):
W = e ∫Ez ds = [ eV ]
dW = eEz ds
→
Fig. 2: Tandem Van de Graaff accelerator of the MPI Institute, Heidelberg. The ion source, delivering negative
ions, is connected to the accelerator on the right-hand side of the picture. The acceleration structure is housed in
a vessel filled with discharge suppressing gas (SF6). At the far end, the dipole spectrometer magnet is visible to
separate different ion species.
It should be emphasized here that a special synchronization between accelerating voltage and
particle beam is not needed as a d.c. voltage is used for acceleration.
42
I NTRODUCTION TO L ONGITUDINAL B EAM DYNAMICS
3 The first radio-frequency accelerator: Wideroe linac

The basic limitation given by the maximum achievable voltage in electrostatic accelerators can be
overcome by applying a.c. voltages. However, a more complicated design is needed to prevent the
particles from being decelerated during the negative half-wave of the radio-frequency (RF) system. In
1928 Wideroe presented for the first time the layout of such a machine (and built it). Figure 3 shows
the principle.
Fig. 3: Schematic layout of an RF linear accelerator (drift tube linac) as proposed by Wideroe in 1928
The arrows show the direction of the electric field for a given moment in time (i.e., the positive
half-wave of the a.c. voltage) and the corresponding polarity applied to the electrodes indicated by the
‘+/−’ sign in the figure. In principle, arbitrarily high beam energies can be achieved by applying the
same voltage to a given number of electrodes, provided that the particle beam is shielded from the
electric field during the negative RF half-wave. Accordingly the electrodes are designed as drift tubes
(‘drift tube linac’) whose length is adopted according to the particle velocity and RF period.
Schematically the problem is shown in Fig. 4.
Fig. 4: During the negative half-wave of the RF system the particles have to be shielded to avoid deceleration
The red area corresponds to the time during which the particle has to be shielded from the
decelerating field direction and is defined by the RF period: tshield = τRF/2. Accordingly the drift tube
length has to be
τ rf
li = vi ·
2
If the kinetic energy (in the classical regime) is given by

1
Ei = mv 2
2
43
B.J. H OLZER
we obtain an equation for the drift tube length that, for a given accelerating voltage U0 and charge q,
depends on the RF frequency νRF and the accelerating step i:
1 i·q·U 0 ·sinψ s
li = · (1)
ν rf 2m
The parameter ψs describes the so-called synchronous phase and can be chosen to be 0° in this
case to obtain highest acceleration performance.
For completeness, it should be mentioned that in 1946 Alvarez improved the concept by
embedding the structure inside a vessel to reduce RF losses. The drift tube linac is typically used as an
accelerator for proton and heavy-ion beams, and beam energies of the order of some tens of Mega
electronvolt (MeV) are obtained. At the time of writing, a new proton linac is installed at CERN to
improve the beam performance for the LHC beams and – as you would expect – again an Alvarez drift
tube linac is used to gain in three stages a kinetic energy of the protons of 150 MeV.
One of the best-known examples of such an accelerator is running at GSI in Darmstadt, used as
a universal tool for the acceleration of (almost) any heavy-ion species. In Fig. 5 the inner structure is
shown, with the drift tubes and the surrounding vessel.
Fig. 5: ‘Unilac’ of GSI
Unlike the d.c. accelerators, synchronization has to be obtained between the particles and the
accelerating RF field, which – as shown above – is represented by the drift tube length, and so in a
certain sense built into the hardware of the system.
4 Cyclotrons
In principle, the drift tube linac concept can be applied to relatively high energies (doing the
calculations in Eq. (1), relativistically correct even beyond the classical regime). The length of the drift
tubes, however, makes the design very inefficient for high energies, and more sophisticated ideas are
used to keep the structure compact and within reasonable financial and technical limits. The idea that
is considered as a next natural step is the application of a magnetic dipole field to bend the particle
orbit into a circle or, better (as we will see), into a spiral.
44
A constant dipole field applied perpendicular to the particle motion will lead to a transverse
deflecting Lorentz force that is given by
F = q·( v × B ) = q·v·B
where we have replaced the cross-product by a simple multiplication.

The condition for a circular orbit is given by the equality of this Lorentz force and the
centrifugal force, and leads to a relation that is of major importance for any circular machine:
m·v 2
q·v·B = → B·R = p / q
(2)
R
The expression B∙R on the right-hand side is called the beam rigidity, as it describes the
resistance of the particle beam to any external deflecting force and clearly depends on the particle
momentum. At the same time it defines – for a given size of the machine (R) and for a given maximum
magnetic field (B) of the dipoles installed – the highest momentum that can be carried by the (circular)
accelerator.
The revolution frequency of the particles
ν q
ω=
z = ·Bz (3)
R m
is constant as long as we can consider the classical approach, i.e., as long as the mass of the particles
can be assumed to be constant. Accordingly the RF frequency applied for the acceleration can be kept
constant and – in the ideal case – set equal to a multiple of the revolution frequency to obtain particle
synchronization.
A schematic layout of a cyclotron is shown in Fig. 6. The accelerating RF voltage is applied
between the two halves of the pill-box-like structure, leading to a step in energy (or better momentum)
twice per revolution.
Fig. 6: Schematic layout of a cyclotron with the two half-pill-box-like electrodes that contain the spiralling orbit,
and the external magnetic field line indicated.
According to Eq. (2), the radius of the particle orbit will increase with increasing momentum
due to the constant magnetic field, and a spiralling orbit is obtained. The maximum energy achievable
45
B.J. H OLZER
in a cyclotron is determined by the strength and geometrical size of the dipole magnet. An example of
such a classical cyclotron is shown in Fig. 7. The vacuum chamber of the cookie-like box, the magnet
coils and the dipole magnet are clearly visible.
Fig. 7: Cyclotron SPIRAL at Ganil

Another limitation, however, should be mentioned. While the revolution frequency of the
particles can be considered as constant in the classical treatment of Eq. (3), the situation changes as
soon as a considerable mass increase is obtained due to the relativistic particle energies. The correct
expression for the revolution frequency ωs , and thus as well for the RF frequency ωRF, therefore is
q
ω
= s (t ) rf ( t )
ω= ·B
γ ( t )·m0
where γ describes the time-dependent relativistic parameter. During the acceleration process the
particle frequency slows down, and to keep the particle motion synchronous with the applied RF
frequency the external RF system has to be tuned to follow this effect. While higher particle energies
can be obtained, the performance of the machine will degrade as a d.c.-like beam acceleration is no
longer possible due to the RF cycling that is needed for each acceleration process.
As before, to refer to the problem of synchronization: the synchronous condition between
particles and accelerating RF voltage in a cyclotron is defined by the length of the spiralling orbit.
5 Radio-frequency resonators and transit time factor

Synchrotrons are ring-like accelerators that use located RF systems for particle acceleration and,
owing to their design principle, can achieve up to now the highest particle energies. The RF voltage –
much like in the case of a drift tube linac – is usually created in a so-called resonator where a standing
RF wave is created, with an electric field vector pointing in the direction of particle motion (Fig. 8).
Fig. 8: Pill box cavity with longitudinal electric field
46
Before going into the detail of the particle dynamics in a synchrotron, a problem has to be
mentioned that is present in all RF-like structures. The gain in kinetic energy (which is equal in this
case to the gain in overall energy) in a d.c. accelerating system is given by
dW
= dE
= eEz ds
with Ez describing the longitudinal component of the electric field, and the overall energy gain W is
obtained via a simple integration over the complete path ds of the particle:
= ∫ Ez ds eV
W e=
The situation changes quite a bit if RF voltage is applied. The electric field is changing as a
function of the RF frequency used:
V
= 0 cos ωt
Ez E= cos ωt
g
and the total energy gain obtained is given by

g 2
eV z
∆W =
g −g
∫ 2
cos ω dz
v
Fig. 9: Typical RF resonator with calculated field vector and gap size g
As usual z refers here to the longitudinal direction, V is the maximum amplitude of the applied
voltage and v is the speed of the particle. The integration refers to the complete distance between the
cavity walls (the gap), and we refer for symmetry reasons to the centre of the RF resonator. Solving
the integral we obtain
sin θ 2
∆W eV = eVT
=
θ 2
The parameter θ is called transit angle and is defined as

ωg
θ=
v
and the transit time factor T, describing the effective available accelerating voltage in an RF system, is
sin θ 2
T=
θ 2
The ideal case of a transit time factor approaching 1 is obtained if the argument θ /2 is as small
as possible, ideally θ /2 → 0, which corresponds either to ω → 0 as in the case of d.c. acceleration or
47
B.J. H OLZER
to g → 0. This means that we obtain the most effective use of the RF system for the smallest gap
distance of the cavity. The simple case shown in Fig. 8 therefore is not an ideal situation. An improved
cavity design will rather look like the one in Fig. 10, where the gap is reduced to the limit of
discharges that will occur if the resulting electric field is pushed too high.
Fig. 10: Example of an optimised cavity design to increase the transit time factor
6 Synchrotrons
A synchrotron is a circular accelerator or storage ring with
• a design orbit of constant radius defined by the arrangement and strength of a number of
dipole magnets that according to Eq. (2) define the particle rigidity (or momentum),
• an RF system, located at a distinct place in the ring and powered at an RF frequency that is
equal to the revolution frequency of the particles or an integer multiple (so-called harmonic)
of it.
For the description of the particle dynamics, we refer to a synchronous particle of ideal energy,
phase and energy gain per turn. As we will see, the synchronization between the RF system and the
particle beam is of major importance in this machine and has to be explicitly included in the design.
To understand its principle, we have to refer briefly to the transverse dynamics of a particle with
momentum error and the resulting dispersive effects (Fig. 11).
Fig. 11: Trajectories for particles of design energy and off momentum
48
While the ideal particle will run on the design orbit defined by the dipole magnets and will
proceed a distance ds, a non-ideal particle running on a displaced orbit (displaced to the outer side of
the ring in the example of Fig. 11) will pass the corresponding distance dl:
dl ρ + x
=
ds ρ
Solving for dl we obtain

 x 
dl= 1 +  ds
 ρ (s) 
and integrating around the machine we get the orbit length of the non-ideal particle, which depends on
the radial displacement x:
 x∆E 
l=
∆E ∫=
dl ∫  ρ (s) ds
 1 +
where we assume that the radial displacement xΔE is caused by a momentum error and the dispersion
function of the magnet lattice
∆p
x∆E ( s ) = D( s ) (4)
p
We obtain an expression for the difference in orbit length between the ideal and dispersive particle,
which is determined by the amount of relative momentum error and the dispersion function in the
storage ring:
∆p  D( s ) 
p ∫  ρ ( s) 
δ l∆E =  ds
The ratio between relative orbit difference and relative momentum error is called the
momentum compaction factor αp and using Eq. (4) it is determined by the integral of the dispersion
function around the ring and the bending radius of the dipole magnets:
δ lε ∆p
=αp
L p
1  D( s) 
L ∫  ρ ( s ) 
αp =  ds
For first estimates, let us assume equal bending radii in all dipoles, so 1/ρ = const and we
replace the integral of the dispersion around the ring by a sum over the average dispersion in the
dipole magnets (outside the dipoles the term 1/ρ = 0, so this assumption is justified for a rough
estimate):
∫ D ( s ) ds =
dipoles
lΣ( dipoles ) · D dipole
So we get a nice and simple expression for the momentum compaction factor that depends only
on the ratio of average dispersion and geometric radius R of the machine:
1 1 1 1
=αp = l∑ ( dipoles ) · D 2πρ· D
L ρ L ρ
49
B.J. H OLZER
2π D
=αp D ≈
L R
Assuming finally that the particles are running at the speed of light, v ≈ c = const, the ratio
between relative error in time is given by the relative change of the orbit length and thus by the
momentum compaction factor and the relative momentum error:
δT δ lε ∆p
= = αp (5)
T L p
6.1 Dispersive effects in synchrotrons

In the case of non-relativistic particles, however, and the problem of synchronization, we need a more
careful treatment, which will give us a relation between the revolution frequency and the momentum
error of the particles. The parameter of interest is the ratio between the relative momentum error and
the relative frequency deviation of a particle:
df r dp
=η
fr p
Given the revolution frequency as a function of machine circumference and speed,

βc
fr =
2π R
we obtain via logarithmic derivation
df r d β dR
= − (6)
fr β R
Remembering that the relative change in radius, i.e., the second term in the expression, is given
by the momentum compaction factor αp ,
dR dp
=α
R p
The momentum is related to the particle energy and its velocity

E0
= βγ
p mv
=
c
and its relative error is obtained as

1
dp d β d (1 − β ) 2
2 −
−1 d β
p
=
β
+ 1 (
=1− β 2 )
β
(1 − β 2 ) 2
−
Introducing the last two equations into Eq. (6), we finally obtain the required relation between
frequency offset and momentum error
df r  1  dp
=  2 −α  (7)
fr  γ  p
50
which for ultra-relativistic particles reduces to the simplified expression of Eq. (5). Accordingly the η
parameter defined above is given as
1
η
= −α
γ2
As an important remark we state that the change of revolution frequency depends on the particle
energy γ and possibly changes sign during acceleration. Particles get faster in the beginning and arrive
earlier at the cavity location (classical regime), while particles that travel at v ≈ c will not get faster
any more but rather get more massive and, being pushed to a dispersive orbit, will arrive later at the
cavity (relativistic regime). The boundary between the two regimes is defined for the case where no
frequency dependence on dp/p is obtained, namely, η = 0 and the corresponding energy is called the
‘transition energy’:
1
γ tr =
α
In general we will design machines in such a way as to avoid the crossing of this transition
energy. As it involves changes of the RF phase unless, the particles lose their longitudinal focusing
created by the sinusoidal RF function, the bunch profile will dilute and get lost. Qualitatively the
longitudinal focusing effect and the problem of gamma transition is explained in Fig. 12.
Fig. 12: Qualitative picture of the phase focusing principle below transition
6.2 The classical regime

Assume an ideal particle that is passing the cavity in time (or phase) the resonator at a certain position,
as indicated by the green spot in Fig. 12. It will see a certain accelerating voltage and correspondingly
receive an energy increase. We call this phase the synchronous phase, as indicated in the plot. A
particle that has smaller energy than the ideal one will travel at lower speed and arrive after the next
turn later, or at a larger phase and thus see a stronger accelerating voltage. It will therefore compensate
the lack in energy and step-by-step come closer to the ideal particle. Just the opposite happens to a
particle that has a positive energy offset. As it is faster than the synchronous particle, it will arrive
earlier at the cavity, see a smaller voltage and again step-by-step will approach the ideal particle. In
both cases a net focusing effect is obtained, which is due to the relation between momentum and speed
and the right choice of the synchronous phase.
For highly relativistic particles the same effect exists but based now on the mass increase with
energy. As visualized in Fig. 13 the high-energy particle (marked in blue) will, due to its higher mass,
run on a longer orbit and – as its speed is constant v ≈ c – it will arrive later at the cavity location.
51
B.J. H OLZER
Fig. 13: Phase focusing situation above transition

As a consequence, the synchronous phase has to be chosen depending on whether we are
running the machine below or above transition. Synchrotrons that have to pass through transition will
have to apply a phase jump to keep the particles bunched.
In this context it is worth having a look at the acceleration mechanism itself. The particle
momentum is defined via the beam rigidity by the dipole field B,
p = eB ρ
and as a consequence a change in particle momentum is reflected by an appropriate change of the B

field:
dp
= eρ B
dt
The momentum increase per turn is therefore given by


 = 2π eρ RB
(∆p )turn= eρ BTr
v
and referring to the energy change rather than the change in momentum we get with
E 2 = E02 + p 2 c 2 ⇒ ∆E = v∆p
the change in energy per turn, which is clearly related to the accelerating voltage and the synchronous
phase ϕs of the particles:
∆Eturn =
∆Wturn = eVˆ sin φs
2π eρ RB =
The following are some remarks that might be worth emphasizing:

• The energy gain depends on the rate of change of the dipole field.
• The number of stable synchronous particles is equal to the harmonic number h. They are
equally spaced along the circumference.
• Each synchronous particle satisfies the relation p = eBρ. They have the nominal energy
and follow the nominal trajectory.
• As long as the particles are not fully relativistic yet, their revolution frequency will
change and so does the RF frequency, which is a multiple of fr to stay in synchronization
during the complete acceleration process.
52
6.3 Frequency change during acceleration

The relation between the revolution frequency and RF frequency is defined by the harmonic number
and depends on the size of the ring and the magnetic dipole field:
f RF
fr
= = f ( B, Rs )
h
Hence, using the beam rigidity relation and the average dipole field to define the radius of the
ideal particle, we get for an average magnetic field <B(t)>
f RF (t ) v(t ) 1 e
= = < B(t) >
h 2π Rs 2π m
f RF (t ) 1 ec 2 r
= B(t )
h 2π Es (t ) Rs
Using the relativistic overall energy

=E2 (m c )
0
2 2
+ p 2c 2
we get finally an expression for the RF frequency as a function of the changing external dipole field:
1
  2
f RF (t ) c  B (t ) 2 
=  
h 2π Rs  ( m c ecr ) + B (t ) 
2 2 2
 0 
At high energies, or, more accurately, as soon as
m0 c 2
B> (8)
ecr
the situation simplifies a lot and the equation above reduces to

f RF (t ) c
≈ const
= (9)
h 2π Rs
so that we can keep the RF frequency constant. It is evident from Eq. (8) that for electron beams in
general the condition (9) is (nearly) always fulfilled while proton or heavy-ion synchrotrons need a
more sophisticated RF control.
7 Synchrotron motion
In the following we will again go through the longitudinal motion. However, unlike the previous
section, we will try to put things on a mathematically more solid ground. As in Fig. 12 we expect a
longitudinal oscillation in phase and energy under the influence of the focusing mechanism explained
above.
The relation between relative frequency deviation and relative momentum error is given by
Eq. (7) as
df r  1  dp
=  2 −α 
fr  γ  p
53
B.J. H OLZER
which translates into the difference in revolution time
dT  1  dp
= α − 2 
T0  γ  p
and leads to a difference in phase at the arrival at the cavity:
∆T
∆ψ
= 2π = ωrf ∗ ∆T
Trf
∆T
= h ∗ ω0 ∗ ∆T = h ∗ 2π
T0
 1  dp
h ∗ 2π  α − 2 
=
 γ  p
h ∗ 2π  1  dE
= α − 2 
β 
2
γ  E
As before, the revolution frequency ω0 and the RF frequency ωRF are related to each other via
the harmonic number h. Hence the difference in energy and the offset in phase are connected to each
other through the momentum compaction factor, or better the η parameter.
Differentiating with respect to time gives the rate of change of the phase offset per turn:
∆ψ h ∗ 2π  1  dE
∆ψ= = α − 2  (10)
T0 β 2T0  γ  E
On the other hand, the difference in energy gain of an arbitrary particle that has a phase distance
of Δψ to the ideal particle is given by the voltage and phase of the RF system:
∆E = e ∗ U 0 ( sin (ψ s + ∆ψ ) − sinψ s )
As before, we describe the phase of the ideal (‘synchronous’) particle by ψs and the phase
difference by Δψ. Referring to small amplitudes Δψ of the phase oscillations, we can simplify the
treatment by assuming
and get for the rate of energy change per turn:

U0
∆E = e ∗ ∆ψ cosψ s
T0
A second differentiation with respect to time delivers

U0
∆E = e ∗ ∆ψ cosψ s (11)
T0
54
Combining Eqs. (10) and (11) we get finally a differential equation for the longitudinal motion
under the influence of the phase focusing mechanism:
U 2π h  1  dE
∆E = e ∗ 0 2  α − 2  cosψ s
T0 β T0  γ  E
For a given energy the parameters in front of the right-hand side are constant and describe the
longitudinal – or ‘synchrotron’ – oscillation frequency. Using therefore
−eU 0 h cosψ s  1 
Ω= ω0 ∗ α − 2  (12)
2πβ E 
2
γ 
we get the equation of motion in the approximation of small amplitudes:

∆E + Ω 2 ∆E = 0
It describes a harmonic oscillation in E–ψ phase space for the difference in energy of a particle
to the ideal (i.e., synchronous) particle under the influence of the phase focusing effect of our
sinusoidal RF function.
As already discussed qualitatively, the expression (12) leads to real solutions if the argument of
the square root is a positive number. Two possible situations therefore have to be considered: below
the gamma transition the η parameter is positive, and above it η is negative. The synchronous phase as
the argument of the cos function therefore has to be chosen to get an overall positive value under the
square root:
γ < γtr η > 0, 0 < φs < π/2
γ > γtr η < 0, π/2 < φs < π
Figure 14 shows the superconducting RF system of the LHC, and the basic parameters,
including the synchrotron frequency, are listed in Table 1.
Table 1: Parameters of the LHC RF system.
Bunch length (4σ) ns 1.06

Energy spread (2σ) 10−3 0.22
Synchrotron rad. loss / turn keV 7
RF frequency MHz 400
Harmonic number h 35640
RF voltage / beam MV 16
Energy gain / turn keV 485
Synchrotron frequency Hz 23
Fig. 14: LHC RF system

Figure 15 finally shows the longitudinal focusing effect that has been observed during the LHC
commissioning phase. On the left-hand side, beam had been injected into the storage ring while the RF
system was still switched off. The bunch, nicely formed by the RF voltage of the pre-accelerator, is
visible only for a few turns and the bunch profile is decaying fast, as no longitudinal focusing is active.
The right-hand side shows the situation with the RF system activated and the phase adjusted. The
injected particles stay nicely bunched and the acceleration process can start.
55
B.J. H OLZER
Fig. 15: Longitudinal beam profile measured during LHC injection. Left: the RF system is switched off, and the
bunch profile is decaying fast. Right: the RF system is activated, and the particles stay bunched.
Bibliography
P. Bryant and K. Johnsen, The Principles of Circular Accelerators and Storage Rings (Cambridge
University Press, Cambridge, 1993).
A. Chao and M. Tigner, Handbook of Accelerator Physics and Engineering (World Scientific,
Singapore, 1999).
D. Edwards and M. Syphers, An Introduction to the Physics of Particle Accelerators, Superconducting
Super Collider Laboratory Report (1990).
F. Hinterberger, Physik der Teilchenbeschleuniger (Springer, Berlin, 1997).
B. Holzer, Lattice design, in CERN Accelerator School: Intermediate Accelerator Physics Course,
CERN 2006-002 (2006).
J. LeDuff, in CERN Accelerator School: 5th General Accelerator Physics Course, CERN 94-01
(1994).
M. Reiser, Theory and Design of Charged Particle Beams (Wiley-VCH, Weinheim, 2008).
M. Sands, The Physics of e+e− Storage Rings, SLAC Report 121 (1970).
P. Schmüser, Basic course on accelerator optics, in CERN Accelerator School: 5th General
Accelerator Physics Course, CERN 94-01 (1994).
K. Wille, Physics of Particle Accelerators and Synchrotron Radiation Facilities (Teubner, Stuttgart,
1992).
56
AC/RF Superconductivity
G. Ciovati∗
Thomas Jefferson National Accelerator Facility, Newport News, Virginia, USA
Abstract
This contribution provides a brief introduction to AC/RF superconductivity,
with an emphasis on application to accelerators. The topics covered include
the surface impedance of normal conductors and superconductors, the residual
resistance, the field dependence of the surface resistance, and the superheating
field.
Keywords: AC/RF superconductivity, accelerators, surface impedance, resid-

ual resistance, superheating field.
1 Introduction
This chapter provides an introductory-level tutorial to AC/RF superconductivity. The emphasis is on
the application to resonant cavities for particle accelerators. In this respect, we will present the basic
theoretical concepts and experimental results related to the low-field surface impedance, the superheating
field, and the field dependence of the surface resistance. All these topics are presented to a greater depth
in the bibliography and some of the references listed at the end of this tutorial.
Approximately 20 years after the discovery of superconductivity in 1911, experimental evidence of
a large change in conductivity at the transition temperature was demonstrated by using Radio-Frequency
(RF) currents [1, 2]. Shortly thereafter, a theory of the electrodynamics of superconductors, based on the
phenomenological two-fluid model, was proposed by Fritz and Heinz London [3, 4]. A new theory of
the electrodynamics of superconductors by Mattis and Berdeen was published in 1958 [5], based on the
Bardeen–Cooper–Schrieffer (BCS) theory, which had been published one year earlier [6]. Experimental
results based on far-infrared transmission through superconducting thin films and supporting the theory
were published by Tinkham et al. in the same period [7, 8].
Regarding the highest AC/RF magnetic field that can be applied to a superconductor, the so-
called superheating field, the earliest theoretical work, based on the Ginzburg–Landau (GL) theory,
dates back to the 1960s [10–12]. Experimental results in the range 90–300 MHz for both type I and type
II superconductors in the vicinity of the critical temperature, Tc , and consistent with the theory, were
published in 1977 [13].
Whereas niobium is the superconductor almost exclusively used to produce resonant cavities for
particle accelerators, superconducting materials with higher critical temperatures are also being used for
RF applications in passive microwave devices, such as filters, resonators, and antennas for mobile com-
munications [14], and to produce microresonators for a variety of applications, such as photon detectors
and quantum circuits [15].
2 Basics of RF cavities
Generally speaking, a resonant cavity is any volume enclosed by metallic walls that contains oscillating
electromagnetic fields. For application to particle accelerators, the electromagnetic energy stored within
the cavity is used to accelerate a charged particle beam. The frequency range relevant for accelerator
applications is RF (3 kHz – 300 GHz).
∗
gciovati@jlab.org
G. C IOVATI
The electromagnetic field inside an RF cavity is the solution to the wave equation:

2 1 ∂2 E
∇ − 2 2 = 0, (1)
c ∂t H
with the boundary conditions n̂ × E = 0 and n̂ · H = 0, where n̂ is the unit vector normal to the surface.
Solutions to Eq. (1) with the specified boundary conditions can be separated into two families of resonant
modes with different eigenfrequencies, based on the direction of the electric and magnetic field:
– TEmnp modes having only transverse electric fields, and

– TMmnp modes having only transverse magnetic fields (but a longitudinal component of the electric
field),
where m, n, and p are indices denoting the number of zeros in the φ, ρ, and z directions, respectively,
in cylindrical coordinates. A useful example of a resonant cavity is a metallic cylindrical waveguide of
length L, shorted by metallic plates at both ends. This geometry is commonly referred to as ‘pill-box’.
The mode used to accelerate charged particles in RF cavities having a geometry resembling that of a
pill-box is the TM010 . The electric and magnetic fields, as well as the resonant frequency of this mode,
can be calculated analytically for the pill-box geometry:

2.405ρ iωt
Ez = E0 J0 e ,
R

E0 2.405ρ iωt (2)
Hφ = −i J1 e ,
η R
2.405c
ω0 = ,
R
where J0 and J1 are Bessel functions of zeroth and firstporder, respectively, R is the pill-box radius, c is
the speed of light, ω is the angular frequency, and η = µ0 /0 ' 377 Ω is the impedance of a vacuum.
Equation (2) shows that the electric field, being at a maximum on-axis, can be used to accelerate charged
particles travelling along the axis of the cavity. A schematic representation of the electric and magnetic
fields inside a pill-box type cavity is shown in Fig. 1.
Fig. 1: Electric and magnetic fields for the TM010 mode inside a pill-box cavity.
Other resonant modes that are sometimes used are the TE011 mode and the TM110 mode. The
first of these has a zero electric field on the cavity surface and is used to study the surface resistance of
superconductors in RF magnetic fields. The second has a transverse component of the electric field on
axis, tilting the beam, which is sometimes necessary in collider accelerators in order to provide a head-on
collision between two beams and thereby increase the luminosity. The deflecting TM110 mode has also
been used in an SRF separator cavity to separate beams of different particles [9].
2
58
AC/RF S UPERCONDUCTIVITY
Although the resonant frequency of the TM010 mode does not depend on the pill-box length, L,
the following condition for synchronism between the beam and the electric field in the cavity sets the
cavity length:
TRF
L = βc , (3)
2
where β is the speed of the particle relative to the speed of light and TRF = 2π/ω0 is the period of
oscillation of the RF field. The condition imposed by Eq. (3) assures that, as an example, a bunch of
relativistic electrons entering the cavity at time t = 0, when Ez = 0, will experience the maximum
acceleration as they travel along the cavity axis.
2.1 Figures of merit

The accelerating field of the cavity, Eacc , is defined as the ratio of the accelerating voltage, Vc , divided
by the cavity length. Vc is obtained by integrating the electric field at the particle’s position as it traverses
the cavity:
Z L
Vc 1
Eacc = = Ez (ρ = 0, z)eiω0 z/c dz . (4)
L L 0
Other important parameters are the ratios of the peak electric and magnetic fields on the cavity surface
divided by the accelerating field, Ep /Eacc and Bp /Eacc , respectively, as they are related to practical
limitations of a cavity’s performance, such as field emission and quench.
The power dissipated as heat in the cavity wall, Pc , and the energy stored within its volume, U,
are given by Z Z
1 1
Pc = < J · E dv = Rs |H|2 da, (5)
2 V 2 S
Z
1
U = µ0 |H|2 dv. (6)
2 V
The quality factor of the cavity, Q0 , is defined, in the same way as for any resonator, as the ratio of the
energy stored divided by the energy dissipated in in one RF period:
R
ω0 U ω0 µ0 V |H|2 dv
Q0 = = R 2
. (7)
Pc S Rs |H| da
Q0 can be calculated from the Breit–Wigner resonance curve as the ratio of the resonant frequency,
divided by the full width at half maximum, as shown in Fig. 2.
Assuming that the surface resistance is uniform over the cavity surface and does not depend on
the amplitude of the applied field, it is possible to define from Eq. (7) a geometry factor, G, that depends
only on the cavity shape (but not its size) and that provides a direct relation between Q0 and Rs :
R
ω0 µ0 V |H|2 dv
G= R
2
= Q0 Rs . (8)
S |H| da
The assumptions on the definition of G are usually valid at low field amplitudes.
The figures of merit for the TM010 mode in a pill-box cavity, calculated from the analytical fields
of Eq. (2), are as follows:
2
Eacc = E0 ,
π
π
Ep /Eacc = = 1.57,
2
π J1 (1.84) A · m−1 Oe (9)
Hp /Eacc = = 2430 −1
= 30.5 ,
2 η MV · m MV · m−1
2.405L 453L/R
G=η = Ω.
2(R + L) 1 + L/R
3
59
G. C IOVATI
Practical RF cavities have more complex shapes than a simple pill-box, and require numerical solvers to
calculate the electric and magnetic fields inside the cavity.
Fig. 2: The normalized stored energy as a function of frequency for a resonator with resonant frequency ω0 and
Q0 = 104 .
2.2 SRF cavity performance

The performance of a superconducting RF cavity is described by a plot of the quality factor as a function
of the accelerating gradient. The state-of-the-art performance of a 1.3 GHz bulk Nb cavity tested multiple
times at 2.0 K is shown in Fig. 3 [16]. The Q0 value at low field corresponds to a surface resistance of
' 8 nΩ, and the maximum Eacc value corresponds to a peak surface magnetic field Bp = µ0 Hp '
185 mT. In the following sections, we will provide a basic description of what determines the Rs at low
field as well as the maximum obtainable Bp value.
Fig. 3: The state-of-the-art performance of a bulk Nb cavity at 1.3 GHz and 2.0 K [16].
4
60
3 Surface impedance
The electromagnetic response of a metal, whether normal or superconducting, is described by a complex
surface impedance defined as follows:
|Ek | |Ek |
Zs = R ∞ = = Rs + iXs , (10)
0 J(x) dx |Hk |
where Rs is the surface resistance and Xs is the surface reactance. In Eq. (10), we have neglected the
displacement current, which is reasonable for good conductors at frequencies less than ∼ 1016 Hz.
3.1 The electrodynamics of normal conductors

The electrodynamics of normal conductors is based on Maxwell’s equations and on material equations
that specify the relations: between the electric displacement, D, and the electric field; between the in-
duction field, B, and the magnetic field; and between the current density and the electric field. Since the
inside volume of a cavity is typically under vacuum, we have D = 0 E, B = µ0 H. A relation between
J and E can be obtained from Drude’s model of ‘nearly free electrons’ in a solid and resulting in Ohm’s
law:
ne2 1
J= E = σE, (11)
mτ 1 + iωτ
where σ = ne2 /mτ is the conductivity and τ ' 10−14 s is the scattering time of the electrons, given
by the ratio of the mean free path, `, divided by the Fermi velocity, vF . In Eq. (11), we have used the
approximation ωτ 1, which is valid at RF frequencies. Using Maxwell’s equations with the material
equations, the following equation for the magnetic field inside a metal is obtained:
∇2 H = iσµ0 ωH. (12)
The solution of Eq. (12) for a semi-infinite slab occupying the positive-x region of space with a magnetic
field of amplitude H0 applied in the y-direction is given by
Hy (x, t) = H0 e−x/δ e−i(x/δ−ωt) , (13)
which describes a wave propagating in the positive-x direction with an exponentially decreasing ampli-
tude. δ is the characteristic decay length, called the ‘skin depth’, and given by
r
2
δ= . (14)
µ0 σω
The electric field, obtained from Ampere’s law, has a behaviour similar to that of the magnetic field:
1+i
Ez (x, t) = − Hy (x, t). (15)
σδ
The surface impedance is then given by
|Ez (x = 0)| 1+i

Zs = = . (16)
Hy (x = 0) σδ
Therefore, Rs = Xs = 1/σδ. If we consider, for example, copper (σ ' 5.8 × 107 S · m−1 ) at 1.5 GHz
and 300 K, we obtain δ = 1.7 µm and Rs = 10 mΩ.
At lower temperatures, the conductivity increases and therefore δ decreases and may become
shorter than the electrons’ mean free path. This implies that the local relation between field and current
given by Ohm’s law is no longer valid at low temperatures, since the distance over which the field varies
5
61
G. C IOVATI
becomes less than the mean free path. A new relationship was introduced by Reuter and Sondheimer in
1948 [17], in which J is related to E over a volume of the size of the mean free path:
Z
3σ R[R · E(r0 , t − R/vF )] −R/` 0
J(r, t) = e dr , (17)
4π` V R4
with R = r0 − r. An effective conductivity, σeff ≈ (δ/`)σ, results from Eq. (17), showing that, unlike
the DC case, increasing the purity of the metal does not improve its conductivity. This phenomenon is
commonly referred to as the ‘anomalous skin effect’. The surface resistance in the extreme anomalous
limit (` → ∞), valid for very good conductors such as copper at low temperatures, is given by
1/3
√ µ0 2
Rn (` → ∞) = 3 ω 2/3 (ρ`)1/3 . (18)
4π
The product ρ` is a material constant and it is 6.8 × 10−16 Ω · m−2 for copper. If one were to operate
a 1.5 GHz copper cavity at cryogenic temperatures such as, for example, 4.2 K, rather than 300 K, the
surface resistance would decrease by a factor of ≈ 0.14, which is not sufficient to justify the cost of a
refrigerator.
3.2 The electrodynamics of superconductors

3.2.1 Theory
Unlike the DC case, superconductors in RF fields do not have zero resistance at finite temperatures.
This is because a time-dependent magnetic field within the penetration depth generates an electric field
(Faraday’s law) that acts on normal electrons, as they are not shielded from it by the superconducting
electrons (which form ‘Cooper pairs’ of mass twice that of a single electron) due to their inertia.
A simple way to describe the electrodynamics of superconductors was given by the London broth-
ers in 1934, based on the phenomenological ‘two-fluid’ model of Gorter and Casimir [18]. According to
the model, the charge carriers are divided into two subsystems: superconducting carriers of density ns
and normal electrons of density nn . The superconducting carriers were associated later on with Cooper
pairs of charge −2e and mass 2me [19]. The normal current Jn and the supercurrent Js are assumed to
flow in parallel, and the total current is the sum of Js and Jn . Js flows with no resistance and follows the
London equations:
∂ 1
Js = E, (19)
∂t µ0 λ2L
1
∇ × Js = − H, (20)
λ2L
p
where λL = m/µ0 ns e2 is the so-called London penetration depth. Equation (19) (the first London
equation) implies perfect conductivity, since a current would flow indefinitely in a superconductor even
for zero electric field, and that an electric field is required to maintain an RF current. Equation (20) (the
second London equation) implies the spontaneous flux exclusion from the bulk of a superconductor and
that an induction field is the source of the supercurrent. Note that Eq. (20) can be written as follows:
1
Js = − A, (21)
λ2L
where A is the magnetic vector potential. Equation (21) represents a local condition between current and
field that is valid if ξ0 λL or ` λL . ξ0 is the coherence length, representing the distance between
electrons forming a Cooper pair.
In the presence of an RF current, J = J0 eiωt , the relation between J and E in a superconductor, is
given by
J = Jn + Js = (σ1 − iσ2 )E, (22)
6
62
where σ1 = nn e2 /mτ is the conductivity of the normal component (the same as the normal-state con-
ductivity) and σ2 = ns e2 /mω is the conductivity of the superconducting component, obtained using
Eq. (20). The electrodynamics of the superconductor become analogous to those of a normal conductor
if one replaces σ with σ1 − iσ2 in the expressions for the skin depth, Eq. (14), and the magnetic field,
Eq. (13). The skin depth becomes
s
2 σ1
δ= ' (1 + i)λL 1 + i , (23)
µ0 ω(σ1 − iσ2 ) 2σ2
where we have made use of the approximation σ1 σ2 , which is valid for superconductors at tempera-
ture T Tc . The magnetic field inside the superconductor becomes

x σ1
−x/λL −i −ωt
Hy (x, t) = H0 e e λL 2σ2
, (24)
and the amplitude of the magnetic field decreases exponentially with a characteristic length λL . For
niobium, λL ' 36 nm, much shorter than the skin depth for copper at 1.5 GHz. The surface impedance
of a superconductor can be obtained by substituting Eq. (23) into Eq. (16). After some calculus involving
complex numbers and with the approximation σ1 σ2 , we obtain the following:
1
Zs = µ20 ω 2 σ1 λ3L + iωµ0 λL . (25)
2
The equivalent circuit for a superconductor is a resistor of resistance Rs in parallel with an inductor of
inductance Ls = µ0 λL , the so-called ‘kinetic inductance’, due to the superconducting charge carriers.
The dependence of Rs on ω and ` in Eq. (25) indicates the following.
– The surface resistance increases quadratically with frequency, and therefore low-frequency cavities
should be considered to reduce the dissipated power.
– The surface resistance increases with increasing purity of the material. An intuitive way to think
about this is that normal-conducting electrons of higher conductivity draw a relatively higher frac-
tion of the total current.
To understand the temperature dependence of the surface resistance, we consider the temperature depen-
dence of ns and nn :
λL (T )2 ∝ 1/ns (T ) ∝ 1/[1 − (T /Tc )4 ], (26)
σ1 (T ) ∝ nn (T ) ∝ e−∆/kB T . (27)
Equation (27), which is valid for T Tc , reflects the creation of normal electrons due to the thermal
break-up of Cooper pairs. From Eqs. (27) and (25), we obtain the following dependence of Rs on
frequency, material purity, and temperature:
Rs ∝ ω 2 λ3L `e−∆/kB T , for T < Tc /2. (28)
The exponential decrease of Rs with temperature indicates that low-temperature operation, such as 2.0 K
for Nb cavities, is preferable to reduce RF losses.
Similarly to the anomalous skin effect in normal conductors, if ξ0 λL and ` λL , the local
relation between current and field is no longer valid. In 1953, Pippard [20] proposed replacing Eq. (21)
with Eq. (29) below, ξ0 playing a role analogous to that of ` in the non-local electrodynamics of normal
conductors: Z
3 R[R · A(r0 )] −R/ξ 0
J(r) = e dr , (29)
4πξ0 λ2L V R4
7
63
G. C IOVATI
with R = r0 − r and 1/ξ ' 1/ξp

0 + 1/`. The dependence of the penetration depth on the mean free path
can be approximated as λ ≈ λL 1 + ξ0 /` and the dependence of Rs on material purity becomes

ξ0 3/2
Rs ∝ 1 + `. (30)
`
Equation (30) shows that Rs ∝ ` increases with increasing mean free path if ` ξ0 (the so-called ‘clean
limit’) as discussed above, and that Rs ∝ `−1/2 increases with decreasing mean free path if ` ξ0 (the
so-called ‘dirty limit’). Therefore Rs (`) has a minimum at ` = ξ0 /2. In 1958, Mattis and Bardeen [5]
obtained from the BCS theory a non-local equation between the total current density (including the
supercurrent and the normal current) and the vector potential:
Z
3 R[R · A(r0 )I(ω, R, T )e−R/` ] 0
J(r) = 2 dr , (31)
4π vF ~λ2L V R4
where I(ω, R, T ) is a function that decays over a characteristic length R ∼ ξ0 . Equation (31) becomes a
product of a kernel function K(q) times the vector potential in the Fourier domain: J(q) = −K(q)A(q)
in one dimension. The surface impedance can be obtained as a function of the kernel K(q), as follows:
iµ0 ωπ
Zs = R ∞ K(q)
, (32)
0 ln(1 + q2
) dq
for diffuse scattering of electrons at the metal surface, such as for the case of a ‘rough’ surface on a scale
of the mean free path, or Z
iµ0 ω ∞ 1
Zs = 2
dq, (33)
π −∞ q + K(q)
for specular reflection of electrons at the metal surface. The calculation of the real and imaginary parts
of K(q) involves the solution of complex integrals, which can only be done numerically. A computer
code that allows the BCS surface impedance to be calculated using Eqs. (32) and (33) was written by
Halbritter in 1970 [21] and a copy is available from the author. An online calculator is also available [22].
An analytical approximation of the BCS surface resistance valid in the local limit, for T < Tc /2 and
ω < ∆/~, is given by [23]

µ20 ω 2 λ3 σ1 ∆ 2.246kB T ∆
Rs ' ln exp − . (34)
kB T ~ω kB T
Considering the case of niobium (λ = 40 nm, σ1 = 3.3 × 108 S · m−1 , ∆/kB Tc = 1.85, Tc = 9.25 K)
at 2.0 K and 1.5 GHz, we obtain RBCS ' 20 nΩ and Xs ' 0.47 mΩ. The ratio of RBCS for Nb at 2.0 K
divided by Rs for Cu at 300 K is ' 2 × 10−6 . Even when the Carnot efficiency ηC = 0.67%, due
2.0 K operation and the technical efficiency of a cryoplant, ηT ' 20%, are included, the reduction in
power consumption by using superconducting cavities instead of normal-conducting ones is still quite
significant (' 103 reduction factor).
3.2.2 Experimental results

Figure 4 shows the surface resistance measured in bulk Nb at 4.2 K over a broad range of frequencies [24].
Small deviations from the BCS theory can be explained by strong coupling effects or an anisotropic
energy gap, in the presence of scattering by impurities or inhomogeneities.
3.2.3 Residual resistance

Improvements in the surface preparation of bulk Nb cavities over the past 40 years have reduced the
typical residual resistance value from ∼ 100 nΩ to ∼ 1 − 10 nΩ. Rres becomes the dominant term
8
64
in the surface resistance at low frequency (<∼ 750 MHz) and low temperatures (<∼ 2.1 K), where
RBCS becomes exponentially small. There are several possibilities contributing to the residual resistance.
Among those there are:
– losses due to trapped magnetic field,

– losses due to normal-conducting precipitates near the surface,
– grain boundary losses,
– metal/oxide interface losses, and
– losses due to normal-conducting electrons in subgap states.
Fig. 4: The surface resistance measured in bulk Nb at 4.2 K as a function of frequency. The solid line is a fit to the
experimental data, while the dashed line results from a calculation based on the BCS theory. Reprinted from [24]
with permission from IEEE.
Figure 5 shows the surface resistance of a Nb thin film deposited on Cu, measured at 4.2 K and 1.5 GHz,
as a function of a parameter related to the purity of the film [25]. The data are consistent with the
existence of a broad minimum of RBCS (`), as predicted by calculations based on the BCS theory. A
measurement of the temperature dependence of the low-field surface resistance of bulk Nb at 1.3 GHz is
shown in Fig. 6 [26]. The data show a clear deviation from the exponential dependence as the temperature
decreases towards 0 K. This additive, T -independent contribution to the surface resistance is the so-
called residual resistance, Rres (not to be confused with the DC residual resistivity ρn = 1/σ1 ). A
well-known contribution to Rres is due to trapped DC magnetic field, due to the incomplete Meissner
effect in technical materials. Figure 7 shows, for example, the low-field Rs (T ) measured for different
DC magnetic field amplitudes applied while cooling the cavity down from 300 K to 4.2 K [28]. Any
DC magnetic field at the cavity location can be trapped in the form of fluxoids pinned by defects in
the material, as the cavity is cooled below Tc . A simple estimate of this contribution can be made by
assuming that all of the DC field, Hext , is trapped in the form of N fluxoids, each carrying one flux
quantum φ0 , and that their normal-conducting cores, each of radius ∼ ξ0 , dissipate according to the
normal state surface resistance. An improved description was given by Gurevich, in which both the
pinning strength and the dissipation due to the motion of the fluxoid’s core under the RF field are taken
into account [27]. Figure 8 shows a schematic representation of a pinned fluxoid with a segment of length
9
65
G. C IOVATI
Fig. 5: The surface resistance measured in Nb thin films at 4.2 K and 1.5 GHz as a function of a parameter related
to the mean free path. The solid line is a result from a calculation based on the BCS theory. Reprinted from [25]
with permission from Elsevier.
Fig. 6: The temperature dependence of the low-field surface resistance of bulk Nb at 1.3 GHz, showing a saturation
to a T -independent value at low temperatures [26].
`s almost normal to the surface of the superconductor. At low RF frequency, the surface RF Meissner
current causes rocking of the whole fluxoid segment. As the RF frequency increases, the RF oscillations
of the segment are mostly localized closer and closer towards the surface, until only a tip of length ∼ λ
vibrates in the high-frequency limit. The RF penetration depth and the pinned fluxoid segment set the
frequency scales. The RF power dissipation due to the fluxoid’s motion is given by
Z `s Z `s
ωF
PRF = u̇(z, t)F (z, t) dz =− Im u(z, ω)e−z/λ dz, (35)
0 ω 2 0
10
66
Fig. 7: The low-field surface resistance as a function of temperature for bulk Nb at 1.5 GHz for different amplitudes
of a DC magnetic field applied during the cavity cool-down. The solid lines are fitted with Rs (T ) = RBCS (T ) +
Rres [28].
where u(z, t) is the fluxoid displacement, obtained by solving the equation of motion, and F = φ0 H0 /λ
is the amplitude of the driving force from the RF field. The result from a calculation of PRF (ω) with
√
Eq. (35) is shown in Fig. 8, indicating a change from ∼ ω 2 dependence at low frequency to ∼ ω at
intermediate frequency and frequency independent at high frequency. Experimental data on bulk Nb in
√
the GHz range show that Rres,mag ∝ ωHext . Rres,mag calculated from Eq. (35) in the intermediate
frequency range is given by r
Hext µ0 ρn ω
Rres,mag = , (36)
Hc 2g
where Hc is the thermodynamic critical field and g is a parameter related to the anisotropy of the super-
conductor. In the case of a 1.5 GHz Nb cavity (ρn ∼ 5 × 1010 Ω · m−1 , Hc = 2000 Oe, 2g = 1), the
residual resistance due to the Earth’s magnetic field (∼ 0.5 Oe) could be as high as ∼ 600 nΩ, about 30
times higher than RBCS at 2.0 K. By applying magnetic shields around superconducting cavities, it is
possible to shield external fields down to ∼ 1 − 10 mOe.
Another well-known contribution to the residual resistance in Nb cavities is due to the precipitation
of normal-conducting niobium hydride islands near the surface, if the bulk H concentration is greater than
∼ 5×10−4 wt.% and if the cool-down rate is < 1 K·min−1 in the temperature range 75–150 K. Figure 9
shows an example of Q0 (Eacc ) measured in a 1.5 GHz bulk Nb cavity at 2.0 K after it was held in the
region 100–150 K for various lengths of time [29]. This problem can be mitigated by degassing the
cavity in a ultra-high-vacuum furnace at 600–800◦ C for 2–6 h. Tunnelling measurements of bulk Nb
samples show the presence of electronic states within the energy gap. The density of states as a function
of energy N () obtained from the measurements can be described by the phenomenological model of
Dynes [35, 36], resulting in a finite density of states at the Fermi level, N (0) ' γNn /∆, where γ is a
damping parameter and Nn is the density of states in the normal state. The contribution to Rres from
normal electrons occupying subgap states can be estimated from the two-fluid model surface resistance
with γσn /∆ replacing σ1 [23]:
Rres,subgap ∼ µ20 ω 2 λ3 σn γ/∆, (37)
Residual resistance values of ∼ 10 nΩ, similar to those obtained in bulk Nb cavities, could result from
Eq. (37) for γ/∆ ∼ 10−3 . The origin of these subgap states is not well understood. They could be due
to intrinsic effects, such as strong impurity scattering or strong electron–phonon coupling, or extrinsic
11
67
G. C IOVATI
Fig. 8: (a) A schematic representation of a pinned fluxoid near the surface of a superconductor in the presence of an
RF field. (b) The normalized power dissipation due to the oscillatory motion of a fluxoid, as a function of frequency
normalized to ω0 = 0 /2ηλ2 , where 0 is the fluxoid line energy and η is the viscous drag coefficient [27].
effects, such as normal-conducting precipitates, defective oxides, or pinned fluxoids.
Fig. 9: The degradation of the quality factor in a 1.5 GHz bulk Nb cavity at 2.0 K after it was held in the region
100–150 K for various lengths of time [29].
3.3 High-temperature superconductors

Superconductors with relatively low Tc , such as Nb, have a ‘s-wave’ character, meaning that the energy
gap is uniform in the momentum space. On the other hand, high-temperature cuprates (Tc ∼ 90 K) have
a ‘d-wave’ character, indicating that the energy gap has four nodes along symmetric directions in mo-
mentum space. This leads to unfavorable consequences for RF application: the temperature dependence
of the surface resistance follows a power law, instead of being exponential, and residual losses are higher.
The Rs (Hp ) dependence also exhibits strong non-linearity. The coherence length is much shorter than
that of Nb (1–2 nm, instead of ∼ 40 nm); therefore the pairing of electrons can easily be disrupted by
defects. Cuprates are also ‘granular’ superconductors, with high grain boundary resistance contributing
to high Rres . All these effects hinder their use for SRF cavity application where low Rs at high RF fields
is required. Figure 10 shows an example of Rs (T ) measured in YBCO samples at 1.7 and 2 GHz [30].
12
68
Superconductors with higher Tc than Nb but still s-wave, such as Nb3 Sn (Tc ' 18 K), NbN, and NbTiN
(Tc ' 17 K), are more promising alternative materials for RF application. Their higher Tc value could
allow operating cavities at 4.2 K instead of 2.0 K, therefore reducing the cost of refrigeration.
Fig. 10: Rs (T ) measured on YBCO samples. Reprinted from [30] with permission from Nova Science Publishers,
Inc.
4 The superheating field

In the previous section, we discussed the physics determining the quality factor of SRF cavities at low RF
fields. However, the feasibility of SRF cavities for particle accelerator applications, particularly those for
high-energy physics, relies on the ability to reach high accelerating gradients, corresponding to high RF
magnetic fields on the cavity surface. In this respect, it is important to understand what is the maximum
RF magnetic field that can be applied to the surface of a superconductor (assumed to be ‘defect-free’),
before a transition to the normal state occurs.
Theoretically, the critical RF magnetic field is considered to be the so-called superheating field,
Hsh , which is the highest field up to which the superconductor remains in the Meissner state, and is
unaltered by the dissipative motion of fluxoids. At H0 = Hsh , the screening surface current reaches the
depairing value, Jd = ns e∆/pF , meaning that the kinetic energy of the superconducting carriers exceeds
the binding energy of the Cooper pairs. The magnetic field at which the Gibbs free energy has the same
value whether the first fluxoid is inside or outside a type II superconductor is the so-called lower critical
field, Hc1 . However, a fluxoid entering the surface of a superconductor has to overcome the so-called
Bean–Livingston surface barrier [31], which arises because of the attractive force between the fluxoid
near the surface and its anti-fluxoid image, at the same distance from the surface but on the opposite
side, required to ensure that the normal component of the current density at the boundary is zero. This
is shown schematically in Fig. 11. The Gibbs free energy of a superconductor with a single fluxoid, as a
function of the fluxoid position, is given by

−x/λ φ0 2x
G(x) = φ0 H0 e − K0 + Hc1 − H0 , (38)
4πλ2 λ
13
69
G. C IOVATI
where the first term is the energy due to the Meissner current, the second term is half of the interaction
energy between the fluxoid and its image, the third term is the fluxoid self-energy, and the last term is
the work done by the source of the applied field. K0 is the zeroth-order Hankel function. The behaviour
of G(x) is shown schematically in Fig. 11: the surface barrier disappears only at H0 = Hsh > Hc1 .
Fig. 11: (a) A schematic representation of a fluxoid near the surface of a superconductor. (b) The Gibbs free energy
as a function of the fluxoid position for different applied fields.
Theoretical calculations of the superheating field as a function of the Ginzburg–Landau (GL) pa-
rameter, κGL , close to Tc have been done since the 1960s using the GL theory [10–12]. The metastability
of the Meissner state leading to the existence of a superheating field represents a local minimum of the
free energy, meaning that its second derivative is positive. The field up to which this metastability condi-
tion is satisfied has been evaluated considering fluctuations of the order parameter and the supercurrent
along the boundary of the superconductor in one or two dimensions [32, 33]. Hsh resulting from the GL
theory (T ≈ Tc ) is given by
Hsh ' 1.2Hc , κGL ≈ 1,

(39)
Hsh = 0.745Hc , κGL 1.
The calculation of Hsh has recently been extended over the whole temperature range, 0 < T < Tc , by
the numerical solution of Eilenberger equations [34, 38]. These equations were obtained from Gorkov’s
formulation of the BCS theory. The calculations were done in the high-κGL limit and as a function of
the mean free path. In the clean limit, Hsh at 0 K is given by
Hsh = 0.845Hc , κGL 1. (40)
A weak dependence of Hsh on ` was found, if scattering was due to non-magnetic impurities, and an
enhancement of Hsh (0 K) up to ' 4.2% at T = 0 was found for πξ0 /` ' 0.6 [38].
Niobium, being a marginal type II superconductor (κGL ≈ 1), is a difficult material to study
theoretically. Assuming the result in Eq. (39) to be valid at low T /Tc , one can estimate the Hsh of Nb to
be ∼ 240 mT at 0 K. On the other hand, the Hsh of Nb3 Sn (κGL ≈ 30) obtained from Eq. (40) is of the
order of 420 mT at 0 K. Results from experiments aimed at measuring the Hsh of bulk Nb and of a thin
Nb3 Sn layer grown on a Nb cavity at 1.3 GHz as a function of temperature are shown in Fig. 12 [39].
The data show that RF magnetic fields above Hc1 have been achieved in both materials, but the highest
field for Nb3 Sn is much lower than the value of Hsh predicted by the theory.
Unfortunately, the surface barrier can easily be suppressed by defects, such as nano-precipitates or
even roughness, making it quite difficult in practice to extend the Meissner state up to Hsh in high-κGL
materials (usually deposited as thin films on a substrate). In 2006, Gurevich proposed the use of multi-
layered films of alternating superconductor–insulator–superconductor deposited on a bulk Nb cavity to
14
70
Fig. 12: Measurements of the maximum RF field achievable in bulk Nb and Nb3 Sn 1.3 GHz cavities as a
function of temperature [39]. The solid, dashed, and dashed–dotted lines show the theoretical expectations for
Hc (T ), Hsh (T ), and Hc1 (T ), respectively. The solid black square shows the highest RF magnetic field ever mea-
sured on Nb cavities at 2.0 K.
achieve the superheating field on the cavity surface [40]. By making the thickness of the superconduc-
tor layer, d, smaller than the penetration depth, the Hc1 of the layer should increase significantly, as
calculated by Abrikosov in 1964 [37]:

2φ0 d
Hc1 = ln − 0.07 . (41)
πd2 ξ
5 Rs (H0 ) dependence due to thermal feedback

As we discussed in Sections 3 and 4, there are well-established theories to calculate the surface resistance
and the RF critical field of superconductors. Unfortunately, there is no well-established theory describing
the dependence of the surface resistance on the amplitude of the RF magnetic field, from low field up
to Hsh . In general terms, the surface resistance is expected to increase with increasing RF field because
the density of thermally activated normal electrons increases. This occurs because the energy gap is
reduced by the kinetic energy of the Cooper pairs to an effective gap ∆eff (vs ) = ∆ − pF vs , where pF is
the Fermi momentum and vs = H0 /λens is the superfluid velocity. A decrease of Q0 with increasing
Eacc is typically observed in SRF cavities, as shown, for example, in Fig. 3. The measured slope of
the Q0 (Eacc ) dependence changes significantly for different cavity treatments and for different cavity
material (e.g. Nb bulk vs. thin film). In this section, we will discuss a thermal feedback model proposed
by Gurevich [41] as a simple mechanism causing an increase of Rs with increasing amplitude of the RF
field.
Let us consider the one-dimensional case of a superconductor of thickness d with a vacuum on
one side (the inner surface of an SRF cavity) and liquid He on the other side (the outer surface of an SRF
cavity). An RF field of amplitude H0 is applied on the vacuum side, parallel to the surface. The RF power
dissipated on the inner surface at temperature Tm is transferred as heat to the He bath of temperature T0 .
The temperature of the outer surface is Ts . The temperature profile is shown schematically in Fig. 13.
The heat balance equations for the inner and outer surfaces can be written as follows:
Z Ts
1 2
Rs (Tm )H0 = κ(T ) dT = hK (Ts , T0 )(Ts − T0 ), (42)
2 Tm
where κ(T ) is the thermal conductivity and hK (Ts , T0 ) is the Kapitza conductance, which is the heat
transfer coefficient between the outer surface and the superfluid He bath (T0 < 2.17 K).
15
71
G. C IOVATI
Fig. 13: (a) A schematic of heat transfer from the inner to the outer cavity surface. (b) The H0 (Tm ) dependence
calculated from Eqs. (43) and (44) for Nb at 1.5 GHz and T0 = 2.0 K. The maximum in the H0 (Tm ) defines the
point of thermal breakdown.
As will be shown below, the overheating of the inner surface is small relative to the He bath
temperature (Tm − T0 T0 ); therefore we can neglect the temperature dependence of κ and hK and use
their values at T0 . This simplifies Eq. (42) to the following:
1 hK κ
Rs (Tm )H02 = (Tm − T0 ). (43)
2 κ + dhK
By substituting the following approximate dependence of RBCS (Tm ),
Aω 2 −∆/kB Tm
RBCS (Tm ) ' e , (44)
Tm
in Rs (Tm ) = RBCS (Tm ) + Rres , Eq. (43) gives a relation between H0 and Tm . This dependence is
plotted, as an example, in Fig. 13 for 3 mm thick Nb at 1.5 GHz, with RBCS (2 K) = 20 nΩ, ∆/kB =
17.1 K, hK = 5 kW · m−2 · K−1 , κ = 10 kW · m−1 · K−1 , and T0 = 2.0 K. The maximum of the
H0 (Tm ) curve corresponds to a thermal quench of the cavity. This point defines the temperature of the
inner surface, Tb , and the magnetic field, Hb , at which thermal breakdown occurs. Tb is the value of Tm
that satisfies dH0 /dTm = 0, where H0 (Tm ) is given by Eq. (43), with Eq. (44) for RBCS . Neglecting
the residual resistance, one obtains
T2
Tb − T0 ≈ 0 . (45)
∆
where Tb − T0 ≈ 0.23 K for Nb at 2.0 K. Substituting Eq (45) into Eq. (43), one obtains
2kB T02 hK κ
Hb2 ≈ , (46)
(κ + dhK )∆eRBCS
where e = 2.718. The breakdown field estimated from Eq. (46) is that which occurs in the case of
uniform heating of the inner surface, without any localized defect. Because both H0 and Rs depend on
the inner surface temperature, Eqs. (43) and (44) provide a dependence of Rs on H0 , and it is therefore
possible to calculate a Q0 (H0 ) curve (Q0 = G/Rs ). This is shown, as an example, in Fig. 14 for a
1.5 GHz bulk Nb cavity with G = 280 nΩ at 2.0 K and the thermal parameters mentioned earlier.
For a thin film of thickness dtf and thermal conductivity κtf deposited on a substrate, the total
thermal conductance becomes hK /[1 + hK (d/κ + dtf /κtf )]. In the case of a 1.5 µm thick Nb3 Sn film
[κtf ' 10−2 W · m−1 · K−1 ] on top of a 3 mm thick Nb cavity, the heat transfer coefficient of the thin
film, dtf /κtf ' 670 W · m−2 · K−1 , is comparable to that of the Nb substrate, d/κ ' 330 W · m−2 · K−1 ,
at 2.0 K.
16
72
Fig. 14: Q0 (B0 ) calculated up to the thermal breakdown field for a 1.5 GHz Nb cavity at 2.0 K with the parameters
mentioned in the text and Rres = 10 nΩ.
6 Summary
Unlike the DC case, superconductors in an RF field have a non-zero surface resistance. The surface
resistance can be easily understood in terms of a two-fluid model and is due to the interaction of the
electric field (exponentially decaying from the surface) with normal-conducting electrons.
At low RF fields, the surface resistance can be expressed as the sum of the BCS surface resistance
and a residual resistance. The low-temperature BCS surface resistance:
– increases quadratically with frequency,

– decreases exponentially with temperature, and
– has a minimum as a function of the material purity.
There are many possible contributions to the residual resistance, and trapped fluxoids and niobium hy-
dride precipitates have been proven to be among them.
The maximum theoretical RF field that can be applied to the surface of a superconductor is the
superheating field, which is of the order of the thermodynamic critical field. Superconductor–insulator–
superconductor multilayer thin films might be a possible way to reach the superheating field for super-
conducting materials other than Nb.
There exists no well-accepted theory of the surface resistance at high RF fields that could be used
to describe the Q0 (Eacc ) curves of SRF cavities. Thermal feedback between RF power dissipation and
surface temperature is a simple extrinsic mechanism causing an increase of the surface resistance with
increasing RF field.
Acknowledgements
I wish to thank Professor A. Gurevich of Old Dominion University for many fruitful discussions and for
providing some of the figures.
References
[1] F.B. Silsbee, R.B. Scott, J.W. Cook and F.G. Brickwedde, Phys. Rev. 39 (1932) 379.
[2] J.C. McLennan, A.C. Burton, A. Pitt and J.O. Wilhelm, Proc. R. Soc. A 136 (1932) 52.
[3] H. London, Nature 133 (1934) 497.
[4] F. London and H. London, Proc. R. Soc. A 149 (1935) 71.
[5] D.C. Mattis and J. Bardeen, Phys. Rev. 111 (1958) 412.
17
73
G. C IOVATI
[6] J. Bardeen, L.N. Cooper and J.R. Schrieffer, Phys. Rev. 108 (1957) 1175.
[7] R.E. Glover and M. Tinkham, Phys. Rev. 104 (1956) 844.
[8] M. Ginsberg and M. Tinkham, Phys. Rev. 118 (1960) 990.
[9] A. Citron, G. Dammertz, M. Grundner, L. Husson, R. Lehm and H. Lengeler, Nucl. Instrum. Meth-
ods 164 (1979) 31.
[10] V.L. Ginzburg, Sov. Phys. JETP 7 (1958) 78.
[11] P.G. de Gennes, Solid State Commun. 3 (1965) 127.
[12] J. Matricon and D. Saint-James, Phys. Lett. A 24 (1967) 241.
[13] T. Yogi, G.J. Dick and J.E. Mercereau, Phys. Rev. Lett. 39 (1977) 826.
[14] J. Gallop, Supercond. Sci. Technol. 10 (1997) A120.
[15] J. Zmuidzinas, Annu. Rev. Condens. Matter Phys. 3 (2012) 169.
[16] F. Furuta et al., Proc. 2006 Europ. Part. Accel. Conf., Edinburgh, Scotland, 2006 (European Physi-
cal Society Accelerator Group, Geneva, Switzerland, 2006), p. 750.
[17] G.E.H. Reuter and E.H. Sondheimer, Proc. R. Soc. A 195 (1948) 336.
[18] C.J. Gorter and H.B.G. Casimir, Physica 1 (1934) 306.
[19] L.N. Cooper, Phys. Rev. 104 (1956) 1189.
[20] A.B. Pippard, Proc. R. Soc. A 216 (1953) 547.
[21] J. Halbritter, FORTRAN-Program for the computation of the surface impedance of superconduc-
tors, Karlsruhe Nuclear Research Center, Technical Report 3/70-6 (1970).
[22] http://www.lepp.cornell.edu/~liepe/webpage/researchsrimp.html.
[23] A. Gurevich, Rev. Accel. Sci. Tech. 5 (2012) 119.
[24] A. Phillipp and J. Halbritter, IEEE Trans. Mag. MAG-19 (1983) 999.
[25] C. Benvenuti, S. Calatroni, I.E. Campisi, P. Darriulat, M.A. Peck, R. Russo, and A.-M. Valente,
Physica C 316 (1999) 153.
[26] B. Aune, R. Bandelmann, D. Bloess, B. Bonin, A. Bosotti, M. Champion, C. Crawford, G. Deppe,
B. Dwersteg, D.A. Edwards, H.T. Edwards, M. Ferrario, M. Fouaidy, P.-D. Gall, A. Gamp, A.
Gössel, J. Graber, D. Hubert, M. Hüning, M. Juillard, T. Junquera, H. Kaiser, G. Kreps, M. Kuchnir,
R. Lange, M. Leenen, M. Liepe, L. Lilje, A. Matheisen, W.-D. Möller, A. Mosnier, H. Padamsee,
C. Pagani, M. Pekeler, H.-B. Peters, O. Peters, D. Proch, K. Rehlich, D. Reschke, H. Safa, T.
Schilcher, P. Schmüser, J. Sekutowicz, S. Simrock, W. Singer, M. Tigner, D. Trines, K. Twarowski,
G. Weichert, J. Weisend, J. Wojtkiewicz, S. Wolff and K. Zapfe, Phys. Rev. ST Accel. Beams 3
(2000) 092001.
[27] A. Gurevich and G. Ciovati, Phys. Rev. B 87 (2013) 054502.
[28] P. Kneisel and B. Lewis, Additional RF-surface resistance in superconducting niobium cavities
caused by trapped magnetic flux, Thomas Jefferson National Accelerator Facility, Technical Report
94-028 (1994).
[29] B. Bonin and R.W. Röth, Proc. 5th Workshop on RF Supercondivity, Hamburg, Germany, 1991, p.
210.
[30] M.A. Hein, Studies in High Temperature Superconductors, Ed. A. Narlikar (Nova Science Publish-
ers, Hauppauge, NY, 1996), vol. 18, pp. 141-216.
[31] C.P. Bean and J.D. Livingston, Phys. Rev. Lett. 12 (1964) 14.
[32] V.P. Galaiko, Sov. Phys. JETP 23 (1966) 475.
[33] L. Kramer, Phys. Rev. 170 (1968) 475.
[34] G. Catelani and J.P. Sethna, Phys. Rev. B 78 (2008) 224509.
[35] R.C. Dynes, V. Narayanamurti and J.P. Carno, Phys. Rev. Lett. 41 (1978) 1509.
18
74
[36] R.C. Dynes, J.P. Carno, J.P. Hertel and T.P. Orlando, Phys. Rev. Lett. 53 (1984) 2437.
[37] A.A. Abrikosov, Sov. Phys. JETP 46 (1964) 1464.
[38] F. Pei-Jen Lin and A. Gurevich, Phys. Rev. B 85 (2012) 054513.
[39] T. Hays and H. Padamsee, Proc. 8th Workshop on RF Superconductivity, Abano Terme, Italy, 1997,
p. 789.
[40] A. Gurevich, Appl. Phys. Lett. 88 (2006) 012511.
[41] A. Gurevich, Physica C 441 (2006) 38.
Bibliography
H. Padamsee. RF Superconductivity: Science, Technology, and Applications (Wiley-VCH, Weinheim,
2009). H. Padamsee, J. Knobloch, and T. Hays. RF Superconductivity for Accelerators (John Wiley &
Sons, New York, 1998).
19
75
Material Properties at Low Temperature
P. Duthil 1
Institut de Physique Nucléaire d’Orsay, IN2P3-CNRS/Université de Paris Sud, Orsay, France
Abstract
From ambient down to cryogenic temperatures, the behaviour of materials
changes greatly. Mechanisms leading to variations in electrical, thermal,
mechanical, and magnetic properties in pure metals, alloys, and insulators
are briefly introduced from a general engineering standpoint. Data sets are
provided for materials commonly used in cryogenic systems for design
purposes.
Keywords: cryogenics, diffusion of heat, electrical conductivity, mechanical

properties.
1 Introduction
The properties of materials are used in specific equations to predict the behaviour of a system. They
depend on chemical composition, crystalline architecture, atomic and charge interactions, heat or
mechanical treatments, etc., and are affected by temperature changes. Thus, understanding the
mechanisms that contribute to relevant properties is important when considering changes in materials
with temperature. For details about transport phenomena properties in matter (e.g. thermal or electrical
properties), one can refer to any textbook on solid-state physics (see, for example, Refs. [1, 2]).
Magnetism or superconducting states also affect material properties. The reader is encouraged to refer
to dedicated lectures (which are too specific to be presented here) at the CERN Accelerator School.
References on data concerning materials used in cryogenics are given in Ref. [3].
2 Thermal properties
The thermal (and electrical) properties of any material are related to the vibrations of its atoms around
their equilibrium positions (in a lattice crystal). The amplitude of these vibrations depends on
temperature and diminishes as the temperature decreases. Note that these vibrations may propagate
within the material at the speed of sound, and are studied as plane waves, with which phonons are
associated. Thermal properties also depend on the movements of negative charges (electrons) and
positive charges (vacancies) if the material is a conductor.
2.1 Heat capacity

The heat capacity C is defined as the quantity of energy (heat) that must be introduced into some
mass of material to increase its temperature T by 1 K. Reversibly, it is the quantity of energy
extracted from this mass of material to decrease its temperature by 1 K. The unit of this extensive
quantity is J·K−1. The intensive property is given by the specific heat c , which is the heat capacity or
thermal capacity per unit of mass (J·kg−1·K−1) or the molar heat capacity (J·mol−1·K−1). The heat
capacity of a material is thus its ability to store or release heat energy. It is an important property in
cool-down or warm-up processes used in the estimation of the energy (and cost) involved, and in the
1
duthil@ipno.in2p3.fr
P. D UTHIL
assessment of the transient states of the heat transfer as it relates to the thermal diffusivity (see
Section 2.2.5). As the temperature tends to zero, so does the heat capacity.
Heat can be supplied at constant volume V or constant pressure p , defining two heat capacities
such that:
 ∂S   ∂U   ∂S   ∂H 
=CV T=
    , C p T=
=     , (1)
 ∂T V  ∂T V  ∂T  p  ∂T  p
where S is the entropy, U is the internal energy, and H is the enthalpy (all having units of joules, J).
For a system comprising the association of several homogenous materials i each having a mass mi :
CV = ∑ mi cV i and C p = ∑ mi c p i . (2)
i i
For solids at low temperatures, the difference C p − CV is negligible (less than 1%), and thus no
distinction will be made in the following between CV and C p .
2.1.1 Crystal lattice contribution

Considering the sum of the energies of the phonon modes, the Debye model enables the expression of
the total energy of the crystal lattice as a function of the Debye temperature θ D :
3 3
 T  θD /T x 3 θ 
= NkB   ∫
U 9= dx 3R D3  D  , (3)
 θD 
x
0 e −1 T 
with
vph  3 N 1/3 3 x t3
θD =   ,
kB  4π V 
D3 =
x3 ∫ 0 et − 1
dt , (4)
where = kB 1.38 ⋅ 10−23 J·K−1 is the Boltzmann constant,

=  1.055 ⋅ 10−34 J·s is the reduced Planck
constant, vph is the speed of sound in the material, N / V is the number of atoms per volume of
material, and D3 is called the ‘third Debye function’. The Debye temperature of some materials is
given in Table A.1 of Appendix A.
For n = 1 mole of material, noting that R = kB N A and that N = nN A , the derivation of Eq. (3)
with respect to the temperature gives the crystal lattice contribution to the molar heat capacity cph (the
subscript ‘ph’ refers to phonons):
3
cph T  θD /T x 4e x
3R
= 3 
 θD 
∫
0
( e x − 1)
2
dx . (5)
It follows that, for any material, the ratio cph / (3R ) can be plotted as a unique function of T / θ D , as
shown in Fig. 1. As T increases, cph / (3R ) tends to 1 (see the black solid curve of Fig. 1) and thus
cph → 3R . For T << θ D , the limit of cph can be found to be (T / θ D )312π 4 / 5 , showing a cubic
dependence of the lattice heat capacity with respect to the temperature (see the grey dashed line of
Fig. 1).
78
M ATERIAL P ROPERTIES AT L OW T EMPERATURE
Fig. 1: cph / (3R ) as a function of T / θ D ; solid black curve and black axis for T / θ D ∈ [0; 2.5] ; dashed grey
curve and grey axis for cph / (3R) ∈ [0;0.01] .
2.1.2 Contribution of free electrons

Thermal excitation induced by heating concerns only a fraction of the electrons for which the energy
is within kBT of the Fermi energy, EF = (  / 2me )(3π 2 N / V ) 2/3 , where=me 9.109 ⋅ 10−31 kg is the
mass of an electron. The total energy of the electrons is U ≈ ( N ⋅ T / TF )kBT , where TF = EF / kB is
called the ‘Fermi temperature’. The electronic heat capacity given by the free-electron-gas model thus
evolves linearly with respect to the temperature:
Ce =∂U / ∂T ≈ (T / TF ) NkB ( unit: J·K -1 ) . (6)
At temperatures much lower than the Fermi temperature, the molar heat capacity (of free
electrons) can be expressed by Ce= ½ ⋅ π 2 NkBT / TF .
2.1.3 Heat capacity of materials

The heat capacities of different materials (metals, thermal insulators, and gases for comparison) are
plotted in Fig. 2. Reference [4] provides the heat capacities of some materials as functions of the
temperature. The tendencies of the curves are explained in the following sections.
Fig. 2: Specific heat c (J·kg−1·K−1) of some materials versus the temperature T compiled from Ref. [5]. Metals
are plotted with solid lines, metallic alloys with dot-dashed lines, and thermal insulators with dotted lines. For
comparison, the specific heats of two gases commonly used in cryogenics are plotted with dashed lines.
79
P. D UTHIL
2.1.3.1 Heat capacity of metals

The heat capacity of metals can be seen as the sum of the contributions from the crystal lattice and the
free electrons. Thus, at low temperature ( T < θ D / 10 and T << TF ), C =Ce + Cph =γ T + AT 3 , where A
and γ are constants related to the material. Experimentally, C / T is plotted versus T 2 ; this yields a
linear line with slope A and intercept γ . Some values of γ are given in Table A.2. For higher
temperatures ( T > 2θ D ), cph  3R , and C increases linearly as the temperature rises.
Below 10 K ( cph << 1 ), the linear electronic term dominates, and thus the heat capacity is
proportional to the temperature.
2.1.3.2 Heat capacity of thermal insulators

For crystallized thermal insulators, the contribution of the phonons is predominant. Therefore,
cph  3R for T > 2θ D and cph = f (T 3 ) for T < θ D / 10 .
2.1.3.3 Heat capacity of superconductors

For superconductors at temperatures below their critical temperature TC , electrons are linked in
Cooper pairs and do not contribute to energy transport. The heat capacity can then be written using
BCS theory as [6]:
cs γ TC a ⋅ e(
=
− bTC
T ) for T < TC , (7)
where γ is the constant coefficient of the electronic contribution measured at T > TC , and a and b are
other constants that depend on the material.
2.1.4 Sensible heat

During a thermodynamic process at constant pressure between two temperatures T1 and T2 , one can
write the specific enthalpy change (unit J·K−1) as follows:
T2
∆h =∫ c p dT .
T1
Thus, the energy involved in the process is E = ∆H = m ⋅ ∆h ; it is called ‘sensible heat’,

referring to a ‘sensible’ change of temperature of the material. The parameter ∆h is seen as a heat
h = c dT
T
stock per unit of mass. The enthalpy ∫O p is plotted versus temperature for different materials in
Fig. 3.
At low temperature, it can be seen that the enthalpies of solid materials, such as G10 (which is a
thermal insulator made of epoxy resin and glass fibres) and stainless steel AISI 304L (EN 10088-
1&2 / 1.4306) for example, are much smaller than those of gases and especially than that of helium. A
consequence is that cooling down materials by removing heat with cold helium vapours (which then
warm up) is more efficient than by vaporizing liquid helium (the vaporization latent heat of helium-4
at a pressure of 1 bar is about 20.9 J·g-1) as smaller quantity of cold helium is thus involved in that
process.
80
h = ∫ c dT (J·kg−1) of various materials versus temperature T. Data were compiled from

T
Fig. 3: Enthalpy O
p
Ref. [5]. Metals are plotted with solid lines, metallic alloys with dot-dashed lines, and thermal insulators with
dotted lines. For comparison, the specific sensible heats of two gases are plotted with dashed lines.
2.2 Thermal conductivity

The thermal conductivity k (unit W·m−1·K−1) relates to the facility with which heat can diffuse into a
material. Fourier’s law yields the quantity of heat diffusing through a unit surface during a unit of time
within a material subjected to a temperature gradient:
 
q =−k ∇T (J·s−1·m−2 ≡ W·m−2). (8)
As an example, we can consider the linear (one-dimensional) support of Fig. 4, with length L,
cross-section A, and with its ends at temperatures TH and TC (subscripts H and C refer to hot and cold,
respectively). Applying Fourier’s equation, writing Q HC = AqHC , which is a constant along the linear
support (conservation of energy), we obtain
L Q HC L TC
∫ dx = Q HC = − ∫ k (T )dT . (9)
0 A A TH
Fig. 4: Conduction within a linear support: considered problem (left) and possible temperature solutions
depending on the thermal conductivity k (right).
81
P. D UTHIL
Thus, if k is constant, the quantity of heat per time unit diffusing within the support is given by
Q HC kA(TH − TC ) / L (unit: W), which is a positive quantity as the heat diffuses along the positive
=
x-direction from the higher to the lower temperature. The resulting temperature profile T(x) is linear.
However, the thermal conductivity depends on temperature, especially in the cryogenic domain,
leading to non-linear profiles (see Fig. 4). Simply put, heat is transported in solids by electrons and
phonons, and the thermal conductivity is seen as the sum of both contributions:
k= ke + kph .
2.2.1 Lattice contribution to the thermal conductivity

Considering the energy change c∆T of a particle travelling over a mean free path  ph between two
temperatures T and T + ∆T , the phonon contribution to the thermal conductivity kph is given by
1
k ph = c ph v ph  phVm , (10)
3
where Vm is the density (mass per unit volume) of the material. The mean free path  ph of a phonon
depends on (i) the Umklapp collisions with other phonons, giving to  ph a variation which is
proportional to e( −½θD /T ) at low temperature tending to a constant as T << θ D , and (ii) the geometrical
defaults and the crystal size limiting the mean free path to a length constant. Thus, for T << θ D , kph is
proportional to cph and is hence a cubic function of T (see Section 2.1.1).
2.2.2 Electronic contribution to the thermal conductivity

The electron contribution to the thermal conductivity is given by
1 1 π 2 nkBT
=ke =ce vF  eVm τ Vm , (11)
3 3 me
where vF is the Fermi velocity ( ½ ⋅ me vF2 =

EF ),  e = vFτ , the mean free path of the electrons, τ is the
mean collision time, and me is the mass of an electron. At low temperature, τ is constant and thus ke
is proportional to the temperature.
2.2.3 Thermal conductivity of materials

The thermal conductivities of different materials are plotted in Fig. 5.
Fig. 5: Thermal conductivity k of various materials compiled from Ref. [5]. Metals are plotted with solid lines,
metallic alloys with dot-dashed lines, and thermal insulators with dotted lines. Two gases are plotted with dashed
lines.
82
2.2.3.1 Thermal conductivity of metals

For pure metals, the electronic contribution dominates for any temperature: ke >> kph . At very low
temperatures, k is proportional to temperature. As T increases, the thermal conductivity is affected by
impurities included within the material. Thus, as a function of temperature, k has a maximum
depending on the purity of the metal: the higher the purity of the metal, the larger this maximum is,
and the lower the temperature of this maximum. Figure 6 plots the thermal conductivities of coppers
with different purities. The purity of the coppers is indicated by the Residual Resistive Ratio (RRR,
see Section 3.1): for ordinary coppers, 5 < RRR < 150; for Oxygen-Free High thermal Conductivity
(OFHC) coppers, 100 < RRR < 200; for very pure coppers, 200 < RRR < 5000.
For non-pure metals or metallic alloys, the phonon contribution may be comparable with the
electronic contribution: the thermal conductivity is thus a monotonous function of temperature that is
roughly linear at low temperature.
Fig. 6: Thermal conductivity kCu of coppers having different purities, indicated by the RRR. Data were compiled
from Ref. [5]. The thermal conductivities of purer coppers are plotted with darker solid lines, whereas those of
less pure coppers are plotted with lighter grey dotted lines.
2.2.3.2 Thermal conductivity of thermal insulators

For crystallized thermal insulators, the electronic contribution is much smaller than that of the lattice
3
contribution. Thus, their thermal conductivity is a function of T .
2.2.3.3 Thermal conductivity of superconductors

If T > TC (normal state), the thermal conductivity ks of superconductors behaves as for metals. If
T < TC (Meissner state), ks is a function of T 3 and is much smaller than that of the normal state. This
abrupt change in thermal conductivity is used to develop thermal interrupters.
2.2.4 Integrals of thermal conductivities

As explained in the preceding sections, because the thermal conductivity depends on temperature,
especially in the cryogenic domain, one must integrate it with respect to the temperature in order to
evaluate the diffusive heat flux between two temperatures (see Eq. (9)): ∫T k (T )dT . Hence, for
T j
engineering purposes, it is convenient to provide integrals of the thermal conductivities evaluated from
83
P. D UTHIL
a reference temperature TREF = Ti . They are tabulated for different materials considering TREF = 1 K in
Table A.3 of Appendix A. The integral ∫T k (T )dT can thus be evaluated by subtraction:
T j
i
Tj Tj Ti
∫ k (T )dT
=
Ti ∫
TREF
k (T )dT − ∫
TREF
k (T )dT .
2.2.5 Thermal diffusivity

Assuming no internal production of heat within a material, the heat diffusion equation is given by
∂T  
Vm (T )c p (T ) = ∇ ⋅ ( k ( T ) ∇T ) . (12)
∂t
In the case of homogeneous materials, the thermal conductivity k reduces to the scalar k, and if it
does not depend on temperature, nor on the density Vm and the specific heat capacity c p , Eq. (12)
simplifies as follows:
∂T k
= ∆T =κ∆T , (13)
∂t Vm c p
where κ (unit: m2·s−1) is the thermal diffusivity of the material. It can be seen as the time rate at
which the (curved) initial temperature distribution is irreversibly smoothed within the material: the
larger the thermal diffusivity, the faster the heat diffuses. It thus allows one to assess the time constant
of the heat diffusion process. In the cryogenic domain, as thermophysical properties of a material
change with temperature, the thermal diffusivity depends on the temperature, i.e. it generally increases
at low temperatures; for metals, the thermal conductivity and the heat capacity are, respectively, linear
and cubic functions of T, while the density is almost constant. The thermal diffusivities of some
materials are plotted in Fig. 7. Note that the heat diffusivity of stainless steel 304L is two orders of
magnitude lower than of G10 at low temperature. It follows that heat will diffuse more slowly in 304L
than in G10 at low temperature; also, 304L will be conductively cooled more slowly than G10 at low
temperature.
Fig. 7: Thermal diffusivity κ (m2·s −1) of various materials versus the temperature T. Metals are plotted with
solid lines and thermal insulators with dot-dashed lines. Data compiled from Ref. [5].
84
2.2.6 Thermal expansion/contraction of solids

The coefficient of expansion is defined as
1  dV 
αV = 
−1
 (unit: K ).
V  dT  P
Being a function of the temperature, the thermal expansion can affect the resistance of an assembly, by
generating large stresses, and/or the mechanical stability of an assembly (buckling). αV is generally
positive and so, at constant pressure, a temperature decrease induces a reduction of the physical
dimensions (size) of a body. Moreover, for a solid body the effect of pressure can be neglected. The
linear coefficient expansion is defined by
1 dL
α= (unit: K−1). (14)
L dT
For a crystallized solid, α varies as cph : at very low temperature it is a cubic function of the
temperature and tends to a constant value as T increases. In practice, the integral of the thermal
expansion coefficient is computed from a reference temperature TREF :
T ∆L L − LREF
∫
TREF
α (T
= ) =
L LREF
(no unit), (15)
where LREF is the length of the body at the reference temperature. Thus, ∆L / L is proportional to T
around the ambient temperature and becomes proportional to T 4 at low temperature (in practice the
coefficient of proportionality at low temperature is negligible). ∫T α (T )dT is plotted for different
T
REF
materials in Fig. 8, with LREF at TREF = 300 K. It can be seen that most of the thermal contraction is
achieved above 50 K or 77 K (boiling point of nitrogen at a pressure of 1 atm).
∫T α (T )dT as a function of the temperature for different solid materials

Fig. 8: ∆L / L =
T
REF =1 K
In cryogenic systems, components can be submitted to large temperature differences: at steady

state because they are linked to both cold and warm surfaces (as cold mass supports, for example), or
85
P. D UTHIL
during cool-down or warm-up transient states. Special care must then be taken with assemblies of
different solid materials experiencing such temperature changes in order to limit stresses and/or their
destruction, as illustrated in Fig. 9. The choice of materials might determine the outcome, and/or
flexibility in the assembly might be achieved by involving dedicated elements (compensation bellows,
elastic or ductile elements).
Fig. 9: Mechanical consequences of cooling a ‘rigid’ assembly of two solid materials having different thermal
expansion coefficients.
3 Electrical conductivity/resistivity
3.1 Electrical resistivity of metals

For metallic conductors, electric charges e are transported by n ‘free electrons’, and the electrical
conductivity is σ = n e e 2 (me vF ) , where me is the mass of an electron and  e denotes the mean free
path of electrons. The electrical resistivity is ρ = σ −1 . At ambient temperature, the electron free path
 e is dominated by electron scattering from thermal vibrations of the crystal lattice (phonons). At low
temperature, it is limited mainly by scattering off chemical and physical crystal lattice imperfections
(chemical impurities, vacancies, dislocations). Considering the two contributions, the electrical
resistivity can be written as (Matthiesen’s rule):= ρ (T ) ρ ph (T ) + ρi , the subscript i denoting
imperfections. ρi is generally not dependent on the temperature but varies from one sample of
material to another due to the imperfection content. However, ρ ph is the same for different samples of
a material as it depends on the frequency of the collisions of electrons with thermal phonons (and
electrons). For T > θ D , the concentration of phonons is proportional to the temperature, and so
ρ ph ∝ T and thus ρ ∝ T . As T → 0 , ρ → ρi , with ρi proportional to T n (with 1 < n < 5 ; see Refs.
[1, 6, 7]). The electrical resistivities of several materials are plotted in Fig. 10.
As ρT →0 (T ) = ρi (0) , an indication of the impurities and crystallographic defects in a metal
(imperfections) is provided by the determination of the Residual (electrical) Resistivity Ratio (RRR):
ρ (273 K) ρ (273 K)
= RRR ≈ . (16)
ρ (0 K) ρ (4.2 K)
86
Therefore, the smaller the imperfection content in a material, the smaller the ρ (0) ≈ ρi (0) and
the larger the RRR (see the example for copper in Fig. 11).
The empirical Wiedemann–Franz law indicates that the ratio ke / σ has almost the same value
for different metals at the same temperature. Theoretically,
2
ke π 2  k B  −2
=   2.445 ⋅ 10 (W·Ω·K ),
= (17)
−8
σT 3  e 
and thus, the more pure a material is, the larger its electronic thermal conductivity (see
Section 2.2.3.1).
Fig. 10: Electrical resistivity ρ (Ω·m) of various materials versus temperature T . Metals are plotted with solid
lines, metallic alloys with dot-dashed lines, and two superconducting alloys with dotted lines.
Fig. 11: Electrical resistivity ρ of different coppers [5]. Impurities and crystallographic defect content are
indicated by the RRR. Electrical resistivities of purer coppers are indicated by darker solid lines, whereas those
of less pure coppers are plotted with lighter grey dotted lines.
87
P. D UTHIL
3.2 Electrical resistivity of semiconductors

For semiconductors, electric charges are transported by conduction band electrons and holes in the
valence band. Around the ambient temperature, lattice vibrations are predominant and electrical
properties are not modified by impurities. The electrical resistivity can be expressed as
ρ (T )= a ⋅ eδ 2 kBT , where a is an experimental constant and δ is the energy band depending on the
material [6]. At low temperature, lattice vibrations are negligible and impurities play an important role
in the transport of charges. Thus, the resistivity of semiconductors is very non-linear: it typically
increases as temperature drops due to there being fewer electrons in the conduction band. An
important application of this property is the use of semiconductors as temperature sensors
(thermistors). Figure 12 presents the electrical resistance of some metallic and semiconductor
thermistors.
Fig. 12: Resistivity of various thermistors (from [8]). Resistivities of metallic sensors (Pt, RhFe) are plotted as
solid black symbols; resistivities of semiconductor sensors (TVO, Allen Bradley, Cernox) are plotted as grey
symbols.
4 Mechanical properties
In solids, atoms are in equilibrium positions depending on their interaction energies. Moving them
from their equilibrium positions requires imposing an external force to overcome the restoring force.
For small displacement amplitudes, the restoring force can be considered as proportional to this elastic
deformation, and the equilibrium position is recovered if the external force is removed. For a larger
displacement (larger external force), a limited number of atomic bonds break. Those dislocations
enable atoms in specific crystal planes to slip one from another, yielding a permanent plastic
deformation that propagates within the material along the densest planes of atoms because the (shear)
stress needed to move increases with the spacing between the planes.
Useful macroscopic mechanical properties can be obtained by performing a tensile test (see
Fig. 13) during which a homogeneous standardized test specimen is subjected to a controlled tension
force F, increased monotonically, and then stretched until it breaks. Considering the initial dimensions
of the specimen (cross-section S0 and length L0), the average linear stress s = F / S0 and strain
ε= ∆L / L0 =( L − L0 ) / L0 are plotted to outline the engineering stress–strain curve. True stress–strain
88
curves can also be plotted considering the volume conservation of the specimen, S0 L0 , and measuring
the real cross-section S, which reduces during necking.
Fig. 13: Tensile test of a specimen: engineering stress–strain curve
4.1 Elastic domain

In the elastic domain, the elongation is proportional to the tensile force (Hooke’s law), and the
specimen can reversibly recover its initial size by removing the force. The proportional coefficient is
the Young modulus (modulus of elasticity) defined as E = s / ε (Pa); it is a measure of the stiffness
E S0 / L0 of the material, and is used in computing structural deflections. The modulus of elasticity is
determined by the binding forces between atoms and is thus generally not affected by impurities
within the lattice crystal (chemical additions to obtain alloys, for example). However, temperature
does affect these forces due to the thermal excitation of the lattice: the modulus of elasticity is slightly
larger at low temperature than around ambient and drops at high temperature (see Fig. 14).
Fig. 14: Tensile stress–strain curve of annealed Ti-6Al-4V at different temperatures (from Ref. [9])
89
P. D UTHIL
The elastic limit or yield stress, Re, is the tensile stress from which elongation of the sample
begins to increase disproportionately with increasing tensile force. A plastic elongation remains after
the removal of the force. The yield point is not always a sharply defined point, especially for hard
materials, as stress concentrations alternate with local line or band deformation steps within the
material. The elastic to plastic transition can exhibit drops and fluctuations on the engineering stress–
strain curve (see the grey solid line in Fig. 13). It leads to the definition of the upper- and lower-limit
yield stresses YSU and YSL. For engineering purposes, offset yield strengths are often used: they are the
tensile stresses from which a permanent plastic deformation of 0.1%, 0.2% or 0.5% remains, and are
respectively referred to as proof strengths RP0.1, RP0.2, and RP0.5. The yield strength generally increases
at low temperature (see Fig. 15).
Resilience is the ability of a material to absorb energy when deformed elastically and to return it
when unloaded. The maximum resilience occurs at the yield point, and is called the ‘resilience
modulus’, U R = ½ Re 2 / E (J·m−3). It is the area under the linear part of the engineering stress–strain
curve. A resilient material (used to make a spring, for example) thus has a large yield stress and a
small modulus of elasticity. As the temperature decreases (the relative variation of Re being generally
greater than that of the modulus of elasticity (see Fig. 15)), the resilience modulus increases at low
temperature.
Fig. 15: Yield strength of various materials changing with temperature (from Ref. [9])
4.2 Plastic domain

The ultimate strength is the maximum tensile stress a material can withstand (under uniaxial loading),
but it is not often used in practice. A material is said to be ductile when it can be plastically deformed
without breaking. Otherwise it is said to be brittle. Ductility allows a material to be submitted to metal
working without breaking, and gives the engineer a security margin when excess loading is applied to
the material. It can be assessed by measuring the elongation and cross-section at fracture. The ductility
of metals and metallic alloys depends on their crystal structure and temperature (see Figs. 16 and 17).
Materials with a face-centred cubic (f.c.c.) crystal structure (such as Cu-Ni alloys, aluminium and its
alloys, austenitic stainless steel, Ag, Pb, brass, Au, Pt, inconel), are ductile even at low temperatures.
Thus, they are favoured in cryogenics applications, as ductility provides some safety margin from
rupture. For materials with a body-centred cubic (b.c.c.) crystal structure (ferritic steels, carbon steel,
90
steel with Ni < 10%, Mo, Nb, Cr, Nb−Ti), a ductile–brittle transition appears at low temperature: the
plasticity capacity is wiped out. For other structures (Zn, Be, Zr, Mg, Co, Ti alloys etc.), no general
trend appears: mechanical properties depend on interstitial components and on the symmetry axis.
Fig. 16: Ductile to brittle transition of some b.c.c. atomic lattice materials when cooled down
Fig. 17: General trend of yields for f.c.c. and b.c.c. materials. Temperature is normalized by the melting
temperature Tm ; stress is normalized by the shear modulus G . Originally from Ref. [10].
The toughness of a material describes its ability to absorb energy and plastically deform until
fracture. A tough material can withstand occasional stresses above the yield stress without fracturing.
The toughness of f.c.c. materials can be judged from tensile tests [11]. The modulus of toughness UT
is the amount of work per unit volume (J·m−3) that can be given to a material without causing it to
fracture, and is the total area under the stress–strain curve. Toughness is thus based on strength and
ductility. Notch or fracture toughness is the ability of a notch specimen to absorb elastic or kinetic
energy up to fracture. A tensile test or a high-strain-rate test, such as the Charpy impact test, are used
to measure it. It is expressed by way of a critical stress intensity factor K (Pa·m0.5), at which point a
thin crack begins to grow. It is the resistance of a material to brittle fracture when a crack exists. As
the temperature decreases, both toughness and fracture toughness decrease. As shown in Fig. 18, the
values of fracture toughness of high-strength nickel and titanium b.c.c. alloys drop at low temperature,
whereas the value remains large for f.c.c. stainless steel 310.
91
P. D UTHIL
Fig. 18: Fracture toughness of materials: plane-strain critical-stress intensity factor for cracking mode I (named
“opening mode” and for which a tensile stress normal to the plane of the crack is applied). From Ref. [9].
5 Magnetic properties

In a vacuum, the magnetic
 field B (unit: T ≡ V⋅s⋅m−² ≡ N⋅A−1⋅m−1) is proportional to the excitation
magnetic field H (unit: V·s·A−1·m−1 ≡ A⋅m−1): B = µ0 H , the proportional   factor being
 the

permeability of free space µ= 0 4π ⋅ 10−7 (N⋅A−2). In a material,
= B µ0 ( H + M ) , where M = χ H
(unit: A⋅m−1) is the magnetization
 and represents how strongly a region of material is magnetized by
the application of H . It is the net magnetic dipole moment per unit volume. Thus,
B =µ0 (1 + χ ) H =µ0 µ r H , where µ r is the relative magnetic permeability. The magnetic moment of a
free atom depends on electron spin, the orbital kinetic moment of the electrons around the nucleus,
and the kinetic moment change induced by the application of a magnetic field. Five types of magnetic
behaviour can be distinguished: diamagnetism and paramagnetism, which are induced by isolated
atoms (ions) and free electrons, and ferromagnetism, antiferromagnetism, and ferrimagnetism, which
occur due to the collective behaviour of atoms.
5.1 Diamagnetic materials

If the magnetic susceptibility χ = µr − 1 < 0 , it causes a diamagnet to create a magnetic field in
opposition to an externally applied magnetic field. When the field is removed, the effect disappears.
Examples
 of diamagnetic
 materials
 are silver, mercury, diamond, lead, and copper. If the (small) field
H is applied, then M = χ H and χ does not depend on temperature. Type I superconductors (such as
copper and niobium) are perfect diamagnets for T < TC .
92
5.2 Paramagnetic materials

A magnetic susceptibility χ = µr − 1 > 0 (with χ small) induces paramagnets (such as aluminium) to
be slightly attracted by an externally applied magnetic field. Since the spin of unpaired electrons in the
atomic orbitals is randomized, the induced permanent magnetic moment (dipoles) has no effect.
However, if a magnetic field is applied, the dipoles tend to align with the applied field, yielding a net
magnetic moment. When the field is removed, the effect disappears. For low levels of magnetization,
χ = C / T , where C is the Curie constant. An application of this dependence of the magnetic
susceptibility with temperature is the principle of magnetic thermometers.
5.3 Ferromagnetic materials

For ferromagnetic materials, such as Fe-Ni-Co alloys (not austenitic steels), magnetization is due to
the contribution of the unpaired electron spins (as for paramagnets) and coupling interactions causing
magnetic moments of adjacent atoms to be parallel to an applied magnetic field and parallel to each
other. Thus, magnetization remains.
Above the Curie temperature θ C , which is generally larger than the ambient temperature, the
Curie–Weiss law yields a paramagnetic behaviour:
= χ cst / (T − θ C ) .
Below θ C , ( M (0) − M (T ) ) / M (0) ∝ T 3/ 2 . When an increase in the applied external magnetic
field H cannot augment the magnetization M, the material reaches its saturation state. Figure 19 shows
the saturation magnetization M s of nickel as a function of (normalized) temperature.
Fig. 19: Magnetization at saturation of nickel as a function of temperature (from Ref. [1])
5.4 Antiferromagnetic materials

For antiferromagnets, the tendency of the magnetic moments of neighbouring atoms is to point in
opposite directions. A substance is antiferromagnetic when all atoms are arranged such that each
neighbour is ‘anti-aligned’. Antiferromagnets have a zero net magnetic moment below a critical
temperature called the Néel temperature, θ N , and thus B = 0 for T < θ N . Antiferromagnets can exhibit
diamagnetic and ferrimagnetic properties for T >= θ N : χ cst / ( T + θ N ) .
References
[1] C. Kittel, Introduction to Solid State Physics, 1st–8th eds. (John Wiley and Sons, New York,
1953–2005).
[2] M. Brousseau, Physique du solide. Propriétés électroniques (Masson, Paris, 1992).
93
P. D UTHIL
[3] J.G. Weisend II et al., A reference guide for cryogenic properties of materials (SLAC-TN-03-
023, Stanford, Calif., 2003).
[4] E.D. Marquardt et al., Cryocoolers 11 (2002) 681.
[5] Cryocomp, Hepak, CRYODATA software, Cryodata Inc. (1999).
[6] A. Bui et al., Techniques de l’ingénieur B 2 (1993) 380.
[7] J.W. Ekin, Experimental Techniques for Low Temperature Measurements (Oxford University
Press, Oxford, 2006).
[8] Amand J.-F. et al., Neutron irradiation tests in superfluid helium of LHC cryogenic
thermometers, ICEC'17 Bournemouth, UK, July (1998).
[9] Atlas of Stress-Strain Curves, 2nd ed. (ASM International, 2002).
[10] M.F. Ashby, Acta Metall. 20 (1972) 887.
[11] ASTM E23, EN-Standard test methods for notched bar impact testing of metallic materials
(1993).
Appendix A: Tabulated quantities
Table A.1: Debye temperatures of various metals
Material θD (K)
Copper 343
Aluminium 428
Titanium 428
Niobium 275
Nickel 450
AISI304 470
AISI316 500
Table A.2: γ coefficient of the linear dependency of the heat capacity of various metals on temperature
Material γ (10-3
J·kg−1·K−2)
Copper 11.0
Aluminium 50.4
Titanium 74.2
Niobium 94.9
Nickel 124.0
94
∫ k (T )dT of different materials; data compiled from [5] and considering TREF = 1 K
Tj
Table A.3: Ti
T SS304 Cu- Cu- Brass Constantan Manganin Invar Ti- Al- 6061- 5083- Nb−
(K) RRR=300 RRR=30 6Al- RRR= T6 T0 Ti
4V 30
2 0.073 69 1.0 0.18 0.124 0.0388 0.17 42.8 3.5 1.06 0.04
4 0.40 3560 345 6.0 1.31 0.773 0.276 0.80 214 17.7 5.61 0.27
6 1.02 8360 807 16.0 3.87 2.12 0.819 1.78 501 41.4 13.7 0.76
8 1.96 14 900 1450 31.0 8.03 4.22 1.7 3.07 900 74.6 25.3 1.5
10 3.3 22 800 2270 51 13.9 7.1 2.95 4.67 1410 118 40.5 2.5
15 8.5 46 600 5130 128 38.2 18.6 7.93 9.91 3190 272 95.2 6.0
20 16.7 72 900 8910 235 74.8 35.9 15.6 16.7 5590 487 173 11.1
25 28.1 95 800 13 500 370 124 59.7 26 24.7 8560 765 273 17.4
30 42.8 115 000 18 400 525 184 89.6 39.1 33.8 11 900 1100 395 25.2
35 61.2 130 000 23 300 697 252 125 54.7 43.8 15 400 1480 538 34.4
40 82.9 140 000 28 000 883 328 166 72.9 54.8 18 900 1900 701 45.0
50 136 155 000 36 200 1280 497 260 117 79.4 25 300 2840 1080 70.4
60 199 164 000 42 900 1730 679 367 170 107 30 500 3900 1540 101
70 271 171 000 48 400 2210 865 483 232 137 34 800 5020 2050 138
77 326 176 000 51 800 2580 997 569 281 160 37 300 5830 2440 167
80 350 177 000 53 300 2740 1050 607 302 171 38 300 6180 2610 180
90 436 182 000 57 800 3320 1250 739 379 207 41 400 7370 3220 228
100 527 187 000 62 000 3950 1440 877 462 245 44 200 8580 3870 280
120 725 196 200 70 270 5330 1847 1165 640 329 49 240 11 040 5280 398
140 940 204 900 78 200 6860 2269 1467 834 422 53 830 13 560 6820 530
160 1170 213 300 86 100 8500 2700 1781 1040 522 58 300 16 130 8490 673
180 1414 221 700 94 000 10 240 3140 2107 1258 630 62 800 18 780 10 270 824
200 1667 229 900 101 800 12 080 3600 2447 1482 744 67 300 21 480 12 170 983
220 1937 238 200 109 600 13 950 4060 2797 1732 865 71 800 24 180 14 170 1150
240 2207 246 300 117 400 16 050 4530 3167 1982 993 76 400 27 080 16 370 1323
260 2487 254 400 125 100 18 150 5000 3557 2242 1127 80 900 30 080 18 570 1503
280 2777 262 500 132 900 20 350 5480 3967 2502 1265 85 400 33 180 20 970 1687
300 3077 270 500 140 600 22 650 5970 4397 2772 1415 90 000 36 380 23 470 1875
95
Higher-order Modes and Heating
B.J. Holzer 1
Abstract
This chapter gives a basic introduction to the problem of wake fields created
in the beam-surrounding environment and the resulting heating effects of
machine components. The concepts are introduced and scaling rules derived
that are exemplified by several observations from operation of the LHC and
other machines.
Keywords: wake fields, heating, accelerators, high-order modes.
1 Introduction
In the treatment of transverse as well as longitudinal beam dynamics, the movement of the particles is
described under the influence of the external fields, in general ignoring – or neglecting – the
interaction of the particles with the fields created directly or indirectly by the beam itself.
The situation visualized in Fig. 1 is therefore highly idealized: it shows the result of a particle
trajectory (green) following the usual rules of focusing and defocusing fields, within the beam
envelope defined by the β-function (red) but without any influence from self or image currents [1].
And at least in the case of intense beams, these additional aspects have to be taken into account.
Fig. 1: A single-particle trajectory in a storage ring
Both the dynamics of individual particles as well as the collective dynamics of the complete
particle distribution within a bunch and between different bunches depend greatly on the interaction
between the electromagnetic field of the particles and their environment; that is, in general, the
properties and geometry of the vacuum chamber. The treatment is usually based on the concept of
impedance and therefore a hopefully useful reminder follows.
While in an ordinary electrical DC circuit, the potential and current are related by Ohm’s law,
U = RI ,
the relation between an AC voltage with a frequency ω and phase ϕU,
1
B.J. H OLZER
=U ac U 0 cos(ωt − φU ) , (1)
and the corresponding current

=I ac I 0 cos(ωt − φI ) , (2)
usually written more elegantly in complex notation, as

U C = U 0 e(iωt −φU ) , U C = I 0 e(iωt −φI ) , (3)
is defined by the impedance Z:

U C U 0 − (φU −φI )
Z
= = e . (4)
IC I0
It is convenient to use the same concept for the relation of the particle beam (representing just
another AC current) and the voltage induced via the interaction of its fields with the beam-surrounding
material; that is, the vacuum chamber.
Before doing so, a second reminder is added here. A charged particle in its rest frame is
surrounded by an electric field that propagates isotropically in all directions. Being accelerated to
larger and larger velocities, this field, observed in the laboratory frame, is Lorentz contracted and the
field lines spread out longitudinally only within an angle of ±1/γ. While for some low-energy heavy
ion rings the relativistic γ-parameter can still be quite low, we will restrict the description here to ultra-
relativistic electron or proton rings. For example, in the case of the LHC running at 7 TeV beam
energy, we obtain 1/γ = 1.3 × 10–4.
Qualitatively, the situation obtained is shown in Fig. 2.
Fig. 2: The field lines of a particle at rest and moving at higher and higher velocities. (Courtesy of D. Brandt,
in [2].)
Due to these transversely extending field lines, an image current is created that is floating inside
the vacuum chamber walls, travelling with the bunch along the accelerator structure. In a uniform
vacuum tube with perfectly conducting walls, the image current is floating without losses and no net
forces are generated that would act back on the bunch. In summary, if an ultra-relativistic beam is
travelling in a perfect orbit, in a perfectly conducting and perfectly smooth vacuum chamber, no
heating effects are obtained (and there are no collective instabilities).
Unfortunately, these three conditions are not realistic. In particular, the vacuum chambers used
in storage rings are designed following a number of different criteria, and they are far from being
superconducting or smooth (even if, nowadays, ‘smooth-fullness’ is increasingly part of a successful
design).
Due to this non-ideal situation, the image charges are retarded, the field lines sketched in Fig. 2
are distorted, and the corresponding ‘wake’ fields act back on the beam (Fig. 3).
98
H IGHER - ORDER M ODES AND H EATING
Fig. 3: The field lines of a particle bunch in a non-uniform vacuum chamber geometry for a sudden change of
the chamber geometry and a cavity-like object.
Both effects – the resistivity of the chamber material and the field retardation – will lead to
longitudinal components of the electrical field and as a consequence they will lead to energy loss of
the beam and at the same time to heating effects in the vacuum system.
1.1 Wake fields and impedances

The interaction of the charged particle beam with its environment can be treated either in the time
domain or in the frequency domain, and the two descriptions are completely equivalent.
In the time domain, the explicit fields are calculated that act back on the beam. In the frequency
domain, the vacuum chamber components are treated as (frequency-dependent) impedances that
describe the relation between the stored beam current and the corresponding voltage. A strong
coupling between beam and environment is expected if the frequency spectrum of the bunched beam
and the impedance of the surrounding chamber have significant components at the same frequencies.
In accordance with the original concept of impedance (Eq. (4)), we set
V (ω ) = − Z (ω ) I (ω ) , (5)
where the minus sign indicates the decelerating effect, leading to energy loss of the particles. The
impedance is, in general, a complex number and depends on the shape and material of each chamber
component.
Further differentiation is useful in this context: for a particular component of the vacuum
chamber, we can obtain wake fields due to sudden changes in the chamber cross-section – these
typically have ‘RF quality factors’ of Q ≈ 1 – or due to cavity-like objects (as on the right-hand side of
Fig. 3) with Q ≫ 1. As in the case of classical oscillators, the quality factor (describing the damping
term of the system) and the bandwidth are related, and so we talk about narrow bandwidths for Q ≫ 1
and broadband systems for Q ≈ 1, where the quality factor Q of the system is defined as the ratio
between the stored energy in the cavity and the energy loss per period [3]:
stored energy
= 2π ×
Q . (6)
energy loss per period
As an example, the resonance curves for three significantly different Q-factors, Q = 1, 10, and
100, in storage ring cavities are shown in Fig. 4. The extreme is obtained in superconducting cavities
where, for example, for the LHC resonators, values as high as Q = 8 × 104 are obtained, leading to an
extremely narrow resonance curve.
99
B.J. H OLZER
Fig. 4: Resonance curves for different quality factors, Q. (Courtesy of E. Jensen, in [3].)
The effect on the beam depends strongly on this Q-factor and thus on the details of the chamber
geometry.
Cavity-like objects, with their high quality factors, have a narrow bandwidth and couple to the
beam in only a small frequency band. However, the fields created by a bunch inside these objects will
stay long enough to react on subsequent bunches or even – after one turn – back on the same bunch.
Low quality factors describe objects such as sudden changes of the vacuum chamber size; they
have a broad frequency bandwidth but in general will not act on subsequent bunches. However, due to
their frequency spectrum they will easily couple back on the bunch that created them.
An extreme example of the first case is given in Fig. 5: it shows the detector beam pipe of the
HERMES experiment [4] that had been installed in the HERA collider. To achieve the highest
luminosity and the best detector resolution, the beam pipe radius at the IP was reduced to about 1 cm,
and careful tapering to the accelerator standard vacuum chamber was needed to limit the energy
deposit due to the wake fields to within tolerable limits.
Fig. 5: The detector beam pipe of the HERMES experiment, representing a low Q-value object
An example of the second extreme in the case of the LHC collider is shown in Fig. 6: for the
detection of particles created during the collision process and to close the detector acceptance as much
as possible, additional detector components are installed (so-called ‘forward spectrometers’) that are
moved on to the beam during stable luminosity operation. Due to the nature of the problem, they
perfectly fulfil the conditions of a cavity-like object.
100
Fig. 6: Forward spectrometer ALFA [5]: the forward detector components are installed in a special vacuum
device and are a good representation of a high Q-factor ‘cavity-like’ object for the beam.
1.2 Longitudinal wake fields

Following the qualitative picture of the wake fields, longitudinal and transverse fields can be
generated if the bunch is passing a sudden change in the vacuum chamber geometry. While the
transverse fields will deflect the beam and might lead to instabilities, the longitudinal wake fields will
lead to energy loss of the particles and create heating of the vacuum components.
In order to obtain some kind of quantitative picture, we observe the situation of a test particle at
the end of the bunch at position z˜ , which will be influenced by the wake fields of the particles ahead
of it. Each charge at position z > z , ahead of our test particle, will create wake fields that influence
our test particle, as shown schematically in Fig. 7.
Fig. 7: A particle at the head of the bunch (position z) creates wake fields that act back on a test particle at the
bunch tail (position z˜ ).
The longitudinal wake function is defined by the integral of the longitudinal field E|| and
normalized to the charge q of the leading particle. Following the notation in Ref. [6], we write
1
W || (z − z˜ ) =
qL
∫ E|| (s,t − ∆z / βc)ds, (7)
where, for causality reasons, only the cases z − z > 0 are relevant and L refers to the complete
interaction length, which can be the length of the vacuum component or, in the extreme case, the
complete circumference of the machine. The integral in Eq. (7) represents the longitudinal potential
that is acting on the test particle and, accordingly, the units of the wake function are [W] = V/C.
Integrating over all particles ahead of our test particle and multiplying by its charge e will result in the
total energy loss of our test particle; in other words, we get the total induced voltage or the wake
potential:
101
B.J. H OLZER
∞
Vhom ( z˜ ) = − e ∫ λ(z ) W || (z − z˜ ) dz , (8)
z˜
where λ describes the line density of the leading particle charges and the negative sign indicates the
decelerating character of the fields. The total energy loss of the bunch is then obtained by integrating
over all slices dz˜ :
∞ ∞
∆U hom ( z˜ ) = − ∫ eλ ( z˜ )dz˜ ∫ λ(z) W (z − z˜) dz .
|| (9)
−∞ z˜
In short, if we know the bunch charge and the geometry of the vacuum chamber, we can
calculate (numerically in most cases) the longitudinal fields and thus the wake function, Eq. (7), and
obtain the resulting energy loss via Eq. (9).
Replacing the charge distribution in Eq. (8) by an instantaneous current at position z˜
I( z˜,t) = Iˆ0 e i(k˜z −ωt ) (10)
we obtain (referring, for simplicity, to one single mode k)
∞
1 z − z
Vhom ( z, t ) =
−
βc z ∫
I ( z, t +
βc
) W|| ( z − z ) dz . (11)
Applying a Fourier transformation of the current, we finally get a relation between the potential
and the beam current:
∞ iω∆z
1 −
Vhom ( t , ω ) =
−I (t,ω ) ∫ e βc
W|| ( z − z ) d ∆z (12)
β c −∞
In full equivalence to the situation of an AC circuit (Eq. (4)), we can now define the
longitudinal impedance in the frequency domain as the parameter that relates current and voltage in
our system,
∞
1
=Z|| (ω)
βc −∞ ∫
e −iω∆z /βc W|| (∆z ) d∆z , (13)
and as before it has the unit ohm (Ω). The longitudinal coupling impedance relates the beam current to
the induced voltage that is created by the wake field and that acts back on the beam:
Vhom ( t , ω ) = − I ( t , ω ) Z|| (ω ) . (14)
To quote from Ref. [6], where these relations are nicely described, “The impedance of the
environment of our beam is the Fourier transform of the wake fields left behind by the particle charges
in this environment.”
There is still work to do, as the key point and source of the problems are the longitudinal field
lines that have to be calculated, and in general this only can be done numerically.
As an example, we present in Fig. 8 a famous historical plot that shows these field lines for a
resistive wall wake field generated by a point charge q [7].
102
Fig. 8: The field lines for a resistive wall wake field generated by a point charge. (Courtesy of K. Bane.)
2 Observations on the beam

While in the end the total energy loss of the bunch is determined by the wake fields in the whole
storage ring and thus the impedance of the complete system is of major importance for beam stability,
large impedances of single accelerator components also have to be avoided. They might lead to local
heating effects and can put severe limits on the operational performance.
During the commissioning phase of the LHC and the run 1 operation with an increasing number
of bunches and single bunch intensities, several ‘hot spots’ in the true sense of the word were found.
An example of the effect of a sharp change in the aperture of the beam-surrounding vacuum
system is shown in Fig. 9. It shows the temperature increase due to wake fields measured at a LHC
beam collimator. By definition, these devices have to represent the locations with the tightest aperture
in the storage ring, as they are supposed to protect any other machine components from unwanted
beam losses. To reduce the problems of longitudinal wake fields, the collimator jaws are usually
tapered to provide a smooth transition between the standard vacuum chamber and the collimator.
Nevertheless, in the case of non-perfect tapering or other imperfections, wakes can be created at these
places and will lead to strong localized energy deposits. The plot shows, during an LHC run lasting for
seven days, a sharp temperature rise during beam injection and ramp, which falls slowly during
collision runs due to the decreasing beam intensity. It has to be emphasized that the peak values in the
plot refer to 50°C and were considered as the maximum tolerable limit for the system.
Fig. 9: The temperature increase measured at a LHC collimator (TCT) during injection, ramp, and stored beam
103
B.J. H OLZER
A quite similar effect has been observed at the ‘TDI’ collimator, which is positioned in the
injection region of the LHC [8] and acts as beam stopper in case of injection failures. Due to local
energy losses and the resulting heating effects, the RF shielding system that should provide a smooth
beam environment has been deformed to such a degree that replacement of the system with an
improved design became unavoidable (see Fig. 10).
Fig. 10: The shielding of the LHC beam stopper, damaged due to wake field effects
In a real accelerator, the contribution of a large number of single elements to the overall wake
field effects will add up. A parameter that is widely used to describe this integrated effect and that is a
property of the complete vacuum system is the so-called ‘energy loss factor’. It is defined as
∆U hom
khom = (15)
e 2 N b2
where ΔUhom describes the total energy loss of the bunch due to the wake field effect and Nb is the
number of particles in the bunch. The loss factor is clearly related to the wake function (Eqs. (7) and
(9)) and we can write
∞ ∞
1
khom ∫ ∫
= 2 λ( z )dz λ( z )W|| ( z − z )dz.
N b −∞ z
(16)
As a general parameter of the system, the loss factor is of considerable importance as it

determines – as we have seen above – the possible heating effects. The complete power loss of the
beam is then obtained by
I 02
Phom = khom (17)
f 0 nb
with nb indicating the number of stored bunches, f0 the revolution frequency, and I0 = nbeNbf0 the
overall stored beam current. Just like losses in a conventional ohmic resistive system, the losses due to
wake effects rise quadratically with the stored beam current.
Two remarks are needed in this context.
− The luminosity optimization of a storage ring collider depends on the beam current I0 and
bunch number nb (see Fig 11), and to optimize for the highest luminosities, the stored beam
current should be distributed over a small number of bunches, nb. This, however, leads in the
end to performance limits due to energy deposit via increased wake effects, and to overcome
104
this problem a compromise has to be found between optimum luminosity and tolerable
heating effects.
1 I02
L=
4 πe 2 ⋅ f 0 ⋅ n b σ x ⋅ σ y
(18)
Fig. 11: Colliding bunches and luminosity in a collider
− The loss factor depends strongly on the bunch length. Figure 12 shows a historical
measurement performed at the SPEAR collider.
Fig. 12: The loss factor as a function of the bunch length, measured at SPEAR [9]
This bunch length effect has also been observed in the LHC, and a dedicated lengthening of the
bunches has been applied, using RF noise, to limit the energy deposit in the critical systems. Figure 13
shows data taken within two subsequent LHC fills of equal beam parameters (such as the energy,
intensity, and number of stored bunches) but for different bunch lengths. The curves show the beam
current (green), the energy (yellow), the bunch length (red), and the collimator temperature (blue). In
the first fill, the bunch is kept sufficiently long and the resulting temperature increase is quite limited.
In the second fill, the bunch lengthening has been switched off and the bunch shortening that occurs
naturally during acceleration leads to a sharp increase in the temperature of the collimator jaws, and
finally to a beam dump triggered by the temperature interlock.
To avoid the limitations due to excessively short bunches in LHC operation, the natural
shortening obtained during the acceleration, which depends on the energy as well as on the cavity
voltage σ z ∝ E1/4 , σ z ∝ V 1/2 , is counteracted by dedicated RF noise. Figure 14 shows the development of
the bunch length during a LHC standard acceleration process. While at the start of the ramp a sharp
decrease of the length is observed, careful ‘shaking’ of the bunches leads to a very reproducible bunch
length of 1.2 ns at flat top.
105
B.J. H OLZER
Fig. 13: The bunch length effect on the loss factor observed in the LHC
Fig. 14: The bunch length at the LHC during the acceleration process: the shortening effect is compensated in a
dedicated way to achieve 1.2 ns at the end of the ramp.
3 Phase shift due to impedance effects

The energy loss caused by the wake field effect, which is described by the loss factor
∆U
khom = 2 hom , (19)
e N b2
will have to be compensated by the RF system of the storage ring and, due to the phase focusing
principle (see Ref. [10]), leads to a shift in the synchronous phase of the beam. As a reminder, we
repeat the qualitative argumentation here (Fig. 15).
106
Fig. 15: The principle of phase focusing and the definition of the synchronous particle
The phase of the synchronous particle, marked by the green spot in the figure, is defined by the
equilibrium between the average energy loss (due to synchrotron radiation, wake field effects etc.) and
the energy gain from the RF system.
In the absence of impedance, the power gain of the bunch is given by
∆P0= eVˆrf ⋅ f 0 N b ⋅ sin φsynch_0 , (20)
where Vˆrf is the RF peak voltage and ϕsynch_0 describes the ideal synchronous phase. The impedance
effects lead to an additional power requirement and, as a consequence, to a new phase ϕsynch_1.
Accordingly, we obtain
∆P= eVˆrf ⋅ f 0 N b ⋅ (sin φsynch_1 − sin φsynch_0 )
≈ eVˆrf ⋅ f 0 N b ⋅ ∆φ1,2 ⋅ cos φsynch_1 . (21)
Measurement of the shift in the synchronous phase can thus be used directly, to determine the
loss factor of the machine.
4 Impedances in an accelerator
For completeness, we will briefly summarize some analytical expressions for the impedances in
typical accelerator structures. A more detailed derivation can be found in Ref. [11].
4.1 Cavity-like structures

While the resonators used in an accelerator clearly present a dominant contribution to the impedance
budget, it is hard to avoid them and the only optimization we can carry out is to reduce higher-order
modes that will be the source of unwanted effects.
For the modelling of ‘cavity-like objects’, however, we can refer to the description that is
typically used to characterize the effect of resonators in an accelerator, namely an LRC circuit
(Fig. 16) with capacity C, inductivity L, and shunt impedance R, which are related to each other via
Rs L Rs
= L = or . (22)
Qω C Q
107
B.J. H OLZER
Fig. 16: An LCR circuit to describe the impedance of a resonator in a storage ring [2]
The impedance obtained is the sum of these three contributions:

1 1 1
= + − i ωC ,
Z|| Rs ωL (23)
which we can write as

Rs
Z|| = . (24)
ω ω 
1 + iQ  r − 
 ω ωr 
In order to avoid unwanted cavity-like structures to the maximum possible extent, a smooth
vacuum system is needed, and whenever this is not possible so-called ‘RF fingers’ are installed to
provide an RF-tight transition between vacuum components of different apertures. As an example,
Fig. 17 shows such a device as used in the LHC storage ring.
Fig. 17: The LHC interconnect device: elastic metal strips are used to provide an ‘RF-tight’ transition between
different vacuum components and to avoid longitudinal wake fields.
Careful design of the vacuum chamber in terms of the geometry and the choice of materials is a
must if high beam currents are desired (in other words, high machine performance). Standard
techniques for the field calculations and construction have been developed and the largest part of the
impedance budget can usually be understood and optimized. However, special devices exist in an
accelerator where even today optimization is not free from problems. As an example, Fig. 18 shows
the injection kicker for the LHC.
108
Fig. 18: The LHC injection kicker, with the ceramic beam tube in the centre
Due to the rapidly rising magnetic field that is needed to inject the new bunch trains from the
transfer line into the storage ring, the vacuum chamber of the kicker magnets cannot be built out of
conducting materials such as copper, steel, and so on. Eddy currents would be created in the surface of
the kicker chamber and would perfectly shield the inside from the external magnetic field. As a
consequence, the usual material chosen is ceramic, which is not really optimal, given the impedance
and local heating effects. Ultimately, the kicker chamber and the surrounding vessel will form a
beautiful ‘cavity-like’ object. A compromise therefore has to be found and the solution is to install a
number of metallic wires, the purpose of which is just to deliver sufficient RF tightness and still allow
the penetration of the kicker field to deflect the beam. As can be expected, an optimum has to be found
– in the theoretical calculations, but also ultimately on the beam.
Figure 19 shows a comparison between the calculated and measured impedances for two
different shielding configurations; one based on 15 wires, the other using 19. The reduction in
impedance using a better shielding concept is as remarkable as the agreement between theory and
measurement [12].
Fig. 19: Measured and calculated impedances for the LHC injection kicker
4.2 Resistive wall impedance

As explained above, the stored beam in an accelerator induces image currents in the vacuum chamber
walls, which travel with the beam. For non-ideal conductivity, however, these image currents create
resistive effects (i.e. losses in the chamber material) and lead to decelerating forces acting on the
109
B.J. H OLZER
particles. These forces are proportional to the beam current itself, and integrating around the storage
ring we get the expression [6]
Z|| (ωn ) R
= 1+ i (25)
n nrw σc δskin
with rw describing the vacuum chamber radius, δskin and σc the skin depth and conductivity of the
material, and n = ωn/ω0 the frequency in units of the revolution frequency.
For a given geometry of the accelerator and the vacuum system, the only parameter that is left
for optimization is the conductivity of the chamber material. In the LHC, therefore, a thin layer of
copper has been used to cover the beam screen in order to reduce the resistive wall effects
(see Fig. 20).
Fig. 20: The LHC beam screen: a thin layer of copper has been used to cover the surface, to reduce the resistive
wall wake field effects.
References
[1] B.J. Holzer, Introduction to transverse beam dynamics, these proceedings.
[2] D. Brandt, Multi particle effects, CAS−CERN Accelerator School, Granada, Spain, 28
October−9 November 2012.
[3] E. Jensen, in Proceedings of the CAS−CERN Accelerator School: RF for Accelerators, Ebeltoft,
Denmark, 8−17 June 2010, CERN-2011-007 (2011).
[4] J. Steward, AIP Conf. Proc. 421 (1997) 69–78.
[5] L. Fabbri, Forward detectors in ATLAS, Proc. 17th Int. Workshop in Deep Inelastic Scattering
and Related Subjects (2009).
[6] H. Wiedemann, Particle Accelerator Physics, Vol. II (Springer−Verlag, Berlin, 1993).
[7] Courtesy of K. Bane, 1991.
[8] E. Metral, Beam induced heating, bunch length, RF and lessons for 2012, Chamonix LHC
Performance Workshop (2012).
[9] P.B. Wilson et al., IEEE Trans. Nucl. Sci. 24 (1977) 1211.
[10] B.J. Holzer, Introduction to longitudinal beam dynamics, these proceedings.
[11] A. Chao, Collective effects in Accelerators (Proc. OPCA School 2010)
and Physics of Collective Beam Instabilities in High Energy Accelerators, Wiley-VCH (1993).
[12] H. Day, Measurements and simulations of impedance reduction techniques in particle
accelerators, CERN-thesis-2013-083.
110
Basics of Low-temperature Refrigeration
A. Alekseev 1
Linde AG, Munich, Germany
Abstract
This chapter gives an overview of the principles of low temperature
refrigeration and the thermodynamics behind it. Basic cryogenic processes -
Joule-Thomoson process, Brayton process as well as Claude process - are
described and compared. A typical helium laboratory refrigerator based on
Claude process is used as a typical example of a low-temperature
refrigeration system. A description of the hardware components for helium
liquefaction is an important part of this paper, because the design of the main
hardware components (compressors, turbines, heat exchangers, pumps,
adsorbers, etc.) provides the input for cost calculation, as well as enables to
estimate the reliability of the plant and the maintenance expenses. All these
numbers are necessary to calculate the economics of a low temperature
application.
Keywords: low-temperature refrigeration, Joule-Thompson process, Brayton

process, Claude process.
1 General principles of refrigeration

If your espresso is too hot, you just wait a minute, the ambient air cools the coffee and you can
enjoy the drink. If your Coca-Cola is too warm, you put some ice into the cup and your drink suddenly
becomes colder. All these actions have some similarity: here, a warm object contacts (directly or
indirectly) the colder object. This thermal contact is an essential requirement for the cooling.
What happens at the interface between warm and cold objects from a thermodynamic point of
view? The heat from the warm object flows to the cold object: from the hot espresso to the colder
ambient air, from the coke to the ice, and so on.
This is our experience: we can cool some object if we have some other material/objects
available that are already a bit colder. This could be cold water or snow. If really low temperatures are
necessary (–100°C or lower), then we use a special device – the so-called ‘refrigerator’ (or ‘cooler’).
1.1 What is a refrigerator?

For us, it is important to understand that every refrigerator has an area (or a surface) for which the
temperature is lower than that of the ambient. For the cooling of an object, we just need to establish
thermal contact between this object and the cold area of the refrigerator, and the heat ( Q 0 in Fig. 1)
flows from the object to the cold surface and the object becomes colder.
1
alexander.alexeev@arcor.de
A. A LEKSEEV
T0 Cooling
object
Q 0
T < T0
Refrigerator
Fig. 1: A refrigerator and a cooling object
Let us note three important comments:

Definition of a cold surface. Often, this is a real solid cold surface, such as a cold finger in a
cryocooler or an evaporator in a household refrigerator. But some refrigerators produce a cold
liquid cryogen (liquid helium, liquid neon, or liquid nitrogen), which – while flowing to/through
the cooling object (e.g. a superconducting magnet) – evaporates and the vapour is then returned
to the refrigerator. In this case, the cryogen liquid surface can be considered as the cold surface.
Refrigerator/liquefier. In low-temperature science, we often use another device called a ‘liquefier’.
A liquefier produces cold liquid that is then drawn off. The thermodynamics is the same for
both the refrigerator and the liquefier, but it is useful to start with refrigerator thermodynamics,
because it is a bit simpler.
Cooling capacity. Some cooling objects do not produce any heat of their own: they just need to be
cooled to a defined temperature – and that is all. But the most cooling objects (magnets, sensors,
current leads) generate heat continuously during operation, and therefore need permanent
cooling. The ‘heat’ in this case is a permanent heat flow from the cooling object to the
refrigerator. This heat flow corresponds to the ‘cooling power’ or ‘cooling capacity’ of the
given refrigerator.
1.2 A refrigerator consumes power

As said above, an important property of every refrigerator is the cold surface for adsorption of the heat
from the cooling object.
Another very important property is that it needs driving power – every refrigerator consumes
power (electrical or mechanical). Take a look at the conventional home fridge – you will see the power
cable for driving the small compressor behind the cabinet.
1.3 The first law of thermodynamics for a refrigerator

I expect that you are aware of the first law of thermodynamics – the conservation of energy principle.
Inelegantly stated, the first law says that what goes into a system must either come out or accumulate.
For steady-state conditions (without accumulation inside the system), the sum of the heat flows going
into the system is equal to the sum of the outlet energy streams. An energy stream is, for example, a
heat flow or mechanical power for driving a compressor, pump, and so on. All material streams
consist of some thermal energy (temperature) or mechanical energy (pressure), and therefore the
thermomechanical energies of material streams have to be taken into account by means of enthalpy
flow. We will try to apply this law to our refrigerator, working in steady-state conditions:
− two energy streams, the heat flow Q 0 from the cooling object and the electric power P, go
into the system;
− ‘steady-state’ means that there is no any accumulation inside this system.
112
BASICS OF L OW- TEMPERATURE R EFRIGERATION
Consequently, according to the first law, at least one additional ‘invisible’ energy flow must
exist, and this heat flow leaves our refrigerator – it goes out from the system – just to balance the
system. This energy flow does indeed exist. It is the waste heat shown in Fig. 2. The usual
abbreviation for this heat flow is Q amb . The first law for a refrigerator can be therefore expressed as
Q 0 + P = Q amb . (1)
Every refrigerator produces waste heat, and this heat is rejected to the ambient. At the rear of
the household fridge, you can find a black grid, made from up of tubes. This device is warm if the
compressor is running. By means of this surface, heat will be removed from the fridge and transferred
to the ambient air. Large refrigerators, such as the helium systems at LHC/CERN, use another method
– they are cooled by water. In this case, the waste heat is transfered to the cooling water. Small-scale
helium refrigerators for laboratories are usually air-cooled.
T0 Cooling
object
Q 0
Cold surface
T < T0
P
Refrigerator
Warm surface
T > Tamb
Q amb
Tamb
AMBIENT
Fig. 2: The first law of thermodynamics for a refrigerator
Summary
Every refrigerator has at least three interfaces:
− a ‘cold surface’ to receive and absorb the heat from the cooling object (to cool the cooling
object);
− a ‘warm surface’ to reject the waste heat to the ambient; and
− ‘power’ fed into the system.
It is important to repeat that the temperature of the ‘cold surface’ is lower than the temperature
of the cooling object, because heat can only flow from a warm object to a cold object.
For the same reason, the temperature of the ‘warm surface’ is higher than that of the ambient,
and the waste heat can flow from this hot surface to the ambient air, which is colder.
113
A. A LEKSEEV
1.4 The analogy between a refrigerator and a water pump

Hans Quack [1] once found some analogies between a refrigerator and a water pump, pumping the
water from a deep-water source (deepness H0 in Fig. 3) to the Earth’s surface.
− The pump consumes some power P (it is usually driven by an electric motor).
− It pumps the water from the deep source to the ambient.
− The suction (inlet) nozzle of the water pump is located a little lower in the source than the
liquid level – this is necessary to guarantee continuous flow to the pump.
− The pressure at the discharge nozzle is a little higher than ambient pressure – otherwise,
water could not flow out.
The refrigerator works in a very similar way.

− The heat Q 0 flows from the cooling object to the cold surface of the refrigerator.
− The refrigerator lifts (elevates) this heat from the cooling temperature to a temperature a
little higher than that of the ambient – this process requires some electrical or mechanical
power P.
− The heat pumped to the temperature higher than that of the ambient (which becomes Q amb )
is rejected to the ambient.
Some engineers use term ‘heat pump’ for a refrigerator, because a refrigerator pumps heat from a low
temperature to the ambient temperature.
T > Tamb H > Hamb

Q amb
Tamb Hamb
AMBIENT
Cooling
object P
T0 H0
Q 0
T < T0 H < H0
Water
Water
Fig. 3: A refrigerator (left) and a water pump
Refrigerator pump
(right)
1.5 The Carnot equation

The first law equation ( Q 0 + P = Q amb ) supplies us with some information about the properties of a
refrigerator. However, this information is not sufficient for complete analysis of the system. For
114
example, we cannot determine the power consumption of a refrigerator with a cooling capacity of
100 W at 100 K. The first law cannot answer this question. The first law only says that:
− if the power consumption were to be P = 100 W, then it would produce waste heat
Q amb = 200 W;
− or if we were to assume a driving power of P = 1 W, then the waste heat produced would be
Q amb = 101 W.
But here is the question: is it actually possible to build a refrigerator with a cooling capacity of
100 W at 100 K that consumes 1 W? The first law cannot answer this question. It is actually answered
by the second law of thermodynamics, in the form of the so-called Carnot equation, which is valid for
an ideal refrigerator – a refrigerator without thermodynamic losses (the Carnot equation in this form
was formulated by Rudolf Clausius, but Sadi Carnot developed the thermodynamic model – the so-
called ‘Carnot engine’ – used later by Clausius):
T −T
PMIN = Q 0 ⋅ amb 0 . (2)
T0
According to the Carnot equation:

1. the power required to drive a refrigerator depends linearly on the cooling capacity Q 0 of the
refrigerator: the larger the required cooling capacity, the higher the amount of power that is
necessary;
2. the power required to drive a refrigerator depends linearly on the ambient temperature Tamb:
the higher the ambient temperature, the higher the amount of power that is necessary;
3. the power required to drive a refrigerator depends on the cooling temperature: the lower the
cooling temperature, the greater is the power that is required.
These three statements do not need special explanation. But the following statement is a little
different:
4. the power required to drive a refrigerator depends on the cooling temperature: the lower the
cooling temperature, the more power is required – and this function is strongly non-linear; it
is 1/x.
According to the Carnot equation: to produce the 100 W cold capacity at 100 K, while assuming an
ambient temperature of 300 K = 27°C, the refrigerator would need at least 200 W:
300 K − 100 K
PMIN = 100 W ⋅ = 200 W .
100 K
This means that a 100 W at 100 K refrigerator with a driving power of less than 200 W cannot exist in
nature:
T −T
P ≥ Q 0 amb 0 . (3)
T0
1.6 A comparison with the water pump and/or vacuum pump

In previous section we discussed the functional analogy between a refrigerator and a water pump.
However, in energy terms, these devices are very different. This is because the power consumption of
the water pump depends on the pressure difference between the inlet and the outlet. The power
consumption of a refrigerator depends on the ratio of the ambient temperature to the cooling
temperature Tamb/T0. For cooling temperatures below 1 K, the ratio Tamb/T0 becomes higher than 300
115
A. A LEKSEEV
and the specific power consumption becomes gigantic. Therefore, as you sometimes hear, we are told
that absolute zero (0 K) is inaccessible/unreachable. Why? Because it requires a lot of energy.
We use sometimes cooling temperatures in the microkelvin region, but either for very small
cooling capacities (usually in the mW range) or for very short periods of time. Otherwise, gigantic
amounts of driving power are required.
1.7 Minimal power consumption for some typical temperatures

The Carnot equation helps us to estimate the minimal power consumption of refrigerators for typical
temperature ranges. Table 1 gives values for the most popular cryogenic temperatures corresponding
to liquid helium, liquid hydrogen, and liquid nitrogen, for a cooling capacity of 100 W.
Table 1: The minimal power consumption for some typical temperatures/cryogens
Liquid nitrogen Liquid hydrogen Liquid helium

(LIN) (LH2) (LHe)
Cooling temperature T0, K 77.4 20.1 4.2
Required cooling capacity Q 0 , W 100 100 100
Minimum power requirement P, W 288 1393 7043

P/ Q 0 2.9 13.9 70.4
The best refrigerator (an ideal refrigerator) producing cold at the liquid nitrogen temperature
level (77 K, or about –200°C) consumes at least 288 W. An ideal 100 W refrigerator working at the
liquid hydrogen level (20 K, or –250°C) needs approximately 1400 W, a value that is four times
higher. And a 100 W refrigerator for liquid helium temperatures (4 K, or –269°C) consumes at least
7 kW.
Exercise: a helium refrigerator for 1.8 K

A laboratory has an older helium system with a cooling capacity of 300 W at 4.5 K. The scientists
would like to upgrade this system and produce the same capacity, but at 1.8 K. The difference between
4.5 K and 1.8 K is only 2.7 K. This seems to be very small and the expectation is that such a kind of
revamp/upgrade would not be very costly, but would have a huge positive impact on the capacity of
the laboratory. Just to be on the safe side, we would like to check what this change (from 4.5 K
to 1.8 K) means for the power consumption of the refrigerator.
For example, we can determine the theoretical minimum power consumption of the existing
system using the Carnot equation and an ambient temperature of 293 K ~ 20 ºC:
PMIN = 300 W × (293 K – 4.5 K)/4.5 K = 19 233 W = 19.2 kW.
We can then calculate the theoretical minimum power consumption for the future 1.8 K system
and compare it with the value for the existing system:
PMIN = 300 W × (293 K – 1.8 K)/1.8 K = 45 833 W = 45.8 kW.
Surprisingly, the new value of ~46 kW is essentially higher – by a factor of 2.4. This is a
dramatic change. Essentially, it is higher than expected. This revamp from 4.5 to 1.8 K essentially
means higher energy consumption and probably a very high cost.
116
1.8 The coefficient of performance, COP

The value of the coefficient of performance (COP) can be used to characterize the efficiency of
refrigerator systems; it is defined as the ratio of the cold capacity to the driving power (or ‘what you
get’/‘what you pay for’):
COP = Q 0 P . (4)
If a helium refrigerator produces 300 W and consumes 75 kW, the COP amounts to the following:
COP = Q 0 /P = (0.3 kW) / (75 kW) = 0.004 = 0.4%.
It is really difficult to say that this system is a high-efficiency system, because a COP value of
0.4% seems to be very small.
Just to get a feeling for this, we can try to estimate the COP for an ideal refrigerator with the
same cooling capacity at the same temperature level using the Carnot equation. From the previous
exercise (4.5 K refrigerator versus 1.8 K), we know that a 300 W at 4.5 K refrigerator needs at least
19.2 kW. The COP for this system amounts to:
COPMAX = Q 0 / PMIN = 0.3 kW / 19.2 kW = 0.0156 = 1.6% .
This is amazing: the best helium refrigerator (allowed by nature in any circumstances) has a COP of
only 1.6%.
From this, we can learn that the COP for a liquid helium system is always very small and it is difficult
to compare helium systems using the COP number only.
1.9 Carnot efficiency

Another value used for the characterization of efficiency is the so-called Carnot efficiency – or Carnot
Fraction (CF) or Figure Of Merit (FOM). This is defined as the COP of a real refrigerator divided by
the COP of an ideal refrigerator (calculated by means of the Carnot equation):
CF = COP / COPCARNOT . (5)
This value makes more sense, because it gives us some feeling about potential improvements.
For our 300 W cooling capacity at 4.5 K:
η e = 0.4% / 1.6% = 0.25 = 25% ,
which is a very high efficiency for a 4.5 K temperature level (see Fig. 4 for comparison with existing
cooling systems).
Fig. 4: The Carnot efficiency for existing cryogenic systems. Reproduced from [2]
117
A. A LEKSEEV
2 Cooling effects
Up to now, we have treated the refrigerator itself as a kind of black box. Now, we will try to look
inside this black box. What happens inside the refrigerator? How does it work? First, we will look at
the available cooling effects. What is necessary to lower the temperature of a gas?
2.1 Joule–Thomson expansion

To begin with, let us consider the very simple experiment shown in Fig. 5:
− we take a conventional gas cylinder (50 l) filled with 200 bar nitrogen at room temperature;
− we open the valve carefully; and
− nitrogen flows into the ambient.
The pressure inside the bottle is high (200 bar), while the pressure outside is atmospheric
pressure (approximately 1 bar abs). Therefore the gas expands from the high-pressure zone to the low-
pressure zone by flowing through the valve. No heat flows, nor any mechanical/electrical energy
flows, are used during this experiment. However, if we measure the temperature of the gas at the
outlet, we will realize that it is lower than room temperature, the temperature difference being
approximately 30°C.
temperature of
nitrogen after
valve expansion to
ambient pressure:
T2 = ca. -10°C
ambient pressure:
ca. 1 bar
gas cylinder filled with nitrogen
temperature T1 = 20°C
pressure ca. 200 bar
Fig. 5: The experimental set-up for the Joule–Thomson cooling demonstration
This temperature change (of a gas when it is forced through a valve or other resistance from
higher pressure to lower pressure) ∆T = T1 − T2 is the so-called ‘Joule-Thomson effect’. This
procedure (the expansion of a gas in a throttle valve, orifice, capillary, porous plug, or other pressure
resistance) is the called ‘throttling’ or ‘Joule Thomson expansion’. The Joule-Thomson effect is
characterized by the so-called ‘Joule-Thomson coefficient’ µ, which is derived as follows:
 ∂T 
µ =   .
 ∂p  h = const
It is very important for an understanding of Joule-Thomson throttling that there are no changes
in the energy of the gas during the throttling procedure – no heat flows, nor any mechanical/electrical
energy, go into the system or leave the system during the throttling. Therefore the thermomechanical
118
energy of the gas – the enthalpy of the gas – during the throttling process is constant: the enthalpy at
the inlet into the valve and the enthalpy at the outlet are identical.
The Joule-Thomson coefficient is a material property, like the density, the specific heat capacity
or any other property. It depends on thermodynamic conditions such as pressure, temperature, and
phase state. And, of course, it differs from substance to substance: if we were to repeat our experiment
with helium (expanding from high pressure to ambient pressure), the temperature at the outlet would
be a little higher than in the bottle. Further, the temperature drop would essentially be smaller
compared to nitrogen.
The Joule-Thomson coefficient is a property of real gases, which means gases that cannot be
described adequately by the ideal gas equation (pV = RT). Theoretically, the enthalpy of an ideal gas
(which only exists theoretically) does not depend on the pressure. Therefore the virtual throttling of an
ideal gas does not cause any temperature drop. And in reality, for example, nitrogen (or ambient air) at
lower pressures (<10 bar) and ambient temperature behaves like an ideal gas, and the throttling of this
nitrogen to lower pressure leads to a negligible small temperature change. However, at high pressure
(100–200 bar) the properties of nitrogen differ from ideal gas behaviour, and this becomes visible in
the temperature change during throttling.
In Fig. 6, the nitrogen expanding process is shown on a temperature–entropy diagram. Point ‘1’
corresponds to the state of the nitrogen in the bottle at the beginning of the experiment: ambient
temperature 300 K and pressure 200 bar.
The line, which describes the expansion process in a valve, is the line of constant enthalpy. This
is because the enthalpy (heat content) of the gas remains constant: we do not see any heat flows into or
out of the expanding gas. At the outlet of the valve, the gas has the same heat content as ahead of the
valve, but the pressure corresponds to the ambient pressure. This point is marked by ‘2’. You can see
from Fig. 6 that the temperature difference indeed amounts to ~30 K.
We can pre-cool the bottle with nitrogen to 200 K (–70°C, corresponding to point ‘3’) and
repeat the experiment. At the corresponding outlet state, marked here with ‘4’, the temperature
difference is essentially higher and amounts to more than 70 K.
1
T1-2
3
T3-4
Fig. 6: Joule-Thomson expansion of N2 in a temperature–entropy (T–s) diagram. The T–s diagram is reproduced
from [6].
119
A. A LEKSEEV
The Joule–Thomson coefficient is relatively high at the conditions close to the two-phase area,
and it is relatively small at higher temperatures and lower pressures.
The temperature–entropy diagram for helium shown in Fig. 7, for temperatures below 10 K,
looks very similar to the diagram for nitrogen. We can see that it is not possible to produce liquid
droplets at the outlet by throttling of helium if the temperature before the throttle is higher than 7.5–
8.0 K. The liquefaction of helium can only happen below this temperature.
Fig. 7: Throttling of helium in a temperature-entropy diagram (dashed line)
2.2 Expansion in a turbine

The Joule–Thomson expansion is a very simple method: all you need is a valve or some other pressure
resistance such as an orifice or capillary. But the transformation of pressure with temperature change
here is inefficient, especially in the gaseous area, because of the small Joule–Thomson coefficient.
From the power generation area, we know another kind of gas expansion – in a turbine. Here,
the hot pressurized gas (or steam) expands in a gas turbine (or a steam turbine) to low pressure. The
turbine drives an electrical generator (see Fig. 8). During the expansion, the temperature of the gas is
reduced. From the power generation point of view, this is only a secondary effect, because the goal of
the expansion process (in a power generation plant) is to produce as much power as possible.
However, this decrease of temperature effect is important for refrigeration.
120
expander turbine
electrical generator
1 T1 , p1
H 1
virtual system boundary

P
T2 , p2
2 H
2
Fig. 8: Expansion in a turbine
The important thing is that the gas turbine produces some power: the more power produced, the
more efficient is the cooling process. This is because of the first law of thermodynamics: the gas
expanding in the turbine does work (produces mechanical power), and its energy content at the outlet
(enthalpy flow H 2 ) is lower than at the inlet to the turbine (enthalpy flow H 1 ) due to the power P
produced:
H 2 = H 1 − P ,
if P > 0, then H 2 < H 1
and therefore the temperature at the outlet T2 is lower than the temperature at the inlet to the turbine
T1 :
T2 < T1 .
An ideal expanding process is shown in Fig. 9, in a temperature–entropy (T–s) diagram. You
can see that the expansion of nitrogen from 200 bar and ambient temperature to 10 bar (not 1 bar, but
10 bar) means a temperature change of more than 140 K. If it is pre-cooled to 200 bar, then the
temperature change is about 80 K. This is indeed much more efficient than the Joule-Thomson
expansion.
The expansion of gas in a turbine works better in the gaseous area, because the heat capacity in
the gaseous area is higher.
1
1-2
DT
3
3-4
DT
Fig. 9: Expansion in an ideal gas turbine expander, shown in a T–s diagram. The T–s diagram is reproduced
from [6].
121
A. A LEKSEEV
3 Basic cycles
3.1 The Joule–Thomson refrigerator
3.1.1 Description of the process

The Joule-Thomson process (the simple Linde process) was developed by Carl Linde for the
liquefaction of air in 1895, and can be used for refrigeration at the nitrogen temperature level. It is a
really simple process and it always works.
The main hardware components required for realization of this process are a multistage
compressor with intercoolers and an aftercooler, a counter-current heat exchanger (Joule-Thomson
heat exchanger) and a throttling valve (Joule-Thomson valve), as shown in Fig. 10.
The most important part is the heat exchanger. This piece of equipment will be discussed later
in Section 4.2.3. The working principle can be explained with the help of Figs. 23 and 24, which show
the so-called ‘plate-fin heat exchanger’. This device consists of several channels divided by aluminium
sheets, inside of the channels you can see corrugated fins. The warm fluid flows through the small
channels, while the cold fluid flows through the other channel in the counter-current direction. This
cools the heat exchanger surface and therefore the warm gas on other side of the surface.
The other important device is a compressor. This machine compresses the gas from 1 bar to
~200 bar. Because it is difficult to compress a gas in one step, it is a multistage machine (usually more
than four stages).
P
Multistage
compressor

Qamb
P
Q amb
1 Tamb
5
Heat exchanger
Throttle valve
2 4
3 T < T0
Heat Q 0 Cold surface (evaporator)
Cooling object T0
Fig. 10: The Joule-Thomson process
3.1.2 The cooling-down procedure

What happens in our refrigerator when the compressor starts to work? It compresses the nitrogen to
200 bar, and in the aftercooler this nitrogen will be cooler to temperature close to the ambient
temperature by means of cooling air or cooling water. Then, the pressurized nitrogen flows into a heat
exchanger.
The first portion of the nitrogen flows through the heat exchanger without any temperature
change, because at the beginning the heat exchanger is warm, its temperature being the ambient
temperature. Therefore the nitrogen at the outlet of the heat exchanger has the same temperature as at
122
the inlet, while the pressure is a little lower than 200 bar because of some pressure losses in the heat
exchanger (point ‘1’ in Fig. 11).
Next, the nitrogen is throttled in the Joule-Thomson valve, the outlet pressure being close to
1 bar (point ‘3a’ in Fig. 11). This process is identical to our experiment with a nitrogen bottle: during
this expansion, the temperature dropped by ~30 K.
This first portion of the nitrogen returns to the compressor through the evaporator and heat
exchanger. This gas is colder than the gas on the high-pressure side of the heat exchanger, and
therefore it cools the warm gas. Consequently, the second portion of the gas arriving at the Joule-
Thomson valve has a temperature that is lower than the ambient temperature (point ‘2b’ in Fig. 11).
After this nitrogen has been throttled in the valve to low pressure, the temperature at the outlet
becomes a little lower than the temperature of the first portion of the nitrogen (point ‘3b’ in Fig. 11)
after throttling. This cold gas flows back through the heat exchanger and cools the warm high-pressure
gas further.
In this way, the heat exchanger is cooled further and further, and at the end of the cooling-down
process the temperature ahead of the throttle valve is ~160 K (–110°C). If throttling is carried out
starting from this temperature, then the nitrogen at the outlet of the valve is so cold that it is partially
liquefied. The surface beyond the throttle valve is cold and we can use it to cool some objects or for
other applications. The heat from the cooling objects evaporates the nitrogen, and this vapour returns
to the compressor through the heat exchanger.
1
5
2b
3a
2c
3b
2d
3c
3d
4
3
Fig. 11: The cooling-down process in a T–s diagram. The T–s diagram is reproduced from [6]
3.1.3 Liquefier versus refrigerator

The kind of cooling system discussed before is the so-called ‘refrigerator’. We can install a small
vessel (separator) after the throttle valve as shown in Fig. 12 on the right, and if some nitrogen
liquefies after throttling, the liquid will accumulate in this vessel. Usually, 5–7% of the nitrogen flow
is liquefied, and the rest (~95%), which is still gaseous, goes through the heat exchanger and cools the
high-pressure stream. If some liquid accumulates in this system, we have to feed the same amount of
123
A. A LEKSEEV
nitrogen into the system just to balance the cycle. The liquid nitrogen can be taken from the system
and used. This kind of cycle is known as a ‘liquefier’.
Q amb P Q amb P
1 1 GAN
5 5 Gaseous
nitrogen
2 4 2 4
Cold surface
(evaporator)
Separator
3 3
Q 0 Liquid
nitrogen
Cooling LIN
object
Refrigerator principle Liquefier principle
Fig. 12: The Joule-Thomson process as a refrigerator or liquefier
3.1.4 The advantages and disadvantages of the Joule-Thomson process

The big advantage of a Joule-Thomson refrigerator is its simplicity. It consists of a relatively low
number of hardware components, with no moving parts in the coldbox. If the heat exchanger is sized
more or less appropriately, it always works.
Another important feature of the Joule-Thomson system is that it produces liquid (liquid
nitrogen in case of nitrogen Joule Thomson process). The cryogenic liquids have several advantages as
coolants.
1. Nitrogen evaporates at a constant temperature. Therefore it is possible to keep the temperature
of cooling objects stable.
2. The density of the liquid is much higher than that of the gas; one can use small cooling
channels.
3. The volumetric heat capacity of the liquid is much higher.
4. The heat transfer coefficients for boiling liquid are high.
5. It is possible to collect the liquid for back-up purposes (safety and availability).
The main disadvantage is the relatively high pressure, which means that a high-pressure
multistage compressor is required. This should be an oil-free compressor, because some oil particles
and/or contaminants, if at low temperature, can freeze out and clog the heat exchanger channels or the
Joule-Thomson valve.
Such a high-pressure oil-free compressor is relatively expensive and requires some maintenance
every 2000–5000 h. Such compressors are only available for relatively small capacities, so that
refrigerators with a capacity above 1 kW need several compressor units and the system then becomes
really expensive.
The biggest disadvantage is that the cooling temperature of a Joule-Thomson refrigerator is
limited to ~54 K (using a mixture of oxygen and nitrogen). To achieve a lower temperature, cryogens
124
such as helium, neon, and hydrogen, and isotopes of these gases are necessary. At ambient
temperature, these gases behave like an ideal gas and therefore the Joule-Thomson coefficient is close
to zero, which means that it is impossible to build a Joule-Thomson refrigerator system based only on
these gases.
Therefore, the market share of pure Joule-Thomson systems is relatively small.
The relatively new development in Joule-Thomson refrigeration is the so-called ‘mixed gas
refrigerator’. This looks like the classic nitrogen Joule-Thomson refrigerator. However, it uses a
special mixture of gases based on nitrogen, methane, ethylene, propane, butane, and some helium/or
neon.
Because of the special properties of this mixture, it is possible to reduce the high pressure in the
cycle to below 20 bars. Another feature is that some components of this mixture – for example,
propane – can dissolve small amounts of compressor oil contaminants (such as solvents). Therefore it
is possible to use small oil-lubricated hermetic compressors from air conditioning systems for this kind
of refrigerator. These compressors are cheap and reliable (20 000 h or more); therefore, the whole
mixed-gas Joule-Thomson system is not as expensive and has a higher availability.
Units for relatively small cooling capacities of approximately 1 W at 80 K (Cryotiger) and for
the biggest cryogenic systems for liquefaction of natural gas (LNG) are commercially available.
3.2 The Brayton process

The Brayton cycle is the second basic cryogenic cycle. It looks like the Joule-Thomson cycle, but it
uses a turbine instead of the Joule-Thomson valve for gas expansion like shown in Fig. 13.
The Brayton cycle does not need very high pressures, because the expansion in the turbine is
much more efficient (in comparison to the Joule-Thomson system). It is possible to work with
pressures of 10–12 bar at the compressor outlet and achieve high efficiency. This is the big advantage.
Due to the fact that this process uses expansion of gas in a turbine and this kind of expansion is
combined with temperature reduction that is always independent of the cryogen used in the cycle, it is
possible to use real low-temperature refrigerants such as helium, neon, and hydrogen and achieve
temperatures below the liquid nitrogen level.
single-stage or
two-stage
Q amb
compressor
P
1 Tamb
5
Heat exchanger
Gas expander
(turbine)
2 4
Pturbine 3 T < T0
Q 0
Cooling object T0
Fig. 13: The Brayton process
125
A. A LEKSEEV
However, this system also suffers from some disadvantages. The first is that the turbine usually
cannot work in the two-phase region. If a partial liquefaction takes place and some of the liquid
particles in the turbine fly with a velocity close to the sonic velocity (200–300 m/s), they can damage
the turbine wheel. Therefore, the normal Brayton process design is such that any liquid at the turbine
outlet is avoided. The consequence is that the classic Brayton cycle cannot produce any liquid: the
cold surface of the refrigerator is cooled by cold gas only. Therefore it is difficult to keep the
temperature stable.
The second problem is that a turbine, even the smallest with a wheel diameter of 2–3 cm,
requires relative large gas flows, and therefore the lower limit for the cooling capacity of a classic
Brayton system amounts to ~500 W. It is really difficult to build a turbine with a smaller cooling
capacity.
11
5
2
4
Fig. 14: Nitrogen Brayton process in a T–s diagram. The T–s diagram is reproduced from [6]
3.3 The Claude process

You can see that the Joule-Thomson expansion is good for the production of liquids, but it is not really
efficient. Expansion in a turbine is very efficient in the gaseous area, but is problematic with regard to
the production of liquids. The next cycle – the so-called ‘Claude cycle’ – is a combination of both the
Joule-Thomson and the Brayton processes, and is shown in Fig. 15. The low-temperature part (in grey)
looks like the Joule-Thomson cycle and is the Joule-Thomson stage of the Claude cycle.
The upper part (1a  2a  3a  4a  5a) is similar to the Brayton cycle. Both cycles are
driven by the same single compressor. The Brayton cycle used here as a kind of pre-cooling cycle: it is
required to reduce the temperature at the inlet into the Joule-Thomson stage. The compressor pressure
is usually a little lower than in a classic Joule-Thomson cycle, but higher than in the classic Brayton
cycle.
The Claude process combines the advantages of the Joule-Thomson and Brayton cycles. It can
produce liquid because of the Joule-Thomson stage and it is very efficient because of super-efficient
pre-cooling based on expansion in a turbine.
126
single- stage or
two- stage compressor
Q amb Q amb
P P
1 Tamb 1 Tamb
5 = 5a 5 = 5a
Gas expander
(turbine)
Brayton stage
Heat exchangers
2a 4a 2a 4a
Pturbine Pturbine
3a 3a
Joule-Thomson stage
throttle valve
2 4 2 4
3 T < To 3 T < To
Q 0 Cold surface Q 0
To
Cooling object To Cooling object
Fig. 15: The Claude process
1a
5a
2a
4a
1
2
3a 5
4
3
Fig. 16: The Claude process in a T–s diagram. The T–s diagram is reproduced from [6]
127
A. A LEKSEEV
4 Helium refrigeration
Table 2 summarizes typical applications for helium refrigeration and provides typical values for the
corresponding cooling capacity and cooling temperature.
− Bulk helium liquefiers are used for the separation of helium gas from helium-containing
natural gas and for its further liquefaction (for transportation purposes) on-site in the field.
However, the current market size is quite small and it is limited to less than three systems
every 10 years.
− The market size for a compact laboratory helium liquefier is larger. This kind of system is
very cost-efficient because of the high level of standardization and the use of low-cost oil-
lubricated screw compressors. This system has become more of a commodity – every other
university in Europe has its own small helium liquefier.
− Helium refrigerators for high-energy physics (for the cooling of magnets, cavities, and other
objects) are most complex helium systems, producing a huge cooling capacity (up to 25 kW)
at different temperature levels down to the extremely low temperature of 1.8 K. Considering
their high power consumption, they are designed to achieve the highest efficiency in order to
reduce the customer’s operating cost.
Table 2: Typical applications for helium cryogenics (Claude process)
Application Cooling temperature Cryogen Cooling capacity

Bulk helium liquefier 4.3 K He 1000–4000 l/h
Laboratory helium 4.2–4.6 K He 20–200 l/h
liquefier
High-energy physics 1.8 K He 1–20 kW
4.4 K
80 K
4.1 The laboratory helium liquefier
4.1.1 Process design

Figure 17 shows a typical process flow diagram of a laboratory helium liquefier.
Compression
The helium gas is first compressed in a single-stage oil-lubricated compressor to 10–14 bar and then
cooled in an aftercooler. The compressor oil is separated by means of of a relatively complex (three- to
four-stage) Oil Removal System (ORS).
Liquid nitrogen pre-cooling
This compressed gas is cooled in the E3100A heat exchanger to ~80 K. This cooling can be supported
by external cooling sources such as liquid nitrogen, if available. As the liquid nitrogen flows through
the separate channels of the heat exchanger, it evaporates and cools the warm helium gas.
From the process design point of view, the liquid nitrogen pre-cooling can be replaced by an
additional Brayton stage (with expansion turbine[s]) at this temperature level. However, pre-cooling
with liquid nitrogen is widely used for the following reasons:
128
(1) low capital expenditure (facilitated only by additional channels in the heat exchanger, along
with liquid nitrogen supply hardware, instead of using highly sophisticated and therefore
expensive cryogenic turbine expanders); and
(2) better overall liquefaction efficiency, since liquid nitrogen is produced by an air separation
plant in a more efficient way compared to cold production in helium plants – this is because
large-scale turbocompressors and expansion turbines used in air separation plants have a
higher efficiency. 2
Pure He Gas Volume
ORS
8
Liquid nitrogen
precooling
Crude-He purification
GN2 LN2
Waste gas
contaminated E3440 E3100A

He
E3450
E3100B
F3110
E3460
T1
E3110A
E3470 E3110B
A3125 T2
A3480 E3480 E3120
E3130
DEWAR
Claude
cycle
Fig. 17: The Process Flow Diagram (PFD) of a laboratory helium liquefier
The Claude part

− The combination of the E3100B, E3110A/B, E3120, and E3130 heat exchangers and the
T1/T2 expansion turbines corresponds to the classical Claude process, described in the
previous section.
− The helium gas is cooled in the E3100B heat exchanger and divided into two streams – the
so-called ‘turbine stream’ and the Joule-Thomson stream.
− The turbine stream is cooled by expansion in two turbines to a low pressure (slightly above
ambient pressure). It flows through the heat exchangers back to the surge line of the
compressor and becomes warm due to heat transfer from the warm Joule-Thomson stream.
− The Joule-Thomson stream is cooled to below 8 K by thermal contact with the cold turbine
stream, before being expanded in a Joule-Thomson valve into a two-phase area and fed into
the liquid helium storage vessel – the so-called Dewar.
2
The isothermal efficiency of a large-scale turbocompressor is usually about 75%, while the isothermal efficiency of a
helium compressor is usually lower than 60%. The isentropic efficiency of an air expansion turbine is typically in the range
of 86–91%, while the isentropic efficiency of a helium turbine is usually lower than 82%.
129
A. A LEKSEEV
− The vapour fraction, separated from the liquid in the Dewar, flows through all of the heat
exchangers, back to the suction line of the compressors; in this way, it cools the warm high-
pressure stream and it becomes warm before re-entering the compressors.
Purification of crude helium
The helium gas feed is often contaminated with water and air gases such as oxygen and nitrogen (and
therefore is often called ‘crude gas’). This crude helium is usually purified prior to liquefaction to
avoid solidification, which can lead to blockage of heat exchangers and clogging of valves.
This task is fulfilled by a purification unit shown on the left-hand side of Fig. 17. It consists of a
dryer (not shown in the figure) and of a low-temperature purifier for the removal of nitrogen and
oxygen. The contaminated helium is cooled in the E3450 and E3460 heat exchangers to ~65–70 K; the
liquiefied portion of the contaminants is separated in a liquid separator (the separated liquid flows
through the heat exchangers into the ambient), while the gaseous contaminants are removed by means
of the A3480 cold adsorber – a vessel filled with materials that adsorb the nitrogen and oxygen.
The cooling duty required for cooling the feed gas is delivered by a small cold slip stream from
the main liquefier cycle; this slip stream cools the crude feed gas stream while flowing through the
E3470, E3460, and E3450 heat exchangers, before it re-enters the helium compressor.
The cold adsorber, as well as the dryer, needs regeneration after a certain working period to
guarantee the adsorption capability. Therefore, these devices are periodically heated for this purpose
(once a week, depending on the feed amount). The collected contaminants are released into the
atmosphere.
Additional hardware components
Two additional hardware components are shown in Fig. 17:
− filter F3110, for the protection of expansion turbines from solid particles;
− the additional cold adsorber A3125 to protect the lowest part of the cycle from
contamination (just for safety, or for long periods of operation).
4.1.2 Capacity
The liquefaction capacity is usually visible as a rising liquid level in the Dewar vessel. The condensed
liquid displaces the vapour helium from the upper part of the vessel into the main helium cycle.
Therefore the liquefaction cycle is actually fed from the two sources simultaneously: from the cold gas
derived from the Dewar vessel (typically 10–15% of the liquefaction capacity, depending on the
vessel) and from the real external feed.
Here are some typical specific consumption values (per litre of liquefied helium) taken from
Ref. [3]:
− 0.22 kW is the theoretical thermodynamic minimal liquefaction energy demand;
− 1.8 kW is the typical energy requirements for a small laboratory liquefier, without liquid
nitrogen pre-cooling;
− 0.9 kW is the typical power demand for a small laboratory liquefier equipped with liquid
nitrogen pre-cooling, which then requires close to 0.5 l of liquid nitrogen supply;
− optimized large-scale helium plants can achieve the specific value of 0.5 kW.
4.2 Key hardware components

The process design is a scientific art on the border between thermodynamics and engineering: it is
about how to design the most energy-efficient and cost-efficient system based on the available
130
hardware components: on one hand, the availability of hardware components limits the creativity of
the engineers; but on the other hand, the thermodynamic simulations performed for conceptually new
processes set the new targets for developing new hardware components or for improving existing
hardware.
Several key hardware components have been developed over recent decades to fulfil some of
the special requirements of helium refrigeration, such as very the low operational temperature and the
related special properties of cryogenic fluids. The main components are compressors, expanders, and
heat exchangers.
4.2.1 Compression
The compression of helium gas is a very interesting process, because of some of the special properties
of this gas:
− helium is a very light substance (molecular weight 4) and has very low density;
− it is a single-atom molecule, which leads to a relatively high isentropic exponent 3 of κ = 1.7.
4.2.1.1 The impact of the isentropic exponent

The following equations describe an ideal isentropic compression process. The outlet temperature Tout
depends on the inlet temperature Tin, the pressure ratio (pout/pin) and the isentropic exponent κ:
κ −1 κ −1
 
p  κ κ  pout  κ
Tout =T in ⋅  out  , Power = M ⋅ R ⋅ T in ⋅  − 1 .
 pin  κ − 1  pin  
 
The impact of the isentropic exponent is illustrated in Table 3. An ideal isentropic single-stage
compression from ambient pressure (1.013 bar) to 10 bar is calculated here for a gas flow of
M = 100 mol/s: the inlet temperature is given by Tin = 300 K. The calculation is done for two
different substances – air and helium; that is, for two fluids with differing values of the isentropic
exponent (κ = 1.4 for air).
Table 3: The ideal isentropic compression of helium in comparison to the compression of air
Air Helium
M 100 100 mol /s
R 8.31 8.31 J/mol/s
Tin 300 300 K
kappa 1.4 1.7
pin 1.013 1.013 bar
pout 10 10 bar
Tout 577 770 K
Tout 304 497 °C
Power 806 949 kW
This difference means that the outlet temperature of 497°C is extremely high in the case of
helium – approximately 200 degrees higher compared to the case for air (Tout = 304°C). The required
3
The isentropic exponent κ describes the relation between the pressure and temperature of a gas during an isentropic process.
A relatively high value of 1.7 means that a small pressure change (pressure ratio) leads to a large temperature change.
For an ideal gas, the isentropic exponent is equal to the ‘heat capacity ratio’, κ = γ = cp/cv; that is, the ratio of the heat
capacity at constant pressure, cp, to the heat capacity at constant volume, cv.
131
A. A LEKSEEV
power of ~950 kW for helium is approximately 20% higher than the compression power for the air
(~800 kW). This very simple calculation indicates very high compression temperatures and makes it
obvious that an efficient cooling procedure is required during compression of the helium.
Low density
The low density of helium is the main reason and explanation for the fact that highly efficient
turbocompressors are not applied for helium compression, although this kind of equipment is always
used in chemical engineering for the compression of gaseous fluids.
A turbocompressor consists of several compressor stages, and every compressor stage includes
three elements: a guide vane, an impeller (wheel), and a diffuser (volute chamber). The gas is
accelerated first in the guide vane, but mainly in the impeller (here, the mechanical energy of the gas is
transformed into kinetic energy, with an end velocity close to 300 m/s), and after the acceleration the
gas is decelerated in the diffuser (a flow channel with a widened cross-section). The kinetic energy of
the gas is transferred into potential energy (pressure). The simplified Bernoulli equation,
2 2
p1 + 12 ρ1 v1 = p2 + 12 ρ 2 v2 ,
describes this process, where p1, p2, ρ1, ρ2, v1, and v2 are pressures, densities, and velocities at the inlet
and outlet of the diffuser.
The pressure difference achieved in a single compression stage therefore depends on the
velocity of the impeller as well as on the gas density. This means that a relatively high pressure
difference (and therefore pressure ratio) can be achieved by compressing a heavy gas (such as argon,
krypton, or xenon). However, the pressure ratio for light gases such as helium is limited to 1.05–1.2.
This means that several compression stages are necessary, instead of a single stage as for heavy gases.
Taking into account a corresponding number of intermediate coolers (necessary after every stage
because of the high compression temperature) along with pressure losses and so on, a turbomachine
for helium would become very complex and expensive.
1000
Piston compressors
100
Pressure, bar
screw compressors turbocompressors
10
1
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Flow, m3/h
Fig. 18: Application ranges for different compressors
As a result, only piston and screw compressors are used for helium compression. Figure 18
shows the limitations for three types of compressors.
132
The screw compressor

“The twin-screw type compressor is a positive displacement type device that operates by pushing the
working fluid through a pair of meshing close-tolerance screws similar to a set of worm gears ... Each
rotor is radially symmetrical, but laterally asymmetric [see Fig. 19] ... The working area is the inter-
lobe volume between the male and female rotors. It’s larger at the intake end, and decreases along the
length of the rotors until the exhaust port. This change in volume is the compression. The intake
charge is drawn in at the end of the rotors in the large clearance between the male and female lobes. At
the intake end the male lobe is much smaller than its female counterpart, but the relative sizes reverse
proportions along the lengths of both rotors (the male becomes larger and the female smaller) until
(tangential to the discharge port) the clearance space between each pair of lobes is much smaller. This
reduction in volume causes compression of the charge before being presented to the intake manifold.”4
All helium screw compressors are oil-lubricated machines. Here, compressor oil is injected into
the compression cavities: it bridges the space between the rotors, both providing a hydraulic seal and
transferring mechanical energy between the driving and the driven rotor. However, the most important
function of the compressor oil is to provide an efficient cooling sink for the hot gas. The oil is
separated from the discharge stream, then cooled, filtered and recycled.
The screw compressor is a low-cost device, because this kind of compressor is a bulk
commodity: they are used to supply compressed air for general industrial applications or in
conventional refrigeration. Compressed air screw compressors are suitable for helium compression in
particular, because of the similar pressures (1 bar at the inlet) and pressure ratio (usually 8–10).
Fig. 19: The screws of a screw compressor
Reciprocation (piston) compressors

Another kind of compressor used in helium refrigeration is the reciprocating compressor. It is used for
small-scale liquefaction (< 30 l/h), as the main helium compressor or for the compression of helium to
high pressure (>30 bar), because screw compressors are not available for this pressure range.
4.2.2 The expansion turbine (or expander or turbine)

It is said that the heart of a helium liquefier is the expansion turbine. This means that an essential part
of the know-how (in helium refrigeration) is about how to build an efficient, reliable, and service-
friendly expansion turbine.
During the expansion from high pressure to low pressure, the gas produces mechanical power;
the enthalpy of the gas (and the temperature) at the outlet is therefore lower than at the inlet.
Theoretically, the mechanical power produced can be integrated into the refrigeration process.
However, in reality this process design option is very uncommon – engineers prefer to dissipate this
energy (transfer it by friction to the heat and reject it into the ambient) rather than using it. This is
4
From Wikipedia, ‘Rotary screw compressor’: see http://en.wikipedia.org/wiki/Rotary_screw_compressor
133
A. A LEKSEEV
because in this way the refrigeration process becomes simple and easy to control; additionally, it is a
slightly less expensive option. All these advantages justify the energy penalty of approximately 0.3–
0.5%.
Figure 20 shows three different turbines in terms of bearing supports. The common features are
as follows.
− The impeller is connected to the shaft, which is supported by two bearings at least.
− The operational temperature of the impeller is relatively low and the bearings work at
temperatures close to ambient or higher; therefore, the distance between these two parts has
to be as long as possible to avoid heat leakage due to conductivity through a solid.
− The two areas (the bearing and the impeller) are separated spatially to avoid cross-
contamination between the main process gas stream and the portion of helium inside the
bearing chamber.
− The impeller of a turbine with gas bearings is usually located below the bearing chamber; in
contrast, the bearing part of a turbine with oil bearings is located below the impeller. This is
necessary to prevent contamination of the impeller by the oil.
Oil bearing (left)
The oil bearing is the option developed in the early 1960s: the oil film guarantees the load-bearing
capacity for both axial and radial loads. The oil is pumped to a high pressure by a separate oil pump
and injected into the bearing gap. Later, it flows down to the bottom of the bearing chamber and
further to a small vessel placed below. This vessel is required for separation of the helium gas and oil.
It works simultaneously as a feed vessel for the oil pump. The helium gas in the bearing space appears
from the main process helium stream, through the gap between the impeller space and the bearing
chamber. This gas is then provided to the suction line of the main helium compressor. It is a kind a
parasitic bypass stream, which affects the efficiency of the whole cycle.
LP Helium
Helium
impeller To
compressor
HP Helium Gas Gas
Helium bearing bearing
leakage Helium
From
leakage
compressor
Gas bearing Gas bearing
Oil
oil bearing
impeller impeller
Oil
bearing HP Helium HP Helium
Helium To
compressor
Oil separator LP Helium LP Helium
Oil pump
OIL BEARING STATIC GAS BEARING DYNAMIC GAS BEARING
Fig. 20: Typical bearing systems
Gas bearing
Later, another solution was found by smart design engineers, based on the use of a gas film instead of
the oil film in the bearing.
First, the so-called ‘static gas bearing’ was introduced to the market: here, the compressed
helium gas from an external source (usually the main helium compressor) is injected into the tiny gap
134
between the shaft and the support. The problem of potential contamination of helium by oil was
elegantly solved in this way.
Subsequently, the so-called ‘dynamic bearing’ was invented. Here, the external high-pressure
gas source and corresponding bypass stream are completely eliminated, on the basis of an appropriate
design of the bearing, which allows the required pressure inside the bearing chamber to build up
internally. The efficiency of gas expansion turbines with dynamic bearings is therefore a little higher
in comparison to turbines with oil bearings and static gas bearings.
Figure 21 shows a modern turbine expander with dynamic gas bearings.
− The high-pressure feed gas enters the unit bottom-up (via two nozzles, on the left and the
right), it expands in the impeller and exits the unit axially downwards (via the nozzle in the
middle).
− The impeller of a small turbocompressor is mounted on the turbine shaft (above the yellow
part in Fig. 21); therefore the turbine impeller drives this turbocompressor, which is part of a
secondary closed cycle (the so-called ‘braking cycle’): the gas compressed here becomes
warm (as result of the compression), the heat is transferred to the cooling water in a compact
heat exchanger (the water cooler), and the cooled gas is then throttled in a valve and flows
again to the compressor inlet. In this way, the mechanical power produced by the turbine is
dissipated into the ambient (the cooling water).
Fig. 21: An expansion turbine with dynamic bearings (Linde)
Fig. 22: A typical turbine cartridge
135
A. A LEKSEEV
4.2.3 Heat exchangers

The heat exchangers in the cryogenic section are almost exclusively of the aluminium plate–fin type.
Figures 23 and 24 “show the structure of a plate fin heat exchanger module: the process streams
are led through passages. Up to 200 of these passages are stapled one on top of the other. The large
number of passages makes it possible to bring several streams into thermal contact within one unit.
The outlet frame is formed by 10–25 mm side bars, which are only interrupted for passages inlets and
outlets. A fluid enters the passage via nozzles and headers. Beginning from here the flow is distributed
with special fins over the entire cross-section of the passage and passed to the main section with heat
transfer fins. The arrangement of the passages as well as fin types can be selected by process design
engineer according to the process requirements” [4].
Fig. 23: Inside the plate–fin heat exchanger
The advantages of plate–fin heat exchangers are as follows:

− high flexibility concerning the number of process streams: several of them can be passed
through the single block, which allows more sophisticated process design;
− a high specific heat exchange surface, which helps to realize a very efficient process because
of the small temperature difference and pressure losses;
− low pressure losses;
− low specific costs.
Table 4: Typical parameters of aluminium plate–fin heat exchangers

Parameter Value(s)
Sizes Up to 1.8 m × 1.5 m × 8.0 m
Specific surface 500–1800 m²/m³
Fin
(plain perforated)
(serrated)
Fin height: 4–10 mm
Fin thickness: 0.1–0.6 mm
Temperature –269°C to +65°C
Pressure < 115 bar
Materials ASTM 3003 (DIN AlMnCu),
ASTM 5083, 6061 (DIN AlMg4.5Mn)
136
Fig. 24: An aluminium plate–fin heat exchanger: 1, core; 2, header; 3, nozzle; 4, width; 5, height; 6, length; 7,
passage outlet; 8, cover sheet; 9, parting sheet; 10, heat transfer fins; 11, distribution fins; 12, side bar; 13, front
bar.
137
A. A LEKSEEV
Figure 25 shows a typical laboratory helium liquefier for universities and academic insitutions.
It consists not only of the liquefier itself, but of all of the necessary infrastructural hardware.
Fig. 25: A laboratory helium system: 1, liquefier; 2, main compressor; 3, oil separator; 5, pressure control panel;
6, main Dewar for LHe; 7, small Dewars; 8, dryer/purifier; 9, instruments; 10, recovery compressor; 11, high-
pressure gas storage; 12, low-pressure gas storage.
Usually, the liquid helium is distributed by relatively small, compact transport Dewar vessels.
This method is used for cooling purposes in numerous experiment set-ups, and consequently liquid
helium evaporates during usage. The gas (contaminated by air and humidity) is usually collected in
low-pressure helium balloons (which look very similar to conventional air balloons – attempts are
made to use rubberized materials to prevent diffusion of the helium through the balloon wall). This gas
is compressed to 150–200 bar with the aid of a small high-pressure compressor (recovery compressor)
and transported back to the liquefier station. The amount of helium recovered can reach relatively high
values, up to 90%, depending on the experience and carefulness of the helium users.
The small transport Dewars are also passed back to the liquefier: the best of them will have a
small portion of liquid helium inside, to guarantee low temperature and minimize cooling downtime
and energy.
The returned helium gas is dryed, purified, liquefied, and transferred into the main storage
vessel, which is a part of the liquefier system. The small Dewars are charged from this vessel on
demand.
4.3 Trends
Karl Loehlein [5] has found the following trends concerning helium refrigeration.
− The requirements concerning the capacity of a single helium refrigeration unit are growing,
particularly with regard to giant projects in high-energy physics, and specially fusion
(ITER).
− The demand for refrigeration at a cooling temperature of 1.8 K is also growing.
− Efficiency is becoming increasingly important at higher cooling capacities.
138
− However, the requirements with regard to the efficiency of small liquefaction systems are
also becoming higher.
− Progress in the development of control and instrumentation is leading to a higher degree of
automation and to simplification of the operation of complex systems.
References
[1] H. Quack, Thermodynamische Grundlagen der Kälteerzeugung, Handbuch zum VDI-
Kryotechnik-Lehrgang (1996).
[2] T.R. Stobridge, Cryogenic refrigerators: an update survey, NBS Technical Note 655 (1974).
[3] K. Ohlig and L. Decker, Industrielle kryogene Anlagen, VDI Wissenforum Kryotechnik (2012).
[4] H.-W. Häring (ed.), Industrial Gases Processing (Wiley-VCH, Weinheim, 2008).
[5] K. Loehlein, Industrielle Kryoanlagen, VDI Wissenforum Kryotechnik (23–25 February 2005).
[6] H. Hausen and H. Linde, Tieftemperaturtechnik (Springer-Verlag, Berlin, 1985).
Bibliography about Cryogenics

F. Pobell, Matter and Methods at Low Temperatures, 2nd ed. (Springer-Verlag, Berlin, 1996).
T.M. Flynn, Cryogenic Engineering (Marcel Dekker, New York, 1997).
K. Timmerhaus and T. Flynn, Cryogenic Process Engineering (Plenum Press, New York, 1989).
139
Design Topics for Superconducting RF Cavities and Ancillaries
H. Padamsee 1
Cornell University, CLASSE, Ithaca, New York
Abstract
RF superconductivity has become a major subfield of accelerator science.
There has been an explosion in the number of accelerator applications and in
the number of laboratories engaged. The first lecture at this meeting of the
CAS presented a review of fundamental design principles to develop cavity
geometries to accelerate velocity-of-light particles (β = v/c ~ 1), moving on
to the corresponding design principles for medium-velocity (medium-β) and
low-velocity (low-β) structures. The lecture included mechanical design
topics. The second lecture dealt with input couplers, higher-order mode
extraction couplers with absorbers, and tuners of both the slow and fast
varieties.
Keywords: accelerators, niobium, gradients, superconducting, design, power

couplers, tuners.
1 General advantages of SRF cavities

There are two textbooks covering the major topics [1, 2]. Many review articles [3–13] are now
available covering the state of the art. There have been 16 international workshops on RF
superconductivity. The proceedings from these workshops carry a detailed and comprehensive
coverage of the substantial work going on, with many excellent tutorials on special subjects. These
proceedings are available in electronic form on the JACOW website (http://jacow.org/). Basics
covered in [1] are not repeated here, although essentials are summarized. Selective examples presented
here are for illustrative purposes only.
Superconducting Radio Frequency (SRF) cavities excel in applications requiring Continuous
Wave (CW) or long-pulse accelerating fields above a few million volts per meter (MV·m−1). We often
refer to the accelerating field as the ‘gradient’. Since the ohmic (power) loss in the walls of a cavity
increases as the square of the accelerating voltage, copper cavities become uneconomical when the
demand for high CW voltage grows with particle energy. A similar situation prevails in applications
that demand long RF pulse length, or high RF duty factor. Here superconductivity brings immense
benefits. The surface resistance of a superconducting cavity is many orders of magnitude less than that
of copper. Hence the intrinsic quality factor (Q0) of a superconducting cavity is usually in the
109 to 1010 range. Characterizing the wall losses, Q0 is a convenient parameter for the number of
oscillations it takes the stored energy in a cavity to dissipate to zero (Q0 is often abbreviated here as
Q). After accounting for the refrigerator power needed to provide the liquid helium operating
temperature, a net gain factor of several hundred remains in the overall operating power for
superconducting cavities over copper cavities. This gain provides many advantages.
Copper cavities are limited to gradients near 1 MV·m−1 in CW and long-pulse (duty
factor > 1%) operation because the capital cost of the RF power and the ac-power related operating
cost become prohibitive. For example, several MW·m−1 of RF power are required to operate a copper
cavity at 5 MV·m−1. There are also practical limits to dissipating such high power in the walls of a
1
hsp3@cornell.edu
H. PADAMSEE
copper cavity. The surface temperature becomes excessive, causing vacuum degradation, stresses, and
metal fatigue due to thermal expansion. On the other hand, copper cavities offer much higher
accelerating fields (∼100 MV·m−1) for short pulse (µs) and low duty factor (<0.1%) applications. For
such applications it is still necessary to provide abundant peak RF power (e.g. 100 MW·m−1), and to
prevent or withstand the aftermath of intense voltage breakdown in order to reach the very high fields.
There is another important advantage that SRF cavities bring to accelerators. The presence of
accelerating structures has a disruptive effect on the beam, limiting the quality of the beam in aspects
such as energy spread, beam halo, or even the maximum current. Because of their capability to
provide higher voltage, SRF systems can be shorter, and thereby impose less disruption. Due to their
high ohmic losses, the geometry of copper cavities must be optimized to provide a high electric field
on axis for a given wall dissipation. This requirement tends to push the beam aperture to small values,
which disrupts beam quality. By virtue of low wall losses, it is affordable to design an SRF cavity to
have a large beam hole, reduce beam disruption, and provide high-quality beams for physics research.
For low-velocity, heavy ion accelerators [7, 8], which must be capable of accelerating a variety
of ion species and charge states, a major advantage of superconducting resonators is that a CW high
voltage can be obtained in a short structure. The desired linac can be formed as an array of
independently phased resonators, making it possible to vary the velocity profile of the machine. An
independently phased array provides a high degree of operational flexibility and tolerates variations in
the performance of individual cavities. Superconducting boosters show excellent transverse and
longitudinal phase space properties, and excel in beam transmission and timing characteristics.
Because of their intrinsic modularity, there is also the flexibility to increase the output energy by
adding higher velocity sections at the output, or to extend the mass range by adding lower velocity
resonators at the input.
2 Figures of merit for cavity performance

The main figures of merit for an accelerating structure are defined and discussed in [1]. These are: RF
frequency, accelerating voltage (Vc), accelerating field (Eacc), peak surface electric field (Epk), peak
surface magnetic field (Hpk), surface resistance (Rs), geometry factor (G), dissipated power (Pc), stored
energy (U), Q value, geometric shunt impedance (Rsh/Q0, often mentioned as R/Q for short), cell-to-
cell coupling for multi-cell structures, Lorentz-Force (LF) detuning coefficient, input power required
for beam power (Pb), coupling strength of input coupler (Qext), higher-order mode frequencies, and
shunt impedances.
We present an in-depth discussion of several important figures of merit. The cavity accelerating
voltage Vc is the ratio of the maximum energy gain that a particle moving along the cavity axis can
achieve to the charge of that particle. The accelerating gradient is defined as the ratio of the
accelerating voltage per cell Vc to the cell length. As the optimal length of the cavity cells is typically
βλ/2, the accelerating gradient is
Vc
Eacc = .
βλ 2
The RF power dissipation in a cavity wall is characterized by the quality factor Q0, which tells
us how many RF cycles (multiplied by 2π) are required to dissipate the energy U stored in the cavity:
2
ω0U ω0 µ ∫ H (r ) dV
Q0
= = V
2
,
Pc ∫ Rs H (r ) dA
A
142
D ESIGN T OPICS FOR S UPERCONDUCTING RF C AVITIES AND A NCILLARIES
where Pc is the RF power dissipated in the cavity. The RF magnetic field H(r) for the excited
eigenmode with angular frequency ω0 = 2πf0 is integrated over the cavity volume V and surface A. The
surface resistivity Rs quantifies the RF power and depends only on the frequency and intrinsic material
properties. It remains the only term in the formula that is material dependent, making it convenient to
write the quality factor as
G
Q0 = .
〈 Rѕ 〉
where G is the geometry factor. The surface resistivity is a function of the RF magnetic field and may
therefore vary along the cavity wall. It must be averaged over the cavity surface. The geometry factor
G is determined only by the shape of the cavity, and hence is useful for comparing cavities with
different shapes. The cavity’s shunt impedance Rsh relates the dissipated power Pc and the accelerating
voltage:
Vc2
Pc = .
Rsh
A related quantity is the geometric shunt impedance Rsh/Q0, or simply R/Q, which depends only on the
cavity’s shape. Two key figures of merit are the ratios of the peak surface electric and magnetic fields
to the accelerating gradient, Epk/Eacc and Bpk/Eacc. A high surface electric field can cause field emission
of electrons, thus degrading performance. A high surface magnetic field may limit the cavity’s
ultimate gradient performance by breakdown of superconductivity, also called quench.
2.1 Design choices

Taking into consideration the above figures of merit, some of the main choices that need to be made
for structure design are: cavity frequency, cell shape, number of cells, beam aperture, operating
gradient, operating temperature, input coupler, and Higher-Order Mode (HOM) coupler types. Two
classes of considerations govern the choices: the particular accelerator application and the
superconducting RF properties. Typical accelerator aspects are: the velocity of the particle(s) under
acceleration, the desired voltage, the duty factor of accelerator operation, and the beam current or
beam power. Other properties of the beam, such as bunch length, also play a role, as these influence
the longitudinal and transverse wakefields, along with higher-order mode impedances. Typical
superconducting properties influencing design choices are the microwave surface resistance at the
chosen frequency, and the peak surface electric and magnetic fields at the design accelerating field.
These properties set the operating field levels, the RF power required, as well as the ac operating
power, together with the operating temperature. Mechanical properties also play a role to ensure
stability under atmospheric loading and temperature differentials, to minimize LF detuning, and to
keep microphonics detuning under control. Finally, input and output power coupling issues interact
with cavity design.
Electromagnetic software packages for modelling accelerating cavities and couplers have been
in existence for decades, first in 2D and later in 3D. Direct simulations of the entire cavity with input
and HOM couplers have been carried out.
In general there are many trade-offs between competing requirements. For example, the higher
the power capability of the input coupler, the larger the allowed number of cells per structure. But the
difficulties of handling long structures set an upper limit on the number of cells. A large number of
cells will also increase the probability of some HOMs remaining trapped inside the structure. A large
beam aperture will improve the propagation of HOMs out of the structure, but will increase the peak
surface electric and magnetic fields.
143
H. PADAMSEE
3 Classification of structures
There are three major classes of superconducting accelerating structures: high-, medium-, and low-β.
Figure 1 shows some practical geometries for each type depending on the velocity of the particles,
spanning the full velocity range of particles [7].
Fig. 1: Superconducting cavities spanning the full range of β (reproduced from [7])
The high-β structure, based on the TM010 resonant cavity, is for acceleration of electrons,
positrons, or high-energy protons with β ~ 1. The cavity gap length is usually βλ/2, where λ is the
wavelength corresponding to the frequency choice for the accelerating structure. Medium-velocity
structures with β between 0.2 and 0.7 are used for protons with energies less than 1 GeV as well as for
ions. At the higher-β end, these resonators are ‘foreshortened’ speed-of-light structures with
longitudinal dimensions scaled by β. Near β = 0.5 spoke resonators with single or multi-gaps become
popular. Spoke resonators operate in a TEM mode, and are so classified. The overlap between
foreshortened elliptical and spoke structures near β = 0.5 involves several trade-offs, which we will
discuss. Elliptical shape cells for β < 0.5 become mechanically unstable as the accelerating gap
shortens and cavity walls become nearly vertical. The choice of a low RF frequency, favoured for ion
and proton applications, also makes the elliptical cells very large, aggravating the structural weakness.
3.1 High-β cavities

A typical high-β accelerating structure consists of a chain of coupled cells operating in the TM010
mode, where the phase of the instantaneous electric field in adjacent cells is shifted by π to preserve
acceleration as a charged particle traverses each cell in half an RF period. Figure 2 shows a nine-cell
accelerating structure [14, 15] developed by the TESLA collaboration and used at FLASH (formerly
the Tesla Test Facility, TTF). The beam enters and exits the structure via the beam tubes. Input
coupler devices attached to ports on the beam tubes bring RF power into the cavity to establish the
field and deliver beam power. Higher-order mode (HOM) couplers extract and damp the HOMs
excited by the beam, and smaller ports carry pick-up probes to sample the cavity field for regulation
and monitoring. The TESLA cavity will be used in the European X-ray Free Electron Laser (XFEL),
and remains a strong candidate for the International Linear Collider (ILC).
144
Fig. 2: Top: photograph of a nine-cell TESLA accelerating structure with one input power coupling port at one
end and one HOM coupler at each end. Bottom: lay-out of the components for the nine-cell TESLA-style
structure.
Single-cell cavities generally used for SRF R&D also find accelerator application, as, for
example, in high current ring colliders, such as CESR, KEK-B, as well as many storage ring light
sources. Figure 3 shows the single-cell CESR and KEK-B cavities [16–18].
Fig. 3: Left: single cell 500 MHz cavity for CESR with waveguide input coupler and fluted beam tube on one
side to remove the first dipole HOMs. The cell length is about 28 cm, slightly shorter than λ/2, to optimize the
cell shunt impedance. Right: single-cell 508 MHz cavity for KEK-B with coaxial input coupler port and large
beam pipe on one side for propagation of HOMs [18].
Most β = 1 structures are now based on the elliptical cavity. The elliptical cell shape [19]
emerged from the more rounded ‘spherical’ shape [20], which was first developed to eliminate
multipacting. The tilt of the elliptical cell also increases the stiffness against mechanical deformations
and provides a better geometry for acid draining and water rinsing.
3.2 Multicell cavities

A multicell cavity is a structure with multiple resonators (cells) electromagnetically coupled together.
For each mode of a single-cell cavity there are N modes of excitation for an N-cell structure. For the
fundamental TM010 mode, there are N TM010-like modes. The accelerating mode is the one which
145
H. PADAMSEE
provides an equal voltage kick to charged particles passing each cell. In this mode the fields in
neighbouring cells are π rad out of phase with each other. Thus, a particle moving at near the speed
of light crosses each cell in half the RF period. The frequencies fm of the modes can be derived from a
circuit model (Fig. 4) of the N-cell cavity as a series of coupled LC resonators, where L and C are the
characteristic inductance and capacitance for each cell.
Fig. 4: Lumped circuit model for a multicell cavity
Here, L and C are related to the single-cell frequency ω0 and the shunt impedance Rsh/Q0 via
Rsh
ω0 = 1/ LC , = 2 L/C .
Q0
The cell-to-cell coupling is related to the inter-cell capacitance Ck via k = C/Ck., and Cb is the
capacitance representing the beam holes. The solution for the coupled Kirchhoff current and voltage
equations yields the dispersion relation for mode frequencies fm for mode m via
 f    mπ  
Ω( m )  m  = 1 + 2k 1 − cos 
=  .
 f0    N 
As shown in Fig. 5, the mode spacing increases with stronger cell-to-cell coupling k, and will
decrease as the number of cells, N. As the number of cells goes to infinity, all points on the dispersion
curve are filled in.
Fig. 5: Dispersion relation (frequency vs. mode number) for a five-cell cavity
146
The cell-to-cell coupling constant k can be obtained from the frequencies of the lowest and
highest frequency modes via
1  (N)
( ) − ( f ( ) ) 
2 2
1
f
k= 2  .
2 ( f ( ) ) − ( f ( ) ) 1 − cos ( π / N ) 
2 2
1 N
For a given amount of stored energy in the cavity, it is necessary to have equal fields in each
cell so that the net accelerating voltage is maximized and the peak surface EM fields are minimized.
This ‘flat’ field profile is achieved when the cells are properly tuned relative to each other. Cell-to-cell
tuning is often needed after initial fabrication when there may be slight deviations in the dimensions of
each cell, or after significant etching, or cell deformation due to electron beam welding or heat
treatment. The field flatness is measured by perturbing each cell in succession using a small metal (or
dielectric) bead while the frequency of the π mode is measured. In practice, the bead can be a small
(relative to the wavelength) segment of a tube on a fishing line suspended through the cavity along the
axis. The relative change in the frequency of each cell is proportional to the relative perturbation of the
stored energy and therefore proportional to E2. From the field profile, the tuning parameters can be
calculated via the circuit model and perturbation theory [1, 21]. Each cell is then tuned by squeezing
or stretching it mechanically.
3.3 History of the elliptical cavity shape

Before 1980, multipacting was the dominant limitation in the performance of β = 1 cavities.
Multipacting (MP) stands for ‘multiple impact electron amplification’. It is a resonant process in
which an electron avalanche builds up within a small region of the cavity surface due to a confluence
of several circumstances. Electrons in the high magnetic field region travel in quasi-circular orbit
segments, returning to the RF surface near to their point of emission, and at about the same phase of
the RF period as their emission. Secondary electrons generated upon impact travel along similar orbits.
Assuming electrons follow simple cyclotron orbits, a simple rule gives the associated magnetic field
for each order of a one-point MP as [22]
f / N = eB / 2πm,
where N is the order of MP, e and m are the charge and mass, respectively, of the electron, and B is the
local magnetic field at the surface. If the secondary emission yield for the electron impact energy is
greater than unity, the number of electrons increases exponentially to absorb large amounts of RF
power and to deposit it as heat to lower the cavity Q. This form of MP is named one-surface or one-
point MP. Depending on the cleanliness of the surface, the secondary emission coefficient of niobium
surfaces prepared by cavity treatment methods is larger than unity for electron impact energies
between 50 and 1000 eV.
With the invention of the round wall (spherical) cavity shape [20], one-surface MP is no longer
a significant problem for velocity-of-light structures. The essential idea is to curve gradually the outer
wall of the cavity. Electron trajectories drift toward the equator of the cavity in a few generations
(Fig. 6). Near the equator, the electric fields are sufficiently low that energy gain falls well below
50 eV, and regeneration stops because the secondary emission coefficient is less than unity. The same
suppression effect is achieved in the elliptical cavity shape, which is generally preferred to the
spherical shape due to added mechanical strength and better geometry for rinsing liquids [19].
147
H. PADAMSEE
Fig. 6: (a) Elimination of one-surface MP by the spherical (elliptical) cell shape. Electrons drift toward the zero
electric field region at the equator, where the electric field is so low that the secondary particles cannot gain
enough energy to regenerate. (b) Two-point MP in a single-cell 1.3 GHz TESLA-shape cavity near
Eacc = 21 MV·m−1. Note the resonant trajectories in the lower half (expanded).
After one-surface MP was cured, two-point MP was discovered in elliptical cavities when
electrons travel to the opposite surface in half an RF period (or in odd-integer multiples of half an RF
period). In the spherical/elliptical cavity geometry, two-point MP survives near the equator of the
cavity. But the electron energies are low (30–50 eV) near the unity cross-over point of secondary
yield, so that the MP is weak and easily processed. The simple rule for two-point MP is
2 f / ( 2 N − 1)= eB / 2πm.
For the elliptical cavities, the peak magnetic field levels of the first-order two-point MP at
various frequencies follow the scaling law [22]:
B [ mT ] = 5 + 55 f [ GHz ] .
This corresponds to multipacting at 76 mT or Eacc = 18 MV·m−1 for the TESLA-shape cavity.

For general analytic approximations to the fields in the equator region, and the resulting rules for two-
point MP in this region, see Ref. [23].
Conditioning times for two-point MP are generally short. During conditioning, MP often grows
sufficiently intense to induce a local thermal breakdown of superconductivity. The location of intense
MP migrates along the equator as the secondary emission coefficient drops in one place due to
electron bombardment. Both MP and its associated breakdown events disappear after sufficient
conditioning, but trapped DC magnetic flux generated by thermoelectric currents during the
breakdown events reduces the Q values, sometimes by as much as a factor of two. Warming up
to >10 K is necessary to remove the trapped magnetic flux. The MP does not reappear since the
surface now has a lower secondary yield.
Multipacting levels become suppressed by electron bombardment, which decreases the
secondary yield over time, most likely by gas desorption, or possible molecular changes in the
adsorbed monolayers. MP can be enhanced again when the secondary emission yield increases due to
adsorbates or condensed gas. Levels which have been successfully processed can recur for short
periods if the cavity is temperature cycled and gases re-condense on the surface. The occurrence of
MP is often accompanied by X-rays when some electrons escape from the MP region into the high
electric field region, where they are accelerated.
148
4 Low-velocity structures
4.1 Medium-β
With the growing interest in accelerators for spallation sources, as for example the Spallation Neutron
Source [24] SNS at Oak Ridge National Laboratory, elliptical resonators have been extended to high
energy (∼1 GeV) proton acceleration using medium-β superconducting cavities (0.6 < β < 0.9).
Medium-β cavities are also important for high current proton linacs for injectors at Fermilab and
CERN, and in the future for energy production via accelerator driven reactors, material irradiation, and
nuclear waste transmutation.
The design of a medium-β structure involves several tradeoffs. The choice of a low frequency
increases the voltage gain per cell, the beam energy acceptance, and the beam quality, at the same time
decreasing RF losses and beam losses. But a low RF frequency increases structure size and
microphonics level, making RF control more challenging. The larger the number of cells, the higher
the voltage gain per structure, but the narrower the velocity acceptance, and the larger the number of
cavity designs needed to optimize the voltage gain with changing particle velocity. In the medium
velocity range, structures must efficiently accelerate particles whose velocities change along the
accelerator. Several structure geometries are therefore needed, each of which is optimized for a
particular velocity range. The lower the velocity of the charged particle under acceleration, the faster it
will change, and the narrower the velocity range of a particular accelerating structure. This implies
that the smaller the value of β of a cavity, the smaller the number of cavities of that β which can be
used in the accelerator.
Efficient acceleration for 0.5 < β < 0.9 is achieved in a straightforward manner by axially
compressing the dimensions of the standard elliptical resonator geometry while maintaining a constant
frequency, as shown in Fig. 7. SNS for example uses two elliptical cavity geometries, one at β = 0.6
between 200 MeV and 600 MeV, and the other at β = 0.8 from 600 MeV to 1 GeV [24, 25]. The lower
limit of usefulness for the compression approach is about β = 0.5, when the vertical flat walls make the
structure mechanically unstable.
Fig. 7: A progression of compressed elliptical cavity shapes at the same RF frequency but for decreasing
β values [25].
As the cells compress to low β geometries, the cavity properties exhibit interesting general
trends [26]: Epk/Eacc increases from its typical value of 2 to 2.5 for β =1 up to 3 or 4 for β = 0.5. For
1 MV·m−1 accelerating field, the peak surface magnetic field near the equator increases from 4 to 5 mT
for β = 1 to 6–8 mT for β = 0.5. The geometric shunt impedance per cell decreases roughly
quadratically as R/Q (Ω) = 120β 2. For constant Eacc, the stored energy (U) per cell is roughly
independent of β. Structure stored energy plays an important role in amplitude and phase control in the
149
H. PADAMSEE
presence of microphonics detuning because the RF power required for phase stabilization depends on
the product of the energy content and the amount of detuning. A typical value is U = 200–250 mJ per
cell at 1 MV·m−1.
4.2 Medium-β spoke resonators

An alternative path to medium-velocity structures with β near 0.5 is via multi-gap spoke resonators
(Fig. 8). Here each spoke element is a half-wavelength resonant transmission line operating in a TEM
mode. Resonant transmission lines developed originally for low-β quarter-wave and half-wave
resonator applications will be discussed later.
(a) (b) (c) (d)
Fig. 8: (a) Spoke and gap profile [27, 28]. (b) 3D sketch [27, 28] and (c) photograph of the first spoke resonator
with β = 0.28, 800 MHz [29]. (d) Multi-spoke resonator [30].
The spoke elements are made elliptical in cross-section to minimize the peak surface fields. The
major axis of the ellipse is normal to the beam axis in the centre of each spoke to minimize the surface
electric field and maximize the beam aperture. A typical beam aperture is 4 cm at 345 MHz. In the
region of the spokes near the outer cylindrical diameter, the major axis is parallel to the beam axis in
order to minimize the peak surface magnetic field. Designs can be optimized by controlling A/B
(in Fig. 8(a)) to reduce Epk/Eacc and C/D to reduce Bpk/Eacc.
In the spoke structure, the cell-to-cell coupling does not rely on the electric field at the beam
holes, as for elliptical cavities, but takes place chiefly via the magnetic field linking cells through the
large openings. As a result, the coupling is very strong (20–30% as compared to 2% for β = 1 elliptical
cavities), which makes the spoke structures robust and the field profiles insensitive to mechanical
tolerances. Half end cells (half-gaps) terminate the structure to derive a flat π mode.
The range of spoke resonator applications continues to be extended into the medium-β regime.
In principle there is no clear-cut transition energy from spoke resonators to elliptical ones. Typically
TM cavities have an inside diameter of about 0.9λ. Spoke structures have outer diameters below 0.5λ.
Thus a spoke cavity can be much smaller than an elliptical cavity at the same frequency, or the spoke
structure can be made at half the frequency for roughly the same dimensions as the elliptical structure.
Choosing a lower frequency allows the option of 4.2 K operation, thus saving capital and operating
costs associated with refrigeration.
4.3 Low-β quarter wave resonators

Low-β resonators have been in use for heavy-ion boosters for more than three decades. The short
independently phased cavities provide flexibility in operation and beam delivery. Applications
continue to expand towards both the lower-β as well as the medium-β range, as with spoke resonators
discussed in Section 4.2. Low-velocity structures must accelerate a variety of ions with different
150
velocity profiles. Different cavity geometries with many gaps have been developed that are suitable
for different beam energies, beam currents, and mass/charge ratios.
The Quarter-Wave Resonator (QWR) derives from transmission-line-like elements and belongs
to the TEM resonator class. Figure 9 shows a coaxial line, λ/4 in length shorted at one end to form a
resonator with maximum electric field at λ/4, where the accelerating gaps are located [31]. Low
frequencies, typically 100–200 MHz, must be used as the active and useful length of the structure is
proportional to βλ. The low frequency results in a large resonator. The typical structure height is about
1 m. The inner conductor, which is made from niobium, is hollow and filled with liquid helium.
Operation at 4.2 K is usually possible due to the low RF frequency.
(a) (b) (c)

Fig. 9: (a) Equivalent circuit current and voltage distributions of a QWR [31]. (b) Two QWRs for SPIRAL-II
with different β values (0.07 and 0.12) [32].
The larger the number of gaps in a QWR, the larger the energy gain, but the narrower the
velocity acceptance. Figure 10 shows the transit time factor for one-gap and two-gap resonators in the
simple approximation of constant field in the gap and zero field outside.
Fig. 10: Normalized transit time factor vs. normalized velocity β/β0, for cavities with different numbers of equal
gaps [31].
Being a non-symmetric structure, the QWR has non-symmetrical electromagnetic fields in the
beam region; this produces undesirable beam steering through electric and magnetic dipole field
components. Compensation can be obtained by gap shaping: the magnetic deflection can be cancelled
by enhancement of the electric deflection [33].
Quarter-wave structures are sensitive to mechanical vibrations because of their large size and
large load capacitance. The related phase stability is an important issue, particularly for the lowest
velocities and for small beam loading, high Qext operation. The mechanical stability problems have
been solved by electronic fast tuners [34] or by the addition of a mechanical damper in the cavity stem
[35, 36]. QWRs have covered a wide overall application range: 48 ≤ f ≤ 160 MHz, 0.001 ≤ β ≤ 0.2,
151
H. PADAMSEE
with two gaps and four gaps. The extensions to the very-low-β regime use a tuning fork
arrangement [37]. Compact and modular, QWRs have proven to be efficient high-performance
resonators. They can achieve reliably 6 MV·m−1.
4.4 Half-wave resonators

A half-wavelength (λ/2) transmission line, with a short at both ends, has maximum voltage in the
middle and behaves as a half-wave resonator (HWR). Figure 11 shows a simple equivalent circuit
along with voltage and current distributions in one of the λ/2 loading elements of a single spoke
resonator [31].
A HWR is equivalent to two quarter waves facing each other providing the same accelerating
voltage as a QWR but with almost twice the power dissipation. Figure 11 shows an example [38]. The
symmetry of the structure cancels steering and opens the use of HWRs at β from 0.1 to 0.5, above the
range customary for QWRs. HWRs also show improved mechanical vibration properties over QWRs.
(a) (b)
Fig. 11: (a) Equivalent circuit current and voltage distributions of a single-spoke HWR [31]. (b) Example of a
HWR [38].
The peak surface electric field occurs at the centre of the loading element. By suitable sizing
and shaping of the cross-section, a surface to accelerating field ratio of 3.3 can be obtained,
independent of β. The maximum Hpk occurs where the loading element meets the outer enclosure, and
is sensitive to the size and shape of the centre conductor. Values of 7 mT·MV−1·m−1 can be obtained
by proper shaping. Most structures are designed with somewhat higher surface field values.
5 Mechanical aspects for structure design

The design of a superconducting cavity must take into account several mechanical aspects: stresses,
vibrations, and Lorentz forces. Codes exist to simulate mechanical properties [39, 40]. The cavity must
withstand stresses induced by the differential pressure between the beam pipe vacuum and
atmospheric pressure. Differential thermal contraction due to cool-down from room temperature to
cryogenic temperatures induces stress on the cavity walls. Mechanical vibrations of the cavity and the
cavity–cryomodule system (microphonics) form another aspect of cavity mechanical design. External
vibrations couple to the cavity and excite mechanical resonances, which modulate the RF resonant
frequency inducing ponderomotive instabilities [41]. These translate to amplitude and phase
modulations of the field, becoming especially significant for a narrow RF bandwidth. Lorentz Force
(LF) detuning becomes important in cavity designs for high-field pulsed operations [42]. Surface
currents interact with the magnetic field to exert a Lorentz force on the cavity wall. This stress causes
a small deformation to change the cavity volume and frequency.
152
5.1 Mechanical stresses

To avoid plastic deformation, the cumulative mechanical stress on the cavity walls must not exceed
the cavity material yield strength, including some engineering margin. The frequency shifts due to
these stresses must be taken into account when targeting the final frequency or tuner settings and tuner
range. Stresses due to the operation of the tuner mechanism should not exceed yield strength while
cold. The mechanical requirements may be dealt with by the proper choice of cavity wall thickness or
by adding stiffening rings or ribs at locations of high strain.
Medium-β elliptical cavities are especially vulnerable to mechanical stresses due to the
flattening of the wall, as mentioned.
5.2 Vibrations
Stiffeners added at appropriate locations raise the cavity mechanical resonant frequencies so that these
no longer couple to the lower-frequency external vibration sources. Dampers introduced in the
mechanical system of cavity and cryomodule reduce the mechanical Q of the resonances. The RF
bandwidth can also be widened by increasing the strength of the input coupler, but this demands
higher RF power and lowers the operating efficiency. Mechanical tuners are usually too slow to
counteract cavity wall deformations from microphonics. The stored energy per cell plays an important
role in amplitude and phase control in the presence of microphonics detuning. When beam loading is
negligible, the amount of RF power required for phase stabilization is given by the product of the
energy content and the amount of detuning. Fast tuners of the piezoelectric or magnetostrictive type
added to the tuning system provide active damping of microphonics together with sophisticated
electronic feedback systems.
5.3 Lorentz force detuning

The cavity wall tends to bend inwards at the iris and outwards at the equator, as shown in
Fig. 12(a) [43].
(a) (b)
Fig. 12: Cavity shape distortions due to LF detuning [43, 44]. (a) Lorentz forces acting on different parts of the
cavity wall. Note the rotated orientation of the cavity. (b) Distortion of the frequency response of the cavity
response at two field levels [45].
The resonant frequency shifts with the square of the field amplitude, distorting the frequency
response [45], as shown in Fig. 12(b). Typical detuning coefficients are a few Hz·MV−2·m−2. This
frequency shift can be compensated for by mechanical tuning once the operation field is reached.
153
H. PADAMSEE
A fast tuner is necessary to keep the cavity on resonance, especially for pulsed operation. However, a
large LF coefficient can generate ‘ponderomotive’ oscillations, where small field amplitude errors
initially induced by any source (e.g. beam loading) cause cavity detuning through the Lorentz force
and start a self-sustained mechanical vibration, which makes cavity operation difficult [46]. LF
detuning is especially important in pulsed operation, where the dynamics of the detuning plays a
strong role.
Stiffeners must be added to reduce the coefficient, as shown in Fig. 1 [42], but these increase
the tuning force. For the TESLA-shape nine-cell elliptical structure (Fig. 1) the LF detuning
coefficient is about 2–3 Hz·MV−2·m−2, resulting in a frequency shift of several kilohertz at
35 MV·m−1, much larger than the cavity bandwidth (300 Hz) chosen for matched beam loading
conditions for a linear collider (or XFEL). Stiffening rings in the nine-cell structure reduce the
detuning to about 1 Hz·MV−2·m−2 [47]. Feedforward techniques can further improve field stability
[48-50]. In CW operation at a constant field, the Lorentz force causes a static detuning which is easily
compensated for by the tuner feedback, but may nevertheless cause problems during start-up, which
must also be dealt with by feedforward in the RF control system.
6 Input couplers
6.1 Requirements and design principles

An input coupler is a device that efficiently transfers RF power from the generator (source) to a beam
loaded cavity by providing a good impedance match between the two, as depicted in the circuit of
Fig. 13(a). The coupler must operate over a wide range of load impedance, which varies from a
matched impedance at full beam loading to full reflection when there is no beam.
(a) (b)
Fig. 13: (a) Equivalent circuit for an input coupler. (b) Coaxial input coupler at the beam pipe of a
superconducting cavity.
For a superconducting cavity, the input coupler is normally inserted at the beam pipe just
outside the end cell of the accelerating structure rather than inside the cell in order to avoid field
enhancements that may lower the quench field, or field perturbations that may initiate MP in the cell.
The beam pipe diameter and spacing between the end cell and the coupler port need to be sufficient
that the antenna does not have to penetrate too far into the beam-line, where it may become a source of
strong wakefields or coupler kicks. The transverse electromagnetic field of the power coupler on the
beam tube can create a small kick, which increases beam emittance [51]. This effect is especially
strong for a cavity at the low-energy end of an accelerator (the injector), where a high average RF
power must be coupled to a vulnerable low-energy beam. In this case, a twin coaxial coupler choice
reduces harmful transverse kick fields, ideally to zero on-axis. The shape and location of the antenna
tip can also be optimized to minimize penetration into the beam pipe.
As an auxiliary device, the coupler design must satisfy numerous requirements and functions. It
must preserve the cleanliness of the superconducting cavity, provide a vacuum barrier between the
cavity and the feeder waveguide, allow some mechanical flexibility for alignment and thermal
contraction during cool-down, permit variable coupling strength (external Q) in desired cases, and
include thermal transitions from room temperature to cryogenic temperature with minimal static and
154
dynamic thermal losses. In addition, the coupler must be equipped with diagnostic elements to allow
safe operation. These requirements call for a careful design from the electromagnetic, mechanical, and
thermal points of view.
Couplers for superconducting cavities have been developed to span RF frequencies from 300 to
2000 MHz, and duty factors from 1 to 100%. There has been remarkable progress in power capability
for both pulsed and CW operation: 300–500 kW of RF power in operating accelerators and up to
2 MW for prototype testing. There have been many review articles on couplers with valuable
references [52−60]. No single coupler design suits all applications. A variety of coupler types have
been explored and developed: coaxial and waveguide; one and two windows; cold and warm
windows. We will discuss the pros and cons of some of these choices.
There are many reasons for progress in power couplers. Extensive simulations take place in the
design phase to predict electromagnetic, thermal, mechanical, and multipacting properties of coupler
geometries. Commercial RF modelling codes (e.g., Microwave Studio [61], HFSS [62]) are available
for 3D simulations with high accuracy to optimize RF transmission, voltage, current, and power
densities. The goal of the RF design is to obtain good transmission properties (minimize reflections
and insertion losses) over a workable bandwidth as well as over possible variations in temperature and
assembly tolerances. The codes model electromagnetic field distributions over the various elements in
the coupler transmission line to establish the best locations for cooling intercepts and window
placement, and to determine the coupling strength, normally given in terms of the external Q (Qext).
For high-current applications, a good coupler design should also ensure that there are no significant
RF fields from higher-order modes that may cause anisotropic heating at the cold window to minimize
thermal stresses.
The detailed electromagnetic field distribution is exported into commercial mechanical analysis
codes (e.g. ANSYS [39], COSMOS [40]) to calculate stress, vibrations, and heating in regions that
bridge ambient and liquid helium temperatures. The goal is to obtain a low cryogenic heat leak by
introducing thermal intercepts at proper locations and temperatures. To minimize RF losses, the
stainless steel parts of the coupler must be coated with high-conductivity copper of optimal thickness
after taking into account the cryogenic heat leak due to conduction, with the goal of minimizing both
static and dynamic heat loads overall. In the warm sections of the coupler, the design should avoid a
large temperature rise at the operating power level so as to keep manageable the stresses due to
thermal expansion and contraction. Cooling designs should take into account the largest possible
anticipated thermal load due to operation in travelling and standing wave modes. The mechanical
design of the coupler needs to be integrated with the cryomodule design, taking into consideration
assembly sequence issues as well as the movement of coupler parts due to cool-down of the module.
These shifts (10–20 mm) are usually accommodated by bellows integrated into the mechanical design
of the coupler.
Equally important is the implementation of clean practices during fabrication and assembly with
high quality control of materials and platings to ensure reliability and high power performance and to
preserve cavity cleanliness. Sharp edges should be eliminated in design and fabrication to avoid field
enhancement which can lead to field emission. For coated parts, an excellent and reliable bond
between film and substrate is essential to stabilize thermally the film, and to prevent particulate
generation, which can be dangerous for field emission if such particles fall into the cavity. Use of a
cold window is advisable for high gradient applications to seal the cavity from the many coupler
components during the early stages of assembly. The cold window should not be in such close
proximity to the cavity that impact from field emitted electrons from the cavity lead to window
charging, arcing, and possible puncturing. An additional warm window is often used as added
protection for vacuum integrity. The space between the two windows must be actively pumped.
Codes are available (see Ref. [63] for a review) to simulate MP in various regions of the coupler
to assist in making the best choices for the geometry, for example the inner and outer conductor
155
H. PADAMSEE
diameters (and impedance of the coaxial line). In cases where a MP band lies close to an operating
point, voltage or magnetic biasing has been developed to disrupt MP resonance conditions for coaxial
and waveguide input couplers, respectively. Degassing the coupler by baking keeps it free of surface
contamination, thus decreasing the secondary emission and thereby the time required to bring a coupler
to the desired power level through the conditioning process.
6.2 Power requirements

The input power requirement, Pf, is determined by the operating cavity voltage, the beam current, and
the RF overhead called for by peak microphonics detuning, and the LF detuning expected [64, 65]:
 R 
2

 I b Qext   2Q δω  2
V2  Q
Pf =c  1 + cos φs  +  ext m  
R  Vc   ω  
4 Qext
Q 
  
 
were Vc is the cavity voltage, Ib is the average beam current, ϕs is the synchronous phase, δωm is the
amplitude of the frequency detuning, and ω is the RF frequency.
Feedback loops provide cavity field stability, reducing the microphonics influence on beam
quality as well as the RF power overhead required to compensate for microphonics detuning [64–68].
Environmental microphonic noise creates fluctuations in the cavity resonance frequency and thereby
produces amplitude and phase modulations of the field, affecting both beam quality and RF system
performance. This is especially true for high-Q superconducting cavities. The optimum Qext for the
power coupler is determined by beam loading:
Qext = Vc2/[(R/Q)Ib cos φs].
Typical loaded Q for beam-loaded applications range from 105 to several 106. In the case of near
zero beam loading, as for example for an Energy Recovery Linac (ERL), the RF power required
depends on the microphonics detuning level and choice of Qext (or loaded Q) as shown in Fig. 14 for
1.3 GHz, seven-cell cavity at 20 MV·m−1 operating field level.
Fig. 14: Peak RF drive power as a function of loaded Q (QL) for a 1.3 GHz, seven-cell cavity at 20 MV·m−1
accelerating gradient. The power is determined by the peak microphonics cavity detuning during cavity
operation [65].
156
Although the required power decreases substantially with increasing QL, running cavities at Qext
in the108 range is challenging because the small cavity bandwidth of a few hertz makes the RF field
extremely sensitive to perturbations of the resonance frequency due to microphonics and LF detuning.
Operating at high QL makes it hard to meet amplitude and phase stability requirements which can be
quite demanding for some applications, such as a high current CW ERL-based light source, where the
relative rms amplitude stability must be better than a few × 10−4 and rms phase <0.1° in order to
achieve the beam quality necessary for a good light source. In addition, Lorentz forces during filling
detune the cavity by several hundred hertz, making necessary precise compensation during turn-on.
Consequently, the highest loaded Qs are presently limited to several 107; but RF control advances are
forthcoming [65] to lower the RF power requirements for CW accelerators.
Many basic choices need to be made in selecting an RF power coupler design. Among the main
factors governing these choices are the RF frequency, the power level, the ease of cooling, the static
heat leak, and the coupling adjustability required. Figure 15 compares the two primary varieties:
waveguides and coaxial couplers.
(a) (b)
Fig. 15: (a) Coaxial coupler used for SNS 800 MHz cavities adapted from the KEK-B coupler. Gas flow or
water flow through the inner conductor is used for cooling [55]. (b) Waveguide coupler for the CESR 500 MHz
cavity used in several storage rings. A planar ceramic disk-shaped window is incorporated in the warm
waveguide, which is of reduced height [67, 68].
6.3 Waveguide input couplers

Rectangular waveguide couplers are not as widely used as coaxial couplers. Examples of waveguide
couplers include CEBAF at 1500 MHz [69] and CESR at 500 MHz [67].
Waveguide coupling is conceptually simpler since it does not require any RF transition between
the output waveguide of the RF power source and the cavity interface. Coaxial couplers generally
incorporate a transition, such as a door-knob-shaped element. Due to the existence of a cut-off
frequency in waveguides, the size of the waveguide coupler is generally larger at a given operating
frequency than for the coaxial case. Because of the larger cross-section, there is a larger contribution
to the infrared heat transfer to the cryogenic environment. The coupling strength depends on the size
and shape of the coupling iris, the location of the waveguide relative to the cavity’s end cell, and the
location of the terminating short (if any) of the waveguide on the opposite side of the beam tube.
Coupling can be adjusted using an external, three-stub tuner waveguide on the air side [70], though the
additional field stress and heating due to the standing waves in the line can become problematic for
heating and breakdown at high average power levels.
One of the main advantages of the waveguide coupler is the need for cooling only one wall. For
1 MW travelling wave power, the peak electric field for a standard waveguide at 1.3 GHz is
157
H. PADAMSEE
400 kV·m−1, whereas for a coaxial line with an outer diameter equal to the small side of the waveguide
it is 800 kV·m−1 [71]. The power density is lower for the waveguide, but the total longitudinal losses
are the same in both cases, about 1 kW·m−1 in copper. For the coaxial line, about two-thirds of this loss
must be cooled from the inner conductor, which is not as readily accessible as the outer conductor.
The losses at the waveguide wall can normally be intercepted at 70 K or 4.5 K, using straps or heat
exchanger piping. Waveguides also offer a higher pumping conductance over a coaxial line. MP
electrons in the coax can be disrupted by an electrical bias of a few kilovolts [72], whereas MP in the
waveguide can be cleared with magnetic bias using a few gauss [73]. However, this approach is not
possible in the superconducting waveguide section due to persistent screening currents which exclude
dc magnetic flux from the waveguide volume. Here, grooving the waveguide wall is a possible option.
The main disadvantage of waveguide couplers is their size, which increases the mechanical and
thermal complexity of interfaces to the cavity and cryomodule. Plating and flanging are also harder for
rectangular waveguides than for round pipes in coax.
6.4 Coaxial input couplers

Not being limited by a cut-off frequency, coaxial couplers are more compact, especially for low-
frequency systems. A variety of window geometries and arrangements are available - to be discussed.
A large range of coupling values can be achieved by proper insertion of the centre conductor into the
line. Also variable coupling can be achieved with a relatively simple adjustment of the inner conductor
penetration via a bellows extension. The centre conductor can be electrically isolated from the outer
conductor using a kapton film to allow the use of bias voltage to disrupt multipacting. Changing the
diameter or impedance of coax lines is a useful method for pushing MP bands to higher power
levels [74]. But these have a higher thermal radiation heat leak and a larger interface to the beam tube.
The sizing of the coax diameters should avoid azimuthal overmoding.
6.5 Windows
A window provides the physical barrier between the cavity vacuum and open waveguide of the power
source, but the barrier must be transparent to microwaves at the operating frequency. Many designs
use two windows. The main arguments for two windows are (i) to preserve the cleanliness of the
cavity by sealing with a first, cold window, and (ii) vacuum safety provided by a second, usually warm
window. Superconducting cavities must be handled and maintained under Class 10–100 clean-room
conditions at all times to be dust free. It is therefore essential to seal the coupler opening of the cavity
with a window at an early stage in the clean room assembly of the input coupler. Placing a window
near the cavity allows a compact cavity assembly for ease of handling after sealing in the clean room.
Being near the cavity means the window is at 70 K or lower, and therefore must have a vacuum on
both sides. Hence the window can be cooled only by conduction, making high average power design
more challenging. Multipacting can occur in the vacuum on both sides of the cold window. If the cold
window is too close to the cavity field, emission electrons from the cavity can charge it up, leading to
arcing and eventually ceramic damage [75, 76].
The second window prevents gas condensation on the cold window, and serves as a backup to
preserve the cavity vacuum in case the cold window develops a leak during operation. The vacuum
between the windows must be pumped separately. The second window is normally incorporated into
the transition from coaxial to waveguide. It can also be a planar waveguide window or coaxial disk
window. The warm part of the coupler, including the second window, is generally assembled after
placing the cavity string into the vacuum vessel, also under clean and dry conditions for faster
processing to high power. Cooling designs for both windows should take into account the largest
possible anticipated thermal load due to operating the window and coupler in a full standing wave
condition swept through 180° phase change.
158
The cold window design is a must for applications aiming for the highest gradients
(>20 MV·m−1), to prevent dust contamination and field emission during subsequent assembly steps.
For high (≥100 kW) average power applications at moderate to low gradients (5–20 MV·m−1), a
single, warm window design is often used with convection cooling or water cooling. A gas barrier
serves as the second window to provide safety for the cavity vacuum in case the main window
develops a leak. In this case, the cavity is exposed only to the dry, dust-free air in between the two
windows. The warm window is located sufficiently far from the cavity cold mass to limit both
conductive and radiation heat leaks into the liquid helium bath. The challenge for the single warm
window design is that a large coupler assembly must be attached to the clean cavity in a clean room.
Several types of ceramic windows are in use. Coaxial couplers use either the cylindrical window [77]
or disk window [78]. Waveguide couplers generally use a planar rectangular ceramic [79] incorporated
within the rectangular waveguide.
6.6 Prime example of a coaxial coupler

Figure 16 shows the geometry of TTF-III (third generation), 1.3 GHz coupler developed at the Tesla
Test Facility (TTF) used for FLASH and the XFEL and possibly the ILC [80]. The coupler is designed
for operation in the pulsed mode (1.3 ms) at several hundred kilowatt power and less than 5 kW
average power. It has an adjustable coupling strength between 1 ×106 and 2 × 107 for 15 mm antenna
movement. There are two cylindrical windows (97.5% Al2O3 with TiN coating), one at 70 K and one
warm near the door-knob transition. The cold part seals the cavity vacuum, and is entirely inserted into
the cryomodule. The warm part has its own separate vacuum. The cold coaxial line has 70 Ω
impedance with 40 mm Outer Diameter (OD), and the warm line has 50 Ω impedance with 62 mm
OD. All stainless steel parts are made of 1.44 mm thick tubes. Copper plating is 30 µm thick on the
inner conductor and 10 µm thick on the outer conductor. There are two heat intercepts: at 4.2 K and
at 70 K.
Fig 16: TheTTF-III coupler
A prime example of the waveguide version is the CESR SRF waveguide coupler
(Fig. 15(b)) [67]. It has a fixed coupling, Qext = 2 × 105, with a factor of three adjustability via a three-
stub waveguide transformer. Magnetic bias by solenoids wound around the normal conducting
waveguide sections helps to suppress MP.
159
H. PADAMSEE
7 Higher-order modes and couplers
7.1 Higher-order modes

When passing through an accelerating cavity, a particle beam excites a wide spectrum of higher-
modes, depending on the impedances (i.e. the R/Q) of those modes. The resulting electromagnetic
field left behind by the beam is called the wakefield. Thus the passage of the beam can deposit
significant power in high impedance, monopole Higher-Order Modes (HOMs). Unless properly
extracted and damped, HOMs can also cause longitudinal beam instabilities and increase the beam
energy spread. The energy lost by the passage a single bunch, charge q, is given by
ω R
U q = kn q 2 , kn = n sh ,
4 Q0
where ωn is the angular frequency of mode n, Rsh/Q0 is the geometric shunt impedance of the
monopole; kn is also referred to as the loss factor of mode n. The total power deposited depends on the
number of bunches per second, or the beam current.
Among the deflecting modes, dipoles have the highest impedance. The energy lost by a charge
to the dipole mode is given by
2 2
2ρ  2  ωn  ωn Rd
U q = kd q   , kd = a   ,
a  c  4 Q0
where ρ is the bunch displacement off-axis, a is the cavity aperture (radius), ωn is the angular HOM
frequency of mode n, and Rd/Q0 is the dipole mode impedance, formally defined in Ref. [1]. Each
dipole mode has two polarizations split with a small frequency difference due to perturbations, such as
the presence of couplers. Dipole modes with high transverse R/Q are harmful for emittance growth of
the beam.
7.2 HOM couplers

The main functions of HOM couplers are to remove the beam-induced power in the monopole HOMs
and to damp the dangerous monopole and dipole modes to avoid energy spread, beam emittance
degradation, and beam blow-up after multiple beam passages. The beam-induced HOM power in
monopole modes must be extracted from the cavity and deposited to higher temperatures to avoid
cryogenic losses. Modes with high shunt impedance and high Q (i.e. high R/Q*Q) are of particular
concern.
As a prime example for monopole mode excitation and power deposition, consider the
European XFEL, which plans to use the TESLA-shaped cavity developed by TTF. The dangerous
modes are damped by two antenna/loop couplers (Fig. 17) placed just outside the end cells on either
end of the cavity [15]. In general, the coaxial HOM coupler has an antenna or pick-up loop to extract
HOM energy from the end cell via the electric or magnetic fields. Inductive and capacitive elements
within the coupler body enhance coupling at the desired frequencies of high R/Q modes, and suppress
coupling to the fundamental mode via a rejection filter with a large rejection ratio, typically more than
–70 dB. The rejection filter must be carefully tuned prior to cavity installation, and the transmission
line must be terminated by a broad-band load. For high average current application, the couplers are
located inside the helium vessel for good cooling.
The measured damping of these modes is shown in Fig. 18(a) [81]. Antenna/loop couplers on
the beam pipe optimized to damp the low-frequency high-impedance modes are generally not as
effective against a broad spectrum of propagating HOMs with frequencies above the cut-off frequency
of the beam pipe. Hence a broad-band beam pipe absorber is needed for applications with a short
160
bunch. The damping for dipole modes is shown in Fig. 18(b) [81]. Most of the dangerous dipole
modes are well damped relative to the beam dynamics requirement of Qext ~ 105.
As with input couplers, HOM couplers must also be placed outside the cells to avoid field
enhancement in the cells, which may lead to premature quench or multipacting. However, some
HOMs have very little stored energy in the end cells due to mode ‘trapping’ within the central cells of
a many-cell structure. Their suppression becomes difficult. Reducing the number of cells and/or
enlarging the iris diameter minimizes the likelihood of trapping by enhancing the cell-to-cell coupling.
Another approach to minimize trapping is to match the end- and inner-cell frequencies by adjusting
the shape of the end cells, which yields a slightly asymmetric cavity.
Antenna/loop-based HOM couplers also introduce kicks, which can spoil the beam emittance
especially at low beam energy. These kicks arise from the wakefields introduced by the coupler
geometry as well as by RF fields and the asymmetry of the coupler locations with respect to the beam
axis. The kicks are reduced by symmetrizing the placement of multiple HOM couplers.
Fig. 17: The TESLA HOM coupler
(a) (b)
Fig. 18: (a) Damping of dangerous monopole HOMs. (b) Damping of the dangerous dipole HOMs
The TESLA HOM coupler was scaled to 805 MHz for SNS at 6% duty cycle operation, and to
1500 MHz for the CEBAF 12 GeV upgrade and CW operation, and also to 3.9 GHz for the third-
harmonic TTF injector cavity [82, 83]. These higher average power applications have met with some
difficulties so that modifications had to be developed. The problem arises due to the heating of the
output antenna by the residual magnetic field of the fundamental mode (several percent of the field on
equator) and the heat leak of the output line. Abnormal heating of HOM couplers can detune the notch
filter and couple out substantial amounts of power from the fundamental mode, leading to thermal
runaway. Other causes for abnormal heating observed are multipacting in the HOM coupler, and
heating from the impact of field emitted electrons emanating from the cavity, or from neighbouring
cavities. Enabling higher duty factor operation (or higher damping) by bringing the coupler tip closer
to the end cell requires improvements in cooling. There is a significant amount of stored energy in the
transmission line coupler. The high electric field regions of the loop coupler are also susceptible to
161
H. PADAMSEE
multipacting and associated heating. The troublesome regions are between the loop and the wall, in the
small gap which defines the notch filter, between the coaxial post and the end wall of the can, and at
several places between the post and the cylindrical wall.
Microwave analysis combined with thermal analysis using codes such as HFSS and ANSYS
have been used to analyze heating difficulties and devise solutions [84, 85]. To keep the output
antenna superconducting, one approach has been to shorten the antenna probe tip, provided the HOM
coupling loss can be tolerated. Another is to enhance the heat conduction at the output connector, for
example by using a larger RF feed-through with sapphire window and cooling copper blocks [86].
Shortening the antenna tip is one way to reduce the fields and suppress MP at the required field levels.
7.3 Waveguide HOM couplers

Waveguides as HOM couplers provide convenient and natural high pass filters that reject the
fundamental mode and require no fine tuning, unlike the fundamental mode rejection filter essential to
coaxial couplers. The waveguide coupler removes HOMs over a wide frequency range. It can handle
high HOM power without heating difficulties. However, the bulkier waveguide adds to structure cost
of the HOM-end groups.
The first waveguide HOM couplers for superconducting cavities were developed at Cornell for
1.5 GHz muffin-tin cavities, and subsequently adopted for the 1.5 GHz Cornell/CEBAF elliptical five-
cell cavities (Fig. 19). The several milliamp beam current of CEBAF results in a small amount of the
HOM power to allow termination of the HOM couplers with waveguide loads inside cryomodules
[87]. In case of higher beam current, the terminating loads must be located outside the cryomodule,
which adds to the mechanical complexity of the waveguide option and introduces additional cryogenic
heat leaks and shielding requirements.
Fig. 19: Waveguide HOM coupler examples for the Cornell/CEBAF cavity
7.4 Beam pipe couplers and absorbers

The cavity beam pipe can be viewed as a transmission line to couple out HOMs with frequencies
above the beam pipe cut-off. The fundamental mode is below cut-off and does not propagate out,
providing a natural rejection filter. As with the waveguide coupler, no tuning is needed. The
cylindrical symmetry of the beam pipe avoids coupler kicks. The diameter of the beam pipe can be
chosen to couple out all monopole and dipole modes. Extraction of the first two dipole modes
demands the largest opening. These high-impedance modes are especially dangerous. But a large
diameter beam pipe reduces the rejection and R/Q of the fundamental mode and enhances the peak
surface fields for operation. These effects can be reduced by introducing a rounded step in the beam
pipe at the iris of the end cell. In some cases, as for KEK-B [88], the largest beam pipe is installed at
only one end of the cavity for the extraction of the first two dipoles. A section of the beam pipe lined
with a microwave absorbing material serves as the load. This can be placed at room temperature
outside the cryostat, or at ~80 K inside. But the presence of numerous absorbing sections along the
beam line reduces the overall real-estate gradient.
162
Beam-line HOM couplers (Fig. 20) are especially suitable for high-current, short bunch
accelerators. The development of these are reviewed in Ref. [1]. Several storage rings now use the
approach: CESR at Cornell, KEK-B in Tsukuba, Taiwan Light Source, Canadian Light Source,
DIAMOND light source, and BEPC-II in Beijing. The HOMs are damped to Q values between 100
and 1000. Measurements of the electromagnetic properties of absorbing materials have been
performed.
Fig. 20: Beam-pipe absorbers lined with ferrite [89]. The CESR load with 3 mm TT211R ferrite tiles bonded to a
sintered copper–tungsten plate with Ag–Sn alloy.
8 Tuners
Frequency tuners are an essential component of acceleration systems. Both slow tuners and fast tuners
fulfil important functions. Recent review talks can be found in Refs. [90−92]. Slow tuners bring a
cavity resonance to the operating frequency, compensating for a variety of effects: cavity dimensional
changes due to evacuation and cool-down, or slow drifts in frequency due to pressure changes in the
helium bath surrounding the cavity. Tuners also compensate for the reactive effects of beam loading in
high-current accelerators to minimize the reflected power. Occasionally cavities need to be detuned to
bypass operation, or for diagnostic purposes.
Slow tuners must cover a wide tuning range (of up to several hundred kilohertz), while
providing a resolution of the order of 1 Hz. Slow tuners are usually motor driven. Fast tuners provide a
smaller tuning range of several cavity bandwidths, but with a control bandwidth of several kilohertz
and slew rates of 1 µm in 100 µs. Together with feed-forward and feedback, these tuners compensate
for static and dynamic LF detuning, especially at high-gradient operation [93]. Fast tuners also have
the potential to control microphonics, typically up to several tens of hertz. Fast tuners are important for
cavities with little beam loading, when operation at high Qext is desirable to minimize RF power. At
high Qext, the bandwidth is sufficiently narrow that microphonic excitations disrupt the cavity
resonance. Typical microphonic noise levels are of the order of few hertz to several tens of hertz with
a frequency spectrum ranging up to a few hundred hertz. The observed spectrum is a result of a
convolution of the spectrum of excitation and the coupling to the mechanical resonances of the
cavities. Typical excitation sources of microphonics are vibrations from pumps and human activity. If
the repetition frequency of the coarse tuner stepping motor matches a mechanical resonance of the
cavity–cryostat system, strong mechanical vibrations can be excited. To avoid microphonic excitations
in general, it is important to ensure that the mechanical resonant frequencies of the structure do not
coincide with the frequency of RF repetition rate
Fast tuners are generally integrated with the slow mechanical tuner and mostly based on
piezoelectric elements [94, 95]. The typical static tuning range is 1 kHz or less.
163
H. PADAMSEE
Tuner designs strive for compactness to avoid wasting beam-line space, disrupting the field
flatness from cell–cell, or tuning neighbouring cavities. The tuner mechanical supports and operating
motors should keep cryogenic heat load to a minimum. Frequency tuners should be free of hysteresis.
Pre-setting is required to avoid the neutral point between tension and compression over the entire
expected range of operation. The hardware must be easy to maintain and repair, ideally without the
need to warm up or disassemble a module, but in practice this has been realized only for a few tuner
choices.
A variety of tuner designs are now available as a result of inventive efforts at a number of
laboratories. One such tuner used at FLASH and the XFEL, called the Saclay/TTF tuner, is shown
in Fig. 21.
(a) (b)
Fig. 21: (a) Basic principle of most slow tuners [96]. (b) Principle of the Saclay tuner used in TTF [97]
Fig. 22 (a) Principle of lever–cam design of the improved Saclay tuner [96]. (b) 3D CAD layout [97]
164
References
[1] H. Padamsee, J. Knobloch and T. Hays, RF Superconductivity for Accelerators (Wiley & Sons,
New York, 1998).
[2] H. Padamsee, RF Superconductivity: Science, Technology and Applications (Wiley-VCH,
Weinheim, 2009).
[3] P. Schmueser, Basic principles of RF superconductivity and superconducting cavities, Proc.
11th Workshop on RF Superconductivity, Travemünde, Germany, 2003, paper MoTo 1.
[4] H. Padamsee, in Frontiers of Accelerator Technology, Eds. S.I. Kurokawa et al. (World
Scientific, Singapore, 1996), p. 383.
[5] D. Proch, in Handbook of Accelerator Physics and Engineering (World Scientific, Singapore,
1999), p. 530.
[6] S. Belomestnykh, Rev. Accel. Sci. Technol. 5 (2012) 147.
[7] M. Kelly, Rev. Accel. Sci. Technol. 5 (2012) 185.
[8] M.P. Kelly, in Handbook of Accelerator Physics and Engineering (World Scientific, Singapore,
2013), p. 681.
[9] H. Padamsee, in Encyclopedia of Electrical and Electronics Engineering, Ed. J.G. Webster
(John Wiley and Sons, New York, 1999).
[10] D. Proch, Rep. Prog. Phys. 61 (1998) 431.
[11] H. Padamsee, Accelerating applications of RF superconductivity—success stories, Proc. 2004
Applied Superconductivity Conf., Jacksonville, FL, 2004, p. 2432.
[12] W. Weingarten, Superconducting cavities—basics, Proc. Joint US-CERN-Japan Int. School,
Frontiers of Accelerator Technology, Eds. S.I. Kurokawa et al. (World Scientific, Singapore,
1994), p. 311.
[13] H. Padamsee and J. Knobloch, Issues in superconducting RF technology, Proc. Joint US-
CERN-Japan Int. School, Frontiers of Accelerator Technology, Eds. S.I. Kurokawa et al.
(World Scientific, Singapore, 1994), p. 101.
[14] E. Haebel et al., Cavity shape optimization for superconducting linear collider, Proc. HEACC
1992, High Energy Accelerator Conference, Hamburg, 1992, p.957.
[15] B. Aune et al., Phys. Rev. ST Accel. Beams 3 (2000) 092001.
[16] H. Padamsee et al., Part. Accel. 40 (1992) 17.
[17] H. Padamsee et al., Accelerating cavity development for the Cornell B-Factory, CESR-B, Proc.
PAC1991, Particle Accelerator Conf., San Francisco, CA, 1991, vol. 2, p. 786.
[18] T. Furuya et al., Proc. 7th Workshop on RF Superconductivity, Gif-sur-Yvette, France, 1995, p.
729.
[19] P. Kneisel et al., Nucl. Instrum. Meth. Phys. Res. 188 (1981) 669.
[20] U. Klein and D. Proch, Multipacting in superconducting RF structures, Proc. Conf. Future
Possibilities for Electron Accelerators, Charlottesville, NC, 1979, p. N1.
[21] P. Schmueser, Tuning of Multi-cell cavities using bead-pull measurements, SRF920925-10,
Cornell University Internal Report.
[22] R.L. Geng, Multipacting simulations for superconducting cavities and RF coupler waveguides
Proc. PAC2003, Particle Accelerator Conf., Portland, Oregon, 2003, p. 264.
[23] V. Shemelin, Phys. Rev. ST Accel. Beams 16 (2013) 012002.
[24] S.-H. Kim, SNS superconducting linac operational experience and upgrade path Proc.
LINAC’08 Linear Accelerator Conference, Victoria, British Columbia, Canada, 2008, p. 11.
[25] C.C. Compton et al. Phys. Rev. ST Accel. Beams 8(4) (2005).
165
H. PADAMSEE
[26] J.R. Delayen, Medium-β superconducting accelerating structures, Proc. 10th Workshop on RF
Superconductivity, Tsukuba, Japan, 2001.
[27] J.R. Delayen, Medium-β superconducting accelerating structures, USPAS, Baton Rouge, LA,
2003.
[28] J.R. Delayen, Low and medium β cavities and accelerators, Proc. 13th Workshop on RF
Superconductivity, Beijing, China (2007), tutorial 4a.
[29] K.W. Shepard et al., Prototype 350 MHz, niobium spoke-loaded cavities, Proc. PAC 1999,
Particle Accelerator Conference, New York, NY, 1999, p. 955.
[30] K.W. Shepard et al., Development of spoke cavities for RIA, Proc. 12th Workshop on RF
Superconductivity, Ithaca, NY, 2005, p. 334.
[31] A. Facco, Tutorial on low beta cavity design, Proc. 12th Workshop on RF Superconductivity,
Ithaca, NY, 2005, p. 21.
[32] G. Devanz, SPIRAL 2 resonators, Proc. 12th Workshop on RF Superconductivity, Ithaca, NY,
2005, p. 108.
[33] P.N. Ostroumov and K.W. Shepard, Phys. Rev. ST. Accel. Beams 11 (2001) 030101.
[34] N. Added et al., Upgraded phase control system for superconducting low velocity accelerating
structure, Proc. LINAC 1992, Linear Accelerator Conference, Ottawa, Ontario, 1992, p. 181.
[35] A. Facco et al., On-line performance of the LNL mechanically damped superconducting low
beta resonators, Proc. EPAC 1998, European Particle Accelerator Conference, Stockholm,
1998, p. 1846.
[36] A. Facco, Part. Accel. 61 (1998) 265.
[37] K.W. Shepard, Superconducting low-velocity linac for the Argonne positive-ion injector, Proc.
PAC 1989, Particle Accelerator Conference, Chicago, IL, 1989, p. 1974.
[38] M.Pekeler, et al, Performance of a Prototype 176 MHz β= 0.09 Half-Wave Resonator for the
SARAF Linac, Proc. 12th Workshop on RF Superconductivity, Ithaca, NY, USA, p. 331 (2005).
[39] ANSYS, Inc., Southpointe, Canonsburg, PA, http://www.ansys.com
[40] COSMOS, Structural Research and Analysis Corp., Santa Monica, CA,
http://www.cosmosm.com
[41] J.R. Delayen, Ponderomotive instabilities and microphonics, a tutorial, Phys. C 441 p. 1 (2006).
[42] J. Sekutowicz, Superconducting high-β cavities, Proc. 13th Workshop on RF Superconductivity,
Beijing, China, 2007, tutorial 2a.
[43] R. Mitchell et al., Lorentz force detuning analysis of the Spallation Neutron Source (SNS)
accelerating cavities, Proc. 10th Workshop on RF Superconductivity, Tsukuba, Japan, 2001,
p. 236.
[44] J.R. Delayen, LLRF control and tuning systems, Proc. 13th Workshop on RF
Superconductivity, Beijing, China, 2007, tutorial 4b.
[45] J.R. Delayen, Low and medium β cavities and accelerators, Proc. 13th Workshop on RF
Superconductivity, Beijing, China, 2007, tutorial 4a.
[46] P. Brown et al., Operating experience with the LEP2 superconducting RF system, Proc. 10th
Workshop on RF Superconductivity, Tsukuba Japan, 2001, p. 185.
[47] M. Liepe et al., Dynamic Lorentz force compensation with a fast piezoelectric tuner, Proc. PAC
2001, Particle Accelerator Conference, Chicago, IL, 2001, p. 1074.
[48] S. Simrock, Advances in RF control for high gradients, Proc. 9th Workshop on RF
Superconductivity, Santa Fe, NM, 1999, p. 92.
[49] S. Simrock, Achieving phase and amplitude stability in pulsed superconducting cavities, Proc.
10th Workshop on RF Superconductivity, Tsukuba, Japan, 2001, p. 231.
166
[50] A. Brandt et al., General automation of LLRF control for superconducting accelerators, Proc.
12th Workshop on RF Superconductivity, Ithaca, NY, 2005 [Phys. C 441 (2006) 263].
[51] V. Shemelin et al., Dipole-mode-free and kick-free 2-cell cavity for the SC ERL injector, Proc.
PAC 2003, Particle Accelerator Conference, Portland, OR, 2003, p. 2059.
[52] M.S. Champion, RF input couplers and windows: performances, limitations, and recent
developments, Proc. 7th Workshop on RF Superconductivity, Gif-sur-Yvette, France, 1995,
p. 195.
[53] D. Proch, Techniques in high-power components for SRF cavities, a look to the future, Proc.
LINAC2002, Linear Accelerator Conference, Gyeongju, Korea, 2002, p. 529.
[54] B. Rusnak, RF power and HOM coupler tutorial, Proc. 11th Workshop on RF
Superconductivity, Travemünde, Germany, 2003, p. 496.
[55] I. Campisi, Fundamental power couplers for superconducting cavities, Proc. 10th Workshop on
RF Superconductivity, Tsukuba, Japan, 2001, p. 132.
[56] I. Campisi, State of the art power couplers for superconducting RF cavities, Proc. EPAC 2002,
European Particle Accelerator Conference, Paris, France, 2002, p. 144.
[57] S. Belomestnykh, Review of high power CW couplers for superconducting cavities, Proc.
Workshop on High-Power Couplers for Superconducting Accelerators, Newport News, VA,
2002, http://www.jlab.org/intralab/calendar/archive02/HPC/papers.htm
[58] S. Belomestnykh, Overview of input power coupler developments, pulsed and CW, Proc. 13th
Workshop on RF Superconductivity, Beijing, China, 2007, paper WE305.
[59] T. Garvey, The design and performance of CW and pulsed power couplers – a review, Proc.
12th Workshop on RF Superconductivity, Ithaca, NY, 2005 [Phys. C 441 (2006) 209].
[60] A. Variola, High power couplers for linear accelerators, Proc. LINAC 2006, Linear Accelerator
Conference, Knoxville, TN, 2006, p. 531.
[61] CST Microwave Studio, CST GMbH, Darmstadt, Germany,
http://www.cst.com/Content/Products/MWS/Overview.aspx
[62] HFSS, Ansoft Corp., Pittsburgh, PA, http://www.ansoft.com/products/hf/hfss/
[63] F. Krawczyk, Status of multipacting simulation capabilities for SCRF applications, Proc. 10th
Workshop on RF Superconductivity, Tsukuba, Japan, 2001, p. 108.
[64] M. Liepe, Microphonics detuning in the 500 MHz superconducting CESR cavities, Proc. PAC
2003, Particle Accelerator Conference, Portland, OR, 2003, p. 1326.
[65] M. Liepe et al., Pushing the limits: RF field control at high loaded Q, Proc. PAC 2005, Particle
Accelerator Conference, Knoxville, TN, 2005, p. 2642.
[66] M. Liepe, RF parameter and field stability requirements for the Cornell ERL prototype, Proc.
PAC 2003, Particle Accelerator Conference, Portland, OR, 2003, p.1329.
[67] S. Belomestnykh and H. Padamsee, Performance of the CESR superconducting RF system
and future plans, Proc. 10th Workshop on RF Superconductivity, Tsukuba, Japan, 2001, p. 197.
[68] S. Belomestnykh et al., Superconducting RF system upgrade for short bunch operation of
CESR, Proc. PAC 2001, Particle Accelerator Conference, Chicago, IL, 2001, p.1062.
[69] J.R. Delayen et al., An RF input coupler system for the CEBAF energy upgrade cryomodule,
Proc. PAC 1999, Particle Accelerator Conference, New York, NY, 1999, p. 1462.
[70] V. Veshcherevich and S. Belomestnykh, Correction of the coupling of CESR RF cavities to
klystrons using three-post waveguide transformers, Report SRF020220-02, Laboratory for
Elementary-Particle Physics, Cornell University (2002).
[71] D. Proch, Techniques in high-power components for SRF cavities, a look to the future, Proc.
LINAC2002, Linear Accelerator Conference, Gyeongju, Korea, 2002, p. 529.
167
H. PADAMSEE
[72] J. Tuckmantel et al., Improvements to power couplers for the LEP2 superconducting cavities,
Proc. PAC 1995, Particle Accelerator Conference, Dallas, TX, 1995, p. 1642.
[73] R.L. Geng et al., Multipacting in a rectangular waveguide, Proc. PAC 2001, Particle
Accelerator Conference, Chicago, IL, 2001, p. 1228.
[74] E. Somersalo et al., Analysis of multipacting in coaxial lines, Proc. PAC 1995, Particle
Accelerator Conference, Dallas, TX, 1995, p. 1500.
[75] B. Yunn and R. M. Sundelin, Field emitted electron trajectories for the CEBAF cavity, Proc.
PAC1993, Particle Accelerator Conference, Washington, D.C., 1993, p. 1092.
[76] L. Phillips et al., New window design options for CEBAF energy upgrade, Proc. PAC 1997,
Particle Accelerator Conference, Vancouver, Canada, 1997, p. 3102.
[77] W.-D. Moeller et al., Development and testing of RF double window input power couplers for
TESLA, Proc. 12th Workshop on RF Superconductivity, Ithaca, NY, 2005, p. 571.
[78] S. Noguchi et al., Recent status of the TRISTAN superconducting RF system, Proc. EPAC
1994, European Particle Accelerator Conference, London, 1994, p. 1891.
[79] E. Chojnacki, Tests and Designs of High-Power Waveguide Vacuum Windows at Cornell,
Proceedings of the 1997 Workshop on RF Superconductivity, Abano Terme (Padova), Italy,
p. 753 (1997).
[80] W.-D. Moeller, High power coupler for the TESLA test facility, Proc. 9th Workshop on RF
Superconductivity, Santa Fe, NM, 1999, p. 577.
[81] J. Sekutowicz, Higher order mode coupler for TESLA, TESLA note 1994-07 (1994).
[82] S. Tariq and T. Khabiboulline, FNAL 3.9 GHz HOM coupler & coaxial cable thermal FEA,
Proc. 12th Workshop on RF Superconductivity, Ithaca, NY, 2005, p. 604.
[83] E. Harms, Status of 3.9-GHz superconducting RF cavity technology at Fermilab, Proc. LINAC
2006, Linear Accelerator Conference, Knoxville, TN, 2006, p. 695.
[84] G. Wu et al., Electromagnetic simulations of coaxial type HOM coupler, Proc. 12th Workshop
on RF Superconductivity, Ithaca, NY, 2005, p. 600.
[85] N. Solyak, New design of the 3.9 GHz HOM coupler, TTC Meeting, KEK, Sept. 25–28, 2006.
[86] C. E. Reece et al., High thermal conductivity cryogenic RF feedthroughs for higher order mode
couplers, Proc. PAC 2005, Particle Accelerator Conference, Knoxville, TN, 2005, p. 4108.
[87] I. Campisi, Artificial dielectric ceramics for CEBAF’s higher-order mode loads, Proc. 6th
Workshop on RF Superconductivity, Newport News, VA, 1993, p. 587.
[88] T. Furuya et al., Superconducting accelerator cavity for KEK B-Factory, Proc. 7th Workshop
on RF Superconductivity, Gif-sur-Yvette, France, 1995, p. 729.
[89] S. Belomestnykh et al., Comparison of the predicted and measured loss factor of the
superconducting cavity assembly for the CESR upgrade, Proc. PAC 1995, Particle Accelerator
Conference, Dallas, TX, 1996, p. 3394.
[90] S. Simrock, Review of slow and fast tuners, Proc. 12th Workshop on RF Superconductivity,
Ithaca, NY, 2005, paper ThA07.
[91] E.F. Daly, Overview of existing mechanical tuners, ERL Workshop, 18–23 March, 2005, oral
communication.
[92] S. Noguchi, Review of new tuner designs, Proc. 13th Workshop on RF Superconductivity,
Beijing, China, 2007, paper WE303.
[93] S. Simrock, Control of microphones and Lorentz force detuning with a fast mechanical tuner,
Proc. 11th Workshop on RF Superconductivity, Travemünde, Germany, 2003, paper TuO09.
[94] G. Devanz, Active compensation of Lorentz force detuning of a TTF 9-cell cavity in
CRYHOLAB, Proc. LINAC 2006, Linear Accelerator Conference, Knoxville, TN, 2006, p. 598.
168
[95] M. Liepe, et al., Dynamic Lorentz force compensation with a fast piezoelectric tuner, Proc. PAC
2001, Particle Accelerator Conference, Chicago, IL, 2001, p.1074.
[96] S. Noguchi, Review of new tuner designs, Proc. 13th Workshop on RF Superconductivity,
Beijing, China, 2007, paper WE303.
[97] P. Bosland, Tuning systems for superconducting cavities at Saclay, SOLEIL Workshop, 2007,
2007/ESLS-RF/ESLS-RF-PRESENTATIONS/07-ESLS07-PBosland.pdf.
169
SRF Cavity Fabrication and Materials
W. Singer 1
Deutsches Elektronen-Synchrotron DESY, Hamburg, Germany
Abstract
The technological and metallurgical requirements of material for high-
gradient superconducting cavities are described. High-purity niobium, as the
preferred metal for the fabrication of superconducting accelerating cavities,
should meet exact specifications. The content of interstitial impurities such
as oxygen, nitrogen, and carbon must be below 10µg/g. The hydrogen
content should be kept below 2µg/g to prevent degradation of the Q-value
under certain cool-down conditions. The material should be free of flaws
(foreign material inclusions or cracks and laminations) that can initiate a
thermal breakdown. Defects may be detected by quality control methods
such as eddy current scanning and identified by a number of special
methods. Conventional and alternative cavity fabrication methods are
reviewed. Conventionally, niobium cavities are fabricated from sheet
niobium by the formation of half-cells by deep drawing, followed by trim
machining and Electron-Beam Welding (EBW). The welding of half-cells is
a delicate procedure, requiring intermediate cleaning steps and a careful
choice of weld parameters to achieve full penetration of the joints. The
equator welds are particularly critical. A challenge for a welded construction
is the tight mechanical and electrical tolerances. These can be maintained by
a combination of mechanical and radio-frequency measurements on half-
cells and by careful tracking of weld shrinkage. The established procedure is
suitable for large series production. The main aspects of quality assurance
management are mentioned. Another cavity fabrication approach is slicing
discs from the ingot and producing cavities by deep drawing and EBW.
Accelerating gradients at the level of 35–45 MV·m–1 can be achieved by
applying Electropolishing (EP) treatment. Furthermore, the single-crystal
option (grain boundary free) is promising. It seems that in this case, high
performance can be achieved by a simplified treatment procedure.
Fabrication of the accelerating structure from a seamless pipe as a cost-
effective alternative is briefly described. This technology has yielded good
results in single-cell cavities and is already available for multi-cell
structures.
Keywords: superconducting cavities, fabrication, materials, quality

assurance, accelerators.
1 Introduction
Niobium, having the highest critical temperature and critical magnetic field of all pure metals (critical
temperature 9.3 K; superheating field of ~240 mT) has for many years now been the preferred metal
for the fabrication of superconducting RF cavities [1–8]. This is because niobium is chemically inert
1
waldemar.singer@desy.de
W. S INGER
(at room temperature, the surface is covered by a protecting pentoxide layer), it can easily be
machined and deep drawn, and it is available on the market, in the required amounts, in bulk and sheet
material form. In the past three decades, improvements of Electron-Beam Melting (EBM) and
purification techniques have made good progress [9, 10]. The purity of industrially produced niobium
has steadily increased due to the reduction of concentrations of metallic impurities such as Ta and W
and of the interstitially dissolved elements carbon, oxygen, and nitrogen. The application of ultra-high-
vacuum technology additionally increases purity. Several companies worldwide are currently in a
position to produce tonnes of niobium for high-gradient cavities.
A few European companies already have about 30 years’ experience in mechanical fabrication
of superconducting cavities. New companies in the USA and Asia are joining this community,
especially in the context of anticipating the construction of an International Linear Collider (ILC) that
would require about 20 000 superconducting cavities. The new trend is not only mechanically to
produce cavities in the industry, but also to ask industry to perform cavity surface treatment and
provide laboratories with cavities ready for installation in accelerating cryo-modules.
The main aspects of niobium production as a base material and the manufacturing of
superconducting resonators from this material are described below.
2 Niobium as a superconducting material for cavity fabrication
2.1 From ore to semi-finished product

Niobium and tantalum always occur in association with one another in nature [11–14] (Table 1).
The main mineral, pyrochlore, is mostly processed by primarily physical processing technology
to obtain niobium oxide, with a concentration from 55% to about 60%. The most important reserves
are calcium niobate from Brazil (about 80%) and Canada (about 10%), and zinc or tin slags containing
niobium and tantalum from Zaire, Nigeria, and Russia, as well as stibiotantalite from West Australia.
The second most important source of Nb-producing ore is niobite(columbite)–tantalite [(Fe,
Mn)(Nb,Ta)2O6]. When tantalum significantly exceeds niobium by weight, it is tantalite; and when
niobium significantly exceeds tantalum by weight, it is columbite. As a mineral with a ratio of
Nb2O5:Ta2O5 from 10:1 to 13:1, columbite occurs in Brazil, Nigeria, and Australia, as well as other
countries of Central Africa. Usually, niobium is recovered when the ores are processed for tantalum.
The world’s largest niobium deposit, located in Araxá, Brazil, is owned by Companhia
Brasileira de Metalurgia e Mineração (CBMM). The reserves are enough to supply current world
demand for about 500 years, about 460 million tons. The weathered ore mine contents are between 2.5
and 3.0% Nb2O5. The mining is carried out by open-pit extraction. The ore is crushed and magnetite is
magnetically separated from the pyrochlore. By chemical processes, the ore is concentrated with
regard to its Nb content (50–60% of Nb2O5).
Table 1: The most important niobium minerals
Composition Percentage of Percentage of Percentage of

Mineral
Nb2O5 + Ta2O5 Nb2O5 Ta2O5
NaCaNb2O6F 38–73 26–73 0.2–22
Pyrochlore
(Fe,Mn)(Nb,Ta)2O6 75–81 47–78 0.1–34
Niobite
(columbite)
(Fe,Mn)(Ta,Nb)2O6 81–86 2–27 53–84
Tantalite
Niobium ores are mainly processed into concentrates by opening up with hydrofluoric acid.
Then the niobium and tantalum oxides are separated from one another. The method used for Ta and
172
SRF C AVITY FABRICATION AND M ATERIALS
Nb separation (liquid–liquid extraction or solvent extraction) is based on their relative solubility in two
different immiscible liquids. The best method to produce raw material on an industrial scale is liquid–
liquid extraction using methyl isobutyl ketone (MIBK: C6H12O). Niobium is recovered as niobium
oxide Nb2O5 via neutralization of the niobium fluoride complex with ammonia to form the hydroxide,
followed by calcination to the oxide. Generally, the tantalum values in solution are converted into
tantalum oxide (Ta2O5). Filtration of the liquid mixture and further processing via solvent extraction
using MIBK produces highly purified solutions of tantalum and niobium. The tantalum content in Nb
oxides separated by liquid–liquid extraction is at a level below 500 µg/g (100 µg/g is reachable).
Classical routes from niobium oxide to metal (see Fig. 1 [10]) consist of carbothermic reduction of
Nb2O5 or aluminothermic reduction according to the equations
Nb 2 O5 + 7C → 2NbC + 5CO,
5NbC + Nb 2 O5 → 7Nb + 5CO,
3Nb 2 O5 + 10Al → 5Al2 O3 + 6Nb,
followed by Electron-Beam Melting (EBM).
The production of high-grade niobium with a small Ta concentration can be performed via the
sodium reduction of purified K2NbF7:
K 2 NbF7 + 5Na → Nb + 2KF + 5NaF.
An optional route for niobium fabrication is powder metallurgy. Niobium can be mechanically
pulverized using cooling from 950°C in a hydrogen atmosphere. After the hydriding treatment, the
metals are crushed, ground, sieved and dehydrided in a vacuum. The powder can then be mechanically
pressed to compacts and used for EBM.
Fig. 1: An overview of the niobium production and purification processes [10]
173
W. S INGER
2.2 Purification
2.2.1 Refining by EBM

As a result of increasing demand for refractory metals in the past few decades, the electron-beam
furnace has been developed into a reliable, efficient apparatus for melting and purification [10, 15–
18]. Evaporation of most impurities by EBM in a vacuum of typically better than 3 × 10–4 mbar for the
first melt and 2 × 10–6 mbar for the last melt is very effective for niobium because of its high melting
point (2468°C). Therefore, it is possible to produce tons of extremely pure niobium industrially.
The principle of EBM equipment, which is now routinely available in the industry, is shown in
Fig. 2 [10].
Fig. 2: A schematic of the EBM equipment [17]
As the ingot is melting, molten metal globules drop into a pool on the ingot, which is contained
in a water-cooled copper cylinder (crucible). For the first melt, the powder, the granulate and the solid
materials are pressed to form an electrode that is fed in horizontally. The material is fed in vertically in
the subsequent melts. Additional purification occurs by repeatedly re-melting under vacuum. In part,
the energy of the electron beam is used to melt the ingot and in part it is used to maintain the pool of
metal liquid. During melting, all gases and impurities that have a melting temperature lower than that
of niobium are evaporating. The ingot is continuously guided through the crucible. The rate of
guidance is co-ordinated with the rate of the material melting to ensure complete melting of the feed
material and proper refining. The melting temperature is a compromise between maximization of
purification and minimization of material losses by evaporation.
Melted ingots sometimes demonstrate a non-homogeneous distribution of impurities from top to
bottom. The skin of the ingot contains more impurities than the inside. Machining away the skin and
cutting away a fraction of the ingot bottom are recommended for a purer final product.
The furnace design known as the e-beam cold hearth furnace could be an effective alternative to
the method described above. This design provides maximum exposure of the melt to the vacuum
environment by melting in a rectangular hearth and at the same time casting a slab shape. Such a shape
is more suitable for rolling than the round ingot.
174
The contents of the interstitial impurities O, N, and C that significantly influence the properties
of the niobium can be reduced down to a level below 10 µg/g; an H content close to 1 µg/g is
reachable. This content achieved by melting should be maintained during fabrication and treatment.
Four to six melting steps are generally necessary to reach the Residual Resistivity Ratio (RRR) = 300
level, with a few µg/g of interstitial impurities; an RRR of up to 500 can ultimately be achieved
(Fig. 3 [18]).
600
500
400
300
200
RRR
100
0
1st Remelt. 2nd Remelt. 3rd Remelt. Annealed
Sheet
Min RRR Max RRR Average RRR
Fig. 3: The RRR values from the first, second and third re-melted ingot and annealed sheet [18]
Complicated processes of evaporation, degassing, solvation, decomposition, and diffusion take

place during EB melting of the ingot (Fig. 4 [10]). It can be seen that for H2 and N2, thermodynamic
equilibrium between the gas phase and the interstitially dissolved H and N atoms is established
according to the reactions H2 (g) ↔ 2 H (in Nb) and N2 (g) ↔ 2 N (in Nb). For oxygen, a balance
between the uptake of O2 on the one hand and H2O and oxide evaporation as NbO or NbO2 on the
other is achieved. Unlike oxygen, carbon desorbs only in the form of CO.
Fig. 4: A schematic of the gassing and degassing reactions between niobium and residual gases in vacuum at
high temperatures [10].
2.2.2 Purification by electrolysis in molten salts

Fundamental investigations of the electrodeposition of Nb from LiF–NaF–KF melts containing
K2NbF7, at T = 700–800°C, have shown that this electrolyte is suitable for material refining
(electrolytic refining) [9]. Niobium becomes extremely pure in relation to metallic elements; for
example, the tantalum content can be reduced by this treatment from 400 to 3.5 µg/g (see Fig. 5).
175
W. S INGER
Subsequent experiments have shown that similar results can be achieved using niobium fluoride with
potassium fluoride (10–20%) and sodium fluoride (5–15%), together with an equimolecular mixture of
chlorides (NaCl–KCl) for the remainder [19]. This rather expensive method was applied in the former
Soviet Union in the 1980s for the industrial production of niobium, as a refining step before EBM. An
RRR purity level of about 1000 was achieved in Nb sheets that were usable for cavity production.
Fig. 5: The purification of niobium by electrolysis in molten salts [9]
Fig. 6: The behaviour of the hydrogen partial pressure during annealing of niobium [21]
176
2.3 Post-purification
In some cavity surface treatment processes, it is very difficult to completely avoid contamination by
interstitials. For example, contamination by hydrogen is almost inevitable during Electropolishing
(EP) or Centrifugal Barrel Polishing (CBP). Furthermore, additional purification of niobium increases
the expectations of better performance. The possible purification of already mechanically produced
cavities is described below.
2.3.1 Hydrogen degassing

Hydrogen degassing can be achieved by common annealing. A plot of the hydrogen partial pressure
versus temperature normally shows two peaks around 350°C and 600–650°C. It can be assumed that
the peak at 350°C is caused by surface-trapped hydrogen, while the other peak indicates the bulk of
the hydrogen contamination. There are some differences of opinion concerning the optimal hydrogen
degassing temperatures. Annealing at 600°C for 10 h has been proposed in the literature [20].
Investigation [21] (see Fig. 6) shows that heat treatment at 800°C for 5 h will result in a larger
depletion of hydrogen from the material than 600°C for 10 h. Annealing at 800°C for 2 h is applied for
European X-ray Free Electron Laser (XFEL) cavities [22], which in our opinion is rather optimal and
allows a hydrogen level of <1 µg/g to be reached. Annealing at 800°C has some advantages: the
degassing of the oven is rather fast, and the niobium becomes completely recrystallized at this
temperature. In addition, the lattice presents fewer defects after such annealing. The improved lattice
is not so sensitive to further hydrogen contamination.
2.3.2 Solid-state gettering

One needs a very high temperature and very low partial pressures in order to degas the oxygen and
nitrogen [10]. Such annealing does not make much sense for cavity application. During
recrystallization annealing at 750–800°C and with a conventional vacuum of ~10–6 mbar, some surface
contamination by O and N takes place [10]. The contamination depth can be estimated on the basis of
the diffusion law [23] and, as shown in Fig. 7, its penetration is about 150–200 µm.
Fig. 7: The calculated penetration depth of oxygen into Nb during recrystallization annealing [7]
This contamination can be removed by purification heat treatment, often called solid-state
gettering (see, e.g., Ref. [24]). The getter metal Me (usually Ti) is vapour deposited on the surface of
the niobium at high temperature. The bonding enthalpies of this metal to the interstitial impurities,
such as oxygen, nitrogen, or carbon, should be higher than that of Nb. The creation of the MexO, MexN
177
W. S INGER
and MexC compounds reduces the concentration of the interstitial impurities on the surface of the Nb.
Moreover, the Me protects against the absorption of the residual gas from the furnace environment. On
the other hand, the high temperature intensifies the diffusion of the interstitial impurities from the
inside to the surface, and as a result also allows purification of the bulk of the niobium. This procedure
has another positive effect for cavity performance: the annealing itself, at high temperature,
additionally homogenizes the niobium (dissolves small segregations of different types; for example,
residues of oxides, clusters of foreign materials, etc.). The temperature and duration of the purification
annealing depend on the evaporation rate of the getter material and the diffusion rate of the impurities.
This technique is, in principle, capable of improving the RRR by a factor of 10.
Several metals, such as Ti, Y, Zr, and Hf, that have higher bonding enthalpies with oxygen,
nitrogen, and carbon than niobium, can be used successfully for niobium post-purification [1, 2, 24].
Titanium has been applied for solid-state gettering of FLASH cavities at DESY, with annealing
parameters of 1400°C for 4 h (or a combination of 1400°C for 1 h plus 1350°C for 3 h). The RRR of
nine cell resonators after gettering normally reaches values of 500–600 [7, 23].
Post-purification with Ti increases the RRR, while as described above, common annealing at the
same temperatures without Ti reduces it (Fig. 8(a) and (b)) [23]. A definite distribution of the
interstitial impurities in the cross-section of the cavity wall has to be expected after such purification.
The calculated concentration of O, N, and C decreases from the centre to the surface, as shown in
Fig. 9 [23]. Similar RRR behaviour is to be expected. A rough method to estimate the RRR
distribution of a purified sample was proposed in [23]. A niobium sample after purification heat
treatment was etched layer by layer, and the RRR was measured after each step. The RRR distribution
of such a sample is shown in Fig. 10. It can clearly be seen that the RRR close to the surface is much
higher than that inside the sample, which is in good correlation with the distribution of the interstitial
impurities in the cross-section of niobium after post-purification (Fig. 9).
Fig. 8: The dependence of RRR on the annealing temperature for Nb annealed (a) with Ti and (b) without Ti
Post-purification potentially allows higher gradients to be achieved. Nevertheless, it has two
drawbacks that narrow the range of application of this procedure. First, it reduces the cavity stiffness.
The stress–strain curves [7] show that raising the annealing temperature from 800ºC to 1400ºC reduces
the elongation at break, the yield and the tensile strength. Secondly, titanium diffuses into niobium
during post-purification. It has been shown for FLASH cavities that the Ti diffusion depth is about
60 µm and is deeper in grain boundaries than in grains [7]. This polluted layer of Ti compounds must
be removed (as a rule, chemically).
178
Fig. 9: The oxygen distribution from the centre to the surface of an Nb sheet after post-purification
Fig. 10: The RRR distribution inside an Nb sheet after post-purification
179
W. S INGER
2.4 Specifications of high-purity niobium

The requirements of high-purity niobium for high-gradient superconducting cavities are listed in
Table 2.
Table 2: Technical specifications for niobium applied for the fabrication of European XFEL 1.3 GHz
superconducting cavities [22].
Electrical and mechanical properties The contents of the main impurities in

µg/g
Residual Resistivity Ratio, > 300 Ta ≤ 500 H ≤2
RRR
Grain size ≈ 50 µm W ≤ 50 O ≤ 10
–
Yield strength, Rp 0.2 50 < Rp 0.2 < 100 N·mm Mo ≤ 50 N ≤ 10
2
Tensile strength >140 N·mm–2 Ti ≤ 50 C ≤ 10

Elongation at fracture > 30% Fe ≤ 30
Vickers hardness, HV10 ≤ 60 Ni ≤ 30
A common indicator of purity, the Residual Resistivity Ratio (RRR), is chosen as RRR > 300.
The main interstitially dissolved impurities that act as scattering centres for unpaired electrons
and reduce the RRR and the thermal conductivity are oxygen, nitrogen, hydrogen, and carbon. Oxygen
is dominant due to the high affinity of Nb with oxygen. The concentrations of O, N, and C should be
kept below 10 µg/g. The influence of hydrogen on the RRR is not so significant, but the hydrogen
content should be kept low (less than 2 µg/g) in order to prevent hydride precipitation and degradation
of the Q-value of the high-RRR cavities under certain cool-down conditions (hydrogen Q decease [1–
3, 5]).
Among the metallic impurities, tantalum has the highest concentration (~500 µg/g). As
described above, this element accompanies niobium in most ores. An impurity level of 500 µg/g is
normally harmless for cavity performance [25], since tantalum is a substitutional impurity and does
not substantially affect the behaviour of niobium. Next in abundance among the substitutional
impurities are metals such as tungsten, titanium, molybdenum, iron, and nickel, usually at levels less
than (30–50) µg/g.
The method of fabrication of Nb sheets at the Tokyo Denkai Company is shown in Fig. 11 as an
example [26]. The sheets should be free of defects (foreign material inclusions or cracks and
laminations) that could initiate a thermal breakdown. Intermediate and final recrystallization annealing
for 1–2 h at 700–800°C in a vacuum furnace at a pressure of 10–5–10–6 mbar has to be performed in
order to reach full recrystallization, a uniformly small grain and sufficient mechanical properties for
subsequent cavity production (Fig. 12(a)). These conditions can be reached by ensuring a high degree
of deformation (> 65%) homogeneously distributed in the bulk of the Nb sheet before annealing
[12, 27]. When the deformation is not uniform, striped patterns become visible in the grain structure
(Fig. 12(b)). Such sheets have a reduced formability by deep drawing and an increased failure rate.
180
Fig. 11: The fabrication of Nb sheets at the Tokyo Denkai Company [26]
Fig. 12: Uniform and homogeneous grains (a) and non-uniform grain structure (b)
Acceptance tests of the material in industry include the RRR measurement, microstructure
analysis, analysis of interstitial and metallic impurities, hardness measurement, tensile tests, and
examination of surface roughness [22].
Stringent vacuum requirements during EBM and recrystallization heating, and well-controlled
conditions during forging, rolling, machining, and grinding, allow the contents of the interstitial
impurities in the niobium to be kept at acceptably low levels to ensure the high performance of RF
cavities.
2.5 Analytics and Quality Control (QC) of niobium

As mentioned above, significant progress in the production of high-purity niobium with an RRR up to
500 has been achieved during the past 20 years on an industrial scale. For example, ~20 tons of Nb
with RRR > 300, required for the European XFEL, was produced within 2 years. This progress would
be unimaginable without the availability of the appropriate analytical methods and devices.
Three different groups of impurities, easily evaporating metallic elements (substitutional
impurities: Al, Fe, Cr, etc.), non-easily evaporating metallic elements (substitutional impurities:
181
W. S INGER
Ta, W), and non-metallic elements (interstitial impurities: H, C, N, O) influence the properties of high-
purity niobium. In particular, the content of gases and carbon, and niobium–gas reactions, play a
remarkable role during material purification, cavity fabrication, and treatment. Of central importance
is the quantitative detection of very low concentrations of gases and carbon at the µg/g level in
niobium and the data of their local distribution. A short overview of the analytical determination of
these impurities is given below.
2.5.1 Interstitial and substitutional impurities
2.5.1.1 The RRR and thermal conductivity

The RRR is a common indicator of the level of purity. RRR values do not allow a resolution of the
influence of certain elements on electron scattering, but indicate material purity with high sensitivity.
This is why most laboratories and companies working with high-purity niobium use the RRR as the
first criterion of the level of purity, which can be measured easily and quickly [28].
As is well known from the electron theory of metals, in general cases the electrical resistivity of
metals (Mathiessen’s rule) at low temperature can be described in the following terms [27]:
ρ= ρ res + ρ ph (T ) + ρ m , (1)
where the first term is the residual resistivity at T = 0 K, caused mainly by electron–impurity scattering
and scattering on lattice defects ( ρ= res ρimp + ρdef ); the second term in Eq. (1) represents the
electron–phonon scattering, and ρ m is responsible for the increase of resistivity in a magnetic field.
Scattering of conduction electrons on the lattice (phonon scattering) is absent at T = 0 K due to

the zero fluctuations of the atoms in the lattice. The ρdef term plays an important role on work-
hardened niobium and can reduce the RRR by a factor of up to two. For recrystallized niobium, the
ρdef contribution is small. In this case, the total resistivity of Eq. (1) in the absence of a magnetic field
consists only of electron–impurity scattering and electron–phonon scattering contributions.
Resistivity caused by scattering of conduction electrons on homogeneously distributed
‘chemical’ defects (foreign atoms) is proportional to their concentration. The influence of the most
important impurity atoms, such as O, N, H, C, Ta, and Zr, on electron–impurity scattering is analysed
in some of the reference works [29]. The most popular is the data of Ref. [9], determined on the basis
of resistance measurements on niobium voluntarily contaminated by impurities:
∆ρi
=ρ ρ ph (T ) + ∑ Ci . (2)
∆Ci
The resistance coefficients ∆ρi / ∆Ci in Eq. (2) are given in Table 3.
Table 3: Residual resistance coefficients of different elements [9]
Impurity atoms Δρi/ΔCi

O 2.64
N 3.49
C 3.33
Ta 0.12
Zr 0.6
182
In absolutely pure metals that have a lattice without structural defects at temperatures close to
0 K, the resistivity tends to be zero.
The RRR is defined as the following ratio:
ρ (300 K)
RRR = ,
ρ (4.2 K)
where ρ(300 K) and ρ(4.2 K) are the resistivity of Nb at room and liquid helium temperatures,
respectively, at standard atmospheric pressure [30]. The superconducting behaviour of Nb (below
TC = 9.3 K, ρ (4.2 K) = 0) has to be taken into account for RRR determination. The electrical
resistivity at 4.2 K is obtained by extrapolation from the value when the material is in a non-
superconducting state (Fig. 13) or is obtained due to suppressing the superconducting state of Nb at
4.2 K by application of a high magnetic field. Therefore, in practical implementation, special
approaches lead to different measurement methods of RRR for Nb [28, 30].
700
RRR=12
600
500
R (arb.unit)
400
RRR=180
300
RRR=400
200
100
0
4 6 8 10 12 T (K)
Fig. 13: The typical low-temperature behaviour of the electrical resistivity for Nb of different grades of purity
A high thermal conductivity in the cavity wall is needed (at least 10 W·m–1·K–1 at 2 K) to guide
the dissipated Radio-Frequency (RF) power to the liquid helium coolant. For bulk niobium cavities,
this requires niobium of high purity [31].
To measure the thermal conductivity, the steady-state method is usually used. The conductivity,
k, is given by the formula
dT
Q = − kA .
dx
By measuring the temperature difference dT over a distance dx, with heat input Q and sample
cross-section A, the thermal conductivity k can be easily derived.
Some curves of the temperature dependence of the thermal conductivity for Nb with RRR
between 120 and 760 are shown in Fig. 14. The strong dependence of the thermal conductivity on the
purity of the Nb (RRR) can be seen. The rule of thumb giving the simplified relationship between the
183
W. S INGER
thermal conductivity and the RRR at liquid helium temperatures is sometimes useful for practical
applications [1, 2]:
λ (4.2 K) = C[W ⋅ m −1 ⋅ K −1 ] × RRR,

C ≈ 0.25 − 0.14.
The following empirical formula can be used to estimate the thermal conductivity of
superconducting niobium more precisely, over a wide temperature range [32]:
−1 −1
 ρ 295 K   1 1 
λ (T=
, RRR, G ) R ( y )  + aT 2  +  + … (3)
 L × RRR × T   D[exp( y )]T
2
BGT 3 
The first term in Eq. (3) describes the scattering of electrons by impurities, lattice defects, and
phonons; while the second term describes the scattering of phonons by the electrons and the grain
boundaries.
1000
Nb Thermal Conductivity, W/m K
100 RRR120
RRR270
RRR400
RRR525
10 RRR760
1
1 10
Temperature, K
Fig. 14: The typical behaviour of the thermal conductivity of polycrystalline niobium at low temperature
The best-fit values for the parameters are to be found in Ref. [32] and are listed in Table 4.
Table 4: The parameters for calculation of the thermal conductivity
Parameter Best-fit value

L 2.05 × 10–8 W·K–2
A 7.52 × 10–7 m·W–1K–1
α 1.53
B 4.34 × 103 W·m–2K–4
1/D 2.34 × 102 mK3·W–1
184
2.5.1.2 Oxygen, nitrogen, carbon, and hydrogen determination

The classical methods for O, N, C, and H determination consist of extraction of these impurities from
niobium in the form of gases, subsequent separation, and determination of quantities. Mostly, vacuum
fusion extraction, inert gas fusion extraction, or hot vacuum extraction [10, 33] are used.
In vacuum fusion extraction applied for O, N, and H, a metal sample of defined weight is
melted in a vacuum inside a crucible. The crucible is often graphite, so that O is removed as CO. In
inert gas fusion, a carrier gas – Ar, for example – is used to sweep out the extracted gas. In hot vacuum
extraction, a solid sample is heated in an ultra-high vacuum until the gases completely leave the
material.
The released gases are measured by the increase of pressure using a thermal conductivity cell or
other pressure-sensitive sensors. In the case of CO, the gas is converted to CO2 by passing over a hot
CuO catalyst and the quantity of CO2 is measured by the pressure change or by infra-red absorption.
The C content of Nb can be similarly determined in a fusion apparatus, using a flowing stream of O2 to
convert C into CO.
For the determination of oxygen, vacuum extraction using the platinum-flux sandwich
technique is recommended. An etched sample has to be packed into platinum foil that has previously
been degassed in purified helium. The platinum/niobium ratio should be about 10:1.
Industrial devices for O, N, C, and H determination are available on the market: well-known
companies are LECO and Horiba. The devices can guarantee 8–10% accuracy [(0.1–0.3) µg/g in
absolute values] for hydrogen and 15–25% accuracy [(0.5–2) µg/g] for nitrogen and oxygen.
For the analysis of metallic impurities at levels below (5–50) µg/g, a number of special
techniques based on spark mass spectrometry, proton or neutron activation and separation procedures
are available. For routine analysis, the main techniques used are Optical Emission Spectrometry
(OES), Atomic Absorption Spectrophotometry (AAS), Atomic Emission Spectrometry (AES), Mass
Spectroscopy (MS), and X-ray fluorescence analysis.
For strong durable production of high-purity niobium, the industrial laboratory should be
equipped with facilities for RRR measurement, scanning apparatus, interstitial impurity analysis (H,
N, O, and C), analysis of metallic impurities (Ta, W, and another refractory metals, such as Fe and
Ni), metallography, tensile testing, and testing for hardness, HV, and surface quality.
Currently, the companies that are in a position to produce high-purity niobium for
superconducting accelerators (five companies worldwide) are mainly equipped with devices for these
purposes. For additional investigation that is sometimes required, it is very desirable to have access to
analytical equipment for the following: thermal conductivity testing, neutron Activation Analysis
(NAA), radiation fluorescence analysis, SEM, EDX, SIMS, XPS, AUGER, X-ray radiography,
neutron radiography, texture analysis, and bulge testing (see also Refs. [1–3]).
2.5.2 Local defects in the sheets for half-cells

Experiments show that a small spot in just one of the sheets for half-cells can damage the performance
of the complete cavity [34]. Thus total quality control of Nb sheets regarding foreign material
inclusions and other types of local defects such as cracks, pores, and delamination is very desirable for
stable cavity performance.
185
Table 5: Defect diagnostics in Nb sheets
Method Principle of the method Penetration depth Resolution Destructive or Test time for a sheet Remarks
non-destructive measuring 265 × 265 × 2.8
mm
X-Ray radiography Difference in X-ray Complete Depends on the Can be non- About 30 min The ‘shadow picture’ depends on the
absorption between defect size of component destructive difference in density and atom number
and Nb (10 µm – 1 mm)
Neutron radiography Difference in neutron Complete Depends on the Can be non- About 60 min The ‘shadow picture’ depends on the
absorption between defects size of component destructive special quality of isotopes, and good
and Nb (10 µm – 1 mm) detection of light elements
Ultrasonic scanning Reflection of sound waves Complete (needs a Up to 50 µm Non-destructive About 30 min Non-homogeneity in metals is difficult
at interface coupling) to detect
Eddy current scanning Electromagnetic induction Depends on frequency Up to 100 µm Non-destructive About 30 min
(from micrometres to
millimetres)
SQUID scanning The Josephson effect Complete Up to 30 µm Non-destructive About 30 min SQUID scanning devices are not
186
available on the industrial level
Neutron activation Irradiation with thermal Complete Detection of Non-destructive About 15 h Efficient for tantalum inclusions, some
analysis neutrons, measurement of clusters with sizes parts per million of Ta in Nb can be
γ-spectrum up to 100 µm detected
Synchrotron Excitation by white beam 1 µm – 100 µm Up to 1 µm Can be done Some hours for inspection of K-lines (energy about 0–80 keV),
fluorescence analysis (analysis of fluorescence non- a 20 × 20 mm area sensitivity up to a few µg/g of impurity
(SYRFA) energy spectrum) destructively content
Synchrotron Energy selection in the About 10 µm An area of 12 × 12 Can be done 5 h for inspection of a 150 × L-lines (energy about 0–10 keV),
fluorescence analysis primary beam, observation mm can be tested non- 100 mm area sensitivity up to a few µg/g of impurity
(XAFS) of the absorption edge in one step destructively content
Microhardness testing Intrusion of the diamond Depends on load value About 50 µm Conditionally Residual marks of few Sensitivity depends on the differences
pyramid non- destructive micrometres that will be in hardness between Nb and inclusion
removed during subsequent
cavity preparation
W. S INGER
The usual conventional surface checks of Nb sheets for cavities are visual inspection,
anodization [35] and looking for discolouration, water soaking, and rust traces. It should be taken into
account that removal of 100–200 µm thick layers will take place later during cavity preparation, so
that inner defects located close to the surface will become uncovered. This means that quality control
should be carried out both on the surface and inside the Nb, at a depth of up to 500 µm close to the
surface area.
Some important requirements for quality control are as follows.
− It should be non-destructive.
− It should be total. At least one side of the sheet should be scanned. The penetration depth of
the signal should not be less than 0.3–0.5 mm.
− It should be fast. The scanning time for one sheet should not exceed 0.5 h.
− It should have a high resolution; defects with a size of 100–500 µm should be detectable.
− It should have sufficient sensitivity to elements with small differences of properties
compared to pure Nb (e.g. tantalum).
An overview of the advantages and disadvantages of different quality control methods is given
in Table 5.
Up to now, only eddy current scanning has been successfully applied on a large scale for quality
control of Nb sheets. Modern eddy current facilities can scan large areas at a rather high speed. High
resolution can be achieved by optimizing the electrical and mechanical parameters of the probe. It is
possible to detect defects of a size in excess of 100 μm at a depth of a few hundred micrometres.
According to simulations of the thermal breakdown (quench), carried out in Ref. [36], a local defect
with a size of ~100 µm in high-purity niobium with RRR > 300 caused the quench at Eacc to be close
to 25 MV·m–1. In this case, the requirements of the European XFEL for an accelerating gradient
Eacc > 23.6 MV·m–1 could be fulfilled.
An eddy current scanning device for defect diagnosis of niobium sheets has been successfully
developed and applied for the European XFEL at DESY. About 16 000 Nb sheets for the European
XFEL have been scanned, and sheets with detected flaws have been sorted out. A typical example of
foreign material inclusion (Ta), probably imbedded into an Nb sheet during rolling, can be seen in
Fig. 15. Eddy current scanning shows a remarkable local peak, the 3D microscope image indicates a
protrusion, and the results of non-destructive element analysis suggested a tantalum inclusion.
Fig. 15: An example of foreign material inclusion (Ta) detected in Nb sheets: (a) eddy current scan; (b) a 3D
microscope image; (c) the results of non-destructive element analysis.
187
W. S INGER
Further improvement of the scanning system in order to detect smaller defects in niobium can
be done using SQUID-based methods. SQUID sensors are more sensitive in comparison to
conventional eddy current pick-up coils [37]. Unfortunately, the industrialization of these devices for
this specific purpose has not yet been realized.
3 Mechanical fabrication of elliptical cavities

The conventional fabrication procedure consists of the deep drawing and Electron-Beam Welding
(EBW) of the parts into a cavity assembly [22]. This procedure is well established and has been used
in industrial fabrication for about 30 years.
3.1 Deep drawing and EBW

Half-cells are produced from niobium discs, pressed into shape using male and female dies. The dies
are usually fabricated from an anodized aluminium alloy. Special hard bronze AMPCO is also suitably
applicable to niobium. Deep drawing of half-cells requires high accuracy, because a resonant
frequency is generated in this manufacturing step. If the wall thickness deviations in the sheets are less
than ±0.1 mm, the use of stiff forming tools only is advisable. Application of an elastic polymer for
the female part of the tool allows stable shaping of the inner side of the half-cell independently from
the sheet’s thickness tolerances. To avoid surface damage, the stamp has to be free from dents,
blisters, and scratches.
While establishing the form of the deep drawing tooling, the spring-back of the niobium sheet
material has to be taken into consideration. Deep drawing is sensitive to the mechanical properties of
niobium. If the grains are too large, an ‘orange peeling’ roughening effect occurs. If the material is not
completely recrystallized, it tears during deep drawing, or deviation from roundness of the half-cell
will take place. To achieve good mechanical properties for high-RRR Nb requires the proper choice of
annealing temperature and time. The final annealing of 2.8 mm thick Nb sheets normally occurs at
750–800°C in a vacuum oven, at a pressure of 10–5–10–6 mbar for 1–2 h. Niobium also has a low
degree of work hardening, which is advantageous for mechanical forming. In almost all cases when
the proper mechanical properties of the sheet are achieved, cavity parts can be deep drawn to final
shape without intermediate annealing. The accuracy of the half-cell is controlled by tactile 3D
measurement and by sandwiching the half-cell between two niobium plates and measuring the
resonance frequency.
Two half-cells are joined at the iris with an EB weld, to form a dumb-bell. The iris EBW is
usually done partially from the outside and partially from the inside. The next step is the welding of
the stiffening ring. Weld shrinkage may lead to a slight deformation of the cell, which needs to be
corrected. Afterwards, frequency measurements are made on the dumb-bells to determine the amount
of trimming at the equators.
The dumb-bells are visually inspected. Defects and foreign material imprints from previous
fabrication steps are removed by grinding. In modern EBW, the iris welds are very smooth, but it is
still good practice also to perform the extra step of grinding the iris weld in order to ensure a smooth
inner surface in this high-electric-field region and to avoid geometric field enhancement.
Beam tubes are either purchased as extruded seamless tubes or rolled from sheets and EB
welded [38]. Flanges for the beam tubes are machined. The flanges are located in regions of low
magnetic field and can be made from an Nb−Ti alloy (~55% Ti).
After proper cleaning, eight dumb-bells and two end group sections are assembled in a precise
fixture to carry out the equator welding, which is done from the outside. Manufacturers’ experience
has shown that all equator welds can be done in one action (one evacuation of the EB chamber),
without deterioration of the quality.
188
EBW is usually done in several steps (for more details, see Ref. [22]). The weld parameters are
chosen to achieve full penetration. A slightly defocused beam in a circular or elliptic pattern (see
Fig. 16), and the use of ~50% of the beam power during the first weld pass and 100% of the beam
power in the second pass, allows a smooth weld seam to be obtained. The cavity welding parameters
should be adapted to each EBW machine individually. The welding parameters of the DESY EBW
machine are listed in Table 6.
Fig. 16: A schematic of the EBW of niobium sheets: 1, the electron beam (Р0, power of beam; r, radius of
slightly defocused beam on the surface; L, scanning amplitude; V, beam velocity); 2, Nb sheets; 3, melting zone
(z, depth of the melting zone).
Welding from the inside (RF-side) is recommended, wherever possible. Welds at the equator
and iris of cells and at the high-order mode HOM coupler parts are especially critical, because they
will be exposed to high magnetic or electric fields. Therefore, thorough cleaning by ultrasonic
degreasing, chemical etching, ultrapure water rinsing, and cleanroom drying is mandatory – clean
conditions must be assured during welding. Touching the weld preparation area after the last cleaning
must be strictly avoided.
The required high quality of the welding seam is in fact very important for performance. This
has been demonstrated many times. A direct verification was done, for example, in Ref. [39], due to
the investigation of samples separated from the quench areas of cavities with low performance.
Experience shows that sometimes holes can be burned through by EBW. The frequently asked
question is whether such a hole is repairable. In Ref. [40], it was shown that the reparation of the
burned hole is critical from the point of view of gas contamination. A definite area around the hole is
contaminated by oxygen, nitrogen, and hydrogen. It should be removed and replaced by a disc of high-
purity Nb before new EBW can be done. Some companies have recently developed a burned hole
repair procedure and demonstrated that accelerating gradients up to 30 MV·m–1 are reachable in a
repaired cavity.
189
W. S INGER
Table 6: Welding parameters for the single-cell cavities of the DESY EBW machine [41]
Parameter Iris Equator
Beam direction From From

From outside
inside outside
Maximal duration in ambient air between etching and
welding (h) 8 8 8
Percentage of current required for full penetration – 68 – 56 50 100

Angle (degrees) – 45 – 0 0 0
Accelerating voltage (kV) – 150 – 150 150 150
EB current (mA) – 8 – 7 7 13
Beam pattern – Circle – Circle Circle Circle
Frequency (Hz) – 4000 – 4000 4000 4000
Amplitude (mm) – 1.5 – 1.5 1.5 1.5
Focus position set (Δ) – –25 – –25 –45 –45
Welding velocity (mm·s–1) – 6 – 6 6 6
Slope up (angle, degrees) – 5 – 10 5 5
Overlap (angle, degrees) – 5 – 5 5 5
Slop down (angle, degrees) – 25 – 30 25 25
Pressure in working chamber (mbar) < 1 × 10–6 < 1 × 10–6
Cool-down time (min) – 120 – 120 – 120
3.2 Purity degradation during EBW

Electron-beam welding is a rapid process; the temperature goes up and down quickly. The temperature
exceeds the melting point of niobium at the welding seam itself, but the temperature in the heat-
affected zone depends on the welding parameters, the material thickness, the thermal conductivity of
the material, and so on. The process of absorption and desorption of gases during EBW is very
complex, and in any case is not in equilibrium. It is in principle possible, but difficult, to make precise
calculations of the resulting RRR and the RRR distribution in the welding seam area, but the
appropriate data are still lacking. In spite of that, several reasonable experimental results are available.
Since niobium is a strong getter material for interstitial impurities, it is important to carry out
the EB welds in a sufficiently good vacuum. The RRR distribution and RRR dependence on pressure
in the EB-welded samples from different companies – Dornier, Ettore Zanon, and ACCEL (now RI) –
show that the RRR degrades by up to 35% if the total pressure is in the range of 10–3–10–4 mbar. For a
pressure between 5 × 10–5 and 10–5 mbar, the RRR degradation is smaller (about 10%) [40]. Tests have
shown that RRR300 niobium suffers less than 10% RRR degradation in welding at a pressure lower
than 5 × 10–5 mbar. Such a vacuum is achievable with industrial equipment and is recommended for
the fabrication of superconducting cavities [22]. It is worthwhile to check the RRR degradation of
high-purity Nb in the EBW chamber with an ultra-high vacuum (total pressure < 10–5 mbar). The
investigation in Ref. [40] shows that for the welding seam itself (Fig. 17), the RRR degradation started
at a pressure in excess of ~5 × 10–6 mbar. At a pressure below 5 × 10–6 mbar, the RRR in the welding
seam even improves, from ~350 to 370–380. The improvement of the RRR is maximal for the
pressure region from 10–8 to 5 × 10–7 mbar. It is interesting that the improvement in the RRR almost
190
does not depend on the pressure in this region. This means that the effort to reach a vacuum better than
5 × 10–7 mbar does make sense, at least for the above-mentioned welding parameters.
Fig. 17: The RRR in the welding seam versus pressure
On the other hand, it is easy to realize that RRR degradation takes place in the heat-affected area
(see Fig. 18). The RRR degradation is maximal in the region ~10–15 mm away from the welding seam
(from 350 to 320–330).
380
2.3E-8 mbar
370 2.0E-6 mbar
2.3E-7 mbar
360
6.5E-8 mbar
350 Reference
RRR
340
330
320
310
-25 -20 -15 -10 -5 0 5 10 15 20 25
position (mm)
Fig. 18: The RRR for the EBW area versus distance from the welding seam at different pressures (DESY EBW
facility).
3.3 The RRR and the hydrogen distribution in the weld area
The hydrogen content measured by heat extraction in the welding seam as well as in the heat-affected
zone [40] shows a good correlation with the RRR, as can be seen in Fig. 19. Reduction of the RRR
follows with enhancement of the hydrogen content. Evidently, hydrogen absorption does not take
place directly in the welding seam, but in the areas with rather moderate temperatures (the heat-
affected zone). It is well known, from pressure-concentration isotherms of hydrogen for Nb in a
steady-state condition (Fig. 20), that at temperatures above 500 K and under the common partial
pressures of hydrogen, the hydrogen content in Nb becomes less than 1 µg/g. At lower temperatures,
absorption of hydrogen will take place according to Fig. 20. Obviously, the heat-affected area fulfils
these last conditions, which then result in an enhanced hydrogen content.
191
W. S INGER
Fig. 19: The RRR distribution compared with the hydrogen content in the weld area (pressure 2.3 × 10–8 mbar)
Fig. 20: Pressure-concentration isotherms for hydrogen in Nb in the steady-state condition [10]
An enhanced hydrogen content at the weld area is in good agreement with the results of
Refs. [42, 43], where the hydrogen disease of Nb cavities was explored. It was pointed out that the
weld area is susceptible to hydrogen disease even if the chemical treatment of the cavity was carried
out correctly. This illustrates why the annealing of cavities at 800°C during treatment (hydrogen
degassing) is reasonable.
192
An example of Rest Gas Analysis (RGA) during welding is shown in Fig. 21 [44]. The start of
welding is normally accompanied by a significant rise of the hydrogen partial pressure in the chamber.
The partial pressure of the water decreases at the same time, as can be seen in Fig. 21. Probably, the
electron beam brings about the decomposition of water molecules, which produces a lot of hydrogen.
Fig. 21: An example of the partial pressure in the EB chamber during the welding of a Nb 300 sample
3.4 Cavity length adjustment procedure

The main principle of the length adjustment procedure follows Ref. [45]. Uncertainty in the shape of
half-cells has to be corrected before completion of the final cavity. The correction should fulfil the aim
that the cavity finally has the right frequency and the right length. The procedure takes advantage of
the fact that the frequency change due to a shape change is different for regions with a high magnetic
field (equator) and a high electric field (iris). A trimming of the half-cell near the equator (∆Le)
increases the frequency; a compression of the half-cell in the iris region (∆Lz) decreases the frequency.
The frequency-sensitivity factors are listed in Table 7 [45].
Table 7: Sensitivity factors
Equator,Length,
∆Le ∆Lz
δf [MHz·mm–1] –5.3 5.4
The length adjustment method consists of the following steps.
1. Frequency measurement of the half-cells.
2. Dumb-bell assembly (welding of iris and stiffening rings).
3. Frequency measurement of the dumb-bell.
4. Trimming of the dumb-bell at the equator, if necessary.
5. Positioning of the dumb-bell in the cavity.
6. Cavity production (equator welding of dumb-bells and end groups).
7. Frequency and length measurement of the cavity.
8. Calculation of the expected cavity length after tuning.
193
W. S INGER
The individual steps of the length adjustment procedure for the half-cell, dumb-bell, end group,
and complete cavity are described in detail in the specification for the production of European XFEL
cavities [22]. Descriptions of the quality-assurance and other documentation can also be found in those
documents.
3.5 Quality management and documentation

Some main principles of quality management that have to be kept in mind during cavity production are
mentioned below [22, 46].
The contractor (the cavity manufacturer) and all its subcontractors have to create and maintain a
Quality Management (QM) system according to DIN ISO 9001. Their quality control system has to be
independent of the manufacturer. The list of subcontractors must include evidence that each
subcontractor has a certified QM system according to DIN ISO 9001, or must contain information on
how quality is assured and audited there. The orderer (the institution placing the contract for cavity
production) may check a subcontractor’s qualifications by carrying out quality audits and holding
adjustment and determination discussions.
The orderer normally requests some quality checks including protocolling in addition to the
Quality Assurance/Quality Control (QA/QC) measures carried out by the contractor. Protocols must
be reported at a time close to the period of component and cavity fabrication.
For each completed cavity, a traceability report has to be prepared, in which the serial numbers
of all incorporated semi-finished products and manufacturing groups are stated in the form of a parts
list, including position information. A conclusion concerning the raw material used should be possible.
The manufacturer has to prepare a Conformity Certificate (CC, confirmation of conformity) for
each completed cavity. The CC confirms that the cavity has been produced according to specification
and that all requirements have been checked and that they comply with the orderer’s demands. A Non-
Conformity Report (NCR) must be prepared if the properties of a certain component deviate from
specification. In the NCR, the additional procedure has to be proposed by the contractor. The NCR has
to be accepted by the orderer.
The experience of European XFEL cavity production shows that use of the Engineering Data
Management System (EDMS) as a central repository for all engineering information and for the
complete documentation relating to the cavity fabrication process, and prompt exchange with the
manufacturer, are advisable [47, 48] (see Fig. 22). This implies that all documents (inspection sheets,
specifications, drawings, etc.) that are created and maintained during the RF cavity manufacturing
process should be made available electronically in the EDMS. The exact formats and implementation
of the test protocols have to be agreed on between the contractor and the orderer. The complete
information that must be reviewed by the manufacturer or approved by the orderer shall be controlled
by the system. Cavity producers have access only to relevant documents and data. For statistical
analysis, a database system, as is operational at DESY [49], is very convenient.
Fig. 22: An example of cavity data management using EDMS and a data bank
194
3.6 Initial experiences in the production of superconducting cavities for the European XFEL
[50]
The production of superconducting cavities for the European XFEL includes the following:
mechanical fabrication, the use of material provided by DESY, prior EP treatment, ethanol rinsing,
outside etching, 800°C annealing, tuning to resonant frequency, final surface treatment by Buffered
Chemical Polishing (BCP) or EP, High-Pressure Water (HPR) rinsing, 120°C baking, integration of
the cavity into a Helium Tank (HT), assembly of the HOM, pick-up and high-Q antennas, and
shipment to DESY for the vertical RF test.
The fabrication of 800 serial cavities, the largest in the history of cavity production, is going on
at two companies – Research Instruments (RI) in Germany and Ettore Zanon in Italy – on the ‘build to
print’ principle and is planned to be finished in 2015.
A new infrastructure, mostly for cavity treatment, is being created at both companies. The
infrastructure comprises EBW equipment, ISO 7 and ISO 4 cleanrooms with cleaning, rinsing, and
etching facilities, Ultra-Pure Water (UPW) production systems, HPR rinsing, 800°C annealing
furnaces, tools for cavity integration in the HT, a 120°C final baking oven, a Slow-Pumping Slow-
Venting (SPSV) vacuum system, and systems for visual inspection of the cavity’s internal surface.
DESY provided both companies with an in-house developed sophisticated machine for cavity
tuning at room temperature (Cavity Tuning Machine (CTM)) and equipment for RF measurement of
dumb-bells and end groups, called HAZEMEMA.
DESY and colleagues from INFN/LASA are monitoring the production process. The main
principle of the supervision of the production is that the cavities have to be built strictly according to
the XFEL specifications, but a performance guarantee is not required. The supervision consists of: a
quality control plan (also for the Pressure Equipment Directive, or PED – see below); the internal QA
and QM systems of the companies; NCRs; and regular visits to the company by DESY expert teams.
In addition, there are regular ‘Project Meetings’ at the company location (approximately every month,
depending on the production progress and quality).
Information flow from the companies to DESY/INFN is organized to be paperless, as described
above, utilizing the EDMS. Transfer of documents from data systems to EDMS is fully automated.
Another important issue for European XFEL cavity production is the implementation of the
Pressure Equipment Directive (PED). The fabrication experiences of the European XFEL cavity with a
helium tank as a pressure-bearing component according to European requirements are briefly
described below.
To avoid a pressure test on a complete cryo-module with eight cavities prepared for electron
acceleration and dressed with power couplers and other accessories, the following options (modules) B
and F are chosen for cavity testing according to European Directive PED/97/23/EC.
Module B (EC type examination) includes: examination of the design; FEM calculations;
qualification of welding processes; qualification of other PED-relevant processes (annealing, deep
drawing, forming); production and destructive examination of test pieces; supervision of production
qualifications on the first eight series cavities; and PED-relevant testing methods for series production
of the cavities.
Module F (product verification) is dedicated to series production and includes mainly visual
inspections, monitoring of documents, and a pressure test for each cavity. The contracted ‘notified
body’ (TUEV NORD) advises and tracks the process.
An important stage of this activity was the qualification of the test pieces. A test piece is
composed of two cells with a helium vessel, without end groups, representing all pressure-bearing
parts and welds of the cavity. It is built using exactly the same manufacturing methods and welding
195
W. S INGER
parameters that are used in the series production. Two test pieces per company were produced and
successfully qualified by destructive examination.
Semi-finished products (cavity material) for pressure-bearing sub-components of cavities and
helium tanks also have to be qualified and purchased, according to PED 97/23/EC, by companies
qualified for this task. These PED activities on material for cavities and helium tanks consist of
qualification of cavity materials Nb40, Nb300, Nb - Ti, Ti Grade 1, and Ti Grade 2 itself (creation of
the Particular Material Appraisal, or PMA); certification of the QM system at the companies
producing the cavity material and the sub-components for the helium tanks; and supervision of the
procurement of the semi-finished material products (traceability, marking, 3.1 test certificates, etc.).
Meanwhile, the first hundreds of series cavities have been produced and treated using the newly
qualified infrastructure and shipped to DESY for cold RF testing. Most cavities shipped up to the
present time have immediately satisfied the XFEL specifications.
4 Large-grain and single-crystal niobium
4.1 Introduction
The idea of using ingot material for cavity fabrication in the form of sheets sliced from an ingot and
producing cavities by means of deep drawing and EBW technology was introduced at Jefferson Lab
(JLab). It turned out that the sliced material had sufficiently good mechanical properties and could be
formed into half-cells despite the large grains. This manufacturing approach, for Large-Grain (LG)
cavities, has attracted worldwide interest in the past few years [51–54]. This option allows the long
production chain from large-grain ingot to fine-grain sheet to be eliminated, and seems to be more
cost-effective than the conventional fine-grain method. A lot of studies on material characterization,
cavity fabrications, and testing around the world have appeared since that time. An overview of these
investigations can be found in Ref. [51].
4.2 Material, fabrication, and RF performance

Up to now, industrial slicing of the discs has been developed by two companies: W.C. Heraeus in
Germany and the Tokyo Denkai Company in Japan [55–57].
One of the issues to be solved was efficient cutting of the discs. W.C. Heraeus have developed a
wire saw cutting procedure that can be cost-effective, assuming that the company can carry out the
slicing of many tens of discs simultaneously while keeping the material purity high (RRR > 300), and
achieving tight thickness tolerances (better than ±0.1 mm) and a high surface quality (Ra < 1.6). The
principle and the main part of the W.C. Heraeus machine for cutting of the large-grain ingot by wire
sawing can be seen in Fig. 23. The procedure developed at Tokyo Denkai is similar [56].
Fig. 23: The principle and the machine for cutting of the large-grain ingots by wire sawing at W.C. Heraeus.
(Courtesy of W.C. Heraeus.)
196
It can be expected that disc material from ingots will be less vulnerable to contamination by
foreign material or other types of defects, because it is taken directly from homogeneous ingot
material that has been very slowly re-melted many times. Several types of flaws (delamination,
imbedded particles, oxides, etc.) that can occur during forging or rolling are avoided in this case.
Experience has shown that the eddy current scanning usually applied to fine-grain sheets is useless
[52, 55].
The introduction of ingot niobium for SRF cavities has attracted the interest of many
institutions. Work on fabrication and testing of such cavities has been carried out at Jefferson Lab,
DESY, KEK, Michigan State University, Cornell University, Peking University, the Institute of High
Energy Physics of China, and BARC of India.
Investigations with single-cell cavities have shown that a rather high accelerating gradient Eacc
of usually 25–35 MV·m–1 can be achieved by rather simple BCP treatment with
HF(40%):HNO3(65%):H3PO4(85%) in the volume ratio 1:1:2. Especially comprehensive
investigations have been undertaken at JLab on single-cell cavities produced with material from
different companies, of different LG dimensions, with different tantalum contents (300–1200 µg/g),
and different shapes and resonant frequencies (for more details, see Ref. [51]). It has been shown that
even using BCP treatment, an accelerating gradient of 45 MV·m–1 is achievable. For high
performance, approximately 100 µm of material has to be removed from the surface when
conventional procedures such as BCP and hydrogen degassing are used.
In contrast to most laboratories, which have made use of their own fabrication and rapid
prototyping capabilities for exploring ingot niobium, DESY has collaborated closely with industry.
After encouraging results gained on the single-cell LG cavities, the main aim was to analyse the
potential of ingot material for large-scale applications such as the European XFEL. Eleven nine-cell
LG cavities of TESLA shape were produced at ACCEL Instruments GmbH – now Research
Instrument GmbH (RI) – from W.C. Heraeus material.
It was shown that it is feasible to build not only single-cell, but also nine-cell, cavities from LG
material without significant difficulties. Deep drawing of the half-cells was done using the same tools
as for fine-grain material. The grain boundaries were noticeably pronounced, with steps of up to
0.5 mm. The deep-drawn half-cells had a quadrangular or oval shape, and sometimes did not meet the
required tolerance of ±0.2 mm. It turned out that the presence of large central crystals in the discs and
their crystallographic orientation had a big influence on the shaping. It is well known that main atom
plane slipping for Body-Centred Cubic (BCC) metals takes place in the (110) planes in the [111]
direction; therefore, for the discs with (100) orientation, a more pronounced anisotropy and
quadrangular shape after deep drawing were expected, in agreement with observations.
The pronounced shape deviation in the half-cells generated some difficulties for assembly of
half-cells, and especially of dumb-bells for welding. RI overcomes these difficulties by using a special
tool that ensures precise joint assembly of the male and female half-cells. Application of the DESY
length adjustment procedure [45] allowed the correct cavity length and the required frequency of the
fundamental mode to be achieved in all of the nine-cell cavities.
After the BCP treatment (removal of a 100 µm surface layer inside and a 20 µm layer outside,
annealing at 800°C for 2 h, final BCP of 20 µm inside, followed by baking at 125–135°C for 48 h),
accelerating gradients up to 30 MV·m–1 in π-mode measurement (Bp = 110–130 mT) and up to
35 MV·m–1 in pass band measurement have been achieved in a stable and reproducible manner for all
11 nine-cell LG TESLA-shaped cavities [52].
After additional EP consisting of main EP of about 50–70 µm, removal followed by an ethanol
rinse, an additional standard 800°C firing, and a final EP consisting of ~50 µm removal, ethanol rinse,
six final HPR cycles, and baking at 120°C, the performance of the cavities was significantly improved,
up to 31–45.5 MV·m–1 at Q0 values above 1010, limited mostly by breakdown. Figure 24 shows the
197
W. S INGER
plot of the unloaded quality factor versus the acceleration gradient Q0(Eacc) at 2 K for these cavities.
Enhancement of the acceleration gradient, typically by more than 10 MV·m–1 after EP in comparison
to BCP was observed.
Fig. 24: The final Q0(Eacc) performance of the LG cavities AC112-AC114, and AC151-AC158 at 2 K after EP
Another interesting aspect of LG cavity behaviour that needs to be stressed is the rather high
unloaded quality factor, Q0, after EP treatment, even at high gradients. As can be seen in Fig. 25, Q0
approaches 3.5 × 1010 at 2 K, at moderate accelerating gradients. Figure 25 compares the Q0 values of
11 EP-treated LG cavities with the Q0 values of 15 XFEL prototype cavities (AC115–AC129) treated
according to an XFEL recipe [46]. As can be clearly seen, the Q0 value for the LG cavities is ~25–30%
larger than for conventional fine-grain cavities, which indicates the high potential of LG cavities in
applications requiring high Q0 values; for example, in Continuous Wave (CW) applications. The
superiority of the Q0 value of LG cavities after BCP treatment is less pronounced.
Fig. 25: A comparison of the unloaded quality factor Q0 at 2 K for 11 EP-treated LG cavities (red) with Q0 at
2 K for XFEL prototype cavities (AC115–AC129, blue, best result) treated according to an XFEL recipe (partly
Final EP and partly BCP Flash [24] were applied).
198
The complete chain of the LG cavity technique, beginning with material production and ending
with cavity installation into a cryo-module, has been successfully tested at DESY. Two LG cavities
are installed and operational in the FLASH accelerator at DESY at the present time. The first
European XFEL cryo-module made from LG cavities has been produced and tested, and it fulfils the
specifications requirements.
4.3 Advantages and disadvantages of LG cavities

Based on experience and on discussions of the past few years at workshops and conferences [51–54,
58], the advantages (pros) and disadvantages (cons) of LG cavities are summarized below.
Pros
1. LG discs are more cost-effective than fine-grain sheets (by 32%, according to the estimate for
XFEL pre-series cavities).
2. The wire saw procedure allows us to achieve high surface quality and thickness tolerances in the
discs.
3. The increased thermal conductivity close to 2 K due to the phonon effect helps us to lead the
heat away from hot spots.
4. Simplified quality control is possible. There is no danger that during the many production steps
from ingot to sheet, the material will be polluted (no RRR degradation). Eddy current scanning
is avoidable.
5. An accelerating gradient of 25–30 MV·m–1 can be reached by simple preparation with BCP
only. The best result of 45 MV·m–1, reached after EP, is one of the best results for this type of
cavity.
6. The onset of Q-drop in large-grain cavities is typically at 10% higher accelerating gradients.
7. It is sufficient to bake the BCP-treated cavity at 120°C for only 12–24 h.
8. The complete chain of the LG cavity technique, beginning with material production and ending
with cavity installation into a linear accelerator, is proven.
9. The quality factor Q0 of EP-treated LG cavities is ~25–30% larger compared to similar fine-
grain cavities.
10. The wire saw method causes much less stress at the surface of the disc compared to rolled
sheets (it reduces the influence of the damage layer on performance).
Cons
1. LG is currently not usable for mass production. For example, the industry is not in a position to
produce the required amount of ~20 tons of LG material for the European XFEL in 2–3 years.
2. Only one company has industrial experience of LG cavity production.
4.4 Single-crystal cavities

Why does EP treatment allow the attainment of better performance for LG cavities than BCP? The
surface quality of large grains is comparable for BCP and EP treatment. The surface roughness of the
BCP-treated large grains depends on the crystal orientation, but is on the same level as EP-treated
fine-grain material (hundreds of nanometres [52], and see Fig. 26).
199
W. S INGER
Fig. 26: (a) A light microscope image of the grain boundary triple junction of LG Nb after 100 µm BCP on the
previously mechanically polished sample: the orientation type is shown on the grains. (b) An atom force
microscope image of the same area: the steps on the grain boundaries are within of 1.5–15 µm for this
orientation constellation.
It seems that the difference in performance between BCP and EP treatment of LG cavities is
caused by the Grain Boundaries (GBs). The reduced Eacc of BCP-treated cavities can be explained, for
example, by the magnetic field enhancement at grain boundary edges and earlier penetration of the
external magnetic field in the niobium, which has a geometrical nature [1, 2]. In other words, GBs can
be considered as planar weak links with a reduced critical current density. In addition, GBs enhance
the possibility of hydrogen absorption and diffusion in the areas in which impurities gather [1–3].
From this point of view, it would be reasonable to take into consideration the behaviour of the
single-crystal cavities (cavities without grain boundaries). A single-crystal cavity with no grain
boundaries definitely has the potential to improve cavity performance and simplify the treatment
procedure substantially, because by nature it does not contain any interruption of the crystal lattice
orientation.
Is it possible to produce cavities consisting of a one single crystal? Indeed, a fabrication method
for single-crystal cavities has been proposed and a few single-cell single-crystal cavities have been
produced recently [59, 60]. The following aspects have been demonstrated on samples and taken into
consideration for the fabrication proposal (for more details, see Refs. [59, 60]).
− Definite enlargement of the single-crystal disc diameter is possible without destroying the
single-crystal structure.
− The single crystals retain the crystallographic structure and, after shaping of the cavity half-
cell from a disc by deep drawing, the orientation perpendicular to the surface remains.
− Appropriate heat treatment will not destroy the deformed single crystal.
− If the orientation of the crystals is matched, two single crystals will grow together by EBW.
Especially important for cavity fabrication by EBW are the last points, which allow production
of cavities as a complete single crystal. It has turned out that two single crystals will grow into one
single crystal if the crystallographic orientations are matched at the EBW seam with an accuracy
approaching 3°. The results of metallographic analyses of cross-sections of niobium samples can be
seen in Fig. 27(a) and (b). Whereas unmatched orientations produce a pronounced grain boundary
200
(Fig. 27(a)), matched orientations of both single crystals grow together without an interface (grain
boundary free) (Fig. 27(b)).
Fig. 27: (a) The EBW connection of two single crystals without matching of crystal orientation (the grain
boundary is pronounced). (b) The EBW connection of two single crystals after matching their orientations.
Several single-crystal cavities have been produced at JLab (of 2.3 and 2.8 GHz resonant
frequency, without enlargement of a single crystal) and at DESY (1.3 GHz, by applying the
enlargement procedure). It has been demonstrated that, for BCP-treated single-crystal cavities, a
similar performance as that after EP treatment of LG cavities, namely an accelerating gradient at the
level of 40 MV·m–1, is reachable. The main results for the performance of single-crystal cavities are
summarized in Table 8.
Table 8: A summary of the test results for the single-crystal single-cell cavities produced at JLab and DESY
Cavity number Eacc, max Bpeak, max Q0(Bpeak,max) Treatment

and resonant (MV·m–1) (mT)
frequency
1 (2.3 GHz) 38 162 4.0 × 109 200 µm BCP, 800°C/3 h, HPR,
120°C/48 h
2 (2.3 GHz) 45 160 7.0 × 109 200 µm BCP, 800°C/3 h, HPR,
120°C/24 h
1AC6 (1.3 GHz 41 177 1.2 × 1010 250 µm BCP, 750°C/2 h, 120 µm
EP, HPR,135°C/12 h
1AC8 (1.3 GHz) 38.9 168 1.8 × 1010 216 µm BCP, 600°C/10 h, HPR,
120°C/12 h
5 (2.8 GHz) 38.5 166 7.6 × 109 170 µm BCP, HPR 120°C/12 h
A single crystal applied up to now for cavity fabrication had a crystal orientation close to (100).
In this context, the question arises of how cavity performance can be influenced by crystal orientation.
This issue has not been investigated until now and is definitely relevant. Anisotropy of the
superconducting energetic gap 2Δ/Tc in the 4d and 5d transition metals, including Nb, is well known
[61]. As one of the main parameters of superconductivity, Δ should influence the main physical
properties of the material (e.g. the critical magnetic field Hc and the surface resistance). Therefore, it is
not unimaginable that the fabrication of single-crystal cavities with a preferred orientation could lead
to further improvement of performance.
201
W. S INGER
On the other hand, the enlargement procedure for a single-crystal disc is costly. Several attempts
have been made in the industry to produce, in a stable manner, rather large single-crystal discs with
diameters of 200–300 mm. Generally, the procedure for single-crystal creation is similar to the well-
known vertical Bridgman procedures for single-crystal growth; that is, partially melted seed, an axial
temperature gradient, and a movable interface between the solid and liquid phases are applicable.
Nevertheless, the industry is still not in a position to produce such single crystals of high-purity
niobium.
5 Fabrication of seamless cavities

The idea of fabricating seamless elliptical cavities is attractive. It is to be expected that this technology
could not only reduce the production costs, but also improve the cavity’s accelerating performance.
Some efforts have been made in the past through the application of explosive forming [62], spinning
[63], and hydroforming [64]. Explosive bonding was successfully applied to Cu tubes, but failed for
bulk niobium. Spinning was started in the 1990s at INFN Linearo, and up to now it has been
demonstrated that a single-cell spun cavity can reach an Eacc similar to that for a welded cavity up to
40 MV·m–1.
Most efforts in recent years have been focused on hydroforming, and this technique has been
applied to elliptical cavities from 300 MHz to 3.9 GHz in many laboratories: CERN, Cornell in the
1980s, CEA SACLAY in the 1990s, and KEK and MSU (Michigan State University). Many
laboratories now successfully produce cavities in copper. A hydroformed 1.3 GHz single-cell cavity
from CEA SACLAY, made from Nb of RRR = 20, has reached Eacc = 18 MV·m–1 after post-
purification [65].
In the 1990s, DESY started and established a highly consistent hydroforming programme for
TESLA-like cavities [66–68]. The DESY forming procedure consists of two stages: reduction of the
tube diameter in the iris area and subsequent expansion of the tube by hydroforming. One starts with a
tube of intermediate diameter between iris and equator. The hydroforming expansion generally
consists of three steps: determination of the stress–strain properties of the tube material, computer
simulation of the expansion, and the hydroforming itself.
The hydraulic two-dimensional bulging of the disc into the spherical form was used to produce
a stress–strain diagram. The values of the stress, σ, and the strain, ε, are determined by measurement
of the pressure, p, the radius of curvature, r, and the thickness, t, in the zenith of the sample during the
deformation procedure. The numerical simulation of the hydraulic expansion of the tube was done at
DESY using the finite element code ANSYS [68]. Non-linear elasto-plastic behaviour in accordance
with the stress–strain curve and isotropic hardening rules were taken into account. The calculations
were carried out on the basis of the experimentally determined stress–strain characteristic of the tubes
to be hydroformed, and resulted in a relationship between the applied internal pressure versus the axial
displacement with radius growth (the path of the expansion) for the hydroforming process.
It was found that the strain before necking of niobium can be increased by using a periodic
stress fluctuation (pulse regime). Tensile tests have shown that elongation before necking is almost
30% higher by application of a pulse method in comparison with a monotonous stress increase. In
addition, the strain rate during the hydraulic expansion of Nb should also be taken into account.
Experiments have shown that the deformation procedure should be rather slow. Around 10% of strain
before necking can be achieved by keeping the strain rate below 10–3 s–1.
During the actual forming process, an internal pressure is applied to the tube and simultaneously
an axial displacement, forming the tube into an external mould; deformation is controlled by a PC,
which follows the established simulation protocol. The main criterion for hydraulic expansion is the
theoretically determined relationship between the internal pressure and the axial displacement. Further
202
correction of the expansion parameters can be done on basis of comparison of the theoretical and
experimental growth of the tube diameter.
The development of the production of seamless Nb tubes for hydroforming was done in
collaboration with a number of companies. Tube production [69] by spinning, and back extrusion in
combination with flow forming and deep drawing, were checked. A uniform, small-grain, and
homogeneous texture is required to provide high plastic deformation of the tube during hydroforming.
The tubes produced from the Nb sheet (spinning or deep drawing, with subsequent flow forming)
show higher strain before the onset of necking compared to those produced by back extrusion
combined with flow forming. Several single-cell units of 1.3 GHz TESLA-shape niobium cavities
have been produced at DESY. Hydroformed single cells have reached an accelerating gradient Eacc
approaching 35 MV·m–1 after BCP, and approaching 42 MV·m–1 after EP (for an example, see
Fig. 28).
Fig. 28: The Eacc performance of hydroformed single-cell cavities
The work of the past few years has been concentrated on multicell and nine-cell cavities.
Several seamless two- and three-cell units have been produced. An accelerating gradient Eacc of 30–
35 MV·m–1 has been reached after BCP, and an Eacc approaching 40 MV·m–1 has been reached after
EP. Hydroformed three-cell units combined with nine-cell niobium cavities have been completed at
Ettore Zanon. The accelerating gradient reached was Eacc = 30–35 MV·m–1 (Fig. 29). One cavity has
been successfully integrated into a cryo-module and is operational in the FLASH accelerator at DESY.
Fig. 29: The Q0(Eacc) performance of hydroformed nine-cell TESLA-shape cavities
203
W. S INGER
References
[1] H. Padamsee, J. Knobloch, and T. Hays RF Superconductivity for Accelerators (John Wiley &
Sons, New York, 1998; 2nd ed. 2009).
[2] H. Padamsee, RF Superconductivity: Science, Technology, and Applications (Wiley-VCH,
Weinheim, 2009).
[3] C. Antoine, Materials and Surface Aspects in the Development of SRF Niobium Cavities
(EuCARD–CERN Monographs, vol. 12, 2012).
[4] H. Stuart (Ed.), Niobium: Proceedings of the International Symposium, 8–11 November 1981,
San Francisco, CA (Metallurgical Society of the American Institute of Mining, Metallurgical,
and Petroleum Engineers, Warrendale, PA, 1984).
[5] W. Singer, SC cavities: material, fabrication and QA (tutorial), Proc. 13th Int. Workshop on RF
Superconductivity, Peking University, Beijing, China (2007).
[6] W. Singer, SC cavities: material, fabrication and QA (tutorial), Proc. 14th Int. Workshop on RF
Superconductivity, SRF2009, Berlin–Dresden, 20–25 September 2009.
[7] W. Singer, Metallurgical and technological request for high purity niobium in SRF application,
in Hydrogen in Matter: A Collection from the Papers Presented at the Second International
Symposium on Hydrogen in Matter (ISOHIM), Uppsala, 13–17 June 2005, Eds. G.R. Myneni
and B. Hjörvarsson (AIP Conference Proceedings 837, AIP Publishing, Melville, NY, 2006),
pp. 51–63.
[8] D. Proch, P. Schmueser, W. Singer, and L. Lilje, Niobium in superconducting RF cavities, Proc.
Int. Symposium Niobium, Orlando, Florida, 2–5 December 2001.
[9] K. Schulze, Preparation and characterization of ultra-high-purity niobium, J. Metals 33 (1981)
33–41.
[10] K. Schulze, O. Bach, D. Lupton, and F. Schreiber. Purification of niobium, in Niobium:
Proceedings of the International Symposium, 8–11 November 1981, San Francisco, CA, Ed. H.
Stuart (Metallurgical Society of the American Institute of Mining, Metallurgical, and Petroleum
Engineers, Warrendale, PA, 1984), pp. 163–224.
[11] Symposium 2000. Proc. Int. Symposium on Tantalum and Niobium: Production, Processing,
Applications, Electronics and Ceramics, 22–25 October 2000, San Francisco, CA, (Tantalum–
Niobium International Study Center, 2000), ISBN 92 9093 003 9.
[12] L. Gmelin, Gmelins Handbuch der Anorganischen Chemie: Niob, Teil A (Springer-Verlag,
Berlin, 1974).
[13] L. Gmelin, Gmelins Handbuch der Anorganischen Chemie: Niob Legierungen, Teil B
(Springer-Verlag, Berlin, 1974).
[14] J. Schlewitz, Niobium and Niobium Compounds (Reprint, Wah Chang, 1996).
[15] M. Hörmann, The production of high thermal conductivity niobium on a technical scale for high
frequency superconductors. W.C. Heraeus, Hanau (1988).
[16] H. Umezawa, Niobium production at Tokyo Denkai, in First International Symposium on the
Superconducting Science and Technology of Ingot Niobium, JLab, Newport News, VA, 22–24
September 2010, Eds. G.R. Myneni, G. Ciovati, and M. Stuart (AIP Conference Proceedings
1352, AIP Publishing, Melville, NY, 2011), pp. 79–83.
[17] R.A. Graham, Single crystal technology for making RRR niobium sheet, in Single Crystal –
Large Grain Niobium Technology: International Niobium Workshop, 30 October – 1 November
2006, Araxa, Brazil, Eds. G.R. Myneni and A.M. Hutton (AIP Conference Proceedings 927,
AIP Publishing, Melville, NY, 2007), pp. 21–35.
[18] R.A. Bosch, Production of high RRR niobium discs for SURA/CEBAF, Fansteel Inc. (1993).
[19] L. Voronenko, A. Eljutin, K. Kovalev, and F. Kovalev, Verfahren zur Hochreinniobproduction.
Patent C 22 B 34/24 (DE 101 12 822 A 1), 11 October 2001.
204
[20] G.R. Myneni and S.R. Agnew, Elasto-plastic behaviour of high RRR niobium: effects of
crystallographic texture, microstructure and hydrogen concentration, in Hydrogen in Materials
and Vacuum Systems: First International Workshop on Hydrogen in Materials and Vacuum
Systems, Newport News, VA, 11–13 November 2002, eds. G.R. Myneni and S, Chattopadhyay
(AIP Proceedings 671, AIP Publishing, Melville, NY, 2003).
[21] C. Chapman, Niobium heat treatment study report, Fermi National Accelerator Laboratory
Report, TD-04-036, 13 August 2004.
[22] Specification documents for production of European XFEL 1.3 GHz cavities, DESY (2009).
[23] W. Singer, X. Singer, and H.M. Wen, Mat. Techn. 91 (2003) 13–18.
[24] H. Safa, D. Moffat, B. Bonin, and F. Koechlin, J. Alloys Compounds 232 (1996) 281–288.
[25] P. Kneisel, G. Ciovati, G.R. Myneni, W. Singer, X. Singer, D. Proch, and T. Carneiro, Influence
of Ta content in high purity niobium on cavity performance, Particle Accelerator Conference,
Knoxville, TN, 16–20 May 2005.
[26] H. Umezawa, Mat. Techn. 7–9 (2003) 33–37.
[27] R.W. Cahn and P. Haasen (Eds.), Physical Metallurgy, 4th revised and enhanced ed., 3 vols
(North-Holland, Amsterdam, 1996).
[28] W. Singer, A. Ermakov, and X. Singer, RRR-measurement techniques on high purity niobium,
TTC Report 2010 (http://flash.desy.de/reports_publications/tesla_reports/ttc_reports_2010).
[29] R.W. Röth, Grenzfeldstärken und Anomale Verlustmechanismen in Supraleitenden
Beschleunigungsresonatoren aus Niob, Dissertation, Wuppertal, 1993 (WUB-DIS 92-12).
[30] A. Ermakov, A.V. Korolev, W. Singer, and X. Singer, A New approach for RRR determination
of niobium single crystal based on AC magnetic susceptibility, in First International
Symposium on the Superconducting Science and Technology of Ingot Niobium, JLab, Newport
News, VA, 22–24 September 2010, Eds. G.R. Myneni, G. Ciovati, and M. Stuart (AIP
Conference Proceedings 1352, AIP Publishing, Melville, NY, 2011), pp. 178–185.
[31] W. Wasserbäch, Phil. Mag. A 38(4) (1978) 401–431.
[32] F. Koechlin and B. Bonin, Superconductor Science and Technology 9 (1996) 453-460
[33] G. Hoerz, H. Speck, E. Fromm and H. Jehn, Gases and Carbon in Metals, Max-Planck-Institut
fuer Werkstoffwissenschaften, Stuttgart, Germany, Physics Data ISSN 0344-8401, 1981, Nr.5-
8, pp. 1-116.
[34] W. Singer, D. Proch, and A. Brinkmann, Particle Accelerators 60 (1998) 83–102.
[35] DESY specification of welded 1.3 GHz superconducting resonators for TTF/VUV-FEL. 09,
DESY EDMS Document, No. 303237 B.1.1 (2005).
[36] D. Reschke, Thermal model calculations for 1.3 GHz TTF accelerator cavities, Proc. 9th
Workshop on RF Superconductivity, Santa Fe, NM (1999), pp. 385–396.
[37] W. Singer, A. Brinkmann, D. Proch, and X. Singer, Physica C 386 (2003) 379–384.
[38] X. Singer, Mat. Techn. 7–9 (2003) 28-32.
[39] W. Singer, X. Singer, S. Aderhold, A. Ermakov, K. Twarowski, R. Crooks, M. Hoss, F. Schölz,
and B. Spaniol, Phys. Rev. ST–AB 14 (2011) 050702.
[40] W. Singer, X. Singer, J. Tiessen, H.M. Wen, and F. Schölz, RRR degradation and gas
absorption in the electron beam welding area of high purity niobium, in Hydrogen in Materials
(AIP Proceedings 671, AIP Publishing, Melville, NY, 2003), pp. 162–175.
[41] A. Schmidt, A. Brinkmann, J. Iversen, A. Matheisen, D. Reschke, M. Schäfer, W. Singer, V.
Sousa, J. Tiessen, and D. Vermeulen, 1.3 GHz niobium single‐cell fabrication sequence, TTC
Report 2010‐01, pp. 1–11.
205
W. S INGER
[42] J. Knobloch, The ‘Q disease’ in superconducting niobium RF cavities, in Hydrogen in Materials

(AIP Proceedings 671, AIP Publishing, Melville, NY, 2003), pp. 133–150.
[43] J. Knobloch and H. Padamsee, Enhanced susceptibility of Nb equator welds to the hydrogen
related Q-virus, Proc. Eighth Workshop on RF Superconductivity, Abano Terme (Padova),
Italy, 6–10 October 1997.
[44] S. Anakhov, X. Singer, W. Singer, and H. Wen, Gas and RRR distribution in high purity
niobium EB welded in ultra-high vacuum, in Hydrogen in Matter: A Collection from the Papers
Presented at the Second International Symposium on Hydrogen in Matter (ISOHIM), Uppsala,
13–17 June 2005, Eds. G.R. Myneni and B. Hjörvarsson (AIP Conference Proceedings 837,
AIP Publishing, Melville, NY, 2006), pp. 71–84.
[45] G. Kreps, D. Proch, and J. Sekutowicz, Half-cell and dumb-bell frequency testing for the
correction of the TESLA cavity length, Proc. 9th Workshop on RF Superconductivity, Santa Fe,
NM (1999), pp. 499–504.
[46] W. Singer, S. Aderhold, A. Brinkmann, R. Brinkmann, J. Iversen, G. Kreps, L. Lilje, A.
Matheisen, W.-D. Moeller, D. Reschke, A. Schmidt, J.K. Sekutowicz, X. Singer, H.Weise, and
P.M. Michelato, Preparation phase for 1.3 GHz cavity production of the European XFEL,
IPAC2010, Kyoto, Japan, 23–28 May 2010, THOARA02.
[47] J. Bürger, J.A. Dammann, L. Hagge, J. Iversen, A. Matheisen, and W. Singer, Physica C:
Supercond. 441(1–2) (2006) 268–271.
[48] J. Dammann, L. Hagge, J. Iversen, J. Kreuzkamp, and W. Singer, Towards PLM-based quality
assurance in the fabrication of superconducting cavities for the European XFEL, IPAC2010,
Kyoto, Japan, 23–28 May 2010.
[49] P.D. Gall, A. Goessel, V. Gubarev, and J. Iversen, A database for superconducting cavities,
Proc. 13th International Conference on RF Superconductivity SRF2007, 14–19 October 2007,
Beijing, China, TUP02.
[50] W. Singer, The challenge and realization of the cavity production and treatment in industry for
the European XFEL, Proc. 16th International Conference on RF Superconductivity SRF2013,
France, Paris, 23–27 September 2013, MOIOA03.
[51] P. Kneisel, G. Ciovati, P. Dhakal, K. Saito, W. Singer, X. Singer, and G. R. Myneni, The rise of
ingot niobium as a material for superconducting radiofrequency accelerating cavities,
Accelerator Physics, Cornell University Library, April 2013.
[52] W. Singer, S. Aderhold, A. Ermakov, J. Iversen, D. Kostin, G. Kreps, A. Matheisen, W.-D.
Moller, D. Reschke, X. Singer, K. Twarowski, H. Weise, and H.-G. Brokmeier, Phys. Rev. ST–
AB 16 (2013) 012003-1.
[53] G.R. Myneni, G. Ciovati, and M. Stuart (Eds.), First International Symposium on the
Superconducting Science and Technology of Ingot Niobium, JLab, Newport News, VA, 22–24
September 2010 (AIP Conference Proceedings 1352, AIP Publishing, Melville, NY, 2011).
[54] G.R. Myneni and A.M. Hutton (Eds.), Single Crystal – Large Grain Niobium Technology:
International Niobium Workshop, 30 October – 1 November 2006, Araxa, Brazil (AIP
Conference Proceedings 927, AIP Publishing, Melville, NY, 2007).
[55] W. Singer, S. Aderhold, J. Iversen, G. Kreps, A. Matheisen, X. Singer, K. Twarowski, H.
Weise, M. Pekeler, F. Schölz, B. Spaniol, and E. Stiedl, Advances in large grain resonators for
the European XFEL, in First International Symposium on the Superconducting Science and
Technology of Ingot Niobium, JLab, Newport News, VA, 22–24 September 2010, Eds. G.R.
Myneni, G. Ciovati, and M. Stuart (AIP Conference Proceedings 1352, AIP Publishing,
Melville, NY, 2011), pp. 13–24.
206
[56] H. Umezawa, K. Takeuchi, F. Furuta, T. Konomi, K. Saito, K. Nishimura, Single Crystal

Niobium Development, Proc. 1st International Particle Accelerator Conference, IPAC10, Kyoto,
Japan, May 23-28, 2010, MOPEB073, www.jacow.org.
[57] K. Saito, F. Furuta, H. Umezawa, Multi-wire Slicing of Large Grain Ingot Material. Proc.
14thInternational Conference on RF Superconductivity. SRF09, Berlin-Dresden, Germany,
September 20-25, 2009, THOAAU05, www.jacow.org.
[58] A. Ermakov, I. Jelezov, X. Singer, W. Singer, G. B. Viswanathan, V. Levit, H. L. Fraser, H.
Wen and M. Spiwek, J. Phys.: Conf. Ser. 97 (2008) 012014.
[59] W. Singer, X. Singer, and P. Kneisel, A single crystal niobium RF cavity of the TESLA shape,
in Single Crystal – Large Grain Niobium Technology: International Niobium Workshop, 30
October – 1 November 2006, Araxa, Brazil, Eds. G.R. Myneni and A.M. Hutton (AIP
Conference Proceedings 927, AIP Publishing, Melville, NY, 2007), pp. 133–140.
[60] X. Singer, W. Singer, J. Schwellenbach and M. Pekeler, Verfahren zur Herstellung von
Hohlkörpern von Resonatoren, Patent No. 10 2006 021 111, DESY, Hamburg, 2 August 2007.
[61] S.V. Vonsovsky, Yu. A. Izymov, and E.Z. Kurmaev, Superconductivity of Transition Metals
(Springer-Verlag, Berlin, 1982).
[62] T. Fujino, H. Inoue, K. Saito, S. Noguchi, M. Ono, E. Kako, T. Shishido, A. Kubota, and S.
Koide, Status of the seamless L-band cavity fabrication at KEK, Proc. 7th Workshop on RF
Superconductivity, SRF95, Gif sur Yvette, France, October 1995.
[63] V. Palmieri, Progress on spun seamless cavities, Proc. 12th Workshop on RF Superconductivity,
SRF2005, 10–15 July 2005, Ithaca, NY, TuP68.
[64] M. Koc (Ed.), Hydroforming for Advanced Manufacturing (Woodhead, Cambridge, 2008).
[65] C. Antoine, J. Gaiffier, H. Chalaye, J.F. Roche , Hydroforming at Saclay: first issues, Proc. 8th
Workshop on RF superconductivity, Abano Terme (Padova), Italy, 1997.
[66] W. Singer and I. Zhelezov, Verfahren zur Herstellung von schweissnahtlosen
Hochfrequenzresonatoren, Patent No. 10 2007 037 835, DESY 22607, Hamburg, Germany,
2007.
[67] W. Singer, Physica C: Supercond. 441(1–2) (2006) 89–94.
[68] I. Gonin, H. Kaiser, D. Proch, and W. Singer, Finite element simulation of the TESLA-cavity
hydroforming process, Proc. 7th Workshop on RF Superconductivity, SRF95, Gif sur Yvette,
France, October 1995.
[69] K. Lange (Ed.), Umformtechnik. Handbuch für Industrie und Wissenschaft, Band 1–4 (Springer-
Verlag, Berlin, 1984).
207
How to Achieve the Best SRF Performance: (Practical) Limitations and

Possible Solutions 1
C. Z. Antoine 2
CEA IRFU, Gif-sur Yvette Cedex, France
Abstract
This chapter presents in the first part the requirement for the surface
preparation of RF Niobium cavities and its justification in term s of physical
origin of limitation (e.g. cleanliness and field emission, influence of the
surface treatments, morphology and surface damage, etc.). In the second part
we discuss the different models describing the ultimate limits of SRF
cavities, and we present one of the possible ways to overcome Niobium
monopoly toward higher performances.
Keywords: RF Niobium cavities, materials, surface, performance.
1 How to get a ‘good’ cavity

Whatever the application, a ‘good’ cavity is one that exhibits the highest possible accelerating field
along with the highest possible quality factor. For a 1.3 GHz ‘Tesla shape’ cavity, that would be a
quality factor in excess of 1010 and an accelerating gradient higher than 40 MV·m–1, which
corresponds to a surface magnetic field of ~170 mT.
The recipe for obtaining a good cavity is pretty well known, although up to the present time we
still don’t know why it works, and most of all, why it doesn’t always work!
Basically, one needs to form high-purity, bulk niobium sheets to shape half-cells, then weld
them together by electron beam melting, get rid of the internal surface damaged layer and any other
surface impurities through (electro-) chemical polishing, and then finally bake the cavity at low
temperature. In the next section, we will try to describe the origin of each of these specifications in the
light of what we know about the physical and chemical properties of the material.
2 Field emission and particulate contamination

Field emission is an emission of electrons from a metal surface subjected to high electric fields
(see Fig. 1). The phenomenon was studied by Fowler and Nordheim in 1928, and is attributed to the
passage of electron tunnelling through the surface barrier.
Identified as a major cause of the limitation to the breakdown voltage of electrical devices due
to the strains that it initiates, field emission has for decades been the object of many studies in d.c. RF
studies started somewhat later.
1
Most of the information in this tutorial is extracted, with permission, from “Materials and surface aspects in the
development of SRF niobium cavities”, EUCARD series on Accelerator Science Vol. XII (2011).
2
claire.antoine@cea.fr
C. Z. A NTOINE
Fig. 1: The modelling of electron trajectories in a cavity
In accelerators, field emission is responsible for the unwanted absorption of RF power, dark
current, and possible breakdown. In cavities, with fields of some tens of megavolts per metre, the
measured emission current density is several orders of magnitude greater than predicted by the
Fowler–Nordheim theory.
The observed current law follows a modified Fowler–Nordheim model (Eq. (1)), with two
adjustable parameters: Ae, analogous to a surface, and β, a dimensionless parameter that corresponds
to the enhancement factor of the electric field. 3 Typical values are Ae ~ 10–15 m2 and β ~ 50–500. Φ is
the work function of the metal:
1.54 × 10−6 β 2 E 2  6.83 × 109 Φ1,5 
I DC ( E ) = Ae exp −  . (1)
Φ  βE 
Numerous experiments in d.c. and RF have shown a correlation between localized breakdown
and field-emitting particles [1-16]. In particular, it has been shown that dust particles and scratches are
good candidates for emission sites, and that the nature and the shape of the particles play a paramount
role, independent of the substrate material.
Dust particles were quickly suspected, but it was apparently not possible to get a proper agreement
with the Fowler–Nordheim law: their apex had a surface of some square micrometres (10–12 m2) and their
βapparent was less than 10 (see Ref. [17]). Other influences have been explored, in particular the role of the
oxide layer. Indeed, this has a very special ‘amorphous-microcrystalline’ structure with stacks of local
defects, which can form ion- and/or electron-conducting channels that may possibly play an active role
in the field emission (see, e.g., Refs. [18-21]).
However, it has been shown that increased thickness of the oxide layer – up to several hundreds
of nanometres – due to anodizing does not change the nature or intensity of the field emission [22].
2.1 The influence of a particle’s morphology on its emissivity

Specific devices inserted into the electron microscope allow us to identify in situ transmitter sites, and
to obtain their image as well as their chemical composition, using EDX on samples. The field emission
does indeed seem to be associated with the presence of particles or scratches on the surface. However,
not all particles emit; it depends on their shape and size. An intentional surface contamination with
particles of a similar kind but of different shape (see Fig. 2(a)) allows us to show that spherical
particles, with a modest field increase factor (β ~ 3), emit far less than particles with more complex
forms [23].
3
Note that this electric field enhancement factor is similar to the magnetic field enhancement factor described in Section 3.
They are both due to morphology (see below). A βmagn of about 1.5–2.0 has dramatic consequences for the transition in the
high magnetic field region, whereas the βel values observed in electron field emission are much higher (50–500) and cannot
be related to surface roughness.
210
H OW TO ACHIEVE THE B EST SRF P ERFORMANCE : (P RACTICAL ) L IMITATIONS AND P OSSIBLE . . .
The presence of dust particles and roughness can locally increase the electric field. Roughness
due to machining or etching does not play much of a role: direct topological evaluation of beta as seen
in Section 3 and also as measured with a tunnel microscope by Niederman [17] shows that micron
roughness induces betas of less than 2 to 10. Higher betas can only be attributed to the combination of
defects at several size scales (see Fig. 2(b)). Due to their fractal nature, most of the natural dust
particles exhibit nanometric defects that add up to the general shape factor, and field enhancement
factors of the order of several hundreds are easily achievable [10, 23, 24]. Moreover, in Ref. [25], it
was shown that when the current density reaches 1011–1012 A·m–2, the emitting area (in the Fowler–
Nordheim equation) decreases to a few square nanometres, a fact compatible with the melting of the
emitter apex and the formation of a Taylor cone.
Fig. 2: (a) Nickel particles of varying morphology. (b) Modelling of the field increase factor due to the
superposition of protuberances. This model allows us to understand why the perfectly spherical particle emits far
less than the other.
Note the following about breakdown: in superconducting cavities, when breakdown occurs, it
usually results in the formation of localized craters of a few microns, and it can either improve or
decrease the cavity performance [9, 16]. Studies conducted on samples confirm this result, showing
that processing is efficient only in 50% of the cases. In the other 50% of cases, breakdown results in a
protruding particle welded on the surface, which is a very stable emitting site, and in a degradation of
the cavity performance [9, 16]. With the improvement of cleaning techniques, surface fields up to 80–
90 MV·m–1 are currently observed without any breakdown or measurable field emission inside
superconducting cavities [2], whereas breakdown is a common feature in normal conducting cavities.
The obvious difference between superconducting and normal conducting cavities is the operating
temperature; a thermally activated phenomenon such as electromigration being negligible at low
temperature [26].
2.2 Influence of assembly and vacuum handling

In the early 1990s, the dust created during cavity assembly was often cited as responsible for the field
emission phenomena observed during the testing of cavities, but there were no formal proofs. The
incidents were listed in the specifications of the experiments: leakage, problems during installation,
change of seal, and so on. A statistical study of hundreds of test results stored at the laboratory showed
that the threshold for the appearance of electrons was about 4–5 MV·m–1 lower in the case of a
problem during assembly. This effect can only be shown statistically: indeed, only a small portion of
211
C. Z. A NTOINE
the surface may be at the origin of the emission, where the electric field is the strongest. The dust
particles accumulate in a random way and are not necessarily localized in the ‘danger zone’.
Conversely, it will take just a single dust particle in that particular zone in order for the electrons to be
emitted. Our work has been able to confirm, on a substantial basis, the intuitions that were generally
held by the experimenters. This has led us to reconsider the role of dust particles [30]. Vacuum
handling also calls for some precautions: in Ref. [31], general operations such as gasket assembly, pre-
pumping and venting, and steady-state pumping have been monitored with a particle counter installed
downstream inside the vacuum system. This shows that any human operation, shock or strong
vibration liberates a lot of particles. Only steady-state pumping is rather benign, with the exception of
start-up or possible arcing. Any valve manipulation generates particles if the pressure differential
between the two sides is high.
3 What kind of niobium?
3.1 High-purity, bulk niobium

High purity is not required for superconductivity where a rather low mean free path leads to the lowest
surface resistance. The surface resistance in RF is defined as a function of the temperature, as follows
(see Fig. 3, left):
RS = RBCS + Rres , (2)
where
ω 2 −∆ / kT
RBCS = A(λ4L , ξ F , , ρ n ) e . (3)
T
Here, A is a constant depending on λL (the penetration depth of the London field), ξ (the coherence
length of the Cooper pairs), (the mean free path of the quasi-particles), and ρn (conductivity in the
normal state); ω is the RF frequency and ∆ is the superconducting gap. There exists a component Rres,
which does not depend on the temperature. Its origin is not very clear, though it seems to be related to
the conductivity of the material in the normal state, ρn. RBCS is due to the scattering of the remainder of
the normal electrons of the superconductor over the lattice (where ‘BCS’ refers to Bardeen–Cooper–
Schrieffer theory; see Section 4.1).
A high Residual Resistivity Ratio (RRR) is required for thermal stability rather than for good
superconducting properties. As can be seen in Fig. 3, at the centre, most of the bulk Nb cavities, with
RRR~300, exhibit a higher RBCS than thin film Nb cavities with RRR ~15-30. In Fig. 3, on the right,
one shows that in presence of defects, the quench field is higher with higher RRR.
Fig. 3: From left to right: a typical surface resistance for a 1.3 GHz cavity at low field; the influence of the mean
free path l on the BCS component of the surface resistance; and a thermal calculation showing the stabilization
effect of a high RRR (after [32, 33]).
212
In the 3–15 K range, there is a direct relationship between the RRR and thermal conductivity
[34]. At 4.2 K, the thermal conductivity from niobium is roughly equal to RRR/4 [2]. A
superconductor is intrinsically a bad thermal conductor, as some of the (electrical and thermal)
conduction electrons are paired into Cooper pairs and thus can no longer contribute to the heat
transfer. To improve the thermal conductivity, it is essential to get rid of the main scattering sources;
that is, interstitial light elements in the metal matrix. At lower temperatures (≤ 2 K), the major
conduction mechanism is not related to electron propagation, but to phonon propagation. In such
situations, the scattering sources are rather crystalline defects and thus fully recrystallized samples
exhibit a large phonon peak, even with a rather low RRR. Examples of thermal conductivity curves vs
RRR is given in Fig. 4.
Fig. 4: The thermal conductivity of various RRR samples (after [34]). For well-recrystallized samples or
monocrystals, the phonon peak can reach several tens of W·m–1·K–1 between 1.5 and 2 K, almost independent of
the RRR.
Sputtered thin films, on the other hand, usually have a high Q0 value at low field because of
their lower mean free path, but they exhibit an early Q-slope, which confines them to low-field
applications (see Fig. 5). We will see in Section 3.6 that this situation is now changing.
Fig. 5: A comparison between typical curves of niobium-sputtered films and bulk Nb cavities
213
C. Z. A NTOINE
3.2 Surface treatments

Empirically, an abrasion of at least 100–200 µm of the surface is necessary in order to obtain optimal
cavity performance (see the discussion on the damaged layer below). Two main treatments exist,
namely Buffered Chemical Polishing (BCP) and ElectroPolishing (EP) (see Fig. 6). EP has supplanted
BCP somewhat, at least in the hope of reaching record accelerating gradients, but is more difficult to
perform reproducibly. A huge quantity of information is available in the literature about the
optimization of each process, as well as on the research into alternative ‘recipes’. The reader is invited
to consult Superconducting Radio Frequency (SRF) tutorials for details. In this chapter, we will
concentrate on alternative routes that seem to be the most promising in terms of cost.
e-
Nb5+ e-
Nb5+
Nb0 NbF5
NO3-
e- NbF5
Nb0 2H+ + 2e-

H2
Nb5+
6 Nb + 10 HNO3 → 3 Nb2O5 + 10 NO↑ + 5 H2O Nb → Nb5+ + 5 e- → Nb2O5
Nb2O5 + 10 HF ⇔ 2 NbF5 + 5 H2O Nb2O5 + 10 HF ⇔ 2 NbF5 + 5 H2O
Fig. 6: A comparison between chemical polishing (left) and electropolishing (right). In both cases, niobium is
oxidized into Nb5+. In the case of chemical polishing, oxidation occurs because of the presence of a strong
oxidant (NO3–) in the solution, while in electropolishing oxidation occurs because of the bias applied to the
anode. Because of the presence of water, the stable form of Nb is Nb2O5; but HF decomposes the oxides into
NbF5, which is soluble in the solution.
Indeed, the removal of some hundreds of microns by electropolishing is costly and presents
reproducibility issues. Chemical polishing leads to a detrimental roughness and in order to get a
‘smooth’ surface it is necessary to electropolish again by about 100 µm. Something similar happens
when we apply industrial mechanical polishing techniques such as tumbling. The industrial process
applied until recently left a damaged layer of about 100 µm, as can be inferred by the necessity to
further etch the surface (by EP or BCP) [35-37]. It would be more interesting to find a reliable method
for the first 1–200 µm and then finish the treatment with a light electropolishing operation (10–
20 µm).
Very recently, the tumbling process was adapted in order to reduce the damaged layer, by
including some of the features of the mechanical polishing applied in metallography and microscopy,
which is known to produce a reduced damaged layer.
This chapter summarizes the main features concerning chemical and electrochemical treatments
of the cavity surface. Basically, the (electro-)chemical reactions are the same, but the localization of
the electrochemical cell is different: in chemical polishing, the oxidation/reduction occurs due to local
differences of potential; while in EP, the potential is applied externally.
214
3.2.1 Buffered Chemical Polishing (BCP)

Composition
− ~2 Volumes of H3PO4 (buffer, very viscous).
0 5+
− ~1 Volumes of HNO3 (oxidant, transforms Nb into Nb ).
5+
− ~1 Volumes of HF (complexant of Nb , dissolves the oxide layer formed by HNO3 into
NbF5).
− Variation of composition allows us to adjust the etching rate.
Pros
− Easy to handle, middle stirring is necessary.
− Fast etching rate.
− Very reproducible.
Cons
− This is not ‘polishing’, but ‘etching’: all crystalline defects are preferentially attacked (etching
pits, etching figures).
− Grains with various orientations are not etched at the same rate, which induces roughness!
max –1
− Except for a few cases, Eacc ~ 25–30 MV·m .
Caution!
− Do not process at temperatures higher than 25°C.
− Risk of runaway.
3.2.2 Electropolishing (EP)

Composition
− ~9 Volumes of H2SO4 (buffer, very viscous).
5+
− ~1 Volume of HF (complexant of Nb , dissolves the oxide layer formed due to the high
0
potential applied to Nb ).
− Variation of composition allows us to adjust the etching rate.
Pros (in ideal conditions, i.e. viscous layer present)
− This is really ‘polishing’, not sensitive to crystallographic defects – it produces a smooth
surface.
− Should not be sensitive to the cathode–anode distance – the same etching rate everywhere.
max –1
− It gives (but not always!) the best ever Eacc ~ 45 MV·m (Tesla shape → ~180 mT).
Cons
− It is not possible to reach an ideal state in most of our processing conditions.
− Very sensitive to stirring condition, temperature, and aging of the mixture.
− Not very reproducible.
− Safety issues (acid mixture sensitive to water, H2 evolution, etc.).
Caution!
− If T increases: the etching rate increases but there is also a risk of pitting, H loading and HF
evolution.
− If V increases: the etching rate increases but there is also a risk of pitting, the generation of
sulphur particles and sensitivity to the cathode–anode distance.
3.2.3 New tumbling developments

Tumbling, or Centrifugal Barrel Polishing (CBP), was initially an industrial process. It has been
applied to cavities for decades at KEK or DESY, but was not optimized to reduce the damaged layer.
Work on this topic has been resumed at Fermilab, with extra care being taken to use less aggressive
polishing media, inspired by the slurry used in metallography. Although the process is not fully
optimized, and still has a lot of intermediate steps, it has been possible to prepare mirror-like finish
cavities with a very good RF performance after CPB and only 20 µm of EP. The removal of welding
215
C. Z. A NTOINE
defects, including some pits, seems very effective, and improvements in performance have been
observed (CBP + 20 µm EP showed improved performance compared to heavy EP) [38, 39].
Performing CBP saves a huge quantity of acid mixture and is preferable from the safety point of
view. The evolution of the process to hydrogen-free processing, inspired by the work at Higuchi, is
under way [40]. Nevertheless, this process is not applicable to every possible cavity shape. Studies
should be conducted to see if the polishing step can be applied initially on the niobium sheets before
forming.
Thorough control might become prohibitive when dealing with large batches, so once a vendor
has been qualified, we can consider taking only random samples. The statistical efficiency of such
sampling is well established in industry and can be relied on (see, e.g., Ref. [41]).
3.3 Surface damage
3.3.1 ‘Skin-pass’, the damaged layer, and recrystallization

The surface state of the niobium sheets is directly linked to the industrial procedure: after lamination,
the sheets are never completely flat. They have to be straightened by a superficial lamination process
that is called ‘skin-pass’. Skin-pass tends to concentrate damage (i.e. dislocations, displaced atoms,
etc.) near the surface (see Figs. 7 and 8).
Differential strain has very bad consequences during the recrystallization steps that will be
performed afterwards: the material is strongly deformed at the surface but much less so in its core. In
the case of weak deformations, we are close to critical hardening and thus we risk an exaggerated
growth of the grains in the core, while the grains near the surface remain small.
Moreover, this situation is aggravated due to texture. Indeed, the texture at the core of well-
crystallized niobium is rather like type (111), while the superficial texture because of the skin-pass is
more like (100) and spreads in across about 200 µm. This latter orientation does not seem to
recrystallize as well as the (111) 4 It will become part of the so-called ‘damaged layer’ of 100–200 µm,
which has to be removed from the surface of the cavity by chemical abrasion. This also explains the
patchy nature of the distribution of dislocations: some grains recover well during recrystallization,
while others retain damage, depending on their initial orientation (see the discussion below on the
influence of baking).
Fig. 7: Examples of crystalline defects frequently encountered in materials
4
T. Bieler (MSU), K.T. Hartwig (Texas AM), R. Crooks (Black Lab); personal communications. Note that recrystallization
also occurs during the 800°C, 2 h treatments usually applied to get rid of hydrogen after heavy (electro-) chemical polishing.
216
Fig. 8: A finite element simulation of the 2% reduction of a 3.5 mm sheet with 1 cm diameter rolls. The strain is
concentrated in the near-surface region (red). The localized strain exceeds the average by a factor of 5. (Courtesy
of Non-Linear Engineering, LLC and [43].)
Fig. 9: A cross-section of a Nb sheet as received (after mechanical–chemical polishing of the cut surface). The
top view is the SEM. The middle view shows the false colour map obtained from the electron backscattered
diffraction diagram. Electrons penetrate ~100 nm of the surface and when the crystal is too disordered, no
diffraction pattern can be interpreted, and only a black spot is shown. The lower view is a pole figure. A perfect
crystal would only exhibit dots. When the material is too disordered, no clear pole figure can be observed. On
the left-hand side, one can observe the core of the sheet. The grains are large, with a uniform size. They are
slightly elongated, which is often the case for a rolled sheet with incomplete recrystallization. The pole figure is
also classical for a rolled sheet with a heavy texture. On the right-hand side, one can observe very small grains,
many of them so disordered that no diffraction is observed. The pole figure is blurred, indicating a highly
disordered material. The first 100 µm of the surface are the most affected, but the damaged layer extends 200–
300 µm deep under the surface.
217
C. Z. A NTOINE
The depth of the damage may also increase somewhat during deep drawing, although the
deformation generally remains below 30% and is easily recovered. Additional thermal strain can also
appear if the welding is not properly conducted. On the other hand, a series of uncontrolled chemical
polishing operations (‘pickling’/brightening by the manufacturer, treatments before welding, etc.) that
can reduce it are applied during the fabrication process. Up to now, there has not been a systematic
study aimed at minimizing this layer, although it might have important consequences for the surface
treatments applied to cavities for large-scale production.
Up to the present day, manufacturers are still far from being able to control the production of Nb
in a reproducible way, and in the context of large orders (ILC type), there is certainly reason to tighten
our specifications on some realistic values and remain highly watchful as to the material that is being
delivered.
3.3.2 Damage, the baking effect, and superconductivity

Until recently, it was very difficult to make a link between crystallographic observations and the
superconducting properties of the cavities. Recent studies by A. Romanenko and collaborators have
opened up this pathway.
Romanenko worked on samples cut out of cavities (small grain or large grain, BCP or EP)
where temperature mapping allowed the selection of cold and hot spots, and he made a thorough
comparison of hot versus cold spots from the structural, morphological, and chemical points of view
[44-47].
It ensued that the main difference between cold and hot spots was observed by Electron
Backscatter Diffraction (EBSD), which is an electron diffraction technique that probes to more or less
the same thickness as the field penetration depth. Indexation of the diffraction pattern gives the exact
orientation of each grain, but one can also gain access to an evaluation of the local misorientation and
dislocation density. Romanenko observed that hot spots tend to exhibit higher misorientation angles
and a higher dislocation density.
Figure 10 shows the distribution of the misorientation angles for a picture formed on various
types of samples, before and after baking. Except for small-grain BCP, misorientation shifts are
reduced upon baking. Small-grain cavities indeed show no improved performance after baking.
If the residual rolling strain is at the origin of the patchy repartition of hot spots, then this
behaviour is very coherent with the nature of the damaged layer that we have described above:
because of the surface texture, some grains with specific orientations resist recrystallization and retain
a high density of dislocation.
Now we must consider how a high density of dislocations can interfere with superconductivity.
Here again, Romanenko has explored a very interesting aspect. He found that Nb samples cut from
cavities that have undergone cold RF tests exhibit the characteristic features of hydride precipitates,
and that in accordance with the density of dislocation, the density of hydrides is higher on hot spots
than it is on cold spots. Indeed, hydrogen is known to segregate along dislocations, forming ‘Cottrell
clouds’. This local high concentration of hydrogen promotes the formation of hydrides on cold RF
testing. Nevertheless, these characteristic flakes are not systematically observed on H-rich niobium
and their appearance probably depends a lot on the surface preparation and temperature cycles.
The presence of hydrides has also been demonstrated by Raman spectroscopy, where the
characteristic vibrational bands from NbH and NbH2 have been observed on cold as well as hot
spots [48].
A. Grassellino has also observed the same samples in the presence of a magnetic field with
muon spin rotation spectroscopy [47].
218
Fig. 10: Local misorientation shifts in cavity samples due to mild baking. Measurements of the misorientation
distribution are performed on the same sample before and after baking. Except for small-grain BCP,
misorientation shifts are reduced upon baking. Small-grain BCP cavities indeed show no improved performance
after baking. (Courtesy of A. Romanenko, after [45].)
In this technique, positive, polarized muons, with their spin aligned to the beam direction, are
implanted on the surface of the sample (typically 300 µm). In a metal, muons tend to position
themselves in an interstitial site of the crystalline lattice. The frequency of precession of the muon’s
magnetic moment is of the order of a number of microseconds and depends on the local magnetic
field. Muons decay in some hundreds of microseconds into a positron and two neutrinos. The positron
is preferentially emitted in the direction of the muon spin at the moment of its disintegration. Taking
into consideration the initial direction of the muon magnetic moment and the time interval between the
moment of injection and the moment of muon decay, one can calculate how the precession frequency
is influenced by the local magnetic field. Moreover, the decay signal also contains information about
the magnetic volume fraction related to a particular frequency. Basically, the measurement is put in
the form of a signal A(t), where the frequency of the oscillation gives the amplitude of the local field,
while its amplitude is a function of the magnetic volume fraction. This technique can be used to probe
– among other things – the field penetration in superconductors.
Several ‘hot’ and ‘cold’ samples have been studied in the presence of a magnetic field
perpendicular to the surface. In principle, this field geometry is different from the field repartition
inside RF cavities, where the field is parallel to the surface, but we will see in Section 3.4 that the
existence of a perpendicular field component is possible due to the surface morphology.
The onset of flux entry measured by A. Grassellino strongly correlates with the onset of RF
losses as measured during the RF test by thermometry. Figure 11 shows a typical example for a
sample cut out of a ‘hot spot’. After baking, the onset of field penetration is shifted towards higher
field for hot as well as cold spots, which is highly consistent with Romanenko’s observations on
dislocation density.
219
C. Z. A NTOINE
It is very difficult at this stage to evaluate the exact mechanism that affects superconductivity.
Niobium hydrides are poor superconductors or normal conductors, depending on their exact
stoichiometry. One can suppose that they will favour preferential field penetration. Nevertheless,
hydrides, the density of dislocations and other types of defects are reduced by baking in a similar way,
and the presence of hydrides might be a collateral effect rather than the origin of the losses.
This series of experimental observations shows indubitably that dislocations might play a
paramount role in the superconducting behaviour of niobium, and that evaluation of their origin as the
study of a possible cure to lower their density might be necessary to increase the reproducibility of
cavity production.
Fig. 11: Left: the amplitude of the asymmetry signal A(t), which is proportional to the volume fraction of the
sample not containing magnetic flux. When vortices start to enter the sample, this fraction decreases. In this
example, one can clearly see that baking increases the onset of field penetration, while subsequent polishing
restores the sample close to its initial behaviour. Right: the behaviour of the cavity from which the sample was
cut out (after [47]).
3.3.3 Tunnelling spectroscopy (point contact tunnelling)

Point contact spectroscopy is a conductance measurement performed by bringing the tip from a tunnel
microscope into contact with the surface. When the tip is just in contact with the surface, the
measurement yields high resistance due to the insulating oxide layer. In this case, the interface layer
between the superconductor and the oxide is also probed during the measurement. On the other hand,
it is possible to produce a low-resistance measurement by pushing the tip across the oxide layer. In this
case, the superconductor is probed somewhat more deeply and can be considered to be representative
of the bulk material. By comparing the low- and high-resistance results, one can identify problems
related to the actual surface of the superconductor.
The first measurements on monocrystals of niobium with the same RRR as the material of the
cavities did not show any dramatic difference in the gap compared to pure metal: 1.55 meV for both
low and high impedance. The low-impedance (i.e. bulk) measurements show a purely BCS behaviour.
In the high-impedance measurement, however (with the oxide), there is an increase in the
density of states of the quasi-particles (smearing) that cannot be explained as being merely an effect of
the measurements. Moreover, the signal is enlarged compared to a purely BCS behaviour, which
means that the mechanism involved is not the proximity effect but comes from a breaking of the
Cooper pairs [49].
An additional correction needs to be introduced. At this stage, the only mechanism that
adequately reflects these results is the theory of Shiba [50] that refers to the inelastic scattering of
220
quasi-particles on magnetic impurities in the case of strong coupling: the curve is adjusted by means of
a so-called ‘pair-breaking parameter’, Γ, that is related to the concentration of the diffusion centres.
After baking, the parameter Γ decreases sharply. It is also very interesting that there is no
difference at the interface between vacuum baking and baking in air, just as is observed on cavities.
In addition, recent statistical analysis on the point contact tunnelling spectrum measured on
chemically polished cavity-grade Nb samples reveals that the pair-breaking parameter, Γ, depends on
the resistance of the junction: low-resistance junctions (R ≤ 100 Ω) exhibit a bulk Nb superconducting
gap of 1.55 meV and almost no pair breaking, whereas higher-resistance junctions (R ≥ 1 kΩ) show
higher values of Γ (see Fig. 12). A preliminary interpretation of these results is that magnetic
impurities are localized in the uppermost defective Nb2O5 oxide and/or at the interface with the
underlying sub-oxides forms (NbO2, NbO).
The origin of the baking effect described earlier might also be found in a local reorganization at
the interface, inducing a lower concentration of localized magnetic moments or a weaker coupling
with the underlying Nb superconductor. Because of the low dimensionality of the interface and the
highly defective nature of the Nb2O5, diffusion can still make a difference even at low temperature.
The same approach has been applied to coupons cut out of cavities (cold and hot spots).
In the case of hot spots, the density of state is so high at the Fermi level that it forms a central
peak in the conductance spectra (around zero bias). This peak remains even if the superconductivity is
killed by the presence of an external magnetic field, and can be attributed to the Kondo effect, due to
the presence of localized magnetic moments [51, 52]. This feature appears for ~30–50% of the
junctions measured on hot spots, as compared to ~5% on samples of cold spots.
The existence of these magnetic impurities has been now confirmed by SQUID magnetometry.
The interesting fact is that the density of the magnetic impurities depends strongly on the initial
surface preparation of the samples and the integrated magnetic signal intensity scales with the surface
area of the sample, which confirms that they are localized in a very superficial surface layer [51, 52].
We should consider these results in the light of the magnetic measurements described in [53], that also
brought to the fore the existence of magnetic impurities in the first 10 nm of the niobium surface.
Fig. 12: Various examples of high-impedance conductance curves measured at 1.6 K. All curves show a
superconducting gap ∆ equal to 1.55 meV, as expected for pure Nb. (a) In black, conductance curves measured
at 1.65 K at the surface of a monocrystalline niobium sample that was electropolished before baking; in red, the
fit of the experimental data with the Shiba model – the adjustment is done using a parameter Γ that takes into
account the inelastic scattering of quasi-particles by magnetic impurities. In blue, for the sake of comparison, the
expected BCS signal for the same superconducting gap ∆. (b) A comparison of the same sample before (in blue)
and after baking (in green). The insert shows an enlargement of the 0–2.5 mV part. After baking, the scattering is
reduced and the shape of the spectrum becomes more BCS-like (red arrows). (c) A typical conductance curve
observed on hot spots (in black). Deconvolution shows that it is probably the superposition of a localized state at
zero field to the superconducting gap.
221
C. Z. A NTOINE
It is indeed known that niobium oxides are sub-stoichiometric and have a variable density of
oxygen vacancies. During baking, radical changes occur in the metal/oxide interface: Nb2O5
dissociates into NbO2 and NbO, as we were able to establish by other means [54]. One can imagine
that the density of magnetic centres decreases considerably during this transformation. However, in
comparison to earlier work [55], the magnetic signal intensity is too important to be caused just by
sub-stoichiometric Nb2O(5 – x) 4d1 electrons (roughly by a factor of between 10 and 50, depending on
the surface treatment), in qualitative agreement with earlier work done by Casalbuoni [53]. Up to now,
the additional contribution remains a mystery. It might be possible that O–H–Nb vacancy complexes
present close to the surface (see Section 4.2.4 from [56]) can exhibit some magnetic behaviour, but
this still needs to be explored.
The existence of magnetic impurities results in a non-zero density of state at the Fermi level (the
so-called ‘gapless superconductivity regime’), which is liable to produce dissipations and suppress the
superheating field [52, 57, 58]. In particular, in Ref. [58], a simple model is considered with uniform
spatial distribution of magnetic and non-magnetic impurities (i.e. within ξ from the surface), which
has the advantage of being valid at arbitrary frequency, κ (the Ginzburg–Landau parameter, GL),
temperature and scattering rate compared to ∆. The authors computed the surface impedance (RS) in
the presence of magnetic impurities in the Shiba approximation. This reveals a saturation of RS at low
temperature, suggesting that magnetic impurities can be responsible for an appreciable fraction of the
residual resistance.
One might wonder how these results correlate with the observation of early field penetration in
high dislocation density areas, as discussed in Section 3.3.2. One can only speculate, but areas with a
high dislocation density are liable to oxidize faster than the remaining surface. For instance, the
emergence of a screw dislocation on the surface provides a small group of isolated atoms (see Fig. 13),
which is highly reactive and would provide a perfect nucleation site for oxide growth [59]. Pitting and
etching also occur preferentially in high dislocation density areas [60-62]. Moreover, angle-resolved
XPS studies show that the thickness of the oxide layer is not uniform [19]. All these ‘ingredients’ are
compatible with oxide becoming thicker or more defective in the emerging dislocation areas.
Fig. 13: A schematic of the emergence point of a screw dislocation
We believe that here lies an interesting direction for further research, not only from a practical
point of view (process optimization), but also for a deeper fundamental understanding of this aspect of
superconductivity.
3.4 Surface morphology and quench

Even when the dissipation phenomena at high field are poorly understood, it is relatively easy to
establish the quench: this is the generalized transition of the superconductor to the normal state. The
cavity then finds itself detuned and most of the incident power is reflected. Because of the high BCS
resistance, at high frequency the generalized heating is often at the origin of the quench. But at
1.3 GHz the thermal instability can in general be attributed to the presence of a localized defect (of the
size of some tens of micrometres in diameter) that may provoke a localized increase in the magnetic
222
field and/or in temperature and thus lead to a quench. Chasing defects the size of some microns on
surfaces the size of a square metre turns out to be quite a challenge: there are a great many possible
defects, but only some effectively influence the functioning of the cavity. Why and how? This is the
both practical and fundamental issue at stake in this chapter.
3.4.1 Surface morphology

The topography of the surface can also be invoked to explain the quench, as it might locally increase
the magnetic field; for example, on the steps that appear at the grain boundaries (see Fig. 14). If the
local field exceeds HC, an area of the material may transit to normal state and start to dissipate
strongly. This phenomenon was originally proposed at Cornell to explain the thermal dissipation at
high field (Q-slope) [63], which was invalidated later by the studies on baking [64].
Fig. 14: The morphology brought about by the different surface treatments greatly depends on the initial
crystalline state. Chemical ‘polishing’ is not a true polishing: it reveals crystallographic defects such as grain
boundaries or the emergence of dislocation lines. Not all the grains etch at the same speed. Therefore steps
appear that are about 20% of the size of the grains. At a nanometric scale, though, it is not possible to distinguish
the effect of one treatment from the other.
Still, this approach remains of interest to explain the quench, provided that the local increase of
the field can be evaluated correctly. The thermal calculations show that the quench is due to a
localized defect with sides no larger than some microns, and therefore very difficult to localize on a
macroscopic object such as a cavity.
We therefore decided to attempt a morphological characterization of the surface, to see whether
there might be a correlation between the quench and a particular kind of surface defect. There were
two problems, as follows.
− How do we measure the surface near a real quench, given that we cannot measure the RF
behaviour of small samples directly? 5 We need to develop an in situ and non-destructive
measuring technique.
− Which morphological parameter should we consider? As we can see in Fig. 15, the
conventional roughness parameters are not well adapted to our problem; because very
different profiles for the electromagnetic properties may have the same roughness values
(arithmetic Ra or quadratic Rq). Nevertheless, profile (b) should lead to a far larger increase
of the field on the peaks than profile (a).
5
Many attempts have been made to design cavities that can withstand samples. In fact, most of these cavities are at higher
frequencies and do not give a large enough precision to extrapolate to the behaviour of niobium at 1.3 GHz.
223
C. Z. A NTOINE
Fig. 15: Two examples of surface morphology with the same roughness parameters, but very different behaviour
with respect to the EM field.
3.4.1.1 Temperature maps

Several temperature-mapping systems exist. Here, we give the example of a superfluid helium
temperature-measuring arm that was developed some years ago. It allows detection of the position of
the quench in situ by means of thermal diffusion through the cavity wall (see Fig. 16). The distance
between the sensors, of the order of a centimetre, and the angular resolution of the motor give a
precision of 2–3 degrees at the level of the equator. Once the location of the quench has been
determined, we may study the inner surface. Optical methods such as endoscopy are not well adapted:
at low magnification it is not possible to distinguish a detail of a size of ~10 μm; at high
magnification, there is not enough depth of field. We therefore need to apply other methods (a
mechanical sensor (profilometer), electron microscopy, etc.) without having to cut a part of the cavity!
There are several methods for making replicas, but in our case it was necessary to ensure fidelity better
than a micron, on uneven surfaces. During the RF tests of the cavities, we made a temperature map in
order to localize the quench, followed by making a replica at the site of the quench and at two
reference surfaces (far from the quench), to see if we could find a correlation between the quench and
a certain kind of defect or a specific morphology.
Fig. 16: The mobile temperature-measuring arm and the map obtained during an RF test of a cavity at 2 K. To
obtain this sort of map, one needs to place oneself at a field value just before the quench.
As we saw before, conventional roughness parameters are not suitable. We need to measure a
quantity that takes into account the curvature radius of the observed topography, and the increase of
the field is caused by sharp features on the surface.
The measured curvature radius depends on the scale of observation. 6 To start with, we have
studied the phenomenon at a scale of one tenth of a micron, for which we can use profilometry. We
then could extend the study if a higher resolution proved necessary.
6
The dependence of the data on the scale of measurement is a widespread problem, but is often ignored with respect to
roughness. An approach using fractals allows the scale to be overcome: one may, for example, express quadratic roughness
as xn, where x is the scale of observation and n is a fractal number. This approach has been successfully applied to lightly de-
polished silicon wafers [50], but so far we have not succeeded in applying it to polycrystalline niobium, probably due to the
great variability of the morphology of each grain.
224
3.4.1.2 Topological analysis

There exist topological methods that allow us to make a global estimate of the surface state that is
more accurate than classical roughness. 7 Among these, there is analysis by ‘equivalent conformal
elliptical structure’ [66]. In this technique, the surface relief is discomposed into small pyramids and
the surface of each face is projected on to a plane perpendicular to the three directions of the space
(with the same origin). In each direction, the projections tend to exhibit elliptical shapes, which allow
a 3D ellipsoid to be reconstructed. The three axes of the obtained ellipsoids are topologically related to
the surface morphology and it is clear that the ‘steep’ steps will tend to increase the vertical axis of the
ellipsoid. This technique provides a tool that helps us to characterize the tendency of the surface to
exhibit sharp features.
Moreover, the demagnetization factor D, which is the inverse of the (magnetic) field
enhancement factor β, is analytically calculable for an ellipsoid (see Section 3.4).
In the example shown in Fig. 17, we analysed un-annealed material (grain Ø ~ 70 µm) with
steps of 9 µm. If one measures the roughness at a much smaller scale, and/or on material with larger
grains, the differences are far less pronounced and the measurements on etched niobium are close to
that of electropolished metal. This clearly shows that the roughness is mainly due to differences in the
etching rate of the grains and to the resulting steps in the case of BCP.
Fig. 17: Equivalent ellipsoids deduced from measurement of electropolished (EP) and chemically polished
(BCP) surfaces, respectively. First, the surface is measured using a profilometer on a 50 × 50 µm2 surface with
9 µm steps. Then the topological analysis is performed on the acquisition file. We have also indicated the
quadratic roughness (σ), the average field enhancement factor (β) and the value of the half-axis C. The analysed
surface contains ~1000 grains.
The values of the increase of the average field (some per cent) are not enough to explain a
quench, but they do give a qualitative assessment of the roughness: obviously BCP produces ‘sharper’
reliefs than EP. This comprehensive approach allows for a general qualification of a surface treatment.
The same approach can be used to qualify an individual step. This has been done by INFN-Legnaro to
assess the influence of distribution of steps on the field; each step is modelled by a half-ellipsoid 8.
They showed that only a single step, with a high form factor and placed perpendicularly with respect
to the field, is needed to trigger a quench.
We arrived at the same conclusions using 2D modelling of the field with FemLab™, but also by
means of direct measurements on a step (see below).
7
The work of Tian et al. [65] H. Tian, et al., "A novel approach to characterizing the surface topography of niobium
superconducting radio frequency (SRF) accelerator cavities". Applied Surface Science, 2010., based on power spectral
density, also allows us to characterize the surface morphology at different scales.
8
V. Palmieri, C. Roncolato, personal communication
225
C. Z. A NTOINE
3.4.2 Replicas at the quench site

To study the morphological evolution of the quench site during successive treatments, a method of
replica-making proved to be the most practical. We opted for a technique validated at the University of
Besançon, which guaranteed us the desired level of reliability. This is a very inexpensive method that
has already proven itself before (on Formula 1 engines!). Following a short period of collaborative
work to verify the validity of the method on niobium samples, we adapted the technique to cavities, by
developing a tracking system suited to the convex and closed shape of the cavity. This technique has
also been used more recently to study etching pits found in the heat-affected zone of the welding seam
[67].
Figure 18 illustrates the various stages of this technique: having spotted the quench on the
outside of the cavity, it is positioned in such a way that the meridian passing through this point is in a
low position. We introduce a little calibrated ball that allows us to identify the meridian and that gives
us a size scale. Then, with the aid of a mirror, we take a photograph of the inner surface. There is a
slight error due to the fact that we project a curved surface on to the plane of the photograph, but the
degree of accuracy is sufficient for our needs. Next, we place a soft polymer (of the ‘dental
impressions’ type) to make a negative replica of the surface. We then transpose this replica positively
on a hard polymer, which is analysed using a profilometer. The area we want to study is identified
with the help of photographs and the placement of coordinates thereupon. As one can see on the
measurements, the existence of a grain that is prominent compared to the others is evident, although
because of the lighting it is very difficult to distinguish it in the photographs.
This technique offers a significant advantage: it is very inexpensive and relatively easy to
implement. It has allowed us to open up a new direction in the exploration of the inner surface of
cavities that until then had been neglected.
From the topography of the surface, we introduced the real profile of the steps in a 2D model
that allows us to reproduce the local increase of the field due to the morphology. This model can still
be improved, because it does not take into account the finite size of the grain in the direction
perpendicular to the field, but it allows us to assess the overall influence of the morphology.
3.4.3 Modelling the generated field

The models are 2D and refer to a step placed on an infinite plane. The magnetic field is parallel to the
surface of the plane and perpendicular to the step. The calculations show that the smaller the radius of
curvature, the larger is the increase of the field [68, 69]. But measurements taken directly on cuts of
the replica show that the radius stays large (~50 µm), a value that we have retained in the subsequent
calculations. It seems that the next most important parameters are the slope and the height of the edge
of the grain. Figure 19 shows the different results that were obtained: one enters a profile determined
from the experimental measurement (a). One then determines the average increase of the field with
respect to the mean field established far from the disturbance (b). One then can calculate the
contribution of the step to the dissipated power (blue curve in (c)) compared to the power dissipated
across the basal plane (green curve).
It is clear that the influence of the step arises suddenly when approaching the transition field.
This behaviour is quite similar to that observed in the cavities, where a set of hot spots can be seen to
appear on the surface associated with the Q-slope. But at the moment of the quench, a single point (not
necessarily the hot spots already shown previously) becomes much hotter and causes the generalized
transition. In the case of the grain that is studied in Fig. 18 (for a sheet with a thickness of 2.8 mm) a
thermal calculation shows that one can stabilize two normal areas on the ‘noses’ of the steps (an area
with a diameter of ~1 µm, where T < TC, but H > HC) up to ~142 mW. With but 1 mW more power,
the material quenches completely.
226
Figure 20 shows the same type of calculation (2D), but now coupling the RF behaviour to the
thermal behaviour of Nb. These calculations show that if the dissipation is correctly evacuated to the
helium bath, it is possible to maintain an superconducting state up to a certain field, but that a slight
increase in the field (a few tenths of a millitesla) induces thermal runaway.
Fig. 18: The different stages in the characterization of the internal surface of cavities. Because of the artificial
enlargement of the vertical axis, the ‘prominent’ grains are easier to highlight. At normal scale, a step of the size
of 10 µm on a grain that is several hundreds of micrometres in size is quite difficult to distinguish.
Fig. 19: A calculation of the effect of the surface morphology on the local increase of the magnetic field and the
power dissipation (see text). (a) A micrograph of a cutting of the replica along the dotted line on the contour
(measured before cutting). (b) The profile used for calculations. (c) Calculations of the local increase in the field
induced by the defect. (d) The power dissipation of the cavity free of defects (in green) and of the default only
(in blue). We see that the default becomes dominant at high field and can induce a quench.
227
228
Fig. 20: The amount of power that can be transferred to the bath is clearly limited by the interface transfer for a given thermal conductivity. An increase in the purity of the
Nb allows us to reduce the interstitial content, which acts as scattering centres for (thermal) conduction electrons. It increases the thermal conductivity and allows us to
transfer the dissipated power better, and hence to obtain a higher field for an ‘equivalent’ defect. The quench happens on the edge of the defect because of both morphological
field enhancement and temperature enhancement: thermomagnetic quench at H < HC and T < TC.
C. Z. A NTOINE
Figure 21 shows the evolution of this quench after a chemical treatment of 20 µm. The
temperature maps show that the site of the quench has moved by several centimetres. However, this
chemistry is not sufficient to fundamentally alter the topography of the grain under consideration;
hence the factor with which the field increases. But the new site clearly shows a new step, for which
the field increase factor is higher than at the former quench location. Strictly speaking, we cannot
prove that the step did not already exist beforehand. It is, however, easy to show that it is always the
step with the highest β that will quench first.
The surface morphology thus explains some premature quenches observed on cavities treated by
BCP, where enhanced roughness appears close to the welding seam, due to recrystallization of large
grains with different orientations.
The influence of grain boundaries and the morphology on field penetration have also been
studied by magneto-optics [70] and critical current measurements.
The study of bi-grains shows that there is indeed a preferential penetration of the field at the
grain boundaries, but only when the field is perfectly aligned with the boundary plane, and for (static)
fields much higher than the peak fields obtained in RF; that is, under conditions that are quite far from
the configuration of the cavities. In combination with the results on the ‘monocrystal’ cavities, this
shows that, although grain boundaries are an area of weakened superconductivity, they do not
dominate the dissipative phenomena in SRF [70].
An experiment on a monocrystal in which a groove was deliberately dug, however, shows that
when the field is applied parallel to the surface but perpendicular to the groove, a significant vertical
field component appears on the edges of the groove (see Fig. 22). This proves that our 2D model is
realistic. To try to quantify the influence of a grain with finite and realistic dimensions, we also need
3D modelling [70].
Fig. 21: (a) The first site of a quench (EQ = 24.72 MV·m–1 at 1.7 K). (b) The same zone after 20 µm of BCP (the
quench has moved elsewhere). (c) The location of the new quench (EQ = 24.94 MV·m–1 at 1.7 K).
229
C. Z. A NTOINE
Fig. 22: The effect of an artificial groove on the field distribution on the surface of a monocrystalline sample
(after [70]). Left: optical microscopy and 3D profilometry. Right, top: images obtained by magneto-optical (MO)
contrast for field parallel (centre) or perpendicular (right) to the groove. Bottom: schematic repartition of the
field at the origin of the MO contrast.
We remark that the results obtained on monocrystalline cavities are very recent and quite
unexpected. Indeed, in numerous superconducting applications, the grain boundaries exhibit
weakened, or even an absence of, superconductivity, and have a predominant influence on the
properties of the superconductor. We have therefore tried to explore this aspect in some detail.
3.4.4 Welding and roughness

The grain size greatly affects the roughness. Near the weld, the grains of the heat-affected zone have
diameters close to 0.5–1.0 cm. The heights of the steps, as well as the parameters of the equivalent
ellipsoids, increase correspondingly. The local increase of the field, of the order of 40–50%, starts to
become significant.
Using the ‘ellipsoid’ approach described in Section 3.4.3, we were nevertheless able to estimate
the demagnetization coefficient of a single step or a surface. Indeed, the computation of the
demagnetization factor for an ellipsoid is relatively easy:
1  m 
Da = 1 − arccos m , (4)
1− m2  1− m2 
where m = c/a.
Table 1 shows the evaluation of the demagnetization factor estimated after the ellipsoid
parameters as directly measured on surfaces with various roughness characteristics. Note that from the
topological point of view, a hole is equivalent to a protrusion, so the same approach can be used to
study etching pits.
230
Table 1: Roughness parameter and demagnetization C factors for ellipsoids measured on standard surfaces
Parameter Etching (BCP) Electropolishing (EP)

Small- Annealed, Thermally Mean value Weld defect*
grained affected zone
far from the C ~ 50 µm
material (near the weld)
weld
2A ~ 200 µm
Φ grains 70 µm 1–2 mm 0.5–1.0 cm 1 mm → 1 cm -
Ra 1–2 µm 4–8 µm 40–80 µm ~1 µm -

C ~300 ~90–100 ~350 ~70 50
β = 1/D 1065 1028 1.4 1018 1.9!*
*A defect associated with a hot spot on a cavity quenching at 15 MV·m , observed at Fermilab [67, 71] (cf. Fig. 23).
–1
These simple calculations can also be confirmed using the finite element approach. Figure 23
shows the modelling of an ellipsoid of similar dimensions and shows similar modelling for a sharp-
edged fracture. Here again, we can observe that the most important part is the area with the sharpest
curvature.
Similar 3D simulations have also been conducted at Cornell [72] and show that field
enhancement factors between 1.5 and 2 can easily be reached in defects that are very small compared
to the cavity size.
In the case of an electric field, enhancement by a factor of 50–100 is often necessary to model
field emission, and it is doubtful that surface morphology plays an important role; whereas a magnetic
field and enhancement by a factor of 2 will have serious consequences for the transition field.
Fig. 23: The finite element simulation of the field enhancement factor for an ellipsoid of 50 µm × 100 µm.
(Courtesy of Z. Insepov.)
Note that the surface morphology does not matter much for field emission, whereas it can be the
dominant effect for the appearance of a quench.
Cavities treated by electropolishing, followed by a moderate baking, show quench thresholds
that are systematically higher than those of cavities treated in a conventional way, by chemical
polishing and baking (35–40 MV·m–1 and 25–30 MV·m–1, respectively). Only a few individual
231
C. Z. A NTOINE
cavities (in thousands of tests) have succeeded in reaching ~40 MV·m–1 after chemical polishing, but it
is not possible to check the reproducibility of these results. Among these were mainly seamless
cavities or cavities without boundaries (prepared by hydroforming or spinning; monocrystalline
cavities) [73, 74].
It is therefore our belief that all the techniques that maintain welds and grain boundaries are
likely to give rise to a premature quench in case of BCP treatment. ‘Large-grain’ cavities do not come
with any particular advantage in the case of BCP surface treatment, except in terms of the cost of the
niobium supply. Electropolishing does not make all the steps on the surface disappear, but as the
radius of curvature on the steps is much larger, the field enhancement factor is not so high. That could
explain the relatively better results obtained with this surface treatment.
3.5 Conclusion on surface defects

When the accelerating gradient is below 40 MV·m–1 (~170 mT), one can be sure of the presence of a
local defect, wherever it is an inclusion, a pit or relief, or a local variation of composition. If the cavity
quenches at less than 15–20 MV·m–1, then this defect is 50–100 µm sized [33] and can be seen with
the naked eye (see Table 2).
The origin of the inclusion is seldom the initial Nb sheet due to its very high purity; but since it
is a very soft material, the dust particles commonly found in a workshop environment will tend to
become embedded. Special care has to be taken with regard to the cleanliness and smoothness of the
tools that may come in contact with Nb, a habit that is not yet common in many industrial plants and
that deserves close attention.
Pits can originate from by means of different mechanisms.
(1) During solidification of a melted area. Light elements (C, O, N, etc.) tend to diffuse towards
the melted part during cool-down, and thus get concentrated in the last liquid parts. If the
vacuum environment during melting is not good, then it can reach the point at which the light
elements form bubbles inside the material. The typical feature of such a pit is sharp edges (due
to faceting during crystallization), and high concentrations of C and O in the walls of the pit
([75] and personal unpublished results). The critical steps are the melting of the ingot and the
welding.
(2) Pit corrosion during etching. If strain remains in local spots, it will favour pit corrosion. The
typical features of an etching pit are that it is ~100 µm in depth, 50 µm in diameter, smooth,
and very shiny [76]. These local strain areas form in particular with thermal stress, such as can
appear after welding if the cool-down conditions are too abrupt.
One can see that welding is a critical part of the cavity fabrication process that calls for particular care.
232
Table 2: Examples of typical large defects
3.6 Niobium: from bulk to thin films

The deposition of thin films on to copper is not a recent idea. Indeed, from the superconducting point
of view, only a few hundreds nanometers of superconducting material is necessary. Copper presents a
good mechanical support with fine thermal conductivity. Such a technology would decrease the cavity
fabrication costs considerably. Indeed, several accelerators worldwide are based on magnetron-
sputtered Nb films (e.g. LHC and Soleil). This technique is well adapted for circular machines where
high accelerating gradients are unnecessary, but until recently, although a large amount of R&D has
been conducted, thin film technology has not been applicable to high-gradient applications
(see Fig. 24). This is probably related to the fact that sputtered films have a highly defective crystalline
structure: a lot of displaced/foreign atoms, very small grains, high dislocation density, and so on.
233
C. Z. A NTOINE
Fig. 24: A comparison between typical sputtered Nb films deposited on copper and bulk Nb cavities
Only recently, progress in deposition techniques has allowed the deposition of ‘bulk-like’ films
that might result in a breakthrough in the fabrication of high-performance Nb cavities (see, e.g.,
Refs. [77-79]).
Nevertheless, bulk as well as bulk-like niobium are close to the ultimate limits of this
superconductor (transition to the normal state). Only a different type of superconductor can overcome
this limitation (see below).
4 After (bulk) niobium

After nearly 30 years, despite several attempts at using other superconductors, bulk niobium is still
unmatched for high-gradient applications. Nevertheless, recent results seem to indicate that we are
about to reach the material’s ultimate limits, notably with several observations, since 2008, of the
location of quenches randomly fluctuating for a given field. This shows that we are no longer limited
by a point defect leading to a premature transition, but that several areas of the surface have the same
transition probability. This field is found around 43 ± 2 MV·m–1 (1830 ± 85 Oe, ‘Tesla shape’,
RRR 300 9), a weaker field than predicted by the earliest theories.
For a long time, the failure of these attempts was thought to be related to difficulties in
preparing the superconductors themselves. Indeed, the superconductors with the highest values of TC
are generally compounds with several possible phases and the superconducting phase often
corresponds to a fairly narrow region of concentration. As soon as we have a region with a lot of
defects (e.g. grain boundaries), the stoichiometry and the regularity of the lattice are no longer
guaranteed, which may result in a degraded superconductivity.
4.1 Ultimate limits: recalls

In principle, the theoretical maximum of the accelerating field is obtained when the magnetic
component of the electromagnetic field at the surface of the cavity reaches the superconductor’s
transition field at the operating temperature. The precise real limit is not known: the exact mechanisms
of radiofrequency dissipation at the operating temperature (2–4 K) are not well understood, and most
of the established models are valid only close to TC.
Niobium is a classical type II superconductor, which means that it is well described by the BCS
theory (Bardeen–Cooper–Schrieffer), and its extension GLAG (Ginzburg, Landau, Abrikosov,
9
The record accelerating field has been reached at Cornell on a low-losses cavity made of Nb RRR 500. With that peculiar
shape, the exceptional 59 MV·m–1 corresponds to 2065 Oe.
234
Gor’kov), which describes the type II behaviour. The specific description of a superconductor in a.c.
(RF) has been proposed by Gorter and Casimir (1934) with the two-fluids model: charge carriers are
divided into two subsystems, the superconducting carriers (Cooper pairs) of density ns and the normal
electrons of density n.
A type II superconductor can exhibit three states. At low temperature, it is in the Meissner state.
In the presence of an external magnetic field, a screening current appears at the surface of the
superconductor (at its penetration depth λ) and produces a magnetic moment opposite to the external
field up to its first critical field, HC1, where HC1 corresponds to the transition from the purely
superconducting state to a mixed state, where both normal and superconducting areas coexist. The
normal areas consist of field lines called ‘vortices’ (‘vortex’ in the singular), surrounded by screening
currents. The second critical field, HC2, corresponds to the transition from the mixed state to the
normal state. Other transition fields that can be considered are HC, the critical thermodynamic field,
and HSH, the superheating field (see Fig. 4).
The question of which transition field (HC, HC1, HC2, or HSH) should be considered is still
controversial and will be discussed in this section. For other details on the general and mathematical
description of superconducting states, the reader is invited to consult textbooks in superconductivity
(see also G. Ciovati’s tutorial in this volume).
4.1.1 Superheating field

It is possible to observe a metastable state where superconductivity persists at higher fields than HC
(type I superconductors) or HC1 (type II), up to the so-called ‘superheating’ field HSH. In the 1960s, the
field HSH was estimated on the basis of thermodynamic considerations related to the surface energy
[2]. It follows the dotted curve in Fig. 4(a). It was measured near TC on several type I and Type II
superconductors in pulsed resonators for various frequencies [80]. It is easier to observe this effect in
RF than in d.c., because apparently surface defects play a lesser role in nucleating the RF magnetic
transition than they do in nucleating the d.c. transition [80].
A superconductor can be described by its Ginzburg–Landau parameter, which is the ratio of the
field penetration depth to the coherence length of the Cooper pairs: κ = λ/ξ.
According to the values of κ, the ‘superheating’ field can be approximated by the following
expressions [2]:
0.89
H SH ≈ H C if κ << 1, (5)
κ GL
H SH ≈ 1.2 H C if κ ~ 1, (6)
H SH ≈ 0.75H C if κ ≫ 1, (7)
where HC is the critical thermodynamic field.

From the superheating model, one should expect Eacc > 50–60 MV·m–1 for niobium (with
Q = 1011, at 2 K and 1.3 GHz).
For several decades, the commonly accepted explanation was that the field reverses every 10–9
s, whereas it takes 10–6 s to reach the nucleation of a normal zone. Therefore there is not enough time
to see it nucleate. But even in his paper [80], Yogi claims that vortex nucleation dominates at lower
temperatures and that individual vortex nucleation takes less than 10–9 s. Except very close to TC, he
always observed the RF transition at a lower field intensity than the ideal HSH value, and the
temperature dependence indicated a line nucleation model; that is, vortex nucleation.
235
C. Z. A NTOINE
Fig. 25: The peculiarities of RF superconductivity. (a) In RF (dotted curve), one observes superconducting
behaviour at higher fields than in d.c. situations (continuous curves). The ‘superheating’ field HSH is defined
following thermodynamic arguments (see Ref. [2]), but obviously other less-defined mechanisms occur
concurrently. (b) The resistance of a RF superconductor is not zero. At low field, the behaviour of the surface
resistance follows that predicted by the BCS theory up to about 2 K, after which it gets dominated by the residual
resistance Rres. At higher field, there is no valid model yet.
Note that HSH does not depend on HC1 or on HC2. Materials that are good superconductors for
applications at direct current (e.g. magnetic coils) are most often ‘bad’ in RF. Indeed, the pinning
centres of vortices (dislocations, precipitates, etc.) that allow one to obtain high values of HC2 are very
dissipative defects in RF.
4.1.2 Vortex nucleation

Nevertheless, the superheating model is based on the premise that the superconductor is ‘defect free’,
a state difficult to attain in the ‘real’ world. Recently, the applicability of this model has been
questioned. If defects are present at the surface, the penetration of individual vortices can be
considerably faster (it is now estimated at ~10–13 s [81-83]).
Penetration of vortices has now been proposed to explain the appearance of the ‘Q-slope’ [83,
84] – that is, high field dissipation (see below) – and the limitation of the performance of cavities [85].
Vortices have to overcome a surface barrier (resulting from the conjunction of Meissner currents and
surface image vortices), which only disappears at H = HC. The surface barrier is reduced by the
presence of defects [82].
RF 10
When defects are present, then the Meissner state is expected to disappear at H C1 . Note that
RF DC
H C1 is expected to be slightly higher than H C1 [87]. The ultimate field in this case would be directly
linked to the thermodynamic field HC, which is always difficult to measure precisely for type II
superconductors.
DC
The origin of the presence of some vortices between H C1 and HC is not clear. Perhaps we are
11
dealing here with residual magnetic field lines that remained trapped during the cooling of the cavity.
10
A. Gurevich, personal communication.
11
Cryostats are magnetically shielded to protect them from the geomagnetic field.
236
The origin of the ultimate limitations in RF is far from being settled, especially at high fields
and low temperature, where some approximations of the general theory are no longer valid [88]. The
theoretical study of this subject has only just begun, partly because it is only nowadays that the
cavities have intrinsic properties that are good enough to be able to test hypotheses dating back more
than 50 years. However, recent propositions by A. Gurevich seem to point towards early magnetic
field penetration issues and maybe a superheating field is not the right criterion for practical
applications.
4.1.3 Criteria for choosing a ‘good’ RF superconductor

A priori, there are two partially contradictory aspects that come into play – the surface resistance (in
order to have the highest possible quality coefficient Q0) and the ‘superheating’ field HSH, which
affects the maximally possible accelerating field:
ω 2 −∆ / kT
RBCS = A(λ4L , ξ F , , ρ n ) e . (8)
T
We see that in order to decrease RBCS, we need to attain a maximum gap ∆ (which requires a
high TC, because TC ~ 1.87∆/kB), a minimal penetration depth λL and good electric conductivity at
normal ρn. A low RBCS is a necessary but not a sufficient condition to get a low surface resistance. We
must also ensure that we have a very weak residual resistance. For example, high-TC ceramic
superconductors, with their weak links (e.g. grain boundaries in normal state), have a residual
resistance at RF that is ~100 times higher than the resistance of copper. Therefore they are of no
interest for RF applications, despite their high TC [89].
We are therefore striving to develop metal compounds with high values of TC. Unfortunately,
known high-TC compounds often also have a very high λL (see Table 3). Getting a high-gradient
accelerator automatically results from a compromise.
Table 3: The properties of several superconductors
ρn λL
Material TC (K) HC (T)* HC1 (T)* HC2 (T)* Type
(µΩ·cm) (nm)*
Pb 7.1 0.08 n.a. n.a. 48 I
Nb 9.22 2 0.2 0.17 0.4 40 II
NbN 17.1 70 0.23 0.02 15 200 II
NbTiN 16.5 35 0.03 151 II
Nb3Sn 18.3 20 0.54 0.05 30 85 II
V3Si 17 II
Mo3Re 15 0.43 0.03 3.5 140 II
II – two
Mg2B2 40 0.43 0.03 3.5 140
gaps
YBCO 93 1.4 0.01 100 150 d-wave
*At 0 K.
Compounds such as NbN or Nb3Sn do seem very attractive, but the attempts that have been
made to prepare them, either by thermal means (from bulk niobium cavities) or by sputtering, have not
yielded the expected results: they remain far behind bulk niobium. For a review of these trials, see
Ref. [90] and Fig. 52.
237
C. Z. A NTOINE
Fig. 26: A comparison between the maximum magnetic field for a Nb3Sn cavity obtained by Snvap diffusion in
Wuppertal, and a bulk Nb cavity tested at Saclay. The appearance of a slope on the Nb3Sn cavity around 30–
35 mT may be related to the HC1 value of Nb3Sn 50 mT at 0 K, while the slope for Nb around 100–120 mT
might be related to its HC1 value (170 mT at 0 K). The Nb3Sn curve is determined after Ref. [7], with geometrical
ratios Hp/Eacc = 4.7 mT for 1.5 GHz and Hp/Eacc = 4.26 mT for 1.3 GHz.
Now if we consider that the penetration of vortices is promoted by surface defects, a

superheating field is not liable to be reached in realistic conditions. Even on bulk niobium, where
much progress has been done on surface preparation, we have several hints that crystalline defects as
common as dislocations can promote field penetration (see Section 2.6). Even if the density of the
defects can be reduced by appropriate surface preparation, not all of them can be removed. For
instance, the density of dislocations in a well-recrystallized metal is still higher than 104 cm·cm–3 at
room temperature (while it can reach 1012 cm·cm–3 for heavily deformed metal).
The other direction that has been explored consists of depositing a thin film (a couple of
microns) on a copper cavity (by sputtering, plasma, etc.), which is much cheaper and more favourable
from a thermal point of view. These techniques are well suited to flat surfaces. In the cavity, layers
generally come with a columnar structure, with porosity, defects, very small grains (100 nm), high
residual stress, and a HC1 value that is lower than that of the equivalent bulk material. It has been
shown on niobium films (~1–5 µm) that the porosity increases with the roughness of the substrate, and
this seems to cause an increase of the dissipation with the field (see, e.g., Ref. [91] and references
therein). In this configuration, the material is entirely in a mixed state. It has been thought for a long
time that these vortices were well anchored on the material defects and thus hardly dissipated. But as
we have already mentioned before, the recent work of A. Gurevich has shown that the BCS resistance
at high field is highly non-linear, and of magnetic origin [82, 84]. Moreover, grain size seems to
influence the frequency at which vortices can be untrapped by an RF field [92]. This might explain
why the technology of films on copper never produced very high field gradients.
This work perhaps also provides a good explanation why the Nb3Sn cavities did not lead to the
expected results, as we can see in Fig. 26: at low field, the Q0 value is very good, which means that the
residual resistance and the BCS resistance are weak and that we did obtain the desired phase. But after
no more than just a couple of MV·m–1, the Q0 collapses and becomes far worse than that of the Nb.
Note that recent progress in deposition techniques seems to open the way for niobium thin films
with bulk-like properties [93]. It will certainly lower the fabrication cost, but we will still be limited by
the performance of niobium.
238
4.1.4 The importance of HC1

So far, only the superheating field had been considered for the choice of superconductor, because it
was assumed that the Meissner effect would be maintained above HC1. But in realistic conditions,
many surface defects still exist and can ease early penetration of vortices. If the actual limitation in
cavities is related to the penetration of vortices via surface defects, and thus to the HC1 of the material,
it is not surprising that only niobium performs well: niobium has the highest value among all known
superconductors!
It therefore seems unlikely that we will improve the performance of cavities using conventional
superconductors. Only specific structures such as those proposed by Gurevich [94] are likely to go
beyond the performance of niobium.
4.2 Superconducting nano-composites: an innovative path for the future of SRF

The recent work of A. Gurevich on the SRF has opened up a whole new approach, that might allow us
to finally smash the monopoly of bulk niobium [94].
It involves the improvement of the performance of bulk niobium by ‘shielding’ it, using very
thin films. Here, the order of magnitude is no longer the micron, but the nanometre. We will show that
it will not only be possible to obtain higher-gradient accelerators, but also to diminish the cryogenic
losses, depositing 30–50 nm of a high-TC superconductor such as Nb3Sn or MgB2. An insulating layer
of some 5–15 nm is required to decouple the two superconductors (in the sense of a Josephson
junction). The presence of bulk niobium is still necessary, as the multilayers are too thin to fully
screen the magnetic field.
The increase of the transition field originates from a property of very thin films: in this case, the
field around a potential vortex decreases as d/π instead of λ (multiple interactions with the anti-vortex
images), where d is the film thickness and λ is the penetration depth of the field. The lower critical
field is much higher:
Φ0  λ 
H C1 =  ln + 0.5  (9)
4πλ  ξ 
becomes
2Φ 0  d 
H 'C1 = 2 
 ln − 0.07  , (10)
πd  ξ 
where Φ0 is the magnetic flux quantum and ξ is the Cooper pair coherence length. The surface barrier
still exists, but it is also higher:
Φ0
HS ≈ H C = (11)
2 2πλξ
becomes
Φ0
H 'S ≈ H 'C = . (12)
2πdξ
239
C. Z. A NTOINE
As an example:
− for NbN, where ξ = 5 nm, HC1 = 0.02 T, and HC = 0.23 T, a layer of 20 nm will give
H’C1 = 4.2 T and HS = 6.37 T;
− for N3Sn, where ξ = 3 nm, HC1 = 0.05 T, and HC = 0.54 T, a layer of 50 nm will give
H’C1 = 1.4 T;
− recall that for Nb, ξ = 38 nm, HC1 = 0.17 T, and HC = 0.18 T [82, 94].
The surface resistance is also expected to diminish with the use of superconductors with high
TC. High(er)-TC superconductors have a BCS resistance that is much weaker than that of niobium. The
experimental issue is then to find conditions of fabrication that are gentle enough to not introduce too
many defects: extremely uniform layers, clean interfaces, few grain boundaries, little residual stress,
and not too many foreign or displaced atoms, so that the residual resistance stays low. Figures 27 and
28 show recent experimental measurements on samples deposited using multilayers [95-97]. This
topic now needs to be systematically developed, first to determine the optimum number/thickness
distribution for the multilayer and, secondly, to adapt the deposition techniques to the building of such
layers inside cavities.
5 Conclusion
Progress in SRF technology over the past 20 years has led us to the ultimate limits for niobium
technology, laboratory-wise. A lot of work is still needed to be able to produce, at high yield, a large
number of accelerator cavities with the same performance. Mastering the surface state – in terms of
cleanliness, purity and crystalline perfection – seems to be of paramount importance. New
technologies are on the threshold of supplanted bulk niobium. In the short term, new thin-film
technologies will allow bulk-like Nb material to be deposited, rendering the SRF technology much
cheaper, while in the longer term, multilayer structures, applicable to the upgrading of existing Nb
structures, as well to new ones, will hopefully allow us to go towards higher field with reduced
cryogenic losses.
Fig. 27: The magnetic behaviour (HC1) of various single and multilayer samples with 25 nm layers of NbN. The
field at which vortices enter the sample is systematically higher than the Nb-only reference sample,
demonstrating the screening effect of the layers.
240
Fig. 28: The first RF test of a multilayer sample (4 nm × 25 nm of NbN). The samples constitute the removable
bottom of a TE011 3.88 GHz cavity. The comparison is made with a ‘good’ bulk Nb cavity measured at 1.3 GHz
and scaled to 3.88 GHz with an ω2 relationship. At low temperature, the residual resistance is dominant, but at
this stage it is difficult to know its exact origin (indium seal or sample). At temperatures higher than 3 K (BCS
regime), the multilayer sample exhibits a much lower surface resistance than pure Nb. In this configuration, half
of the current flow inside the layers and the remaining half of the current is expected inside the niobium. Other
recent experimental measurements on similar films can be found in, for example, Ref. [98] (NbN) and
Refs. [99, 100] (MgB2).
References
[1] R. Little and W. Whitney, Journal of Applied Physics 34(8) (1963) 2430-2432.
[2] H. Padamsee, J. Knobloch, and T. Hays, RF Superconductivity for Accelerators, 2nd edn
(Wiley-VCH, Weinheim, 2008).
[3] A. Descoeudres et al., Physical Review Special Topics-Accelerators and Beams 12(3) (2009)
032001.
[4] G. R. Werner et al., Voltage breakdown and the processing mechanism, Proc. 11th Workshop on
RF Superconductivity, SRF 2003, Tavemunde, Germany, 2003.
[5] G. R. Werner et al., Voltage breakdowns on Nb and Cu surfaces, Proc. 10th Workshop on RF
Superconductivity, SRF 2001, Tsukuba, Japan, 2001.
[6] A. Zeitoun-Fakiris and B. Juttner, J. Phys. D: Appl. Phys. 24 (1991) 750-756.
[7] T. Wang, C. Reece and R. Sundelin, DC field emission studies on Nb, Proc. 10th Workshop on
RF Superconductivity, SRF 2001, Tsukuba, Japan, 2001.
[8] I. Brodie, Journal of Applied Physics 35(8) (1964) 2324-2332.
[9] J. Tan, Etude de l'émission électronique par effet de champ en haute fréquence, thesis,
Université Paris-6, 1995.
[10] G.R. Werner, Probing and modeling voltage breakdown in vacuum, Thesis, Cornell University,
2004.
[11] E. Mahner et al., Field emission measurements on niobium cathodes of high purity, Proc. 5th
Workshop on RF Superconductivity SRF 1991, Hamburg, Germany, 1991.
[12] R. Noer et al., Journal of Applied Physics 59(11) (1986) 3851-3860.
[13] P. Niedermann et al., Journal of Applied Physics 59(3) (1986) 892-901.
241
C. Z. A NTOINE
[14] T. Junquera et al., Study of luminous spots observed on metallic surfaces subjected tohigh RF
fields, Proc. of the 1995 Particle Accelerator Conference, 1995, p. 1632.
[15] S. Maissa, Etude des émissions lumineuses associées aux émissions électroniques dans les
cavités hyperfréquences, thesis, Orsay Paris XI, France, 1996.
[16] J.M. Jimenez, Etude de l'émission électronique par effet de champ en courant continu sur des
cathodes étendues, thesis, Clermont-II, France, 1994.
[17] P. Niedermann and O. Fischer, IEEE Trans. Dielectr. Electr. Insul. 24(6) (1989) 905–915.
[18] E.S. Crawford and J.S. Anderson, Phil. Trans. R. Soc. Lond. A (Math. Physi. Sci.) 304(1485)
(1982) 327–364.
[19] J. Halbritter, Appl. Phys. A 43 (1987) 1–28.
[20] M.L.A. Robinson and H. Roetschi, J. Phys. Chem. Solids, 29 (1968) 1503–1510.
[21] J.S. Sheasby and B. Cox, J. Less-Common Metals 15 (1968) 129–135.
[22] C. Chianelli et al., Very low current field electron emission from anodized niobium, Proc. 5th
Workshop on RF Superconductivity, DESY, Hamburg, Germany (1991).
[23] M. Jimenez et al., J. Phys. D: Appl. Phys. 26(9) (1992) 1503–1509.
[24] M. Jimenez and R. Noer, J. Phys. D: Appl. Phys. 27 (1994) 1038–1045.
[25] M. Luong, Etude émission électronique par effet de champ sur des surfaces larges en régime
statique et hyperfréquence, thesis, Universite Paris VI – Pierre et Marie Curie, 1997.
[26] C.Z. Antoine, F. Peauger, and F. Le Pimpec, Nucl. Instrum. Methods Phys. Res. A 670 (2012)
79–94.
[27] C.Z. Antoine et al., Dust contamination during chemical treatment of RF cavities: symptoms
and cures, Proc. 5th Workshop on RF Superconductivity, DESY, Hamburg, Germany (1991).
[28] C.Z. Antoine et al., Avoiding dust contamination during chemical treatment of RF cavities? 3rd
EPAC, Berlin (1992).
[29] C.Z. Antoine et al., Rôle de la contamination dans les cavités pour accélérateurs de particules,
Contamin Expert '93, Paris (1993).
[30] C.Z. Antoine, Statistical analysis of the risk of dust contamination during assembling of RF
cavities, Proc. 6th Workshop on RF Superconductivity, CEBAF, USA (1993).
[31] J. Martignac et al., Particle contamination in vacuum systems, Proc. 7th Workshop on RF
Superconductivity, DIST CEA, Gif sur Yvette (1995).
[32] H. Safa, High field behavior of SCRF cavities, Proc. 10th Workshop on RF Superconductivity,
Tsukuba, Japan (2001).
[33] H. Safa, An analytical approach for calculating the quench field in superconducting cavities,
Proc. 7th Workshop on RF Superconductivity, DIST CEA, Gif sur Yvette (1995).
[34] F. Koechlin and B. Bonin, Supercond. Sci. Tech. 9 (1996) 453–460.
[35] G. Issarovitch et al., Development of centrifugal barrel polishing for treatment of
superconducting cavities, Proc. 11th Workshop on RF Superconductivity, SRF 2003, Lübeck,
Germany (8–12 September 2003).
[36] T. Higuchi et al., Investigation on barrel polishing for superconducting niobium cavity, Proc.
7th SRF Int. Workshop, DIST CEA, Gif-sur-Yvette (1995)
[37] K. Saito, Recent developments in SRF cavity cleaning techniques at KEK, SRF 1999, Santa Fe
(1999).
[38] C. Cooper, Centrifugal barrel polishing of cavities worldwide, Proc. 15th Int. Conf. on RF
Superconductivity, Chicago (2011).
[39] C. Cooper and L. Cooley, Mirror smooth superconducting rf cavities by mechanical polishing
with minimal acid use, Fermi National Accelerator Laboratory (FNAL), Batavia, IL (2011).
242
[40] T. Higuchi and K. Saito, Development of hydrogen-free EP and hydrogen absorption

phenomena, Proc. 11th Workshop on RF Superconductivity, SRF-2003, Lübeck, Germany (8–
12 September 2003).
[41] W.A. Shewhart, Economic Control of Quality of Manufactured Product, vol. 509 (American
Society for Quality Control, Milwaukee, WI, 1980; first published Van Nostrand, Toronto,
1931).
[42] T. Bieler (MSU), K.T. Hartwig (Texas AM), and R. Crooks (Black Lab), personal
communications.
[43] C.Z. Antoine and R. Crooks, Reducing electropolishing time with chemical–mechanical
polishing, Proc. 14th Int. Conf. on RF Superconductivity, SRF 2009, DBB Forum, Berlin (20–
25 September 2009);
available at http://accelconf.web.cern.ch/AccelConf/srf2009/html/author.htm
[44] A. Romanenko, Review of high field Q-slope, surface measurements, Proc. 13th Workshop on
RF Superconductivity, Beijing, China (2007).
[45] A. Romanenko and H. Padamsee, The role of near-surface dislocations in the high magnetic
field performance of superconducting niobium cavities, Supercond. Sci. Tech. 23 (2010)
045008.
[46] O.S. Romanenko, Surface characterization of Nb cavity sections – understanding the high field
Q-slope, Thesis, Cornell University, 2009.
[47] A. Grassellino, Muon spin rotation/relaxation studies of niobium for SRF applications, Proc.
15th Int. Conf. on RF Superconductivity, Chicago (2011).
[48] J. Zasadzinski et al. Raman spectroscopy as a probe of surface oxides and hydrides of niobium,
Proc. 15th Workshop on RF Superconductivity, SRF 2011, Chicago, IL, USA (2011).
[49] T. Proslier et al., Appl. Phys. Lett. 92 (2008) 212505.
[50] H. Shiba, Prog. Theor. Phys. 40 (1968) 435; Prog. Theor. Phys. 50 (1973) 50.
[51] T. Proslier et al., IEEE Trans. Appl. Supercond. 99 (2011) 2619–2622.
[52] T. Proslier et al., Localized magnetism on the surface of niobium: experiments and theory, Bull.
Am. Phys. Soc. 56 (2011).
[53] S. Casalbuoni et al., Nucl. Instrum. Methods Phys. Res. A 538 (2005) 45–64.
[54] M. Delheusy et al., Appl. Phys. Lett. 92 (2008) 101911.
[55] R.J. Cava et al., Phys. Rev. B 44(13) (1991) 6973–6981.
[56] C. Antoine, Materials and surface aspects in the development of SRF Niobium cavities.
EUCARD series on Accelerator Science, ed. R.S. Romaniuk and J.P. Koutchouk. Vol. 12, 2012.
[57] F.P.-J. Lin and A. Gurevich, Effect of impurities on the superheating field of Type II
superconductors, Bull. Am. Phys. Soc. 57(1) (2012).
[58] M. Kharitonov et al., Surface impedance of superconductors with magnetic impurities, ArXiv
preprint arXiv:1109.3395 (2011).
[59] J.P. Mercier, W. Kurz, and G. Zambelli, Introduction à la science des matériaux, vol. 1 (PPUR,
Lausanne, 1999).
[60] L.E. Samuels, Metallographic Polishing by Mechanical Methods, 4th edn (electronic resource)
(ASM International, Materials Park, OH, 2003).
[61] S. Hashimoto, S. Miura, and T. Kubo, Dislocation etch pits in gold, J. Mat. Sci. 11(8) (1976)
1501–1508.
[62] X. Zhao, G. Ciovati, and T. Bieler, Characterization of etch pits found on a large-grain bulk
niobium superconducting radio-frequency resonant cavity, Phys. Rev. ST–AB 13(12) (2010)
124701.
243
C. Z. A NTOINE
[63] J. Knobloch et al., High-field Q slope in superconducting cavities due to magnetic field
enhancement at grain boundaries, Proc. 9th Workshop on RF Superconductivity, Santa Fe, NM
(1999).
[64] B. Visentin, Review on Q-DROP mechanisms, Proc. Int. Workshop on Thin Films and New
Ideas for Pushing the Limits of RF Superconductivity, Legnaro National Laboratories (Padua),
Italy (2006).
[65] H. Tian et al., A novel approach to characterizing the surface topography of niobium
superconducting radio frequency (SRF) accelerator cavities, Appl. Surf. Sci. (2010).
[66] C. Roques-Carmes et al., Int. J. Mach. Tools Manufact. 38(5–6) (1998) 573–579.
[67] M. Ge et al., Supercond. Sci. Tech. 24 (2011) 035002.
[68] S. Berry et al., Topologic analysis of samples and cavities: a new tool for morphologic
inspection of quench site, Proc. 11th Workshop on RF Superconductivity, SRF-2003, Lübeck,
Germany (8–12 September 2003).
[69] S. Berry, C. Antoine, and M. Desmons, Surface morphology at the quench site, EPAC 2004,
Lucern, Switzerland (2004).
[70] A.A. Polyanskii et al., Review of magneto-optical result on high purity Nb for superconducting
RF application, Proc. Int. Workshop on Thin Films and New Ideas for Pushing the Limit of RF
Superconductivity, Legnaro National Laboratories (Padua), Italy (2006).
[71] R. Ricker and G. Myneni, J. Res. NIST 115(5) (2010) 353–371.
[72] V. Shemelin and H. Padamsee, Magnetic field enhancement at pits and bumps on the surface of
superconducting cavities, TTC-Report 2008-007.
[73] W. Singer et al., Hydroforming of NbCu clad cavities at DESY, Proc. 10th Workshop on RF
Superconductivity, Tsukuba, Japan (2001).
[74] X. Singer et al., Hydroforming of multi-cell niobium and NbCu-clad cavities, Thomas Jefferson
National Accelerator Facility, Newport News, VA (2009).
[75] C.Z. Antoine, RF material investigation by sample analysis, Particle Accelerators 60(1–4)
(1997).
[76] D. Landolt, Electrochim. Acta 32(1) (1987) 1–11.
[77] A. Valente-Feliciano et al., Development of Nb and alternative material thin films tailored for
SRF applications, Proc. Int. Part. Accel. Conf., New Orleans, Louisiana (2012), pp. 2444–2446.
[78] C. James et al., Superconducting Nb thin films on Cu for applications in SRF accelerators, IEEE
Trans. Appl. Supercond. 23(3) (2013) 3500205.
[79] A. Anders, Deposition of niobium and other superconducting materials with high power
impulse magnetron sputtering: concept and first results, Proc. 15th Int. Conf. on RF
Superconductivity, Chicago (2011).
[80] T. Yogi, G. Dick, and J. Mercereau, Phys. Rev. Lett. 39(13) (1977) 826–829.
[81] E.H. Brandt, Electrodynamics of superconductors exposed to high frequency fields, Proc. Int.
Workshop on Thin Films and New Ideas for Pushing the Limit of RF Superconductivity,
Legnaro National Laboratories (Padua), Italy (2006).
[82] A. Gurevich, Physica C 441(1–2) (2006) 38–43.
[83] A. Gurevich and G. Ciovati, Phys. Rev. B 77(10) (2008) 104501–104521.
[84] P. Bauer et al., Physica C 441 (2006) 51–56.
[85] K. Saito, Theoretical critical field in RF application, Proc. 11th Workshop on RF
Superconductivity, SRF-2003, Lübeck, Germany (8–12 September 2003).
[86] A. Gurevich, personal communication.
[87] G. Lamura et al., J. Appl. Phys. 106 (2009) 053903.
244
[88] See e.g

http://tdserver1.fnal.gov/project/workshops/RF_Materials/talks/Hasan_Sethna_superheating%2
0field.ppt
[89] H. Piel, Nucl. Instrum. Methods Phys. Res. A 287(1–2) (1990) 294–309.
[90] http://tdserver1.fnal.gov/project/workshops/RF_Materials/talks/A-M_Valente-
Feliciano_NewMaterialsOverview.ppt
[91] M. Fouaidy et al., New results on RF properties of superconducting niobium films using a
thermometric system, Proc. EPAC, European Physical Society, Paris (2002).
[92] M. Mathur, D. Deis, and J. Gavaler, J. Appl. Phys. 43(7) (1972) 3158–3161.
[93] See e.g. presentations from Anders, Krishnan or Valente- Feliciano at thin film workshop 2010
or SRF 2011
[94] A. Gurevich, Appl. Phys. Lett. 88 (2006) 012511.
[95] C.Z. Antoine, J.C. Villegier, and G. Martinet, Appl. Phys. Lett. 102(10) (2013) 102603.
[96] C.Z. Antoine et al., IEEE Trans. Appl. Supercond. 21(3) (2011) 2601–2604.
[97] C.Z. Antoine et al., Phys. Rev. ST–AB 13 (2010) 121001.
[98] W. Roach et al, IEEE Trans. Appl. Supercond. 23(3) (2013) 8600203.
[99] T. Tajima et al., MgB2 thin film studies, Proc. 15th Int. Conf. on RF Superconductivity,
Chicago (2011), pp. 287–292.
[100] C. Zhuang et al., MgB2 thin films on metal substrates for superconducting RF cavity
applications, J. Supercond. Novel Magnetism (2013) 1–6.
245
Superconductivity for Magnets
R. Flükiger 1
Abstract
The present state of development of a series of industrial superconductors is
reviewed in consideration of their future applications in high field
accelerator magnets, with particular attention on the material aspect. The
discussion is centred on Nb3Sn and MgB2, which are industrially available in
a round wire configuration in kilometre lengths and are already envisaged
for use in the LHC Upgrade (HL-LHC). The two systems Bi-2212 and R.E.-
123 may be used in magnets with even higher fields in future accelerators:
they are briefly described.
Keywords: systems, Nb3Sn, MgB2, Bi-2212, R.E.-123, superconducting

wires, critical current densities.
1 Introduction
Superconducting high field magnets are based on wires or tapes consisting of a variety of materials. In
this manuscript, selected current-carrying properties of various commonly used superconducting wires
for accelerators are described. The main subjects will be the two systems Nb3Sn and MgB2; round
wires in these materials can already be produced industrially in long lengths. A brief presentation of
the two other known systems, Bi-2212 and YBaCuO, will follow. Bi-2223 tapes are also available in
kilometre lengths, but are not envisaged for use in future accelerators and are not included here.
2 Multifilamentary Nb3Sn wires: preparation techniques
Nb3Sn is in the cubic phase, with Tc = 18.2 K, and belongs to the category of Low-Temperature
Superconductors (LTS). Its upper critical field Bc2 varies between 25 T (measured in binary Nb3Sn)
and 30 T (after alloying with Ti or Ta). Industrial Nb3Sn wires always contain a certain content of
additives (~1.5 at.% Ti or ~3 at.% Ta) in order to achieve a maximum value of Bc2.
Nb3Sn multifilamentary wires can be prepared using several techniques. The most common
ones are (a) the ‘bronze route’ (the reaction of Cu-Sn bronze and Nb), (b) the Powder-In-Tube (PIT)
technique (where Nb3Sn is formed by a reaction between Nb and NbSn2), and (c) the RRP technique
(a direct reaction between Nb and Sn, based on the internal Sn diffusion mechanism). In contrast to the
bronze route, the two other techniques, PIT and RRP, do not require intermediate heat treatments, as
represented schematically in Fig. 1.
1
rene.flukiger@cern.ch
R. F LUKIGER
Fig. 1: Schematic representation of the processing for bronze route and RRP Nb3Sn wires
Out of the three processing methods followed for the fabrication of Nb3Sn wires, only the two
with the highest Tc values have been retained at CERN for further studies in consideration of the LHC
Upgrade (or HL-LHC) quadrupoles: the RRP wires from Oxford Instruments (USA) and the PIT wires
from Bruker (Hanau, Germany) (see Figs. 2 and 3). The investigations at CERN involved wires with
diameters between 0.7 and 1.0 mm, with Cu/non-Cu ratios ranging from 1.15 to 1.25, with unit lengths
reaching 800 m. The RRR ratio, important for the thermal stability of the wire, was 150 in both types
of wire. The filament diameters were between 40 and 60 mm for the RRP wires and between 35 and
50 mm for the PIT wires.
Fig. 2: Cross-sections of (a) bronze route, (b) PIT, and (c) RRP multifilamentary Nb3Sn wires. The non-Cu
critical current densities up to 16 T are shown in (c) for RRP wires. (Courtesy of J.M. Parrel, OST.)
Fig. 3: The two Nb3Sn wire types tested at CERN in view of LH-LHC quadrupoles: RRP wires from Oxford
Instruments (USA) and PIT wires from Bruker (Hanau, Germany).
248
S UPERCONDUCTIVITY FOR M AGNETS
2.1 Uniaxial tensile stress

In view of the use of Nb3Sn wires in large magnets with very high currents and fields, the question
arises about the effect of the strong Lorentz forces under these extreme conditions. The 3D stress
situation in a wire has been treated by analysing two cases: (a) the uniaxial tensile stress and (b) the
compressive stress. Although Nb3Sn has the highest electronic density of states, N(EF), among all
A15-type compounds, it is thought that the variation of this property is of minor importance in
understanding the effects of Lorentz forces. Indeed, it has been shown [1] that the effect of uniaxial
applied stress on Jc is mainly correlated to changes in the phonon spectrum. It was found that the main
variation of Jc under uniaxial tensile stress is primarily sensitive to non-hydrostatical stress
components, the hydrostatical components having a markedly lower influence.
It has been shown by neutron diffraction measurements [2] that, during the cooling process, the
A15 lattice undergoes an increasing degree of elastic distortion into a tetragonal phase: the distortion
starts at ~500 K and is maximum at 4.2 K, where (1 − c/a) ~ 0.2%. When applying a stress in the axial
direction, it was thus argued that the precompression would gradually diminish, to reach a ‘stress-free’
state at a given stress value. This was confirmed by a direct measurement of the crystallographic
changes of the A15 phase inside the filaments when applying tensile stresses at 10 K, first performed
by means of X-ray diffraction, on a flat, monofilamentary bronze route Nb3Sn wire [2]. The bronze:Nb
ratio of this sample was 7:1, which explains the relatively high value of εm = 0.5% (Fig. 4), at which
the measured value of Ic reaches a maximum (also shown in Fig. 4). This particular design was chosen
for the direct observation of the effect by diffractometry. The same authors showed later by neutron
diffraction that this behaviour is also representative for multifilamentary Nb3Sn wires.
Fig. 4: Critical current Ic and lattice parameters a and c on a monofilamentary flat Nb3Sn wire vs. the uniaxial
applied strain ε. The variation of a is directly measured, and c is calculated after correction for the volume
compression ∆V(ε): ∆V(0) = 0.6% and ∆V(1.0%) = 0.8% [2].
Various measuring devices have been developed to measure Jc(ε), the variation of Jc as a
function of the applied strain ε: the linear strain rig [3], the Pacman (University of Twente), and the
modified Walters spiral, which was developed at the University of Geneva [4] and later also at NIST,
Boulder, CO, USA. The Walters spirals for the measurement of Jc(B,ε) on wires and tapes are shown
in Fig. 5.
249
R. F LUKIGER
Wire
Tape
Fig. 5: Modified Walters spirals developed at the University of Geneva to measure the effect of uniaxial tensile
strain on round Nb3Sn wires (left) and R.E.-123-coated conductor tapes (right) up to 21 T at 4.2 K. The
characteristics of both Walters spirals are Ic ≤ 1000 A, with wire lengths ≤ 0.8 m, thus allowing a strong Ic
criterion: 0.1 or 0.01 µV·cm−1 [4].
The uniaxial force on the wire or tape is obtained by a rotation of the central axis of the spiral.
The variation of Jc(ε) can be understood by considering that the Nb3Sn filaments in the wire are under
precompression, as a result of the differential thermal contraction between the Nb3Sn phase and the
surrounding Cu or Cu-Sn bronze. Since the linear thermal contraction value α of Nb3Sn is
8 × 10−6 K−1, i.e. considerably lower than the value for Cu or Cu-Sn (18 × 10−6 K−1), the cooling from
~1000 K (the reaction temperature) to 4.2 K (the operation temperature) results in a compression of
the matrix on the filaments (precompression). The resulting reversible distortion has a strong effect on
the physical properties, in particular on the critical current density values Jc. For a given temperature,
the latter can now be described as depending on two parameters: Jc(B,ε) (see Fig. 6).
Fig. 6: Normalized Ic vs. applied strain for a RRP Nb3Sn wire at T = 4.2 K and for different fields. Ic behaves
reversibly in the shown strain window (εm = 0.25%). The critical currents at zero applied strain (0.1 µV·cm−1)
are marked by the dashed line: 278.4 A (12 T), 156.6 A (15 T), 95.9 A (17 T), 50.4 A (19 T). (From [5].)
The question about the possible effect of uniaxial tensile stresses on flux pinning can be
answered by representing the normalized pinning force Fp/Fp,max as a function of the reduced magnetic
field b = B/Bc2; as shown [5] for a RRP wire, all data fall on a universal curve, reflecting that the
pinning mechanism is not influenced by the application of uniaxial tensile strain (see Fig. 7). This
confirms that the effects of uniaxial stress on Jc are essentially due to the change of Bc2.
250
Fig. 7: Normalized pinning force vs. reduced magnetic field b = B/Bc2 for a RRP Nb3Sn wire from Oxford
Instruments (OST). All data fall on a universal curve, indicating that the pinning mechanism is not influenced by
the application of uniaxial tensile strain. (From [6].)
It can be shown that a stronger criterion for Ic (defined by µV·cm−1) leads to an earlier detection
of nanocracks: the longer wire length in the Walters spiral allows a deeper insight, as shown by Fig. 8,
on a bronze route Nb3Sn wire of 0.8 mm diameter produced by Furukawa for ITER [6]. It follows that
the early detection of the irreversible strain limit εirr (the beginning of the nanocrack formation)
depends on the Ic criterion. It follows from Fig. 8 that Ic does not depend on the strain criterion up to
0.6–0.7%. However, the effects of the various criteria are seen after releasing the strain ε, as shown
in Fig. 8.
Fig. 8: Variation of Ic/Ico vs. uniaxial strain ε at 4.2 K and 13 T for a Furukawa bronze route wire, showing the
earlier detection of the irreversible strain limit due to nanocracks for the stronger Ic criterion. (From [6].)
A correlation between different parameters, e.g. temperature, field and strain dependence
superconducting parameters, can be found by a parametrization proposed by L. Bottura (CERN),
starting with a formula due to Summers:
2 2
CNb3Sn (ε )  B    T 2 
Jc ( B, T , ε ) = 1 −  1 −    ,
B  Bc2 (T , ε )    Tc0 (ε )  
251
R. F LUKIGER
Bc2 (T , ε )   T     T   
2 2
 T  
=1 −     1 − 0.31   1 − 1.77 ln    ,
Bc20   Tc0 ( ε )     T c0 ( ε )  
   T c0 ( ε )  
  
 
(
CNb3Sn (ε ) = CNb3Sn,0 1 − α Nb3Sn ε )
1.7 1/2
,
c20 ( ε )
B= (
Bc20m 1 − α Nb3Sn ε
1.7
),
Tc0
= (
(ε ) Tc0m 1 − α Nb3Sn ε )
1.7 1/3
,
where α Nb3Sn is 900 for ε = −0.003, Tc0m is 18 K, Bcm0 is 24 T, and CNb3Sn,0 is a fitting parameter equal
to 60 800 AT1/2·mm−2 for a value of Jc = 3000 A·mm−2 at 4.2 K and 12 T.
2.2 Effect of transverse stresses

The effect of transverse stress on the superconducting properties, and in particular on the current-
carrying capability, is much stronger than that of uniaxial stress. This is illustrated in Fig. 9, which
shows the dependence of the ratio Ic/Ic0 of a bronze route wire as a function of the applied stress,
where σa is the uniaxial stress and σt is the transverse stress. It is clearly seen that a maximum of Ic/Ic0
is reached for transverse stresses σt of the order of 20 MPa, in contrast to uniaxial stresses, where σa is
observed at 150 MPa; thus, σt(Icm) << σa(Icm) [7].
A particularly important problem that arises when considering the use of Nb3Sn wires in
accelerators concerns their behaviour under the effect of transverse stress produced by the large
Lorentz forces. This problem has been studied at the University of Geneva, within the framework of
the Ph.D. work [8] by G. Mondonico, who studied the variation of Jc in three load cases. These cases
are:
1) a wire pressed between two parallel walls (this is the worst possible case and is never
encountered);
2) a wire simultaneously pressed by four walls without epoxy;
3) same as case 2), but with epoxy (thus simulating the stress on a wire in a Rutherford cable).
Fig. 9: Comparison between uniaxial tensile stress and compressive stress on the same bronze route wire,
showing a considerably stronger effect due to the compressive stresses [7].
252
The result of this investigation is represented in Fig. 10; it shows that the epoxy has a very
beneficial effect, rendering the stress distribution more hydrostatic, thus reducing the transverse stress
acting directly on the wires in the Rutherford cable.
Fig. 10: Ic/Ic0 vs. σt at 19 T and 4.2 K for the PIT wire in the three different compressive stress configurations.
The irreversible limits are indicated by dashed lines. (From [8].)
2.3 Conclusions about Nb3Sn wires

As the present contribution is in the form of a lecture, it contains a limited choice of subjects and
references, just necessary to present the main physical properties of Nb3Sn wires. The main features of
the behaviour of Nb3Sn wires can be summarized as follows.
• Small solenoids for NMR based on Nb3Sn can reach 23.5 T (1 GHz). In large dipoles or
quadrupoles (e.g. for LHC Upgrade), the achievable field is limited to 13–15 T, due to the
limited available space, large Lorentz forces, and thermal stability requirements.
• The amount of Nb3Sn wire in a magnet increases strongly with the produced field: at 20 T, it
is five times more than for 12 T.
• The critical current density at small fields is dictated by flux pinning due to A15 grain
boundaries (size: 50–100 nm). At high fields, Jc is mainly influenced by the value of Bc2.
• Both internal Sn (RRP) and Powder-in-Tube (PIT) wires satisfy or almost satisfy the
conditions for LHC Upgrade accelerator magnets: Jc = 1500 A·mm−2 at 4.2 K/15 T. Bronze
route wires do not reach these high Jc values, but are best suited for the ‘persistent mode’
operation of NMR magnets.
3 The MgB2 system

The MgB2 phase (Tc = 39 K) crystallizes into a hexagonal lattice (Fig. 9), with a = 0.30834 nm and c
= 0.35213 nm. This compound has been chosen for the high current connection between the surface
and the dipoles and quadrupoles of the LHC Upgrade. These connections will carry a current of
13 000 A, at an operation temperature of 10 K. The phase diagram of the Mg-B system (in Fig. 11)
shows the presence of a gas phase, which has to be taken into account when preparing MgB2 wires.
253
R. F LUKIGER
Fig. 11: The hexagonal structure of the MgB2 phase
Fig. 12: The Mg-B high-temperature phase diagram, showing a gaseous domain above 1100°C
The superconducting properties of the compound MgB2 are anisotropic, similar to those of
High-Temperature Superconductor (HTS) compounds. However, the anisotropy of MgB2 is much
smaller, and its behaviour is more like that of Low-Temperature Superconductor (LTS) compounds.
MgB2 has some peculiar properties; in particular, it exhibits two superconducting gaps. This important
fact has been proven by a series of physical measurements, e.g. by low-temperature calorimetry, and
has been theoretically described. For the description of the current-carrying capability, however, the
larger gap is the more important one, and most superconducting properties can be described by the
BCS model, where the electron–phonon interaction leads to Cooper pairs.
3.1 The irreversibility field, Birr

An ‘irreversibility field’ Birr can be found in high-temperature superconductors which is markedly
different from the value of the upper critical field Bc2. This is not the case for Nb3Sn, where Birr and Bc2
almost fall together. The determination of Birr can be based on the measurements of the electrical
resistivity R(T). The definition is somewhat arbitrary: usually, the values are taken at 10% and 90%, or
at the normal-state resistivity value Rn, for Bc2 and Birr, respectively. The dependence of Bc2 and Birr on
the temperature for binary ex situ MgB2 wires is illustrated in Fig. 13.
Fig. 13: Dependence of Birr(T) and Bc2(T) in a binary MgB2 wire (field // to ab plane)
254
3.2 The anisotropy of Bc2 in MgB2

The physical properties of MgB2 show a marked anisotropy. As can be seen from Fig. 14 for single
crystals, the value of Bc2⊥ perpendicular to the ab planes is markedly lower than Bc2// , the value
parallel to ab. In carbon alloyed MgB2, the anisotropy is lowered. At sufficiently high carbon contents
(too high for the optimization of the critical current density) it disappears.
.
Fig. 14: Bc2(T) for Nb−Ti, Nb3Sn, and alloyed MgB2 single crystals. The latter show a marked anisotropy with
respect to the applied field.
Fig. 15: Anisotropy of the upper critical field Bc2 of binary MgB2 wires
3.3 Effect of alloying on Jc

A strong enhancement of Jc was observed after adding SiC nanopowders to the initial Mg+B powder
mixtures [9]. The effect of SiC additives on the transport Jc of MgB2 wires at 4.2 K is shown in Fig.
16: Jc increases at high fields, but decreases at low fields. The effect of additives on Jc of MgB2 has
been the subject of numerous investigations [10]. Single-crystal observations show that the
substitution of C on the B lattice sites leads to an enhanced residual resistivity ρ0, which is correlated
to the observed enhancement of the upper critical field, Bc2, and the irreversibility field, Birr, in MgB2.
The enhancement of ρ0 with lattice disorder in alloyed MgB2 wires was clearly recognized as the
dominant effect enhancing the critical current density, Jc, at high magnetic fields. The vortex pinning
behaviour of MgB2 is only slightly affected by the presence of C on the B sites. This behaviour is
similar to that encountered in Nb3Sn: alloying by Ta and Ti leads to a reduction of the electronic mean
255
R. F LUKIGER
free path and thus to an enhancement of Bc2, with the consequence of an enhancement of Jc at high
fields. At lower fields, however, Jc is always lower in alloyed wires [11]. The reasons for the decrease
of pinning force at low fields are not explicitly known. However, there are reasons to think that the
grain boundaries are affected by the presence of additives, in analogy to the case of Nb3Sn wires.
There is a direct consequence of the behaviour illustrated in Fig. 16 in view of the LINK application in
the LHC Upgrade: since the MgB2 wires in this application at 10 K will be exposed to very low fields
only, binary rather than ternary alloyed MgB2 wires constitute the logical choice. In this particular
case, the preference will be given to ex situ wires [12], which in addition are also mechanically the
most robust ones.
Fig. 16: Comparison between the variation of Jc vs. the applied field B for binary and SiC added in situ MgB2
wires [11].
3.4 Fabrication of MgB2 wires

Three main routes for the fabrication of MgB2 wires are established, based on the Powder-in-Tube
(PIT) method: (a) the ex situ method [13] (Columbus, Genoa, Italy), (b) the in situ method [14]
(Hypertech, Columbus, Ohio, USA), and (c) the Internal Mg Diffusion process (IMD) [15, 16]. Wires
of several kilometre lengths have been fabricated by both ex situ and in situ processing; IMD wires are
still being optimized. The fabrication steps are shown schematically in Fig. 17, which emphasizes the
fact that the main difference between ex situ, in situ, and IMD processing resides in the initial powder
mixture introduced to the metallic tubes. The tubes may consist of Fe, Ni or Monel, with Nb as a
barrier material.
Recently, it was found that tubes consisting of pure Ti do not require a barrier, which is an
advantage for the fabrication procedure.
In ex situ wires, the powder mixture consists of previously reacted MgB2 powder particles
annealed at 900°C at the end of the deformation process [13]. The in situ process starts with a mixture
of Mg and B powders, the reaction at 650°C occurring at the end of the deformation process [14]. In
the IMD process, first introduced by G. Giunchi [16], Mg rods are introduced inside a ‘tube’
consisting of pressed boron powder, the reaction occurring after deformation. The main processing
steps of bundling and deformation to the wire are similar for the three methods. There is an important
difference between the mass density inside the MgB2 filaments obtained by these three methods: in
IMD wires the mass density is highest and is very close to 100%. The mass density in ex situ and in
situ wires is considerably smaller: ~75% and ~50%, respectively. It will be shown in the following
paragraph that in MgB2 wires—prepared by powder metallurgical processing—the mass density is
correlated to Jc.
256
Fig. 17: The three different methods of MgB2 wire fabrication by powder metallurgical processing
Fig. 18: Typical cross-sections of MgB2 multifilamentary wires prepared by three different industrial processes:
(a) ex situ, (b) in situ, and (c), (d) IMD processing [11].
Ex situ wires can also be prepared in a round shape. Various MgB2 wire configurations tested at
CERN for the LINK project in LHC Upgrade are shown in Fig. 17 (from the presentation by
A. Ballarino at this CAS).
.
Fig. 19: Various cross-sections of round, Cu-stabilized ex situ MgB2 wires produced by Columbus for the LINK
project at CERN [17].
3.5 Effect of cold pressing on Jc of MgB2 wires

Wires with powder metallurgically prepared filaments contain a certain amount (depending on the
fabrication path) of voids. As mentioned in Section 3.4, IMD wires present almost no voids, but in ex
situ and in situ wires they are of the order 25 and 50 vol.%, respectively. Attempts to increase the
mass density by performing the final heat treatment under high Ar pressure (0.2 MPa) or in hard metal
anvils (>10 GPa) yielded a moderate increase of Jc only: these methods are not applicable for
industrial purposes because the gain in Jc is too small to cover the costs of the pressure application. An
alternative to these processes consists in submitting the wires to a high pressure at room temperature
after the final deformation, just before the high-temperature treatment. Cold high-pressure processing
[18], consisting of a simultaneous application of high pressure on four sides of square wires, has led to
an enhancement of mass density (up to 90%) and thus of Jc for in situ MgB2 wires [19], as a result of
better connectivity. As shown on the left of Fig. 20, an enhancement of Jc at 4.2 K by a factor 2 at all
fields was obtained after cold pressing at 1.48 GPa; the same Jc value can be obtained at fields 2 T
257
R. F LUKIGER
higher after cold pressing. On the right of Fig. 20, considerably higher enhancements of Jc at 20 and
25 K are obtained on cold pressed wires, by factors 5 and 8, respectively.
Fig. 20: Variation of the fields B(104)// (field values where Jc = 1 × 104 A·cm−2) at 4.2, 20, and 25 K) for C4H6O5
alloyed MgB2 tapes (Monel sheath, Nb barrier) after cold densification at various pressures. Reaction conditions:
600°C/4 h (600°C/4 h) [19].
3.6 Present state of Jc vs. B in MgB2 wires

The highest values of Jc known so far have been obtained by Li et al. [20], who used a modified IMD
method, yielding almost 100% dense filaments. Their values are shown in Fig. 21. The highest value
is Jc = 1.1 × 105 A·cm−2, corresponding to Jc,eng = 1.67 × 104 A·cm−2at 10 T, where Jc,eng is the overall
critical density. It follows that MgB2 magnets could reach 10 T at 4.2 K. Based on the present data, it
is expected that at 20 K magnets producing fields up to 5 T could also be fabricated. Each one of the
three known methods used for the fabrication of industrial MgB2 wires presents individual advantages.
Ex situ wires show a high homogeneity over long wire lengths and have their best performance at low
fields, whereas the advantage of in situ and IMD processing consists of an easier introduction of
carbon-based additives and thus in higher Jc values at high fields.
Fig. 21: Critical current density of MgB2 prepared by a modified IMD technique. The left figure shows Jc in the
filament; the one on the right shows the engineering Jc [19].
258
The present limits of Jc can be estimated from the data for highly textured thin films: Jc// values
for highly textured thin films with B//ab are very similar to those of Nb3Sn; on the contrary, Jc for the
same thin films, but measured with the field perpendicular to the ab plane, exhibits the lowest values
of all. From Fig. 22, it can be estimated that the expected value of Jc for a round, alloyed MgB2 wire at
high field, e.g. 16 T, will not exceed 1 × 104 A·cm−2: this indicates that the values of Li et al. [19] in
Fig. 19 for fully dense IMD wires are very close to the maximum that can be reached with this
compound. Possible ways of increasing Jc by artificial pinning are still under study.
Fig. 22: Comparison of Jc vs. B for MgB2 wires and thin films. Thin films: Matsumoto et al. [21]; in situ tapes:
Hässler et al. [22]; cold pressed in situ wires: Flükiger [18]; IMD wires: Togano [15].
3.7 Conclusions for MgB2 wires

In spite of the relatively low value of Bc2 and the inherent anisotropy of the MgB2 phase with respect
to other superconductors, the following advantages of MgB2 wires can be recognized.
Advantages:
• abundant constituents (Mg and B);
• no chemical toxicity;
• low cost with respect to HTS materials (comparable to Nb−Ti);
• applicability at 4.2 ≤ T ≤ 25 K;
• persistent mode operation in MRI magnets up to 25 K.
Disadvantages:
• poor thermal stability with respect to those of Nb−Ti or Nb3Sn;
• applicability of MgB2 wires is limited to relatively low fields: ~10 T at 4.2 K and ~4 T at
20 K.
In future, the compound MgB2 may be considered for the following applications:
• level measurement of liquid H2 containers (e.g. hydrogen-powered cars);
• hydrogen-cooled high current leads at T ≈ 20 K (Russia);
• LINK project (CERN): 13 000 A current leads at ~10 K (total: >1000 km of MgB2 wire)—
see the talk by A. Ballarino at the present CAS Summer School [17];
• wind generators; a first generator study for 10 MW is under construction (European project).
At present, MgB2 is a ‘niche’ superconductor, but the niche is apparently quite large. Going by
the projections of Columbus, it is at least 10 times cheaper than HTS conductors. Also, it can be made
by the PIT process in round wire or in tape form, thus exhibiting a flexible architecture. And, although
259
R. F LUKIGER
it cannot challenge either Nb-Ti or Nb3Sn at 4 K, it can challenge them at 20–30 K. Because of its low
cost, it clearly challenges any HTS conductor for low-field applications in this temperature domain
too, where MRI and high-current bus bar applications need an affordable and capable superconductor.
MgB2 is viable here, where the cost of HTS rules it out for such applications. It is reasonable be
optimistic about further significant advances in the application of MgB2 conductors.
4 The system Bi-2212

Bi-2212 is the only HTS that allows the fabrication of round wires with isotropic properties. This leads
to new possibilities, e.g. it would be possible to build large accelerator magnets at fields well above
15 T, in particular dipoles or quadrupoles. The results presented here have been obtained by
D. Larbalestier’s group at Florida State University. Additional data can be found in the publications of
Malagoli et al. [23], Kametani et al. [24], Jiang et al. [25] and Scheuerlein et al. [26].
These works have progressed in collaboration with a manufacturer, Oxford Instruments (OST).
All data shown here are extracted from the studies of these working groups. The fabrication process
consists of a series of deformation steps of a silver tube filled with previously reacted Bi-2212 powder
followed by a final high-temperature heat treatment at T ≤ 888°C.
After several years of development with relatively deceptive Jc values, progress was achieved
after observing that the Bi-2212 filaments post heat treatment were not wire-like, but were rather in
the form of hollow ‘tubes’ which exhibited a great number of holes. Large bubbles form on melting
and holding at Tmax, resulting in large holes in the wall of the Bi-2212 ‘tubes’, obviously leading to a
strong reduction of Jc. The origin of these holes is the formation of a gas phase in this narrow
temperature range. A series of high-pressure treatments was tried in order to avoid the formation of
holes in the Bi-2212 walls. It was found that a moderate external Ar pressure was sufficient to reach a
sizeable increase of Jc, as shown in Fig. 25. The progress in Bi-2212 round wires is visualized in
Fig. 21, which compares performance at 4.2 K with those of the other known superconductors
available in a round shape.
Fig. 23: Example of a round multifilamentary Bi-2212 wire, produced by Oxford Instruments OST (USA),
before heat treatment, showing the filament configuration. The matrix consists of silver.
260
Fig. 24: Heat treatment performed on the Bi-2212 wires after the final deformation [23–26]
Fig. 25: Increase of the engineering critical current density JE (overall critical current density) of round Bi-2212
wires as a function of the applied isostatic pressure [27].
It can be seen that at 4.2 K, Jc (layer) of Bi-2212 wires is higher than for any of the others for
fields exceeding 14 T. For the industrial application, the cross-over with Nb3Sn wires is at 18 T.
Further progress is expected in the future.
Fig. 26: Jc vs. applied field (layer and engineering Jc values) at 4.2 K of various round wires based on known
superconducting materials [27].
261
R. F LUKIGER
4.1 Conclusions for Bi-2212 wires

On the condition that the recent progress made will lead to the fabrication of homogeneous round
wires of kilometre lengths, the following comments can be made [23–27].
Advantages:
• Bi-2212 is the only HTS material available in round wires;
• the multifilamentary configuration is OK;
• excellent thermal stability (Ag sheath);
• at 4.2 K, Bi-2212 exhibits the highest Jc values at B > 18 T of all round
shaped superconducting wires;
• after further progress, it may be envisaged that Bi-2212 can be used for fields B > 20 T.
Disadvantages:
• higher costs due to processing and Ag sheath;
• bubble problem: in principle this is solved, but industrial tests on long wire lengths are still
required;
• poor mechanical stability: may be the most important obstacle at high-field applications;
• (accelerators): new strategies for mechanical reinforcement must be developed.
5 Coated conductor tapes based on R.E.Ba2Cu3Oy superconductors

Magnet applications for high-energy physics are still a major driver for the further development
of superconducting technology. High-temperature superconductors (HTS) with very high Bc2 values
are promising with respect to large magnets generating fields exceeding 20 T at 4.2 K, as required in
future accelerators following LHC Upgrade. The current-carrying properties of R.E.Ba2Cu3Oy
(REBCO) coated conductors with Tc = 92 K are well suited to fulfil these needs. However, these
materials cannot yet be obtained as round wires, due to the high anisotropy, which follows directly
from the layered crystal structure (see Fig. 27) of the compound YBa2Cu3O7−x. (In the applications, the
Y in this formula will be replaced by a Rare Earth (R.E.) metal, due to its higher current-carrying
properties, so YBa2Cu3O7−x is usually abbreviated to R.E.BCO.)
Fig. 27: The crystal structure of the HTS compound YBa2Cu3O7−x
262
In contrast to the three systems presented before, R.E.BCO conductors can at present only be
prepared in the shape of flat tapes, called coated conductors. It follows that particular techniques must
be developed to fabricate dipoles or quadrupoles with R.E.BCO. In the following, the main properties
of coated conductors are very briefly reviewed in view of their possible application in accelerators at
4.2 K. A variety of methods have been developed by various manufacturers to fabricate coated
conductors; two of them are represented in Fig. 28.
Fig. 28: Schematic of RABITs and IBAD fabrication techniques for R.E.BCO (given in the figure as YBCO)
coated conductors.
The main difference between the RABITS and the IBAD processes concerns the substrate,
which consists of biaxially textured Ni foils and by untextured Hastelloy, respectively. As a
consequence, RABITS wires exhibit lower uniaxial tensile stresses than IBAD wires, where
irreversible strains for Ic up to 0.7% can be reached. The multilayered architecture of coated
conductors fabricated using RABITS and IBAD is shown in Fig. 29; it is easily seen that, due to
various problems linked to mechanical stability and to the need for intermediate buffer and stabilizer
layers, the total cross-section of the superconductor is limited to values around 1%. In order to
enhance the critical current density, it is thus necessary to enhance the HTS layer thickness. However,
it is very difficult to maintain a perfect biaxially textured layer for thicknesses exceeding 1 µm; this
has only recently been possible, after very sophisticated and costly improvements of the deposition
parameters.
Fig. 29: Multilayered architecture of RABITS and IBAD coated conductors. (Thanks to M. Rupich from AMSC
and to Selvamanickam from SuperPower.)
263
R. F LUKIGER
It is customary to define the critical current density by the unit A·cm−1, where the value is
normalized for a 1 cm wide tape (Jcw = Jc·cm−2 × 1 cm = A·cm−1), and no longer depends on the tape
width. The variation of the critical current density as a function of the HTS layer thickness up to 6 µm
at 77 K is represented in Fig. 30. These values are obtained on short samples; the values on industrial
coated conductor lengths (≤500 m) obtained by various manufacturers are presently in the range 500–
600 A, with a tendency towards higher values.
Fig. 30: Ic vs. thickness of the HTS superconducting layer at 77 K at zero field. Maximum critical current
density obtained Jc = 1040 A·(cm-width)−1 for a 6 mm thick layer. Definition: Jcw = Jc·cm−2 × 1 cm = A·cm−1 is
normalized for a tape width of 10 mm. (Courtesy: Fujikura (Japan).)
The variation of Jc for a GdBCO tape of Fujikura (Japan) with an initial value of 600 A·cm−1 as
a function of applied fields up to 31 T and at various temperatures between 4.2 and 77 K is shown in
Fig. 31, illustrating the wide potential of this kind of coated conductor tape.
Fig. 31: Jc vs. B and T for a HTS coated conductor (Fujikura) with Jc = 600 A·cm−1. Measurements taken from
the group of Professor Kiss (Kyushu University), in collaboration with Florida State University and Tohoku
University.
264
The last property discussed in this manuscript concerns the AC losses in coated conductors. Due
to the tape geometry, it is impossible to fabricate conductors with the classical multifilamentary
configuration. A common way to lower AC losses is to configure the tapes into a Roebel geometry,
shown schematically in Fig. 32. Recently, a step towards a ‘multifilamentary’ architecture consisted in
the so-called ‘striation by laser scribing’ technique, a carefully adjusted laser process whereby the
whole tape width is subdivided into three or more ‘filaments’.
Fig. 32: Schematic representation of Roebel + striation for the reduction of AC losses in coated conductors
Fig. 33 Reduction of AC losses by simultaneous Roebel + striation of a coated conductor tape. (Source: Fujikura
(Japan).)
The combination of Roebel + striation, also shown in Fig. 32, indeed represents a simulation of
a multifilamentary arrangement. It has led to a reduction of the AC losses, as shown in Fig. 33. The
reduction of AC losses can be obtained in tapes by means of 6–10 striations. In cables, it is possible to
use the Roebel technique in combination with striations.
6 Conclusions for coated conductor wires

Due to limited time, only a small number of properties of HTS coated conductors have been discussed
in the present lecture. Some conclusions about the future possibilities of these conductors can be
drawn. The requirements for HTS coated conductors can be summarized as follows.
Current density:
• Improvement with thicker biaxially textured HTS layers.
• Improvement by replacing Y by rare earths (R.E.) (dopants).
265
R. F LUKIGER
• Improvement by introducing artificial pinning centres.

Mechanical properties:
• Substrate must be strong enough to withstand the high-temperature formation of
REBaCuO.
• The tape as a whole must be strong and flexible enough to be wound into cables and
coils at 300 K.
• The tape must withstand longitudinal and transverse stresses during operation.
Electrical stability:
• Excess current must be carried in the Ag layer and in the Al, Cu,… outer layers.
Thermal stability:
• Heat transfer to the coolant must be optimized.
AC losses:
• The architecture must be modified to minimize AC losses (Roebel, striations).
In order to achieve a wide market penetration, a certain number of tasks have to be accomplished by
coated conductor tapes:
• higher homogeneity of Jc over whole tape length;
• thicker HTS layer Ic values;
• reproducible production of >1 km lengths;
• enhanced pinning by nano-additives;
• reduced anisotropy by nano-additives.
There is no doubt that all these tasks will be accomplished, taking into account the important
efforts undertaken by the industry. However, a major factor monitoring the application volume of
coated conductors will be the production costs, which are at present one order of magnitude too high.
References
[1] D.M.J. Taylor and D.P. Hampshire, Report No. 8, EFDA03-1126, 2006 (unpublished); available
online at http://www.dur.ac.uk/superconductivity.durham/publications.html.
[2] W. Goldacker and R. Flükiger, Adv. Cryo. Eng. 28 (1982) 361.
[3] J. Ekin, Experimental Techniques for Low Temperature Measurements (Oxford University
Press, New York, 2006), pp. 412–467.
[4] D. Uglietti, B. Seeber, V. Abächerli, A. Pollini, D. Eckert and R. Flükiger, Supercond. Sci.
Technol. 16 (2003)1000.
[5] G. Mondonico, B. Seeber, C. Senatore, R. Flükiger, V. Coarto, G. De Marz and L. Muzzi, J.
Appl. Phys. 108 (2010) 093906.
[6] D. Uglietti, B. Seeber, V. Abächerli, N. Banno and R. Flükiger, IEEE Trans. Appl. Supercond.
15 (2005) 3652.
[7] W. Specking, W. Goldacker and R. Flükiger, Adv. Cryog. Eng. 34 (1988) 569.
[8] G. Mondonico, Ph.D. thesis, University of Geneva, 2013.
[9] S.X. Dou, S. Soltanian, J. Horvat, X.L. Wang, S.H. Zhou, M. Ionescu, H.K. Liu, P. Munroe and
M. Tomsic, Appl. Phys. Lett. 81 (2002) 3419.
[10] E.W. Collings, M.D. Sumption, M. Bhatia, M.A. Susner and S.D. Bohnenstiehl, Supercond.
Sci. Technol. 21 (2008) 103001.
[11] R. Flükiger and H. Kumakura, in 100 Years of Superconductivity, Eds. H. Rogalla and P.H. Kes
(CRC Press, Taylor and Francis, 2011), p. 758
266
[12] A. Malagoli, C. Bernini, V. Braccini, C. Fanciulli, G. Romano and M. Vignolo, Supercond. Sci.
Technol. 22 (2009) 105017.
[13] G. Grasso, A. Malagoli, D. Marrè, E. Bellingeri, V. Braccini, S. Roncallo, N. Scati and A.S.
Siri, Physica C 378–381 (2002) 899.
[14] M.A. Susner, Y. Yang, M.D. Sumption, E.W. Collings, M.A. Rindfleisch, M.J. Tomsic and J.V.
Marzik, Supercond. Sci. Technol. 24 (2011) 012001.
[15] K. Togano, J.M. Hur, A. Matsumoto and H. Kumakura, Supercond. Sci. Technol. 22 (2009)
015003.
[16] G. Giunchi, S. Ceresara, G. Ripamonti, A. DiZenobio, S. Rossi, S. Chiarelli, M. Spadoni, R.
Wesche and P.L. Bruzzone, Supercond. Sci. Technol. 16 (2003) 285.
[17] A. Ballarino, LINK project, presentation at CAS 2013.
[18] R. Flükiger, M.S.A. Hossain, C. Senatore and M. Rindfleisch, Physica C 471 (2011) 222011.
[19] M.S.A. Hossain, C. Senatore, M. Rindfleisch and R. Flükiger, Supercond. Sci. Technol. 24
(2011) 075013.
[20] G.Z. Li, M.D. Sumption, M.A. Susner, Y. Yang, M. Kongara, M.A. Rindfleisch, M.J. Tomsic,
C.J. Thong and E.W. Collings, Supercond. Sci. Technol. 26 (2012) 115023.
[21] A. Matsumoto, Y. Kobayashi, K.-I. Takahashi, H. Kumakura and H. Kitaguchi, Appl. Phys.
Express 1 (2008) 021702.
[22] W. Hässler, P. Kovac, M. Eisterer, A.B. Abrahamsen, M. Herrmann, C. Rodig, K. Nenkov, B.
Holzapfel, T. Melisek, M. Kulich, M. v. Zimmermann, J. Bednarcik and J.C. Grivel, Supercond.
Sci. Technol. 23 (2010) 065011.
[23] A. Malagoli, C. Bernini, V. Braccini, G. Romano, M. Putti, X. Chaud and F. Debray,
Supercond. Sci. Technol. 24 (2011) 075016.
[24] F. Kametani, T. Shen, J. Jiang, C. Scheuerlein, A. Malagoli, M. Di Michiel, Y. Huang, H. Miao,
J.A. Parrell, E.E. Hellstrom and D.C. Larbalestier, Supercond. Sci. Technol. 24 (2011) 075009.
[25] J. Jiang, W.L. Starch, M. Hannion, F. Kametani, U.P. Trociewitz, E.E. Hellstrom and D.C
Larbalestier, Supercond. Sci. Technol. 24 (2011) 082001.
[26] C. Scheuerlein, M. Di Michiel, M. Scheel, J. Jiang, F. Kametani, A. Malagoli, E.E. Hellstrom
and D.C. Larbalestier, Supercond. Sci. Technol. 24 (2011) 115004.
[27] D. Larbalestier, oral presentation at CERN, November 2012.
267
Magnetic Design of Superconducting Magnets
E. Todesco 1,
Abstract
In this paper we discuss the main principles of magnetic design for
superconducting magnets (dipoles and quadrupoles) for particle accelerators.
We give approximated equations that govern the relation between the
field/gradient, the current density, the type of superconductor (Nb−Ti or
Nb3Sn), the thickness of the coil, and the fraction of stabilizer. We also state
the main principle controlling the field quality optimization, and discuss the
role of iron. A few examples are given to show the application of the
equations and their validity limits.
Keywords: magnets for accelerators, superconducting magnets, magnet

design.
1 Introduction
The common thread of these notes is to provide some analytical guidelines with which to outline the
design of a superconducting accelerator magnet. We consider this for both dipoles and quadrupoles:
the aim is to understand the trade-offs between the main parameters such as the field/gradients, the
free aperture, the type of superconductor, the operational temperature, and the current density. These
guidelines are rarely treated in handbooks: here, we derive a set of equations that provides us with the
main picture, with an error that can be as low as 5%. This initial guess can then be used as a starting
point for fine tuning by means of a computer code to account for other field quality effects not
discussed here, such as iron saturation, persistent currents, eddy currents, etc. Examples will be given
in the text to guide the reader around the equations.
In Section 2, we introduce the sector coil design for electromagnets, derive the multipolar
expansion, and write down the conditions for field quality that must be imposed on the wedges and on
the angles of a sector coil. We also present the main equations for an electromagnet, relating the
central field and the gradient to the current density and the coil width. In Section 3, we discuss the
superconducting case; first we propose simple expressions for the critical surfaces of Nb−Ti and
Nb3Sn. We then focus on the peak field in a sector coil; this allows us to derive a complete equation
giving the relation field/gradient versus coil width, aperture, superconductor type, and quantity of
stabilizer. In Section 4, we analyse the role of iron, and discuss its beneficial effects on lowering the
current density and keeping the magnetic flux within the magnet. In Section 5, we discuss an
alternative design based on block coils, and in Section 6 we summarize the different steps taken in
magnet design.
2 Electromagnets based on sector coils

In this section, we discuss the main equations for an electromagnet, starting with the Biot–Savart law,
which yields the field produced by a current line. The section deals with a generic electromagnet, i.e. it
is independent of the specific features of superconductivity: we only consider how to make a ‘good’
1
ezio.todesco@cern.ch
E. T ODESCO
dipole or quadrupole field with current lines carrying a given current density. After having defined the
multipolar expansion for a current line, we focus on the computation of the field given by a sector coil
producing a dipolar field. This is the main building block that can be used to construct the so-called
cos-theta layout, based on a keystoned cable and wedges. We then derive the conditions for the
positions and number of wedges required to satisfy the accelerator field quality constraints. The
quadrupole case is also considered, to show the similarities and the differences.
2.1 Biot–Savart law and the multipole expansion

An infinitesimal current line dr produces an infinitesimal magnetic field dB according to the equation
 
  µ 0 I × dr
dB (r ) = , (1)
4π r 3

where I is vector of the current intensity, r is the distance from the current line, and µ0 = 4 × 10−7 (N·A−2)
is the vacuum permeability. For a line in the s direction of infinite length, integration yields
 
  µ0 I × r
B(r ) = , (2)
2π r 2
and the problem becomes two-dimensional. The two components are given by
µ I y − y0
Bx ( x, y ) = − 0 ,
2π ( x − x0 )2 + ( y − y0 )2
(3)
µ I x − x0
B y ( x, y ) = 0 ,
2π ( x − x0 )2 + ( y − y0 )2
and, using the complex notation z = x + iy and B = By + iBx (see Fig. 1), we obtain
µ0 I ( x − x0 ) − i ( y − y0 )
By ( x, y ) + iBx ( x, y ) = . (4)
2π ( x − x0 )2 + ( y − y0 )2
Since
a − ib a − ib 1
= = , (5)
a 2 + b2 ( a + ib )( a − ib ) ( a + ib )
the field can be written as
µ0 I
B( z ) = , (6)
2π ( z − z0 )
where the current line is in z0 and the field is given at position z.
B=By+i Bx I
z0=x0+i y0
z=x+iy
Fig. 1: Jean-Baptiste Biot (left); the field generated in z by an infinite current line placed at z0 (center); and Felix
Savart (right).
270
M AGNETIC D ESIGN OF S UPERCONDUCTING M AGNETS
The above formulation, based on complex notation, besides being compact, can easily produce a
multipolar expansion: because
1 ∞
=1 + z + z 2 + z 3 + ... = ∑ z n −1 for z < 1 (7)
1− z n =1
(remember: the expansion is valid inside the unitary circle), one can write
n −1
µ0 I µI 1 µI ∞  z
B( z ) == − 0 − 0 ∑ 
= for z < z 0 . (8)
2π ( z − z0 ) 2π z0 1 − z 2π z0 n =1  z0 
z0
When the multipolar expansion is carried out, we lose something: Eq. (6) is valid everywhere (except
at the singularity in z0), but the right-hand side of Eq. (8) is valid only for |z|<|z0|). So this
expression is good for estimating the field seen by the beams, but not for estimating the field in the
magnet coil or in the iron.
In any current-free region D the field has null divergence, and can be expanded in multipoles:
∞
B ( x, y ) = ∑ ( Bn + iAn ) ( x + iy ) x, y ∈ D .
n −1
(9)
n =1
To avoid coefficients with physical dimensions depending on the order of the multipoles, we usually
define a reference radius Rref. Moreover, the main component is factorized so that the normalized
multipoles (bn, an) become dimensionless (see, e.g., Ref. [1], p. 50); for a dipole,
n −1
∞
 x + iy 
B ( x, y ) 10−4 B1 ∑ (bn + ian ) 
=  x, y ∈ D . (10)
n =1  Rref 
Here, bn and an are called the normal and skew multipoles, respectively. The reference radius Rref has
no physical meaning (it is just a choice of units to express the multipoles, and has nothing to do with
the convergence radius of the series), and is usually chosen to be 2/3 of the aperture. For an
accelerator magnet, typical acceptable field quality deviations from the ideal field are of the order of
0.01% at 2/3 of the aperture radius. This is why a 10−4 is also factorized in Eq. (10). In this way the
normalized multipoles have values of the order of unity, providing a more compact expression and
simplifying the compilation of tables (no exponential format is needed).
Comparing Eqs. (8) and (10), we obtain the main field of a current line,
Iµ 1 Iµ x
B1 = − 0 Re   = − 0 2 0 2, (11)
2π  z0  2π x0 + y0
and the multipoles are

n −1
I µ0  Rref 
Bn + iAn =
−   ,
2π z0  z0 
n −1
(12)
I µ0 104  Rref 
bn + ian =
−   .
2π z0 B1  z0 
This notation may seem a bit obscure, but in the following we will see how easily the conditions for
field quality in a electromagnet can be derived from these expressions.
271
E. T ODESCO
Equations (12) have a meaningful implication: since the position of the current line z0 is larger
than the reference radius Rref, the argument of the power is smaller than one, and the multipoles vanish
as a power law. Plotting the logarithm of the absolute value of the multipoles against the multipole
order, one obtains a linear regression with slope Rref/|z0| (see Fig. 2). The aim of magnetic design (see
Refs. [1], pp. 45–63, [2]) is to arrange the current lines around the free aperture so that the
contributions (12) compensate for each other. The target is to have multipoles of a few units for low
order, n ≤ 3, and less than one unit for the higher orders.
The requirements on field quality can be translated into the requirements on the positioning of
the current lines in the cross-section; typically, the required precision is of the order of 0.1 mm. The
multipole variation created by a misplacement of the current line of ∆z0 is given by
n −1
I µ0  Rref 
∆( Bn + iAn ) n
=   ∆z0 , (13)
2π z0 2  z0 
and has the same power law decay as a current line. This implies that the random components of the
field quality due to the position of the current lines also have a power law decay. This can be used to
judge the precision of the measurement, corresponding to the appearance of a plateau where data no
longer follow the Biot–Savart decay (see Fig. 2, right).
10.00
1.00
|bn|, |an|
0.10
0.01
0 5 10 15
Multipole order n
Fig. 2: Decay of multipoles of a current line with order plotted on a semilogarithmic scale (left), and an example
of a magnetic measurement (right) of the random components of field quality in a Nb3Sn short model HQ,
showing the measurement precision at around 10−2 units [3].
2.2 Dipoles based on sector coils

It is well known that a cos-theta distribution of current density provides a perfect dipolar field (see,
e.g. Refs., [1], pp. 45–63, [2, 4]). This means that there is a maximal current on the midplane, with
opposite sign on the left and right sides of the aperture, and zero current at 90° and 270° (see Fig. 3,
left). This can be approximated in a real winding by a sector coil with layers and wedges (see Fig. 4).
Here we will present analytical expressions for the sector coil (see Fig. 3, right). Sector coils provide a
very good approximation of an accelerator magnet, allowing a quasi-analytical approach that shows all
the main features of magnet design [5, 6]. Moreover, equations can be easily implemented in
spreadsheets, and solutions can be used as a starting point for more refined computations with finite
element codes and/or optimizers.
A current line in a sector coil of width w and angle α, without wedges (see Fig. 3, right), yields
a main field
Iµ 0  1  Iµ cos θ
B1 = − Re  = − 0 , (14)
2π  z0  2π ρ
272
where we have used polar coordinates z = ρ exp(iθ). Replacing the current I by the integration
element, we find the equation for the main field as follows:
jµ α r+w
ρ dρ 2 j µ0
I → j ρ dρ dθ −2 0 ∫ cos θ dθ ∫
B1 = =
− w sin α . (15)
2π −α r
ρ π
Fig. 3: A cos-theta distribution of current (left), and a sector coil without wedges (right)
Fig. 4: Approximations of cos-theta coil with sector coils; wedges (left) and layers (right) [4]
So, for a dipole sector coil of aperture r and coil width w, we have that
• the field is proportional to the current density j;
• the field is proportional to the coil width w;
• the field is independent of the coil aperture r.
We can verify that the configurations shown in Figs. 3 and 4 have the following expansion:
 −4  z2 z4 z6 
B =B1 1 + 10 b3 2 + b5 4 + b7 6 + ...  , (16)
 
  Rref Rref Rref 
since the top–bottom symmetry guarantees a2n+1 = 0, the left–right antisymmetry guarantees that
b2n = 0, and the up–left/down–right antisymmetry guarantees that a2n = 0. The computations are a little
cumbersome, but are a good exercise in familiarizing oneself with the multipolar expansion. The odd
normal components are called the ‘allowed’ multipoles in a dipole. We can see now the full force of
the multipolar expansion: the conditions for a perfect dipole are reduced from a two-dimesional
vectorial field to a few real coefficients (b3, b5, b7,…). Usually a truncation at order 11 is used for
tracking particles, so in the end the field quality condition for a dipole corresponds to cancelling only
five coefficients.
For the same sector dipole as that just analysed, the integration of high-order multipoles n > 2
(the case n = 2 will be considered in the quadrupole subsection) yields
273
E. T ODESCO
 
 n −1
j µ R 2 sin(nα ) (r + w) − r
n −1
j µ  R  2 sin(nα ) 
2− n 2− n
1  . (17)
Bn =
− 0 ref
− 0 r  ref 
= − 1
π n 2−n π  r  n(2 − n)   w  n − 2 
 1 +  
  r 
Therefore, a sector coil of 60º has b3 = 0, i.e. the first-order non-linearity is cancelled. The main field
is given by
3 j µ0
B1 = − w~6.9 × 10−7 j  A ⋅ m −2  w [ m ] =
6.9 × 10−4 j  A ⋅ mm −2  w [ mm ] . (18)
π
The 60º sector coil is the equivalent of the Helmholtz coil, i.e. the simplest winding with a ‘good’ field
quality. Unfortunately this is not ‘good’ enough since b5 is too large. We have
 
 4 
1 sin 5α  Rref  r  1
b5 = 4
10 × − 1 , (19)
15 sin α  r  w   w
3 
 1 +  
  r 
which for a typical case of w = r and α = 60º gives b5 ~ 100 units, whereas the requirements from
particle dynamics are a few units at most. In the limit of a thin coil, w→0, we obtain the simplified
expression
4
1 sin 5α  Rref 
b5 104 ×
=  r  , (20)
5 sin α  
as given in Ref. [1], p. 53.
The interesting observation that can be derived from this simple case is that for w/r >> 1
normalized multipoles are proportional to r/w. This means that a dipole with a thick coil w w.r.t. to a
small aperture r will have ‘naturally low’ multipoles. Conversely, a dipole with large aperture and low
field (as in the Relativitic Heavy Ion Collider main dipoles [7], where r/w ∼ 0.25) will have a field
quality that is more difficult to optimize.
To set two multipoles to zero requires two free parameters. The designers of the Tevatron [8]
chose to have a two-layer sector coil with the same cable width; see Fig. 5. Therefore, B3 and B5 could
be set to zero:
 1 1  1 1 
B3 ∝ sin(3α1 )  −  + sin(3α 2 )  − = 0,
r + w r   r + 2w r + w 
 1 (21)
1  1 1 
= B5 sin(5α1 )  − 3  + sin(5α 2 )  =
−  0.
 ( r + w )  ( r + 2 w ) ( r + w) 
3 3 3
r 
This is a system of non-linear equations in the angles α1 and α2. It has solutions only for w/r < 0.5 (see
Fig. 6); this is indeed the case for the Tevatron main diplole coil, which has w/r ∼ 0.2, corresponding
to the angles 37º and 72º. Using the above expressions, one can compute that this produces b7 = 20
units, which in Tevatron times was considered not to be critical. During the 1980s, the concern for
high-order non-linearities became more and more important. SSC [9], HERA [10], and LHC [11]
dipole coils make use of wedges, leading to more free parameters, better to optimize the field quality
(see Fig. 7, left).
274
60
72°
50
40
37°
y (mm)
w
α1 20
α2
0
0 20 40 60
0
0
r 50
x (mm)
Fig. 5: Coil layout with two layers and no copper wedges (left), and coil used for Tevatron main dipoles (right)
Fig. 6: The solutions for the angles of the inner layer α1 and the outer layer α2 vs. the ratio w/r and the coil
adopted in Tevatron dipoles.
60
50.0
45.0
40 40.0
y (mm)
35.0
30.0
20 25.0 α3 α
2
20.0
α1
15.0
10.0
0 5.0
0 20 40 60
x (mm) 0.0
Fig. 7: Coil of the main HERA dipole (left) and coil layout with one layer and one wedge (right)
275
E. T ODESCO
With one layer and one wedge (see Fig. 7, right), there are three free parameters α1, α2, and α3, and in
the non-normalized multipole equations we simply have to add the expression for each sector coil:
µ0 jRref
2
sin(3α1 ) − sin(3α 2 ) + sin(3α 3 )  1 1
B3  − ,
π 3 r + w r 
µ0 jRref
4
sin(5α1 ) − sin(5α 2 ) + sin(5α 3 )  1 1
B5  − 3 , (22)
π  ( r + w ) r 
3
5
µ0 jRref
6
sin(7α1 ) − sin(7α 2 ) + sin(7α 3 )  1 1
B7  − 5
.
π  ( r + w ) r 
5
7
(Note that, conversely, the normalized multipoles are not linear, so it is not permitted to add the
contribution of each sector to calculate the final values for the whole coil.) In setting b3 and b5 to zero,
we now have a one-parameter family of solutions. Among them, we have three solutions with integer
angles that are easy to remember: (α1, α2, α3) = (24º, 36º, 70º), (36º, 44º, 64º), (48º, 60º, 72º). The
solution ~(43.2º, 52.2º, 67.3º) sets b7 to zero.
It is now easy to continue: with two wedges, there are five free parameters, and we can set b3–
b11 to zero. The equations for the angles are as follows:
sin(3α1 ) − sin(3α 2 ) + sin(3α 3 ) − sin(3α 4 ) + sin(3α 5 ) =
0,

0,

0, (23)

0,

0.
Also, in this case there is one solution with angles 0°–33.3°, 37.1°–53.1°, 63.4°–71.8° (see Fig. 8,
left), which is the basis for the RHIC dipole coil layout, where one additional wedge included to
recover the correct keystoning of the cables (see Fig. 8, right).
60
40
y (mm)
20
0
0 20 40 60
x (mm)
Fig. 8: Coil with two wedges which cancels b3 to b11 (left), and coil used on the RHIC dipole (right), where the
straight lines indicate the angles of the coil.
276
We now arrive at the goals of this section: estimating the proportionality coefficient between the
field, the current density, and the coil width for a few layouts (see Eq. (18) for the 60° sector coil).
This allows us to compare how much field we can produce for a given coil width. We define the
coefficients γ and γ0 as
B1 ≡ γj ≡ γ 0 wj , (24)
and for a 60° sector coil, see Eq. (18), we calculate

2µ π
γ0 = − 0 sin = 4 3 × 10−7 ≈ 6.9 × 10−7 T ⋅ m ⋅ A −1  . (25)
π 3
The same coefficient can be estimated for layouts with one or more wedges, yielding similar results
(see Table 1). It is interesting to note that the addition of one wedge (which allows us to reduce b5 to
zero) to layouts with the same angular width (i.e. the same quantity of superconductor) costs only ~5%
of the field (from 6.93 × 10−7 to 6.63 × 10−7; see the first two rows of Table 1). So, in general, the
requirements on field quality imply only a small reduction in the field that can be obtained from a
generic coil winding.
Table 1: Features of six different layouts based on sector coils of increasing complexity
Layout No. wedges Total angular width γ0 Zero harmonics

(degrees) (T ∙ m /A)
[0°−60°] 0 60.0 6.928E-06 b3
[0°−48°, 60°−72°] 1 60.0 6.625E-06 b 3 ,b 5
[0°−24°, 36°−60°] 1 48.0 5.480E-06 b 3 ,b 5
[0°−36°, 44°−64°] 1 56.0 6.335E-06 b 3 ,b 5
∼[0°−43.2°, 52.2°−67.3°] 1 58.3 6.535E-06 b 3 ,b 5 ,b 7
∼[0°−33.3°, 37.1°−53.1°, 63.4°-71.8°] 2 57.7 6.411E-06 b 3 ,b 5 ,b 7
Example: The coil of the LHC main dipole [11] is ∼30 mm wide and has 350 A·mm−2 (inner
layer) and 430 A·mm−2 (outer layer) as the overall current densities. Applying Eq. (24), using
γ0 = 6.9 × 10−7 T·m·A−1, one obtains 3.5 T and 4.3 T, thus giving a total of 7.8 T, close to the 8.3 T
obtained by an accurate model.
2.3 Quadrupoles based on sector coils

The sector layout for a quadrupole is shown in Fig. 9. The arrangement of currents produces a null
field in the centre, so B1 = 0, and the second-order harmonic is given by
Iµ R  1 
B2 = − 0 ref Re  2  . (26)
2π  z0 
We define the gradient as
B2 Iµ  1  I µ cos 2θ
G≡ − 0 Re  2  =
= − 0 , (27)
Rref 2π  z0  2π ρ 2
where again we switch to polar coordinates z = ρ exp(iθ); on integrating over the sector coil, we obtain
jµ α r+w
ρ dρ 2 j µ0  w 
I → j ρ dρ dθ , −4 0 ∫ cos 2θ dθ ∫
G= −
= ln 1 +  sin 2α . (28)
2π −α r ρ 2
π  r
277
E. T ODESCO
The gradient is proportional to the current density and to the (natural!) log of one plus the ratio of the
coil width and the aperture. This means that
• the gradient is proportional to the current density j;
• the gradient depends on the ratio of the coil width w and the aperture r;
• for large coil width, the gradient increases with the logarithm of the coil width ∝log w.
30 32°
22°
20
18°
y (mm)
10
0
0 10 20 30 40 50 60
x (mm)
Fig. 9: Sector layout for a quadrupole (left), and one-eighth of the RHIC main quadrupole coil, with straight
lines indicating the 18°–22°–32° angles (right).
The multipolar expansion is now achieved by factorizing the main component, i.e. B2
n −1
∞
 x + iy 
10 × B2 ∑ (bn + ian ) 
B ( x, y ) = −4
 x, y ∈ D , (29)
n =1  Rref 
and the allowed multipoles are b2n+2:

  z5 z9 z13 
B= B2 1 + 10−4 b
 6 5 + b + b + ...  . (30)
 10 9 14 13 
  Rref Rref Rref 
The first piece of good news is that the field quality constraints are easier to calculate for a
quadrupole, as there are fewer low-order harmonics to be minimized: keeping the rule of truncation at
order 11, one needs to minimize only two harmonics, b6 and b10. The second piece of (very) good
news is that the equations for the quadrupole field quality are very similar to those for the dipole case,
namely any angular solution for a dipole cancelling the first n allowed harmonics will also cancel the
first n allowed quadrupolar harmonics, provided that angles are divided by a factor of 2 (see Table 2).
Therefore, a 30° sector coil cancels b6, a one-wedge coil [0°–24°, 30°–36°] cancels b6 and b10, and so
on. So we have already solved the quadrupolar equations for the field quality in the previous
subsection (and also for sextupoles, decapoles, etc.)!
Table 2: Features of six different layouts based on sector coils of increasing complexity
Layout No. wedges Total angular width γ0 Zero harmonics

(degrees) (T ∙ m /A)
[0°−30°] 0 30.0 6.928E-06 b6
[0°−24°, 30°−36°] 1 30.0 6.625E-06 b 6 ,b 10
[0°−12°, 18°−30°] 1 24.0 5.480E-06 b 6 ,b 10
[0°−18°, 22°−32°] 1 28.0 6.335E-06 b 6 ,b 10
∼[0°−21.6°, 26.1°−33.7°] 1 29.2 6.535E-06 b 6 ,b 10 ,b 14
∼[0°−16.7°, 18.6°−26.6°, 31.7°-35.9°] 2 28.9 6.411E-06 b 6 ,b 10 ,b 14
278
As in the dipole case, we present coil layouts (which are usually selected via numerical codes) that
can be understood in terms of the above analytical equations and solutions. We present three
examples:
• The one-layer coil of the main RHIC quadrupole [7] that closely follows the [0°–18°, 22°–32°]
layout (see Fig. 9, right).
• The two-layer coil of the Tevatron main quadrupole [8], which has an inner layer with a [0°–12°,
18°–30°] layout, and the outer layer is a 30° coil (the outer layer has little impact on higher orders
so a wedge to minimize b10 is not required); see Fig. 10, left.
• The two-layer coil of the LHC main quadrupole [11], which has an inner layer with a [0°–24°,
30°–36°] layout, and the outer layer is a 30° coil with a small wedge to recover the Roman
arch structure; see Fig. 10, right.
30 30° 30 36° 30°
24°
18° 20
20
y (mm)
y (mm)
10 12° 10
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
x (mm) x (mm)
Fig. 10: One-eighth of the HERA main quadrupole coil, with straight lines indicating the 12°–18°–30° angles
(left), and one-eighth of the LHC main quadrupole coil, with straight lines indicating the 24°–30°–36° angles
(right).
We conclude this section as we did the previous one, by estimating the proportionality
coefficient between the field, the current density, and the coil width for a few layouts. We define
 w
G ≡ γ 0 j ln1 +  , (31)
 r
and we can easily verify that the values for the constant γ0 are the same as in the dipole case (provided
that angles are divided by 2; see Table 2).
Example: The LHC main quadrupole coil has an aperture of 28 mm radius, a width of 31 mm,
and 430 A·mm −2 overall current density. Applying Eq. (31), using γ0 = 6.6 × 10 7 T·m ·A −1, we obtain
210 T·m, close to the 220 T·m obtained by an accurate model. Remember, this is a natural logarithm!
3 Superconducting magnets
The preceding considerations are valid for any electromagnet based on sector coils. We now consider
the case of superconducting sector coils made of Nb−Ti and/or Nb3Sn. We start by defining
approximations for the critical surfaces, which are easier to handle than the usual fits and which allow
analytical solutions. The magnet is limited not by the field in the centre (dipole) or at the edge of the
aperture (quadrupole), but by the peak field in the coil; for this reason, a whole section is dedicated to
the estimation of this quantity. Finally, the intersection of the electromagnet loadline B(j) with the
superconductor characteristics allows us to derive an analytical expression for the field/gradient as a
function of the quantity of coil and the type of superconductor.
279
E. T ODESCO
3.1 Critical surfaces

Superconductors are able to tolerate a given combination of current density and field, in an
approximately triangular zone of the (B, j) space, at a fixed temperature [12]. For Nb−Ti, a good
approximation is the linear function
jsc ( B
= ) s (b − B) , (32)
where b and s are two free parameters that fit the critical surface. The fit is accurate over a few tesla
(see Fig. 11). Typically, the slope s ~ 5.0 × 107 A·m2·T−1, i.e. lowering the field by 1 T gains
500 A·mm−2. At 1.9 K and 9 T, typical values are ~2000 A·mm −2, thus yielding b ~ 13 T. At 4.2 K
and 6 T, typical values are as well around 2000 A·mm−2, thus yielding b ~ 10 T, i.e., the whole critical
surface at 1.9 K is shifted left by about 3 T A sketch of the critical surface through the Bottura fit [13]
and the linear approximation is given in Fig. 11.
For Nb3Sn, the linear approximation is not valid since there is a non-negligible positive
curvature [14] (see Fig. 11). A good approximation [6] that provides an analytical solution when
intersecting with the magnet loadline is a shifted hyperbola:
b 
( B ) s  − 1 .
jsc= (33)
B 
This is an empirical fit corresponding to a pinning force linear in the field,

F ( B )= jsc ( B ) ⋅ B= s ( b − B ) , (34)
and works very well over a large range of fields (see Fig. 11). Again, s and b are two free parameters,
with the same dimensions as in the previous case (a current density per tesla and a field, respectively).
Using s ~ 3.15 × 109 A·m−2·T−1 and b ~ 22 T at 4.2 K, we obtain a current density of 2700 A·mm−2 at
12 T and of 1450 A·mm−2 at 15 T, which are currently close to the best achievable values.
Extrapolation at 1.9 K can be achieved with s ~ 3.3 × 109 A·m−2·T−1 and b ~ 24 T. As before, these are
indicative values, and one can work out the best parameters for each critical surface and range of
interest.
3500
3000
2500
Nb-Ti
jsc(A/mm2)
2000
1.9 K
1500 Nb3Sn
Nb-Ti 1.9 K
1000 4.2 K
Nb3Sn
500 4.2 K
0
0 5 10 15 20
B (T)
Fig. 11: Critical surfaces of Nb−Ti and Nb3Sn; linear fit and hyperbolic fit are shown. (The thin lines in the
Nb3Sn case are barely visible as they are very close to the Kramer fit [14].)
280
The insulated coil contains only a fraction, κ, of superconductor, with κ usually in the range
0.25–0.40. So the equation for the overall current density (including stabilizer, voids, and insulation) is
j ( B) = κs (b − B) for Nb−Ti (35)
and
b 
j ( B ) = κs − 1
B  for Nb3Sn. (36)
3.2 Peak field in the coil

The superconducting magnet is limited by the combination of current density and magnetic field in the
coil. Because of this, an estimate of the peak field in the coil is required. For a dipole, we define λ as
the ratio between the peak field Bp and the central field B1:
Bp ≡ λ B1 . (37)
One can show that λ is a function of the ratio of the coil aperture and the coil width, i.e. w/r. A rough
estimate based on the fit of many superconducting coils of accelerator magnets [6] is given by
r
λ= 1 + a , (38)
w
with a ~ 0.04 (see Fig. 12). This means that for r ~ w (as in the LHC main dipole), the field in the coil
is about 4% larger than the central field. The estimate becomes rather pessimistic for w/r < 0.5, where
it has been shown that an improved result may be achieved with careful coil optimization. For
example, for the FAIR dipole with r = 75 mm and w = 15 mm, the proposed coil layout has λ = 1.13
instead of the 1.20 expected from our fit.
1.3 TEV MB HERA MB

SSC MB RHIC MB
LHC MB Fresca
1.2
MSUT D20
λ [adim]
HFDA NED
1.1
1.0
0.0 0.5 1.0 1.5 2.0
equivalent width w/r
Fig. 12: Ratio between peak field and central field in several accelerator magnet dipoles, and hyperbolic fit with
a = 0.04.
For a perfect quadrupole, the peak field on the border of the aperture is rG, so one can define a
similar dimensionless ratio λ as follows:
Bp ≡ λ Gr . (39)
281
E. T ODESCO
In addition, (i) λ is a function of w/r, and (ii) λ increases for w/r << 1; however, there is a new feature:
for larger and larger coil widths, one does not reach the ideal case λ =1, rather there is another
divergence. The following fit has been proposed [5]:
r w
λ= 1 + a1 + a−1 , (40)
w r
with a1 = 0.04 and a-1 = 0.11, based on the data of several coils (see Fig. 13). These fits have to be
considered as approximated analytical expressions, allowing for the fact that λ ≠ 1.
1.5 ISR MQ TEV MQ HERA MQ

SSC MQ LEP I MQC LEP II MQC
1.4 RHIC MQ RHIC MQY LHC MQ
LHC MQM LHC MQY LHC MQXB
1.3
λ [adim]
LHC MQXA
1.2
1.1
current grading
1.0
0.0 0.5 1.0 1.5 2.0
aspect ratio w eq/r (adim)
Fig. 13: Ratio of peak field and gradient times aperture in several accelerator magnet quadrupoles, and fit of
Eq. (40).
3.3 Field versus coil width in superconducting dipoles

One of the central choices in magnet design is the selection of the width of the coil, i.e. how much
superconductor to use in the magnet. For a generic electromagnet, this choice affects the current
density and the operational field through Eq. (24). In a superconducting magnet, current density and
field are not independent variables due to the constraints imposed by the critical surface (see Fig. 11).
A magnet with small coil width will produce little field for a given current density, so the loadline will
be steep. This magnet will reach the critical surface in a ‘low field’ and large current density region
(see Fig. 14, left). On the other hand, a magnet with large coil width will produce a high field for the
same current density, and the loadline will have a very gentle slope. This magnet will reach the critical
surface at high field and low current density.
We now make these arguments more quantitative by computing the operational limits for the
Nb-Ti case. On intersecting the loadline
Bp = λγ 0 jw (41)
with the linearized critical surface (32) and solving for j, we obtain
κ s [b − λγ 0 jw] =
κ s b − Bp ( j )  =
j= κ sb − κ sλγ 0 jw → j (1 + κsλγ 0 w) = κsb . (42)
282
1200
1200
low B
high j 800
800
j(A/mm2)
(Bss, jss) (Bp,ss, jss)
j(A/mm2)
high B 400
400
low j
0 0
0 5 10 15 0 5 10 15
B (T) B (T)
Fig. 14: Magnet loadlines (red line) and Nb−Ti critical surface, for large andsmall coil widths (left), and
definition of short sample current density, central field and peak field (right).
The so-called short sample condition, i.e. the limit of the magnet given by the conductor
performance (see Fig. 14, right), is obtained:
κ sγ 0 w κ sλγ 0 w κs
Bss = b Bp,ss = b jss = b. (43)
1 + κ sλγ 0 w 1 + κ sλγ 0 w 1 + κ sλγ 0 w
The main feature of these expressions is that for w→∞ the central field reaches the asymptotic limit b
as w/(1+w). This sets a practical limit on the maximum field achievable with Nb−Ti of ~10 T, not at b
~ 13 T. The remaining 2–3 T can be achieved only with very large coils, which become more and
more expensive. The LHC dipole coil, with its coil of 30 mm width, is therefore at the limit for Nb−Ti.
To gain 1 T in the short sample field, we would need to increase the coil width from 30 to 50 mm (see
Fig. 15). Equations (43) also give the dependence on the filling ratio, on the critical surface properties,
and the second-order dependence on the magnet aperture.
Fig. 15: Short sample field versus coil width and aperture for Nb−Ti (left) and Nb3Sn (right); filling factor
κ = 0.3.
Example: Before switching to the Nb3Sn case, we apply Eqs. (43) to the case of the LHC main
dipole [dip]. We have w = 31 mm, r = 28 mm, and therefore
r 28
λ =1+ a = 1 + 0.04 = 1.036 , (44)
w 31
from Eq. (38). The filling factor equals 0.28 and 0.33 in the inner and outer layers, respectively, so we
choose an average value κ = 0.3. Using s = 5.0 × 108 A·m−2 and b = 13 T, we have
283
E. T ODESCO
κ sγ 0 w 0.3 × 5.0 ×108 × 6.6 ×10−7 × 0.031

=Bss = b = 13 9.6 T , (45)
1 + κ sλγ 0 w 1 + 0.3 × 5.0 ×108 ×1.036 × 6.6 ×10−7 × 0.031
which is very close to the real value of 9.7 T, which also includes the iron effect (see next section),
and
B 9.6
jss = ss = =4.7 ×108 A ⋅ m −2 , (46)
γ 0 w 6.6 ×10 × 0.031
−7
i.e. 470 A·mm−2. (Remember to keep the SI units, even though they are not practical for our case of
coil width and aperture in metres, current density in ampere per metre squared.)
For Nb3Sn, Eq. (35) for the critical surface must be replaced by Eq. (36). The hyperbolic fit
gives a second-order equation that can be solved explicitly:
κ sγ 0 w  4b  Bss
=Bss  + 1 − 1 , jss = , Bp,ss = λ Bss . (47)
2  κ sλγ 0 w  γ 0w
For w→∞ in this case, we reach the field limit b; saturation is slower w.r.t. Nb-Ti (see Fig. 15) due to
the positive curvature of the critical surface.
Example: We consider the case of an 11 T dipole, with a coil geometry very similar to the LHC
main dipole: r = 30 mm, w = 31 mm. The peak to bore field ratio is given by
r 30
λ = 1+ a = 1 + 0.04 = 1.039 . (48)
w 31
The short sample field is around 14 T:
γ 0 w =6.6 ×10−7 × 0.031 =2.05 ×10−8 ;
0.3 × 3.3 ×109 × 2.05 ×10−8  4 × 24 

Bss =  9 −8
+ 1 − 1  13.8 T , (49)
2  0.3 × 3.3 ×10 ×1.039 × 2.05 ×10 
which becomes 11 T when the 20% margin is imposed. The short sample current is 670 A·mm−2:
B 13.8
jss = ss = =6.7 ×108 A ⋅ m −2 . (50)
γ 0 w 6.6 ×10−7 × 0.031
3.4 Field versus coil width in superconducting quadrupoles

The same approach can be used to work out the quadrupole case [6]. For the Nb−Ti,
κ sγ 0 ln(1 + w / r ) Gss
Gss = b jss = Bp,ss = λ rGss , (51)
1 + κ rsλγ 0 ln(1 + w / r ) γ 0 ln(1 + w / r )
where λ is given by Eq. (40). A plot giving the short sample gradient versus the coil width and the
aperture is shown in Fig. 16. The Nb3Sn technology gives ~50% more gradient for the same aperture
w.r.t. Nb-Ti.
284
Fig. 16: Short sample gradient versus coil width and aperture for Nb−Ti and Nb3Sn; filling factor κ = 0.3
Example: We apply this equation to the case of the large-aperture Nb-Ti quadrupole developed
for the LHC upgrade. It has aperture r = 60 mm and coil width w = 31 mm (two layers of the LHC
main dipole cable), and therefore the peak field to bore field ratio is given by
r w 60 31
λ = 1 + a1 + a−1 = 1 + 0.04 + 0.11 = 1.13 . (52)
w r 31 60
The log factor is
 w  31 
ln1 +  = ln1 +  = 0.42 , (53)
 r  60 
and therefore one has
0.3 × 5.0 ×108 × 6.6 ×10−7 × 0.42

Gss
= 8 −7
=123 141 T ⋅ m −1 , (54)
1 + 0.3 × 0.06 × 5.0 ×10 ×1.13 × 6.6 ×10 × 0.42
quite close to the real value of 149 T·m−1. The current density and peak field are
141
jss = −7
=5.1×108 A ⋅ m −2 and Bp,ss = 1.13 × 0.06 ×141 = 9.6 T . (55)
6.6 ×10 × 0.42
For the Nb3Sn case, the equations become more cumbersome, but are still manageable in a
spreadsheet:
κ sγ 0 ln(1 + w / r )  4b 
=Gss  + 1 − 1 , (56)
2  κ rsλγ 0 ln(1 + w / r ) 
and the same expressions hold for the short sample peak field and current density as in Eq. (51).
Example: We consider the case of the Nb3Sn quadrupole for the LHC upgrade QXF, having
r = 75 mm and w = 36 mm, and therefore
r w 75 36
λ = 1 + a1 + a −1 = 1 + 0.04 + 0.11 = 1.14 . (57)
w r 36 75
The log factor is
 w  36 
ln 1 +  = ln 1 +  = 0.39 , (58)
 r  75 
and therefore we have
285
E. T ODESCO
κ sγ 0 ln(1 + w / r ) = 0.32 × 3.3 ×109 × 6.6 ×10−7 × 0.39 = 274 T ⋅ m −1 , (59)
274  4 × 24 
G
=ss  1 173 T ⋅ m −1 ,
+ 1 −= (60)
2  274 ×1.13 × 0.075 
very close to the real value of 169 T·m−1. The current density and peak field are
173
jss = −7
=6.7 ×108 A ⋅ m −2 and Bp,ss = 1.14 × 0.075 ×173 = 14.7 T . (61)
6.6 ×10 × 0.39
4 The role of iron

Iron has several functions: (i) to keep the magnetic flux within the magnet, (ii) to prevent fringe fields
in the tunnel that may endanger the electronics, and (iii) to reduce the current density in the magnet via
the creation of a virtual coil. Since the iron can retain a maximum of Bs ~ 1.5–2 T, for a dipole the iron
thickness t needed for shielding is
rB1
t~ . (62)
Bs
For example, in the LHC dipole we have r = 28 mm and B1 = 8.3 T, yielding t ~ 100 mm. For a
quadrupole, the shielding condition is given by
r 2G
t~ . (63)
2 Bs
So, for the LHC main quadrupole, G = 220 T·m−1, r = 28 mm, and t ~ 40 mm.
Inside the aperture, the iron acts as a virtual second-layer coil (see Fig. 17). According to the
current image method (see, e.g., [1], p. 53), a thick ring of iron at a distance RI from the centre of the
aperture produces an image current of a current line ρ, which is at a distance ρʹ and carries current Iʹ,
where
RI2 µ −1
ρ'= , I'= I~I. (64)
ρ µ +1
This approximation is valid for non-saturated iron. Above 2 T, iron becomes saturated, and the effect
of the virtual coil decreases. In this case, a finite element code [15, 16] should be used to construct the
calculations. Geometries that do not satisfy the circular symmetry in the non-saturated case can be
treated as a first approximation with an radius of an iron ring. For non-saturated iron, integrating over
a sector coil of width w in an aperture r leads to a virtual coil of inner radius R1 and outer radius R2
(see Fig. 17, right), giving an additional field
∆B1 = γ 0 J ( R2 − R1 ) , (65)
286
ρ'
RI
R1 R2
ρ
RI
Fig. 17: Impact of iron: current image produced by iron at RI (left), and sector coil with virtual coil provided by
iron (right).
Note that the current densities of the virtual coil J and of the coil j are not the same, since the total
current is the same but the area of the virtual coil is much larger:
(r + w) 2 − r 2
[ ] (
j (r + w) 2 − r 2 = J R22 − R12 ) → J (R 2 − R1 ) = j
R 2 + R1
; (66)
therefore, substituting Eq. (66) into Eq. (65) we obtain,
(r + w) 2 − r 2
∆B1 = γ 0 j , (67)
R2 + R1
and, since R2 = RI2/r and R1 =RI2/(r + w),
(r + w) 2 − r 2 γ 0 j (r + w) 2 − r 2 γ 0 j
∆B1 γ 0 j
= = = wr (r + w) . (68)
R2 + R1 RI2 1 + 1 RI2
r r+w
Finally, after all this algebra, we obtain a beautiful, simple expression: the iron simply increases the
field of a fraction f, which is a function of the aperture, of the coil width, and of the inner radius of the
iron:
∆B1 r (r + w)
f ≡ =2 . (69)
B1 RI
This equation is a generalization of the estimate for a thin coil [1], p. 54, to a thick coil. In the most
favourable case w << r and RI = r + w (thin coil surrounded by iron without spacers or collars), the
iron can double the main field. Typically, one needs some space between the coil and the iron (collars
or spacers 10 to 50 mm thick, see Fig. 18), and typical gains are in the range 10–50%. The largest gain
is not only for small collars, but also for thin coils, as in the RHIC case (see Fig. 19). In the LHC case,
the gain is ~17%.
287
E. T ODESCO
Fig. 18: Cross-section of the RHIC dipole [7] (left), with thin spacers, and of the LHC dipole [11] (right), with
thick collars.
+15% +20%
+25%
3 +30%
RI/r (adim)
TEV MB HERA MB
+40% SSC MB RHIC MB
2 +50% LHC MB Fresca
MSUT D20
Forbidden zone
HFDA NED
1
0.0 0.5 1.0 1.5 2.0
equivalent width w/r
Fig. 19: Impact of iron: increase in main field vs. iron radius and coil width
Iron not only increases the field for a given current density, but also lowers the magnet loadline,
thereby increasing the short sample field: the effect of including the iron is to increase the coefficient
γ0 of the fraction f (see Eq. (69)). For example, in a Nb-Ti dipole we have
κ sγ 0 (1 + f ) w r (r + w)
Bss = b, f = . (70)
1 + κ sλγ 0 (1 + f ) w RI2
Example: We can recompute the LHC dipole case adding the 17% to obtain
0.3 × 5.0 ×108 × 6.6 ×10−7 ×1.17 × 0.031
Bss = 13 9.9 T , (71)
1 + 0.3 × 5.0 × 108 × 1.036 × 6.6 × 10−7 × 1.17 × 0.031
i.e. 3% more of that estimated without iron. Indeed, there is a beneficial decrease of the short sample
current, which decreases from 470 A·mm−2 (see Eq. (46)) to
B 9.9
jss = ss = =4.1×108 A ⋅ m −2 ; (72)
γ 0 w 6.6 ×10 ×1.17 × 0.031
−7
this reduces stresses and provides a greater margin for magnet protection.
Example: It is interesting to consider the extreme case of the RHIC dipole (r = 40 mm,
w = 10 mm, filling factor κ = 0.22, operational temperature 4.2 K); without iron, we would have a
short sample field of
0.22 × 5.0 ×108 × 6.6 ×10−7 × 0.010
Bss = 9 3.6 T , (73)
1 + 0.22 × 5.0 × 108 × 1.16 × 6.6 × 10−7 × 0.010
288
and with the iron we obtain

0.22 × 5.0 ×108 × 6.6 ×10−7 ×1.56 × 0.010
Bss = 9 4.4 T , (74)
1 + 0.22 × 5.0 × 108 × 1.16 × 6.6 × 10−7 × 1.56 × 0.010
i.e. a considerable gain of 20% additional short sample field. The short sample current without iron is
B 4.4
jss = ss = =5.4 ×108 A ⋅ m −2 , (75)
γ 0 w 6.6 ×10 × 0.010
−7
and with the iron it decreases by 20%, from 540 A·mm−2 to 430 A·mm−2:
B 4.4
jss = ss = =4.3 ×108 A ⋅ m −2 . (76)
γ 0 w 6.6 ×10 ×1.56 × 0.010
−7
5 Other layouts
The sector coil is the workhorse of accelerator magnets. All superconducting magnets installed in an
accelerator are based on the cos-theta layout, which is achieved using sector coils. Alternative layouts
[17, 18] have been proposed and implemented in short models, with mixed results. Here we briefly
discuss the block coil, and how to extend the previous equations to the case of a layout not based on
cos-theta. The winding of a block coil is similar to that of a solenoid: cables are perpendicular to the
horizontal midplane, rather than being parallel as in the cos-theta layout. Moreover, cables are not
keystoned, and they form rectangular blocks arranged in layers. If a layer is above the beam tube, it
can be wound as a flat racetrack coil, but if it is close to the midplane the end must be opened to make
space for the beam tube (flared ends). An example is the beautifully simple layout of HD2 (see
Fig. 20).
Fig. 20: HD2, a dipole based on the block concept
At present, there is no consensus on the feasibility and advantages of the block layout. We make
the following remarks, although we are aware that some of them are not universally shared.
• The main feature is that the cos-theta layout is self-supporting, with no need for an internal
structure. On the other hand, the block needs an inner tube to support the forces acting on the
coil, reducing the aperture available to the beam.
• The block layout is not viable for a small coil to aperture ratio w/r, as in the RHIC magnet. If
the coil is thin, it does not make sense to arrange the conductors in a block geometry. On the
other hand, for large w/r >> 1 (~2 in HD2), the block coil becomes an interesting option.
289
E. T ODESCO
• Both layouts are equivalent from the point of view of field quality optimization. Multipoles
can be cancelled with a few free parameters, i.e. by adding wedges. Note that the HD2 case
does not need wedges because the ratio w/r is large; see Eq. (19).
• The block layout advantage of a simple coil geometry, squared and without wedges, is
apparent: the winding of cos-theta coils with wedges is a well-mastered technology, including
the ends. On the other hand, for the block layout the mechanical support of the coil ends and
the transition in the flared region have not yet been mastered. To be fair, one also has to take
into account the fact that we are comparing a few block short models with several thousands
of cos-theta magnets.
• The forces accumulate on the midplane in a cos-theta geometry, whereas in a block coil they
occur mainly in the horizontal direction.
• With respect to the cos-theta layout, the block has a somewhat larger value of λ and a lower
γ0, so the design is less effective [19]. This difference vanishes for very large coils.
• The block layout is very flexible, allowing the number of turns in the outer part to be adjusted,
whereas for a cos-theta layout one has to add a whole layer (or change the cable width). Since
the level curves of the field are mostly parallel to the vertical direction, in the block layout it is
easier to have material grading.
One can use the equations derived in this paper for block layouts (or for any layout) by replacing the
coil width w by an equivalent coil width we, defined as the width of the 60º sector coil which has the
same quantity of superconductor of the ‘exotic’ design, and the same aperture r. Let A be the cross-
sectional surface area of the coil; we equate it to the surface area of a sector coil as follows:
2π
A≡ (r + we ) 2 − r 2  . (77)
3 
From this definition, we obtain
3A 
2  we 
2
  3A 
= r 1 +  − 1 , i.e. we = 1 + − 1  r . (78)
2π  r    2π r 2 
Example: HD2 has a 20 mm aperture, and makes use of rectangular insulated cable of 22.2 mm
width and 1.62 mm thickness, i.e. 36 mm2. There are 24 and 30 turns in each layer, thus giving a total
coil cross-sectional surface area of 7800 mm2. This gives
 3 × 7800 
we =  1 + − 1 × 20= 44 mm . (79)
 2π × 20 × 20 
6 Putting it all together

The starting point for magnet design is the aperture. This is usually a factor that we accept as an input
to our calculations, i.e. it is a requirement coming from accelerator physicists. The main choices are:
• technology and operational temperature, which fix a maximum attainable field;
• bore field or gradient;
• current density;
• width of the coil, i.e. how much superconductor to use;
• fraction of stabilizer, which dilutes the superconductor.
290
These five quantities are related via equations that can be implemented in a spreadsheet, namely (as
discussed in Section 3), Eqs. (43) and (47) for dipoles, and Eqs. (51) and (56) for quadrupoles, for the
Nb−Ti and Nb3Sn cases, respectively. Figures 14 (dipoles) and 15 (quadrupoles) give these coupled
dependences in the parameter space for accelerator magnets, and can be a useful instrument for a
quick estimate of what can be done, what is better not to try, and what is impossible. Out of the above-
mentioned quantities, the current density is probably the most relevant parameter in magnet design: it
affects many different aspects, such as protection, mechanical structure and stresses, and stability.
Large current densities provide more compact magnets, but as can be observed from examining the
history of accelerator magnets, most cases range between 300 and 500 A·mm−2 (see Figs. 21 and 22).
Fig. 21: Bore field and current density (both at 20% margin) vs. coil width in dipoles
Fig. 22: Peak field and current density (both at 20% margin) vs. coil width in quadrupoles
Given the choice of field (peak field for quadrupoles) and coil width, one has to satisfy field
quality conditions using the material presented in Section 2. An adequate number of layers and
wedges are needed to tune the field harmonics; on the other hand, the unnecessary complexity of too
many layers and wedges should be avoided. The main principles of field quality optimization for
dipoles and quadrupoles, pointing out the main features, have been discussed. When taking into
account that we have a finite size cable and imperfect keystoning, plus other effects not discussed
here, such as persistent current and saturation, one has to implement a second fine tuning of the field
quality. This is achieved via a code, sometimes through optimization algorithms, or by hand. After
construction, a third iteration is usually necessary to apply another fine tuning of the field quality. The
capability to go through successive approximations is a fundamental approach of magnet design.
Acknowledgements
We wish to thank D. Tommasini, M. Juchno, and S. Izquerdo Bermudez for their critical reading of
the manuscript and valuable comments. These notes are based on material prepared for the USPAS
lectures, in collaboration with H. Felice, P. Ferracin, and S. Prestemon.
291
E. T ODESCO
References
[1] K.H. Mess, P. Schmuser and S. Wolff, Superconducting Accelerator Magnets (World Scientific,
Singapore, 1996).
[2] F.M. Asner, High Field Superconducting Magnets (Clarendon press, Oxford, 1999), pp. 132–154.
[3] J. Di Marco et al., IEEE Trans. Appl. Supercond. 24 (2014) in press.
[4] M.N. Wilson, Superconducting Magnets (Clarendon press, Oxford, 1983), pp. 27–32.
[5] L. Rossi and E. Todesco, Phys. Rev. STAB 10 (2007) 112401.
[7] M. Anerella et al., Nucl. Instrum. Methods A499 (2003) 280.
[8] W.E. Cooper et al., Fermilab Report TM-1183 (1983).
[9] R. Kreutz, in Proc. European Particle Accelerator Conf. (1992), p. 1423.
[10] S. Wolff, in The Superconducting Magnet System for HERA, Eds. C. Marinucci and P.
Waymuth (SIN, Zurich,1995).
[11] L. Rossi, IEEE Trans. Appl. Supercond. 14 (2004) 153.
[12] A. Devred, Practical low-temperature superconductors for electromagnets, CERN 2004-006
(2004).
[13] L. Bottura, IEEE Trans. Appl. Supercond. 10 (2000) 1054.
[14] E.J. Kramer, J. Appl. Phys. 44 (1973) 1360.
[15] OPERA, Vector fields.
[16] S. Russenschuck, Field Computation for Accelerator Magnets: Analytical and Numerical
Methods for Electromagnetic Design and Optimization (Wiley-VCH, Zurich,2010).
[17] R. Gupta, in Proc. Particle Accelerator Conf. (1997), p. 3344.
[18] G. Sabbi et al., IEEE Trans. Appl. Supercond. 15 (2005) 1128.
[19] L. Rossi and E. Todesco, IEEE Trans. Appl. Supercond. 19 (2009) 1186.
292
Mechanical Design of Superconducting Accelerator Magnets
F. Toral 1
CIEMAT, Madrid, Spain
Abstract
This paper is about the mechanical design of superconducting accelerator
magnets. First, we give a brief review of the basic concepts and terms. In the
following sections, we describe the particularities of the mechanical design
of different types of superconducting accelerator magnets: solenoids, cos-
theta, superferric, and toroids. Special attention is given to the pre-stress
principle, which aims to avoid the appearance of tensile stresses in the
superconducting coils. A case study on a compact superconducting cyclotron
summarizes the main steps and the guidelines that should be followed for a
proper mechanical design. Finally, we present some remarks on the
measurement techniques.
Keywords: superconducting accelerator magnets, mechanical design, pre-

stress, electromagnetic forces.
1 Introduction
The designer of a superconducting magnet will be concerned about achieving a very good magnetic
field quality and protecting the magnet in case of quench, but he or she should not forget that
mechanical failures are the cause of performance loss in superconducting magnets, compared with that
predicted by the electromagnetic computations.
Superconducting accelerator magnets are characterized by large fields and current densities. As
a result, coils experience large stresses, which have three important effects.
i) Quench triggering: the most likely origin of quench is the release of stored elastic energy when
part of the coil moves or a crack suddenly appears in the resin. Due to the low heat capacity
of materials at low temperatures, the resulting energy deposition is able to increase the
temperature of the superconductor above its critical value.
ii) Mechanical degradation of the coil or the support structure: if the applied forces/pressures are
above a given threshold (yield strength), plastic deformation of the materials takes place.
iii) Field quality: the winding deformation may affect the field quality.
The parts of the magnet are produced and assembled at room temperature, but their working
temperature is about −270ºC. The designer must consider carefully the differential thermal
contractions of materials during cool-down and operation.
Taking into account the aforementioned aspects, the mechanical design will aim to:
i) avoid tensile stresses on the superconductor;
ii) avoid mechanical degradation of the materials;
iii) study the magnet life cycle: assembly, cool-down, energizing, and quench.
1
fernando.toral@ciemat.es
F. T ORAL
Figure 1 shows the strategy that the designer should follow during the mechanical analysis of a
superconducting accelerator magnet.
Calculate electromagnetic forces
Design a support structure

(if necessary) to avoid
tensile stresses in the
conductors
No
Is it compatible with the
assembly process cooling-down and
quench?
Yes
End
Fig. 1: Strategy for the mechanical design of superconducting magnets
2 Basic concepts
Some basic concepts from electromagnetism and elasticity theory will be reviewed in this section,
with special attention paid to the particular expressions used in the following sections.
2.1 Electromagnetic forces

A charged particle q moving with speed v in the presence of an electric field E and a magnetic field B
experiences the Lorentz force, which is given by
   
F [ N=] q ( E + v × B ) . (1)
In the same way, a conductor element carrying current density j in the presence of a magnetic
field B will experience the force density
  
f L  N ⋅ m −3  =
j×B. (2)
The Lorentz force is a body force, i.e. it acts on all the parts of the conductor, as does the
gravitational force. The total force on a given body can be computed by integration:
 
FL [ N ] = ∫∫∫ f L dv . (3)
The magnetic energy density u stored in a region without magnetic materials (µr = 1) in the
presence of a magnetic field B is
 
B ⋅ H B2
u  J ⋅ m  =
−3
= . (4)
2 2 µ0
The total energy U can be obtained by integration over all the space, by integration over the coil
volume, or by knowing the so-called self-inductance L of the magnet:
 
B⋅H   1 2
U [ J=
] ∫∫∫ d=v ∫∫∫
A ⋅ j d=
v LI . (5)
all 2 coil 2
294
M ECHANICAL D ESIGN OF S UPERCONDUCTING ACCELERATOR M AGNETS
The stored energy density may be understood as a ‘magnetic pressure’, pm (see Eq. (6)). In a
current loop, the magnetic field line density is higher inside: the field lines try to expand the loop, like
a gas in a container. The magnetic pressure is given by
B2
pm  N ⋅ m −2  =. (6)
2 µ0
Fig. 2: Magnetic field lines created by a current loop (graph courtesy of www.answers.com)
2.2 Stress and strain

In continuum mechanics, stress is a physical quantity which expresses the internal pressure that
neighbouring particles of a continuous material exert on each other. As shown in Fig. 3(a), when the
forces are perpendicular to the plane, the stress is called normal stress (σ); when the forces are parallel
to the plane, the stress is called shear stress (τ). Stresses can be seen as the way a body resists the
action (compression, tension, sliding) of an external force. A tensile (pulling) stress is considered as
positive, and is associated with an elongation of the pulled body. As a consequence, a compressive
(pushing) stress is negative and is associated with a body contraction [1]. The normal and shear
stresses are given by
F
σ= [ Pa ] z  N ⋅ m −2  (7)
A
and
F
[ Pa ]
τ= [N ⋅ m ] .
y −2
(8)
A
xy
Fz Fz
δ
z
Fy
A
y
l0
x
(a) (b)
Fig. 3: (a) Normal and shear stresses. (b) Strain
A strain ε is a normalized measurement of deformation representing the displacement δ between

particles in the body relative to a reference length l0 (see Fig. 3(b)):
δ
ε= . (9)
l0
295
F. T ORAL
According to Hooke’s law (1678), within certain limits, the strain ε of a bar is proportional to
the exerted stress σ. The constant of proportionality is the elastic constant of the material, the so-called
modulus of elasticity E, or Young modulus:
δ σ F
ε= = = . (10)
l0 E AE
The Poisson ratio ν is the ratio of ‘transverse’ to ‘axial’ strain:

ε transversal
ν= − . (11)
ε axial
When a body is compressed in one direction, it tends to elongate in the transverse direction.
Conversely, when a body is elongated in one direction, it gets thinner in the other direction. The
typical value is around 0.3.
A shear modulus G can be defined as the ratio of the shear stress τ and the shear strain γ:
τ xy E
G= = . (12)
γ xy 2(1 + ν )
The proportionality between stress and strain is usually more complicated than suggested by
Hooke’s law (see Eq. (10)). Figure 4 shows the stress–strain graph for a typical material. The
following individual points should be noted.
i) Point A shows the limit of proportionality. The first section of the curve is a straight line, in
accordance with the linear behaviour described by Hooke’s law.
ii) Point B is known as the yield point. It is usually defined as the point where the permanent
deformation is 0.2%.
iii) Point C shows the ultimate strength. Beyond this point, the strain increases, even at lower
stresses.
iv) Label D corresponds to the fracture point.
σ C
D
B
O
ε
Fig. 4: Stress vs. strain graph for a typical material [1]
Several failure criteria are defined to estimate the failure/yield of structural components. One of
the most broadly used is the equivalent (von Mises) stress σv, given by
( σ 1 − σ 2 ) 2 + (σ 2 − σ 3 ) + (σ 3 − σ 1 )
2 2
σv = , (13)
2
where σ1, σ2, and σ3 are the principal stresses.
296
2.3 Material properties

The properties of a material must be well known to perform a proper mechanical design.
Superconducting windings are composites, i.e. mixtures of different materials. As a first approach, the
magnet designer uses smeared-out properties of the winding, taking into account the volumetric
fraction of each material and its distribution. For example, Table 1 shows the main properties of some
materials commonly used in Nb−Ti coils. Figure 5(a) depicts a simple Nb−Ti coil wound with round
wire, and Fig. 5(b) shows the model used to obtain the smeared-out properties (see Table 2),
calculated by numerical methods. Material properties are strongly dependent on temperature. The
designer may find some dispersion in the values depending on the source: Refs. [2], [3], and [4] can be
used as general references.
Table 1: Physical properties of some typical materials used in Nb−Ti coils (at 4.2 K)
Young
Material Poisson Shear modulus Integrated contraction References
modulus
ratio (GPa) (296 to 4.2 K)
(GPa)
Nb−Ti 77 0.3 20 1.87E-3 [5]
Nb−Ti wire 125 0.3 48 2.92E-3 [6]
Copper 138 0.34 52 3.15E-3 [7]
Varnish insulation 2.5 0.35 0.93 10.3E-3 [8]
Epoxy 7 0.28 2.75 6.40E-3 [9]
z θ
r
(a) (b)
Fig. 5: (a) Simple solenoid winding. (b) Sub-model used to obtain the smeared-out mechanical properties
Table 2: Smeared-out mechanical properties of Nb−Ti winding (at 4.2 K)
Young modulus Poisson Shear modulus Integrated thermal

(GPa) ratio (GPa) contraction
(296 to 4.2 K)
θ 94 2.99E-3
r 35 3.90E-3
z 35 3.93E-3
rθ 0.08
zθ 0.08
rz 0.35 24
297
F. T ORAL
Note the significant differences between the integrated thermal contractions of the materials
shown in Table 1. The magnet assembly is always made at room temperature. Obviously, the magnet
designer needs to analyze the induced stresses due to the different contraction coefficients of glued or
clamped parts during the cooling down. Special attention must be paid to the degradation of insulating
materials.
3 Solenoids
Solenoids will be the first type of coils to be reviewed, due to their simple geometry. In this case, the
analytical expressions are easily deduced. Numerical methods will then be described, and some
remarks on their advantages and risks will be included.
3.1 Thin-wall solenoids

In an infinitely long solenoid carrying current density j, the field inside is uniform and outside is zero.
Lorentz forces push the coil outwards in a purely radial direction, creating a hoop stress σθ on the
wires. We assume that the wall thickness is very small.
First, using Ampère’s law, we can compute the field inside the solenoid:
 
∫
B ⋅ dl = µ 0 I ⇒ B0 ∆z = µ 0 jw∆z . (14)
Assuming that the coil’s average field is B0/2, the magnetic pressure pm is given by the
distributed Lorentz force FL:
  B  
FL=
r ur f Lr ( a ∆θ∆zw )=
ur j 0 ( a ∆θ∆zw )=
ur pm ( a ∆θ∆z ) ur , (15)
2
B02
pm = , (16)
2 µ0
where fLr is the density of the Lorentz force in the radial direction. It is important to note that the
magnetic pressure increases with the square of the field. For example, in an infinitely long solenoid
with a central field of 10 T, the windings undergo a pressure of 398 atm!
The simplest stress calculation is based on the assumption that each turn acts independently of
its neighbours. Based on the equilibrium of forces on half a solenoid, as shown in Fig. 6(b), one may
compute the hoop stress σθ :
π 2
=2F ∫=
π
−
f cos θ aw dθ
2 Lr 2 f Lr aw,
(17)
B02 B02 a a
F = f Lr aw = pm a ⇒ σ θ w = a ⇒ σθ = = pm .
2 µ0 2 µ0 w w
In general, the peak stress occurs in the innermost turn, where the magnetic field is also
maximum. Using Eq. (2), and assuming that B is the field at the innermost turn, located at radius a, the
peak stress can be calculated as
σ max = BJa ∝ J 2 . (18)
The reader should note that the peak stress increases with the square of the current density. In
our example, assuming an inner radius of 10 cm and a thickness of 10 mm, the peak stress is about
400 MPa. It is too high for a Nb3Sn winding (yield stress ∼150 MPa) and possibly even for a Nb−Ti
winding (yield stress ∼500 MPa), assuming a filling factor of 70%. In that case, how should one build
298
a robust solenoid able to create a central field of 10 T? The solution is to exert a pre-stress on the
winding that correspondingly decreases the tensile azimuthal stress σθ.
 
a w F f Lr
B0 pm
∆z ∆θ

F
(a) (b)
Fig. 6: (a) Uniform magnetic field inside a solenoid. (b) Radial pressure on the solenoid due to Lorentz forces
3.2 Thick-wall solenoids

Figure 7(a) shows the magnetic field map of a typical thick solenoid winding, and Fig. 7(b) depicts the
Lorentz forces when it is energized. The wall thickness is not negligible compared with the length.
The electromagnetic forces tend to push the coil:
− outwards in the radial direction (Fr > 0);
− towards the mid-plane in the axial direction (Fy < 0 in the upper half coil, and the opposite in the
lower half).
Fig. 7: (a) Magnetic field lines created by a thick solenoid winding. (b) Lorentz forces (graphs from [10])
Figure 8 shows the azimuthal stress distribution for long solenoids with two different shape
factors. The shape factor is the ratio of the outer and the inner radii. The label σʹθ is given to the
curves depicting the hoop stress when we assume that the turns act independently, which is a poor
approximation. The label σθ shows the hoop stress calculated when we assume that adjacent turns
press on each other, developing radial stresses. Note that thin solenoids perform negative radial
stresses, whereas thick solenoids show regions with positive radial stresses, i.e. tensile stresses. In the
latter case, there is a risk of resin cracking or wire movement, which could trigger a quench. In
summary, long and thin solenoids are mechanically more stable than thick ones.
299
F. T ORAL
(a) (b)
Fig. 8: (a) Azimuthal stress σθ distribution in long solenoids with shape factors (a) 1.3, (b) 4.0 [10]
3.3 Application example

Figure 9(b) shows the set-up prepared to test a superconducting solenoid with a superconducting
switch. This magnet should be used as a mock-up of the Alpha Magnetic Spectrometer (AMS) main
magnet to test its power supply [11]. It is wound with an enamelled Nb-Ti round wire. In the first
cool-down, the learning curve was very slow (see Fig. 10). The magnet quenched at very low currents,
below half the short sample limit, while the nominal working point was at 75% on the load line. After
warm-up, the (re)-training did not improve; indeed, a slight de-training (quench current lower than in
the previous quenches) was observed in the first quench. It is clear that a mechanical problem limits
the magnet performance. It was decided to stop the training tests and to analyze the mechanics
carefully.
Length 123.75 mm
Inner diameter 188.72 mm
Outer diameter 215 mm
Number of turns 1782 Thermal shield
Nominal current 450 A SC solenoid

Peak field 5.89 T G11 bobbin
Bore field 4.25 T Spacer
Current density 493 A·mm-2 SC switch
Working point 75%

Self inductance 0.5 H Cryostat
(a) (b)
Fig. 9: (a) Solenoid parameters. (b) Solenoid test set-up
300
Trainingimán
Entrenamiento testCRISA
300
250
(A)(A)
200
Intensidad
150
Current
100
Re-training
50
0
0 2 4 6 8 10
Quench number
Nº de quench
Fig. 10: Learning curve of the superconducting solenoid: first training and re-training
The solenoid was wound on a G10 bobbin and wet impregnated with epoxy resin. The
integrated (from 300 down to 4.2 K) thermal contraction coefficient of the glass fibre and the winding
are quite different, 280E-5 and 392E-5, respectively. The detailed mechanical analysis was not made
before the magnet fabrication because it was not considered necessary for this small test coil.
However, even for such a small magnet, the wrong mechanical design may spoil the performance.
When the Finite Element Method (FEM) numerical analysis was performed, tensile stresses up to
46 MPa were detected in the coil ends (see Fig. 11(a)). These are able to crack the epoxy resin,
triggering the premature quenches. It was decided to turn the bobbin core, and to split it into two
different parts (see Fig. 11(b)). Now the coil is working mainly under compression. When the magnet
was cooled down, the training improved significantly (see Fig. 12, yellow and light blue dots). It
finally reached the nominal current after a few quenches.
(a) (b)
Fig. 11: Axial (vertical) stress: (a) continuous G10 bobbin; (b) split G10 bobbin
Fig. 12: Learning curve of the superconducting solenoid: complete training
301
F. T ORAL
3.4 Numerical vs. analytical methods

In this section, analytical and numerical methods are applied to the calculation of the stresses present
in solenoid windings. Analytical expressions are valid only for simplified models, while numerical
methods are able to model very precisely the real magnet, i.e. the
− anisotropic material properties;
− complicated/detailed geometry;
− ‘sophisticated’ boundary conditions: sliding/contact surfaces, joints;
− load steps: assembly, cooling down, energizing;
− transient problems.
There is a common temptation to forget about the analytical approach and start the analysis
directly with the numerical simulations. This is not the most effective strategy, however. The
analytical methods have to be used first, because they allow us to:
− understand the problem and the physics behind it;
− make a first estimate of the solution;
− simplify the numerical simulation;
− check and understand the results of the numerical simulations.
4 Cos-theta accelerator magnets

This type of magnet is the most common in particle accelerators, since the superconductor efficiency
is very high (the current distribution is very close to the aperture) and permits very high magnetic
fields to be achieved. The geometry is relatively complicated, especially at the coil ends.
4.1 Lorentz forces

The Lorentz forces in an n-pole magnet tend to push the coil:
− towards the mid plane in the vertical/azimuthal direction (Fy, Fθ < 0);
− outwards in the radial–horizontal direction (Fx, Fr > 0).
(a) (b) (c)
Fig. 13: (a) Cos-theta dipole winding; (b) field map on the coil (c) electromagnetic forces [12]
At the coil ends, the Lorentz forces tend to make the coil longer. The forces are pointing
outwards in the longitudinal direction (Fz > 0).
302
In short, the electromagnetic forces try to expand the coil, as in a current loop. However, the
coil by itself is unable to support the magnetic forces in tension. These forces must be counteracted by
an external support structure.
In order to estimate the value of these forces, three different approximations can be considered
for any n-pole magnet (see Fig. 14) [12]:
− Thin shell: the current density may be expressed as J = J0 cos nθ (ampere per unit
circumference), where J0 is a constant. This is the simplest model. It allows us to estimate
orders of magnitude and proportionalities.
− Thick shell: the current density may be written as J = J0 cos nθ (ampere per unit area). This
model may be used to get a first-order estimate of forces and stresses.
− Sector: the current density J is constant (ampere per unit area). The sector spans an angle
θ = 60º(30º) for a dipole(quadrupole), to make zero the first magnetic field harmonic. This
model may be used to obtain a first-order estimate of forces and stresses.
(a) (b) (c)

Fig. 14: Current density in the different winding approximations: (a) thin shell, (b) thick shell, and (c) sector
4.1.1 Electromagnetic forces on a thin shell

Beth’s theorem states that the complex force on a current element (per unit length in the longitudinal
direction z) is equal to the line integral of magnetic pressure around the boundary of that element in
the complex plane:
 B2
F = Fy + iFx = − ∫ dz . (19)
2 µ0
For a cylindrical current sheet, the total force [N·m−1] on half a coil is [10]:
 1 π 2n 2
= F
2 µ0 ∫0
( 2
Bin − Bout )i ae iθ d θ . (20)
If the current density is given by J = J0 cos nθ [A·m−1], and assuming an iron with infinite
permeability µ = ∞ placed at radius R (see Fig. 15), the density force f [N·m−2] is given by [12–14]
µ J 2  
2n 2n
  a   −iθ 2 n −1 a 
f = f x + if y = 0 0  1 + 2    e ( ) − e ( ) + 2   eiθ  .
iθ 2 n +1
(21)
8    R    R 


303
F. T ORAL
y  
t Fy n
Fx
a θ
x
R iron
Fig. 15: Electromagnetic forces on a cylindrical current sheet surrounded by iron with infinite permeability
The tangential and normal components of the density force may be obtained by calculating the
dot products with the tangent and normal unit vectors, t and n, respectively:
   iθ µ0 J 02  a  2 n
fr = f ⋅ n = f ⋅ e =   1 + cos ( 2nθ )  , (22)
4 R 
   µ0 J 02  
2n
a
fθ = f ⋅ t = f ⋅ ei(π  sin ( 2nθ ) .
2 +θ )
= − 1 +   (23)
4   R  
For a dipole, the force on half the coil is given by

µ0 J 02  1  a  
2
π 2
Fx  N ⋅ m −1=
 ∫0 ( f r cos θ − fθ sin θ )a d=
θ  +  a; (24)
2  3  R  
π 2 µ0 J 02 1
∫0 r
Fy  N ⋅ m −1  = ( f sin θ + fθ cos θ )a d θ =
− a. (25)
2 3
It is proportional to the bore radius and the square of the current density (and field). The term
containing the iron radius R is the contribution from the iron, which can be easily distinguished from
the contribution from the conduction current.
In a rigid magnet structure, the force determines an azimuthal displacement of the coil and
creates a separation at the pole (see Fig. 26). The structure should withstand Fx. Meanwhile, Fy
provides a compression on the coil itself, with a maximum stress at the mid-plane. If one thinks of a
coil working as a “roman arch”, where all the hoop forces fθ accumulate on the mid-plane, the total
force Fθ transmitted on the mid-plane per unit length of the magnet is
π 2 By2
Fθ = ∫ fθ a dθ = − a. (26)
0 µ0
Furthermore, one can consider a real winding as a set of current sheets and solve the problem by
superposition. This method allows us to compute the magnetic field, the stored magnetic energy, and
the electromagnetic forces [13]. As an application example, Fig. 16 shows one coil of a corrector
quadrupole prototype magnet developed for LHC, the so-called MQTL. It is split into a set of thin
shells. Table 3 shows good agreement in the results using three different methods: the superposition of
thin shells, a BEM–FEM numerical calculation (ROXIE [15]), and FEM numerical calculation.
304
Normalized current density

(a)
Angular position (deg)

(b)
Fig. 16: (a) MQTL winding: detailed view of the subdivision in thin shells. (b) Current density in one of the thin
shells.
Table 3: Magnetic field, stored energy, and forces in MQTL magnet using different computation methods
Thin
Magnitude ROXIE FEM Units
shells
Gradient 2.99 3.00 2.99 T·m−1
b6 98.4854 98.4616 98.1640 1E-4
b10 1.1899 1.1871 1.4283 1E-4
b14 0.0152 0.0152 0.2352 1E-4
Bmax 0.366 0.368 0.379 T
L 0.0547 0.0546 0.0548 mH·m−1
Fx 51.74 50.30 48.955 N
Fy −118.26 −116.57 −115.27 N
4.1.2 Electromagnetic forces on a thick shell

Assuming that the current density is J = J0 cos nθ, where J0 is measured in A·m−2, the shell inner
radius is a1, the outer radius is a2, and no iron is present, the radial and azimuthal components of the
magnetic field Bi inside the aperture of an n-pole magnet are given by [12, 14]
µ J  a 2 − n − a12 − n 
Bri = − 0 0 r n −1  2  sin nθ , (27)
2  2−n 
µ0 J 0  a 2 − n − a12 − n 
Bθ i = − r n −1  2  cos nθ . (28)
2  2−n 
The radial and azimuthal components of the field at the coil may be written as follows:
µ0 J 0  n −1
 a22 − n − r 2 − n 

 2+n 2+n 

 + 1  r 1−+ na1  sin nθ ,
Br = − r   (29)
2   2−n  2+n r
   
305
F. T ORAL
µ0 J 0  n −1
 a22 − n − r 2 − n 

 2+n 2+n 

 − 1  r 1−+ na1  cos nθ .
Bθ = − r (30)
2   2−n  2+n r 
   
The radial and azimuthal components of the electromagnetic force density (measured in N·m −3)
acting on the coil are:
µ0 J 02  n −1
 a22 − n − r 2 − n 

 2+n 2+n 

 − 1  r 1−+ na1  cos2 nθ ,
f r = − Bθ J = r (31)
2   2−n  2+n r 
   
µ0 J 02  n −1
 a22 − n − r 2 − n 

 2+n 2+n 

 + 1  r 1−+ na1  sin nθ cos nθ .
fθ = Br J = − r   (32)
2   2−n  2+n r
   
The Cartesian components of the Lorentz force density may be computed using the following
expressions:
f x = f r cos θ − fθ sin θ , (33)
f y = f r sin θ + fθ cos θ . (34)
In the particular case of a dipole, the field inside the coil is given by
µ0 J 0
By = − (a2 − a1 ) . (35)
2
The components of the total force acting on the coil per unit length are given by
µ0 J 02  7 3 1  a2 10  3 1 2

Fx =  a2 +  ln + a1 − a2a1  , (36)
2  54 9  a1 3  2 
µ 0 J 02  2 2  a1 1  3 
 a2 +  ln − a1  .
3
Fy = − (37)
2  27 9  a2 3  
A very simple approximation of the maximum stress at the mid-plane is given by

Fy
σy = . (38)
b−a
4.1.3 Electromagnetic forces on a sector coil

Assuming a uniform current density J = J0 perpendicular to the cross-section plane, inner radius a1,
outer radius a2, a span angle φ such that the first allowed field harmonic is null (i.e. φ = 60º for a
dipole), and no iron is present, the polar components of the magnetic field inside the aperture are
[12, 14] as follows:
2µ J  ∞
r 2n  1 1  
Bri =− 0 0 ( a2 − a1 ) sin φ sin θ + ∑  n −1 − n −1  sin ( 2n + 1) φ sin ( 2n + 1) θ  , (39)
π  n =1 ( 2n + 1)( 2n − 1)  a1 a2  
306
2µ J  ∞
r 2n  1 1  
− 0 0 ( a2 − a1 ) sin φ cos θ + ∑
Bθ i =  n −1 − n −1  sin ( 2n + 1) φ cos ( 2n + 1) θ  . (40)
π  n =1 ( 2n + 1)( 2n − 1)  a1 a2  
The radial and azimuthal components of the field in the coil are given by
2 µ J  ∞ 
a   
2 n +1
r
Br =− 0 0 ( a2 − r ) sin φ sin θ + ∑ 1 −  1   sin ( 2n − 1) φ sin ( 2n − 1) θ  , (41)
π    r   ( 2n + 1)( 2n − 1)
n =1  
2 µ0 J 0  ∞ 
 a1   
2 n +1
r
Bθ =
− ( a2 − r ) sin φ cos θ − ∑ 1 −    sin ( 2n − 1) φ cos ( 2n − 1) θ  . (42)
π    r   ( 2n + 1)( 2n − 1)
n =1 

In the case of a dipole, the polar components of the Lorentz force density are given by
2 µ0 J 02  r 3 − a13 
f r = − Bθ J = + sin φ (a2 − r ) − cos θ , (43)
π  3r 2 
2 µ0 J 02  r 3 − a13 
fθ = Br J = − sin φ (a2 − r ) + sin θ . (44)
π  3r 2 
The Cartesian components of the total force acting on the coil per unit length are given by
2 µ 0 J 02 3  2π − 3 3 3 a 2 3 4π + 3 3 π 
Fx = +  a2 + ln a1 + a1 − a 2 a12  , (45)
π 2  36 12 a1 36 6 
2 µ0 J 02 3  1 3 1 a1 3 1 3 
Fy = − a2 + ln a1 − a1  . (46)
π 2 12 4 a2 12 
4.1.4 Axial electromagnetic forces on the coil ends

The virtual displacement principle establishes that the variation of stored magnetic energy U with the
magnet length equals the axial force Fz pulling from the ends, as long as the rest of the dimensions are
kept constant:
∂U
Fz = . (47)
∂z
That is, the stored magnetic energy per unit length equals the axial force: in the LHC main dipoles, it
is about 125 kN per coil end.
If the coil is approximated as a thin shell, the axial force Fz may be written as follows:
2n
µ0π   a   2 2 By 2
2
Fz = 1 +    J 0 a = πa . (48)
4n   R   µ0
The axial force in a dipole increases with the square of the magnetic field and the aperture. For
the same current density, the end forces on a quadrupole coil are half those measured in a dipole.
Similar expressions for thick shell and sector approximations may be found in Ref. [12].
307
F. T ORAL
4.2 Pre-stress
As pointed out in the previous sections, one of the main concerns of the mechanical designer is to
avoid tensile stresses on the superconducting conductors when they are powered. The classical
solution is to apply a pre-compression. This method was implemented in ancient times, for example in
the Roman arch bridge (Fig. 17(a)). In the case of cos-theta winding configurations, the external
structure usually applies a radial inward compression (Fig. 17(b)), which is transformed into an
azimuthal compression inside the coil which counteracts the formation of tensile stresses that would
otherwise appear under the action of the electromagnetic forces.
(a) (b)
Fig. 17: (a) Roman arch aqueduct in Segovia (Spain) (courtesy of http://commons.wikimedia.org). (b)
Electromagnetic forces on a quadrupole coil, counteracted by a radial inward compression.
The simplest structure that will provide external pre-compression is a cylindrical shell. It is
usually made from aluminium, since its high thermal contraction eases the assembly (less interference
is necessary to provide a given pressure at cold conditions), as will be seen later. The maximum stress
in an aluminium shell at cool-down is about 200–300 MPa. As a first approach, one can assume that
the radial Lorentz force behaves as a uniform pressure (see Fig. 18(a)) or, alternatively, take its
horizontal component:
F [ N ⋅ m] P ⋅ a FLx
σ=θ [ MPa ] = , σ=θ [ MPa ] . (49)
δ δ δ
For n-pole magnets, one can compute the bending moment in a thin cylinder under radial forces
separated by an angle of 2θ and the corresponding hoop stress as [16]:
Fa  cos ( x ) 1 − δ a 
= M  − ,
2  sin (θ ) θ 
(50)
6M
σθ = 2 .
δ
 δ
F
x
p 2θ
F
a

F
Fig. 18: (a) Cylindrical shell under outward radial pressure. (b) Cylindrical shell for an n-pole magnet
308
The maximum compressive stress at the coil must be checked. It usually takes place at the mid-
plane, as shown in previous sections. If it is too high for the insulation, the most common solution is to
reduce the current density accordingly.
In the following sections, some particular aspects of pre-stress application in real magnets will
be reviewed depending on the magnet field value.
4.2.1 Low field magnets

We will refer to magnets as ‘low field’ magnets if the coil peak field is below 4 T. Coils are usually
made with monolithic wires and then fully impregnated. This is, for example, the situation for most of
the LHC corrector magnets. The easiest way to provide the pre-compression is by means of an outer
aluminium shrinking cylinder. It is very convenient to place the iron as close as possible to the coils,
i.e. inside the shell, to enhance the field. However, the iron cannot be constructed as a hollow cylinder
or ring laminations, because the iron contracts less than the aluminium and the coils would become
loose inside the iron yoke. A clever lamination layout, the so-called ‘scissors’ lamination, was
developed at CERN [17]: eccentric paired laminations with different orientations apply the inward
pressure alternatively on neighbouring coils (see Fig. 19). An additional advantage of this system is its
low price for series production, as the laminations can be accurately produced by fine blanking.
The MQTL was the longest corrector magnet produced for LHC using scissor-type laminations.
Some interesting lessons can be drawn from the prototyping phase [18]. First, it is worth noting that a
few holes have been drilled in the iron to maintain a good field quality even with moderate iron
saturation. The first allowed multipole, b6, varies with the current when the iron becomes saturated.
Holes in the iron help to achieve a similar magnetic field map at low and high operating currents, i.e.
the variation of b6 with current is reduced. Figure 20 shows that elliptical holes are a better choice than
circular ones from the point of view of the mechanics, since the concentration of radial stresses on the
inner edge of the hole is lower.
The second interesting feature appeared during the training test of the second prototype (see
Fig. 21). The learning curve was very slow (see curves with the label ‘V1’), starting with a first
quench at very low current, about 220 A, while nominal current was 550 A. Also, there was no
improvement when the magnet was cooled down from 4 to 1.9 K: one can conclude that there is a
mechanical problem which limits the magnet performance. The interference of the shrinking cylinder
was increased (see curves with the label ‘ModA’) to provide a higher pre-stress to the coils, but the
magnet behaviour did not improve significantly. The de-training that happened occasionally suggests
‘slip-stick’ movements between the iron laminations and the coil package due to axial electromagnetic
forces as a likely origin of the poor training.
Position of Shrink ring

maximum
width
Coils Scissors laminations
(stars mark
position of
maximum width)
Scissors
laminations Coils
(iron)
Position of
Shrink ring maximum
width
Fig. 19: Scissor laminations to provide pre-stress on the coils of LHC corrector magnets
309
F. T ORAL
SHELL
G-11
spacer
IRON
G-11 bandage
COIL
(a) (b)
Fig. 20: Radial stress distribution for circular (a) and elliptical (b) iron holes
Re-training / training
Retraining/Training QuenchesMQTL
of MQTL
650
I [Amps]
600
[A]
550
at quench
at Quench
500
450
level
400
level
350
Current
MQTL2ModA @ 1.9K
Current
300 MQTl2ModA @ 4.4K

MQTL2V1 @ 4.4K
250
MQTL2V1 @ 1.9 K
200
0 2 4 6 8 10 12 14 16 18 20 22 24
Quench Number
Fig. 21: Training tests of MQTL second prototype
An alternative method of providing the coils with the necessary pre-stress is based on iron
blocks rather than scissor laminations. In the case of the superconducting combined magnet prototype
developed for TESLA500 project, the iron was split into four sector blocks (see Fig. 22), whose radii
were calculated to fit with the coil package and the shell at cold conditions [19]. The coils are glued
together with glass-fibre spacers and wrapped around with a glass-fibre bandage. The main magnet is
a quadrupole, and two corrector dipoles, horizontal and vertical, are glued around the quadrupole
coils. All are cos-theta type windings.
Fig. 22: Iron yoke split into four sector blocks
310
Figure 23 shows the hoop stress distribution when the coils are free, without any external
support, compared with that when the coils are pre-compressed with an aluminium shell. In the former
case, some tensile (positive) stresses appear in the region of contact of the coil and the central spacer,
which is also the peak field region, i.e. the area more prone to trigger a quench. On the contrary, on
fitting the shrinking cylinder, the full coil is under compression when the magnet is powered.
G-11 SPACER GLASS-FIBRE

BANDAGE
COIL
G-11 SPACER
(a) (b)
Fig. 23: Hoop stress distribution in the coil assembly of the TESLA500 magnet: (a) without pre-compression,
(b) with pre-compression.
A very slow learning curve was recorded during the first training test (see Fig. 24), including a
premature quench at about half the nominal current (100 A). Seventeen quenches were necessary to
power the magnet at nominal current. The outer diameter of the shell was measured to evaluate the
quality of the pre-compression, noting that two of the blocks had lost part of the pre-stress (see
Fig. 25). The outer shells were disassembled and the interference was increased by gluing thin
stainless steel sheets on the outer radius of the blocks with lower compression. It was checked that the
shell outer diameter increased as expected, producing a symmetrical layout. Effectively, an important
enhancement took place during the second training test: the third quench was already above nominal
current. The magnet improved smoothly up to 130 A. In the re-training, the first quench was at a lower
current, but still above nominal current. The most likely factors that still limit the magnet performance
are the following:
− the absence of a structure to support the longitudinal electromagnetic forces, or
− the three layers of glued coils with glass-fibre spacers, which are relatively soft and have
anisotropic properties. These are especially important when a bandage is wrapped around
each layer of the finished coils: because it is applied manually, this could increase the
inhomogeneity.
140
130
120
110
CURRENT (A)
100
90
80
training 4K quad
training 4K combined
70
training 4K quad
training 2K quad
60
retraining 4K quad
50 first campaign 4K
0 5 10 15 20 25 30 35
QUENCH NUMBER
Fig. 24: Training tests of TESLA500 magnet prototype
311
F. T ORAL
1 1
24 2 24 2
23 3 23 3
267.15 267.15
22 4 22 4
267.10 267.10
21 267.05 5 21 267.05 5
267.00 267.00
20 6 20 6
266.95 266.95
19 266.90 7 19 266.90 7
18 8 18 8
17 9
17 9
16 10 h=285 mm
16 10
h=150 mm
h=285 mm 15 11
15 11 h=15 mm
h=150 mm
(b)
14 12
(a)
14 12
13 h=15 mm 13
Fig. 25: Measurements of outer shell diameter at different angular positions before (a) and after (b) increasing
the interference.
In short, coils of low field magnets are usually kept under compression by outer aluminium
shells, fitted with some interference.
4.2.2 High field magnets

In this section, we refer to magnets as ‘high field’ as those with coil fields in the range from 4 to 10 T.
Conductors are usually Nb−Ti cables with polyimide tape insulation, mostly of Rutherford type.
In the case of the Tevatron main dipole, the nominal field in the aperture is 4.4 T. When the
coils are powered, the electromagnetic forces compress the cables azimuthally towards the mid-plane
and radially against the external support structure (see Fig. 26). Assuming an infinitely rigid structure
without pre-stress, the pole turn would move off about 100 µm, with a stress on the mid-plane of
−45 MPa, at nominal current (see Fig. 27) [12].
(a) (b)
(c)
Fig. 26: (a) FEM model of Tevatron main dipole. (b) The coil moves off the pole when powered. (c) Coil cross-
section: two layers of Rutherford cables [12].
(b)
(a)
Fig. 27: Tevatron main dipole coil powered at nominal current: azimuthal displacements (a) and stresses (b) [12]
312
Figure 28 shows the azimuthal stress and displacement of the pole turn (i.e. the one with the
highest field) in different pre-stress conditions at several current levels. The total displacement of the
pole turn is proportional to the pre-stress. A full pre-stress condition (−33 MPa) minimizes the
displacements and, probably, the quench triggering.
(a) (b)
Fig. 28: Azimuthal stress (a) and displacement (b) of the pole turn of the Tevatron main dipole in different pre-
stress conditions at several current levels [12].
The practice of pre-stressing the coil has been applied to all accelerator large dipole magnets:
Tevatron [20], HERA [21], SSC [22, 23], RHIC [24] and LHC [25]. The pre-stress is chosen in such a
way that the coil remains in contact with the pole at the nominal field, sometimes with a ‘mechanical
margin’ of more than 20 MPa (see Fig. 29).
Fig. 29: Azimuthal stress at the pole turn for different coils of the main dipoles of large particle
accelerators [12].
In high field magnets, the pre-stress is usually provided by means of collars. Collars were
implemented for the first time in the Tevatron dipoles. Since then, they have been used in all but one
(RHIC) of the high field cos-theta accelerator magnets and in most of the R&D magnets. They are
composed of stainless steel or aluminium laminations of a few millimetre thickness. The collars take
care of the Lorentz forces and provide a high accuracy for coil positioning. Shape tolerance is about
±20 μm. A good knowledge of the coil properties (initial dimensions and modulus of elasticity) is
mandatory to predict the final coil status: both coils and collars deform under pressure.
313
F. T ORAL
Collars usually consist of two paired pieces with different geometries (see Fig. 30). The
uncompressed coils are oversized with respect to the collar cavity dimension. The collars have holes
or key slots which are aligned when both the collars and the coils are pressed at the nominal value. At
that position, some bolts or keys are pushed through to lock the assembly. Once the collaring press is
released, the collars experience a ‘spring back’ due to the clearance of the locking feature and
deformation. The pre-stress may also change during cool-down due to the different thermal
contraction of the collars and coils.
(a) (b)
Fig. 30: Paired collars (a) and assembly with LHC main dipole coils (b) [26]
For fields above 6 T, it is usually necessary that the rest of the structure contributes to support
the Lorentz forces. For instance, at nominal field, a LHC dipole experiences a horizontal force of
1.7 MN·m−1 and a vertical one of −0.75 MN·m−1 per quadrant. The stainless steel outer shell is split
into two halves which are welded around the yoke at high tension (about 150 MPa) to withstand those
forces. It is worth noting that when the yoke is placed around the collared coil, a gap (vertical or
horizontal) remains between the two halves. This gap is due to the collar deformation induced by coil
pre-stress. If necessary, during the welding process the welding press can impose the desired curvature
on the cold mass. In the LHC dipole, the nominal sag is 9.14 mm.
Fig. 31: (a) Cold mass of LHC main dipole. (b) Vertical press with automatic welding for the assembly of the
outer shell [27].
End plates, which are applied after shell welding, provide axial support to the coil under the
action of the longitudinal electromagnetic forces. A given torque may be applied to the end bolts. In
some cases, the outer shell can also act as a liquid-helium container.
314
Fig. 32: (a) TQ quadrupole: thick rods hold the longitudinal Lorentz forces (courtesy of P. Ferracin). (b) Sketch
of the SSC dipole end plates. (Courtesy of A. Devred.)
4.2.3 Very high field magnets

Let us consider as ‘very high field’ magnets those with coil peak fields beyond 10 T. These are R&D
objects. All the collared magnets presented in the previous section are characterized by significant coil
pre-stress losses (see Fig. 33):
− the coil reaches the maximum compression (about 100 MPa) during the collaring operation;
− after cool-down, the residual pre-stress is about 30–40 MPa.
What would happen if the ‘required’ coil pre-stress after cool-down were greater than 100 MPa?
Following the same approach, the compression on the coil would be too high during the collaring.
Fig. 33: Maximum compressive stress on the coils during the different assembly steps [12]
An alternative solution has been proposed and developed in the framework of the US LHC
Accelerator Program (LARP). It is based on the use of bladders during the magnet assembly [28].
Figure 34(a) shows the TQ quadrupole cross-section. The coils are surrounded by the iron, which is
split into four pads and four yokes, which remain open during all magnet operations. An outer
aluminium shell contains the cold mass. The initial pre-compression is provided by water-pressurized
bladders and locked by keys. During cool-down, the coil pre-stress significantly increases due to the
high thermal contraction of the aluminium shell. Figure 34(b) shows how the maximum compressive
315
F. T ORAL
stress on the coil is similar to that on the collared magnets, but this maximum takes place after cool-
down and is available to counteract the electromagnetic forces. A small spring back occurs when
bladder pressure is reduced, since some clearance is needed for key insertion.
R&D work is ongoing to prove that magnets assembled with this method:
− are able to provide accelerator field quality, and
− may be fabricated with lengths of several metres.
One of the magnets for the ongoing LHC upgrade is being designed following this approach:
the MQXF quadrupole (140 T·m−1 gradient in 150 mm aperture).
(a) (b)
Fig. 34: (a) TQ quadrupole cross-section. (b) Maximum compressive stress on the coils during the assembly
steps [12].
Another novel stress management system developed at Texas A&M University is based on
intermediate coil supports [29]. Each coil block is isolated in its own compartment and supported
separately (see Fig. 35). Lorentz forces exerted on multiple coil blocks do not accumulate, but rather
are transmitted to the magnet frame by the Inconel ribs and plates. A laminar spring is used to pre-load
each block.
Fig. 35: (a) TAMU dipole coil blocks. (b) Detailed view of a coil block [30]
4.2.4 Pre-stress: controversy

As we have seen, the pre-stress aims to avoid the appearance of tensile stresses in the coils and limit
the movement of the conductors. This raises the question: what is the correct value of the pre-stress?
In Tevatron dipoles, it was found that there was not a good correlation between small coil
movements (<0.1 mm) and the magnet learning curve. In the LHC short dipole program, the coils
316
were unloaded at 75% of the nominal current, without degradation in the performance. In LARP TQ
quadrupoles, two different behaviours were detected (see Fig. 36):
− with low pre-stress, the coils were unloaded but kept a good quench performance;
− with high pre-stress, the learning curve was a stable plateau, but with a small degradation.
TQS03a low prestress

ess
str
pre
low
S 03a TQS03b high prestress
TQ
ess
restr
ig hp
h
3b
S0
TQ
Fig. 36: (a) TQ coil stress vs. current. (b) TQ training tests with low and high pre-stress [12]
In LHC corrector sextupoles (MCS), a specific test program was run to find the optimum value
of pre-stress [31]. Coils were individually powered under different pre-compressions immersed in the
same field map as the magnet by means of a custom set-up (see Fig. 37). The conclusions were the
following:
− The learning curve was poor in free conditions.
− Training was optimum with low pre-stress and around 30 MPa. However, degradation was
observed for high pre-stress (above 40 MPa).
− The nominal pre-stress for series production was 30 MPa.
In conclusion, there is not an exact value for the correct value of pre-stress, but experience
shows that too high a pre-stress can degrade the magnet performance.
1200
60MPa AZIMUTHAL
1000 PRE-STRESS
50MPa AZIMUTHAL
CURRENT(A)
800 PRE-STRESS
600 10MPa AZIMUTHAL

PRE-STRESS
400 3MPa AZIMUTHAL
PRE-STRESS
200 0MPa AZIMUTHAL
PRE-STRESS
0
1 3 5 7 9 11 13 15
Nr.OF QUENCHES
(a) (b)
Fig. 37: (a) Custom set-up to test individual MCS coils at different pre-stresses, provided by auxiliary
superconducting coils. (b) Training tests with different pre-stresses [31].
317
F. T ORAL
5 Superferric accelerator magnets

The stress distribution in the coils of a superferric magnet is different from that in the cos-theta
magnets: when powered, the coil experiences in-plane expansion forces, and it is usually attracted by
the iron (see Fig. 38(a)). The force density is not as high as in cos-theta magnets, because fields are
moderate.
In small magnets, simple support structures (such as wedges, see Fig. 38(b)) are sufficient to
hold the coils and prevent wire movement, since coils are usually fully impregnated.
wedge
coil iron
pole
(a) (b)
Fig. 38: (a) Lorentz forces on superferric octupole coil blocks. (b) Support wedges in between two coils [32]
Large superferric magnets are very common in fragment separators (NSCL-MSU, RIKEN,
FAIR) and particle detectors (SAMURAI, CBM). Usually, the iron is warm. Then, the Lorentz forces
on the coil are counteracted by a stainless steel casing, which is also the helium vessel. In some cases,
parts of these forces may be transferred to the external structure by means of low-heat-loss supports
(see Fig. 39).
GFRP
SUS304
Cryostat cylinder Support
cylinder
(SUS304) yoke link
80K
shield
(Al)
20K coil
shield 160x180
(Al)
pole
GFRP Heater (Cu wire)

(a) Coil vessel
Cross-sectional view
(SUS316L)
(b)
Fig. 39: (a) Lorentz forces on Samurai magnet coils. (b) Cross-section of the cryostat [33]
6 Toroids
In toroids, as the magnetic pressure varies along the coil it is subjected to strong bending forces. If one
wants to simplify the support structure, the following strategy must be followed [10]:
318
− Each coil experiences a net force towards the centre because the field is strongest there: it is
wise to flatten the inner edge of the coil so that it leans on the support structure.
− The rest of the circumference will distort to a shape working under pure tension, where no
bending forces are present. This tension must be constant around the coil. Assuming R to be
the distance to the centre and ρ to be the local radius of the curvature of the coil, the
condition for local equilibrium is given by
= ( R ) I ρ constant,
T B=
R
B ( R ) = B0 , (51)
R0
{ }
32
1 + ( dR dz )
2
TR
=ρ = = KR.
2
d R dz 2
B0 R0 I
There is no analytical solution. Figure 40 shows a family of solutions. Toroids are used in large
Tokamak fusion reactors, whose coil shapes resemble those depicted in Fig. 40. Nominal currents are
usually very large. Indeed, the most commonly used cable is the so-called cable-in-conduit (CICC).
The superconducting strands are free, enclosed within a stainless steel pipe, with a double objective: to
host the coolant flow through the voids in between the strands, and to support the electromagnetic
forces on the conductors. The use of this type of cable leads to some peculiarities regarding the
mechanical calculations. We will review some of these aspects using the EDIPO magnet [34] as an
example. It is a superferric dipole designed and fabricated to characterize cables for ITER coils. The
nominal bore field is 12.5 T. The overall magnet length is 2.3 m. The Lorentz forces are huge: 1000,
500 and 400 tons in the horizontal, vertical, and longitudinal directions, respectively. The magnet is
not collared. These forces are contained both by the low carbon iron laminations and by the outer
stainless steel shell (see Fig. 41).
Fig. 40: Numerical solutions for toroid coil profile with constant tension and zero bending moment [10]
319
F. T ORAL
(a) (b)
Fig. 41: (a) Tresca stress in EDIPO magnet. (b) Friction cone and horizontal displacements map. (Courtesy J.
Lucas.)
It is worth pointing out that two different finite element models were used for the mechanical
analysis:
− a general model, with few details, used to study the support structure deformation due to
cool-down and Lorentz forces;
− a sub-model of the coil used to analyze the local stresses on the conductors, mainly in the
insulation, which is wrapped around each CICC pipe. The Lorentz forces are transferred as
internal pressures from the global model, and the contact with the support structure is
modeled as a boundary condition (see Fig. 42).
This approach is very efficient, since there is no need to go into the details of the complete
magnet model.
(a) (b)
Fig. 42: (a) Sub-model used to analyze the stresses on the CICC insulation. (b) Tresca stress map. (Courtesy of
J. Lucas.)
7 Case study: Advanced Molecular Imaging Techniques (AMIT) cyclotron

Let us finish our review on the mechanical design of superconducting magnets with a comprehensive
example, the superconducting compact cyclotron of the AMIT project [35]. It includes a 4 T
superconducting magnet with warm iron. The iron pole radius is 175 mm. The main components are
depicted in Fig. 43. We will pay special attention to the mechanical design strategy and the flow of
decisions.
320
COILS
SUPPORT
IRON
HELIUM VESSEL
CRYOSTAT
(b)
(a)
Fig. 43: (a) Cyclotron cross-section (courtesy of L. Garcia-Tabares). (b) 3D open model to show the coils and
the iron poles. (Courtesy of D. Obradors.)
During the electromagnetic design, it was observed that the axial Lorentz forces between the
coils could be attractive or repulsive, depending on their relative position and the operating current. It
is easier to manage attractive forces, since the support structure would work under compression in that
case, but it was not compatible with the dimensional constraints (compactness). Within the available
space, the optimal design was that with the smallest repulsive forces, about 100 kN per coil.
In the conceptual design, the designer must analyze the requirements on the support structure
(see Fig. 44).
− Radial Lorentz forces Fr will induce a pressure on the winding. Since the coil is relatively
thick, it is very likely that tensile (positive) radial stresses will appear in the inner layers of
the coil. In any case, they will induce high hoop stresses in the superconductor. These forces
will be held by an outer aluminium shell, fitted with a given interference. Due to the high
thermal contraction of aluminium, the interference will be small and the assembly will be
easy.
− Axial Lorentz forces Fz will pull both coils towards the iron, thus inducing positive axial
stresses in the windings. These forces will be held by a casing: when the coil is powered, it
will press on the casing. Therefore, it is very important to guarantee the flatness of both
contact surfaces, to avoid wire movements and, indeed, the quench triggering. These forces
will induce bending moments at the corners of the support structure (holes are necessary to
introduce the cyclotron vacuum chamber). A numerical model is needed to analyze the
minimum fillet radius necessary to limit the stress concentration on those corners.
− The support structure will also be the helium vessel. The maximum pressure will take place
in case of quench, when the helium suddenly boils off. A thermo-hydraulic model will be
used to determine the necessary cross-section of the helium flow to limit the pressure and,
indeed, the stresses induced on the vessel.
− Finally, one should choose the material. The structure will be made using non-magnetic
stainless steel, which fulfils all the requirements: high magnetic field, low temperature
operation, high stresses, and liquid-helium tightness. The best steel grades are 1.4429 and
1.4435, the second one being easier to be procured in small quantities on the market.
321
F. T ORAL
Fz
Fr
Fig. 44: Support structure for AMIT cyclotron: coils are inside the red casings, which are welded to the central
part. (Courtesy of J. Munilla.)
Stresses in the coil and aluminium shell have been calculated in warm, cold, and energized
conditions. Figure 45 shows the distribution of radial and hoop stresses. Radial stresses in the coil are
always negative (compressive), whereas hoop stresses are positive, but are limited to 50 MPa in the
coil and 150 MPa in the shell.
Fig. 45: Stresses on the coils at different assembly steps as a function of the distance to the coil inner radius.
(Courtesy of J. Munilla.)
The stress distribution in the stainless steel support structure has been calculated using FEM.
The maximum values are located in the corners (see Fig. 46) due to the bending moments induced by
the axial electromagnetic forces. Their value is of the order of 80 MPa, which is an acceptablevalue
for the steel.
Max 80 MPa
(a) (b)
Fig. 46: (a) Lorentz forces on the coil. (b) Von Mises stresses in the coil casing (Courtesy of J. Munilla)
322
When the coils are not centred with the iron poles, some forces will arise. These forces have
been calculated in three directions (see Fig. 47). The x and y horizontal axes are different due to the
presence of the vacuum chamber hole through the iron yoke. The forces are in the direction of the
misalignment with a positive slope, i.e. trying to increase the off-axis error.
Fig. 47: (a) Magnetic forces vs. misalignments of coils with respect to the iron yoke (Courtesy of J. Munilla)
Stresses in the casing supports have been calculated in both centred and off-axis conditions (see
Fig. 48). Upper rods will develop larger stresses because the coils hang from them.
In summary, the mechanical design interacts with the electromagnetic and the cooling
calculations. The design decisions must take into account all these aspects to find the best trade-off.
The first calculations are more general, even using analytical expressions at the very beginning. Once
the concept is fixed, we come into the details, mainly with the use of numerical calculations and
models. In the same way, a 2D approach is taken first, then a 3D design is realized, which is more
time consuming.
(a) (b)
Fig. 48: (a) Von Mises stresses in the G10 rods at nominal current. (b) Von Mises stresses in the G10 rods at
nominal current and a 0.5 mm misalignment in the y-direction. (Courtesy of J. Munilla.)
8 Measurement techniques
8.1 Stress
The most widespread technique for stress measurement is the capacitive gauge. The basic principle is
to measure the variation of capacity induced by a pressure exerted on a capacitor. Figure 49(a) shows
a typical layout of a capacitive gauge.
Let S denote the area of the two parallel electrodes, with δ the thickness of the dielectric and ε
the electric permittivity; then the capacity C is given by the following well-known expression:
C = εS / δ . (52)
323
F. T ORAL
When a pressure σ is applied, the capacity will change as follows, due to the deformation of the
dielectric:
εS
C= . (53)
δ (1 − σ E ) 
The gauges must be calibrated to achieve a good accuracy. The capacity can be measured as a
function of pressure and temperature, as shown in Fig. 49(b) [36].
(a)
(b)
Fig. 49: (a) Typical layout of a capacitive gauge. (b) Capacitance measurement as a function of pressure at
different temperatures [36].
8.2 Strain
The basic principle is to measure the variation of the resistance induced by a strain in a resistor [37].
The gauge consists of a wire arranged in a grid pattern bonded on the surface of the specimen (see
Fig. 50). The strain experienced by the test specimen is transferred directly to the strain gauge. Once
the gauges are glued, several thermal cycles may help to get rid of noisy or unstable measurements
and to guarantee a good stress transfer. The gauge responds with a linear change in electrical
resistance. The gauge sensitivity to strain is expressed by the gauge factor, which is the ratio of the
resistance variation to the elongation:
∆R / R
GF = (54)
∆l / l .
The gauge factor is usually about 2. Gauges are calibrated by applying a known pressure to a
stack of conductors or a beam. The temperature and magnetic field effects can be compensated for by
measuring a nearby gauge which is not under stress.
(a) (b)
Fig. 50: (a) Typical layout of a strain transducer [37]. (b) Strain gauges glued onto an aluminium shell [12]
324
8.3 Coil physical properties

The elastic modulus E is given by
dσ dσ
E
= = l0 , (55)
dε dl
where σ is the applied stress, ε is the specimen strain, dl is the displacement, and l0 is the initial length.
The elastic modulus E is measured by compressing a stack of conductors, usually called a ten-
stack, and measuring the induced deformation. The stress–displacement curve is not linear and
presents a significant difference between the loading and unloading phases (see Fig. 51). The elastic
modulus depends on the pressure applied and on the ‘history’ of the loading. It is also dependent on
the temperature.
The thermal contraction is given by
lw0 − lc0
α= , (55)
lw0
where lw0 and lc0 are the unloaded height of the specimen at room and cold temperature, respectively.
Figure 51(b) shows a set-up to measure the thermal contraction by comparison with a well-known
aluminium reference. It can be also evaluated using the stress loss in a fixed cavity.
Fig. 51: (a) Typical stress–displacement curve of a cable stack [12]. (b) Set-up for thermal contraction
measurement [38].
Acknowledgement
The author warmly thanks Paolo Ferracin (CERN) for his support and helpful comments during the
preparation of this lecture. His USPAS lectures [12] can be used by the interested reader to continue
learning about the mechanical design of superconducting accelerator magnets.
References
[1] S. Timoshenko, Strength of Materials (D. Van Nostrand Company, New York, 1930).
[2] http://cryogenics.nist.gov, 2013.
[3] http://www.cryodata.com, 2013.
[4] P. Duthil, Material properties at low temperature, these proceedings.
[5] Y. Iwasa, Case Studies in Superconducting Magnets (Plenum Press, New York, 1994).
[6] M. Wu and M. Tigner, Eds., Handbook of Accelerator Physics and Engineering (World
Scientific, Singapore, 1999).
325
F. T ORAL
[7] T. M. Flynn, Cryogenic Engineering (Marcel Dekker, New York, 1997).

[8] Formvar®
[9] INTA, Introduction to structural composites, Lectures held at Instituto de Tecnicas
Aeroespaciales (INTA), Madrid (2009).
[10] M. Wilson, Superconducting Magnets (Clarendon Press, Oxford, 1983).
[11] S. Sanz, Development of a superconducting solenoid to test AMS power supply, CIEMAT
Internal Report (2006).
[12] H. Felice, P. Ferracin, S. Prestamon and E. Todesco, Superconducting magnets for particle
accelerators, USPAS lectures, Units 10, 14, and 19, 2011. Available at
http://etodesco.home.cern.ch/etodesco/, 2013.
[13] F. Toral et al., Further developments on Beth’s current-sheet theorem: computation of magnetic
field, Energy and mechanical stresses in the cross-section of particle accelerator magnets, Proc.
18th International Conference on Magnet Technology, Morioka City, 2004 [IEEE Trans. Appl.
Supercond., 14(2) (2004) 1886].
[14] R. Meuser, Magnetic field and stresses for thin and thick cos-nθ windings, Engineering Notes
M5251, M5254 and M5266, Lawrence Berkeley Laboratory
[15] S. Russenschuck, ROXIE – a computer code for the integrated design of accelerator magnets,
Sixth European Particle Accelerator Conf., Stockholm, Sweden (June 1998).
[16] W. Young, R. Budynas, Roark’s Formulas for Stress and Strain (McGraw-Hill, New York,
2001).
[17] A. Ijspeert and J. Salminen, Superconducting coil compression by scissor laminations, LHC
Report 47 (1996).
[18] F. Toral, PhD Thesis, Universidad Pontificia Comillas, Madrid, 2001.
[19] F. Toral et al., Fabrication and testing of a combined superconducting magnet for the TESLA
test facility, Proc. 19th International Conference on Magnet Technology, Genova, 2005 [IEEE
Trans. Appl. Supercond., 16(2) (2006) 231].
[20] K. Koepke et al., IEEE Trans. Magnet. 15(l) (1979) 658–661.
[21] S. Wolff, AIP Conf. Proc. 249 (1992) 1160−1197.
[22] A. Devred, AIP Conf. Proc. 249 (1992) 1309–1372.
[23] T. Ogitsu et al., IEEE Trans. Appl. Supercond. 3(l) (1993) 686–691.
[24] J. Muratore, BNL, private communications, 2011.
[25] P. Lebrun et al. (Eds.), LHC Design Report, CERN–2004–003, volume I (2004).
[26] L. Rossi, Superconducting magnets, CERN Academic Training (15–18 May, 2000).
[27] L. Rossi, The LHC from construction to commissioning, FNAL seminar (19 April, 2007).
[28] S. Caspi et al., IEEE Trans. Appl. Supercond. 11(1) (2001) 2272–2275.
[29] A. Mc Inturff et al., Fabrication of TAMU3, a Wind&React stress-managed 14T Nb3Sn block
coil design, EUCAS 2007 [J. Phys. Conf. Series 97 (2008) 012128].
[30] C.L. Goodzeit, Superconducting accelerator magnets, USPAS (January 2001).
[31] M. Bajko et al., Training study on superconducting coils of the LHC sextupole corrector
magnet, LHC Report 236 (1998).
[32] P. Abramian et al., IEEE Trans. Appl. Supercond. 23(3) (2013), article 4101204.
[33] H. Sato et al., IEEE Trans. Appl. Supercond. 23(3) (2013), article 4500308.
[34] A. Portone et al., IEEE Trans. Appl. Supercond. 21(3) (2011) 1953−1959.
[35] C. Oliver et al., Optimizing the radioisotope production with a weak focusing compact
cyclotron, Proc. Cyclotron Conf. (2013).
326
[36] N. Siegel et al., Design and use of capacitive force transducers for superconducting magnet
models for the LHC, LHC Project Report 173, 1998.
[37] C. L. Goodzeit et al., IEEE Trans. Magnet. 25(2) (1989) 1463–1468.
[38] K.P. Chow and G.A. Millos, IEEE Trans. Appl. Supercond. 9(2) (1999) 213–215.
Bibliography
H. Felice, P. Ferracin, S. Prestamon and E. Todesco, Superconducting magnets for particle
accelerators, http://etodesco.home.cern.ch/etodesco/, USPAS (2011).
Proceedings of the CAS−CERN Accelerator School: Magnets, Bruges, Belgium, 16−25 June 2010,
edited by Daniel Brandt, CERN-2010-004 (2010).
Proceedings of the CAS−CERN Accelerator School: Superconductivity and cryogenics for
accelerators and detectors, Erice, Italy, 8−17 May 2002, edited by Stephan Russenschuck and
Giovanna Vandoni, CERN-2004-008 (2004).
Proceedings of the CAS−CERN Accelerator School: Superconductivity in particle accelerators,
CERN−96−03 (1996) and CERN−89−04 (1989).
M. Wilson, Superconducting Magnets (Clarendon Press, Oxford, 1983).
K. H. Mess, S. Wolff, and P. Schmüser, Superconducting Accelerator Magnets (World Scientific,
Singapore, 1996).
327
Heat Transfer and Cooling Techniques at Low Temperature
B. Baudouy 1
CEA Saclay, France
Abstract
The first part of this chapter gives an introduction to heat transfer and
cooling techniques at low temperature. We review the fundamental laws of
heat transfer (conduction, convection and radiation) and give useful data
specific to cryogenic conditions (thermal contact resistance, total emissivity
of materials and heat transfer correlation in forced or boiling flow for
example) used in the design of cooling systems. In the second part, we
review the main cooling techniques at low temperature, with or without
cryogen, from the simplest ones (bath cooling) to the ones involving the use
of cryocoolers without forgetting the cooling flow techniques.
Keywords: heat-transfer, cooling techniques, low-temperature, cryogen.
1 Introduction
Maintaining a system at a temperature much lower than room temperature implies that the design of
your cooling system must ensure its thermal stability in the steady-state regime; that is, the
temperature of your system must remain constant in the nominal working condition. It also requires a
certain level of thermal protection against transient events; that is, your temperature system must stay
below a certain value to avoid problematic situations such as extra mechanical constraints due to the
thermal expansion of materials with temperature. The ultimate goal is to minimize the heat load from
the surroundings and to maximize the heat transfer with a cooling device.
To be able to achieve this objective, identification of the different heat loads on the system is
required, as well as knowledge of the fundamental laws of heat transfer, the thermo-physical properties
of materials and fluids such as the density (kg·m–3) and heat capacity (J·kg–1·K –1), and thermal
conductivity (W·m–1·K–1) and the cooling techniques such as conduction, radiation or forced flow
based techniques.
The objective of the present chapter is to give an introduction to heat transfer and cooling
techniques at low temperature, exemplified with practical cases and data.
2 Heat transfer at low temperature
2.1 Thermal conduction
2.1.1 Conduction in solids

Thermal conduction is a heat transfer without mass transfer in solids. The relationship between the
heat flux density and the temperature gradient, as presented in Fig. 1, is given by the Fourier law,
 
q= − k (T )∇T , (1)
1
bertrand.baudouy@cea.fr
B. BAUDOUY
where q is the heat flux density (W·m–2), k is the thermal conductivity (W·m–1·K–1), and T is the
temperature. In one dimension, Eq. (1) can be written as
L Thot
dT dx
q = -k (T ) ⇒Q ∫ = ∫
k (T )dT , (2)
dx 0
A T cold
where A is the cross-section of the domain and Q is the power (W). In the case where A is constant,
then Eq. (2) is simplified to
Thot
dT Q 1
q = -k (T ) ⇒ = ∫ k (T )dT . (3)
dx A LT
cold
Fig. 1: The definition of heat flux in a one-dimensional domain
Thot
∫Tcold
k (T )d T is the thermal conductivity integral and knowledge of this property is of great importance
in the thermal design of low-temperature devices, because the thermal conductivity of most materials
varies strongly with temperature. Figure 2 depicts the evolution of the thermal conductivity integral
with temperature for the common materials used at low temperature [1]. As expected, the thermal
conductors possess a larger thermal conductivity integral.
105
104
Conductivity Intergal (W/m)
103
102
101
Cu (RRR=200)
Cu (RRR=20)
Acier 304L
100 Al 6061
NbTi
G10
Epoxy
10-1 Polyamide (Nylon)
Pyrex
10 100
Temperature (K)
Fig. 2: The thermal conductivity integral of commonly used materials at low temperature [1]
330
H EAT T RANSFER AND C OOLING T ECHNIQUES AT L OW T EMPERATURE
One of the principal uses for the thermal conductivity integral is in the determination of heat
losses and heat interception between room temperature and the low temperature of the system under
consideration. A typical example is the computation of the heat input into a liquid helium bath cooled
system, which is suspended by three 304 stainless steel rods from the 300 K top flange (cf. Fig. 3(a)).
If the rods are 1 m long and 10 mm in diameter, then according to [1], the heat leak is
A 300 2.36 × 10−4
=Q4K k
=
∫ SS (T ) d T 3.07 × 103 0.7 W .
= (4)
L 4.2 1
This corresponds to a consumption of 1 l·h–1 of liquid helium. Note that if the rods are made of copper
(Residual Resistivity Ratio, RRR = 20) with a conductivity integral of 1.26 105 W·m–1, then Q4K ≈ 20
W; or alternatively, in G10 (epoxy fiberglass tape) with an integral of 167 W·m–1, then Q4K ≈ 26 mW.
One has to keep in mind that the choice of the material also depends on other criteria, such as the
mechanical, magnetic, and environmental conditions, and so on.
To reduce the heat load on the helium bath, heat interception by another cold source at an
intermediate constant temperature (thermalization) is usually used. Historically, the most commonly
used method is a heat sink cooled by boiling nitrogen; possibly replaced in recent years by a heat sink
temperature-regulated by a cold stage of a cryocooler. In the case of boiling nitrogen (at 77 K) with an
interception located at one third of the length from the top of the cryostat (see Fig. 3(b)), the heat leak
to the liquid helium is reduced to
77
A 2.36 × 10−4
=Q4K =
L 4.2∫k (T ) d T =
0.75
325 0.1 W , (5)
which corresponds to a seven-fold reduction of liquid helium consumption. The heat arriving at the
nitrogen reservoir is computed as follows:
300
A 2.36 × 10−4
Q77K
=
L ∫
77
k=
(T ) d T
0.25
2.75
= × 103 2.6 W , (6)
which is equivalent to a liquid nitrogen consumption of 0.06 l·h–1. This example is simplified, but it
shows the principle of heat interception. The optimization of the heat interception depends on many
parameters, such as the thermalization temperature, the material properties, the geometry of the
system, and so on.
Fig. 3: The heat interception concept at 77 K
331
B. BAUDOUY
The thermal design at low temperature requires evaluation of the thermal losses of the system.
To do so, one must calculate the thermal resistance of the structural material. In the steady-state
regime and without internal dissipation, a thermal resistance Rth can be defined from Eq. (3) as
Z2 T Z2
dz
2
T −T 1 dz  K 
Q ∫
Z
A( z )
= - ∫ k (T )dT = k (T1 − T2 ) ⇒ Rth = 1 2 =
T
Q k ∫ A( z) W  ,
Z
(7)
1 1 1
where k is the average conductivity over the temperature range considered and A(z) is the cross-
section as shown in Fig. 4. For simple configurations, the resistance can be defined easily, as for a slab
with a constant section (A = A1 = A2), Rth = L/ k A (for a cylindrical wall between two radii, R1 and R2,
with a length L, Rth = ln(R2/R1)/ k 2πL). In the same manner, a convective boundary thermal resistance
can also be simulated by considering a heat transfer coefficient h and a heat transfer area A as
Rth = 1/hA.
Fig. 4: A heat flux tube based on the surfaces A1 and A2
When several materials enter into the design of the structural component, it is necessary to
account for all of them. When the components are thermally in series, then different values of Rth are
in series, so Rtotal is simply ∑Ri. For the case of a composite wall with different heat transfer
coefficients at the boundaries, as described in Fig. 5(a), the heat flux is computed as follows:
Th − Tc Th − Tc
=Q = A . (8)
( ) ( ) (
∑ Ri (1 hh ) + L1 k1 + L2 k2 + L3 k3 + (1 hc )
i
)
In the case of parallel components, the Rth are in parallel, so 1/Rtotal = ∑1/Ri. Then, the heat flux
is as follows:
Th − Tc
Q=A . (9)
(1 hh ) + ( L1 ) ( ) ( )
k1 + 2 L2 (k2 + k3 ) + L4 k4 + (1 hc )
Here, L2 = L3 and A2 = A3 = A/2.
Fig. 5: ‘Parallel’ (a) and ‘series/parallel’ (b) composite walls with heat transfer coefficients at the boundaries
332
In the treatment of the thermal resistance, we assumed a perfect thermal contact between the
different components. But this is rarely the case, since the two surfaces of the materials are not in a
perfect and full contact. The rugosity creates an imperfect contact, generating a temperature drop
(cf. Fig. 6) due to the local contact creating constriction of the flux lines. Additionally, the phonon
scattering at the solid–solid contact (Kapitza resistance) and the heat transfer via eventual interstitial
elements increase the contact thermal resistance significantly. As before, this resistance is defined as
Rc = (T2 – T1)/Q. Resistance Rc depends on the surface condition, the nature of the materials, the
temperature, the interstitial materials, the compression force, and so on. It is proportional to the force
applied on the contact, but not to the pressure (the number of contact points increases with the force).
It reduces with increasing force but increases with decreasing temperature by several orders of
magnitude from 200 to 20 K. At low temperature, Rc can be the largest thermal resistance source, but it
can be reduced by firm tightening of the two pieces or the insertion of conductive and malleable fillers
(charged grease, indium, or surface coatings). Finally, it is worth saying that thermal contact resistance
modelling is very difficult; therefore the use of experimental data is recommended, as in the example
presented in Fig. 7 [2], where the thermal conductance values (i.e. the inverse of Rc) are presented.
Fig. 6: The concept of contact thermal resistance
Fig. 7: The thermal conductance of different solid–solid joints as a function of temperature. Reproduced from [2]
with permission of Oxford University Press.
333
B. BAUDOUY
It is often important to include the transient conduction process in a cryostat design at low
temperature. In thermal conduction, transient processes are approached using an energy conservation
equation that leads to a diffusion equation of the following type:
∂T 
ρC
∂t
( )
=∇. − k (T )∇T + Q  W ⋅ m −3  , (10)
where ρ is the density and C is the specific heat of the solid. On the left-hand side of the equation, the
term accounts for the thermal inertia. The first term on the right-hand side accounts for the conduction,
and the second one, Q, is a heat source. In 1D with constant thermal properties, Eq. (10) is simplified
to
∂T k ∂ 2T k
= +Q* ⇒ D
=  m 2 ⋅ s −1  , (11)
∂t ρ C ∂x 2
ρC  
in which the thermal diffusivity D = k/ρC is defined. D allows the evaluation of the characteristic
diffusion time τ due to a thermal perturbation in the solid medium. This thermal time constant is
obtained by solving Eq. (11) and is defined as
4 L2
τ≈ , (12)
π2 D
where L is the characteristic thermal length of the solid. Equation (12) is the time constant when the
temperature has reached two thirds of the final temperature. To reach 95% of the final temperature, the
thermal time constant is 3τ. As can be seen from the equation, τ is a function of the length and of the
thermal diffusivity and it differs from one material to another (Table 1). One can note the large
increase of the thermal diffusivity with temperature.
Table 1: The thermal diffusivity of different materials at different temperatures in cm2·s–1 [1]
300 K 77 K 4K
Cu OFHC
1.2 3.2 11 700
(RRR = 150)
Pure Al (RRR = 800) 1 4.7 42 000
Commercial Al (6061) 0.7 1.3 1 200
SS 304 L 0.04 0.05 0.15
Nb−Ti 0.03 0.02 0.51
2.1.2 Conduction in liquids

Liquids are bad thermal conductors, at room temperature and at low temperature alike. Furthermore,
their thermal conductivity usually decreases with temperature. Table 2 presents the thermal
conductivity of some common cryogens at atmospheric pressure [3, 4]. In most heat transfer situations
at low temperature, the thermal conduction in a liquid is considered negligible compared to convection
or phase change phenomena. There is one exception to this rule; namely, the superfluid helium. This
topic will be treated in another section.
Table 2: The thermal conductivity of some cryogens at atmospheric pressure (W·m–1·K–1)
O2 (T = 90 K) N2 (T = 77 K) H2 (T = 20 K) He (T = 4.2 K)
0.152 0.14 0.072 0.019
334
2.1.3 Conduction in gas

The heat transfer between two surfaces in a gas is of interest to evaluate the heat leak or to characterize
thermal switches, devices that exchange heat with gas in certain conditions. There are two different
heat transfer regimes depending on the ratio between the mean free path of the gas molecules, λ, and
the distance L between the two surfaces involved in the heat transfer:
 R = universal gas constant = 8.31 J ⋅ mol−1 ⋅ K −1 ,
RT  23 −1
λ= 2  N A = Avogadro's number = (6.02 × 10 ) ⋅ mol , (13)
2π d N A p 
d = molecule diameter.
When λ ≫ L, it corresponds to the free molecular regime, whereas when λ ≪ L, it corresponds
to the hydrodynamic regime. The free molecular regime is obtained at low residual pressure, and the
heat transfer depends on the residual gas pressure and is independent of L. The heat leak, Q, in the free
molecular regime is described by Kennard’s law:
 γ +1 R ∆T Cp
= α
Q A=  p with γ , (14)
 γ − 1  8π M T Cv
where M is the molar mass of the gas, A is the surface area receiving the heat flux, and α is an
accommodation coefficient [2]. Coefficient α relates the degree of thermal equilibrium between the gas
and the wall. Its value ranges from α ≤ 0.5 for helium, to α ~ 0.78 for argon and α ~ 0.78 for nitrogen.
The hydrodynamic regime is obtained at high residual gas pressure and the heat transfer is independent
of pressure and described by a Fourier law. The thermal conductivity is thus derived from the kinetic
theory. Some values at atmospheric pressure are presented in Table 3, and Fig. 8 presents a very useful
example of heat transfer in helium gas between two copper plates separated by 1 cm [2].
Table 3: The thermal conductivity, k [mW·m–1·K–1] at 1 atm
4
T (K) He H2 N2
300 150.7 176.9 25.8
75* 62.4 51.6 7.23
20 25.9 15.7
5 9.7
Fig. 8: The heat transfer in helium gas between two copper plates separated by 1 cm. Reproduced from [2] with
permission of Oxford University Press.
335
B. BAUDOUY
2.2 Thermal radiation
2.2.1 General laws

Any surface at finite temperature absorbs, reflects, and emits electromagnetic radiation. The
wavelengths associated with thermal radiation span from 0.1 to 100 μm, as shown in Fig. 9.
Fig. 9: The electromagnetic radiation spectrum, showing the thermal radiation zone in yellow
The emitted radiation, Φ, consists of a continuous non-uniform distribution of monochromatic

components. This distribution and the magnitude of the emitted radiation depend on the nature and the
temperature of the emitting surface. Φ also has a directional distribution (Fig. 10). Therefore to
characterize the radiation heat transfer (as a function of temperature and surface), both the spectral and
the directional dependence must be known. On a real body, incident radiation is absorbed, transmitted
through the body or reflected; therefore the law of energy conservation requires 1 = α + τ + ρ, where α
represents the spectral absorption component, τ the spectral transmission component, and ρ the spectral
reflection component.
Fig. 10: Spectral and directional distribution
When describing the radiation characteristics of real surfaces, it is useful to start with the
concept of an ideal surface: the blackbody. The blackbody is a perfect emitter and absorber. It absorbs
all incident radiation regardless of the wavelength and direction. At a given temperature and
wavelength, the emission is maximum and the blackbody is a diffuse emitter (no directional
dependence). The emissive power (W·m–3) of the blackbody per unit wavelength and per surface
(hemispherical) as a function of wavelength is described by Planck’s law:
0 C1  2π hC02 =
C1 = 3.742 ×10−16 Wm 4 ⋅ m −2 ,
Eλ = with  (15)
(
λ 5 eC2 λT
)
−1 C2 =
 1.4388 ×10−16 Wm 4 ⋅ m −2 .
hC0 / k =
336
The emissive power is presented in Fig. 11 as a function of wavelength and temperature. It

exhibits a maximum at a certain temperature, given by Wien’s law:
λmaxT 2898 μm ⋅ K −1.
= (16)
Fig. 11: The emissive power of a spectral blackbody
Integrating Planck’s law with respect to wavelength, the so-called Stefan–Boltzmann law is
obtained, defining the blackbody total hemispherical emissive power at temperature T:
∞
C
∫0 λ 5 ( eC 1λT − 1) d λ = σ T
0 4
E = with σ = 5.67 ×10−8 W ⋅ m −2 ⋅ K −4 . (17)
2
For a real surface material, the emissive power is only a fraction of the blackbody. The ratio of
the real surface to the blackbody emissive power is called the emissivity, ε. One can define the
spectral, monochromatic directional emissivity as ε(λ, θ, ϕ, T) and the spectral emissivity as ε(λ, T).
But for simplicity, the total emissivity, ε(T), is most commonly used. Then, the emissive power for a
real surface material is defined as
E 0 = εσ T 4 . (18)
In general, emissivity decreases with temperature but increases with oxidation, impurities, dirt,
and so on. To achieve the lowest emissivity value, it is recommended that highly polished and highly
conductive clean surfaces (e.g. gold, silver, copper, or aluminium) should be used. Table 4 presents the
total emissivity of various materials used at cryogenic temperatures compared to 3M black paint at
different temperatures [5]. Figure 12 also presents useful emissivity data for various materials between
two surfaces at different temperatures [6]. These data help in the design of the thermal radiation shield,
from room temperature down to liquid helium temperature.
Table 4: The total emissivity of various metals at three different temperatures
300 K 78 K 4.2 K
3M black paint (80 μm) on a copper surface 0.94 0.91 0.89
Polished aluminium (33 μm roughness) 0.05 0.023 0.018
Polished copper (41 μm roughness) 0.10 0.07 0.05
304 Polished stainless steel (27 μm roughness) 0.17 0.13 0.08
337
B. BAUDOUY
Fig. 12: The total emissivity from ambient temperature down to liquid helium temperature (4.2 K). Reproduced
from [6] with permission of Springer.
2.2.2 Radiation exchange between two surfaces

The heat transfer by radiation between two enclosed surfaces; from one at temperature T1 with an
emissivity ε1 and a surface A1, to another at T2 with ε2 and A2, can be written as follows:
 1 − ε1 1 1− ε2 
q12 =σ (T14 − T2 4 ) /  + + , (19)
 ε1 A1 A1F12 ε 2 A2 
where q12 is the heat transfer rate from surface 1 to surface 2. F12 is the ‘view factor’, which is the
fraction of the heat leaving surface 1 that is intercepted by surface 2. A useful relationship can be used
to determine the view factors in complex geometry as the reciprocity relation arising from the
definition of the view factor [7]:
Ai Fij = A j F ji . (20)
338
This relation is essential to determine one view factor from knowledge of the other. Another
important view factor relation pertains to the enclosure surface. Again, from the definition of the view
factor, one can show in this case that
N
∑
j =1
Fij = 1 . (21)
In determining the heat balance in the apparatus design at low temperature, the most useful
special examples of Eq. (19) are the large parallel plates,
A
= A= 1 A2 , σ A1 (T14 − T2 4 )
q12 = ;
F12 = 1, 1 1 (22)
+ −1
ε1 ε2
and the long concentric cylinders,
A1 r1
= , σ A1 (T14 − T2 4 )
A2 r2 q12 = .
1 1 − ε 2  r1  (23)
F12 = 1, +
ε1 ε 2  r2 
2.2.3 Shielding and multilayer insulation

To understand the philosophy of ‘passive’ thermal shielding from room temperature to low
temperature, simple evaluation using the blackbody relation= q σ (Twarm
4 4
− Tcold ) is sufficient, without
knowledge of the emissivities and view factors. The heat transfer density from 300 K to 77 K using
this equation is 457 W·m–2 and from 300 K to 4.2 K, q = 459 W·m–2. In other words, from 77 K to
4.2 K, q = 2 W·m–2. The heat transfer rate from 77 K to 4.2 K is 200 times lower than that from 300 K
to 77K (and 4.2 K). Obviously, an intermediate surface held at intermediate temperature is required to
reduce the heat load at low temperature. Using Eq. (22) with the assumption that the surfaces and
emissivities are identical for the different reflecting surfaces, one can calculate the heat flux density q
as a function of the number of intermediate surfaces for passive shielding, as shown in Table 5.
Table 5: Passive shielding heat transfer as a function of the intermediate surface number
εσ 4 4 1 σε 4 4 1 σε 4 4
=q (Twarm Tcold
−= ) q (Twarm Tcold
−= ) q (Twarm − Tcold )
2−ε 2 2−ε n +1 2 − ε
4 4 4 4
Twarm + Tcold Twarm + Tcold
T4 = Ti 4 Tcold
= 4
+
2 i +1
339
B. BAUDOUY
In practice, passive thermal shielding is implemented using a multilayer insulation (MLI)

system (also called superinsulation), an assembly of reflective films (usually aluminium or aluminized
polyester film) separated by insulating interlayers (polyester, glass-fibre nets, or paper), operated under
vacuum, as shown in Fig. 13. The reflecting layers reduce heat transfer by radiation, the insulating
interlayers reduce heat transfer by conduction between reflecting layers, and the high vacuum reduces
convection and residual gas conduction.
Fig. 13: A depiction of a multilayer insulation (MLI) system
Under a good vacuum, a typical value of the heat transfer for a 20-layer MLI system is around
1–3 W·m–2 from 300 K to 80 K (but can reach 5 W·m–2 if compressed) and less than 100 mW·m–2
between 80 K and 4 K. To optimize the use of MLI, the maximum number of layers per centimetre
should be between 20 and 30, and isothermal contact points are required. But the efficiency of the MLI
can deteriorate due to bad installation; for example, gaps in the MLI blankets, overlapping, with the
inner (cold) side in contact with the outer (hot) side, and mechanical stress. For comprehensive
application data and empirical formulas, see Ref. [8].
2.3 Convection
2.3.1 Introduction
Once again, we will present the basic notion of the physics of convection to extract the useful
information on cryogenic design. In convection heat transfer, the heat can be transferred in the fluid by
movement of matter. It is a quantity of energy that is advected within the fluid. The movement of
matter can be created externally by a pump or a pressurization system. Hence we talk about forced
convection. In this regime, the equations of motion in the Boussinesq approximation are useful to
present the dimensionless numbers and their physical significance. It will be also useful to evaluate
heat transfer coefficients and flow regimes. In the steady-state regime, the equations of motion in the
Boussinesq approximation are as follows:
Continuity ∇.v =
0 ∇.v* =
0
ρ v.∇v = −∇p + µ∇ 2 v + f Dimensionless v* .∇v* = −∇p* + Re −1 ∇ 2 v*
Navier–Stokes (24)
Energy ρ Cv.∇T = k ∇ 2T + Q *
v .=
∇T * −1
Re Pr ∇ T −1 2 *
From the dimensionless equations, the flow is characterized by two important numbers: the
Reynolds number, Re, which is the ratio of the inertia to the viscous forces,
ρ Lv
Re = , (25)
µ
and which characterizes the flow regime; and the Prandtl number, Pr, which is the ratio of the
momentum to the thermal diffusivity,
340
µC
Pr = , (26)
k
and which characterizes the properties of the fluid. It is essential to know the regime in which the flow
is taking place, since the surface heat transfer and the friction both depend strongly on it. When
Re < 2300, it is in the laminar regime, where the viscous forces dominate, creating an ‘ordered’ flow
with streamlines. In this regime, the surface heat transfer is low and so is the surface friction. The
turbulent regime is reached for ReL > 5 × 105, while ReD > 4000 for a plate and a tube-shaped
geometry, respectively. Here, the inertia forces dominate, and the flow becomes highly irregular
(velocity fluctuation), as shown in Fig. 14 (velocity fluctuation). The surface heat transfer and the
friction are higher than those in the laminar regime.
Fig. 14: The development of a velocity boundary layer in a fluid in contact with a plate
When the fluid movement is created internally, by a decrease or increase of the fluid density or
by the buoyancy effect, it is called ‘natural convection’. In the steady-state regime, the corresponding
equations of motion in the Boussinesq approximation are as follows:
∇.v =
0 ∇.v* =
0
Dimensionless
ρ v.∇v = −∇p + µ∇ 2 v + β∆Tg v* .∇v* = −∇p* + Re −1 ∇ 2 v* + Gr Re −2 T * (27)
2
ρ Cv.∇T = k ∇ v* .=
∇T * Re −1 Pr −1 ∇ 2T *
In natural convection, the main dimensionless number is the Grashof number,
g β∆TL2
Gr = , (28)
µ2
which is the ratio of the buoyancy to the viscous forces. Looking at the dimensionless equation, one
can deduce that when Gr × Re–2 ≫ 1, forced convection is negligible. If Gr × Re–2 ≈ 1, then the flow is
in a mixed convection state, where natural and forced convections are equally important. The Grashof
number has the same role in natural convection as Re in forced convection. Turbulence has a strong
effect, as in forced convection, and is reached for Gr × Pr ≥ 105.
When there is an interaction with solid surfaces in convection, the quantity of energy transferred
in (or out) of the fluid to solids must be evaluated. The heat transferred to solid elements is modelled
by Newton’s law:
=q h (Ts − T∞ ) , (29)
341
B. BAUDOUY
where h is the heat transfer coefficient [W·m–2·K–1], Ts the temperature of the solid and T∞ is the
temperature far from the solid. At the boundary as depicted in Fig. 15, the local heat flux is qn = k.∇Tn;
that is, the heat flux going through the conductive layer of the fluid also obeys Newton’s law. This
representation leads to the dimensionless Nusselt number, defined as
hL ∂T *
Nu = = . (30)
k ∂n* 0
Fig. 15: The thermal boundary layer at the wall of a flow with a plate
The Nusselt number is to the thermal boundary what the friction coefficient is to the velocity
boundary. The Nusselt number is defined as Nu = f(Re, Pr, L) for forced convection and Nu = f(Gr,
Pr, L) for natural convection. Their form is different for a turbulent or a laminar regime, and in the
following sections we will present the correlations that are useful for design at low temperature,
depending on the regime and the configuration of the flow.
2.3.2 Natural convection

The heat flux at the wall is computed with correlations of the type Nu = f(Gr, Pr, L), where the
thermophysical properties are usually established at an average temperature between the solid and the
fluid temperatures. The heat transfer coefficients are of the order of 10–100 W·m–2·K–1. In natural
convection, the simplest correlations are expressed as Nu = c × (Gr × Pr)n, in which n = 1/4 for a
laminar regime and n = 1/3 for a turbulent regime. The correlations are not very different from those
for classical fluids, but few data exist for cryogenic fluids, since two-phase phenomena take over even
at low heat fluxes. Table 6 shows various correlations taken from the literature [9-12].
Table 6: Nusselt correlations for cryogenic fluids
c n
Turbulent supercritical helium (vertical orientation) 0.615 0.258
Turbulent liquid nitrogen (various orientations) 0.14 1/3
Turbulent liquid hydrogen (various configurations) 0.096 0.352
2.3.3 Forced convection

As for natural convection, the correlations used for non-cryogenic fluids are suitable at low
temperature in the forced convection regime. For turbulent flows in pipes, the Dittus–Boetler
correlation, Nu = 0.023Re0.8Pr0.4, is used for hydrogen [13]; a modified version, Nu = 0.022Re0.8Pr0.4,
is used for supercritical helium [14]; and another modified version, Nu = 0.027Re0.8Pr0.14/3(μf/μw)0.14, is
used for nitrogen [15]. The heat transfer coefficients can go up to several kW·m–2·K–1. A laminar flow
in pipes is very rare, except in porous media, and it is too specific for details to be given here.
342
2.3.4 Boiling convection

In boiling convection, heat is transferred between a surface and the fluid by the conjunction of a phase
change and the vapour bubble movement in the vicinity of the surface. Heat transfer combines natural
convection in the liquid, latent heat to be absorbed for the bubble formation, and the bubble
hydrodynamics. The heat transfer process depends on the bubble growth rate, the detachment
frequency, the number of nucleation sites, and the surface conditions. In pool boiling, several regimes
can be identified, as shown in Fig. 16. Before the onset of boiling, natural convection takes place; and
since the boiling heat transfer is extremely efficient, when boiling is activated, the wall temperature
increase is slowed down. After the onset of boiling, the evolution of nucleate boiling is encountered;
that is, from partially to fully developed nucleate boiling, where the vapour content and structure are
continuously increasing. At one point, called the ‘critical point’, the vapour production is so high that
the vapour structures coalesce and form a blanket of vapour at the heating surface. This results in a
new regime called ‘film boiling’, where the heat is primarily transferred by conduction through the
vapour film, explaining the large increase of the wall temperature. Table 7 shows the different heat
flux and temperature increases at the critical points for various cryogenic fluids [10, 12, 16, 17].
Fig. 16: The boiling regimes for a flat horizontal surface
Table 7: Characteristic points of the boiling curve
ΔTc (K) qc (kW·m–2) qr (kW·m–2)
Helium 1 10
Nitrogen 10 100
Hydrogen 5 100 10
There are several heat transfer correlations that can fit experimental data within one order of
magnitude and here we will present the most popular one – the Kutateladze correlation
q = f(p) × ΔT2.5, which works for most cryogenic fluids:
0.6 0.125 3/2
h σ 
1/2
 qCp ρ  σ 1/2    ρ  2  σ 3/2   p 
−4
 =
 3.25 × 10     g         , (31)
kl  g ρ    hv ρ v k  g ρ     µ   g ρ    (σ g ρ )1/2 
  
343
B. BAUDOUY
where σ is the tension surface and p is the pressure. The subscripts ℓ and v stand for ‘liquid’ and
‘vapour’, respectively.
Figure 17 shows an example of experimental data for flat horizontal boiling heat transfer in
nitrogen and the comparison with the Kutateladze correlation [10]. The critical heat flux can also be
computed with Kutatekadze correlations, which once again give very good results compared to
experimental data for several cryogenic fluids, such as helium, nitrogen, oxygen, and hydrogen
[12, 18]. It is written as follows:
1/4
qc 0.16hv ρ 1/2
v  σ g ( ρ − ρ v )  ,
=  (32)
where hℓv is the latent heat of vaporization. It is worth noting that this correlation is not valid when the
fluid is sub-cooled; that is, when the pressure above the heated surface is higher than the saturation
pressure. For this peculiar case, the following correlation can be used [19]:
 3/4
qc,sub ρ  Cp ∆Tsub 
= 0.2 1 + 0.15    σ. (33)
qc,sat 
  ρv  hv 

Last but not least, two-phase forced flow heat transfer modelling must take in account the
boiling heat transfer, which depends on the surface heat transfer, and the forced convection, which
depends on the vapour quality (x = ṁv/ṁt) and the total mass flow rate (ṁt). Usually, boiling tends to
be dominant for low vapour quality and high heat flux, while forced convection tends to be dominant
for high vapour quality and a large mass flow rate. Several general correlations specific to cryogenic
fluids emerge. The best approach is to try more than one such correlation to evaluate the heat transfer
rate. The most successful ones are the superposition method (Chen) [20], the intensification model
(Shah) [21], and the asymptotic model (Liu and Winterton, n = 2) or the Steiner–Taborek (n = 3)
correlation. The latter is considered by some authors as the most accurate and it includes the cryogenic
fluids (helium, hydrogen, nitrogen, oxygen, etc.) [22].
Fig. 17: Boiling curves for horizontal pool boiling for nitrogen, reprinted from [10] with permission of Elsevier
344
3 Cooling techniques at low temperature
3.1 Introduction
Nowadays, cooling a device to low temperature can be achieved with or without cryogenic fluids. A
cooling method is called a ‘wet method’ when a cryogenic fluid is used in contact with the device or a
‘dry method’ when a cryocooler is used without any fluid as coolant. A third technique is the ‘indirect
cooling method’, when a cryogenic fluid is used without direct contact with the device but only
through intermediate components. Figure 18 presents a good summary of all three methods and their
variants that are used to reach the low temperature.
The ‘dry system’ method – or simply the ‘cryogen-free cooling method’ – relies solely on
conduction to cool down a system. The cold source in this method is a cryocooler. The main reason for
using such a method is to avoid dealing with a cryogenic fluid. Due to the performance limitations of
the cryocoolers available today, however, they are only suitable for systems subjected to a small heat
load (typically, a few watts at 4 K) and with slow transient perturbation processes. Typical examples
are room-temperature bore magnets or MRI magnets.
In the ‘indirect cooling’ method, there is no direct contact between the cryogen and the system;
yet the heat conduction in the system is still of importance, but one now has to consider the surface
heat transfer between the fluid and the cooling loop on the system. This method can be generally
achieved with an external flow of cryogen as a cold source. The main reason for using this method is
the reduction of the cryogenic fluid inventory while keeping a high fluid–solid heat transfer
coefficient. It is suitable for moderate or well-distributed heat load systems and with slow transient
perturbation processes. Typical examples are the large detector magnets, such as CMS or Atlas at
CERN.
Fig. 18: Schematic examples of various cooling methods
345
B. BAUDOUY
The principal reason to use a ‘wet system’ (or direct contact) cooling method is to evacuate a
large heat load or maintain a uniform temperature within the system. The direct contact between the
cryogenic fluid and the system not only ensures a large heat transfer rate, but also allows the liquid to
serve as an enthalpy reserve to overcome large transient heat perturbations. This can be achieved with
a liquid bath, a stagnant coolant (He II pressurized or saturated), or a single-phase flow (supercritical).
The first examples that come to mind are the accelerator magnets, although CICC magnets (ITER,
45 T NHMFL magnet), He II cooled magnets (Tore Supra, Iseult), and superconducting cavities are
also cooled using the ‘wet system’ method (bath cooling).
3.2 Different methods of cooling
3.2.1 Baths
The most common cooling method is to use a liquid bath in which the system is immersed. The
simplest of these methods is the saturated bath, where the fluid at the free surface is at saturation
(T ≈ Tsat). It is a direct or indirect cooling method with no net liquid mass flow and where the main
heat transfer process is essentially due to latent heat of vaporization. The main advantages are the
simplicity of the cryogenic design and operation, the high heat transfer due to nucleate boiling, and,
therefore, an almost constant surface temperature. If there is sub-cooling due to a hydrostatic pressure
head, as is depicted for the helium cooling bath in Fig. 19, then there is an extra ΔTsub that can be used
before boiling is reached. The disadvantage of this technique, on the other hand, is that a large quantity
of cryogen has to be handled, particularly in case of a quench of a cryomagnetic system (risk of
pressure rise). A limited range of temperature along the liquid–gas saturation curve is available (He,
4.2 K; H2, 20.4 K; N2, 77.3 K; etc.). If q > qcr, then the heat transfer coefficient in film boiling is an
order of magnitude smaller than that in a nucleate boiling case and non-uniform cooling can result due
to film vapour formation.
Fig. 19: An example of a helium cooling bath method
Superfluid helium (He II) bath cooling techniques are frequently employed to maintain a low
temperature in superconducting accelerator cavities or certain accelerator magnets (below 2.17 K). The
method exploits the extraordinary heat transfer capability of He II. The equivalent heat transfer
conductivity is of the order of k ≈ 105 W·m–1·K–1 and is independent of the flux, gravity, and surface
orientation. The main thermal barrier of this cooling technique is the surface thermal resistance; the
Kapitza resistance. For copper immersed in He II, the Kapitza resistance Rk = 3 × 10–4 K·m2·W–1 [23];
and for Kapton, Rk = 10–3 K·m2·W–1 [24]. In the use of a saturated bath, the main risk is the handling
of the volume below atmospheric pressure (leaks inducing air contamination). For pressurized baths
most of the cryostat is above atmospheric pressure (as in the case of LHC magnets) which eliminates
346
the risk of leaks. To take advantage of the cooling capacity of superfluid helium (i.e. stay in the
superfluid state), the temperature superheat in the magnet has to be lower than ΔTλ (see Fig. 19).
Going lower in temperature is obviously costly for a large heat load (oversizing of pumping unit) and
also complicates the design and operation. A detailed presentation of superfluid helium heat transfer is
done in a different chapter of this volume.
3.2.2 Forced flow

Forced flow is one of the methods used to reduce the amount of cryogen, especially in indirect cooling
composed of a network of peripheral tubes. Another advantage is the adjustable heat transfer rate with
mass flow rate. An example of the forced flow of hydrogen is shown in Fig. 20 [13]. High heat transfer
can be achieved, of the order of 104 W·m–2, in different cryogens in single phase. For a single-phase
flow, the main disadvantages include: the pressurization system, or the implementation of the
circulation pump and its maintenance at low temperature; the implementation of the heat exchanger
system to sub-cool the fluid; and the temperature range limitation due to finite sub-cooling. For the
example given in Fig. 20, the maximum allowable ΔT is 5 K prior to reaching the two-phase boiling
regime (the onset of nucleate boiling).
Cooling in the supercritical state can be grouped in the single-phase category and one of its main
advantages is a heat transfer coefficient comparable to that of pool boiling in helium. The heat transfer
characteristics are q ≈ 104 W·m–2 for ΔT ≈ 1 K for helium [14]. But one needs to have a ‘heavier’
cryogenic installation to ensure a periodic re-cooling for operation (a series of large magnets) and to
maintain a pressure above Pcrit (2.25 bars Absolute for He) in the system (usually 3 < Ploop < 6 bars
Absolute along the cooling circuit). Another advantage of cooling with supercritical fluid is the lack of
hydraulic instabilities in the single-phase flow compared to a two-phase flow.
However, the use of two-phase forced flow cooling does have some advantages. The first is the
guarantee of having an almost isothermal flow due to the high heat transfer, even at high vapour
quality, in this flow regime. For the example of a 10 mm diameter horizontal tube in helium, the
following cooling characteristics can be obtained: qmax ≈ 3 × 103 W·m–2 for a mass flow rate of
6 × 103 kg·s–1 and for T ≈ 1 K with no initial sub-cooling [25]. The main drawbacks are the limited
range of temperature cooling (LHe < 5.2 K, 14 < LH2 < 20.4 K and 65 <LN2 < 77.3 K, for example)
and the non-uniform cooling if the vapour and the liquid are not homogeneous in their motion.
Moreover, if the heat flux q > qcr, then there is film boiling and the heat transfer rate becomes an order
of magnitude smaller.
Fig. 20: The boiling curve in forced hydrogen flow, reprinted from [13] with permission of Elsevier
347
B. BAUDOUY
3.2.3 Natural and two-phase circulation loops

Circulation loops are an auto-tuned mass flow rate system in which the flow is created by the weight
unbalance between the heated branch and the feeding branch of the loop due to vaporization or
decreased vapour density, as described in Fig. 21. The main advantage is that there is no need for a
circulating system (centrifugal pump, forced pressure differential, etc.). We can distinguish two types
of circulation loops: the ‘Open Loop’, where the boil-off goes out of the system (as in Fig. 21) to be
re-liquefied somewhere else before refilling the reservoir permanently to avoid a dry-out; and the
‘Closed Loop’, where the vapour is re-condensed in a closed reservoir with a heat exchanger.
The operating principle is the same if a single-phase fluid is used in the circulation loop, where
density variation creates fluid motion. In nitrogen, vertical single-phase natural convection has been
studied and the heat transfer coefficient as found in Fig. 21 is q ≈ 4 kW·m–2, with a mass flow rate of
40 g·s–1 and ΔT ≈ 3 K for a Ø10 mm tube [26].
In a vertical configuration, two-phase flow circulation loops also have high heat transfer rates.
In nitrogen, the heat flux is around 104 W·m–2 Ø10 mm tube for Ø10 mm and a mass flow rate of
40 g·s1 for ΔT ≈ 2.5 K [25], whereas for helium the heat flux is around 103 W·m–2 for Ø10 mm and a
mass flow rate of 20 g·s–1, and ΔT ≈ 0.3 K [27]. When the heated section is horizontal, the heat
transfer is as high as that for the vertical case, but there are flow instabilities at low heat flux flow [28].
The natural circulation loop in liquid helium can be modelled easily, with good accuracy, using the
homogeneous flow model [29, 30].
3.2.4 Cryogen-free cooling and the coupled system

Today, cryocoolers can provide sufficient power to cool down and maintain a small cryomagnetic
system at low temperature. In certain applications, one can use a conductive thermal link between the
cold source (the cryocooler) and the low-temperature device. This is known as the cryogen-free
cooling method. Even if the obvious advantages of easy implementation (no liquid, no heat exchanger,
no transfer line, etc.) are very attractive, one has to be very accurate in the thermal design, since these
cryocoolers provide a finite cooling power at a prescribed temperature. Therefore if the real power to
be extracted exceeds the maximum power of the cryocooler, the working temperature will inevitably
be higher than expected. The other technical issue to keep in mind is that a thermal link is necessary to
distribute the cooling power over the entire system with this ‘point-source’ of cold that is the
cryocooler. For a fully conductive thermal link, the thermal diffusion in the thermal link limits the
cooling for transient events.
One of the solutions to overcome the disadvantages of the cryogen-free cooling method is to use
a thermal link with a fluid. Several methods are under development and we will briefly present some
of them.
Capillary-pumped devices have been used for many temperature ranges. The operating principle
is identical whatever the temperature range. A flow is created by capillary pressure in a porous
medium at the liquid/vapour interface. The simplest system is a heat pipe, with a wick inside the pipe
serving as porous media. An typical example of such a system has been presented by Kwon [31]. This
wick-based heat pipe, developed at the nitrogen temperature range, can transfer around 50 W between
the cold source and the system to be cooled with a temperature difference of around 10 K. Here, the
large temperature difference between the cold source and the warm source is the main disadvantage.
More sophisticated cooling systems have been designed, especially for space applications; for
example, the cryogenic loop heat pipe, which is basically a heat pipe in a loop configuration to create a
mass flow rate. At nitrogen temperature, the most powerful ones can transmit 40 W for ∆T = 6 K
between the cold source and the system [32].
348
Fig. 21: A schematic of a vertical circulation loop: the boiling curve in a 1-m high nitrogen loop [26]
Another phase-change based thermal link is the Oscillating Heat Pipe (OHP). An OHP generally
consists of a capillary tube, wound in a serpentine manner, connecting the ends to the inlets. The
simplicity of its construction and the versatile geometry are the main advantages. The OHP utilizes the
pressure change due to volume expansion and contraction during a phase transition to excite the
oscillation of liquid slugs and vapour bubbles between the evaporator and the condenser. This system
can be used with different fluids and exhibit large equivalent heat transfer conductivity (Table 7). As
for the other capillary-pumped devices, the main drawback is the large temperature difference between
the condenser (cooling part) and the evaporator (heating part), as shown in Table 8 [33].
The so-called ‘vertical thermosyphon’ is also an option to consider when creating a thermal link.
The working principle is somewhat similar to the above-mentioned circulation loop and is based on the
weight unbalance, but here there is a single vertically oriented tube. The liquid flows down the wall
and the vapour flows in a counter-current manner to the liquid at the centre of the tube. One example
in nitrogen can be found in [34]. This thermosyphon is able to transmit 20 W with a ΔT of 10 K at
nitrogen temperature.
Table 8: The performance of an OHP for different fluids [33]
Fluid Heat Temperature of Temperature of Keff

input (W) cooling part (K) heating part (K) (kW·m ·K )
–1 –1
H2 0–1.2 17–18 19–27 0.5–3.5

Ne 0–1.5 26–27 28–34 1–8
N2 0–7 67–69 67–91 5–18
349
B. BAUDOUY
Fig. 22: (a) A schematic of a couped circulation loop with a cryocooler. (b) A typical boiling curve reprinted
from [35] with permission of Elsevier.
Finally, a small natural circulation loop coupled with a cryocooler can be used to serve as a self-
sustaining thermal link at cryogenic temperatures. The principle of this link is shown in Fig. 22. It is
mainly composed of a condenser/phase separator cooled by the second stage of a cryocooler.
Connected to the condenser, one can find a loop made of the heated branch (heat exchanger) and the
feeding branch. Liquid helium flows through this loop and is re-condensed in the condenser. The heat
transfer coefficient for a 4 mm diameter tube loop, in helium around 4.2 K, is 5000 W·m–2·K–1, with a
critical heat flux of 500 W·m–2 [35].
References
[1] Cryocomp, Eckels Engineering, v. 3.06. Cryodata Inc., Florence, SC 29501, USA.
[2] J.W. Ekin, Experimental Techniques for Low Temperature Measurements (Oxford University
Press, Oxford, 2006).
[3] R.F. Barron, Cryogenic Heat Transfer (Taylor & Francis, Philadelphia, PA, 1999).
[4] J.C. Johnson. A compendium of the properties of materials at low temperatures, part 1. WADD
Tech. Rep. 60-56 (US Government Printing Office, Washington, DC, 1960).
[5] K.H. Hawks and W. Cottingham, Total normal emittances of some real surfaces at cryogenic
temperatures, in Advances in Cryogenic Engineering, Ed. K. Timmerhaus (Plenum Press, New
York, 1970), Vol. 16, pp. 467–474.
[6] W. Obert, Emissivity measurements of metallic surfaces used in cryogenic applications, in
Advances in Cryogenic Engineering, Ed. R. Fast (Plenum Press, New York, 1982), Vol. 27, pp.
293–300.
[7] F.P. Incropera and D.P. DeWitt, Fundamentals of Heat and Mass Transfer, 5th ed. (John Wiley
& Sons, New York, 2002).
[8] T. Nast. Radiation heat transfer, in Handbook of Cryogenic Engineering, Ed. J.G. Weisend
(Taylor & Francis, Philadelphia, PA, 1998), p. 186.
[9] M.A. Hilal and R.W. Boom, Int. J. Heat Mass Transfer 23 (1980) 697–705.
350
[10] J.A. Clark, Cryogenic heat transfer, in Advances in Heat Transfer, Eds. T.F. Irvine and J.P.
Hartnett (Academic Press, New York, 1968), pp. 325–517.
[11] D.E. Daney, Int. J. Heat Mass Transfer 19(4) (1976) 431–441.
[12] Y. Shirai et al., Cryogenics 50(6–7) (2010) 410–416.
[13] H. Tatsumoto et al., Phys. Procedia 36 (2012) 1360–1365.
[14] P.J. Giarrantano et al., Forced convection heat transfer to subcritical helium I, in Advances in
Cryogenic Engineering, Ed. K. Timmerhaus (Plenum Press, New York, 1974), Vol. 19, pp.
404–416.
[15] K. Ohira et al., Cryogenics 51(10) (2011) 563–574.
[16] R.V. Smith, Cryogenics. 9(1) (1969) 11–19.
[17] M. Kida et al., J. Nucl. Sci. Technol. 18(7) (1981) 501–513.
[18] D.N. Lyon, Boiling heat transfer and peak nucleate boiling fluxes in saturated liquid helium
bertween the λ and critical temperatures, in Advances in Cryogenic Engineering, Ed. K.
Timmerhaus, (Plenum Press, New York, 1964), Vol. 10, pp. 371–379.
[19] Y.A. Kirichenko et al., Cryogenics 23(4) (1983) 209–211.
[20] J.C. Chen, Ind. Engng Chem. Proc. Des. Dev. 5(3) (1966) 322–339.
[21] M.M. Shah, ASHRAE Trans. 82(2) (1976) 66–86.
[22] H. Steiner and J. Taborek, Heat Transfer Engng 13(2) (1992) 43–69.
[23] A. Iwamoto et al., Cryogenics 41(5–6) (2001) 367–371.
[24] B. Baudouy, Cryogenics 43(12) (2003) 667–672.
[25] M. Mahé, Etude des propriétés d'échange thermique de l'hélium diphasique en convection
forcée, Ph.D. thesis, Université Pierre et Marie Curie, Paris, 1991.
[26] B. Baudouy, AIP Conf. Proc. 1218 (2010) 1546–1553.
[27] L. Benkheira et al., Int. J. Heat Mass Transfer 50(17–18) (2007) 3534–3544.
[28] B. Gastineau et al., IEEE Trans. Appl. Supercond. 22(3) (2012) 900–1004.
[29] B. Baudouy, AIP Conf. Proc. 1434 (2012) 717–723.
[30] B. Baudouy et al. Cryogenics 53 (2013) 2–6.
[31] D.W. Kwon and R.J. Sedwick, Cryogenics 49 (2009) 514–523.
[32] Y. Zhao et al., Int. J. Heat Mass Transfer 54 (2011) 3304–3308.
[33] K. Natsume et al., Cryogenics 51 (2011) 309–314.
[34] A. Nakano et al., Cryogenics 38(12) (1998) 1259–1266.
[35] Y. Song et al., Int. J. Heat Mass Transfer 66 (2013) 64–71.
Bibliography
Journal and conference proceedings
Advances in Cryogenic Engineering, Vols. 1–57, Proceedings of the Cryogenic Engineering and
International Cryogenic Materials Conference (USA).
Cryogenics, Elsevier Science (http://www.journals.elsevier.com/cryogenics/).
Proceedings of the International Cryogenic Engineering Conference (Europe/Asia).
Monographs
R.F. Barron, Cryogenic Heat Transfer (Taylor & Francis, Philadelphia, PA, 1999).
351
B. BAUDOUY
J.W. Ekin, Experimental Techniques for Low Temperature Measurements (Oxford University Press,
Oxford, 2006). http://www.oup.com
T.M. Flynn, Cryogenic Engineering (Marcel Dekker, New York, 1997).
W. Frost, Heat Transfer at Low Temperature (Plenum Press, New York, 1975).
S.W. Van Sciver, Helium Cryogenics, 2nd ed. (Springer, New York, 2012).
J.G. Weisend (Ed.), Handbook of Cryogenic Engineering (Taylor & Francis, Philadelphia, PA, 1998).
Databases
− Cryocomp, Eckels Engineering, v. 3.06. Cryodata Inc., Florence SC 29501, USA.
− Hepak, Gaspak, MetalPak, Cryodata Incorporated.
− NIST Database (http://cryogenics.nist.gov).
352
Cryostat Design
V. Parma 1
Abstract
This paper aims to give non-expert engineers and scientists working in the
domain of accelerators a general introduction to the main disciplines and
technologies involved in the design and construction of accelerator cryostats.
Far from being an exhaustive coverage of these topics, an attempt is made to
provide simple design and calculation rules for a preliminary design of
cryostats. Recurrent reference is made to the Large Hadron Collider magnet
cryostats, as most of the material presented is taken from their design and
construction at CERN.
Keywords: superconductivity, cryogenics, vacuum, heat transfer, MLI,

construction technologies.
1 Introduction
The term cryostat (from the Greek word κρύο meaning cold, and stat meaning static or stable) is
generally employed to describe any container housing devices or fluids kept at very low temperatures;
the notion of ‘very low temperature’ generally refers to temperatures that are well below those
encountered naturally on Earth, typically below 120 K.
The very first cryostats were used in the pioneering years of cryogenics as containers for
liquefied gases. The invention of the first performing cryostats is generally attributed to Sir James
Dewar (Fig. 1), and hence cryostats containing cryogenic fluids are nowadays also called dewars. In
1897 Dewar used silver-plated double-walled glass containers to collect the first liquefied hydrogen.
Even though the heat transfer phenomena were not well mastered during his époque, Dewar
understood the benefits of thermal insulation by vacuum pumping the double-walled envelope, as well
as shielding thermal radiation by silver-plating the glass walls.
H. Kamerlingh Onnes further developed glass-blowing which became the enabling technology
for making dewars for his laboratory in Leiden. He introduced this technique as one of the specialities
in the school of instrumentation he founded, the Leidse Instrumentmakersschool, which still exists
today.
Since those times, the evolution of cryostats has been led by specific needs for the variety of
applications. Today, cryostats can be found in a large range of applications spanning from industrial
products to specific devices for scientific research instruments.
Two representative examples of cryostats produced in large industrial series are cryostats for
superconducting magnet medical devices such as magnetic resonance imaging (MRI), with a typical
worldwide production of a thousand units per year, and storage reservoirs for industrial cryogenic
fluids (LN2, LO, LH) produced in several thousands of units per year. Figure 2 illustrates a few of
these examples.
1
vittorio.parma@cern.ch
V. PARMA
A possible breakdown of cryostats based on their applications is given in Fig. 3, where those for
accelerator applications are detailed. Cryostats are needed in most particles accelerators, from linacs to
synchrotron machines, where superconductivity is the enabling technology. Under test cryostats we
list those that are specifically developed for testing accelerator devices, though standard commercial
test cryostats are also available for a number of laboratory applications. In this course we will
concentrate on cryostats for accelerator devices, housing either superconducting magnets or
superconducting Radio Frequency (RF) cavities, and cooled with liquid helium.
Fig. 1: Sir James Dewar (1842–1923), and his double-walled glass dewar
(a) (b)
Fig. 2: (a) An MRI device; (b) cryogenic storage tanks
Fig. 3: Cryostats and their applications
354
C RYOSTAT D ESIGN
2 Cryostat requirements
When designing cryostats for accelerator superconducting devices, the engineer is confronted with the
interaction between various technical disciplines, some of which are the domain of specialists, like
superconductivity, cryogenics and vacuum (Fig. 4). In these domains has to be developed a general
understanding in order to be able to interact with specialists.
The basic function of a cryostat is to house and thermally insulate a superconducting device,
while providing all the interfaces for its reliable and safe operation (cryogen feeding, powering,
diagnostics instrumentation, safety devices, etc.) The basic technical competencies for a cryostat
design engineer are mechanical engineering and heat transfer. But their application at cryogenic
temperatures calls for specific competencies on thermal and mechanical properties of materials at low
temperatures, which make the work very specific and a discipline in itself. Cryostat design also
includes system integration of complex devices that have to be housed inside the insulation vacuum
and often at cryogenic temperatures: typical examples are beam instrumentation such as Beam
Position Monitors (BPM), cryogenic instrumentation (temperature, pressure, level, and mass flow
sensors) and control devices (valves, servo-motors, and mechanical transmissions), electrical circuits
for magnets’ and cavities’ powering and protection, etc. All these items or devices need to be designed
for integration so that they can be assembled, operated and maintained when needed and, in most
cases, designed not to hinder the overall thermal performance of the cryostats.
Fig. 4: Cryostat design, and related disciplines
An illustrative example of a cryostat for accelerators is one of the LHC dipoles (Fig. 5).
The main functionality of the cryostat is to support and position precisely the heavy dipole
magnets while limiting to a minimum the thermal heat loads from ambient temperature to the 2 K
temperature of the magnets. The magnets are supported on support posts made of low thermal
conductivity material and are placed inside vacuum vessels in order to remove heat loads from gas
convection and conduction; thermal shields cooled to a higher temperature than that of the magnets,
and wrapped with reflective Multilayer Insulation (MLI), further contribute to reducing the radiation
heat loads to the magnets.
The precise alignment of the magnets within the accelerator is obtained by the fine adjustment
of the external jack supporting the cryo-magnet assembly while its position is measured by surveying
355
V. PARMA
alignment targets mounted on the vacuum vessel. The positioning accuracy and stability of the magnet
inside the cryostats is the main requirement for which the supporting system has been designed.
Fig. 5: Cross-section view of the LHC dipole cryostat
Table 1 summarises the main requirements of the supporting and alignment system for the main
dipoles and quadrupoles of the LHC, where sub-millimetre positional reproducibility and stability as
well as alignment precision is sought.
Table 1: Main requirements of the supporting and alignments system of the LHC main magnets
Property/requirement Dipole Quadrupole

Weight 300 kN 65 kN
Cold mass in cryostat positioning accuracy (after assembly):
Radial ± 1 mm ± 0.5 mm
Vertical ± 1 mm ± 0.5 mm
Longitudinal ± 2 mm ± 1 mm
Positioning reproducibility–stability of cold mass in cryostat (in operation, during lifetime):
Radial ± 0.3 mm (rms) < ± 0.3 mm (rms)
Vertical ± 0.3 mm (rms) < ± 0.3 mm (rms)
Longitudinal ± 1 mm (rms) < ± 1 mm (rms)
Radial tilt ± 0.3 mrad (rms) < ± 0.3 mrad (rms)
External supporting system:
Alignment range ± 20 mm ± 20 mm
Movement resolution ± 0.05 mm ± 0.05 mm
Considering the extremely high electrical power consumption of the cryogenic plants for heat
extraction at cryogenic temperatures (about 1 kW W−1 from 1.8 K to 300 K), the static heat inleak to
the cold masses at 1.8 K had to be minimized in the most efficient and economical way. During the
course of the development of the LHC cryostats and its components, systematic laboratory heat load
measurements were performed to validate designs, manufacturing and assembly methods, completed
by prototype thermal performance assessments. The thermal heat loads per unit length reached for the
LHC cryostats are illustrated in Table 2.
356
C RYOSTAT D ESIGN
Table 2: Thermal budget per unit length of the LHC cryostats (W m−1)
T
50–75 K 4.6–20 K 1.9 K
−1
Static heat loads (W m ) 7.7 0.23 0.21
The following chapters present a general, but non-exhaustive, coverage of these disciplines. The
LHC cryostat will often be used as a case study throughout the course. The reader is referred to the
bibliography and references for further developments, see Refs. [1–18].
3 Heat transfer for cryostats

In the following we shall consider the heat transfer phenomena that are relevant to cryostats. Solid
conduction, which is generally the dominating contribution in supporting systems and all cryostat
feed-throughs (current leads, instrumentation wires, beam pipes, etc.); residual gas conduction, which
can introduce a non-negligible contribution in the case of a non-perfect vacuum insulation; thermal
radiation, which is normally the dominating heat load contribution; and the use of multilayer
insulation and thermal shielding as the protection measures normally employed.
3.1 Solid conduction

Fourier’s law expresses heat conduction in solids as being proportional to a temperature gradient
through k, which is the thermal conductivity of a material:
𝑄̇ = −𝑘(𝑇) ∙ 𝐴 ∙ grad(𝑇) . (1)
At cryogenic temperatures k is generally temperature-dependent and non-linear, implying that

numerical integration over temperature is necessary in most practical cases.
Figure 6 shows that thermal conductivity for solid materials spreads over more than five orders
of magnitude, underlining the importance of the selection of materials where thermal conductivity is a
property of interest. We can also notice that, for a single material, the trend is a reduction of thermal
conductivity with temperature, as would be expected by the reduction with temperature of phonon
vibrations in the material structure. This is, however, not the case in metals, where the main
contribution to conductivity is the free movement of conduction electrons inside the crystalline lattice.
In this case electron scatter with phonons reduces with temperature, hence increasing the mobility of
electrons, resulting in an increase of thermal conductivity. At lower temperatures a second scatter
mechanism with impurities prevails and thermal conductivity reduces once again.
In high-purity metals, the increase of thermal conductivity is very pronounced (more than one
order of magnitude for pure coppers, as can be seen in Fig. 6); but in metal alloys, where impurities
dominate, thermal conductivity shows a monotonic decrease according to the relation:
𝑘(𝑇) ≅ 𝑇/𝑎𝑖 (2)
where ai is a constant depending on the level of impurities of the material.

It is interesting to note, from electron conduction theory and in an analogy with electrical
diffusion that, for metals, thermal conductivity k and electrical resistivity σ follow the Wiedemann–
Franz law:
𝑘
≅𝐿∙𝑇 (3)
𝜎
357
V. PARMA
Fig. 6: Thermal conductivity (W m−1 K−1) of selected materials as a function of temperature (K). (Ref. [12].)
which states that the ratio between these two properties is, in good approximation, a constant for all
metals at the same temperature. The proportionality constant L, known as the Lorenz number, is equal
to:
𝐿 = 2.44 × 10−8 (W Ω K −2 ) . (4)
Electrical resistivity measurements, which are easier to make, therefore offer a more practical but
indirect measure of thermal conductivity.
Let us now reconsider Fourier’s law, but in one dimension, the case of a beam of cross-section
A, and length l. Equation (1) in its integral form becomes (dropping the minus sign and keeping in
mind that heat flows from high to low temperature):
𝐴 𝑇
𝑄̇ = 𝑙
∙ ∫𝑇 2 𝑘(𝑇) d𝑇 (5)
1
which expresses the conduction heat flow when the two tips of the beam are kept at temperatures T1
and T2. We can therefore note that the integral part is a property of the material in use and of the
358
C RYOSTAT D ESIGN
temperature range. For engineering applications, the use of tables of conductivity integrals is often
practical. Table 3 gives an example for some of the most commonly used materials in cryostats:
Table 3: Thermal conductivity integrals (W m−1) for selected technical materials (between indicated
temperatures and 4.2 K).
20 K 80 K 290 K
OFHC copper 11,000 60,600 152,000
DHP copper 395 5,890 46,100
Aluminium 1100 2,740 23,300 72,100
Aluminium 2024 160 2,420 22,900
Stainless steel AISI 304 16.3 349 3,060
Typical glass-fibre/epoxy composite G-10 2 18 153
One can appreciate immediately that when designing a support for a superconducting device operating
at 4.2 K, choosing a G-10 type glass-fiber epoxy composite can reduce conduction heat loads by a
factor of about 20 as compared to steel (this is of course only an indicator in the choice of the material
because other performance metrics like strength or stiffness have to be included).
Let us now consider once again the case of a beam of area A and length L, as illustrated in
Fig. 8, where heat is transferred along x, and where an internal energy is deposited inside its volume.
Fig. 7: One-dimension solid beam, with heat deposition q and conduction heat flux along x
We can write the energy balance equation for an element x + Δx, introduce Fourier’s law describing
longitudinal heat transfer, and introduce the change of internal energy related to the heat capacity of
the element as:
∂  ∂T  ∂T
 k·  + q =ρc· (6)
∂x  ∂x  ∂t
where the first term represents longitudinal conduction, 𝑞̇ is the internal heat deposited, and the term
on the right-hand side represents the thermal inertia of the element related to its heat capacity (ρ is the
density and c is the specific heat capacity).
If k is considered constant (valid for small changes of T), we can rewrite the equation as:
∂ 2T q 1 ∂T k
2
+ =· with α = . (7)
∂x k α ∂t ρc
359
V. PARMA
Thermal diffusivity α indicates how fast a thermal perturbation develops along the beam. A few cases
of special interest are detailed below.
3.1.1 Steady-state beam with no heating

Of particular interest is the steady-state case (𝜕𝑇/𝜕𝑡 = 𝑜) in which there is no internal heat
deposition (𝑞̇ = 0). This is, for example, the case for any solid element having its extremities fixed in
temperature (T0 and TL). Equation (6) is then simply re-written:
∂  ∂T 
 k (T ) .  = 0 . (8)
x
∂ ∂x
We can distinguish two cases of interest:

(i) k is constant. By integrating and imposing the boundary conditions at the extremities, we
find a linear temperature profile along the beam, given by:
T=
(T − T )
T0 + L 0 ⋅ x (9)
L
and a constant heat flux along the beam, given by:

˙
A
=q k . (TL − T0 ) ; (10)
L
(ii) k is linear with T (in the example of impure metals, like steels and Al alloys). Let us
assume k = T/a. By integrating and imposing the boundary conditions at the tips, we find a
quadratic temperature profile along the beam, given by:
TL2 − T02
T=T02 + ⋅x (11)
L
and a constant heat flux along the beam, given by:
=q
1
A/ L⋅
TL2 − T02 ( ) . (12)
2 a
These two solutions are graphically shown in Fig. 8.
Fig. 8: Temperature profiles for solutions (i) and (ii)
3.1.2 Steady-state beam with uniform heat deposition

Another relevant case is that of a structure that is thermally connected to a heat sink at a given
temperature, and which is subject to uniformly distributed heating (Fig. 9). This is the case, for
360
C RYOSTAT D ESIGN
example, of a thermal shield that is actively cooled by a cryogenic line at a fixed temperature at one
point, and on which radiation heat is uniformly distributed and transferred through solid conduction
from the hottest point to the cooling point.
Fig. 9: Uniformly heated beam, cooled at one tip
Let L be the beam length, t its thickness, and w its width (perpendicular to the page). The beam is
cooled at T0 in x = 0. The uniform heat deposition along the beam is q (W m−2).
Equation (6) can be re-written as:
d  dT  q
 k (T ) ⋅ + =0. (13)
dx  dx  t
Considering k to be constant (which is usually the case for thermal shields, as they should be designed
to be quasi-isothermal), the equation can be integrated to yield the following temperature profile and
heat flux along x:
q q⋅L
T ( x ) = T0 − ⋅ x2 + ⋅x , (14)
2 ⋅ k ⋅T k ⋅T
q ⋅ w( x − L) .
q ( x) = (15)
From Eq. (14), we can deduce the maximum ΔT along the beam, which may be the design objective
achieved by providing a minimum thickness t. As previously mentioned, thermal shields are normally
designed to be quasi-isothermal (typically within 5–10 K). So, by resolving t as a function of ΔTmax we
obtain the practical design formula to choose the minimum thickness:
q ⋅ L2
t≥ ⋅ (16)
2k ⋅ Δ ⋅ Tmax
3.2 Residual gas conduction

Heat transfer by convection between surfaces at different temperature inside cryostats is negligible at
low pressure, but residual gas conduction remains an important contribution that depends on the level
of vacuum, the gas species, and the geometry and temperatures involved.
During normal operation, the large cryogenic cryo-pumping capacity of the cold surfaces of the
internal components (essentially the surfaces of cryogen reservoirs and thermal shields) keeps pressure
low enough (below 10−3 Pa) to provide an adequate insulation vacuum; vacuum pumping groups can
also be stopped without observing any increase in pressure. Residual gases inside the cryostat or air
leaks are condensed on the cold surfaces. Cryo-condensation depends on the gas species and on the
temperature of the cold surfaces. More precisely, the condensation process happens when the gas
pressure is higher than its vapour pressure at the temperature of the condensation surface. Figure 11
shows the vapour pressure curves for gases detected inside the LHC cryostats, which are
representative of an industrial construction level of surface cleanliness (not Ultra High Vacuum,
cleaned) and making use of MLI (mostly contaminated with water but also other types of hydrocarbon
species). One can observe that at 4.2 K (liquid helium temperature), most of the gas species (with the
exception of H2 and He) can be cryo-condensed on the cold surfaces and the residual pressure always
remains below 10−3 Pa. In the presence of He, having a vapour pressure well above the 10−3 Pa
361
V. PARMA
threshold even at 4.2 K or lower temperature, other pumping means must be employed to keep a good
insulation vacuum (turbomolecular pumps, cryo-adsorbing materials, cryo-pumping in a molecular
regime, etc.).
Fig. 10: Vapour pressures of common gases in the LHC insulation vacuum
Let us now consider heat transfer from residual gas conduction in the case of a degraded vacuum in
cryostats. We are especially interested in the case of helium because practically all superconducting
accelerator devices (magnets and accelerating cavities) are cooled in cryogenic helium, hence with the
risk of helium leaks inside the cryostats.
Heat transmission between two surfaces separated by a gas follows two different regimes
depending on the ratio between the mean free path of gas molecules λ and the distance L between the
two surfaces. When λ << L, the viscous regime applies, and heat transmission is described in terms of
thermal conduction k, and does not depend on pressure. In this case heat flux is inversely proportional
to L as in the case of solid conduction. Decreasing the residual gas pressure, the molecular regime is
reached as λ >> L. The molecules travel from the warm to the cold surface and heat transfer becomes
proportional to residual gas pressure and is independent of wall distance L.
From gas kinetic theory, λ is given by:
η T
=λ 115 ⋅ (17)
p M
with viscosity η in Pa.s, pressure p in Pa, temperature T in K, and molecular weight M in g mol−1.
In the molecular regime, residual gas conduction, from a warm surface A2 at temperature T2, to a
cold surface A1 at temperature T1 obeys Kennard’s law:
𝑄̇ = 𝐴1 ∙ 𝛼(𝑇) ∙ Ω ∙ 𝑝 ∙ (𝑇2 − 𝑇1 ) (18)
with pressure p in Pa, temperatures in K, Ω is a coefficient depending on the gas species

(2.13 W m−2.Pa.K for helium), and α(T) is an accommodation coefficient that depends on the gas
species, temperatures, and geometry of the surfaces. For simple geometries, as in the case of parallel
or concentric walls, it is given by the expression:
α1 ⋅ α 2
α= (19)
A
α 2 + α1 (1 − α 2 ) ⋅ 1
A2
362
C RYOSTAT D ESIGN
where α1 and α2 , for helium and air at various temperatures, is given in Table 4:
Table 4: Accommodation coefficients for helium and air
T (K) Helium Air

300 0.3 0.8
80 0.4 1
20 0.6 -
4 1 -
We can note that residual gas conduction is proportional to pressure and to the temperature difference
between walls.
3.3 Radiation
Thermal radiation is by far the most important contribution to the thermal budgets for cryostats, and
deserves a special attention.
The surface of a body at a given temperature emits and absorbs electromagnetic radiation. When
receiving electromagnetic radiation, a body absorbs a fraction α (absorptivity), reflects a fraction β
(reflectivity), and transmits a fraction τ (transmissivity), with energy conservation imposing the
condition α + β + τ = 1. When τ = 0, the body is said to be opaque, meaning that all energy is either
absorbed or reflected. Also, when β = 0, and from energy conservation α = 1, then the body is said to
be black. In this case, the body absorbs all energy.
In general, the opaque body is a good approximation in most of our applications in cryostats.
Conversely, the black body approximation almost never applies but is often used as a convenient
simplification for calculating the maximum emitting radiation power of a body, and for first-order heat
load calculations. But there is one case in cryostats where a surface behaves like a black body: an
orifice in a thermal shield, through which radiation shines, and through multiple reflections is
gradually absorbed by the internal surfaces (Fig. 11). The cryostat engineer has to be very careful,
when designing thermal shields, in avoiding gaps and slots, which may be detrimental to the thermal
performance. When gaps are unavoidable, typically when thermal contraction compensation gaps must
be introduced, practical solutions can be adopted to avoid light shining through: for example the use of
MLI pads covering slots; or if MLI cannot be used (in UHV applications) gaps should be surrounded
by traps of high-absorptivity materials (special coatings, for example) to absorb light on the thermal
shield wall, and reduce multi-path reflection on the inside.
Fig. 11: Black body approximation of an aperture in a cavity
A black body emits a radiation flux with an emissive power, which is a function of the wavelength and
depends on the temperature of the body, as illustrated in Fig. 12. It is interesting to note that the
363
V. PARMA
maximum emissive power occurs at a wavelength inversely proportional to temperature, according to

Wien’s law:
2898
𝜆max = 𝑇
(μm K –1 ) (20)
and that at temperatures of practical interest in cryostats (from 300 K to vanishingly low
temperatures), the maximum occurs at wavelengths in the far-infrared region.
Fig. 12: Emissive power of a black body
By integrating over the whole wavelength spectrum, we can calculate the total emissive power:
∞
Eb(T )= ∫ Eb , λ dλ = σ ⋅ T 4 (W m −2 ) (21)
0
with σ being the Stefan-Boltzmann’s constant:

σ 5.67 ×10-8 (W m 2 K 4 )
= (22)
providing the important result that the total emissive power is therefore proportional to the fourth
power of temperature. From simple calculation, we can conclude that a black body at room
temperature (293 K) emits about 420 W m−2. Considering that the vacuum vessel of a cryostat is at
room temperature, and assuming that a thermal shield has a 1-cm2 gap, results in a black body heat
load through the gap of ~10 mW. The value may appear small, but when compared to the extremely
high electrical power cost of heat extraction at cryogenic temperatures (about 1 kW W−1 from 1.8–
300 K), 10 mW deposited at 1.9 K requires about 10 W of cooling power, and is therefore three orders
of magnitude larger.
Let’s now consider heat exchange between two black bodies at different temperature and facing
each other with a given orientation, as illustrated in Fig. 13.
Fig. 13: Scheme of two surfaces exchanging radiation heat
364
C RYOSTAT D ESIGN
The heat exchange balance between the two can be computed as the difference between radiation
emitted from the first body and absorbed by the second, and the radiation emitted by the second body
and absorbed by the first. This is given by:
q1 − 2 σ (T 14 − T 2 4 ) A1F 12
= (23)
where the geometrical view factor F 12 is the fraction of the total radiation leaving the first body,
which is intercepted and absorbed by the second and which depends upon the relative orientation of
the two surfaces. It can be demonstrated that reciprocity applies, and that A1F 12 = A2 F 21 .
View factors are quite tedious to calculate, but values can be found in the literature. Figure 14
gives view factors for selected geometries.
Fig. 14: View factors for given geometries
365
V. PARMA
Real surfaces do not behave as black bodies but rather as grey-diffuse bodies. Only a fraction of the
black body radiation is emitted (i.e. grey-body), and the intensity of emission is uniform in all
directions, (i.e. diffuse-emitting body).
By introducing the total hemispheric emissivity, which defines this fraction of the emitted
radiation with respect to that of a black body:
E (T )
ε (T )
= ≤ 1. (24)
Eb(T )
Equations (21) and (23) become, respectively:

E (T ) = ε ⋅ σ ⋅ T 4 , (25)
ε ⋅ σ (T 14 − T 2 4 ) A1F 12 .
q1 − 2 = (26)
For real materials, the grey-diffuse model is a satisfactory approximation. Emissivity depends on the
material but also on the surface finish and cleanliness. Clean and well-polished metallic surfaces have
small emissivity, whereas non-metallic surfaces have higher emissivity. For this reason thermal shields
in cryostats are normally made of metallic surfaces (copper or aluminium), and with a clean and
reflective surface finish. Table 5 gives hemispherical emissivity values for a selection of materials.
Table 5: Hemispheric emissivity for a selection of materials
Temperature
(K)
4 20 80 300
Copper mechanically polished 0.02 0.06 0.1
Copper black oxidised 0.8
Gold 0.01 0.02
Silver 0.005 0.01 0.02
Aluminium electropolished 0.04 0.08 0.15
Aluminium mechanically polished 0.06 0.1 0.2
Aluminium with 7 μm oxide 0.75
Magnesium 0.07
Chromium 0.08 0.08
Nickel 0.022 0.04
Rhodium 0.08
Lead 0.012 0.036 0.05
Tin 0.012 0.013 0.05
Zinc 0.026 0.05
Brass, polished 0.018 0.029 0.035
Stainless steel, 18-8 0.2 0.12 0.2
Glass 0.94
Ice 0.96
Oil paints, any colour 0.92–0.96
Silver plate on copper 0.013 0.017
Aluminium film 400A on Mylar 0.009 0.025
Aluminium 200A on Mylar 0.015 0.035
Nickel coating on copper 0.027 0.033
It should be noted that our appreciation of the visual appearance of the surface finish is limited
to the visible range, and this may be misleading. For instance, a reflective metallic surface, which has
been coated by a transparent resin, still appears reflective. But since the resin behaves like a black
body at room temperatures or lower, it will absorb radiation rather than reflect it. This is the case of
366
C RYOSTAT D ESIGN
some aluminized tapes available on the market and commercialized for their low-emissivity properties
of practical use in cryostats (for shielding parts, or joining MLI blankets), and in which the aluminium
reflective layer is deposited onto a transparent polyester tape, therefore behaving like a black body
despite the shiny appearance.
Emissivity varies with temperature. For metals at cryogenic temperatures emissivity reduces
almost proportionally with temperature, which enhances the low-emissivity properties of cryogenic
cooled thermal shields in cryostats.
Of particular interest to cryostat design are the geometries with two enclosed envelopes, in
particular with cylinders, and parallel flat plates.
For enclosed cylinders, with A1 and A2 being the inner and outer surfaces respectively, and kept
at temperatures T1 < T2, from energy balance and making use of Eq. (26), the power exchanged is
given by:
σ A1(T 2 4 − T 14 )
q1 − 2 = F 12 . (27)
1 A1  1 
+  − 1
ε 1 A2  ε 2 
This equation provides interesting design hints. Assuming A2 is the vacuum vessel of a cryostat and A1
is a given internal cryogenic component (or a thermal shield), reducing radiation heat loads to A1 can
be obtained by reducing the size of the vessel A2 (the radiation source), as well as reducing both ε1 and
ε2 (in particular ε1, which has the largest effect). For this reason thermal shields should be made with
low-emissivity surfaces (aluminium, copper or Ni plating), whereas the vacuum vessel, dominated by
other constraints (construction materials and fabrication technology) is normally not prepared for low
emissivity.
For parallel flat plates of area A, which is also a convenient approximation for surfaces of large
radius placed together at a close distance, the power exchange, when T2 < T1, is given by:
. (28)
If we now insert a third intermediate floating screen between the two flat surfaces, and we assume that
ε1 = ε2 = ε, a good approximation when materials are similar and their temperatures are not too
different, the heat exchange becomes:
σ (T 14 − T 2 4 )
q1 − 2 = , (29)
2
2( − 1)
ε
which is therefore reduced to half of the thermal radiation between the two parallel surfaces.
This is a useful hint on how radiation heat loads in a cryostat can be reduced by inserting one
(or more) floating reflective screens between the vacuum vessel and the thermal shield. This
introduces the topics of thermal shielding and MLI, presented in the following sections.
3.4 Thermal shielding

Introducing one intermediate floating shield in a cryostat reduces by half the heat load to the cold
surface. If a dedicated cryogenic circuit actively cools the shield, its effectiveness can be improved by
two mechanisms. First, the emissivity of the shield decreases at lower temperatures, therefore the heat
exchanged with the vacuum vessel is reduced as well as the heat radiation between the thermal shield
367
V. PARMA
and the cold surface. Secondly, heat extraction by active cooling of the thermal shield at a higher
temperature than that of the cold surface results in a lower cryogenic cooling cost (as explained in
Section 4).
Thermal shielding cooled at a temperature in the range of 50–80 K is a universal practice. In the
LHC cryostats (Fig. 5), the thermal shield is made of aluminium, and is supported on the magnet
support posts, for which it also provides heat interception. The shield is composed of an aluminium
bottom tray extrusion, integrating a cooling line at 50–80 K, which provides mechanical stiffness to
15-m long assemblies, and is closed on its top part by a rolled and welded 2.5-mm thin aluminium
sheet. Given the very large quantity of material required, aluminium was preferred to copper for its
lower cost while still offering a reasonably good thermal conductivity.
Thermal radiation protection of the thermal shields is enhanced by the use of MLI, discussed in
the next section.
3.5 Multilayer insulation protection

MLI is based on the principle of multiple radiation reflection obtained by inserting reflective layers
between the warmer radiating surface (typically the vacuum vessel) and the colder surfaces (thermal
shield or cold surface of the internal device).
The result of Eq. (29) can be generalized to N reflective layers, obtaining a reduction of
radiation by a factor (N + 1). In practice, reflective layers are packed in blankets, and thermal contact
between adjacent layers would be inevitable; for this reason, the reflective layers are interleaved with
insulating spacers, as schematically shown in Fig. 15, limiting thermal conduction.
Fig. 15: MLI structure, showing reflective layers interleaved with insulating spacers
Modelling the thermal heat exchange through MLI can be quite complex, material (reflective and
insulating layers) dependent, as well as strongly influenced by the application of the blanket in the
cryostats. A simplified engineering model considers two main contributions, one accounting for
radiation, proportional to the difference of temperatures at the fourth power, and a second accounting
for the residual solid conduction across the blanket thickness proportional to the temperature
difference between the warm and the cold surface and to the average temperature:
 β  α T1 + T2
qMLI=  ⋅ (T14 − T24 )  + ⋅ ⋅ (T1 − T2 ) (30)
 N +1  N +1 2
368
C RYOSTAT D ESIGN
where β and α are corrective factors that should be obtained experimentally. Values obtained for the
MLI system of the thermal shield of the LHC cryostats are β = 3.741 10−9 and α = 1.401 10−4.
MLI in accelerator cryostats generally consists of aluminised thin polyethylene films as
reflective surfaces, and paper, glass-fibre, or polyester insulating nets as insulating spacers.
The efficiency of MLI also depends on the practical aspects of how it is mounted. The packing
density of the layers plays an important effect. Figure 16 shows schematically that with increasing
packing density, the solid conduction contribution increases. There is a somewhat optimal packing
density, which can be found between 15 and 25 layers per centimetre.
Fig. 16: Total heat flow as a function of MLI packing density
MLI efficiency also depends on practical implementation rules. The ideal packing density
should be preserved after installation of the blankets. When mounted in horizontal accelerator
cryostats, MLIs tend to be compressed by their weight on the top part of the thermal shields, whereas
it tends to be looser at the bottom. A good rule is to check that an acceptably correct packing remains
after installation. Also, when preassembled MLI blankets are to be mounted around circular
geometries, one should include in the design the larger circumferential dimension of the outermost
layers with respect to innermost ones. Differential thermal contractions between the MLI and its
support should be also considered to preserve the correct packing when cold. Finally, a typical mistake
is to stack too many layers in a reduced space between warm and cold surfaces, ending up with
unwanted thermal conduction shorts. A good rule is to leave a minimum gap of at least a few
centimetres between the warm surface and the outermost MLI layer.
The choice of the ideal number of MLI layers is not a straightforward one, and is very much
dependent on the materials employed (film material and thickness, single or double-aluminisation,
aluminisation thickness, type of spacers, etc.); a general tendency is to introduce as many layers as can
possibly fit in the available space, sometimes resulting in thermal contact shorts, but also in an
increase of material and assembly costs.
Extensive testing was carried out at CERN on MLI types and blanket assemblies for the LHC
magnet cryostats. An optimal and cost-effective solution was chosen for the MLI blankets of the
thermal shield (Fig. 17): 30 reflective layers (double-aluminised polyethylene terephthalate film,
coated with a minimum of 400 Å of aluminium on each side) interleaved with polyester net. An
additional ten-layer MLI blanket is mounted on the cold mass, essentially to reduce heat loads from
residual helium gas conduction in case of degraded vacuum from helium leaks, as well as thermally
insulating the cold surfaces in case of an accidental air venting of the cryostat.
MLI installation precautions were carefully studied for the LHC cryostats, also including
industrial production and assembly considerations to be compatible with the large series (about 2000
dipole and quadrupole cryostats).
369
V. PARMA
MLI blankets were preassembled in industry, and assembled in the magnet cryostats at CERN
(Fig. 17). An important investment in tooling for series production of blankets allowed the
manufacturer to reduce manpower costs.
(a) (b) (c)
Fig. 17: (a) MLI stack in a blanket; (b) stitching table for industrial production; (c) MLI blanket on a dipole cold
mass and thermal shield.
The thermal performance of the LHC cryostat system was experimentally measured at various
stages of the design and construction of the machine, but final thermal performance measurements
were made during commissioning of the machine.
From these results the performance of the MLI system could be deduced, yielding the following
figures:
• Thermal efficiency of 30 layers of LHC MLI between 300 K and 50 K: ~1 W m−2;
• Thermal efficiency of 10 layers of LHC MLI between 50 K and 1.9 K: ~50 mW m−2.
The literature reports laboratory tests providing lower values than these, but the interest of the above
figures is that they include all the inefficiencies resulting from the real implementation of MLI
protection on the large-scale accelerator.
To conclude this first part dedicated to thermal studies of the cryostats, the reader can find in
Appendix A a case study based on the LHC cryostat with a numerical application of the formulae
presented on radiation protection, thermal shielding, MLI, and residual gas conduction. Appendix B
presents the experimental validation of the thermal performance of the LHC cryostats during the
commissioning of the machine.
4 Cryogenics considerations
Helium is the cryogenic fluid of choice in accelerator technology because its thermo-physical
properties make it the only one suitable to operate superconducting devices below 10 K. When
operating below the lambda point, at temperatures below 2.17 K, helium offers the possibility of
enhancing the field performance of magnets made in Nb-Ti, while superconducting RF acceleration
cavities can profit from a reduction of the Bardeen, Cooper and Schrieffer (BCS) contribution to the
surface resistance with a reduction in dynamic losses, especially at high frequency.
Accelerator magnets are usually operated with sub-cooled liquid (i.e. pressurized above its
saturation point) ensuring a good liquid coverage and penetration around the superconducting coils.
Operating in pressurised liquid is also beneficial in reducing the risk of electrical insulation breakdown
in gaseous helium and avoids contamination of helium in the case of leaks. Superconducting Radio
Frequency (SRF) cavities are normally operated in saturated helium, where pool-boiling cooling is
preferred for technical simplicity of the cooling schemes.
Figure 18 illustrates on a phase diagram the helium operating point of a number of state-of-the-
art linear and circular superconducting accelerators.
370
C RYOSTAT D ESIGN
Fig. 18: He phase diagram and operation point of several superconducting machines
Cooling at higher temperatures than the operating point of the superconducting device is needed
in cryostats for thermal shields and intermediate temperature heat interception in cryostat components.
A temperature of about 80 K is often chosen, with the advantage that liquid nitrogen can be employed,
a far more cost-effective cryogenic fluid.
From Table 6, comparing the main thermo-physical properties of helium and nitrogen, we note
that the latent heat of evaporation of nitrogen is one order of magnitude larger than that of helium,
meaning that for a given heat load the boil-off mass flow is ten times lower than for helium, making it
a more effective coolant when large enthalpies have to be extracted, typically when cooling down
massive equipment from ambient temperature down to 80 K. This is the case for the initial cool-down
phase of the LHC magnets, where the helium circulated in the magnets is pre-cooled by liquid
nitrogen. On the contrary, helium vapour has seven times higher enthalpy between its boiling
temperature and 300 K (1550 kJ kg−1), meaning that its vapours still have a large cooling capacity that
can be effectively used for cooling components with a thermal conduction path from room temperature
to 4.2 K, as explained in the following sections.
Table 6: Main thermo-physical properties of helium and nitrogen
Property Units 4He N2

Boiling T (at 1 atm) K 4.2 77.3
Critical temperature K 5.2 126.1
Critical pressure 105 Pa 2.23 33.1
Latent heat of evaporation (at 1 atm) kJ kg−1 21 199
−1
Enthalpy between T boiling and 300 K kJ kg 1550 233
−3
Liquid density (boiling at 1 atm) kg m 125 810
−3
Saturated vapour density (at 1 atm) kg m 17 4.5
Gas density (at 1 atm, 273.15 K) kg m−3 0.18 1.25
Liquid viscosity (at boiling T) μPa s 20 17
371
V. PARMA
4.1 Efficiencies of helium refrigeration and liquefaction

Heat loads are extracted from the cryostat either through boil-off of the liquid cryogens and pumping
of vapours, or through forced circulation of cryogens (typically in their supercritical state) in cooling
circuits with heat exchangers. In the first case, isothermal cooling can be maintained by ensuring
constant pressure by vapour pumping over the saturated bath. This is normally adopted to keep the
constant operating temperature of the superconducting device. The second case provides non-
isothermal cooling and is generally the way that thermal shields and heat intercepts are cooled; a
temperature increase along the cooling line is inevitable but the cooling circuit can be designed to keep
it conveniently small.
In both cases the heat extracted from the cryostats is transported, through a mass flow of
cryogens, to the refrigerator, where heat is eventually rejected to the ambient. The cryogen is then
recompressed, cooled, and circulated back to the cryostat and its utilities.
In the ideal case, refrigeration requires a minimum of work, according to Carnot’s principle, to
extract a given heat from a temperature Tc to a heat sink at Tw; in the T-S diagram this corresponds to
the area contained between two adiabatic and two isothermal transformations of a reversible cycle
(Fig. 19.)
Fig. 19: (a) Refrigeration between a cold source Tc and a warm one Tw; (b) the ideal Carnot cycle
A corresponding coefficient of performance (COP) can be defined as the ratio between the
extracted heat from the cold source and the associated extraction work:
Qc
COP = . (31)
W
In the ideal case of the Carnot cycle, making use of the first and second principles of thermodynamics,
the COP attains its maximum value:
Tc
COPmax = (32)
Tw − Tc
372
C RYOSTAT D ESIGN
simply depending on temperatures or, by combining Eqs. (31) and (32), calculate the minimum work
needed to extract a given heat Qc:
Tw − Tc
Wmin
= Qc ⋅ . (33)
Tc
Refrigerators eventually dump heat to ambient temperature, so by taking Tw = 293 K, we can easily
calculate the minimum specific heat extraction work, also equivalent to the specific heat load
extraction power in W/W, for a refrigerator working between the liquid temperature of commonly
used cryogens and room temperature (Table 7).
Table 7: Minimum refrigeration power at different liquid temperatures
Fluid Temperature Wmin Qc−1

(K) (W/W)
Liquid N2 77 2.8
Liquid H2 20.4 13.4
Liquid He 4.2 68.4
Liquid He, (operating T of LHC magnets) 1.8 161.8
Extracting heat at 2 K rather than at 80 K implies a cost increase by a factor of more than 50.
This is the reason why heat loads to the lowest temperature in a cryostat (the operating temperature of
the superconducting device) have to be minimized, and why heat interception is thermodynamically
more convenient at higher temperatures, which explains why thermal shields are cooled at higher
temperatures, in the 50–80 K range.
In the real case, refrigerators have an efficiency even lower than the Carnot efficiency due to
irreversibilities in the machine. Technical efficiency of refrigerators is only a fraction of the Carnot
factor, approaching 0.3 for modern large refrigeration systems at 4.2 K. Therefore the net refrigeration
efficiency at 4.2 K is only 0.0028; in other words cooling 1 W at 4.2 K costs an extraction power of
about 360 W. In the case of the LHC magnets, which are operated at 1.8 K, where the efficiency is
about 0.15 of Carnot, 1 W of heat load costs about 1 kW of extraction power (Fig. 20).
Fig. 20: COP as a function of cooling temperature (K) and percentage Carnot efficiencies
373
V. PARMA
In cryogenic plants, a distinction can be made between operation in pure refrigeration, where
isothermal cooling of the utilities load is provided, and pure liquefaction, in which liquid is produced
and used for non-isothermal cooling of the utilities load, as illustrated in Fig. 21. Accelerator
cryogenic plants generally provide a combination between these operating modes, depending on the
specific needs, with some possibility of trading between the two.
(a) (b)
Fig. 21: (a) Simplified refrigeration versus (b) liquefaction cycles
When operating in liquefaction mode, the 4.5 K liquid produced can be used for isothermal boil-
off cooling making use of its latent heat of evaporation, followed by non-isothermal cooling up to
room temperature, making use of the large remaining cooling capacity of its vapours. Room
temperature gaseous helium is then transferred back to the cryogenic plant for recompression and re-
liquefaction. As schematically shown in the T–S plots in Fig. 21, the cooling capacity in the
liquefaction mode is about 75 times higher than in refrigeration.
Let us now compare the thermodynamic efficiency of liquefaction and refrigeration at 4.2 K. By
combining the first and second principles of thermodynamics, and introducing entropy S, we can
express the work of refrigeration as:
W= Tw ⋅ ∆Sc − Qc . (33)
For liquefaction, we can split Eq. (33) into two contributions, the first one for pre-cooling helium from
293 K down to 4.5 K, and the second one for condensation; furthermore, considering that heat
exchanged at constant pressure is equivalent to enthalpy H, we can write:
Wliq= Tw ⋅ ∆S precool − Qprecool + Tw ⋅ ∆Sconden. − Qconden.
. (34)
= Tw ⋅ ∆Sprecool − ∆H precool + Tw ⋅ ∆Sconden. − ∆H conden.
From tables of thermodynamic properties of helium, we can extract the entropy and enthalpy figures
and calculate the work of liquefaction, which turns out to be 6.6 kW per 1 g s−1 of liquefied gas.
By taking the Carnot minimum specific heat refrigeration work of about 68 W/W, we can now
conclude that 1 g s−1 of helium liquefaction is about equivalent to 100 W iso-thermal refrigeration at
4.5 K. We now have the ingredients for comparing the thermal efficiency of vapour cooling and
isothermal cooling of cryostat heat intercepts, presented in the following sections.
374
C RYOSTAT D ESIGN
5 Heat intercepts
In the previous section we have seen that the cost of refrigeration increases dramatically when
working at temperatures close to absolute zero. For this reason, the thermal design of cryostats,
especially for systems operating with superfluid helium, should aim at containing heat loads reaching
the lowest temperature level that may be achieved by thermal shielding and by heat interception.
Thermal shielding was extensively covered in the previous sections, and the next sections are
dedicated to heat interception.
5.1 Conduction heat intercepts

Solid conduction in cryostat components from room temperature to the lowest temperatures can be
reduced by thermalizing at higher temperatures at intermediate and adequately chosen locations. The
available intermediate temperature levels often depend on the architecture of the machine and its
cryogenic system, but the use of a thermal shield cooled at a temperature in the range of 50–80 K is a
universal practice, so this temperature level is also available for heat intercepts. In the LHC, a second
cooling circuit at a temperature level of 5–20 K is also used to cool the beam screens inside the beam
tubes; the same circuit is also used for a second level of heat interception on most components, in
particular on the magnet support posts. See Fig. 22.
Fig. 22: Two heat intercepts at intermediate temperatures
By minimizing the following cooling cost function f (L1, L2), sum of three conductive
contributions weighted with three factors C1, C2, and C3, which depend on the cost of cryogenic
cooling at the three temperature levels, we can find the optimal locations L1, L2 of the heat intercepts:
 
Tc
A A A
80K 8K
min  f ( L1, L 2) = C1⋅

 L1 ∫ k (T )dT + C 2 ⋅
Tw
L 2 − L1 ∫ k (T )dT + C 3 ⋅
80K
L − L2 ∫ k (T )dT  . (35)
8K
For the LHC support posts, with Tc = 2 K, the optimal positions of two heat intercepts, at 5 K and
75 K, were chosen to minimize the cost function with C1 = 16 W/W, C2 = 220 W/W, and
C3 = 990 W/W as cost factors at 75 K, 5 K, and 1.8 K, respectively.
Table 8 compares the cryogenic cost for heat interception under various configurations. The
two-heat intercepts solution chosen offers a cryogenic cooling cost saving of more than one order of
magnitude as compared to that without heat intercepts.
Table 8: Heat loads with and without heat intercepts and total cost of heat extraction per LHC support post
Number of heat intercepts (in optimal positions) Q1.8K Q5K Q75K Qelec
(W) (W) (W) (W)
No intercept 2.79 - - 2790
1 (at 75 K) 0.54 - 6.44 638
2 (at 75 K and at 5 K) 0.047 0.42 7.1 252
375
V. PARMA
Figure 23 illustrates an LHC support post in glass-fibre reinforced epoxy (GFRE) composite
material with two aluminium heat intercepts positioned in optimal positions.
Fig. 23: LHC support post in GFRE, with two heat intercepts
The heat intercepts are 10-mm thick aluminium plates close-fitted and glued to the external wall
of the composite column. The heat intercepts shrink-fit onto the composite column during cool-down
ensuring an improved heat exchange due to contact pressure.
Active cooling of the bottom intercept at 50–65 K is ensured by an all-welded aluminium
connection to the thermal shield, while the top intercept is a cast aluminium plate integrating a
stainless steel cryogenic line at 5–10 K.
5.2 Vapour cooling

The large enthalpy stored in helium gases evaporating from a bath can be usefully employed for
cooling supports, cryostat necks, RF couplers and current leads. Taking a support of length L and
cross-section A (Fig. 24), considering perfect exchange with the escaping gas (i.e., assuming identical
temperature of wall and gas), the heat balance equation gives:
dT • •
k (T ) ⋅ A ⋅ = Q + m Cp ⋅ (T − Tl ) (36)
dx
•
where Q is the residual heat flow reaching the liquid bath at temperature Tl.
Fig. 24: Vapour cooled support with perfect cooling
376
C RYOSTAT D ESIGN
Assuming that the vapour boil-off is entirely due to this heat load (so-called self-sustained
cooling), and introducing the latent heat of evaporation Lv, we can state:
• •
Q = m⋅ Lv . (37)
Equation (36) yields:

•
A Tw k (T )
Q= ⋅∫ ⋅ dT . (38)
L T l
1+
( T − Tl ) ⋅ Cp
Lv
In this form, the expression of the heat load is equivalent to that of solid conduction but for the
denominator in the integral, which being larger than one results in an attenuation factor with respect to
the conduction case.
Attenuation factors for liquid helium cooling can be in the range of 15 to 30 as reported in
Table 9 for a selection of technical materials.
Table 9: Thermal conduction integrals and corresponding reduced thermal conductivity integral in self-sustained
helium cooling from 4.2–293 K.
Material Thermal conductivity integral Reduced thermal conductivity integral

(W cm−1) (W cm−1)
OFHC copper 1520 110
Aluminium 1100 728 40
AISI 300 stainless steel 31 0.92
Vapour cooling is widely used for electrical conductors feeding current from room temperature to
superconducting devices (Fig. 25). When designing current leads the aim is to minimize the heat load
introduced when transmitting a given current into the cryostat. The heat load in this case has a second
source in addition to pure conduction, which is the ohmic loss in the lead. The goal of the design is
therefore to minimize both heat conduction and electrical resistance.
Fig. 25: Vapour-cooled electrical conductor
The thermal balance of an element of the conductor dx (Fig. 25) can be written as:
2
d  dT  • dT I
 k (T ) ⋅ A ⋅  − f ⋅ m⋅ Cp (T ) ⋅ + ρ (T ) ⋅ = 0 . (39)
dx  dx  dx A
where I is the current in the conductor, ρ is its electrical resistivity, and where f is a cooling efficiency
factor that can vary between 0 and 1, where 0 corresponds to no cooling and 1 to perfect heat
exchange between gas and wall.
377
V. PARMA
For current lead materials that follow the Wiedemann–Franz law (most metals and alloys),
thermal conductivity and electrical resistivity are bound by the relation:
ρ (T ) ⋅ k (T ) =
L0 ⋅ T , (40)
which can be inserted into Eq. (39), and solved numerically to calculate the heat load per unit current
for varying values of f as plotted in Fig. 26. Perfectly cooled leads can reach the minimum value of
~ 1 W kA−1. It should be noted that this result is independent of the conductive material of the
conductor. An even better performance can be achieved by using materials that do not follow the
Wiedemann–Franz law, for example High-Temperature Superconductors (HTS), having close to zero
resistivity and which behave as relatively bad thermal conductors up to high temperatures.
Fig. 26: Heat load per unit current in vapour-cooled leads
Vapour cooling is also employed for RF power couplers. An illustrative example is the SPL
cryo-module, under development at CERN (Fig. 27). The external conductor of the coaxial RF coupler
is made out of a double-walled stainless steel tube with a copper-plated internal diameter of 100 mm; a
forced circulation of helium vapour from 5–293 K inside the double-walled envelope, regulated in
mass flow and output temperature, allows intercepting conduction of heat as well as resistive RF
power deposited in the copper-plated wall.
Fig. 27: General assembly view of the SPL short cryo-module with vapour-cooled RF couplers
A comparison between heat intercept and vapour-cooled solutions is given in Table 10, where
one can appreciate the efficiency of those solutions adopted (cases E and F).
378
C RYOSTAT D ESIGN
Table 10: Cooling efficiency comparison of heat intercepts and vapour cooled solutions for the SPL RF coupler
Case Q at 2 K Wel Q at 8 K Wel Q at 80 K Wel Vapour Q equivalent Wel Total

(W) (W) (W) (W) (W) (W) rate at 4.5 K (W) Wel
(W) (1 g s−1 = W) (W)
A No intercept 11.629 11512.71 11,513
B One 1.816 1,797.84 39.513 632.208 2,4301
optimized
and perfect
intercepts at
80 K
C Two 0.129 127.71 2.64 580.8 26.816 429.056 1,138
optimized
and perfect
intercepts at
80 K and 8
K
D 4.5 K self- 0.031 30.69 0.019 1.9 407 438
sustained
vapour
cooling
E Real case, 0.1 99 0.04 4 880 1,039
He vapour
cooling 4.5–
300 K
F Real case, 0.5 495 0.04 4 880 1,435
He vapour
cooling,
4.5–300 K,
RF power
on
G Real case, 22 21,780 0 0 0 21,780
no He
vapour
cooling, RF
power on
6 Mechanical and construction aspects
6.1 Choice of materials for cryostats

A cryostat is composed, as a minimum, of the following main components:
• Vacuum vessel;
• Cryogenic vessels containing superconductor devices and/or containing cryogens;
• Thermal shield;
• Supporting systems.
Vacuum and cryogenic vessels are sheet-metal constructions, with thicknesses typically in the 1–
15 mm range. Materials preferentially range from low-carbon construction steels to stainless steels,
though aluminium is also sometimes employed. Vacuum vessel materials operate at room temperature
but must be qualified to withstand sudden cool-down to about −70°C in case of accidental rupture of
379
V. PARMA
the insulation vacuum. Charpy qualification tests at low temperature have to demonstrate adequate
energy absorption.
Cryogenic vessels are generally made of austenitic stainless steels. The austenitic structure does
not undergo any ductile-to-brittle transition at cryogenic temperature. The 304-type (18Cr8Ni on a Fe
basis) is the most common grade in use. Its low-carbon steel version 304L (C ≤ 0.030%), is preferred
because of its enhanced corrosion resistance and ductility in welded structures. Austenite is non-
magnetic, but for special applications where a fully austenitic and stable structure is sought even at
cryogenic temperatures (high field quality in superconducting magnets, for instance), 316-type steels
and, in particular, 316LN should be used. The 316LN type is also commonly used in bellows
convolutions owing to its formability and ductility.
Thermal shields must be in high thermal conductivity material. Copper and aluminium are the
materials of choice, preferentially as alloys for their better mechanical properties and
manufacturability while still preserving good thermal conductivity. Pure coppers are preferred only for
demandingly high thermal conductivity where temperature homogeneity matters. Aluminium alloys
are by far the most used materials in thermal shields for accelerator cryostats. Series 6000 (Al-Mg-Si),
combines good manufacturability (can be extruded) and weldability, with excellent thermal
conductivity at 80 K (close to Al1100, pure aluminium).
Materials for supporting systems usually depend on their design. Column-type supports are
extensively used in accelerator cryostats, and are the choice for the LHC cryostats. Fibre-reinforced
plastic materials are employed because they are good thermal insulators, but their types and material
composition, as well as manufacturing processes, cover a wide range of mechanical and thermal
properties. Epoxy-based composites, in particular cryogenic-grade glass-fibre composite materials
G10 and G11 are widely employed. Injection-moulded plastics, reinforced with short fibres (Ultem)
are also materials that have been used.
The LHC support posts are composed of Resin Transfer Moulding (RTM) monolithic 4 mm-
thick tubular columns in GFRE material. This material was chosen for its low thermal conductivity-to-
stiffness ratio while remaining a commercially available material to keep reasonable costs. Supporting
systems based on suspension rods can be made of stainless steel or titanium alloys.
The choice of the materials and the design of the components are primarily dictated by their
requirements, but one should not neglect their manufacturability and the capability of having them
produced in an industrial context, in particular for large series production, where cost containment
from economy of scale and commercial competition can be substantial.
Whenever possible, international standards and norms should be employed and referred to when
specifying materials, manufacturing processes, and quality control. The use of commercially available
materials, sometimes at the cost of relaxing excessively stringent initial requirements, can be an
advantageous strategy to contain costs and widen the number of potential suppliers to foster
competition. For instance for cryogenic fluid envelopes 316 LN stainless steel (particularly indicated
for superconducting magnet envelopes for preserving low magnetic susceptibility) can be suitably
replaced with 304L (a general-purpose grade) at half the material cost. Contrarily, when special
applications call for specific material composition and semi-finished material forms, one should make
a careful selection and tighten requirements as compared to standard industrial ones. In vacuum
flanges, for instance, which are normally machined out of forged bars or cold-rolled plates, non-
metallic inclusions in the base material can be stretched and align in directions along which leaks may
develop. For applications with stringent tightness requirements, suitable specifications should limit the
allowed inclusion content, and the material checked according to microscopic test methods. Specifying
the melting process, such as imposing remelting such as Vacuum Arc Remelting (VAR) or
ElectroSlag Remelting (ESR), are ways of ensuring the homogeneity of the material. Specifying that
380
C RYOSTAT D ESIGN
flanges are made through three-dimensional forging is also beneficial to avoid inclusions aligning in
preferential directions.
When dealing with vacuum and pressure vessels, including piping and thin-walled flexible
components (bellows and flexible hoses), the design engineer should not neglect safety aspects and
should refer to the relevant regulations in force in the country of construction and operation of the
equipment.
6.2 Thin shells under pressure

Circular tubes are the most commonly used geometries for internal pressure. Simple formulas can be
used in a preliminary design to calculate the main stresses in thin shells (where t < r/10, with thickness
t and radius r).
The circumferential stress in the presence of an internal pressure p is given by:
p⋅r
σ1 = (40)
t
whereas the axial force induced by pressure on the closed extremities introduce an axial stress that is
half the circumferential stress:
p⋅r
σ2 = (41)
2⋅t
and the stress perpendicular to the wall can be neglected: .
Considering the Tresca failure criterion, and introducing the admissible stress σa, which is a
property of the material, we can write:
p⋅r
σ a ≥| σ 1 − σ 3 | = , (42)
t
which can be used to calculate, for a given material, pressure, and radius, the minimum thickness t.
This formula holds for both internal and external pressure. Taking for example a vessel of a diameter
of 1 m, and with σa = 150 MPa (value for 304L, with a safety factor of 1.5 on Rp1.0), for a pressure of
1 bar (0.1 MPa), we can calculate t > 0.3 mm.
But when subject to external pressure, thin shells must be designed against radial buckling,
which is a non-linear phenomenon depending on initial geometrical imperfections and which can lead
to a sudden collapsing of the cylinder when a critical pressure is exceeded, as illustrated in Fig. 28.
Fig. 28: Radial buckling of a thin tube
381
V. PARMA
For a thin tube of infinite length, the critical pressure is given by:
3
E t
pcr = ⋅   (43)
4(1 − υ )  r 
2
with E and υ the Young’s modulus and Poisson ratio of the material, respectively.
As a rule of thumb, thickness and radius should satisfy the ratio t/r > 0.012. To cope with
geometrical imperfections, a minimum safety factor of 3 should be included, so the design rule
becomes:
t
≥ 3.7% . (44)
r
For instance, for the LHC cryostat vacuum vessel, with a radius of about 0.5 m, the thickness should
be at least 18.5 mm (in fact it could be reduced to 12 mm by introducing stiffening ribs).
6.3 Supporting systems in accelerator cryostats

Supporting systems in accelerator cryostats are designed to support and precisely position
superconducting devices. Magnets generally require alignment precisions within few tenths of a
millimetre, whereas SRF cavities are less demanding, but still require precisions within fractions of a
millimetre. Alignment must be maintained irrespective of thermo-mechanical movements induced by
the cool-down and warm-up transients, and drifts throughout the thermal cycles in the lifetime of an
accelerator.
Adjusting the position of each entire cryostat by means of moving them on their external
supporting system is the easiest way of aligning the superconducting device, provided the cryostat
internal supporting system keeps its positioning stability and is reproducible in time.
The LHC supporting system is based on this principle, as are most of the large accelerators (for
instance the Hadron Electron Ring Accelerator (HERA), Relativistic Heavy Ion Collider (RHIC), and
Spallation Neutron Source (SNS).
In some machines, the internal supporting systems can be adjusted from outside the cryostat; in
this case the positioning stability requirement is less stringent but a position monitoring system of the
superconducting device should be included if adjustment in cold operation is desired. For instance, the
HIE Isolde cryomodules, under construction at CERN, integrate a motorized adjustment of the internal
structure supporting the cavities and solenoids, and a CCD-based optical camera monitoring their
internal position through view ports (Fig. 29).
Fig. 29: HIE ISOLDE cryomodule with optical monitoring and external adjustment of cavities/solenoid
382
C RYOSTAT D ESIGN
Various supporting solutions can be employed; a non-exhaustive overview is given below. The
LHC is based on column-type compression posts, in GFRE material; RHIC makes use of a double
column-type compression support in Ultem™, a polyetherimide thermoplastic material (Fig. 30).
Fig. 30: RHIC dipole on column-type supports
A supporting system making use of vertical suspension and transversal adjustment tie rods was
adopted for the HERA dipoles (Fig. 31). The position of the magnet could be adjusted from outside
the cryostat, but only when warm.
Fig. 31: HERA dipole, with suspension tie rods
The X-Ray Free Electron Laser (XFEL) cryomodule also makes use of column-type posts, but
in a suspended configuration (Fig. 32).
Fig. 32: XFEL cryomodule, with suspended column-type supports
383
V. PARMA
The SNS cryomodules, based on the Continuous Electron Beam Accelerator Facility (CEBAF)
design, makes use of a frame to which cavities are suspended and adjusted vie tie rods, and which is
suspended inside the vacuum vessel (Fig. 33).
Fig. 33: SNS cryomodule with internal frame and tie rods suspension of cavities
Finally, the Superconducting Proton Linac (SPL) cryomodule, under development at CERN,
proposes the use of the double-walled tube of the external conductor of the RF power coupler as the
main mechanical support for the cavities, with the addition of a secondary supporting element between
adjacent cavities (Fig. 34).
Fig. 34: SPL cryomodule. RF coupler's double-walled tube as main cavity support
6.4 Welded and brazed assemblies

In the fabrication of cryostat envelopes and components, welding of steel and aluminium is a common
technology, which achieves two basic and distinct functions: providing structural reinforcement in an
assembly, or ensuring leak-tight joining. The latter creates a non-dismountable assembly and is chosen
for permanent assemblies. Nevertheless, welding can be a cost-effective solution, when compared to
flanged assemblies, for joints that only need occasional dismounting. For instance, the LHC magnet
interconnections, where the cryogenic piping contains the electrical connections, are all welded,
despite the occasional need for replacing magnets.
Leak-tight welding requires special precautions in preparation and execution, and must follow
very strict protocols and Quality Control (QC) checks. The preferred and most common technology is
Tungsten Inert Gas (TIG), though Metal Inert Gas Welding (MIG), is also in use (Fig. 35).
384
C RYOSTAT D ESIGN
(a) (b)
Fig. 35: Welding technologies. (a) TIG; (b) MIG
When TIG is employed in leak-tight welding of thin-walled structures, filler material is often
not necessary. In designing weld joints, one should follow prescribed geometries (Fig. 36 shows some
of these rules). Leak-tight welds should always be on the vacuum side of the component. Structural
reinforcing welds in the presence of a leak-tight weld shall always be made on the external side and
discontinuously, not trapping closed volumes between the two welds. One of the keys to successful
welding is an adequate protection by backing gas (argon or helium). Adequate protection is ensured by
the TIG torch on the welding side, but proper protection should also be provided on the back side.
Design of welded seams must be carefully chosen to avoid sources of impurities and defects,
especially for UHV applications. Non-destructive inspections like X-rays are often necessary on
structural welds, which should be designed to be accessible for inspection.
Fig. 36: Welding geometries for vacuum leak-tight welds
Welds are normalized and should follow welding procedure qualification, especially if the
welds have a structural function on pressure vessels (Table 11 reports the European Norms is force).
In this case the welds’ design should conform to the pressure vessel codes in use; welders are also
385
V. PARMA
required to be qualified for the execution of specific weld geometries. Qualification of welders is
carried out by accredited qualification bodies.
Table 11: European Norms relevant to steel and aluminium welds and welders
Steel Aluminium
Welding procedure EN ISO 15614-1:2004 EN ISO 15614-1:2005

approval Specification and qualification of welding Specification and qualification of
procedures for metallic materials— welding procedures for metallic
Welding procedure test—Arc and gas materials—Welding procedure test—Arc
welding of steels and arc welding of and gas welding of steels and arc welding
nickel and nickel alloys of aluminium and alloys
Qualification of EN 287-1:2004 Qualification test of EN ISO 9606-2:2004 Qualification test

welders welders—Fusion welding—Steels of welders—Fusion welding—
Aluminium and aluminium alloys
Qualification of EN 1418:1998 Welding personnel—Approval testing of welding operators for fusion
welding operators welding and resistance weld setters for fully mechanized and automatic welding of
metallic materials
Brazing is an assembly alternative for materials that cannot be welded, or for dissimilar
materials (copper on steel, for instance). It consists of heating a filler material just above its melting
point so that it can flow by capillary action between the two materials to be joined. The use of flux
agents is essential to avoid oxides from forming and degrading the quality of adhesion. A proper
cleaning after brazing is essential to eliminate flux residues, which can provoke corrosion and possibly
cause leaks in the base materials.
Brazing can be carried out in furnaces under controlled atmosphere, but this limits the size of
the parts to be assembled and is costly.
A list of available European Norms covering brazing is given below:
• EN 13134:2000 Brazing—Procedure approval;
• EN 13133:2000 Brazing—Brazer approval;
• EN 12797:2000 Brazing—Destructive tests of brazed joints;
• EN 12799:2000 Brazing—Non-destructive examination of brazed joints;
• EN ISO 18279:2003 Brazing—Imperfections in brazed joints.
6.5 Pressure vessels and construction standards

Since 2002, the Pressure European Directive 97/23/EC (PED) has become an obligatory legislation
throughout the EU. It applies to all equipment with an internal maximum allowable pressure ≥0.5 bar,
in which case the equipment is considered to be a pressure vessel. Vessels must be designed,
fabricated, and tested according to the essential requirements defined in the PED (design, safety
accessories, materials, manufacturing, testing, etc.), and must be subject to a qualification pressure test
under 1.43 times the maximum allowable pressure.
The category of the pressure vessel (from I to IV) depends on the stored energy, expressed as
pressure × volume in bar.l, according to the diagram in Fig. 37. The category establishes the
conformity assessment procedure to be adopted. Vessels not falling under the categories are to be
386
C RYOSTAT D ESIGN
designed in accordance with sound engineering practice. From category I and above, CE marking
becomes obligatory and the level of conformity assessment increasingly stringent (Table 12).
Fig. 37: Category of pressure vessels according to stored energy expressed in bar.l
Table 12: Category of pressure vessels and associated conformity assessments
Category Conf. assessment module Comment

SEP None The equipment must be designed
and manufactured in accordance
with sound engineering practice.
No CE marking and no
involvement of notified body.
I A CE marking with no notified body
involvement, self-certifying.
II A1 The notified body will perform
unexpected visits and monitor final
assessment.
III B1 + F The notified body is required to
approve the design, examine and
test the vessel.
IV G Even further involvement of the
notified body.
When designing pressure equipment, the engineer should make use of, as far as applicable, standards
and norms, as this approach gives presumption of conformity with the PED.
A list of useful norms for cryostat design and fabrication is given below:
• EN 13458-1:2002 Cryogenic vessels—Static vacuum insulated vessels—Part 1: Fundamental
requirements;
• EN 13458-2:2002 Cryogenic vessels—Static vacuum insulated vessels—Part 2: Design,
fabrication, inspection and testing + EN 13458-2:2002/AC:2006;
• EN 13458-3:2003 Cryogenic vessels—Static vacuum insulated vessels—Part 3: Operational
requirements + EN 13458-3:2003/A1:2005;
• EN 13445-1:2009 Unfired pressure vessels—Part 1: General;
387
V. PARMA
• EN 13445-2:2009 Unfired pressure vessels—Part 2: Materials;

• EN 13445-3:2009 Unfired pressure vessels – Part 3: Design;
• EN 13445-4:2009 Unfired pressure vessels – Part 4: Fabrication;
• EN 13445-5:2009 Unfired pressure vessels – Part 5: Inspection and testing;
• EN 13445-8:2009 Unfired pressure vessels – Part 8: Additional requirements for pressure
vessels of aluminium and aluminium alloys.
Other design codes such as the French CODAP or the American ASME Boiler and Pressure Vessel
Code can be used, but proof of conformity is at the charge of the manufacturer.
Furthermore, useful European Norms for semi-finished products are listed in Table 13.
Table 13: List of European Norms for semi-finished products
Product European Norm

Plate and sheets EN 10028-1:2007+A1:2009 Flat products made of steels for pressure purposes—Part
1: General requirements
EN 10028-3:2009 Flat products made of steels for pressure purposes—Part 3:
Weldable fine grain steels, normalized
EN 10028-7:2007 Flat products made of steels for pressure purposes—Part 7: Stainless
steels
Tubes EN 10216-5:2004 Seamless steel tubes for pressure purposes—Technical delivery
conditions—Part 5: Stainless steel tubes
EN 10217-7:2005 Welded steel tubes for pressure purposes—Technical delivery
conditions—Part 7: Stainless steel tubes
Forged blanks EN 10222-1:1998 Steel forgings for pressure purposes—Part 1: General requirements
for open die forgings
EN 10222-5:1999 Steel forgings for pressure purposes—Part 5: Martensitic, austenitic
and austenitic-ferritic stainless steels
Castings EN 10213:2007 Steel castings for pressure purposes
Pipe fittings EN 10253-4 Butt-welding pipe fittings—Part 4: Wrought austenitic and austenitic-
ferritic (duplex) stainless steels with specific inspection requirement
Bars EN 10272:2007 Stainless steel bars for pressure purposes
Aluminium EN 12392:2000 Aluminium and aluminium alloys –Wrought products—Special
requirements for products intended for the production of pressure equipment (choose
materials included in the list given in EN 13445-8 section 5.6)
7 Pressure relief protection systems

Cryostats contain large cold surfaces, cryogenic fluids, and sometimes large stored energy (e.g.
energized magnets), with the potential risk of sudden liberation of energy through thermodynamic
transformations of the fluids, which can be uncontrolled and lead to a dangerous increase of pressure
inside the cryostat envelopes.
The consequence, in the case of a rupture of the envelopes, may be serious for personnel
(injuries from deflagration, burns, and oxygen deficiency hazard) as well as for the equipment.
Performing a thorough risk analysis is essential to identify and understand risk hazards that may cause
a pressure increase and in order to assess consequences, define mitigation actions, and design adequate
safety relief devices to limit pressure accordingly.
A non-exhaustive list of potential sources of pressure increase is:
• compressors connected to cryogenic lines;
388
C RYOSTAT D ESIGN
• connection to higher pressure source (e.g. HP bottles);

• heating of ‘trapped’ volumes (typically in a circuit between valves) during warm-ups;
• helium leak to insulation vacuum, with consequent increased conduction/convection heat
loads to cryogenic liquid vessels;
• cryo-condensed air leaks on cold surfaces and consequent pressure increase and increased
conduct/convection heat loads during warm-ups;
• heating/vaporization of cryogens from sudden release of stored energy in superconductor
device (e.g. quench or arcing in a superconductor magnet circuit);
• uncontrolled air/nitrogen venting of insulation vacuum with sudden condensation on cold
surfaces;
• uncontrolled release of cryogenic fluid to higher temperature surfaces (thermal shield and
vacuum vessel), and consequent pressure increase and increased conduction/convection heat
loads to cold surfaces.
The last three of these causes are generally those that define the sizing conditions for the overpressure
relief devices in cryostats for superconducting devices, and are developed in the following section.
The applicable European Norm is EN 13648 Safety devices for protection against excessive pressure.
7.1 Design of pressure relief devices

The design of pressure relief devices should be made for each cryostat envelope that is to be protected;
it includes the vacuum vessel and all contained cryogenic reservoirs and circuits.
The design approach can be subdivided as follows:
a) Risk analysis and mitigation:
• Make a thorough risk analysis and evaluate risk hazards;
• Identify mitigation measures (e.g. protection of exposed bellows and flanged connections);
• Identify severity of consequences and appreciate probability of the event;
• Define the Maximum Credible Incident(s) (MCI) and design the safety relief system
accordingly. The safety relief system must be designed to keep pressure rise within the limits
of the Maximum Allowable Working Pressure (MAWP).
b) Design steps for MCI:
• Estimate the heat exchange and its conversion to mass flow rates to be discharged;
• Check the sizing of piping (generally designed for normal operation) to the relief device and
increase if necessary;
• Choose the type of safety device (burst disks, valves, plates) and size the safety device (DN
and set pressure). Make use of safety device manufacturers’ formulas and charts;
• Size recovery piping downstream of safety device and check venting needs in the building
where the release occurs (Oxygen Deficiency Hazard issue).
7.1.1 Design of pressure relief devices for cryogenic fluid vessels

Cryogenic fluid vessels are, in general, pressure vessels, where the MAWP can be in the range of a
few bar (for SRF cavity vessels) up to a few tens of bar (for magnet vessels).
389
V. PARMA
Let’s consider that the hazard leading to the MCI is a breach in the insulation vacuum, leading
to a fast venting with air of the vacuum vessel volume, with sudden condensation heat transfer on the
cold surfaces, and boil-off of the cryogen with an increase of pressure (Fig. 38).
Fig. 38: Sudden air venting of the insulation vacuum volume, pressure increase in cryogenic vessel, relieving
through its safety device.
The heat flux exchange on the cold surface has been measured experimentally, and reference
values for different insulation configurations can be found in literature. The following conservative
values can be used:
• 0.6 W cm−2 for a vessel protected by ten layers of MLI;
• 4 W cm−2 for a bare surface.
The safety relief device should be designed to be able to relieve a mass flow equivalent to the highest
heat load.
The mass flow, Qm can be calculated by following the approach described in EN13468-3.4.
Formulas are given to distinguish between cases in which pressure remains below or exceeds
the critical pressure (p = 2.23 bar for helium).
The same norm provides, under conservative assumptions, formulas for calculating the required
flow relief cross-section, depending on whether the flow is critical or sub-critical.
This is the minimum cross-section area of the safety relief device to be chosen. Figure 39
illustrates a few examples of commercially available pressure relief devices: a burst disk, calibrated to
break at a given pressure difference, and a reversible spring-loaded valve.
(a) (b)
Fig. 39: Pressure relief devices. (a) burst disk; (b) spring-loaded valve
390
C RYOSTAT D ESIGN
7.1.2 Design of pressure relief devices for vacuum vessels

According to the Pressure European Directive 97/23/EC, in order not to be considered a pressure
vessel, vacuum vessels should be designed to keep the internal MAWP within 1.5 bar absolute (i.e.
Δp < 0.5 with respect to atmospheric pressure) in all working conditions, including an accidental
event. The safety relief devices are therefore to be designed to keep within these values.
Often the MCI corresponds to the rupture of a cryogenic circuit: the cryogenic liquid floods the
vacuum vessel, vaporizes and expands in contact with the warm walls, the internal pressure increases
until the safety relief device opens, relieving the fluid to atmospheric pressure (Fig. 40). The vacuum
vessel safety device is to be designed to relieve a mass flow equal to the highest flow from the cryogenic
vessel, but at a warmer temperature, while keeping the vacuum vessel pressure within 1.5 bar absolute.
Fig. 40: Breach in the cryogenic fluid vessel, expansion in the vacuum vessel and relief through its safety device.
Calculating the maximum mass flow from the cryogenic fluid vessel to the vacuum vessel
consists of first estimating the size of the breach in the cryogenic circuit (for instance rupture of a
bellows), then calculating the mass flow through the orifice, depending on the flow conditions and
pressure inside the cryogenic vessel.
Conservatively, the cross-section of the vacuum vessel safety device should be able to discharge
the same mass flow, but at a warmer temperature. This cross-section is highly dependent on the relief
temperature, which is difficult to estimate. In the first instance one should consider the most
conservative assumption of discharge at room temperature. If the sizing results are excessively large
(often the case), one should review this assumption by taking into account more realistic (but still
conservative) assumptions limiting the warm-up of the fluid along the evacuation path, for instance
assuming that the thermal shield protects the fluid from heat-up too much. Once the minimum relief
cross-section is calculated, this should be used to select the safety relief device. Figure 41 illustrates
the DN200 spring-loaded safety relief plate mounted on each LHC dipole cryostat.
Fig. 41: LHC dipole cryostat. DN200 spring-loaded safety relief plate
8 Summary
After more than one century of developments and applications since the first laboratory cryogenic
reservoirs of J. Dewar, cryostats are nowadays industrially produced. From a first intuitive
understanding of heat exchange, we now have a solid understanding of the phenomena involved. An
391
V. PARMA
overview of the basic thermal and mechanical design aspects has been presented, trying to provide
simple calculation tools for preliminary design.
If cryostats are built in industrial series for commercial applications, accelerator cryostats
remain a niche in the field, and are designed for very specific applications in superconductivity, most
of the time requiring state-of-the-art enabling technologies like cryogenics and vacuum. Some aspects
of these disciplines were treated. Mechanical and construction issues have been mentioned, as well as
standards and norms, which apply when cryostats are fabricated in an industrial context. The LHC
cryostat has been extensively used as a case study throughout this paper. For further developments, the
reader is invited to refer to the bibliography presented hereafter.
Acknowledgements
The work presented in this course is essentially the result of contributions from a number of colleagues
and the work done during the design and construction of the LHC. I wish to acknowledge in particular
for the material provided and for their contributions in preparing this course: R. Bonomi,
P. Cruikshank, Ph. Lebrun, Y. Leclercq, A. Poncet, D. Ramos, and G. Vandoni.
References
[1] A. Bejan, Heat Transfer (Wiley & Sons, New York, 1993).
[2] Cryogénie, ses applications en supraconductivité, IIF/IIR, various authors. Techniques de
l’ingénieur, (1995, Paris).
[3] M. Wilson, Superconducting Magnets. Oxford University Press (Oxford, 1983).
[4] R.R. Conte, Éléments de Cryogénie. Masson & Cie, Éditeurs, (Paris, 1970).
[5] S.W. Van Sciver, Helium Cryogenics, The International Cryogenics Monograph Series
(Plenum Press, New York 1986).
[6] K. Mendelssohn, The Quest for Absolute Zero (McGraw Hill, New York, 1966).
[7] R.B. Scott, Cryogenic Engineering (Van Nostrand, Princeton, 1959).
[8] G.G. Haselden, Cryogenic Fundamentals (Academic Press, London, 1971).
[9] R.A. Barron, Cryogenic Systems (Oxford University Press, New York, 1985).
[10] B.A. Hands, Cryogenic Engineering (Academic Press, London, 1986).
[11] K.D. Timmerhaus and T.M. Flynn, Cryogenic Process Engineering (Plenum Press, New York,
1989).
[12] J. Ekin, Experimental Techniques for Low Temperature Measurements (Oxford University
Press, 2006, 2007, 2011).
[13] Proceedings of the CAS-CERN Accelerator School: Superconductivity and Cryogenics for
Particle Accelerators and Detectors, Erice, Italy, 8−17 May 2002, edited by Stephan
Russenschuck and Giovanna Vandoni, CERN−2004−008 (CERN, Geneva, 2004).
[14] Proceedings of ICEC and CEC/ICMC conferences.
[15] W. Lehman and G. Zahn, Safety aspects for LHe cryostats and LHe transport containers,
ICEC7, London, 1978.
[16] G. Cavallari et al., Pressure protection against vacuum failures on the cryostats for LEP SC
cavities, 4th Workshop on RF Superconductivity, Tsukuba, Japan, 14–18 August, 1989.
(pag.781-803).
[17] M. Wiseman et al., Adv. Cryog. Eng. 39 (1994) 997.
392
C RYOSTAT D ESIGN
Appendix A. Case study: Thermal radiation design of the Large Hadron Collider
cryostats
Let’s consider the design of a tubular cryostat as in the case of the LHC. The aim is to analyse the
effects, in a step-by-step sequence, of insulation vacuum, thermal shielding, and MLI on heat loads.
Consider the following data:
• Magnet cold mass: 316 L (stainless steel), diameter 0.6 m, T = 2 K;
• Vacuum vessel: low-carbon construction steel, diameter 1 m, T = 293 K;
• Heat loads budgets: ~ 0.2 W m−1 at 2 K; ~ 5 W m−1 at 80 K;
We will consider both the cases with a good insulation vacuum (P = 10−3 Pa (10−5 mbar)), and a
degraded insulation vacuum (P = 10−1 Pa (10−3 mbar)).
All calculations in the following are made for a cryostat unit length of 1 m.
a) Bare cold mass, good insulation vacuum

Considering:
• Emissivity cold mass: εCM = 0.12
• Emissivity vacuum vessel: εVV = 0.2
From Eq. (27):
σ ACM (TVV 4 − TCM 4 )
Q= (A.1)
1 ACM  1 
+  − 1
ε CM AVV  ε VV 
We calculate the heat load to the cold mass:

HLCM = 63 W (A.2)
which is far too high with respect to the budget.
b) One aluminium foil around cold mass, good insulation vacuum

With:
• Emissivity of Al foil (at 2 K): εCM = 0.06
We now calculate:
HLCM = 40 W (A.3)
which remains too high.
c) 30 MLI layers around cold mass, good insulation vacuum

Considering a heat load from 290 K with 30 MLI layers (calculated with the MLI formula): 1.2 W m−2
we now calculate:
HLCM = 2.3 W (A.4)
Despite the large gain, we are still one order of magnitude off budget.
393
V. PARMA
d) Bare cold mass, actively cooled thermal shield, good insulation vacuum
Let us now introduce an intermediate thermal shield in aluminium, actively cooled to 80 K, but still
consider a bare cold mass.
• Thermal shield diameter: 0.8 m
• Thermal shield T = 80 K
• Emissivity of Al (at 80 K): εTS = 0.1
Using the equation for radiation, we calculate the heat load to the cold mass and a second heat load to
the thermal shield HLTS:
HLCM = 0.26 W  close to budget (A.5)
HLTS = 79 W  above budget (5 W m−1) (A.6)
e) Adding 30 MLI layers around the thermal shield, good insulation vacuum
Considering a heat load from 290 K with 30 MLI layers (calculated with MLI formula): 1.2 W m−2:
HLCM = 0.26 W  close to budget (A.7)
HLTS = 3.0 W  within budget (A.8)
f) Adding one Al foil around the cold mass, good insulation vacuum
With:
• Emissivity of Al foil (at 2 K): εCM = 0.06
We now calculate:
HLCM = 0.18 W  within budget (A.9)
HLTS = 3.0 W  within budget (A.10)
This configuration satisfies the heat load budgets, provided the insulation vacuum is good.
g) Adding one Al foil around the cold mass, degraded insulation vacuum
Let us now consider case f) but in a case where, due to a helium leak, the insulation vacuum is
degraded. By making use of Eqs. (18) and (19), we can calculate the residual conduction contribution
to heat loads.
In a first instance let us consider P = 10−3 Pa (10−5 mbar), which is still a relatively good
vacuum. In this case:
Qres = 0.15 W (A.11)
And the total load is:

Q = Qrad + Qres = 0.18 + 0.15 = 0.33 W (A.12)
Though exceeding the budget the contribution from residual conduction remains limited.
Let us now consider P = 10−1 Pa (10−3 mbar), which is a bad vacuum. In this case:
Qres = 15 W (A.13)
which is two orders of magnitudes higher and unacceptably high.
394
C RYOSTAT D ESIGN
h) Adding 10 MLI layers around the cold mass, good and degraded insulation vacuum
By adding 10 MLI layers on cold mass, and using measured data in good vacuum (<10−3 Pa):
50 mW m−2, we calculate:
HLCM = 0.09 W  well within budget (A.14)
HLTS = 3.6 W  within budget. (A.15)
Under degraded vacuum (~10−1 Pa), and measured data with MLI: ~2 W m−2, we calculate:
HLCM = 3.8 W  above budget (A.16)
HLTS = 3.6 W  within budget. (A.17)
Even though the budget at 2 K is exceeded, it can be noted that MLI on the cold mass reduces heat
loads by a factor of 4 in the case of degraded vacuum. Furthermore, MLI is normally wrapped around
cold surfaces in order to reduce, by a factor of about 7, condensation heat fluxes in the case of
accidental cryostat venting with air (bare surface: q ~4 W cm−2; 10 layers of MLI: q ~0.6 W cm−2).
Table A.1 summarizes the various configurations calculated.
Table A.1: Summary of heat loads for various configurations
Case 2 K heat loads 80 K heat loads

(W) (W)
a) Bare cold mass 63 NA
b) Cold mass with one Al foil 40 NA
c) Cold mass with 30 MLI layers 2.3 NA
d) One thermal shield at 80 K, no MLI 0.26 79
e) 30 MLI layers on thermal shield 0.26 3.0
f) As e) + one Al foil on cold mass 0.18 3.0
g) As f) but degraded vacuuma Up to 15 >33.0
h) +10 MLI layers on cold mass 0.09b 3.6
3.5c >3.6
a
10−1 Pa; bin good vacuum; cin degraded vacuum.
395
V. PARMA
Appendix B. Experimental assessment of the thermal performance of the Large Hadron

Collider cryostats
The following presents the assessment of the static heat loads to the 1.9 K cold mass and to the 50–65 K
thermal shielding, carried out during the first commissioning period. Results are then compared with
budgets. A sensibility analysis of parameters like insulation vacuum pressure and cryogenic line
temperatures is also discussed.
Cryogenic cooling loop of the arc cells

The cold masses containing the superconducting magnets are cooled via a separate cryogenic circuit,
featuring an internal bayonet heat exchanger operating at low-pressure saturated helium (16 mbar), in
which liquid is fed from the cryogenic line C and vapours are pumped via the cryogenic line B back to
the cryogenic plant. This cooling loop extends over a total length of 107 m, covering two quadrupoles
and six dipoles, and ends in a helium phase separator for collection of liquid in excess (Fig. B.1). The
common pressurized 1.9 K static helium bath of the cold masses, a so-called cryogenic subsector,
extends over the length of two cooling loops (214 m) and is delimited by plugs.
Fig. B1: Schematic cryogenic layout of a standard arc cell
A schematic of the heat flow in the LHC cryostat is presented in Fig. B.2.
(a) (b)
Fig. B.2: (a) Cross-section of a dipole cryostat; (b) schematic of heat flows and heat intercept temperatures
396
C RYOSTAT D ESIGN
HeII calorimetric measurements of heat load to the cold mass

The static heat load received by the cold masses in a cryogenic subsector, considered as a closed
system, can be estimated by the change in its internal energy over time, during a warm-up, while
active cooling via the heat exchanger is stopped. A cryogenic sector can be substantially considered a
closed system when the valves to the cryogenic distribution line are closed and leak-tight. This
assumption can be checked by calculating the density of helium, on the basis of measured pressure and
temperature in the system, and verifying that it remains constant during the isochoric transformation.
Moreover, since the internal energy of the cold masses is derived from point measurements of
the temperature and pressure in the system, a uniform temperature is required; this condition is
ensured at temperatures below the λ point, taking advantage of the very high heat transport properties
in superfluid helium. Calculating the internal energy also requires knowledge of the heat capacity of
the cold masses, and in particular related to their helium content, which represents the dominating
contribution at cryogenic temperatures. The advantage of the measurement method described below is
that it allows estimation of both the helium content and the static heat load by calorimetric
measurement.
After stopping the supply of sub-cooled helium to the heat exchanger in one or more subsectors,
a natural warm-up (phase A) starts. After about 1.5 h, a faster warm-up is forced by injecting a known
electrical power (typically 40 W per subsector) into the cold masses via built-in heaters, giving a
second trend of about 1 h in the temperature increase (phase B). Fig. B.3 illustrates the evolution of
temperatures in the cold masses during one of these tests. The constant calculated value of helium
density is also plotted.
Fig. B.3: Temperature evolution and calculated helium density in one subsector during a test.
During this evolution, and before reaching the λ line, the set of two energy balance equations can be
written:
WCM
=
∆U CM
=
A
∑ m ⋅ ∆u
i i i
A
(B.1)
∆t A ⋅ Ls ∆t A ⋅ Ls
WCM + W=
∆U CM B
=
∑ m ⋅ ∆ui i i
B
(B.2)
IN
∆t B ⋅ Ls ∆t ⋅ Ls B
397
V. PARMA
where u = u(p,T) is the specific internal energy, mi and ∆ui j are the mass content and the variation of
specific internal energy of the cold mass i-th component, respectively (different materials and
unknown helium content) during the phase j (j being A or B), lasting ∆tj, while Ls is the subsector
length.
Equation (B1) corresponds to the natural warm-up under a static load per unit length WCM
(phase A), whereas Eq. (B2) corresponds to the forced warm-up, under the sum of static and electrical
loads per unit length WCM + WIN (phase B). Solving the system of Eqs. (1) and (2) yields the two
unknowns, namely helium average content m He and WCM.
Once the helium content is determined, natural warm-ups were sufficient to measure static heat
loads elsewhere in other cryogenic subsectors whenever they occurred during the LHC
commissioning. Measurements were made acquiring data every minute during periods of a few hours
on a number of subsectors. Insulation vacuum pressure and temperatures of lines C′ and E were also
monitored in order to assess their influence on heat loads.
Table B.1 reports test results for a number of measurement runs carried out on specific
subsectors, together with parameters to which heat loads are correlated, namely temperatures of the
lines E and C′ and insulation vacuum pressure. The analysis presented hereafter assumes no sector-
specific dependence of the results.
Table B.1: Test results for heat loads to the 1.9 K cold mass
Cold mass heat Properties within subsector

Test Cryogenic subsector/ load in Line E Line C′ Insulation
Test #
method machine sector subsector (K) (K) vacuum
(mW m−1) (Pa)
1 A+B 15_17/sector 7 right 130.4 ± 15.1 52.2 ± 5 10 ± 1 0.9·10−5
2 A+B 23_25/sector 7 right 126.5 ± 14.7 50.5 ± 5 10 ± 1 0.9·10−5
3 A+B 27_29/sector 7 right 126.0 ± 14.6 49.6 ± 5 10 ± 1 4.2·10−5
4 A 15_13/sector 3 left 206.8 ± 24.0 51.5 ± 5 20 ± 1a 4.3·10−5
5 A 11_13/sector 2 right 184.9 ± 21.4 43.5 ± 5 20 ± 1a 5·10−5
6 A+B 17_19/sector 6 left 133.1± 15.1 43.5 ± 5 7±1 5.3·10−5
7b A+B 11_13/sector 5 right 141.3 ± 16.7 53 ± 5 5.5 ± 1 3.9·10−5
31/sector 7 right to 6÷8±1
8 A 271.6 ± 31.5 50 ± 5 1·10−4 ÷ 1·10−3 c
33/sector 8 left (unstable)
a
Line C′ not cooled during the test. bMagnet quench occurred in this subsector a few hours earlier, possibly
increasing heat loads. cDegraded vacuum due to leaks.
From tests 1, 2, 3, 6, and 7, the average helium content has been estimated to be
26.8 ± 3.1 l m−1.
Heat loads in all tests except 4, 5, 7, and 8 (discussed below) are more than 36% lower than the
budget estimate of 0.21 W m−1. This can be in part explained by the lower than nominal temperature in
line E, thus reducing conduction heat through the support posts and radiation heat from the thermal
shield to the cold mass. However test 7, which has the closest to nominal temperatures in lines E and C′,
yields a heat load which remains 32% lower than nominal. A clear correlation between temperature of
line E and heat load is seen by comparing tests 5 and 4, or tests 6 and 7, where an increase of 8–10 K,
while keeping all other variables fixed, translates into an increase in heat load of 6% (12% when line C′
is also off-nominal).
When line C′ is not cooled (tests 4 and 5), an increase in heat loads up to 39% is observed (comparing
tests 5 and 6). This large effect results from different additional heat loads, since line C′ intercepts heat from
398
C RYOSTAT D ESIGN
supports and also from other components. About one half of the additional heat load is imputable to the cold
mass support posts, while the remaining contribution could be explained by the additional heat through other
components, like power leads feed-through and beam-screens supports.
Finally test 8, in a subsector with degraded vacuum due to helium leaks, yields heat loads which
are more than twice as high than those of test 2 where temperatures of the cryogenic lines are
comparable.
Heat load to the thermal shield

In stationary conditions, the heat intercepted by the thermal shield is given by the enthalpy change
during the non-isothermal heating of gaseous helium along line E. For each cryogenic subsector, the
heat load to the thermal shield per unit length is:
WE
=
∆H s
=
( m E ⋅ ∆hs ) (B.3)
Ls Ls
where ∆hs = ∆hs(p,T) is the specific mass enthalpy change along the Ls-long segment of line E
corresponding to the considered subsector, as a function of temperature and pressure readings, while
m E is the helium mass flow evaluated from the overall measured pressure drop ∆p along line E.
Measurements were made in stable conditions on a period of three consecutive days, at a
sampling time of one minute, taking averages to reduce measurement uncertainties.
Only one test was carried out in one sector in stable condition with temperature of line E
between 45 and 55 K (5-10 K lower than nominal). The resulting heat load along the sector varies
between 3.1 and 4.3 W m−1, in average 18% lower than budgeted. Since the temperature of line E was
lower than nominal, these estimates are to be considered as upper limits.
In summary, even taking into account the uncertainties of the methods used to assess the heat
loads in nominal conditions of the thermal shield and of the cold mass helium bath, the performances
of the LHC cryostat are well within design and measured values on prototype strings of magnets. Heat
loads at 1.9 K are about 25% lower than nominal when operating the cryostat thermal shield and heat
intercepts at temperatures close to nominal. The heat loads on the thermal shield, estimated on one
sector only, are also considerably below nominal (-18%), and would further decrease operating it at its
nominal temperature.
These results also confirm that the performance of the LHC magnet cryostats has not been
impaired by machine installation and in particular the interconnection assembly performed in the
tunnel.
399
Cable Stability
L. Bottura 1
Abstract
Superconductor stability is at the core of the design of any successful cable
and magnet application. This chapter reviews the initial understanding of the
stability mechanism, and reviews matters of importance for stability such as
the nature and magnitude of the perturbation spectrum and the cooling
mechanisms. Various stability strategies are studied, providing criteria that
depend on the desired design and operating conditions.
Keywords: cable stability, perturbation spectrum, cooling mechanisms,

superconducting magnets.
1 Introduction
The first superconducting magnets built in the decade around 1960 out of the newly available
superconducting materials (mainly Nb–Zr, Nb3Sn, and Nb–Ti, in the form of tapes or large
monofilamentary strands) had their first transition to the normal state – that is, a quench – long before
reaching the expected critical current, largely disappointing the constructors. This happened in spite of
the success in the development of pure superconducting materials. The situation has been illustrated
by Chester [1] in an excellent review article on the status of the development of superconducting
magnets:
“[...] the development of superconducting solenoids and magnets has been far from straightforward,
mainly because the behaviour of the materials in coils frequently did not accord with the behaviour of
short samples. [...] The large number of coils [...] wound from Nb–Zr and Nb–Ti wire, and [...] Nb3Sn,
revealed several intriguing and very frustrating characteristics of these materials in magnets.”
An example of this behaviour is shown in Fig. 1, reporting the history of the maximum current
reached in a superconducting solenoid wound with Nb–Zr wire. At the first powering, the magnet
reached only ~12 A, after which it quenched. At the following attempt to power the magnet, the
current that could be reached before quench was higher. This process continued at each attempt, and
the maximum current that could be reached increased quench after quench, slowly approaching a
plateau – in the example of Fig. 1, at around 28 A. This behaviour became known as training [2], and
the curve reported in Fig. 1 is called the training curve. The plateau current reached, however, was
still below the expected maximum current-carrying limit of the cable. As shown in Fig. 2, coils wound
from 0.25 mm Nb–Zr wire could achieve only a small fraction of the critical current of the single wire.
1
luca.bottura@cern.ch
L. B OTTURA
Fig. 1: The training curve for an early Nb–Zr superconducting solenoid. (Reproduced from [1].)
Fig. 2: The typical critical current of a Nb–Zr superconducting wire (curve A) compared with that in coils
producing the same field (curve B): 10 kOe = 1 T. (Reproduced from [1].)
The current limitation observed was originally thought as originating from bad spots in the wires or
cables, and thus attributed to bad homogeneity in the quality of the superconductor. This idea
produced the concept of degradation of the performance of the conductor. Although training clearly
showed that a physical degradation could not be responsible for the bad performance, the misleading
name remained as an inheritance of the imprecise understanding. Particularly puzzling was the fact
that the degradation depended on the coil construction and on its geometry. Again quoting Chester [1]:
“The prediction of the degraded current for any new shape or size of coil proved to be impossible and,
for a time, the development of coils passed through a very speculative and empirical phase.”
A principle not yet fully understood at the time was that of stability of the cable with respect to
external disturbances. Insufficient stability and large external disturbances were the key issues in the
failure of the early experiments on superconducting magnets. It has since become understood that a
superconducting magnet is always subject to a series of energy inputs of very different natures, time-
scales, and magnitudes, the so-called disturbance spectrum [3]. The energy input in the
superconducting cable increases its temperature and can be sufficient to take the superconducting
material above critical conditions, where it becomes resistive and Joule heating is generated. The
region that has transited to the normal resistive state is the so-called normal zone in the magnet. Most
materials at cryogenic temperature have a small heat capacity (ideally vanishing at absolute zero), and
the temperature margin, the difference between the operating temperature Top and the temperature at
402
C ABLE S TABILITY
which current sharing starts, Tcs, must be kept small to achieve an economic design. The result is that
the energy necessary to produce a normal zone can be small, typically in the range of tens to hundreds
of microjoules for a few cubic millimetres of strand.
If not prevented by other mechanisms, the temperature in the normal zone increases further and
the normal front propagates, so that the superconductor cascades from its nominal operating point into
an irreversible process, leading to a complete loss of superconductivity in the magnet; that is, the
magnet experiences a quench. This sequence of events is shown schematically in Fig. 3.
Fig. 3: An event tree following an external energy input, and leading from stable operating conditions back to
stable operation or to a magnet quench. The stability design and analysis are concentrated on the shaded area in
the event tree.
Even if properly protected against damage, a magnet quench is an undesirable event in terms of
availability and cost. A well-designed magnet will not quench under normal operating conditions. The
study of stability has evolved through many years of experimentation and analysis towards an
understanding of the processes and mechanisms whereby a superconductor remains (or not) within its
operating region, thus ensuring stable operation of the magnet; that is, free from quenches. With
reference to the schematic representation of Fig. 3, stability is therefore mainly concerned with the
phases in the event tree enclosed in the shaded area.
In spite of the substantial progress in understanding and improvement in the manufacturing
techniques, stability is still one of the limiting factors for high-performance magnets. As an example,
Fig. 4 reports the sequence of training quenches for an accelerator magnet, showing that, to achieve
high performance, training may still be necessary, and still the short-sample limit is not reached. The
training behaviour can differ significantly depending on manufacturing details such as the quality of
the fitting of the mechanical structure and the coil windings.
403
L. B OTTURA
Fig. 4: A sequence of training quenches for a short model of an LHC dipole magnet, showing the initial, short
training period necessary to reach nominal operating conditions and a plateau current reached after several
quenches. Note that the plateau is below the expected short-sample limit of the cable.
This chapter deals with cable stability under most common conditions found in superconducting
magnet design. The first step in a sound design is to estimate the envelope of the perturbations that
will be experienced by the magnet. Typical energies and time-scales of the perturbation spectrum are
reviewed in Section 3.2. The conductor can then be designed to accommodate these perturbations by
means of a sufficiently large stability margin, using the concepts discussed in Sections 3.6 (adiabatic
stabilization), 3.7 (cryostability), 3.8 (cold-end recovery and the equal-area theorem), 3.9 (well-cooled
operation for force-flow cooled conductors), and 3.10 (minimum propagating zone). Appropriate
examples based on existing magnets are given in each section. In order to develop the relevant
concepts and techniques, we will introduce in Section 3.3 a general form of the heat balance for a
superconducting cable, and in Sections 3.5 and 3.6 we will discuss the details of heat generation
during current sharing and steady state, and transient heat transfer to a cryogenic coolant. Finally, in
Section 3.11, we will discuss some advanced topics (current distribution) in stability design and
analysis.
2 The perturbation spectrum

Several mechanisms can cause the generation of heat in a superconducting cable carrying a current in
a magnetic field and operating in a cryogenic environment. These perturbations are distributed over a
wide spectrum, with large differences in magnitude and time-scale depending on the origin of the
perturbation itself. Some disturbances – for example, the mechanical energy release due to a small
motion of a superconducting wire – can be extremely localized in space and time, and can initiate only
normal zones of small volume. Other disturbances can affect large portions of a superconducting
magnet and last a significant time – for example, a.c. loss for pulsed operation or heat deposition from
nuclear processes in a superconducting accelerator magnet – and thus they can potentially produce
normal zones with large volume. The specific perturbation spectrum for a particular application
depends on the design of the superconducting magnet and on the operating conditions of the system,
and is difficult to generalize.
In this section, we give a sample of most common disturbances, and their associated energy,
volume and time-scales. Stability analysis and design is mostly concerned with transient perturbations
of limited duration. Hence they can be characterized by the total energy deposited during the transient
404
C ABLE S TABILITY
in units of [J]. For the comparison of different phenomena acting on different volumes, it is useful to
reason in terms of energy density with reference to the total cross-section of the conductor. For
practical reasons, this quantity is usually quoted in units of [mJ·cm–3], as the typical values then range
from fractions of the unit to a few hundreds.
As is the case for heat leaks from room temperature, continuous heat deposition is characterized
by the heating power, measured in units of [W] for localized inputs, in units of [W·m–2] for surface
heat loads, or in units of [W·m–3] for volumetric loads. These perturbations, although very important
for the overall performance of a system, are usually not the concern for stability. They are dealt with
by proper sizing of the thermal insulation and cooling system. It is nonetheless interesting to show the
corresponding energy deposited over the typical time-scales of interest for stability, also discussed
here in units of [mJ·cm–3] to provide comparable dimensions.
2.1 Flux jumps

A small heat input into a superconductor in a magnetic field – due, for example, to any of the reasons
discussed below – causes a decrease of the critical current density Jc through the temperature increase
(all technical superconductors have a negative Jc(T) slope). In adiabatic conditions, the magnetization
of the superconductor (proportional to the current density) also decreases, resulting in a penetration of
the external magnetic field in the superconducting bulk. A part of the energy stored in the
magnetization profile is therefore dissipated resistively within the superconductor. The energy release
caused by the decrease of the magnetization can be sufficient to cause an irreversible transition of the
wire to the normal state – a flux jump.
In order to estimate the maximum possible energy release during a flux jump, we can make the
assumption that during the transient the magnetization of a superconducting filament disappears
because of the process described above. To simplify matters in this study, we approximate the filament
as a plane slab with thickness D identical to the diameter of the filament. Following a ramp of
increasing magnetic field up to a value B much larger than the penetration field, and neglecting the
influence of transport current, the shielding currents fill the slab, in the fully penetrated state.
Fig. 5: The profile of the magnetic field and current density in a fully penetrated superconducting filament,
ideally represented by a slab of thickness D along x and infinite extension in the other two dimensions. The area
shaded in the plot of the field profile disappears during a complete flux jump.
The magnetic field contribution δB due to the shielding currents has a dependence on the
position x, given by
δ B = µ0 J c x , (1)
where Jc is the critical current density of the superconductor. The field profile is shown in Fig. 5.
Following a complete flux jump, the energy stored in this field profile is dissipated resistively inside
405
L. B OTTURA
the filament. The average energy dissipation per unit volume of the filament can then be
calculated as follows:
D /2
2 δ B2 µ0 J c2 D 2
D ∫0 2 µ0
=Q′′′ = dx . (2)
24
From the above expression, we can estimate the energy density released during a flux jump in a
typical superconductor such as Nb–Ti. Assuming that Jc is 5000 A·mm–2 and that the filament
diameter D is 50 µm, we obtain an energy density of 3 mJ·cm–3. This energy can be released
extremely rapidly, typically in tens of microseconds up to fractions of milliseconds, and is sufficient to
increase adiabatically the temperature of a Nb–Ti strand by few degrees.
The magnetization of a fully penetrated superconducting filament is proportional to the size of
the filament, and the energy stored in the trapped magnetic field is proportional to the square of the
filament size, as shown by Eq. (2). Because of this, flux jumps were a principal problem at times when
superconducting material technology did not allow the production of a fine subdivision of the
superconductor in the wire. In fact, flux jumps were among the first perturbations recognized to be
responsible for performance degradation, and thus were studied intensely in the late 1960s and early
1970s (see, e.g., the review by the Rutherford Laboratory Superconducting Applications Group [4]),
leading to one of the first quantifications of the idea of a disturbance spectrum. Because of this, the
early considerations of stability were often interleaved with flux jump theory.
Fine subdivision in small filaments is the most obvious cure for flux jumping. On the one hand,
it reduces the energy that is dissipated, as is clear from Eq. (2), and therefore it eases the so-called
adiabatic stabilization. On the other hand, fine subdivision means an increase of the superconductor
surface, making it easier to remove the heat generated by the flux penetration in the bulk
superconductor efficiently – so-called dynamic stabilization. Nowadays, flux jumps are no longer a
problem for standard production of low-temperature superconducting wires (typically based on Nb–Ti
and Nb3Sn materials). Flux jumps may still play some role in high-temperature superconductors
operated at low temperatures (around and below 20 K), although the steadily improving technology is
quickly making this statement obsolete.
2.2 Mechanical events

A superconducting magnet is always subjected to stresses, from pre-loading at assembly, differential
thermal contractions at cool-down, or resulting from the electromagnetic forces at operation. The
forces acting on a superconducting cable can induce small movements. In some cases, the stress can be
large compared to the elastic and failure limits of the materials, and displacement can take place as a
result of material yield or fracture. Any displacement causes a change in the stress state associated
with a release of a part of the mechanical energy stored. The release of the mechanical energy can
happen locally, through micro-slips constrained by friction, material yielding, vibration, or local
cracking [5].
We can appreciate the amount of mechanical energy associated with one of the above events by
estimating the energy release due to a hypothetical strand motion, as shown schematically in Fig. 6. A
strand operating at a current Iop in a transverse field Bop results in a Lorentz force F' per unit length of
the strand of
F ′ = I op Bop . (3)
Taking typical values for the bending magnets of a particle accelerator, an Iop of 400 A and a Bop
of 10 T, the force per unit length is F' = 4 kN·m–1. This force is reacted against the other wires in the
winding pack and the structure of the magnet. However, even in a tightly packed winding, the wire can
move through small distances because of the geometrical tolerances on the wire dimensions and the
406
C ABLE S TABILITY
limitations on the control of the winding geometry. Movements δ of a few micrometres over a strand
length l of a few millimetres are not uncommon if the winding pack is not fully impregnated. The
work W performed by the Lorentz force during a movement δ can be calculated as follows:
W = F ′δ l . (4)
A movement of 10 µm of a 1 mm length of a single strand under the conditions given above is

associated with work W = 40 µJ. This work corresponds to an energy density of
F ′δ
W ′′′ = , (5)
Astrand
where Astrand is the total strand cross-section. For a strand with a 1 mm2 cross-section, typical of the
application considered, the energy density is 40 mJ·cm–3. The mechanical work is partially dissipated
as friction against the other wires and partially as a resistive loss induced by the electric field on the
moving wire. The percentage of energy dissipation depends on the detail of the process and cannot be
estimated easily, ranging from only a few per cent to large fractions of the above estimate. These
events take place on a few cubic millimetres of cable, and are fast, with typical times that generally
range from tens of microseconds to a few milliseconds.
Fig. 6: A schematic representation of strand motion in a magnetic field. The strand length l is subjected to a
transverse electromagnetic load F' per unit length and moves by a distance δ in the direction of the force.
In a few cases, the thermal or magnetic stresses become large enough to cause some part of the
structural materials to yield, or to fracture, depending on the embrittlement at low temperature. Under
these circumstances, more massive disturbances are produced, such as de-bonding or shear failure of
the insulation or displacement of a part of the magnet. Heat is released in these processes through
friction during motion, or once the movement is stopped. These massive processes are, however, rarely
taking place at the single-cable level but, rather, at the mechanical interface between the coil winding
and the supporting structures. Hence they are generally produced at some distance from the
superconducting wire, and energy reaches the superconductor only after a diffusion process through
structural components and/or the insulation. For this reason, although the energy release is large,
potentially reaching thousands of mJ·cm–3, and affects large coil volumes, the time-scale for the
energy release is long and thus the associated power is small. In spite of this, particular care must be
applied when designing critical areas in a coil system, such as the interface between the coil winding
and the mechanical structure (e.g. coil flanges and formers for solenoids), interfaces between winding
parts (e.g. mating surfaces of segmented magnets), and coil interconnections (e.g. soldered joints
between windings), aiming on the one hand to avoid displacements and on the other to minimize the
energy release following the unavoidable deformations.
407
L. B OTTURA
2.3 Electromagnetic transients

In several cases, superconductors must be either designed for pulsed operation (e.g. transformers,
power cables, SMES systems) or must be able to withstand transient changes of the self and
background magnetic fields (e.g. the effect of a plasma disruption in a superconducting magnet system
of a thermonuclear experiment, or the tripping of normal resistive inserts in a hybrid solenoid). In any
case, all superconducting magnets, whether designed for steady-state or pulsed operation, must be
ramped to the operating conditions. Hence the operation of a superconductor in a magnet is always
associated with more or less severe conditions with regard to the variation of the field seen by the
cable.
Any field change, on the other hand, produces energy dissipation through hysteresis or coupling
a.c. loss in superconducting filaments, strands or cables. The time-scale of the energy deposition in
this case is governed by the dynamics of the magnetic field as well as the time constant of the induced
persistent or coupling currents, and can cover a large span of characteristic time, from a few
milliseconds up to quasi-steady-state heating conditions for a.c. operation. The energy deposited by
a.c. loss can also vary largely, depending on the field change, its time-scale and the characteristics of
the superconductor.
In well-designed cables, the a.c. loss deposited during pulsed operation or fast field transients
(typical field variations of a few teslas per second, lasting 10–100 ms) ranges from a few mJ·cm–3 to a
few hundreds of mJ·cm–3. Larger energy deposition is usually avoided by a small filament diameter,
thus reducing the hysteresis loss, and the placement of resistive barriers in the strands and within the
cable, thus reducing the coupling loss. Continuous operation with a.c. excitation results in a distributed
power deposition ranging from 0.1 W·m–3 of cable to 100 W·m–3 for the most demanding conditions.
The typical time necessary to reach regime conditions for continuous operation is 0.1–1.0 s.
2.4 Heat leaks

All superconducting magnets operate at temperatures in the cryogenic range. A common source of
heat is therefore represented by leaks entering through the thermal insulation of the magnet, the
supporting structure, current leads and instrumentation wires. Although minimized by proper design,
heat inleaks cannot be completely avoided and are removed by direct or indirect cooling. Heat through
the insulation and supports in normal conditions, as well as loss of cooling or degraded cooling in
abnormal operating conditions, generally result in broadly distributed heat sources of potentially large
deposited energy but low power and nearly steady-state characteristics. Heat leaks from
instrumentation wires or current leads can lead to more localized energy inputs with, however, long
time-scales of energy deposition.
The typical heat loads due to common heat leaks have a broad range of values depending on the
size and operating conditions of the magnet system. The expected values for an accelerator magnet are
a few watts, while for the magnetic system of a fusion reactor the heat load represented by current
leads and heat leaks can amount to a few kilowatts. The corresponding power density with reference to
the cable volume in the magnet ranges from 0.1 to 10 W·m–3. Higher power densities can be reached
for localized inputs, such as a high-resistance superconducting joint or a badly thermalized
instrumentation wire. The characteristic time for the establishment of regime conditions for the heat
load associated with current leads or heat leaks and degraded cooling is from several seconds up to
several hundreds of seconds.
2.5 Other sources

One additional source of energy in superconducting magnet systems used either in nuclear physics
experiments or thermonuclear reactors is the heat deposited by nuclear interactions during normal and
abnormal conditions. One of the most severe examples is the nuclear shower following a partial or
408
C ABLE S TABILITY
complete loss of beam in the steering magnets of a particle accelerator, leading to energy deposition of
a few mJ·cm–3 to a few tens of mJ·cm–3 over short periods of time, in the millisecond range, and
localized over a length of a few metres, which is small compared to the overall accelerator size, but
which still affects large volume portions of a magnet.
Superconducting magnet systems for thermonuclear reactors sustain a neutron flux generated by
the fusion reaction. As for the continuous heat loads discussed in the previous section, although the
nuclear heat can be large, the heat is deposited over long pulses, several seconds in duration, and has
thus nearly steady-state characteristics. The corresponding power density ranges from 10 to 100 Wm–3.
2.6 A summary of the perturbation spectrum

Figure 5 shows a summary of the energy that can be deposited by the various mechanisms discussed
above in a superconducting magnet as a function of the characteristic time for energy deposition, the
perturbation spectrum. The values reported are intended to represent guidelines for the typical orders
of magnitude, and do not necessarily apply to a specific magnet system; nor do they all appear
simultaneously in a given magnet systems, and they should therefore be used with caution. The
representation of Fig. 7 clearly shows, however, that on the time-scale of fast energy deposition
(below 1 ms) the dominating mechanisms are those associated with wire and conductor motion. Flux
jumps with superconducting strands based on present-day technology are masked by the more
energetic mechanical events. At intermediate time-scales, in the range of a few milliseconds to a few
hundreds of milliseconds, the dominant energy perturbation in pulsed magnets is generated by a.c.
loss. For magnets operated in steady state in nuclear environments – for example, particle accelerators
– the a.c. loss is negligible and the main concerns in this time-scale are events associated with particle
showers and nuclear heat. Finally, if the magnet system operates in a steady-state condition and in a
quiet environment – for example, MRI and laboratory magnets – the dominant events are those
associated with conductor motion at the fast time-scale. For longer times, above 1 s, the dominant
perturbations are generated by the steady-state heat loads (heat leaks, nuclear heat, a.c. loss) and are no
longer a direct stability concern, as on this time-scale the heat is usually evacuated by the cooling
system.
Fig. 7: A typical spectrum of the energy perturbations during normal operation of superconducting magnets. The
energy deposited by different processes discussed here is plotted as a function of the characteristic time of
energy deposition. Values are indicative and are intended for comparison of orders of magnitude.
409
L. B OTTURA
3 Heat balance
The temperature of a superconducting cable changes following the energy input associated with one of
the perturbations discussed in the previous section. The evolution of the temperature of the cable is
governed by a transient heat balance containing, in the most general case:
− the heat-generation term representing the external perturbation;
− the Joule heating term, which appears as soon as the superconductor exceeds the current-
carrying capability;
− the heat sink associated with the enthalpy of the cable;
− heat conduction along the cable and across the winding; and
− heat exchange with a coolant, with a possibly limited heat capacity, and either stagnant or
flowing along the cable.
This situation is schematically depicted in Fig. 8.
Fig. 8: Terms contributing to the heat balance governing the evolution of the temperature of a superconductor
following the energy input associated with one of the perturbation sources discussed.
The specific form of the heat balance depends on the details of the cable considered.
Accordingly, the temperature evolution, and in the final analysis the stability of the superconductor,
differs for different cables. The governing equation is relatively simple in the case when the conductor
can be approximated as an adiabatic strand, in which case the conduction and cooling terms disappear
and analytical treatment is appropriate. On the other hand, in the case of a cable as used for large-scale
applications cooled by a forced flow of helium, the terms of the equation can have a complex
mathematical structure and the solution of the heat balance requires extensive numerical treatment.
In order to the discuss the main concepts necessary in order to understand the most important
features of superconductor stability, we focus on the ideal case of a superconducting cable operating at
a current Iop in a background magnetic field Bop and with an initial temperature Top. The external
energy input has a power density qext ′′′ with reference to the whole cable cross-section A, consisting of
a superconductor portion Asc and a stabilizer, a low-resistance material the function of which will be
discussed later, of cross-section Ast:
=A Asc + Ast . (6)
Although other materials such as resistive barriers or structural materials could be included, we
will neglect them for the moment, for the sake of simplicity. The cable temperature T evolves
410
C ABLE S TABILITY
following the perturbation, but is assumed to be uniform across the cable transverse dimension. In
addition, we make the hypothesis of uniform current distribution among the strands of the cable. The
cable is cooled by a helium bath at temperature The. Once again, for the sake of simplicity, the helium
bath is considered as stagnant and with an infinite heat capacity, so that The is constant. The heat
exchange between the cable and the helium takes place over a wetted perimeter w and is characterized
by a heat transfer coefficient h.
Under these conditions, the temperature of the cable evolves in accordance with the following
equation, which is obtained from the balance of heat sources and heat sinks:
∂T ∂  ∂ T  wh
C ′′′ + qJ′′′+
= qext k − (T − The ) . (7)
∂t ∂ x  ∂ x  A
In the above balance, C and k are the cable volumetric heat capacity and the thermal
conductivity, defined as a weighted average of the properties of the superconductor and of the
stabilizer (subscripts ‘sc’ and ‘st’, respectively):
Asc ρsc csc + Ast ρst cst
C= , (8)
A
Asc ksc + Ast kst

k= , (9)
A
where csc and cst are the specific heats, ρsc and ρsc are the mass densities, and ksc and kst are the thermal
conductivities. The term qJ′′′ stands for the heat per unit cable volume generated when the
superconducting material is driven above critical conditions, in the so-called current sharing regime.
This is the topic of the following section, and for the moment we simply assume that qJ′′′ is zero when
the superconductor is operating below critical conditions, and different from zero above.
A sample of a typical temperature evolution following an energy perturbation in a
superconducting cable as described by the heat balance equation (Eq. (7)) is shown in Fig. 9. The
external perturbation inducing the thermal transient is assumed to deposit its energy over a time of
10 ms. Without entering into the details of the results reported in the figure, we note that initially the
temperature of the superconductor increases sharply as a consequence of the heat input provided by
the perturbation. After the end of the energy pulse, the temperature initially drops under the effect of
the heat conduction and cooling heat fluxes. If the conduction and cooling are sufficient, the cable
recovers the superconducting state and eventually returns to stable operating conditions; that is, it
follows the curve marked as recovery in the plot of Fig. 9. If the cooling is insufficient, after the drop
the cable temperature starts to rise again under the predominant contribution of Joule heating,
eventually leading to the quench shown by the curve indicated as thermal runaway in Fig. 9.
The balance among the non-linear Joule heating and the cooling terms is extremely delicate and
can be displaced by small changes in the perturbation energy. On the basis of the qualitative features
discussed above, it is possible to define an energy margin ∆Q’’’ as the minimum energy density that
the external source needs to provide to the cable to cause a thermal runaway. An energy input larger
than ∆Q’’’ causes a thermal runaway, while an energy input smaller than ∆Q’’’ leads to a recovery.
For consistency with our discussion on the perturbation spectrum, we measure the energy margin in
units of [mJ·cm–3]. In the literature, the energy margin is often also quoted as a stability margin, with
the same definition as above. For perturbations of known and limited distribution in space, it can be
useful to refer to the minimum quench energy, ∆Q (MQE), which corresponds to the integral in space
of the energy margin and is thus measured in units of [J].
411
L. B OTTURA
Fig. 9: The qualitative evolution of the temperature in a superconducting cable for an energy perturbation just
below and just above the energy margin. The former leads to a recovery, while the latter results in a thermal
runaway.
4 Current sharing
An ideal superconductor becomes resistive if one of the three parameters current density, temperature,
or magnetic field exceeds the boundary of the critical surface Jc(B,T), and in these conditions the
current flow is associated with resistance and Joule heating in the cable. This is schematically shown
in Fig. 10, which presents the temperature dependence of the critical current Ic of a superconducting
cable, defined as the product of the critical current density and the superconductor cross-section:
I c = J c Asc . (10)
When the cable operates at a temperature Top below the critical current, the material is
superconducting. This situation can also be ideally maintained at a temperature above Top provided
that the operating current is still smaller than the critical current. The temperature at which the critical
current equals the operating current is called the current sharing temperature, Tcs. Above Tcs, the
superconductor develops a resistance and the current flow is associated with Joule heating. The
difference between the current sharing temperature and the operating temperature is often referred to
as the temperature margin, ∆T.
Technical low- and high-Tc superconductors have a high resistivity in the normal state compared
to normal conductors in cryogenic conditions. As an example, Nb–Ti has a normal state resistivity ηsc
of the order of 6.5 × 10–7 Ω·m, while copper and aluminium have typical low-temperature resistivities
ηst of a few 10–11 Ω·m to 10–10 Ω·m, depending on the degree of purity and on the background
magnetic field. As is intuitive, and is discussed later in detail, a decrease in heat generation is always
beneficial for stability. For this reason, in addition to protection and manufacturing considerations,
superconducting strands are nearly always built as a composite containing both superconducting
material and a normal conducting stabilizer material with low resistivity in intimate mechanical,
thermal and electrical contact.
The normal conducting stabilizer acts as a low-resistance shunt in the case of transition of the
superconductor to the normal state, in what is called the current sharing process. A good
representation of current sharing can be achieved considering the superconductor and the stabilizer as
parallel resistors, as shown schematically in Fig. 11. For the moment, we ignore the details of the
current distribution within the strand cross-section, a safe hypothesis in the time-scales of interest for
stability for common strands, but insufficient when dealing with super-stabilized conductors, as will
be discussed in a later section.
412
C ABLE S TABILITY
Fig. 10: The dependence of the critical current on temperature and the definition of the current sharing
temperature.
Fig. 11: Current sharing in a composite strand consisting of a superconductor and a normal conducting material.
The situation is shown for operation below the critical surface of the superconductor (left), in current sharing
conditions, at a temperature between the current sharing temperature and the critical temperature (centre), and
above the critical temperature (right).
For an operating temperature below Tcs, the whole operating current Iop flows in the
superconductor, with zero resistance and no Joule heating. For an operating temperature between Tcs
and Tc, the superconductor develops a longitudinal resistive voltage. Under this voltage, a part of the
current is transferred from the superconductor to the stabilizer. The current in the stabilizer also
produces a longitudinal resistive voltage, and in equilibrium conditions this is equal to the voltage in
the superconductor. The amount of current transferred depends on the electrical characteristics of the
superconductor (in normal state) and of the stabilizer. As discussed above, the normal state resistivity
of the superconductor is much larger than that of the stabilizer. This corresponds to the voltage–
current characteristic schematically shown in Fig. 12, with zero resistivity up to the critical current and
infinite resistivity above.
Fig. 12: The voltage–current characteristic of an ideal superconductor. The resistivity is large in the normal state,
for a ratio of operating current to critical current above 1. In these conditions, the longitudinal electric field has
an ideally infinite value.
413
L. B OTTURA
In this condition, the current transferred to the stabilizer is exactly the current in excess of the
critical current, or
I=
st I op − I c , (11)
while the superconductor carries the current Ic. The longitudinal electric field E in the stabilizer (and in
the superconductor) is given by
ηst
E = I st , (12)
Ast
and the Joule heat power density in the cable can be calculated as follows:
EI st + EI sc EI op ηst I op ( I op − I c )
qJ′′′
= = = . (13)
A A Ast A
Finally, for an operating temperature above Tc, the critical current of the superconductor is zero and
the whole current flows in the stabilizer. In this case, the longitudinal electric field is
ηst
E = I op
Ast (14)
and the Joule heat power density is given by

2
ηst I op
qJ′′′max = . (15)
Ast A
Fig. 13: Current sharing between the superconductor and the stabilizer as a consequence of a temperature change
The situation described above is shown schematically in Fig. 13. We see by direct comparison
of Eqs. (13) and (15) that the maximum Joule heating is reached in the condition T > Tc, as expected.
So far, we have made no assumptions with regard to the temperature dependence of the critical
current, and the expressions derived above are quite general. It is, however, customary to take a line
approximation for the Ic(T) dependence, writing that
Tc − T
I c ≈ I op . (16)
Tc − Tcs
414
C ABLE S TABILITY
In this case, we can write explicitly the temperature dependence of the longitudinal electric field
in the current sharing regime:
T − Tcs ηst
E = I op , (17)
Tc − Tcs Ast
which is also a linear function of temperature, rising from zero at the current sharing temperature Tcs to
its maximum at the critical temperature Tc. The Joule heat power density is given by
0 for T < Tcs ,

  T − Tcs 
=qJ′′′ qJ′′′max   for Tcs < T < Tc , (18)

  Tc − Top 

qJ′′′max for T > Tc ,
where the maximum Joule heating power density is defined as in Eq. (15). The functional dependence
reported in Eq. (18) is represented schematically in Fig. 14. Note the non-intuitive linear increase of
the Joule heating power between Tcs and Tc.
The ideal situation discussed so far provides a fair description of most situations, with a simple
caveat on the fact that the stabilizer resistivity entering the expressions above usually has a strong
dependence on the magnetic field below 20 K and on temperature above 20 K. Once this dependence
is taken into account, the result of Eqs. (15) and (18) can be used for most practical cases.
Fig. 14: The Joule heating power as a function of temperature, normalized to the maximum value reached for
operation above Tc.
In particular situations – for example, when dealing with operating conditions a few per cent
away from the critical current, or when considering large cables manufactured with from several
hundreds to a thousand strands – it is necessary to correct the above result to achieve an accurate
description. In these particular cases, it is no longer sufficient to use the asymptotic voltage–current
characteristic shown in Fig. 12. A better description for the resistive transition of the superconductor is
achieved using, for instance, the power-law approximation given by
n
I 
E = E0  sc  , (19)
 Ic 
415
L. B OTTURA
where E0 is the electric field used as a criterion for the definition of the critical current and n is the
exponent defining the sharpness of the transition. The power law has the form shown in Fig. 15.
Fig. 15: The normalized voltage–current characteristic of a superconductor characterized by a power-law

dependence of the resistive voltage on the current, with exponent n. The voltage–current characteristic is plotted
for different values of the exponent n.
A low value of the exponent n corresponds to a shallow and broad transition, while a high value
of the exponent n gives a sudden transition. Note also that a low exponent n corresponds to the
appearance of a resistive voltage before reaching the critical current. We should therefore expect that a
small portion of the operating current is already transferred to the stabilizer below the critical
conditions. Finally, the ideal limit used so far is achieved when the exponent n tends to infinity.
The current sharing between the superconductor described by the power law and the stabilizer
can be computed by equating the longitudinal electric field as done previously. The electric field in the
superconductor is given by Eq. (19). For the stabilizer, Eq. (12) still holds, but we rewrite it as
follows:
ηst
E
= (I op − I sc )
Ast
. (20)
The equilibrium condition is that

n
I  ηst
E0  sc= (I op − I sc ) , (21)
 Ic  Ast
which cannot be solved analytically for an arbitrary value of the exponent n. It is, however, possible to
solve Eq. (21) numerically, obtaining the value of the current in the superconductor and in the
stabilizer. The Joule heat power density is then computed as in Eq. (13), where this time the simplest
form is obtained as follows:
EI st + EI sc EI op
=qJ′′′ = . (22)
A A
416
C ABLE S TABILITY
Fig. 16: The normalized Joule heat dependence on temperature for a superconductor with a power-law voltage–
current relation. The normalized heat generation is plotted for different values of the exponent n and compared to
the ideal case obtained for an infinite value of n. The inset shows a detail of the heat generation in the vicinity of
the current sharing temperature Tcs.
A sample of the numerical solution obtained for normalized heat generation is shown in Fig. 16.
We note that low values of the power-law exponent n correspond to less heat generation in the current
sharing regime, between Tcs and Tc. The implications of this fact for stability will be discussed later.
On the other hand, a peculiar feature to be remarked in the case of low n is that the Joule heating
already starts before reaching Tcs, which is consistent with the appearance of an electric field already,
below the critical conditions, as discussed earlier. In fact, early current sharing and Joule heating at
low n can be a limiting factor for operation at a high fraction of the critical current, and for this reason
a high value of the exponent n is considered an indicator of the good quality of the strand or cable.
Finally, note how the linear limit given by Eq. (18) is approached when n is large, as expected.
5 Heat transfer
With the sole exception of superfluid helium, heat transfer in cryogenic fluids has been found to be
very similar to that predicted by standard thermodynamics. Proper allowance must be made for the
fact that the thermophysical properties at the operating point of relevance are very different from those
of room- and high-temperature coolants. Apart from this, however, the correlations available in the
literature for the various room- and high-temperature heat transfer regimes are essentially also valid in
cryogenic conditions with small adaptations. For low-temperature superconducting magnets two
regimes are of particular relevance, namely boiling heat transfer to a stagnant bath of atmospheric
pressure, saturated liquid helium (temperature around 4.2 K, pressure approximately 1 bar) and forced-
flow convection of supercritical helium (temperature in the range of 4.5 K, pressure above 3 bar). For
these two regimes, we will give practical correlations and typical values of heat transfer that will be
used later in the discussion. In addition, in more recent years, following advancements in the
technology necessary to produce superfluid helium, cooling in a bath of stagnant, sub-cooled
superfluid helium at atmospheric pressure is used in several small and large-sized applications. Heat
transfer in superfluid helium is peculiar and we will therefore discuss a simple approximation to this
process.
5.1 Boiling heat transfer

Cooling of the first large-sized superconducting magnets, such as the Big European Bubble Chamber
(BEBC), was achieved by submerging the magnet in a saturated bath of stagnant helium at
atmospheric pressure and a temperature of 4.2 K. Heat transfer in these conditions is associated with
417
L. B OTTURA
phase transition from liquid to vapour, the boiling process. The heat transferred from the heated
surface to the helium in the boiling regime has a complex but known behaviour, shown schematically
in Fig. 17.
Fig. 17: The heat flux for boiling heat transfer regimes: the transition between nucleate and film boiling
conditions is unstable.
In the figure, we report a collection of results of correlation fits to measured heat flux data as a
function of the temperature difference between the heated surface and the bulk of the bath [6, 7]. If the
temperature difference is small, heat transfer takes place in the nucleate boiling regime, where the heat
transferred is proportional to approximately the third power of the temperature difference. For
temperature differences in the range of 0.5 K, the heat transfer reaches a crisis point at which
conditions become unstable. The maximum heat flux that can be reached is the Peak Nucleate Boiling
Flux (PNBF), which depends on the nature of the heated surface. The material, surface roughness,
surface coating, and surface orientation with respect to gravity can affect the heat transfer and PNBF
by a factor of 2 [8]. The value of the PNBF for helium, however, has never been found to exceed
10 kW·m–2. If the heat flux is increased above the PNBF, the surface becomes covers with a film of
vapour and the heat transfer degrades. This regime is called film boiling and is characterized by a
linear dependence of the heat flux on the temperature difference. The transition from nucleate to film
boiling and back is hysteretic, takes place at constant heat flux and is characterized by a sudden jump
in the temperature difference at the surface.
From the typical data of Fig. 17, it is possible to calculate an effective heat transfer coefficient,
h, as the ratio of the heat flux to the temperature difference ∆T between the heated surface at
temperature Ts and the helium at temperature The. This has been done below, leading to the following
approximate expressions:
hNucleate = 405 + 37 ×103 (Ts − The ) ,
1.4
(23)
hFilm 592 (Ts − The )

−0.077
= , (24)
which hold for saturated helium at atmospheric pressure. The result of the above expressions is shown
in the compilation of Fig. 18, together with other heat transfer mechanisms discussed later. The
nucleate boiling regime is associated with extremely high values of h, changing rapidly and ranging
from a few hundreds of W·m–2·K–1 to well above 10 kW·m–2·K–1 for a temperature difference of
0.5 K. As soon as the PNBF is reached, however, the equivalent heat transfer drops to a constant
value, of the order of 400–600 W·m–2·K–1.
418
C ABLE S TABILITY
Fig. 18: A summary of the equivalent heat transfer coefficient for different heat transfer regimes in helium
During the transient processes of interest for stability, the heat transfer to a bath of saturated
helium is substantially different with respect to the steady-state behaviour discussed above. In
measurements, it is found in particular that nucleate boiling persists at much higher heat fluxes than
those observed in steady-state conditions, more than one order of magnitude higher than the PNBF
discussed above. This is the effect of thermal diffusion in the helium in direct contact with the heated
surface [9]. The heat transfer crisis is reached in transient conditions when the helium volume affected
by thermal diffusion has absorbed an energy equivalent to the latent heat of evaporation, at which
point a transition to film boiling takes place. Also, the equivalent heat transfer coefficient during
transient boiling can reach extremely high values, of the order of 50 kW·m–2·K–1, probably limited by
Kapitza resistance at the solid wall [9].
5.2 Forced flow

Steady-state heat transfer to a turbulent forced flow of supercritical helium appears to be well
approximated by a correlation of the Dittus–Boelter form, as shown by Yaskin [10] and Giarratano
[11]. A best fit of the available data is obtained with the following expression, which includes a
correction for large temperature gradients at the wetted surface:
0.716
K T 
hDB = 0.0259 He Re0.8 Pr 0.4  he  . (25)
Dh  Ts 
The steady-state forced-flow heat transfer for a common Reynolds number in cable cooling
pipes (Reynolds in the range of a few 104 to 105) is usually of the order of 1000 W·m–2·K–1. A large
temperature difference, of several degrees kelvin, between the heated surface and the bulk causes an
appreciable degradation of the above values.
As for the boiling conditions, strong variations of the heat transfer are observed during fast
transients. Giarratano [12] and Bloem [13] measured the transient heat transfer to a forced flow of
supercritical helium in dedicated measurements on short test sections (see Fig. 19).
419
L. B OTTURA
Fig. 19: The transient heat transfer coefficient in supercritical helium, measured by Bloem [13]. (Reprinted from
Ref. [13] with kind permission from Butterworth–Heinemann Journals, Elsevier Science Ltd, The Boulevard,
Langford Lane, Kidlington OX5 1GB, UK.)
The experiments showed an initial peak in the heat transfer coefficient at early times, below 1
ms. At later times, in the range of some milliseconds to about a hundred milliseconds, the initial peak
decreased approximately with the inverse of the square root of time. This behaviour can be explained
in terms of the diffusion of heat in the thermal boundary layer. Using the analytical solution of
diffusion in a semi-infinite body (the helium) due to a heat flux step at the surface, the effective heat
transfer coefficient can be computed as (Bloem [13])
1 π K he ρ he che
hBLQ = , (26)
2 t
where Khe is the heat conductivity of helium. The above expression is shown to fit the experimental
data properly for times longer than a millisecond and until the thermal boundary layer is fully
developed. At early times, Eq. (26) would tend to predict an exceedingly high heat transfer coefficient,
consistent with the assumptions of the analytical calculation. In reality, the early values of h are found
to be limited by the Kapitza resistance [14] at the contact surface of the strand, which gives a
significant contribution only when the transient heat transfer coefficient is of the order or larger than
10 kW·m–2·K–1 (or in the case that the wetting helium is in the superfluid state as discussed later). At
later times, usually around 10–100 ms, the thermal boundary layer is fully developed and the steady-
state value of h is approached. An empirical expression for the heat transfer during the transient,
describing the transition from transient to steady-state conditions, can be obtained as follows:
h = max {hBLQ , hDB } , (27)
giving good agreement with the experimental results, and showing how for short pulses the heat
transfer coefficient only depends on the helium state and not on the flow conditions.
During the flow transients generated by the heating-induced flow, the two processes are
combined; that is, the boundary layer changes in thickness during the thermal diffusion process.
Experimental measurements in these conditions, and in particular on transient heat transfer over long
lengths, pose some significant problems and results are so far not available. This issue is important, as
increased turbulence in the flow can contribute to the stability margin. It is not clear whether the
phenomenon has a local nature or depends on the heated length and the time-scales involved in the
establishment of the expulsion of helium from the normal zone.
420
C ABLE S TABILITY
5.3 Superfluid helium

Helium undergoes a quantum transition, very similar to the phenomenon of superconductivity, when it
is cooled below the so-called lambda point – that is, 2.17 K – at ambient pressure. In this state it
becomes superfluid helium, characterized by very low viscosity and an exceedingly high thermal
conductivity that allows removal of heat at high rates both at the surface interface with solid materials
(e.g. a superconducting strand) as well as in the fluid bulk. For this reason, superfluid helium is used
as a coolant in high-performance magnetic systems based on Nb–Ti or Nb3Sn, where the low
operating temperature is used to boost the critical current density. An additional advantage is that the
thermal conductivity of a superfluid helium bath can be used to evacuate the heat loads in the
magnetic system without requiring fluid convection. The high heat transfer rate of superfluid helium
can be described in relevant conditions by an internal convection of two fluid species, a normal
component and a superfluid component that carries no entropy. The two species move in counter-flow
to maintain the total density, thus also achieving net energy transport.
In the case of normal helium considered in the previous sections, heat transfer is controlled in
most situations by the vapour film (boiling heat transfer) or by the thermal resistance of the boundary
layer (convection). In superfluid helium, on the contrary, the heat fluxes that can be sustained in the
bulk fluid are so large that the thermal resistance at the solid/fluid interface becomes important. The
dominant mechanism for the heat resistance at the interface is the mismatch in the propagation of the
phonons. This interface thermal resistance is usually called the Kapitza resistance, and is in principle
present at all operating temperatures. The heat flux across the Kapitza resistance can be roughly
approximated by the following expression:
(
′′= σ Tsn − Then ,
qKapitza ) (28)
where the exponent n is in the range of 3–4. The constant σ depends on the nature and state of the
material, and for a value of n = 4 its value can range from 200 to 400 W·m–2·K–4. With this choice of
n, the equivalent heat transfer coefficient at the surface is then given by
( )( )
hKapitza =σ Ts2 + The2 Ts + The . (29)
Typical values of the equivalent heat transfer coefficient are plotted in Fig. 18 for comparison
with the other heat transfer regimes. At low temperature (below 2 K), the Kapitza resistance is
relatively large, corresponding to a heat transfer coefficient in the range of 5 kW·m–2·K–1, and is
usually the main limit for heat transfer. Already at 4.2 K, though, the equivalent heat transfer
coefficient is extremely high, above 50 kW·m–2·K–1, thus explaining why the Kapitza resistance is
generally not a limiting factor in heat transfer to normal helium.
6 Stabilization strategies
6.1 Adiabatic stabilization

The lower limit for the energy margin of a superconductor can be obtained by considering that the
cable responds adiabatically to the external energy input. This is the case in the absence of a cryogenic
cooling fluid and whenever the volume affected by the external energy input is so large that heat
conduction at the boundary of the normal zone can be neglected. The absence of cooling by a stagnant
or flowing cryogenic fluid, most frequently liquid helium, is typical of small windings, either wound
from a dry superconductor or impregnated with organic resins. Dry or impregnated windings
conductively cooled using a cryocooler as a heat sink are becoming increasingly attractive due to the
appeal of cryogen-free operation. Cooling happens in this type of winding on a time-scale that is much
longer than the time of relevance for discriminating between recovery and thermal runaway. Hence for
421
L. B OTTURA
these magnets, the cooling term is absent in the heat balance. In this case, the energy balance
simplifies to the following:
∂T
′′′ + qJ′′′ .
C = qext (30)
∂t
As expressed by Eq. (18), any excursion of the superconductor above the current sharing
temperature Tcs will cause the appearance of Joule heating, resulting in an inevitable thermal
runaway. 2 In this case, the current sharing temperature provides the boundary between recovery and
thermal runaway. The stability margin corresponds to the energy necessary to increase the
superconductor temperature from operating conditions to Tcs. Within the approximations considered so
far, this can be calculated by integrating Eq. (30), taking into account that below Tcs the Joule heating
is zero:
∞ Tcs
∫ qext′′′ dt =
0
∫ CdT .
Top
(31)
The integral on the left-hand side of Eq. (31) corresponds to the energy margin, while the
integral on the right-hand side is the difference in the volumetric specific enthalpy between the
operating temperature and the current sharing temperature. We can thus write that for an adiabatic
superconductor the energy margin is as follows:
′′′ H (Tcs ) − H (Top ) ,
∆Q= (32)
with the definition

T
H (T ) = ∫ C (T ′ ) dT ′ . (33)
0
For this reason, the mechanism underlying adiabatic stability, namely the heat capacity of the
superconductor, is also referred to as enthalpy stabilization.
In order to estimate the orders of magnitude of the energy margin in adiabatic conditions, it is
necessary to examine the typical values of the heat capacity and the specific volumetric enthalpy of the
typical materials present in a superconducting winding. This is done in Figs. 20 and 21, respectively.
The functional dependence of the volumetric heat capacity is different among pure metals (copper,
aluminium), alloys (stainless steel), superconductors (Nb–Ti and Nb3Sn), and organic composites
(resin and typical insulators) because of the different weights of the electronic and phonon
contributions to the specific heat. However, in spite of the large range of values, a general feature
exhibited by all materials is the decrease of the specific heat approaching absolute zero. The
consequence is that the enthalpy difference for a given temperature margin ∆T is smaller at lower
temperatures, as can be inferred by taking the volumetric specific enthalpy difference for a fixed
temperature interval from Fig. 21. This effect is particularly important, nearly one order of magnitude,
when considering operation in superfluid helium, at 1.8–2.0 K, as compared to operation in
atmospheric pressure liquid helium, at 4.2 K.
2
Strictly speaking, in the case of a practical superconductor with a voltage–current characteristic described by the power-law
relation, Joule heating is always positive, although small, even below current sharing. In this case, the superconductor would
never be stable, even in the absence of an external energy input. The steady-state Joule heating is, however, small and is
removed by the cooling system – which is inevitably present – that acts on times much longer than the time-scales of interest
for stability, but sufficient for maintaining the steady-state operating temperature.
422
C ABLE S TABILITY
Fig. 20: The volumetric heat capacity for typical materials used in low-temperature superconducting magnets
Fig. 21: The volumetric specific enthalpy for typical materials used in low-temperature superconducting
magnets, obtained by integrating the data of Fig. 20.
The adiabatic energy margin can be estimated using the data in Fig. 21, by adding the individual
contributions to the heat capacity of all materials participating in the temperature excursion. We can
do this by taking the weighted average of the volumetric specific enthalpy:
∆Q′′′
= ∑ f ( H (T ) − H (T ) ) ,
i i cs i op (34)
where the fi is the materials fraction of the ith component characterized by the specific volumetric
enthalpy Hi.
The energy margin in Eq. (34) depends on the operating temperature Top and on the operating
field Bop and current Iop through the current sharing temperature. It is hence possible to scan the energy
margin over the whole operating space for a given superconductor once the material fractions are
fixed. A typical strand for low-temperature applications has a stabilizer to superconductor ratio in the
range of 1.0–1.5. As we have made the assumption that the external energy input is distributed over a
large volume of the winding, we also include the electrical insulation in the calculation. Typical
fractions of materials for impregnated windings are in the range of 30% superconducting material
(fsc = 0.3), 40% stabilizer (fst = 0.4), and 30% insulation (fin = 0.3).
In Fig. 22, we report the result of the calculation of Eq. (34) for Nb–Ti and Nb3Sn as
superconducting material operating initially at a temperature of 4.2 K. The calculation has been
423
L. B OTTURA
performed for several values of the operating fields, and is plotted as a function of the operating ratio
of the critical current Iop/Ic. We can see at once the clear difference between the two materials, due to
the fact that Nb3Sn has a higher critical temperature and thus exploits a region of higher heat capacity
for stabilization. Also, it is clear that approaching the limits of performance of the conductor (8–10 T
for Nb–Ti and 16–20 T for Nb3Sn), the adiabatic energy margin for an efficient use of the
superconductor (operating fraction of the critical current around 0.8) becomes extremely small
(below 1 mJ·cm–3) and the magnet will not be able to withstand even the smallest perturbation without
quenching.
Fig. 22: The adiabatic energy margin for a typical winding, with 30% superconductor, 40% copper, and 30%
insulation, wound either using Nb–Ti (left) or Nb3Sn (right). The calculation has been performed with an initial
operating temperature of 4.2 K, for different values of the operating field (steps of 2 T between the curves) and
as a function of the ratio of the operating to the critical current.
Another interesting result can be obtained by performing the above calculation for different
initial operating temperatures. It is particularly instructive to examine the effect of reducing the
operating temperature, which is often done to raise the critical current density in the hope of increasing
the operating field of a magnet. The results of the calculation at 4.2 K, discussed in Fig. 22, are
compared in Fig. 23 to the results obtained for an initial operating temperature of 1.8 K. Because a
temperature change affects the critical current, to allow a direct comparison the curves are plotted as a
function of the operating current density in the superconductor. We see from the comparison of the
adiabatic energy margin that sub-cooling has essentially no effect at low field, as in any case the
dominant contribution to the energy margin is due to the heat capacity in close proximity to Tcs.
It is only at high field (8–10 T for Nb–Ti and 16–18 T for Nb3Sn) that the effect of sub-cooling
becomes appreciable. However, the order of magnitude of the absolute gain in stability is at most a
few mJ·cm–3, and at best comparable with the expected mechanical and electromagnetic perturbations
in a typical magnet.
Fig. 23: The adiabatic energy margin for a typical winding, with 30% superconductor, 40% copper, and 30%
insulation, wound either using Nb–Ti (left) or Nb3Sn (right), and computed for initial operating temperatures of
4.2 K and 1.8 K. The calculation has been performed for different values of the operating field (steps of 2 T
between the curves) and as a function of the operating current density to allow direct comparison of the results.
424
C ABLE S TABILITY
For the above reasons, enthalpy stabilization is not sufficient to make the best use of the current-
carrying potential of superconductors, and other means have been devised to cope with the
perturbation spectrum, especially in large magnet systems.
6.1.1 The adiabatic stability of an MRI magnet

One of the most widespread and well-known present-day applications of superconductivity is in the
magnets developed for Magnetic Resonance Imaging (MRI). These magnets are solenoids with a very
good field homogeneity and a large bore, as large as 1 m in diameter and 1.5 m long, to allow full-
body scans of human beings. Typical field levels in the bore of the solenoid are at present in the 1–2 T
range. For cost and maintenance reasons, these magnets are built with high operating current densities,
and with little or no cryogen in the winding pack. They are essentially adiabatic, and therefore they
must be carefully designed to avoid training and quenches.
A typical MRI magnet winding pack is subdivided into a series of thin coaxial, possibly nested
solenoids for shielding, which produce the field and correct for winding and geometrical errors. To
obtain a good field homogeneity, the winding geometry must be tightly controlled and carefully
maintained. In addition, the contribution from the magnetization of the superconductor must be
minimized. Because of these requirements, MRI magnets are generally wound from single wires, with
medium-size superconducting filaments. The winding pack is impregnated so that it forms a single
rigid unit and the wires are constrained in position. Cooling is indirect, by conduction through the
winding pack. The thin winding pack allows heat removal under a small temperature gradient.
Fig. 24: Typical Nb–Ti strands for superconducting MRI magnets
A typical MRI magnet is wound with Nb–Ti wire, well adapted to the field range required, of
the types shown in Fig. 24. The wire has a high copper/Nb–Ti ratio, of the order of 5–10, mostly for
protection reasons because of the large inductance of the coil. Wires for MRI magnets are produced
with a round or rectangular shape (to ease winding), and have external dimensions of the order of 1–
2 mm. The Nb–Ti filaments have a typical diameter of 50 µm. At low field, they are delivered with a
guaranteed critical current density in the Nb–Ti cross-section in the range of 5000 A·mm–2. These
must be compared to operating current densities in the range of 200 A·mm–2 in the strand; that is, of
the order of 1000–2000 A·mm–2 with reference to the Nb–Ti cross-section. We can see at once that
MRI magnets are designed with large operating margins to increase their reliability. Still, additional
care is necessary.
In the adiabatic case, we can estimate the stability margin as the enthalpy of the wire from
operating conditions to the current sharing temperature. To do this we take in account the contribution
of copper, the main component in the wire, and Nb–Ti, and we use the diagram reported in Fig. 21.
Assuming an operating temperature of 4.2 K, and a current sharing temperature of 7.5 K (consistent
with the operating current and field range given above), the 3.3 K temperature margin corresponds to
an enthalpy change of 6.14 mJ·cm–3 for copper and 46.7 mJ·cm–3 for Nb–Ti. The total adiabatic
energy margin is obtained using the weighted sum of Eq. (34) and gives approximately 13 mJ·cm–3.
This value is larger than our estimate of the energy release due to movement of the conductor, but
does not leave much contingency to cope with uncertainties in the actual temperature margin and other
possible energy inputs. Resin impregnation of the winding pack, as mentioned previously, is a
425
L. B OTTURA
common practice to avoid movements and thus to minimize energy release through wire motion. Note
that cracking of the impregnation resin during cool-down and energization – resins undergo a large
thermal contraction from room temperature to 4.2 K, but have little tensile strength at cryogenic
temperatures – can also be a source of localized energy release. This is generally avoided by filling
void volumes in the winding pack with fillers and fibre cloths or ropes that increase the tensile
strength.
6.2 Cryostability
Early superconducting coils had a wide spectrum of large perturbations, significantly above the
summary presented in Fig. 5. This was either because the strands and tapes used were prone to flux
jumping, or because the mechanical design was not adapted to avoid movements, slips, insulation
cracks, and the associated energy releases during energization. The adiabatic energy margin, as
discussed in the previous section, was by no means sufficient to accommodate the energy
perturbations. A small and localized normal zone had, in addition, no chance of recovering, because
the Joule heating of the superconducting material in normal state was extremely high, and therefore
the coils quenched prematurely. Based on this observation, Krantowitz and Stekly [15] and Stekly and
Zar [16] added a high electrical conductivity shunt backing the superconductor, a pure copper
stabilizer, and exposed this material to a liquid helium bath of large volume and thus constant
temperature. The effect was dramatic, improving the performance of the magnet and paving the way to
large-sized applications of superconductivity.
This result was achieved due to two beneficial effects: on the one hand, the Joule heat
generation in the case of transition was largely decreased by the stabilizer; while on the other, the
superconductor was efficiently cooled through boiling heat transfer. Cryogenic stabilization, or
cryostability, was achieved when the steady-state composite temperature that would be attained with
the full operating current flowing in the stabilizer was below the critical temperature of the
superconductor. In this case the initial normal zone, caused by an arbitrary energy source, would
shrink and eventually disappear.
The cryostability condition can be best understood by considering the 1D heat balance of
Eq. (7) again. As the cryostability condition applies independent of the length of the conductor
subjected to an energy perturbation, we can neglect the heat conduction term, and we obtain the
following:
∂T wh
C
∂t
′′′ + qJ′′′−
= qext
A
(T − Top ) , (35)
where we have made use of the fact that the helium temperature remains constant, equal to the initial
operating temperature. To achieve cryostability, we are seeking the condition for which, in steady
state, following the end of the energy pulse, the heat generated by the normal zone is equal to or less
than the heat removal at its surface. The cryostability condition is hence obtained when
wh
qJ′′′ ≤
A
(T − Top ) . (36)
For the Joule power, we take the linear approximation of Eq. (18), which reaches the maximum
given by Eq. (15) when the superconductor is at the critical temperature. Assuming for the moment a
constant heat transfer coefficient, corresponding to a linear increase of the heat flux with temperature,
we obtain that the cryostability condition is given by
ηst
I op2 ≤ h (Tc − Top ) . (37)
wAst
426
C ABLE S TABILITY
Cryostable operation is obtained when Eq. (37) is satisfied, while in the event of higher
generation or lower cooling than implied by Eq. (37), the superconductor is not cryostable. To class
the mode of operation, Stekly has introduced a dimensionless coefficient α, defined as follows [16]:
ηst I op2
α= . (38)
hwAst (Tc − Top )
Operation is cryostable for α ≤ 1, and the maximum operating cryostable current, IStekly, is
given by
hwAst (Tc − Top )
IStekly = . (39)
ηst
The cryostability condition expressed by Eqs. (37) and (38) has a simple graphical interpretation
shown in Fig. 25, obtained by tracing the heat generation and the heat removal per unit of cooled
conductor surface as a function of operating temperature. In the case of the curve marked ‘(a)’ in the
plot, the operation is cryostable, as at any point, the heat generation is less than the heat removal. The
limiting condition is reached with the curve marked ‘(b)’, where the heat generation exactly matches
the cooling at the critical temperature. Cryostability is violated in the case of the curves marked ‘(c)’
and ‘(d)’, for which a sufficiently large perturbation will raise the superconductor temperature to a
region where the heat generation exceeds the cooling, thus leading to an instability. Note that once the
cooling condition has been selected, it is the heat generation curve that varies with the design changes,
while the heat flux to the helium is a constant.
Fig. 25: A graphical interpretation of the cryostability conditions in the case of a constant heat transfer
coefficient between the superconductor and the coolant. The graph has been obtained by plotting the heat
generation and the removal per unit of a cooled conductor surface. Several heat generation values have been
plotted: (a) in the cryostable region, (b) at the cryostable limit, and (c) and (d) not cryostable.
In reality, the heat transfer depends strongly on the cooling conditions, as discussed earlier. In
the case of boiling helium, the typical heat flux is shown in Fig. 17. The equivalent heat transfer
coefficient h is non-linear, with initially high values (typically 1000–10 000 W·m–2·K–1) in the
nucleate boiling regime, and drops to a minimum of the order of 500 W·m–2·K–1 at the onset of film
boiling. The cryostability condition formulated above is fulfilled when, under any conditions, the heat
removal exceeds the heat generation; that is, when the maximum possible heat generation is less than
the minimum possible heat removal. This situation is shown in Fig. 26 for the curves marked (a) and
(b). The equivalent Stekly criterion in the case of variable heat transfer is obtained taking for h the
minimum value along the boiling curve. As for the case of Fig. 25, curves (c) and (d) in Fig. 26 violate
the cryostability condition. Note, however, that intermediate stability conditions at higher heat
generation values than allowed by cryostability could exist. Taking as an example curve (c) in Fig. 26,
427
L. B OTTURA
we see that under small perturbations (a small temperature increase), the heat removal is still larger
than the generation. A conductor operating in this condition would therefore recover from sufficiently
small energy inputs, but it would be unstable for large enough energy depositions.
Fig. 26: A graphical interpretation of the cryostability conditions in the case of boiling heat transfer between the
superconductor and the coolant. Cryostability is obtained for curves (a) and (b), but not for the generation
curves (c) and (d).
6.2.1 Cryostability: the BEBC magnet

Cryostable magnets were among the first to be built soon after the formulation of this principle, in the
early 1970s. A dramatic example was the Big European Bubble Chamber (BEBC) at CERN [17], a
4.7 m bore split solenoid with a 0.5 m gap, producing a maximum field in its centre of 3.5 T,
corresponding to a maximum field at the conductor of 5.1 T, and storing an energy of 800 MJ (see
Fig. 27).
Fig. 27: One of the two coils forming the BEBC split-solenoid magnet at CERN
Each coil was wound in 20 pancakes out of a flat monolithic conductor with a thickness of
3 mm and a width of 61 mm, schematically shown in Fig. 28. This conductor was itself a composite,
containing 200 untwisted Nb–Ti filaments with a diameter of about 200 µm in an OFHC copper
matrix. The conductor had a total Nb–Ti area of approximately 6.5 mm2, and about 176.5 mm2 of
copper cross-section. The nominal operating current of the conductor was 5700 A, corresponding to an
operating current density in the composite of about 30 A·mm–2. Adjacent conductors in a pancake
were separated by insulation and by a copper spacer that allowed helium to wet the outer surface of the
composite. Only one broad face of the composite was wetted (the other face was pressed against the
insulation and a reinforcing steel strip), thus resulting in a wetted perimeter of 61 mm.
428
C ABLE S TABILITY
Fig. 28: A schematic view of the composite structure of the BEBC conductor
In order to estimate the cryostability condition, we take for boiling helium cooling, at 4.2 K, the
average characteristics derived from Fig. 17 and given by Eqs. (23) and (24). The minimum heat
transfer coefficient is of the order of 600 W·m–2·K–1. In a field of 5 T, copper has an electrical
resistivity of approximately 3.4 × 10–10 Ω·m, while Nb–Ti has a critical temperature of the order of
7.4 K. With these values, the Stekly parameter α is 0.55; that is, the conductor operates largely in the
cryostable regime. This situation is shown in Fig. 29, which compares the heat removal and heat
generation.
Fig. 29: A plot of the heat generation and heat removal per unit cooled surface of the BEBC conductor in
nominal operating conditions. The conductor is cryostable, as in all possible conditions the heat removal largely
exceeds the heat generation.
Indeed, the BEBC coil could be energized up to the operating current without problems, in spite
of the fact that the Nb–Ti filaments were larger than the maximum size that was stable against a flux
jump. In fact, because the filaments were not twisted in the composite, even larger magnetization was
associated with the currents that flowed in the electromagnetically coupled filaments, excited by the
field ramp. Owing to the cryostable operating regime, it was possible to suppress the large
magnetization produced by these coupling currents using heaters that temporarily quenched the
conductor. The conductor recovered as soon as the heaters were switched off, a rather bizarre use of
cryostability.
429
L. B OTTURA
6.3 Cold-end recovery

So far, all discussions have concerned a portion of superconductor long enough for all end effects to
be neglected. In reality, perturbations happen often over finite lengths. We should hence expect that
the conditions at the end of the resulting normal zone could help in cooling and thus provide additional
stability. In these conditions, stability analysis becomes a complex matter. Nevertheless, a very simple
and elegant treatment has been found by Maddock, James, and Norris [18], which identifies steady-
state equilibrium conditions taking into account the effect of heat conduction along the
superconductor. The situation examined is the case of a superconducting wire with a normal zone in
the centre at a temperature Teq, exchanging heat with a helium bath at constant temperature and
sufficiently long that the two ends of the wire are at equilibrium temperature Top with the coolant. Heat
is transported by conduction from the normal zone (the warm end) to the extremity (the cold end). The
temperature distribution along the wire is shown in Fig. 30, together with the corresponding heat
generation and removal. Only one half of the wire is plotted because of the assumed symmetry.
Fig. 30: The temperature distribution and corresponding Joule heat generation and cooling plotted as a function
of the length for a superconducting wire with a normal zone in the centre. Only half of the wire length is shown
because of symmetry.
The heat balance in this condition is given by Eq. (7), where in principle all terms must be
retained. If we look for the equilibrium condition, however, the heat balance simplifies to
∂  ∂ T  wh
qJ′′′+ k − (T − The ) =
0. (40)
∂ x  ∂ x  A
Maddock and co-workers introduced a new variable S, defined as

∂T
S =k , (41)
∂x
which represents the heat flux along the superconductor. S has the property that
∂S ∂S ∂ T S ∂S
= = . (42)
∂x ∂T ∂ x k ∂T
430
C ABLE S TABILITY
We can substitute the relation (41) in the steady-state heat balance of Eq. (40) and obtain the
following relation:
 wh  ∂S
k  (T − The ) − q′′′J  =
S . (43)
 A  ∂T
Equation (43) can be integrated directly, yielding the following integral relation between heat
generation by Joule heating and cooling:
Teq Seq
 wh 
∫T k  A (T − The ) − q′′′J  =
S
∫ SdS . (44)
op op
If we now make the assumption that the normal zone is sufficiently long to have reached the
equilibrium condition, so that Teq is given by
AqJ′′′
T=
eq Top + , (45)
wh
then the heat conduction will be zero both at the warm and at the cold end, and the relation between
heat generation and cooling (for constant heat conductivity) will be simply
Teq
 A 
∫ h (T − T ) − w q′′′ =
Top
he J 0. (46)
Equation (46) is the so-called equal-area theorem, which states that for equilibrium, no net area
should be enclosed between the heat generation and cooling curves plotted as a function of
temperature. This very interesting result can be examined graphically on the same representation used
in Figs. 25 and 26 to determine cryostability, and shown in Fig. 31 for the two cases of linear and
boiling heat transfer. The point at temperature Teq corresponds to the intersection of the generation and
cooling curves. The generation curve reported, although not cryostable, is still an equilibrium
condition, as the area enclosed between generation and cooling is zero. The excess heat generation in
the normal region is compensated by excess cooling in the superconducting region. Heat conduction
functions as the vector of this heat from one region to the other.
Fig. 31: A graphical interpretation of the equal-area condition, in the case of a constant heat transfer coefficient
between the superconductor and the coolant (left) and in the case of boiling heat transfer (right). A
superconductor characterized by the heat generation curves plotted is stable, although the cryostability condition
is violated, as the net area enclosed between the generation and the cooling is zero.
For the linear heat transfer case, with constant h, it is possible to determine the value of the
operating current that corresponds to the equal-area condition. To do this, we note that the equilibrium
condition corresponds to the situation in which we have
Teq − Tc = Tcs − Top , (47)
431
L. B OTTURA
which derives from the similarity of the two shaded triangles in Fig. 31. From Eq. (45), we also have
that the equilibrium temperature is given by
ηst I op2
T=
eq Top + . (48)
whAst
We can combine Eqs. (47) and (48) to obtain the operating current IMaddock corresponding to
stable operation under the equal-area condition:
hwAst (Tc − Top ) + (Tcs − Top ) 
I Maddock = . (49)
ηst
As expected, this current is higher than the cryostable current IStekly given in Eq. (39), implying
that the superconductor can be used more effectively than when limited by cryostability. The value of
the Stekly parameter for operation in the equal-area condition is given by
α = 1+
(T − T ) ,
cs op
(50)
(T − T )
c op
which approaches a value of 2 for operation at a small fraction of the critical current (when Tcs ≈ Tc).
In reality, the temperature variation of thermal conductivity and the non-linear character of the heat
transfer can cause significant variations from the above limits, with a net effect that in general is
towards a higher stable operating current and a corresponding Stekly parameter.
6.3.1 The equal-area condition for the BEBC magnet

In the case already examined above for the BEBC solenoid, we have computed the value of the
nominal heat generation and cooling, and verified that the BEBC magnet was operating in the
cryostable regime. It is possible, using the same assumptions, to estimate the maximum operating
current that could have been achieved in steady-state stable conditions as dictated by the equal-area
theorem. Of course, a change in operating current would result in a simultaneous change in the
magnetic field produced, of the critical properties of the Nb–Ti superconductor and of the resistivity of
the copper. Linearizing the properties in the vicinity of the nominal working point, we make the
following approximations:
magnetic field at the conductor: B=

op 0.9 ×10−3 I op [T ] ,
 Bop − 10.8 
critical temperature: Tc 7.4 1 −
=  [K] ,
 5.1 − 10.8 
 Bop − 10.8 
critical current: =I c 13 000 1 −  [A] ,
 5.1 − 10.8 
from which it is possible to compute the current sharing temperature in any working conditions.
432
C ABLE S TABILITY
Fig. 32: The equal-area condition for the BEBC magnet as compared to the cryostable operating condition
discussed earlier.
We can now take the same heat transfer curve as used in Fig. 29 and vary the current Iop
parametrically until the equal-area condition is reached. This situation is shown in Fig. 32. The
corresponding value of the operating current is approximately 7650 A, for an operating field of 6.84 T.
Ideally, this should have been the upper limit for stable steady-state operation of the BEBC magnet.
The equal-area condition guarantees stable operation at a current above the cryostability limit
even in the case of a large portion of the winding going normal. This allows the designer to increase
the allowable operating current density with a beneficial reduction in the amount and cost of
superconductor and stabilizer used for construction of the coil. In fact, once established, the
temperature profile, which obeys the equal-area theorem, is stable. If we take the example of the
central point, and we refer to the generation and cooling fluxes plot of Fig. 31, a small temperature
increase will cause the cooling to become larger than the generation and the conductor will evolve
back to the equilibrium temperature. For a small temperature decrease, the cooling will be less than the
generation and the temperature will increase back up to the balance point. In the final analysis, the
equal-area theorem guarantees that the coil will not suffer from thermal runaway, whatever the energy
input, 3 but, as for the cryostability condition, it does not quantify the energy margin of the conductor.
Because, in addition, both the cryostability and the equal-area conditions apply to a long length of
superconductor initially brought into normal conditions, the energy margin for a conductor operating
in these conditions is in practice much larger than any perturbation expected during operation; that is,
infinite from an engineering point of view.
6.4 Well-cooled operation of CICCs

We have assumed so far that the cooling takes place in a helium bath, providing an ideally infinite heat
sink. For some applications, it is advantageous to cool the superconductor using a forced flow of
helium, in which case the amount of helium available for stabilization is no longer infinite. Various
superconductor configurations have been developed around this concept, of which the most successful
from the stability point of view is the Cable-in-Conduit Conductor (CICC). The development of
CICCs was largely motivated by the observation that cryostable pool boiling magnets (i.e. satisfying
the Stekly criterion) are known to have a low operating current density, and thus a large size and cost.
It was also clear, however, that large-sized magnets operating in noisy mechanical or electromagnetic
environments (e.g. operating in rapidly changing magnetic fields or subject to significant stress cycles)
3
In reality, for large enough energy inputs, the temperature of the superconductor can become sufficiently large that the
stabilizer resistivity, assumed constant so far, starts to increase sensibly. This condition, however, requires large energies and
is not relevant to our discussion.
433
L. B OTTURA
require a minimum energy margin to withstand typical perturbations that cannot be absorbed
adiabatically in the small heat capacity of the conductor.
Helium is the only substance known to have a large heat capacity at low temperature. This is
shown in Fig. 33, which presents the volumetric heat capacity (the product of density and specific heat
at constant pressure) for different values of pressure in the supercritical regime. Comparing the values
of Fig. 33 to those for solid materials in Fig. 20, we see that helium can provide a heat capacity two to
three orders of magnitude larger than the solid materials in the range of 4–10 K that is typical for low-
temperature superconductors. The volumetric enthalpy is shown in Fig. 34 and also demonstrates, by
comparison with Fig. 21, the large heat sink that could be provided by helium.
Fig. 33: The volumetric heat capacity of helium at different pressures: the peak corresponds to the crossing of
the pseudo-critical line in the supercritical regime.
Fig. 34: The volumetric specific enthalpy for helium at different pressures in the supercritical regime
The key idea behind the CICC was hence to give access to the large heat sink represented by the
amount of helium in the cooling circuit of the cable, thus increasing substantially the adiabatic energy
margin discussed previously. At the same time, the aim was to increase the heat transfer from the
superconductor to the helium, so that the cryostability limit would be pushed to higher current
densities.
The CICC concept evolved from the Internally Cooled Superconductor (ICS), which had found
applications in magnets of considerable size in the late 1960s and early 1970s (see, in particular, the
434
C ABLE S TABILITY
work of Morpurgo [19]). In an ICS, the helium is all contained in the cooling pipe, very much like
standard water-cooled copper conductors. The conductor can be wound and insulated using standard
technology, and the magnet is stiff both mechanically and electrically, a considerable advantage for
medium and large-size systems requiring, with the increasing amount of stored energy, high discharge
voltages. Control of the heat transfer and cooling conditions is achieved using supercritical helium,
thus avoiding the uncertainties related to a flowing two-phase fluid. A major drawback of this concept,
however, was the fact that in order to achieve good heat transfer (and thus stability and a high
operating current density), the helium would theoretically have to flow in the early ICS layouts at
astronomical flow rates. The advantage of the increase of the wetted perimeter obtained by subdivision
of the strands was already clear at the beginning of the development of the ICS (Chester [1]). Hoenig
[20–22] and Dresner [23–25] developed models for the local recovery of an ICS after a sudden
perturbation, where they found that for a given stability margin, the mass flow required would be
proportional to the 1.5th power of the hydraulic diameter. This consideration finally brought Hoenig,
Iwasa, and Montgomery [20, 21] to present the idea for the first CICC prototype, shown in Fig. 35.
Fig. 35: The original concept of the CICC, as presented by Hoenig et al. [21]. (Reproduced from [21] by
permission of Servizio Documentazione CRE–ENEA Frascati. Copyright 1975 CRE–ENEA Frascati.)
Although many variants have been considered, the basic CICC geometry has changed little
since. A bundle conductor is obtained by cabling superconducting strands, with a typical diameter in
the millimetre range, in several stages. The bundle is then jacketed; that is, inserted into a helium-tight
conduit that provides structural support. Supercritical helium flows in the conduit, within the
interstitial spaces of the cable. With the cable void fractions of about 30–40% that are commonly
achieved, the channels have an effective hydraulic diameter of the order of the strand diameter, while
the wetted surface is proportional to the product of the strand diameter and their number. The small
hydraulic diameter ensures a high turbulence, while the large wetted surface achieves high heat
transfer, so that their combination gives the known excellent heat transfer properties.
Strictly speaking, although it can satisfy the Stekly criterion (see later) a CICC cannot be
considered as cryostable, because the amount of helium available for its stabilization (which represents
the dominant heat capacity) is in any case limited to the volume in the local cross-section. The
consequence is that a large enough energy input will always cause a quench, a behaviour that Dresner
[25, 26] defines as meta-stable. Rather, the question concerns the magnitude of the energy margin
∆Q’’’ for a given configuration and operating condition. In the initial studies, the energy input was
thought to happen suddenly, and initial experiments and theory concentrated on this assumption.
Throughout this section, we will extend the definition to an arbitrary energy deposition time-scale.
Finally, in spite of the fact that the cryostability concept does not apply to CICCs, we will see that the
Stekly criterion, in its original form of a power balance at the strand surface, still plays a fundamental
role in its stability.
Stability in CICCs is different from the theories discussed so far, for the following reasons:
− the largest heat sink providing the energy margin is the helium, and not the enthalpy of the
strands themselves or conduction at the end of the heated length;
435
L. B OTTURA
− this heat sink is limited in amount;

− finally, the helium behaves as a compressible fluid under energy inputs from the strands,
implying additional feedback on the heat transfer coefficient through heating-induced flow.
As a consequence, two of the main issues in CICC stability are the heat transfer from the strand
surface to the helium flow and the thermodynamic process in the limited amount of helium.
Measurements of the stability margins of CICCs started early in their history [27–31]. The
original idea of reducing the necessary flow in order to obtain the desired stability margin was
frustrated as soon as the first experimental data were obtained: the stability margin was largely
independent of the operating mass flow, as was recognized by Hoenig [28, 29] (see the results reported
in Fig. 36), and soon duplicated by Lue and Miller [31]. These results indicated that the heat transfer at
the wetted surface of the strands during a temperature excursion was only weakly correlated to the
steady-state mass flow and the associated boundary layer. In later experiments, Lue, Miller, and
Dresner [32, 33] could observe multiple stability regions, both as a function of the operating current
and of the operating mass flow (a typical stability margin showing the dual behaviour curve is shown
in Fig. 37).
As discussed by Dresner [34] and Hoenig [30], during a strong thermal transient the heat
transfer coefficient h at the strand surface changes mainly for two reasons (see also the earlier
discussion on heat transfer): (a) thermal diffusion in the boundary layer (a new thermal boundary layer
is developed and thus h increases compared to the steady-state value); and (b) induced flow [35] in the
heated compressible helium (associated with increased turbulence and thus again an increase in h).
The concurrence of these two effects explains the weak dependence of ∆Q’’’ on the steady mass flow
and (at least qualitatively) the multivalued stability behaviour for different pulse powers.
The typical behaviour of the energy margin in CICCs was found through measurements to be a
function of the operating current (see the vast amount of data presented in Refs. [36–41]). Such
behaviour is shown schematically in Fig. 38. For a sufficiently low operating current, a region with a
high stability margin – termed here, after Schultz and Minervini [42], the well-cooled operational
regime – is observed. In this regime, the stability margin is comparable to the total heat capacity
available in the local cross-section of the CICC, including both the strand material and the helium,
between the operating temperature Top and the current sharing temperature Tcs. With increasing
current, a fall in the stability margin to low values, the ill-cooled regime, is found. In this regime, the
stability margin is lower than in the well-cooled regime by one to two orders of magnitude and
depends on the type and duration of the energy perturbation.
Fig. 36: The energy margin of a Nb–Ti CICC and a Nb3Sn CICC as a function of the steady-state helium flow,
measured by Hoenig et al. [28] (reproduced from [28] by permission of the IEEE. Copyright 1979 IEEE).
436
C ABLE S TABILITY
Fig. 37: The energy margin of a Nb–Ti CICC as a function of the operating current, measured by Lue et al. [33].
The experiment was performed on a single triplex CICC of length Lsample = 3.8 m, with a strand diameter
φw = 1 mm, under zero imposed flow (vHe) at a helium pressure of pabs = 5 bar. The background field was
B = 6 T, and resistive heating took place in τh = 16.7 ms. (Reproduced from [33] by permission of the IEEE.
Copyright 1981 IEEE.)
The transition between the two regimes was identified by Dresner [34] to be at a limiting
operating current, Ilim:
Ast wh (Tc − Top )
I lim = . (51)
ηst
The above definition of the limiting current Ilim is equivalent to the Stekly criterion of Eq. (39).
Equation (51) sets a condition necessary for recovery: the heat transfer from the strand to the helium
must be larger than the Joule heat generation. This condition is satisfied for operating currents below
Ilim; that is, in the well-cooled regime. On the other hand, above Ilim, in the ill-cooled regime, a normal
zone will always generate more heat than it can exchange to the helium, and therefore no recovery will
be possible once the strand temperature is above Tcs.
This explains the behaviour of the energy margin below and above Ilim. In the well-cooled
regime, recovery is possible as long as the helium temperature is below the current sharing
temperature Tcs. Therefore the energy margin is of the order of the total heat sinks in the cable cross-
section between the operating temperature Top and Tcs, obviously including the helium. In the ill-
cooled regime, an unstable situation is reached as soon as the strands are current sharing, and therefore
the energy margin is of the order of the heat capacity of the strands between Top and Tcs plus the energy
that can be transferred to the helium during the pulse. In practical cases, the heat capacity of the
helium in the cross-section of a CICC is the dominant heat sink by two orders of magnitude and more,
and this explains the fall in the stability margin above Ilim.
437
L. B OTTURA
Fig. 38: The schematic behaviour of the stability margin as a function of the cable operating current
The transition between the well-cooled and ill-cooled regimes happens in reality as a gradual
fall from the maximum heat sink values to the lower limit (Miller [39]). Defining the limiting fraction
ilim of the critical current Ic as ilim = Ilim/Ic, the typical extension of this fall is of the order of (ilim)1/2. An
intuitive explanation for this fall can be given, again using the power balance at the strand surface. For
the derivation of Eq. (51), it was assumed that the helium has a constant temperature Top. In reality,
during the transient, the helium temperature must increase as energy is absorbed, so that the power
balance is displaced; that is, power can be transferred only under a reduced temperature difference
between the strand and the helium. Two limiting cases can be defined. The first is the ideal condition
of the helium at constant temperature, giving the limiting current of Eq. (51) – for which, however, the
energy absorption in the helium is negligible. Operation at (and above) Ilim is necessarily associated
with a stability margin at the lower limit – the ill-cooled value. The second limiting case is found
when the Joule heat production can be removed even when the helium temperature has increased up to
Tcs. This second case is obtained for a current of (and below)
low Ast wh (Tc − Tcs )
I lim = , (52)
ηst
low
which we call the lower limiting current, by analogy to Eq. (51) and due to the fact that I lim is always
low
less than Ilim. For operation at (and below) I lim , the full heat sink can be used for stabilization and the
low
stability margin is at the upper limit – the well-cooled value. Between the two values Ilim and I lim , the
stability margin falls gradually, sometimes showing a multiple stability region in the close vicinity of
Ilim. The multiple stability region extends over a small area that is not of interest for the safe design of
a stable CICC. Therefore this feature is usually neglected.
The dependence of the stability margin on the background field B is rather obviously explained
by the influence on the critical and current sharing temperatures. A higher B causes a drop both in the
limiting current (through a decrease of Tc and an increase of ηst) and in the energy margin (through a
decrease in Tcs). Therefore, as expected, ∆E drops as the field increases. An interesting feature,
however, is that the limiting current only decreases with Tc1/2 ; that is, with a dependence on B weaker
438
C ABLE S TABILITY
than that of the critical current. At large enough B, we will always have that Ilim is larger than Ic and
the cable will reach the critical current in well-cooled conditions.
The stability margin depends on the duration of the heating pulse, as shown experimentally by
Miller et al. [31] and reported in Fig. 39. A change in the heating duration for a given energy input
corresponds to a change in the energy deposition power. In the well-cooled regime – that is, for low
operating currents in Fig. 39 – the heat balance at the end of the pulse is in any case favourable to
recovery, and therefore the energy margin does not show any significant dependence on the pulse
length. When the conductor is in the ill-cooled regime, the power removal capability is limited. For
short heating pulse durations, the heating power increases and conductor reaches Tcs faster than for
lower powers, corresponding to longer heating durations. Therefore the energy margin increases at
increasing pulse length until it becomes comparable to the total heat capacity (as in the well-cooled
regime). This effect is partially balanced for very fast pulses, because the heat transfer coefficient can
exhibit very high values (see earlier discussion) that could shift the well-cooled/ill-cooled transition at
higher transport currents, and thus in principle higher energy margins should be expected in this range.
However, the high input powers in this duration range tend to heat the conductor above 20 K, into a
temperature range where the stabilizer resistivity grows quickly and the power balance is thus strongly
influenced. This effect causes saturation of the energy margin for extremely fast pulses (well below
1 ms in duration).
Fig. 39: The dependence of the stability margin for a CICC (indicated on this plot as ∆H) on the heating time-
scale (τh), as measured by Miller et al. [31]. The parameters varied in the experiment, indicated in the inset, are
the transport current in the sample, Is, the helium flow velocity, vHe, and the helium pressure, p. (Reproduced
from [31] by permission of the IEEE. Copyright 1979 IEEE.)
The dependence on the remaining operating conditions, typically the operating temperature and
pressure, is not easily quantified. The reason is that the helium heat capacity in the vicinity of the usual
operational regimes (operating pressure pop of the order of 3–10 bar and operating temperature
Top ~ 4–6 K) varies strongly with both pop and Top. This affects both the heat sink and the heat transfer
coefficient (through its transient components). An increasing temperature margin under constant
operating pressure gives a higher ∆E. But a simultaneous variation of pop and Top, under a constant
temperature margin, can produce a large variation (typically of the order of a factor of 2 in the range
given above) in ∆E (Miller [39]).
439
L. B OTTURA
A mention must be made of the case in which the operating point is in the superfluid helium
(He-II) range. The main difference compared to operation in He-I is the high heat transfer capability
associated with superfluid helium. The presence of He-II has thus two effects. First, the power balance
at the strand surface is drastically changed, being displaced towards the well-cooled condition. In
addition, a significant heat flux leaks at the end of the heated region, thus making available a larger
heat sink than the volume strictly contained in the heated region only. As an example, Lottin and
Miller [41] measured the stability margin of a 2 m long conductor in an operating temperature range
from 1.8 to 4.2 K. For this length the end effects are small, so that the experiment is a good basis to
show the influence of the surface heat transfer.
The stability margin in the case of He-II operation behaves at low current in a way similar to
what would be expected in the case of He-I operation. In fact, at low current, the current sharing and
critical temperatures are well above the transition temperature Tλ from He-II to He-I (around 2 K).
Heating of the strands up to current sharing implies that the surrounding helium undergoes the He-II to
He-I phase transition, and the stability margin is thus governed by heat transfer in He-I. At the ill-
cooled transition, however, the stability margin shows a peculiar behaviour. Owing to the large heat
transfer capability in He-II, the power balance at the strand surface remains favourable for recovery as
long as the wetting helium is in the He-II phase. Therefore, in a first approximation, the full heat sink
between the initial operating point and the transition temperature Tλ is still available at levels of the
operating current at which the conductor would have become ill-cooled for operation in He-I. In other
words, the conductor can still be considered as well-cooled for temperature excursions up to Tλ. As the
helium undergoes a phase transition at temperature Tλ, the available heat sink is significant, of the
order of 200 mJ·cm–3 of the helium volume. Finally, with increasing current, the power balance can
again become unfavourable, as soon as the heat flux limits in He-II are reached. There, the final
transition to the ill-cooled regime of operation takes place. This behaviour is shown in Fig. 40,
following measurements by Lottin and Miller [41].
Fig. 40: The stability margin of a Nb–Ti CICC as a function of the operating current, measured by Lottin and
Miller [41], at different temperatures in supercritical and superfluid helium. (Reproduced from [41] by
permission of the IEEE. Copyright 1983 IEEE.)
440
C ABLE S TABILITY
6.5 Minimum propagating zones

The discussion so far has examined normal zones extending over large lengths of superconductor,
ideally as large as the whole winding length. In reality, normal zones are established by small energy
inputs over limited lengths of superconductor. In order to withstand these inputs, the cryostability and
equal-area conditions would require a much too severe limitation on the operating current density. A
better design criterion can be identified by resorting to the concept of the minimum propagating zone,
originally defined by Wipf [43]. This concept was developed to a great extent by Wilson and Iwasa
[44]. Following Wilson, we again consider the case of a superconducting wire with a normal zone in
the centre. Using the auxiliary variable S to represent the heat flux along the wire, because of
symmetry at the centre of the normal zone we must have that:
S =0, (53)
irrespective of the temperature reached by the superconductor. In this case we can again use the equal-
area theorem and state that an equilibrium condition is defined by Eq. (46) (neglecting the variation of
the thermal conductivity with temperature) where now, however, the central temperature T’eq is no
longer Teq but, rather, is defined as the temperature at which the equal-area condition is satisfied. This
situation is shown graphically in Fig. 41.
Fig. 41: A graphical interpretation of the equal-area theorem in the case of heat generation above the maximum
steady state allowed by the equal-area theorem. The superconductor temperature profile with maximum
temperature T’eq is an equilibrium point, as it also satisfies the equal-area condition, but it is unstable.
For a choice of heat generation above that allowed for steady-state equilibrium by the equal-area
theorem, the central temperature T’eq must be lower than Teq. The corresponding temperature profile in
space can be found by integrating numerically Eq. (46). A family of such profiles, as produced by
Wilson [44], is shown in Fig. 42. Each curve in the family corresponds to a different generation curve,
and thus to a different equilibrium temperature. A property of the temperature profiles thus obtained is
that for a given heat generation and cooling condition, any normal zone with a temperature profile
below the one obtained will collapse because the cooling exceeds the generation, and the
superconductor will recover from the local transition. If the temperature profile of a normal zone is
above that obtained by the equal-area condition, then the normal zone will grow in time, leading to
thermal runaway. Hence the normal zone identified by this modified equal-area condition represents
an unstable equilibrium point, determining the boundary between recovery and thermal runaway.
Because of this, it has been called the Minimum Propagating Zone (MPZ) [43].
441
L. B OTTURA
The MPZ concept makes it possible to estimate the energy margin of the superconductor against
short energy perturbations. It was observed by Wilson that if an energy input has a dimension in space
smaller than the MPZ length, then the temperature profile evolves quickly towards the MPZ profile.
This led him to postulate that the energy margin can be estimated as the energy necessary to
instantaneously establish the MPZ. This can be regarded, in fact, as the minimum energy necessary in
all conditions to quench the conductor, or the Minimum Quench Energy (MQE), and is therefore a
conservative estimate for the energy margin. Any energy input happening on a finite time-scale will be
associated with heat transport and will result in an energy margin larger than the MQE.
Heat conduction in more than one dimension, as used to establish the heat balance, has a similar
effect, providing additional cooling for the MPZ. A demonstration of this is shown in Fig. 43, which
reports the measured energy margin versus the results of calculations in 1D and 2D geometry. The
agreement with the 2D calculation is evident. Also it is clear that, as expected, the energy margin is
larger than the MQE as computed from the above theory.
Fig. 42: The family of temperature profiles corresponding to MPZs obtained for different Joule heat generation
conditions. Each curve represents the unstable equilibrium boundary between recovery and thermal runaway.
(Reproduced from [44].)
442
C ABLE S TABILITY
Fig. 43: The measured and computed energy margins for a small solenoid equipped with heaters, plotted as a
function of the dimensionless Joule heat generation (after [44]). The computed curves refer to the MQE
following either a 1D or a 2D calculation. (Reproduced from [44].)
6.6 Transient stability in the general case

At present, fully cryostable magnets are rarely the preferred designer choice. In an efficient magnet
design, the cable operating current density must be kept high to make the magnet cross-section as
small as possible. For a specified field or stored energy, and thus a given magnetomotive force, a
maximum current density results in decreased material and production costs. As we have shown
previously, a cryostable magnet needs a large amount of copper stabilizer – compared to the amount of
superconductor – and a large amount of helium providing an ideally infinite heat sink. Therefore, a
cryostable magnet has an intrinsically low operating current density.
On the other hand, cryostability implies that the conductor is stable against any disturbance
spectrum, independent of the magnet details and the operating mode. In reality, the variety of
conductor designs and of magnet winding techniques, together with the variety of operating
requirements, results in a wide range of possible disturbance spectra. A cryostable conductor design is
therefore, in general, excessively safe. Indeed, most magnets presently designed and built are not
cryostable at the operating point, but they can still be operated reliably. The common feature of these
magnets is that their stability margin is above the disturbance spectrum experienced during operation.
443
L. B OTTURA
The first step in a sound design is thus to estimate the envelope of the perturbations that will be
experienced. Subsequently, the conductor can be designed to accommodate these perturbations by
means of a sufficiently large stability margin. Note that this process can imply iterations, as the
disturbance spectrum can depend on the conductor and the coil design themselves.
Depending on the energy release dominating the disturbance spectrum, the different
stabilization principles discussed in the previous sections can be used. A magnet operated in steady-
state mode, with a tightly packed winding, affected by small mechanical disturbances localized in time
and space (e.g. in the case of fully impregnated windings) may rely on the heat sink provided by the
small enthalpy margin of the superconductor and stabilizer themselves: an adiabatic winding. To
stabilize larger perturbations, the additional heat sink provided by helium may be necessary. Bringing
helium into close contact with the conductor thus increases its stability margin, provided that the heat
transfer at the wetted surface is efficient in the time-scale of the energy deposition considered.
Magnets with small amounts of added helium (or other heat sinks) are called quasi-adiabatic, as they
would in any case behave adiabatically for a fast enough time-scale. The stability margin can be made
larger by increasing the heat sink (e.g. the amount of helium) and its efficiency in absorbing heat
inputs (i.e. the heat transfer). This is typically the route followed in CICCs for large, pulsed magnets
that are designed for use in energy storage or thermonuclear fusion applications. The disturbance
spectrum is dominated in these cases by electromagnetic energy coupling through a.c. losses, which
are generally much larger than the enthalpy margin of the superconducting wire itself. Several options
are possible to increase the amount of helium and the heat transfer. In a forced-flow conductor, for
instance, the helium flows in channels inside the conductor, and the strands are subdivided to increase
their wetted perimeter and improve turbulent heat transfer. Another option is to use superfluid helium,
which has an exceedingly high heat transfer rate, in close contact with the wire. In any case, the
superconducting cable is in a meta-stable situation; namely, it can be quenched by a large enough
energy input. The art consists in reaching the desired stability margin for reliable operation with
maximum operating current density.
The main difficulty lies in the fact that the calculation of the energy margin associated with a
perturbation of arbitrary distribution in space and time is a complex matter. All of the theories
discussed so far have underlying approximations and limits, and the only way to attack the general
case is by numerical simulation of the non-linear heat balance. Even so, the calculation remains a
difficult task, involving accurate computation of heat conduction and possibly compressible helium
flow in complex geometry, taking into account the non-linear material properties. In practice, the
numerical calculation of the stability margin is the virtual analogue of an experiment, proceeding by
trial and error to refine the approximation between the lower perturbation boundary, leading to
recovery, and the upper boundary, resulting in a quench. The techniques discussed in the previous
sections, involving verification of the power balance and of the enthalpy margin, provide approximate
calculations that are usually sufficient for scoping calculations and design, and to start the search for
more intense numerical calculations. The following examples give the typical logic sequence followed
to achieve stable operation of magnetic systems in all operating conditions.
6.6.1 The transient stability of the EU–LCT coil

The Euratom–Large Coil Task (EU–LCT) coil was built in the framework of the Large Coil Task
project, a multinational effort to demonstrate the feasibility of a toroidal field system for a
thermonuclear fusion reactor [45]. The coil was wound in a D-shape, using the two-in-hand technique
in seven double pancakes. The winding pack was epoxy impregnated under vacuum and enclosed in a
thick steel casing, which provided the main support against the electromagnetic forces (see Fig. 44).
At the nominal operating current of 11 400 A, the maximum field produced in the winding during full-
array tests in the IFSMTF test facility was of 8.1 T, and the stored energy was about 100 MJ. The
conductor itself, shown in Fig. 45, was obtained by Roebel-cabling 23 rectangular Nb–Ti strands (with
copper stabilizer) around a central steel foil, and encasing this core in a steel jacket, producing a flat
444
C ABLE S TABILITY
cable that was 10 mm thick and 40 mm wide. The helium could flow between each strand within the
leak-tight jacket. Each strand, 2.35 × 3.1 mm2 in size, contained 774 Nb–Ti filaments with a nominal
diameter of 45 µm.
Fig. 44: The EU–LCT coil in its casing
Fig. 45: The EU–LCT cable
The current density in the strands was around 70 A·mm–2 in nominal conditions. This value is
more than twice as high as the one for the BEBC conductor described earlier, and with an increase in
the operating field from about 5 T in the BEBC magnet to about 8 T in the EU–LCT coil. The cooled
perimeter of this complex configuration was estimated to be of the order of 165 mm, and at the
nominal flow conditions the heat transfer coefficient was approximately 600 W·m–2·K–1. If we
calculate the Stekly coefficient for these specific conditions, we obtain a value of α ≈ 4, considerably
above the cryostable limit.
The disturbance spectrum during the operation of a TF coil in a fusion experiment is expected to
be dominated by a.c. loss deposition during the field change associated with the sudden instability of
the plasma column, or plasma disruption. Tests were performed on the EU–LCT coil, pulsing an
external coil and producing field changes up to 0.3 T with a time-scale of 0.5 s. This deposited in the
conductor energy of the order of 15 mJ·cm–3 of strand, without causing a quench [45]. Calculations
and measurements showed that for heat inputs in a short time-scale (0.5 ms), the stability margin was
of the order of 10–30 mJ·cm–3 of strand [46] in conditions comparable to the operating point of the
445
L. B OTTURA
cable. Over longer time-scales the stability margin increased, as more time was available to transfer
heat to the helium.
No measurements are available for the conditions of the field pulse test quoted above, but a
rough estimate, considering that the stability margin scales as the square root of the time-scale of the
energy deposition [47], results in a minimum stability margin of the order of 100 mJ·cm–3, well above
the energy deposited by a.c. loss. Indeed, the coil never had a spontaneous quench during testing.
6.6.2 The Tore Supra toroidal field magnet

The Tore Supra [48] is a tokamak built in the 1880s at the Centre d’Etudes de Cadarache (France). Its
toroidal field (TF) magnet is completely superconducting, and operates in a stagnant superfluid helium
bath. The TF magnet is composed of 18 circular coils, wound out of a monolithic composite conductor
in 26 double pancakes. The double pancakes are separated by spacers that maintain electrical
insulation but allow the free flow of helium around the conductor and ensure a helium percentage in
the winding pack of the order of 50% of the conductor volume. The winding pack is kept under
compression by an external steel casing that provides the tightness for the superfluid helium bath,
which is maintained at a temperature of approximately 1.8 K and a pressure of about 1 bar in normal
operating conditions. At the operating current of 1400 A, the maximum field produced on the winding
pack is of 9 T, for a stored energy in the TF magnet of 610 MJ. The conductor (see Fig. 46) is a
rectangular wire, of dimensions 2.8 × 5.6 mm2, with 11 000 Nb–Ti filaments of 23 µm diameter in a
mixed copper and CuNi matrix. The nominal Nb–Ti cross-section is 4.6 mm2 and the copper cross-
section is 10 mm2. At the operating conditions, the current density in the wire is approximately
90 A·mm–2.
Fig. 46: The Tore Supra strand
For the Tore Supra, the tolerance against the disturbance spectrum was formulated with the
requirement that that the conductor must be able to recover:
− after a localized (length of the order of some millimetres) temperature excursion up to 30 K,
or
− after a global (one full pancake) temperature excursion to 15 K, or
− after a plasma current disruption when the conductor is subjected to a field change of 0.6 T
in 10–20 ms.
Stability in superfluid helium has peculiar characteristics as compared to the situation of a
conductor wetted by boiling or supercritical normal helium. The main difference is the large heat
transfer capability of superfluid helium [49]. At small heat fluxes, the heat transport in superfluid
helium is virtually infinite and the heat transfer coefficient h from the conductor to the helium is
mostly governed by the Kapitza resistance at the wetted surface, with rather large values, in the range
of several thousands of W·m–2·K–1 [49]. The picture is different for large heat fluxes. In both steady-
state and transient conditions, there is an upper limit to the heat flux that can be supported by
superfluid helium before reaching the transition to the normal state, the so-called lambda line. This
limit depends both on the helium state and on the geometry. In the case of 1 bar sub-cooled superfluid
446
C ABLE S TABILITY
helium, the operating condition of the Tore Supra, a normal helium film forms at the wetted surface as
soon as the peak heat flux is exceeded [50]. At the same time, the heat transfer drops while the
conductor temperature rises sharply.
The consequence is that for small heat fluxes – for example, those deriving from mechanical
energy releases – the heat removal is such that the helium heat capacity available for stabilization can
be used completely. Larger energy depositions can be tolerated until the associated heat flux is below
the maximum allowable value. This limits the available heat sink, as seen from the conductor side, to a
fraction of the total helium volume. In the case of the Tore Supra, calculations and experiments were
performed to guarantee that the conditions given above could be satisfied. In particular, a 60 m long
cable had been tested in conditions comparable to the operation of the TF coil [51]. It was found that
at the nominal operating current of 1400 A, the cable was stable against a field pulse (1 T in 8 ms)
comparable to the one required in the design specifications. The a.c. loss deposited by this field pulse
was around 35 mJ·cm–3 of wire, and in these conditions no normal zone could be detected. The
average heat flux associated with such an a.c. loss is approximately 5 kW·m–2. This value, for the
geometry of the cooling channel of the Tore Supra conductor, is well below the critical heat flux limit,
which can be estimated to be of the order of 100 kW·m–2 [49]. As the heat flux does not limit heat
transfer, practically all the helium enthalpy from the operating temperature to the lambda transition is
used for stabilization. The typical helium enthalpy from 1.8 K to Tλ is of the order of 300 J·m–3 of the
helium volume, which is approximately 150 J·m–3 of the strand volume. This last value is a good
estimate of the stability margin in normal operating conditions.
7 Summary and advanced topics

This chapter has presented the basic considerations and models that go into the achievement of stable
superconductors. Overall, we can see the strategies presented above as a trade-off between the desired
performance and the allocated margin. One way to see this is to look at the schematic representation in
Fig. 47, where the various stabilization strategies discussed are plotted in terms of the typical range of
energy margins versus the typical range of operating current densities for which the strategy can be
applied. The reader is warned that, as for the perturbation spectrum, this is only an order-of-magnitude
representation, and exceptions can deviate considerably from the ranges identified there. Overall,
however, we see that a high operating current density is invariably associated with a small energy
margin. This implies that much effort must be put into the control and reduction of the perturbation
spectrum.
Fig. 47: A scatter plot of the typical range of energy margins versus operating current densities corresponding to
the various stabilization strategies discussed in this chapter.
447
L. B OTTURA
Enough is known about the mechanisms determining stability so that, in conjunction with other
constraints, superconductors can be designed and optimized successfully. However, this does not
mean that the field is not open to new areas of research. As new magnet designs are proposed, and as
more stringent requirements are imposed on the designer, areas of further study continue to open up:
in particular, work towards improving our understanding of stability under transient operating
conditions, and the interaction of magnetic and thermal instabilities.
The details of transient local heat transfer are not fully known, nor understood, especially in
complex flow geometries such as are often used for CICCs with cooling passages. As heat transfer
plays such an important role in the determination of the stable behaviour of a superconductor, this
point is somewhat surprising, but must be understood in terms of the difficulties inherent in the precise
measurement of flow and heat transport in a cryogenic fluid.
Stability depends in a synergistic manner on the d.c. and a.c. operating conditions of the cable in
the coil. This is a main direction of research in the field of stability. In particular, in view of the
applications to pulsed magnets, the interaction of stability, current distribution and a.c. losses in the
cable is one of the main topics. The so-called ramp-rate limit of operation for pulsed magnets (a
decrease in the maximum achievable current at increasing field change rate) is an outstanding example
of this synergistic interaction. The appearance of such a phenomenon, explained so far in terms of
non-uniform current distribution and a degradation of the stability margin of the cable, has alerted us
to the difference between d.c. stability, with a constant operating current and background field, and
a.c. stability of the cable.
The distribution – and redistribution – of current among the strands and within the cable can
have dramatic effects on stability. This statement applies to most cables used in technical applications
(flat cables, CICCs and super-stabilized cables). Certainly, the general solution of the thermal,
hydraulic and electromagnetic behaviour of a cable can be regarded as a formidable task. For this
reason, most of the efforts to understand current distribution in multistrand cables have been limited so
far to the purely electromagnetic problem, neglecting the intrinsic coupling with the thermal behaviour
[52–54]. Only recently there have been more general attempts to consistently solve the coupled
electromagnetic and thermal problem, and models have been presented for triplet of strands [55] and
flat accelerator cables [56].
During a thermal transient, the current in a quenched strand tends to redistribute to the
neighbouring strands, driven by the voltage of the normal zone. The redistribution takes place across
the transverse contact resistance (or at the joints in the case of insulated strands). The variation in the
strand current induces a change in the Joule heating rate, coupling back to the temperature evolution.
To model the redistribution process, mutual inductive coupling of strands must be taken into account,
while capacitive effects are negligible. Because a cable is strongly non-isotropic and because it has
discrete contacts at the strand crossing, the first natural approach to a model of the current distribution
is the use of an electrical network modelling the strands as uniform current density sticks, coupled
inductively and through localized cross-resistances (see, e.g., [52, 54]). This network approach is
solved by Kirchhoff’s voltage and current laws, and requires that appropriate current loops are set for
each degree of freedom in the cable cross-section. It is very detailed, providing information on each
strand crossover contact, but it can result in a very large number of equations that are not conveniently
coupled to a system of partial differential equations such as those given above.
One alternative, which has been used extensively for analytical studies, is to approximate the
cross-contacts as a continuous transverse conductance (see, e.g., [53]). A typical example is that of an
ideal two-strand cable. In this case, the governing equations become identical to those for an electrical
transmission line with negligible capacitance, a well-known problem in electromagnetics. This semi-
continuum approach is also useful for stability studies.
448
C ABLE S TABILITY
Super-stabilized superconductors, used in large magnets with low-intensity perturbation

spectrum (detector magnets for high-energy physics, or SMES magnets), are a special field that is
complex, but rather well understood. In super-stabilized conductors, a large amount of high-
conductivity material is added in parallel to the cable for protection. The distance of the stabilizer from
the multifilamentary area, and its low resistivity, result in an increase of the current diffusion time out
of the superconductor into the stabilizer. This effect is negligible within a strand, but becomes
appreciable in the limit of large segregated stabilizers, when this time can become comparable or
larger than the time-scale of the evolution of the thermal transient. The cable is said to be super-
stabilized if the time needed for current distribution is comparable to or larger than the time of flight of
the normal zone along the same section of conductor. In this type of cable, the power dissipated by
Joule heating during a transition to the normal state is initially much higher than the value reached
after the current diffusion has taken place. After complete current diffusion, the heating decreases to
the asymptotic steady-state value corresponding to a uniform current distribution. The variation of
Joule heating associated with the current diffusion affects the recovery of the cable. Furthermore, the
current diffusion can cause multiple stability boundaries, as well as stationary and travelling normal
zones. Stability models for super-stabilized cables are obviously focused on the effect of current
distribution inside the massive stabilizer. Continuum models are commonly used to describe this
process [57, 58]. The details of the superconducting cable, as well as heat transfer to the helium, are
lesser issues.
Finally, the old problem associated with flux jumps, dating back to the beginning of the history
of superconducting magnet technology, should not be forgotten. Indeed, the push for higher fields in
compact magnets such as accelerator dipoles and quadrupoles drives the need for current density to
very high values in Nb3Sn, in excess of 3000 A·mm–2 at 12 T and 4.2 K. At the same time, for
manufacturing reasons, the filaments of these high-Jc materials are as large as 100 µm, and at present
cannot be made much smaller than 30 µm. Such a high Jc and a large diameter causes flux jumps at
low field (0–2 T), as should be expected. An unexpected additional problem is associated with a newly
identified self-field instability, which can appear at intermediate and high fields (4–12 T) in strands of
large critical current and diameter, of the order of 1 mm or larger [59]. Such strands are required for
large-scale applications of high-field magnets, to help increase the final size of the cable and ease
protection. The drawback of a large strand diameter, associated with the very high values of Jc quoted
earlier, is that the transport current tends to remain confined in a very thin skin of filaments, at the
periphery of the multifilamentary composite. A simple way to understand this instability is to consider
this current distribution as generating a magnetic moment that can collapse and trigger an instability,
in much the same way as a flux jump.
References
[1] P.F. Chester, Rep. Prog. Phys. 30 (Part II) (1967) 561–614.
[2] M.A.R. LeBlanc, Phys. Rev. 124 (1961) 1423.
[3] M.N. Wilson, Superconducting Magnets (Plenum Press, New York, 1983).
[4] Rutherford Laboratory Superconducting Applications Group, J. Phys. D3 (1970) 1517–1585.
[5] Y. Iwasa, Conductor motion in the superconducting magnet – a review, IIF-IIR-Commission
A1/2 – Saclay (1981), pp. 125–136.
[6] H. Brechna, Superconducting Magnet Systems (Springer Verlag, Berlin, 1973).
[7] D.N. Lyon, Int. J. Heat Mass Transfer 7 (1964) 1097.
[8] C. Johannes, Paper B 3, Cryogenic Engineering Conf., Boulder, CO (1970).
[9] C. Schmidt, Appl. Phys. Lett. 32(12) (1978) 827.
[10] L.A. Yaskin, M.C. Jones, et al., Cryogenics 17 (1977) 549–552.
449
L. B OTTURA
[11] P.J. Giarratano, V.D. Arp, and R.V. Smith, Cryogenics 11 (1971) 385–393.
[12] P.J. Giarratano and W.G. Steward, Trans. ASME 105 (1983) 350–357.
[13] W.B. Bloem, Cryogenics 26 (1986) 300–308.
[14] G. Krafft, Heat transfer below 10 K, in Cryogenic Engineering, Ed. B.A. Hands (Academic
Press, London, 1986), pp. 171–192.
[15] A.R. Krantowitz and Z.J.J. Stekly, Appl. Phys. Lett. 6(3) (1965) 56–57.
[16] Z.J.J. Stekly and J.L. Zar, IEEE Trans. Nucl. Sci. 12 (1965) 367–372.
[17] E.U. Haebel and F. Wittgenstein, Big European Bubble Chamber (BEBC) magnet progress
report, Proc. 3rd Int. Conf. on Magnet Technology, DESY, Hamburg (1970), pp. 874–895.
[18] B.J. Maddock, G.B. James, and W.T. Norris, Cryogenics 9 (1969) 261–273.
[19] M. Morpurgo, The design of the superconducting magnet for the ‘Omega’ Project, in Particle
Accelerators (Gordon and Breach, London, 1970), Vol. I.
[20] M.O. Hoenig and D.B. Montgomery, IEEE Trans. Magn. 11(2) (1975) 569.
[21] M.O. Hoenig, Y. Iwasa, and D.B. Montgomery, Supercritical-helium cooled ‘bundle
conductors’ and their application to large superconducting magnets, Proc. 5th Magn. Tech.
Conf., Frascati (1975), p. 519.
[22] M.O. Hoenig, Y. Iwasa, D.B. Montgomery, and A. Bejan, Supercritical helium cooled
cabled, superconducting hollow conductors for large high field magnets, Proc. 6th Int. Cryog.
Eng. Conf., Grenoble (1976), p. 310.
[23] L. Dresner, IEEE Trans. Magn. 13(1) (1977) 670.
[24] L. Dresner and J.W. Lue, Design of forced-cooled conductors for large fusion magnets, Proc.
7th Symp. on Eng. Probs. of Fusion Res., Knoxville (1977), vol. I, p. 703.
[25] L. Dresner, Cryogenics 20 (1980) 558.
[27] M.O. Hoenig and D.B. Montgomery, Cryostability experiments of force cooled
superconductors, Proc. 7th Symp. on Eng. Probs. of Fusion Res., Knoxville (1977), vol. I, p.
780.
[28] M.O. Hoenig, D.B. Montgomery, and S.J. Waldman, IEEE Trans. Magn. 15(1) (1979) 792.
[29] M.O. Hoenig, Cryogenics 20 (1980) 373–389.
[30] M.O. Hoenig, Cryogenics 20 (1980) 427–434.
[31] J.R. Miller, J.W. Lue, S.S. Shen, and J.C. Lottin, IEEE Trans. Magn. 15(1) (1979) 351.
[32] J.W. Lue, J.R. Miller, and L. Dresner, J. Appl. Phys. 51(1) (1980) 772.
[33] J.W. Lue and J.R. Miller, IEEE Trans. Magn. 17(1) (1981) 757.
[34] L. Dresner, IEEE Trans. Magn. 17(1) (1981) 753.
[36] J.R. Miller, J.W. Lue, S.S. Shen, and L. Dresner, Adv. Cryog. Eng. 26 (1980) 654.
[37] J.W. Lue and J.R. Miller, Adv. Cryog. Eng. 27 (1982) 227.
[38] J.V. Minervini, M.M. Steeves, and M.O. Hoenig, IEEE Trans. Magn. 21(2) (1985) 339.
[39] J.R. Miller, Cryogenics 25 (1985) 552.
[40] T. Ando, M. Nishi, Y. Takahashi, K. Yoshida, and S. Shimamoto, Investigation of stability in
cable-in-conduit conductors with heat pulse duration of 0.1 to 1 ms, Proc. 11th Int. Cryog.
Eng. Conf., Berlin (1986), p. 756.
[41] J.C. Lottin and J.R. Miller, IEEE Trans. Magn. 19(3) (1983) 439.
450
C ABLE S TABILITY
[42] J.H. Schultz and J.V. Minervini, Sensitivity of energy margin and cost figures of Internally
Cooled Cabled Superconductors (ICCS) to parametric variations in conductor design, Proc.
9th Magn. Tech. Conf., Zurich (1985), pp. 643–646.
[43] A.P. Martinelli, Wipf, Proc. Appl. Sup. Conf., Annapolis (1972).
[44] M.N. Wilson and Y. Iwasa, Cryogenics 18 (1978) 17–25.
[45] D.S. Beard, W. Klose, S. Shimamoto, and G. Vecsey (Eds.), The IEA Large Coil Task,
Fusion Engineering and Design, 7 (1988).
[46] C. Schmidt, Cryogenics 24 (1984) 653–656.
[47] C. Schmidt, Cryogenics 30 (1990) 501–510.
[48] A. Torossian, TF-coil System and Experimental Results of Tore Supra, Fusion Engineering
and Design, 20 (1993), pp. 43–53.
[49] S.W. Van Sciver, Helium Cryogenics (Clarendon Press, Oxford, 1986).
[50] P. Seyfert, J. Lafferranderie, and G. Claudet, Cryogenics 22 (1982) 401–408.
[51] R. Aymar, C. Deck, P. Genevey, F. Lefevre, C. Leloup, C. Meuris, S. Palanque, A. Sagniez,
and B. Turck, IEEE Trans. Magn. 17(5) (1981) 2205–2208.
[52] A.A. Akhmetov, A. Devred, and T. Ogitsu, J. Appl. Phys. 75(6) (1994) 3176–3183.
[53] L. Krempaski and C. Schmidt, J. Appl. Phys. 78(9) (1995) 5800–5810.
[54] A.P. Verweij and H.H.J. ten Kate, IEEE Trans. Appl. Supercond. 5(2) (1995) 404–407.
[55] N. Amemiya and O. Tsukamoto, IEEE Trans. Appl. Supercond. 5(2) (1995) 218–221.
[56] M.N. Wilson and R. Wolf, IEEE Trans. Appl. Supercond. 7(2) (1997) 950–953.
[57] A. Devred, J. Appl. Phys. 65(10) (1989) 3963–3967.
[58] C.A. Luongo, R.J. Loyd, and C.L. Chang, IEEE Trans. Magn. 25(2) (1989) 1576–1581.
[59] B. Bordini and L. Rossi, IEEE Trans. Appl. Supercond. 19(3) (2009) 2470–2476.
451
Cooling with Superfluid Helium
Ph. Lebrun 1 and L. Tavian

Abstract
The technical properties of helium II (‘superfluid’ helium) are presented in
view of its applications to the cooling of superconducting devices,
particularly in particle accelerators. Cooling schemes are discussed in terms
of heat transfer performance and limitations. Large-capacity refrigeration
techniques below 2 K are reviewed, with regard to thermodynamic cycles as
well as process machinery. Examples drawn from existing or planned
projects illustrate the presentation.
Keywords: superfluid helium, cryogenics, superconductivity, refrigeration

techniques.
1 Introduction
Once confined to low-temperature physics laboratories, superfluid helium2 has become a technical
coolant for advanced superconducting devices, to the point that it is now implemented in industrial-
size cryogenic systems, routinely operated with high reliability. There are two classes of reason that
call for the use of superfluid helium as a coolant for superconducting devices; namely, the lower
temperature of operation, and the enhanced heat transfer properties at the solid/liquid interface and in
the bulk liquid.
The lower temperature of operation is exploited in high-field magnets [1, 2], to compensate for
the monotonously decreasing shape of the superconducting transition frontier (the ‘critical line’) in the
current density versus magnetic field plane, shown in Fig. 1 for some superconducting materials of
technical interest. In this fashion, the current-carrying capacity of the industrial Nb–Ti
superconducting alloys can be boosted at fields in excess of 8 T, thus opening the way for their use in
high-field magnet systems for condensed-matter physics [3–5], nuclear magnetic resonance [6, 7],
magnetic confinement fusion [8, 9], and circular particle accelerators and colliders [10–12]. In the case
of high-frequency superconducting devices such as acceleration cavities [13], the main drive for
superfluid helium cooling is the exponential dependence of the BCS losses on the ratio of the
operating temperature to the critical temperature. Accelerators based on this technology, such as
medium-energy, high-intensity machines [14–16], electron linacs feeding free-electron lasers [17], and
future high-energy lepton colliders [18–20], operate in a temperature range that minimizes the capital
costs and the overall energy consumption. This issue is schematized in Fig. 2.
The technical heat transfer characteristics of superfluid helium basically derive from peculiar
transport properties [21, 22]. Its low bulk viscosity enables superfluid helium to permeate to the heart
of magnet windings, while its very large specific heat (typically 105 times that of the conductor per
unit mass, 2 × 103 per unit volume), combined with excellent heat conductivity at moderate heat flux
(103 times that of cryogenic-grade OFHC copper) can produce powerful stabilization against thermal
1
philippe.lebrun@cern.ch
2
Strictly speaking, we are referring to the second liquid phase of helium, called He II, which exhibits the unusual bulk
properties associated with superfluidity and is therefore also called a ‘superfluid’. This is not to be confused with the entropy-
less component of the phenomenological two-fluid model accounting for the behaviour of He II, for which some authors
prefer to retain the term ‘superfluid’.
P H . L EBRUN AND L. TAVIAN
disturbances. In order to fully exploit these properties in both steady-state and transient regimes – for
example, for power heat transport over macroscopic distances as well as intimate stabilization of
superconductors inside magnet windings – an elaborate thermo-hydraulic design of the cooling
circuits, conductor, insulation, and coil assemblies is required. This often conflicts with other technical
or economic requirements of the projects and acceptable trade-offs have to be found.
In the following, we will only address the specific issues of cryogenic technology pertaining to
the use of superfluid helium as a technical coolant, namely different cooling methods as well as
processes and machinery for sub-lambda temperature refrigeration [23]. Reference is made to
companion lectures for cryogenic techniques which – however important in system design – are not
superfluid-helium specific, such as conventional heat transfer and cryostat design [24, 25].
Fig. 1: The critical current density of technical superconductors
500 BCS Losses

450 Carnot Efficiency
400 Refrigeration Cost
350
Arbitrary Units
300
250
200
150
100
50
0
1 1.5 2 2.5
T [K]
Fig. 2: The optimal operating temperature of RF superconducting cavities
454
C OOLING WITH S UPERFLUID H ELIUM
2 Different cooling methods
2.1 Pressurized versus saturated superfluid helium

A glance at the phase diagram of helium (Fig. 3) clearly shows the working domains of saturated
helium II, reached by gradually lowering the pressure down to below 5 kPa along the saturation line,
and pressurized helium II, obtained by sub-cooling liquid helium at any pressure above saturation, and
in particular at atmospheric pressure (about 100 kPa).
Although requiring one more level of heat transfer and additional process equipment – in
particular, a pressurized-to-saturated helium II heat exchanger – pressurized helium II cooling confers
several important technical advantages [26]. The avoidance of low-pressure operation in large and
complex cryogenic systems clearly limits the risk of air in-leaks, and resulting contamination of the
process helium. Moreover, in the case of electrical devices, the low dielectric strength exhibited by
low-pressure helium vapour [27] in the vicinity of the minimum of the Paschen curve (Fig. 4) [28]
leads to the additional risk of electrical breakdown at fairly low voltage. Operating in pressurized
helium II avoids this kind of problem.
10000
Solid
1000 Supercritical
Pressure [kPa]
Liquid II
Liquid I Critical
100 Point
Pressurized He II Saturated He I
(subcooled liquid)
10
Vapour
Saturated He II
1
1 2 3 4 5 6
Temperature [K]
Fig. 3: The phase diagram of helium
100
Breakdown voltage [kV]
10
0.1
1 10 100 1000 10000 100000
Pressure x distance [Pa.m]
Fig. 4: The Paschen curve for helium at 300 K
455
However, the most interesting and specific aspect of pressurized helium II in the operation of
superconducting devices stems from its capacity for cryogenic stabilization. As a sub-cooled
(monophase) liquid with high thermal conductivity, pressurized helium II can absorb in its bulk a
deposition of heat, up to the temperature at which the lambda line is crossed, and local boiling starts
only then, due to the low thermal conductivity of helium I. Quasi-saturated helium II, which is in fact
slightly sub-cooled due to the hydrostatic head below the surface of the liquid bath, may only absorb
heat deposition up to the point at which the saturation line is crossed and change of phase occurs. The
enthalpy difference from the working point to the transition line is usually much smaller in the latter
case. The argument, developed in Ref. [29], typically yields an order of magnitude better performance
in favour of pressurized helium II.
2.2 Bath cooling

The first attempts to lower the temperature of a helium bath at atmospheric pressure were made with
the purpose of operating superconducting magnets at temperatures below 4.2 K [30]. Thermal
stratification in the ‘Roubeau’ bath (Fig. 5) permitted the lambda point to be reached, but the large
heat exchange area with the sub-cooled helium I volume, combined with the high thermal conductivity
of the helium II bath, prevented the latter from cooling down further. By inserting a restriction to
thermal conduction, in the form of an insulating plate with a minimum residual helium cross-section,
the ‘Claudet’ bath [31] enabled the helium II volume to be sub-cooled well below the lambda point, to
a temperature at which its heat transport properties are maximal. Both systems need an external source
of refrigeration below 2 K, usually in the form of a saturated helium II heat exchanger coupled to the
pressurized bath.
Fig. 5: Bath cooling with pressurized helium II: the Roubeau bath (left) and the Claudet bath (right)
2.3 Conduction cooling

In the following, we shall only consider conductive heat transport in helium II at heat fluxes of
technical interest (typically above 1 kW·m–2). For most practical geometries, this means working in
the ‘turbulent’ regime, with full mutual friction between the components of the two-fluid model [32].
In this regime, helium II exhibits a large, finite, and non-linear bulk heat conductivity, the value of
which depends both on temperature and heat flux. While the general patterns of this behaviour can be
predicted by the Gorter–Mellink [33] theory, 3 practical data useful for engineering design has been
established in a number of experiments [34–39].
3
In 1949, C.J. Gorter and J.H. Mellink introduced the idea of an interaction producing mutual friction between the
components of the two-fluid model, to account for the observed transport properties of helium II.
456
Consider conduction in one dimension; for example, in a tubular conduit of length L, the ends of
which are maintained at temperatures TC and TW . The steady-state heat flux q is given by
q n L X (TC ) − X (TW ) ,
= (1)
where the best experimental fit for n is 3.4, and X(T) is a tabulated function of temperature, physically
analogous to a conductivity integral [34]. A plot of this function reveals that the apparent thermal
conductivity of helium II goes through a maximum at around 1.9 K (Fig. 6).
Fig. 6: The thermal conductivity integral and apparent thermal conductivity of pressurized superfluid
helium [34].
As an example, the heat flux transported by conduction between 1.9 K and 1.8 K in a 1-m long
static column of helium II is about 1.2 W·cm–2; that is, three orders of magnitude higher than would be
conducted along a bar of OFHC copper of the same geometry! The non-linearity with respect to heat
flux also results in a much weaker dependence of conduction upon length or thermal gradient. Figure 7
shows the steady-state conduction Q in superfluid helium between 1.9 K and 1.8 K versus the static
column length L for different equivalent nominal diameters of the column. This abacus clearly shows
that while the heat flux conducted in a solid is directly proportional to the thermal gradient applied,
doubling the conduction length in a column of helium II only reduces the heat flux by some 20%.
Fig. 7: Steady-state conduction in pressurized superfluid helium
457
The variation of X(T) also implies that, for each value of the cold boundary temperature TC ,
there exists a maximum possible heat flux at which TW reaches the lambda point, and the helium
column ceases to be superfluid. Values of this limiting heat flux, which also weakly depends on L,
range from a fraction to a few units of W·cm–2, for practical cases of interest. This clearly places an
intrinsic limitation on the applicability of helium II conduction for quasi-isothermal cooling of long
strings of superconducting devices in an accelerator. Transporting tens of watts over distances of tens
of metres would then require a temperature difference of several hundred millikelvins and a large
cross-section of helium, which is both impractical and thermodynamically costly. For a more precise
estimate, consider a uniformly heated tubular conduit of length L, operating between temperatures TC
and TW , and apply the helium II steady-state conduction equation to this fin-type geometry. After
integration,
( n + 1)  X (TC ) − X (TW ) ,

qtotal n L = (2)
where q total is the total heat flux flowing through the section at temperature TC , near the heat sink.
Figure 8 shows the steady-state conduction Q tot in superfluid helium of a cryomagnet string with
linear heating ξ between 1.9 K (the temperature of the warmest magnet) and 1.8 K (the temperature
at the heat sink). As an example, the cooling by conduction of a 50-m long cryomagnet string, with a
uniform linear thermal load of 1 W·m–1, would require a helium II cross-section of 90 cm2; that is, a
10.7-cm diameter conduit. In view of such constraints, the conduction-cooling scheme originally
considered for the LHC project [40] was later abandoned, in favour of the more efficient one described
in Section 2.5.
Conduction through static pressurized superfluid helium, however, remains the basic process for
extraction and local transport of heat from the LHC magnet windings, across their polyimide-wrap
electrical insulation. Although the polyimide tape, which constitutes the insulation of the
superconducting cable, is wrapped in two layers with a half overlap (Fig. 9) in order to achieve
sufficient mechanical toughness and dielectric strength, this still preserves sufficient percolation paths
for helium II conduction to significantly improve the heat transfer, well above the solid conduction
across the sole polyimide [41].
Fig. 8: The steady-state conduction cooling of a cryomagnet string with a linear applied heat load
458
Fig. 9: The heat transfer across the polyimide-wrap insulation of a superconducting cable
The high thermal conduction in helium II can also be exploited to ensure quasi-isothermality of
helium enclosures of limited spatial extension, such as the helium bath of a superconducting magnet
under test. Knowledge of temperature changes at any point in the bath permits us to assess the
enthalpy changes of the system, and thus to perform calorimetric measurements. This technique
proves very convenient for measuring minute heat in-leaks [42] or substantial energy dissipation [43],
such as produced by ramping losses or resistive transitions in superconducting magnets.
2.4 Forced-flow convection of pressurized superfluid helium

To overcome the limited conduction of helium II in long strings of cryogenic devices, the obvious
issue is to create a forced circulation of the fluid in a cooling loop, thus relying on convective and
advective heat transfer. One can then benefit from an additional control parameter, the net velocity
imparted to the bulk fluid. In the following, we shall only discuss convection in channel diameters of
technical interest; that is, typically greater than a few millimetres. The flow induced by a pressure
gradient across a hydraulic impedance is then essentially determined by the bulk viscosity of the fluid.
Assuming that internal convection between the components of the two-fluid model is independent of
the net velocity reduces the problem to the behaviour of a flowing monophase liquid with high, non-
linear thermal conductivity. The steady-state convective heat transport Q between two points 1 and 2
of the cooling loop is then given by the difference in enthalpy H of the fluid flowing with a mass flow-
rate m :
=Q m ( H 2 − H1 ) . (3)
An estimate of the potential advantage of forced convection over conduction can be made, using
the same geometry and temperature boundary conditions as described in Section 2.2. Consider
helium II pressurized at 100 kPa, flowing in a heated pipe of length 1 m and cross-section 1 cm2, and
assume that its temperature increases from 1.8 K at the pipe inlet to 1.9 K at the outlet. It is easy to
show that for flow velocities above 0.2 m·s–1, advective heat transport exceeds conduction.
The above calculation, however, neglects the pressure drop along the flow. A glance at the
pressure–enthalpy diagram of helium (Fig. 10) reveals a positive Joule–Thomson effect [44]: the
enthalpy of the fluid increases both with increasing temperature and pressure, so that an isenthalpic
expansion results in a temperature increase. For example, pressurized helium II flowing across a
pressure gradient of 50 kPa will warm up from 1.8 K to 1.9 K, in the absence of any applied heat load.
The magnitude of this effect requires precise knowledge of the thermo-hydraulic behaviour of
helium II, in order to validate its implementation in long cooling loops [45].
459
. 1000
Q
.
1 m 2 1.7 1.8 1.9 2.0 2.1 K
1
100
Pressure [kPa]
2' 2
T1 = 1.8 K T2 = T2' = 1.9 K 2'
Saturation
10
Q1→2= m ( H 2 − H1 ) > 0
Q1→2=′ m ( H 2′ − H1 =
) 0 1
0 1 2 3 4
Enthalpy, H [J/g]
Fig. 10: The pressure–enthalpy diagram and forced-flow convection in superfluid helium
Following early work [46, 47], several experimental programmes have investigated the heated
flow of pressurized helium II in pipes and piping components [48, 49], culminating with the 230-m
long test loop at CEA Grenoble [50, 51], which gave access to high Reynolds numbers and extended
geometries characteristic of accelerator string cooling loops. In parallel with that work, mathematical
models were developed for calculating combined conductive and convective heat transport processes
in complex circuits [52, 53], and were validated through experimental results. Pressure drop and heat
transfer – both steady-state and transient – in flowing pressurized helium II may now be safely
predicted for engineering purposes, using well-established laws and formulas.
The implementation of forced-flow cooling requires cryogenic pumps operating with
pressurized helium II. Although most of the experimental work has been performed using positive
displacement – that is, bellows or piston-pumps originally developed for helium I [54] – the thermo-
mechanical effect, specific to the superfluid, may also be used for driving cooling loops by means of
fountain-effect pumps [55–58]. In spite of their low thermodynamic efficiency [59], a drawback of
limited relevance for using them as circulators, which usually have to perform low-pressure pumping
work, fountain-effect pumps are light, self-priming and have no moving parts, assets of long-term
reliability in, for example, on-board applications in space [60]. At higher heat loads, they have been
considered [61] and tested [62] for forced-flow cooling of superconducting magnets: the overall
efficiency of the process may then be improved by configuring the cooling loop so as to make use of
the heat load of the magnet proper to drive the thermo-mechanical effect in the pump [63]. Practical
implementation of these techniques for cooling superconducting magnets has, however, been limited,
in particular by the need to provide other means of flow circulation during pre-cooling with normal
helium above the lambda point.
2.5 Two-phase flow of saturated superfluid helium

The conductive and convective cooling systems described above both transport heat deposited or
generated in the load over some distance through pressurized helium II, up to a lumped pressurized-to-
saturated helium II heat exchanger acting as quasi-isothermal heat sink. This is achieved at the cost of
460
a non-negligible – and thermodynamically costly – temperature difference, thus requiring the heat sink
to operate several hundred millikelvins below the temperature of the load.
A more efficient alternative is to distribute the quasi-isothermal heat sink along the length of the
accelerator string. In this fashion, the conduction distance – and hence the temperature drop – in the
pressurized helium II is kept to a minimum, typically the transverse dimension of the device cryostat.
This leads to the cooling scheme proposed for the LHC at CERN, schematized in Fig. 11: the
superconducting magnets operate in static baths of pressurized helium II at around atmospheric
pressure, in which the heat load is transported by conduction to the quasi-isothermal linear heat sink
constituted by a copper heat exchanger tube, threading its way along the magnet string, and in which
flowing two-phase saturated helium II gradually absorbs the heat as it vaporizes [11].
Saturated He II, flowing Heat exchanger tube
Pressurized He II, static
SC magnet Helium vessel SC bus-bar connection

Fig. 11: The principle of the LHC superfluid helium cooling scheme
Although potentially attractive in view of its efficiency in maintaining long strings of magnets at
quasi-uniform temperature, this cooling scheme departs from the well-established wisdom of avoiding
long-distance flow of two-phase fluids at saturation, particularly in horizontal or slightly inclined
channels. Moreover, no experimental data was originally available on flowing saturated helium II, and
very little for other cryogenic fluids in this configuration. Following the first exploratory tests [64],
which demonstrated the validity of the concept on a reduced geometry, a full-scale thermo-hydraulic
loop [65] permitted us to establish the stability of horizontal and downward-sloping helium II flows, to
observe partial (but sufficient) wetting of the inner surface of the heat exchanger tube by the liquid
phase, due to flow stratification, and to address process-control issues and develop strategies for
controlling uniformity of temperature at strongly varying applied heat loads, given the low velocity of
the liquid phase. As long as complete dry-out does not occur, an overall thermal conductance of about
100 W·m–1·K–1 can be reproducibly observed across a DN40 heat exchanger tube, made of industrial-
grade deoxidized phosphorus copper.
Once the wetting of the inner surface of the tube is guaranteed, the heat transfer from the
pressurized to the saturated helium II is controlled by three thermal impedances in series: solid
conduction across the tube wall, and Kapitza resistance at the inner and outer interfaces between tube
wall and liquid (Fig. 12). While the former can be adjusted, within technological limits, by choosing
the tube material and wall thickness, the latter, which finds its origin in the refraction of phonons at the
liquid/solid interfaces and is thus strongly temperature dependent, usually dominates below 2 K [66].
461
100 Heat exchange surface

limit q
radiation Solid He II
Phonon es)
ts (C l ea n Surfac
10 en
Experim T(x)
hK [kW/m2.K]
∆Ts
1
)
y Surfaces
ents (Dirt
Experim
0.1
ov theory
Khalatnik Valid for small heat flux
0.01 (when ∆T ≪ T)
1.3 1.5 1.7 1.9 2.1
T [K]
Fig. 12: The Kapitza conductance at a copper/helium II interface
The final validation of the two-phase helium II flow cooling scheme for the LHC has been
performed successfully on a 100-m long test string, equipped with full-scale prototype cryomagnets,
operated and powered in nominal conditions [67, 68]. At varying heat loads exceeding 1 W·m–1, all
magnets in the string were maintained in a narrow range of temperature, a few tens of millikelvins
above the saturation temperature of the flowing helium II. Thermal buffering provided by the
pressurized helium II baths contributed to limiting temperature excursions, at the cost of introducing
strong non-linearities and time delays in the system, which must be coped with by elaborate, robust
process control [69, 70]. As a complement to that applied work focused towards LHC, more
fundamental experimental studies have been conducted on specially instrumented test loops at CEN-
Grenoble, comprehensively equipped with diagnostics and a transparent section for visual observation
and interpretation of the flow patterns [71–73]. As long as the vapour velocity remains sufficiently low
to maintain stratified flow (up to a few metres per second), the engineering design of such a cooling
scheme rests on a few simple sizing rules [74]. At higher vapour velocity, entrainment and atomization
effects complicate the flow pattern and impact on the heat transfer and pressure drop [75].
This type of cooling scheme may also be used for extracting much higher linear heat loads,
typically about 10 W·m–1, as present in the low-beta quadrupoles of the high-luminosity insertions of
the LHC [76, 77], at the expense of a larger-diameter heat exchanger tube to limit the saturated vapour
velocity and thus preserve flow stratification.
3 Refrigeration cycles and equipment

The properties of helium at saturation (see Fig. 3) make it necessary to maintain an absolute pressure
below 1.6 kPa on the heat sink of a 1.8 K cryogenic system. Bringing the saturated vapour up to
atmospheric pressure thus requires compression with a pressure ratio exceeding 80; that is, four times
that of refrigeration cycles for ‘normal’ helium at 4.5 K. Figure 13 shows the basic scheme for
refrigeration below 2 K. A conventional refrigerator produces liquid helium at 4.5 K, later expanded
down to 1.6 kPa in a Joule–Thomson expansion stage. The gaseous helium resulting from liquid
vaporization is compressed above atmospheric pressure and eventually recovered by the 4.5 K
refrigerator. We will therefore start by presenting the Joule–Thomson expansion stage.
462
Three types of cycles, sketched in Fig. 13, can be considered [78, 79] for producing refrigeration
below 2 K:
− the ‘warm’ compression cycle, based on ambient-temperature sub-atmospheric compressors;
− the ‘cold’ compression cycle, based on multistage cold compressors all the way up to
atmospheric pressure; and
− the ‘mixed’ compression cycle, based on a combination of cold compressors in series with
ambient-temperature sub-atmospheric compressors.
We will then proceed to discuss the thermodynamics and machinery for these three types of cycle.
Fig. 13: Generic process cycles for refrigeration below 2 K
3.1 The Joule–Thomson expansion stage

The efficiency of the Joule–Thomson expansion of liquid helium, say from 0.13 MPa and 4.5 K, down
to 1.6 kPa and 1.8 K, can be notably improved if the liquid is previously sub-cooled by the exiting
very-low-pressure vapour (Fig. 14). This is performed in a counter-flow heat exchanger, sub-cooling
the incoming liquid down to 2.2 K by enthalpy exchange with the very-low-pressure saturated vapour.
This heat exchanger has to produce a limited pressure drop, particularly in the very-low-pressure
stream. A maximum pressure drop of 100 Pa is generally acceptable, corresponding to a few per cent
of the absolute saturation pressure. The design of such heat exchangers for large flow-rate [80] is not
straightforward, and their qualification impractical. As a consequence, the LHC cryogenic system
features several hundred small-sized (5–20 g·s–1) heat exchangers, distributed around the ring. This
avoids the transportation of sub-cooled helium over long distances, saving one header in the ring
distribution line. It also permits qualification and reception testing of the heat exchangers on a test
stand of reasonable size. Following prototyping, technical validation of different solutions [81, 82] and
commercial selection, these heat exchangers have been series produced by industry and are now in
operation in the LHC tunnel.
463
.
m LHe Subcooled liquid @ 2.2 K
1
Saturated liquid @ 4.5 K
Sub-cooling HX 1000 LHe
.
1 m
1'
1' 1
Pressure [kPa]
Saturation
100 dome
3 2' 2 3
10
. 0 .
Sat. He II Q 3 Q Sat. He II
1 2' 2
1.8 K, 16 mbar
1.8 K, 16 mbar
0 10 20 30 40
13 % GHe produced Enthalpy [J/g] 46 % GHe produced
during expansion during expansion
Fig. 14: The efficiency of the Joule–Thomson expansion
3.2 The ‘warm’ compression cycle

For low-power refrigeration – for example, in small laboratory cryostats – this is achieved by
means of standard Roots or rotary-vane vacuum pumps (Fig. 15), handling the very-low-pressure
gaseous helium escaping from the bath after it has been warmed up to ambient temperature through a
heat exchanger and/or an electrical heater. This technology may be pushed to higher flow-rates using
liquid-ring pumps, adapted for processing helium by improving the tightness of their casing and
operating them with the same oil as that of the main compressors of the 4.5 K cycle [83], or oil-
lubricated screw compressors operating at low suction pressure (Fig. 15). In any case, compression at
ambient temperature is hampered by the low density of the gaseous helium, which results in large
volume flow-rates and thus requires large machinery, as well as in costly, inefficient heat exchangers
for recovering the enthalpy of the very-low-pressure stream.
Fig. 15: Sub-atmospheric compressors: (a) a combination of Roots and rotary-vane vacuum pumps; (b) the
compound screw.
All these compressors are positive-displacement machines having volumetric characteristics.
Screw compressors are routinely used in helium refrigeration and their implementation in a 1.8 K
cycle therefore follows from current practice. Special attention, however, has to be paid to the
protection against air in-leaks: in particular, the motor shaft and its rotary sealing must be located on
the discharge side to operate above atmospheric pressure.
464
A first limit to the use of sub-atmospheric screw compressors stems from volumetric flow
requirements: the biggest available machines have a swept volume of about 4600 m3·h–1, so that higher
flow-rates require parallel arrays. Moreover, the isothermal efficiency – defined as the ratio of
isothermal compression work to the effective compression work of the machine – decreases markedly
with the suction pressure, as shown in Fig. 16, thus precluding their use at very low pressure in
efficient process cycles.
Fig. 16: The isothermal efficiency of warm sub-atmospheric compressors
3.3 The ‘cold’ compression cycle

The alternative process is to perform compression of the vapour at low temperature; that is, at its
highest density. The pumps and recovery heat exchangers get smaller in size and less expensive, but
the compression work is then injected in the cycle at low temperature, so that the inevitable
irreversibilities have a higher thermodynamic weight. Moreover, the pumping machinery that handles
the cold helium must be non-lubricated and non-contaminating, which seriously limits the choice of
technology. Hydrodynamic compressors, of the centrifugal or axial–centrifugal type, have been used
in large-capacity systems [84]. Their pressure ratio, limited to 2–3.5 per stage, however, makes it
necessary to arrange them in multistage configurations [85, 86], thus narrowing the operational range
of the system, in particular for start-up or off-design modes.
Depending on the operating temperature (2.0 K or 1.8 K), the ‘cold’ compression cycle requires
at least four or five stages in series in order to perform the overall pressure ratio of 45–80. The
compressed helium is directly returned to the cold low-pressure (LP) stream of the 4.5 K refrigerator.
The main drawback of this cycle concerns turndown capability. The cold compressor set has to
guarantee the same pressure ratio for any load. A typical operating field for hydrodynamic
compressors (Fig. 17) displays the pressure ratio as a function of the reduced flow m* and the reduced
speed N*. The working area is limited on the left-hand side by the stall line, on the right-hand side by
the choke line, and on top by the maximum rotational speed of the drive. At constant pressure ratio,
the compressor can handle a flow reduction of only about 20% before reaching the stall line. Below
80% of nominal, additional vapour generation by electrical heating must be used to compensate for the
load reduction. Such a cycle is therefore not very compliant to turndown, and its operating cost is not
optimized for part-load operation.
465
This led CERN to conduct, in view of the LHC project, a R&D programme on cold
compressors, procuring from specialized industry three prototype hydrodynamic compressors of
different designs [87–90] to investigate critical issues such as drive and bearing technology, impeller
and diffuser hydrodynamics, and mechanical and thermal design, as well as their impact on overall
efficiency [91]. The choices eventually retained for the LHC series machines [92–94] are three-phase
electrical induction motor drives working at room temperature, with a rotational speed varying from
200 to 700 Hz, active magnetic bearings working at room temperature, axial–centrifugal (three-
dimensional) impellers, and fixed-vane diffusers (Figs. 18 and 19).
Operating
3.5 range m T Pin
m* = × in × 0
line
(20 to 25 %)
3
Forbidden m 0 Tin 0 Pin
ll
Sta
area
Pressure ratio
and
2.5 Tin 0
* N
Design N= ×
2 point N0 Tin
e
lin
e
with: m = mass-flow
ok
1.5
Ch
Tin = inlet temperature

N1* N2* N 3* Pin = inlet pressure
1 N = rotational speed
0 1 subscript 0 = design condition
Reduced flow, m*
Fig. 17: The typical operating field of a hydrodynamic compressor
Fig. 18: Axial–centrifugal cold compressor cartridges for the LHC
466
Fig. 19: The isentropic efficiency and a typical cross-section of a cold compressor
The thermodynamic efficiency of cold compressors, determined by hydrodynamic design as

well as by the limitation of internal leakage and of heat in-leaks along the drive shaft, has significantly
improved (Fig. 19). Here, the relevant estimator is the isentropic efficiency, defined as the ratio of the
compression work in the adiabatic, reversible case, to the real one. Recent machines reach 75%
isentropic efficiency at their design point.
3.4 ‘Mixed’ compression cycles

For large systems, oil liquid-ring pumps or lubricated screw compressors may be used in series with
cold compressors, in ‘mixed’ compression cycles. Cold compressors are well suited for the lower
stages, while the presence of volumetric machines in the upper stages permits independent adjustment
of the flow-rate or wheel inlet conditions, thus improving load adaptation [95, 96].
In ‘mixed’ compression cycles, the number of cold compressor stages can be reduced to three,
depending on the swept volume and the number of warm sub-atmospheric machines. The compressed
helium can be returned to the 4.5 K refrigerator at different levels.
− At the warm medium-pressure (MP) side (connection #1 in Fig. 13(d)). This requires the use
of screw compressors having a sufficient built-in pressure ratio. In this case, the enthalpy of
the gas at the outlet of the cold compressors has to be recovered by the heat exchangers of
the 4.5 K refrigerator. The main advantage of this solution is that the same oil-removal and
final cleaning systems can be used for the warm sub-atmospheric compressors and for the
booster stages of the 4.5 K refrigerators, thus minimizing the investment cost of the system.
− At the warm low-pressure (LP) side (connection #2 in Fig. 13(d)). This solution is
compatible with the use of either screw compressors or liquid-ring pumps. The enthalpy of
the cold gas at the outlet of the cold compressors also has to be recovered by the heat
exchangers of the 4.5 K refrigerator. In this case, the warm sub-atmospheric stage requires
its own oil-removal system.
− At the cold low-pressure side (connection #3 in Fig. 13(d)). This is required when the
enthalpy of the cold gas at the outlet of the cold compressors cannot be recovered by the
heat exchangers of the 4.5 K refrigerators (the LHC case) [92]. In this case, the warm sub-
atmospheric stage requires its own oil-removal and final cleaning system (coalescing filters
and charcoal adsorbers), increasing the investment cost.
467
The main advantage of the ‘mixed’ cycle resides in its turndown capability. With sub-
atmospheric compressors having volumetric characteristics, the pressure at the outlet of the cold
compressors decreases linearly with the flow-rate; that is, if the temperature and rotational speed do
not change, the reduced flow-rate m* stays constant, thus keeping the working point fixed in the
operating field. Such a cycle can then handle a large dynamic range – for example, a value of 3 for the
LHC – without any additional electrical heating. Moreover, the total pressure ratio of the cold
compressor train is lowered and the speed of some machines can then be reduced, thus decreasing the
total compression power and operating cost, and improving the overall efficiency of the cycle.
Another operational advantage concerns the possibility of maintaining the load in cold standby
with the cold compressors freewheeling and all compression performed, though at much reduced flow,
by the warm machines. This mode allows repair or exchange of a cold-compressor cartridge without
emptying the helium from the system. In addition, the load adaptation provided by the warm
volumetric machines proves very useful during transient modes such as cool-down and pump-down, in
which the cold compressors operate far from their design conditions.
The only drawback of this cycle concerns the risk of air in-leaks due to the presence of sub-
atmospheric circuits in air. Helium guards are recommended to prevent pollution of the process helium
[97].
3.5 The application range of low-pressure helium compression techniques

The practical ranges of application of the different techniques are shown in Fig. 20, setting a de facto
limit for warm compression above 20 000 m3·h–1, or typically 300 W at 1.8 K. The diagram also
illustrates the large span of refrigeration power and the diversity of projects using superfluid helium.
The investment and operating costs of large superfluid helium refrigeration systems can be assessed
from basic thermodynamics and practical scaling laws derived from recent experience [98], thus
providing input for the technical–economical optimization of such systems.
100
"Warm" p )
FNAL/JAERI m
"Cold" pu 5 °C
m 1
uu @
Suction pressure [kPa]
"Mixed" ac
/h
ts v m3
o 0
Ro '00
Screw compressor 0
10 (2 BNL
PU
W
Superfluid SE RN
AC CE
M CEBAF TEST TESLA CEBAF
SNS
CERN TEST
CENG ELBE
LHC
TORE SUPRA
1
TTF CERN CCU
1 10 100 1000
Mass-flow rate [g/s]
Fig. 20: The range of application of low-pressure helium compression techniques
468
4 Other technological aspects

The preceding sections have addressed the main specific aspects of superfluid helium technology,
namely the properties of helium II, cooling methods, and refrigeration below 2 K. There is, of course,
a lot more to the design and construction of a fully operational helium II cryogenic system for cooling
superconducting devices in a particle accelerator, but it is less specific – if at all – to superfluid
helium, and can be dealt with by adopting the design rules and good practice required by more
conventional liquid helium technology. Still, three aspects are singled out in what follows, as they are
often the subject of questions asked by newcomers to the field.
4.1 Helium II to vacuum leak-tightness

The low viscosity of helium II and the thermo-mechanical effect it displays across ‘superleaks’ often
raise the question of special techniques for designing and building tight vessels, pipelines and
enclosures for helium II systems. Experience has shown that the state-of-the-art design and
construction rules applicable for cryogenic equipment operating with normal liquid helium are
sufficient to ensure helium II leak-tightness. This may be explained by the fact that a minute crack in
the wall of the helium II vessel of a cryostat, surrounded by insulation vacuum, is physically different
from a ‘superleak’; that is, porosity across a plug separating two helium II enclosures. In the former
case, the leaking helium will vaporize when it reaches saturation pressure some way along the crack
and the leakage rate will eventually be controlled by the flow of vapour downstream – as it would be
for a normal helium leak.
The practical approach used for ensuring leak-tightness of the LHC helium II enclosure –
essentially the ‘cold masses’ of the superconducting magnets and their interconnections – was to
enforce an all-welded austenitic stainless steel construction, using automatic welding to reduce human
error. While TIG welding was used for low-thickness material – for example, the cryomagnet
interconnections [99] – a combination of STT (root pass) and MAG (filling passes) was used for the
10-mm thick AISI 316 LN wall of the magnet ‘cold mass’ [100]. Local and global leak detection
following the pressure test was systematically performed as an integral part of the quality assurance
procedure [101].
Specific strategies were applied in exceptional cases when an all-welded construction could
not be used. The copper heat exchanger tubes running along the LHC magnet strings [102] have
brazed end-sleeves of austenitic stainless steel, so that they can be TIG-welded to each other when
interconnecting the cryomagnets; by design, the brazed joint is located, upon assembly of the magnet,
inside the all-welded helium enclosure, so that it only has to provide helium-to-helium tightness.
Another case was cold reception tests, for which the cryomagnets were connected to the test benches
by demountable, double-ring metal seals, providing a vacuum-monitored and spectrometer leak-
detected guard volume.
4.2 Air to sub-atmospheric helium leak-tightness

Even when operating with pressurized superfluid helium, there will be sections of the helium circuits
below atmospheric pressure: the cold parts of these circuits, vacuum insulated for thermal reasons,
therefore also benefit from a vacuum guard that prevents any contamination of the helium circuit by
air in-leak from the atmosphere. The room-temperature parts of the circuits are usually in contact with
the atmosphere, and any leak will result in contamination. In welded circuits, such leaks may occur
across demountable seals, instrumentation (e.g. pressure sensors), or non-tightly closing relief valves.
Corrective measures against such leaks, as schematized in Fig. 21, make use of helium guard volumes:
the helium guard may be a dedicated vessel enclosing the potentially leaky component, or the space
between double, concentric seals.
469
PT
He guard header
HP helium
Evacuation
for periodic rinsing
I
VLP circuits
Safety valves Instrumentation and components Static seal
not completely welded, with double
without double O-ring joints O-ring joints
Fig. 21: Helium guards against air in-leaks into sub-atmospheric helium circuits
Concerning the protection of cryogenic control valves operating at sub-atmospheric pressure,

such as the Joule–Thomson expansion valves mentioned in Section 3.1, it is advisable to install these
valves with the higher-pressure side not under the seat as recommended by general practice but, rather,
on the stem side where the sealing interface with the atmosphere occurs; an imperfect sealing would
then result in helium leakage to the atmosphere, rather than air in-leak into the helium circuit. For
cryogenic valves that have sub-atmospheric pressure on both sides, double sealing with an
intermediate helium guard is required on the stem interface with the atmosphere, as an application of
the general protection principles described above.
4.3 Precision thermometry

Low-temperature precision thermometry is always challenging, whether it uses the primary
thermometers of the International Temperature Scale ITS-90 [103] or more practical secondary
thermometers that have to be calibrated against them. In a cryogenic process at superfluid helium
temperature, errors in temperature measurement may result in significant irreversibilities; for example,
by requiring an increase of the temperature gradients for transporting heat so that they include error
bars. As an example, the total temperature gradient along a 3.3 km long sector of the LHC, from the
warmest magnet at the end of the sector to the refrigeration plant at 1.8 K, is only 100 mK, including
all heat transfer processes and uncertainties; allowing a 10% uncertainty on this value requires the
thermometers to have an absolute precision better than ±5 mK in this temperature range.
The primary thermometer of the ITS-90 in the superfluid helium domain is the helium vapour
pressure curve: measurement of temperature on a saturated bath is therefore reduced to that of
pressure, provided that all sources of parasitic heads in the measurement system – for example,
hydrostatic – are correctly identified and compensated. In other cases, one generally uses resistive
temperature sensors, usually based on semiconductor materials showing high non-linearity. Once the
selected sensors have demonstrated their stability over time, thermal cycles, and variation of
environmental conditions – for example, ionizing radiation – they need to be individually calibrated to
achieve precision [104]. The calibration data must then be reduced to be practically implemented in
readout electronics and/or monitoring software.
Finally, it must be remembered that the best thermometer can only measure its own
temperature! This is particularly important for thermometers operating in vacuum, for which good
thermal coupling to the object to be measured, protection against parasitic heat in-leaks, whether
conductive (along readout wires) or radiative, and limitation of internal Joule heating by the readout
current are essential to the quality of the measurement. Thermometric blocks integrating all these
functions [105] have been successfully developed, to be installed in the field by non-specialized
personnel.
470
5 Conclusion
The operation of superconducting devices in particle accelerators below 2 K, using superfluid helium
as a technical coolant, has now become state-of-the-art, as exemplified by the excellent operational
records of CEBAF at the Thomas Jefferson National Accelerator Facility, SNS at the Oak Ridge
National Laboratory, and the LHC at CERN. The specific aspects of superfluid helium technology –
addressed in this chapter – can be combined with standard cryogenic practice to design, build, and
operate complete helium II systems of industrial size. The superfluid-helium cooled particle
accelerator projects in construction or under study, such as the European Spallation Source (ESS) and
the International Linear Collider (ILC), however, represent major challenges and opportunities for
further progress, in view of their large size, their complexity, and the quest for reliability and
efficiency.
References
[1] G. Claudet and R. Aymar, Adv. Cryog. Eng. 35A (1990) 55–67.
[2] S.W. van Sciver, Cryogenics 32 ICEC Suppl. (1992) 320–327.
[3] H.J. Schneider-Muntau and J.C. Vallier, IEEE Trans. Magn. 24 (1988) 1067–1069.
[4] J.R. Miller, M.D. Bird, S. Bole, A. Bonito-Oliva, Y. Eyssa, W.J. Kenney, T.A. Painter,
H.J. Schneider-Muntau, L.T. Summers, S.W. van Sciver, S. Welton, R.J. Wood,
J.E.C. Williams, E. Bobrov, Y. Iwasa, M. Leupold, V. Stejskal, and R. Weggel, IEEE Trans.
Magn. 30 (1994) 1563–1571.
[5] S.W. van Sciver, J.R. Miller, S. Welton, H.J. Schneider-Muntau and G.E. McIntosh, Adv.
Cryog. Eng. 39A (1994) 375–380.
[6] W.D. Markiewicz, I.R. Dixon, C.A. Swenson, W.S. Marshall, T.A. Painter, S.T. Bole, T. Cosmus,
M. Parizh, M. King and G. Ciancetta, IEEE Trans. Appl. Supercond. 10 (2000) 728–731.
[7] NMR magnets by Bruker, http://www.bruker.com/products/mr/nmr/magnets/magnets.html
[8] R. Aymar, G. Claudet, C. Deck, R. Duthil, P. Genevey, C. Leloup, J.C. Lottin, J. Parain,
P. Seyfert, A. Torossian and B. Turck, IEEE Trans. Magn. MAG 15-1 (1979) 542–545.
[9] G. Claudet, G. Bon Mardion, B. Jager and G. Gistau, Cryogenics 26 (1986) 443–449.
[10] L.R. Evans, LHC accelerator physics and technology challenges, Proc. PAC99, Eds.
A. Luccio and W.W. MacKay (IEEE, Piscataway, NJ, 1999), pp. 21–25.
[11] Ph. Lebrun, Cryogenics 34 ICEC Suppl. (1994) 1–8.
[12] Ph. Lebrun, IEEE Trans. Appl. Supercond. 10 (2000) 1500–1506.
[13] G. Ciovati, AC/RF superconductivity, these proceedings.
[14] C.H. Rode, CEBAF cryogenic system, Proc. PAC95 (Am. Phys. Soc./IEEE, Piscataway, NJ,
1995), pp. 1994–1998.
[15] C. Rode and the JLab SNS team, The SNS superconducting linac system, Proc. PAC2001,
Eds. P. Lucas and S. Webber (IEEE, Piscataway, NJ, 2001), pp. 619–623.
[16] ESS Technical Design Report, European Spallation Source, Lund (2013).
[17] The European X-Ray Free-Electron Laser Technical Design Report, DESY 2006-097,
Hamburg (July 2007).
[18] F. Richard, J.R. Schneider, D. Trines and A. Wagner (Eds.), TESLA Technical Design
Report, DESY 2001-011 and ECFA 2001-209 (2001).
[19] ILC Technical Design Report (2013).
[20] T.J. Peterson, M. Geynisman, A. Klebaner, V. Parma, L. Tavian and J. Theilacker, Adv.
Cryog. Eng. 53 (2008) 1565–1572.
471
[21] F. Vinen, Physical properties of superfluid helium – a general review, Proc. Workshop on the
Stability of Superconductors in Helium I and Helium II, IIF/IIR Bulletin, Commission A1/2,
Saclay, 1981/6 (1981), pp. 43–51.
[22] F. Vinen, Physics of superfluid helium, Proc. CAS Superconductivity and Cryogenics for
Accelerators and Detectors, CERN-2004-008 (2004), pp. 363–374.
[23] L. Tavian, Large cryogenic systems at 1.8 K, Proc. EPAC 2000, Eds. J.L. Laclare,
W. Mitaroff, Ch. Petit-Jean-Genaz, J. Poole, and M. Regler (Austrian Academy of Sciences
Press, Vienna, 2000), pp. 212–216.
[24] B. Baudouy, Heat transfer and cooling techniques, these proceedings.
[25] V. Parma, Cryostat design, these proceedings.
[26] G. Bon Mardion, G. Claudet, P. Seyfert and J. Verdier, Adv. Cryog. Eng. 23 (1977) 358–362.
[27] B. Fallou, J. Galand and B. Bouvier, Cryogenics 10 (1970) 142–146.
[28] H. Winkelnkemper, Z. Krasucki, J. Gerhold and T.W. Dakin, Electra 52 (1977) 67.
[29] B. Rousset and F. Viargues, Cryogenics 34 ICEC Suppl. (1994) 91–94.
[30] P. Roubeau, C. R. Acad. Sci. Paris 273-14B (1971) 581–583.
[31] G. Claudet, Bath cryostats for superfluid helium cooling, in Handbook of Applied
Superconductivity, Part D9, (IOP Publishing, Bristol, 1998), pp. 795–800.
[32] V. Arp, Cryogenics 10 (1970) 96–105.
[33] C.J. Gorter and J.H. Mellink, Physica 15 (1949) 285–304.
[34] G. Bon Mardion, G. Claudet and P. Seyfert, Cryogenics 19 (1979) 45–47.
[35] S.W. van Sciver, Heat transfer in superfluid helium II, Proc. ICEC8, Ed. C. Rizzuto (IPC
Science & Technology Press, Guildford, 1980), pp. 228–237.
[36] P. Seyfert, Practical results on heat transfer to superfluid helium, Proc. Workshop on the
Stability of Superconductors in helium I and helium II, IIF/IIR Bulletin, Commission A1/2,
Saclay, 1981/6 (1981), pp. 53–62.
[37] S.W. van Sciver, Adv. Cryog. Eng. 27 (1982) 375–398.
[38] G. Claudet and P. Seyfert, Adv. Cryog. Eng. 27 (1982) 441–449.
[39] P. Seyfert, Results on heat transfer to He II for use in superconducting magnet technology,
Proc. ICEC9, Eds. K. Yasukochi and H. Nagano (Butterworth, Guildford, 1982), pp. 263–268.
[40] G. Claudet, F. Disdier, Ph. Lebrun, M. Morpurgo and P. Weymuth, Preliminary study of a
superfluid helium cryogenic system for the Large Hadron Collider, Proc. ICFA Workshop on
Superconducting Magnets and Cryogenics, BNL 52006, (Brookhaven National Laboratory,
USA, 1986), pp. 270–275.
[41] C. Meuris, B. Baudouy, D. Leroy and B. Szeless, Cryogenics 39 (1999) 921–931.
[42] H. Danielsson, Ph. Lebrun and J.M. Rieubland, Cryogenics 32 ICEC Suppl. (1992) 215–218.
[43] S. Caspi, The use of calorimetry in superfluid He II to measure losses in superconducting
magnets, Proc. ICEC9, Eds. K. Yasukochi and H. Nagano (Butterworth, Guildford, 1982),
pp. 347–350.
[44] P.L. Walstrom, Cryogenics 28 (1988) 151–156.
[45] G. Claudet, F. Disdier, A. Gauthier, Ph. Lebrun, M. Morpurgo and J. Schmid, Conceptual
study of the superfluid helium cryogenic system for the CERN Large Hadron Collider, Proc.
ICEC12, Eds. R.G. Scurlock and C.A. Bailey (Butterworth, Guildford, 1988), pp. 497–504.
[46] A. Kashani and S.W. van Sciver, Adv. Cryog. Eng. 31 (1986) 489–498.
[47] A. Kashani and S.W. van Sciver, Transient forced convection heat transfer in He II, Proc.
ICEC11, Eds. G. Klipping and I. Klipping (Butterworth, Guildford, 1986), pp. 654–658.
472
[48] P.L. Walstrom and J.R. Maddocks, Adv. Cryog. Eng. 33 (1988) 449–456.
[49] P.L. Walstrom, J.G. Weisend II, J.R. Maddocks and S.W. van Sciver, Cryogenics 28 (1988)
101–109.
[50] B. Rousset, G. Claudet, A. Gauthier, P. Seyfert, Ph. Lebrun, M. Marquet, R. van Weelderen
and J.L. Duchateau, Cryogenics 32 ICEC Suppl. (1992) 134–137.
[51] B. Rousset, G. Claudet, A. Gauthier, P. Seyfert, A. Martinez, Ph. Lebrun, M. Marquet and
R. van Weelderen, Cryogenics 34 ICEC Suppl. (1994) 317–320.
[52] H.A. Snyder and A.J. Mord, Adv. Cryog. Eng. 37A (1991) 81–88.
[53] A.J. Mord, H.A. Snyder and D.A. Newell, Cryogenics 32 (1992) 291–299.
[54] M. Morpurgo, Cryogenics 17 (1977) 91–93.
[55] A. Hofmann, A. Khalil, H.P. Krämer, J.G. Weisend, R. Srinivasan and B. Vogeley,
Investigations on fountain-effect pumps for circulating pressurized helium II, Proc. ICEC11,
Eds. G. Klipping and I. Klipping (Butterworth, Guildford, 1986), pp. 312–316.
[56] S.W.K. Yuan and T.C. Nast, Adv. Cryog. Eng. 33 (1988) 457–464.
[57] P. Kittel, Adv. Cryog. Eng. 33 (1988) 465–470.
[58] A. Hofmann, A. Khalil and H.P. Krämer, Adv. Cryog. Eng. 33 (1988) 471–478.
[59] P. Kittel, Losses in fountain-effect pumps, Proc. ICEC11, G. Klipping and I. Klipping
(Butterworth, Guildford, 1986), pp. 317–322.
[60] G. Klipping and H.D. Denner, Adv. Cryog. Eng. 35A (1990) 81–93.
[61] A.J. Mord and H.A. Snyder, Adv. Cryog. Eng. 39A (1994) 797–804.
[62] A. Hofmann, W. Herz, E. Suesser, B. Vogeley, T. Voekel and G. Zahn, Adv. Cryog. Eng.
39B (1994) 1813–1820.
[63] A.J. Mord and H.A. Snyder, Cryogenics 32 (1992) 461–465.
[64] J. Casas-Cubillos, A. Cyvoct, Ph. Lebrun, M. Marquet, L. Tavian and R. van Weelderen,
Cryogenics 32 ICEC Suppl. (1992) 118–121.
[65] A. Bézaguet, J. Casas-Cubillos, Ph. Lebrun, M. Marquet, L. Tavian and R. van Weelderen,
Adv. Cryog. Eng. 39A (1994) 649–656.
[66] D. Camacho, S. Chevassus, C. Policella, J.M. Rieubland, G. Vandoni and R. van Weelderen,
Thermal characterization of the He II LHC heat exchanger tube, Proc. ICEC17, Eds. D. Dew-
Hughes, R.G. Scurlock, and J.P. Watson (IOP Publishing, Bristol, 1998), pp. 647–650.
[67] A. Bézaguet, J. Casas-Cubillos, B. Flemsaeter, B. Gaillard-Grenadier, Th. Goiffon,
H. Guinaudeau, Ph. Lebrun, M. Marquet, L. Serio, A. Suraci, L. Tavian and R. van
Weelderen, Adv. Cryog. Eng. 41A (1996) 777–784.
[68] A. Bézaguet, J. Casas-Cubillos, H. Guinaudeau, B. Hilbert, Ph. Lebrun, L. Serio, A. Suraci,
L. Tavian and R. van Weelderen, Cryogenic operation and testing of the extended LHC
prototype magnet string, Proc. ICEC16/ICMC, Eds. T. Haruyama, T. Mitsui, and K. Yamafuji
(Elsevier Science, Amsterdam, 1997), pp. 91–94.
[69] B. Flemsaeter, Contribution to the dynamic analysis and optimal control of the superfluid
helium cooling loop for the LHC magnet string, Diploma thesis, Norwegian Institute of
Technology, Trondheim (1995).
[70] B. Flemsaeter, E. Blanco, J. Casas-Cubillos, C. de Prada and S. Saelid, Applying advanced
control techniques for temperature regulation of the LHC superconducting magnets, Proc.
ICEC17, Eds. D. Dew-Hughes, R.G. Scurlock, and J.P. Watson (IOP Publishing, Bristol,
1998), pp. 631–634.
473
[71] B. Rousset, A. Gauthier, L. Grimaud, A. Bézaguet and R. van Weelderen, Thermohydraulic

behaviour of He II in stratified co-current two-phase flow, Proc. ICEC16/ICMC, Eds.
T. Haruyama, T. Mitsui, and K. Yamafuji (Elsevier Science, Amsterdam, 1997), pp. 519–522.
[72] B. Rousset, A. Gauthier and A. Grimaud, Cryogenics 37 (1997) 733–737.
[73] L. Grimaud, A. Gauthier, B. Rousset and J.M. Delhaye, Cryogenics 37 (1997) 739–744.
[74] Ph. Lebrun, L. Serio, L. Tavian and R. van Weelderen, Adv. Cryog. Eng. 43A (1998) 419–426.
[75] B. Rousset, B. Jager, E. di Muoio, L. Puech, P. Thibaud, R. Vallcorba, R. van Weelderen and
P.E. Wolf, Adv. Cryog. Eng. 47B (2002) 1311–1318.
[76] Y. Huang, J. Kerby, T. Nicol and T. Peterson, Adv. Cryog. Eng. 43A (1998) 403–410.
[77] R. Byrns, Y. Huang, J. Kerby, Ph. Lebrun, L. Morrison, T. Nicol, T. Peterson, R. Trant,
R. van Weelderen and J. Zbasnik, The cryogenics of the LHC interaction region final-focus
superconducting magnets, Proc. ICEC17, Eds. D. Dew-Hughes, R.G. Scurlock, and
J.P. Watson (IOP Publishing, Bristol, 1998), pp. 743–746.
[78] G. Gistau-Baguer, High-power refrigeration at temperatures around 2 K, Proc.
ICEC16/ICMC, T. Haruyama, T. Mitsui, and K. Yamafuji (Elsevier Science, Amsterdam,
1997), pp. 189–194.
[79] F. Millet, P. Roussel, L. Tavian and U. Wagner, Adv. Cryog. Eng. 43A (1998) 387–394.
[80] E. Kashtanov, V. Pleskatch, K. Polkovnikov, A. Shembel, V. Sytnik, S. Zintchenko,
B. Krakovsky, O. Krasnikova, O. Popov, H. Burmeister, B. Petersen and H.O. Roggenbuck,
Large low-pressure heat exchanger for the TTF cryogenic system, Proc. ICEC18, Ed. K.G.
Narayankhedkar (Narosa, New Delhi, 2000), pp. 315–318.
[81] F. Viargues, G. Claudet and P. Seyfert, Cryogenics 34 ICEC Suppl. (1994) 325–328.
[82] P. Roussel, A. Bézaguet, H. Bieri, R. Devidal, B. Jager, R. Moracchioli, P. Seyfert and
L. Tavian, Performance tests of industrial prototype subcooling helium heat exchangers for
the Large Hadron Collider, Adv. Cryo. Eng. 47B (2002) 1311-1318
[83] G.M. Gistau and G. Claudet, The design of the helium refrigerator for Tore Supra, Proc.
ICEC10, Eds. H. Collan, P. Berglund, and M. Krusius (Butterworth, Guildford, 1984),
pp. 288–291.
[84] G.M. Gistau, J.C. Villard and F. Turcat, Adv. Cryog. Eng. 35B (1990) 1031–1037.
[85] G.M. Gistau, Y. Pecoud and A.E. Ravex, Adv. Cryog. Eng. 33 (1988) 675–681.
[86] C.H. Rode, D. Arenius, W.C. Chronis, D. Kashy and M. Keesee, Adv. Cryog. Eng. 35A
(1990) 275–286.
[87] L. Decker, K. Löhlein, P. Schustr, M. Vins, I. Brunovski, L. Tucek, Ph. Lebrun and L. Tavian,
A cryogenic axial–centrifugal compressor for superfluid helium refrigeration, Proc.
ICEC16/ICMC, Eds. T. Haruyama, T. Mitsui, and K. Yamafuji (Elsevier Science,
Amsterdam, 1997), pp. 195–198.
[88] L. Decker, A. Kündig, K. Löhlein, W. Pürtschert, B. Ziegler, Ph. Lebrun, L. Tavian,
I. Brunovsky and L. Tuček, Adv. Cryog. Eng. 43A (1998) 637–642.
[89] M. Bonneton, L. Tavian, G. Gistau-Baguer, F. Turcat and P. Viennot, Adv. Cryog. Eng. 43A
(1998) 643–650.
[90] N. Saji, H. Asakura, S. Yoshinaga, K. Itoh, T. Nogaku, A. Bézaguet, J. Casas, Ph. Lebrun, and
L. Tavian, A 1 kPa centrifugal cold compressor for the 1.8 K helium refrigeration system,
Proc. ICEC17, Eds. D. Dew-Hughes, R.G. Scurlock, and J.P. Watson (IOP Publishing,
Bristol, 1998), pp. 295–298.
[91] A. Bézaguet, Ph. Lebrun and L. Tavian, Performance assessment of industrial prototype
cryogenic helium compressors for the Large Hadron Collider, Proc. ICEC17, eds. D. Dew-
Hughes, R.G. Scurlock, and J.P. Watson (IOP Publishing, Bristol, 1998), pp. 145–148.
474
[92] S. Claudet, P. Gayet, B. Jager, F. Millet, P. Roussel, L. Tavian and U. Wagner, Specification
of eight 2400 W @ 1.8 K refrigeration units for the LHC, Proc. ICEC18, Ed.
K.G. Narayankhedkar (Narosa, New Delhi, 2000), pp. 207–210.
[93] B. Hilbert, G.M. Gistau-Baguer and F. Dagut, 2.4 kW @ 1.8 K refrigeration units for the LHC
project: the Air Liquide system, Proc. ICEC18, Ed. K.G. Narayankhedkar (Narosa, New
Delhi, 2000), pp. 211–214.
[94] H. Asakura, J. Boesel, T. Honda, A. Kuendig, K. Kurtcuoglu, A. Meier, M. Mori, A.E. Senn
and S. Yoshinaga, Four 2400W/1.8 K refrigeration units for CERN-LHC: the IHI/Linde
system, Proc. ICEC18, Ed. K.G. Narayankhedkar (Narosa, New Delhi, 2000), pp. 215–218.
[95] J.Ph. Guignard, Contribution à l'étude de la stabilité de fonctionnement et de l'adaptation de
charge des compresseurs centrifuges cryogéniques multiétages, Travail d’option ‘Machines-
Energétique’, Ecole des Mines de Paris (1993).
[96] M. Kauschke, C. Haberstroh and H. Quack, Adv. Cryog. Eng. 41A (1996) 921–926.
[97] F. Minot, B. Gravil, A. Bocquillon, C. Dugas, D. Henry, J.L. Maréchal, S. Nicollet,
P. Prochet, J.P. Serries and R. Simon, Problems encountered with subatmospheric circuits on
the Tore Supra cryogenic system, Proc. ICEC17, Eds. D. Dew-Hughes, R.G. Scurlock, and
J.P. Watson (IOP Publishing, Bristol, 1998), pp. 609–612.
[98] S. Claudet, Ph. Lebrun and L. Tavian, Towards cost-to-performance optimisation of large
superfluid helium refrigeration systems, Proc. ICEC18, Ed. K.G. Narayankhedkar (Narosa,
New Delhi, 2000), pp. 203–206.
[99] F. Bertinelli, P. Borowiec, A. Kotarba, S. Olek and Z. Sułek. The quality control of the LHC
continuous cryostat interconnections, Proc. EPAC08 Genoa, JACoW (2008), pp. 2398–2400.
[100] P. Fessia, C. Lanza and S. Sgobba, IEEE Trans. Appl. Supercond. 14 (2004) 1185–1188.
[101] N. Bourcey, P. Campos, P. Chiggiato, P. Limon, A. Mongelluzzo, G. Musso, A. Poncet and
V. Parma, Adv. Cryog. Eng. 53A (2008) 325–332.
[102] F. Bertinelli, G. Favre, L.M.A. Ferreira, S. Mathot, L. Rossi, F. Savary and E. Boter, Design
and fabrication of superfluid helium heat exchangers tubes for the LHC superconducting
magnets, Proc. EPAC 2004 Lucerne, JACoW (2004), pp. 1837–1839.
[103] H. Preston-Thomas, Metrologia 27 (1990) 3–10.
[104] Ch. Balle, J. Casas-Cubillos, N. Vauthier and J.P. Thermeau, Adv. Cryog. Eng. 53B (2008)
965–972.
[105] Ch. Balle and J. Casas-Cubillos, Adv. Cryog. Eng. 41 (1996) 1715–1721.
Bibliography
F. London, Superfluids, vol. II, Macroscopic Theory of Superfluid Helium (John Wiley & Sons, New
York, 1954; republished by Dover, New York, 1964).
Proceedings of the CERN Accelerator School ‘Superconductivity and cryogenics for accelerators and
detectors’, CERN-2004-008 (2004).
Proceedings of the CERN Accelerator School ‘Superconductivity in particle accelerators’, CERN
Report 89-04 (1989).
Proceedings of the CERN Accelerator School ‘Superconductivity in particle accelerators’, CERN
Report 96-03 (1996).
Proceedings of the symposium ‘A half century of superfluid helium’. Adv. Cryog. Eng. 35A (1990) 1–93.
S.W. van Sciver, Helium Cryogenics (Plenum Press, New York, 1986; 2nd ed. 2012).
475
J. Wilks, The Properties of Liquid and Solid Helium (Clarendon Press, Oxford, 1967).
J. Wilks and D.S. Betts, An Introduction to Liquid Helium (Clarendon Press, Oxford, 1987).
476
Superconductor Dynamics
F. Gömöry 1
Institute of Electrical Engineering, Slovak Academy of Sciences, Bratislava, Slovakia
Abstract
Superconductors used in magnet technology could carry extreme currents
because of their ability to keep the magnetic flux motionless. The dynamics
of the magnetic flux interaction with superconductors is controlled by this
property. The cases of electrical transport in a round wire and the
magnetization of wires of various shapes (circular, elliptical, plate) in an
external magnetic field are analysed. Resistance to the magnetic field
penetration means that the field produced by the superconducting magnet is
no longer proportional to the supplied current. It also leads to a dissipation of
electromagnetic energy. In conductors with unequal transverse dimensions,
such as flat cables, the orientation with respect to the magnetic field plays an
essential role. A reduction of magnetization currents can be achieved by
splitting the core of a superconducting wire into fine filaments; however,
new kinds of electrical currents that couple the filaments consequently
appear. Basic formulas allowing qualitative analyses of various flux dynamic
cases are presented.
Keywords: type II superconductors, flux pinning, critical state model.
1 Introduction
The main motivation for using type II superconductors (in these materials the magnetic flux is divided
into flux quanta called flux lines) in electromagnets originates in their ability to carry large electrical
currents without dissipation. This property, which enables the generation of stationary magnetic fields
beyond the reach of traditional technology based on metallic conductors, is the consequence of a
mechanism that hinders the motion of magnetic flux inside such a superconductor [1, 2]. Flux lines
interact with pinning centres that usually are imperfections in the composition or in the structure. This
phenomenon is illustrated schematically in Fig. 1, which shows a superconducting plate with finite
thickness in the x-direction and very large dimensions along the other two axes exposed to a magnetic
field parallel to the z–y plane. For simplicity, just one kind of pinning centre is shown on the right-
hand side of Fig. 1 (see the white objects of irregular shape randomly distributed in the volume of the
superconductor). When the applied magnetic field, Bex, is switched on, the magnetic flux enters the
plate in the form of flux lines, each carrying the elementary quantum of magnetic flux, Φ0 = 2 × 10–15
Wb. A rough estimate of the distance between the flux lines can be obtained by assuming the absence
of any mechanism hindering the flux penetration, i.e. neglecting also a flux pinning. In that case, the
lines would distribute uniformly and the flux density inside the plate would be roughly equal to the
external applied field, Bex. Let us assume that the flux lines form the quadratic lattice shown in Fig. 2.
Its line spacing could be then calculated as
Φ0
a= , (1)
Bex
1
elekgomo@savba.sk
F. G ÖMÖRY
resulting in the applied field Bex = 1 T at distance a = 45 nm. In other words, there are a very large
number of flux lines interacting with a superconducting wire with a typical diameter of 1 mm. This
conclusion remains valid after detailed calculations, indicating that a hexagonal lattice would be
preferred. Now consider the influence of flux pinning. Because the pinning centres obstruct the flux
penetration, we will find the highest density of flux lines close to the sample surface, and it will
decrease inwards. By applying the Maxwell equation in quasi-static conditions (i.e. neglecting the
displacement current) and taking into account all the symmetries, we find for the plate shown in Fig. 1
that
dB y ( x )
= − j z ( x) . (2)
dx
Fig. 1: Left: slab from a hard superconductor exposed to a parallel magnetic field Bex. Right: the existence of
imperfections (white spots in the upper figure) causes a non-uniformity in the concentration of magnetic flux
lines (lower figure). Magnetic flux cannot move freely inside a hard superconductor because the imperfections
act as pinning centres. Macroscopic currents resulting from the gradient magnetic flux density are indicated in
the left-hand picture.
Fig. 2: Regular lattice of flux lines created in a type II superconductor when exposed to the magnetic field in the
z-direction. The central parts (black) denote the normal cores surrounded by circulating supercurrents.
478
S UPERCONDUCTOR DYNAMICS
This expression can be interpreted in the following way: a macroscopic electrical current is
equivalent to a gradient in the magnetic flux density. It is valid for both normal conductors and
superconductors. In a metal the currents (called the eddy currents) dissipate because of normal metal
resistivity and the field gradients disappear with time. Only in AC fields with periods shorter than the
dissipation time will field gradients persist, leading to the so-called ‘skin effect’. However, in the
superconductor shown in Fig. 1 the gradient of the density of flux lines—thus of the flux density
itself—persists in the quasi-static conditions of switching on the external field. Thus Eq. (2) predicts
the existence of persistent macroscopic currents flowing in the z-direction. The density of the flux
lines is lower than that equivalent to the applied field. In other words, the magnetic field in a
superconducting plate is ‘screened’ by these currents, which produce a magnetic field with a polarity
that opposes the applied field Bex.
This simple example shows that the ‘freezing’ of the flux line positions by pinning forces is
essential to reach the large stationary densities of electrical current necessary in the construction of
electromagnets able to generate static magnetic fields exceeding 10 T. However, the mechanism that
serves to keep the distribution of the magnetic field and current density unchanged opposes any
change in the magnetic field. Such superconductors are referred to as ‘hard’ because in a cyclic
magnetization process they exhibit hysteresis resembling the behaviour of a magnetic material with
large coercivity. The mechanism of flux pinning, however, causes hysteresis of the magnetic flux
distribution with respect to the current supplied to the magnet winding. As a consequence, the
dynamics of the magnetic flux movement in an electromagnet made from a hard superconductor is
controlled by the flux pinning mechanism.
2 Critical state model

The interaction of flux lines with pinning centres is understood for some simple cases, and there is
extensive research activity dedicated to this issue. Nevertheless, the complexity of superconducting
materials and the variety of pinning centres encountered make it difficult to establish a procedure
allowing the prediction of the macroscopic volume density of the pinning force directly from
microscopic calculations. It is more practical to formulate a phenomenological model describing the
macroscopic behaviour of a hard superconductor. Its simplest version was developed by Bean [3], and
is commonly known as the ‘critical state model’. It is valid on a macroscopic scale, ignoring the
details of a magnetic field distribution around an individual flux line. Working with an average taken
over many flux lines, the critical state model states that in any (macroscopic) part of a hard
superconductor one can find either no electrical current or a current with density equal to the so-called
‘critical current density’, jc. In the original formulation its value is constant and it characterizes fully
the properties of a material that is a hard superconductor, i.e. a type II superconductor with flux
pinning. Several later adjustments were incorporated in order to include the properties of
superconductors in a more realistic way, e.g. the suppression of the critical current density by the
magnetic field [4] or its dependence on the electrical field [5, 6]. With these modifications the results
can be better compared to experimental data; however, the results are no longer analytical. Therefore
the original version with jc = const. is still used to obtain a basic understanding of the problem, and we
will use this in the following.
The actual value of the current density in a hard superconductor is controlled by the preceding
history. No current flows in the regions where the magnetic flux vortices had not penetrated before,
and in the rest of the superconductor the flux line density gradient provides a macroscopic current
density according to Eq. (2). Because the density of the flux lines controls the magnetic field in a
superconductor, the critical state model states that
0 in the places where B = 0,
j = (3)
 jc elsewhere.
479
F. G ÖMÖRY
Interestingly, the electrical field that is commonly used to calculate electrical currents in normal
metals is not explicitly found in this expression. However, because another Maxwell equation states
that any change in magnetic field is accompanied by the appearance of an electrical field, all the
regions in a hard superconductor in which a non-zero current density appears will have had to
experience a finite electrical field during the transient process of magnetic field penetration. In fact,
this is the way to include the history of the magnetization process into the analysis of flux dynamics in
hard superconductors.
In spite of its simplicity, the critical state model defined by Eq. (3) has provided many useful
predictions that can also be reached in an analytical way.
2.1 Magnetic flux dynamics at current transport

Let us consider a long wire with circular cross-section (its length, l, significantly exceeds its
radius, R), made from a hard superconductor characterized by the critical current density jc. The
maximum value of DC current that the wire can carry without resistance is
I c = jc π R 2 . (4)
When the current I1 (lower than Ic) is supplied to the wire in the stand-alone conditions (no
additional magnetic field applied), only a part of its cross-section will carry the electrical current and
experience a non-zero magnetic field. The process of injecting I1 (through the terminations that are at
the wire ends, far from the portion of wire under investigation) is accompanied by the appearance of a
temporary electrical field that provokes electrical currents that could not be deleted because of flux
pinning. The only solution compatible with the critical state equation (2) is the one shown in the
middle part of Fig. 3: a current with density jc fills the outer shell with a cross-section that is just right
to collect the necessary value of transported current. The existence of parallel paths along the wire axis
with zero current density (in the centre) and non-zero current density (in the outer shell) does not
violate any law of electromagnetism, because in the stationary state the electrical field is zero in the
current-free central zone as well as in the outer shell—the latter is a consequence of superconductor
zero resistivity. Nevertheless, in the transient process of establishing the transport current, there has
been an electrical field in the outer shell. Calculating its distribution in space and time together with
the time evolution of the current density distribution allows the determination of the cost of
electromagnetic energy (furnished by the supply of wire current) necessary to establish the final
distribution [7]. A further increase in current to the value I2 > I1 (but still below Ic) will produce
distributions similar to that shown in the centre of Fig. 3, with the subsequent shrinking of the central
current-free zone. Eventually, when the transport current reaches Ic, all the wire section will be filled
by the critical current density, jc.
Fig. 3: Series of distributions of magnetic field (lines of constant vector potential) and electrical current density
(grey scale) calculated for a round wire of radius R = 0.5 mm and a critical current of Ic = 100 A during the first
increase of transported current.
480
Probably the most striking consequence of the critical state model is reached when the wire
current decreases. Because the critical current density once induced in a hard superconductor cannot
be cancelled, the only way to reduce the total current transported in the wire is by reversing the current
polarity in a part of it. Similar to the case of current increase, it is the outer shell where this process
starts, as can be found through a detailed analysis of the electrical field at the transition of total current
in the wire from Ic to I3 < Ic. The series of distributions at decreasing current is shown in Fig. 4. Note
that the entire section of wire is filled by the critical current density even when the total current is
reduced to zero. The difference between this distribution and the original one shown on the left of
Fig. 3 is due to flux pinning. History dependence (hysteresis) in the current distribution indicates that,
from the thermodynamic point of view, the process is irreversible and that a part of the
electromagnetic energy provided by the current supply has been converted to heat. This quantity is
called the AC loss because it is commonly determined in a cyclic process, in this example at the
transport of alternating current with sinusoidal waveform.
(grey scale) calculated for a round wire of radius R = 0.5 mm and a critical current of Ic = 10 A during the
reduction of transported current from Ic down to zero.
(grey scale) calculated for a round wire of radius R = 0.5 mm and a critical current of Ic = 100 A during the
sweep of transported current from 80% of Ic, to −80%, and then to zero. The central parts (white) denote the
current- and field-free neutral zone where the electrical field remained zero throughout.
Usually a superconducting wire operates at transport currents that are always lower than the
critical current. Then the central part of the wire remains free of current at all times. Considerations
within the critical state model state that no magnetic field (no flux lines) has penetrated to this so-
called ‘neutral zone’, and the electrical field remained zero also during the change of field and current
distribution. This property can be used to evaluate the AC loss in the wire carrying an AC current
Iac(t) = Iasinωt with the amplitude Ia and frequency f =ω/2π. In fact, the voltage measured on such a
481
F. G ÖMÖRY
wire can be derived from the path integral of the electrical field on the rectangular loop shown in
Fig. 6. Because the inner leg of the loop is in the neutral zone, the electric field on that part is zero.
Two lines perpendicular to the wire axis do not contribute either, because the electrical field in the
axial direction could not exist in this geometry. All the voltage induced in the loop due to the change
of the linked magnetic flux with time is on the outer surface of the wire and can be picked by two
voltage taps:
∂Φ
U =− . (5)
∂t
This relation is very useful in expressing the loss of electromagnetic energy during one cycle with
period T = 2π/ω, with the result
= W ∫= UI ac dt ∫ I ac dΦ , (6)
T
which is simply the area of loop enclosed by the Φ(Iac) dependence during one cycle of AC current.
The loops for two cycles are shown in Fig. 7. The absence of time as an independent variable means
that the shape of this loop does not depend on the frequency f = ω/2π. This feature is common for all
cases for which the flux pinning controls the flux penetration. The accompanying dissipation is
therefore called the hysteresis loss.
Fig. 6: The voltage measured by a pair of taps on the surface of a round wire formed of hard superconductor
during transport of electrical current. The voltage can be calculated from the change in magnetic flux enclosed in
the rectangular loop consisting of two radial sections perpendicular to the wire axis, with the longitudinal section
enclosed in the neutral zone and the line connecting two voltage taps. Electrical fields on all parts except the
wire surface are zero.
Fig. 7: Dependence of magnetic flux enclosed between the wire surface and the neutral zone, Φ, on the wire
current, Iac, at an amplitude equal to 80% of Ic (squares) and at Ic (diamonds). It begins at (0,0) with the initial
part shown in Fig. 3; after the first turn in the current ramping, repetitive cycles with closed loops follow.
482
2.2 Magnetic flux dynamics in transient magnetic field

The example of AC transport by a round wire has demonstrated the main features of magnetic flux
dynamics in a hard superconductor. Closer to the actual conditions in the winding of an electromagnet
is the case when such a wire is exposed to an external magnetic field that changes in time. In contrast
to the previous case, the solutions cannot be found analytically. Fortunately, numerical methods
incorporating the critical state principles are available [8–10].
2.2.1 Wire with circular cross-section

Figure 8 presents the results of calculations for some significant points in the magnetization cycle of a
round wire exposed to a transversal magnetic field. Braking of the magnetic flux penetration is
equivalent to the appearance of a current loop (two currents of opposite direction) generating a
magnetic field that opposes the change in field inside the superconductor. At the first field increase the
central part remains free of currents. This would persist in cases when the maximum field (the
amplitude of the field in the case of cyclic AC magnetization) does not exceed the so-called
‘penetration field’, Bp, defined as the field at which all sections of the wire are filled with critical
current density.
In order to characterize the process of magnetic flux penetration into the wire exposed to an
external magnetic field, the dependence of its magnetization, M, on the actual value of the external
field, Bex, is commonly used. This is illustrated in Fig. 9, where the loop calculated for the wire shown
in Fig. 8 is presented. At the penetration field, Bp, the sample magnetization reaches the saturation
magnetization, Ms. Its value can be obtained from the following consideration, illustrated by Fig. 10,
which shows the current distribution in a wire with its length parallel to the z-coordinate exposed to
the magnetic field in the y-direction [11]. The magnetic moment will have only the y-component that
is at any instant of the magnetization cycle calculated by performing the integration (see also
Appendix A)
m = ∫ − j ( x, y )dxdy (7)
S
over the wire cross-section, S. When the saturation state has been reached—this is the situation
actually shown in Fig. 10—this expression can be rewritten as
∫ − xjc dxdy +
ms =
x<0
∫
x >0
2 jc ∫ xdxdy =
xjc dxdy =
x >0
jc Sx , (8)
where x is the average distance of the x-coordinate from the axis of symmetry (in Fig. 10 this is the y-
axis) in the wire cross-section, i.e.
1
x = ∫ x dS . (9)
SS
The quantity 2 x may be interpreted as the wire dimension transverse to the applied field. It
can be evaluated in a similar way as for a wire of any shape S, although the integration could be more
complicated when the boundary between two orientations of current density does not coincide with
any of the principal coordinate axes. The saturation magnetization is given by
ms
M= s = jc x . (10)
S
4R 4R
For a round wire with radius R, one obtains x = and M s = jc , respectively.
3π 3π
483
F. G ÖMÖRY
(grey scale) calculated for a round wire of radius R = 0.5 mm and critical current of Ic = 100 A during the
application of an external magnetic field, Bex, in a direction transverse to the wire axis. Actual values of Bex (in
milli-tesla) are indicated above the plots.
Fig. 9: Dependence of magnetization on the applied field calculated for the round wire of Fig. 8
Fig. 10: Saturation of round wire with screening currents
484
2.2.2 Wire with elliptical cross-section—effect of magnetic field orientation

The importance of the transverse dimension can be nicely illustrated by the following example.
Assume a wire with an elliptical cross-section, the relation between the main axes (commonly called
the aspect ratio) being a:b = 10:1. Two typical cases of its orientation with respect to the applied field
are shown in Fig. 11. When the field is parallel to the minor axis—this is called the ‘perpendicular’
orientation—the values of the transverse dimension and the saturation magnetization will be
4a 4a
= x⊥ = ; M s jc , (11)
3π 3π
respectively. Rotating the wire by 90° will lead to the configuration called ‘parallel’, with the
following values of transverse dimension and saturation magnetization:
4b 4b
= x|| = ; M s jc . (12)
3π 3π
Because a >b, the values for the perpendicular field are larger than those for the parallel field, by
exactly the same ratio as the ellipse axes.
Fig. 11: Exposure of wire with elliptical cross-section to a perpendicular or parallel magnetic field
Let us now analyse the difference in the magnetic flux dynamics between two orientations of a
wire with an elliptical cross-section. For this purpose we compare the magnetization loops calculated
in two cases (by a numerical procedure) and plotted in Fig. 12. The difference in the loop ‘thickness’
controlled by the saturation magnetization is spectacular. This is because the same ratio (a/b = 10)
exists for the values of Ms. Much less prominent is the difference in the penetration fields, Bp, which
can be identified as the field occurring when the magnetization reaches its saturation value. The flux
lines before entering the wire deform in order to adapt to become roughly parallel to the surface. This
buckling creates more line tension in the perpendicular orientation; therefore the push to enter the wire
is bigger, and the penetration process is quicker than in the parallel case. However, this difference is
far lower (a factor of about 2 in this case) than in the saturation magnetization. Because the dissipation
of the electromagnetic energy in one cycle—commonly called the AC loss—per unit length of wire is
given by a formula indicating that it is proportional to the loop area,
Ql = ∫ mdBex [J·m −1], (13)
the AC loss in the perpendicular field is much bigger than in the parallel field.
485
F. G ÖMÖRY
The main consequences for electromagnets manufactured from wires containing hard
superconductors, arising from the existence of magnetization currents in these materials, are as
follows:
• magnetization currents generate a macroscopic magnetic field that is not proportional to the
supplied current, e.g. after switching off the current a remanent field remains in the bore;
• ramping of the magnetic field generates a dissipation that warms up the magnet winding, so a
cooling system must be designed to remove this AC loss.
In a magnet designed to use the transport capacity of a superconducting material in a reasonable way,
the magnetization currents saturate the wires in a substantial portion of the winding. In other words,
the local magnetic field is far larger than the penetration field of the wire, Bp. Then the essential
parameter controlling the behaviour is the saturation magnetization. The volume density of the AC
loss  i.e. the dissipation Q released in the superconductor volume V  in the cycle with AC field
amplitude Ba >> Bp can be estimated as
Q
= 4 M s Ba . (14)
V
Two cases for the orientation of the magnetic field with respect to a wire of elliptic cross-section with
aspect ratio 10 will then result in the prediction of an AC loss 10 times higher in the perpendicular
case because of the difference in the saturation magnetization.
Fig. 12: Magnetization loops calculated for a wire of elliptical cross-section, where a = 1 mm, b = 0.1 mm,
Ic = 100 A, exposed to the magnetic fields of parallel (dashed) and perpendicular (solid) orientation with respect
to the major semi-axis.
2.2.3 Slab in parallel field

The distributions shown in Figs. 8 and 11 and the magnetization loops plotted in Figs. 9 and 12 have
been calculated numerically because there is no analytical solution available for the case of a wire
magnetized in a perpendicular field. One can avoid setting up a numerical model by using the
486
analytical approximation of these results [12]. On the other hand, analytical formulas are available for
the case of a superconducting slab of thickness w, as shown in Fig. 1. Magnetic flux profiles and
current distributions obtained at a cyclical change of magnetic field applied parallel to the slab are
plotted in Fig. 13 for the field amplitude just reaching the penetration field value and in Fig. 14 for
larger field amplitudes. The value of Bp is obtained with the help of Eq. (2) under the assumption that
the field just reaches the centre of the slab, which is at a distance w/2 from the surface:
w
Bp,slab = µ0 jc . (15)
2
The saturation magnetization is easily calculated from Eq. (10) by taking into account that, for a slab
with thickness w in a parallel applied field, x = w 4 :
w Bp,slab
M= s,slab j=
c . (16)
4 µ0
The volume density of loss in cyclic magnetization in an AC field with amplitude Ba was derived for
this case [3] as follows:
 2 Ba3
 for Ba < Bp ,
Q 1  3 Bp
=  (17)
V µ0  4 2
2 Bp Ba − 3 Bp for Ba > Bp ,
with Bp = Bp,slab given by Eq. (15).
Fig. 13: Series of profiles of magnetic field (upper part) and electrical current density (lower part) at the first
increase (left) and subsequent reduction (right) of the magnetic field applied in a parallel direction to a slab
infinite in the y–z plane. A particular case of maximal field equal to the penetration field, Bp, is shown.
487
F. G ÖMÖRY
Fig. 14: Similar to Fig. 13, but the field increases from Bp (left) and is then reduced to a negative value (right).
Note that the entire section of the slab is filled with critical current density.
The plot of this dependence for slabs with different thicknesses and critical current densities is
shown in Fig. 15. Among other things, it illustrates the fact that two different slabs exhibit the same
volume density of loss provided the penetration field, Bp, is the same. An important feature of loss
dependence is the change of slope when the amplitude trespasses the value of Bp: from Q ∝ Ba3
observed below penetration, it reduces to Q ∝ Ba for large amplitudes Ba >> Bp in accordance with
the general formula (14). In spite of fact that the slab geometry is far from describing a wire used to
wind a superconducting magnet, Eq. (17) was widely used to predict the loss in superconducting
magnets developed for a generation of magnetic fields varying in time, such as dipoles for particle
accelerators.
Fig. 15: Dependence of the volume density of AC loss calculated with the help of Eq. (17) for four slabs
differing in thickness w and critical current density jc. The losses calculated for two slabs with identical values of
the penetration field calculated from Eq. (15) are the same.
488
Nowadays, with increased computing power and numerical methods developed for the
calculation of magnetic flux dynamics in hard superconductors, such approach serves only for
qualitative considerations and rough estimations. Interestingly, the results [e.g. 13, 14] found by such
computations always follow the general features shown in Fig. 16: there is a change in the slope at
amplitudes comparable to the penetration field and an inverse order of loss level below and above this
‘knee’. At low amplitudes, a smaller Bp means higher losses; at large amplitudes, it results in lower
losses.
Fig. 16: Electrical loop created by the currents coupling two superconducting filaments when exposed to the
magnetic field Bex changing in time (in this particular case, it is increasing).
3 Multicore wires and coupling currents

One important consequence of Eq. (10) is that the only way to reduce the magnetization currents
without lowering the transport capacity of a superconducting wire—as well as the AC loss at large AC
field amplitudes given by Eq. (14)—is to minimize the dimension of the wire that is perpendicular to
the applied field. In the case of flat conductors, such as Rutherford cables or tapes coated with high-
temperature superconductors, this means one must avoid the exposure of the magnetic field
perpendicular to the flat face. This is not always possible, and therefore a remedy was developed that
involved splitting the superconducting core into many filaments. Imagine dividing a wire of circular
cross-section with radius R0 into N circular filaments. Maintaining the critical current requires that the
total cross-section of the superconducting material be the same, so the filament radius will be
R N = R0 N . Then, according to Eq. (10), the saturation magnetization will be reduced by a
factor N . Therefore the wires used in magnets generating a pulse or AC magnetic field are
composites containing fine (5–50 µm in diameter) superconducting filaments in a metallic matrix. The
matrix—besides assisting thermal and mechanical stabilization—allows a transfer of electrical current
in the transverse direction and the optimal distribution of this current among the superconducting
filaments. However, another phenomenon influencing the macroscopic magnetization and the AC loss
will appear at the ramp of the magnetic field: electrical currents connecting the filaments across the
metallic matrix in closed loops, commonly called the coupling currents.
The mechanism of magnetically induced coupling is illustrated schematically for just two
filaments in Fig. 16. The current loop provoked by the change in applied magnetic field consists of the
part running along the superconducting filaments, but it must be closed either across the normal
conducting matrix or at the ends of filaments. Let us now compare the flux dynamics in two perfectly
coupled parallel round filaments for the case when the filaments would be completely insulated.
Distributions of current density and magnetic field calculated for these two cases are shown for some
representative instants of the magnetization cycle in Fig. 17. When the connection between filaments
is perfect, all the current running along one filament will return through the second one:
∫ jdS = − ∫ jdS ,
1,coupled 2,coupled
(18)
489
F. G ÖMÖRY
where the indices 1 and 2 denote the cross-sections of the two filaments, respectively. In the case of
non-existing coupling,
= ∫ jdS =
1,uncoupled
∫ jdS 0 ,
2,uncoupled
(19)
i.e. all the current induced by the field change must return through the same filament. Let us evaluate
the transverse dimension in these two cases. For uncoupled round filaments, each with radius R, the
value of x is the same as for the single round wire with radius R:
4R
x2,uncoupled = . (20)
3π
In the case of perfect coupling, the maximum magnetic moment is determined by the distance between
filaments, d, and one can estimate that
d
x2,coupled = . (21)
2
Fig. 17: Distributions of magnetic field (lines of constant vector potential) and electrical current density (grey
scale) calculated for a pair of round wires during the change in applied magnetic field, Bex in a direction
transverse to the plane connecting the axis of the two wires. The upper part shows the result obtained when
assuming a perfect galvanic connection at the terminations of the wires; the lower part shows the distributions
calculated for two insulated wires.
Magnetization loops calculated by a numerical finite element method [10] are given for these
two cases in Fig. 18. The proportion of values for the saturation magnetization is given by the ratio
of x , which is evaluated using Eqs. (20) and (21). Using the values R = 1 mm and d = 4 mm, one finds
that Ms in the coupled case should be 4.7 times higher than in the uncoupled case, which is in good
agreement with the numerically calculated result.
The existence of coupling currents means that splitting a single superconducting core into many
filaments is not in itself sufficient to depress the magnetization currents and reduce AC loss. An
additional measure is required to uncouple the filaments. This was achieved by means of a
transposition through twisting the whole filamentary zone. The polarity of the electrical field induced
by the change in the external magnetic field alternates in the half-loops between filaments, as shown in
Fig. 19. The net voltage generated along one filament is therefore negligible, and thus there is no
driving force to create interfilament currents. However, within one half of the twist pitch, lp, a
potential difference between parallel filaments remains, leading to a current traversing the matrix. This
490
mechanism can be interpreted as a diffusion of magnetic flux opposed by coupling currents [1] with
time constant
2
µ  lp 
τ= 0   , (22)
2ρt  2π 
where ρt is the effective transverse resistivity of the multifilament composite. This quantity is
estimated from the resistivity of the matrix, ρm, and the volume fraction occupied in the wire by the
superconductor, λ, taking into account that superconducting filaments provide shorts for currents:
1− λ
ρt = ρm . (23)
1+ λ
This is the lower limit of the effective transverse resistivity. For cases when the interfaces between the
filaments and matrix create obstacles for the current flow, e.g. because the formation of an oxide layer,
the values of transverse resistivity could be significantly higher.
Fig. 18: Magnetization loops evaluated from the distributions shown in Fig. 17. Full line: coupled wires; dashed
line: uncoupled wires.
In contrast to the magnetization currents generated by the flux pinning in a superconductor, the
coupling currents represent a ramp-rate-dependent phenomenon. The screening field created by the
coupling currents changes in time, with the characteristic time constant given by Eq. (22). This can be
measured e.g. as the delay of the magnetic field in the closed vicinity of the wire with respect to the
applied field, as illustrated in Fig. 20. Then the magnetization loops will be ramp-rate dependent and
one must always indicate the waveform of the magnet current used in experiment, otherwise the
interpretation of results is not possible.
Because the coupling currents are controlled by the normal resistance of the matrix, they exhibit
properties similar to normal eddy currents. In particular, the behaviour at any cyclic change can be
predicted on knowing the response at various frequencies. The volume density of the AC loss due to
491
F. G ÖMÖRY
coupling currents caused by the external magnetic field Bex = Ba sin (ωt ) is predicted by the
following formula [15]:
Qc Ba2 χ 0 πωτ
= , (24)
Vw µ0 1 + ωτ
where Vw is the volume of the whole wire (both superconductor and matrix) and χ0 is the magnetic
susceptibility for a completely screened wire [16]. The latter quantity could be determined
experimentally at low temperature and small Ba, or calculated from the shape of the wire. For the wires
with cross-sections that could be approximated by an ellipse with axes a and b, respectively, placed in
a magnetic field oriented in parallel to the minor semi-axis, b, its value approaches χ0 = 1 + a/b [17].
Accordingly, for the round wire, χ0 = 2. In weak magnetic fields with amplitudes well below the
penetration field, the coupling loss should be corrected by this factor, taking into account a total
expulsion of flux from the superconducting filaments; Eq. (24) is modified to
Qc Ba2 χ πωτ
= (1 − λ ) 0 . (25)
Vw B << B µ0 1 + ωτ
a p
In reality one often obtains a result that lies between the predictions of Eqs. (24) and (25), as shown in
Fig. 21. In this plot the AC loss is expressed in terms of the imaginary part of the complex magnetic
susceptibility [18],
Q µ0
χ '' = . (26)
Vw π Ba2
Such a representation is similar to the ‘loss function’ but allows better quantitative comparison.
Fig. 19: The polarity of the electrical field induced between two twisted wires alternates every half-pitch
Fig. 20: The magnetic field inside or in the vicinity of the wire, Bi, compared to the applied field, Bex, increasing
with constant ramp rate, is delayed in time. From this delay, the time constant of the coupling currents, τ, can be
estimated.
492
Fig. 21: Imaginary part of the complex magnetic susceptibility measured on a multifilamentary wire compared
with two predictions that differ in the extent of filament penetration by an applied magnetic field. The solid line
has been calculated from Eq. (24), assuming complete filament penetration; the dashed line is from Eq. (25),
derived for completely screened filaments.
4 Conclusions
The pinning of magnetic flux, which is necessary to secure an exceptionally high current transport
capacity in type II superconductors, is at the same time an obstacle when a variation in the magnetic
field occurs. This is because a dissipation in the cyclic regime (e.g. when transporting AC current)
appears. Also, a superconducting wire in a DC magnet must undergo a change in magnetic field at the
ramp necessary to reach the operating field.
The dynamics of magnetic flux penetration into a hard superconductor (i.e. a type II
superconductor able to pin the magnetic flux in its volume) is described by a complex process that
depends on the properties of the superconducting material as well as the architecture of the
superconducting wire and the orientation of magnetic field. Many basic features of this process can be
modelled on a macroscopic scale with the help of the critical state model, assuming that, in a hard
superconductor, the local value of the current density could be either zero or jc.
Infiltration of a magnetic field provokes magnetization currents, which have negative
consequences on the quality of the magnetic field generated by the superconducting magnet. These
currents also lead to a dissipation of electromagnetic energy. The most practical way of minimizing
these currents is through a reduction of the superconductor’s dimension transverse to the magnetic
field. This is why composite wires for pulsed magnets contain fine superconducting filaments, which
in turn should be twisted in order to reduce the coupling currents induced in the interfilament loops.
The basic principles erquired to understand these problems have been explained in this paper.
The current state of numerical methods allows detailed calculations that assume a more realistic
description of the superconductor properties than is possible with the original critical state model with
jc = constant. For example, a magnetic-field dependence of the critical current density and its
anisotropy, and/or non-uniformity of jc in the volume of the superconductor, can be included in order
to interpret the experimental data obtained from industrially produced materials [19]. There is
currently a great deal of activity directed towards a more exact prediction of flux dynamics in non-
traditional superconductors, and the topic is far from being completely managed. A recent review of
the achievements in this field is given in [20].
493
F. G ÖMÖRY
Appendix A
The quantity used here to define the magnetic fields is the induction, B, in units of tesla (T), where
1 T = 1 V·s·m−2. This is the quantity measured by magnetic field sensors and an exerting force on a
moving charged particle. In a material with a magnetic response, one finds loops of electrical currents

(see Fig. A1, left panel). The area of the loop, S , and the circulating current, I, determine its magnetic
moment
 
m = IS [Am2]. (A.1)
The magnetization of a sample with volume V is defined as the volume density of magnetic moments
 ∑m 
M= [A·m−1]. (A.2)
V
Quite often in the literature on superconductors an alternative definition is used that expresses the
magnetization in tesla. This is obtained by multiplying Eq. (A.2) by the permeability of vacuum,
µ0 = 4π·10−7 H·m−1. In all other aspects they are identical.
Fig. A1: Magnetic moment of a circular loop (left) and the magnetization of a macroscopic sample (right)
Acknowledgement
This work was supported in part by the VEGA grant agency under contract No. 1/0162/11, by the
European Commission EURATOM project FU07-CT-2007-00051 and co-funded by the Slovak
Research and Development Agency under the contract No. DO7RP-0003-12.
References
[1] M.N. Wilson, Superconducting Magnets (Clarendon, Oxford, 1983).

[2] W.J. Carr Jr, AC Loss and Macroscopic Theory of Superconductors, 2nd ed. (Taylor and
Francis, London, 2001).
[3] C.P. Bean, Phys. Rev. Lett. 8 (1962) 250–253.
[4] Y.B. Kim, C.F. Hempstead and A.R. Strnad, Phys. Rev. Lett. 9 (1962) 306–309.
[5] E.H. Brandt, Phys. Rev. B 54 (1996) 4246–4264.
[6] N. Amemiya, S. Murusawa, N. Banno and K. Miyamoto, Physica C 310 (1998) 16–29.
[7] W.T. Norris, J. Phys. D 3 (1970) 489–507.
[8] A.M. Campbell, Supercond. Sci. Technol. 20 (2007) 292–295.
[9] C. Gu and Z. Han, IEEE Trans. Appl. Supercond. 15 (2005) 2859–2862.
[10] F. Gömöry, M. Vojenčiak, E. Pardo, M. Solovyov and J. Šouc (2010), Supercond. Sci.
Technol. 23 (2010) 034012.
494
[11] M. Ashkin, J. Appl. Phys. 50 (1979) 7060–7066.

[12] F. Gömöry, R. Tebano, A. Sanchez, E. Pardo, C. Navau, I. Hušek, F. Strýček and P. Kováč,
Supercond. Sci. Technol. 15 (2002) 1311–1315.
[13] E. Pardo, D.X. Chen, A. Sanchez and C. Navau, Supercond. Sci. Technol. 17 (2004) 83–87.
[14] T. Yazawa, J.J. Rabbers, B. ten Haken, H.H.J. ten Kate and Y. Yamada, Physica C 310 (1998)
36–41.
[15] A.M. Campbell, Cryogenics 22 (1982) 3–16.
[16] K. Kwasnitza and S. Clerc, Physica C 233 (1994) 423–435.
[17] P. Fabbricatore, S. Farinon, S. Innocenti and F. Gömöry, Phys. Rev. B 61 (2000) 6413–6421.
[18] P. Fabbricatore, S. Farinon, S. Incardone, U. Gambardella, A. Saggese and G. Volpini, J. Appl.
Phys. 106 (2009) 083905.
[19] F. Gömöry, in High Temperature Superconductors (HTS) for Energy Applications, Ed.
Z. Melhem (Woodlhead Publishing Ltd., Oxford, 2012), pp. 216–256.
[20] F. Grilli, E. Pardo, A. Stenvall, D.N. Nguyen, W. Juan and F. Gömöry, IEEE Trans. Appl.
Supercond. 24 (2014) 8200433.
495
Vacuum Technology for Superconducting Devices
P. Chiggiato 1
Abstract
The basic notions of vacuum technology for superconducting applications
are presented, with an emphasis on mass and heat transport in free molecular
regimes. The working principles and practical details of turbomolecular
pumps and cryopumps are introduced. The specific case of the Large Hadron
Collider’s cryogenic vacuum system is briefly reviewed.
Keywords: vacuum technology, outgassing, cryopumping, LHC vacuum.
1 Introduction
Vacuum is necessary during the production of superconducting thin films for RF applications and for
the thermal insulation of cryostats. On the other hand, vacuum systems take an advantage from the
low temperatures necessary for superconducting devices. This chapter focuses on the principles and
the main definitions of vacuum technologies; some insights about gas and heat transfer in a free
molecular regime are given. Only turbomolecular pumping and cryopumping are described, since they
are the most relevant for superconducting applications. Pressure measurement is not included because,
in general, it is not considered a critical issue in such a domain. A comprehensive introduction to
vacuum technology may be found in the books listed in the references.
2 Basic notions of vacuum technology

The thermodynamic properties of a rarefied gas are described by the ideal gas equation of state:
PV = N moles RT (1)
where P, T and V are the gas pressure, temperature, and volume, respectively, and R is the ideal gas
constant (8.314 J K−1 mol−1 in SI units); Nmoles is the number of gas moles. From statistical physics,
Eq. (1) may be rewritten in terms of the total number of molecules N in the gas:
PV = NkBT (2)
where kB is the Boltzmann constant (1.38 × 10−23 J K−1 in SI units). Pressure units are pascals
(1 Pa = 1 N m−2) or millibars (1 mbar = 100 Pa). Quantity of gas may be expressed as the number of
molecules, number of moles or, by Eq. (1), pressure-volume units at a given temperature. The latter
units are used in most cases in vacuum technology. The quantities of gas expressed in pressure-
volume units are converted to the number of molecules by dividing them by 𝑘B 𝑇
(1 Pa m3 = 2.5 × 1020, 1 mbar l = 2.5 × 1019, both at 295 K).
In vacuum systems, pressures span several orders of magnitude (Table 1). Degrees of vacuum
are defined by upper and lower pressure boundaries. Different degrees of vacuum are characterized by
different pumping technologies, pressure measurement, materials, and surface treatments. In general,
vacuum systems for particle accelerators operate in the high and ultra-high vacuum.
1
paolo.chiggiato@cern.ch
P. C HIGGIATO
Table 1: Degrees of vacuum and their pressure boundaries
Pressure Pressure
boundaries boundaries
(mbar) (Pa)
Low vacuum (LV) 1000–1 105–102
Medium vacuum (MV) 1–10−3 102–10−1
High vacuum (HV) 10−3–10−9 10−1–10−7
Ultra-high vacuum (UHV) 10−9–10−12 10−7–10−10
Extreme vacuum (XHV) <10−12 <10−10
The mean speed of gas molecules ⟨𝑣⟩ is calculated by the Maxwell–Boltzmann distribution [1]:
8kBT 8 RT
=v = (3)
π⋅m π⋅M
where m is the mass of the molecule and M is the molar mass. The unit of both masses is kg in SI.
Assuming that the density of molecules all over the volume is uniform, the molecular impingent rate ϕ
onto a surface of unit area is calculated:
1 1 8kBT
=φ = n v n (4)
4 4 π⋅m
where n is the volume number density (n = N V−1).

The shortest conceivable time to form a single layer of adsorbed molecules, i.e. when all
molecules impinging on a surface adhere with it, is called the monolayer formation time tML.
Assuming typically 1019 adsorption sites per m2:
1019 mT
tML [ s ]= ≈ 93 ×106 . (5)
φ P
For nitrogen (m = 4.6 × 10−26 kg) at room temperature and pressures of the order of 10−4 Pa, tML is of
the order of 1 s.
In any physically limited vacuum system, molecules collide between each other and with the
walls of the vacuum envelope/container. When the gas density is so low that molecular collisions are
improbable, there is a drastic change in the transport phenomena. This important feature is described
by the Knudsen number Kn, namely the ratio between the average distance between two consecutive
points of molecular collision 𝜆̅ [1], i.e. the molecular mean free path, and a characteristic dimension
D of a vacuum system, for example the diameter in cylindrical beam pipes:
λ
Kn = . (6)
D
As reported in Table 2, the values of Kn delimit three different gas flows: free molecular, viscous, and
transitional.
498
VACUUM T ECHNOLOGY FOR S UPERCONDUCTING D EVICES
Table 2: Gas dynamic regimes defined by the Knudsen number
Kn range Regime Description
Kn >0.5 Free molecular flow Molecule–wall collisions dominate
Kn <0.01 Continuous (viscous) Gas dynamics dominated by intermolecular

flow collisions
0.5 < Kn < 0.01 Transitional flow Transition between molecular and viscous flow
In typical beam pipes (D ~10 cm), the transition to molecular regime is obtained for pressures lower
than about 10−1 Pa. Vacuum systems of accelerators operate in the free molecular regime, except for
short transfer lines and ion sources. The free molecular flow regime characterizes and determines the
pumping and pressure-reading mechanisms. Since there is no interaction between molecules, pumps
and instruments must act on single molecules.
In stationary conditions, the number densities (n1 and n2) and pressures (P1 and P2) in two
connected volumes at different temperatures (T1 and T2) are correlated by means of Eq. (7):
n1 T P T
= 2; 1 = 1 . (7)
n2 T1 P2 T2
In general, it can be written that:

P
= constant; n T = constant. (8)
T
Therefore, in equilibrium, in a vessel at lower temperature there is a higher density and a lower
pressure. With reference to Fig. 1, Eq. (7) can be obtained by equating the impingement rate (see
Eq. (4)) from volume 1 to volume 2 and vice versa. This phenomenon is usually referred to as thermal
transpiration.
T1, P1, n1 T2, P2, n2

A
Fig. 1: Schematic drawing of two volumes communicating through a thin, small wall slot
2.1 Gas transport in free molecular flow
2.1.1 Gas conductance

In the free molecular regime and isothermal conditions, the net gas flow Q between two points of a
vacuum system is proportional to the pressure difference (P1 − P2) at the same points:
Q = C (P1 − P2 ) (9)
where C is called the gas conductance of the vacuum system between the two points. In the free
molecular regime, the conductance does not depend on pressure. It depends only on the mean
molecular speed and the vacuum system geometry. If the gas flow units are expressed in terms of
pressure–volume (e.g. mbar l s−1 or Pa m3 s−1), the conductance is reported as volume per unit time
(i.e. l s−1 or m3 s−1) (litre is represented by the letter l).
499
P. C HIGGIATO
The conductance of a wall slot of area A between two vessels is given by Eq. (10). It can be
obtained by calculating the difference between the impingement rates for volumes 1 and 2 in
isothermal conditions, with reference to Fig. 1.
1 T
C= Av ∝A (10)
4 m
Therefore, the conductance of the wall slot is inversely proportional to the square root of the molecular
mass. For equal pressure drop, the gas flow of H2 is the highest. The conductances C′ of a unit-
surface-area wall slot for different gas molecules are listed in Table 3.
Table 3: Wall slot conductances C′ of 1 cm2 surface area, for common gas species, at room temperature
Gas H2 He CH4 H2O N2 Ar
𝑪′ 𝐚𝐭 𝟐𝟗𝟑 𝐊 (l s−1 cm−2) 44 31.1 15.5 14.7 11.75 9.85
For complex geometries, the transmission probability τ is introduced. The conductance of any duct
connecting two vessels at the same temperature is given by Eq. (11).
C = C ′A1τ 1→2 (11)
where C′A1 is the conductance of the duct aperture seen as a wall slot in the first vessel, and 𝜏1→2 is
the probability that a molecule once entered into the duct from the first vessel is transmitted to the
other vessel.
The transmission probabilities depend only on vacuum component geometry. Approximate
formulas are reported in many vacuum technology textbooks, for example in Refs. [2–5]. For long
tubes (D >> L, where D is diameter and L is length) the transmission probability and the resulting
conductance are:
4D
τ≈ (12)
3 L
D3
C ≈ 12.3 . (13)
L
In Eq. (13) D and L are expressed in cm, C in l s−1. Conductances of more complicated components
are calculated by test-particle Monte Carlo methods (TPMC). The reference TPMC software at CERN
is MolFlow+ [6].
Vacuum components are installed either in series, i.e. traversed by the same gas flow, or in
parallel, i.e. equal pressures at the extremities. In the former case, the inverse of the total conductance
CTOT is given by the sum of the inverse of the single conductances Ci:
1 N
1
=∑ . (14)
CTOT 1 Ci
In the latter case, the total conductance is the sum of the single conductances:
N
CTOT = ∑ Ci . (15)
1
500
2.1.2 Pumping speed

In vacuum technology, a pump is any ‘object’ that removes gas molecules from the gas phase. A
vacuum pump is characterized by its pumping speed S, which is defined as the ratio between the
pumped gas flow QP (pump throughput) and the pump inlet pressure P:
QP
S= . (16)
P
The pump throughput can be written as the gas flow through the cross-section of the pump inlet
(surface area AP) multiplied by the capture probability σ, i.e. the probability for a molecule that enters
the pump to be definitely removed and never more to reappear in the gas phase of the vacuum system
(see Eq. (17)). In the literature, σ is also called the Ho coefficient:
1
QP φ=
= APσ AP n v σ . (17)
4
Considering Eqs. (2) and (10), it turns out that:

P
QP = AP C ′ σ . (18)
kBT
From the definition of the pumping speed and converting the throughput into pressure–volume units:
S = AP C ′σ . (19)
Therefore, the pumping speed is equal to the conductance of the pump inlet cross-section multiplied
by the capture probability. The maximum theoretical pumping speed of any pump is obtained for
σ = 1 and it is equal to the conductance of the pump inlet cross-section. Table 4 reports some values
of the maximum pumping speed of lump pumps for typical diameters of the pump inlet. Because S
depends on C′, and so on the inverse of the square root of the molecular mass, the maximum
theoretical pumping speed is that for H2.
The pumping speed given by the suppliers of vacuum pumps is called the nominal pumping
speed; it refers to the pump inlet. The effective pumping speed Seff is that acting directly in the vacuum
vessel of interest. The effective pumping speed is lower than the nominal pumping speed owing to gas
flow restrictions interposed between the pump and the vessel.
Table 4: Maximum pumping speeds in l s−1 for different diameters of circular pump inlets
ID H2 N2 Ar
(mm) (l s−1) (l s−1) (l s−1)
36 448 120 100
63 1371 367 307
100 3456 924 773
150 7775 2079 1739
The effective pumping speed is calculated equating the net gas flow from the vessel and that removed
by the pump. Taking into account Eqs. (9) and (16), with reference to Fig. 2, one obtains:
Q = C1 ( P1 − P2 ) = SP2 = Seff P1 (20)
and so:
1 1 1
= + . (21)
Seff S C
501
P. C HIGGIATO
As a result, for 𝐶 ≪ 𝑆, one finds 𝑆eff ≈ 𝐶. In other words, the effective pumping speed does not
depend on the installed pump if the conductance of the interposed connection is very low. This
conclusion is of primary importance in the design of efficient vacuum systems.
Fig. 2: Schematic drawing of a gas flow restriction of conductance C interposed between a pump of pumping
speed S and a vacuum vessel.
2.1.3 Outgassing
Several gas sources contribute to the total gas load in a vacuum system. In addition to intentional gas
injections and air leaks, outgassing of materials plays a crucial role in the global gas balance.
Materials release gas molecules that are both adsorbed onto their surfaces and dissolved in their bulk
[7]. The outgassing is distributed uniformly all over the vacuum-exposed surfaces; its rate is in general
reported for unit surface area (q).
The outgassing rate is reduced by dedicated surface treatments. For metals, in general, several
surface treatments are available, each aiming at removing a specific set of contaminants. Gross
contamination and the sorption layer (hydrocarbons, Cl compounds, silicone greases, etc.) are
removed by solvent and detergent cleaning. Thick oxide layers (1–10 nm thick) are removed by
chemical pickling. The damaged metal skin – full of dislocations, voids, and impurities – is removed
by acid etching or electropolishing.
For all materials, after thorough surface treatment, water vapour dominates the outgassing
process. Its outgassing rate is inversely proportional to the pumping time (t) for smooth metals. The
following empirical relationship is in general applied:
3 ×10−9  mbar l 
qH2O ≈ . (22)
t [ h ]  s cm 2 
For organic materials, in particular polymers, in the first phase of pumping, the water vapour
outgassing rate is inversely proportional to the square root of the pumping time 𝑞H2 O ∝ 1⁄√𝑡, the
absolute value being strongly dependent on the nature and history of the material. For example, the
Multi-Layer Insulation (MLI) used for the LHC insulation vacuum is made of polymer foils. There are
several such layers amounting to 250 m2 of MLI per metre of the LHC. When a typical insulation
vacuum sector is exposed to ambient air for several weeks, it takes 200 hours of pumping with an
effective pumping speed of 100 l s−1 to attain a pressure of about 5 × 10−3 mbar at room temperature;
from such data an outgassing rate of 10−9 mbar l s−1 cm−2 per layer of MLI is obtained. Such a pressure
would have been achieved, after the same pumping time, with a stainless steel sheet as wide as two
soccer fields per metre of LHC sector.
The water vapour outgassing is reduced by in situ heating in vacuum (bake-out). Such a
thermal treatment is very effective for metals if it is carried out for at least 12 h at temperatures higher
than 120°C. After bake-out, H2 becomes the main outgassed species for metals; the outgassing rate can
be regarded as constant at room temperature. It can be reduced by heat treatment at higher
temperatures either in situ, or ex situ in a vacuum furnace [7].
502
As shown in Table 5, the value of outgassing rate spans several orders of magnitude. The right
choice of materials and treatments is essential in the design of vacuum systems to limit the gas charge.
The outgassing features should be taken into account at the very beginning of the design process when
materials are selected.
Table 5: Values of outgassing rates for selected materials used in vacuum technology. The reported heat
treatments are carried out both in situ and in vacuum.
Material q Main gas
(mbar l s−1 cm−2) species
Neoprene, not baked, after 10 h Order of 10−5 H2O

of pumping [8]
Viton, not baked, after 10 h of Order of 10−7 H2O
pumping [8]
Austenitic stainless steel, not 3 × 10−10 H2O

baked, after 10 h of pumping
Austenitic stainless steel, baked 3 × 10−12 H2

at 150°C for 24 h
OFS copper, baked at 200°C for Order of 10−14 H2

24 h
In particle accelerators, gas molecules can also be desorbed by electron, photon, and ion collisions
induced by the beam. In modern high-energy accelerators, such a gas source is generally preponderant.
Significant reduction of desorption yields is obtained by gradually accumulating high doses of particle
collisions. Another very efficient method is coating the inner surface of the beam pipe with non-
evaporable getter thin films that, after in situ activation, achieve a very clean and pumping surface.
2.1.4 Pressure profiles

The calculation of the pressure profile along vacuum systems is an essential task of vacuum experts
and should be tackled at the design phase. In general, the contributions to the total pressure of
localized and distributed gas sources are considered separately and finally added. This is possible
because in most cases the equations that describe pressure profiles are linear. This may not be true if
the pumping speed is pressure-dependent.
In the case of a localized gas source, the pressure in a vacuum vessel is obtained by taking into
account Eq. (16) and the intrinsic pressure limitation P0 of the installed pumping system:
Q
P = + P0 . (23)
S
The pumping speed S is given either by the supplier or preliminarily measured; P0 is the pressure
attained in the system without any gas load. When a restriction of conductance C is interposed
between the pump and the vessel, the effective pumping speed Seff is considered and Eq. (23) becomes:
Q Q (C + S )
P= + P0 = + P0 . (24)
Seff CS
When many vessels are interconnected, the flow balance is written in each vessel (node analysis). This
analysis leads to a system of linear equations, from which the pressure values in each vessel are
calculated. As an example, with reference to Fig. 3, in the first vessel, the injected gas flow (Q) is
503
P. C HIGGIATO
either pumped (P1S1) or transmitted to the second vessel (C1(P1 − P2)). This latter flow is pumped in
the second vessel or transmitted to the third vessel, and so on. Thus:
1 1 + C1 ( P1 − P2 ) ,
Q =PS
C1 ( P1 − P2=
) C2 ( P2 − P3 ) + P2 S2 ,
(25)
C2 ( P2 − P3 =
) C3 ( P3 − P4 ) + P3 S3 ,
C3 ( P3 − P4 ) =
P4 S 4 .
When a second localized gas flow is settled, the node analysis is repeated. The contributions of each
localized gas flow to the pressure values are then added. If the cross-section of the interconnection
ducts is constant, it can be shown that the pressure varies linearly between two connected vessels.
Fig. 3: Schematic drawing of four interconnected vacuum vessels. In each vessel, the gas flow balance is written
(node analysis).
In case of distributed outgassing, a local gas balance is written. The resulting equations are typical of
transport phenomena:
∂P( x, t )
q ( x) = −C ( x ) ,
∂x
(26)
∂ 2 P ( x, t ) ∂P( x, t )
C ( x) = .
∂x 2 ∂t
where 𝑞(𝑥) is the gas flow at abscissa 𝑥. 𝐶̅ (𝑥) is the conductance of the unit length of a vacuum vessel
having a constant cross-section as that at abscissa 𝑥; 𝐶̅ (𝑥) is constant and equal to 𝐶𝐿 if the vessel has
a constant cross-section, total conductance C, and length L. In a steady state, the second space
derivative of the pressure is zero; consequently, the pressure profiles are parabolic. A typical example
is a uniform beam pipe, with distributed outgassing, that is pumped at both extremities by lump
pumps. The pressure reaches a maximum 𝑃max in the middle of the beam pipe and minimum 𝑃min at
the pumping positions. The difference Δ𝑃 between the two extremes is given by:
Q
∆P= Pmax − Pmin= (27)
8C
where Q is the total outgassing rate of the beam pipe. Nowadays, pressure profiles are also easily
obtained by TPMC simulation.
2.2 Heat transport by gas

Gas molecules participate in the global heat transfer process transferring energy from the warmer
walls to the colder ones. The contribution of gas convection is erased at relatively high pressures; for
this reason it is not considered here.
The energy that a gas molecule can convey depends on how the energy is stored in the
molecule, i.e. on the molecular movements that are active at a certain temperature. Translational
movements in the three dimensions are common to all molecules. Molecules formed with more than
504
one atom can also rotate around the centre of mass. Vibrational modes are in general inactive at
cryogenic temperature. Kinetic theory shows that the thermal energy is equally shared among each
active molecular movement: each active degree of freedom carries the same average energy, which is
equal to ½kB T.
The specific heat capacity at constant volume (cv) of a single molecule with f active degrees of
freedom is:
kB
cv = f . (28)
2m
The molar specific heat capacity at constant volume (cMv) is obtained from Eq. (28) multiplied by the
mass of one mole, i.e. NAm:
N A kB R
cMv f=
= f . (29)
2 2
For an ideal gas, specific heat capacities at constant pressure (cp and cMp) can be calculated [9]:
k
cp =
cv + B , cMp =
cMv + R . (30)
m
The specific heat capacity ratio is known as the isentropic expansion factor γ :
cp cMp 2
γ= = = 1+ . (31)
cv cMv f
2.2.1 Gas thermal conduction in viscous regime

In a viscous regime, the conduction heat flow is given by the Fourier equation:
qth = −kth ∇T . (32)
The gas thermal conductivity kth is given by Eq. (33):

1
k=
th ( 9γ − 5) ⋅ η ⋅ cv (33)
4
where η is the gas viscosity:

5 1 mkBT
η= (34)
16 δ 2 π
and δ is the molecular diameter. Table 6 reports selected values of δ, m, and f for simple molecules.
Table 6: Mass, diameter, and active degree of freedom at room temperature (translational and rotational) for
typical gas molecules found in HV and UHV.
m δ f
(kg) (pm)
N2 4.6 × 10−26 370 5
H2 3.3 × 10−27 293 5
H2O 3.0 × 10−26 275 6
CO 4.6 × 10−26 376 5
CH4 2.7 × 10−26 382 6
505
P. C HIGGIATO
The heat flow between two parallel plates, at temperatures T1 and T2 and distance L may be
approximately obtained by Eq. (35):
T −T
qth ≈ k th 1 2 (35)
L
where the gas conductivity is calculated for the temperature of the warmer plate. For example, the heat
flow between two plates at 300 K and 80 K at a distance of 10 cm is about 55 Wm−2 when the residual
gas is nitrogen; the thermal conductivity is 25.3 mW K−1m−1.
The conduction heat flow in viscous regimes does not depend on the gas pressure: lower
pressure means less energy carriers but a longer mean free path; qth is inversely proportional to the
distance between the walls of the vessel.
2.2.2 Gas thermal conduction in molecular regime

In the molecular regime, gas molecules interact only with the walls of the vessel. The molecules carry
thermal energy directly from one wall to another. The fraction of energy transferred after collision
depends on the temperature and on the nature of the molecules and surfaces. The transfer process is
quantitatively described in terms of a coefficient of thermal accommodation α, which is a measure of
the efficiency of energy exchange between a gas flow and a solid surface:
E2' − E1'
α1 = (for surface 1) ,
E2' − E1
E1' − E2'
α2 = (for surface 2) . (36)
E1' − E2
In Eq. (36), 𝐸2′ and 𝐸1′ are the thermal energies leaving, each second, the unit area of surface 2 and
surface 1, respectively; 𝐸1 and 𝐸2 are the values that 𝐸1′ and 𝐸2′ would have if thermal equilibrium
were achieved in all collisions at surfaces 1 and 2, respectively. In other words, the coefficient of
thermal accommodation is the average fraction of the maximum possible energy exchange at the
walls.
The equation for the heat flow in molecular regime was given by Kennard [2]; the subscript 2
refers to the warmer surface:
γ + 1 R T2 − T1
qth = α P. (37)
γ − 1 8Mπ T
Therefore, the heat flow is proportional to the pressure and does not depend on the distance between
the two walls.
Here, α is the overall accommodation coefficient:
α1α 2
α= . (38)
A
α 2 + α1 (1 − α 2 ) 1
A2
Eq. (37) is valid for concentric spheres, coaxial cylinders, or parallel planes. In the case of two parallel
slabs, of the same area, made of the same material, Eq. (38) simplifies to:
α1
α= . (39)
2 + α1
506
Thermal accommodation coefficients for most materials with technical surface finishing are about 0.9,
0.8, and 0.4 for argon, nitrogen, and helium, respectively (Table 7) [2, 10, 11, 12]. If an adsorbed or
condensed layer of gas is formed, α tends to 1. The lower values of α for the lightest molecules
counteract the mass dependence of Eq. (37) so that the same magnitude of heat flow is expected for
He and H2 as for N2 on technical surfaces. This is not the case in a viscous regime where the thermal
conductivity of H2 is about one order of magnitude higher than that of nitrogen.
Table 7: Selected values of the thermal accommodation coefficient for platinum and stainless steel AISI-304 at
room temperature [2, 10, 11, 12].
α
Pt AISI 304 Pt
Gas technical surface technical atomically
at room surface clean surface
temperature
N2 0.77 0.8
H2 0.29 0.15
Ar 0.86 0.87 0.55
CO 0.78
He 0.38 0.36 0.03
The temperature T in the denominator of Eq. (37) is an effective temperature of the non-
equilibrium gas [13]. If the pressure P is measured between the two walls, T is given by the following
equation:
1  A1  1  A2  1
=   +   (40)
T  A1 + A2  T1'  A1 + A2  T2'
where 𝑇1′ and 𝑇2′ are the effective temperatures of the gas molecules after collision with surface 1 and
2, respectively. They can be expressed in terms of the real temperatures of the two walls, i.e. T1 and T2.
 A1 
α1T1 (1 − α 2 ) + α 2  + α 2 (1 − α1 ) T2
 A2 
T1' = ,
A1
α 2 + α1 (1 − α 2 )
A2
A1
α1T1 (1 − α 2 ) + α 2T2
' A2
T2 = . (41)
A1
α 2 + α1 (1 − α 2 )
A2
Alternatively, if the pressure P is measured by an external gauge, T is the temperature at the gauge
position, generally room temperature. In fact, because the ratio 𝑃⁄√𝑇 is constant throughout the
vacuum system, the values of P and T must be chosen at the same location.
As an example, two plates at 300 K and 80 K at a distance of 10 cm are considered. As in
Section 2.2.1, the residual gas is N2. The heat flow in the molecular regime is given by the
approximated equation:
qth ≈ 265 ⋅ P (W m−2). (42)
507
P. C HIGGIATO
The heat radiation contribution calculated by the Stefan-Boltzmann equation gives a value of
8.9 W m−2; to be realistic, values of thermal emissivity equal to 0.4 (stainless steel) and 0.02 (clean
copper) are chosen for the surfaces at 300 K and 80 K, respectively. As a result, the gas conduction
contributes to the same extent as radiation when the pressure is in the 10−2 Pa range. To reduce the gas
heat transport to less than 10 times the radiation value, pressure in the 10−3 Pa range are needed.
Figure 9 shows the different heat transport contributions, including in the viscous regime.
Fig. 9: Heat power transmission between two parallel plates at a mutual distance of 10 cm. The temperatures of
the plates are 300 K and 80 K. The plates are made of stainless steel (thermal emissivity 0.4) and clean copper
(thermal emissivity 0.02), for the warmer and colder surfaces, respectively. The residual gas is nitrogen.
3 Gas pumping in molecular regimes

Vacuum pumps in the molecular regime are classified into two families: momentum transfer pumps
[2] and capture pumps [14]. Both act on each molecule singularly since no momentum and energy
transfer is possible between molecules in this pressure range. In the first family, molecules receive a
momentum component pointing towards the pump outlet (foreline) where the gas is compressed and
evacuated by pumps working in the viscous regime (e.g. rotary vane, diaphragm, and scroll pumps). In
this set of pumps, the turbomolecular ones are the most used for superconducting applications. The
second family removes gas molecules by fixing them on a surface exposed to the vacuum. The
molecular adsorption is based on chemical bonding (getter pumps) or physisorption at cryogenic
temperatures. The latter may be a by-product of cooling at temperatures lower than the
superconducting transition; it is the dominant pumping principle in beampipes embedded in
superconducting magnets. Only turbomolecular pumps and cryopumping are considered in this
chapter.
3.1 Turbomolecular pumps

In turbomolecular pumps (TMP), gas molecules are steered by rapidly rotating blades. The rotating
blades are tilted with respect to the rotational axis. As a result, oblique channels are formed between
successive blades. The rotor is inserted between two static surfaces defining two interspaces (see
508
Fig. 10). When a molecule comes from interspace-1, the angular distribution of the molecular velocity
seen from the blades is preferentially oriented towards the blades’ channels. Conversely, for those
coming from interspace-2, the velocity vectors point preferentially towards the blades’ wall, therefore
increasing backscattering. A gas flow is consequently generated from interspace-1 to interspace-2.
This mechanism works only if the angular distribution of the molecular speed as seen from the blade is
significantly deformed, i.e. only if the peripheral speed of the blades is at least of the order of the
mean molecular speeds (hundreds of metres per second).
Fig. 10: A single rotor–stator stage in a turbomolecular pump, showing the molecular speed distributions as seen
from the moving blades.
In a real TMP, the gas is compressed by several series of rotating blades. Each series of rotating blades
is followed by a series of static blades. Molecules transmitted through the rotating blades’ channels hit
the static blades; as a result, the angular distribution of velocity is randomized and the molecules are
ready for the next compression stage. The momentum transfer is effective only if the molecules do not
experience intermolecular collisions after hitting the blades; this is equivalent to saying that the mean
free path has to be larger than the blade distance. As a result, this type of pump works at full pumping
speed only in molecular regimes (P < 10−3 mbar).
The most important characteristics of a molecular pump are the pumping speed S and the
maximum compression pressure ratio that can be achieved between the pump inlet PIN and outlet
POUT:
P 
K 0 =  OUT  . (43)
 PIN 
With reference to Fig. 10, it can be shown [2] that:

S ∝ b ⋅ u ⋅ sin α cos α , (44)
 u   
K 0 ∝ exp  ∝ exp u m  . (45)
 v ⋅ b ⋅ sin α   b ⋅ sin α 
   
To maximise the pumping speed, the distance between the blades is larger for the first series of blades
and their inclination is 45°. The maximum compression ratio is obtained by modifying the geometry
of the following series of blades; this is achieved by reducing both the distance between the blades and
their inclination (see Fig. 11). In addition, the tighter blades compensate for the increased gas density
and thus for the resulting lower molecular mean free path.
509
P. C HIGGIATO
Fig. 11: Cut-out view of a turbomolecular pump. Courtesy of Wikipedia

http://en.wikipedia.org/wiki/Turbomolecular_pump).
The pumping speeds of commercial turbomolecular pumps (Fig. 12) are constant in the free molecular
regime (P < 10−3 mbar) and range between 10–3000 l s−1 depending on pump inlet diameter and
mechanical design. As indicated by Eq. (44), pumping speed should not be affected by the mass of the
gas molecules. Indeed, in real pumps the gas dependence of the pumping speed is limited. In terms of
pumping speed, turbomolecular pumps are not selective.
Fig. 12: Pumping speeds of a turbomolecular pump equipped with a DN63CF flange. For a TMP, the pumping
speeds are constant in the free molecular regime and the nature of the pumped gas has a limited effect, if
compared with other type of pumps. (Courtesy of Pfeiffer Vacuum.)
510
Fig. 13: Maximum compression ratio of two different turbomolecular pumps for N2 and H2. (Courtesy of
Pfeiffer Vacuum.)
As expected, the maximum compression ratio is the lowest for H2; in classical designs it is about 103
(see Fig. 13). Currently, values up to 106 are achieved by integrating a molecular drag pump (Gaede or
Holweck drag stage) to the rear of the blade sets [2]. The TMP ultimate pressure is of the order of
10−10–10−11 mbar for baked and all-metal vacuum systems. Lower pressures may be obtained if
another type of pump removes the TMP H2 back-streaming.
At the pump outlet, the compressed gas is evacuated by mechanical pumps (often called
backing pumps) operating in the viscous regime. The turbomolecular pump and its backing pump are
in general assembled in a single unit that includes power supplies, controls, and instrumentation.
The TMP can evacuate high gas flow at relatively high pressure without selectivity and memory
effects. For example, a DN100 TMP can withstand a continuous gas flow up to 2 × 10−1 mbar l s−1
without damage.
The use of dry pumps for TMP backing has surmounted the risk of oil back-streaming from
rotary vane pumps. The complete removal of all lubricated mechanical bearings has been obtained by
magnetic rotor suspension.
The main drawback of a TMP is related to possible mechanical failures leading to definitive
damage to the high-speed rotor. In addition, in case of unwanted rotor deceleration caused by a power
cut or the rotor seizing, the vacuum system has to be protected by safety valves and dedicated pressure
sensors against air back-stream. TMPs have limited application in radioactive environments due to
possible damage to their electronics and power supplies. Some radiation-resistant TMPs are now
available; for a significant cost, the whole of the electronics are moved beyond the radiation shielding
by means of long cables. The application of TMP is limited in areas close to magnets; the residual
magnetic field has to be lower than 5 mT.
3.2.3 Cryopumping
Cryopumps rely on three different pumping mechanisms:
• Cryocondensation. This mechanism is based on the mutual attraction of similar molecules at low
temperature (Fig. 14). The key property is the saturated vapour pressure Pv, i.e. the pressure of
the gas phase in equilibrium with the condensate at a given temperature [15]. The lowest
pressure attainable by cryocondensation pumps is limited by the saturated vapour pressure.
Among all gas species, only Ne, H2 and He have Pv higher than 10−11 Torr at 20 K. The Pv of H2
511
P. C HIGGIATO
at the liquid He boiling temperature is in the 10−7 Torr range, and is 10−12 Torr at 1.9 K. The
quantity of gas that may be cryocondensed is very large and limited only by the thermal
conductivity of the condensate.
• Cryosorption. This is based on the attraction between gas molecules and substrates (Fig. 14).
The interaction forces with the substrate are much stronger than those between similar
molecules. As a result, providing the adsorbed quantity is lower than one monolayer, the sojourn
time is much longer, and gas molecules are pumped at pressures much lower than the saturated
vapour pressure. A significant quantity of gas may be pumped below one monolayer if porous
materials are used; for example, in one gram of standard charcoal for cryogenic application,
about 1000 m2 of surface are available for adsorption. The important consequence is that
significant quantities of H2 and He may be pumped at 20 K and 4.3 K, respectively. In general,
submonolayer quantities of all gas species are effectively cryosorbed at their own boiling
temperature.
• Cryotrapping. In this mechanism, the molecules of a low-boiling-temperature gas are trapped in
the condensation layer of another gas. This is possible because the interaction energy between
dissimilar molecules may be much higher than that between similar molecules. The trapped gas
has a saturated vapour pressure several orders of magnitude lower than in its pure condensate.
Typical examples are Ar trapped in CO2 at 77 K and H2 in N2 at 20 K.
(a) (b)
Fig. 14: Schematic drawing depicting (a) cryocondensation, where the leading mechanism is intermolecular
interaction; and (b) submonolayer cryosorption, where the mechanism is molecule–substrate interaction.
Modern cryopumps exploit the first two mechanisms. Cryocondensation takes place on a cold surface,
in general at 80 K for water vapour and at 10–20 K for the other gas species. The cryosorption of He,
H2, and Ne is localized on a hidden surface coated with a porous material. This part of the pump is
kept out of the reach of the other type of molecules, i.e. they have a probability close to 1 of being
intercepted and adsorbed by another surface before reaching the cryosorber (see Fig. 15). The
refrigeration is obtained by He gas cooled by a Gifford–McMahon cryocooler.
(a) (b)
Fig. 15: Drawings of: (a) a generic cryopump; (b) a closer view to the cryosorption surface where the porous
material is fixed.
512
Cryopumps having pumping speeds in the range 800–60,000 l s−1 are commercially available
(Table 8). For condensable gas molecules, the capture probability is close to 1 (e.g. for water vapour).
The maximum gas capacity (also called maximum gas intake) for the condensable gas is limited only
by the thermal conductivity of the condensate. To avoid a thick condensate layer and excessive
thermal load, cryopumps should be started in the molecular regime (P < 10−3 mbar). The quantity and
properties of the porous material determine the maximum gas intake of a cryosorbed gas. In general, it
is orders of magnitude lower than that for a condensable gas (see Table 8).
Table 8: Pumping speeds and maximum gas capacities of a commercial cryopump (Oerlikon-Leybold 800 BL
UHV); the pump inlet diameter is 160 mm.
H2O N2 Ar H2 He
S (l s−1) 2600 800 640 1000 300
Capacity (Torr l) 225,000 225,000 3225 375
Courtesy of Oerlikon-Leybold.
Cryopumps require periodic regeneration to evacuate the gas adsorbed or condensed; in this way, the
initial pumping speed is fully recovered. To do so, the cryopumps are warmed up to room temperature
and the released gas is removed by mechanical pumps (mobile TMP in particle accelerators). During
regeneration, the pump is separated from the rest of the system by a valve.
Excessive gas adsorption on the cryosorber leads to performance deterioration. A partial and
much faster regeneration (1 h against more than 10 h) may be carried out at temperatures lower than
140 K in such a way as to remove the sorbed gas without releasing water vapour from the pump stage
at higher temperature.
3.2.4 Cryocondensation and cryosorption in the LHC cryogenic vacuum system

In the LHC arcs, pressures of 10−8 mbar (for hydrogen, measured at room temperature) are needed to
fulfil the beam-gas-scattering lifetime requirement. The cold bore, in direct contact with the cold mass
of the magnets, is cooled to 1.9 K. At such a temperature, vapour pressures lower than 10−12 mbar are
reached for all gases, except He, for which up to 1015 molecules cm−2 can be cryosorbed before
reaching the pressure limit. However, direct interaction between the LHC’s proton beams and the cold
bore would provoke cryogenic, vacuum, and beam stability issues [16–19].
The adopted solution is the insertion of a tubular shield, called a beam screen (see Fig. 16),
between the beam position and the cold bore. The beam screen is cooled by helium gas at a
temperature gradually increasing from 5 K to 20 K. The beam screen intercepts the heat flow due to
image currents, beam losses, and photon and electron collisions before it reaches the cold mass of the
magnets. The drawback of such a solution is the inevitable reduction of the beam pipe aperture.
The beam screen is provided with holes, called pumping slots, to ensure cryocondensation of
hydrogen and cryosorption of He on the cold bore at a saturated vapour pressure compatible with the
lifetime requirement. The saturated vapour pressures of all other gas species are below the pressure
limit between 5–20 K; therefore, they can be efficiently cryocondensated on the beam screen.
However, beam-induced photon and electron bombardment on the beam screen can desorb the
cryopumped gas, which eventually is cryocondensated on the cold bore after migration through the
pumping holes. Owing to the shape and shielding of the pumping slots, the gas pumped on the cold
bore is no longer affected by the beam.
513
P. C HIGGIATO
(a) (b)
Fig. 16: Pictures of the LHC’s beam screen. (a) The beam screen is shown inserted in a tube having the same
diameter of the cold bore. The internal surface is covered by co-laminated copper to reduce the losses generated
by image currents. (b) The beam screen outside the cold bore is shown. The pumping slot shield aims at reducing
the transmission to the cold bore of electrons and radiation generated by the beam. The two cooling capillaries
are laser spot-welded on the beam screen.
In the LHC’s long straight sections, the superconducting magnets are cooled at 4.5 K. At this
temperature, the saturated vapour pressure of H2 does not fulfil the LHC’s pressure requirements.
Consequently, H2 has to be pumped by cryosorption onto porous materials as already implemented in
cryopumps (see Fig. 15). For such purpose, woven carbon fibre plates developed at the Budker
Institute of Nuclear Physics (BINP) are fixed on the external surface of the beam screens [20, 21].
Such a cryosorber may pump more than 1017 H2 molecules cm−2 before reaching equilibrium pressures
at 10−8 mbar, up to 20 K (see Fig. 17). The regeneration, which is possible at temperatures close to
80 K, is done during the magnet warm-ups.
(a) (b)
Fig. 17: Sample of (a) cryosorber made of (b) woven carbon fibres. This material is used to adsorb H2 in the
beam pipes of magnets that are cooled at 4.5 K instead of 1.9 K. It is inserted between the pumping slot shields
and the beam screen (courtesy of V. Baglin; G. Jesse of CERN MME group made the pictures).
References
[1] Sir James Jeans, An Introduction to the Kinetic Theory of Gases (Cambridge University Press,
Cambridge, 1967).
[2] K. Jousten, Ed. Handbook of Vacuum Technology (Wiley-VCH, Weinheim, 2008).
514
[3] J.M. Lafferty, Ed. Foundations of Vacuum Science and Technology (Wiley-Interscience, New
York, 1998).
[4] A. Berman, Vacuum Engineering Calculations, Formulas, and Solved Exercises (Academic
Press, San Diego, 1992).
[5] D.J. Santeler, J. Vac. Sci. Technol. A 4 (1986) 348.
[6] R. Kersevan and J.-L. Pons, J. Vac. Sci. Technol. A 27 (2009) 1017. Available at http://test-
molflow.web.cern.ch/.
[7] P. Chiggiato, CERN-ATS-Note-2012-048 TECH (in French). Vacuum in particle accelerators,
CAS, 2006; available at http://cas.web.cern.ch/cas/Spain-2006/Spain-lectures.htm.
[8] R.N. Peacock, J. Vac. Sci. Technol. 17 (1980) 330.
[9] J.O. Hirschfelder, C. Francis Curtiss and R.B. Bird, Molecular Theory of Gases and Liquids
(Wiley, New York, 1954).
[10] D.J. Rader et al., SANDIA Report SAND 2005-6084.
[11] W.M. Trott et al., Rev. Sci. Instrum. 82 (2011) 035120.
[12] Y. Demirel and S.C. Saxena, Int. J. Energy Res. 20 (1996) 327.
[13] R.J. Corruccini, Vacuum 7 (1959) 19.
[14] K.M. Welch, Capture Pumping Technology (North-Holland Elsevier, Amsterdam, 2001),
p. 106.
[15] R.A. Haefer, Cryopumping Theory and Practice (Oxford Science Publications, Oxford, 1989).
[16] O. Gröbner, Vacuum 60 (2001) 25.
[17] O. Gröbner, Proc. CAS Vacuum Technology, Denmark 1999, p. 291; available at
http://cds.cern.ch/record/455985/files/open-2000-288.pdf
[18] V. Baglin et al., CERN-ATS-2013-006 ; available at
https://cds.cern.ch/record/1507613/files/CERN-ATS-2013-006.pdf.
[19] V. Baglin, Proceedings of EPAC 2002, Paris, France; available at
http://accelconf.web.cern.ch/AccelConf/e02/PAPERS/WEPDO010.pdf.
[20] V. Anashin et al., Vacuum 75 (2004) 293.
[21] V. Anashin et al., Proc. 2005 Particle Accelerator Conference, Knoxville, Tennessee, p. 791;
available at http://accelconf.web.cern.ch/AccelConf/p05/PAPERS/TPAP002.PDF
515
Manufacturing and Testing of Accelerator Superconducting Magnets
L. Rossi 1
Abstract
Manufacturing of superconducting magnet for accelerators is a quite
complex process that is not yet fully industrialized. In this paper, after a
short history of the evolution of the magnet design and construction, we
review the main characteristics of the accelerator magnets having an impact
on the construction technology. We put in evidence how the design and
component quality impact on construction and why the final product calls for
a total-quality approach. LHC experience is widely discussed and main
lessons are spelled out. Then the new Nb3Sn technology, under development
for the next generation magnet construction, is outlined. Finally, we briefly
review the testing procedure of accelerator superconducting magnets,
underlining the close connection with the design validation and with the
manufacturing process.
Keywords: superconducting magnets, accelerators, LHC, applied

superconductivity, accelerator industrialization, magnet construction.
1 Introduction
The manufacture of superconducting (SC) magnets for an accelerator is mostly confined to large
laboratories, where magnets are conceived and designed and where research and development (R&D)
on small models and prototypes are carried out. It is only when it comes to series production for large
projects that industry becomes involved. The lack of continuity between large projects, the need for
special tools and the strong interlink between manufacture and design with a high degree of specific
knowledge make it very difficult for industry to be able to maintain a team with the necessary
expertise for accelerator magnet manufacturing. Actually, we can say that the number of companies
capable of offering a real service to the laboratories in this domain has decreased with respect to 30
years ago, there probably being only three or four in the world at present.
2 A short history of the manufacturing of SC magnets

As soon as NbZr and Nb3Sn became available, see Ref. [1], researchers and engineers tried to make
magnets for proton accelerators (for electrons the required field is typically 0.1–1 T so there is no need
for superconductivity). Meanwhile, the newly discovered Nb–Ti became rapidly the option of
preference; the use of SC magnets for synchrotrons, with fast ramping fields (0.1–0.5 T/s) was
pursued, see for example the GESSIE collaboration [2]. However, all of these efforts have been
carried out in research institutions. Unlike other types of magnets and applications, such as solenoids
for research, power application, nuclear magnetic resonance (NMR) and magnetic resonance imaging
(MRI), etc., accelerator magnet technology remained confined to research laboratories.
At the end of the 1970s CERN was able to boost the luminosity in the Intersecting Storage
Rings (ISR), featuring 25 GeV proton energy per beam, by means of large-aperture SC quadrupoles,
1
lucio.rossi@cern.ch
L. ROSSI
of about 5 T peak field, placed in the low-β insertions [3]. In 1980 the eight, 1 m long, ISR
quadrupoles were the first SC magnets in routine operation in an accelerator. Their cold masses were
the first, small, SC magnet series industrially manufactured for accelerators.
Despite the success of the ISR quads, with the decision to develop the resistive Super Proton
Synchrotron (SPS) and the Large Electron–Positron Collider (LEP) as the main accelerators for CERN
for the 1980s and 1990s, the spotlight on the development of accelerator magnets passed to the United
States, where two large new projects were ‘racing’ to be approved. At Brookhaven National
Laboratory (BNL), Upton, NY, a design team was working on a 200+200 GeV proton–proton collider
of 3.8 km based on dipole magnets operating at a 4 T field, later upgraded to 5 T. The R&D for this
project was very important and many prototype magnets were actually built [4]. However, for various
reasons the BNL project, called Isabelle and later the Colliding Beam Accelerator (CBA), was
terminated in its infancy in 1983. Meanwhile, the Fermi National Accelerator Laboratory (known as
Fermilab, Batavia, IL) in 1972 successfully commissioned its brand new 200 GeV synchrotron (based
on resistive magnets, in a 6.9 km ring), which was later pushed to 500 GeV. The director,
R. R. Wilson, launched a vigorous R&D program aimed at designing and manufacturing SC dipoles of
4–5 T in order to double the energy of the synchrotron [5]. His dream was realized in 1983, when the
largest SC magnet system—composed of 774 dipoles rated for 4.2 T and 240 quadrupoles—was able
to push the proton beam to a record energy of 517 GeV, to later reach almost 1000 GeV or 1 TeV,
hence the name of Tevatron. The machine was later transformed into a collider, since the magnet bore
(75 mm) was large enough, hosting two counter-circulating beams of protons and antiprotons. This
gave access to ∼2 TeV centre-of-mass energy: the Tevatron SC magnets kept Fermilab the leading
high-energy physics laboratory until 2010, when the Large Hadron Collider (LHC) broke this record.
The Tevatron was the first very-large-scale demonstration of the possibility of
superconductivity, with its 4.2 K cryogenics as a practical engineering system. From the design point
of view the Tevatron SC magnets feature collared coils (CC) enclosed in the cryostat while the iron
yoke stays at ambient temperature. This solution was possible because the forces were still
manageable by the austenitic steel collars, but poses serious issues of field alignment and force
interaction between coils and yoke. This solution was adopted because it shortens the cool-down time
from 10–15 days to 2–3 days, thus reducing the machine downtime for interventions. In modern
colliders the total number of collisions is the most important performance parameter, after collision
energy. Because of this, reducing the downtime is as important as increasing the collision rate
(luminosity). In the case of the Tevatron, given the numerous interventions that this pioneering
machine required, the room-temperature yoke was probably a good choice. Anyway, one should
remark that Tevatron magnets were built and assembled in the laboratory; industry was just providing
mechanical and electrical components [6].
Although all construction was in the lab, given the large number of dipoles (774), the design
and construction processes marked the beginning of industrially-oriented manufacturing with:
• use of ‘simple’ and relatively cheap Rutherford cable as conductor;
• use of collars to pre-stress the coils and contain the electromagnetic forces;
• use of laminations for collars, iron laminations and for manufacturing and assembly tooling;
less well known, because it does not show up in the design, this last innovation was critical to
achieve the requested tolerance at an acceptable cost.
The next large project was the Hadron Electron Ring Accelerator (HERA), which went into operation
in 1991 at the DESY laboratory in Hamburg, Germany [7]. A single proton beam guided by SC
magnets was made to collide against an electron beam. Of similar size and field to the Tevatron, the
HERA dipoles featured an important difference: use of cold iron around collars that were made of
aluminium alloy, to gain pre-stress during cool-down thanks to higher thermal contraction of
aluminium than that of stainless steel. A cold iron yoke, following the Isabelle design, was the
518
M ANUFACTURING AND T ESTING OF ACCELERATOR S UPERCONDUCTING M AGNETS
invariable choice of all main projects after the success of HERA. While the Tevatron magnets were
manufactured in the laboratory, HERA was the first industrially built large project. About 420, 9 m
long, dipoles and 220 quadrupoles, in addition to most of the 660 corrector magnets, were
manufactured by European industry.
During the 1990s the Relativistic Heavy Ion Collider (RHIC) was built at BNL in the tunnel of
the decommissioned Isabelle [8]. The field level was relatively modest for SC magnets: about 3.5 T.
RHIC had a limited budget, so the challenge was to keep the cost low rather than achieve record-
breaking characteristics. The 264 dipoles are each nearly 10 m long with an 80 mm bore, with cold
iron construction. The most important characteristic is that the iron laminations, in addition to being
the magnetic yoke for flux return, are used to restrain the electromagnetic forces, i.e. they are
effectively the collars. The coils are surrounded by simple, thin collars, serving as spacers and for coil
positioning and are made from a plastic material for cost saving. The iron is therefore very near to the
collars, thus contributing noticeably to the field while the low field helps to keep saturation effects
under control. The magnets are single-aperture so, to work as a collider, two complete, independent
rings had to be built. The RHIC magnets have been built to print in Industry with a strong involvement
from the laboratory.
In the US, after the termination of Isabelle and the start of the Tevatron, even before the design
and construction of RHIC, a study was launched to design and build the ‘definitive’ machine: the
Superconducting Super Collider (SSC), which had the goal of colliding protons against protons at
20 TeV per beam, an enormous leap for the discipline [9].
From 1983 until 1993, the year of its cancellation by the US Congress, the R&D for the 87 km
long proton–proton collider marked the development of magnets and the first attempt of
industrialization on a giant scale. To fill the 87 km tunnel more than 7700 dipoles, 15 m long and of 50
mm bore, were needed, and these were designed to reach 6.6 T at 4.4 K.
The design was based on austenitic steel free-standing collars surrounded by a cold iron yoke
with no mechanical function. Among the breakthroughs carried out in the context of the SSC R&D
have to be accounted: i) the increase of Jc from 2000 to >2700 A/mm2, 5–4.2 K, in conjunction with
high-quality fine filaments of 5–6 μm; ii) the industrialization of high-quality cabling; iii) a new cable
insulation based on full polyimide; and iv) a better way of collaring that was more suitable for
industrial production. The SSC project was stopped in 1993 for various reasons, including the
continuous increase of the cost and the presence of a less expensive, although more modest, program
aimed at the same physics territory: the LHC.
R&D for the LHC at CERN started at the end of the 1980s [10] with serious funding allocated
only after 1992, incorporating many features of various previous design studies and R&D (especially
from the SSC). However, to take full advantage of the existing 26.7 km long LEP tunnel, CERN
pushed Nb–Ti magnet technology to its extreme. Design innovations were: i) use of the two-in-one
design first proposed by BNL [11]; the LHC was actually developed for the original ‘twin’ dipole
variant, where the two channels are fully coupled both magnetically and mechanically; and ii) a
coolant temperature of 1.9 K by using HeII to boost Nb–Ti performance and make use of the superior
conductivity and heat transfer properties offered by superfluid helium.
The implications of the two-in-one layout on magnet design would go outside the scope of this
work. However, its success is such that currently all magnets for future hadron colliders are designed
with a dual bore, even for ion versus proton collisions, which is somewhat surprising given the
difference in mass-to-charge ratio.
In Fig. 1 and Table 1 are reported a synoptic view of the cross-section of the main dipoles of the
projects discussed above.
519
L. ROSSI
Tevatron HERA RHIC LHC
Fig. 1: Cross-section through the main SC dipoles of the SC colliders
Table 1: Features of the main SC dipoles of the SC colliders
Tevatron HERA RHIC LHC

Maximum beam TeV 0.98 + 0.98 0.82 (0.92) 0.25 + 0.25 7+7
energy 0.1/n + 0.1/n (4 + 4)
Dipole field T 4.3 4.65 (5.22) 3.5 8.3 (4.8)
Aperture mm 76 75 80 56
Magnetic length m 6.1 8.8 9.5 14.3
Stored energy MJ 320 250 60 8000
Weight (cold mass) kg 3,607 27,000
Quantity # 774 422 264 1232
Iron yoke Warm Cold Cold Cold
Operating K 4.2 4.5 4.6 1.9
temperature
Operating He supercritical He saturated He forced flow HeII at 1 bar
condition
Structure Single bore, Single bore, cold Single bore, cold Twin bore, cold
warm iron, iron, straight iron, curved iron, curved
straight
Force retaining Stainless steel Al alloy collars Yoke ‘collars’ Stainless steel
collars collars + shrink
cylinder via
yoke
Operating current A 5,027 (5,640) 5,050 11,850
3 Cabling and insulation

Here we do not discuss the cable from the SC property point of view, but just as an element, and a
critical one, of magnet construction. Regarding insulation, we will not discuss the electrical properties,
rather we draw out the practical implications for coil technology.
3.1 Rutherford cable

The cable is a fundamental unit of the construction. A Rutherford cable is composed of twisted
strands, see Fig. 2. Cabling can in part add or remove the strand twisting, which contributes to the
mechanical stability of the cable. In Fig. 3 a cabling machine used for LHC cable production is shown.
The most important characteristics of the cable from the point of view of the winding are the
dimensions and mechanical stability under tension and bending.
520
Fig. 2: Composition of the LHC dipole Rutherford cable
(a) (b)
Fig. 3: Cabling machine for the LHC Rutherford cable: (a) view of the large rotating wheels carrying the strand
spools; (b) view of the ‘turk head’ where the cable is formed.
3.1.1 Dimensions
The collar concept relies on enclosing the coil package inside a cavity whose dimensions are given by
the collars. The typical accuracy required by magnet designers for the coil package is about 50 μm or
better. Since the number of turns in a coil (see Fig. 4) is of the order of 20–40, in principle the
accuracy required of a single unit, the cable, would be 1 μm. Actually, the accuracy of a strand is of
that order and the accuracy of the cable cannot be as good. However, in series production what matters
is not, to a certain extent, the absolute value. The average thickness of a cable is a value that, within
limits, can be corrected or compensated for with a systematic action (for example, by adjusting the
size of the collars or by shims between the collars and the coil): therefore, what is important is that the
variation is kept within very tight limits. In Fig. 5 the actual variation in one of the cable production
lines for the LHC is shown. It shows an amazing uniformity of better than ±2 μm. However, this is not
the thickness ‘seen’ by the coil manufacturers. The values in Fig. 5 are measured just at the end of the
cabling line, before any manipulation, before the thermal treatment for inter-strand resistance control
and before the cable is unwound from the spool and then wound onto the coil. Being a composite and
not a solid piece, the cable size depends on the conditions of measurement in a very subtle way. So in
the coil winding process it is important to have a means of accepting a different average value (also
10–20 μm, which translates to 0.5 mm of the total thickness) while the handling, tooling and
processing must ensure that the transverse pressure and longitudinal tension be always uniform. Only
in this case can we profit from the uniformity shown in Fig. 5 for manufacturing coils that have
521
L. ROSSI
dimensions as uniform as possible. As a last remark it should be remembered that the Rutherford cable
must always be measured at the reference transverse pressure, the one used for coil-curing (Nb–Ti) or
treatment (Nb3Sn), and in the correct conditions of side confinement, in order to have consistent data.
Fig. 4: LHC twin dipole CC with an expanded view of a quarter of a SC coil
Fig. 5: Cable thickness measurements from a production line for LHC dipole inner cables. More than 1000 km
of cable are represented here.
The mechanical stability of the Rutherford cable is also very important. A Rutherford cable is a
loose ensemble of strands, plastically deformed at each folding (but with some spring-back if a
thermal treatment is not applied), with no glue between the strands. A Rutherford cable is stable if it is
contained by means of suitable tooling or if it is tensioned with the right force: too small a force will
leave the strands free from each other and a too high tension will drive the cable near to an instability
point and the cable can suddenly buckle into a round cord. This second case is a disaster that is usually
detected during winding: depending on the degree of buckling and the length, and upon the skill of the
operators, the cable may be recovered. The first type of instability is subtler: during winding, a strand
can bulge out, especially in the place where winding tension changes, such as in the bends. If the
defect is identified during winding it can be repaired, by experienced operators, by unwinding and
then pushing back the strand(s) popping out and restarting with more care: it takes time and increases
costs. However, the strands are below the insulation when the coil is wound, and the defect may well
522
pass undetected. In such a case the cable may be degraded, more or less severely, during curing or
thermal reaction. With Nb–Ti the degradation in terms of critical current is not too severe [12];
however, the real danger is possible damage to the insulation. Cable mechanical stability must be
tested a priori and specified at the bidding time, as was done for the LHC. We conclude this section
on cable stability by remarking that the Rutherford cable has a rotation, or transposition sense: it is
built with left-hand or right-hand torsion, see Fig. 6. It is easy to realize that if the coil is wound with
the same rotation sense of the cable torsion, the cable itself tends to increase its stability, while if it is
wound in the opposite way its stability is decreased. If each layer is wound independently this is not a
problem: winding sense and cable torsion sense can be adjusted. However, if the coil is built with the
double pancake technique, i.e. without a joint in the layer jump, one layer is wound in a favourable
sense and one in a less-favourable sense. For the LHC this was an issue at the beginning of
prototyping when the cable was not yet optimized. During manufacturing it remained a concern only
for the main quadrupoles, wound in a double pancake: in general, the series production LHC cable was
so well optimized and mechanically stable that this did not cause much trouble, with the exception of
the cable from just one manufacturer.
(a) (b)
Fig. 6: (a) Left-hand and (b) right-hand torsion cables for a LHC dipole. Cable (a) is from LHC production,
while (b) dates from 1996 and was used in the first LHC 15 m long prototype (CERN–INFN collaboration). At
that time cables had a long transposition pitch and were not oxidized.
The cable is by far the most precious component of a SC magnet: the cable makes about 20-
30% of the cost of a Nb–Ti magnet; this cable cost fraction can reach 50-60% of the total cost for
Nb3Sn magnets. Therefore, the first manufacturing issue is avoiding the waste of cable. A good
practice is using dummy copper cable (better if alloyed to have better mechanical properties) for
debugging and then tooling up for the production process. Another, even more useful, practice is the
use of super-dummy cable (low-value superconductor, of degraded SC performance but of near-to-
final mechanical properties) for fine adjustments to the whole coil manufacturing process. In many
cases a super-dummy coil and coil assembly may also be useful for testing the entire cold mass
assembly and even when debugging the testing station and procedures.
3.2 Insulation
For insulation, dielectric strength put aside, the problems are of two types. The first one concerns the
ability of reaching the desired geometrical tolerance. Each cable is covered with insulation of a
thickness of 70–150 μm. If we take an average of 2 × 125 μm for an LHC dipole (each cable has two
faces) and compare it with the cable thickness (see Fig. 5) we see that it is about 10% of the total
thickness. However, while the cable can be accurate at a level of ∼0.1%, manufacturers of insulation
sell it with a variability of 5% or worse. So the uncertain insulation thickness may dominate the
523
L. ROSSI
geometry of the coil. Measures can be taken, by working with the supplier to better qualify the
process, both by intrinsic improvement of the uniformity and by sorting production into batches of
different average thickness and small variance. It is important anyway to measure the thickness of
stack of many insulation layers and, finally, to measure the thickness of insulated cable stacks
(typically 10 cables) under transverse pressure and conditions similar to those experienced during
curing and/or thermal treatment. During curing of polyimide or epoxy pre-impregnated insulated coils,
insulation flows at high temperatures, partially filling the interstices between adjacent strands, thus
impacting both on cooling and on the final effective dimensions of the coils.
A phenomenon that might be of considerable importance for large projects is insulation creep.
Creep is usually important only at high temperatures; however, if a magnet is stored for a few years in
improper conditions, as happened to the main LHC dipoles because of delays to the test station and
later due to tunnel availability for installation, creep might play a role.
Another practical issue for insulation based on polyimide, such as the LHC main magnets, is the
careful control of the polyimide glue on the external layer. Glue, which is activated by curing at about
180°C, is necessary to handle the coil until collaring takes place. The tape on the glue is wound with a
gap to create the very thin channels (2 mm wide, 0.02 mm thick) in which superfluid helium can flow
to cool the winding. The correct curing of the glue happens in a relatively narrow window of
temperature and humidity. This last factor is usually neglected but it can have a detrimental effect. In
particular, low humidity can, surprisingly, disable the curing effect. A careful check of the uniformity
of the quality for the glue was important during LHC construction. Due to some alarming results,
sampling of the glue quality, done via a peel-off test, was reinforced, which led to the detection of a
serious degradation of the gluing factor with time. Certain deliveries had to be refused, and a stop to
coil production due to a lack of adequate insulation was avoided with a very small margin.
To conclude this section about insulation, we would like to pass on our experience accumulated
in many large and small projects.
i. One must be generous with insulation, inter-turn, inter-layer and to ground: an insulation
defect, in the presence of high Jc and large amount of stored energy, can suddenly and
unexpectedly damage the magnet, without any precursor. Usually, a mechanical default
and/or quench bad behaviour degrade the performance of the magnet, and only rarely do
they prevent the magnet from working at some useful level. However, an electric fault may
easily make the magnet completely unusable.
ii. Insulation must be the responsibility of the magnet manufacturer, avoiding additional
interfaces, outsourcing and dilution of responsibility. In the LHC project, CERN provided
all insulation material. However, the magnet manufacturer was responsible for a careful
inspection of the insulation process (done by subcontractors), as well as a thorough check of
the conformity of the insulation to the specifications and for its suitability for the actual
construction and operating condition of the magnet.
4 Nb–Ti technology and the LHC experience

Nb–Ti technology owes a large fraction of its success to the ductility of the Nb–Ti cable: the single
wire behaves like a copper-reinforced wire and can survive mishandling or process mistakes. Really
large errors and huge mishaps are needed to produce damage to a Nb–Ti wire or cable.
The construction of an accelerator magnet can be subdivided in two main parts: CC and cold
mass completion (CMC). Here we review the various steps for each part.
524
4.1 Collared coil
4.1.1 Coil winding

The single coil unit is a pole coil that can be composed of a single layer or double layers. In the latter
case the coil can be fabricated with the double pancake technique if a single unit length is to be
employed for the whole coil, or by joining two separately wound coil layers. In this case, also called a
‘false double pancake’, a different cable can be employed allowing current density grading. In
practice, accelerator coils are composed of one or two layers. There are a few cases of four-layer coils,
which are usually composed of two double pancakes. Coil winding must be very accurate (the tooling
must ensure about 20 μm accuracy over 15 m, which can be done at an affordable cost only by
assembling laminated components; see Section 2). A double pancake is very suitable when the
azimuthal extension of the two layers is similar, otherwise the insulation curing (see below) may
become difficult. The most critical issues are as follows.
i. Avoid damaging insulation with sharp-edged metal (but plastics may also be dangerous).
ii. Ensure correct, uniform tension of the insulated cable both in the straight parts, in the ends
(submitted to a 3D bend) and when tension is transferred from the spooler to the retaining
device. These are usually pneumatically activated pistons, pushing on the straight part after
or before the end bend is engaged.
iii. Avoiding strand pop-out (see the section above on the Rutherford cable).
iv. Make good end parts. The saddle-shape ends needed for our magnets are actually one,
probably the main, source of quench. Although optimized codes to design them are
available, making a good end is still an art requiring more than one iteration.
v. Formation of the transition between layers, usually called layer jump.
4.1.2 Coil curing

This is a moderate thermal treatment (150–190°C) under pressure to cure, or polymerize, the glue that
is put on the external layer of the cable insulation. By gluing the layer in the right position, the coil
acquires a consistency and a shape to withstand all manipulations until the coil is collared.
Polymerization is the second critical step, not only because of the sensibility of the component to its
quality and to various environment conditions, but also because the accuracy of the polymerization
tooling determines the precision of the final coil. For this reason the polymerization pressure is very
high, not far from the pressure needed for collaring, and much care is taken in the design of the curing
mould (or cradle). For LHC coils pre-polymerization happens under transverse (azimuthal) pressure
up to 100 MPa at 100–135°C and then final curing under 80 MPa transverse pressure at 190–193°C.
The pressure corresponds to a vertical pressure of 2.5 MN/m (single coil). In the case of a two-layer
coil (or two double-pancake coils, or a four-layer one) one can choose between the following options.
i. Winding the outer layer on the top of the inner one and then polymerize the complete coil.
This is the only possibility for a real double pancake with no electrical junction between the
two layers.
ii. Polymerizing the inner layer right after winding, then winding the outer layer onto the top of
the inner one, finally polymerizing the whole coil. In this way the inner layer is polymerized
twice: however, this ensures better control of the accuracy of the absolute value of the radius
and the variation of the second layer. Also, this procedure gives more flexibility to
production lines for a large quantity, allowing better tooling specialization. This procedure
also allows the assembly of quench heaters on the top of the inner layer if necessary.
However, it requires soldering the cables of the two layers, always a delicate operation liable
to weaken the electrical integrity of the coil, becoming a possible source of heating,
525
L. ROSSI
reducing the stability margin (in the case of a bad joint). In any case, the electrical joint is a
singular point in the mechanics of the coil assembly.
4.1.3 Pole assembly

This is the work required to make the coils ready for collaring: installation of coil instrumentation;
assembly of the outer layer on the top of the inner one, if separately polymerized; soldering of the
cable for connecting the inner to the outer layer (if not a double pancake); and insertion of the
interlayer insulation, which is usually a critical component that must ensure electrical insulation while
allowing a generous flow of coolant for the required heat removal from the outer coil. The following
step is the assembly of a critical component: the quench protection heaters. They are usually placed on
the outer side of the external layer; in the LHC main dipoles there are two of them, for redundancy,
which means that the coil azimuthal extension of a single quench heater is less than half. Quench
heaters, a metallic strip encapsulated between two polyimide layers, must be thin for fast heat
diffusion into the superconductor, however they must also be robust enough to withstand a pressure of
150 MPa and thermal contraction, which are contradictory requirements that require a compromise to
be found. Various layers of ground insulation, typically in polyimide thin foils, are then placed and a
coil protection sheet is eventually laid onto the ground insulation to make sure that the high-pressure
during collaring is uniformly distributed, avoiding peak stress concentrations.
4.1.4 Collaring
Collaring is a critical operation because, in most types of design, it encases the coils in their force-
retaining structure and also gives the final shape to the coils, thus playing a big role in the main
performance indicators of a SC magnet: quench behaviour and field quality. The collaring operation is
carried out under powerful presses, see Fig. 7, and in the case of the LHC about 300 MN are applied to
the 15 m long collaring to obtain the target pre-stress [13]. All components are manufactured to an
accuracy level of 20–30 μm, and assembly aims at 50 μm final accuracy. The accuracy of the collar
profiles through production of the LHC dipoles is shown in Fig. 8: about 12 million collar pieces have
been manufactured in special austenitic steel through a fine blanking technique. In order to preserve
this accuracy of components in the finished magnet, the hardened steel beam that transmits the force
from the press pneumatic pistons to the collar–coil package is a single piece, 300 mm thick, 700 mm
wide, machined along the 15 m length to an accuracy of 20 μm.
(a) (b)
Fig. 7: (a) A 15 m long collaring press for LHC dipoles, capable of 2100 tonnes/m (∼20 MN/m) force; (b)
insertion of a CC assembly into the press for performing the collaring operation.
526
Fig. 8: Evolution of two important dimensions of the LHC dipole collars through production (Value No.
indicates the batch number), showing the ability to stay, after an initial correction, in the desired range of
±20 μm.
The coil must be very accurate too, and its radial thickness and azimuthal extension must be
near to the absolute target value, to avoid being damaged by the collar closing. But even the most
precise coils cannot become their final size until pressure is applied. In other words, coils change size
during collaring, passing from zero to peak stress of 150 MPa (see [13, Fig. 19]). The movement of the
conductors can range from one to several millimetres, in the case of coils with spongy insulation.
Collaring must be accomplished slowly, massaging the coils through cycles: during each cycle the
press is released after locking rods or locking keys are inserted when the dimensions are near to those
of the final value. Collaring is best accomplished by carefully choosing the point of pressure
application, in a way to minimize the stress lost after releasing the force by the press; and the job of
keeping pre-stress is passed over to the locking rods or the locking keys. Further discussion depends
on the details of the coils, the collars’ shape and the press.
A final word on collaring is about the style of manufacture. For a dipole a horizontal axis press
naturally pushes in a vertical direction. However, this choice is less evident for quadrupole. To respect
the four-fold symmetry a solution is to have a press with a vertical axis, and to apply pressure along
two perpendicular directions at the same time. Devised at the Commissariat à l’Energie Atomique
(CEA), in France, many years ago, this system has been applied to most quadrupoles produced so far,
with excellent results. However, it inconveniently requires positioning the coils in a vertical direction;
therefore, the height available above the floor must be two times the length of the magnets, to allow
the coil to be inserted into the press. Quadrupoles are usually shorter than dipoles: the LHC main
quadrupoles are 3.5 m long, for example, and this is not too difficult. It would be complicated for the
next generation of IR quadrupoles for the LHC where low-β quads (if in Nb–Ti) more than 10 m long
would be required, unless the magnet is further subdivided into subunits. For such a size, a horizontal
press like that used for a dipole would be better to avoid enormous vertical infrastructure. Of course
this technique requires that the uniaxial (vertical) force is transmitted to two orthogonal axes at 45%
from the vertical. The system, first proposed for the low-β of the SSC [14], and also for the first design
of the low-β quad for the LHC upgrade [15], has to be carefully tuned in detail to avoid a rise in
unwanted spurious components.
An interesting alternative path is the use of the iron yoke as a collar. For dipoles this is a
possibility for low-field magnets, where iron saturation is not severe: iron must be radially very near to
the coils to be effective, and saturation may degrade field quality. The use of such system for
527
L. ROSSI
quadrupoles has been carried out for the MQXA magnets, a type of LHC low-β quadrupoles, by the
KEK, the High Energy Accelerator Research Organization, in Japan [16]. As with a horizontal press,
the problem is that the main force is uniaxial: a system to project it on two perpendicular axes 45°
from the vertical is necessary. Again, the results have been very good, but not without some
corrections for the rise of unwanted harmonics.
4.1.6 Coil enclosure

The CC package is quite robust and can be easily handled. Many measurements are carried out on this
package (see below), which then has to be completed with the insertion of the cold bore tube (CBT),
which is the beam pipe. Since aperture is one of the most precious assets of an accelerator magnet, the
room left between the CBT and coils is usually minimal. For the LHC dipoles, which are 15 m long,
the solution was to assemble the coils before collaring around the CBT, so it remains trapped inside
the coils after collaring. The disadvantage of this solution is that an intervention on the CBT requires a
de-collaring of the magnet, which is a big operation and not without a risk of damaging the coils
themselves. In the case of the LHC quadrupole the smaller coil length against that of the dipole (3.3 m
against 14.3 m), and the slightly larger margin between coils and CBT (gained by reducing the
insulation of the CBT, which was possible because of the lower quench voltage of quadrupoles than of
dipoles) has allowed insertion of the CBT inside the coils after collaring. The operation could be
repeated for the very few quadrupoles for which it was necessary, but not without damaging the CBT
insulation, which had to be replaced (insulation is so hard to remove from the CBT that its re-use is
not convenient).
The collar package is then finished by closing the coil ends with a plate: this plate is the
interface between the coil ends on which the longitudinal force is exerted (see Fig. 9) and the blocks or
screws that oppose this force by loading the end flange of the cold mass. This CC flange also has to
allow an exit for the instrumentation wire and the quench heaters’ wiring.
Fig. 9: Picture of an LHC dipole CC, finished with the end plate at its extremity
4.2 Cold mass completion
4.2.1 Yoke assembly

The first operation is to surround the CC with their magnetic circuit. Accuracy of the iron lamination
composing the yoke is typically 50 μm, which also calls for a fine blanking manufacturing technique.
This is very important, especially when for cost-saving reasons the thickness is rather large: in the
LHC dipoles it is almost 6 mm (in resistive magnets the laminations are typically 1–1.5 mm). If some
528
parts of the yoke contribute to the mechanical structure (see, for example, the iron insert in the LHC
dipole cross-section, Fig. 10), their accuracy is similar to that of one of the collars, ∼20–50 μm. Iron
laminations are covered with paint, or a passivation (phosphatation) or oxidation layer, in order to
avoid eddy current build-up and rusting: however, rusting typically occurs on the lamination edge, on
the surface exposed to the cut. An important feature of the yoke is that the filling factor must be under
control: in relative terms for the field homogeneity between magnets, since the iron yoke contributes
approximately 15–20% to the total field, and in absolute terms because it affects the spring-back after
magnet curvature (see below).
Fig. 10: Cross-section through an LHC dipole (magnetic circuit or cold mass)
4.2.2 Restraining cylinder/helium vessel

The external cylinder has the primary function of being the lateral surface of the cylinder that
constitutes the helium vessel, as in the case of the HERA dipoles. However, different functions can be
assigned to the external cylinder by the magnet designer. For example, in the case of the LHC main
quadrupoles it works as an inertia cylinder: it ensures straightness along the same axis of the main
quadrupoles and the corrector package (octupoles, trimming quadrupoles, sextupoles, and orbit
correctors’ dipoles). However, it is not required to contribute to coil pre-stress. For this reason it is a
16 mm thick 316LN steel single-piece cylinder, with precise pins for magnet precise positioning. Such
a choice is naturally coupled with vertical assembly, requiring large infrastructure (see Fig. 11).
In the case of the LHC main dipole, the cylinder is required to assist the collars by exerting an
additional pre-stress on the coils, via the yoke. Such an assembly is also called ‘line fitting’ because
the relevant surfaces must be very precise and fit (see Fig. 12). The pre-stress is given via shrinkage
provided by welding two half-cylinders that have an azimuthal length shorter than the azimuthal
length of the outer yoke. A discussion of this point and of its kinematics is beyond the scope of this
paper. It is sufficient to say that, to ensure the required uniform shrinkage in the presence of a
geometry that cannot be as precise as desired with 16 m long open half-shells, CERN has developed a
new welding methodology by adapting the surface tension transfer (STT) technique to austenitic steel
and to synchronous double-side topology. STT is an advanced fast method for oil pipes, and it has
been developed for carbon iron with one single-side welding. This has been a great effort because a
fully robotized system integrated into the huge presses (220 MN over 15 m with an accuracy of 50 μm
529
L. ROSSI
for a beam cross-section of 600 mm × 300 mm but free frame of 1.5 m × 1 m) had to be developed and
industrially commissioned where the cold mass construction was taking place. The welding press and
welding operation is shown in Fig. 13.
Fig. 11: System to turn the LHC Quadrupole cold mass vertically (courtesy of Accel, Germany)
Fig. 12: Quadrant cross-section through an LHC dipole, with the precise (25–50 μm) fitting surfaces indicated
with dashed lines.
This step of the longitudinal shrinking cylinder welding also serves to curve the magnet. A
sagitta of 9 mm is generated on the 15 m long, 600 mm diameter cylinder. The magnets are welded
with an over-bend sagitta of about 15 mm; then the spring-back due to welding shrinkage and to the
530
spring effect of the iron laminations (which are more compressed on the inner side) reshape the
magnet at, or near to, the target curvature. Actually it was almost impossible, despite all attempts, to
attain the severe ±1 mm sagittal tolerance and even more difficult to achieve the ±0.3 mm tolerance
required at the extremities. However, an ingenious way of assembling the magnet inside the cryostat
and a sorting during installation has allowed achievement of and even exceeding the target alignment
tolerances, with the result that today the LHC machine is better aligned than expected, with beneficial
consequence for the beam dynamics.
(a) (b)
Fig. 13: Automated dipole welding in the large press during series production: (a) front view of the welding
press, with one dipole under the press enclosed in the cradle mould during welding, and the next dipole waiting
to be inserted; (b) side view of the robotized welding operation.
4.2.3 Connections and bus bars

The various coils need to be connected to form the circuit and then connected to the by-pass line for
protection (bus bars and diode), see Fig. 14. The connections must be stabilized and secured against
electromagnetic forces. Lack of space makes it difficult to integrate the various components; and even
standard processes and operations, such as TIG welding, may become very difficult, requiring special
procedures and very skilled personnel. In addition, stability and reliability of connections that have to
carry 13 kA under various conditions (including a 50 mm elongation when passing from 300 K to
2 K), and submitted to a violent heating and temperature gradient in the case of quench, should not be
underestimated. In the LHC dipole this part is less than perfect and it may eventually be a limitation of
the whole accelerator.
The various connections to the diode are not well engineered, and the surface resistance under
certain conditions may increase outside the control limits; the diode package and the various copper
buses have an electrical insulation that is far from robust. Naked surfaces are dangerously close to the
enclosing steel box (ground), making this system vulnerable to shorts caused by metal chips. Indeed,
the quantity of steel chips and burrs that have been trapped in the magnet is much greater than
expected. This is certainly an unexpected drawback of having a thick welded cylinder in the presence
of a diode circuit that is not well protected and in the lowest point of the magnet. During the first
commissioning of the LHC, following the strong He flushing, necessary for the cool-down, we
experienced a few electrical short-circuits. We expect that occasionally, after future cold-down–warm-
up cycles or after the strong mass flow induced by a quench, the LHC dipoles may again occasionally
experience this type of short.
4.2.4 Corrector magnets

It is normal to assemble a good fraction of the correcting circuit directly on the main dipoles, to have a
chance of locally correct systematic effects (either geometric or persistent currents). Since these are
531
L. ROSSI
small magnets, they are assembled around the beam tube and are called ‘spool pieces’ to distinguish
them from other correctors or higher-order magnets that are placed in the corrector package that is
usually placed together with the main quadrupoles. In the LHC all dipoles have sextupole correctors
(providing an integrated sextupole strength equal to that given by the whole of the lattice sextupoles)
on one extremity of the cold mass; in addition, each second dipole features, on the other extremity, a
small octupole–decapole correction package.
The main difficulty is placing the magnetic axis of the spool piece at the right position, with a
tolerance that for the LHC is ±0.3 mm. The connections are not difficult but their numbers are
enormous—about 60,000 inside the cold masses: since some families of corrector magnets are series-
powered over the whole sector (⅛ of the LHC), one single bad connection kills ⅛ of the correcting
strength, which is not acceptable for LHC operation.
(a) (b) (c)

Fig. 14: Extremities of an LHC dipole (before closing the cool mass), showing: (a) the connections between
various coils; (b) the lyra-shaped bus on the other extremity of the dipole; (c) the corrector magnet assembly.
4.2.5 3D part
The finishing of the end part is technically less interesting from the point of view of the magnet
design; it is, however, critical for the success of the magnet’s operation. We review here the main
operations.
i. Insertion of the Heat EXchange (HEX) tube, the complex copper-stainless steel vacuum
brazed tube where the superfluid helium (HeII) is actually formed and that takes the
thermal load of the magnets at 1.9 K.
ii. Force retainer closing flange. This is a very thick flange that is placed against the coil
closing flange, to support the longitudinal force. This stainless steel flange is 50 mm thick
in the LHC since it has to support a force of the order of ∼10 tonnes without bending. It is
pre-loaded against the end covers (see below) with adjustable screws.
iii. End covers and weldings. LHC dipoles and quadrupoles are closed longitudinally by end
covers, which are almost all dish-shaped: welded to the external cylinder they form the
HeII enclosure. In the LHC there is more than 40 km of thick stainless steel welding that is
HeII-tight at 20 bar (pressure peak during quench). In addition, there are some 40,000
welds of various bellows of various sizes and other delicate welds, such as the non-fully
penetrating fillet welding on the beam tube that separates the superfluid helium from the
beam ultra-high vacuum and from the cryostat insulation vacuum. All must be HeII-tight
under 20 bar operation pressure (quench) and many have severe alignment and
dimensional tolerances. To have success a strict campaign of qualification of the process
and of the people executing the work was put in place, using advanced technology. It is
interesting to note the special development for the end covers of the dipoles. The end
covers were produced with powder sintering technology, the first application on 40 mm
thick 316LN of 600 mm diameter, which has won a special prize for CERN and the
Finnish company METSO.
532
iv. Cabling and instrumentation. In the LHC magnet series construction, the instrumentation
was reduced to a minimum, in order to reduce risk and cost; a decision that we partly
regretted at a later stage, during magnet testing and accelerator commissioning. However,
numerous wires for quench detection (redundant), for temperature measurement, for
quench heater powering, and for diode control had to be properly installed using a minimal
space. The exit port of the LHC instrumentation is not really satisfactorily arranged, also
because of the tight space, and numerous interventions were required to fix problems
during test and commissioning.
In the CMC, and especially in the 3D part, the complexity rather than the technical difficulty plays a
major role. It is the realm of integration rather than technology, and somehow appears less
challenging. For this reason it is often engineered too late, when good solutions are no longer possible.
If there is an invariable lesson coming from all projects, it is that the integration studies should have
started at an earlier stage.
5 LHC construction experience
5.1 Quality assurance and quality control

A quality assurance and quality control (QA/QC) plan can only follow a compiled list of the
manufacturing steps, an analysis of what is really important to measure, a decision about what are the
acceptable tolerances, and a clear definition of what to do when the measured values are outside the
target values. All of this can be done only if the design is properly elucidated in all aspects. The
technical specification must reflect this analysis.
A good relationship with the manufacturing team is important to transmit the idea that QA/QC
is not there to impede production; it is there, rather, to make sure that things go well and that the
product is acceptable. Incredible as it appears, manufacturing contractors frequently perceive that
measurements and checks slow production down: the result is that manufacturing may be of too poor a
quality and needs to be redone, and painful extra costs are eventually generated by disassembly,
corrective action and re-assembly.
On the other hand, the client often tends to ask for all possible measurements, just in case,
without setting a proper data analysis process with clear acceptance criteria and often without proper
models that can provide guidance on minimizing rejections and the length of production stoppages.
In the case of the LHC dipoles, the technical specifications [17] contained all checkpoints, the
data were transferred to CERN on-line for immediate analysis and a team of analysts was always
available to give an answer in half a day to a few days for the most difficult cases.
Also, resident inspectors, provided by a professional QA/QC company through a global contact
for the whole LHC project, were placed by CERN at each production site, complementing the
numerous visits of CERN technicians, specialist engineers and project engineers.
The most important measurements, apart the electrical ones, have been the coil size and the
magnetic measurements. The coil azimuthal size allowed the determination of the thickness of the
shims that are placed between coil and collar at the pole position (about 60°), see Fig. 4. The coil size
varied typically ±50 μm, with some systematic fluctuation. This almost always allowed the use of
nominal shim values, giving a big advantage for the uniformity of the harmonic content.
In Fig. 15 the values of the shim thicknesses used for the whole of production are reported [18].
The undulation of each single coil has been contained in 20 μm root mean square, a very good value
for 15 m long coils with uniform shims.
533
L. ROSSI
Fig. 15: Thickness of the shims placed between coils and collar pole, determining azimuthal extension and stress
upon the coils.
As part of the QA plan, magnetic measurements were introduced both at the CC level and after
completion of the magnetic circuit with the iron yoke. At the beginning they were thought mainly to
steer production toward the beam dynamic target, i.e. to guarantee that the bending strength per unit
current, B1 × Lmagnetic/I (also called the integrated transfer function), the sextupole b3, the decapole b5
and the septupole b7 were inside the tolerance band. Two small fine-tunings of the cross-section, via
changes of the copper wedge profile and thickness, have been necessary to maintain b3 and b5 during
production in the desired limits, as shown in Fig. 16 for the sextupoles [18]. Magnetic measurements
were revealed to be a powerful tool for finding, by means of a magnetic model of the coils, the exact
position of a single cable or a coil block inside the CC when visual inspection and other methods were
no longer possible. A detailed account of the errors and imperfections that have been intercepted with
this method is reported in [19]. The accumulated number of important defects on the CC (requiring a
coil de-collaring and corrective actions), intercepted through magnetic measurements, through dipole
production showed a high rate of faults during two periods. The first period is reasonably related to the
learning phase of the process when passing from prototype to production. After that initial phase, the
production goes rather smoothly without any large problems, until a certain point when the fault rate
has a steep increase. Later on, the sudden increase was correlated with the introduction into the
workshop of new people, hired to staff the new production lines that had been installed to meet the
required dipole production rate (see the following section on learning curves).
Fig. 16: Evolution of the normalized sextupole, b3, of the main LHC dipoles during construction. The vertical
lines indicate the points when the two fine-tunings of the coil cross-sections were introduced.
534
5.2 Learning curves

Dipole production has been analysed in terms of so-called ‘learning curves’ [20]. A learning curve
represents the production time for a unit when expressed as a percentage reduction of the production
time for a unit when doubling the production number. When producing the first N units, a time ∆t is
required. A learning percentage (ρ) reduction of 90% means that to produce a further N units it takes
0.9∆t; then to produce a further 2N units a time 0.9⋅(1 + 0.9) ∆t is necessary, and so on. This reflects
an exponential decrease that can flatten very quickly for exponent near unity. In Fig. 17 we report the
behaviour for the construction of the dipole for one of the three production lines. Table 2 indicates
where LHC production is situated inside the panorama of industrial products.
Fig. 17: Learning curves for LHC main dipole CC in one production line. The fitted line indicates the learning
percentage (lp or ρ).
Table 2: Learning percentage (ρ) for LHC main dipole production compared with those of various other
production items.
Industry Learning percentage (ρ)
(%)
Complex machine tools for new models 75–85
Repetitive electrical operations 75–85
Shipbuilding 80–85
LHC magnets 80–85
RHIC 85
Aerospace 85
Purchased parts 88–88
Repetitive welding operations 90
Repetitive electronics manufacturing 90–95
Repetitive machining or punch-press operations 90–95
Raw materials 93–96
It is interesting to examine the graph of the production forecast for single coils for one particular
production line. Coil manufacturing may be the most delicate operation, heavily relying on the skill
and training of personnel. The green (upper) curve in Fig. 18 expresses the projection of the company
at the time of the plan (and costing forecast). The red and blue curves show the reality. It is interesting
to note the correlation, done rather late in production, between the spikes indicating a sudden increase
in the unit production time and the injection of new personnel into manufacturing, almost a textbook
case. This shows the weight of the human factor in SC magnet production. Its success does not depend
on design alone: proper tooling, a well-defined process and, above all, experienced or well-trained
personnel play a major role in the quality and rate of production.
535
L. ROSSI
5.3 Industrialization and construction strategy

LHC dipole production can be taken as an example of an industrialization strategy based on a non-
negligible investment with a very good result. Starting at the end of the 1980s, a relatively small, albeit
slowly increasing, continuous program took place in close connection with industry. It may be
legitimate to retrospectively wonder whether CERN went to industry too early. However, this helped
to keep industry in the game and to gather support for the LHC from the CERN member states, by
evoking the large industrial return that the LHC magnets entailed (in terms of money and technology
transfer).
Fig. 18: Evolution of time needed for coil manufacture for LHC dipoles in one line (Alstom–Jeumont
consortium), and its correlation with the introduction of new personnel.
Companies manufacturing LHC main dipoles were: Jeumont–Schneider and Alstom, later the
Alstom–Jeumont consortium in France; Ansaldo Componenti, later Ansaldo Superconduttori Genova
and then ASG in Italy; and Noell Pressaug, later Babcock Noell Nuclear and then BNG in Germany.
An early involvement of Austrian Elin was discontinued in 1994 after a short model (in Nb3Sn) and a
long prototype; and an attempt to involve Oxford Instruments (UK) did not go beyond a short model.
In Fig. 19 the first LHC dipoles, designed with a 50 mm bore, aluminium collars and 10 m long cold
mass and developed in the CERN–INFN collaboration [21], is shown on the CERN test bench in 1994.
Fig. 19: The first LHC main dipole, 10 m long, on the test bench at CERN in June 1994, during the successful
test campaign.
536
The other large magnets were fabricated without a long R&D phase between CERN (and
collaborating institutes such as CEA) and industry. The company Accel in Germany was contracted to
manufacture the series production of the main quadrupoles and part of the insertion quadrupoles. The
remainder of the insertion quadrupoles have been manufactured by Tesla Engineering (UK). The
special low-β quadrupoles have been provided as an ‘in-kind contribution’ by Fermilab (US), who
manufactured the magnets in its own premises, and KEK (Japan), who, after an R&D phase carried
out in the laboratory, contracted construction to Toshiba.
Coming back to our story of the main dipoles, CERN kept the three companies in the business
by continually assigning contracts: also, due to limited resources, CERN renounced manufacture of
long magnets on its own site at the beginning. However, around 1995–1996 it became clear that at
least one production line was required for cold mass completion in-house (CERN building 181, later
called the CERN Magnet Assembly Facility (MAF)) to set up the many technical details.
As discussed above, these details may be less attractive or challenging from a technological
point of view, but they were all needed to carefully set up and consolidate the following: insulation of
the cold bore tube, the iron assembly method, pole and aperture connections, lyra and bus bars,
shrinking shell welding technique, end-cover technology, corrector assembly, the curvature formation
technique, alignment metrology, etc. These are the most critical from the list of technologies that had
to be developed or set up in the CERN assembly facility.
A picture of CERN MAF around the year 2000 is given in Fig. 20. In 2001, with all work
transferred to industry for dipole mass production, the CERN MAF was converted to an assembly hall
for the cold mass assembly of the LHC insertion magnets.
Fig. 20: CERN MAF (building 181, formerly the ISR experimental hall), with all equipment for dipole cold
mass assembly around year 2000.
The strategy for procurement of the main components of the dipoles underwent a slow, but
steady, change between the early years (1988–1992) and the final contracts for series construction,
issued in 2001. In the initial years CERN was ordering turn-key finished magnets, completed in their
537
L. ROSSI
cryostats, very much like HERA dipoles. However, for many reasons this idea had to be changed to
arrive at a point where all main components were procured by CERN and supplied to the magnet
manufacturers. In certain cases CERN procured even the raw materials, such as the low-carbon steel
(iron) for the yokes, which was then given to various manufacturers for formation of the laminations
via fine blanking; finally, the laminations were distributed to the various companies in charge of the
construction of the magnets (dipoles and quadrupoles). CERN became a supplier to its own suppliers,
but not without risk.
We can summarize the main advantages below.
• Advance purchasing. CERN could purchase critical components with a long lead time for
procurement well in advance of placing the contract for magnet manufacture. A typical
example is the low-carbon steel for yoke laminations, purchased at the end of 1997 just at
the second and final approval stage of the project (the main magnet contracts were assigned
in 1999 and 2001).
• Economies due to large-scale purchasing. By purchasing the quantities of materials for all
of the dipoles (and in certain cases for quadrupoles too) better prices were obtained. The
economy applies also to the component qualification process and follow-up checks
(QA/QC). It is difficult to quantify the saving, because it also includes overhead savings that
were also associated with a further step in purchasing, if components had been included in
the main tender: probably, however, it is a considerable fraction (30–40%) of the total cost
of components (∼500 million CHF).
• Technical compliance and uniformity. It is always difficult to able to specify in a clear and
objective way all of the characteristics that are actually needed in a component. We have
examples of the fact that by assigning the components to a magnet manufacturer, although
with an apparently detailed specifications, the magnets ended up with deviations from the
range of acceptability without a real breach of the specifications (quench heater steel from a
dipole manufacturer, beam tube fillet welding from a quadrupole manufacturer, etc.).
• QA/QC on components. It is almost impossible for different purchasers to have the same
QA/QC and a follow-up with the same intensity and criteria of selection. No QA can
substitute for skill and attentive inspectors. With CERN purchasing all components we
ensured a homogeneity of characteristics that is very useful for series production, where
uniformity of characteristics is in practice very important to avoid mistakes and non-
conformities. When we had a deviation from specification we could evaluate it correctly,
without suspicion that the non-conformity was the emerging part of a submerged iceberg,
and we could also accept important non-conformities (for example, in the SC cables, in the
collar profiles, etc.) because we could convince ourselves that such non-conformities would
not adversely affect the final performance.
• Security of supply. By keeping in hand the procurement of components we could anticipate
problems of quality, see above, but also of shortages. For example, we could avoid being
trapped by a big strike in the steel foundry (the unique supplier of all low-carbon steel for
the whole project), by removing all raw iron plates that were in stock there with an
exceptional transportation arrangement (about 400 trucks were organized in a real blitz) to a
temporary store under our control. In another case we helped to start up a new company for
production of critical components (quench heaters). We could feel that production from one
of the two production centres was of increasingly bad quality and we anticipated its
stoppage due to a lack of profit. The new company then became the required second
production centre for quench heaters. All of these actions could have hardly be seen by the
magnet manufacturers and it would have been very surprising to see them acting as we did
538
(a supplier is always protected by the clause of force majeure in the case of a national strike,
or of a sudden stop by a subcontractor).
• In-kind contributions and contract distribution. CERN is an international organization
financed by 20 member states (MS); and for the LHC project various non-member states
(NMS) made an important contribution. By keeping component procurement in-hand CERN
could spread in an optimal way the distribution of the contracts among MS, thus favouring a
balanced return, without however compromising on the technical aspects and without
spending more than the minimum (with a very few exceptions, such as upon a contract for
iron laminations). If global contracts, including tooling and components, had been assigned
to magnet manufacturers, making these a kind of general contractor, this balance would have
been much more difficult to reach and certainly less optimized. On this line alone the
centralization of procurement could make the best use of the in-kind contribution by the
NMS. For example, supplying the Nb–Ti and Nb sheets, from the US, to superconductor
manufacturers is probably an absolute ‘prima’ in the panorama of superconductor
production, impossible without centralization and an assumption of responsibility by CERN.
Other very successful examples are the supply of all polyimide insulation foils and tapes for
all coils, as well as the supply of austenitic steel for the collars for all dipoles: both produced
by Japanese companies and procured by CERN for the various component or magnet
manufacturers in Europe.
The list of the inevitable drawbacks associated with such a strategy is given below.
• Liability for delays. This is the first price to pay: make sure that delivery is on time or at
least ‘just in time’. Many times we really went to the borders of a just-in-time deadline;
however, for the dipoles we managed to avoid any stoppages (which would have cost about
1.5 million CHF per week). We had one case of large delays for the main quads (corrector
magnet supply): despite all actions an extra cost of several million CHF had to be paid, a big
number in absolute terms, but small compared with the total savings. One has to say that as
the final client we could not apply to ourselves, or toward our magnet manufacturers, the
clause of force majeure that contractors have toward the client.
• Additional workload for procurement. The CERN Magnet group organized a
Superconductor section and a Components Centre section, which together with the
purchasing office took care of the procurement for the SC cables and the other components
(a few other components were procured by the Magnet Laboratory and Insulation section
and by the Cryostat and Integration group, later merged with the Magnet group). Interfaces
between CERN and magnet manufacturers became more complex and we had to manage
together the components and the magnet manufacturing schedules. All of these could be
evaluated in an increase of personnel to follow-up the components: about six full-time
equivalents (FTE) × 5 years for SC and about four FTE × 5 years for the Component Centre,
i.e. about 50 FTE-years, excluding the Procurement office and Finance office personnel.
• Technical responsibility and interfaces. Each important component was specified by CERN.
However, with procurement CERN became fully responsible for their technical quality, too,
with respect to assembly problems or performance risk. Clearly distinguishing whether a
failure is due to a component imperfection rather than a non-conforming use of it by the
magnet manufacturer is not trivial, at all, in many cases. Examples of such interfaces were
especially numerous in coil construction: short circuits may have been caused by burrs and
chips on the copper longitudinal wedges or defects of the insulating tapes (both supplied by
CERN) or by metallic inclusions during winding in a non-proper environment.
• QA documentation responsibility. By supplying components CERN also become entangled
in the each company’s QA plan, which had to be fed by the component data. A delay in QA
539
L. ROSSI
documentation would have become a cause of trouble to the company. In a way, this
entangled situation forced CERN to become very efficient with the QA plan to avoid
incurring penalties (which happened in reality, fortunately however at a very minor level).
On the other hand CERN could be strict in imposing the respect of the QA plan by the
magnet manufacturers, having shown its good will in this respect.
• Logistics. The extra workload also meant managing advanced logistics. Not only did CERN
have to work with a coordinated plan, being careful in not favouring/disfavouring any of the
magnet manufacturers, but it had to organize the transport, storage and the QA documents’
availability for a great number and variety of components. In total 150,000 tonnes of
materials were transported all over Europe (and from North America and Asia). For more
than 4 years we had to follow at least five large trucks per day in Europe, having reserves for
any inconvenience, including stolen materials (which also happened, of course, involving
precious components such as the SC cable).
Altogether the success of the strategy was overwhelming. Of course this meant that any lack of
performance of a magnet at acceptance was not necessarily due to bad manufacturing. CERN was not
only responsible for the design and assembly procedure, but also for components, whose behaviour is
critical for the magnets’ performance. We want to underline here that it is clear, from what has
previously been said, that industry has been very cooperative, an attitude to be credited to the quality
of the industry involved and that could be gained and reinforced thanks to the long R&D and
industrialization phases, during which a strong, mutual trust between CERN and industry was built.
The path from reaching 9 T in a single short dipole, which happened in 1987, and from the
success of the first 9 T long prototype, which happened in 1994, to the point where design and
technology was mature enough to contract construction of a large number of 8–8.5 T operational field
full-size dipoles, which happened in 2002, was longer than expected. In comparison, the period of
construction was relatively short, despite the large number of units: the last dipole and last quadrupole
were delivered to CERN on 7 November 2006. The time for installation and commissioning, also
because of the incident of September 2008 [22] that plagued machine start-up, was also longer than
expected, resulting in full operation of the LHC from 30 March 2010. The LHC magnet timeline is
reported in Fig. 21.
Fig 21: LHC magnet time line
540
In this paper we do not mention the specific task of cryostat manufacture. In comparison with
magnet construction the cryostat is mainly a thermo-mechanical construction, with far fewer problems
of tolerance and accuracy. However, the cryostat is the point of the integration of the magnet in the
cryogenic system and the interface with the electrical circuit as well as with the other services. This
complexity needs to be discussed in the early stage of the magnet design, taking the necessary
feedback, and requires a close collaboration between the magnets’ designers and cryostat engineers. In
the end, it is not the magnet that has to work correctly, but the magnet system in its complexity that is
more by far than that of a single magnet.
6 Novel construction technology: Nb3Sn and high-temperature superconductors

So far all existing accelerators have employed Nb–Ti-based magnets, whose construction issues have
been discussed in the previous chapter. However, it is more than 20 years since studies were
performed to increase the performance and to open the way for fields in excess of 10 T. To this aim a
Nb3Sn superconductor is necessary, of course, with all of the technology consequences entailed by this
extremely brittle compound that necessitates a thermal treatment for its formation.
A non-exhaustive list of manufacturing issues when Nb3Sn technology is applied to accelerator
magnets is given below.
i. Use of wind and react technology (first winding the coil with a ductile conductor, then the
entire coil is treated at 650–700°C to react the conductor, forming brittle Nb3Sn) is
mandatory because the coil radius of curvature is 10–50 mm which, with a cable of 1–
1.5 mm thickness means more than 1–2% strain, much beyond the 0.3% safety limit and
even the 0.6–0.7% irreversible degradation threshold.
ii. The wind and react route is very demanding: all coils components (mainly insulation, but
not this alone) must withstand the 650°C reaction temperature, maintaining their properties
and avoiding decomposition into carbon. A lot of information is available about various
types of possible glass fibres that can be used: S-glass or, better, its non-military grade, less
expensive, S2-glass variant, or the similar European standard R-glass. The more common E-
glass could be used, but its annealing point is dangerously close to 650°C and caution must
be taken in the presence of high stresses. Glass tape or braid being a weave, from the
electrical point of view behaves basically like an inter-turn spacer; solid insulation is
actually ensured by the resin impregnation (usually under vacuum), which is a critical
operation for electric, mechanical and thermal properties. Glass-epoxy can be reinforced
with mica-glass composite tape. Choice of the sizing agent for the glass is also critical to
avoid excessive carbon formation. A mild grey coloration of the reacted coil is generally
acceptable (carbon increases its resistance by a factor ∼1000 when cooled from 300 to 2 K).
iii. Coils and their tooling should allow for expansion–contraction over almost 1000°C (from
300 to 970 K and then down to 4 K) while keeping an accuracy of 30–50 μm. The SC cable
after Nb3Sn formation at 650°C shows a completely different behaviour in terms of thermal
expansion and this must be taken into account to avoid over-stressing. The actual behaviour
is strongly dependent on the copper content and on the particular process used to obtain
Nb3Sn: bronze route, classical internal-tin-diffusion, rod-restacking process, powder-in-tube,
etc.
iv. The degradation of the critical parameters of Nb3Sn due to strong transverse stress is almost
unavoidable and a large effort in conductor development, coil technology and magnet
structural design is necessary just to mitigate this effect. Of course this is not easy because
Nb3Sn is used for high fields, B > 10 T, when the force and stress build-up become
considerable.
541
L. ROSSI
v. The field quality (and sometimes quench performance, too) is affected by large filaments,
very large Jc at low field, low conductivity in the stabilizer, large intra-cable coupling
currents, etc. These effects requires mitigation measures such as, for example, a high-
resistance core inside the Rutherford cable, in the form of stainless steel tape, which affects
the mechanical and stability properties of the conductor and winding.
Use of Nb3Sn for accelerator magnets is definitely not straightforward. The Conductor
Development Program of DOE (CDP-DOE) launched in the US in 1998 [23] was aimed at the
following specific goals for building an accelerator much beyond the existing Nb3Sn then available
(1995), for fusion and high-field solenoid programs: i) Jc of at least 3000 A/mm2 at 12 T, 4.2 K was
the primary goal, which was attained quite soon; ii) then an effective filament size of 30–50 μm (for
field quality) and high Residual Resistivity Ratio (RRR) in the stabilizer became the main goals and
this is still work in progress.
Considering construction technology, a number of companies are able to produce Nb3Sn
solenoids (for high-field laboratory magnets and NMR spectroscopy); a few specialized companies are
accomplishing the task of manufacturing the huge Nb3Sn coils for the ITER project; however, no
company today possesses the technology for manufacturing Nb3Sn accelerator magnets. It is one of the
challenges for the next project, High Luminosity LHC, to be able to set a mature and solid Nb3Sn
technology for accelerator magnets while starting the necessary technology transfer to industry.
In Fig. 22 a mosaic of the tooling and coil construction at various stages is shown for the Nb3Sn
11 T dipole project (Fermilab–CERN collaboration) [24]. The 11 T dipole is designed to use a
‘classical’ collar structure for force containment. In the context of the large-aperture quadrupole R&D
for LHC Accelerator R&D Program (LARP), a US program oriented to the High Luminosity LHC
upgrade [25], and for their own internal high-field dipole program, Lawrence Berkeley National
Laboratory (LBNL) designers have devised a new concept called ‘key and bladder’. Here, collar
function is reduced to a minimum: they are very thin and used just for coil assembly.
The restraining forces are generated by an external solid shrinking cylinder, via inflation of
high-pressure bladders. The insertion of a suitable key maintains the pre-stress when the bladders are
de-pressurized and finally removed. The restraining forces are smaller than those usually applied by a
collar, but they increase during cool-down because the aluminium cylinder contracts more than the
Nb3Sn coils. More details can be found in [13, 26].
Clearly this technology can also be applied to Nb–Ti coils. However, it is particularly
interesting for Nb3Sn magnets, because coils wound with this conductor have a higher (two to three
times) elastic modulus than coils made out of Nb–Ti; in addition, Nb3Sn coils are also less
dimensionally accurate. A collar structure controls the coil size, and stress comes as a consequence
(assuming collars to be of infinitely rigid structure), while a key and bladder structure controls the
force, and coil size comes as a consequence.
Owing to the higher coil rigidity and reduced dimensional accuracy, collar structures for Nb3Sn
windings may result in excessively high pre-stress. Such a dangerous situation is much less probable
in the case of a key and bladder structure. However, this is a new technology that has still to be
demonstrated in long, operational quality, magnets.
542
Fig. 22: Quenches of LHC dipole 1002 at virgin training curve and after the first and second thermal cycles
(indicated by vertical lines).
7 Testing
A SC magnet that has not yet been tested is worth virtually nothing. The scope of magnet testing
radically changes between the R&D phase and the construction phase. During R&D the scope of the
test is to assess the limit of the design and the merits of the various choices, and assess the best
technology in terms of performance and reliability.
In the construction phase, and particularly with a massive production series, the test has two
main goals: i) to serve as ultimate QA tools, to establish whether the whole magnet system is
performing as specified: quench performance is important in this respect but it is ‘only’ one of the
many important aspects of the system’s performance; ii) to serve as an acceptance test, with all of its
financial and legal implications that a laboratory has toward the manufacturer.
Here we will briefly discuss tests only in relation to validation of design and construction: a
review of the LHC magnet test can be found in [27].
7.1 Test for series production

Magnet testing in operating conditions is the ultimate QA tools because not only is it necessary to
ascertain whether the work was done well: it is also necessary to intercept various mistakes, such as
defects in components, fabrication errors and design faults or weaknesses that cannot be evidenced in
any way other than the final test in full operating conditions. Examples of such weak points are: a too
high resistance in the SC joints; premature quenches due to mis-positioning of the SC cable; wear of
insulation because of cool-down induced movements, and so on. The importance of testing as the
ultimate QA measure is evidenced by the fact that it may, and usually does, trigger corrective actions
in the manufacturing chain in order to remove, or at least mitigate, the mistake and other sources of
non-conformity. Therefore, testing in operating conditions should really be part of the production
cycle. The promptness of the feedback must be proportional to the production rate: taking the LHC as
an example, when each company was at the maximum production rate of three dipoles per week, the
3-month contractual delays between magnet delivery and testing meant that 40 dipoles were produced
between the eventual discovery of a systematic defect and the implementation of the corrective actions
543
L. ROSSI
in the production chain. However, 40 dipoles are almost 10% of the whole production: readiness in
testing is one of the best ways to avoid huge extra costs and delays in production.
The result of the test as an acceptance test is to pay the manufacturing company upon positive
results, and to release it from any further responsibility, except of course for hidden defects, for which
should be agreed a clause of responsibility lasting for a few years, enforced by a solid bank guarantee.
A positive acceptance test means that the magnet is qualified for operation. The temptation to
understand the design limits, by pushing magnets beyond what is needed for operation, should be
resisted because it is mixing acceptance testing and the R&D phase. Once a magnet is accepted by the
group/team clearly in charge of it, a further analysis of the magnet’s properties, to verify the quality of
the magnet and decide the proper installation slot, may be demanded by a suitable body. For the LHC
this was the Magnet Evaluation Board, which was off-line from the magnet acceptance chain.
7.2 Testing during the magnet research and development phase

In the R&D phase the magnet is pushed beyond the operating conditions, to learn about design choices
and actual performance limits and, therefore, operational margins.
Before describing the test it is useful to explain that accelerator magnets, when pushed near the
critical current limit and/or the force-retaining structural limit, have a typical behaviour called
‘training’. Quench happens at a level of current that is lower, and in certain cases much lower, than the
maximum theoretical current, which is the critical current along the load line, also called the short
sample limit because it is inferred from measurements of a short cable sample cut from the extremities
of the unit length. Assessing training and limits is one major goal of testing an accelerator magnet.
The test starts with a cool-down that must be carried out in a well-thought-out manner, with an
acceptable temperature gradient to avoid excessive thermal stresses. The electrical insulation is then
checked: this check has to be done at each phase, starting with reception from the manufacturer, at
room temperature. Electrical checks are critical to assessing magnet integrity and must be really well
thought out, at a voltage that should slightly decrease between successive tests. The protection circuit
must then be tested at a low current. Finally, the power test can proceed, by raising the current with a
ramp rate equal to or similar to that in operation.
Usually the magnet undergoes a sudden transition (quench), losing SC status at a value of
current well below the design one. The current is quickly removed (100 ms), however the magnet
heats up in 1 or 2 seconds, reaching ambient temperature in a hot spot. The magnet is cooled again and
re-energized until another quench is recorded at a value of current usually higher than the previous
one: it looks as though the magnet is being ‘trained’ toward higher values. The sequences of quenches
are called training curves, and a typical sequence is reported in Fig. 22. After the training the magnet
reaches flat values. The eventual difference between the target value and the flat value is called
degradation (more than 5–10% would be considered a serious fault of design or fabrication). Usually
the flat value is above the operating conditions, as in Fig. 22, since the magnet is actually designed to
have some margin, to account for shortfalls and degradation during series construction. Referring to
Fig. 22, it can be seen that at the LHC we defined an operational target (nominal field value, 86% of
the short sample limit) and also an ultimate field value (93% of the short sample limit), corresponding
to the limit of mechanical design, bus bar system, connections, etc.
One further important performance parameter is the so-called ‘memory’. After the training
curve, the magnet is completely warmed up (the warm-up induced by quench alone is never
complete). The magnet is cooled down again and is ready to repeat a number of powering-to-quench
cycles. This is the second training curve. One can then repeat the thermal cycle and have a third
training curve. It can be seen from Fig. 22 that this dipole magnet, actually one of the best of the pre-
series, is well qualified for its operational field and is marginally good for its ultimate field: 9 T is
reached and also passed at the second and third training curves; however, some loss of memory
544
happens around its ultimate value. The implications of commissioning accelerator operation go
beyond the scope of this paper. Here it suffices to say that some of the LHC dipoles exhibit a small
loss of memory around the nominal value. This will probably imply that the nominal value in the LHC
(8.33 T, corresponding to a 7 TeV/beam) will be achieved later in the future because, even if only 30%
of the LHC shows some loss of memory, this means that some 400 quenches will be required in the
tunnel during re-commissioning of the LHC in 2014–2015. It may well happen that, if the number of
re-training quenches is greater than our expectation, the LHC main dipoles may be limited to working
at around 80–83% of their short sample value, as in all previous accelerators. This would mean
limiting the beam energy to 6.5–6.8 TeV/beam.
Another important goal of the test is to measure the field and the harmonic content of the
magnetic field in the region used by the beam. This property is usually called field quality. A good
part is usually measured at room temperature, i.e. at a very low current and without SC effects, as
mentioned in the Section 5.1. Usually these measurements give a very good prediction of the field
quality due to the geometry of the conductors. However, measurements in operational conditions, at
cold, need to be done: i) to assess the exact form of the warm–cold correlation for geometric field
quality; and ii) to assess the behaviour of the SC effects. For the LHC a large number of studies
allowed changing the strategy for magnetic measurements in an advantageous way: also, due to
pressure posed by a delay in testing, the cold field measurements have been reduced from an initial
100% down to about 15% (see [18, 28]). It is worth saying that the excellent field quality of the
magnets, and complete knowledge of its operation, are some of the most important ingredients of the
great success of the LHC in its first years of operation.
8 Conclusions
Manufacturing a SC magnet of accelerator quality is a unique challenge. As good as the design may
be, a magnet that is manufactured badly with low-quality or bad components and tooling will perform
poorly, impeding a correct evaluation of the design merits. In this paper we have tried to summarize
the main principles of accelerator magnet construction, and to recollect the main lesson gathered from
the LHC adventure, underlining the close tie among design, manufacturing process, tooling and
testing. Superconductivity calls for a total quality process: different from room-temperature
technology, even an apparently minor detail can have detrimental effects on the magnet’s
performance. In this respect the engineering design and the construction and testing represent a true
intellectual challenge, especially in large projects where resources are limited and great pressure is put
on schedules. The main lessons we can draw from LHC magnet construction can be summarized in
two points:
i. one must not compromise on quality and on controls as any hidden defects will show up, at
the worst possible time;
ii. even more importantly, especially when resources are limited as is the case in all practical
projects, ‘the best is the enemy of the good’; when a solution is sufficient to the scope, one
has to resist any improvements since, in a complex system such as a SC magnet, a change in
one part will likely affect, sometimes in a subtle way, the whole system.
References
[1] J.E. Kunzler, IEEE Trans. Magn. 23 (1987) 396.
[2] M.A. Green, GESSS Machine Design Committee Reports, Report KFK 1764,
Kernforschungs Zentrum Karlsruhe (1972).
[3] J. Billan et al., IEEE Trans. Nucl. Sci. 26. (1979) 3179.
545
L. ROSSI
[4] P.F. Dahl et al., Superconducting magnet models for Isabelle, Proc. PAC, 1973, p. 688,
published by IEEE and available on Jacow.org
[5] H.T. Edwards, Ann. Rev. Nucl. Part. Sci. 35 (1985) 605.
[6] A. Tollestrup and E. Todesco, Rev. Accel. Sci. Technol. 11 (2008) 185.
[7] S. Wolff, IEEE Trans. Magn. 24 (1988) 719.
[8] M. Anerella et al., Nucl. Instrum. Methods A499 (2003) 280.
[9] J. Strait et al., IEEE Trans. Magn. 25 (1989) 1455.
[10] R. Perin, IEEE Trans. Magn. 27 (1991) 1735.
[11] J.P. Blewett, 200-GeV intersecting storage accelerator, Proc. 8th Int. Conf. on High Energy
Accelerators, Switzerland, 1971, Eds. M.H. Blewett and N. Vogt-Nielsen (CERN,
Geneva), p. 501. http://lss.fnal.gov/conf/C710920/
[12] G. Baccaglioni et al., IEEE Trans. Magn. 30 (1994) 1827.
[13] L. Rossi and L.L. Bottura, Rev. Accel. Sci. Technol. 5 (2012) 51.
[14] G. Spigo, G. Cunningham, G. Goodzeit, D. Orrell, J. Turner and J. Jayakumar, Design and
performance of a new 50 mm quadrupole magnet for the SSC, 5th IISSC, Supercollider 5,
San Francisco, 1993 (Plenum, New York, 1994), p. 647.
[15] G. Ambrosio, F. Ametrano, G. Bellomo, F. Broggi, L. Rossi and G. Volpini, Preliminary
proposal of a Nb3Sn quadrupole model for the low-β insertions of the LHC, INFN Report
N. INFN/TC-95/25 (1995).
[16] Y. Ajima et al., Nucl. Instrum. Methods Phys. Res. A550 (2005) 499.
[17] Technical specification for the supply of 1158 cold masses of the superconducting dipole
magnets for the Large Hadron Collider, CERN IT-2997/LHC/LHC, LHC Project document
No. LHC-MB-CI-0006, EDMS 312601.
[18] E. Todesco, Report on field quality in the main LHC dipoles, CERN Report, EDMS
807803 (2006).
[19] C. Vollinger and E. Todesco, IEEE Trans. Appl. Supercond. 16 (2006) 204.
[20] P. Fessia, F. Regis and L. Rossi, IEEE Trans. Appl. Supercond. 17 (2007) 1101.
[21] E. Acerbi, M. Bona, D. Leroy, R. Perin and L. Rossi, State of the construction of two INFN
full length superconducting dipole prototype magnets for the Large Hadron Collider
(LHC), Proc. EPAC 92, Berlin, 1992 (Edition Frontiér, 1992), p. 1420.,
http://www.jacow.org
[22] L. Rossi, Supercond. Sci. Technol. 23 (2010) 034001.
[23] R. Scanlan and D. Dietderich, IEEE Trans. Appl. Supercond. 13 (2003) 15361.
[24] A. Zlobin et al., Status of a single-aperture 11 T Nb3Sn demonstrator dipole for LHC
upgrades, Int. Particle Accelerator Conference, New Orleans, USA, (2012), p. 1098,
http://www.jacow.org
[25] G. Ambrosio et al., IEEE Trans. Appl. Supercond. 21 (2001) 1858.
[26] P.A. Bish et al., A new support structure for high field magnets, Lawrence Berkeley
National Laboratory report, LBNL-47796 SC-MAG, p. 738.
[27] A. Siemko and P. Pugnat, IEEE Trans. Appl. Supercond. 18 (2008) 126.
[28] L. Bottura, V. Granata, S. Fartoukh and E. Todesco, A strategy for sampling the field
quality of the LHC dipoles, Proc. EPAC, Lucerne, 2006, p. 1606., http://www.jacow.org
546
Current Leads, Links and Buses
A. Ballarino 1
Abstract
Electrical transfer from a room temperature power source to a
superconducting system is done via conventional or superconducting current
leads and superconducting buses or links. The principles of optimization of
these devices are presented, with emphasis on the cryogenic, electrical, and
superconductor related aspects that drive choices for a system.
Keywords: current leads, superconductors, high-temperature

superconductors, powering, superconducting magnets, accelerators.
1 Introduction
The powering of a superconducting system requires transfer of current from ambient temperature,
where the current is generated, to the cryogenic environment. This electrical transmission, which takes
place inside a cryostat, is performed by the current leads, i.e. devices that transport current in a
significant gradient of temperature and represent, therefore, one of the major sources of heat inleak
into the cryostat.
Designing cryogenic devices requires precise knowledge of material properties at the operating
temperatures, some of which may vary by orders of magnitude between ambient and cryogenic
conditions. Thermal conductivity in metals and alloys spans across several orders of magnitude and is
dominated by thermal conduction by electrons, scattered by lattice phonons and imperfections.
Electrical resistivity also varies by several orders of magnitude depending upon the type of material,
degree of alloying, and purity. It decreases with temperature and it shows a typical plateau
corresponding to a residual resistivity value due to electron scattering by impurities and lattice
imperfections. Metals follow the Wiedemann–Franz–Lorenz (WFL) law, which establishes an inverse
proportionality between thermal conductivity (k) and electrical resistivity (ρ) at a given absolute
temperature (T):
𝑘(𝑇)𝜌(𝑇) = 𝐿0 𝑇 (1)
where L0 is a constant, the Lorenz number, which was calculated by Sommerfeld in 1927 using Fermi–
Dirac statistics:
π2 𝑘B 2
𝐿0 = � � = 2.45 ∙ 10−8 W Ω K −2 (2)
3 𝑒
where kB is Boltzmann’s constant and e the electron charge [1].

Good electrical conductors are also good thermal conductors, and the transfer of electrical
current in metals that operate in a gradient of temperature, for which one needs to minimize both
ohmic losses and thermal conduction, is bound by the existence of a minimum heat inleak into the
cryogenic environment. For current leads operating between ambient and liquid helium temperature,
this minimum heat inleak (QC,min) per unit current is about 47 W kA−1 for a conduction-cooled lead and
1.1 W kA−1 for a gas-cooled lead operated in self-sustained mode (self-cooled lead). This value is
independent of material properties, and it represents an optimum performance in terms of minimum
1
amalia.ballarino@cern.ch
A. BALLARINO
heat inleak associated with the transmission of a given current inside a cryostat. For a cryogenic
system, minimum heat inleak means minimum liquid cryogen consumption or minimum refrigeration
power load. The achievement of this optimum performance is the main objective of the design of a
current lead.
The WFL law is a useful correlation that can be used to derive the thermal conductivity of
metals as a function of temperature from the electrical resistivity, the latter being easier to measure
with precision. However, it is known that highly-alloyed materials depart from the WFL law, and in
many metals at low temperature the Lorenz number deviates from the theoretical Sommerfeld value
[2]. This means that for the optimization of a current lead one should preferably use the measured
temperature-dependent material properties instead of the WFL relationship. The value of the minimum
heat inleak only changes by a few per cent, but the use of the material properties enables correct
definition of the specific geometry that enables the achievement of the optimum performance.
Before the discovery of High-Temperature Superconductors (HTS), current leads were made
from metals or alloys. Design optimization in this case is mainly associated with the calculation of the
optimum geometry of the lead for a selected material and for a given current to be transferred.
Discovery of high-temperature superconductivity and production of technical HTS opened an
innovative way to improve the performance of conventional current leads by incorporating in them
materials that escape the WFL law in view of their low thermal conductivity and ideal zero resistivity
in the superconducting state. HTS current leads, also called binary leads, enable the reduction of the
heat inleak into the liquid helium bath by a factor of up to ten with respect to the minimum obtainable
with conventional leads [3]. The corresponding total saving in cooling power is about 30% [3].
Studies on the possibility of integrating HTS materials in leads started in the 1990s [4–6], when
prototypes were also made [7–10], and the first large-scale application of HTS current leads was with
the Large Hadron Collider (LHC) machine, where more than 1000 HTS leads are used to feed the
superconducting magnet circuits of the accelerator [11]. Thanks also to the developments carried out
for this large-scale application, HTS leads have today become a recognized standard technology. The
International Thermonuclear Experimental Reactor (ITER) will use HTS current leads for feeding the
superconducting coils of the tokamak being built at Cadarache, France [12].
2 Conventional current leads

Depending on the cooling mode, conventional current leads can be classified into conduction-cooled
or gas-cooled (Fig. 1). In the first case, the lead operates in vacuum, while in the second case there is
convective heat transfer. If the heat conducted by the lead into the liquid cryogen bath generates the
gas that is then used for cooling the lead itself, the gas-cooled lead is termed self-cooled. In all cases,
the optimization requires the solution of the steady-state heat balance equation that includes energy
generation due to the ohmic heating associated with the transfer of current.
(a) (b) (c)
Fig. 1: Cooling scheme for a conventional current lead: (a) conduction-cooled current lead without heat sinks;
(b) conduction-cooled current lead with intermediate heat sinks; (c) self-cooled current lead.
548
C URRENT L EADS , L INKS AND B USES
2.1 Conduction-cooled conventional current leads

Under the assumption that the total length of the lead (L) is much larger than its cross-section (A), the
mono-dimensional heat balance equation can be considered. The minimization of the heat inleak at the
cold end of a conventional lead (QC) is obtained by imposing a zero slope in the temperature profile
(T(x)) at ambient temperature (x = L), which corresponds to zero thermal conduction at the warm end
of the lead.
The minimum heat inleak (QC,min) per unit current (I) is almost independent of material
properties. This can be easily demonstrated for a conduction-cooled lead under the assumption that it
follows the WFL law. From the solution of the mono-dimensional steady-state heat balance equation:
d d𝑇 1
d𝑇
�𝑘(𝑇)𝐴 � +
d𝑥
ρ(𝑇) 𝐴 𝐼 2 = 0 (3)
introducing:
d𝑇
𝑄(𝑇) = 𝑘(𝑇)𝐴 (4)
d𝑥
one can calculate QC per unit current transferred:
𝑄𝐶
𝐼
= �𝑄𝐻2 + 𝐿0 𝐼 2 (𝑇𝐻2 − 𝑇𝐶2 ) , (5)
which is minimized when the heat conducted at the warm end of the lead (QH) is zero:
𝑄𝐶,min
𝐼
= �𝐿0 𝐼 2 (𝑇𝐻2 − 𝑇𝐶2 ) . (6)
The geometry that enables the achievement of optimum performance is dependent upon
material properties and it can be obtained from:
𝐿 𝑇 𝑘(𝑇)
� � = ∫𝑇 𝐻 . (7)
𝐴 opt 𝐶 2
�(𝑄𝐶,min −𝐿0 𝐼2 �𝑇 2 −𝑇𝐶2 �)
From Eq. (7) it appears that, for a given material, the optimum thermal performance at a given
current corresponds to a constant length-to-cross-section ratio. The optimum geometry of a lead is
often conveniently reported in terms of shape factor (SF), which applies to all optimized leads made
from a specific material, regardless of current:
𝐼𝐿
SF = � � . (8)
𝐴
The mono-dimensional temperature profile T(x) of an optimized conduction-cooled lead can be

obtained from numerical integration of Eq. (7) for a given operating current and a selected geometry,
i.e. length and corresponding cross-section. The total length of a current lead is often imposed by the
requirements from integration inside a cryostat, and the cross-section is derived from Eq. (8).
It should be noted that while QC,min at the current for which the lead is optimized does not
depend on material properties, the heat conducted by the lead when it is either in stand-by mode
(I = 0 A) or it operates at a current lower than nominal does depend on material properties. Leads
made from materials with high thermal conductivity, e.g. copper with Residual Resistivity Ratio
(RRR) above 100, in stand-by conditions conduct, inside the cryogen bath, significantly higher heat
549
A. BALLARINO
than those made from materials with low thermal conductivity, e.g. phosphorus deoxidized copper
(RRR < 10) or brass. This is an important input for the selection of the material, in particular for
systems that are not continuously powered. Considerations of stability in transient conditions also
favour the selection of materials with lower thermal conductivity, at the cost of a larger cross-section.
An optimized conduction-cooled current lead operating between room temperature and the
liquid helium bath has a heat inleak at 4.2 K of about 47 W kA−1. Such a high thermal load limits the
application of this type of lead to systems operated at low currents. Improvement of thermal
performance can be obtained by additional extraction of heat at higher temperatures using intermediate
heat sinks (Fig. 1(b)), which are provided either by cold stages of cryo-coolers or by dedicated gas-
cooled heat exchangers.
Conduction-cooled current leads with intermediate heat sinks were developed for the powering
of the LHC orbit corrector magnets. These circuits operate at a maximum current of 60 A. A hybrid
conductor consisting of a tapered copper-plated brass rod, electrically insulated and enclosed in a
stainless steel tube, transfers the current across the technical vacuum from room temperature to the
1.9 K superfluid helium bath where the magnets are operated [13]. Heat sinking is provided by two
heat exchangers cooled by helium gas available in the LHC cryogenic system at about 50 K and 20 K,
respectively (Fig. 2). With this design, the heat inleak at 60 A is reduced to about 0.19 W, to be
compared with the 2.8 W that would be conducted by a lead operating at the same current and with no
intermediate heat sinking.
Fig. 2: LHC orbit corrector conduction-cooled current leads (path indicated by arrow) integrated inside the LHC
cryostat. The leads operate in the cryostat vacuum insulation between room temperature (RT) and 1.9 K. They
are thermalized against two heat sinks cooled by He gas, available in the LHC cryogenic system, which are
maintained at about 50 K and 20 K.
Conduction-cooled current leads are often chosen for their simplicity, provided the cryogenic
system can afford the associated thermal losses. They do not require an active control of flow and they
can have, for low current ratings, a compact design that incorporates mechanical flexibility to enable
easy integration in a cryostat [13]. Drawbacks are the higher heat inleak and, for systems that include
intermediate heat sinks, the difficulty of providing a good thermalization, in the vacuum insulation of
the cryostat, at the location where the heat is extracted from the electrically insulated conductor.
550
2.2 Self-cooled conventional current leads

The optimization of a self-cooled conventional current lead requires the solution of a system of
differential equations expressing the heat balance between the conductor and the coolant. In a steady
state and in the mono-dimensional case, neglecting thermal conduction in the gas, the system is:
d d𝑇 ρ(𝑇)𝐼2
�𝑘(𝑇)𝐴 � =− + 𝑃ℎ(𝑇)(𝑇 − ϑ) ,
d𝑥 d𝑥 𝐴
dϑ
𝑚𝑐p (ϑ) d𝑥 = 𝑃ℎ(𝑇)(𝑇 − ϑ) (9)
where T is temperature of the conductor, k and ρ are the conductor thermal conductivity and electrical
resistivity, A and P are the conductor cross-section and wetted perimeter, ϑ is the temperature of the
gas, cp is the specific heat of the gas, m is the gas mass flow rate, and h is the heat exchange coefficient
between conductor and coolant.
In self-cooled conditions, the heat conducted into the liquid bath (x = 0) generates the mass flow
rate needed for cooling the lead:
𝑘𝐴 d𝑇
𝑚= (10)
𝐶𝐿 d𝑥
where CL is the latent heat of vaporization of the liquid cryogen.

The solution of Eq. (9) requires the knowledge of the current lead geometry for the definition of
P and h. A first optimization of the lead can be made by assuming a perfect heat exchange between
conductor and gas (h→∞). With this assumption, the heat balance equation becomes:
d d𝑇 ρ(𝑇)𝐼2 d𝑇
�𝑘(𝑇)𝐴 � + − 𝑚𝐶p (𝑇) =0. (11)
d𝑥 d𝑥 𝐴 d𝑥
Taking as boundary conditions the temperatures at the warm end (x = L, T = TH) and at the cold
(x = 0, T = TC) end of the lead and applying Eq. (10), it is possible to calculate from Eq. (11) the
minimum heat inleak per unit current [14] and the corresponding temperature profile characterized by
a zero slope at the warm end of the conductor (x = L). For a self-cooled lead operating between room
temperature and liquid helium, the minimum heat inleak is about 1.1 W kA−1. As for a conduction-
cooled lead, this optimum performance is almost independent of material properties, and the geometry
of the lead is defined by an optimum SF. Figure 3 reports the heat conducted at 4.2 K by two self-
cooled leads, one made from stainless steel and the other from pure copper, as a function of the SF
[15]. The minimum heat inleak is about 1.1 W kA−1 in both cases, while the optimum SF differs
because of different material properties. Moving away from the optimum SF implies always higher
heat inleaks and higher mass flow rates for cooling. Shape factors smaller than the optimum
correspond to leads that operate in an over-cooled regime – this is the case, for instance for leads that
have a too-large cross-section for a given current and length – while an SF smaller than optimum
corresponds to leads that are sub-cooled. This latter case is critical in that leads having a too-small
cross-section for a given current and length suffer thermal runaway – steady-state operation with local
temperatures higher than room temperature is only possible within a small range of geometries.
Since the SF depends on the properties of the material, use should be made of the real conductor
properties as a function of temperature over the full range of operation. It is clear from Fig. 3 that
leads made from pure metals are more sensitive to variations of geometry around the optimum. The
final optimization of a lead has to be done via the solution of Eq. (9) taking into account the detailed
geometry selected in the design phase.
Table 1 summarizes the minimum heat inleak per unit current of conventional conduction-
cooled and gas-cooled current leads operated between room temperature and either liquid helium or
liquid nitrogen temperature. In Fig. 3 is reported the temperature profile calculated and measured on
551
A. BALLARINO
self-cooled current leads optimized for transferring 18 kA between room temperature and the liquid
helium bath [15].
Fig. 3: Dependence of heat inleak on shape factor for two gas-cooled (GC) conventional current leads made
from stainless steel (SS) and copper (Cu ETP) [15]. The leads transfer current to a liquid helium bath. The
minimum heat inleak is the same while the optimum SF differs because of different material properties.
Table 1: Minimum heat inleak (W kA−1) of conventional current leads
Type of lead QC,min (4.2 K) QC,min (77 K)

Conduction-cooled 47 45
a
Gas-cooled 1.1 23b
a
Helium gas cooling; bnitrogen gas cooling
3 HTS current leads

A HTS current lead consists of a resistive part, made from a normal conductor, and a superconducting
part made from a high-temperature superconducting material (Fig. 5). The resistive part operates
between room temperature (TH) and the maximum operating temperature of the HTS (THTS), while the
HTS part provides the electrical transfer in the temperature region from THTS to the cryogenic
environment (TC). The use of HTS materials offers the potential for reducing the refrigeration
requirements to values significantly lower than those achievable with conventional optimized current
leads. The total gain in terms of reduction in the exergetic cost of the refrigeration depends on the final
design of the HTS current lead. The latter is influenced by the availability of cryogens for the choice
of the cooling mode, and by the electrical requirements of the superconducting system, which include
the definition of protection needs and the conditions to be met in transient operation.
The most efficient cooling schemes for HTS leads are based on independent cooling of the
resistive and HTS parts. For a self-cooled HTS current lead relying entirely on cooling via a mass flow
generated by the heat conducted by the lead itself into a helium bath, the reduction in exergetic cost of
the refrigeration with respect to a conventional current lead is limited to about 15%. This is because
the cooling is driven by the resistive part, which determines the helium boil-off requirements [4],
while the HTS part is over-cooled. Self-cooling using a nitrogen bath or forced flow cooling by helium
gas are both efficient options for cooling the resistive part of an HTS lead. The reduction in total
cooling power obtained by adopting these cooling schemes is up to about 30% of a conventional self-
cooled lead, while the reduction in heat conducted into the helium bath is a factor of up to 10 [11].
Conduction cooling is limited to applications that usually rely on cryo-coolers and run at low currents
(I ≤ 1.5 kA). In all cases, the HTS part of the lead can be either conduction-cooled or gas-cooled.
552
Fig. 4: Temperature measured and calculated on a CERN self-cooled current lead, 1.2 m long, optimized and
operated at 18 kA [15]. The upper curve indicates the qualitative change in the optimum temperature profile
corresponding to a change in material properties – Cu with lower RRR.
The optimization of a HTS current lead consists of two parts: the optimization of the resistive
part and of the superconducting part (Fig. 5).
(a) (b) (c)
Fig. 5: Cooling scheme of: (a) HTS conduction-cooled current lead; (b) HTS self-cooled current lead; and (c)
HTS gas-cooled current lead. THTS is the temperature at the top end of the HTS part and THe is the inlet
temperature of the gas cooling the resistive part of the lead.
3.1 HTS current leads: resistive part

The optimization of the resistive part of an HTS lead requires the solution of Eq. (9) according to the
description in Section 2. If self-cooling is provided by nitrogen boil-off, optimization is identical to
that of a conventional self-cooled lead and the performance is as reported in Table 1.
Cooling via forced flow of helium gas opens possibilities for different operating temperature
ranges and optimization schemes [11]. As opposed to self-cooled conventional current leads, where
the helium mass flow rate is bound to the heat load by Eq. (10) and where the bottom end of the lead is
at the same temperature of the liquid cryogen, in the gas-cooled resistive part of a HTS lead THTS can
assume a wide range of operating values. The latter depend on the inlet temperature of the coolant
(THe) and on the maximum operating temperature of the superconductor. In the case of cooling with
helium gas, THTS can assume any value in the range up to about 85 K. It can be shown [15] that also in
this case the minimum mass flow rate required for the cooling of the conductor is the one that gives a
temperature profile characterized by a zero slope at room temperature. As for conventional self-cooled
current leads, for a given THe there is an optimum SF. The SF depends both on material properties and
on THTS.
553
A. BALLARINO
Figure 5 maps the optimum performance of the helium gas-cooled resistive part of a HTS lead
optimized for different operating conditions [11]. The minimum mass flow rate per unit current is
calculated for a wide range of inlet temperatures of the coolant and of operating temperatures of the
HTS. The performance and the operating conditions of the LHC and of the ITER HTS current leads
are also reported. In the LHC, cooling is done via helium gas entering at about 20 K, and the top end
of the HTS is maintained at 50 K [9]. In the case of the ITER HTS leads, cooling is done via helium
gas entering at about 50 K, and the top end of the HTS is maintained at 65 K [12]. The inlet
temperature of the gas depends on the cryogenic system. The operating temperature of the HTS is the
result of an optimization that takes into account that, for a given THe, higher THTS requires lower mass
flow rates but a larger cross-section of superconductor. There is a trade-off, which is a system choice.
Figure 7 reports the SF for each of the operating points in Fig. 6. The optimum geometry of the
resistive part of the lead is calculated using the material properties of copper with RRR = 70. The SF
depends on material properties and on operating conditions (THTS and THe).
Fig. 6: Minimum mass flow rate per unit current (mI) required for cooling the resistive part of a HTS lead as a
function of the operating temperature of the HTS (THTS) and of the inlet temperature of the helium gas (THe).
Operating conditions and optimum performance of the resistive part of the LHC and ITER HTS current leads are
also reported.
Fig. 7: Shape factor of the gas-cooled resistive part of a HTS lead as a function of the maximum operating
temperature of the HTS (THTS) and of the inlet temperature of the helium gas (THe). The geometry is calculated
using the material properties of copper with RRR = 70.
The resistive parts of HTS leads are heat exchangers whose design can be based on the use of a
large number of parallel wires or small tubes, or foil, soft-soldered to the copper terminals. However,
for high current ratings, the alternative option based on a helical or meander flow of coolant presents
554
several important advantages, namely: (i) reliable and consistent low resistance joints to the terminals
can be easily made using standard techniques like electron-beam welding; (ii) the mass of the fins that
creates the heat exchange area adds to the mass per unit length of the lead without increasing the heat
conduction, thus increasing the thermal inertia of the lead, and time to thermal runaway, should the
coolant flow be interrupted; and (iii) the heat exchanger is a well-defined component that can be
readily specified and manufactured. For these reasons, recent large-scale applications have adopted the
latter design for series production of current leads [11, 12].
The final optimization of the resistive part of a current lead designed for transferring high
currents requires 2D and possibly 3D finite element calculations for the detailed definition of the
geometry and the modelling of the thermal, electrical, and fluid dynamic performance [16].
3.2 HTS current leads: superconducting part

The thermal load (QC) of the superconducting part of a lead is given by the sum of the thermal
conduction and of the power dissipation of the electrical joint at the cold end of the lead (TC), where
the HTS element is connected to a bus transferring the current to the superconducting device. Thermal
conduction depends on the temperature gradient, on the conductor cross-section, on the length of the
HTS element, and on the thermal conductivity of the material. THTS has to be chosen in accordance
with Fig. 6. The final operating value, selected in the admitted temperature range, has to take into
account the increased superconductor cross-section at higher temperatures, with consequent higher
heat load and cost of the device. The length of the HTS part is for practical and economic reasons
limited to a maximum of about 0.5 m [17, 12]. The HTS part can be conduction-cooled or self-cooled.
Cooling with gas enables a significant reduction of the heat load into the liquid cryogen [9, 15].
HTS conductors for application in current leads must have low thermal conductivity. Protection
requirements in case of resistive transition of the HTS impose the need for a shunt, parallel to the
superconductor, able to transfer the current, after detection of a quench, during the time needed for
discharging the electrical circuit. For this reason, the use of HTS leads may not offer any advantage in
superconducting circuits that have very long time constants and that cannot provide a fast discharge in
case of resistive transition of the HTS element (e.g. large detector magnets).
For protection reasons, HTS conductors with a metal matrix are preferred to bulk conductors, in
particular for current leads transferring high currents. Bi-2223 has to date been the preferred choice for
large-scale applications. It is produced as a multi-filamentary tape conductor with a silver alloy matrix.
The typical cross-section of the tape is about 4 mm × 0.2 mm. The alloying of silver with gold enables
a significant reduction of the thermal conductivity of the matrix with respect to pure silver (Fig. 8). Bi-
2223 for use in current leads is produced specifically with a suitable gold alloyed matrix. The typical
fraction of gold is 3% atomic.
Fig. 8: Thermal conductivity of silver-gold matrix of Bi-2223 tape used in current leads [18]. Percentages of
gold are atomic.
555
A. BALLARINO
The large anisotropy of the Bi-2223 conductor and the consequent strong dependence of the
critical current density on the field orientation shall be taken into account in the optimization of the
geometry of the HTS part. Bi-2223 tapes are more sensitive to magnetic fields oriented perpendicular
to the plane of the tape – a dependence that is particularly marked at higher temperatures. In the LHC
current leads, for instance, up to nine Bi-2223 tapes are vacuum-soldered together to form rugged [19]
superconducting components. The stacks are then disposed in a cylindrical configuration that orients
the self-field parallel to the plane of the tapes [17].
YBCO tape is in principle also suitable for application to HTS current leads. It could be used
without the copper stabilizer, which is usually soldered or electro-plated onto the tape, to limit the
main thermal conduction to the contribution of the metallic substrate and buffer layers. An additional
stabilizer with low thermal conductivity then has to be provided along the HTS part in order to meet
the protection requirements that are defined by the characteristics of the electrical circuit.
4 HTS current leads: operation in pulsed mode

The previous sections describe the optimization of current leads operated at constant current. For the
optimization of the resistive part of HTS leads operated in pulsed mode, the system of non-linear
second-order partial differential equations expressing the heat balance between the lead and the gas in
transient conditions in Eq. (12) must be solved numerically:
𝜕 𝜕𝑇 ρ(𝑇)𝐼(𝑡)2 𝜕𝑇
�𝑘(𝑇)𝐴 � + − 𝑃ℎ(ϑ)(𝑇 − ϑ) = 𝑐c (𝑇)νc 𝐴 ,
𝜕𝑥 𝜕𝑥 𝐴 𝜕𝑡
𝜕ϑ 𝜕ϑ
𝑃ℎ(ϑ)(𝑇 − ϑ) − 𝑚𝑐p (ϑ) 𝜕𝑥 = 𝑐p (ϑ)νg (ϑ)𝐴g 𝜕𝑡 . (12)
In Eq. (12), t is the time, cc, νc and cp, νg are the specific heat and density of the conductor and
coolant, respectively, and Ag is the coolant cross-section. The system in Eq. (12) can be solved
numerically taking as boundary conditions T(x = 0,t) and ϑ(x = 0,t) and as initial conditions the
temperature profile at t = 0 of the coolant and of the conductor (T(x,t = 0) and ϑ(x,t = 0)). For pulsed
operation of superconducting systems having ramp rates of interest for some accelerators magnets –
several kA s−1 – the optimum performance is obtained by using a variable cross-section – larger in the
top part of the lead [20].
AC losses in the Bi-2223 tape, which include hysteresis losses in the superconductor and eddy
current losses in the silver alloy matrix, are a small fraction of the thermal conduction [21].
5 Current leads: operation requirements

A conventional self-cooled current lead and the gas-cooled resistive part of an HTS lead need to be
protected against thermal runaway. The origin of the runaway can be a loss or a reduction of coolant
flow rate due to operational issues. Also, conduction-cooled current leads need to be protected in case
of improved performance and reduced cross-section via the use of intermediate heat sinks. Time
constants in the case of thermal runaway are long – several seconds – and protection is given by the
voltage measured across the devices that trigger a current discharge when a critical threshold is
reached – typically of the order of 100 mV. The superconducting part of an HTS current lead has to be
protected in case of resistive transition. The voltage measured across the HTS element serves as a
quench protection signal.
The mass flow rate of self-cooled conventional leads or the gas-cooled resistive part of HTS
leads needs to be controlled. In HTS leads, THTS is the signal typically used for the control of the valve
556
regulating the flow [22]. In conventional leads, a temperature sensor located toward the warm end is
often used for control purposes.
6 Current leads for persistent mode operation

Persistent current operation is adopted in some applications, such as magnets for magnetic resonance
imaging systems (MRI), where high stability of magnetic field (field decay <10−8 h−1) has to be
guaranteed. After having been energized, the magnet is operated in a closed loop, with power supply
turned off and current leads detached from the lower section. A superconducting switch inside the
magnet, which is maintained above critical temperature during powering of the system, ensures the
circuit continuity in the superconducting loop during persistent mode operation. To guarantee
uniformity of current in time, resistive joints in the circuit must have sufficiently low resistance.
Current leads for this type of application must be light, easily extractable, and equipped with retractors
for automation and control of the break-away lead configuration. To meet these requirements, their
optimization can be done for transient mode operation since transfer of current – and associated losses
– are limited to the energizing phase of the system.
7 Buses and links

In an accelerator system, a current lead is connected at its cold end to a superconducting bus that
provides the electrical link to the magnet. Superconducting buses also interconnect magnets that
belong to the same electrical circuit. The bus is operating at a constant temperature in a liquid helium
bath, and it consists of a Nb-Ti strand or cable designed for transporting the required current in the
maximum field that the superconductor experiences. The design of a superconducting bus must take
into account stability and protection requirements.
Resistive transitions in a bus can be induced by local increase in temperature due to transient
disturbances, like movements or beam losses, or by resistive joints in the regions where splices
between superconducting cables are made. If a normal zone appears and heat generation exceeds the
available cooling, the normal zone propagates longitudinally along the bus with a speed given by the
magnitude of the longitudinal quench velocity. Copper stabilizer is connected in parallel to the
superconductor and acts as a bypass for the current during the discharge of the circuit that follows the
detection of a resistive transition. The cross-section of the copper stabilizer, which has to be such as to
limit the peak temperature reached during the transient, must be continuous along the bus length to
avoid burnout of the cable when the resistive zone reaches the area of discontinuity. Such burnout can
have major consequences if the bus is part of a circuit with high stored energy: an open circuit
generates an electrical arc and a discharge of the energy in the cryostat cold mass. The LHC incident,
in September 2008, was caused by a resistive splice in the main dipole Nb-Ti bus near an area with a
discontinuity in the copper stabilizer [23].
A superconducting bus can be located in a dedicated cryostat and cooled by its own cryogenics.
The system comprising the bus, the cryostat, and the cryogenic instrumentation and control equipment
is called the superconducting link. Superconducting links made from HTS and MgB2 superconductors
and cooled by forced flow of helium gas are being developed [24, 25]. These links, which take
advantage of properties of materials that remain superconducting at higher temperatures than classical
superconductors, permit new concepts to be developed for power transmission systems. They have the
potential of providing in the near future an attractive alternative way of delivering current to
superconducting magnet circuits.
557
A. BALLARINO
References
[1] A. Sommerfeld, Naturwissenschaft 15 (1927) 825.
[2] G. Kumar et al., J Mat. Sci. 28 (1993) 4261–4262.
[3] A. Ballarino, Physica C 372–376 (2002) 1413–1418.
[4] A. Ballarino and A. Ijspeert, Expected advantages and disadvantages of high-Tc current leads
for the Large Hadron Collider orbit correctors, Adv. Cryo. Eng. 39 (1994) 1959-1966.
[5] R. Heller and J. R. Hull, IEEE Trans. Appl. Supercond. 5 (1995) 797–800.
[6] R. Wesche and A. Fuchs, Cryogenics 34 (1994) 145–154.
[7] T. Uede et al., IEEE Trans. Appl. Supercond. 3 (1993) 400–403.
[8] A. Ballarino and L. Serio, Test results on the first 13 kA prototype HTS leads for the LHC,
Proc. PAC 1999 2 (1999) 1405-1407.
[9] A. Ballarino, Physica C 372–376 (2002) 1413–1418.
[10] T. Isono et al., IEEE Trans. Appl. Supercond. 9 (1999) 519–522.
[11] A. Ballarino, Physica C 468 (2008) 2143–2148.
[12] P. Bauer et al., IEEE Trans. Appl. Supercond. 22 (2013) 4800304.
[13] A. Ballarino, Conduction-cooled 60 A resistive current leads for the LHC dipole correctors,
LHC Project Report 691 (2004).
[14] M. Wilson, Superconducting Magnets (Clarendon Press, Oxford, 1983), pp. 256–278.
[15] A. Ballarino, Ph.D. thesis, Politecnico di Torino, 1997.
[16] M. Sitko, B. Bordini and A. Ballarino et al., IEEE Trans. Appl. Supercond. 23 (2013)
4900805.
[17] A. Ballarino, S. Mathot and D. Milani, 13000 A HTS current leads for the LHC accelerator:
from conceptual design to prototype validation, Proc. EUCAS, 2003 [Supercond. Sci. Technol.
17 (2004), LHC Project Report 696.
[18] H. Fujishiro et al., IEEE Trans. Magn. 30 (1994) 1645–1650.
[19] A. Ballarino, L. Martini, S. Mathot, T. Taylor and R. Brambilla, IEEE Trans. Appl.
Supercond. 17 (2007) 3121–3124.
[20] A. Ballarino, IEEE Trans. Appl. Supercond. 17 (2007) 2282–2285.
[21] S. Ginocchio, A. Ballarino, E. Perini and S. Zannella, IEEE Trans. Appl. Supercond. 17 (2007)
224–227.
[22] A. Ballarino, IEEE Trans. Appl. Supercond. 23 (2013) 4801904.
[23] L. Rossi, Supercond. Sci. Technol. 23 (2010) 034001.
[24] A. Ballarino, IEEE Trans. Appl. Supercond. 21 (2011) 980–983.
[25] A. Ballarino, Supercond. Sci. Technol. 27 (2014) 044024.
558
Large Superconducting Magnet Systems
P. Védrine 1
CEA Saclay, Gif sur Yvette Cedex, France
Abstract
The increase of energy in accelerators over the past decades has led to the
design of superconducting magnets for both accelerators and the associated
detectors. The use of Nb−Ti superconducting materials allows an increase in
the dipole field by up to 10 T compared with the maximum field of 2 T in a
conventional magnet. The field bending of the particles in the detectors and
generated by the magnets can also be increased. New materials, such as
Nb3Sn and high temperature superconductor (HTS) conductors, can open the
way to higher fields, in the range 13–20 T. The latest generations of fusion
machines producing hot plasma also use large superconducting magnet
systems.
Keywords: superconducting magnets, fusion machine, detector magnets,

particle accelerators; superconductors.
1 Overview of large-scale applications of superconductivity

Nowadays, superconductivity is essential in the construction of magnets for physics. By using
superconducting magnets, the Joule effect can be eliminated, because the electrical resistance below a
critical temperature is zero. The magnetic field generated in very large volumes can also be increased
compared with that in conventional magnets. The operating cost of the machine can be reduced,
despite the additional equipment needed to maintain the magnet at low temperature, and the size of the
magnets is decreased because of the high current density in the superconducting materials.
The materials used in the design of superconducting magnets for physics are metallic alloys
such as Nb−Ti or inter-metallic compounds such as Nb3Sn. They can carry current densities as high as
3000 A·mm−2 at 12 T and 4.2 K.
In particle physics accelerators, the increase in the collision energy leads to higher bending
fields and higher focusing gradients, because the particle energy of a beam is proportional to its radius
multiplied by the dipole bending field. A normal conducting magnet design is limited both to fields
below 2 T and to a gradient below 20 T·m−1, which would lead to a very large synchrotron. The first
large accelerator built with superconducting magnets was the Tevatron at the Fermi National
Laboratory near Chicago in 1983 [1]. The recent discovery of the Higgs boson at the Large Hadron
Collider (LHC) highlights the role of superconductivity in this machine, which has been in operation
at CERN since 2008 [2].
Superconducting magnets are used not only in the accelerator, but also in the associated
detectors. The first superconducting magnet in a detector came with the 10 in. bubble chamber magnet
built at the Argonne laboratory (1964) [3]. The most popular design for detectors is a solenoid with
diameters of a few metres and a length of tens of metres. However, dipole or toroidal configurations
1
pierre.vedrine@cea.fr
P. V ÉDRINE
have also been built for some detectors. The ATLAS detector built for the LHC is one of the most
famous examples of a toroidal magnet [4].
Controlled thermonuclear fusion is also a very interesting large-scale application and requires
magnetic confinement of plasma with very high fields (>13 T) in large volumes. Several machines
have been built in the past and remain in operation. The superconducting magnets for the International
Thermonuclear Experimental Reactor Programme (ITER) built in Cadarache, France, are now under
construction [5].
Superconducting magnets are also used in many areas other than high-energy physics or
fusion. Superconducting materials must be used everywhere that high fields in large volumes are
required, or where increased current densities or a reduction of the Joule effect is desired.
Superconducting solenoids are now a common tool in research laboratories, providing fields from a
few tesla to more than 20 T in bores from 30 to 300 mm. Superconducting magnets for magnetic
resonance imaging are used commonly in hospitals, and new research projects are pushing the limits
of image resolution [6]. Nuclear magnetic resonance spectroscopy systems used to obtain information
about molecules rely on high-field superconducting magnets. Other applications, such as power
applications (rotating electrical machines, transformers, fault current limiters, power transmission
lines), transportation (ship and aircraft propulsion systems, magnetically levitated trains, railway
traction transformers), industrial applications (inductive heating, magnetic separation), and energy
storage are all under investigation [7].
The market for superconductivity, for an annual total of €5 million, is today dominated by the
magnetic resonance imaging market, and this is expected to increase slowly during the next decade.
Applications with high-temperature superconductors represent less than 1% of the total.
2 Accelerator magnets: review of the four major projects (Tevatron, HERA, RHIC,
and LHC)
2.1 Introduction
The challenge in manufacturing superconducting magnets for accelerators is to ensure reproducibility,
safety, and reliability of hundreds of magnets mounted along kilometres of a circular accelerator, in a
cryogenic environment.
The design of superconducting magnets is very different from that of conventional magnets.
The quality of the magnetic field is obtained principally from the position of the conductors around the
bore tube, and the tolerances on the positioning must be better than 20 µm. An iron yoke is present to
reduce the fringe field outside the coil and also to enhance the field in the useful aperture of the
magnet. Some specific problems arise in the use of superconducting magnets, for example quenches,
training, and protection.
The Lorentz forces, created by the combination of the high current density in the coils and the
flux density, tend to squeeze the coils azimuthally towards the mid-plane with a magnitude of many
100 kN·m−1. These forces have the tendency to move the coil from its end stops at the pole plane. In
the same way, the radial forces tend to force the coil outwards with a maximum displacement at the
mid-plane. Axially, the coils tend to elongate. All these forces have to be maintained in a strong
mechanical structure to avoid deformation or movement of the coils, which would lead to premature
quenches or a bad field quality.
One way to prevent movement of the conductors under the azimuthal forces is to apply a pre-
compressive force during the clamping of the collars, which maintains the coils in a rigid cavity. The
coils act like springs with a non-linear hysteretic response, and the Lorentz forces have to pass beyond
the pre-load value to begin to displace the conductors significantly. The value of the pre-compression
560
L ARGE S UPERCONDUCTING M AGNET S YSTEMS
required at room temperature depends on the mechanical arrangement and on the relative thermal
contraction of the selected materials between the room and helium temperatures.
Generally, the coil consists of two layers of turns of Rutherford cables, sometimes different
between one layer and another. The cable is wound with a tension of about 500 N on a mandrel
assembled with accurately stamped laminations. Once the first layer is ready, it is cured under pressure
in a mould also made with stamped laminations. This operation is performed at the curing temperature
of the resin contained in the insulation of the cable. The second layer of the coil is wound separately
on a second mandrel or on top of the first layer after its curing. The second solution has the advantage
of avoiding a spliced joint between the layers.
After the curing of the second layer, the coil is ready to be assembled with the other coils in
common collars (punched laminations a few millimetres thick), made of aluminium alloy or austenitic
stainless steel. The clamping of the collars is insured by rods or keys. The dimensions of the collars
are set to ensure the pre-load required for the magnetic forces and the correct geometrical shape of the
coil.
The stacking of the iron laminations can take place around the collared coil. The iron yoke is
generally compressed onto the collars and fixed by clamps. The gap between the top and bottom (or
left and right) iron laminations is open at room temperature and designed to be closed at helium
temperature.
An outer cylinder in one or two parts, made of stainless steel or Al alloy, is welded or slid
longitudinally around the yoke to align the entire assembly and to provide an additional pre-load on
the coil.
If no precautions have been taken, when a magnet quenches the entire stored energy will be
dissipated in the area of the quench initiation point. In high-field accelerator magnets, the need for a
high overall current density implies a low copper to superconducting ratio (about 2:1). We are
therefore very far from cryogenic stabilization, and most of the heat dissipated in the normal zone will
heat the quench initiation point. The only way to decrease the temperature is to obtain a faster current
drop by spreading the quench rapidly over the entire coil in which the stored energy can be absorbed.
2.2 Tevatron and HERA

The ﬁrst complete superconducting synchrotron was the Tevatron built at FNAL near Chicago [8].
The Tevatron was commissioned in 1983 and closed in 2011; during operation in 1995 it was used to
discover the top quark. The 1 TeV proton–antiproton ring was made of about 774 superconducting
dipoles and 216 quadrupoles, wound with Nb−Ti conductors and assembled with a circumference of
6.3 km. The second accelerator, an electron–proton ring called HERA (Hadron-Elektron-Ring-Anlage,
or ‘Hadron Electron Ring Facility’) with a 820 GeV proton beam, was commissioned at DESY near
Hamburg (Germany) in 1991 and closed in 2007. It used about 422 dipoles and 256 quadrupoles, also
over a circumference of 6.3 km [9].
Figure 1 (left) shows the Tevatron accelerator complex and (right) the HERA accelerator.
Table 1 summarizes the main characteristics of the dipoles and quadrupoles of these two machines.
Figure 2 shows as an example the Tevatron and HERA dipoles.
561
P. V ÉDRINE
Fig. 1: Tevatron accelerator complex (left) and HERA accelerator (right)
Table 1: Magnet parameters of the Tevatron and HERA proton ring
Dipole Quadrupole
Tevatron HERA Tevatron HERA
Operating temperature (K) 4.6 4.5 4.6 4.5
Central field (T) 4.4 4.68
Gradient (T·m−1) 75.8 91.2
Field length (m) 6.11 8.82 1.679 1.861
Nominal current (A) 4400 5027 4400 5027
Inner coil diameter (mm) 76.2 75 88.9 75
Number of coil layers 2 2 2 2
Superconducting material Nb−Ti Nb−Ti Nb−Ti Nb−Ti
Filament diameter (µm) 8 14 8 19
Collar material Stainless steel Al alloy Stainless steel Stainless steel
Yoke Warm Cold Warm Cold
Cryostat length (m) 6.4 9.77 2.31 3.98
Total number 774 422 256 224
562
Fig. 2: Tevatron dipole (left) and HERA dipole (right) magnet cross-section
The superconducting magnets of both machines use Nb−Ti as the superconductor. The Tevatron
used a wire with filaments of 9 µm for the dipole and the quadrupole, whereas HERA used a 14 µm
filament for the dipole and a 19 µm filament for the quadrupole. The wire surfaces were quite
different. The Tevatron wire surface was tinned and oxidized in alternating wires, whereas it was only
silver tinned in HERA.
The cables for the Tevatron magnets were of Rutherford type, but with different numbers of
strands: 23 and 24 strands of 0.648 mm cable for the dipole and quadrupole, respectively. The cross-
section was 1.067/1.372 × 7.671 mm2. The Cu/Sc ratio was 1.8:1. The minimum critical current
density was 1800 A·mm−2 at 4.2 K, 5 T. The insulation of the cable comprised one layer of polyimide
film of 25 µm wrapped with an overlap of 58%. In addition, the cable was wrapped with fibreglass
tape impregnated with curable B-stage epoxy.
The cable for HERA was a 24-strand cable with a cross-section of 1.28/1.67 × 10 mm2. The
Cu/Sc ratio was 1.8:1, and the minimum critical current density was 2200 A·mm−2 at 4.2 K, 5.5 T.
The coils were wound in two layers and connected by splicing, except on the HERA
quadrupole, where the second layer was wound on the first after curing.
Laminated stainless steel in the case of the Tevatron dipole, or a laminated aluminium collar for
the HERA dipole, was used to clamp the coils together and apply the required pre-compression. The
collars were able to withstand the magnetic forces for the Tevatron, but not for the HERA dipole,
above a field of 5.9 T. At these levels of field, the yoke helped to counterbalance the forces.
The Tevatron magnets used a ‘warm iron’ yoke outside the cryostat (see Fig. 2), whereas the
HERA magnets used a ‘cold hybrid iron’ yoke. In the case of the ‘warm iron’, the iron was at room
temperature and not cooled to cryogenic temperature. The effect of the iron on the field was small
because the ratio of the radii was low and the saturation negligible. The disadvantage was that the
centring of the coil inside the yoke was very difficult and required cryogenic support to limit the heat
conduction. A misalignment of the yoke could create very huge non-symmetric magnetic forces on the
conductors, together with a field distortion. For the ‘cold hybrid iron’, the iron was cold, but a gap was
left between the coils and the yoke where the clamping collars held the electromagnetic forces, which
allowed good centring. The effect of the iron saturation was moderate, but the contribution to the main
field remained important.
All Tevatron magnets were measured over two different excitation cycles: one at the nominal
ramp rate and the other at double the ramp rate. Magnets quenched between 3900 A and 4900 A, with
a mean value around 4500 A.
563
P. V ÉDRINE
Training of the HERA magnets was negligible, and the maximum quench current was reached
after only a few quenches. Measurements of the quench currents at 4.75 K gave a mean of 6458 A for
the dipoles and of 7383 A for the quadrupoles.
2.3 The RHIC project

The Relativistic energy Heavy-Ion Collision (RHIC) project [10] is a colliding beam facility that has
been in operation at Brookhaven National Laboratory since 2000. It is dedicated to studies of nuclear
phenomena in RHICs at a maximal energy of 100 GeV/u. The collider is designed with two
independent, identical quasi-circular concentric superconducting magnet rings (3.8 km in
circumference) (Fig. 3). The two beams can intersect at six locations around the ring. A standard cell
consists of two dipoles of 9.46 m and two combined magnet units comprising a quadrupole of 1.13 m,
a 0.75 m sextupole, and a 0.58 m assembly of multipole correctors.
The magnets are installed in an existing tunnel. As the radius of curvature is relatively large
compared with the target energy, the magnetic field of the dipole is only 3.45 T.
The dipole magnet is designed with a one-layer coil containing four blocks of conductors,
separated by three wedges to improve the field quality (Fig. 4). The diameter of the bore is 80 mm.
Table 2 summarizes the main parameters of the RHIC magnets.
The superconducting cable, 9.73 mm wide with a keystone angle of 1.2° (1.166 × 9.73 mm), is a
‘Rutherford-type’ cable with 30 strands of 0.648 mm in diameter. Each wire is a composite of 6 µm
Nb−Ti filaments in a matrix of pure copper with a Cu/Sc ratio of 2.25:1. The minimum specified
critical current density is 2600 A·mm−2 at 5 T, 4.2 K, but a value of about 2800 A·mm−2 was achieved
during the manufacture of the wire. Standard polyimide/fibreglass insulation is used for the cable. The
coil, which is made of 32 turns arranged in four blocks, is surrounded by a spacer made with an
injection-moulded mineral-loaded phenolic glass, which acts as an insulator and creates a distance
from the surrounding yoke (Fig. 4). The iron yoke serves to clamp the coils, create the pre-
compression, and withstand the magnetic forces. The outer helium vessel is made with two half
stainless steel tubes welded longitudinally around the yoke. The coil end forces are also retained by
thick endplates anchored at the shell.
Fig. 3: RHIC collider
564
Table 2: Main parameters of the RHIC magnets
Arc dipole Arc

quadrupole
Operating temperature (K) 4.6 4.6
Central field (T) 3.45
Gradient (T·m−1) 71.8
Nominal current (A) 5050 4720
Field length (m) 9.45 1.13
Inner coil diameter (mm) 80 80
Cryostat length (m) 9.728 3.023
Total number 264 276
The quadrupole magnets employ a similar design. This quadrupole is assembled with a
corrector magnet at one end and a sextupole magnet at the other in one cryostat.
Of the magnets, 20% were cold tested and quenched. They reached the ‘plateau’ rapidly, within
a few quenches. The initial quench current and the plateau quench current of each magnet exceeded
the 5 kA operating current.
Fig. 4: Cross-section of RHIC dipole cold mass
2.4 The LHC project

The Large Hadron Collider (LHC) is a European project superconducting synchrotron, which is
installed in the 27 km Large Electron–Positron (LEP) tunnel at CERN [2]. It is designed to accelerate
two independent beams of protons in opposite directions up to 7 TeV, to study quark–quark
interactions and to explore rare processes such as the existence of the Higgs boson. The structure of
the LHC comprises eight arcs separated by eight straight sections (Fig. 5). The two beams alternate
from the outer to inner channels in successive arcs, crossing each other’s paths at eight points, six of
which could be used for physics experiments. The circumference of the collider is fixed by the
dimensions of the LEP tunnel, in which available space is limited.
565
P. V ÉDRINE
Fig. 5: LHC collider
These two constraints have led to the development of the concept of a high-field
superconducting magnet of 8.33 T assembled in a structure referred to as ‘two-in-one’. The two
magnetic apertures are placed horizontally in the same yoke and cryostat, leading to a smaller cross-
section (Fig. 6). The choice of Nb−Ti as the superconductor has also led to the use of superfluid
helium to reach the high field demanded. The use of superfluid helium to cool the magnets to below
2 K has some advantages. Superfluid helium has a very large heat conductivity, which helps to remove
heat, and it also has the ability to penetrate the coils through every pore, and therefore to improve the
coil cooling. The standard half-cell comprises three dipoles of 14.3 m and one main quadrupole of
3.1 m. Two sets of combined corrector magnets are also installed in each half-cell: a combined
octupole/quadrupole corrector and a combined decapole/sextupole/dipole corrector. The R&D
program for the LHC has focused on the fabrication of numerous models and prototypes.
The main characteristics of the LHC collider dipoles can be found in Table 3.
Table 3: Main parameters of the LHC main dipole and quadrupole magnets
Dipole Quadrupole
Operating temperature (K) 1.9 1.9
Central field (T) 8.33
Gradient (T·m−1) 222
Operating current (A) 11 850 11 850
Field length (m) 14.3 3.1
Inner coil diameter (mm) 56 56
Distance between apertures (mm) 194 194
Number of coil layers 2 2
Superconducting material Nb−Ti Nb−Ti
Collar material Stainless steel Stainless steel
Yoke Cold Cold
Total number 1232 386
566
The dipole coils [11] are made with two different cables, one for each layer. The inner layer
has 28 strands of 1.065 mm diameter with 7 µm filaments. The cross-section is
1.736/2.064 × 15.1 mm2. The Cu/Sc ratio is 1.7. The outer layer has 36 strands of 0.825 mm diameter
with 6 µm filaments. The cross-section is 1.362/1.598 × 15.1 mm2. The Cu/Sc ratio is 1.9. The main
quadrupole cable is of the same dimensions as that used for the outer layer of the dipole coils. The
critical current density at 10 T and 1.8 K in all these cables is above 1530 A·mm−2.
The design of the coils is classical with two layers of cables. The Cu/Sc ratio and the cross-
section of the cables are adapted on the inner and outer layers to optimize the current density. The
insulation scheme is composed of half-overlapped polyimide tape, reinforced with a B-stage
polyimide adhesive film.
The force containment consists of coil clamping collars, the iron yoke, and the shrinking
cylinder. They contribute to produce the necessary azimuthal pre-compression in the coils and to
withstand the electromagnetic forces in one single structure for the two channels. The collars are made
of high-strength austenitic steel. They are locked by rods near the mid-plane. The collars are common
for the two apertures.
The iron yoke is split vertically into two parts with a gap at room temperature. The gap is
needed to compensate for the difference in thermal contraction of the iron and of the collar material
during the cooling. The gap is closed at 2 K. The electromagnetic forces at full excitation are shared
between the collars and the yoke, and the gap remains closed.
The shrinking cylinder is also the outer shell of the helium tank. This cylinder is welded around
the yoke with an interference to generate a tensile stress, which participates in the pre-compression of
the coils throughout the current excitation.
Fig. 6: Cross-section of an LHC dipole
Table 3 presents the main characteristics of the lattice quadrupole [12, 13]. Figure 7 shows a
cross-section of the cold mass of the quadrupole. The quadrupole design is based on two-in-one
geometry. The two quadrupoles of each unit are combined in a focusing/defocusing configuration, to
minimize the saturation effects in the yoke. Stainless steel collars alone withstand the electromagnetic
forces and the pre-compression. A set of eight tapered keys locks the collars in position. The yoke is
made of single-piece laminations, interlocked by pins and keys, with the two quadrupole assemblies
fixing their relative positions. A rigid stainless steel tube (inertia tube) is placed around the outside of
the yoke and aligns the magnet. Holes, through which dowels can travel, are machined very precisely
in the inertia tube. These dowels also pass through the keys, which are placed into slots in the yoke,
such that the yoke is aligned by the inertia tube.
567
P. V ÉDRINE
Fig. 7: Cross-section of an LHC quadrupole
Most of the magnets reached the nominal current of 11 850 A (about 86% of the short sample
limit) with two quenches. Series production magnets have been built by four European companies:
Alstom (France), Babcock Noell (Germany), and ASG (Italy) for the dipoles; and ACCEL Instruments
(Germany) for the quadrupoles.
3 Detector magnets: from the first bubble chambers to CMS and ATLAS
3.1 Introduction
Large colliders have emerged in which the experiments can only be carried out at intersection points.
The number of detectors installed in the collider is limited by the number of possible intersection
points. Therefore, the detector must be designed to carry out the widest range of physics experiments
for all the possible directions around the colliding beams. For this purpose, an axially symmetrical
configuration is advantageous, and naturally the use of solenoid magnets was selected first. This
constraint has led to a new class of magnets for detectors called ‘large thin-wall solenoids’ [14]. These
solenoids provide a huge magnetic field volume (1–4 T) surrounding the colliding beam, filled with a
panoply of all-purpose detectors for the analysis of collision events.
A solenoid producing a field in the direction of the circulating beams does not create first-order
perturbations on the particles’ trajectories and does not require complicated compensating devices.
From a technological point of view, a solenoid presents the simplest structure, and, in view of its two-
fold symmetry, it is naturally self-supporting and thus raises no difficult mechanical problems. This
configuration is particularly favourable in the case of a superconducting solenoid, because it does not
require sophisticated cold supports and it allows a simple cryostat structure. The size of the solenoids
can be increased to very large dimensions to allow a large volume of free space in which to insert the
detectors into the magnetic field. Dipole or toroidal configurations are also used in some detectors.
The toroidal field shape gives the best momentum resolution at low angles, and the magnetic field is
always transverse to the particle momentum.
Following the above remarks, it is necessary to stress a few requirements specific to the
application considered here. The question of transparency is related to the types of particles liable to
cross the magnet and reach the detectors situated outside. In practice, two situations can be found
according to the relative location of the solenoid with respect to the e–γ calorimeter, which is one of
the essential components of the detectors. If the solenoid is inside this calorimeter, it is relatively small
in dimension, but it must be made as thin as possible to minimize interactions with particles (typically
a fraction of a radiation length, corresponding to a few centimetres of aluminium). If it is outside, it
becomes very large in size, but it is not so restricted in material thickness—within the limits imposed
by the absorption of hadron particles. Thus, a choice has to be made between the two options, which
results from the difficult compromise between considerations of the physics, technical feasibility, and
overall cost.
568
The minimization of the radial thickness of the magnet is also important to save space for the
detectors. Other requirements concern the mechanical interfaces in a crowded environment, the weight
within the limitation of handling and lifting equipment, and also the cost and reliability.
Generally, an indirect cooling or forced-flow solution has been adopted to keep the temperature
of the magnet down. A pool-boiling technique, ensuring cryostatic stability, requires huge and costly
helium vessels, incompatible with the requirements of geometry and space. Forced flow in the
conductor is sometimes adopted, but this leads to very complex cryogenic systems and serves only to
provide an additional reserve of enthalpy for the stability of the conductor. In indirect cooling, loops
with helium flowing within are attached to the outer cylinder of the coil. The heat is removed by
conduction from the coil to these loops.
The means by which the circulation of the coolant is obtained has also evolved. First, forced-
flow modes used the refrigerator Joule–Thompson loop, with the inconvenience of relying completely
on the refrigerator. Any shutdown of the refrigerator, however short, leads to a shutdown of the field
and often to a quench. Next, forced flow by means of cold pumps with a reserve of helium was
introduced. In this case, one depends much less on the refrigerator, but still on the reliability of the
pumps. Some more recent projects use natural convection with a ‘thermo-siphon’ method, working
with a reserve of helium placed at the top of the cryostat. This last method has the advantage of being
largely independent of the refrigerator and also of not having any moving parts, the reliability of
which is always limited.
In the coil design, the first constraint is to ensure safe and reliable operation. The conductor is
chosen to have a large safety margin, where the working current is often half the critical current. The
conductor is generally a Rutherford cable embedded in a pure aluminium stabilizer. The amount of
stabilizer is determined mainly based on criteria for protection, and the overall critical current ranges
from 30 to 60 A·mm−2, which is about one order of magnitude lower than that currently used for
accelerator magnets. Almost all the superconducting solenoids for detectors use the inner winding
method, where the first layer of the conductor is layer-wound inside an outer support cylinder. The
other layers of the conductors are then wound inside the last layer. The coils are tightly clamped, and
their support member is impregnated with epoxy, to form a monolithic structure that ensures good
thermal conduction and prevents internal movement of the conductors.
The large dimensions of these solenoids lead to huge hoop stresses that cannot be borne by the
conductors alone. For a given field and current density, the hoop stress is proportional to the radius of
the solenoid. Therefore, an external support cylinder has to be used at the outer edge of the conductor
of the outer layer to limit the tensile strain in the conductor to a value smaller than 0.1%. This cylinder
is made of an aluminium alloy with high yield strength. Some conductors are also reinforced with an
aluminium alloy outer shell bonded to the pure aluminium stabilizer, to take a part of the large hoop
stress developed in some designs.
The stability of a coil is its capacity to absorb or remove possible thermal disturbances, such as
conductor motion and deposition of energy by particles. However, such disturbances are difficult to
predict as they depend greatly on the quality of the coil’s manufacture. We can design a conductor
with a very large stability margin, fully cryostable, but which is detrimental to the size of the magnet.
Another approach is to limit the large mechanical disturbances by using an epoxy impregnated coil.
The design must also avoid local stress concentrations and maintain a stress level far below the
initiation of cracks in the epoxy. The conductor itself helps the stabilization with its enthalpy and the
high thermal conductivity of the aluminium.
Quench protection is the major requirement in the design of detector magnets, because the
stored energy can be as high as 1–3 GJ. During a quench, a large amount of the stored energy is
extracted during the dumping of the current in the external resistances with a maximum voltage
generally smaller than 1 kV.
569
P. V ÉDRINE
At the end of the quench, the maximum temperature reached in the hottest point of the coil must
be limited to 100 K, because below this temperature the thermal expansion of the materials is very low
and the increase in temperature does not lead to mechanical problems. The Joule heating in the
transited length of the conductor is absorbed consistently by the enthalpy of the monolithic conductor
itself. This temperature is a function of the dump time constant and the characteristics of the materials.
However, despite its higher resistivity, the support cylinder plays a positive role in the protection. This
tube is strongly magnetically coupled with the solenoid, such that, when external discharge begins, a
part of the energy is transferred to the tube with eddy currents generated by the high dB/dt. The tube is
then heated as a whole, which helps the quench to propagate through the rest of the solenoid (‘quench
back’).
3.1.1 Past detector magnets

The first 12 ft superconducting bubble chamber magnet was constructed at Argonne National
Laboratory in 1968 [15]. The magnet was a split pair solenoid with a vertical axis and gap to allow the
beam entry. Bath-cooled coils with Nb−Ti copper stabilized conductors operated in cryostable mode.
A 15 ft version (Fig. 8, left) was built at Fermilab in 1972 [16].
In 1971 at CERN, the Big European Bubble Chamber (BEBC) was built [17], which included a
superconducting magnet (Fig. 8, right) made with 20 pancakes of Nb−Ti conductor (3 × 61 mm with
200 untwisted filaments of 200 µm). The nominal current was 5700 A. The inter-layer insulation
included slots to allow boiling He to flow between the conductors. Table 4 gives the main parameters
of the first bubble chamber superconducting magnets.
Table 4: First bubble chamber superconducting magnets
Argonne Fermilab CERN

12 ft 15 ft BEBC
Field (T) 1.8 3 3.5
Winding inner diameter (m) 4.8 4.27 4.7
Stored energy (MJ) 80 400 800
Conductor dimensions (mm) 2.54 × 50 3.8 × 38 3 × 61
Current (A) 2000 5000 5700
Number of pancakes 30 43 20
Fig. 8: Fermilab 15 ft bubble chamber (left) and BEBC (right)
570
By the end of the 1970s and during the 1980s, a new type of solenoid for the detectors was
developed. To minimize the amount of matter in the coil and its cryostat, low-mass materials were
substituted for the usual materials (e.g., an aluminium-stabilized conductor instead of copper and
stainless steel). The current density in the conductor was increased (no more adiabatic stability),
indirect cooling by external pipes was implemented, and an intrinsic protection with aluminium shunts
and quench back tube was designed to allow a quasi-uniform distribution of the stored energy in case
of quench.
A list of such solenoids and their main characteristics is given in Table 5.
DELPHI [18] and ALEPH [19] were the two detectors for the LEP using superconducting
coils. They both had to produce a very uniform field in the large volume of the central detector, and
for that purpose they included additional compensating windings at the two ends of the main solenoid.
Both use Al-stabilized conductors. A new feature for this type of conductor compared with the older
ones is the use of flat Rutherford cable as an insert instead of a monolithic superconductor.
Table 5: Characteristics of the last generation of superconducting solenoids in operation today
Name ALEPH DELPHI CLEO2 [20] ZEUS[21] H1[22]
Accelerator LEP LEP CESR HERA HERA
Laboratory CERN CERN Cornell DESY DESY
Designed by Saclay RAL Cornell Oxford. Milan RAL

University.
Manufactured by Saclay RAL Oxford Inst. Ansaldo RAL
Inner bore (m) 4.96 5.2 2.88 1.72 5.2
Outer bore (m) 5.98 6.2 3.48 2.22 6.08
Winding length (m) 6.35 6.8 3.48 2.5 5.16
Overall length (m) 7 7.4 3.78 2.8 5.75
Conductor (mm2) 35 × 3.6 24 × 4.5 16 × 5 15 × 4.3 26 × 4.5
Stabilizer Al Al Al Al Al
Cold mass (t) 25 25 7 3.4
Conductor mass (t 8 7 2.4 7
Current (A) 5000 5000 3300 5000 5500
Design field (T) 1.5 1.2 1.5 1.8 1.2
Stored energy (MJ) 137 108 25 16 120
Cooling method Thermo- Forced-flow Thermo- Forced-flow Forced-flow

siphon pumps siphon pumps pumps
Radiation length (X0) 2 7.4 0.9 4
Year of completion 1987 1988 1988 1989 1989
571
P. V ÉDRINE
The ALEPH winding is made in two halves, each of 3.2 m, whereas the DELPHI coil is split
into shorter modules of 1.5 m length. The impregnation technique is also different: DELPHI uses a
‘prepreg’ insulated conductor and ALEPH is vacuum impregnated. The two solenoids use the ‘inner
winding’ method, in which the conductor is layered from an external spool on the inner radius of the
external support cylinder through a number of fixtures. The cooling circuit is laid in a similar way for
both magnets with loops on the support cylinder. However, the refrigeration systems are different. For
DELPHI, liquid helium is circulated in forced flow with cold pumps, whereas ALEPH uses a thermo-
siphon.
3.2 Detector for LHC projects

The energy stored in the detector magnets for the LHC is one order of magnitude larger than in
previous generations of magnets because of their size and magnetic field. They all use a cable
embedded in aluminium with indirect cooling as the conductor. All solenoids use an iron yoke. The
final problem is that the magnets must be assembled on site because of their size.
3.2.1 The CMS detector for LHC

The Compact Muon Solenoid (CMS) [22] (Fig. 9, left) is a high-field superconducting magnet (4 T)
with a massive return yoke (12 000 ton) (Fig. 9, right), which forms part of the external muon
spectrometer. The level of field is required to achieve a high resolution compatible with the high
luminosity of the LHC.
The main characteristics are a nominal magnetic field of 4 T at the operating current of
19.5 kA, in a 5.9 m diameter and 12.5 m long warm bore, leading to a stored energy of 2.7 GJ. These
characteristics make this superconducting solenoid the largest and most powerful ever designed. The
CMS solenoid has a larger stored energy per unit mass of coil (12 kJ·kg−1) than any large
superconducting magnet ever built.
The coil comprises a four-layer winding to increase the number of ampere-turns, divided into
five identical sections of 2.5 m length, each internally wound and vacuum impregnated before final
assembly. The outer part of these modules (a 50 mm thick aluminium alloy shell) is used as an outer
winding mandrel and as a mechanical reinforcement structure, but also as a cooling wall and quench
back tube during cool-down, energization, and fast discharge of the coil. The cold mass total weight is
about 225 t. It is supported inside the vacuum tank through a system of radial and longitudinal tie-bar
supports.
The conductor is made of three parts, which are assembled during the fabrication process: a
Rutherford cable made of 32 strands of 1.288 mm diameter is co-extruded inside a high-purity
aluminium matrix to form what is called the insert; two aluminium alloy sections are then welded on
each side of the insert using the continuous electron beam technique to give the mechanical
reinforcement to the conductor structure. The critical current of the conductor at 5 T and 4.2 K is
55 600 A. Cable dimensions are 20.63 × 2.34 mm2, and the conductor dimensions are 64 × 22 mm2.
Indirect cooling of the cold mass is performed using a two-phase helium thermo-siphon.
The five modules were delivered over a period of one year up to January 2005 and superposed
vertically around the internal vacuum chamber and also supporting the internal thermal shields. Then,
after the external thermal shields were put in place, the assembly was tilted into the horizontal position
and inserted into the external vacuum chamber, and positioned in the magnetic yoke in August 2005.
Following the successful testing of the detector magnet at its nominal field rating of 4 T in the CERN
surface facility between August and November 2006, the detector was disassembled to be lowered into
the underground cavern. The magnet, together with its central soft iron shielding ring, was lowered
into the tunnel in February 2007. The magnet was ramped up to its nominal field level in November
2008, with no serious problems occurring.
572
Fig. 9: The complete CMS detector (left); the CMS solenoid in its iron yoke (right)
3.2.2 The ATLAS detector for LHC

ATLAS is one of the two experiments dedicated to the search for the Higgs boson installed on the
LHC ring at CERN.
The ATLAS detector contains two separate magnetic systems [4]: a thin-wall superconducting
solenoid close to the central detector and a set of large superconducting toroidal magnets for the outer
muon spectrometer (Table 6).
The complete toroidal system comprises three separate toroids: a large central ‘barrel’ toroid
(Fig. 10) and two ‘end cap’ smaller toroids inserted at each end of the barrel to achieve complete
angular coverage. Each toroid is made with eight ‘racetrack’ coils laid radially around the beam axis.
Each coil is built independently and consists of two double pancakes, epoxy impregnated, together
with pure aluminium cooling plates, which conduct the heat to the liquid helium loops (classical
indirect cooling scheme). A central web structure is clamped rigidly to the two pancakes and contains,
via pure compression, the internal magnetic forces applied to the conductors. The conductor is a
Rutherford cable co-extruded in a pure aluminium matrix.
Fig. 10: ATLAS detector (left); general view of the barrel toroid (right)
573
P. V ÉDRINE
Table 6: Main parameters of the superconducting toroids and solenoid for ATLAS
Barrel End cap Solenoid
Inner bore (m) 9.4 1.65 2.46
Outer diameter (m) 20.1 10.7 2.63
Overall length (m) 25.3 5 12.5
Number of coils 8 2×8 1
Stored energy (MJ) 1080 2 × 250 40
Operating current (kA) 20.5 20.5 7.7
Peak field (T) 3.9 4.1 2.6
Total weight (t) 830 2 × 239 5.7
Conductor
Overall size (mm2) 57 × 12 41 × 12 30 × 4.25
Type Rutherford cable + pure Al co- Rutherford cable + high-

extruded strength pure Al co-
extruded
The main challenge for these magnets and especially the ‘barrel toroid’, is their integration in
the complex detector set-up. The magnet must leave the maximum open space for the detectors, and
therefore the support structure must be as invisible as possible to the particles.
The eight coils for the central ‘barrel toroid’, developed by CEA Saclay in collaboration with
LASA (INFN) and CERN, were tested one at a time, above ground, by applying a 22 000 A current. In
2005, they were assembled in the ATLAS cavern, 100 m below ground level, using an aluminium
structure. Arranged in a star configuration and positioned with precision of a few millimetres, the coils
occupy a volume equal to that of a six-storey building. The structure supporting the 1400 ton muon
detector can withstand considerable magnetic forces.
The 2 T central toroid, which is a magnet developed by the High Energy Accelerator Research
Organization (KEK) in Japan, was cooled to 4 K for the first time in June and July 2006. A 21 000 A
test current was then applied during the night of 9 November 2006. In view of the success of the first
performance test, the central toroid was connected to the end toroids, which were developed by the
Rutherford Appleton Laboratory in collaboration with NIKHEF and CERN. The two end toroids are
drawn to the central toroid by a force of 240 t.
The three magnets were brought up to their rated current on 4 August 2008. The toroids were
then tested at the same time as the central solenoid.
All these performance tests were successful, and ATLAS is now the largest superconducting
magnet system in the world.
574
4 Fusion magnets: review of machines built and under construction
4.1 Introduction
The thermonuclear plasmas produced in fusion machines need to be confined by strong magnetic
fields. In the early days, resistive magnets were used in small-sized machines and in pulse-mode
operations. However, the total electrical power needed to energize the magnets was sometimes greater
than 1 GW, which drove the need for machines using superconducting magnets.
The development of superconducting magnets for fusion started in the 1970s, and today all
large fusion projects present a superconducting magnet system and no longer use resistive magnets. A
detailed description of the history of fusion machines can be found in Ref. [23].
4.2 Baseball magnets

The Baseball I and II (Fig. 11) superconducting magnets were developed in the 1960s for magnetic
fusion experiments at Lawrence Livermore National Laboratory [24]. The Baseball II superconducting
magnet was designed to generate a maximum magnetic field of 7.5 T (for a central field strength of
2 T). The designed nominal current was 2400 A. The conductor was a monolith of 6.35 mm2, made
with a composite of 24 Nb−Ti 0.6 mm filaments embedded in a copper matrix. The magnet weighed
13 t and has been operated at central fields of up to 1.5 T.
Fig. 11: Baseball II superconducting coil
4.3 The Mirror Fusion Test Facility

The Mirror Fusion Test Facility (MFTF), built at Lawrence Livermore National Laboratory, is an
advanced experimental fusion device designed as an intermediate step between the existing mirror
machines and an experimental fusion reactor. MFTF was completed in 1986, but the overall project
was cancelled later that year. The MFTF magnet system [25] contained a total of 26 large
superconducting coils made with yin-yang pairs and solenoids.
The yin-yang magnet pair (Fig. 12, left) had an average major radius of 2.5 m and an average
minor radius of 0.75 m. A peak magnetic field of 7.68 T occurred on the windings at the minor radius.
The magnetic field dropped rapidly to 4.2 T at the mirrors and to 2.0 T in the centre.
The first yin-yang magnet was tested successfully in February 1982 to its full design field
(7.68 T) and current (5775 A). The coil weighed 341 ton, and the stored energy was 410 MJ. The coil
current density was 25 A·mm−2. The 480 Nb−Ti 0.2 mm diameter filaments were embedded in a
copper matrix (Cu/Sc ratio of 1.7). The conductor (Fig. 12, right) was wrapped and soldered in an
embossed and perforated sheath of high-purity copper to improve the stability and the cooling. The
overall dimensions were 12.4 × 12.4 mm. The coil was wound with 58 layers of 24-turn conductors.
575
P. V ÉDRINE
The magnet was cooled in a 4.3 K, 1.3 bar helium bath. The magnet weighted 375 ton for a
stored energy of 410 MJ.
Fig. 12: MFTF yin-yang coils (left) and conductor (right)
4.4 T-15 Tokamak

The Tokamak T-15 [26, 27] was built in the Kurchatov Institute, Moscow, Russia, at about the same
time (1983–1988) as the Tore Supra in Cadarache (see Section 4.5), using 24 TF coils (Fig. 13, left)
with over 11 ton of Nb3Sn material. It was the largest Nb3Sn-based device in the world prior to the
start of construction on the ITER.
The coils were wound with a react and wind conductor made with a flat cable of 10 Nb3Sn
1.5 mm strands, first heat treated on a drum and then temporarily straightened to be bonded to two
copper pipes of 0.4 m diameter by an electroplated Cu layer (Fig. 13, right). The critical current of the
100 km conductor was 3.9 kA at 6.5 T. The overall dimensions were 17.4 × 6.5 mm. T-15 produced its
first plasma in 1988, and it was shut down in 1995 due to the poor economic environment in Russia at
that time. The TF coil system was made of 24 circular coils with an average diameter of 2.4 m and a
stored energy of 384 MJ.
Fig. 13: T-15 TF coils (left) and conductor (right)
4.5 Tore Supra

Tore Supra is a superconducting Tokamak installed at the CEA Cadarache by the Euratom-CEA
Association. It delivered its first plasma in April 1988. The major plasma radius is 2.25 m and the
minor radius is 0.7 m. The overall diameter is 11.5 m and the height is 7.5 m.
576
Each Toroidal Field (TF) coil [28] (Fig. 14, left) is made of 26 double pancakes of conductor
separated by insulating spacers, which provide both the mechanical cohesion of the winding and
secure a very large volume of superfluid helium at 1.8 K and 1 bar in direct contact with the
conductor. Each coil winding is encased in a strong stainless steel casing. The 18 cases form an inside
and an outside vault, which resists the centripetal forces produced by the toroidal field.
The Nb−Ti conductor (Fig. 14, right) is a monolithic bare conductor with 23 µm filaments
embedded in a copper matrix. The dimensions are 2.8 × 5.6 mm. The weight of the superconductor is
about 45 t for a total magnet weight of 160 t. The stored energy is 610 MJ.
Full performances of the magnet were reached on 8 November 1989. The current in the TF
coils was increased to 1455 A corresponding to 9.3 T on the conductor (design values: 1400 A and
9 T, respectively) and 4.5 T at the plasma centre. Tore Supra coils are now operated at 90% of the
design values (4 T on the plasma). Tore Supra demonstrates that a TF superconducting system can be
operated routinely in the severe conditions of a Tokamak.
Fig. 14: Tore Supra TF coils (left) and conductor (right)
4.6 Experimental Advanced Superconducting Tokamak (EAST)

EAST is the Experimental Advanced Superconducting Tokamak built at the Institute of Plasma
Physics of the Chinese Academy of Sciences (CASIPP) in Hefei [29]. EAST’s major radius is 1.7 m
and its minor radius is 0.4 m. The machine’s overall dimensions are 10 m height and 7.6 m diameter.
The total weight is 414 ton. The toroidal field at the plasma centre is 3.5 T, the operating current is
14.3 kA, and the total stored energy is 300 MJ.
The magnetic system comprises 16 D-shaped TF coils (Fig. 15), one central solenoid split into 6
individual coils, and 4 large Poloidal Field (PF) coils. The peak field on the TF coil is 5.8 T. The
largest PF coil has an outer diameter of 7.6 m. Coils are cooled with supercritical helium at 4.5 K, and
the total cold mass of the superconducting magnet system is 194 ton.
The TF conductor is an Nb−Ti/Cu Cable in Conduit Conductor CICC. The diameter of the
superconducting strands is 0.85 mm. The temperature margin is 1.88 K. The jacket material of the
cable is 316LN stainless steel with 1.5 mm wall thickness. The dimensions of the CICC are 20.4 ×
20.4 mm. The turn electrical insulation consists of multiple layers of polyimide and fibreglass tapes
impregnated with epoxy resin.
The assembly of the Tokamak was completed at the end of 2005, and the first engineering
commissioning was carried out at the beginning of 2006. The EAST machine is now in operation.
577
P. V ÉDRINE
Fig. 15: EAST TF coil assembly
4.7 SST-1
SST-1 (Steady State Superconducting Tokamak) is a plasma confinement experimental device built at
the Institute for Plasma Research in Gandhinagar, India. The project started in 1994 and the
integration of the system was completed in 2012. The first plasma campaigns have already started,
following engineering validation of the Tokamak.
The magnet System of SST-1 [30] is an assembly of 16 superconducting D-shaped TF coils
(Fig. 16) with an average diameter of 1.8 m, 9 superconducting PF coils, and a pair of resistive PF
coils. TF magnets generate a 3.0 T field at the major radius of 1.1 m. The minor plasma radius is
0.2 m. The diameter of the machine is 4.4 m and its height is 2.6 m. The total weight is 160 ton.
Magnets are cooled with supercritical helium at 4 bar and 4.5 K. The operating current is 10 kA and
the peak field on the coils is 5.1 T. Each of the TF coils consists of six double pancakes, each pancake
having nine turns. The TF system stores 56 MJ.
The conductor is an Nb−Ti CICC. The PF superconducting magnets are wound from the same
CICC as that used for the TF magnets. The conductor consists of 1350.86 mm diameter NbTi/Cu
strands with a high copper to superconductor ratio of 4.9:1. These strands are twisted in four stages
before being jacketed inside a conduit made of stainless steel, which has a cross-section of 14.8 ×
14.8 mm and a void fraction of 40% inside the cable space for liquid helium to flow.
Fig. 16: SST-1 TF coils
4.8 KSTAR
The Korean Superconducting Tokamak Advanced Reactor (KSTAR) (Fig. 17, left) is an advanced
plasma, steady-state Tokamak experiment built at the Korean Basic Science Institute in Daejon, South
Korea [31]. The system has been in operation since 2008.
578
The 16 TF coils (Fig. 17, right) provide a 3.5 T magnetic field at the centre of the plasma. The
major radius is 1.8 m, for a minor radius of 0.5 m. The TF coil dimensions are 4.2 m height and 3 m
width. The overall dimensions of the machine are 8.8 m diameter and 8.6 m height. The peak field of
the TF coils is 7.2 T. The PF system has 13 coils, 7 in a Central Solenoid (CS) stack and 6 outer PF
coils.
All conductors use CIC superconductors with cooling by forced-flow supercritical helium with
an inlet temperature of 4.5 K and an inlet pressure of 5 bar. The TF coils use 0.78 mm diameter Nb3Sn
strands in a 2.8 mm thick Incoloy 908 conduit. Conductor dimensions are 25.65 × 25.65 mm. The
cable pattern is 34 × 6 of 486 strands. The conductor current in the TF coils is 35.2 kA and the stored
energy is 470 MJ. The critical current density of the Nb3Sn strands at 12 T and 4.2 K is higher than
750 A·mm−2. CS and PF coils use Nb3Sn in an Incoloy 908 conduit, except for the outer PF coils that
use NbTi strands in a 316LN conduit.
Fig. 17: KSTAR machine (left); TF coil (right)
4.9 JT-60SA
The magnet system for the JT-60SA [32] is developed within the framework of the ITER ‘Broader
Approach’. An agreement has been signed between Japan and Europe to upgrade the Japanese JT-60
Tokamak to a superconducting Tokamak, JT-60SA (Fig. 18), as a ‘satellite’ facility to ITER, to
develop operating scenarios and address key physics issues for an efficient start of the ITER
experiment and for research towards the future DEMOnstration Power Plant for Fusion (DEMO).
Europe is in charge of supplying TF coils for plasma confinement in the machine, as well as other
components (power supplies, current leads, cryogenic plant, ECRH, and cryostat), as an in-kind
contribution to the project.
The major plasma radius is 3.1 m and the minor plasma radius is 1.15 m. The toroidal magnetic
field at the plasma centre is 2.68 T for a peak field on the conductor of 6.5 T. The operating current is
25.7 kA and the temperature margin is 4.6 K. The total mass of the magnet is about 1300 t and the
total magnetic energy is 1060 MJ.
Eighteen TF coils, a CS, and seven Equilibrium Field (EF) coils are at the heart of the Tokamak
system. The four CS modules use Nb3Sn-type superconductors, whereas the TF and EF coils are made
of Nb−Ti. All conductors for the TF, CS, and EF coils are cooled with supercritical helium with a coil
inlet temperature of 4.5 K. The TF coils use NbTi superconductor at 5.65 T. The TF conductor is a
CICC with a circular multistage cable comprising 486 strands cabled without a central spiral. The
minimum temperature margin is 1.2 K under normal operating conditions and 1.0 K after a plasma
disruption. The operating current is 25.7 kA for the TF coils. The TF conductor has an outer
dimension of 22 × 26 mm, which is deliberately not square in order to optimize the winding pack (six
double pancakes with six turns) shape in the TF coil case. The CS operates at high field and uses
Nb3Sn superconductor.
579
P. V ÉDRINE
Fig. 18: JT-60SA Tokamak
4.10 Wendelstein 7-X

Stellarators use a single-coil system with no longitudinal net current in the plasma, and hence operate
without a transformer (continuous operation and inherent stability). The Wendelstein 7-X (W7-X)
machine is a low-shear stellarator under construction at IPP Greifswald, Germany [33, 34]. The major
radius is 5.5 m and the minor radius is 0.53 m. The magnetic field on the axis is 3 T and the stored
energy is 600 MJ. The machine has a mass of 725 t. The assembly of the W7-X magnets (Fig. 19, left)
are at an advanced stage, and the first plasma is scheduled for 2015.
The magnet system is formed by 70 superconducting coils: 50 Non-Planar Coils (NPC) (Fig. 19,
right) with 5 different geometries are wound from 960 m of Nb−Ti conductor each, and 20 Planar
Coils (PC) with 2 different geometries are wound from 390 m of conductor each. The coil windings
are manufactured with a superconducting Nb−Ti cable enclosed in an aluminium jacket. After vacuum
pressure impregnation, the windings are embedded in cast stainless steel casings, which are then
equipped with cooling systems consisting of copper strips or plates soldered to stainless steel tubes.
Following factory tests, all coils are tested at cryogenic temperature in a facility at CEA Saclay [35]as
part of the final and formal acceptance procedure.
Fig. 19: W7X machine assembly (left); NPC (right)
580
4.11 ITER
ITER (Fig. 20, left), the International Thermonuclear Experimental Reactor Programme, should
demonstrate the scientific and technological feasibility of fusion power by achieving extended burn of
D–T plasmas with steady state as the ultimate goal. The reactor is being built at Cadarache near Aix-
en-Provence in France, and the first plasma is planned for 2020. The other programme goals are to
integrate and test all essential fusion power reactor technologies and components, and to demonstrate
safety and environmental acceptability of fusion for future machines. The major radius of the machine
is 6.2 m and the minor radius is 2 m. The magnetic field at the centre of the plasma is 5.3 T.
The ITER magnetic field is composed of four systems [5]: the toroidal magnetic field system,
the CS, the PF system, and the Correction Coils (CC). They all use Nb−Ti- and Nb3Sn-based
conductors. The total mass of the system is about 10 000 t. Table 7 lists the main parameters of the
coils.
The conductors are CICC made up of superconducting and copper strands assembled into a
multistage, rope-type cable inserted into a conduit of butt-welded austenitic steel tubes (Fig. 20, right).
Table 8 gives the main parameters of the ITER conductors.
The coils are cooled with supercritical helium at an inlet temperature of 4.5 K.
Fig. 20: ITER Reactor (left); TF conductor (right)
Table 7: ITER coil characteristics
TF CS PF
Number of coils 18 1 (6 modules) 6
Dimensions (m) 14 × 9 12 × 4 8–24
Conductor type Nb3Sn CICC Nb3Sn CIC Nb3Sn CIC
Quantity (km) 88 42 65
Conductor total weight (t) 826 728 1224
SC strand weight (t) 384 122 224
Operating current (kA) 68 45 45
Operating temperature 5 4.5 4.5
(K)
Peak field (T) 11.8 12.8 6 (on PF6)
Stored energy (GJ) 41 7 4
Coil total weight (t) 6540 974 2163
581
P. V ÉDRINE
Table 8: ITER conductor parameters
TF CS PF
Conductor Nb3Sn Nb3Sn Nb−Ti
CIC configuration 226 × 6 226 × 6 240 × 6
Conductor dimension (mm) 43.7 49
Number of superconducting strands 900 576 1440
Number of copper strands 522 288
Diameter of superconducting strands 0.82 0.83 0.73
(mm)
Jacket material 316LN JK2LB 316LN
Critical current/strand (A) @12 T, 190 220
4.2 K
Critical current/strand (A) @5.6 T, 303
4.2 K
Total mass (t) 826 745 1224
Operating temperature margin (K) 0.8 0.8 1.6
Helium fraction (%) 29.7 33.5 34.2
5 Conclusion
Important developments have been made in the technology over the last 40 years for the large-scale
applications of superconductivity in terms of field strength, scale, field volume, and stored energy.
The development of new conductors has increased their capabilities to withstand high current
densities and large mechanical forces and stresses. New coil winding configurations and new coil
assembly methods enable the engineering of large magnets with sizes and stored energies never
achieved before.
The next step will be to use Nb3Sn and HTS materials to increase the magnetic field level and
magnetized volumes. Robust R&D is needed to enhance the conductor mechanical strength and to
protect the coils against quenches, which will provide an opportunity for the next generation of large-
scale superconducting magnets to emerge.
Acknowledgements
I wish to thank all those people who have contributed to this paper, and particularly: Jean-Luc
Duchateau (CEA), Arnaud Devred (ITER), François Kircher (CEA), Elwyn Baynham (RAL), Akira
Yamamoto (KEK), Lucio Rossi (CERN), Luca Bottura (CERN), and Paolo Ferracin (CERN) for all
their contributions of material.
References
[1] G. Dugan, The Tevatron: status and outlook, Proc. European Particle Accelerator Conf., Rome,
1988 (World Scientific, Singapore, 1989), vol. I, p. 223.
[2] The LHC Design Report, vol. I The LHC Main Ring, CERN-2004-003, CERN (2004).
582
[3] C. Laverick and G.M Lobell, Rev. Sci. Instrum. 36 (1965) 825–830.
[4] H.H.J. ten Kate, Int. J. Mod. Phys. A 25(15) (2010) 2933–2954
[5] N. Mitchell et al., IEEE Trans. Appl. Supercond. 22 (2012) 4200809.
[6] P. Védrine et al., IEEE Trans. Appl. Supercond. 22(3) (2012) 6001104.
[7] S. A. Gourlay, IEEE Trans. Appl. Supercond. 18(3) (2008) 1671–1680.
[8] S. D. Holmes and V. D. Shiltsev, Ann. Rev. Nucl. Part. Sci. 63 (2013) 435–465.
[9] R. Meinke, IEEE Trans. Mag. 27(2) (1991) 1728–1734.
[10] M. Anerella et al., The RHIC Magnet System for NIM, 610-20(AM-MD-311), BNL (2001).
[11] L. Rossi, IEEE Trans. Appl. Supercond. 13(2) (2003) 1221–1228.
[12] M. Peyrot et al., IEEE Trans. Appl. Supercond. 10(1) 170–173.
[13] T. Tortschanoff et al., IEEE Trans. Appl. Supercond. 16(2) (2006) 281–284.
[14] A. Yamamoto, Nucl. Instrum. Meth. A 453 (2000) 445–454.
[15] J. Purcell, The superconducting magnet system for the 12-foot bubble chamber, ANL Report
ANL/HEP 6813 (1968).
[16] J. Purcell et al., Superconducting magnet for the 15 foot NAL bubble chamber, ANL/HEP
7215, Argonne National Laboratory, Argonne National Laboratory, Argonne, IL (1973).
[17] E.U. Haebel and F. Wittgenstein, Big European Bubble Chamber (BEBC) magnet progress
report, Proc. Int. Conf. Bubble Chamber Technology, Argonne National Laboratory, June 1970
(Argonne National Laboratory , Argonne Ill, 1971), vol. 2, pp. 1126–1149.
[18] R. Apsey, IEEE Trans. Mag. 21(2) (1985) 490–493.
[19] J. Baze et al., IEEE Trans. Mag. 24(2) (1988) 1260.
[20] H. Hirabayashi, IEEE Trans. Mag. 24(2) (1988) 1256–1259.
[21] E. Acerbi et al., IEEE Trans. Mag. 24(2) (1988) 1354–1357.
[22] A. Hervé, Mod. Phys. Lett. A 25(20) (2010) 1647–1666.
[23] J.L. Duchateau, P.P. Komarek and B. Turck, in 100 Years of Superconductivity (Taylor &
Francis, London, 2011).
[24] C.D. Henning et al., IEEE Trans. Nucl. Sci. 18(4) (1971) 290–295.
[25] T.A. Kozman et al., IEEE. Trans. Mag. 19(3) (1983) 859–866.
[26] N.A. Chernoplekov, Fusion Eng. Des. 20 (1993) 55–63.
[27] N.A. Chernoplekov and N.A. Monoszon, IEEE Trans. Mag. 23(2) (1987) 826–830.
[28] B. Turck and A. Torossian, Fusion Eng. 1(1995) 393–398.
[29] J. Wei et al., IEEE Trans. Appl. Supercond. 20(3) (2010) 556–559.
[30] S. Pradhan, Superconducting magnets of SST-1 Tokamak, C and S Papers Series—International
Atomic Energy Agency IAEA CSP; FT/3-4Rb, Vilamoura, Portugal (2005).
[31] K. Kim et al., Nucl. Fusion 45 (2005) 783–789.
[32] Y. Yoshida et al., J. Plasma Fusion Res. SER. 9 (2010) 214–219.
[33] F. Wagner et al., Physics, technologies and status of the Wendelstein 7-X device, C and S
Papers Series—International Atomic Energy Agency IAEA CSP; FT/3-5, Vilamoura, Portugal
(2014).
[34] H. Viebke et al., Fusion Eng. Des. 84(7–11) (2009) 1925–1927.
[35] K. Hertel et al., IEEE Trans. Appl. Supercond. 20(3) (2010) 534–537.
583
Case Studies on Superconducting Magnets for Particle Accelerators
P. Ferracin 1
Abstract
During the CERN Accelerator School ‘Superconductivity for accelerators’,
the students were divided into 18 groups, and 6 different exercises (case
studies), involving the design and analysis of superconducting magnets and
RF cavities, were assigned. The problems covered a broad spectrum of
topics, from properties of superconducting materials to operation conditions
and general dimensions of components. The work carried out by the students
turned out to be an extremely useful opportunity to review the material
explained during the lectures, to become familiar with the orders of
magnitude of the key parameters, and to understand and compare different
design options. We provide in this paper a summary of the activities related
to the case studies on superconducting magnets and present the main
outcomes.
Keywords: superconducting magnets, dipole, quadrupole, case study.
1 Introduction
The CERN Accelerator School ‘Superconductivity for accelerators’, held in Erice from April 24 to
May 4 2013, had among its objectives to provide an overview of superconducting RF systems and
superconducting magnets for accelerators, to explain the fundamental properties of superconducting
materials, and to cover the basic physical principles behind their behaviour, as well as their design and
fabrication. In parallel to the series of lectures, a set of cases studies was assigned to the participants,
with the goals of complementing with practical work the theories described in the classes and applying
analytical formulas, scaling laws, and plots presented in the lectures to solve problems of design and
analysis.
A total of six exercises were assigned: four on superconducting magnets and two on RF
superconducting cavities. For each of the two areas, a series of subtopics was identified: in the case of
superconducting magnets, the work dealt with superconducting strands and cables, magnetic design,
operational margins, and mechanical design. In the case of RF cavities, thin films, local defects, and
tests of the properties were investigated.
In this paper we give a summary of the case studies for the superconducting magnets, firstly
providing a description of the problems and of the learning objectives, then reporting on how the
activities were organized, and finally discussing the main results and outcomes.
2 Case study problems

The four problems for the case studies on superconducting magnets were formulated as follows.
1. Low-β Nb3Sn quadrupole magnets for the HL-LHC
1
paolo.ferracin@cern.ch
P. F ERRACIN
The Large Hadron Collider (LHC) will run at 6.5–7 TeV, providing 300 fb−1 of integrated
luminosity before the end of the decade. In the 12 years of operation following 2021, CERN is
planning to upgrade the LHC to obtain ten times more integrated luminosity, i.e. 3000 fb−1. Part
of the upgrade relies on reducing the beam sizes in the Interaction Points (IPs) by increasing the
aperture of the low-β quadrupole magnets. Currently, the LHC interaction regions feature
Nb−Ti quadrupole magnets with a 70 mm aperture and a gradient of 200 T/m.
Exercise: Design a Nb3Sn superconducting quadrupole magnet with an aperture of 150 mm,
operating at 1.9 K, and aimed at the upgrade of the LHC Interaction Regions.
2. Low-β Nb-Ti quadrupoles for the HL-LHC
Exercise: Design a Nb−Ti superconducting quadrupole magnet with an aperture of 120 mm,
operating at 1.9 K, and aimed at the upgrade of the LHC Interaction Regions.
3. High-field large-aperture magnet for a cable test facility
High-field (Bbore > 10T)) magnets are needed to upgrade existing accelerators in Europe and to
prepare for new projects on a longer time-scale. Nb3Sn is currently the right candidate to meet
those objectives because of its superconducting properties and its industrial availability. Over
the very long term, further upgrades could require dipole magnets with a field of around 20 T: a
possible solution is to combine an outer Nb3Sn coil with an inner coil of High Temperature
Superconductor (HTS), which both contribute to the field. In addition, a high-field dipole
magnet with a large aperture could be used to upgrade the Facility for REception of
Superconducting CAbles FRESCA test facility at CERN, with the aim of meeting the strong
demands to qualify conductors at higher fields.
Exercise: Design a superconducting dipole with a 100 mm aperture that is capable of reaching
15 T at 1.9 K (~90% of Iss).
4. 11 T Nb3Sn dipole for the LHC collimation upgrade
The second phase of the LHC collimation upgrade will enable proton and ion beam operation at
nominal and ultimate intensities. To improve the collimation efficiency by a factor of 15–90,
additional collimators are foreseen in the room temperature insertions and in the Dispersion
Suppression (DS) regions around points 2, 3, and 7. To provide a longitudinal space of about
3.5 m for additional collimators, a solution based on the substitution of a pair of 5.5 m, 11 T
dipoles for several 14.3 m, 8.33 T LHC Main Bending dipoles (MB) is being considered.
Exercise: Design a Nb3Sn superconducting dipole with a 60 mm aperture and an operational
field (80% of Iss) at 1.9 K of 11 T.
For each of the four exercises, a set of more specific questions was assigned as follows.
• Determine the maximum bore field/gradient and coil size (using sector coil scaling
laws).
• Define strands and cable parameters.
− Strand diameter and number of strands, Cu/SC ratio, pitch angle, cable width,
cable mid-thickness and insulation thickness, filling factor κ.
• Determine load-line (no iron), ‘short sample’ conditions, operational conditions (80% of
Iss), and margins:
− jsc_ss, jo_ss, Iss, Bbore_ss. or Gss, Bpeak_ss;
− jsc_op, jo_op, Iop, Bbore_op, or Gop, Bpeak_op;
− margins in T, jsc, Bpeak.
586
C ASE S TUDIES ON S UPERCONDUCTING M AGNETS FOR PARTICLE ACCELERATORS
• Compare ‘short sample’ conditions, ‘operational’ conditions, and margins if the same
Nb3Sn (Nb−Ti) magnet uses a Nb−Ti (Nb3Sn) superconductor.
• Define a possible coil lay-out to minimize field errors.
• Determine electromagnetic forces Fx and Fy and the accumulated stress on the coil mid-
plane in the operational conditions (80% of Iss).
• Evaluate dimensions of iron yoke, collars, and shrinking cylinder, assuming that the
support structure is designed to reach 90% of Iss.
In addition, the students were asked to compare and evaluate different designs and technological
options currently under investigation in the superconducting magnet community.
• High-temperature superconductor: YBCO vs. Bi2212.
• Superconducting coil design: block vs. cos-theta.
• Support structures: collar based vs. shell based.
• Assembly procedure: high coil pre-stress vs. low coil pre-stress.
3 Learning objectives
The general purpose of the case studies was to guide the students towards the conceptual design of a
superconducting magnet, and, more specifically, towards the definition of its key parameters and
dimensions by the use of analytical formulas and scaling laws provided during the lectures. Starting
with a set of magnet specifications, the exercises were conceived in such a way that a ’first-order’
characterization of different magnet components (strand, cable, coil, yoke, support structure) had to be
performed, and the physics behind different design options analysed.
Another characteristic of the case studies was that the specifications chosen for each problem
were those of magnets currently under development. The idea was for the students to face the design
issues that magnet designers working on different projects are dealing with right now, and for them to
be aware of the activities being performed by the different magnet groups around the world.
The first exercise referred to MQXF [1], a Nb3Sn quadrupole magnet being designed by CERN
and the US LARP collaboration [2] for the Interaction Regions of the High-Lumi LHC [3]. The
specifications of the second problem were based on MQXC [4, 5], a Nb−Ti quadrupole magnet
developed by CERN and CEA Saclay for a future upgrade of the LHC low-β quadrupoles [6] and
currently under test [7]. The 11 T magnet [8, 9], planned for the upgrade of the LHC collimation
system and under fabrication and test at FNAL and CERN, was used to define the third problem. The
final problem referred to FRESCA2 [10, 11], a large-aperture Nb3Sn dipole which is being developed
by a CERN–CEA Saclay collaboration for the upgrade of the FRESCA cable test facility.
Each of the exercises started with a request to define the overall coil dimensions for the
specified field and aperture by using the scaling laws presented in the lecture ‘Magnetic design of SC
magnets’ by E. Todesco. A detailed description of these laws can be found in [12, 13]. Then, the
students were supposed to determine the general cable properties by using plots and data that
corresponded to typical superconducting cables provided in the lectures ‘Superconductors for magnets
I-II’ by R. Flukiger and ‘Superconducting cables’ by P. Bruzzone. Once the sizes of the cable and coil
were defined, the following step was to compute magnet load-lines and field-current–temperature
margins, as well as to compare the limit performance of Nb−Ti vs. Nb3Sn magnets utilizing the
parameterization of the critical curves for Nb−Ti [14] and Nb3Sn [15] superconductors. The lecture on
magnetic design by E. Todesco also contained analytical formulas to determine the angles of the coil
blocks and Cu wedges to minimize field errors, to establish the dimensions of the iron yoke, and, as a
587
P. F ERRACIN
result, to outline a preliminary coil and magnet lay-out. The electromagnetic forces and coil stresses
were computed by using the sector-coil formulas discussed in the lecture ‘Mechanical design of SC
magnets I-II’ by F. Toral, who adopted the analytical approaches presented in [16–21]. The calculation
of the forces allows an estimation of the thickness of outer cylinder, thus completing a simplified
conceptual design of the magnet.
4 Organization of the activities

The students were divided into 18 working groups with 5 to 6 students per group. As there were 6
exercises, the same problem was assigned to 3 groups. The group members were chosen to have
different backgrounds and expertise, and they all came from different universities/institutions. Two
afternoons were devoted to the case studies, and all the activities took place in a common room. Such
a set-up significantly facilitated the interaction among the different groups and between students and
teachers.
The schedule of the school was organized so that all the material required for the exercises was
presented and explained in lectures that preceded the case study afternoons. This approach resulted in
the exclusions from the case study topics of several subjects that were treated in the latter lectures of
the school, such as, for example, quench protection, cryogenics, and fabrication issues. On the last
day, the students were requested to give a 10 minute presentation summarizing the work to the other
groups and lecturers.
5 Outcomes and conclusions

The afternoon sessions dedicated to the case studies were characterized by an extremely active
participation by all the students, with a very high level of interaction within each group, among
groups, and with the lecturers. Indeed, the teachers clarified some critical aspects of the problems and
corrected possible errors during the initial set-up of the solutions. In addition, the lecturers assisted in
finding the useful formulas for the exercise amongst the large quantity of material provided during the
10 days of classes. All 18 groups completed the assignments after about 8 to 10 hours of work, and
they summarized the results in the final presentations.
Although the general cable and coil parameters were easily identified, computation of the
magnet limits through the interception of the load-line with the superconductor critical curves required
more supervision. It turned out that concepts such as filling factor, overall current density and short-
sample current, which are routinely used by magnet designers, are not straightforward for newcomers,
and they needed to be defined properly and thoroughly explained during the classes. The students
demonstrated particular interest in the representation of the field in a series of harmonics and in the
minimization of the field perturbations via the subdivision of the coil in blocks and Cu wedges.
Concerning the mechanical design, the computation of the forces in operational condition was
extremely useful in pointing out the general level of stress in a superconducting coil (~30–70 MPa for
Nb−Ti, 100–150 MPa for Nb3Sn). Finally, the discussion about alternative design and material options
currently considered within the superconducting magnet community allowed the students to formulate,
and include in their presentations, interesting comparisons between the possible choices, to outline the
pros and cons, and, sometimes, to express their preferences.
The final presentations constituted a unique opportunity for the groups to describe the work
carried out during the previous days and to underline the most important conclusions of the design
analysis. The various talks were well organized, concise, and informative, and they demonstrated the
effectiveness of the case studies as a tool to review the material covered during the lectures, pick the
key analytical formulas, identify the magnitude of the different parameters, and outline a preliminary
magnet lay-out.
588
C ASE S TUDIES ON S UPERCONDUCTING M AGNETS FOR PARTICLE ACCELERATORS
References
[1] E. Todesco, H. Allain, G. Ambrosio, F. Borgnolutti, F. Cerutti, D. Dietderich, L. Salvatore
Esposito, H. Felice, P. Ferracin, G. Sabbi, P. Wanderer and R. Van Weelderen, IEEE Trans.
Appl. Superconduct. 23 (2013) 4002405.
[2] S. A. Gourlay, G. Ambrosio, N. Andreev, M. Anerella, E. Barzi, R. Bossert, S. Caspi, D.R.
Dietderich, P. Ferracin, R. Gupta, A. Ghosh, A.R. Hafalia, C.R. Hannaford, M. Harrison, V.S.
Kashikhin, V.V. Kashikhin, A.F. Lietzke, S. Mattafirri, A.D. McInturff, F. Nobrega, I.
Novitsky, G.L. Sabbi, J. Schmazle, R. Stanek, D. Turrioni, P. Wanderer, R. Yamada and A.V.
Zlobin, IEEE Trans. Appl. Superconduct. 16 (2006) 324.
[3] L. Rossi and O. Bruning, High Luminosity Large Hadron Collider: a description for the
European Strategy Preparatory Group, CERN ATS 2012–236 (2012), p. 23.
[4] S. Russenchuck, B. Auchmann, J.C. Perez, D. Ramos, P. Fessia, M. Karppinen, G. Kirby, T.
Sahner and N. Schwerg, IEEE Trans. Appl. Supercond. 21 (2011) 1674.
[5] G. A. Kirby et al., IEEE Trans. Appl. Supercond. 22 (2012) 4001804.
[6] R. Ostojic, LHC Interaction Region Upgrade – Phase I, LHC Project Report 1094 (2008).
[7] G.A. Kirby, B. Auchmann, M. Bajko, M. Charrondiere, N. Bourcey, V.I. Datskov, P. Fessia, J.
Feuvrier, P. Galbraith, A. Garcia Tabares, J. Garcia-Perez, P. Granieri, P. Hagen, C. Lorin, J.C.
Perez, S. Russenschuck, T. Sahner, M. Segreti, E. Todesco and G. Willering, IEEE Trans. Appl.
Supercond. 23 (2013) 4002105.
[8] M. Karppinen, N. Andreev, G. Apollinari, B. Auchmann, E. Barzi, R. Bossert, V.V. Kashikhin,
A. Nobrega, I. Novitski, L. Rossi, D. Smekens and A.V. Zlobin, IEEE Trans. Appl. Supercond.
22 (2012) 4901504.
[9] A.V. Zlobin, N. Andreev, G. Apollinari, B. Auchmann, E. Barzi, R. Bossert, M. Karppinen, F.
Nobrega, I. Novitski, L. Rossi, D. Smekens, D. Turrioni and R. Yamada, IEEE Trans. Appl.
Supercond. 22 (2012) 4001705.
[10] A. Milanese, M. Devauz, M. Durante, P. Manil, J.C. Perez, J.M. Rifflet, G. de Rijk and F.
Rondeaux, IEEE Trans. Appl. Supercond. 22 (2012) 4002604.
[11] P. Ferracin, M. Devaux, M. Durante, P. Fazilleau, P. Fessia, P. Manil, A. Milanese, J.E. Munoz
Garcia, L. Oberli, J.C. Perez, J.M. Rifflet, G. de Rijk, F. Rondeaux and E. Todesco, IEEE
Trans. Appl. Supercond. 23 (2013) 4002005.
[14] L. Bottura, IEEE Trans. Appl. Supercond. 10 (2000) 1054.
[15] L.T. Summers, M.W. Guinan, J.R. Miller and P.A. Hahn, IEEE Trans. Magn. 27 (1991) 2041.
[16] R. Meuser, Magnetic field for a thick cos nθ winding, Engineering Note M5254, Lawrence
Berkeley National Lab. (1978).
[17] R. Meuser, Stresses in a thick cos nθ winding, Engineering Note M5256, Lawrence Berkeley
National Lab. (1978).
[18] A. Asner, Cylindrical aperture multipoles with constant current density sector windings CERN
Internal Report, SI/Note MAE/69-15 (1969).
[19] P. Fessia, F. Regis and E. Todesco, IEEE Trans. Appl. Supercond. 17 (2007) 1269.
[20] P. Fessia, F. Regis and E. Todesco, IEEE Trans. Appl. Supercond. 19 (2009) 1203.
[21] H. Felice, P. Ferracin, S. Prestemon and E. Todesco, Superconducting magnets for particle
accelerators, US Particle Accelerator School, http://www.uspas.fnal.gov.
589
Participants
ALBERTY, L. CERN, Geneva, CH
ALTINKOK, A. A. Izzet Baysal University, Bolu, TR
ARIMOTO, Y. KEK, Ibaraki, JP
ARNAU IZQUIERDO, G. CERN, Geneva, CH
AULL, S. CERN, Geneva, CH
BAGRETS, N. Karlsruhe Institute of Technology, Karlsruhe, DE
BAJAS, H. CERN, Geneva, CH
BAYER, C. KIT, Eggenstein-Leopoldshafe, DE
BAYLISS, V. STFC Rutherford Appleton Lab, Didcot, UK
BEDNAREK, M. CERN, Geneva, CH
BERTUCCI, M. INFN, Milan, IT
BETEMPS, R. CERN, Geneva, CH
BONOMI, R. CERN, Geneva, CH
BRODZINSKI, K. CERN, Geneva, CH
BROWN, M. Florida State University, Tallahassee, US
CHAIBI, M. TTI, Santander, ES
CHECCHIN, M. National Institute of Nuclear Physics, Legnaro, IT
CORNACCHINI, A. CERN, Geneva, CH
DALLOCCHIO, A. CERN, Geneva, CH
DARVE, C. European Spallation Source, Lund, SE
DASSA, L. CERN, Geneva, CH
DEVAUX, M. CEA Saclay, IRFU/SACM, Gif sur Yvette Cedex, FR
DOBOS, D. CERN, Geneva, CH
DZITKO, H. CEA SACLAY, Ponfanger, FR
ELEFANT, F. CEA Saclay, Gif sur Yvette Cedex, FR
EOZENOU, F. CEA Saclay, IRFU/SACM, Gif sur Yvette Cedex, FR
FERNANDES, M. CERN, Geneva, CH
FERRAND, G. CEA Saclay, IRFU/SACM, Gif sur Yvette Cedex, FR
FURCI, H. CEA Saclay, Gif sur Yvette Cedex, FR
GABOURIN, S. CERN, Geneva, CH
GADE, P. V. Institut Fur Technische Physik, Eggenstein-Leopoldshafe, DE
GEITHNER, O. GSI, Darmstadt, DE
GIANNELLI, S. CERN, Geneva, CH
GILLEY, G. STFC, Didcot, UK
GLOWA, N. EPFL-CRPP, Villigen-Psi, CH
HAGEN, P. CERN, Geneva, CH
HE, S. Institute of Modern Physics, Lanzhou, CN
IIO, M. KEK, Ibaraki, JP
INGLESE, V. CERN, Geneva, CH
IZQUIERDO BERMUDEZ, S. CERN, Geneva, CH
JECKLIN, N. CERN, Geneva, CH
JENSEN, E. CERN, Geneva, CH
JUCHNO, M. CERN, Geneva, CH
KAR, S. Interuniversity Accelerator Centre, New Delhi, IN
KARIO, A. Institut Fur Technische Physik, Eggenstein-Leopoldshafe, DE
KLEINDIENST, R. Helmholtz-Zentrum-Berlin, Berlin, DE
LACKNER, F. CERN, Geneva, CH
LOUVET, M. Synchrotron-soleil, Saint Aubin, FR
LOZANO BENITO, M. CERN, Geneva, CH
591
MAEDER, J. GSI, Darmstadt, DE
MAIANO, C. INFN, Segrate, IT
MARTINELLO, M. National Institute of Nuclear Physics, Legnaro, IT
MARTINEZ DE ALVARO, T. CIEMAT, Madrid, ES
MIERAU, A. GSI, Darmstadt, DE
MIKULAS, S. CERN, Geneva, CH
MONDINO, I. CERN, Geneva, CH
MUNILLA LOPEZ, J. CIEMAT, Madrid, ES
MURANAKA, T. CERN, Geneva, CH
NAVARRO-TAPIA, M. CERN, Geneva, CH
NIU, X. Institute of Modern Physics, Lanzhou, CN
PAPKE, K. CERN, Geneva, CH
PASQUET, R. CEA Saclay, Gif sur Yvette Cedex, FR
PEREZ BERMEJO, J. TTI - NORTE, Aljaraque, ES
PORHIEL, A. SIGMAPHI, Vannes, FR
PRIEBE, A. CERN, Geneva, CH
PRINCIPE, R. CERN, Geneva, CH
PURUSHOTAMAN, S. GSI, Darmstadt, DE
ROBERTS, W. TRIUMF, Vancouver, CA
ROGER, V. CERN, Geneva, CH
ROGEZ, V. CERN, Geneva, CH
SANTIAGO KERN, R. Uppsala University, Uppsala, SE
SAPINSKI, M. CERN, Geneva, CH
SEGAL, C. National High Magnetic Field Laboratory, Tallahassee, US
SHORNIKOV, A. CERN, Geneva, CH
SPINA, T. CERN, Geneva, CH
STECKERT, T. CERN, Geneva, CH
STODEL, M. GANIL/CNRS, Caen cedex, FR
SUBLET, A. CERN, Geneva, CH
SUGANO, S. KEK, Ibaraki, JP
TAN, F. Institute of Modern Physics, Lanzhou, CN
TAN, J. CERN, Geneva, CH
TERENZIANI, G. CERN, Geneva, CH
TRUBLET, T. CEA Saclay, IRFU/SACM, Gif sur Yvette Cedex, FR
TULU, E. University of Rostock, Rostock, DE
VALLCORBA, R. CEA Saclay, IRFU/SACM, Gif sur Yvette Cedex, FR
VALUCH, C. CERN, Geneva, CH
VOGT, J. Helmholtz-Zentrum-Berlin, Berlin, DE
WEGNER, R. CERN, Geneva, CH
XU, M. Institute of Modern Physics, Lanzhou, CN
YUE, W. Institute of Modern Physics, Lanzhou, CN
YUREVICH, S. Physical Technical Institute of The National Academy of Sciences of Belarus,
Minsk, BY
ZHANG, P. CERN, Geneva, CH
ZHENG, S. Institute of Modern Physics, Lanzhou, CN
ZICKLER, T. CERN, Geneva, CH
592

CERN-Superconductivity For Accelerators

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CERN-Superconductivity For Accelerators

Uploaded by

Copyright:

Available Formats

CERN–2014–005

ORGANISATION EUROPÉENNE POUR LA RECHERCHE NUCLÉAIRE

CERN EUROPEAN ORGANIZATION FOR NUCLEAR RESEARCH

Superconductivity for Accelerators

This report should be cited as:

A contribution in this report should be cited as:

10:00 R. Bailey G. Ciovati H. Padamsee R. Flukiger F. Toral V. Parma L. Bottura F. Gomory

12:30 A D. Larbalestier B. Holzer C. Antoine R R. Flukiger F. Toral V. Parma P. Lebrun L. Rossi U

Keywords: thermodynamics, expansion of gas, heat machines, energy, phase

2 Equilibrium and equation of state

4 Work, heat, and the first law of thermodynamics

Fig. 2: Work of a closed system (a) and an open system (b)

4.2 Internal energy

4.3 First law of thermodynamics and energy balances

This can be rewritten as follows:

Another function of state, called enthalpy, is then defined as

5 Entropy and second law of thermodynamics

where ΔS e = ∫ δ Q / T relates to the heat exchange and S i is an entropy production term:

5.1 Entropy of an isolated system

5.2 Principle of increase of entropy

6 Thermodynamic temperature and the zeroth law of thermodynamics

At thermodynamic equilibrium, maximization of the entropy requires that δ S = δ S e + S i = 0 + 0 = 0

8 Relating entropy to functions of state

8.1 Entropy with internal energy

8.2 Thermodynamic potentials

dU T dS − p dV , we deduce an appropriate relation between the first derivatives of T and p :

It turns out that

The relationship between C p and CV is expressed by Meyer’s relation:

which yields, for an ideal gas,

where γ is the ratio of the heat capacities:

11 Some thermodynamic transformations

11.1 Adiabatic expansion of a gas

11.2 Joule–Kelvin (Joule–Thomson) expansion of a gas

where α is the coefficient of thermal expansion:

For the ideal gas, α T = 1 and thus (∂T / ∂p ) H =

11.3 Thermodynamic transforms of an ideal gas

energy balance (first law): ∆U = U f − U i = 0 ⇒ W + QC + QH = 0 ;

12.1 Heat engine

i.e., making use of equation (21),

12.2 Heat pump or refrigerator

12.3 The Carnot cycle

Fig. 6: Carnot cycle

Fig. 8: Vapour compression heat machine

13 Maximum available work

The term −TEXT S i represents a destruction of exergy.

For a stationary state and a closed system, it reduces to

14 States of matter and phase changes

Fig. 9: p −T phase diagram

which yields, by subtraction,

0 = (vgas − vliq )dp − ( sgas − sliq )dT ,

15 Thermodynamics of magnetic systems: type I superconductor phase transition

Experimentally, Hˆ CM (T ) = Hˆ CM (0) − Hˆ CM (0) ⋅ (T / TCM ) 2 (see Fig. 11(b)), dHˆ CM2

However, the heat capacity C = δ Q / dT =T dS / dT = −T (∂ 2G / ∂T 2 ) p , a second derivative of the

Therefore at T = TCM the superconducting → normal transition is a second-order transition.

16 Second-order phase transitions

An example of a second-order transition is the normal–superfluid transition of liquid helium-4.

Appendix A: Thermodynamic processes for an ideal gas

Process Isochoric Isobaric Isothermal Adiabatic Polytropic

Work WAB − p (VB − VA ) VB γ −1

Introduction to Transverse Beam Dynamics

Keywords: accelerators, transverse beam dynamics, ring geometry,

2 Geometry of the ring

As an example let us consider a magnetic B field of 1 T, which is a conservative size. The