REVIEWS OF MODERN PHYSICS, VOLUME 76, OCTOBER 2004

Physics of magnetically confined plasmas

Allen H. Boozer
Department of Applied Physics and Applied Mathematics, Columbia University,
New York, New York 10027, USA
共Published 28 January 2005兲

The physics of magnetically confined plasmas has had much of its development as part of the program
to develop fusion energy and is an important element in the study of space and astrophysical plasmas.
Closely related areas of physics include Hamiltonian dynamics, kinetic theory, and fluid turbulence. A
number of topics in physics have been developed primarily through research on magnetically confined
plasmas. The physics that underlies the magnetic confinement of plasmas is reviewed here to make it
more accessible to those beginning research on plasma confinement and for interested physicists.

CONTENTS Appendix: General Coordinate Systems 1136

References 1138

I. Introduction 1071
II. Fusion Energy 1072 I. INTRODUCTION
III. Magnetic-Field Lines 1074
A. Relation to Hamiltonian mechanics 1075 A plasma is a gas in which charged particles are of
B. Methods of forming toroidal magnetic surfaces 1079 sufficient importance for the gas to be a good electrical
C. Evolution of magnetic-field lines 1082 conductor. Ordinary matter becomes ionized and forms
IV. Confinement Systems 1085 a plasma at temperatures above about 5000 K, and most
V. Efficiency of Magnetic Confinement 1087 of the visible matter in the universe is in the plasma
A. Equilibrium limits 1088 state. The high electrical conductivity implies that cur-
B. Stability limits 1092 rents can flow in a plasma. These currents can interact
1. Current-driven instabilities 1093
with magnetic fields to produce the forces that are
2. Resistive stability 1095
needed for confinement.
3. Robustness of magnetic surfaces 1096
The physics of plasma confinement using magnetic
C. Stability analyses 1097
fields has been driven intellectually by the program to
develop fusion energy. Fusion has provided a focus for
1. Expressions for ␦W 1098
the research, but much of the physics that has been de-
2. Pressure-driven ballooning modes 1099
veloped has far broader applications—most obviously to
3. Minimization of ␦W 1100
space and astrophysical plasmas. In addition, the physi-
D. Interaction of plasmas with coils 1101
cal insights and concepts are of intrinsic scientific impor-
1. Freedom of coil design 1102
tance.
2. Plasma response to coil changes 1103
A number of topics in physics have had their primary
VI. Radiation and Transport 1106
development through research on plasma confinement.
A. Electromagnetic radiation 1106
These include:
B. Kinetic theory 1108
C. Landau damping and quasilinear diffusion 1111 共1兲 The relation between the field lines of divergence-
D. Drift kinetic theory 1113 free fields and Hamiltonian mechanics.
1. Alfvén’s guiding-center velocity 1114 共2兲 The constraints of magnetic helicity conservation on
2. Magnetic-moment conservation 1116 the rapid evolution of magnetic fields.
3. The drift Hamiltonian 1117
共3兲 Collisionless relaxation phenomena.
4. The action invariant J 1118
E. Particle trajectories and transport 1119 共4兲 Simplified kinetic equations that are based on adia-
1. Confinement of particle trajectories 1120 batic invariants of classical mechanics.
2. Transport at low collisionality 1122 共5兲 The theory of small-amplitude and short-
3. Power required for maintaining fields 1123 wavelength turbulence, called microturbulence.
F. Microstability 1124
G. Gyrokinetic theory 1128 共6兲 The experimental observation and theoretical expla-
H. Alfvén instabilities 1131
nation of transport barriers where plasma microtur-
1. Continuum Alfvén wave damping 1132
bulence is stabilized by a strong gradient in the
2. The weakly damped gap mode 1132
plasma flow.
3. The particle-Alfvén wave interaction 1134 共7兲 The demonstration for shear Alfvén waves that con-
VII. Plasma Edge 1135 tinuum modes are rapidly damped but discrete
Acknowledgments 1136 modes exist that are weakly damped and can be de-

0034-6861/2004/76共4兲/1071共71兲/$50.00 1071 ©2005 The American Physical Society

1072 Allen H. Boozer: Physics of magnetically confined plasmas
stabilized by particles with a speed near the phase
velocity of the waves. Shear Alfvén waves propagate
by twisting the magnetic-field lines that are embed-
ded in a plasma.
One of the goals of this review is to make these and
other topics in the physics of magnetic confinement ac-
cessible to a broader audience.
The physics of magnetic confinement is covered in a
number of books. Probably the most complete is Toka-
maks 共Wesson, 2004兲, which emphasizes the relation be-
tween theory and experiment. Fundamentals of Plasma
Physics and Controlled Fusion 共Miyamoto, 1997兲 gives
the details of many of the classic derivations. The
Theory of Toroidally Confined Plasmas 共White, 2001兲 of-
fers insights on a number of fundamental phenomena,
and a standard textbook is Introduction to Plasma Phys-
ics 共Goldston and Rutherford, 1995兲. The International
Thermonuclear Experimental Reactor 共ITER兲 design
team summarized the physics of tokamaks in a series of
articles in the December 1999 issue of the journal
Nuclear Fusion. The first of these articles is an overview
共ITER Physics Basis Editors et al., 1999兲.
Despite the number of books and articles written on
plasma confinement a review of the fundamental physics
that includes derivations of major results has been lack-
ing up till now. This review is designed to fill that gap
while being accessible to the general physics community.
Several important types of confined plasmas are not
covered in this review. Magnetically confined plasmas
occur in many space and astrophysical situations, and
much of the physics is shared with laboratory plasmas.
However, this review does not provide sufficient context
for understanding the breadth of applications to space
and astrophysical plasmas. Some texts that do provide
such context include Basic Space Plasma Physics by
Baumjohann and Treumann 共1996兲, Physics of Space
Plasmas by Parks 共1991兲, and Plasma Astrophysics by
Tajima and Shibata 共1997兲. Bisnovatyi-Kogan and Love-
lace 共2001兲 have written a major review of the plasma
issues associated with accretion disks, and Ferrari 共1998兲
has reviewed extragalactic jets. Laboratory plasmas in
which all of the magnetic-field lines in the plasma inter-
sect the chamber walls are also not covered. Plasmas
either flow out of such open systems at a thermal speed
or have a confinement time comparable to a collision
time. The fast Z pinch is an open plasma confinement
system, which has prominence from its use as a driver
for inertial fusion 共Ryutov et al., 2000兲. Open plasma
confinement systems are also important for materials
processing 共Lieberman and Lichtenberg, 1994兲.
Other major plasma topics that are not covered are
waves, which are of particular importance for plasma
heating, and the methods of measuring plasma param-
eters, which is the topic of plasma diagnostics. Standard
books on waves in plasmas are those of Stix 共1992兲;
Brambilla 共1998兲; and Swanson 共2003兲. Plasma diagnos-
tics are based on a broad range of physics principles.
The standard text is Principles of Plasma Diagnostics
共Hutchinson, 2003兲, and reviews have been written by
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
Gentle 共1995兲 and the ITER Physics Export Group on
Diagnostics et al. 共1999兲 in the Nuclear Fusion series on
ITER.
What is covered in this review is the physics of the
magnetic confinement of plasmas that are flowing slowly
compared to sonic speeds and at each point in the
plasma are close to thermodynamic equilibrium, which
implies the confinement is long compared to collision
times. Although the fundamental topic is not fusion, an
understanding of the basic features of magnetically con-
fined fusion plasmas is useful for placing research on
plasma confinement in context. These features are ex-
plained in Sec. II.
A review of the physics of magnetically confined plas-
mas contains too much material to be absorbed at a uni-
form level. Therefore, paragraphs marked by bullets are
given near the beginning of most sections. Readers are
encouraged to read just the bulleted sections of the en-
tire review before working through the details of the
sections. The main idea in each section can be deter-
mined by reading the first sentence of each paragraph. A
result is generally given before the derivation, so the
derivations can be skipped. However, the derivations are
sufficiently complete that a reader should be able to re-
construct them using a table of identities of vector cal-
culus. The review is designed to be read at various levels
of detail, and that is the way it should be read.
II. FUSION ENERGY
• The goal of research on fusion energy is a commer-
cially viable source of energy. The primary research
effort is on the fusion of two isotopes of hydrogen,
deuterium and tritium, to form an isotope of helium,
an alpha particle, and a neutron.
• Fundamental considerations imply that in a fusion
power plant the particle distribution functions would
be close to local Maxwellians with a temperature of
about 20 keV and a density of approximately 2
⫻ 1020 nuclei/ m3, and the plasma would have the
shape of a torus with the minor radius of the torus a
few meters in length.
• Technical limits on the magnitude of the magnetic
field that can be used for confining fusion plasmas
make the efficiency of the utilization of the magnetic
field a central issue.
• Most of the freedom in the design of fusion plasmas
is in the plasma shape, which means freedom in the
design of the coils that surround the plasma. About
50 shape parameters can be controlled, though only
about four of these are consistent with an axisym-
metric torus.
Fusion is a potential source of energy, which is essen-
tially unlimited in quantity and produces no greenhouse
gases. The fusion reaction that appears technically easi-
est to harness is between two isotopes of hydrogen, deu-
terium and tritium, with the product being ordinary he-
lium 共an alpha particle兲 and a neutron. Deuterium
 Allen H. Boozer: Physics of magnetically confined plasmas
occurs naturally in water in sufficient abundance as to be
an essentially unlimited resource. Tritium has a half-life
of 12 years and must be produced through an interaction
of the neutron with isotopes of lithium. Consequently,
the fuel for fusion is deuterium and lithium. The waste
products of fusion energy production need not be radio-
active in principle, but in practice some radioactive
products will be produced. The level, lifetime, and na-
ture of the radioactive waste are dependent on the use
of appropriate materials and the cleverness of the de-
sign. The fusion reaction is easily terminated, so a run-
away reaction is not a safety concern. Sheffield 共1994兲
has written a general review of fusion systems and Baker
et al. 共1998兲 have discussed fusion systems with a focus
on the important issue of the materials that would be
used in fusion power plants.
The development of fusion power is paced by both
physics and engineering considerations. A magnetically
confined plasma with an essentially self-sustaining fusion
burn has not yet been produced, but this is thought to be
essentially an issue of the size of the experiments that
have been undertaken. A number of proposals have
been made for a burning plasma experiment; the most
ambitious currently under consideration is the Interna-
tional Tokamak Experimental Reactor 共ITER; Aymar,
2000兲. Many studies have claimed fusion power could be
economically competitive, but the compatibility of the
physics and the engineering with energy at an acceptable
cost will remain a primary issue until demonstration
power plants are built.
Fundamental considerations imply that toroidal fusion
power systems would have a power output of order a
gigawatt. First, the structure surrounding the plasma
must have a minimum thickness, about 1.5 m, for the
fusion neutrons to convert lithium into tritium and heat
as well as shield the external world from radiation. Sec-
ond, until material limitations arise at a power density of
several megawatts per square meter of chamber wall
area, power becomes cheaper the higher the power den-
sity. These two considerations plus the need for a toroi-
dal plasma, which is discussed in Sec. III, imply that fu-
sion energy is most economical in units of approximately
one gigawatt of electrical power. A larger or smaller unit
size would imply a larger or a smaller aspect ratio of the
toroidal plasma.
The electrostatic repulsion of the deuterium and the
tritium nuclei sets the energy scale at which fusion reac-
tions occur, which is a thermal energy of a few tens of
kilovolts, which is a few hundred million degrees kelvin.
For the plasma to maintain its burning state, the rate of
production of high-energy alpha particles must be suffi-
cient to offset the energy losses of the plasma through
electromagnetic radiation and thermal transport through
diffusion; see Sec. VI. Since neutrons are electrically
neutral they have no significant interaction with the
plasma or the magnetic field. The maximum rate of en-
ergy production, at fixed plasma pressure, occurs at a
temperature of approximately 20 kV, or 2.3⫻ 108 K. For
a steady burn at a temperature of 20 kV, the number
density of nuclei times the energy confinement time
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1073
must equal 2 ⫻ 1020 共nuclei/ m3兲 sec. At that temperature
the deuterium and tritium are completely ionized and,
therefore, form a plasma. These considerations would be
changed if the plasma reaction rate could be substan-
tially increased. Fisch and Herrmann 共1994兲 have pro-
posed a method for increasing the reaction rate by en-
hancing the transfer of energy from the fusion alphas to
the fusing ions.
The required power density on the chamber walls for
economic electric power determines the plasma density,
about 2 ⫻ 1020 nuclei/ m3, and the plasma pressure, about
ten atmospheres.
Plasmas of interest for magnetically confined fusion
are in a paradoxical collisionality regime. The rate for
Coulomb collisions, which relax the particle distribution
functions to Maxwellians, times the energy confinement
time is about 100 for the ions and 10 000 for the elec-
trons. The rapidity of collisions implies the distribution
functions are close to Maxwellian. However, the mean
free path of the thermal particles, about 10 km, is enor-
mous compared to the size of the plasma. This paradoxi-
cal collisionality regime places requirements on the
quality of the collisionless particle trajectories in order
to have adequate confinement; see Sec. VI.E.1.
Three considerations set a minimal level for the
magnetic-field strength, which is a few tesla.
共1兲 The magnetic-field pressure must be significantly
larger than the plasma pressure to provide a stable
force balance; see Sec. V.
共2兲 A charged particle moves in a circle about magnetic-
field lines, and the radius of this circle, the gyrora-
dius ␳, must be sufficiently small that the high-
energy alpha particles remain in the plasma and
heat it; see Sec. VI.E.1.
共3兲 The thermal transport coefficients, which are re-
duced by an increase in the magnetic-field strength,
must be sufficiently small to obtain the required en-
ergy confinement time; see Sec. VI.F.
Engineering considerations make it desirable to use
the lowest magnetic-field strength that is consistent with
the physics requirements. The technical limitations that
arise for magnetic fields larger than about 10 T make
magnetic confinement fusion more difficult than if
higher fields could be used. If higher fields were avail-
able, power losses from electron cyclotron radiation,
which set an upper limit on the electron temperature,
would be of increasing importance. A discussion of elec-
tron cyclotron losses is given in Sec. VI.A and by Alba-
jar, Bornatici, and Engelmann 共2002兲.
In addition to fusion using magnetic fields to confine
the plasma, a large research program is being pursued
on inertial confinement fusion. The physics is described
in the book Inertial Confinement Fusion by Lindl 共1998兲.
Inertial confinement means the plasma is confined for a
time of order a / Cs, with Cs the speed of sound and a the
plasma radius. Confinement is determined by the bal-
ance between the inertial and the pressure forces. Mag-
netically confined plasmas are in essentially a static bal-
1074 Allen H. Boozer: Physics of magnetically confined plasmas
ance between the magnetic and the pressure forces. The
characteristic plasma density during the burn period of
an inertially confined plasma is of order 1032
particles/ m3, which is about 12 orders of magnitude
larger than the density of a magnetically confined fusion
plasma. The size of the pellets that are imploded in in-
ertial confinement has a millimeter scale, while magnetic
fusion plasmas have a scale of meters. However, the cen-
tral plasma temperatures are roughly the same in mag-
netically and inertially confined fusion plasmas in order
to obtain an optimal reaction rate.
The focus of much of the research on magnetic fusion
during the last decade has been on understanding and
extrapolating of transport processes in confined plasmas.
Energy transport means the loss of energy by plasma
processes rather than by electromagnetic radiation. Con-
fined plasmas are usually, but not always, in a state in
which microturbulence dominates the transport pro-
cesses. Microturbulence 共Sec. VI.F兲 means the fluctua-
tions have a wave number that is greater than, or of
order of, the inverse of an ion gyroradius ␳i and an am-
plitude of order the ion gyroradius to system size ␳i / a
⬍ 10−2. The extrapolation of microturbulent transport
from existing to future experiments and to fusion power
plants is essential for the scientific planning of fusion
research and is an issue in the feasibility of fusion power.
The focus of innovation in fusion research is on ways
to obtain more control over fusion plasmas and on im-
proved fusion systems. A plasma equilibrium is deter-
mined by the shape of the plasma, the magnetic-field
strength, and the profiles of the plasma current and pres-
sure; see Sec. V.A. In a power plant, economics dictates
that only about 5% of the fusion power can be used for
control. Consequently, the plasma current and pressure
profiles are largely determined by internal plasma phe-
nomena. The largest element of control that the de-
signer has over the plasma and its performance is on the
plasma shape. Control of the plasma shape is actually
control over the design of the coils that surround the
plasma. As shown in Sec. V.D.1, coil constraints limit the
designer to about 50 shape parameters, of which only
four 共aspect ratio, ellipticity, triangularity, and square-
ness兲 are consistent with an axisymmetric torus. The axi-
symmetric tokamak 共Sec. IV兲 is the most advanced
plasma confinement configuration and is the configura-
tion of choice for all major designs for experiments with
a fusion burn. Careful consideration of the four axisym-
metric shape parameters is known to be essential for
attractive tokamaks. However, tokamaks set more than
90% of the available shape parameters to zero. The use
of a larger set of shape parameters allows the designer
to sidestep issues in tokamak design, such as current
profile control, and is being studied under the topic of
stellarator research; see Sec. IV. Other elements of
plasma control that may be available include the density
profile, through clever fueling techniques, and the inter-
action between the plasma edge and the surrounding
walls.
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
III. MAGNETIC-FIELD LINES
• Near-Maxwellian plasmas are in force balance be-
ជp
tween the pressure and the magnetic forces, ⵜ
ជ.
= ជj ⫻ B
• The magnetic-field lines confining a near-Maxwellian
plasma must lie in the surfaces of constant pressure,
Bជ ·ⵜ
ជ p = 0. These surfaces can be spatially bounded
only if they have the topological form of a torus. The
pressure p共␺t兲 is a function of ␺t, which is the flux of
the toroidally directed magnetic field that is enclosed
by the pressure surfaces.
In plasmas of fusion interest, the ions and the elec-
trons are in near-Maxwellian distributions, which implies
the plasma has the pressure of an ideal gas, p = nT,
where T is the temperature in energy units and n = ne
+ ni is the sum of the number of electrons and ions per
cubic meter. Plasma confinement implies a pressure gra-
dient, and ⵜ ជ p is a force per unit volume. This force is
balanced by the electromagnetic force produced by the
cross product of the current density in the plasma and
the magnetic field,
ជ.
ⵜជ p = ជj ⫻ B 共1兲
Equation 共1兲 gives the force equilibrium of a near-
Maxwellian plasma and places fundamental constraints
on magnetic confinement systems.
A confined near-Maxwellian plasma is constrained to
have the topological form of a torus. Why is this? Equi-
librium implies B ជ ·ⵜ
ជ p = 0; a magnetic-field line must lie in
a constant-pressure surface through its entire length. For
example, a constant-pressure surface cannot be a sphere
and satisfy Eq. 共1兲. The magnetic-field lines in a
constant-pressure surface resemble strands of hair. The
hair on a topological sphere, such as a person’s head,
always has a crown, a place where the hair spirals out
from a point. A crown is a point at which Eq. 共1兲 cannot
hold. A theorem of topology says a nonsingular vector
field Bជ 共xជ 兲 can be everywhere tangent to a spatially
bounded function, namely, p共xជ 兲, in only one shape, the
torus.
Since toroidal surfaces are central to the whole theory
of plasma confinement, it is important to have a method
for describing them as a basis for terminology. The sim-
plest description uses 共R , ␸ , Z兲 cylindrical coordinates
with spatial positions given by
xជ 共r, ␪, ␸兲 = R共r, ␪, ␸兲R̂共␸兲 + Z共r, ␪, ␸兲Ẑ. 共2兲
Expressed in terms of the Cartesian unit vectors, the
radial unit vector of cylindrical coordinates is R̂共␸兲
= x̂ cos ␸ + ŷ sin ␸ and its derivative dR̂ / d␸ is the unit
vector ␸ˆ . Simple circular toroidal surfaces are given by
R = Ro + r cos共␪兲 and Z = −r sin共␪兲. The constant Ro is the
major radius, r is the local minor radius, and ⑀ = r / Ro,
which must be less than one, is the local inverse aspect
ratio. A function, such as xជ 共r , ␪ , ␸兲, which determines the
 Allen H. Boozer: Physics of magnetically confined plasmas
FIG. 1. 共Color兲 Magnetic fluxes and currents defined using the
cross-sectional area for the toroidal flux ␺t and the current I
and using the central hole of the torus for the poloidal flux ␺p
and the current G. The poloidal angle is ␪ and the toroidal
angle is ␸. 共R , ␸ , Z兲 are ordinary cylindrical coordinates.
positions in space that are associated with three quanti-
ties 共r , ␪ , ␸兲, defines an 共r , ␪ , ␸兲 coordinate system. The
Appendix gives the theory of general coordinates that is
required for understanding this review. The polar angle
of cylindrical coordinates, ␸, is also the toroidal angle.
The angle ␪ is called the poloidal angle, and r is a radial
coordinate that labels the various toroidal surfaces.
A. Relation to Hamiltonian mechanics
ជ 共xជ 兲, or any other
• The field lines of a magnetic, B
divergence-free field are the trajectories of a Hamil-
tonian. 共A short tutorial on Hamiltonians is given
below just after the bullets.兲 If ␸共xជ 兲 is a toroidal
ជ ·ⵜ
ជ ␸ ⫽ 0, then the magnetic-field lines
angle, and B
are given by a one-and-a-half-degree-of-freedom
Hamiltonian, which is the poloidal magnetic flux
␺p共␺t , ␪ , ␸兲. The canonical momentum is the toroidal
flux ␺t, the canonical coordinate is a poloidal angle ␪,
and the canonical time is the toroidal angle ␸; see
Fig. 1. The full Hamiltonian system consists of the
Hamiltonian, ␺p共␺t , ␪ , ␸兲, and the transformation
function, xជ 共␺t , ␪ , ␸兲, which gives the spatial location
of each canonical coordinate point 共␺t , ␪ , ␸兲.
• When the magnetic-field lines lie in toroidal surfaces,
the canonical coordinates can be chosen so that the
poloidal flux is a function of the toroidal flux alone,
␺p共␺t兲. The twist of the field lines, which is called the
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1075
FIG. 2. 共Color兲 The solution to Eq. 共15兲 plotted for an m = 3
magnetic island in the ␸ = 0 plane. The surface 兩s兩 = 1 is the
island separatrix, and the limit 兩s兩 → 0 gives the island O point.
The half width of the island is ␦.
rotational transform, is ␫ ⬅ d␺p / d␺t. When the rota-
tional transform is a rational number ␫ = n / m, the
field lines close on themselves after n poloidal and m
toroidal transits of the torus. The coordinates in
which the poloidal flux is a function of the toroidal
flux alone are called magnetic coordinates and trivi-
alize the solution of the magnetic differential equa-
ជ ·ⵜ
ជ f = g, which arises frequently in plasma phys-
tion B
ics.
• If a magnetic field Bជ 共xជ 兲 that forms perfect surfaces is
0
perturbed by a field ␦B ជ , then the magnetic surfaces
can split to form islands 共Fig. 2兲, where the rotational
transform is a rational number, ␫ = n / m. If
ជ ·ⵜ
ជ ␺ /B
ជ ·ⵜជ ␸ = 兺b exp关i共n␸ − m␪ 兲兴, the width of
␦B
t 0 mn m
the island that splits the surface ␫ = n / m is propor-
tional to 冑兩bmn兩. If islands from different rational sur-
faces 共␫ different rational numbers兲 are sufficiently
wide to overlap, the magnetic-field lines in that re-
gion will come arbitrarily close to every point in a
volume of space rather than lying on surfaces. Such
field-line trajectories are said to be stochastic.
Plasma confinement is destroyed in regions of sto-
chastic field lines.
• The opening of an island is a singular process in a
toroidal plasma. Let p0共xជ 兲 be any function that satis-
ជ ·ⵜ
fies B ជ p = 0. Without an island dp / d␺ is gener-
0 0 0 t
ally nonzero near the resonant rational surface ␫
= n / m, but with an arbitrarily small island only the
derivative dp0 / d␺h can be nonzero, where the helical
flux is defined by d␺h ⬅ 共␫ − n / m兲d␺t.
The nested toroidal surfaces of magnetic-field lines
are reminiscent of the tori formed by particle trajecto-
ries of an integrable Hamiltonian H共p , x , t兲 that is peri-
odic in time with T the period, H共p , x , t + T兲 = H共p , x , t兲.
Such Hamiltonians are said to have one-and-a-half de-
grees of freedom. An excellent reference for the Hamil-
tonian mechanics used in this review is the book Regular
1076 Allen H. Boozer: Physics of magnetically confined plasmas
and Chaotic Dynamics 共Lichtenberg and Liberman,
1992兲.
The Hamiltonian formulation of mechanics follows
from Newton’s equations of motion, dpជ / dt = −ⵜ ជ V共xជ , t兲
and dxជ / dt = pជ / m. Let H共pជ , xជ , t兲 = p / 2m + V共xជ , t兲; then
2
Newton’s equations of motion are reproduced by the set
of ordinary differential equations dpជ / dt = −⳵H / ⳵xជ and
dxជ / dt = ⳵H / ⳵pជ . Any set of ordinary differential equa-
tions of this form is said to be Hamiltonian with xជ the
canonical coordinate, pជ the canonical momentum, and t
the canonical time, regardless of the functional form of
the Hamiltonian, H共pជ , xជ , t兲. If xi are the components of
the canonical coordinate xជ , and pi are the components of
the canonical momentum pជ , then Hamilton’s equations
are dxi / dt = ⳵H / ⳵pi and dpi / dt = −⳵H / ⳵xi. The super-
scripts number the components and are not powers. In
Hamiltonian systems with one-and-a-half degrees of
freedom xជ and pជ have only one component each.
Magnetic-field lines are the trajectories of a one-and-
a-half-degree-of-freedom Hamiltonian 共Kerst, 1964;
Whiteman, 1977; Boozer, 1983; Cary and Littlejohn,
1983兲. A magnetic-field line moves through 3-space
共x , y , z兲, so the same number of coordinates are involved
as in the 共p , x , t兲 space of Hamiltonian mechanics. The
only problem is that Cartesian coordinates 共x , y , z兲 are
not the canonical coordinates for magnetic-field lines. To
obtain canonical coordinates, one must first write the
magnetic field in what is called the symplectic form
共Boozer, 1983兲,
ជ = ⵜជ ␺ ⫻ ⵜជ ␪ + ⵜជ ␸ ⫻ ⵜជ ␺ ,
2␲B
t p
with ␪ and ␸ arbitrary coordinates except B ជ ·ⵜ
ជ ␸ ⫽ 0. In
practice, ␺t is generally the magnetic flux of the toroi-
dally directed field, ␪ is a poloidal, and ␸ is a toroidal
ជ ·ⵜ
ជ ␸ ⫽ 0, points in 3-space can be
angle 共Fig. 1兲. When B
described using 共␺t , ␪ , ␸兲 as coordinates, which means
points in space are given as the function xជ 共␺t , ␪ , ␸兲. Us-
ing 共R , ␸ , Z兲 cylindrical coordinates 关Eq. 共2兲兴, one de-
fines points in space by giving R and Z as functions of
the 共␺t , ␪ , ␸兲 coordinates, xជ 共␺t , ␪ , ␸兲 = R共␺t , ␪ , ␸兲R̂共␸兲
+ Z共␺t , ␪ , ␸兲Ẑ.
The 共␺t , ␪ , ␸兲 coordinates are the canonical coordi-
nates of the magnetic-field lines. The mathematics of
general coordinate systems is described in the Appen-
dix. When B ជ ·ⵜ
ជ ␸ ⫽ 0, the Jacobian J of the canonical
ជ ␺ · 共ⵜ
ជ ␪⫻ⵜ
ជ ␸兲
coordinates is finite, that is, 1 / J ⬅ ⵜ t
ជ ជ
= 2␲B · ⵜ␸. In these coordinates, magnetic-field lines,
ជ 共xជ 兲, are easily shown
which are the solutions to dxជ / d␶ = B
to have the canonical form
d␺t ⳵ ␺ p共 ␺ t, ␪ , ␸ 兲
=− ; 共4兲
d␸ ⳵␪
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
d ␪ ⳵ ␺ p共 ␺ t, ␪ , ␸ 兲
= 共5兲
d␸ ⳵␺t
using Eqs. 共A4兲 and 共3兲. The canonical momentum is the
toroidal magnetic flux, ␺t = 兰B ជ · daជ , which is the mag-
␸
netic field integrated over the cross section of a
constant-␺t torus 共Fig. 1兲. The canonical coordinate is a
poloidal angle ␪, and the canonical time is a toroidal
angle ␸. The Hamiltonian is the poloidal flux that goes
down through the hole in a constant-␺p torus, ␺p
= −兰B ជ · daជ 共Fig. 1兲. The field-line Hamiltonian ␺ can be
␪ p
a function of clock time t in addition to ␺t, ␪, and ␸ 共Sec.
III.C兲, but clock time is a parameter in the Hamiltonian
description and not one of the canonical variables.
The derivation of the symplectic representation of the
magnetic field, Eq. 共3兲, clarifies the interpretation. Let
共r , ␪ , ␸兲 be an arbitrary set of well-behaved coordinates.
For example, let the radial and vertical coordinates of
cylindrical coordinates, Eq. 共2兲, have the form
R共r , ␪ , ␸兲 = Ro + r cos ␪ and Z = −r sin ␪, which makes r
constant on circular toroidal surfaces. In three dimen-
sions a vector can have only three independent compo-
nents, so any vector can be written in the form
冉 冊 冉 冊
ជ = ⵜជ g + ␺ ⵜជ ␪ ␸
A t − ␺pⵜជ . 共6兲
2␲ 2␲
The functions of position, g, ␺t, and ␺p, represent the
three components of A ជ . If A
ជ is interpreted as the vector
potential of the magnetic field, then its curl gives the
symplectic representation of the magnetic field 关Eq. 共3兲兴.
The theory of one-and-a-half-degree-of-freedom
共3兲
Hamiltonian systems 共Lichtenberg and Lieberman,
1992兲 is the same as that of magnetic-field lines. A
magnetic-field line has three fundamentally different
types of trajectories.
共1兲 It can close on itself after traversing the torus m
times toroidally, in the ␸ direction, and n times po-
loidally, in the ␪ direction.
共2兲 A field line can come arbitrarily close to every point
on a toroidal surface as the number of toroidal tra-
versals goes to infinity.
共3兲 A field line can come arbitrarily close to every point
in a nonzero volume of space as the number of tor-
oidal traversals goes to infinity.
The first of these possibilities is topologically unstable;
an arbitrarily small perturbation can destroy the closure
of a set of field lines. For example, the closure of a pure
toroidal field Bជ = 共␮ G / 2␲R兲␸ˆ , with G the current, can
0
be destroyed by an arbitrarily small field in the vertical
direction BZẐ. The second possibility, each field line
coming arbitrarily close to every point on a toroidal sur-
face, is the one desired for magnetic confinement. The
Kolmagorov-Arnold-Moser theorem 共Kolmogorov,
1954; Arnold, 1963; Moser, 1967兲 says this possibility is
topologically stable over most of a volume filled by field
lines if the perturbation is sufficiently small. The state-
 Allen H. Boozer: Physics of magnetically confined plasmas
ment that a magnetic field lies in toroidal surfaces means
the second possibility, except on isolated surfaces of zero
measure. The third possibility, magnetic-field lines com-
ing arbitrarily close to every point in a volume, implies
the absence of magnetic confinement in that region. Un-
fortunately, this is the generic situation for a magnetic
field. Consequently, magnetic confinement depends on
the formation of very special magnetic fields, fields that
have lines that lie on nested surfaces through the bulk of
the plasma volume.
In regions of magnetic confinement of a near-
Maxwellian plasma, the constraint B ជ ·ⵜ
ជ p = 0 holds. This
constraint greatly simplifies the field-line trajectories. In
the language of Hamiltonian mechanics the pressure p is
an isolating constant of the motion. Whenever an isolat-
ing constant of the motion exists in a one-and-a-half-
degree-of-freedom Hamiltonian system, the canonical
coordinates can be chosen so the Hamiltonian is a func-
tion of only one canonical variable, the canonical mo-
mentum, rather than three. These canonical coordinates
are called action-angle variables and are central to
Hamiltonian perturbation theory. For magnetic-field
lines, action-angle coordinates are known as magnetic
coordinates 共Hamada, 1962兲. Their existence is easily
ជ ·ⵜ
ជ p = 0, the magnetic field can be
demonstrated. Since B
written in arbitrary 共p , ␪ , ␸兲 coordinates as B ជ =B ⵜជ
1 p
ជ ជ ជ ជ ជ
⫻ ⵜ␪ + B2ⵜ␸ ⫻ ⵜp. The constraint that ⵜ · B = 0 implies
⳵B1 / ⳵␸ = −⳵B2 / ⳵␪, which means the expansion coeffi-
cients must have the forms 2␲B1 = ␺⬘t 共p兲共1 + ⳵␭ / ⳵␪兲 and
2␲B2 = ␺p⬘ 共p兲 − ␺⬘t 共p兲⳵␭ / ⳵␸. The magnetic field, therefore,
has the form
ជ = ⵜជ ␺ ⫻ ⵜជ ␪ + ⵜជ ␸ ⫻ ⵜជ ␺ 共␺ 兲,
2␲B
t m p t
where the magnetic poloidal angle ␪m ⬅ ␪ + ␭. In mag-
netic coordinates 共␺t , ␪m , ␸兲, the field-line Hamiltonian,
which is the poloidal flux ␺p, is a function of the toroidal
flux ␺t alone. The rotational transform ␫共␺t兲, the greek
letter iota, and its reciprocal, the safety factor q共␺t兲, are
defined by
1 d␺p
␫⬅ ⬅ .
q d␺t
In general, the rotational transform is used in the stel-
larator and the safety factor in the tokamak literature.
We use ␫ to avoid confusion with the use of q for elec-
trical charge.
When magnetic coordinates exist, ␺p共␺t兲, the field-line
trajectories are said to be integrable and can be given
explicitly in terms of the initial conditions: ␺t = ␺0 and
␪m = ␪0 + ␫共␺0兲␸. The initial poloidal angle of the field
lines ␣ ⬅ ␪0 / 2␲ can be used as a coordinate in place of
␪m. If this is done
ជ = ⵜជ ␺ ⫻ ⵜជ ␣ ,
B 共9兲
t Magnetic coordinates are central to the theory of
which is called the Clebsch representation. The theory of
this representation predates Clebsch 共1859兲, and in the
mathematics literature ␺t and ␣ are called Euler poten-
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1077
tials. A history has been given by Stern 共1970兲. The
magnetic-field lines are orthogonal to both the toroidal
flux ␺t and to ␣ since B ជ ·ⵜ
ជ ␺ = 0 and B ជ ·ⵜជ ␣ = 0, so both are
t
constants of the motion of the Hamiltonian. However,
the constancy of ␣ only prevents a field line from coming
arbitrarily close to all points on a ␺t surface if ␫ is the
ratio of two integers, ␫ = n / m, in other words, a rational
number. When iota is a rational number, the field lines
close on themselves after m toroidal circuits, which
means in the ␸ direction, and n circuits in the poloidal,
which means in the ␪ direction. If ␫ is a rational number,
␣ is what is called in Hamiltonian mechanics an isolating
invariant because it isolates the trajectory so it can ap-
proach only a fraction of the spatial points it could oth-
erwise reach. However, when ␫ is an irrational number,
the trajectory of a single field line comes arbitrarily close
to every point on a constant-␺t surface, and one says ␣ is
a nonisolating invariant.
The Clebsch representation of the magnetic field, Eq.
共9兲, exists in any region of space in which all field lines
pass through a plane. Let 共r , ␣兲 be any set of coordinates
in that plane, and let the trajectory of a field line that
passes through the point 共r , ␣兲 be xជ 共r , ␣ , ᐉ兲, where ᐉ is
the distance along a field line from the plane. By con-
struction both r and ␣ are constant along a field line,
ជ ·ⵜ
ជ r = 0 and B
ជ ·ⵜ
ជ ␣ = 0, so Bជ = f共r , ␣ , ᐉ兲ⵜជ r⫻ⵜ ជ ␣. Since Bជ is
B
divergence free, ⳵f / ⳵ᐉ = 0. If we define ␺t so ⳵␺t / ⳵r
ជ =ⵜ
= f共r , ␣兲, then B ជ ␺ ⫻ⵜ
ជ ␣.
t
Various choices are commonly made for the third co-
ordinate of Clebsch coordinates. The most common is
the distance along the field lines ᐉ. The dual relations of
general coordinates, Eq. 共A7兲, imply B ជ =ⵜជ ␺ ⫻ⵜជ␣
共7兲 t
= 共1 / J兲⳵xជ / ⳵ᐉ, where J is the Jacobian of 共␺t , ␣ , ᐉ兲 coor-
dinates. Since ᐉ is the distance along the lines, 共⳵xជ / ⳵ᐉ兲2
= 共JB兲2 = 1, which implies that the Jacobian of 共␺t , ␣ , ᐉ兲
coordinates is 1 / B. Any vector in three dimensions can
be written in the covariant form, B ជ = B̃ ⵜ ជ ␺ + B̃ ⵜជ␣
␺t t ␣
ជ ᐉ. The orthogonality relation of general coordi-
+ B̃ᐉⵜ
ជ · ⳵xជ / ⳵ᐉ = B. That is,
nates, Eq. 共A3兲, implies B̃ᐉ ⬅ B
共8兲
ជ = Bⵜជ ᐉ + B̃ ⵜជ ␺ + B̃ ⵜជ ␣ .
B 共10兲
␺t t ␣
Another choice for the third coordinate is the magnetic
scalar potential ␾, which is defined by the indefinite in-
tegral ␾共␺t , ␣ , ᐉ兲 ⬅ 兰Bdᐉ along each field line. Using this
coordinate, an arbitrary magnetic field can be written as
ជ = ⵜជ ␾ + B ⵜជ ␺ + B ⵜជ ␣ .
B 共11兲
␺t t ␣
The Jacobian of 共␺t , ␣ , ␾兲 coordinates is 1 / 共ⵜ ជ␺
t
ជ ជ
⫻ ⵜ␣兲 · ⵜ␾ = 1 / B .
2
magnetically confined plasmas because they allow a
simple solution to a differential equation that frequently
arises, the magnetic differential equation B ជ ·ⵜ
ជ f=g
1078 Allen H. Boozer: Physics of magnetically confined plasmas
共Kruskal and Kulsrud, 1958; Newcomb, 1959兲. In mag-
netic coordinates, Eqs. 共7兲 and 共8兲, this equation has the
form
冉 ⳵
⳵␸
+ ␫ 共 ␺ t兲
⳵
冊
⳵␪m
f=
g
ជ · ⵜជ ␸
B
, 共12兲
which can be solved algebraically using the Fourier ex-
ជ ·ⵜ
pansion of g / B ជ ␸ = 兺␥ ei共n␸−m␪m兲. The Fourier expan-
mn
sion coefficients of the function f are fmn = −i␥mn / 共n
− ␫m兲.
A practical example of the use of the magnetic differ-
ential equation is the effect of a magnetic perturbation
␦Bជ on a field that has perfect surfaces, B ជ . To find the
0 d␫
perturbed magnetic surfaces one solves for a function
p = p + ␦p such that 共Bជ + ␦Bជ 兲·ⵜ
ជ p = 0. The unperturbed
0 0
ជ ·ⵜជ p = 0. Consequently
system has perfect surfaces so B 0 0
the first-order result is given by the magnetic differential
equation
ជ · ⵜជ ␦p = − ␦B
ជ · ⵜជ p .
B 0 0 共13兲
For simplicity, assume the Fourier expansion
␦Bជ · ⵜជ ␺
t
= bmn sin共n␸ − m␪m兲,
ជ ជ
B0 · ⵜ ␸
where ␺t is the toroidal flux enclosed by a surface of
constant p0. Then one obtains a finite solution near the
resonant rational surface, ␺t = ␺mn, on which ␫共␺mn兲
= n / m only if the gradient of p0 vanishes there. More
precisely, as ␺t → ␺mn, one must let dp0 / d␺t = 共n − ␫m兲c0
with c0 a constant. Then ␦p = c0bmn cos共n␸ − m␪m兲, and
near the rational surface p0共␺t兲 = p0共␺mn兲 − c0m共d␫ /
d␺t兲共␺t − ␺mn兲2 / 2. The equation for the perturbed sur-
faces is p0共␺t兲 + ␦p = const. Using the identity cos共2x兲 = 1
− 2 sin2共x兲, the equation for the perturbed surfaces near
the resonant rational surface is
冑 再 冉 冊冎
s 4bmn
␺t − ␺mn = 2 2 n␸ − m␪m
兩s兩 d␫ s − sin
m 2
d␺t
The constant s labels the surfaces of constant p, which
are the magnetic surfaces in the perturbed configuration.
Equation 共15兲 for perturbed magnetic surfaces has two
topologically distinct regions, which can be studied by
holding ␸ fixed and varying ␪m 共Fig. 2兲. First, the sur-
faces that have a surface label 兩s兩 艌 1 cover the full range
of ␪m. The surfaces with s ⬎ 1 and s ⬍ −1 are different
sets of surfaces, which are distorted by the perturbation
but not fundamentally changed. However, the surfaces
that satisfy 1 ⬎ s ⬎ −1 cover only a limited ␪m range and
are said to form a magnetic island. The two signs of s
= ± 兩s兩 give two parts of the same surface for 1 ⬎ s ⬎ −1.
Each term bmn in a Fourier expansion of a magnetic
perturbation, Eq. 共14兲, produces an island if there is a
surface ␺t = ␺mn on which ␫共␺mn兲 = n / m, which is called a
resonant rational surface. The half-width of the island in
toroidal flux is
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
␦mn ⬅
冑冨 冨4bmn
m
d␫
. 共16兲
d␺t
Islands can also be calculated using the Hamiltonian for
the field lines. Equation 共3兲 implies the function b
⬅ ␦Bជ ·ⵜ
ជ ␺ /B
ជ ·ⵜ
ជ ␸ = −⳵␺ / ⳵␪. Therefore, if ␺ = ␺¯ 共␺ 兲
t 0 p p p t
+ „␺˜ p共␺t兲…mn cos共n␸ − m␪M兲, then bmn = −m共␺˜ p兲mn and Eq.
共16兲 can be rewritten as
冑冨 冨
4共␺˜ p兲mn
␦mn ⬅ . 共17兲
d␺t
The Chirikov criterion 共Chirikov, 1979兲 says that if the
half-widths of islands from different resonant rational
surfaces become comparable to the separation between
these resonant surfaces, the magnetic-field lines become
stochastic, which means a single field line comes arbi-
trarily close to every point in a finite volume of space.
A qualitative understanding of the formation of is-
lands and the breakdown of magnetic surfaces can be
共14兲 gained from a study of what is called the standard map
共Chirikov, 1979兲. Given an initial poloidal angle ␪0 and
radial position ⌿0, the standard map assumes that after
one toroidal circuit the poloidal angle of a field line be-
comes ␪1 = ␪0 + ⌿0 and the radial coordinate becomes
⌿1 = ⌿0 + k sin共␪1兲. When k = 0, the radial coordinate is
2␲ times rotational transform, ⌿ = 2␲␫共␺t兲. The param-
eter k is proportional to the perturbation. A trajectory
of the standard map is found by iteration from the origi-
nal position 共␪0 , ⌿0兲. That is, the Nth point along the
trajectory is at ␪N = ␪N−1 + ⌿N−1 and ⌿N = ⌿N−1
+ k sin共␪N兲. The standard map has the essential property
for modeling the field lines of a divergence-free field,
which is a unit Jacobian ⳵共⌿N , ␪N兲 / ⳵共⌿N−1 , ␪N−1兲 = 1. If a
large collection of trajectories are followed that were
. 共15兲 initially in a small area 共␦⌿0兲共␦␪0兲, then the area occu-
pied by the trajectories remains the same forever. For
k ⫽ 0, islands appear at ⌿ = 2␲n / m, where n and m are
integers, with the width of the islands scaling as 冑k for
small k. For k ⬍ 0.9716. . ., Greene 共1976兲 has shown that
trajectories of the standard map can cover only a limited
range of ⌿. However, for larger values of the perturba-
tion k, some trajectories cover an unlimited ⌿ range
共Fig. 3兲. Such trajectories are said to be stochastic. The
breakdown of the ⌿ surfaces with increasing k is analo-
gous to the breakdown in the magnetic surfaces with an
increasing perturbation.
In the presence of a perturbation that produces an
arbitrarily small island, any function p0共xជ 兲 that satisfies
Bជ ·ⵜជ p = 0 must have the form dp / d␺ = 共n − ␫m兲c . This
0 0 0 t 0
constraint on p0 implies that the opening of an island is
a singular process in a toroidal equilibrium. In the ab-
sence of an island, a function such as the pressure can
have a nonzero derivative with respect to the toroidal
flux, dp0 / d␺t, but an arbitrarily small island means the
 Allen H. Boozer: Physics of magnetically confined plasmas 1079

derivatives near the island can only be nonzero using the

helical flux, dp0 / d␺h ⫽ 0, where the helical flux is defined
by d␺h ⬅ 共␫ − n / m兲d␺t. The singularities that arise due to
the modification of the pressure gradient by a small
resonant perturbation are known as the Glasser effect
共Glasser et al., 1975兲. The resolution of these singulari-
ties is an area of active research and is usually discussed
under the topic of neoclassical tearing modes; see, for
example, Rosenberg et al. 共2002兲. The singularities asso-
ciated with the gradient in the electric potential can pre-
vent a small island from opening in a rotating plasma;
see Sec. V.B.3.

B. Methods of forming toroidal magnetic surfaces

• A spatial region in which the magnetic-field lines lie
in toroidal surfaces requires either a net toroidal cur-
rent flowing within the region or helical shaping of
the bounding toroidal surface.
To form magnetic surfaces, a tokamak 共Fig. 4兲, uses a
net toroidal current in the plasma, while a stellarator
共Figs. 5–7兲 uses helical shaping and in some cases a net
toroidal current as well. Both of these plasma confine-
ment systems are discussed in Sec. IV. Ignoring the to-
pologically unstable case of all field lines closing on
themselves, magnetic-field lines form toroidal surfaces
only when there is both a toroidal and a poloidal mag-
netic field. The toroidal magnetic field can be produced
simply by external coils, while the poloidal field is more
difficult.
A pure toroidal magnetic field is produced by a wire
carrying a current G along the z axis of cylindrical coor-
dinates,
FIG. 3. 共Color兲 Iterates of the standard map plotted for 共a兲 k
= 0.3 and 共b兲 k = 1.1. Islands arise near rational values of ⌿ / 2␲
ជ = ␮0G ⵜជ ␸ = ␮0G ␸ˆ .
B 共18兲 for small values of k, but the trajectory followed by iterating
␸
2␲ 2␲ R the map remains bounded in ⌿ space. For k ⬎ 0.9716, trajecto-
ries exist that cover all values of ⌿. The large fuzzy region in
An infinite straight wire is not a practical coil set, but 共b兲 is a single stochastic trajectory found by one set of itera-
tions. The periodicity of the standard map was used to plot all
exactly the same magnetic field is produced by any cur-
iterates in the 0 艋 ␪ / 2␲ ⬍ 1 and 0 艋 ⌿ / 2␲ ⬍ 1 region.
rent distribution that surrounds the toroidal region of
interest, runs in a constant-␸ plane, and is axisymmetric
共no ␸ dependence兲. The toroidal field due to external small, one needs the separation between the toroidal
coils is given by a multivalued scalar potential, Bជ =ⵜជ ␾, field coils to be significantly less than the minor radius of
that obeys Laplace’s equation, ⵜ2␾ = 0, which has the so- the plasma, which means N must be somewhat larger
lution ␾ = ␮0G␸ / 2␲. It is impossible to make the toroi- than 2␲ / ⑀c, with ⑀c ⬅ a / Rc the inverse aspect ratio of the
dal field coils perfectly axisymmetric because that would toroidal surface that forms the plasma edge at r = a. The
preclude any direct access to the plasma. The effect of N physical reasons that the toroidal ripple must be limited
separate toroidal field coils can be approximated by as- are discussed in Sec. VI.E on particle trajectories and
suming the coils have N vertical legs located at an outer transport.
radius Rc with the current returning along the z axis, The production of the poloidal, ␪-directed, magnetic
that is, R = 0. The magnetic potential becomes field is more difficult. Of course a poloidal field can be

冋 冉 冊 册
produced by a coil that encircles the z axis at a radius Ro
␮ 0G R N
cos N␸ and carries a current I. Close to the coil, r / Ro Ⰶ 1, the
␾= ␸+ , 共19兲
2␲ Rc N poloidal field is B␪ = ␮0I / 2␲r, and the poloidal flux obeys
⳵␺p / ⳵r = 2␲RoB␪共r兲. The toroidal magnetic flux is ap-
which gives a sinusoidal ripple in the strength of the proximately ␺t = ␲r2B␸, so the rotational transform, ␫
toroidal field. In order to make the ripple sufficiently ⬅ d␺p / d␺t, is approximated by

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1080 Allen H. Boozer: Physics of magnetically confined plasmas
FIG. 4. 共Color兲 A tokamak showing the axisymmetric plasma
and the coils necessary to support it. Magnetic surfaces exist in
a tokamak only when the toroidal plasma current I is nonzero.
R oB ␪
␫⬇ . 共20兲
rB␸
A circular coil produces magnetic surfaces, but this con-
figuration by itself is not suitable for confining fusion
plasmas. Connections are needed between the coil and
the outside world so that the heat produced by the slow-
ing of the fusion neutrons within the coil structure can
be removed. Such connections would have to pass
through the hot fusing plasma and would destroy its con-
finement.
Although a circular coil cannot be used to produce the
poloidal magnetic field, much the same effect can be
produced by a current in the plasma that is parallel to
the magnetic field, which is called the net plasma cur-
ជ / ␮ . This method was suggested by I.
rent, ជjnet = k共xជ 兲B 0 B␪ = ␮0I / 2␲r and I the current enclosed by a surface of
Tamm and A. Sakharov in the Soviet Union in the early
1950s and is used to form the poloidal field in tokamaks.
FIG. 5. 共Color兲 The coils and some magnetic-field lines of the
Large Helical Device 共LHD兲. The rotational transform in
LHD is due to the wobble of the magnetic-field lines produced
by the helical coils. A typical plasma has an average major
radius of 3.6 m. Figure courtesy of the National Institute for
Fusion Science, Japan.
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
FIG. 6. 共Color兲 The W7-X stellarator with its helically shaped
plasma and the coils required to support it. Magnetic surfaces
exist in a stellarator even in the absence of a plasma. The
average major radius of a plasma is 5.5 m. Figure courtesy of
the Max-Planck-Institut für Plasmaphysik, Garching, Ger-
many.
If the plasma pressure gradient is negligibly small, the
net current is the total current. Such equilibria are called
force free and obey the equation
ជ = k共xជ 兲B
ⵜជ ⫻ B ជ. 共21兲
Taking the divergence of both sides of this equation, one
finds that k共xជ 兲 is constrained to be constant along the
magnetic field,
ជ · ⵜជ k = 0.
B 共22兲
This implies that a gradient in k ⬅ ␮0j储 / B, with j储
ជ / B, can only occur in regions of good magnetic
⬅ ជj · B
surfaces where k共␺t兲 is a function of the enclosed toroi-
dal flux. If the magnetic surfaces are essentially circular,
then the rotational transform is given by Eq. 共20兲 with
radius r. Usually k can be approximated by a constant
FIG. 7. 共Color兲 The surface of the proposed National Compact
Stellarator Experiment 共NCSX兲 quasiaxisymmetric stellarator,
shown with three toroidal circuits of a magnetic-field line. The
rotational transform ␫ is just under 2 / 3, so the line almost
closes. Despite the strong helical shaping, the particle drifts are
similar to those of an axisymmetric tokamak. Much of the ro-
tational transform in NCSX is due to the torsion of the mag-
netic axis. The average major radius of a plasma is 1.44 m.
Figure courtesy of the Princeton Plasma Physics Laboratory,
Princeton, NJ.
 Allen H. Boozer: Physics of magnetically confined plasmas
for sufficiently small r. The rotational transform in that
region is ␫ = Rok / 2. For radii r larger than the current
channel, the rotational transform is ␫ ⬀ 1 / r2.
Stellarators 共Spitzer, 1958兲 have magnetic surfaces and
a nonzero rotational transform even in the absence of a
plasma. Stellarator magnetic surfaces must have helical
shaping. The existence of magnetic surfaces without any
enclosed currents is demonstrated starting with the axi-
symmetric magnetic surfaces formed by a coil encircling
the z axis at a radius Ro, which is designed so the rota-
tional transform ␫ is an integer Np in the center of the
region of interest. If this system is modified by a mag-
netic perturbation in which bm=1, n=Np is nonzero, an is-
land is formed 关Eq. 共16兲兴. Since the magnetic surfaces
inside an island do not cover the full range of theta 共Fig.
2兲, room exists for connections and supports for the coil.
The magnetic surfaces inside the island fill a volume and
have no current on them. The integer Np is the number
of periods of this type of stellarator, which is called a
Heliac 共Boozer et al., 1983兲.
The poloidal field in a stellarator can be produced in
two ways in the absence of a plasma. The first way uses
helical wobbles of the magnetic-field lines that are
driven by helical currents in coils 共Spitzer, 1958兲. The
rotational transform in the Large Helical Device 共Ya-
mada et al., 2001兲 is produced entirely by the field-line
wobble produced by helical coils 共Fig. 5兲. The magnetic
field that is produced by the helical wobbles acts, when
averaged over the wobbles, as if there were a force-free
current that filled the space around the coils and that
became exponentially small with increasing distance
from the coils. To understand this, consider a weak he-
lical perturbation to a spatially constant magnetic field,
Bជ / B = ẑ + ⵜ
ជ ␾, with ␾共x , y , z兲 = ⌬共x兲cos共k z − k y兲 in Car-
0 z y
tesian coordinates. Since ⵜ2␾ must be zero, the distance
⌬共x兲 ⬀ exp共kx兲 with k2 = k2y + k2z. The magnetic-field lines
are given by dx / dz = Bx / Bz ⯝ k⌬ cos共¯兲 or x ⯝ x0
+ 共k⌬ / kz兲sin共¯兲, and dy / dz = By / Bz. That is, dy / dz
= ky⌬ sin共¯兲 / 兵1 − kz⌬ sin共¯兲其, which when expanded
to keep the important second-order terms in ⌬ is
dy / dz ⯝ ky⌬ sin共¯兲 + 兵共kyk2 / kz兲 + 共kykz兲其⌬2 sin2共¯兲. The
magnetic-field lines have an oscillation in the ŷ direction,
␦y = 共ky / kz兲⌬ sin共¯兲, but also a systematic drift in that
direction, which is equivalent to a y-directed magnetic
field,
冉 冊
具By典 1 ky k2
= 1 + 2z 共k⌬兲2 .
B0 2 kz k = 共␮0G / 2␲兲ⵜជ ␸ , with G a constant. The potential ␸ ,
The second-order terms in the dx / dz equation give no
systematic drift. Consequently, field lines wobble in and
out of a constant-x surface but have a drift in the y di-
rection. The rotational transform per period in the y and
x directions is ␫p ⬅ 共ky / kz兲具By典 / Bz, or
冉 冊冉 冊
1 ky 2
k2z
␫p = 1+ 共k⌬兲2 .
2 kz k2
There is an upper limit on the transform per period
since k⌬ must be small compared to 1. The effective
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1081
force-free current that fills the space around the helical
coil is ␮0ជjeff = 2k具By典ẑ.
The second way to produce a poloidal field without a
plasma is through torsion in the magnetic axis. The mag-
netic axis is the field line at ␺t = 0 and is the axis of the
poloidal angle. The torsion ␶ of a curve measures the
extent to which the curve fails to lie flat on a plane. The
rotational transform due to torsion is
L
␫ = 具␶典 , 共25兲
2␲
with 具␶典 the torsion of the magnetic axis averaged along
its length L. Torsion was used in the early 1950s to pro-
duce the poloidal field in Spitzer’s Figure-8 stellarators
共Spitzer, 1958兲. The optimization of the particle trajecto-
ries in a stellarator 共Sec. VI.E.1兲 requires the use of tor-
sion. Indeed, torsion gives a major part of the rotational
transform in both the Wendelstein-7X 共W7-X兲 共Fig. 6;
Beidler et al., 1990兲, and the National Compact Stellar-
ator Experiment 共NCSX兲 共Fig. 7; Zarnstorff et al. 2001兲
stellarator designs. The derivation of the relation be-
tween the torsion and the rotational transform uses the
Frenet formulas 共Mathews and Walker, 1964兲 to analyze
the behavior of field lines near the magnetic axis, with ᐉ
the distance along the axis. The magnetic axis, xជ 0共ᐉ兲, is
the closed curve about which the field lines wind. The
derivative of a curve is its tangent, b̂0 ⬅ dxជ 0 / dᐉ, which is
a unit vector along the magnetic-field line that forms the
axis. The derivative of the tangent is db̂0 / dᐉ = ␬␬ˆ , with
␬␬ˆ the curvature of the axis. The derivative of the cur-
vature unit vector is d␬ˆ / dᐉ = −共␬b̂0 + ␶␶ˆ 兲, with ␶␶ˆ the tor-
sion. The derivative of the torsion unit vector is d␶ˆ / dᐉ
= ␶␬ˆ . The Frenet unit vectors are mutually orthogonal
and satisfy b̂0 ⫻ ␬ˆ = ␶ˆ . Mercier 共1964兲 used the Frenet unit
vectors to establish a coordinate system near an arbi-
trary magnetic axis. We shall simplify his analysis by
making the magnetic surfaces circular near the axis,
which excludes the rotational transform from helical
wobbles. That is, we choose coordinates xជ 共r , ␪ , ᐉ兲
= xជ 0共ᐉ兲 + r cos␪␬ˆ + r sin ␪␶ˆ . Using the methods of gen-
eral coordinates given in the Appendix, one finds
ⵜជ r = cos␪␬ˆ + sin ␪␶ˆ , ⵜជ ␪ = 共−sin ␪␬ˆ + cos␪␶ˆ 兲 / r + ␶b̂ / 共1
0
ជ ᐉ = b̂ / 共1 − ␬r cos␪兲. The magnetic field
− ␬r cos␪兲, and ⵜ
0
in the absence of plasma currents has the form B ជ
共23兲
p p
which can be interpreted as a toroidal angle, obeys
ⵜ 2␸ ⬅ ⵜជ ·ⵜ
ជ ␸ = 0. If the magnetic strength is constant
p p
along the magnetic axis, then for r → 0 the toroidal angle
␸p = 2␲共ᐉ + ␬␶r3 sin ␪ + ¯ 兲 / L, with L the length of the
axis. For r → 0, the magnetic-field-line equations
are d␪ / d␸ = Bជ ·ⵜ
ជ ␪/B
ជ ·ⵜជ ␸ = ␶L / 2␲, and dr / d␸
p p p
共24兲 =Bជ ·ⵜ
ជ r/B
ជ ·ⵜ
ជ ␸ = 0. Consequently, the rotational trans-
p
form near the axis is ␫ = 具␶典L / 2␲, with the axis average of
the torsion 具␶典 ⬅ 养 ␶dᐉ / 养 dᐉ.
The magnetic configuration due to the coils alone is
1082 Allen H. Boozer: Physics of magnetically confined plasmas
called the vacuum configuration. It is in principle easy to
design vacuum configurations that have magnetic sur-
faces of any desired shape, but unless the surfaces have
helical shaping the rotational transform will be zero. To
design a vacuum configuration, first assume there is an
ជ = ␮ G␸ˆ / 2␲R, which pro-
axisymmetric toroidal field, B
␸ 0
duces the toroidal magnetic flux. One then defines the
desired shape of an outermost magnetic surface using
共R , ␸ , Z兲 cylindrical coordinates, xជ s = Rs共␪ , ␸兲R̂共␸兲
+ Zs共␪ , ␸兲Ẑ, by giving the appropriate Fourier-series rep-
resentation of Rs and Zs. The unit normal to this surface
is n̂ ⬀ 共⳵xជ s / ⳵␪兲共⳵xជ s / ⳵␸兲. Coils just outside the plasma can
in principle produce any desired normal magnetic field
on the surface xជ s. If the coils are designed so that they
produce a normal field equal but opposite in sign to
ជ , then the surface xជ will be a magnetic surface,
n̂ · B
␸ s
since the magnetic-field lines do not cross it, n̂ · B ជ = 0.
When vacuum magnetic fields produce one magnetic
surface, the volume enclosed by that surface is generally
dominated by regions of good surfaces rather than is-
lands and stochastic regions. Practical limitations on the
design of magnetic-field configurations are discussed in
Sec. V.D.
C. Evolution of magnetic-field lines
• Clock time is a parameter in the Hamiltonian de-
scription of evolving magnetic-field lines. For an
evolving magnetic field, the Hamiltonian,
␺p共␺t , ␪ , ␸ , t兲, as well as the coordinate transforma-
tion function, xជ 共␺t , ␪ , ␸ , t兲, depend on time as well as
the canonical coordinates. A magnetic field evolves
ideally 共without a topology change兲 if the field-line
Hamiltonian can be made time independent by an
appropriate choice of the coordinate transformation
function xជ 共␺t , ␪ , ␸ , t兲.
• The mathematical condition for an ideal evolution of
ជ 共xជ , t兲 is that a function ⌽ 共xជ , t兲 exist such that
B
B
ជ ជ ជ ជ
B · ⵜ⌽ = −E · B, where the electric field is determined
B field lines are independent of time if the canonical coor-
by Faraday’s law, ⳵B ជ / ⳵t = −ⵜ
ជ ⫻E
ជ . Magnetic-field lines
evolve ideally under more general conditions than
the tying together of the plasma and the field, and a
generic magnetic field always evolves ideally in a suf-
ficiently localized spatial region.
• The self-entanglement of magnetic-field lines is mea-
ជ ·B
ជ d3x. A spiky
sured by the magnetic helicity K ⬅ 兰A
current profile causes a rapid dissipation of energy
relative to magnetic helicity. If the evolution of a
magnetic field is rapid, then it must be at constant
helicity.
Ideally a fusion plasma would be in a steady state, but
the theory of the evolution of the magnetic field is im-
portant for finding the conditions for 共1兲 establishing the
field configuration, 共2兲 maintaining the magnetic field,
and 共3兲 finding the conditions under which rapid changes
in the magnetic field can occur. The evolution of the
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
magnetic field means the evolution of the magnetic-field
lines. The equations for the evolution of magnetic-field
lines are also of interest for the evolution of other intrin-
sically divergence-free fields such as the vorticity field of
fluid mechanics.
The evolution of a magnetic field B ជ 共xជ , t兲 is called ideal
if it is consistent with the field’s being embedded in an
ideal, zero-resistivity fluid moving with a velocity uជ 共xជ , t兲,
which can be interpreted as the velocity of the magnetic-
field lines. An ideal evolution does not change the topol-
ogy of the magnetic-field lines. The conditions for an
ideal evolution are obtained by noting that the electric
ជ 共xជ , t兲 associated with an evolving magnetic field
field E
Bជ 共xជ , t兲 is given by Faraday’s law to within an arbitrary
additive gradient of a potential. Mathematics implies
that an arbitrary vector E ជ 共xជ , t兲, and hence the electric
field, can be written as
ជ + uជ ⫻ B
E ជ = − ⵜជ ⌽ + Vⵜជ
B 冉 冊 ␸
2␲
共26兲
ជ ·ⵜ
ជ ␸ is nonzero. The
in any region of space in which B
function V, called the loop voltage, is constant along
each magnetic-field line and is given by
冕 L
−L
ជ · dᐉជ
E
冕 冉 冊
V ⬅ lim L , 共27兲
␸
ⵜជ · dᐉជ
L→⬁
2␲
−L
where dᐉជ is the differential distance along B
ជ . The com-
ponent of Eជ parallel to Bជ is balanced by ⌽ and V, and
the components of E ជ perpendicular to B ជ are balanced by
uជ . If the loop voltage V is zero, the evolution is ideal.
The relation between the conservation of topological
properties and the vanishing of the loop voltage V will
be shown in this paragraph, which can be skipped on a
first reading. The topological properties of the magnetic-
dinates 共␺t , ␪ , ␸兲 can evolve in such a way that the field-
line Hamiltonian does not change, 共⳵␺p / ⳵t兲c = 0. The
subscript c means the partial derivative is performed at a
fixed point in canonical coordinates. This conclusion fol-
lows from two statements. First, the trajectories of
magnetic-field lines in canonical coordinates are deter-
mined by the field-line Hamiltonian ␺p共␺t , ␪ , ␸ , t兲 alone
关Eqs. 共4兲 and 共5兲兴. Second, topological properties are un-
changed by a continuous temporal variation in a coordi-
nate transformation function, xជ 共␺t , ␪ , ␸ , t兲. Consequently
the topological properties of field lines are determined
by the field-line Hamiltonian alone, and these properties
cannot change unless the field-line Hamiltonian changes.
The evolution equation for the magnetic-field-line
Hamiltonian is obtained from the evolution of the vec-
tor potential in canonical coordinates. Using only the
theory of general coordinates 共Appendix兲, one can write
共Boozer, 1992兲 the time derivative of an arbitrary vector
 Allen H. Boozer: Physics of magnetically confined plasmas 1083

ជ =␺ ⵜជ ជ ជ ជ + vជ ⫻ B
ជ =␩
J · ជj .
A t 共␪ / 2␲兲 − ␺pⵜ共␸ / 2␲兲 + ⵜg in a coordinate system E 共32兲
defined by the function xជ 共␺t , ␪ , ␸ , t兲 关Eq. 共A19兲兴 as
In plasmas the resistivity tensor is diagonal but with dis-

冉 冊 冉 冊
⳵Aជ
⳵t xជ
=−
⳵␺p ជ ␸
ⵜ
⳵t c 2␲
ជ + ⵜជ s,
+ uជ ⫻ B 共28兲
tinct values along, ␩储, and across, ␩⬜, the magnetic field.
In a quiescent plasma, ␩⬜ is about twice ␩储, but the mi-
croturbulence that is present in most confined plasmas
enhances ␩⬜ by several orders of magnitude while leav-
where we have let B ជ ⬅ⵜ
ជ ⫻Aជ and introduced the velocity ing ␩储 essentially unchanged 共see Sec. VI.F兲. Substituting
of a fixed point in canonical coordinates, Ohm’s law into the general expression for the electric
field, Eq. 共26兲, one finds that
⳵xជ 共␺t, ␪, ␸,t兲
uជ ⬅
⳵t
. 共29兲
Vⵜជ 冉 冊
␸
2␲
ជ =␩
+ 共vជ − uជ 兲 ⫻ B J · ជj + ⵜជ ⌽B . 共33兲

The subscript xជ means a fixed spatial point, and the sub-

script c means a fixed point in the canonical 共␺t , ␪ , ␸兲
ជ · uជ , Eq. 共A20兲, is
coordinates. The function s = 共⳵g / ⳵t兲c − A
the generating function for infinitesimal canonical trans-
formations of Hamiltonian mechanics. Even when the
ជ / ⳵t兲 ជ = 0, the
vector potential is independent of time, 共⳵A x a conducting fluid obeys two distinct conservation laws
canonical coordinates can change, uជ ⫽ 0, using the free-
dom of infinitesimal canonical transformations s. The re-
lation between Eqs. 共26兲 and 共28兲 can be demonstrated
ជ = −⳵A
ជ / ⳵t − ⵜ
ជ ⌽. The two equations have the same
using E
content with the identification
The loop voltage, V = 兰␩储j储dᐉ / 兰ⵜ ជ 共␸ / 2␲兲 · dᐉជ , vanishes if
the parallel resistivity is zero. If the resistivity tensor is
zero, the magnetic field and the plasma move together,
ជ = vជ · B
uជ = vជ , with the choices ⌽B = 0 and uជ · B ជ.
The evolution of a magnetic field that is embedded in
when the resistivity vanishes, ␩ J = 0: 共1兲 the ideal evolu-
ជ
tion of B, and 共2兲 the tying of the magnetic-field lines to
the fluid, uជ = vជ . A distinction between these two laws is
rarely made in the literature, but the first holds under
more general conditions. The breaking of the ideal evo-

再冉 冊 冎
⳵␺p
⳵t c
− V ⵜជ
␸
2␲
= ⵜជ 兵s − 共⌽B − ⌽兲其. 共30兲
lution of Bជ , which changes the magnetic-field-line topol-
ogy, is determined by the parallel resistivity ␩储 alone.
The flow of the plasma relative to the magnetic-field
lines, vជ − uជ , depends on the perpendicular resistivity ␩⬜
If V = 0, then ␺p共␺t , ␪ , ␸ , t兲 can be made independent of
and in general on the parallel resistivity as well.
time by the choice of infinitesimal generating function
The distinction between the two conservation laws for
s = ⌽B − ⌽. In a system with magnetic surfaces ␺p共␺t , t兲,
magnetic evolution is especially clear in a type-II super-
the loop voltage is the change in the poloidal flux out-
conductor with a melted flux lattice 共Huebener, 1979兲.
side a given toroidal flux surface,
The technically important superconductors are type-II,
which means the magnetic field can penetrate into the
⳵␺p共␺t,t兲
V= . 共31兲 material, which lowers the magnetic-field energy, but the
⳵t magnetic-field lines must lie in narrow flux tubes. These
Remarkably, the evolution of a magnetic field is ge- flux tubes can form a lattice, which holds the tubes in
nerically ideal in a sufficiently small spatial region even place, and in the technically important superconductors
when the field is embedded in a resistive fluid or is in a the flux lattice is rigid. When the lattice is rigid, the su-
vacuum. This follows from the observation that locally perconductor has zero resistivity. However, the flux lat-
ជ + uជ ⫻ B
ជ = −ⵜ
ជ ⌽ ; a nonzero loop voltage V is needed tice can melt, which allows the flux tubes to move
E B
through the superconductor with a velocity, uជ − vជ , pro-
only if the field lines 共a兲 are followed an infinite distance,
portional to the applied force. This motion is equivalent
共b兲 close on themselves, or 共c兲 intercept surfaces on
to a perpendicular resistivity ␩⬜. Therefore a type-II su-
which the potential ⌽B must obey fixed boundary con-
perconductor with a melted flux lattice has a tensor re-
ditions. The evolution is locally ideal 共Boozer, 1992兲
sistivity with ␩储 = 0 but ␩⬜ nonzero. The first conserva-
even near a null field, B ជ = 0, provided the matrix of first
tion law, the ideal evolution of B ជ , holds rigorously in a
derivatives Bij ⬅ ⳵Bi / ⳵xj has a nonzero determinant,
superconductor since ␩储 = 0. However, if the flux lattice
which is the generic situation at a magnetic-field null. In
of a superconductor melts, the perpendicular resistivity
other words, an arbitrarily small perturbation can make
becomes large, and the second conservation law, the ty-
a zero determinant of Bij nonzero at a null.
ing of the field lines to the superconductor, is not even
The simplest model of a nonideal evolution treats the
approximately valid.
plasma as a moving conductor. A Lorentz transforma-
On rational magnetic surfaces the rotational trans-
tion in the nonrelativistic limit implies the electric field
form is a rational number, ␫ = n / m, and magnetic-field
ជ
in a conductor moving with velocity vជ 共xជ , t兲 is E =E ជ
mov lines close on themselves. On these surfaces the loop
ជ . In a simple conductor E
+ vជ ⫻ B ជ J · ជj, with ␩
J the re-
mov = ␩ voltage V can have a different value on each magnetic-
sistivity tensor, and Ohm’s law becomes field line. When this occurs, the field-line Hamiltonian

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1084 Allen H. Boozer: Physics of magnetically confined plasmas

␺p will develop a term that goes as ei共n␸−m␪兲, and the ited by the properties of magnetic helicity 共Woltjer,
rational surface will split to form islands. In a highly 1958兲. Magnetic helicity is defined as

冕
conducting fluid, this is associated with the important
phenomenon of fast reconnection. The currents that K共t兲 ⬅ ជ ·B
A ជ d3x. 共36兲
arise in a highly conducting fluid, ␩ → 0, to preserve the
topology are surface currents, which means they are In a system with perfect magnetic surfaces, the helicity is
delta functions—nonzero only on the rational surface K共␺t , t兲 = 兰共␺t␫ − ␺p兲d␺t with ␫共␺t , t兲 ⬅ ⳵␺p / ⳵␺t the rota-
but infinite there. Such currents lead to rapid dissipa- tional transform. The helicity is a measure of the topo-
tion, so perturbations can open islands on a very short logical entanglement of the magnetic-field lines.
time scale compared to the global resistive time, ␶␩ The importance of magnetic helicity comes from its
⬅ 共␮0 / ␩兲a2, of a plasma of radius a. The actual time scale ជ ·Eជ d3x. If the current density in a
rate of dissipation, 2 兰 B
in a plasma is a subtle question, a question on which
plasma has a large spatial variation, as it does in turbu-
recent progress has been made. See Rogers et al. 共2001兲
lent situations, the rate of loss of magnetic energy,
and reviews by Bhattacharjee et al. 共2001兲 and by Priest
and Forbes 共2000兲. ជ d3x, is rapid in comparison to the rate of helicity
兰jជ · E
In astrophysical and space plasmas, the magnetic-field dissipation. It is relatively easy to change the magnetic
lines often enter and leave the volume of interest. In energy quickly compared to the resistive time scale, ␶␩
such systems nonideal effects in the magnetic field, such ⬅ 共␮0 / ␩兲a2, of a plasma of radius a. However, if the mag-
as reconnection, are far more subtle than in systems with netic energy changes rapidly compared to ␶␩, the change
two periodic directions, where reconnection is focused must be at an essentially constant helicity.
onto the rational surfaces 共Boozer, 2002兲. A nonzero The time derivative of the helicity can be put in a
loop voltage V may be required by boundary conditions. convenient form using E ជ = −⳵Aជ / ⳵t − ⵜ
ជ ⌽,
Nonideal effects can also arise if neighboring magnetic-
field lines separate from each other exponentially with
distance along the lines. In this case, a solution ⌽B共xជ , t兲
dK
dt
=−2 冕 ជ ·E
B ជ d 3x + S + S .
es in 共37兲

to E ជ ·Bជ = −B ជ ·ⵜជ ⌽ may exist but have poor analytic prop-

B
The volumetric term, −共dK / dt兲d ⬅ 2 兰 B ជ ·E
ជ d3x, is the he-
erties due to exponentially large gradients across the
licity dissipation. The two surface terms are the external
field lines. Except when magnetic-field lines lie on sur-
sources of helicity: the electrostatic source
faces, the exponential separation of neighboring lines is
a generic property. Neighboring field lines satisfy
dxជ / d␶ = Bជ 共xជ 兲 and d共xជ + ␦ជ 兲 / d␶ = B
ជ 共xជ + ␦ជ 兲 with ␦ជ → 0. Letting Ses ⬅ − 2 冖 ជ · daជ
⌽B 共38兲
Bij ⬅ ⳵Bi / ⳵x and ⌬ij共ᐉ兲 ⬅ 兰共Bij / B兲dᐉ, one finds the sepa-
j
and the inductive source
ration between the lines at distance ᐉ down either line is
␦ជ 共ᐉ兲 = exp共⌬
genvalues of ⌬
J 兲 · ␦ជ . Unless the magnitudes of all of the ei-
0
J 共ᐉ兲 are exponentially small for large ᐉ, Sin ⬅ − 冖 A
ជ
ជ ⫻ ⳵A · daជ .
⳵t
共39兲
neighboring field lines separate exponentially.
If the bounding surface is a perfect conductor, the exter-
Similar equations to the magnetic evolution equations
nal sources of helicity vanish. When the bounding sur-
can be derived for the vorticity, ␻ ជ ⬅ⵜ ជ ⫻ vជ , of a fluid. The
face is a magnetic surface, the inductive source is Sin
equivalent of Ohm’s law for the vorticity is the Navier- = ␺td␺p / dt − ␺pd␺t / dt. The poloidal and toroidal loop
Stokes equation, ជ · 共⳵xជ / ⳵␪ 兲d␪ and V
voltages are defined as Vp ⬅ 养 E m m t

⳵vជ ⬅ 养E ជ · 共⳵xជ / ⳵␸兲d␸, so S = V ␺ + V ␺ , the loop voltages

+ vជ · ⵜជ vជ = − ⵜជ w共p兲 − ␯ⵜជ ⫻ ␻
in t t p p
ជ. 共34兲 times the fluxes. The toroidal loop voltage is equal to the
⳵t
loop voltage V if the poloidal loop voltage, Vp = −⳵␺t / ⳵t,
This form of the equation assumes the density ␳ is a is zero.
function of the pressure alone with dw共p兲 ⬅ dp / ␳共p兲. The magnetic energy, WB = 兰共B2 / 2␮0兲d3x, has the time
ជ vជ = −vជ ⫻ ␻ derivative dW / dt = −兰jជ · E ជ d3x. When the standard
The kinematic viscosity is ␯. The term vជ · ⵜ B
ជ 共v2 / 2兲, so the Navier-Stokes equation can be written Ohm’s law, Eq. 共32兲, is used, with ␩ J having distinct par-
+ⵜ
ជ replaced by allel and perpendicular components, −dWB / dt is the
in a form analogous to Ohm’s law with E sum of two terms: the dissipative loss of energy
⳵Aជ / ⳵t, 共dWB / dt兲d = 兰共␩储j2储 + ␩⬜j⬜
2
兲d3x and a nondissipative trans-

冉 冊
fer of energy between the magnetic field and the fluid in
⳵vជ 1 ជ 兲d3x. Retaining only the
ជ − ⵜជ w + v2 − ␯ⵜជ ⫻ ␻
= vជ ⫻ ␻ ជ. 共35兲 which it is embedded, 兰vជ · 共jជ ⫻ B
⳵t 2 dissipative term, −共dWB / dt兲d 艌 兰␩储j2储 d3x. In contrast, the
volumetric dissipation of helicity is 2 兰 B ជ ·E
ជ d 3x
The Hamiltonian of the vorticity field lines can be made
time independent if the viscosity vanishes, ␯ = 0. ជ d3x.
= 2 兰 ␩储ជj · B
The rapidity of the evolution of magnetic fields is lim- The relative rates of dissipation of energy and helicity,

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas
冉 冊
dWB
艌
1
冉 冊
1 dK 2
共40兲
冕
−
dt 2 dt
, ជ ·E
d d
␩ 储 B 2d 3x
can be demonstrated 共Berger, 1984兲 using the Schwarz
inequality. This inequality says the average of the square
of a function is at least as large as the square of the
average of the function. The appropriate average for the
relative dissipation rates is the ␩ average, which is de-
fined as 具f典␩ ⬅ 兰共␩储B2fd3x兲 / 兰共␩储B2d3x兲. The rate of mag-
netic energy dissipation satisfies −共dWB / dt兲d
艌 共兰␩储B2d3x兲具共j储 / B兲2典␩, while the rate of helicity dissipa-
tion satisfies 共dK / dt兲d = 2共兰␩储B2d3x兲具j储 / B典␩. The Schwarz
inequality, 具共j储 / B兲2典␩ 艌 具j储 / B典␩2 , then implies Eq. 共40兲. The
greater the spatial variation in j储 / B, the stronger the
Schwarz inequality becomes. In Sec. V.B.1, we shall find
that unstable, or turbulent, plasmas develop very spiky,
Dirac delta-function-like, spatial distributions of j储 / B. In
such situations, the dissipation of the magnetic energy is
extremely rapid in comparison to the dissipation of he-
licity.
Magnetic energy can be dissipated rapidly compared
to helicity, so highly unstable plasmas evolve to mini-
mize their energy for fixed helicity. As Woltjer 共1958兲
showed, this evolution relaxes the current density to the
form ជj = 共k / ␮0兲Bជ , where k is a spatial constant. This
property of the state of minimum energy with fixed he-
licity is demonstrated by varying A ជ in 兰共B2 / ␮
0
− kA ជ ·B
ជ 兲d3x. The constant k, which is called a Lagrange
multiplier, is chosen to make the helicity after minimiza-
tion equal to its initial value. The terms involving ␦B ជ
ជ ជ
= ⵜ ⫻ ␦A are integrated by parts, and the boundary
terms 养B ជ ⫻ ␦Aជ · daជ can be ignored if the boundary is a
ជ ⫻B
ជ = kBជ.
perfect conductor. One finds ⵜ
The physical importance of helicity conservation was
demonstrated by Taylor 共1974兲. He showed that turbu-
lent periods in the reversed-field pinch plasma confine-
ment device 共Prager, 1999兲 lead to flattened current pro-
files with a more quiescent plasma. Since Taylor’s
important work, the relaxation of the current profile to
form a more quiescent plasma has been known as a Tay-
lor relaxation. Taylor’s work demonstrated that helicity
conservation should be a central element of any theory
of rapidly evolving magnetic fields.
A plasma can transition between two states with mag-
netic surfaces but with different helicity distributions,
K共␺t兲, in a time short compared to the resistive time ␶␩.
However, when this occurs helicity conservation implies
that the magnetic surfaces must have broken in the in-
termediate state.
Spiky current profiles cause a rapid loss of magnetic
ជ d3x. On the time scale of energy dissipation,
energy, 兰jជ · E
the magnetic helicity can be transported but not dissi-
pated. The absence of helicity dissipation implies 2E ជ ·B
ជ confinement systems and has been the basis for a
ជ number of proposals for experiments in which deu-
must be the divergence of a flux, the helicity flux Fh with
2Eជ ·Bជ =ⵜជ ·F
ជ . Energy dissipation, which in the Schwarz
h
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1085
inequality argument takes place through the parallel cur-
rent 兰共j储 / B兲B ជ d3x, implies −兰F
ជ ·ⵜ
ជ 共j储 / B兲d3x must be
h
ជ = −␭ ⵜ ជ
positive. This condition is satisfied if F h h 共j 储 / B兲
共Boozer, 1986兲. The positive coefficient ␭h is called the
hyper-resistivity. The gradient of j储 / B is a source of free
energy, which can drive instabilities much as the pres-
sure gradient can; see Sec. V.B.1.
If a system has a large source of free energy other
than the magnetic-field energy, the helicity flux F ជ can
h
have terms that add energy to the magnetic field. Such
terms are needed in a magnetic dynamo, where mag-
netic fields are generated by taking energy from another
source, such as a fluid flow. Taylor relaxation in the
reversed-field pinch demonstrates the existence of a dis-
sipative term in the helicity flux, such as −␭hⵜ ជ 共j储 / B兲, but
not a term that can add energy to the magnetic field,
which is needed for a dynamo. The role of helicity and
its conservation in limiting the forms of magnetic dy-
namo theories remains controversial.
The external sources of helicity, the electrostatic Ses
and the inductive Sin, are important for creating the
plasma currents needed for magnetic surfaces in axisym-
metric devices. The inductive source is usually a toroidal
loop voltage Vt supplied by varying the magnetic flux in
a solenoid that goes through the central hole of the torus
while an essentially constant toroidal flux is supplied by
toroidal field coils. If the loop voltage Vt is small, the
plasma is relatively quiescent, and this is the mode of
operation of the tokamak. In the reversed-field pinch,
the loop voltage is made sufficiently large that the
plasma has periods of turbulence, which relax the cur-
rent profile to form transiently quiescent states. Electro-
static helicity injection Ses requires magnetic-field lines
that penetrate conducting plates, or electrodes. A given
magnetic-field line goes from a plate held at one voltage
to a plate held at a different voltage with the voltage
difference ⌽h driving a current along the magnetic-field
line. If ⌽h is small, the magnetic-field structure is af-
fected only slightly by the presence of the voltage. When
⌽h is large, the currents driven by the electric potential
difference produce a field that is large compared to the
field Bh that penetrates the plates. When ⌽h is large,
unstable plasma states can be created that undergo a
Taylor relaxation to plasmas that can have closed mag-
netic surfaces 共Jarboe et al., 1983; Raman et al., 2003;
Tang and Boozer, 2004兲.
IV. CONFINEMENT SYSTEMS
• The two most prominent magnetic configurations for
confining plasmas are tokamaks and stellarators. An
ideal tokamak is axisymmetric and uses a net toroi-
dal current to produce magnetic surfaces. The toka-
mak has the most extensive database of all magnetic
terium and tritium are burned under conditions simi-
lar to those of a fusion power plant. Stellarators use
1086 Allen H. Boozer: Physics of magnetically confined plasmas
FIG. 8. 共Color兲 The cross section of the NSTX spherical torus
共ST; Peng, 2000兲. All modern tokamaks 共Fig. 9兲 and ST devices
have a similar plasma cross section. The field lines that go from
the bottom of the plasma to the wall form a divertor; see Sec.
VII. Figure courtesy of the Princeton Plasma Physics Labora-
tory, Princeton, NJ.
helical shaping to produce at least part of the rota-
tional transform.
• The primary control a machine designer has over the
performance of a confinement system is the shape of
the outermost plasma surface. About 50 properties
of the plasma shape can be controlled but only about
four of these are consistent with axisymmetry: aspect
ratio, ellipticity, triangularity, and squareness.
In the world research effort on magnetically confined
plasmas, the tokamak, Fig. 4, and the stellarator, Figs.
5–7, are by far the largest programs. Tokamak plasmas
are axisymmetric and carry a net toroidal current in or-
der to form the magnetic surfaces. Stellarator plasmas
have the form of a torus with helical shaping, which pro-
duces some or all of the rotational transform ␫. Of the
two, the tokamak has been studied more and more data
have been amassed concerning it. It has been the basis
for a number of proposals for experiments in which deu-
terium and tritium are burned under conditions similar
to those of a fusion power plant. In addition, the toka-
mak has a variant, the spherical torus or spherical toka-
mak 共ST; Fig. 8兲. The ST is like a tokamak at a very tight
aspect ratio, ⑀a ⬅ a / Ro, much closer to unity, which has
physics advantages 共Peng, 2000; Sykes, 2001兲 as well as a
smaller unit size for fusion systems. For tokamaks a tight
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
aspect ratio gives the most desirable physics properties
but for stellarators the larger the aspect ratio Ro / a, the
easier it is to design desirable physics properties. The
helical shaping of the stellarator allows one to design
around certain issues of the tokamak and ST. Many of
these issues are associated with the maintenance and
stability of the net plasma current, issues that can be
avoided in the stellarator.
In addition to the tokamak and stellarator, many
other magnetic configurations are being actively pursued
in the world fusion program. Sheffield 共1994兲 has re-
viewed many of these configurations. The most promi-
nent are the reversed-field pinch 共Ortolani and Schnack,
1993; Prager, 1999兲, the spheromak 共Bellan, 2000兲, and
the magnetic dipole 共Garnier, Kesner, and Mauel, 1999;
Kesner et al., 2001兲.
The largest tokamaks have been the Joint European
Torus 共JET兲, which is at Culham, England 共Keilhacker et
al., 1999, 2001兲; the Tokamak Fusion Test Reactor
共TFTR兲, which was at Princeton, NJ 共Hawryluk et al.,
1998兲; and the Japanese Tokamak JT-60U 共Kamada et
al., 1999兲. The largest tokamak that is operating in the
U.S. is the DIII-D tokamak at General Atomics, Fig. 9,
where an investigation is underway to make tokamak
plasmas that are better suited for high-pressure, steady-
state operation 共Chan et al., 2000; Petty et al., 2000兲.
Both the JET and the TFTR tokamaks produced power
above the 10-megawatt level by fusing deuterium and
tritium 共Hawryluk, 1998兲. However, as expected from
the time of their design in the early 1970s, both toka-
maks required a high level of externally injected power,
comparable to or greater than fusion power, in order to
maintain the required plasma temperature, and neither
was, therefore, what is meant by a burning plasma ex-
periment.
The largest stellarator 共Yamada et al., 2001兲 is the
Japanese Large Helical Device 共LHD; Fig. 5兲. A stellar-
ator of comparable scale, Wendelstein-7X 共W7-X兲, is be-
ing constructed in Greifswald, Germany 共Beidler et al.,
1990兲. The W7-X stellarator, Fig. 6, has essentially no
net current and an unusually small current flowing par-
allel to the magnetic-field lines. This implies that the
shape of the magnetic surfaces is similar with and with-
out plasma. The W7-X design has many innovative fea-
tures of engineering and physics, which have been stud-
ied in the smaller W7-AS stellarator 共McCormick et al.,
2003兲 at Garching, Germany. The National Compact
Stellarator Experiment 共NCSX兲, Fig. 7, which is under
construction at the Princeton Plasma Physics Laboratory
共Zarnstorff et al., 2001兲, is quasiaxisymmetric; see Sec.
VI.E.1. Quasiaxisymmetry implies that the particle or-
bits closely resemble those of a tokamak even though a
large fraction of the rotational transform comes from
helical shaping. Quasiaxisymmetric stellarators have the
feature that they can be designed with an arbitrarily
small level of helical shaping, so there is a continuous
connection between quasiaxisymmetric stellarator de-
signs and tokamaks. The potential importance is that
quasisymmetry allows a minimal modification of the to-
kamak while giving the freedom to design around phys-
 Allen H. Boozer: Physics of magnetically confined plasmas
FIG. 9. 共Color兲 Cross section of the DIII-D tokamak. The
distance from the machine center line to the plasma center is
approximately 1.7 m. The plasma cross section of all modern
tokamaks is similar to that of DIII-D. The field lines that go
from the top of the plasma to the wall form a divertor; see Sec.
VII. Figure courtesy of Edward Lazarus, from Lazarus et al.,
1990.
ics issues that could potentially affect the use of the to-
kamak for practical fusion power.
The primary control that a machine designer has over
the plasma properties is the shape of the plasma. As
shown in Sec. V.A, a plasma equilibrium is specified
共Bauer, Betancourt, and Garabedian, 1984兲 by giving its
toroidal flux content, which is the ␺t enclosed by the
plasma surface; the rotational transform profile ␫共␺t兲; the
pressure profile p共␺t兲; and the plasma shape xជ s
= Rs共␪ , ␸兲R̂共␸兲 + Zs共␪ , ␸兲Ẑ. The optimal shape for the
plasma surface in both tokamaks and stellarators is far
from a simple torus with a circular cross section, Rs
= Ro + a cos ␪ and Zs = −a sin ␪. Since a tokamak design is
by definition axisymmetric, both Rs and Zs can depend
only on ␪. A frequently used expression for the shape of
a tokamak plasma is Rs共␪兲 = Ro + a cos共␪ + ⌬ sin ␪兲 and
Zs共␪兲 = −␬a sin ␪. The parameter ␬ is the elongation or
ellipticity, and ⌬ is the triangularity. Stellarators have a
much larger design space since Rs and Zs can depend on
both the poloidal angle ␪ and the toroidal angle ␸.
The number of plasma shape parameters that can be
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1087
controlled by coils at a sufficient distance from the
plasma for a fusion power system is limited to about
four for a tokamak and about ten times that number for
a stellarator 共see Sec. V.D.1兲. The number of free shape
parameters for the tokamak is sufficiently small that
they all have names: aspect ratio, ellipticity, triangularity,
and squareness, though squareness is not defined by an
agreed-upon shape function. The performance of toka-
maks is markedly improved by a careful choice of these
parameters relative to a circular cross section. The num-
ber of free parameters that are available to stellarator
designers is so large that extensive runs with optimiza-
tion codes are needed to choose design points. Many of
these optimization codes and techniques were devel-
oped by Jürgen Nührenberg and co-workers as part of
the W7-X design effort 共Nührenberg et al., 1995兲.
V. EFFICIENCY OF MAGNETIC CONFINEMENT
• The cost of fusion power depends on the efficiency
with which the magnetic field can be utilized and on
the fraction of the power output that must be used to
maintain the plasma and the magnetic field, which is
called the recirculating power fraction.
• The efficiency of magnetic-field utilization depends
on 共1兲 the ratio of the pressure of the plasma to that
of the magnetic field, ␤ ⬅ 2␮0p / B2, and 共2兲 the ratio
of the magnetic field at the plasma to that at the
coils. The plasma pressure, or ␤, can be limited by
equilibrium, stability, or transport issues.
• The power required to maintain the net current of a
tokamak places significant constraints on the design.
The cost of fusion power depends on the efficiency of
the confinement. Engineers tend to emphasize the ratio
of the power output of a fusion system to its mass. But
more physics-oriented efficiency measures are 共1兲 the
field strength on the coils needed to confine a given
plasma and 共2兲 the ratio of the power output to the
power required to sustain the current in the plasma and
coils, or more generally, the recirculating power fraction.
The power required to sustain the plasma current is dis-
cussed in Sec. VI.E.3.
The magnetic field at the coils that is needed to con-
fine a given plasma depends on two numbers. The first is
the plasma beta, which is the ratio of the plasma pres-
sure to the magnetic-field pressure,
2 ␮ 0p
␤⬅ , 共41兲
B2
and the second is the ratio of the magnetic-field strength
on the coils to that in the plasma. The limits on the
plasma beta due to equilibrium considerations are dis-
cussed in Sec. V.A, and limits from the existence of un-
stable perturbations of large spatial scale are discussed
in Secs. V.B and V.C. The considerations that set the
ratio of the magnetic-field strength on the plasma to that
on the coils are discussed in Sec. V.D. Section VI will
consider limitations on plasma beta, and hence the
magnetic-field strength, due to transport.
1088 Allen H. Boozer: Physics of magnetically confined plasmas
A. Equilibrium limits
• A plasma equilibrium is determined by the shape of
the outermost plasma surface, the toroidal flux en-
closed by that surface, and the profiles of the plasma
pressure, p共␺t兲, and rotational transform, ␫共␺t兲
⬅ d␺p / d␺t. The rotational transform of a magnetic
field is the average number of poloidal transits of the
torus a field line makes per toroidal transit. In toka-
maks, ␫ is generally less than unity.
• The plasma ␤ ⬅ 2␮0p / B2 is limited by the distortions
to the plasma shape caused by the current density j储
parallel to the magnetic field that arises to make
ជ · ជj = 0. That is, B
ⵜ ជ ·ⵜ
ជ 共j储 / B兲 = −ⵜ
ជ · ជj , with ជj given by
⬜ ⬜
the force exerted by the plasma, ⵜp = j ⫻ B ជ ជ ជ . These
distortions can be important when beta is just a few
percent, though much higher betas can be achieved
by careful plasma design. In asymmetric plasmas,
these distortions of the plasma shape can cause the
breakup of the magnetic surfaces that are needed for
plasma confinement.
ជ ·ⵜ
• A magnetic field that lies in surfaces, B ជ p = 0, has a
simple contravariant representation, Eq. 共7兲, which
means a representation that uses cross products of
pairs of coordinate gradients. A magnetic field asso-
ciated with a plasma equilibrium simultaneously has
a simple covariant representation, Eq. 共57兲 or Eq.
共58兲, which means a representation that uses coordi-
nate gradients. Coordinates in which the magnetic
field has both a simple contravariant and a simple
covariant representation simplify the analysis of
plasmas.
If at each point in a stationary plasma the plasma is
close to thermodynamic equilibrium, then the force ex-
erted by a stationary plasma is accurately approximated
by its pressure gradient −ⵜ ជ p. The equilibrium between
the pressure and electromagnetic forces is ⵜ ជ p = ជj ⫻ B
ជ,
which is known as the equilibrium equation for a
plasma.
A toroidal plasma equilibrium is specified by giving
共1兲 the shape of the magnetic surface that bounds the
plasma, xជ s共␪ , ␸兲, with ␪ and ␸ arbitrary poloidal and to-
roidal angles, 共2兲 the total toroidal magnetic flux in the
plasma, 共3兲 the pressure profile p共␺t兲, and 共4兲 the rota-
tional transform profile ␫共␺t兲 共Bauer, Betancourt, and
Garabedian, 1984兲. This result is proven by showing that
equilibria are extrema of the energy in a toroidal region
bounded by a fixed surface xជ s共␪ , ␸兲 when the ideal,
which means dissipationless, plasma constraints are ob-
served. The energy of the plasma plus the magnetic field
is
W= 冕 冉
plasma
B2
+
p
2␮0 ␥ − 1
d3x,冊 共42兲
where ␥ is the adiabatic index. The variation that is car-
ried out is in the shape of the magnetic surfaces
xជ 共␺t , ␪ , ␸兲 with the shape of the outermost plasma sur-
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
face xជ s共␪ , ␸兲 held fixed. That is, xជ 共␺t , ␪ , ␸兲 → xជ 共␺t , ␪ , ␸兲
+ ␰ជ , where ␰ជ is the displacement of the surfaces with the
displacement ␰ជ zero on the plasma surface.
Using the expressions that are derived below for the
variation in magnetic-field energy and the variation in
the pressure in response to a small plasma displacement
␰ជ , one finds that
冕
␦W = ជ 兲 · ␰ជ d3x,
共ⵜជ p − ជj ⫻ B 共43兲
plasma
which means W has equilibria, ⵜ ជ p = ជj ⫻ B
ជ as its extrema.
The minimization of the energy W gives the shape
xជ 共␺t , ␪ , ␸兲 of all of the magnetic surfaces in the plasma.
The best-known code for finding equilibria by minimiz-
ing the energy is called VMEC 共Hirshman, van Rij, and
Merkel, 1986兲. It should be noted that the adiabatic in-
dex ␥ does not appear in the equilibrium equation, so ␥
can be chosen arbitrarily. The choice ␥ = 0 is frequently
made because with this choice dp / dt = 0. When dp / dt
= 0, the pressure profile p共␺t兲 is unchanged by variations
␰ជ in the shape of the surfaces.
This paragraph gives the derivations of the variation
in magnetic field and the plasma pressure in response to
a small plasma displacement ␰ជ and can be skipped on a
first reading. These derivations have four parts:
共1兲 First, the relation between the total and the partial
time derivative must be explained. In fluid mechan-
ics the total time derivative dg / dt gives the rate of
change of any function g共xជ , t兲 in the frame of refer-
ence of the moving fluid, while the partial time de-
rivative 共⳵g / ⳵t兲xជ gives the time derivative at a fixed
spatial point. The partial derivative 共⳵g / ⳵t兲xជ is usu-
ally written as ⳵g / ⳵t. The rate of change of the po-
sition vector xជ is the flow velocity of the fluid, vជ
⬅ dxជ / dt. The chain rule implies dg / dt = ⳵g / ⳵t
ជ g, so the total time derivative is
+ 共dxជ / dt兲 · ⵜ
dg ⳵g
⬅ + vជ · ⵜជ g. 共44兲
dt ⳵t
共2兲 Second, the total time derivative of the canonical
coordinates must be shown to be zero in an ideal
fluid. The total time derivative is zero because the
coordinates are carried with the fluid, though the
partial time derivatives are nonzero if the fluid is
moving. This result is proven using the theory of
general coordinates 共Appendix兲, which implies that
each canonical coordinate of the transformation
function xជ 共␺t , ␪ , ␸ , t兲 obeys
d␺t
= 共vជ − uជ 兲 · ⵜជ ␺t , 共45兲
dt
with uជ ⬅ 共⳵xជ / ⳵t兲c 关Eq. 共29兲兴. In an ideal plasma, uជ
= vជ , so the canonical coordinates satisfy d␺t / dt = 0.
 Allen H. Boozer: Physics of magnetically confined plasmas 1089

ជ , in an
共3兲 Third, the variation in the magnetic field, ␦B
ideal plasma must be found for a small plasma dis-
placement ␰ជ . The variation is
ជ = ⵜជ ⫻ 共␰ជ ⫻ B
␦B ជ 兲.
This equation follows from Ohm’s law, Eq. 共32兲,
ជ + vជ ⫻ B
ជ = 0, plus Faraday’s
with zero resistivity, E
law, ⳵B ជ / ⳵t = −ⵜជ ⫻E ជ , and vជ = ⳵␰ជ / ⳵t. Actually vជ
⬅ dxជ / dt = ⳵␰ / ⳵t + vជ · ⵜ␰ជ , but for a sufficiently small
ជ ជ
displacement, ␰ជ → 0, one has vជ = ⳵␰ជ / ⳵t. The variation
in the magnetic energy is determined using a vector
ជ · ␦B
ជ = 共␰ជ ⫻ B
ជ 兲·ⵜ
ជ ⫻Bជ +ⵜ
ជ · 兵共␰ជ ⫻ B
identity to write B
⫻Bជ 其. The divergence term makes no contribution
because the plasma surface has a fixed shape, which
means ␰ជ = 0 on the bounding surface. As shown in
Sec. III.C, the rotational transform retains its initial
profile, ␫共␺t兲, through all variations in the plasma
shape when the resistivity is zero.
共4兲 Fourth, the pressure variation and the displacement
must be related. A pressure variation in an ideal
plasma conserves the entropy per particle, which im-
plies 共d ln p / dt兲 / 共d ln n / dt兲 = ␥. The plasma number
density obeys the continuity equation, ⳵n / ⳵t
ជ · 共nvជ 兲 = 0, so dp / dt = −␥pⵜ ជ · vជ . This means the
+ⵜ
change in the pressure at a fixed point in space, ␦p
⬅ 共⳵p / ⳵t兲xជ ␦t, is
␦p = − ␰ជ · ⵜជ p − ␥pⵜជ · ␰ជ .
The variation in the pressure can be rewritten in a form
that makes that part of the energy variation obvious
␦p = 共␥ − 1兲␰ជ · ⵜជ p − ⵜជ · 共␥p␰ជ 兲. The 共␥ − 1兲 term in ␦p cancels
the similar factor in W and the divergence term in ␦p
makes no contribution to the change in the energy since
the change in the shape ␰ជ is zero on the plasma bound-
ary.
An important theorem of equilibrium theory is that
given an arbitrary set of magnetic surfaces, xជ 共␺t , ␪ , ␸兲,
and an arbitrary rotational transform profile ␫共␺t兲, the
magnetic force ជf ⬅ ជj ⫻ B
ជ can always be written as ជf
B B
ជ ␺ with f 共␺ , ␪ , ␸兲 a known function. The angles ␪
= f ␺tⵜ
t ␺t t
and ␸ are arbitrary poloidal and toroidal angles, but ␺t is
the toroidal magnetic flux enclosed by a magnetic sur-
face. If xជ 共␺t , ␪ , ␸兲 and ␫共␺t兲 are consistent with an equi-
librium, the function f␺t depends on ␺t alone, and the
equilibrium pressure profile p共␺t兲 is given by dp / d␺t
= f . To prove this theorem, note that if ⵜ ជ ␺ ⫻ ជf
␺t
ជ ·ⵜ
ជ ␺ 兲jជ − 共jជ · ⵜ
ជ ␺ 兲B
ជ is zero, then ជf = f ⵜជ
⬅ 共B ␺t ␺t. Since
冉 冊 ជ ជ ជ
冉 冊 ជ
ជ = 1 + ⳵␭ ⵜ ␺t ⫻ ⵜ ␪ + ␫ − ⳵␭ ⵜ ␸ ⫻ ⵜ ␺t ,
B
⳵␪ 2␲ ⳵␸ 2␲
共48兲
which is proven using the argument that led to Eq. 共7兲.
共46兲 ជ ␺ = 共ⵜជ ⫻Bជ 兲·ⵜជ ␺ =ⵜ ជ · 共B
ជ ⫻ⵜ ជ ␺ 兲=0
The constraint ␮0ជj · ⵜ t t t
determines ␭ as the solution to a Poisson-like equation
that has derivatives only in the ␪ and ␸ coordinates. Pe-
riodicity in ␪ and ␸ implies that ␭ has a definite expres-
sion, which can be determined magnetic surface by mag-
netic surface. The function ␭ is the difference between
the arbitrary and the magnetic poloidal angles, ␪m = ␪
+ ␭, and determines how the field lines wind through the
ជ兲 magnetic surfaces. Once ␭ is determined, one has a defi-
nite expression for B ជ , which gives the magnetic force ជf .
B
The calculation of the curl of the magnetic field is com-
plicated but is explained in the Appendix.
The equilibrium with profiles ␫共␺t兲 and p共␺t兲 has been
shown to be an extremum of W = 兰兵共B2 / 2␮0兲 − p共␺t兲其d3x.
The extremum can also be obtained by allowing arbi-
trary variations in both the ␭ function of Eq. 共48兲 and
the shape of nested magnetic surfaces xជ 共␺t , ␪ , ␸兲 with the
outermost plasma surface held fixed. These variations,
unlike the variation in shape that led to Eq. 共43兲, are
unconstrained and do not use Eqs. 共46兲 or 共47兲. This
alternate method of extremizing the energy follows from
the magnetic field’s obeying the ideal constraints if ␫共␺t兲
is conserved and if the ␺t surfaces of xជ 共␺t , ␪ , ␸兲 remain
nested. The same Poisson-like equation for ␭ is obtained
when 兰B2d3x is minimized with respect to ␭.
共47兲
One can always minimize the energy W over any set
of shape functions xជ 共␺t , ␪ , ␸兲, so it may at first appear
that equilibria always exist. This is not true for two rea-
sons. First, the set of shape functions that are considered
may not be rich enough to find an extremum in which f␺t
is a function of ␺t alone, as is required for equilibrium.
Second, if one considers an arbitrarily rich set of shape
functions, the parallel component of the current, j储
ជ / B, may become singular at the rational surfaces.
⬅ ជj · B
Rational magnetic surfaces are defined by the rotational
transform’s being the ratio of two integers, ␫ = n / m.
These are surfaces on which the magnetic-field lines
close on themselves after m toroidal and n poloidal cir-
cuits of the torus.
The parallel current determines much of the theory of
force balance of a plasma embedded in a magnetic field.
The parallel current is given by ⵜ ជ · ជj = Bជ ·ⵜ
ជ 共j储 / B兲 + ⵜ
ជ · ជj
⬜
= 0, with the perpendicular current determined by the
force the magnetic field exerts on the plasma, ជfB ⬅ ជj ⫻ B ជ.
The equation for variation of the parallel current along
t B the field lines is

冉 冊
t t B
ជ ·ⵜ
ជ ␺ = 0 by the definition of magnetic surfaces, the re- ជ ជ
B t ជ · ⵜជ j储 = ⵜជ · fB ⫻ B ,
B 共49兲
ជ ␺ = 0 can
sult is proven if the equilibrium constraint ជj · ⵜ B B 2
t
be imposed. The most general divergence-free field,
ជ ·B
ⵜ ជ = 0, that has the required surfaces, Bជ ·ⵜ ជ ␺ = 0, and which is a magnetic differential equation, Eq. 共12兲, for
t
rotational transform profile, ␫共␺t兲, is the parallel current. In an equilibrium plasma, ជf = ⵜ
ជ p.
B

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1090 Allen H. Boozer: Physics of magnetically confined plasmas

The subtlety of the parallel current in a plasma equi- ជ ⫻B

of the current, ជj = ⵜ ជ / ␮ , Eq. 共53兲. The derivation of
0
librium is clarified by the equation for j储 / B in 共␺t , ␣ , ␾兲 the contravariant representation of the magnetic field,
Clebsch coordinates, ជ ·B
ជ = 0 and 共2兲
Eq. 共7兲, depended on two conditions: 共1兲 ⵜ

冉冊 冉 冊
⳵ j储
⳵␾ B
=−
dp ⳵ 1
d␺t ⳵␣ B2
. 共50兲 that a nonconstant function ␺t exist with B ជ ·ⵜ
These two conditions also hold for the current associ-
ជ ␺ = 0.
t

ated with a plasma equilibrium, since it is divergence-

On closed magnetic-field lines u ⬅ 养 d␾ / B2 = 养 dᐉ / B
ជ ␺ = 0, so the current
free and in equilibrium satisfies ជj · ⵜ
must be independent of ␣ for a single-valued j储 / B to t
exist. To have a nonzero pressure gradient on a rational associated with an equilibrium must have the contravari-
surface, u must have the same value on every field line ant representation
of that surface. In 共␺ , ␣ , ␾兲 coordinates, B ជ =ⵜជ␾ ជ ជ ជ ជ
ជj = − ⳵Gtot ⵜ ␸ ⫻ ⵜ ␺t + ⳵Itot ⵜ ␺t ⫻ ⵜ ␪m ,
t
ជ ␺ +B ⵜ
ជ 共53兲
+ B ␺tⵜ t ␣ ␣, Eq. 共11兲, which can be simplified in an ⳵␺t 2␲ ⳵␺t 2␲
equilibrium plasma to B ជ =ⵜជ ␾+B ⵜជ ␺ since ⵜជ ⫻Bជ ·ⵜ
ជ␺
␺t t t
= 0. The divergence of the perpendicular current is where the radial derivatives of the total poloidal and
ជ · 兵共B
ជ ⫻ⵜ
ជ p兲 / B2其 = 共dp / d␺ 兲兵ⵜ
ជ ␾ · 共ⵜ
ជ ␺ ⫻ⵜ
ជ ␣兲其⳵共1 / B2兲 / ⳵␣, total toroidal current are
ⵜ t t
so ⵜ ជ · j = B2共dp / d␺ 兲⳵共1 / B2兲 / ⳵␣. The divergence of the ⳵Gtot dG共␺t兲 ⳵␯共␺t, ␪m, ␸兲
⬜ t
= + 共54兲
parallel current is B ជ ·ⵜ
ជ 共j储 / B兲 = B2⳵共j储 / B兲 / ⳵␾. The con- ⳵␺t d␺t ⳵␸
ជ
straint that ⵜ · j = 0 gives Eq. 共50兲. and
A solution of Eq. 共49兲 for j储 / B is the sum of 共1兲 a
special solution, which varies over the magnetic surface ⳵Itot dI共␺t兲 ⳵␯共␺t, ␪m, ␸兲
= + . 共55兲
and is given by any solution to the magnetic differential ⳵␺t d␺t ⳵␪m
equation, and 共2兲 a homogeneous solution in which j储 / B
is constant along field lines, B ជ ·ⵜ
ជ 共j储 / B兲 = 0. Both the spe- The function G共␺t兲 = 兰jជ · daជ ␪m is the poloidal current in
h
cial and the homogeneous solution can be singular at the region exterior to a constant-␺t surface, which is the
rational surfaces. A singularity of the special solution, current through the hole in the torus, Fig. 1, while
共j储 / B兲s = jPS / B, where jPS is the Pfirsch-Schlüter current I共␺t兲 = 兰jជ · daជ ␸ is the toroidal current in the region interior
共Pfirsch and Schlüter, 1962兲, can arise if the pressure gra- to a constant-␺t surface, which is the current through a
dient dp / d␺t is nonzero at a rational surface on which cross section of the torus 共Fig. 1兲. The contravariant rep-
养dᐉ / B varies. This singularity is discussed more exten- resentation of the current implies the magnetic field has
sively below. The homogeneous solution, 共j储 / B兲h = jnet / B, the covariant repesentation
where jnet is the net current, has the form
ជ = ␮0 兵G共␺ 兲ⵜជ ␸ + I共␺ 兲ⵜជ ␪ − ␯ⵜជ ␺ + ⵜជ F其.
B 共56兲
j t t m t
2␲
␮0 net = k共␺t兲 +
B
兺m,n
kmnei共n␸−m␪m兲␦共␺t − ␺mn兲, 共51兲
We shall find that the function ␯共xជ 兲 is determined by the
with ␺mn defined by ␫共␺mn兲 = n / m and ␦共␺t − ␺mn兲, the equilibrium equation. The function F共xជ 兲 is determined
Dirac delta function. Each nonzero constant kmn implies by the constraint that ⵜ ជ ·B
ជ = 0 and the boundary condi-
a surface current on a rational surface, which produces a tions on the magnetic field.
normal magnetic field that cancels the resonant part of The contravariant representation of the magnetic
the normal field due to external perturbations and pre- field, Eq. 共52兲, is unchanged if one defines new poloidal
vents the formation of an island. and toroidal angles by ␪m = ␪n + ␫␻ and ␸ = ␸n + ␻, where
A more complete exploration of the special solution
␻ is any well-behaved function of position. If these
for j储 / B of Eq. 共49兲, the Pfrisch-Schlüter current, re- forms are substituted into the covariant representation
quires a covariant representation, Eq. 共56兲, of Bជ in gen- of the magnetic field, Eq. 共56兲, one obtains an equation
eral magnetic coordinates. Magnetic coordinates are de- of the same form but with ␯ replaced by ␯n ⬅ ␯
fined by B ជ having the contravariant representation, + 共dG / d␺t + ␫dI / d␺t兲␻ and with F replaced by Fn ⬅ F
+ 共G + ␫I兲␻. There are two obvious choices for ␻. The
ជ = ⵜជ ␺ ⫻ ⵜជ ␪ + ␫共␺ 兲ⵜជ ␸ ⫻ ⵜជ ␺ ,
2␲B 共52兲 first makes ␯n = 0, and the resulting coordinates are
t m t t
called Hamada coordinates 共␺t , ␪H , ␸H; Hamada, 1962兲,
which follows from Eq. 共7兲 and ␫ ⬅ d␺p / d␺t. The general
in which
covariant representation can be obtained from the cova-
ជ =ⵜ
ជ ␾+B ⵜ ជ
riant expression in Clebsch coordinates, B ␺t ␺ t, ជ = ␮0 兵G共␺ 兲ⵜជ ␸ + I共␺ 兲ⵜជ ␪ + ⵜជ F其.
that was used in the derivation of Eq. 共50兲 for B t H t H 共57兲
2␲
⳵共j储 / B兲 / ⳵␾. However, the derivation is almost as simple
starting from fundamental principles. The second makes Fn = 0, and the resulting coordinates
The covariant representation of the magnetic field, are called Boozer coordinates 共␺t , ␪B , ␸B; Boozer, 1981兲,
Eq. 共56兲, follows from the contravariant representation in which

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas 1091

ជ = ␮0 兵G共␺ 兲ⵜជ ␸ + I共␺ 兲ⵜជ ␪ + ␤ ⵜជ ␺ 其.

B 共58兲 G
dI
−I
dG
G
⳵␯
−I
⳵␯
t B t B * t
2␲ j储 d␺t d␺t ⳵␪B ⳵␸B
= + . 共64兲
The cross product between the contravariant repre- B G + ␫I G + ␫I
sentation of the current, Eq. 共53兲, and of the magnetic
The first term on the right-hand side is jnet / B = k共␺t兲 / ␮0
field, Eq. 共52兲, must equal the pressure gradient. This
with jnet the net current, and the second term is jPS / B
equality gives an equation for ␯, Eq. 共63兲, and an equa-
with jPS the Pfirsch-Schlüter current. Equation 共64兲 for
tion for the average equilibrium on a magnetic surface,
j储 / B is obtained from the dot product of Eqs. 共53兲 and
Eq. 共62兲. Using arbitrary magnetic coordinates
共58兲.
共␺t , ␪m , ␸兲,
In axisymmetric equilibria the expression for the

冉 冊
Pfirsch-Schlüter current can be simplified to
ជ ជ
ជ = − B · ⵜ ␸ ⳵Gtot + ␫ ⳵Itot ⵜជ ␺ .
ជj ⫻ B
2␲ ⳵␺t ⳵␺t
t 共59兲 jPS 1 G
B
=
dV
−
␫ G + ␫I d␺t B冉 2␲ dp
ជ · ⵜជ ␸ d␺t 冊
, 共65兲

ជ p = 共dp / d␺ 兲ⵜ
Since ⵜ ជ␺ ,
t t ជ ·ⵜ
with dV / d␺t = 养 d␪m / B ជ ␸. This equation for j follows
PS
from Eq. 共64兲 with ⳵␯ / ⳵␸B = 0 and Eq. 共63兲, which gives
⳵Gtot ⳵Itot 2␲ dp
+␫ =− . 共60兲 an expression for ⳵␯ / ⳵␪B. Equation 共65兲 is written in a
⳵␺t ⳵␺t ជ · ⵜជ ␸ d␺t
B form that is valid in any coordinates in which ␸ is the
toroidal angle and ␺t is the toroidal flux.
The contravariant form for the magnetic field, Eq. 共52兲, Even in simple axisymmetric equilibria, the Pfirsch-
implies that the Jacobian of magnetic coordinates, 1 / J Schlüter current jPS provides a significant limitation on
ជ ␺ ⫻ⵜ
= 共ⵜ ជ ␪ 兲·ⵜ
ជ ␸, is the plasma pressure. At large aspect ratio, Ro / r Ⰷ 1, the
t m
vacuum toroidal field is dominant, that is, G Ⰷ ␫I, and
1/Bជ ·ⵜ
ជ ␸ ⬇ R2 / 共R B 兲. The Pfirsch-Schlüter current, Eq.
1 o o
J= , 共61兲 共65兲, is in the toroidal direction and approximated by
ជ · ⵜជ ␸
2␲B
2共R − Ro兲 1 dp
jPS ⬇ − , 共66兲
with the volume enclosed by the magnetic surfaces ␫Bo r dr
V共␺t兲 = 兰Jd␺td␪md␸. If Eq. 共60兲 is averaged over the po-
loidal and toroidal angles, one obtains the average equi- with R − Ro = r cos ␪. If the pressure is assumed to de-
librium equation 共Kruskal and Kulsrud, 1958兲, pend on the minor radius r as p = p0共1 − r2 / a2兲, one finds
the Pfirsch-Schlüter current produces a magnetic field in
dG dI dV dp the vertical or Ẑ direction, ⳵␦BZ / ⳵R = −␮0jPS, which at
+␫ =− , 共62兲 the plasma edge, 共R − Ro兲2 = a2, is ␦BZ ⬇ −2␮0p0 / 共␫Bo兲.
d␺t d␺t d␺t d␺t
This vertical field opposes the poloidal field due to the
which is a coordinate-independent equation. The func- plasma current, B␪ ⬇ ⑀a␫Bo, on the inboard side with ⑀a
tion ␯ satisfies a magnetic differential equation, ⬅ a / Ro. When the vertical field becomes sufficiently
strong to cancel the poloidal field on the inboard side,

冉 冊 冉 冊
the equilibrium has reached its pressure limit. In other
⳵ ⳵ dV dp
+␫ ␯= − 共2␲兲2J . 共63兲 words, equilibria must have a volume-averaged beta,
⳵␸ ⳵␪m d␺t d␺t 具␤典 ⬅ 2␮0具p典 / Bo2 with 具p典 = p0 / 2, which satisfies

The function ␯ and hence the current are singular at a 1

rational surface ␫ = n / m unless either the resonant Fou-
rier term of 2␲J = 1 / Bជ ·ⵜ
ជ ␸ is zero or the pressure gradi-
ent dp / d␺t is zero.
Since Eq. 共63兲 for ␯ holds in any set of magnetic coor-
dinates, it must hold in Hamada coordinates, in which
␯ = 0. This means the Jacobian of Hamada coordinates is
independent of the angles, 共2␲兲2JH = dV / d␺t, but also
that the coordinate transformation to Hamada coordi-
nates is singular if the function ␯ is singular when one
uses the cylindrical angle ␸ as the toroidal angle.
The Jacobian of Boozer coordinates can be obtained
by dotting the covariant and contravariant representa-
tions of Bជ together, J = ␮ 共G + ␫I兲 / 共2␲B兲2. The parallel
B 0
current is particularly simple in these coordinates,
具 ␤ 典 ⱗ ⑀ a␫ 2 . 共67兲
2
Stability limits the average rotational transform in a to-
kamak to ␫ ⬇ 1 / 2 and the inverse aspect ratio is ⑀a
⬇ 1 / 3, so 具␤典 ⱗ 4% when the magnetic surfaces are cir-
cular. Actual tokamaks with additional shaping have
achieved 具␤典 greater than 10%, and ST plasmas have
achieved 具␤典 approaching 50%. However, the poor scal-
ing of the beta limit with aspect ratio limits tokamaks to
having a tight aspect ratio. The weakening of the poloi-
dal field on the inboard side and the strengthening on
the outboard side cause the magnetic surfaces to shift
outward, a phenomenon called the Shafranov shift, so
the same flux passes between the plasma axis and the
inboard and outboard edges.

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1092 Allen H. Boozer: Physics of magnetically confined plasmas
The equilibrium of an axisymmetric plasma can also
be found using the Grad-Shafranov equation 共Lüst and
Schlüter, 1957; Shafranov, 1958; Grad and Rubin, 1959兲.
Using 共R , ␸ , Z兲 cylindrical coordinates, an arbitrary axi-
symmetric, divergence-free field can be written in a
mixed covariant-contravariant representation as 2␲B ជ full electron density 共neutral plus ionized兲 moving at the
ជ ␸+ⵜជ ␸⫻ⵜ ជ ␺ , which follows from ⵜ ជ · 共ⵜ
ជ ␸兲 = 0 in
= ␮0Gⵜ p only eight orders of magnitude larger. Actually, the local
cylindrical coordinates. The curl of this mixed represen-
ជ G⫻ⵜ
tation gives 2␲ជj = ⵜ ជ ␸ + ␸ˆ 共⌬ * ␺ 兲 / 共␮ R兲, where
p 0
⌬ * ␺p ⬅ R 冉 冊
⳵ 1 ⳵␺p
⳵R R ⳵R
+
⳵ ␺p
⳵Z2
2
共68兲
is called the Grad-Shafranov operator and ␸ˆ is a unit
ជ fluid, as they are in the outer solar convective zone, they
vector of cylindrical coordinates. From ជj ⫻ B
ជ ␺ one finds G is a function of ␺ alone and
= 共dp / d␺p兲ⵜ p p tion distances 共Thiffeault and Boozer, 2003兲. These cur-
dG dp
⌬ * ␺p = − ␮20G − ␮0共2␲R兲2 , 共69兲
d␺p d␺p
which is the Grad-Shafranov equation.
Equation 共49兲 for the variation in the parallel current
also has important implications for nontoroidal plasmas
embedded in magnetic fields. For example, the photo-
sphere of the sun becomes sufficiently tenuous with alti-
tude that the forces exerted by the magnetic field on the
plasma must be small, ជfB ⬅ ជj ⫻ Bជ → 0. In this situation,
which is called a force-free equilibrium, the parallel cur-
rent satisfies the conservation law that j储 / B be constant
along magnetic-field lines. The constancy of j储 / B implies
that stars, like the sun, with outer convective zones and
a magnetic field, must have a corona, a region in the
upper solar atmosphere of very-high-energy electrons
compared to the energy that would be expected from
thermal diffusion 共Boozer, 1999兲. The reason is that the
current j = envs is the product of the electron charge,
number density, and streaming velocity vs of the elec-
trons relative to the ions. If the streaming velocity ex-
ceeds the electron thermal velocity ve ⬅ 冑Te / me, the
electrons will break away from a Maxwellian distribu-
tion and reach whatever energy is necessary to carry the
current. This is the phenomenon of runaway electrons
共Dreicer, 1960; Connor and Hastie, 1975兲, which arises
from the collision frequency’s dropping as one over the
speed of a particle cubed, Eq. 共134兲, when the speed is
larger than the thermal velocity. In the absence of a
corona in a star like the sun, the density of the atmo-
sphere drops exponentially with height, n共z兲
⬀ exp共−z / h兲, due to gravitational force balance, ⵜ ជ p = ␳gជ ,
with a scale height h ⬇ 100 km. Although the fractional
ionization increases with altitude, the total number of
electrons that are potentially available to carry the cur-
rent rapidly drops. The magnetic field above the sun has
structures with a scale of 104 km, spanning a range of
atmospheric densities of e100. Since j储 / B is approximately
constant due to the rapid drop in the atmospheric pres-
sure, the streaming parameter vs / ve would increase by a
factor of roughly e100 ⬇ 1043 without a corona. This is of
course impossible without the streaming parameter be-
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
coming greater than one and the electrons running away.
At the bottom of the photosphere, magnetic fields are
observed to have variations of 0.1 T over 103 km, so lo-
cal current densities have to be greater than j
⬇ 102 A / m2. The current that could be carried by the
electron thermal speed is of order 1010 A / m2, which is
current density on magnetic-field lines emerging from
the sun should have a value orders of magnitude larger
than the observed current density of 102 A / m2, which by
the observation process involves a spatial average. The
reason a large current density should be expected is that
when field lines are churned around in a conducting
develop strong parallel currents with very short correla-
rents flow all along the field lines, Eq. 共49兲, relaxing to
their equilibrium values via shear Alfvén waves; see Sec.
VI.H. Runaway electrons are not generally considered
as an explanation for the actual solar corona. Whether
the currents along the magnetic-field lines that penetrate
the solar surface have sufficient strength and shortness
of correlation length to cause the actual solar corona are
questions that are not easily answered observationally.
B. Stability limits
• A magnetically confined plasma cannot be in ther-
modynamic equilibrium, so a potential for instability
always exists. Instabilities can be driven by the pres-
sure gradient or the net current in the plasma.
• The pressure gradient is destabilizing when the cen-
ter of curvature of the magnetic-field lines, ␬ជ
ជ b̂ with b̂ ⬅ B
⬅ b̂ · ⵜ ជ / B, is in the direction of higher
pressure 共bad curvature兲 and stabilizing when the
curvature has the opposite sign 共good curvature兲.
• Stability calculations can be done by preserving the
constraints of an ideal, dissipationless, plasma evolu-
tion, which prevent islands from opening but give
singular currents, or by keeping smooth current dis-
tributions, which allow islands to open. Through lin-
ear order in the perturbation theory, there is no dis-
tinction in these two types of analyses except near
rational surfaces ␫ = n / m. However, the ideal analysis
always predicts greater stability.
A magnetically confined plasma always has the ther-
modynamic free energy that is required for instability.
The maximum entropy state of the charged particles
that form a plasma is a Gibbs distribution, exp共−H / T兲,
but the particle energy H = 21 mv2 + q⌽ is independent of
ជ 共xជ 兲. A plasma that is carrying a cur-
the magnetic field B
rent cannot be in a Gibbs distribution and therefore al-
ways has thermodynamic free energy. The stability of
current-carrying plasmas depends on the existence of
constraints on the plasma motion that prevent the re-
duction in the free energy.
 Allen H. Boozer: Physics of magnetically confined plasmas
The free energy that is available for instabilities al-
lows toroidal plasmas to spontaneously break their sym-
metry by kinking. This kinking is driven by the gradients
in the parallel current distribution, k ⬅ ␮0j储 / B, and in the
pressure. Both drives limit the achievable plasma beta
共ITER Physics Expert Group on Disruptions, Plasma
Control, and MHD et al., 1999兲. Books by Bateman
共1978兲 and Freidberg 共1987兲 discuss plasma stability un-
der the assumptions of ideal magnetohydrodynamics
共MHD兲, which means in the absence of dissipative ef-
fects.
Instabilities driven by the gradient in the parallel cur-
rent distribution, d共j储 / B兲 / d␺t, limit the magnitude of the
poloidal magnetic field or more precisely the rotational
transform, and both equilibrium and stability limits in-
volve the strength of the poloidal magnetic field.
The pressure limit due to plasma stability for a toka-
mak plasma is approximated with remarkable success,
even for highly shaped plasmas, by the Troyon limit
共Troyon and Gruber, 1985兲. The volume-averaged ␤
⬅ 2␮0p / B2 can be no greater than
␤n ␮0 I
具␤典 = , 共70兲
10 4␲ aB
where I is the net toroidal current and a is the half-width
of the plasma in the Z = 0 plane. The Troyon coefficient,
also called beta normal, is a number which is ␤n ⬇ 3 in
somewhat optimized plasmas but can be a factor of 2
larger. The results of tokamak experiments are often ex-
pressed by giving the ␤n that was achieved.
The stability of a plasma to a pressure gradient is
largely determined by the relative orientation of the
pressure gradient ⵜ ជ p and the curvature of the magnetic
ជ b̂, where b̂ ⬅ B
field lines, ␬ជ ⬅ b̂ · ⵜ ជ / B is a unit vector
along the magnetic field. For example, the curvature of a
circular line b̂ = ␸ˆ 共␸兲 of 共R , ␸ , Z兲 cylindrical coordinates
is ␬ជ = −R̂ / R. The pressure gradient directly enters the
energy change due to a displacement ␰ជ of an equilibrium
plasma through −兰共␰ជ · ⵜ ជ p兲共␰ជ · ␬ជ 兲d3x 关Eq. 共92兲兴. Good cur-
vature means ␬ជ · ⵜ ជ p ⬍ 0 and bad curvature means ␬ជ · ⵜ ជp
⬎ 0. Good curvature is stabilizing because that geometry
compresses plasma motion down the pressure gradient.
Unfortunately, topology implies the field-line curvature
must be bad somewhere on a toroidal magnetic surface,
and pressure-driven instabilities localize in these places.
Localization is stabilizing because it implies a bending of
the magnetic field. If the bad curvature regions are suf-
ficiently localized, the pressure gradient does not desta-
bilize the equilibrium.
If the plasma is assumed to be ideal, which means zero
resistivity, then the evolution equations for the magnetic
field imply the rotational transform ␫ is a time-
independent function of ␺t. In the presence of a pertur-
bation, this assumption generally leads to a singular net
plasma current, ␮0ជjnet = kB ជ with B ជ ·ⵜ
ជ k = 0. One can avoid
a singular net current by assuming that the distribution
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1093
of net current k共␺t兲 is independent of time instead of
␫共␺t兲. The assumption of a fixed net current is a resistive
stability analysis.
Remarkably, linear perturbation theory makes no dis-
tinction between an ideal analysis, ␫共␺t兲 independent of
time, and a resistive analysis, k共␺t兲 independent of time,
except at resonant rational surfaces, ␫ = n / m. In an ideal
analysis a singular current arises at resonant rational sur-
faces and in a resistive analysis an island opens. Away
from resonant rational surfaces, the change in the rota-
tional transform scales as the amplitude of the magnetic
perturbation squared and, therefore, cannot enter a lin-
ear stability analysis. To see this consider the perturbed
field-line Hamiltonian ␺p = ␺¯ p共␺t兲 + ␺1 cos共n␸ − m␪兲. One
can solve for the magnetic-field lines to second order in
the perturbation ␺1 and find that the rotational trans-
form is changed by an amount
再 冉 冊 冎
m d 1 d␺21 m␫⬙␺21
⌬ ␫ 共 ␺ t兲 = + , 共71兲
4 d␺t n − ␫m d␺t 共n − ␫m兲2
where ␫⬙ ⬅ d2␫ / d␺2t . The perturbation ␺1 is linear in the
perturbation amplitude for a sufficiently small perturba-
tion, so ⌬␫ is quadratic in the perturbation amplitude
where n − ␫m is nonzero.
1. Current-driven instabilities
• The strongest effect of the net plasma current on
stability is near the rational surfaces, ␫ = n / m. Defin-
ing a helical flux by d␺h ⬅ 共␫ − n / m兲d␺t, a gradient in
the net current is stabilizing if d共j储 / B兲 / d␺h is positive
and destabilizing if d共j储 / B兲 / d␺h is negative.
• Only perturbations with low poloidal mode number
m can be driven unstable by the net plasma current.
These instabilities are called kinks.
• A current density proportional to a Dirac delta func-
tion generally arises on the rational surfaces, ␫
= n / m, in the presence of perturbations that conserve
the constraints of an ideal, dissipationless, plasma.
The amplitude of these delta-function currents is
proportional to the jump in the resonant Fourier
term of the radial displacement of the plasma,
ជ␺ 兴 .
关␰ជ · ⵜ t mn
The stability properties of a force-free equilibrium,
ជ
ⵜ⫻Bជ = k共␺ 兲B
ជ , illustrates the effect of the distribution of
t
net plasma current k共␺t兲 on plasma stability. The funda-
mental results are that the net current affects stability
primarily through the gradient dk / d␺t. Near a rational
surface ␫ = n / m the gradient of the net current distribu-
tion enters through the quantity
m dk
, 共72兲
n − ␫m d␺t
which is destabilizing when positive and stabilizing when
negative. In other words, plasma stability is extremely
sensitive to the gradient in the current distribution k
⬅ ␮0j储 / B near rational surfaces, but the effect on stabil-
1094 Allen H. Boozer: Physics of magnetically confined plasmas

ity of the gradient is opposite on the two sides of the Ro m dk0

rational surface. For example, in a tokamak, a low-order ␦k = ␦A储 , 共73兲
rB0 n − ␫m dr
rational surface just outside the plasma is very destabi-
lizing because dk / d␺t is negative near the plasma edge ជ ⫻ ␦B
and the equation ⵜ ជ = ␦kB
ជ is
0

冉 冊
while ␫ is positive but becoming smaller, d␫ / d␺t ⬍ 0. The
destabilizing effect of the current gradient is obtained on 1 d dÃ储 m2Ã储 Ro m dk0
r − 2 =− Ã储 . 共74兲
the plasma side of the rational surface, but there is no r dr dr r r n − ␫m dr
current to provide a stabilizing effect on the opposite
side of the rational surface. This instability, which is The rotational transform and the current distribution
are related by
called an external kink, is particularly pronounced when
q = 1 / ␫ is just above an integer at the plasma edge. 1 dr2共␫ − ␫v兲
The stability of force-free equilibria is very sensitive = Rok0共r兲, 共75兲
r dr
to the gradient of the current distribution k due to the
ជ ·ⵜ
ជk where ␫v共r兲 is the rotational transform of the vacuum, or
properties of the magnetic differential equation B
current-free, magnetic field. In a tokamak, ␫v is zero, but
= 0, which must hold for both the perturbed and the un- it is nonzero in a stellarator.
perturbed equilibrium. In the perturbed equilibrium, k Externally driven currents are required to perturb the
ជ + ␦B
= k 共␺ 兲 + ␦k and 共B ជ 兲·ⵜ
ជ k = 0. The unperturbed equi- plasma, but it is mathematically simpler to view these as
0 t 0
ជ ⫻B
librium is ⵜ ជ = k 共␺ 兲B ជ , and the magnetic perturba- lying just outside the plasma. To understand the need for
0 0 t 0
ជ ⫻ ␦Bជ =Bជ ␦k + k ␦B
ជ , where Bជ ·ⵜ
ជ ␦k externally driven currents, note that solutions to Eq. 共74兲
tion is given by ⵜ 0 0 should be well behaved as r → 0, which leaves only one
ជ ជ
= −␦B · ⵜk . That is, ␦k is given by a magnetic differential free boundary condition. The solution to Eq. 共74兲 out-
0
equation and is proportional to dk0 / d␺t. The perturbed side of the plasma, r ⬎ a, is Ã储 ⬀ 1 / rm with the assumption
current distribution ␦k can become very large near a that there are no perturbing currents away from the
resonant rational surface, ⬀1 / 共n − ␫m兲, because of the plasma boundary. The matching condition at the plasma
properties of the magnetic differential equation. The boundary at r = a is that Ã储 be continuous. However, the
largeness of ␦k is what allows unstable perturbations to
radial derivative of Ã储 cannot also be made continuous,
exist.
for that would be a second boundary condition. The
Instabilities due to the gradient in the net current are
driven by the parallel part of the perturbed vector po- jump in the derivative, which is denoted by 关⳵Ã储 / ⳵r兴,
ជ 储 ⬅ 共b̂ · ␦A
ជ 兲b̂ with b̂ ⬅ B
ជ / B. This is proven by gives the current that must be externally driven at the
tential, ␦A
edge of the plasma to produce the magnetic perturba-
noting that the three independent components of any tion.
vector, including the perturbed vector potential, can be The stability of a magnetic perturbation to a plasma is
written at each spatial point as ␦A ជ = 共␦A / B兲Bជ determined by the sign of the power that is needed to
B
+ ␦ A ␺ⵜជ ␺ +ⵜ ជ ␦g. If one chooses a gauge in which drive the current on the plasma surface that supports the
t
ជ ·ⵜជ ␦g = 0, then ␦A = ␦A储 and ␦B ជ ·ⵜ
ជ k 共␺ 兲 = 共ⵜ
ជ perturbation. Positive power means stability and nega-
B B 0 t tive power instability. The surface current that is re-
⫻ ␦A ជ 储兲 · ⵜ
ជ k . This representation of ␦A
ជ must be used quired to support the perturbation is
0

冋 册
with care. Given a change in the poloidal flux, ␦␺p
ẑ ⳵␦A储
ជ · 共⳵xជ / ⳵␸兲d␸, the change in the toroidal flux, ␦␺
= − 养 ␦A ␦ជj = − ␦共r − a兲 , 共76兲
t
␮0 ⳵r
= 养 ␦Aជ · 共⳵xជ / ⳵␪兲d␪, is constrained if ␦g is a single-valued
function of ␪ and ␸. One can satisfy the constraint where it is assumed that there are no currents outside of
the plasma surface, ␦A储 ⬀ 共a / r兲m for r ⬎ a. This follows
Bជ ·ⵜ
ជ ␦g = 0 and have arbitrary changes in the poloidal and
from Ampère’s law, ⵜ2␦A储 = −␮0␦jz. The power per unit
toroidal fluxes if ␦g has the form ␦g = 共␪ − ␫␸兲⌬共␺t兲. length in z that is needed to drive the magnetic pertur-
The stability of force-free equilibria can be studied bation is

with greater simplicity when the aspect ratio is very
large, Ro / r → ⬁. In a large-aspect-ratio torus with circu-
lar magnetic surfaces, one can let ␸ = z / Ro replace the
unit vector ␸ˆ by ẑ, and describe the magnetic field using
ជ = B 关ẑ + ␫共r兲共r / R 兲␪ˆ 兴,
共r , ␪ , z兲 cylindrical coordinates, B 0 o
where the rotational transform ␫ is assumed to be of
order unity. The toroidal flux is ␺t = ␲B0r2. The impor-
tant part of the perturbed vector potential is the parallel
ជ = ␦A储ẑ, so ␦B
ជ =ⵜ
ជ ␦A储 ⫻ ẑ. If the parallel
component, ␦A
component of the vector potential is written as ␦A储
= Ã储共r , t兲sin共n␾ − m␪兲, then
冋 册
␲a ⳵Ã储 ⳵Ã储
pd = − , 共77兲
␮0 ⳵t ⳵r
where the quantities are to be evaluated at r = a. To
prove this, note that the power per unit volume going
ជ · ␦ជj. The electric
into the electromagnetic fields is −E
field associated with the magnetic perturbation ␦B ជ =ⵜជ
ជ = −ẑ⳵␦A储 / ⳵t,
⫻ 共␦A储ẑ兲 is obtained using Faraday’s law, E
which implies Eq. 共77兲. In the absence of a plasma, Ã储
= A共t兲共r / a兲m for r ⬍ a and Ã储 = A共t兲共a / r兲m for r ⬎ a, which

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas 1095

requires a power pd = 共m␲ / ␮0兲dA2 / dt per unit length, Ro m

which is positive as one would expect. To make the ␰r = − ␦A储 . 共78兲
rB0 n − ␫m
power pd negative and obtain an instability,
共⳵2Ã储 / ⳵r2兲 / Ã储 must be negative in some part of the
ជ r. The re-
The radial component means ␰ជ · r̂, where r̂ ⬅ ⵜ
lation between ␰ and ␦A储 is proven using ␦B = B ជ ·ⵜ
ជ␰ ,
plasma. The condition that 共⳵2Ã储 / ⳵r2兲 / Ã储 be negative im- r r 0 r
plies the stability properties of the current-driven modes which follows from ␦B ជ =ⵜជ ⫻ 共␰ជ ⫻ Bជ 兲, Eq. 共46兲, and ␦B
0 r
given in the discussion of Eq. 共72兲 and that only pertur- ជ ␦A储, which follows from ␦B
= ␪ˆ · ⵜ ជ =ⵜជ ␦A储 ⫻ ẑ. The current
bations with low values of m can be unstable. flowing on a rational surface, r = rmn, Eq. 共76兲, can then
The restriction of current-driven modes to low mode be written in terms of the jump in the radial displace-
numbers comes from the stabilizing effect of the term ment,
m2Ã储 / rm in Eq. 共74兲, which tends to make 共⳵2Ã储 / ⳵r2兲 / Ã储
ẑ d␫ rB0
positive. Although current-driven instabilities must be ␦ជj = − ␦共r − rmn兲关␰r兴. 共79兲
low m, pressure-driven perturbations have comparable ␮0 dr Ro
stability properties at all m. The dominant drive for Much of the theory of instabilities in a cylinder, in-
pressure-driven instabilities is the Pfirsch-Schlüter cur- cluding the existence of jumps in the radial displace-
rent that arises in the perturbed plasma state. The desta- ment, was developed by Newcomb 共1960兲. Alan Glasser
bilization of short-wavelength perturbations by the pres- in unpublished work during the 1990s extended New-
sure gradient is discussed in Sec. V.C.2 on ballooning comb’s results to general axisymmetric equilibria. Glass-
modes. er’s work can be easily generalized so it applies to
In a tokamak the vacuum transform ␫v vanishes, and a plasma equilibria with arbitrary scalar pressure
very simple solution to Eqs. 共74兲 and 共75兲 exists for any 共Nührenberg and Boozer, 2003兲.
k0共r兲 when m = 1. This solution is ␦A储 = 关␫共r兲
− n兴rA sin共n␾ − ␪兲 where A is a constant. Define rn by 2. Resistive stability
␫共rn兲 = n. In an ideal stability analysis, the general m = 1
• The stability of a plasma with nonzero resisitivity can
solution for r ⬍ rn can be matched to an external solution be tested by considering the perturbation driven by a
Ã储 = 0 for r ⬎ rn. The matching gives a strong surface cur- delta-function current on a rational surface. If this
rent, ␮0j̃z = n␦共r − rn兲共d ln ␫ / d ln r兲rnA. This m = 1 pertur- current increases when power is removed, 兰jជ · E ជ d 3x
bation is called a sawtooth instability, which in a simple ⬎ 0, then the plasma is unstable to the formation of
analysis has zero growth rate without plasma dissipation an island with a width that increases linearly in time.
and an infinite growth rate with dissipation. The rota- Such instabilities are called tearing modes.
tional transform in the center of a tokamak whose cur- • The stability of a plasma to perturbations that are
rent is maintained by a loop voltage tends to rise until it sufficiently slow for all current perturbations to relax
exceeds unity. After the central transform exceeds unity, can be tested by making the distribution of the force-
strong m = 1, n = 1 kinklike relaxation oscillations occur free current a time-independent function of the tor-
that maintain the central transform just above, but close oidal magnetic flux ␺t.
to, unity. These relaxation oscillations are called saw-
teeth due their appearance on electron temperature di- In the presence of a perturbation, a resonant surface
agnostics where they were first observed. Reviews on current naturally arises on each resonant surface, which
the m = 1 mode have been written by Migliuolo 共1993兲 means a current density that is proportional to ␦共␫
and Hastie 共1997兲. − n / m兲cos共n␸ − m␪兲. The relaxation of this singular cur-
The location of the resonant rational surface, ␫共rmn兲 rent density due to plasma resistivity leads to an island
and potentially to a resistive instability. The paper that
= n / m, is a special radial position in the differential
sparked research on resisitive instabilities is that of
equation for Ã储 关Eq. 共74兲兴. In ideal theory an island must Furth, Killeen, and Rosenbluth 共1963兲.
be avoided, which means ␦B ជ · r̂ = −共m / r兲Ã储 cos共n␸ − m␪兲 Magnetic islands are produced by the resonant part of
must vanish there. The surface current that arises on the the perturbed parallel component of the vector poten-
resonant surface in an ideal analysis to prevent an island tial. That is,
from opening is given by Eq. 共76兲 with a replaced by rmn. ជ · ⵜជ ␺
␦B t ⳵␦A储/B ⳵␦A储/B
The relaxation of this current dissipates energy and al- = ␮0G共␺t兲 − ␮0I共␺t兲 , 共80兲
lows an island to open. ជ · ⵜជ ␸
B ⳵␪ ⳵␸
A jump in the resonant component of a radial dis-
with 共␺t , ␪ , ␸兲 magnetic coordinates in which B ជ has a
placement, ␰ជ · ⵜ
ជ ␺ , is the signature of a singular surface
t
simple covariant representation 关Eq. 共58兲兴. A resonant
current on a rational surface ␫ = n / m in an ideal pertur-
Fourier term in ␦B ជ ·ⵜ
ជ ␺ /B
ជ ·ⵜ
ជ ␸, Eq. 共14兲, is related to the
bation analysis. The demonstration is easiest in a large- t
aspect-ratio torus. The perturbed parallel component of half-width of an island by Eq. 共16兲. The derivation of Eq.
the vector potential ␦A储 can be replaced by the radial 共80兲 is straightforward, using the representation for the
component of the displacement, perturbed vector potential ␦A ជ = 共␦A / B兲B ជ + ␦A ⵜជ␺
B ␺ t

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1096 Allen H. Boozer: Physics of magnetically confined plasmas
+ⵜជ ␦g with the gauge ␦g chosen so ␦A储 = ␦A and using
B i
Eq. 共58兲 for B ជ in ␦A
ជ 储 = 共␦A储 / B兲Bជ . One can easily show
that a change in the gauge does not affect the resonant
components of Eq. 共80兲.
The rate of growth of an island can be calculated us-
ing a Rutherford 共1973兲 analysis. The component of
Ohm’s law along the resonant magnetic-field line, ␫
= n / m, is E储 = −⳵␦A储 / ⳵t = ␩␦j储. As the island grows the
force-free current near the rational surface flows on the
magnetic surfaces of the island, so the width of the cur-
rent channel ␦R is proportional to the width of the mag-
netic island. For a slowly growing island, one can find
the correct current distribution by solving the induction
equation 共Boozer, 1984a兲. Ampère’s law, ␮0␦j储 = −ⵜ2␦A储,
implies that the width ␦R of a narrow current channel
can be defined by ␦j储 = −2⌬⬘␦A储 / 共␮0␦R兲, with
冋 册
1 ⳵Ã储
⌬⬘ ⬅ . 共81兲
Ã储 ⳵r
The width of the island and the current channel are pro-
portional to 冑兩Ã 兩, so ␦R2 ⬀ 兩Ã 兩, and
储 储
d␦R ␩
= ⌬⬘ .
dt ␮0
If ⌬⬘ ⬎ 0, the width of the island grows until the current
singularity disappears, that is, ⌬⬘ → 0.
In the limit of a resistive analysis with no singular
currents, both Ã储 and ⳵Ã储 / ⳵r must be continuous. The
singularity of the right-hand side of the differential
equation for Ã储, Eq. 共74兲, is sufficiently weak that one
can integrate straight through, even numerically, without
real difficulty, which makes both Ã储 and ⳵Ã储 / ⳵r continu-
ous. Said more precisely, the solution to f⬙ = fx / 共x2 + ␦2兲 is
well behaved in the limit as ␦ → 0. Because the relax-
ation of the singular currents that arise in an ideal analy-
sis always takes energy from the field, a perturbation
always takes more energy to drive, and is therefore
more stable, in an ideal than in a resistive analysis.
3. Robustness of magnetic surfaces
• Plasma rotation prevents an arbitrarily small exter-
nal perturbation from opening a magnetic island in-
side a plasma.
• The pressure-driven net parallel current, which is
called the bootstrap current, can make an otherwise
stable plasma unstable to the opening of an island.
This instability is called the neoclassical tearing
mode.
An arbitrarily small external magnetic perturbation
cannot force an island to open in a rotating plasma 共Fitz-
patrick and Hender, 1991兲. To understand why, we must
first understand the equations for plasma rotation. The
ជ vជ
force balance equation of an ion species is minivជ · ⵜ
ជ p = qn 共E
ជ + vជ ⫻ B
ជ 兲. If the plasma is rotating slowly
+ⵜ i i
compared to the thermal speed, 冑T / mi, but rapidly com-
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
pared to the ion diamagnetic speed, 兩ⵜ ជ p / qn B兩, ion force
i
balance is approximated by Eជ + vជ ⫻ Bជ = 0 with the electric
ជ = −ⵜ
field given by a potential, E ជ ⌽. The magnitude of
ជ p / qnB兩 ⬇ 共␳ / a兲冑T / m . The
the diamagnetic speed is 兩ⵜ i i i
ion gyroradius is ␳i = 共mi / qB兲冑T / mi with ␳i / a ⬇ 1 / 500
under fusion conditions. The equation vជ ⫻ B ជ =ⵜ
ជ ⌽, plus
the requirement that a steady flow be divergence-free,
lead to two magnetic differential equations, B ជ ·ⵜ
ជ ⌽=0
ជ ·ⵜ
ជ 共v储 / B兲 = ⵜជ · 共ⵜ
ជ ⌽⫻B ជ / B2兲. These equations act as
and B
constraints on the two independent directions of plasma
rotation, poloidal and toroidal. Plasma rotation is deter-
mined by two functions that depend on only ␺t. These
functions are ⌽共␺t兲 and the solution to the homogeneous
ជ ·ⵜជ 共v储 / B兲 = 0. The equation B ជ ·ⵜ
ជ ⌽ = 0 holds
equation B
under very general conditions because of the ease with
which electrons flow along the magnetic-field lines. If
the magnetic surfaces are good and the rotation is slow
compared to the ion thermal speed, the electric potential
is accurately given by a function of the toroidal flux
alone, ⌽共␺t兲.
Even a small stationary island forces one component
of the plasma rotation to be zero; only rotation parallel
共82兲
to the resonant field lines is allowed. On the outermost
surface of an island, the separatrix, Fig. 2, the magnetic
ជ ·ⵜ
ជ ⌽ = 0 forces d⌽ / d␺ to be zero.
differential equation B t
This follows from the general result discussed at the end
of Sec. III.A that in the vicinity of an arbitrarily small
ជ ·ⵜ
ជ ⌽ = 0 can only
island any function ⌽ that satisfies B
have a nonzero derivative relative to the helical flux,
which is defined by d␺h ⬅ 共␫ − n / m兲d␺t. Without the is-
land ⌽ can be an arbitrary function of ␺t.
The zeroing of a component of rotation on a surface,
which is required if an island is to exist, requires a
torque. If an external magnetic perturbation is not
strong enough to transport this torque to the current
that produces the perturbation, ␦B ជ , then an island can-
x
not open. The z component of that torque, Tz, is associ-
ated with toroidal rotation. From the plasma perspec-
tive, any external magnetic perturbation in a tokamak
can be viewed as being created by a surface current that
flows in a thin axisymmetric surface that surrounds the
plasma. Torque balance between the plasma and the
perturbing external current allows an exact calculation
ជ 兲d3x, that
of the total electromagnetic torque, 兰xជ ⫻ 共jជ ⫻ B
the perturbation exerts within the plasma, by carrying
out an integral over the external current-carrying sur-
face. The z component of the torque is
冖
1 ជ 兴 · ␸ˆ ␦B
ជ · daជ .
Tz = R关␦B x i 共83兲
␮0
surf
ជ is the field produced by currents internal to the
Here ␦B
i
plasma. ␦Bជ is the external magnetic perturbation, and
x
ជ 兴 is its jump across the current-carrying layer that
关␦B
x
ជ . Equa-
represents the external current that produces ␦B x
 Allen H. Boozer: Physics of magnetically confined plasmas
tion 共83兲 is derived by noting that the surface current
that flows in the current-carrying layer is ␦Jជ = n̂
ជ 兴 / ␮ , where n̂ is the normal to the surface, and the
⫻ 关␦B x 0 vation of magnetic surfaces is an important design and
force per unit area in the torodial direction is 共␦Jជ
⫻ ␦Bជ 兲 · ␸ˆ . The torque that can be exerted by the current-
i
carrying layer has an upper limit. Since ␦B ជ for a stable
i
plasma is proportional to the external perturbation ␦B ជ
x Expert Group on Disruptions, …, 1999兲. Disruption
共Sec. V.D.2兲, the maximum torque scales as 共␦Bx兲 . If the
2
torque that is required to introduce an island is greater
than this limit, then the island cannot open. Equation
共83兲 also contains the essence of the phenomenon of the
torque on a plasma due to a magnetic-field error becom-
ing much larger as a plasma approaches marginal stabil-
ity. This is because the ratio ␦Bi / ␦Bx becomes large as
marginal stability is approached 共Sec. V.D.2兲.
How does the torque that an island induces prevent
the island from opening? For simplicity assume the
plasma pressure is zero; then the force balance is ជf
ជ , where ជf is the viscous force exerted by the
= ␦ជj ⫻ B
vis
plasma. The parallel current obeys B ជ ·ⵜ
ជ 共␦j储 / B兲 = −ⵜ
ជ · 共B
ជ force of interaction with a vertical field. Changes in the
⫻ ជfvis / B2兲. To prevent an external magnetic perturbation,
such as that of Eq. 共14兲, from opening an island, the
parallel current must have the form ␦j储 / B
= kr共␺t兲cos共n␸ − m␪m兲 so it can cancel the perturbing field
on the rational surface. Without the viscous force this
current can only flow on the rational surface ␫ = n / m,
which means kr共␺t兲 must be a delta function. With a vis-
cous force, 共n − ␫m兲kr共␺t兲 is given by the resonant Fou-
rier component of ⵜ ជ · 共B
ជ ⫻ ជf / B2兲. Since 共n − ␫m兲 van-
vis
ishes at the rational surface, even a small viscous force
can spread out and also maintain the current that is re-
quired to prevent an island from opening.
If a plasma is initially confined on perfect magnetic
surfaces, then it is a complicated question whether sur-
faces will split forming islands due to either instabilities
or small external perturbations. Rotational shielding of
islands, which we discussed earlier, is only one example
of several important effects. In the modern literature,
these effects are primarily discussed under the topic of
neoclassical tearing modes; see, for example, Rosenberg
et al. 共2002兲.
A neoclassical tearing mode causes a magnetic island
to open through the formation of a strong gradient in
the net plasma current near a rational surface. An island
removes the pressure gradient near its resonant rational
surface 共Sec. III.A兲, and zeros the bootstrap current,
which is a net parallel current due to the pressure gradi-
ent 关Eq. 共220兲兴. The zeroing of the bootstrap current in
the vicinity of the rational surface produces a large gra-
dient d共j储 / B兲 / d␺h, which is of the same sign on both
sides of the rational surface, where d␺h ⬅ 共␫ − n / m兲d␺t.
Depending on the relative signs of the bootstrap current
and d␫ / d␺t, this effect either strongly favors, or mitigates
against, the formation of an island 共Sec. V.B.1兲. In a to-
kamak with ␫ peaked on axis, the formation of an island
is favored.
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1097
Magnetic surface quality does seem to have some hys-
s
teresis: good surfaces tend to stay good and surfaces
split by islands tend to stay split by islands. The preser-
s operational consideration in both tokamaks and stellara-
tors.
In tokamaks instability can lead to a catastrophic loss
of surfaces, which is called a disruption 共ITER Physics
avoidance is critical to the success of magnetic fusion.
Stellarators are far more immune to disruptions. There
are two reasons for this. First, the free energy in a stel-
larator equilibrium tends to be smaller because of the
contribution of the vacuum magnetic field to the rota-
tional transform. Second, the vacuum field in a stellar-
ator strongly centers the plasma in its surrounding
chamber. In a tokamak the plasma is centered in its
chamber by a vertical, z-directed, magnetic field. The
toroidal current that provides the poloidal field in a to-
vis kamak produces a self-force in the R̂ direction, which is
called the hoop stress. This must be balanced by the
plasma equilibrium cause a change in the plasma posi-
tion, which may result in the plasma’s striking the cham-
ber walls, with a rapid loss of the plasma energy and
current. Such an event in a power plant could cause se-
vere damage to the walls.
C. Stability analyses
• Instability of an equilibrium can be demonstrated by
finding a plasma displacement ␰ជ 共xជ 兲 that reduces the
energy W of the magnetic field and the plasma 关Eq.
共42兲兴.
• The two components of the plasma displacement
that lie in a magnetic surface can be expressed in
ជ ␺ , for dis-
terms of the radial displacement, ␰␺ ⬅ ␰ជ · ⵜ t
placements that minimize the energy W.
• The change in the energy ␦W produced by a plasma
displacement is a quadratic operator on ␰␺ for per-
turbations that minimize the energy W.
The theory of long-wavelength plasma stability is
closely related to the theory of equilibria. Equilibria can
be found by minimizing the energy, Eq. 共43兲, while im-
posing either torodial symmetry for tokamaks or period-
icity for stellarators, xជ 共␺t , ␪ , ␸ + 2␲ / Np兲 = xជ 共␺t , ␪ , ␸兲,
where Np is the number of periods. These equilibria are
unstable if the energy can be lowered by a more general
shape function xជ 共␺t , ␪ , ␸兲. Mathematically, equilibria are
extrema and stable equilibria are minima of the energy
共Lundquist, 1951; Bernstein et al., 1958兲. The long-
wavelength instabilities found by minimizing the energy
are called magnetohydrodynamic 共MHD兲 instabilities as
distinct from the microinstabilities that have a wave-
length comparable to an ion gyroradius.
1098 Allen H. Boozer: Physics of magnetically confined plasmas

1. Expressions for ␦W plasma. One finds that 兰共␦ជj ⫻ Bជ 兲 · ␰ជ d3x = −兰共␦B兲2d3x / ␮ ,

0
If xជ e共␺t , ␪ , ␸兲 gives the shape of the magnetic surfaces where the last volume integral is over all of space. Using
in an equilibrium plasma, the stability of that equilib- Eq. 共47兲,
rium can be tested by considering the shape function
␰ជ · ⵜជ ␦p = − ␰ជ · ⵜជ 共␰ជ · ⵜជ p兲 − ⵜជ · 共␥p␰ជ ⵜជ · ␰ជ 兲 + ␥p共ⵜជ · ␰ជ 兲2 .
xជ 共␺t , ␪ , ␸兲 = xជ e共␺t , ␪ , ␸兲 + ␰ជ with ␰ជ a small perturbing dis-
placement. The displacement causes a change in the 共89兲
magnetic field, Eq. 共46兲, and a change in the pressure, Using Eq. 共46兲 one can easily show ␦B ជ ·ⵜ
ជ p=B
ជ ·ⵜ
ជ 共␰ជ · ⵜ
ជ p兲,
Eq. 共47兲. The change in the energy, Eq. 共42兲, due to a
small change from equilibrium, Eq. 共43兲, is quadratic in
ជ 兲 = −␰ជ 储 · ⵜ
which implies ␰ជ 储 · 共jជ ⫻ ␦B ជ 共␰ជ · ⵜ
ជ p兲. The integral
the displacement and given by over the plasma volume of a pure divergence vanishes
because the plasma pressure and pressure gradient van-

␦W =
1
2
冕plasma
ជ − ជj ⫻ ␦B
共ⵜជ ␦p − ␦ជj ⫻ B ជ 兲 · ␰ជ d3x, 共84兲
ish at the plasma edge.
Equation 共88兲 for w␰⬜ can be rewritten 共Furth et al.,
1965兲 as
with ␦ជj ⬅ ⵜ ជ ⫻ ␦B ជ / ␮ . The integral is over the volume oc- w ␰⬜ = w + + w ± , 共90兲
0
cupied by the plasma, for outside the plasma ␦p, ␦ជj, and where
ជj are zero. The factor of 2 arises from the force operator
共 ␦ B ⬜兲 2 B 2 ជ ជ
Fជ 关␰ជ 兴 ⬅ ⵜ
ជ ␦p − ␦ជj ⫻ Bជ − ជj ⫻ ␦B
ជ being linearly dependent on w+ ⬅ + 共ⵜ · ␰⬜ + 2␰ជ ⬜ · ␬ជ 兲2 共91兲
2␮0 2␮0
the displacement ␰ជ . Partial integrations demonstrate that
the force operator has the important property of being is always positive but the term
self-adjoint, 兰␰ជ 1 · Fជ 关␰2兴d3x = 兰␰ជ 2 · Fជ 关␰1兴d3x, where ␰ជ 1 and ␰ជ 2 1 j储 ជ ជ 兲 · ␦B
ជ − 共␰ជ · ␬ជ 兲共␰ជ · ⵜជ p兲
are two arbitrary displacements 共Bernstein et al., 1958; w± ⬅ − 共␰⬜ ⫻ B ⬜ ⬜ ⬜ 共92兲
2B
Bernstein, 1983兲. This implies that if the displacement is
has an indefinite sign. The curvature of the magnetic-
expanded in a set of vectors, ␰ជ = 兺cjuជ j, then ␦W
field lines is
= 兺c*i Wijcj where Wij is a Hermitian matrix.
As shown below, Eq. 共84兲 for ␦W can be written as the ␬ជ ⬅ b̂ · ⵜជ b̂, 共93兲
sum of a vacuum energy and a plasma energy 共Bernstein
et al., 1958兲, where b̂ ⬅ B ជ / B.
The derivation of Eqs. 共90兲–共92兲, which is given in this
␦W = ␦Wv + ␦Wp . 共85兲 paragraph and can be skipped, follows from an expres-
sion for the parallel component of the perturbed mag-
The vacuum energy is
netic field,

Wv ⬅ 冕 共␦B兲2 3
ext 2␮0
d x, 共86兲
ជ · ␦B
B ជ = − 共ⵜជ · ␰ជ + 2␰ជ · ␬ជ 兲B2 + ␮ ␰ជ · ⵜជ p.
⬜ ⬜ 0 ⬜ 共94兲
The derivation of this identity uses the vector identity
with the integral performed over the region exterior to B ជ ⫻ 共␰ជ ⫻ B
ជ ·ⵜ ជ 兲 = −␰ជ · 兵共ⵜ
ជ ⫻B
ជ 兲⫻B ជ · 共B2␰ជ 兲 plus Am-
ជ 其−ⵜ
⬜ ⬜ ⬜
the plasma. The plasma contribution is ជ ⫻B
père’s law, ⵜ ជ = ␮ ជj, and force balance ⵜ
ជ p = ជj ⫻ B
ជ.
0

冕 冉 冊 Force balance plus Ampère’s law imply ⵜ ជ 共B2 + 2␮ p兲

1 ⬜
w␰⬜ + ␥p共ⵜជ · ␰ជ 兲2 d3x,
0
␦Wp = 共87兲 = 2B2␬ជ . Combining results one obtains Eq. 共94兲. The
plasma 2
identity of Eq. 共94兲 can be written in a second form,
where which is

ជ 兲2 1
ជ 储兲2
共␦B B2 ␰ជ ⬜ · ⵜជ p 1 ជ ជ
共␦B 1ជ = ⌬2 + ␦B储 − 共␰⬜ · ⵜ p兲⌬, 共95兲
w ␰⬜ ⬅ + 共ⵜជ · ␰ជ ⬜兲共␰ជ ⬜ · ⵜជ p兲 − ជ兲
␰⬜ · 共jជ ⫻ ␦B 2␮0 2␮0 2B 2
2␮0 2 2
共88兲 where

has no dependence on the parallel part of the displace- ⌬ ⬅ ⵜជ · ␰ជ ⬜ + 2␰ជ ⬜ · ␬ជ . 共96兲

ment, ␰ជ 储 ⬅ 共b̂ · ␰ជ 兲b̂, where b̂ ⬅ B
ជ / B. Equation 共87兲 implies The last term in Eq. 共88兲 for w␰⬜ can be rewritten using
that to minimize ␦W one should choose the parallel the fact that the vector triple product in three dimen-
component of the displacement so that ⵜ ជ · ␰ជ = 0. sions is nonzero only when one of the vectors comes
The derivation of Eqs. 共85兲–共88兲 is given in this para- from each of the three independent directions. Conse-
graph, which can be skipped. The derivation follows ⬜
ជ 兲 = ␰ជ · 共jជ储 ⫻ ␦B
quently ␰ជ · 共jជ ⫻ ␦B ⬜
ជ 兲 + ␰ជ · 共jជ ⫻ ␦B
⬜ ⬜
ជ 储兲, which
⬜
ជ 兲 · ␰ជ in ␦W can be inte-
from noting that the term 共␦ជj ⫻ B ជ 兲 = 共j储 / B兲␦B
can be rewritten as ␰ជ ⬜ · 共jជ ⫻ ␦B ជ · 共␰ជ ⫻ B
⬜ ⬜
ជ兲
grated over all of space since ␦ជj = 0 outside of the + 共␦B储 / B兲␰ជ · ⵜ
⬜
ជ p.

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas 1099

A minimization of ␦W is really a minimization of a 兰C2d3x through the variation ␦␩ gives 2 兰 兵C ជ ·ⵜ

ជ
ជ ␺ that involves radial
quadratic functional of ␰␺ ⬅ ␰ជ · ⵜ t ⫻ 共␦␩ⵜជ ␺ 兲 / 2␲其d3x. An integration by parts then gives
t
derivatives only up to the first order. If the perturbing the desired result.
displacement is written as
⳵xជ ⳵xជ ជ,
␰ជ = ␰␺ + ␩ + ␮B 共97兲 2. Pressure-driven ballooning modes
⳵␺ ⳵␪
• Perturbations of arbitrarily large wave number can
then the only place the coefficient ␮共␺ , ␪ , ␸兲 enters ␦W is be destabilized by a pressure gradient that is in the
in the positive definite term involving 共ⵜ ជ · ␰ជ 兲2. A minimi- same direction as the field-line curvature. These in-
ជ · ␰ជ = 0, which is a magnetic differ- stabilities are called ballooning modes since they bal-
zation of ␦W makes ⵜ
loon out, that is, they have their largest amplitude at
ential equation for ␮ in terms of the ␩ and ␰␺ coeffi- the least stable place on a constant-pressure surface.
cients.
The ␩ coefficient of ␰ជ can be expressed in terms of ␰␺
• Plasma instabilities that have a high wave number
have a short wavelength in the constant-pressure sur-
for perturbations that minimize ␦W. A perturbation that
faces across the magnetic-field lines but a long wave-
minimizes ␦W must give a perturbed equilibrium, and in
length along the magnetic-field lines. For ballooning
ជ ⫻ ␦B
a perturbed equilbrium the current, ជj + ⵜ ជ / ␮ , must modes this anisotropy arises to minimize the stabiliz-
0
be orthogonal to the toroidal flux. At a given spatial ing effect of field-line bending. For microinstabilities
point the perturbation causes the toroidal flux to go 共Sec. VI.F兲, the anisotropy arises to avoid stabiliza-
from ␺ to ␺ − ␰ជ · ⵜ
t t
ជ ␺ , which follows from d␺ / dt
⬜ t t
tion by Landau damping 共Sec. VI.C兲.
= ⳵␺t / ⳵t + 共⳵␰ជ / ⳵t兲 · ⵜ ជ ␺ and the constraint that d␺ / dt = 0 in
t t The expression for ␦W is greatly simplified when the
an ideal plasma. See the discussion of Eq. 共45兲. There- perturbations that are considered have a short wave-
fore ␦W minimizing perturbations obey the constraint length across the magnetic-field lines. Such instabilities
␦ជj · ⵜជ ␺t = ជj · ⵜជ 共␰ជ ⬜ · ⵜជ ␺t兲, where ␦ជj ⬅ ⵜជ ⫻ ␦B ជ / ␮ . Now are called ballooning modes 共Todd et al., 1977; Connor,
0
ជ 共␰ជ · ⵜ ជ ␺ 兲=ⵜ ជ · 关共␰ជ · ⵜជ ␺ 兲jជ兴. The normal to the surface is Hastie, and Taylor, 1978兲. Despite their short wave-
ជj · ⵜ ⬜ ⬜
t t length across the magnetic-field lines, the least stable
n̂ ⬅ ⵜ ជ ␺ / 兩ⵜ ជ ␺ 兩, so one can write ជj = n̂ ⫻ 共jជ ⫻ n̂兲 and perturbations of this type have a long wavelength along
t t
ជ ជ ជ
共␰⬜ · ⵜ␺t兲j = ⵜ␺t ⫻ 关共jជ ⫻ n̂兲␰ជ ⬜ · n̂兴. This implies ជj · ⵜ ជ 共␰ជ · ⵜ
⬜
ជ␺ 兲
t
the magnetic field. These disparate scales can be effi-
ជ ␺ ·ⵜ ជ ⫻ 关共jជ ⫻ n̂兲␰ជ · n̂兴. Letting ciently represented using a concept from geometric op-
= −ⵜ ⬜
t ជ ·ⵜ
tics, the eikonal S共␣兲, which satisfies B ជ S = 0. Let
ជ ⬅ ␦B
C ជ + 共␰ជ · n̂兲共␮ ជj ⫻ n̂兲 共98兲
⬜ 0
ជ cos„S共␣兲…,
␰ជ ⬜ = ⌶ 共101兲
one finds that
ជ 兲 · ⵜជ ␺ = ⵜជ · 共C
ជ ⫻ ⵜជ ␺ 兲 = 0 where ␣ is a Clebsch coordinate 关Eq. 共9兲兴. For
共ⵜជ ⫻ C t t 共99兲 ballooning-mode calculations it is conventional to
is the condition for the current in the perturbed equilib- ជ =ⵜ
choose the Clebsch coordinates so B ជ ␣⫻ⵜ
ជ ␺ , where
p
rium to lie in the perturbed flux surfaces. This constraint ␺p is the poloidal magnetic flux and
can be imposed ␺t surface by ␺t surface, since the only
derivatives that arise are in ␪ and ␸. The two constraints 2␲␣ = ␸ − q共␺p兲兵␪m − ␪0共␺p兲其, 共102兲
ⵜជ · ␰ជ = 0 and Eq. 共99兲 determine both ␮共␺ , ␪ , ␸兲 and
t
␩共␺t , ␪ , ␸兲 in terms of ␰␺共␺t , ␪ , ␸兲. with q共␺p兲 ⬅ 1 / ␫共␺p兲 ⬅ d␺t / d␺p the safety factor. The ei-
ជ ⫻C ជ 兲·ⵜ ជ ␺ = 0 must vanish for a konal S could be a function of both ␣ and ␺p, but this
The constraint that 共ⵜ t provides no generality beyond that in the function
␦W minimizing perturbation can be obtained directly ␪0共␺p兲. The eikonal can be written as
from the energy principle if w␰⬜ is written as
ជ2 S = 2␲N␣ , 共103兲
C
w ␰⬜ = − wd , 共100兲
2␮0 where N is the toroidal mode number of the perturba-
tion. The short-wavelength limit means N → ⬁. The ex-
ជ · ⵜ 兲n̂ 共Bernstein et al., 1958;
where wd ⬅ 共␰ជ · n̂兲2共jជ ⫻ n̂兲 · 共B pression that one obtains is

冕再 冉 冎
Bernstein, 1983兲. For a divergence-free perturbation the
only place the coefficient ␩ enters the energy is through
ជ , and there it enters only through ␦B
C ជ . The contravari-
ant representation of the magnetic field in magnetic co-
␦W
共2␲N兲2
=
2␮0
ជ · ⵜជ F
共ⵜជ ␣兲2 B
B
冊 2

− ␬˜
dp 2 3
d␺p
F d x,

ordinates, Eq. 共52兲, plus the dual relations imply 共104兲

ជ =ⵜ
共⳵xជ / ⳵␪兲 ⫻ B ជ ␺ / 2␲. Consequently the part of C ជ that
t where the relevant component of the field-line curvature
depends on ␩ is C ជ =ⵜ ជ ⫻ 共␩ⵜ ជ ␺ 兲 / 2␲. The minimization of is
␩ t

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1100 Allen H. Boozer: Physics of magnetically confined plasmas

ជ ⫻ ⵜជ ␣ If the ballooning expression for ␦W is extremized, one

␬ជ · B
␬˜ ⬅ 共105兲

再 冎
. obtains the differential equation
B2
ជ · ⵜជ ␪ ⳵F
⳵ 共ⵜជ ␣兲2 B 2␬˜ dp
The perpendicular displacement of an unstable pertur- + F = 0, 共109兲
bation must have the form of Eq. 共108兲, which defines ⳵␪ ␮0 B2 ⳵␪ ជ · ⵜជ ␪ d␺p
B
the function F.
where B ជ ·ⵜ
ជ = 共B
ជ ·ⵜ
ជ ␪兲⳵ / ⳵␪ and the Jacobian of 共␺ , ␪ , ␣兲
The derivation of the short-wavelength form for ␦W is p
given in this paragraph and can be skipped for those not coordinates is 1 / Bជ ·ⵜ
ជ ␪ for any poloidal angle ␪. Equa-
interested in the mathematical details. Using Eq. 共101兲 tion 共109兲 predicts instability if a solution F crosses F
for the perpendicular displacement, derivatives of the = 0 twice. To understand why this prescription works,
displacement enter ␦W, Eqs. 共90兲–共92兲, through note that if one of the F’s in the F2 term of Eq. 共104兲 is
replaced by the F from Eq. 共109兲, one finds after an
ⵜជ · ␰ជ ⬜ = cos共S兲ⵜជ · ⌶
ជ − sin共S兲⌶
ជ · ⵜជ S 共106兲 integration by parts that ␦W = 0. Now add a term
and 兰⌬sF2d␺pd␪d␣ / 2 to both sides of the ballooning expres-
sion for ␦W / 共2␲N兲2, Eq. 共104兲, with ⌬s a constant. Tak-
ជ 兲其 .
␦B⬜ = cos共S兲兵ⵜជ ⫻ 共⌶ ⫻ B 共107兲 ing ␦Wm / 共2␲N兲2 ⬅ ␦W / 共2␲N兲2 + 兰⌬sF2d␺pd␪d␣ / 2 to its
⬜
extreme, one finds that Eq. 共109兲 is modified by the zero
Note that ␦B = cos共S兲ⵜ ជ ⫻ 共⌶ ⫻ B ជ 兲 − sin共S兲兵ⵜជ S ⫻ 共␰ជ ⫻ B
ជ 兲其, on the right-hand side becoming ⌬sF. One solves this
⬜
ជ S ⫻ 共␰ជ ⫻ Bជ 兲其 = −共␰ជ · ⵜជ S兲B
ជ does not con- modified equation. If the solution has two places where
but the term 兵ⵜ ⬜ ⬜
F = 0 when ⌬s ⬎ 0, one replaces an F in the modified
tribute to the magnetic perturbation perpendicular to
equation, ␦Wm / 共2␲N兲2, with the F from the differential
the field lines. When these expressions are inserted into
equation. One then finds ␦Wm / 共2␲N兲2 = 0, so
Eqs. 共90兲–共92兲, the terms proportional to ⵜ ជ S and hence
␦W / 共2␲N兲2 = −兰⌬sF2d␺pd␪d␣ / 2, which is negative. That
N appear only in a positive definite term that is propor- is, one has found a perturbation that reduces the energy,
ជ · ␰ជ + 2␰ជ · ␬兲2. Consequently instability is pos-
tional to 共ⵜ so instability is predicted.
⬜
sible in the limit as N → ⬁ only if one chooses ␰ជ · ⵜ ជS
⬜
= 0. By choosing a 1 / N correction to ␰ជ ⬜ appropriately, 3. Minimization of ␦W
one can eliminate the positive definite term 共ⵜ ជ · ␰ជ
⬜ Given any positive definite normalization of the dis-
+ ␰ជ · ␬兲2 from ␦W altogether, since no other term is af- placement, 储␰2储, a plasma is unstable if one can find a
fected in the lowest nontrivial order by the 1 / N correc- displacement such that
tion. The perpendicular displacement for unstable short-
␦W
wavelength modes must have the form ␭⬅ 共110兲
储 ␰ 2储
ជ ⫻ ⵜជ S
B
␰ជ ⬜ = F 共108兲 is negative. The sign of ␭ is independent of the normal-
B2 ization. However, different normalizations provide dif-
ferent types of information. Three normalizations will
since it must be perpendicular to both the magnetic field be discussed, the kinetic-energy norm, the ␰␺ norm, and
and ⵜ ជ S. One then finds ␰ជ ⫻ B ជ = Fⵜ ជ S and ␦B ជ =ⵜជ F⫻ⵜ ជ S. the surface norm.
⬜
The parallel current term in Eq. 共92兲 vanishes because The conventional normalization of the displacement is
共␰ជ ⬜ ⫻ Bជ 兲 · ␦B
ជ = Fⵜ
⬜
ជ S · 共ⵜ
ជ F⫻ⵜជ S兲 = 0. As discussed earlier, the kinetic-energy norm, 储␰2储k ⬅ 兰␳␰2d3x / 2, which makes
the parallel current cannot be a source of instability for 冑共−␭兲 the growth rate of the instability. This norm is
short-wavelength perturbations. The perpendicular part poorly behaved when the plasma is stable because then
of the magnetic perturbation is ␦B ជ = 共b̂ · ⵜ
ជ F兲共b̂ ⫻ ⵜ
ជ S兲 a minimization of ␭ can mean a maximization of 储␰2储k.
⬜
ជ / B. Putting the pieces together one obtains As we have seen, the coefficient ␮ of the displacement,
with b̂ ⬅ B Eq. 共97兲, is determined by a magnetic differential equa-
Eq. 共104兲.
tion, which means the Fourier component of ␮ that reso-
The form for ␦W for short-wavelength perturbations nates with a rational surface can be made arbitrarily
implies that unstable modes must have a long wave-
large in that neighborhood with little change in ␦W.
length along the magnetic-field lines in order to mini-
That is, one can always find a continuum of modes that
mize the 共B ជ ·ⵜជ F兲2 term. If one can choose ␪ in ␣ so makes ␭ → 0 for an essentially fixed ␦W. The existence
0
␬˜ dp / d␺t is positive, then ␦W predicts instability if the of this continuum 共Grad, 1973; Goedbloed, 1998兲 makes
variation of F along the field lines is not taken into ac- it difficult to find the points of marginal stability. Even in
count. The 共B ជ ·ⵜ
ជ F兲2 term can stabilize the mode in the unstable region, the displacement that minimizes ␦W
places where ␬˜ dp / d␺t is positive, but the mode ampli- is usually not the perturbation that minimizes ␭. Conse-
tude is naturally larger there. That is, the plasma dis- quently a minimization of ␭ using displacements that are
placement balloons out at the locations where ␬˜ dp / d␺t divergence-free and satisfy 共ⵜ ជ ⫻C
ជ 兲·ⵜ
ជ ␺ = 0 does not give
t
is positive. the correct growth rate for unstable modes.

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas
The singularity of the displacement ␰ជ near rational
surfaces is particularly important for the theory of per-
turbations that are rotating relative to the plasma. A
singular plasma displacement rotating through a plasma
gives an infinite correction to the energy. This singularity
can be resolved 共Betti and Freidberg, 1995兲, but at the
cost of an imaginary contribution to the energy, which
represents the torque between the perturbation and the
plasma. Though not noted by Betti and Freidberg
共1995兲, the resolution of the singularity also changes the
real part of the energy 共Boozer, 2003兲, which implies that
the critical value of the plasma parameters required for
the stability of the perturbation are very sensitive to ro-
tation.
A second choice for the normalization of the displace-
ment is similar to the kinetic energy except only the
component ␰␺ is retained, 储␰2储␺ ⬅ 兰␳共␰␺兲2d3x / 2. The
quantity ␭ has no special interpretation, but the con-
tinuum problem in finding points of marginal stability is
eliminated.
A third choice of normalization uses ␰␺ but only its
value on the plasma surface, 储␰2储s = 养 共␰␺兲2wd␪d␸, with
w共␪ , ␸兲 ⬎ 0 an arbitrary weight function. This normaliza-
tion has pathological features for internal plasma modes,
which have a displacement that reduces the energy while
keeping the plasma boundary fixed. For these modes ␭
→ −⬁. However, if the minimizing ␭ is bounded, the dis-
placements ␰␺ of the plasma surface that are associated
with a spectrum of ␭’s give the modified equilibria of the
plasma in the presence of external perturbations.
An instability with a negative but bounded ␭ with a
fixed perturbation of the surface shape, the 储␰2储s norm, is
called an external mode. External modes would be sta-
bilized if a perfect conductor were close enough to the
plasma surface. A magnetic perturbation cannot pen-
etrate a perfect conductor, so the component of the
magnetic-field line displacement ␰ជ that is normal to the
conductor must vanish. Actual plasmas are surrounded
by chamber walls, which are conductors. An instability
that would be stabilized if these conducting structures
were perfectly conducting is called a wall mode. Wall
modes can grow on the resistive time scale of the con-
ducting structures, which is many orders of magnitude
slower than the growth rate determined by plasma iner-
tia. Due to their slow growth, resistive wall modes can
be stabilized by plasma rotation and by feedback 共Sec.
V.D.2兲. The stabilization of wall modes is considered an
important topic for tokamak and ST plasmas that have
the profile of net current that is required for steady-state
operation.
When one uses a fixed perturbation of the boundary,
the 储␰2储s norm, the Fourier coefficient of ␰␺ in magnetic
coordinates can jump at each resonant rational surface,
␫ = n / m, and the magnitude of the jump gives the magni-
tude of the surface current on the rational surface. Once
the expansion coefficients ␩ and ␮ of the displacement,
Eq. 共97兲, have been eliminated from ␦W in favor of ␰␺,
radial derivatives of the displacement arise only in the
form ⳵b / ⳵␺t where
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1101
ជ · ⵜជ ␺
␦B t
b⬅ . 共111兲
ជ ជ
B · ⵜ␸
ជ =ⵜ ជ ⫻ 共␰ជ ⫻ B
ជ 兲, implies that b = 共⳵ / ⳵␸
Equation 共46兲, ␦B 0
␺
+ ␫⳵ / ⳵␪m兲␰ . If one allows the resonant Fourier coeffi-
␺
cient of ␰mn to change by a fixed amount over a distance
␦␺t about a rational surface ␫ = n / m, the resonant Fou-
rier coefficient of ⳵b / ⳵␺t has a well-defined limit on each
side of the rational surface as ␦␺t → 0. One can show that
the jump in the resonant coefficient of ␰␺ or in ⳵b / ⳵␺t is
proportional to the surface current on the rational sur-
face 共Nührenberg and Boozer, 2003兲. For a large-aspect-
ratio torus, this was shown in the derivation of Eq. 共79兲.
The amplitude of the jumps in ␰␺ at rational surfaces
gives a measure of the width of the islands that would
open in the absence of a singular current. For a large-
aspect-ratio torus with circular surfaces, ␺t ⬀ r2 with r the
minor radius, one can easily show that the half width of
the island that would arise in the absence of the singular
current is
共␦r兲mn = 冑冏 2rmn
m
冏
关共␰r兲mn兴 . 共112兲
One can use an ideal ␦W method to assess and improve
the quality of equilibria and to find the required pertur-
bations to the plasma surface to remove islands
共Nührenberg and Boozer, 2003兲.
D. Interaction of plasmas with coils
• Between the plasma and the surrounding coils, the
magnetic field has the form B ជ =ⵜជ ␾. Since ⵜ
ជ ·B
ជ = 0,
the interaction of a plasma with currents outside the
plasma must be through solutions to Laplace’s equa-
tion, ⵜ2␾ = 0, which greatly constrains the form of the
interaction.
• Practical coils can control only a certain number Nf
of features of the plasma shape. For axisymmetric
systems Nf ⬇ 4, but for stellarators Nf ⬇ 50.
• Important long-wavelength instabilities of tokamaks
would be stabilized if the surrounding chamber walls
were perfectly conducting. Such instabilities are
called resistive wall modes since the instabilities can
grow on a time scale determined by the resistivity of
the chamber walls. Their slow growth permits feed-
back stabilization. The interaction of the plasma with
the perturbing currents in walls and in feedback coils
can be calculated by coupled circuit equations. The
plasma circuit elements have a determinate response
to changes in the other circuit elements.
The external coils that are required for the magnetic
confinement of a plasma have two distinct functions.
First, they provide the required toroidal magnetic flux
␺t. Second, they ensure B ជ · n̂ is zero on the plasma sur-
face. These two functions are separated in tokamak coil
design: the toroidal field coils provide the flux and the
1102 Allen H. Boozer: Physics of magnetically confined plasmas
poloidal field coils ensure B ជ · n̂ = 0 共Fig. 4兲. However, in
stellarators these functions are often coupled in a single
coil set 共Fig. 6兲. Nevertheless, in discussions of the effi-
ciency and practicality of stellarator coils, it is often use-
ful to separate the two functions.
Critical issues in coil design are the ratio of the mag-
netic field at the coils to that on the plasma, the com-
plexity and the forces associated with the required coil
currents, and the production of a field with the required
symmetry while providing for ports, for plasma access,
and for discrete coils.
A point that is often emphasized in the magnetic fu-
sion literature is that, if all else is equal, the power pro-
duced by a fusion plant scales as the square of ␤
⬅ 2␮0p / B2. A point less frequently made is that, under
the same assumptions as those that give the ␤2 scaling,
the power production scales as the fourth power of the
ratio of the magnetic field on the plasma to that on the
coils. If the plasma ␤ is limited, then the optimal plasma
temperature is the one that maximizes the fusion power
production at a fixed plasma pressure p = nT. The power
from the fusion of deuterium and tritium scales as pDT
= n2fDT共T兲, with n the plasma number density and fDT a
function of the plasma temperature. The maximum of
pDT at fixed pressure occurs at d共fDT / T2兲 / dT = 0. This
condition implies that near the optimal temperature,
which is about 20 keV, the fusion power density pDT is
proportional to the pressure squared, or equivalently, to
␤ 2.
Between the coils and the plasma, the magnetic field
is curl-free,
ជ = ⵜជ ␾ ,
B 共113兲
as well as divergence-free, so
ⵜ2␾ = 0. 共114兲
The interactions of the coils with the plasma must occur
through solutions to Laplace’s equation, which places
strong constraints on the form of this interaction. Much
of the theory of coil design, such as the work of Merkel
共1987, 1988兲 on the coil design for the W7-X stellarator,
is an application of the theory of Laplace’s equation.
1. Freedom of coil design
The choice of plasma shape is the major determinant
of the quality of a confined plasma. Unfortunately, the
theory of Laplacians says that a general plasma shape
cannot be supported by distant coils. What can be done,
through the design of coils, is to enforce a certain num-
ber of conditions on the plasma shape. This number,
which is the number of degrees of freedom in the coil
design Nf, is in practice about four for tokamaks and
about 50 for stellarators.
The theory of Laplace’s equation implies that it is
mathematically impossible for a distant coil set to pro-
ជ · n̂ on the plasma
vide a generic normal magnetic field B
surface, even when B ជ · n̂ is a smooth analytic function.
This statement is proven by giving a generic normal field
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
that cannot be produced by distant coils. Consider the
cylindrical problem in which the radial magnetic field
that must be produced by coils, Br共␪兲 = 兺Bm cos共m␪兲, is
given on the surface r = a. Between the coils and the
plasma the magnetic field obeys Eqs. 共113兲 and 共114兲, so
␾ = 兺共aBm / m兲共r / a兲m cos共m␪兲. In general, a maximum
value of r / a exists for which this Fourier series con-
verges, which means there is a maximum value of r / a for
which a solution exists. To find this maximum, consider
the convergence properties of the Fourier coefficients
Bm = 兰Br共␪兲cos共m␪兲d␪ / ␲. If Br is an analytic function of
␪ this integral can be performed using the method of
residues of complex analysis. If ␪p is the distance of the
nearest pole of Br共␪兲 from the real axis, then as m → ⬁
the Fourier coefficients have the form Bm
= Bc exp共−m␪p兲. In other words, the Fourier coefficients
of a generic function Br共␪兲 decay exponentially as
m → ⬁. Now consider the coefficients Bm共r / a兲m. As
m → ⬁ these coefficients are proportional to
exp兵m关ln共r / a兲 − ␪p兴其, which means they diverge exponen-
tially with m if r / a ⬎ exp共␪p兲.
Even when distant coils can in principle produce the
required magnetic field on the plasma surface, practical
coils may not exist. Laplace’s equation implies that the
intrinsic difficulty of distant coils is exponentially depen-
dent on the wave number k of the distribution of the
normal magnetic field that they are producing on the
plasma surface. The number of independent distribu-
tions of magnetic field that can be produced by practical
coils is comparable to the number of Fourier compo-
nents of the magnetic potential ␾, Eq. 共113兲, that have a
wave number smaller than some critical value km. In a
large-aspect-ratio circular tokamak, k = m / a and the ra-
tio of magnetic field on the coils at r = b to that on the
plasma at r = a is 共b / a兲共m−1兲. For a practical separation
between the coils and the plasma in a fusion power
plant, b / a ⬇ 1.7, the ratio of the field at the coil to that
on the plasma is five to one for m = 4. Practical tokamak
coils can control about four properties of the plasma,
corresponding crudely to the poloidal Fourier harmonics
up to Mm = 4, so tokamak coil designs have four degrees
of freedom, Nf = 4. For stellarators the wave number of a
magnetic-field distribution is k ⯝ 冑共m / a兲2 + 共n / Ro兲2,
where a is the minor and Ro is the major radius of the
torus. To preserve the Np-fold periodicity of the stellar-
ator, the toroidal mode number of a field distribution
must satisfy n = nhNp, where nh is an integer. Summing
over the possible combinations of m and nh that satisfy
k 艋 Mm / a, one finds that the number of degrees of free-
dom of stellarator coils is
2
Mm
Nf ⬇ 1.8 , 共115兲
⑀p
where the inverse aspect ratio per period of the stellar-
ator ⑀p ⬅ Npa / Ro. Typically ⑀p ⬇ 1 / 2, so with Mm = 4, stel-
larator coils have about 50 degrees of freedom.
Efficient coils for stellarators should exist, since the
number of degrees of freedom in stellarator coil design,
 Allen H. Boozer: Physics of magnetically confined plasmas
Nf ⬇ 50, is larger than the number of shape parameters
Ns that are needed to achieve reasonably optimized
plasma shapes. The COILOPT code 共Strickler, Berry, and
Hirshman, 2002兲 tackles this problem directly by varying
parameters that define the coils until a combined opti-
mum is found for the coils and the plasma. However, the
large number of degrees of freedom and the nonlinear-
ity of the optimization process means the design of stel-
larator coils for optimal plasmas is extremely subtle, and
it is important to understand the choices that are made
during this optimization.
The fundamental choices in the optimization of a stel-
larator are those of the plasma shape and the coils. The
optimized plasma shapes, xជ s共␪ , ␸兲, are found using a set
of Ns parameters 共Nührenberg et al., 1995; Neilson et al.,
2000兲. For example, the shape parameters could be the
Fourier coefficients of Rs共␪ , ␸兲 and Zs共␪ , ␸兲 of a repre-
sentation of the surface in 共R , ␸ , Z兲 cylindrical coordi-
nates. The optimization process for the plasma shape
determines Ns shape parameters, so plasma optimization
can only determine Ns constraints on the coils. Practical
limits exist on the number of shape parameters that can
be determined. These limits arise not only from the in-
creasing difficulty of optimization with additional pa-
rameters, but also from the numerical accuracy of the
codes that determine the physics properties associated
with a given plasma shape. The constraints on the coils
should be chosen to ensure the Ns known shape param-
eters have their desired values but with no constraints
imposed for fitting unknown shape parameters. This can
be done by studying the relation between changes in the
plasma shape and small changes in the fields due to the
coils, which is discussed in Sec. V.D.2.
2. Plasma response to coil changes
The relation between perturbations to the plasma
shape and small changes in the magnetic field due to the
coils is important for stellarator coil design, the elimina-
tion of islands due to magnetic-field errors, and the de-
sign of feedback systems for resistive wall modes in to-
kamaks.
A small change in the plasma shape is equivalent to a
normal displacement ␰ជ · n̂ of the plasma surface and has a
simple and unique relation to the normal component of
the perturbed magnetic field on the original plasma sur-
face ␦B ជ · n̂. To prove this, we first note that a tangential
displacement, which means ␰ជ · n̂ = 0 but ␰ជ ⫽ 0 with
xជ s共␪ , ␸兲 → xជ s共␪ , ␸兲 + ␰ជ 共␪ , ␸兲, does not change the plasma
shape but does change the 共␪ , ␸兲 coordinate system that
describes that shape. In an ideal plasma ␦B ជ =ⵜជ 共␰ជ ⫻ B
ជ 兲,
Eq. 共46兲, and ␦B ជ ·ⵜ
ជ ␺ =Bជ ·ⵜ
ជ 共␰ជ · ⵜ
ជ ␺ 兲, which relates ␰ជ · n̂
t t
ជ sary to know the change in the normal magnetic field on
and ␦B · n̂ since the normal to the plasma surface is n̂
ជ ␺ / 兩ⵜ
ជ ␺ 兩. Actually the equation ␦B ជ =ⵜ
ជ ⫻ 共␰ជ ⫻ B
ជ 兲 holds
=ⵜ t t
whether or not the magnetic field is embedded in a per-
fectly conducting plasma, provided the rotational trans-
form on the surface is an irrational number. The reason
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1103
ជ =ⵜ
is ␦B ជ ⫻ ␦A
ជ . But any vector, including the perturbed
ជ = ␰ជ ⫻ B
ជ
vector potential, can be written in the form ␦A
ជ ␦g provided a function ␦g共xជ 兲 exists that satisfies the
+ⵜ
magnetic differential equation B ជ ·ⵜ
ជ ␦g = B
ជ · ␦A
ជ . This
equation for ␦g is solvable when the rotational trans-
form ␫ is an irrational number and the perturbation does
not include a loop voltage. A loop voltage was consid-
ered by Lüst and Martensen 共1960兲; also see the discus-
ជ =ⵜ
ជ ⫻ 共␰ជ ⫻ B
ជ兲
sion of Eq. 共26兲. The general validity of ␦B
on irrational surfaces is another way of stating that the
rotational transform is not changed in linear order by a
perturbation ␰ជ except near a rational surface 关Eq. 共17兲兴.
Functions of ␪ and ␸, such as the magnetic-field per-
turbation normal to the original plasma surface ␦B ជ · n̂,
are conveniently expressed using orthonormal functions.
A set of dimensionless functions fi共␪ , ␸兲 on the plasma
surface are orthonormal if
冖
fif*j wda = ␦ij , 共116兲
where w ⬎ 0 is an arbitrary weight function and da is the
area element of the plasma surface. The normal compo-
nent of the magnetic perturbation on the original plasma
surface is expanded in the fi’s in the form
ជ · n̂ = w 兺 ⌽ f* .
␦B 共117兲
j j
The expansion coefficients have units of flux
冖
⌽i = ជ · n̂da.
f i␦ B 共118兲
The derivation of the Ns constraints on stellarator
coils that are required to fit Ns parameters in the plasma
shape illustrates a response analysis. A small change in
each of the Ns shape parameters produces a small
change in the plasma shape, which is a normal displace-
ment ␰ជ · n̂ of the plasma surface, and determines a small
s
change in the normal magnetic field ␦B ជ · n̂. The N func-
s s
ជ
tions ␤ 共␪ , ␸兲 ⬅ ␦B · n̂ can be transformed into a set of
s s
not more than Ns orthonormal functions fs共␪ , ␸兲 by a
ជ · n̂ can be
Gram-Schmidt process such that any of the ␦B s
expanded in terms of the fs’s 关Eq. 共117兲兴. The constraints
ជ · n̂da = 0, then ensure that through lin-
on the coils, 养f B s
ear order in the field error the Ns plasma shape param-
eters are reproduced but with no constraints on the coils
from unknown shape parameters. Good stellarator coils
exist if the Ns field distributions that are needed are in
the solution space of the Nf field distributions that effi-
cient stellarator coils can produce.
In a number of problems involving coils, it is neces-
ជ · n̂, that is re-
the plasma surface due to the coils, ␦B x
quired to produce a given normal magnetic field on the
ជ · n̂. The difference between ␦B ជ · n̂ and
plasma surface, ␦B
ជ · n̂ is the normal field produced by the perturbed
␦B x
1104 Allen H. Boozer: Physics of magnetically confined plasmas
plasma current ␦Bជ · n̂. The decomposition of ␦B ជ · n̂ into a
i
part produced by coil currents and a part produced by
internal plasma currents can be carried out if there is
enough information to solve Laplace’s equation for the
magnetic potential ␾ between the plasma and the con-
ducting structures. Two boundary conditions are re-
quired. In a cylinder, a solution to Laplace’s equation
has the form ␾ = 兵␾i共a / r兲m + ␾x共r / a兲m其cos共m␪兲 and
boundary conditions must determine both ␾i, which
gives the effect of currents inside the plasma, and ␾x,
which gives the effect of currents outside the plasma.
The response of the plasma to perturbations deter-
mines the relation between the perturbed normal field
ជ · n̂, and the normal field due
due to external current, ␦B x that break these symmetries.
to the internal perturbed plasma currents, ␦Bជ · n̂. Since
i
only two boundary conditions are required to determine
a unique solution to Laplace’s equation, any two inde-
pendent pieces of information will suffice. One example
is the relation between the perturbed normal and tan-
gential magnetic-field components. This information is
contained in ␦B ជ at the location of the unperturbed
plasma surface. Since ␦B ជ is continuous at the plasma
surface, but ␦B ជ =ⵜជ ␾ outside the plasma, the vector ␦B ជ Nr sets of integers n / m that give the rational numbers
gives two functions of information on the plasma sur-
ជ ␾ = n̂ · ␦B
face, ␾ and n̂ · ⵜ ជ , and not three as one might first
assume.
Given a plasma model, a linear relation exists at the
location of the orginal plasma surface between a small
change in the normal component of the magnetic field
produced by external currents ␦B ជ · n̂ and the perturbed
x
normal field ␦Bជ · n̂. This linear relation is conveniently
expressed using matrices. The fluxes that are the expan-
sion coefficients of the perturbed normal field on the
plasma surface form a matrix vector ⌽ ជ ⬅ 养 ជf共␦B
ជ · n̂兲da.
The matrix vector ជf has the same functions fi共␪ , ␸兲 as its
components. Analogously, the fluxes that are the expan-
sion coefficients of the normal field on the plasma sur-
face due to perturbed coil currents form a matrix vector
⌽ជ . The linear relation between ␦Bជ · n̂ and ␦B
ជ · n̂ deter-
x x
mines the permeability matrix,
⌽ J·⌽
ជ =P ជ . 共119兲
x
The permeability matrix P J is a property of the plasma
and for an ideal plasma can be determined using a ␦W
J is known, it is straightforward to
stability code. Once P
uniquely relate any small perturbation in the plasma
shape to the change in the normal magnetic field on
plasma surface that is produced by the coils.
Tokamak and stellarator plasmas are particularly sen-
sitive to magnetic perturbations that resonate with low-
order rational surfaces and cause islands to open. The
coils can be modified to eliminate the islands in one
plasma state 共Hudson et al., 2002兲, though error correc-
tion or trim coils may be needed to eliminate islands
over a broad range of plasma states. The issue of islands
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
would not exist in a tokamak with perfect axisymmetry,
but it does exist even for an ideal stellarator if the num-
ber of periods is small, Np ⱗ 6. As the plasma equilib-
rium changes, a magnetic field that was nonresonant can
develop resonant components. However, there are a
relatively small number Nr of low-order rational sur-
faces that are consistent with the periodicity of a stellar-
ator. This means that any set of trim coils that obeys the
periodicity of the stellarator and has Nr sets of leads can
control the islands provided the matrix between the
fields produced by the trim coils and the resonant fields
is nonsingular. In practice, tokamaks are not precisely
axisymmetric and stellarators are not precisely periodic,
so additional coils may be needed to correct field errors
The information required to determine whether a
given set of trim coils can control the islands of a plasma
equilibrium can be found using an ideal ␦W stability
code. As discussed in Sec. V.C.3, perturbations at the
ជ␺
edge of an ideal plasma, ␰ជ · n̂, can cause a jump in ␰ជ · ⵜ t
at the rational surfaces. These jumps imply islands
would open in a resistive plasma with the island width
proportional to the square root of the jump. If there are
associated with these jumps, then there are Nr specific
forms for ␰ជ · n̂ on the plasma surface that are propor-
tional to the jumps but otherwise have the smallest pos-
sible amplitude, 养共␰ជ · n̂兲2wda. All other small displace-
ments of the plasma edge cause no jumps and, therefore,
no islands. These N specific forms for ␰ជ · n̂ on the plasma
r
surface give Nr distributions of normal magnetic field,
␦B ជ · n̂ = n̂ · ⵜ
ជ ␾ , and have N associated perturbed tan-
r r r
gential fields on the plasma surface, which give the mag-
netic scalar potentials ␾r there. From the ␾r and the
ជ ␾ , one can construct the N orthonormal functions
n̂ · ⵜ r r
fr共␪ , ␸兲 that give the externally produced fluxes
养fr␦B ជ · n̂da that control the islands. These fluxes are lin-
x
early related to the currents in the trim coils. If there are
Nt independent sets of trim coils, which means indepen-
dent currents It, then the fluxes required for island con-
trol obey 养f ␦B ជ · n̂da = 兺 M I with the mutual induc-
r x t rt t
tance matrix Mrt an Nr ⫻ Nt matrix. If this matrix has Nr
nonzero eigenvalues, the trim coils can control the
island-causing resonances, though practicality requires
that none of the Nr eigenvalues be excessively small.
In tokamaks, important kinklike instabilities, the re-
sistive wall modes, can be stabilized by conducting struc-
tures, such as the chamber wall that surrounds the
plasma; see Sec. V.C.3. These instabilities grow on the
resistive time scale of the wall and can be feedback sta-
bilized using external coils. The permeability matrix P J
contains all of the plasma information that is needed for
calculating the stability and the feedback of wall modes.
The circuit representation of the plasma that is associ-
ated with the permeability matrix has been used to de-
sign and interpret feedback systems for a number of ma-
jor tokamak experiments 共Bialek et al., 2001兲. Prior to
 Allen H. Boozer: Physics of magnetically confined plasmas
the development of the circuit representation of resistive
wall modes, Lazarus, Lister, and Neilson 共1990兲 devel-
oped a closely related theory for the feedback of the
vertical instability of tokamaks. The vertical instability
causes a Ẑ motion of the plasma and limits the degree to
which tokamak plasmas can be shaped to achieve higher
beta operation. A number of groups have developed
methods for studying feedback of resistive wall modes,
starting with the early work of Bishop 共1989兲. Recent
discussions of feedback techniques have been given by
Fitzpatrick 共2001兲, Bondeson et al. 共2002兲, Chance et al.
共2002兲, and Boozer 共2003兲.
The energy that is required to drive resistive wall
modes comes from the plasma, so a representation of
the effects of the perturbed plasma current on the exter-
nal circuits is critical. Since these effects must be propa-
gated through a solution to Laplace’s equation, the ef-
fects can be represented by the magnetic field normal to
the plasma surface that is produced by the perturbed
currents inside the plasma, ␦B ជ · n̂. This means the effects
i
of the perturbed plasma current can be represented by a
surface current flowing on the plasma surface that pro-
duces the same normal field at the location of the unper-
turbed plasma surface as the perturbed plasma current.
A surface current can be represented as a matrix vec-
tor with elements that are discrete currents Ji. The nor-
mal field on the surface due to these currents is linear, so
it can be represented using Eq. 共118兲 as a magnetic-flux
ជ =L J · Jជ . The matrix L J is called the
matrix vector ⌽ J p p is the wall inductance, Mpw = Mwp is the mutual induc-
plasma inductance and depends only on the geometric
shape and the size of the surface. The surface current
that gives the same normal field on the unperturbed
plasma surface as the perturbed plasma current is
J −1 · 共P
Iជp = L J −J ជ .
1兲 · ⌽ 共120兲
p x
The representation of a surface current by a matrix vec-
tor Jជ is demonstrated by noting that the current flowing
on a surface ␺t = ␺s has the form
ជjs = ␦共␺t − ␺s兲ⵜជ ␬ ⫻ ⵜជ ␺t . 共121兲
This is the most general vector that is divergence-free,
ជ ␺ = 0, and is zero except in the
lies in the surface ជjs · ⵜ t
surface ␺t = ␺s. The current potential ␬共␪ , ␸兲 can be ex-
panded in the orthogonal functions, ␬ = 兺J*i fi共␪ , ␸兲. Each
element Ji of the matrix vector Jជ has units of amperes
and can be viewed as a current in a circuit. Actually, the
general expression for the current potential ␬共␪ , ␸兲 is the
sum of the single-valued current potential 兺J*i fi plus two
non-single-valued terms. One of these terms is
−␪Is / 共2␲兲, which gives the net toroidal current Is in the
surface, and the other is ␸Gs / 共2␲兲, which gives the net
poloidal current Gs in the surface. The currents Is and
Gs are not usually retained in analyses of wall modes.
When the plasma is ideal, the permeability matrix P J plification and the phase shift are given by si and ␣i
has real eigenvalues that are customarily written as −1 / si
共Boozer, 2003兲. The stability coefficients si are propor-
tional to the energy required to drive the perturbation
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1105
with a positive si implying a negative energy. An un-
J has a negative
stable wall mode arises if the matrix P
eigenvalue, which means a positive si. The relation be-
J and the energy is demonstrated by considering
tween P
the power required to drive a perturbing current Jជ on a
surface infinitesimally outside of the plasma. That power
is P = −兰jជ · ␦Eជ d3x. Using Eq. 共121兲, P = − 养 ␬共ⵜ ជ
d s d
⫻ ␦E ជ 兲 · daជ . Since ⳵␦Bជ / ⳵t = −ⵜ
ជ ⫻ ␦E
ជ , the power is P
d
= Jជ † · d⌽ជ / dt. The flux and the current are proportional to
ជ =P
each other, ⌽ J ·L J · Jជ , so the energy ␦W required to
p
drive the perturbation, Pd = d␦W / dt, has the explicitly
real form ␦W = 1 共Jជ † · ⌽
4
ជ +⌽ជ † · Jជ 兲. Since the inductance ma-
J is a positive Hermitian matrix, the energy ␦W is
trix L p
negative only if the permeability PJ has a negative eigen-
value.
The circuit equations for wall modes are particularly
simple in the case of primary importance, a single mode
passing through marginal stability, which means one sta-
bility coefficient s passing through zero. Because of the
singularity of the permeability matrix at s = 0, only the
marginal mode is important, so the matrix vectors can
be approximated by a single matrix element, the ele-
ment that represents the marginally stable perturbation.
For this problem, ⌽ = −⌽x / s, ⌽x = MpwIw, ⌽w = LwIw
+ MwpIp, and d⌽w / dt = −RwIw. ⌽w is the magnetic flux
that penetrates the wall, Iw is the current in the wall, Lw
tance between the wall and the plasma, and Rw is the
wall resistivity. Simple algebra demonstrates that the
flux through the wall is
再 冎
1 + s MpwMwp
⌽w = Lw 1 − Iw . 共122兲
s L wL p
When the effective inductance of the wall, ⌽w / Iw, is
negative, the wall mode grows at a rate proportional to
Rw. By adding sensors and actively driven coils to the
circuit equations, one can stabilize the wall mode by
feedback.
In a rotating tokamak plasma, the eigenvalues of the
permeability matrix 共Boozer, 2003兲 are complex num-
bers 1 / 共−si + i␣i兲. The quantity ␣i gives the toroidal
torque between the plasma and the mode i. The toroidal
torque exerted by an axisymmetric surface that carries a
surface current ជjs is given by ␶␸ = −兰共jជs ⫻ ␦B ជ 兲 · 共⳵xជ / ⳵␸兲d3x,
which implies ␶ = i共N / 2兲共Jជ † · ⌽
ជ −⌽
ជ † · Jជ 兲 when the ␸ de-
␸
pendence of fi is written as exp共−iN␸兲. When an exter-
nal perturbation is applied to a rotating tokamak
plasma, the perturbation on the plasma surface can be
amplified and can have a toroidal phase shift. The am-
共Boozer, 2003兲. Plasma rotation can stabilize the resis-
tive wall mode by dragging the mode toroidally at too
rapid a rate for it to penetrate the wall.
1106 Allen H. Boozer: Physics of magnetically confined plasmas
VI. RADIATION AND TRANSPORT
• A fusion burn of deuterium and tritium requires that
the characteristic time for loss of energy from the
plasma satisfy ␶E ⬇ 4 ⫻ 1021 / niT, where ␶E is in sec-
onds, ni is the number of deuterium-tritium ions per
cubic meter, and T is the plasma temperature in ki-
lovolts. The characteristic energy confinement time
of a plasma with few impurities and a temperature of
about 20 keV is predominately due to thermal en-
ergy transport by plasma processes, not electromag-
netic radiation.
• Energy transport determines the minimum size and
hence the cost of an experiment to study plasmas
that burn deuterium and tritium. For a power plant,
the level of transport must be consistent with the
required system parameters; see Sec. II.
• The energy confinement time ␶E of proposed experi-
ments is estimated using empirical power-law scal-
ings, which have the fundamental assumption that no
critical values are crossed of the parameters that en-
ter the power law.
The scaling of the energy confinement time of toroi-
dally confined plasmas is a primary issue in the feasibil-
ity and cost of a burning plasma experiment 共ITER
Physics Expert Group on Confinement …, 1999兲. By
definition a burning plasma experiment burns deuterium
and tritium while requiring little external power. The
loss of energy from a fusing plasma at 20 keV with few
impurities is dominated by plasma processes, called en-
ergy transport, not electromagnetic radiation. The focus
on proposals for burning plasma experiments has meant
that energy transport has been a major topic for plasma
research.
Energy transport is also an issue for the feasibility of
fusion power, but differences exist between the transport
issues for fusion power and those for a burning plasma
experiment. The size of a burning plasma experiment is
essentially determined by the magnitude of the energy
transport: the smaller the transport, the smaller and
cheaper the required experiment. However, the basic
size of a power plant is determined by issues that have
little to do with plasma transport coefficients. The criti-
cal issues in a power plant are whether the energy trans-
port is consistent with what is required to maintain a
steady burn—neither too large nor too small—and
whether the particle transport is sufficiently large to re-
move alpha-particle ash. In addition, the feasibility of
steady-state tokamak power plants is dependent upon
the natural profiles that the temperature, density, and
current take in a fusion plasma. The performance of
stellarator power plants is far less dependent on the is-
sue of profiles.
Energy transport can provide an upper limit on the
plasma beta, ␤ ⬅ 2␮0p / B2, because transport for given
plasma conditions is generally larger, the smaller the
magnetic field.
Predictions of the energy confinement times of experi-
ments are usually based on empirical scaling relations
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
共ITER Physics Expert Group on Confinement …, 1999兲.
Empirical scaling relations generally assume the math-
ematical form of a power law,
␶E共x1,x2, . . . 兲 = a0xa11xa22 ¯ , 共123兲
so each parameter of the set 共x1 , x2 , . . . 兲 enters multipli-
catively. The constants 共a0 , a1 , . . . 兲 are chosen to obtain
the best fit to the data. Mathematics implies that power-
law scaling is a precise description if and only if the
parameters 共x1 , x2 , . . . 兲 have no characteristic values or
scale. A function is independent of the scale s if f共x / s兲
= f共1 / s兲f共x兲, which implies f共x兲 ⬀ xa. All other functions
have a scale, and the properties of the function depend
on the size of x relative to s. For example, the scale of
sin x is 2␲. The constants of a power law 共a0a1 , . . . 兲 are
easily fit to data by a linear regression of the logarithm
of ␶E against the logarithms of the parameters.
The most commonly used scaling relations are power
laws based on experimental parameters such as input
power, plasma current, and plasma size. Less used but
more scientifically appealing scaling relations are based
on dimensionless parameters 共Connor, 1984兲. The di-
mensionless energy confinement time is the confinement
time times the cyclotron frequency of deuterium, ⍀
⬅ eB / md. Important dimensionless parameters are 共a兲
the gyroradius of deuterium, ␳ ⬅ 共冑Tmd兲 / 共eB兲, divided
by the plasma radius, which is called ␳*; 共b兲 the number
of bounces between collisions a deuterium ion makes
when trapped in the variation in the magnetic-field
strength on a magnetic surface, a ratio called 1 / ␯*; and
共c兲 ␤ = 具2␮0p / B2典. In terms of these parameters,
⍀␶E = a0␳a*␳␯a*␯␤a␤ . 共124兲
A. Electromagnetic radiation
• Although electromagnetic radiation is usually a sub-
dominant energy-loss mechanism for deuterium-
tritium 共DT兲 fusion systems, electromagnetic radia-
tion limits the feasibility of non-DT fusion fuels and
the tolerable level of impurities in all fusion systems.
• The most important types of radiative losses from
magnetically confined plasmas are bremsstrahlung,
cyclotron, and atomic.
Electromagnetic radiation is a subdominant energy-
loss mechanism in a deuterium-tritium 共DT兲 fusion
plasma operating at 20 keV with a low level of impuri-
ties. However, electromagnetic radiation limits the toler-
able level of impurities, the range of temperatures, and
the types of fuel 共Nevins, 1998兲 for which fusion energy
is feasible. Three types of radiative losses are important:
bremsstrahlung, cyclotron, and atomic.
Electromagnetic radiation arises when the current
density ជj has both a nonzero curl and a time derivative.
The time derivative of the current density of a single
particle is ជ̇j共xជ 兲 = qaជ ␦共xជ − xជ p兲, where xជ p共t兲 is the position of
the particle, aជ = d2xជ p / dt2 is its acceleration, and ␦共¯兲 is
the Dirac delta function. The rate at which a nonrelativ-
 Allen H. Boozer: Physics of magnetically confined plasmas
istic particle of charge q loses energy H through electro-
magnetic radiation is given by the Larmor formula,
dH q 2a 2
=− , 共125兲
dt 6 ␲ ⑀ 0c 3
which is derived in the standard electrodynamics texts.
Bremsstrahlung, which in German means radiation
due to deceleration, arises from the electrostatic scatter-
ing of one charged particle by another. The power loss
due to bremsstrahlung is dominated by the scattering of
electrons by ions. The power loss per unit volume due to
scattering by ions of charge Ze with a number density nz
is proportional to Z2nzne冑Te. The typical photon emit-
ted in bremsstrahlung has an energy equal to the elec-
tron temperature. In plasmas of fusion interest there is
negligible reabsorption of bremsstrahlung radiation be-
cause the total radiated power is far below the black-
body level. An accurate derivation of bremsstrahlung is
given by Karzas and Latter 共1961兲.
The basic dependences and power loss of bremsstrah-
lung can be understood starting with the expression for
the acceleration of one charged particle by another, aជ 1
= 共q1q2 / 4␲⑀0m1兲共xជ 1 − xជ 2兲 / 兩xជ 1 − xជ 2兩3. The maximum accel-
eration occurs at the point of closest approach of the
two particles, b ⬅ min共兩xជ 1 − xជ 2兩兲. The characteristic fre-
quency of the bremsstrahlung radiation is ␻ = v / b, where
v is the particle velocity, which implies the characteristic
wave number satisfies kb = v / c. For nonrelativistic par-
ticles, v / c Ⰶ 1, the wavelength of the radiation is long
compared to the distance between the radiating par-
ticles. The time derivative of the current density of two
particles is ជ̇j共xជ 兲 = q1aជ 1␦共xជ − xជ 1兲 + q2aជ 2␦共xជ − xជ 2兲. Since the
wavelength of the radiation is long compared to the dis-
tance between the radiating particles, one can ignore
that distance and let ជ̇j共xជ 兲 = 共q1 / m1 − q2 / m2兲m1aជ 1␦共xជ − xជ 1兲
using Newton’s third law to write m2aជ 2 = −m1aជ 1. The ex-
pression for ជj implies that radiation from electron-
electron scattering vanishes in the nonrelativistic limit
and that bremsstrahlung is primarily a result of electron-
ion scattering.
The change in energy of an electron with speed v that
passes by an ion that has a charge Ze at a distance b is
␦H ⬇ 共dH / dt兲共b / v兲, where dH / dt is given by the Larmor
formula, Eq. 共125兲, with the acceleration a
⬇ 共Ze2 / 4␲⑀0me兲 / b2. The bremsstrahlung power pb emit-
ted per unit volume is given by the number of electrons
per unit volume times the number of ions the electrons
pass per unit time. The number of ions of charge Z that
the electrons pass per unit time at a range between b
and b + db is nzv2␲bdb, so pb ⬇ nzne 兰 ␦Hv2␲bdb. This
integral is proportional to 1 / bmin with bmin the distance
of closest possible approach. In classical mechanics the
distance of closest possible approach is given by
共Ze2 / 4␲⑀0兲 / bmin = mev2 / 2, but for electrons that have en-
ergies of importance for fusion systems this distance is
much smaller than the limit set by quantum mechanics,
which is bminmev ⬇ ប. Quantum effects determine the
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1107
closest possible approach when v / c ⬎ 2Z␣, where the
fine-structure constant ␣ ⬅ e2 / 共4␲⑀0cប兲 ⬇ 1 / 137. The
typical energy of an emitted photon is ប␻ ⬇ បv / bmin
⬇ mev2 ⬇ T. The basic dependences of the emitted
power, Z2nzne冑Te, are obtained by combining the re-
sults.
Cyclotron radiation results from the acceleration of
electrons with velocity vជ by the magnetic field, aជ
ជ . The characteristic frequency of the asso-
= −共e / m 兲vជ ⫻ B
e
ciated motion is the electron cyclotron frequency ⍀e
⬅ eB / me, and this is the frequency at which cyclotron
radiation is emitted by an electron. A straightforward
application of the Larmor formula, Eq. 共125兲, gives the
power emitted per unit volume due to the cyclotron mo-
tion of Maxwellian electrons, pc = ␻pe 2
⍀2e T / 共3␲c3兲. Since
具mv⬜2
/ 2典 is the energy in two independent components
of velocity, and the energy in three independent compo-
nents is 23 T, one has 具mv⬜
2
/ 2 / 典 = T. The plasma frequency
is ␻pe = 冑e2ne / 共me⑀0兲, which is the frequency with which
the electrons would oscillate if displaced en masse from
the ions. Thermodynamics implies that the emission of
electromagnetic waves due to the cyclotron motion can-
not exceed the energy flux leaving a blackbody in the
relevant frequency range. The energy flux leaving a
blackbody at frequencies less than ␻ is Fbb
= ␻3T / 共12␲2c2兲 provided the energy per photon is far
less than the temperature, ប␻ Ⰶ T. This is the Rayleigh-
Jeans limit, which is the relevant limit when ␻ ⬃ ⍀e. To
be consistent with thermodynamics, the plasma must re-
absorb the radiated power within a distance La
⬅ 4␲Fbb / pc = 共⍀e / ␻pe兲共c / ␻pe兲, which is a fraction of a
millimeter in a fusion plasma. Due to relativistic effects,
electrons also emit radiation at harmonics of the cyclo-
tron frequency with the emission at each higher har-
monic reduced by a factor T / 共mec2兲 compared to the
previous harmonic. The power emitted by a region is
comparable to the blackbody flux Fbb up to a frequency
␻ that is set by the first cyclotron harmonic for which the
power can leave the plasma without strong reabsorption.
Accurate calculations of the power losses due to cyclo-
tron emission are complicated. Albajar, Bornatici, and
Engelmann 共2002兲 give recent results and historical ref-
erences.
Atomic radiation arises from electrons switching
atomic levels and from the capture of free electrons into
bound states. Significant atomic radiation only arises if
the electron temperature is low enough for the atom or
ion to have bound electrons. The atomic radiation from
an element with a number density nz has the form pz
= nenzRz共T兲, with Rz共T兲 a function that is large at elec-
tron temperatures at which ionization occurs. The func-
tion Rz共T兲 for a single element can have large variations
with multiple peaks. Under fusion conditions Rz共T兲 is
negligible except for elements with high atomic num-
bers, such as iron. Functions Rz共T兲 have been given by
Post et al. 共1977兲.
At a given impurity level, the power losses due to
bremsstrahlung and atomic radiation depend quadrati-
1108 Allen H. Boozer: Physics of magnetically confined plasmas

cally on the plasma density, as does the nuclear power

input, so the nuclear power input minus the radiative
losses due to bremsstrahlung and atomic radiation can
df ⳵f ⳵
= +
dt ⳵t ⳵xជ
·
dxជ
dt
f + 冉 冊
⳵
⳵pជ
·
dpជ
dt
f . 冉 冊 共128兲

be viewed as an effective power input, peff = n2feff共T兲.
The power loss due to cyclotron emission and absorp-
tion is more complicated and in some features resembles
a diffusive transport process.
Fusion power is feasible only at low impurity levels. In
addition to the enhancement of electromagnetic radia-
tion, the ionization of impurities adds many electrons to
the plasma. These electrons exert a pressure and trans-
port energy but are unrelated to the production of fu-
sion power.
B. Kinetic theory
• Kinetic theory is needed to calculate transport coef-
ficients such as particle diffusivities, thermal conduc-
tivities, and bootstrap currents.
• Collisions in plasmas change the trajectories of par-
ticles diffusively. This is in contrast to collisions in
ordinary gases, which produce large abrupt changes
in the particle trajectories.
• Plasma confinement forces the distribution functions
of the particles to deviate from local Maxwellians.
The rate of entropy production that is required to
hold the distribution functions away from local Max-
wellians allows a simple calculation of transport co-
efficients in the low-collisionality limit.
Transport calculations are carried out using kinetic
theory. The fundamental quantity in kinetic theory is the
distribution function f共xជ , pជ , t兲, which describes the evolu-
tion of a large group of identical particles. In a plasma,
there is a distribution function for the electrons and for
each of the ion species. The distribution function is the
density of particles in phase or momentum space 共xជ , pជ 兲.
In other words, the number of particles in a region of
momentum pជ and ordinary space xជ is the integral of the
If pជ is the canonical momentum of Hamiltonian mechan-
ics, Eq. 共128兲 follows from Eq. 共127兲 because of Hamil-
ton’s equations dxជ / dt = ⳵H / ⳵pជ and dpជ / dt = −⳵H / ⳵xជ . In
the presence of a magnetic field, the canonical momen-
ជ 关Eq. 共191兲兴. If pជ ⬅ mvជ , Eq. 共128兲 fol-
tum is pជ = mvជ + qA
lows because xជ̇ = pជ / m and pជ̇ = q共E ជ + pជ ⫻ B
ជ / m兲, so
共⳵ / ⳵xជ 兲 · xជ̇ = 0 and the momentum space divergence of pជ̇ is
zero, 共⳵ / ⳵pជ 兲 · pជ̇ = 0.
Without collisions, the kinetic equation is called the
Vlasov equation, df / dt = 0. The Vlasov equation is easily
understood. Given the distribution function at t = t0, the
distribution function at t = t0 + ␦t is obtained by advancing
each particle in the distribution along its trajectory, xជ
= xជ 0 + 共dxជ / dt兲␦t and pជ = pជ 0 + 共dpជ / dt兲␦t. If f共xជ 0 , pជ 0兲 is the
given distribution, then for an infinitesimal interval
of time, ␦t = t − t0, the distribution function is f„xជ
− 共dxជ / dt兲␦t , pជ − 共dpជ / dt兲␦t…, which implies ⳵f / ⳵t
= −共dxជ / dt兲 · ⳵f / ⳵xជ − 共dpជ / dt兲 · ⳵f / ⳵pជ . The Vlasov equation is
a hyperbolic, partial differential equation that has one
characteristic. That characteristic is the trajectory of a
particle.
The collision operator is the momentum-space diver-
gence of a flux F ជ of particles through phase space 共xជ , pជ 兲,
⳵ ជ
C共f兲 = − · F. 共129兲
⳵pជ
Collisions are caused by the graininess of the plasma, an
effect that is not directly described by the distribution
function f共xជ , pជ 兲.
The graininess of the plasma leads to complicated
electric fields that scatter particles and cause the colli-
sional flux F ជ . The electric field at position xជ due to a
charge q at xជ 1 is

冕
distribution function over that region, 兰fd3pd3x. A more
detailed treatment of kinetic theory than that given here ជ 共xជ ,xជ 兲 = − q ikជ ikជ ·共xជ −xជ 兲 d3k
E 1 e 1 , 共130兲
can be found in Helander and Sigmar 共2002兲 and in most ⑀0 k2 共2␲兲3
plasma textbooks.
The evolution of the distribution function is given by which is demonstrated by checking that ⵜ ជ ·E
ជ = 共q / ⑀ 兲␦共xជ
0
the kinetic equation ជ ជ ជ ជ ជ ជ
− x1兲, ⵜ ⫻ E = 0, and E共x → ⬁ , x1兲 = 0. Note the Dirac delta
function has the representation
df

冕
= C共f兲, 共126兲
dt ជ d 3k
␦共xជ − xជ 1兲 = eik·共xជ −xជ 1兲 . 共131兲
共2␲兲3
where C共f兲 is the collision operator. The total time de-
rivative of kinetic theory is an extension of the concept The electric field that is obtained from a uniform density
of the total time derivative of fluid mechanics, Eq. 共44兲, ជ 典 ⬅ 兰E
ជ 共xជ , xជ 兲nd3x , is zero. However, the
n of charges, 具E
which operates on functions g共xជ , t兲 in ordinary space, to 1 1
root-mean-square electric field from a uniform density
functions f共xជ , pជ , t兲 in phase space, of charge is not zero,
df ⳵f dxជ ⳵f dpជ ⳵f
= + · + · ,
dt ⳵t dt ⳵xជ dt ⳵pជ
共127兲 具E2典 ⬅ 冕 ជ 共xជ ,xជ 兲 · E
E 1
ជ 共xជ ,xជ 兲nd3x ,
1 1 共132兲

ជ f. Equation 共127兲 can also be written as

where ⳵f / ⳵xជ ⬅ ⵜ and can be written as

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas 1109

具E2典 = n 冉 冊冕q
⑀0
2
1 d 3k
k2 共2␲兲3
. 共133兲 n共xជ 兲 ⬅ 冕 fd3p; 共138兲

This electric field causes accelerations ⌬aជ = 共q / m兲E ជ of momentum,

the particles, which for a fast-moving particle persist for
a time ⌬t = 1 / kv with v the speed of the particle. That is,
a particle has essentially random changes in its velocity,
nmuជ 共xជ 兲 ⬅ 冕 pជ fd3p; 共139兲

⌬vជ = 共⌬aជ 兲共⌬t兲, separated by time intervals of order ⌬t. and energy conservation,

冕
These random changes scatter the velocity at an average p2 3
rate ␯ = 具共⌬v兲2 / ⌬t典 / v2. That is, ␯ = 具共q / m兲2E2⌬t典 / v2, which ⑀共xជ 兲 ⬅ fd p. 共140兲
can be written as 2m

冕
The entropy per unit volume is
nq4 1 1 d 3k
␯⬇
m2⑀20 v3 k3 共2␲兲3
.

The integral over wave numbers is logarithmically diver-

共134兲
s⬅− 冕 f ln共f兲d3p. 共141兲

gent without a cutoff at large or at small wave numbers, This definition of the kinetic entropy is shown below to

冕 冉 冊
give results that are in agreement with the thermody-
d 3k kmax namic entropy
ln ⌳ ⬅ = ln . 共135兲
4␲k2 kmin The Maxwellian distribution fM has the maximum en-
tropy per unit volume s, with a fixed number of particles
The cutoff at large wave numbers kmax is given by the
per unit volume n, momentum per unit volume nmuជ ,
distance of closest possible approach of two charged par-
ticles, bmin = 1 / kmax. As shown in the discussion of and energy per unit volume, ⑀ = 21 nmu2 + 23 nT, which de-
bremsstrahlung, Sec. VI.A, bmin is determined by quan- fines the temperature T,
tum effects at fusion temperatures, bmin = ប / 共mv兲. The
cutoff at small wave numbers is given by the Debye
length, kmin = 1 / ␭D. The Debye length is ␭D
fM共xជ ,pជ 兲 =
n
共2␲T/m兲3/2
exp 冉
−
共pជ − muជ 兲2
2mT
冊. 共142兲

⬅ 冑⑀0T / 共nq2兲. The Coulomb logarithm is ln⌳ ⬇ 17 in
laboratory plasmas.
The Debye length, ␭D ⬅ 冑⑀0T / 共nq2兲, is the shielding
distance for the electrostatic potential of a charge. What
is meant by this? Let ⌽ be the potential due to a charge
Q that is placed in a plasma. The electric force −qnⵜ ជ⌽
on a species with charge q and number density n is bal-
ជ 共nT兲. In
anced by the pressure force of that species ⵜ
equilibrium the temperature is constant, so the equilib-
rium density is n = n⬁ exp共−q⌽ / T兲. The Poisson equation
That is, one takes s + ␭1n + ␭ជ 2 · nmuជ + ␭3⑀ to its extreme by
considering variations in the distribution function ␦f
while treating the ␭’s as constants. When the ␭’s, which
are called Lagrange multipliers, are chosen to obtain the
correct density, momentum, and energy, the Maxwellian
is obtained. To be consistent with thermodynamics, a
collision operator must cause the entropy to increase
except when the distribution function is a local Maxwell-
ian.
A collision operator, C共f兲 = −共⳵ / ⳵pជ 兲 · F, that is simple
but obeys the four conditions has the phase-space flux

再 冎
for the potential is then
Q qn⬁ −q⌽/T ជ =−J
F
␯
· 共pជ − muជ 兲f + mT
⳵f
. 共143兲
ⵜ ⌽ = − ␦共xជ 兲 −
2
共e − 1兲. 共136兲 2 ⳵pជ
⑀0 ⑀0
The collision frequency J ␯ in this simple collision opera-
Far from the charge, q⌽ / T Ⰶ 1, and Poisson’s equation tor is momentum independent. The correct collision op-
becomes ⵜ2⌽ = ⌽ / ␭D 2
. Consequently, for r Ⰷ ␭D, the elec- erator is considerably more complicated and in particu-
tric potential of a charge in a plasma is lar the collision frequency J ␯ is dependent on the energy

⌽=
Q
exp −冉 冊 r
␭D
. 共137兲
of the particles being scattered 关Eq. 共134兲兴.
The simplified collision operator of Eq. 共143兲 is useful
for illustrating the diffusive nature of collisions in plas-
⑀0 r mas. A distribution function that develops complicated
Conservation laws place three conditions on the colli- structures in momentum space can be smoothed arbi-
sion operator C共f兲, and the thermodynamic law of en- trarily quickly. More precisely, if the distribution func-
tropy increase places a fourth condition. The discussion tion changes over a range of momenta ⌬p then that
of these conditions is simpler using phase-space coordi- change is smoothed out at a rate of order ␯共p / ⌬p兲2.
nates in which the momentum is pជ = mvជ rather than the The change in the entropy density s共xជ , t兲 at each point
ជ , of Hamiltonian me- in a plasma is due to an entropy flux F ជ and the creation
canonical momentum, pជ = mvជ + qA s
chanics, so these are the phase-space coordinates that of entropy by collisions ṡc. A differentiation of the en-
will be used. The three conservation laws are for the tropy density s共xជ , t兲, Eq. 共141兲, with the use of df / dt
particle, = C共f兲 and Eq. 共128兲, implies

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1110 Allen H. Boozer: Physics of magnetically confined plasmas

⳵s ជ ជ Equation 共147兲 coupled with Eq. 共146兲 implies the dis-

+ ⵜ · Fs = ṡc . 共144兲 tribution functions must be close to local Maxwellians
⳵t
on each pressure surface to obtain the confinement
ជ ⬅ −兰vជ 共ln f兲fd3p where vជ = pជ / m, and
The entropy flux is F needed for fusion ignition. The required energy confine-
s
the rate of entropy production by collisions is ment time in a fusion power plant is about 102 ion colli-
sion times and 104 electron collision times. The mainte-

ṡc ⬅ − 冕 ln共f兲C共f兲d3p. 共145兲

The definition of entropy per unit volume s共xជ , t兲, Eq.
共141兲, is consistent with the thermodynamic entropy if
and only if it satisfies three conditions.
共1兲 The entropy, 兰sd3x, must be additive. That is, en-
tropy in a region that consists of two parts must be
the sum of the two entropies, a condition that s sat-
isfies.
共2兲 The entropy of an isolated system cannot decrease.
This condition implies ṡc 艌 0 and is a constraint on
the collision operator. If the distribution function is
written in the form f = fM exp共f̂兲 with fm defined so its
n共xជ 兲, uជ 共xជ 兲, and ⑀共xជ 兲 are the same as those of f, then
Eq. 共143兲 for simplified collisional flux gives a posi-
tive entropy production as long as J ␯ is a positive
matrix,
ṡc =
mT
2
冕 ⳵f̂
⳵pជ
·J
⳵f̂
␯ · fd3p.
⳵pជ
共146兲
共3兲 The change in the energy due to a transfer of heat is
T␦S.
By taking the time derivative of the energy per unit vol-
ume ⑀共xជ , t兲 ⬅ 兰共p2 / 2m兲fd3p, one can show, using Eq.
共128兲 and df / dt = C共f兲 to eliminate ⳵f / ⳵t, that the energy
nance of near Maxwellians requires the trajectories of
charged particles to remain close to constant-pressure
surfaces during the time between collisions.
Thermodynamics relates the collisional entropy pro-
duction ṡc and the plasma transport in an even more
complete form. Given ṡc, one can read off the transport
coefficients. For simplicity, assume that at each point in
the plasma the distribution function is well approxi-
mated by a stationary Maxwellian, so the flow velocity
uជ = 0. The textbook thermodynamic equation, dU = TdS
− pdV + ␮dN, relates thermodynamic quantities of an en-
tire system. Plasma studies use the energy ⑀ ⬅ U / V, the
entropy s ⬅ S / V, and the number of particles n ⬅ N / V
per unit volume. Substituting these definitions into the
thermodynamic equation yields 共d⑀ − Tds − ␮dn兲V = −共⑀
− Ts + p − ␮n兲dV. The thermodynamic properties of a
plasma are independent of the plasma volume V, so
both sides of this equation must be zero. The chemical
potential ␮ can be evaluated for a Maxwellian using ␮
= 共⑀ − Ts + p兲 / n, and one finds
␮ n
冉 冊
= c0 + ln 3/2 共148兲
T T
with c0 a constant. Consequently, a stationary Maxwell-
ian has the form
fM = cM exp 冉 ␮ − mv2/2
T
冊, 共149兲

flux is the sum of two parts: a part proportional to the where cM is a constant. The thermodynamic equation
fluid velocity uជ and a part equal to the heat flux F ជ d⑀ = Tds + ␮dn implies that the time rate of change of the
h
ជ is entropy density is ⳵s / ⳵t = 共1 / T兲⳵⑀ / ⳵t − 共␮ / T兲⳵n / ⳵t. Now
⬅ 兰 21 m共vជ − uជ 兲3fd3p, where vជ = pជ / m. The entropy flux F
also the sum of two parts when the plasma is near ther-
s
ជ · ⌫ជ with ⌫ជ the diffusive particle flux, and
⳵n / ⳵t = −ⵜ
modynamic equilibrium, which means f = fM exp共f̂兲 with
ជ ·F
⳵⑀ / ⳵t = −ⵜ ជ with F
ជ the heat flux. Therefore
h h

兩f̂ 兩 Ⰶ 1. One part of the entropy flux is proportional to

ជ / T. In a con-
the fluid velocity uជ and the other part is F h
⳵s
⳵t
ជ · ⵜជ ␮ + F
=−⌫
T
h
T T
冉ជ − ␮ ⌫ជ .
ជ · ⵜជ 1 − ⵜជ · 1 F
h
T
冊 共150兲
tinuum system the temperature is well defined only near
thermodynamic equilibrium, and in that limit the third The divergence term on the right-hand side of Eq. 共150兲
condition on the entropy is satisfied. is the divergence of the entropy flux, but the other terms
Thermodynamics gives a simple but rigorous expres- are due to the irreversible production of entropy, which
means production by collisions. Since the chemical po-
sion for the heating power P required to maintain a sta-
tential and the temperature depend only on the ␺t coor-
tionary, uជ = 0, steady-state plasma,
dinate, the collisional entropy production is

P共␺t兲 = T共␺t兲 冕 ṡcd3x, 共147兲 ṡc = − ⌫

d␮/T
d␺t
+ Fh
d1/T
d␺t
, 共151兲

where the volume integral is over a region enclosed by a ជ ·ⵜ

ជ ␺ and F ⬅ F ជ ·ⵜជ ␺ . The quantities
with ⌫ ⬅ ⌫
flux surface ␺t, and P共␺t兲 is the total heating power in t h h t
−d共␮ / T兲 / d␺t and d共1 / T兲 / d␺t are called thermodynamic
that region. To prove this equation, note that the power
forces, and Fh and ⌫ are the conjugate fluxes. Near ther-
per unit volume that must be added to balance the heat
modynamic equilibrium, the fluxes are proportional to
ជ ·F
flux is ⑀˙ h = ⵜ ជ , while the entropy created by collisions
h the forces, ⌫ = −Dnd共␮ / T兲 / d␺t − Dcd共1 / T兲 / d␺t and Fh
must satisfy ṡ = ⵜ ជ · 共F
ជ / T兲. = Dcd共␮ / T兲 / d␺t + DTd共1 / T兲 / d␺t. The cross terms, the
c h

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas
terms proportional to Dc, have the same coefficient due
to the symmetry of the collision operator, or more gen-
erally due to Onsager symmetry 共Onsager, 1931兲. Conse-
quently
ṡc = Dn 冉 冊
d␮/T
d␺t
2
+ 2Dc
d␮/T d1/T
d␺t d␺t
+ DT 冉 冊
d1/T
d␺t
2
. 共152兲
By calculating ṡc, one can obtain all three independent
transport coefficients.
Much freedom exists in the choice of variables ␰j that
are used to describe the distribution function. The left-
hand side of the kinetic equation df / dt = C共f兲 becomes
df / dt = ⳵f / ⳵t + 兺j共⳵f / ⳵␰j兲共d␰j / dt兲. All that is needed is
knowledge of how each of the variables changes along
the trajectory of a particle, d␰j / dt. For example, f is a
solution of the collisionless kinetic equation df / dt = 0,
the Vlasov equation, if and only if its variables are con-
stants of the motion of the particles, d␰j / dt = 0. In a sys-
tem that is independent of time, f共H兲 is a solution to the
Vlasov equation where H is the Hamiltonian or energy
of a particle.
The calculation of density perturbations using kinetic
theory is subtle when the unperturbed distribution func-
tion is given as a function of a constant of the motion
such as the Hamiltonian, f0共H兲. Consider a collisionless
plasma that is perturbed by a change in the electric po-
tential, ␦⌽. The distribution function in the presence of
the perturbation is f = f0共H兲 + ␦f. The density perturba-
tion that is caused by the perturbation ␦⌽ is not 兰␦fd3v,
but ␦n = 兰␦fd3v + 兰q␦⌽共df0 / dH兲d3v. That is, the total
change in the distribution function is
df0
⌬f ⬅ ␦f + q␦⌽ . 共153兲
dH
To understand the reason, suppose the plasma is per-
turbed by an electric potential ␦⌽ = ⌽̃ sin共kx − ␻t兲. The
kinetic equation of a collisionless plasma, df / dt = 0, can
be solved in two ways, and a subscript of v or H will be
placed on ␦f to indicate which set of variables, 共x , v , t兲 or
共x , H , t兲, is used to find the solution. First, using variables
共x , v , t兲 one can write d␦fv / dt = ⳵␦fv / ⳵t + v⳵␦fv / ⳵x
= −共qE / m兲df0 / dv since dv / dt = qE / m. This equation im-
plies
k q␦⌽ df0
␦fv = . 共154兲
kv − ␻ m dv
Second, using variables 共x , H , t兲 one can write d␦fH / dt
= ⳵␦fH / ⳵t + v⳵␦fH / ⳵x = −共df0 / dH兲dH / dt where the change
in the energy of a particle is dH / dt = q⳵␦⌽ / ⳵t. The solu-
tion is
␻ df0
␦fH = q␦⌽ . 共155兲
kv − ␻ dH
These two solutions are not equal, even after making the
substitution df0 / dv = 共df0 / dH兲mv. Indeed, ␦fv = ␦fH
+ q␦⌽df0 / dH. The resolution of this paradox is that the
unperturbed Hamiltonian H0 goes to H = H0 + q␦⌽. A
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1111
first-order Taylor expansion implies f0共H兲 = f0共H0兲
+ q␦⌽df0 / dH0. The term q␦⌽df0 / dH is called the adia-
batic part of the plasma response.
C. Landau damping and quasilinear diffusion
• A collisionless, but unidirectional, transfer of energy
can occur between a wave and the particles that form
a plasma. The sign of this transfer, which is called
Landau damping, is determined by the sign of the
velocity derivative of the distribution function at the
place where the particle velocity equals the phase
velocity, ␻ / k, of the wave.
• If the energy density of waves has a continuous de-
pendence on their phase velocity, ␻ / k, the waves
cause a collisionless diffusion of the particles whose
velocities resonate with the phase velocities of the
waves. This is called quasilinear diffusion.
In fusion, and many other plasmas of interest, the col-
lision frequency is small compared to the time it takes a
charged particle to cross a distance comparable to the
plasma size. Since collisional effects are weak, it is natu-
ral to study the Vlasov equation, df / dt = 0. This hyper-
bolic partial differential equation has the particle trajec-
tories as its single characteristic. In other words, the
distribution function is carried along by the particle mo-
tion, so any function of the constants of motion is a so-
lution to the Vlasov equation.
Since the Vlasov equation, df / dt = 0, contains only in-
formation about the particle trajectories, which are time
reversible, it is natural to assume that its solutions are
themselves time reversible. This turns out to be false in
practice, with the sense of time determined by the na-
ture of the initial conditions. Solutions to the Vlasov
equation can evolve from smooth initial conditions into
functions with arbitrarily complicated structures. Plasma
observations have finite resolution, and the averaging
that is implicit in the observation process leads to irre-
versibility. These time-irreversible collisionless effects
are important in plasma physics for the damping of ex-
ternally driven waves, which gives plasma heating, and
for amplifying plasma waves, which gives instabilities.
The two collisionless phenomena that will be discussed
are Landau damping 共Landau, 1946兲 and quasilinear dif-
fusion 共Vedenov, Velikhov, and Sagdeev, 1961; Drum-
mond and Pines, 1962兲.
Landau damping is a collisionless, but unidirectional,
transfer of energy between an electromagnetic wave and
a plasma. It is central to the theory of plasma heating by
waves and the destabilization of electromagnetic pertur-
bations by non-Maxwellian distributions of the plasma
species.
To understand Landau damping, assume that at t = 0
the distribution function of a plasma species is indepen-
dent of position, f0共v兲, but a weak electric field is intro-
duced that for t ⬎ 0 has the form
1112 Allen H. Boozer: Physics of magnetically confined plasmas

E = Ek cos共kx − ␻t兲.
The problem has only one spatial coordinate x and one
共156兲
冉冕 f⬘共v兲
␻ − kv
dv 冊 imag
=−i
␲
兩k兩
冉冊
f⬘
␻
k
. 共165兲

velocity coordinate, which is vx, but will be denoted by v Landau damping is closely related to a second phe-
for simplicity. The perturbed Vlasov equation, df / dt = 0, nomenon, quasilinear diffusion, which is important for
with f = f0 + ␦f, is understanding the interaction of waves with plasmas and
the effect of short-wavelength perturbations on the
⳵␦f ⳵␦f q ⳵f0
+v + E = 0. 共157兲 plasma. If one identifies the distribution function f0 of
⳵t ⳵x m ⳵v the Landau damping discussion with the spatially aver-
The solution for the perturbed distribution function ␦f aged distribution function, f0 ⬅ 具f典, then a spatial average
that obeys the initial condition that ␦f = 0 is of the Vlasov equation, df / dt = 0 关Eq. 共128兲兴 with p
= mv, gives

冓 冔
qEk ⳵f0 sin共kx − ␻t兲 − sin共kx − kvt兲
␦f = , 共158兲 ⳵f0 ⳵ q
m ⳵v ␻ − kv =− E␦f . 共166兲
⳵t ⳵v m
which is nonsingular though growing at v = ␻ / k,
Equation 共161兲 gives an expression for 具qE␦f典, which is
⳵f0 qE proportional to ⳵f0 / ⳵v. Therefore
共␦f兲v=␻/k = − t. 共159兲
⳵v m
⳵f0 ⳵ ⳵f0
Particles moving at the phase velocity of the perturba- = D共v兲 , 共167兲
⳵t ⳵v ⳵v
tion, v = ␻ / k, see a time-independent electric field and
are accelerated by it. A spatial average is defined by the where the velocity diffusion coefficient is
limit as L goes to infinity of 具g共x兲典 ⬅ 共1 / 2L兲兰−L
+L
gdx. The q2Ek2 sin关共␻ − kv兲t兴
spatially averaged power going to the plasma is D= . 共168兲
2m2 ␻ − kv

具P典 = 冕 v具qE␦f典dv. 共160兲 This expression for the velocity-space diffusion is not
useful unless there is a spectrum of perturbations with
different k’s. The spatially averaged energy density of a
Using the trigonometric identity sin共kx − kvt兲 = sin共kx
− ␻t兲cos关共␻ − kv兲t兴 + cos共kx − ␻t兲sin关共␻ − kv兲t兴, one has single wave is 共⑀0 / 2兲具E2典 = 共⑀0 / 4兲Ek2 . If 兰E共␻ / k兲dk is the
energy per unit volume of a spectrum of waves, one can
q2Ek2 ⳵f0 sin关共␻ − kv兲t兴 integrate Eq. 共168兲 over k using the integration formula
具qE␦f典 = − . 共161兲 of Eq. 共163兲 to obtain
2m ⳵v ␻ − kv
The velocity integral that must be performed to ob- 2␲q2 E共v兲
D= . 共169兲
tain the power, Eqs. 共160兲 and 共161兲, has the form ⑀0m2 兩v兩

冕 F共v兲
sin关共␻ − kv兲t兴
␻ − kv
dv = 冕冋冉 F
␻
k
1+
␰
␻t
冊册 sin ␰ d␰
␰ 兩k兩
,
The quantity ␻共k兲 / k is the phase velocity of the electric
perturbation, so a perturbation diffuses particles that
have a velocity equal to the phase velocity.
共162兲
Quasilinear diffusion is closely related to the stochas-
where ␰ ⬅ 共kv − ␻兲t. The integral of sin共␰兲 / ␰ from minus ticity of magnetic-field lines that was discussed in Sec.
to plus infinity is ␲, so when ␻t → ⬁ III.A. In principle, the equations of mechanics are re-

冕
versible. However, when stochastic, trajectories that
sin关共␻ − kv兲t兴 F共␻/k兲 have an infinitesimal initial separation increase that
lim F共v兲 dv = ␲ . 共163兲
␻t→⬁ ␻ − kv 兩k兩 separation exponentially with distance along the trajec-
tories. If the trajectories are resolved to finite accuracy,
The spatially averaged power is then then the information is quickly lost that would be

具P典 = − ␲ v
2m兩k兩 ⳵v
冉 冊
q2Ek2 ⳵f0
v=␻/k
. 共164兲
Power is transferred to the plasma 共positive power兲 if
there are fewer particles at high energy than low and to
the electric perturbation 共an incipient instability兲 if there
are more particles at high energy than low.
One can simplify some of the analysis of Landau
damping by writing it in the form ␦f = f̃ exp关i共kx − ␻t兲兴.
The critical step in Landau damping is then the determi-
nation of the imaginary part of what appears to be a real
integral. The theory of Laplace transforms implies the
imaginary part of the integral, the Landau integral, is
needed to launch the time-reversed trajectories.
The spatial averaging that is involved in deriving the
quasilinear diffusion coefficient, destroys information
and allows an entropylike quantity s0 ⬅ −兰f0 ln f0d3p to
increase. However, s0 is not the entropy density, nor
even the spatial average of the entropy density, s
⬅ −兰f ln fd3p. The spatially averaged entropy density 具s典
cannot change if f is a solution to the Vlasov equation,
df / dt = 0 关Eq. 共144兲兴. Let f = f0 exp共f̂兲, then 具exp共f̂兲典 = 1
since 具f典 = f0. Assuming f̂ is small, the relation between
具s典 and s0 is 具s典 = s0 − 21 兰 具f̂2典f0d3p. Consequently s0 must
increase if the magnitude of f̂ does, to avoid a change in

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas 1113

具s典. Actually, the complicated velocity-space structure

that f̂ develops, coupled with the diffusive nature of
plasma collisions, means that f̂ may not reach a large
amplitude even when the fractional change in the
volume-averaged entropy density is of order unity.
Quasilinear diffusion can represent the heating of a
plasma by a fluctuating electric field even in steady state.
Of course for a steady state in the presence of heating,
there must be a cooling term in the kinetic equation,
such as ⳵f0 / ⳵t = ⳵共␯cvf0兲 / ⳵v, where 2␯c共v兲 is the cooling
rate. For this cooling term, the volume-averaged distri-
bution function relaxes to ⳵ ln f0 / ⳵共v2 / 2兲 = −␯c / D, which
has the shape of a Maxwellian with an effective tem-
perature Teff / m ⬅ D / ␯c in velocity regions where D / ␯c is
independent of velocity.

D. Drift kinetic theory

• The single characteristic of the Vlasov operator,
df / dt 关Eq. 共127兲兴, is the trajectories of the particles.
Consequently, one can use approximations to the
particle trajectories to simplify kinetic calculations.
• Charged particles make a circular gyration about
magnetic-field lines. When the radius of gyration is
small compared to the spatial variation of the electric
and magnetic fields, the magnetic moment, ␮
2
= mv⬜ / 2B, is a constant of the motion.
• When the gyration radius is small, the particle trajec-
tory can be accurately approximated by tracking the
center of the circle, xជ g, about which the particle gy-
rates, the guiding center. The guiding center drifts at
a velocity vជ g given by Eq. 共181兲. The equation of mo-
tion is first order, dxជ g / dt = vជ g, and not the usual
second-order equation, md2xជ / dt2, equal to a force.
• The drift velocity of the gyrocenters is also given by
a Hamiltonian, Eq. 共195兲, which has only four ca-
nonical variables 共␪ , ␸ , p␪ , p␸兲, Eqs. 共199兲 and 共200兲,
instead of the six canonical variables of the full tra-
jectories. The form of the Hamiltonian for the drift
motion demonstrates that variation in the magnetic-
field strength on the magnetic surfaces is the primary FIG. 10. 共Color兲 Orbit of a charged particle and its guiding
determinant of the confinement properties of the center: 共a兲 A charged particle gyrates in a circle about a
particle trajectories. straight magnetic-field line while moving along the line; 共b兲 the
instantaneous center of this circle is the guiding center xជ g, and
• Particles with a sufficiently small ratio of their veloc- the difference between the position of the particle xជ and the
ity parallel to the magnetic field, v储, to their total guiding center is the vector gyroradius ␳ជ ; 共c兲 a uniform electric
velocity v are trapped between maxima of the field perpendicular to the magnetic field causes the particle
magnetic-field strength along a magnetic-field line. velocity, and hence its gyroradius, to be slightly larger on one
These trapped particles drift from one field line to side of its gyro-orbit than the other, causing a drift across the
another conserving their action, J = 养 mv储dᐉ. magnetic-field lines.
The Vlasov operator df / dt in the kinetic equation is
determined by the trajectories of the particles, and the culation obscures fundamental properties of the
determination of these trajectories is a major difficulty trajectories. For example, the confinement properties of
in solving the kinetic equation. Charged particles in a particle trajectories are essentially determined by the
magnetic field move in circles with a gyration frequency variation of the magnetic-field strength on the constant-
⍀ ⬅ qB / m, Fig. 10, which makes a direct calculation in- pressure surfaces.
efficient when the circles are small, as they are in a fu- This section derives the asymptotic expressions for the
sion plasma. In addition to being inefficient, a direct cal- particle trajectories 共Alfvén, 1940兲 as well as the associ-

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1114 Allen H. Boozer: Physics of magnetically confined plasmas

ated kinetic theory, which holds when the circles made

by charged-particle orbits in a magnetic field are smaller
than the scale of field variations. The basic results are as
follows: 共1兲 the magnetic moment,
2
mv⬜
␮= , 共170兲
2B
is a constant of the motion with v⬜ the magnitude of the
velocity components perpendicular to B ជ . The invariance
properties of ␮, which is an adiabatic invariant, are dis-
cussed in Sec. VI.D.2. 共2兲 The center of the circle, xជ g,
about which a particle gyrates 共Fig. 10兲, drifts at a veloc-
ity vជ g = dxជ g / dt. This velocity is called the guiding-center
or drift velocity, and dxជ g / dt = vជ g gives a first-order ordi-
nary differential equation for the center’s trajectories.
The guiding-center velocity is given by Eq. 共181兲 as well
as by a drift Hamiltonian, Eq. 共195兲. The Hamiltonian of
the drift motion, Sec. VI.D.3, has only four canonical
variables 共␪ , ␸ , p␪ , p␸兲—the poloidal and toroidal angles
of Boozer coordinates, Eq. 共58兲 and their conjugate mo-
menta 关Eqs. 共199兲 and 共200兲; Boozer, 1984b兴. This Hamil-
tonian formulation was utilized in a code by White and
Chance 共1984兲. Their code has been the basis of numer-
ous investigations of phenomena that depend on particle
drift. In addition to the canonical formulation of the
drift equations, which is emphasized here, Littlejohn
共1981兲 has given an important noncanonical, though FIG. 11. 共Color兲 Magnetization drift: The magnetic field is out
Hamiltonian, treatment based on Lie theory. of the figure. In the presence of a density gradient, the number
The velocity-space coordinates of the guiding-center of particles moving in one direction perpendicular to both the
motion, which simplify the kinetic equation, df / dt magnetic field and density gradient differs from the number of
particles moving in the opposite direction. This leads to a
= C共f兲, are the Hamiltonian H, which is the energy, and
divergence-free flow of the particles in the B ជ ⫻ⵜ ជ n direction
the magnetic moment ␮, where H = mv2储 / 2 + ␮B + q⌽. An
even if all the particles are moving in perfect circles. This flow
average over the phase angle ␽ of the circular gyromo-
is called the magnetization drift.
tion of the particles gives the drift equations. The
velocity-space volume element in cylindrical velocity-
space coordinates is d3v = v⬜dv⬜d␽dv储. Using The second term is called the magnetization current and
共H , ␮ , ␽ , xជ 兲 as coordinates, dH = mv储dv储, so arises, as does a magnetization current in a solid, from
B dHd␮d␽ the magnetic moments of the particles that constitute
d 3v =
m2
兺± 兩v储兩
, 共171兲 the medium 共Jackson, 1999; see Fig. 11兲. The actual mag-
netic moment produced by the circular gyromotion is
where the 兺± means that the sum is taken over the sign
of velocity along the magnetic field, v储. In drift kinetic
theory, one can integrate over the gyrophase ␽, which is ជ = − ␮b̂.
m 共174兲
equivalent to replacing d␽ in Eq. 共171兲 by 2␲. The drift
kinetic equation is obtained by calculating
df共H , ␮ , xជ , t兲 / dt and setting dxជ / dt = vជ g, the drift velocity The definition of the magnetic moment of a current dis-
of the guiding centers, ជ ⬅ 21 兰 xជ ⫻ ជjd3x, which for a charged particle
tribution is m
df ⳵f dH ⳵f moving in a gyro-orbit is m ជ = 共q / 2兲␳ជ ⫻ vជ with ␳ជ the vector
= + vជ g · ⵜជ f + = C共f兲, 共172兲 gyroradius 关Eq. 共177兲兴.
dt ⳵t dt ⳵H
where the energy exchange between the particles and
the electromagnetic fields, dH / dt, is given by Eq. 共196兲.
The use of drift kinetic theory to calculate flows re- 1. Alfvén’s guiding-center velocity
quires care. The current density is not q 兰 vជ gfd3v, as one Alfvén 共1940兲 derived an expression for the guiding-
would naively expect, but center or drift velocity that follows from the expression
ជj = q 冕 冉冕 冊
vជ gfd3v − ⵜជ ⫻ b̂ ␮fd3v . 共173兲
for the trajectories of a charged particle in given electric
and magnetic fields,

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas 1115

dvជ ជ + vជ ⫻ B
ជ 兲. ␬ជ ⬅ b̂ · ⵜជ b̂, 共182兲
m = q共E 共175兲
dt
ជ / 兩B
with b̂ ⬅ B ជ 兩. The curvature of a pure toroidal field
The position of the particle is given by dxជ / dt = vជ . In a
uniform magnetic field with no electric field, a particle ជ
B = 共␮ G / 2␲R兲␸ˆ is ␬ជ = −R̂ / R since d␸ˆ / d␸ = −R̂. The term
0
moves in a circle at the frequency involving ⳵b̂ / ⳵t in the guiding-center velocity is usually
qB dropped, since it is small if the time scale T over which b̂
⍀⬅ , 共176兲 varies is long compared to the characteristic time for a
m
particle to cross the system R / v.
with the circle having a radius 共Fig. 10兲 In time-independent magnetic and electric fields,
Alvén’s expression for the guiding-center velocity can be
b̂ ⫻ vជ written in an alternative form, which simplifies the
␳ជ ⬅ , 共177兲 theory of particle confinement. One can easily show
⍀
共Morozov and Solov’ev, 1966兲 that
where b̂ ⬅ B ជ / 兩B
ជ 兩. The center of the circle xជ , which is
g
called the guiding center, moves with a velocity v储 v储 ជ ជ + ␳储B
ជ兲
vជ g = ⵜ ⫻ 共A 共183兲
⬅ b̂ · vជ parallel to the magnetic field. The position of the B
charged particle is 共Fig. 10兲
has the same drift across the magnetic field as Eq. 共181兲.
xជ = xជ g + ␳ជ . 共178兲 The parallel gyroradius is defined by
The addition of a constant electric field perpendicular
to the magnetic field causes the guiding centers to drift v储
␳储共H, ␮,xជ 兲 ⬅ , 共184兲
ជ ⫻B
in the E ជ direction 共Fig. 10兲. Let ⍀

ជ ⫻B
E ជ
vជ E⫻B ⬅ , 共179兲
B2
which is independent of position and time if E ជ and B
ជ ment ␮ constant. The trajectories are obtained by solv-
ជ ·B
ជ = 0, Eq. 共175兲 can be written as m共d / dt兲共v
are. If E
− vជ E⫻B兲 = q共v − vជ E⫻B兲 ⫻ B ជ . The motion of a charged par-
ticle in a constant electric field that is perpendicular to
the magnetic field lines is identical to the motion without
the electric field in a frame of reference moving with the
velocity vជ = Eជ ⫻B ជ / B 2.
The properties of the trajectories become far more
complicated when the particles move in electric and
magnetic fields that depend on position and time. How-
ever, Eq. 共178兲 can be viewed as defining the guiding
center of a particle in arbitrary magnetic and electric
fields, and the exact motion of the guiding center,
where the energy H = 21 mv2储 + ␮B + q⌽ and the magnetic
moment ␮ are treated as constants. That is, the curl of v储
is calculated holding the energy H and the magnetic mo-
ing first-order equations for the guiding-center position,
dxជ g / dt = vជ g. The only difficulty in the integration is at
turning points of the parallel motion, where v储 passes
through zero and changes its sign. This difficulty can be
removed by the Hamiltonian formulation; see Sec.
VI.D.3.
The terms in Alfvén’s expression for the drift of the
guiding center can be understood by analogy to the E
⫻ B drift 关Eq. 共179兲兴. If a force Fជ is applied to a charged
particle gyrating in a magnetic field, then the same argu-
ment that led to the E ⫻ B drift yields a drift velocity
vជ F⫻B ⬅ Fជ ⫻ B ជ / 共qB2兲. If the magnetic-field lines have a
nonzero curvature ␬ជ , the centrifugal force, F ជ = −mv2储 ␬ជ ,

冉冊
c
dxជ g ជ ⫻ b̂
E d b̂ gives the curvature drift. If the magnetic-field strength
= v储b̂ + + vជ ⫻ , 共180兲 varies, the quantity ␮B acts like a potential energy, giv-
dt B dt ⍀
ing the ⵜB force, F ជ = −␮ⵜ ជ B, and an associated drift.
ⵜB
is obtained using Eqs. 共175兲, 共177兲, and 共178兲. The only part of the derivation of the guiding-center
When the spatial variation of the electric and mag-
velocity, Eq. 共181兲, that is not obvious is the part of vជ g
netic fields is on a scale much longer than the gyroradius
perpendicular to B ជ that follows from the last term in Eq.
兩␳ជ 兩, particles can be tracked by following the velocity of
the guiding center averaged over the rapid gyromotion, 共180兲. This part of vជ g is given by

再 冉 冊冎 冉冊 冉冊
vជ g ⬅ 具dxជ g / dt典. The expression for the guiding-center ve-
locity is 共Alfvén, 1940兲 d b̂ d b̂ d b̂
b̂ ⫻ vជ ⫻ = vជ b̂ · . 共185兲

冉 冊
− v储
2 ជ ជ ⫻B ជ dt ⍀ dt ⍀ dt ⍀
b̂ ⳵b̂
v⬜ ⵜB E
vជ g = v储b̂ + ⫻ v2储 ␬ជ + v储 + + 2 ,
⍀ ⳵t 2 B B ជ + vជ · ⵜ
One can let d / dt = ⳵ / ⳵t + v储b̂ · ⵜ ជ for b̂ ⬅ B
ជ / 兩B
ជ 兩 and
⬜ ⬜
共181兲 ⍀ ⬅ qB / m have no velocity dependence. Since b̂ is a unit
where the curvature of the magnetic-field lines is vector, b̂ · db̂ / dt = 0 and

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1116 Allen H. Boozer: Physics of magnetically confined plasmas

再 冉 冊冎
b̂ ⫻ vជ ⫻
d b̂
dt ⍀
= vជ ⬜ 冉冊
d 1
dt ⍀
−
v储 db̂
⍀ dt
. 共186兲
ជ,
pជ = mvជ + qA

where vជ ⬅ xជ̇ . The gradient of the Lagrangian is ⳵L / ⳵xជ

共191兲

The direction of the velocity perpendicular to the mag- ជ 共vជ · A

= qⵜ ជ 兲 − qⵜ ជ ⌽. A vector identity implies ⵜ ជ 共vជ · A
ជ 兲 = vជ
netic field changes over a gyroperiod, so 具vជ ⬜典 = 0 in low- ជ ⫻A ជ 兲 + vជ · ⵜ
ជAជ , where vជ is not differentiated since it is
est order, and ⫻ 共ⵜ

冓 冔
an independent coordinate in a Lagrangian analysis.
d 1 2
v⬜ 1 Now, dpជ / dt = mdvជ / dt + q⳵A ជ / ⳵t + vជ · ⵜ
ជAជ , so pជ̇ = ⳵L / ⳵xជ is
vជ ⬜ = ⵜជ ⬜ . 共187兲
dt ⍀ 2 ⍀ equivalent to mdvជ / dt = q共E ជ + vជ ⫻ B
ជ 兲.
The gyrophase average is thus The beauty of the Lagrangian approach is the ease of

冓 冔
finding the equations of motion in arbitrary 共␺t , ␪ , ␸兲 co-
v储 db̂ v2储 v储 ⳵b̂ ordinates, xជ 共␺t , ␪ , ␸ , t兲. If L is independent of a coordi-
= b̂ · ⵜជ b̂ + . 共188兲 nate ␸, as it is in axisymmetric plasmas, then p␸ is a
⍀ dt ⍀ ⍀ ⳵t
constant of the motion, ṗ␸ = ⳵L / ⳵␸ = 0. The ␸ component
of the canonical momentum is

2. Magnetic-moment conservation ⳵xជ ជ · ⳵xជ .

p␸ = mvជ · + qA 共192兲
The integration of the guiding-center velocity, Eq. ⳵␸ ⳵␸
共181兲, to obtain xជ g is greatly simplified by the existence This equation is obtained from p␸ ⬅ ⳵L / ⳵␸˙ using dxជ / dt
of an adiabatic invariant, the magnetic moment, which
has the approximate form ␮ = mv⬜ 2
/ 2B. = ⳵xជ / ⳵t + 共⳵xជ / ⳵␺t兲␺˙ t + ¯, which implies ⳵共dxជ / dt兲 / ⳵␸˙
An important general principle of Hamiltonian me- = ⳵xជ / ⳵␸. The poloidal flux is ␺p = −2␲A ជ · 共⳵xជ / ⳵␸兲 using
chanics, which is rarely taught in mechanics courses, is Eq. 共6兲 for A ជ with the gauge g = 0. The constancy of p
␸
that if the parameters that define a periodic motion of a ensures the confinement of particles of sufficiently small
particle change sufficiently slowly, then the action of the velocity. Particles cannot cross the field lines by a dis-
periodic motion is conserved. If pជ ⍀ are oscillating com- tance greater than their gyroradius in the poloidal field,
ponents of the canonical momentum of the particle ជ ⬅ⵜ ជ 共␸ / 2␲兲 ⫻ ⵜ ជ ␺ , alone.
B
Hamiltonian and xជ ⍀ are the conjugate canonical coordi- p p
The definition of the adiabatic invariant that is known
nates, then the action is given by an integration over the as the magnetic moment is the integral
periodic motion, 养pជ ⍀ · 共dxជ ⍀ / dt兲dt. Such invariants are
called adiabatic invariants, and the magnetic moment ␮
is an example. The conservation properties of adiabatic
invariants is a complicated subject 共Kruskal, 1962兲,
␮⬅
q 1
m 2␲
冖 pជ ⬜ ·
dxជ ⬜
dt
dt 共193兲

closely related to the existence of a magnetic surface over a gyroperiod where pជ is the canonical momentum,
共see Sec. III.A兲. The basic result is that if the parameters pជ ⬅ mvជ + qA ជ . The quantity m␮ / q is the action of the gy-
of a Hamiltonian are changed on a time scale T, which is romotion.
slow compared to the frequency ⍀ of a periodic motion, The approximate expression for the adiabatic invari-
then the action of that motion is conserved with expo- ant, ␮ = mv⬜ 2
/ 2B, which is correct to lowest order in the
nential accuracy. That is, the variation of the action is of gyroradius to system size, is obtained from Eq. 共193兲
order exp共−⍀T兲. ជ · 共dxជ / dt兲dt
Canonical momenta are defined using the Lagrangian using 养vជ ⬜ · 共dxជ ⬜ / dt兲dt = 共2␲ / ⍀兲v⬜ 2
and 养A ⬜
formulation of the equations of motion. The Lagrangian ជ
= − 养 A · dxជ ⬜. The minus sign arises because a positive
of a particle in electric and magnetic fields is charge moves about its circular orbit in a clockwise di-
rection, while the convention for a line integral is coun-
m 2 ជ − q⌽, ជ · dxជ = 兰Bជ · daជ = ␲B兩␳ជ 兩2. The
L共xជ̇ ,xជ 兲 = xជ̇ + qxជ̇ · A 共189兲 terclockwise. The integral 养A ⬜
2 approximate expression for the magnetic moment, ␮
2
ជ =ⵜ
ជ ⫻A
ជ , and E
ជ = −⳵A
ជ / ⳵t − ⵜ
ជ ⌽. The ca- = mv⬜ / 2B, follows.
where xជ̇ ⬅ dxជ / dt, B The magnetic moment is proportional to the magnetic
nonical momenta are defined by flux enclosed by the circular gyromotion, so magnetic-
⳵L moment conservation can be viewed as flux conserva-
pជ ⬅ . 共190兲 tion. The conservation of the magnetic moment is also
⳵xជ̇ equivalent to a fixed Landau level of the quantum
theory of particle motion in a magnetic field. That is, the
The time derivatives of the canonical momenta are pជ̇ adiabatic invariance of ␮ follows from the adiabatic ap-
= ⳵L / ⳵xជ . proximation of quantum mechanics. The lowest-order
The Lagrangian gives the standard equations of mo- conservation of ␮ can also be demonstrated by differen-
tion, Eq. 共175兲, in Cartesian coordinates. In these coor- tiating the expression for ␮ of Eq. 共170兲 with respect to
dinates, the canonical momentum, pជ = ⳵L / ⳵vជ , is time.

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas 1117

Just as magnetic surfaces can only be broken by reso-

nant perturbations, the invariant ␮ can only be broken if
there is a resonance between the time variation of the
dH
dt
=q
⳵⌽
⳵t
冉
− vជ ·
⳵Aជ
⳵t
. 冊 共198兲

gyromotion and the gyromotion itself. The Fourier The only subtle term in the derivation of Eq. 共196兲 is
transform of an analytic function 兰f共t兲 exp共−i⍀t兲dt ជ / ⳵t典. This term is calculated using the same
具qvជ ⬜ · ⳵A
⬃ exp共−⍀T兲, where T is the distance of the closest pole ជ · 共dxជ / dt兲dt in the magnetic
of f共t兲 from the real axis; see Sec. V.D.1. T is the charac- technique as that for 养A ⬜ ⬜
teristic time scale for variations. It is sometimes said that ជ / ⳵t典 = −共⍀ / 2兲共⳵B / ⳵t兲兩␳ជ 兩2.
moment. That is, 具vជ ⬜ · ⳵A
the magnetic moment is conserved to all orders in ⑀
⬅ 1 / ⍀t. What is meant is that if the function exp共−1 / ⑀兲 is
Taylor expanded in ⑀ about ⑀ = 0, then every term in the 3. The drift Hamiltonian
Taylor series is zero. The function exp共−1 / ⑀兲 is the most When the scale of the spatial and temporal variation
important example in physics of a function that is not of the magnetic and electric fields is long compared to
zero but has a Taylor series that is identically zero. It the gyroradius and the gyrofrequency of a particle, the
serves as a warning that an expansion in a small param- motion of the particle can be tracked using the drift
eter can be subtle. A complicated variation in the mag- Hamiltonian, which is the Hamiltonian for the guiding-
netic and electric fields across the field lines is irrelevant center motion. This Hamiltonian is the energy, Eq. 共195兲,
to ␮ conservation if the total time derivatives of these H共p␪ , p␸ , ␪ , ␸ , t兲 = 21 mv2储 + ␮B + q⌽. The canonical mo-
ជ , are small compared to the gy-
fields, d / dt = ⳵ / ⳵t + v储b̂ · ⵜ menta of the drift Hamiltonian are
rofrequency ⍀. The irrelevance of variations across the
␮ 0I q
magnetic field to the conservation of ␮ is important for p␪ = mv储 + ␺t , 共199兲
the validity of gyrokinetic theory, which is discussed in 2␲B 2␲
Sec. VI.G. and
The constancy of the magnetic moment implies that
the number of independent variables in a guiding-center ␮ 0G q
p␸ = mv储 − ␺p . 共200兲
calculation is four instead of the six required for the full 2␲B 2␲
particle motion. The four variables can be taken to be
The 共␺t , ␪ , ␸兲 coordinate system is the Boozer coordinate
the three components of the guiding-center position and
system, Eq. 共58兲, with the subscripts omitted to simplify
the energy. The energy, or Hamiltonian, of a charged
the notation.
particle is
To obtain the drift Hamiltonian and its canonical co-
ordinates, we start with a Lagrangian for the guiding-
1 center motion that was given by Taylor 共1964兲,
H ⬅ mv2 + q⌽. 共194兲
2
1 ជ − 共␮B + q⌽兲,
LT共xជ ,xជ̇ 兲 = m共xជ̇ · b̂兲2 + qxជ̇ · A 共201兲
2
Magnetic-moment conservation implies the energy can
also be written as where b̂ ⬅ Bជ / B. To lowest order in the ratio of gyrora-
dius to system size, its trajectories agree with those given
1 by Alfvén’s expression for the guiding-center motion
H = mv2储 + ␮B + q⌽. 共195兲
2 关Eq. 共181兲兴. The validity of the Taylor Lagrangian is
demonstrated by an explicit calculation. The canonical
As shown below, the gyrophase averaged change in the momentum pជ ⬅ ⳵L / ⳵xជ̇ is pជ = mv储b̂ + qAជ . The energy or
T
energy is Hamiltonian is H ⬅ pជ · xជ̇ − LT, which gives Eq. 共195兲. The

冓 冔
time derivative of the canonical momentum is dpជ / dt
dH ⳵ ⳵A储 ជ / dt, but db̂ / dt ⯝ ⳵b̂ / ⳵t + v储␬ជ with
= 共␮B + q⌽兲 − v储 , 共196兲 = mv̇储b̂ + mv储db̂ / dt + qdA
dt ⳵t ⳵t ជ b̂, and dA ជ / dt = ⳵Aជ / ⳵t + vជ · ⵜជA
ជ . The equations of
␬ជ ⬅ b̂ · ⵜ
motion in Lagrangian dynamics are dpជ / dt = ⳵LT / ⳵xជ . The
which can be integrated along with xជ g to obtain the en- ជ 共vជ · b̂兲
ergy H. The parallel velocity is given by the energy and gradient of the first term in LT is zero, that is, ⵜ
magnetic moment, ⯝ 0. This follows from ⵜ ជ 共vជ · b̂兲 = vជ ⫻ 共ⵜ
ជ ⫻ b̂兲 + vជ · ⵜ
ជ b̂, but b̂
ជ ជ
⫻ 共ⵜ ⫻ b̂兲 = −␬ជ and vជ · ⵜb̂ ⯝ v储␬ជ . Using ⵜ共vជ · A ជ ជ 兲 = vជ ⫻ B ជ
v储 = ± 冑2m共H − ␮B − q⌽兲. 共197兲 ជA
+ vជ · ⵜ ជ , one has ⵜ ជ L ⯝ qvជ ⫻ B
T
ជ + qvជ · ⵜ
ជAជ −ⵜជ 共␮B + q⌽兲.
Equation 共196兲 for the gyrophase averaged energy Putting the pieces together, mv̇储b̂ + mv储共⳵b̂ / ⳵t + v储␬ជ 兲
change can be derived by writing the Hamiltonian in its + q⳵A ជ / ⳵t ⯝ qvជ ⫻ B
ជ −ⵜ ជ 共␮B + q⌽兲. The component of the
ជ 兲2 / 共2m兲 + q⌽, with
canonical variables, H共pជ , xជ , t兲 = 共pជ − qA equations of motion that is along the magnetic field is
the particle velocity vជ = 共pជ − qA ជ 兲 / m. Hamilton’s equa- then mv̇储 = ␮b̂ · ⵜ ជ B + qE储, which is consistent with Eq.
tions imply that dH / dt = ⳵H / ⳵t, so 共196兲 for the gyrophase averaged energy change. The

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1118 Allen H. Boozer: Physics of magnetically confined plasmas

components of the velocity that are perpendicular to the

magnetic field agree with Eq. 共181兲 for vជ g.
Given a Lagrangian, it is usually trivial to obtain a
Hamiltonian description of trajectories. However, a
subtlety exists in obtaining a Hamiltonian description of
guiding-center motion from the Taylor Lagrangian. The
canonical momenta plus the coordinates of the Taylor
Lagrangian depend on only four independent variables,
which can be taken to be 共xជ , v储兲. Consequently the
Hamiltonian can have only four variables, two canonical
coordinates and two canonical momenta.
The canonical coordinates of the drift Hamiltonian
are closely related to the Boozer magnetic coordinates,
Eq. 共58兲, so one needs to transform the Taylor Lagrang-
ian into these coordinates. The expression is FIG. 12. 共Color兲 Trapped and passing particles: A particle is
trapped or passing depending on its energy H relative to the
energy ␮Bmax. Bmax is the maximum of the field strength along

LT =
m
2
冉
␮0
˙ + I␪˙
G␾
2␲B
冊 2

+q
␺t␪˙ − ␺p␸˙
2␲
− ␮B − q⌽m ,
a magnetic-field line, and ␮ is the adiabatically conserved mag-
netic moment. The distance along the line is ᐉ. The electric
potential ⌽ is ignored for simplicity.
共202兲
p␣ = q␺ 共203兲
where ⌽m ⬅ ⌽ + s. The function s is determined using Eq. and
共28兲 and gives the effect of the motion of the magnetic p␾ = mv储/B = q␳储 . 共204兲
coordinate system. Usually the distinction between ⌽m
and ⌽ is negligible, and so it will not be retained. The The Hamiltonian is the energy 关Eq. 共195兲兴,
most difficult and subtle point in the transformation of B2 2
the Taylor Lagrangian into Boozer coordinates is the H共p␣,p␾, ␣, ␾兲 = p + ␮B + q⌽. 共205兲
2m ␾
ជ =B
parallel velocity, vជ · B ជ · dxជ 共␺ , ␪ , ␸ , t兲 / dt, which gives
t
˙ A comparison of the trajectories given by the drift
v储 = 共␮ / 2␲B兲关G共␺ 兲␸ + I共␺ 兲␪兴. To prove this, use the
0 ˙ t t Hamiltonian in Clebsch coordinates with those obtained
chain rule to write dxជ / dt = ⳵xជ / ⳵t + 共⳵xជ / ⳵␺t兲␺˙ t + ¯. The from Eq. 共183兲 is instructive. Equation 共183兲 implies

冉 冊
term B ជ · ⳵xជ / ⳵t = B
ជ · uជ , Eq. 共29兲, can be taken to be zero,
d␺ ⳵␳储 ⳵B␣␳储
because the flow of the canonical coordinates along the = vជ g · ⵜជ ␺ = v储B − . 共206兲
field lines can always be chosen to be zero. The time
dt ⳵␣ ⳵␾
derivatives ␪˙ ⯝ 共v储 / B兲B ជ ·ⵜជ ␪ and ␸˙ are larger by the ratio The first of these two terms is reproduced by the Hamil-
of the gyroradius to the system size than the derivative tonian formulation, but the second is not. The change in
ជ = 共Bជ · ⳵xជ / ⳵␪兲␪˙ + 共B
ជ · ⳵xជ / ⳵␸兲␸˙ . The ␺ as the particle moves along its trajectory due to the
␺˙ t, so one can let vជ · B
term −v储B⳵共B␣␳储兲 / ⳵␾ = v储⳵共B␣␳储兲 / ⳵ᐉ is given by ␦␺
ជ ជ ˙
transformation of xជ̇ · A = A · uជ + 共␺ / 2␲兲␪ − 共␺ / 2␲兲␸˙ , while
t p = −B␣␳储. This follows from d␾ = Bdᐉ, with ᐉ the distance
ជ · uជ = −s with the choice of gauge g = 0 关Eq. 共A20兲兴.
A ជ , and dt = dᐉ / v储. B has dimensions of the mag-
along B ␣
Combining the results, one obtains Eq. 共202兲. netic field times a length, which means a length of order
Given the Taylor Lagragian in magnetic coordinates, a, the scale of the plasma. Consequently, the maximal
Eq. 共202兲, the determination of the canonical momenta deviation of ␺ along a trajectory is approximately
and the drift Hamiltonian are straightforward. The ca- ␦␺ / ␺ ⬇ ␳储 / a, which can be viewed as a redefinition of the
nonical momenta are p ⬅ ⳵L / ⳵␪˙ , which gives Eq. 共199兲
␪ guiding-center position, not a systematic drift. The loca-
and p␸ ⬅ ⳵L / ⳵␸˙ , which gives Eq. 共200兲. tion of the guiding center is a question of definition to
It is also useful to have the Hamiltonian for the within a distance of order a gyroradius. That arbitrari-
guiding-center motion in Clebsch coordinates 共␺ , ␣ , ␾兲. ness can be used to simplify the equations for the
In these coordinates the magnetic field has the contra- guiding-center motion.
variant representation B ជ =ⵜជ ⫻ 共␺ⵜ
ជ ␣兲, Eq. 共9兲, and the co-
ជ =ⵜជ ␾+B ⵜ ជ ជ 4. The action invariant J
variant representation B ␣ ␣ + B␺ⵜ␺, Eq. 共11兲.
Dotting the two representations together, one finds that The conservation of the magnetic moment ␮ causes
the inverse of the coordinate Jacobian is B ជ ·ⵜជ ␾ = B2. The the parallel velocity of a particle to pass through zero
Taylor Lagrangian is transformed into Clebsch coordi- and change sign at a point where H = ␮B + q⌽. Particles
ជ =Bជ · 共⳵xជ / ⳵␾兲␾˙ = ␾˙ and vជ · Aជ that are trapped between two such points of high mag-
nates by noting that vជ · B netic field or electric potential are called trapped par-
ជ = ␺␣˙ . The canonical momenta are
= 共dxជ / dt兲 · A ticles 共see Fig. 12兲. They have a periodic oscillation be-

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas 1119

tween the points and hence an adiabatic invariant, which symmetry, or if the magnetic-field strength is the
is customarily written as same at all minima of the field strength in a magnetic

冖
surface.
J= mv储dᐉ. 共207兲 • The electric potential in a confined plasma, with a
temperature T, has the characteristic magnitude 兩⌽兩
J is called the action invariant 共Northrop, 1963兲. In ⬇ T / e. The reason is that one species is generally
Clebsch coordinates 共␺ , ␣ , ␾兲, Eqs. 共9兲 and 共11兲, the ac- more poorly confined, so that a species preferentially
tion is J共H , ␮ , ␺ , ␣兲 ⬅ 养 p␾d␾ = 养 mv储dᐉ 关Eq. 共204兲兴, which leaves the plasma until its pressure gradient is bal-
means J is an action in the standard sense of Hamil- ជ p兩 = 兩enE
ជ 兩. Only a tiny
tonian mechanics. anced by the electric field 兩ⵜ
The derivatives of the action, J共H , ␮ , ␺ , ␣兲 give impor- fraction of the particles are lost while setting up this
tant information about the long-term trajectories of electric field, which is called the ambipolar field, so
trapped particles. The derivative with respect to the en- the plasma is approximately quasineutral.
ergy gives the bounce time ␶b, the time to go from one • A pressure gradient drives a net current along the
turning point to the other. Equation 共197兲 for the paral- magnetic-field lines. This current is called the boot-
lel velocity implies ⳵v储 / ⳵H = 1 / mv储, so strap current.
⳵J
⳵H
= 冖 dᐉ
v储
= 2␶b . 共208兲
In a confined fusion plasma, particles can move a dis-
tance more than a thousand times the size of the plasma
between collisions. Plasma confinement for times long
Writing J = q 养 ␳储d␾ makes the ␺ and ␣ derivatives easier compared to the collision time requires that the trajec-
to evaluate. Equation 共206兲 implies tories of all particles that form a Maxwellian distribution
stay close to the constant-pressure surfaces. In addition,
⳵␳储 vជ g · ⵜជ ␺ ⳵共B␣␳储兲
= + . 共209兲 the alpha particles that are produced by the fusion reac-
⳵␣ v储B ⳵␾ tion must remain confined as they slow from their birth
Therefore energy of 3.5 MeV and heat the plasma 共ITER Physics
Expert Group on Energetic Particle, …, 1999a兲. Two
⳵J
⳵␣
=q 冖 vជ g · ⵜជ ␺
v储
dᐉ = 2q⌬␺ , 共210兲
questions need to be addressed: 共1兲 Do the trajectories
of high-energy particles remain in the plasma? 共2兲 What
effect does the straying of near-thermal particles from
where ⌬␺ is the change in ␺ going from one turning
the pressure surfaces have on the transport coefficients?
point to the next. Similarly,
Before discussing the motion of particles and the as-
⳵J sociated transport phenomena, it is useful to have an
= − 2q⌬␣ . 共211兲 estimate of the radial electric field in a plasma. The char-
⳵␺
acteristic change in the electric potential across a con-
The long-term motion of a trapped particle consists of a fined plasma is 兩q⌬⌽ / T兩 of order unity. This potential
radial drift, difference is associated with only a small net charge den-
d␺ 1 ⳵J/⳵␣ sity qn⌬ with n⌬ / n Ⰶ 1. That is, a confined plasma is gen-
= , 共212兲 erally quasineutral. The reason for the potential differ-
dt q ⳵J/⳵H
ence ⌬⌽ is that one species, ions or electrons, is more
and a precession, poorly confined than the other. The more poorly con-
d␣ 1 ⳵J/⳵␺ fined species leaves the plasma until the electric poten-
=− . 共213兲 tial becomes sufficiently strong to provide confinement
dt q ⳵J/⳵H
for that species d ln p / d␺t = −qnd⌽ / d␺t. If the tempera-
ture were constant, the density and potential would be
E. Particle trajectories and transport related by n ⬀ exp共−q⌽ / T兲. Even a substantial density
drop is associated with a modest change in q⌽ / T. The
• Particles that have a sufficiently large ratio of the net charge density qn⌬ associated with this potential is
velocity parallel to the magnetic field, v储, to the total
given by Gauss’s law, ⵜ2⌽ = −qn⌬ / ⑀0, which implies qn⌬
velocity v can move all along the field lines 共Fig. 12兲
⬇ ⑀0⌬⌽ / a2 with a the minor radius of the plasma. The
and are generally well confined when magnetic sur-
fractional charge imbalance can be written as n⌬ / n
faces exist. Such particles are called passing particles.
⬇ 共␭D / a兲2, where the Debye length is ␭D ⬅ 冑⑀0T / q2n. In
• Particles with a small ratio of v储 / v are trapped be- a fusion plasma, the Debye length is of order a tenth of
tween maxima of the field strength 共Fig. 12兲 and are a millimeter, so the charge imbalance between electrons
well confined only when stringent conditions are met and ions n⌬ / n is extremely small, and the plasma is said
on the variation of the magnetic-field strength in the to be quasineutral. The electric force exerted on the
magnetic surfaces. The trapped particles can be well ជ , is also a factor of 共␭ / a兲2 smaller
confined if the field strength depends on only one overall plasma, qn E ⌬ D
angle in the magnetic surface, as is the case in axi- ជ p.
than the pressure force ⵜ

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1120 Allen H. Boozer: Physics of magnetically confined plasmas
A related concept to quasineutrality is ambipolarity,
which means the electrons and the ions diffuse at the
same rate so there is no radial current. In steady-state
situations, plasma transport is generally ambipolar for
otherwise the plasma would lose either all of the ions or
all of the electrons.
1. Confinement of particle trajectories
The confinement of individual particles is strongly de-
pendent on whether a particle is trapped or passing 共Fig.
12兲. If Bmax共␺t兲 is the maximum magnetic-field strength
on the magnetic surface that contains toroidal flux ␺t,
then a particle is trapped if ␮Bmax 艌 H − q⌽共␺t兲. If a par-
ticle is not trapped then it is said to be passing. For
passing particles, the parallel velocity retains a given
sign between collisions, v储 ⬎ 0 or v储 ⬍ 0, and is never zero.
For a barely trapped particle, the conservation of ␮ im-
plies ␮Bmax = ␮Bmin + mv2储 / 2, where v储 is the parallel ve-
locity at the location of the minimum of the magnetic-
field strength. If one defines 2⑀ ⬅ 共Bmax − Bmin兲 / Bmax,
then at the magnetic-field minimum v储 / v = 冑2⑀ for a
barely trapped particle. In a circular-cross-section toka-
mak, ⑀ ⯝ r / Ro.
The confinement of passing particles is rarely a prob-
lem unless magnetic surfaces are lost. The confinement
of passing particles is easily derived using the form for
the guiding-center velocity of Eq. 共183兲. Since the paral-
lel velocity never vanishes, the motion of a passing par-
ticle is along an effective magnetic field B ជ 共H , ␮ , xជ 兲 ⬅ B
ជ trapped-particle confinement can be derived starting
*
+ⵜ ជ ⫻ 共␳储B
ជ 兲, where the energy H and the magnetic mo-
ជ is di-
ment ␮ are constants of the motion. The field B *
vergence free and differs from the magnetic field Bជ only
by a small term that is proportional to the gyroradius to
the system size. If Bជ has magnetic surfaces, then the
surface quality of the B ជ field can be investigated using
*
the methods of Sec. III.A.
The confinement of trapped particles in toroidal plas-
mas is more difficult than that of passing particles, and
constrains the design of confinement systems. In prin-
ciple, the guiding-center drifts can carry a trapped par-
ticle out of the plasma after it has traveled a distance of
order 共R / ␳兲a, where R is the radius of curvature of the
field lines, ␳ is the gyroradius, and a is the minor radius
of the plasma. This distance is too short for confining
fusion plasmas. The time constant that is associated with
unbounded drift motion is ␶ ⬇ aR / 共␳v兲, which is of order
the time for Bohm diffusion 关Eq 共229兲兴.
Trapped particles are well confined if the magnetic-
field strength along the magnetic-field lines satisfies pe-
riodicity, B共ᐉ兲 = B共ᐉ + L兲, where ᐉ is the distance along a
field line and L is a constant along that line. Magnetic
configurations that satisfy this constraint are called qua-
sisymmetric. If the field strength is given in Boozer co-
ordinates, Eq. 共58兲, then periodicity implies the field
strength can depend on the poloidal and toroidal angles
only though the linear combination ␪h ⬅ ␪ + Nh␸, where
Nh is an integer. That is, the field strength has the form
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
B共␺t , ␪h兲. The helical canonical momentum, ph
⬅ pជ · 共⳵xជ / ⳵␸兲␪h, of the Taylor Lagrangian, Eq. 共202兲, is
conserved,
G − N hI ␺ p + N h␺ t
ph = ␮0 mv储 − q . 共214兲
2␲B 2␲
The expression for ph is obtained using 共⳵xជ / ⳵␸兲␪h
= 共⳵xជ / ⳵␸兲␪ − Nh⳵xជ / ⳵␪, so ph = p␸ − Nhp␪. The conservation
of ph means the excursions that trapped particles make
from the magnetic surfaces are proportional to the gyro-
radius.
An axisymmetric tokamak satisfies the condition of
quasisymmetry with Nh = 0. But the condition of quasi-
symmetry can also be satisfied to high accuracy in stel-
larators in which the magnetic surfaces are not symmet-
ric in ␸. The Quasi-Helically Symmetric stellarator at
the University of Wisconsin, the first operating quasi-
symmetric stellarator, has four periods, Np = 4 and Nh
= 4 共Talmadge et al., 2001兲. The NCSX stellarator, which
is being constructed at Princeton 共Zarnstorff et al., 2001兲
is quasi-axisymmetric, Nh = 0, just as is a tokamak, but
has three periods, Np = 3 共Fig. 7兲. Garren and Boozer
共1991a, 1991b兲 have shown that quasisymmetry cannot
be precisely achieved except in perfect axisymmetry.
However, the required breaking of quasisymmetry,
which is of order the local inverse aspect ratio cubed,
共r / R兲3, can be very small.
A different method than quasisymmetry for obtaining
with the requirement that a deeply trapped particle re-
main close to a flux surface. A condition for good con-
finement of deeply trapped particles is that all the
minima of the field strength Bmin along each field line
occur at essentially the same value of B 共Mynick et al.,
1982兲. Field minima satisfy ⳵B / ⳵ᐉ = 0 and ⳵2B / ⳵ᐉ2 ⬎ 0,
where ᐉ is the distance along a magnetic-field line.
Deeply trapped particles have 共v储 / v⬜兲2 Ⰶ 1 even when
they are near the minimum of the field strength and
have esentially zero action, J = 养 mv储dᐉ = 0 关Eq. 共207兲兴.
Since the action is conserved, the particles must remain
deeply trapped throughout their drift motion. That is,
they must remain at a minimum of the field strength.
The guiding-center drift, Eq. 共181兲, for a deeply trapped
particle is vជ g = 共B ជ / qB2兲 ⫻ ⵜជ 共␮B + q⌽兲 so the particles
drift on surfaces of constant ␮B + q⌽. Since the potential
is a function of the toroidal flux, ⌽共␺t兲 共Sec. V.B.3兲, the
deeply trapped particles stay on a flux surface if
⳵Bmin / ⳵␣ = 0. The derivative ⳵Bmin / ⳵␣ is zero at minima
if all minima of the magnetic field on a ␺t surface have
the same value.
If deeply trapped particles are confined to the mag-
netic surfaces, then it is relatively easy to shape the
variation in magnetic-field strength along the field lines
to confine most of the trapped particles. However, par-
ticles near the boundary between trapped and passing
tend to have bad orbits unless all field maxima are at the
same field strength 共Cary and Shasharina, 1997兲. The
condition of all field maxima being at the same field
 Allen H. Boozer: Physics of magnetically confined plasmas
strength is not as important because a radial varia-
tion in the average field strength, B0共␺t兲
⬅ 冑兰B2Jd␪d␸ / 兰Jd␪d␸, and the electric potential ⌽共␺t兲
cause particles to drift in a way that converts particles
near the trapped-passing boundary into either trapped
or passing particles. In addition, collisions in a plasma
are a diffusive phenomenon, and thermal particles near
the trapped-passing boundary switch rapidly between
the two types of orbits, which means they slowly diffuse
rather than rapidly drift out of the confinement region.
The W7-X stellarator is designed 共Nührenberg et al.,
1995兲 to make the minima of the field strength on a
magnetic surface have the same value. This is accom-
plished by the plasma’s having a pentagonal shape when
viewed along the z axis 共Fig. 6兲. Each magnetic surface
has five straight sections where the field strength is low
and five high-curvature corners where the field strength
is high. It is easy to design all field minima to have the
same field strength, since they occur in the straight sec-
tions. The maxima of the field strength occur in the cor-
ners of the pentagonal shape. The fractional variation in
the field strength at the maxima is approximately
2␦R / R, where ␦R is the half-width of the plasma along
the major radius at the corners, and ␦R is very narrow in
W7-X, ␦R / R ⬇ 1 / 20. The pressure balance in W7-X ap-
proximately satisfies p共␺t兲 + B20共␺t兲 / 2␮0 const, so at the
design beta value 具2␮0p / B2典 = 5% the radial variation in
B0 is sufficient to confine most particles near the
trapped-passing boundary. The principle used to confine
particle orbits in W7-X has been given many names:
linked mirrors, quasi-isodynamic, quasi-omnigenious,
and most recently quasipoloidal symmetry. Quasipoloi-
dal symmetry is broken by the strong curvature in the
corners to first order in the inverse aspect ratio r / R,
while quasisymmetry in its more traditional usage is bro-
ken in third order, 共r / R兲3.
A tokamak with too few toroidal field coils offers an
example of the bad particle confinement that occurs
when the field strength minima on a magnetic surface
have differing values. The field minima are created by
the space between the individual coils. Even when these
minima are shallow, ⳵Bmin / ⳵␣ is large and the particles
trapped in these minima drift out of the device on a
relatively short time scale aR / 共␳v兲. The fractional varia-
tion in the magnetic field due to toroidal asymmetry is
called the toroidal ripple ␦. The approximate expression
for the field strength is B共␺t , ␪ , ␸兲 = B0兵1 − 共r / Ro兲cos ␪
+ ␦ cos共N␸兲其 关Eq. 共19兲兴. Unless the toroidal ripple satis-
fies ␦ Ⰶ 共r / Ro兲共␫ / N兲 it causes a large number of second-
ary minima at varying values of the field strength.
When the minima of the magnetic field on a pressure
surface are not all at the same field strength, the colli-
sionless drift trajectories generally cross a large fraction
of the pressure surfaces. From the conservation of ac-
tion, Sec. VI.D.4, one has
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1121
d␺t ⳵J/⳵␣
=− . 共215兲
d␣ ⳵J/⳵␺t
Since ␣ is the Clebsch angle with a characteristic range
of unity, the radial excursion of particles ⌬␺t is approxi-
mately 共⳵J / ⳵␣兲 / 共⳵J / ⳵␺t兲. This can be reduced by enhanc-
ing the precession, which is the change in the Clebsch
angle per bounce, ⌬␣ = −共⳵J / ⳵␺t兲 / 2q 关Eq. 共211兲兴. The
precession of the deeply trapped particles is propor-
tional to ⳵共␮B + q⌽兲 / ⳵␺t, which for thermal particles is
generally dominated by the radial variation of the po-
tential, ⳵⌽ / ⳵␺t with either sign of the radial electric field
ជ = −共d⌽ / d␺ 兲ⵜជ ␺ enhancing the confinement of drift or-
E t t
bits. However, for superthermal particles a precession
zero, or resonance, can occur in which ⳵共␮B + q⌽兲 / ⳵␺t
= 0. Frequently, ⳵B / ⳵␺t ⬍ 0 for deeply trapped particles,
which means they are in a region of bad field-line curva-
ture. In this case, an electric field that tends to push a
charge species out, q共⳵⌽ / ⳵␺t兲 ⬍ 0, cannot have preces-
sion resonance and provides better confinement of the
individual trajectories than an electric field that pulls
that species in. For very-high-energy particles, such as
fusion ␣ particles, ␮ = 共mv⬜ 2
/ 2兲 / B is sufficiently large
that only the term ␮⳵B / ⳵␺t in the precession is impor-
tant.
The action J is only an adiabatic invariant and is not
conserved if the integrand of Eq. 共207兲 varies on the
time scale of the bounce motion ␶b. Even in the absence
of collisions, the action invariant can be broken in two
ways. First, the drift motion of a particle can take it to a
place where the local minimum in which it is trapped no
longer confines particles with the action that that par-
ticle has. Each region in which a particle can be trapped
has a maximum value of the action, Jmax, that it can
confine. As the particle drifts from one field line to an-
other, the maximum action varies, and if the action J
exceeds Jmax the particle escapes from the region where
it has been trapped. A particle can also be captured by
local minima if it drifts so that Jmax is increasing.
The second way the action J is broken by the particle
drift motion is if a particle drifts a sufficient distance in a
full bounce that the magnetic field near a turning point
changes significantly from one bounce to the next. This
is particularly important for finding the effect of toroidal
ripple on trapped alpha particles in tokamaks and can
determine the required limitation on ripple. The change
in ␺t per full bounce of a particle is 2⌬␺t共␺t , ␣兲 = ⳵J / ⳵␣. If
the change in ␣ during a full bounce, which is −2⳵J / ⳵␺t,
is sufficiently large, the sign of 2⌬␺t共␺t , ␣兲 is essentially
random with each full bounce, and the particle diffuses
collisionlessly with a diffusion rate that is approximately
共⌬␺t兲2 / ␶b, which is proportional to the square of the
ripple amplitude ␦2. This effect 共Goldston et al., 1981兲
can occur at arbitrarily small ripple, even ripple that is
too small to cause secondary minima ␦ Ⰶ 共r / Ro兲共␫ / N兲, if
the precession rate of the alphas is sufficiently large in
the axisymmetric field.
The breaking of the conservation of the action J due
to particle drift motion can give trapped-particle trajec-
1122 Allen H. Boozer: Physics of magnetically confined plasmas
tories a complexity that can only be studied by numeri-
cally integrating the guiding-center equations of motion.
2. Transport at low collisionality
The deviation of trapped-particle trajectories from the
pressure surfaces leads to enhanced transport and to a
net parallel current that is proportional to the pressure
gradient, the bootstrap current. The transport phenom-
ena associated with the deviation of the particle drift
trajectories from the pressure surfaces are known as
neoclassical transport.
Usually the transport rates for particles and energy in
tokamak plasmas are much larger than their neoclassical
values because of microturbulence 共see Sec. VI.F兲, so
the neoclassical transport theory is not as important as
one might think. However, the bootstrap current is im-
portant for steady-state tokamaks and can be a compli-
cation in the design of stellarators. The theory of neo-
classical transport in tokamaks has been reviewed by
Hinton and Hazeltine 共1976兲, and a book on the theory
of collisional transport in toroidal plasmas has been
written by Helander and Sigmar 共2002兲.
To understand neoclassical transport, suppose the
typical deviation of a trapped particle from a pressure
surface is a distance 共⌬r兲t. In a low-collisionality limit,
the distribution function must be consistent with the
Vlasov equation, df / dt = 0. Since the overall confinement
is long compared to a collision time, the distribution
function must also be close to a local Maxwellian, fM
⬀ n共r兲e−H/T共r兲. The deviation from a local Maxwellian is
f̂ = 共⌬r兲t⳵ ln共fM兲 / ⳵r. If this is inserted into Eq. 共146兲 for
the entropy production, ignoring the temperature gradi-
ent for simplicity, one finds that
ṡc =
␯
冑2 ⑀ 共⌬r兲2t 冉 d ln共n兲
dr
冊 2
, 共216兲
where 冑2⑀ is the fraction of trapped particles. For circu-
lar magnetic surfaces, ⑀ = r / Ro, the inverse aspect ratio.
Actually there are two factors of 冑2⑀ in this equation for
the entropy production. The first factor is an enhanced
effective collision rate ␯ / 共2⑀兲 because the collision op-
erator is diffusive, and particles need to be scattered
through velocity space by a distance of only 冑2⑀v to
move all the way across the trapped-particle part of ve-
locity space and become passing particles. The second
factor of 冑2⑀ comes from the trapped particles’ being the
only particles that have a large deviation from the pres-
sure surfaces 共Sec. VI.E.1兲. The rate of entropy produc-
tion is also 关d ln共n兲 / dr兴2 times the diffusion coefficient,
Eq. 共152兲, so the diffusion coefficient is D
⬇ 共␯ / 冑2⑀兲共⌬r兲2t .
For quasisymmetric confinement systems, the devia-
tion of the trapped-particle trajectories from a constant-
pressure surface can be calculated using the conserva-
tion of the helical canonical momentum ph 关Eq. 共214兲兴.
The deviation of a particle from a magnetic surface de-
pends on the variation in its parallel velocity v储, which is
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
maximized by the barely trapped particles. For a barely
trapped particle, v储 / v = ± 冑2⑀. Consequently a barely
trapped particle deviates by an amount
⌬␺bt = 冑2⑀␮0
G − N hI
␳ 共217兲
␫ + Nh
on either side of the flux surface on which its turning
points are located, where ␳ = mv / qB is the gyroradius.
At large aspect ratio, ␺t = ␲Br2, I Ⰶ G = 2␲RoB / ␮0, and
the deviation of a barely trapped particle is
共⌬r兲bt = 冑 2 ␳
⑀ ␫ + Nh
. 共218兲
This deviation of trapped particles is called a banana
orbit because of its shape when projected on a constant-
␸ plane. The diffusion, D ⬇ 共␯ / 冑2⑀兲共⌬r兲2t , is then ap-
proximately
冉 冊
␯ ␳ 2
D⬇ , 共219兲
⑀ 3/2
␫ + Nh
which is called the neoclassical diffusion coefficient 共Ga-
leev and Sagdeev, 1968兲. In tokamaks and quasi-
axisymmetric stellarators, Nh = 0, but Nh is nonzero in
quasihelically symmetric stellarators 共Talmadge et al.,
2001兲.
As shown by Kovrizhnikh 共1969兲, the statement that
Eq. 共219兲 gives the neoclassical diffusion in a quasisym-
metric system is misleading. This equation is approxi-
mately correct for the diffusion of heat, but the diffusion
of particles in a quasisymmetric system can only arise
from unlike particle collisions. The reason is the
momentum-conserving properties of the collision opera-
tor. Three expressions are important: 共1兲 the collisional
entropy production, Eq. 共145兲, 共2兲 the relation between
the collisional entropy production and the transport co-
efficients, Eq. 共152兲, and 共3兲 f̂ = 共⌬␺t兲⳵ ln共fM兲 / ⳵␺t. The de-
viation of a particle from a magnetic surface, ⌬␺t, is cal-
culated using ph conservation, Eq. 共214兲, which for a
density gradient implies f̂ ⬀ 共dn / d␺t兲mv储. The momen-
tum conservation properties of the collision operator
then make the entropy production zero, ṡc = 0, so no par-
ticle transport occurs. For a temperature gradient, f̂ has
additional factors of the velocity, so the entropy produc-
tion ṡc is nonzero, and the diffusion of heat is approxi-
mated by Eq. 共219兲. Particle transport does not vanish in
quasisymmetric systems because ions and electrons can
exchange momentum. For ions this rate of momentum
exchange is a factor of approximately 冑me / mi smaller
than the rate of ion-ion collisions, though for electrons
the rates of momentum exchange with like and unlike
particles are comparable. Momentum conservation
means that the transport of ions and electrons must be
at the same rate, to lowest nontrivial order in
gyroradius-to-system-size ratio, independent of the ra-
dial electric field. This phenomenon is called intrinsic
ambipolarity.
 Allen H. Boozer: Physics of magnetically confined plasmas
FIG. 13. 共Color兲 Trapping of particles: In the presence of a
density gradient, the number of trapped particles with a paral-
lel velocity greater than zero differs from the number with a
parallel velocity less than zero. This variation in density is
transmitted to the passing particles by the diffusive property of
collisions in a plasma. A particle is trapped if −冑2⑀ ⬍ v储 / v
⬍ 冑2⑀ at the minimum of the magnetic-field strength along a
magnetic-field line with ⑀ the variation in the field strength.
Otherwise the particle is passing.
In systems that are not quasisymmetric, the constraint
of ambipolarity sets the radial electric field. In quasisym-
metric systems, the radial electric field, and hence the
rate of toroidal rotation, are only weakly affected by
neoclassical transport.
One neoclassical effect that is very important for
steady-state tokamaks is the bootstrap current 共Bicker-
ton et al., 1971兲. The conservation of ph, Eq. 共214兲, im-
plies that the deviation of a trapped-particle trajectory
from the pressure surface has one sign, say positive,
共⌬r兲t ⬎ 0, when the parallel velocity is positive and an-
other, say negative, when the parallel velocity is nega-
tive. In the presence of a density gradient, dn / dr ⬍ 0,
this implies that at a fixed radius there are more barely
trapped particles with a positive than with a negative
parallel velocity. The passing particles that have a posi-
tive parallel velocity interact through a diffusive colli-
sion operator with passing particles that have a negative
parallel velocity only through the trapped particles. This
and the fact that there are more trapped particles mov-
ing in the positive direction along the magnetic-field
lines than in the negative direction leads to an excess of
passing particles moving in the positive direction 共see
Fig. 13兲. The excess of passing particles moving in the
positive direction produces a net parallel current jb
⬇ qvT共⌬r兲tdn / dr, where vT ⬅ 冑T / m is the thermal speed.
That is,
1 T dn
jb ⬇ 共220兲
冑⑀B ␫ + Nh dr .
The calculation of transport coefficients in non-
quasisymmetric configurations is complicated. Transport
coefficients as well as the effect of collisions on the par-
ticle trajectories can be determined using the Monte
Carlo equivalent to the collision operator 共Boozer and
Kuo-Petravic, 1981兲. The Monte Carlo collision operator
represents the effects of collisions during a time step by
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1123
a change of the velocity coordinates that has a random
component for each particle. The simplest Monte Carlo
collision operator represents the Lorentz collision op-
erator,
␯ ⳵ ⳵f
C共f兲 = 共1 − ␭2兲 , 共221兲
2 ⳵␭ ⳵␭
where ␭ ⬅ v储 / v is the pitch of the particle relative to the
magnetic field. The Monte Carlo equivalent is
␭n = 共1 − ␯␶兲␭o ± 冑共1 − ␭o2 兲␯␶ , 共222兲
where ␭n and ␭o are the new and the old values of the
pitch with the change caused by collisions during a time
interval ␶. The symbol ⫾ means the sign is chosen at
random. Since the particle trajectories are the character-
istics of the operator df / dt, the inclusion of effects of
collisions by Monte Carlo methods means one can find
the solution f to the equation df / dt − C共f兲 = g by the
method of characteristics. This is the basis of ␦f Monte
Carlo studies of transport 共Lin et al., 1995; Sasinowski
and Boozer, 1995兲, in which one calculates transport co-
efficients by using Monte Carlo methods to determine
the deviations from a local Maxwellian of the particle
distribution functions. An alternative to the Monte
Carlo codes for numerical evaluation of transport coef-
ficients is the DKES code 共van Rij and Hirshman, 1989兲
which solves an approximate form of the drift kinetic
equation by a variational principle.
3. Power required for maintaining fields
• The net current in a tokamak can be maintained in
steady state using externally produced radio-
frequency waves. However, the required power is un-
acceptably large for practical fusion power if more
than about a third of the total current is wave driven.
• Tokamaks can be designed so that more than two-
thirds of the net plasma current is maintained by the
bootstrap current, which is the net current driven by
the pressure gradient. The requirement of a large
bootstrap current makes the feasibility of steady-
state tokamaks more sensitive to the pressure profile
given by plasma transport processes than is a steady-
state stellarator.
An important issue for steady-state tokamaks is the
power that is required for maintaining the plasma cur-
rent 共Fisch, 1987; ITER Physics Expert Group on Ener-
getic Particles, …, 1999b兲. The traditional way to main-
tain the current in a tokamak is to have a solenoid
within the central hole of the torus. A change in the
magnetic flux in this solenoid produces a loop voltage,
V = ⳵␺p / ⳵t, Eq. 共31兲, which drives the current. A loop
voltage can only be maintained transiently, so a different
method must be adopted for long-pulse or steady-state
tokamaks.
The most developed method of steady-state current
drive uses waves to maintain a distribution of electrons
in which more electrons are moving in the direction re-
quired for the current than in the opposite direction.
1124 Allen H. Boozer: Physics of magnetically confined plasmas
Unfortunately, too much power is required to maintain
the full tokamak current. However, the total current
need not be externally driven because a pressure gradi-
ent drives a net toroidal current, the bootstrap current
关Eq. 共220兲兴. From the point of view of nonequilibrium
thermodynamics, the bootstrap current arises from a
cross term in the transport matrix. The bootstrap current
can provide most of the current in a tokamak, but the
magnitude and profile of this current is dependent on
the plasma pressure and density profiles. Since plasma
profiles are difficult to control in the absence of external
input power, the requirement of a large bootstrap cur-
rent places an additional uncertainty on the feasibility of
steady-state tokamaks, an uncertainty that does not exist
for steady-state stellarators.
Waves in various frequency ranges can provide the
power to maintain a current. In the frequency range of
the so-called lower hybrid waves, this power was derived
by Fisch 共1978兲 and in the electron cyclotron frequency
range by Ohkawa 共1970兲 and Fisch and Boozer 共1980兲.
The required power to maintain a current using waves
can be approximated by a simple argument 共Boozer,
1988兲. The most efficient steady-state current drive has
the current carried by high-energy, mildly relativistic
electrons that form a tail on the background Maxwell-
ian. The current density is j = entc, where nt is the num-
ber of the tail electrons per unit volume and c is the
speed of light. The power per unit volume that is re-
quired to maintain this current cannot be smaller than
the power required to maintain the tail, pw = 共␥
− 1兲ntmec2␯共␥兲, where ␥ ⬅ 1 / 冑1 − v2 / c2 and ␯共␥兲 is the
slowing-down rate of electrons with kinetic energy 共␥
− 1兲mec2. The power per unit volume pw is proportional
to a driven current of density j. The quantity Ew ⬅ pw / j
has units of volts per meter. For electrons at ␥ ⬇ 2, which
is the most efficient energy for current drive, Ew ⬇ Er
where
Er ⬅
e ln共⌳兲
4␲⑀0共c/␻pe兲
冉
2 ⯝ 0.087
V
m
冊冢 n
1020
1
冣 ,
m3
the background electron density is n, the electron
plasma frequency is ␻pe ⬅ 冑ne2 / ⑀0me, and ln共⌳兲 ⬇ 17 is
the Coulomb logarithm 关Eq. 共135兲兴. Equation 共223兲 is
obtained within a numerical factor if one uses the colli-
sion frequency ␯ of Eq. 共134兲 in the calculation of Ew
⬅ pd / j with v = c.
The power required to maintain a current using sub-
relativistic electrons can be calculated by similar argu-
ments, with the current density j = entv and the power
per unit volume pw = 共ntmev2 / 2兲␯共v兲. The rate of slowing
of high-energy particles ␯共v兲 is proportional to 1 / v3, Eq.
共134兲, so
c2
Ew ⬇ Er . 共224兲
v2
The total power to maintain the current is obtained by
multiplying Ew by 2␲R̄, where R̄ is the average major
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
radius on a magnetic surface, to obtain, Vw ⬅ 2␲R̄Ew,
which has units of voltage. The power is Pw
= 兰Vw共␺t兲共dI / d␺t兲d␺t, where I共␺t兲 is the net toroidal cur-
rent inside a magnetic surface that contains toroidal flux
␺t. The quantity Vw can be tens of volts, which makes the
power requirements for driving the total current—more
than ten megamperes—unacceptable.
The power required to maintain a current using waves
scales differently than p␩ = E储j储, the power required using
a loop voltage, which means an electric field E储 = ␩j储. The
power required for maintaining a current with waves,
pw, is proportional to the driven current, while p␩ scales
as the current squared. For the total current, the ratio
p␩ / pw ⬇ 共mec2 / T兲兵共c / ␻pe兲 / a其 / 冑␤␪, where ␻pe
2
⬅ nq2 / ⑀0me
is the square of the electron plasma frequency and ␤␪
⬅ 2␮0p / B2␪ is the plasma beta in the poloidal field alone.
For small tokamaks p␩ / pw can be larger than one, and
currents are efficiently carried by high-energy electrons,
such as runaway electrons. In a power-plant-scale device
p␩ is much less than pw. The more important ratio is the
ratio of pw to the fusion power, which is ␯T␶E共T /
mec2兲共c / ␻pe兲 / 共a冑␤␪兲, where ␯T is the collision frequency
of thermal electrons and ␶E is the energy confinement
time. In a power plant, this ratio is comparable to unity.
F. Microstability
• Confined plasmas are generally unstable to perturba-
tions that have a wavelength across the magnetic
field comparable to, or smaller than, the ion gyrora-
dius ␳i, but with a wavelength along the field lines
that is comparable to the overall system size. For
near-Maxwellian plasmas, the characteristic growth
rates and frequencies of microinstabilities are Cs / a,
where Cs ⬅ 冑共Te + Ti兲 / mi is the sound speed and a is
the plasma radius.
共223兲
• Microinstabilities lead to microturbulence with an
associated particle diffusion of order the gyro-Bohm
rate, Dg = ␳2i Cs / a. The relative amplitude of the fluc-
tuations in microturbulence is small, roughly equal to
the ratio of the ion gyroradius to system size, ␦n / n
⬇ ␳i / a.
• The most prominent microinstabilities in the modern
literature are the ion and the electron temperature
gradient modes. They are called the ITG or ␩i mode
and the ETG or ␩e mode.
• The logarithmic temperature gradients, d ln T / dr,
can have critical values at which the transport greatly
increases. Plasma temperature gradients may remain
close to these critical values, which makes the tem-
perature throughout the plasma proportional to the
temperature near the plasma edge.
Even when a plasma is stable to perturbations that
have a wavelength comparable to the plasma size, the
plasma may be unstable to perturbations with wave-
lengths comparable to or smaller than the gyroradius of
the ions, ␳i. Such instabilities are called microinstabili-
 Allen H. Boozer: Physics of magnetically confined plasmas
ties. Microinstabilities do not cause a sudden loss of the
plasma equilibrium but can greatly enhance the plasma
transport across the magnetic-field lines. Long-
wavelength instabilities, such as the instabilities dis-
cussed in Sec. V, can be so catastrophic in their effect
that the only issue of interest is whether they are stable
or not. With microinstabilities the primary issue is the
nonlinear, or saturated, state, which is generally turbu-
lent. In other words, the primary issue is the transport
caused by the microturbulence. Unlike the large fluctua-
tions associated with turbulence in ordinary fluids, the
fluctuations associated with plasma microturbulence are
small, ␦n / n ⬇ ␳i / a ⬃ 1 / 500. Horton 共1999兲 has reviewed
the theory of microinstabilities associated with drift
waves, which is the type of microinstability of most rel-
evance to toroidal plasmas. Yoshizawa et al. 共2001兲 have
reviewed the theory of turbulence in fluids and plasmas
with an emphasis on plasma microturbulence.
Although collisions play an important role in some
microinstabilities, generally the issue is whether the Vla-
sov equation, df / dt = 0, is consistent with the growth of
electromagnetic perturbations. Gardner’s theorem
共Gardner, 1963兲 gives a condition under which the
Vlasov-Maxwell equations can have no unstable solu-
tions. Unfortunately, this condition is violated in
current-carrying plasmas, such as magnetically confined
plasmas. Gardner 共1963兲 noted that the distribution
function in the Vlasov equation is the density of a con-
served fluid in phase space 共xជ , pជ 兲. If f共xជ , pជ 兲 has a form
such that the plasma energy is increased by the inter-
change of any packets of this fluid, then no energy can
be removed from the plasma to support electromagnetic
fluctuations, and the system must be stable. If the distri-
bution function of a single plasma species in the direc-
tion r̂ is
F共u,vជ 0兲 ⬅ 冕 ␦关u − r̂ · 共vជ − vជ 0兲兴f共vជ 兲d3v,
Gardner’s theorem says the Vlasov-Maxwell equations
have no unstable solutions if a vជ 0 exists such that
u⳵F / ⳵u 艋 0 for all species, for all values of u, and for all
directions r̂. Distribution functions can be far from local
Maxwellians and satisfy Gardner’s condition for stability.
Unfortunately, Gardner’s condition is not satisfied for a
current-carrying plasma, so microinstabilities are an is-
sue in magnetic confinement.
Rosenbluth and Rutherford 共1981兲 argued that only
low-frequency microinstabilities are energetically pos-
sible when the source of their free energy is the pressure
gradient. By low frequency they mean no higher than
approximately Cs / a, where Cs ⬅ 冑Ts / mi, Ts is the sum of
the electron and ion temperatures, a is the plasma ra-
dius, and mi is the ion mass. The energy per unit volume
associated with the oscillations of an instability is
roughly nmi共␻⌬兲2 with ␻ the frequency and ⌬ the spatial
scale of the microinstability. The maximum energy that
the oscillation can tap from the pressure gradient is ap-
proximately nTs共⌬ / a兲2, where a is the plasma radius.
The first-order term nTs⌬ / a vanishes because the aver-
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1125
age motion of the plasma 具⌬典 is zero in an oscillation.
The instability is not energetically favored unless
nmi共␻⌬兲2 艋 nTs共⌬ / a兲2, or 兩␻兩 艋 Cs / a.
A heuristic model clarifies some of the properties of
the plasma transport caused by the turbulence associ-
ated with fully developed microinstabilities. The per-
turbed electric potential ␦⌽ of a fully developed micro-
instability has a scale across the magnetic-field lines in
the magnetic surface of 1 / k⬜ and ⌬ in the radial direc-
tion. The variation of the potential ␦⌽ along the
magnetic-field lines is very weak, k储 / k⬜ Ⰶ 1, in order to
avoid Landau damping 共Sec. VI.C兲. The variation in the
electric potential causes a drift velocity Eជ ⫻B ជ / B2. The
magnitude of the radial component of this velocity is
␦vr ⬇ k⬜␦⌽ / B. Let ␶c be the correlation time of the mi-
croturbulence, which means the time scale over which
significant changes in the pattern of the perturbed po-
tential ␦⌽ occur. If ␦r ⱗ ⌬, then during a correlation time
particles in the plasma take a radial step, ␦r ⬇ ␦vr␶c, with
equal probability of the step’s being inwards or out-
wards. The random steps cause diffusion with the diffu-
sion coefficient D ⬇ ␦r2 / ␶c, which can be rewritten as D
⬇ k⬜2
共␦⌽ / B兲2␶c.
The correlation time of microturbulence has different
values in two limits. If the microturbulence is extremely
weak the correlation time is determined by the growth
rate of the underlying microinstability, and the theory is
essentially that of quasilinear diffusion 共Sec. VI.C兲.
However, when the microturbulence is fully developed
the correlation time is determined by the turbulence it-
self changing the potential ␦⌽. In this strong-turbulence
limit, the correlation time is determined by how long it
takes particles to diffuse across the potential contours,
␶c ⬇ 1 / 共k⬜2
Ds兲. The diffusion coefficient Ds is the coeffi-
共225兲
cient for diffusion in the magnetic surfaces across the
field lines. In this direction, the E ⫻ B velocity is ␦vs
⬇ 共␦⌽ / ⌬兲 / B, so the diffusion in the magnetic surfaces is
related to diffusion across the surfaces by Ds
⬇ D / 共k⬜⌬兲2. The correlation time is ␶c ⬇ ⌬2 / D, and the
radial step is ␦r ⬇ ⌬. Inserting this expression for ␶c into
D ⬇ k⬜ 2
共␦⌽ / B兲2␶c, one finds
D ⬇ k ⬜⌬ 冏 冏
␦⌽
B
. 共226兲
In strong microturbulence the particle diffusion is linear
in the perturbation amplitude rather than quadratic as it
is in quasilinear diffusion.
The perturbation amplitude that is reached in a mi-
croturbulent plasma is bounded by 兩ⵜ ជ ␦n兩 ⬇ 兩ⵜជ n兩. That is,
the microturbulence cannot create steeper gradients
than the ambient gradient while transferring energy
from the gradients. If one of the species is responding
adiabatically, which means e␦⌽ / T = ␦n / n, the bound on
the fluctuation amplitude is
1126 Allen H. Boozer: Physics of magnetically confined plasmas
e␦⌽ ␦n 1 the zonal flows is the surface-averaged inertial force of
= ⬇ , 共227兲
T n k ⬜a
where 1 / a ⬅ d ln 共n兲 / dr is the scale of the ambient den-
sity gradient. The radial diffusion coefficient is then
⌬ T
D⬇ , 共228兲
a eB
where T / 共eB兲 = ␳iCs, which is the ion gyroradius times
the speed of sound.
The transport caused by microturbulence is depen-
dent on the radial extent ⌬ of the constant ␦⌽ contours.
There are two extreme assumptions about ⌬. The more
pessimistic is that ⌬ is proportional to the size of the
plasma, ⌬ ⬀ a. This assumption makes the diffusion pro-
portional to the Bohm diffusion coefficient, which in the
modern literature is usually defined as
T changes in the potential by moving across the field lines
DB ⬅ , 共229兲
eB
although David Bohm’s unpublished work, which first
1 the ion gyroradius to system size, ␳i / a.
gave this coefficient, included an unjustified factor of 16 .
The confinement time given by Bohm diffusion, ␶B
⬇ a2 / DB, is similar to the time it takes particles to drift
out of a torus due to unconfined particle trajectories,
␶d = a / vg, where a is the plasma radius and vg ⬇ 共␳i / a兲Cs
is the guiding-center drift velocity. The more optimistic
assumption is that ⌬ is of order the ion gyroradius ␳i.
The assumption that ⌬ ⬇ ␳i gives the gyro-Bohm rate,
␳i T ␳2i Cs
Dg ⬅ = . 共230兲
a eB a
The Bohm time is far too short for a fusion power plant.
The gyro-Bohm confinement time, ␶g = a2 / Dg, is mar-
ginal,
a 3B 2
␶g ⬇ 7 ms , 共231兲
共T/10 keV兲3/2
where the plasma radius is in meters and the magnetic
field is in tesla.
What sets the radial scale ⌬ of the perturbed potential
␦⌽? First, consider a microinstability, such as the ion
temperature gradient 共ITG兲 instability discussed below,
in which the electrons have an adiabatic response, which
means ⌬n / n = e⌽ / T, but the ions behave nonadiabati-
cally. In linear theory, ⌬ can be much larger than the ion
gyroradius. However, the long radial contours of the po-
tential are broken up in the nonlinear microturbulent
state by what are known as zonal flows 共Diamond and
Kim, 1991兲. These flows come from the E ⫻ B drift in the
part of potential perturbation that has a nonzero aver-
age over the magnetic surface, ␦⌽̄共␺t , t兲 ⬅ 具␦⌽典. The
surface-averaged fluctuation in the potential, ␦⌽̄共␺t , t兲,
cannot be damped by electrons flowing along the field
lines, unlike the rest of the variation in the potential,
␦⌽ − ␦⌽̄. Indeed, the adiabatic response of the electrons,
if written correctly, is ␦n / n = e共␦⌽ − ␦⌽̄兲 / T. The drive for
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
the fluctuating E ⫻ B velocity. The left-hand side, or in-
ertial part, of the Navier-Stokes equation is ␳共⳵vជ / ⳵t
ជ vជ 兲 = ⳵共␳vជ 兲 / ⳵t + ⵜ
ជ · 共␳vជ vជ 兲, where the continuity equa-
+ vជ · ⵜ
tion ⳵␳ / ⳵t + ⵜ ជ · 共␳vជ 兲 = 0 was used to place the inertial
terms in the second form. The force that drives the zonal
flows is the average over the magnetic surfaces of
具ⵜជ · 共␳vជ vជ 兲典, where vជ = 共B ជ ⫻ⵜ ជ ␦⌽兲 / B2 is the E ⫻ B velocity
of the plasma in the turbulence. The tensor ␳vជ vជ is known
as the Reynolds stress tensor, so zonal flows are driven
by the Reynolds stresses. When the microturbulence is
strong, the zonal flows are sufficiently robust to make
⌬ ⬇ 1 / k⬜. When the ions are the nonadiabatic species,
the perpendicular wave number k⬜ tends to be compa-
rable to the ion gyroradius ␳i. The reason is that when
k⬜ Ⰶ 1 / ␳i the instability grows faster the larger k⬜. How-
ever, when k⬜ Ⰷ 1 / ␳i the ions respond adiabatically to
共Sec. VI.G兲, which removes the drive for the instability.
The typical fluctuation amplitude, Eq. 共227兲, is of order
Microinstabilities, such as the electron temperature
gradient 共ETG兲 mode that is discussed below, also exist
in which the ions respond adiabatically but electrons
nonadiabatically. The ions behave adiabatically when
the wave number of the perturbations satisfies k⬜␳i Ⰷ 1,
so the ions are free to cross the field lines in response to
changes in the potential. For this case zonal flows are
not important because the adiabatic ion response, ␦n / n
⬇ −e␦⌽ / T, is to the full variation in the electric potential
and not just the part that varies on the magnetic sur-
faces, as is the case with adiabatic electrons. The radial
extent ⌬ of the constant-potential contours is still lim-
ited by the ion gyroradius ␳i because otherwise the ions
could not respond adiabatically, although the perpen-
dicular wave number for the ETG mode is comparable
to the electron gyroradius, k⬜␳e ⬇ 1. Because of the
great anisotropy, k⬜⌬ ⬇ ␳i / ␳e, ETG microturbulence can
produce transport comparable to ITG microturbulence
with zonal flows 共Dorland et al., 2000; Jenko et al., 2000兲.
For both, the diffusion is comparable to the gyro-Bohm
rate. However, it should be noted that a number of ex-
periments see a reduction in the diffusion with ion mass
rather than the increase that would be expected from
gyro-Bohm diffusion. This effect, which is discussed by
Bessenrodt-Weberpals et al. 共1993兲, is not understood
theoretically.
An important feature of recent experiments has been
the observation of transport barriers, which are narrow
regions in which transport is greatly reduced. This topic
has been recently reviewed 共Terry, 2000; Wolf, 2003兲.
Despite the narrowness of the transport barriers, the re-
duction in transport is sufficiently great to significantly
enhance the overall plasma confinement. The theoretical
explanation 共Biglari et al., 1990兲 is a stabilization of the
microturbulence by a strong radial gradient, or shear, in
the E ⫻ B flow in the magnetic surfaces. The shearing
rate of the flow is ␥ ⬅ 兩d关共E ជ ⫻Bជ 兲 / B2兴 / dr兩. If the shearing
s
 Allen H. Boozer: Physics of magnetically confined plasmas
rate ␥s is greater than the growth rate of a microinsta-
bility, the instability is stabilized by being torn apart by
the sheared flow and is stabilized. Since the jump in the
potential across a shear layer cannot be greater than
roughly ⌬⌽ ⬇ Ts / e, and the typical growth rate is Cs / a,
one finds the width of a shear layer is roughly ␦s ⬇ 冑a␳i.
Neoclassical diffusion in a tokamak, Eq. 共219兲, divided
by the gyro-Bohm diffusion, Eq. 共230兲, is
Dg ␫2⑀3/2 Cs
⬇ , 共232兲
Dnc ␯ a
which is roughly 102 in a fusion plasma. A simple esti-
mate of the temperature drop that one can obtain across
a transport barrier ⌬T / T is 共␦s / a兲共Dg / Dnc兲. Since
共␦s / a兲共Dg / Dnc兲 is generally greater than unity under fu-
sion conditions, one can obtain a large temperature drop
across a transport barrier. The difference between zonal
flows and transport barriers is that zonal flows have a
radial scale and a time variation determined by the mi-
croturbulence. Transport barriers are quasistatic fea-
tures with the strong variation in the electric potential
related to the strong variation in the pressure, through
the tendency in plasmas for 兩ⵜ ជ p兩 ⬇ 兩enEជ 兩 for one of the
two species.
The diffusion coefficients that have been discussed are
more properly considered transport coefficients for heat
than coefficients for particles. The reason is that if either
species responds adiabatically there can be no particle
transport. Radial particle transport averaged over a
magnetic surface is ⌫r = 具共n + ␦n兲␦vr典. For an adiabatic re-
ជ ⫻ⵜ
sponse, ␦n / n = ± e␦⌽ / T while v = r̂ · 共B ជ ␦⌽兲 / B2, so ␦n
r
and vr are out of phase, which means their average over
the surface is zero. Although individual particles of the
nonadiabatic species diffuse, so that heat can be trans-
ported, the electric field arranges itself so the net par-
ticle flux is zero in order to preserve quasineutrality,
qni ⯝ ene.
Neither species responds fully adiabatically in fully
developed microturbulence, so microturbulence gener-
ally leads to particle transport across the magnetic-field
lines, which is effectively an enhancement of the perpen-
dicular component of the resistivity tensor ␩⬜ 关Eq. 共32兲兴.
The component of the resistivity along the magnetic
field, ␩储, is, however, rarely enhanced by microturbu-
lence, for two reasons. First, the parallel component of
the fluctuating electric field ␦E储 = −ik储␦⌽ is small com-
pared to the electric field across the magnetic-field lines
because 兩k储 / k⬜兩 Ⰶ 1, so little scattering of the parallel
motion of the particles occurs. Second, if any group of
ions and electrons can diffuse rapidly across the
magnetic-field lines, then ␩⬜ is enhanced, but if any
group of electrons can flow freely along the magnetic-
field lines, then ␩储 remains close to its quiescent plasma
value.
Our discussion of microturbulence has assumed that
only the electric potential is perturbed. Microturbulence
couples to shear Alfvén modes 共Sec. VI.H兲, when the
Alfvén frequency, ␻A ⬅ vA␫ / Ro, is comparable to that of
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1127
microinstabilities, Cs / a. Since the Alfvén velocity is vA
⬅ 冑B2 / ␮0min, the coupling occurs when the plasma
pressure satisfies ␤ ⬅ 2␮0p / B2 ⬎ 共a␫ / Ro兲2. Alfvén cou-
pling means that the component of the vector potential
parallel to the magnetic field, ␦A储, is perturbed. If ␦A储 / B
is Fourier decomposed in magnetic coordinates that
have a simple covariant form, Eq. 共58兲, then the mag-
netic surfaces are broken to form an island unless each
Fourier component 共␦A储 / B兲mn vanishes at its rational
surface, ␫ = n / m 关Eq. 共80兲兴. It is unclear whether the reso-
nant components of 共␦A储 / B兲mn are nonzero under stan-
dard plasma conditions, but if they were nonzero they
would produce a qualitative change in the microturbu-
lence due to electron transport along the stochastic
magnetic-field lines. On the other hand, if the Fourier
components 共␦A储 / B兲mn exactly vanish at their resonant
rational surfaces, then the effect of the coupling of the
microturbulence to the shear Alfvén modes would be to
cause the magnetic surfaces to wobble, which makes cal-
culations more difficult but causes no qualitatively new
physical effects.
The microinstability that has received the most atten-
tion in recent years is the ion temperature gradient
共ITG兲 instability, also called the ␩i mode,
d ln共T兲
␩i ⬅ , 共233兲
d ln共n兲
where T共␺t兲 is the ion temperature and n共␺t兲 is the ion
density. The ITG mode, which has been reviewed by
Horton 共1999兲, appears to be responsible for the en-
hanced transport of heat by ions in tokamak plasmas
and can be viewed as a kinetic version of sound waves
that are destabilized by the ion temperature gradient.
The simplest version of the ITG instability 共Kadomt-
sev and Pogutse, 1970兲 occurs in a uniform magnetic
ជ = Bẑ with the plasma having temperature and
field B
density gradients in the x direction. The fluctuations de-
pend on y, z, and time as exp关i共kyy + kzz − ␻t兲兴 and only
weakly on x. In a fluid model, the ions obey the conti-
ជ · 共nvជ 兲 = 0, with the perpendicular
nuity equation, ⳵n / ⳵t + ⵜ
components of the velocity given by vជ ⬜ = Eជ ⫻Bជ / B2. Us-
ing a tilde to denote perturbed quantities, the continuity
equation implies
ñ ky d ln n
␻ + ⌽̃ = kzṽ储 . 共234兲
n B dx
The parallel component of the velocity is given by force
balance along the field lines, nmidv储 / dt = −⳵p / ⳵z
− en⳵⌽ / ⳵z, or
␻minṽ储 = kz共p̃ + en⌽̃兲. 共235兲
The ions are assumed to respond adiabatically, which
means p / n5/3 is carried with the flow, ⳵共p / n5/3兲 / ⳵t
ជ 共p / n5/3兲 = 0, and
+ vជ · ⵜ
1128 Allen H. Boozer: Physics of magnetically confined plasmas

p̃ 5 ñ ky d ln n ITG mode is closely related to the destabilization of bal-

= − 共␩i − 2/3兲 ⌽̃.
p 3 n ␻B dx
The electrons are of sufficiently low mass that they can
move rapidly along the magnetic-field lines and remain
in thermodynamic equilibrium,
ñ e⌽̃
= ,
n Te
which is called an adiabatic response. In giving the elec-
tron response we have assumed that the perturbation
has a long spatial scale compared to the Debye length,
which implies the electron and ion densities are essen-
tially equal in both the unperturbed and the perturbed
states. In other words, the perturbed plasma is quasineu-
tral. Combining results, one finds
共236兲 looning modes by bad curvature 共Sec. V.C.2兲. The ver-
sion of the ITG mode in which curvature effects domi-
nate is called the toroidal branch, and the version in
which curvature effects are subdominant is called the
slab branch.
An important feature of ITG turbulence is that it be-
comes strong only when a critical gradient is reached. In
共237兲
the simplest version of the theory, the critical gradient is
a critical value of ␩i. When curvature effects are re-
tained, turbulence can arise when the radius of curva-
ture R times 1 / LT ⬅ 兩d ln T / dr兩 exceeds a critical value.
The R / LT critical gradient is counterintuitive, since the
weaker the curvature the easier it is for curvature to
destabilize the mode. However, the growth rate of the
mode and the maximum rate of transport become small

再 冉 冊冎
when R is large. In many experiments, in particular

1−
␻*e
␻
=
␻
冉 冊
kz 2
T Te 5 ␻*e
+ +
mi T 3 ␻
␩i −
2
3
, 共238兲
those with a high ion temperature, the plasma is thought
to operate just above the critical gradient. In this situa-
tion, the logarithmic ion temperature gradient,
where the electron drift frequency is d ln T / dr, is given by the critical gradient with the mag-
nitude of the heat flux determined by conditions at the
kyTe d ln n
␻*e ⬅ − . 共239兲 plasma edge. That is, the ion temperature throughout
eB dx the plasma is proportional to the ion temperature near
If the frequency, ␻, of the perturbation is high compared the plasma edge. A similar phenomenon occurs in the
to the drift frequency ␻*e, then Eq. 共238兲 gives sound earth’s atmosphere where the temperature profile is de-
waves termined by the critical gradient for convection,
d ln共p兲 / d ln共n兲 艌 ␥, with the magnitude of the heat flux
Te + 35 T determined by the boundaries of the convection zone.
␻2 = k2z . 共240兲 A second instability, the electron temperature gradi-
mi
ent 共ETG兲 mode, has very similar physics, only the role
However, when the frequency of the perturbation is low of the electrons and ions is reversed. The ETG mode has
compared to the drift frequency, ␻ Ⰶ ␻*e, the frequency wave numbers comparable to the electron gyroradius ␳e.
of the perturbation is On this scale the ions have only weak magnetic effects,

␻2 =
T 2
冉 冊
mi 3
− ␩i k2z ,
which has an exponentially growing ITG mode for ␩i
⬎ 2 / 3. In a plasma such as a fusion plasma in which
collisions are weak, neither the sound wave nor the ITG
mode are treated realistically in this analysis unless
␻ / k储 Ⰷ 冑T / mi, because otherwise strong-ion Landau
damping 共Sec. VI.C兲 implies that only decaying solutions
exist. In other words, the analysis is only realistic for
sound waves if T / Te Ⰶ 1 and for the ITG mode if ␩i Ⰷ 1.
To the extent the electrons respond adiabatically, the ␩i
mode causes ion heat transport but no electron trans-
port and, because of quasineutrality, no particle trans-
port.
Here we have ignored not only kinetic effects, but
also two other important determinants of ITG stability,
shear and the B ជ ⫻ⵜ
ជ B drift. Magnetic shear, the change
in direction of magnetic-field lines from one pressure
surface to another, is stabilizing. Magnetic shear effects
will be considered in the discussion of Alfvén instabili-
ties in Sec. VI.H.2. The B ជ ⫻ⵜជ B drift destabilizes the
ITG drift in regions of bad magnetic-field-line curvature,
which means the center of curvature is on the higher-
pressure side of the field lines. This destabilization of the
共241兲
␳i ⬇ 60␳e, and the ions respond adiabatically, ␦n / n
= −e⌽ / T. The ETG mode may be responsible for the
enhancement of the electron heat transport above its
neoclassical value 共Dorland et al., 2000; Jenko et al.,
2000兲, and the combination of the ITG and the ETG
modes may give the enhanced particle transport that is
observed in experiments.
G. Gyrokinetic theory
• The simplification of the particle trajectories and ki-
netic theory that occurs when the gyroradius ␳ is
small compared to the system size can be extended
to include perturbations of the magnetic and electric
fields that have arbitrarily large wave numbers per-
pendicular to the magnetic field, k⬜␳ arbitrary. This
approximate kinetic theory is called gyrokinetic
theory.
Computational studies of microinstabilities, which are
on the scale of the gyroradius of one of the species, are
carried out using the gyrokinetic equations. This and
other computational studies of plasmas have been dis-
cussed by Tang 共2002兲. The gyrokinetic equations are the
kinetic equations but with the particle velocity repre-
sented by the gyrokinetic drift velocity. The derivation

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas
of these equations was developed by Rutherford and
Frieman 共1968兲, Taylor and Hastie 共1968兲, and Antonsen
and Lane 共1980兲.
The gyrokinetic drift velocity consists of two parts.
The first part is the guiding-center drift of the particles
in the large-scale electric and magnetic fields. This is just
the ordinary guiding-center drift, vជ g, which was dis-
cussed in Sec. VI.D. The second part, ␦vជ g, is the modifi-
cation of the drift of the guiding center by the small-
scale perturbations. The perturbations that arise in
microturbulence are small, of order the gyroradius-to-
system-size ratio ␳ / a, but the drifts that the perturba-
tions cause can be of order of other guiding-center drifts
due to the strong spatial gradients, of order 1 / ␳.
Despite having a wavelength across the magnetic-field
lines comparable to the gyroradius of one of the species,
the perturbations that arise in a microturbulent plasma
have a wavelength along the magnetic field comparable
to the plasma size. The long parallel wavelength arises
to minimize Landau damping 共Sec. VI.C兲. This compli-
cated spatial structure can be accommodated math-
ematically by writing perturbed quantities, such as the
electric potential, in eikonal form,
␦⌽ = ⌽̃共xជ ,t兲eiS共␺t,␣兲 ,
just as one does for ballooning modes. The eikonal
S共␺t , ␣兲 depends on the two Clebsch coordinates, where
ជ =ⵜ
ជ ␺ ⫻ⵜ ជ ␣, Eq. 共9兲, and represents the rapid variation
B t
of ␦⌽ across the magnetic-field lines. ⌽̃ varies on a much
longer spatial scale, which involves the plasma size, and
with a slow time scale, of order the thermal sound speed
divided by the plasma radius Cs / a. Although
kជ ⬜ ⬅ ⵜជ S
is very large, k⬜␳ ⬇ 1, the spatial variation of kជ ⬜ itself is
much longer, involving the overall scale of the plasma.
Three electromagnetic quantities affect the particle
drifts due to their rapid spatial variation: the perturbed
electric potential, ␦⌽ = ⌽̃ exp共iS兲, the perturbed parallel
component of the vector potential, ␦A储 = Ã储 exp共iS兲, and
the perturbed parallel component of the magnetic field,
␦B储 = B̃储 exp共iS兲. As shown below, their effect on the
drifts is given by
冓
␦vជ g ⬅ e−iSg
dxជ g
dt
冔 =−
ikជ ⬜ ⫻ b̂
B
␹,
with the generalized potential
␮B̃储 2J1共k⬜␳兲
␹ = 共⌽̃ − v储Ã储兲J0共k⬜␳兲 +
q k ⬜␳
defined using the time derivative of the particle energy
or Hamiltonian,
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1129
q
⳵␹
⳵t
冓
⬅ e−iSg
dH
dt
. 冔 共246兲
Here Sg means the eikonal is evaluated at the guiding-
center position, and 具¯典 means an average over the gy-
rophase ␽. J0 and J1 are the zeroth and the first cylindri-
cal Bessel functions, ␮ ⬅ mv⬜ 2
/ 2B is the magnetic
moment, which is an adiabatic invariant, and b̂ ⬅ B ជ / B.
The time derivative of the guiding-center position xជ g,
Eq. 共180兲, depends on the electric and magnetic fields at
the true position of the particle, xជ , rather than at the
location of the guiding center, xជ g. The difference be-
tween the true position xជ and the guiding center xជ g is the
vector gyroradius ␳ជ = xជ − xជ g, Eq. 共177兲. The difference be-
tween the eikonal evaluated at the guiding-center posi-
tion and the actual position of a particle is Sg − S
= −kជ ⬜ · ␳ជ .
The Bessel functions that appear in the gyrokinetic
equations measure the importance of the interaction of
a charged particle with the magnetic field. As k⬜␳ → 0
the interaction is strong, and the Bessel functions have
the limits J0共k⬜␳兲 → 1 and 2J1共k⬜␳兲 / 共k⬜␳兲 → 1. For k⬜␳
→ ⬁, the interaction becomes weak, but at a slow rate,
共242兲 1 / 冑k⬜␳. As z ⬅ k⬜␳ → ⬁, the Bessel functions have the
asymptotic forms Jn共z兲 = 冑共2 / ␲z兲 cos关z − 共n + 1 / 2兲共␲ / 2兲兴.
The distribution function of gyrokinetic theory is
evaluated at the guiding-center position of the particles
xជ g rather than their actual position xជ and has the form
f共H , ␮ , xជ g , t兲 = f0共H , xជ g兲 + ␦f, where ␦f = f̃共H , ␮ , xជ g , t兲
⫻exp共iSg兲 and f0 is the equilibrium distribution function.
The gyrokinetic equation for the amplitiude of the
perturbed distribution function, ␦f = f̃ exp共iSg兲, is
共243兲 冉 ⳵
⳵t
冊
+ vជ g · ⵜជ + ikជ ⬜ · vជ g f̃ + Q + FI = CL共f̃兲, 共247兲
where the inhomogeneous term is
⳵␹ ⳵f0
FI ⬅ ␦vជ g · ⵜជ f0 + q , 共248兲
⳵t ⳵H
with f0 the distribution function for the unperturbed
equilibrium. CL共f̃兲 is a linearized collision operator, and
the quantity Q is a quadratic nonlinearity. The dominant
nonlinearity in gyrokinetic theory is generally taken to
ជ ␦f nonlinearity, for which Q has the
be the 共dxជ g / dt兲 · ⵜ
form
共244兲 Q=
b̂ · 具具e−iSgⵜជ 共␹eiSg兲 ⫻ ⵜជ 共f̃eiSg兲典典
B
. 共249兲
The notation 具具¯典典 means an average over the gy-
rophase and a projection of the expression using the or-
thogonality of the solutions to the linear gyrokinetic
equation. The function ␹ exp共iSg兲 comes from convert-
共245兲
ing the electric and magnetic fields that appear in dxជ g / dt
from the particle position to the guiding-center position
plus an average over the gyrophase. The gyrophase av-
erage is discussed below in the derivations of ␦vជ g and ␹.
1130 Allen H. Boozer: Physics of magnetically confined plasmas

The nonlinear part Q of the gyrokinetic equation re-

quires a definition of orthogonality among the solutions
to the linear gyrokinetic equation, which means solu-
J0共z兲 =
1
2␲
冕 ␲

−␲
e−iz cos ␽d␽ , 共253兲

tions that have different eikonals S. For example, in axi- to carry out the integration over the gyrophase ␽, which
symmetric systems different solutions in eikonal form is one of the integration variables in d3v 关Eq. 共171兲兴.
can have different toroidal mode numbers, as discussed The perturbed parallel component of the vector po-
in Sec. V.C.2 on ballooning modes. The gradient of the ជ ⫻ⵜជ
eikonal is the effective perpendicular wave number. As tential can be calculated using Ampère’s law, ⵜ
is well known from Fourier analysis, perturbations with ជ = ␮ ␦ជj, or
⫻ ␦A 0
different wave numbers are orthogonal. The definition
of orthogonality and the evaluation of the required inte-
2
k⬜ Ã储 = ␮0j̃储 . 共254兲
grals to obtain Q is the major subtlety of the study of The perturbed current density is
microturbulence with the gyrokinetic equation. An ap-
proximate method, which is relatively easy to follow, has
been given by Beer, Cowley, and Hammett 共1995兲. j̃储 = q 冕 v储J0共k⬜␳兲f̃d3x, 共255兲
The linear part of the gyrokinetic equation follows
from Eqs. 共244兲 and 共246兲 and the kinetic equation where we have assumed df0 / dH is symmetric in v储.
df / dt = C共f兲, where df / dt = df0 / dt + d␦f / dt. The time de- The equation for B̃储 is given by the force balance
rivative of the perturbed distribution function is d␦f / dt across the field lines, which for short-wavelength pertur-
= 共df̃ / dt兲eiSg + if̃eiSgdSg / dt, and the time derivative of bations has the form k̂ · ⵜជ 共p + B2 / 2␮ 兲 = 0. This follows
⬜ 0
the equilibrium distribution function is df0 / dt from ␮0ជj ⫻ B ជ =Bជ ·ⵜ
ជBជ −ⵜ
ជ B2 / 2 since B
ជ ·ⵜ
ជBជ = B2b̂ · ⵜ
ជ b̂
ជ f . The linear gyrokinetic
= 共df0 / dH兲dH / dt + 共dxជ g / dt兲 · ⵜ ជ ·ⵜ
ជ B is negligible when k Ⰷ k储. The perturbed per-
0 + b̂B ⬜
equation is obtained by multiplying both sides of the
kinetic equation by e−iSg and then averaging over the pendicular pressure is p̃⬜ = 具兰m共vជ · k̂⬜兲2⌬fd3v典, which
gyrophase ␽. The rapidly varying part of ␦f is in the means
eikonal, so f̃ is slowly varying, and df̃ / dt = ⳵f̃ / ⳵t + vជ g · ⵜ
The gyrophase average is 具dSg / dt典 = 具vជ g · ⵜ ជ S 典 = kជ · vជ .
g ⬜ g
ជ f̃.
p̃⬜ = q⌽̃ 冉 冕 ␮
⳵f0 3
⳵H
d v+ 冕 ␮f̃
2J1共k⬜␳兲 3
k ⬜␳
冊
d v B, 共256兲
To have a complete set of equations, relations are
where ␮ ⬅ mv⬜ 2
/ 2B is the magnetic moment and the
needed between the perturbed distribution function f̃ first-order Bessel function obeys
and the perturbed electric potential ⌽̃, the parallel com-
ponent of the vector potential Ã储, and the parallel com-
J1共z兲
z
=
1
2␲
冕 −␲
␲
e−iz cos ␪ sin2 ␪d␪ . 共257兲
ponent of the magnetic field B̃储. For purposes of taking
the averages, the perpendicular velocity of the particle The adiabatic term in p̃, which is the term proportional
will be written as
to ⌽̃, is equal and opposite for the electrons and the ions
provided the Debye length satisfies k⬜␭D Ⰶ 1, so e␦ne
vជ ⬜ = v⬜兵共cos ␽k̂⬜ ⫻ b̂ + sin ␽k̂⬜兲其. 共250兲 = q␦ni. The force balance is p̃⬜ + BB̃储 / ␮0 = const, so
The perturbed electric potential is given by Poisson’s
equation, which can be approximated as k⬜ 2
␦⌽ B̃储 = − ␮0 冕 ␮f̃
2J1共k⬜␳兲 3
k ⬜␳
d v. 共258兲
= 共q / ⑀0兲␦n, or
The remainder of the section gives the derivations of
Eq. 共245兲 for ␹, which is defined by q⳵␹ / ⳵t
q ⬅ 具e−iSgdH / dt典 and ␦vជ g ⬅ 具e−iSgdxជ g / dt典 关Eq. 共244兲兴.
2
k⬜ ⌽̃ = ñ. 共251兲
⑀0 First, we need the approximate relation between the
vector potential and the magnetic field that holds when
k⬜a Ⰷ 1 with a the plasma radius. The magnetic field is
The perturbed density of a species is ␦n = 兰⌬fd3v, where
ជ =ⵜជ ⫻ ␦A
ជ where ␦A ជ
ជ = Ã
⌬f ⬅ q␦⌽共⳵f0 / ⳵H兲 + ␦f 关Eq. 共153兲兴. Since ␦f = f̃ exp共iSg兲 ␦B exp共iS兲. Dotting this expres-
but ␦n = ñ exp共iS兲 with Sg − S = −k⬜␳ cos ␽, the perturbed sion with the unperturbed magnetic field B ជ , one finds
density is that Bជ · ␦B
ជ =ⵜ
ជ · 共␦A
ជ ⫻B
ជ 兲 + ␦Aជ ·ⵜ
ជ ⫻B
ជ . The only term on
the right-hand side that is large is the first term, through

ñ = q⌽̃ 冕 ⳵f0 3
⳵H
d v+ 冕 J0共k⬜␳兲f̃d3v, 共252兲
its dependence on kជ ⬜ ⬅ ⵜ ជ S,

ជ .
␦B储 = − 共ikជ ⬜ ⫻ b̂兲 · ␦A 共259兲
⬜

where we used the integral expression for the zeroth- The components of the perturbed magnetic field that are
order Bessel function, perpendicular to the unperturbed field are given by B ជ

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas 1131

⫻ ␦Bជ =ⵜជ 共B
ជ · ␦A
ជ 兲 − ␦A
ជ ⫻ⵜ
ជ ⫻B
ជ − ␦A
ជ ·ⵜ
ជBជ −B
large term is the first term, which gives
ជ ·ⵜ
ជ ␦A
ជ. The
冓 e−iSgvជ ⬜ ⫻ b̂
d iS
dt
e 冔 2 J1共k⬜␳兲
= − i共kជ ⬜ ⫻ b̂兲v⬜
k ⬜␳
. 共268兲

ជ = ikជ ␦A储 .
b̂ ⫻ ␦B
⬜
Second, we derive the expression for ␹, Eq. 共245兲,
from its definition, Eq. 共246兲. The exact equation for the
change in the energy of the particles is dH / dt = q共⳵⌽ / ⳵t
ជ / ⳵t兲 关Eq. 共198兲兴. The required gyrophase average
− vជ · ⳵A
具e−iSgdH / dt典 is the sum of two terms. The first term is
共260兲 Fourth, we calculate the part of ␦vជ g, Eq. 共244兲, that is
along the perturbed magnetic field. Motion along the
magnetic field is given by v储b̂ + v储␦b̂ with the change in
the direction of the magnetic field, ␦b̂ = ␦B ជ / B. Equa-
⬜
tion 共260兲 implies
ikជ ⬜ ⫻ b̂

冉 冊
␦具e−iSgv储␦b̂典 = v储Ã储J0共k⬜␳兲. 共269兲
⳵⌽̃ ⳵Ã储 B
q − v储 具ei共S−Sg兲典. 共261兲
⳵t ⳵t Putting the pieces together, one finds the effect of the
perturbations on the drift equation 共244兲 with the gener-
The gyrophase average 具exp关i共S − Sg兲兴典 can be written as
alized potential ␹ given by Eq. 共245兲.
具exp共ik⬜␳ cos ␽兲典 = J0共k⬜␳兲. The second term is

冓
− vជ ⬜ ·
⳵ ជ −iS
⳵t
共 ␦ A ⬜e g兲 = 冔
␮ ⳵B̃储 2J1共k⬜␳兲
q ⳵t k ⬜␳
, 共262兲 H. Alfvén instabilities
• Shear Alfvén waves are a twisting of the magnetic-
ជ
which is derived using Eqs. 共250兲 and 共259兲, vជ ⬜ · ␦A field lines. These waves have a continuous spectrum
⬜
ជ
= v k̂ · ␦A sin ␽ + i共v / k 兲␦B储 cos ␽. The gyrophase of possible frequencies but are generally heavily
⬜ ⬜ ⬜ ⬜ ⬜
average 具sin ␽ exp关i共S − Sg兲兴典 is zero because damped. The damping occurs if there is sufficient
field-line shear, which means variation in the direc-
1
2␲
冕 ␲

−␲
eiz cos ␽ sin ␽d␽ = 0, 共263兲
tion of the magnetic-field lines or sufficient variation
in the Alfvén velocity, B / 冑␮0␳ with ␳ the plasma
mass density, transverse to the magnetic field.
and the gyrophase average 具cos ␽ exp关i共S − Sg兲兴典 is
• If the Alfvén velocity varies along the magnetic-field
1
2␲
冕 ␲

−␲
eiz cos ␽ cos ␽d␽ = iJ1共z兲,
where z = k⬜␳ and ␳ = mv⬜ / qB. Assembling the parts,
one obtains 具e−iSgdH / dt典 = q⳵␹ / ⳵t with ␹ given by Eq.
共245兲.
Third, we calculate the part of ␦vជ g, Eq. 共244兲, that is
across the magnetic field. The time derivative of the
guiding-center position is given by Eq. 共180兲. Because of
the slowness of the time variation, the perturbed electric
field is given by the gradient of the potential,
ជ = − ikជ ␦⌽.
␦E
⬜
The contribution of the E ⫻ B drift to ␦vជ g is
共264兲
lines and the magnetic-field lines have shear, then the
continuous spectrum of Alfvén waves has gaps, and
weakly damped discrete 共sharp-frequency兲 shear
Alfvén modes can exist in these gaps.
• Energy is transferred between shear Alfvén modes
and particles with a resonant velocity, ␻ / k储, along the
field lines. A destabilizing transfer of energy from
the particles to the waves occurs if the diamagnetic
drift frequency of the interacting particles, ␻* of Eq.
共290兲, is larger than the frequency ␻ of the Alfvén
mode; otherwise the transfer is stabilizing. Only par-
共265兲
ticles, such as fusion alphas, that have a gyroradius in
the poloidal magnetic field alone comparable to their

冓 冔
density gradient have a destabilizing interaction.
ជ ⫻B
␦E ជ ikជ ⬜ ⫻ b̂ If a magnetic field is twisted and released, the twist
e−iSg =− J0共k⬜␳兲⌽̃, 共266兲 travels down the magnetic-field lines at the Alfvén ve-
B2 B locity,
where we used 具exp关i共S − Sg兲兴典 = J0共k⬜␳兲. The effect of
the perturbation on the term is vA ⬅ 冑 B2
, 共270兲

冓 冔 冓 冔
␮ 0␳
d 1 iB̃储 i共S−S 兲 dS and is called a shear Alfvén wave. Most shear Alfvén
␦ e−iSgvជ ⬜ ⫻ b̂ =− e g vជ ⬜ ⫻ b̂ .
dt ⍀ ⍀B dt perturbations are strongly damped, but certain perturba-
共267兲 tions called gap modes have weak natural damping. The
weakly damped gap modes are susceptible to being
ជ S = kជ · vជ , which means
The large term is dS / dt = vជ ⬜ · ⵜ driven to high amplitude by interactions with particles
⬜ ⬜
dS / dt = k⬜v⬜ sin ␽. Equation 共250兲 implies vជ ⬜ ⫻ b̂ that are moving along the field lines with a velocity com-
parable to the Alfvén velocity. This instability of the gap
= v⬜共k̂⬜ ⫻ b̂兲sin ␽ − v⬜k̂⬜ cos ␽. The gyrophase integral modes goes under the name of the toroidal Alfvén
that involves the second term in vជ ⬜ ⫻ b̂ is zero, and the eigenmode 共TAE兲 instability. The theory of short-
gyrophase integral that involves the first term is calcu- wavelength TAE instability was developed by Cheng et
lated using Eq. 共257兲, al. 共1985兲, and the theory of the more important low-

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1132 Allen H. Boozer: Physics of magnetically confined plasmas
mode-number TAE instability by Cheng and Chance
共1986兲. The concern is that alpha particles produced by
the fusion reaction will drive gap modes to a high am-
plitude causing a loss of the alphas from the plasma be-
fore they can transfer their energy to the bulk plasma
共ITER Physics Export Group on Energetic Particles, …,
1999a兲. Only at a low plasma beta, ␤ ⬅ 2␮0p / B2, is the
Alfvén velocity above the speed v␣ at which fusion al-
phas are produced, ␤ ⯝ 1.8% 共T / 20 keV兲共v␣ / vA兲2, with T
the plasma temperature. Wong 共1999兲 has reviewed the
relation between the experiments and theory for TAE
instabilities. The properties of shear Alfvén modes that
are central to this theory will be discussed in this section.
1. Continuum Alfvén wave damping
Shear Alfvén waves have a continuous spectrum that
is heavily damped 共Tataronis and Grossmann, 1973兲.
The physics is illustrated by a model in which the equi-
librium magnetic field is uniform and in the ẑ direction,
Bជ = Bẑ, but the plasma mass density ␳ depends on x. The
force balance is ␳共x兲⳵vជ / ⳵t = ␦ជj ⫻ B ជ with ␮ ␦ជj = ⵜ
ជ ⫻ ␦B
ជ . The
0 modified Bessel functions of the first, I0共␰兲, and the sec-
perturbation to the magnetic field is given by the ideal
Ohm’s law, ⳵␦B ជ / ⳵t = ⵜ
ជ ⫻ 共vជ ⫻ Bជ 兲. The two components of
velocity that enter the calculation are written as vx
= ṽx共x兲 exp关i共kyy + k储z − ␻t兲兴 and vy = ṽy共x兲 exp关i共kyy + k储z
− ␻t兲兴. The wave number ky is in the surface of constant
Alfvén velocity, and the wave number k储 is along the
magnetic field.
If the model equations for Alfvén waves are analyzed,
keeping all three components of ␦B ជ , one finds
d2ṽx dṽy
关␻2 − k2储 v2A共x兲兴ṽx + vA
2
+ ikyv2A =0
dx2 dx
and
dṽx
关␻2 − 共k2储 + k2y兲v2A共x兲兴ṽy + ikyv2A
dx
= 0.
These equations, which are more general than the deri-
vation given here, can be combined into a single equa-
tion for ṽx,
d
冉 ␻2 − k2储 v2A dṽx
dx ␻2 − 共k2储 + k2y兲v2A dx
+ 冊
␻2 − k2储 v2A
v2A
ṽx = 0.
When the Alfvén velocity is independent of position,
this equation has two types of solutions. One type is the
compressional Alfvén mode with v2Ad2vx / dx2 + 兵␻2 − 共k2储
+ k2y兲vA
2
其vx = 0, which is a propagating wave if k2x
⬅ 共␻ / vA兲 − 共k2y + k2储 兲 艌 0 and evanescent 共exponential de-
2 2
pendence on x兲 otherwise. The other solution is the
shear Alfvén wave with ␻2 = k2储 v2A, which has an arbitrary
dependence on x. When the Alfvén velocity depends on
position, these two types of solutions are coupled, and
the point where v2A共x兲 = ␻2 / k2储 is a singular point, a point
at which energy is absorbed.
Equation 共273兲 has a singular point, which will be de-
noted by x = 0, where v2A共x兲 = ␻2 / k2储 . At the singular
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
point, the velocity ṽy has a 1 / x singularity, which means
there is an infinite amount of energy in the vicinity of
the singular point, an unphysical result. This singularity
in the energy means an arbitrarily large amount of en-
ergy can accumulate, a buildup of energy that can be
represented by letting the frequency be complex, ␻
= ␻0 + i␥. Assume ␥ is small, let x = 0 be the point where
k2储 v2A = ␻20, and let
2 ␻ 0␥
␦⬅ 共274兲
k储 dv2A/dx
2
evaluated at x = 0. Then near its singular point Eq. 共273兲
can be approximated as
d dṽx
共x − i␦兲 = 共x − i␦兲k2yṽx , 共275兲
dx dx
while Eq. 共272兲 implies kyṽy = idṽx / dx.
To calculate the power per unit area flowing into the
singular point of Eq. 共273兲, we study the two indepen-
dent solutions of Eq. 共275兲 in the vicinity of the singular
point. These solutions are the cylindrical zeroth-order
ond, K0共␰兲, kind, where ␰ ⬅ ky共x − i␦兲. As ␰ → 0, I0 → 1
and K0 → −ln ␰. The regular, or I0, part of the solution is
undamped, but the singular, or K0, part is damped. Fo-
cusing on the singular part of the solution, let
ṽx = VsK0共␰兲, 共276兲
where Vs is a constant. Then near ␰ = 0, ṽy = −iVs / ␰. The
average power per unit y − z area that goes into the ve-
locity singularity is P = Re兵兰␳ṽ*y共⳵ṽy / ⳵t兲dx / 2其. Here we
used the easily proven result that if a function f共y兲 is the
共271兲 real part of fc exp共ikyy兲 and another function g共y兲 has
the same form, then the y average of f共y兲g共y兲 is
Re兵f*c gc其 / 2. Since ⳵ṽy / ⳵t = −i␻ṽy, P = 共␥ / 2兲 兰 ␳ṽ*yṽydx. The
integral that must be performed has the form
共272兲
冕 −⬁
⬁
dx ␲
= ,
x2 + ␦2 兩␦兩
共277兲
so P = 共␲ / 2兲共␥ / 兩␦兩兲␳0Vs2 / k2y where ␳0 = ␳共x = 0兲. In other
words, the power per unit area going into the sigularity
共273兲
is
P= 冏
␲ ␳0Vs2 d ln vA共x兲
␻0 2 . 冏 共278兲
2 ky dx
2. The weakly damped gap mode
The existence of weakly damped gap modes requires
both a shear in the magnetic-field lines across the lines
and a variation in the magnetic-field strength along the
magnetic-field lines. An equation for the shear Alfvén
mode, which contains both shear and variation in the
Alfvén velocity, is relatively simple when the perturba-
tion is localized to a magnetic-field line. This occurs
when the perpendicular wave number k⬜ is large in
comparison to the wave number along the magnetic field
 Allen H. Boozer: Physics of magnetically confined plasmas 1133

lines k储. The assumption k⬜ Ⰷ k储 leads to the eikonal ap- The energy density of a shear Alfvén wave,

冉 冊
proximation as in the discussion of ballooning modes,
␳k⬜
2
1 ⳵␦⌽* ⳵␦⌽
Sec. V.C.2, and gyrokinetics, Sec. VI.G. w= ␦⌽ * ␦⌽ + 2 , 共282兲
All the perturbed quantities in a shear Alfvén wave 4B2 ⍀ ⳵␨ ⳵␨
can be expressed in terms of the perturbation to the is the sum of the kinetic energy ␳vជ * · vជ / 4 and the mag-
electric potential ␦⌽ = ⌽̃ exp共iS兲, where S共␣兲 = 2␲m␣ is netic energy ␦B ជ * · ␦B
ជ / 4␮ . The factors of four, where
0
the eikonal and m is a large integer. The perpendicular two is expected, come from the use of complex arith-
wavelength is k ⬅ ⵜ ជ S. The Clebsch angle ␣ = 共␪ metic, as discussed just above Eq. 共277兲.
⬜ m
ជ =ⵜ
− ␫␸兲 / 2␲ is defined so B ជ ␺ ⫻ⵜជ ␣ 关Eq. 共9兲兴. The equa- A shear in the magnetic-field lines, s ⫽ 0, causes a
t
tion for the perturbed electric potential, which is derived rapid loss of energy from Alfvén waves if the field
below, is strength is constant, ⑀ = 0. Consider the Fourier trans-

冉 冊
form of Eq. 共279兲 with k⬜ 2
/ B replaced using Eq. 共281兲.
2
⳵ k⬜ ⳵␦⌽ k2
= − ⬜ ⍀共␨兲2␦⌽, 共279兲 That is, let ␦⌽ = 兰␾˜ exp共i␬储␨兲d␬储, which implies d␦⌽ / d␨
⳵␨ B ⳵␨ B ˜ and ␨␦⌽ → −id␾
→ i␬储␾ ˜ / d␬储. Then when the variation of
where ␨ is a dimensionless coordinate along the the Alfvén speed along the field lines is ignored, ⍀2
magnetic-field line. The distance along a line is ᐉ = const, Eq. 共279兲 implies
= 共L / 2␲兲␨ with L a characteristic distance. ⍀ is a dimen-
sionless, or normalized, frequency, d d␾
˜
s2 共⍀2 − ␬2储 兲 = 共⍀2 − ␬2储 兲␾
˜. 共283兲
⍀共␨兲2 ⬅ ␻2 冉 L
2 ␲ v A共 ␨ 兲
冊 2
, 共280兲
d␬储 d␬储
This equation is singular at the place where ␬储 = ⍀, which
is the equation for a shear Alfvén wave in the absence of
with ␻ the frequency of the perturbation ⌽̃ ⬀ exp共−i␻t兲. shear, s = 0. The energy density in ␬储 space, w␬, is defined
The characteristic form is ⍀2共␨兲 = ⍀20共1 + 2⑀ cos␨兲. In a so 兰w␬d␬储 = 兰wd␨, and is given by

冉 冊冉 冊冉 冊
large-aspect-ratio tokamak, ⑀ = r / Ro gives the variation
in the magnetic-field strength, and L = 2␲Ro / ␫ is the pe- ␲ k⬜ 2
␬2储 d␾
˜ * d␾
˜
w␬ = ␳ 1+ ␾˜ * ␾˜ + s2 共284兲
riodicity length of the field strength. The period of the 2 B 0 ⍀2 d␬储 d␬储
magnetic-field strength in a tokamak is ␪m = 2␲, but
with the subscript naught on 共k⬜ / B兲20 implying evalua-
along a field line ␪m = ␫␸ with dᐉ = Rod␸.The frequency
tion at ␨ = 0. Near the singular point ␬储 = ⍀, the dominant
␻A ⬅ 共2␲ / L兲vA is known as the Alfvén frequency, so ⍀
component in the energy density w␬ scales as the square
= ␻ / ␻A. For a large-aspect-ratio tokamak, ␻A = vA␫ / Ro.
A magnetic field has global shear if d␫ / d␺t is nonzero. of sd␾˜ / d␬储, and an infinite amount of energy is located
When the field is sheared, the characteristic dependence near this point in ␬储 space. These equations and the reso-
of k⬜2
/ B on ␨ is lution of their singularity exactly follow the derivation
that led to Eq. 共278兲 for the dissipation of Alfvén wave
2
k⬜ energy. The singular point, ␬储 = ⍀, of Eq. 共283兲 represents
⬀ 共1 + s2␨2兲, 共281兲
B the transfer of energy to the region ␨ → ⬁, which in
steady state has infinite energy content, while the singu-
with s ⬀ d␫ / d␺t. This follows from 2␲ⵜ ជ ␣=ⵜជ ␪ − ␫ⵜ
ជ␸ lar point k储vA = ␻ of Eq. 共273兲 represents the transfer of
m
ជ
− ␸共d␫ / d␺t兲ⵜ␺t with ␸ ⬀ ␨. For a large-aspect-ratio toka- energy to an arbitrarily small region in x, which means
mak, ␨ = ␫␸ and s = d ln␫ / d lnr. to kx → ⬁.
This paragraph contains the derivation of Eq. 共279兲 The existence of gaps in the spectrum of Alfvén waves
for ␦⌽. The derivation starts with the ideal Ohm’s law, and the existence of weakly damped modes in those
␦Eជ + vជ ⫻ B
ជ . The magnetic perturbation of a shear Alfvén gaps is easier to demonstrate if the perturbed quantities
wave is perpendicular to the main field, and when k⬜ are expressed in terms of u ⬅ 共ik⬜ / 冑B兲␦⌽, which places
the energy density in a nonsingular form,
→ ⬁ Eq. 共260兲 shows that ␦B ជ = ikជ ⫻ b̂␦A储 with b̂

冉 冏 冏冊
⬜ ⬜
ជ / B. Therefore ␦Eជ = −b̂⳵␦A储⳵t − ⵜ
ជ ␦⌽, which implies vជ ␳ 1 du s␨ 2
⬅B w= 兩u兩2 + 2 − su . 共285兲
ជ ⫻ ikជ 兲␦⌽ / B2 and ⳵␦A储 / ⳵t = −共B ជ ·ⵜជ ␦⌽兲 / B. The force 4B ⍀ d ␨ 1 + s 2␨ 2
= 共B ⬜
ជ implies ␦ជj = 共B
balance ␳⳵vជ / ⳵t = ␦ជj ⫻ B ជ ⫻ ␳⳵vជ / ⳵t兲 / B2 and The equation for u has the characteristic form
⬜
that ⵜ ជ · ␦ជj = k2 ␳共⳵␦⌽ / ⳵t兲 / B2. Since the current is ⳵ 2u s2
⬜ ⬜
ជ ·ⵜ
ជ 共j储 / B兲 = −ⵜជ · ␦ជj . 2 − u = − 共1 + 2⑀ cos ␨兲⍀20u. 共286兲
divergence-free, B Therefore
⬜ ⳵␨ 共1 + s2␨2兲2
ជ ·ⵜ
ជ 共j储 / B兲 = −k2 共␳ / B2兲⳵␦⌽ / ⳵t.
B ⬜ Ampere’s law gives The solution to this equation for values of ⍀0 is given in
2
k⬜ ␦ A 储 = ␮ 0j 储 ,
and ⳵ ␦⌽ / ⳵t = −␻ ␦⌽.
2
Let 2 2 ជ ·ⵜ
B ជ Fig. 14. The equation obtained from transforming
= 共2␲B / L兲⳵ / ⳵␨ with L a characteristic length along the Eq. 共279兲 is ⳵2u / ⳵␨2 − ␴共␨兲u + ⍀2共␨兲u = 0, where ␴共␨兲
field line. ⍀2 is defined by Eq. 共280兲. Combining results, ⬅ 共冑B / k⬜兲⳵2共k⬜ / 冑B兲 / ⳵␨2. Using k⬜ / 冑B ⬀ 冑1 + s2␨2 and
one obtains Eq. 共279兲. ⍀2共␨兲 = ⍀20共1 + 2⑀ cos␨兲, one obtains the characteristic

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1134 Allen H. Boozer: Physics of magnetically confined plasmas
FIG. 14. 共Color兲 The solution to Eq. 共286兲, plotted for three
values of the eigenvalue: red, ⍀0 = 0.42 gives a solution in the
continuum, green, ⍀0 = 0.4775 gives a singular solution, and
blue, ⍀0 = 0.487 gives a discrete mode. The parameters s = 0.5
and ⑀ = 0.2 are the same for all three solutions.
equation for u. If ⍀2 = const, then the solution is u
→ exp共i⍀␨兲 for either s → 0 or s␨ → ⬁. This means as ␨
→ ⬁ the Fourier transform of the shear Alfvén wave sat-
isfies ␬储 → ⍀. Since the amplitude of u is constant as ␨
→ ⬁, the solution u → exp共i⍀␨兲 has infinite energy.
Now consider the effect of a variation in the Alfvén
speed along the magnetic-field lines. The characteristic
equation for u, Eq. 共286兲, reduces to the Mathieu equa-
tion 共Bender and Orszag, 1978兲,
d 2u
+ 共1 + 2⑀ cos x兲⍀20u共x兲 = 0,
dx2
for either zero shear, s = 0, or for 兩s␨兩 Ⰷ 1. The Mathieu
equation has two types of solutions, depending on the
value of the eigenvalue ⍀20. For certain ranges of ⍀20, the
regular regions, the equation yields oscillatory solutions,
and for others, the singular, or gap, regions, there are
exponentially growing solutions 共Fig. 14兲. The smallest
⍀20 ⬎ 0 region that is singular occurs in the vicinity of
⍀20 = 1 / 4 for 兩⑀兩 Ⰶ 1, which means ␻ = ␻A / 2 with ␻A
⬅ vA␫ / Ro the Alfvén frequency. More precisely, the sin-
gular, or gap, region is ⍀20 = 41 共1 ± ⑀兲 for 兩⑀兩 Ⰶ 1. The regu-
lar ranges of ⍀20 give continuum shear Alfvén waves.
These solutions extend over the full range of ␨ and,
therefore, require an infinite energy to drive. The singu-
lar regions, or gaps, are values of ⍀20 for which physical
shear Alfvén waves do not exist. The energy in a singu-
lar solution is in the region 兩␨兩 → ⬁, since the amplitude
of u increases exponentially with 兩␨兩.
The characteristic equation for u, Eq. 共286兲, also has
spatially bounded solutions, which means the solution is
concentrated in the region where 兩s␨兩 is small 共Fig. 14兲.
These solutions are the gap modes because they occur in
the singular gap regions of the Mathieu equation and
have little intrinsic damping. They are the modes of in-
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
terest for Alfvén instabilities. Since they have finite en-
ergy, they can be driven unstable by their interactions
with high-energy particles. The existence of discrete
modes in the gaps of a frequency spectrum is a well-
known phenomenon in condensed-matter physics. For a
simple discussion see Allen et al. 共2003兲.
3. The particle-Alfvén wave interaction
The interaction of the particles that form the plasma
with the shear Alfvén wave is given by the kinetic equa-
tion. This interaction is weaker than one might at first
expect for modes, such as the discrete or gap mode, that
have no parallel electric field. When E储 = 0, particles in
resonance with a shear Alfvén wave, v储 = ␻ / k储, do not
exchange energy with the wave unless there is a cross-
field drift vd in the equilibrium magnetic field.
To simplify the analysis of the interaction of particles
with a shear Alfvén wave, the perturbed electric poten-
tial will be assumed to have the form
␦⌽ = ⌽̃ei共k z+kyy−␻t兲 .
储
共288兲
The distribution function f of the species that has a reso-
nant interaction with the Alfvén wave can be written as
the sum of an equilibrium distribution and a perturba-
tion, f = fr共H , x兲 + ␦f and obeys the Vlasov equation
df / dt = 0. The time-averaged power to the Alfvén wave
per unit volume will be shown to be
冉 冊 冉冊
␲ k2yvd2 ␻* q2 2 ␻
pw = −1 ⌽̃ f̄r , 共289兲
2 兩k储兩 ␻ Tr k储
where vd is the guiding-center drift velocity of particles
in the ŷ direction in the unperturbed magnetic field,
共287兲
kyTr ⳵ ln fr
␻* ⬅ 共290兲
qB ⳵x
is the diamagnetic drift frequency, f̄r共vz兲 ⬅ 兰frdvxdvy, and
Tr ⬅ −⳵ ln fr / ⳵H is the effective temperature of the inter-
acting particles, which have a Hamiltonian H.
The power to the shear Alfvén wave, Eq. 共289兲, has a
destabilizing sign only if ␻* is greater than ␻, which is
equivalent to the requirement that the gyroradius ␳␪ of
the interacting particles in the poloidal field alone be
sufficiently large to be comparable to their density gra-
dient. The condition on the poloidal gyroradius is de-
rived by first noting that the gap mode has a typical
frequency ␻ = ␻A / 2 with the Alfvén frequency ␻A
⬅ vA␫ / Ro. For the interacting species Tr ⬃ mrv2A, and ky
⯝ m / r with m the poloidal mode number. Defining the
poloidal gryroradius ␳␪ ⬅ 共Ro / ␫r兲␳, one finds that the
condition ␻ * ⬎ ␻ implies ␳␪d ln nr / dr ⲏ 1 / m. Only par-
ticles, such as fusion alphas, that have a poloidal gyrora-
dius comparable to their density gradient can destabilize
Alfvén modes. Extremely high m-number Alfvén modes
are not of such great concern because of the small radial
extent of the perturbations that they produce.
To derive the power going to an Alfvén wave, the
time derivative of both the unperturbed and the per-
 Allen H. Boozer: Physics of magnetically confined plasmas 1135

turbed parts of the distribution function are needed. The • The plasma edge can be defined by a solid object,
time derivative of the unperturbed part of the distribu- called a limiter, or by a separatrix between the
tion function fr is given by magnetic-field lines that lie on toroidal surfaces and
on open magnetic-field lines that intercept the cham-
dfr dxg ⳵fr dH ⳵fr ber walls. Most modern plasma confinement devices
= + , 共291兲
dt dt ⳵x dt ⳵H define the edge with a separatrix. This method of
defining the plasma edge is called a divertor.
which depends on dH / dt = ⳵H / ⳵t and the x̂ component
of the guiding-center velocity, dxg / dt. The time rate of The edge is the interface between the hot plasma and
change of the Hamiltonian is the walls. This interface must remove impurities, such as

冉 冊 冉 冊
the helium ash, and the waste heat. Ideally the waste
⳵H ⳵␦⌽ ⳵␦A储 k储v储 ⳵␦⌽ heat is removed by electromagnetic radiation because
=q − v储 =q 1− , 共292兲
⳵t ⳵t ⳵t ␻ ⳵t that avoids hot spots on the walls. In addition, the edge
is more important to plasma confinement than one
where we used ␦E储 = −⳵␦⌽ / ⳵z − ⳵␦A储 / ⳵t = 0, which means might expect because transport often appears to obey a
␻␦A储 = k储␦⌽. The relation between ␦A储 and ␦⌽ also gives critical-gradient theory, which means the temperature
the guiding-center velocity in the x direction, throughout the plasma is proportional to the edge tem-
dxg ␦Ey
dt
=
B
+ v储
␦Bx
B
=− 1−
␻ B ⳵y
冉
k储v储 1 ⳵␦⌽
, 冊 共293兲
perature.
In modern plasma confinement devices the plasma
edge is generally defined by the separatrix between
and the time derivative magnetic-field lines that lie on toroidal magnetic-

冉 冊
surfaces and open magnetic field lines that go into spe-
dfr ␻* q␦⌽ cially designed regions on the chamber walls 共Fig. 9兲.
= i共␻ − k储v储兲 1 − fr , 共294兲
dt ␻ Tr This method of defining the plasma edge is called a di-
vertor 共ITER Physics Expert Group on Diverter Model-
where we have written ⳵ ln fr / ⳵H = −1 / Tr and used the
ling …, 1999; Loarte, 2001兲.
diamagnetic drift frequency 关Eq. 共290兲兴.
Simple features of divertors can be understood from a
The time derivative of the perturbed part of the dis-
ជ vជ + ⵜ
fluid model, ␳vជ · ⵜ ជ p = ជj ⫻ B
ជ , with ⵜ
ជ · 共␳vជ 兲 = 0 and ␳
ជ ␦f,
tribution function ␦f is given by d␦f / dt = ⳵␦f / ⳵t + vជ g · ⵜ
= mn the mass density of the plasma. The plasma flow is
which can be written as d␦f / dt = −i共␻ − k储v储 − kyvd兲␦f,
where vd is the drift of particles in the y direction in the rapid along the magnetic field. Let ␳vជ = ⌫b̂ + ␳vជ ⬜, then
unperturbed magnetic field. Since f obeys the Vlasov
equation, df / dt = 0, the perturbed distribution function is
given by d␦f / dt = −dfr / dt, or
B 冉冊
ជ · ⵜជ ⌫ = S,
B
共297兲

冉 冊冉
␦f = 1 −
␻*
␻
1+
k yv d
冊 q␦⌽
␻ − k 储 v 储 − k yv d T r
fr . 共295兲
where the source S ⬅ −ⵜ ជ · 共␳vជ 兲. Equation 共297兲 gives the
⬜
flux of plasma ⌫ along a magnetic-field line. Since the
flow is essentially parallel to the magnetic field, the par-
The time-averaged power per unit volume going to allel component of the force balance is ⌫b̂ · ⵜ ជ 共⌫ / ␳兲
ជ 典. The perturbed electric
the Alfvén wave is pw = 具−␦ជj · ␦E ជ p = 0, which can be written as
+ b̂ · ⵜ
ជ = −ik ␦⌽ŷ, and the part of ␦j = 兰v ␦fd3v that
field is ␦E y y d
is in phase is the imaginary part, which is given by the d共⌫/␳兲 dp
⌫共ᐉ兲 + = 0, 共298兲
Landau integral, Eq. 共165兲, dᐉ dᐉ

i␦jiy = − i
␲
兩k储兩
冉 冊 冉冊
qkyvd2 1 −
␻* q␦⌽ ␻
␻ Tr
f̄r
k储
, 共296兲
dependent variable,
ជ . If the density ␳共ᐉ兲 is used as the in-
where d / dᐉ ⬅ b̂ · ⵜ

where f̄r共vz兲 ⬅ 兰frdvxdvy. The time-averaged power per

unit volume to the Alfvén wave is pw = 具−␦ជj · ␦E ជ典
d⌫2
d␳
冉 ⌫2
= − 2␳ Cs2 − 2 ,
␳
冊 共299兲

ជ ␦⌽典, which implies p = k j̃ ⌽̃ / 2. Consequently

= 具␦ជj⬜ · ⵜ i
w y y where Cs2 ⬅ dp / d␳. That is, the effective sound speed is
the power to the wave per unit volume is given by Eq. Cs2 ⬅ 共dp / dᐉ兲 / 共d␳ / dᐉ兲. Equation 共299兲 is well known
共289兲. from the theory of nozzles in fluid mechanics and says
the density drops as the flux ⌫ increases when the flow
starts with zero speed, ⌫ / ␳ = 0. The maximum flux that
VII. PLASMA EDGE can be obtained is ⌫ = ␳Cs, which implies a flow at the
• Control of the plasma edge is important to 共a兲 main- speed of sound Cs. If the divertor chamber exerts a suf-
tain plasma purity, 共b兲 limit the maximum heat flux to ficiently small back pressure, then the plasma flow will
the chamber walls, and 共c兲 achieve a high edge tem- reach the sonic rate as it flows down the field lines. In
perature if the plasma transport is associated with a the other limit in which the back pressure pb keeps the
critical temperature gradient. flow slow compared to the sound speed, the flux reaches

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1136 Allen H. Boozer: Physics of magnetically confined plasmas
⌫2 = 2␳0共p0 − pb兲 with p0 the pressure where ⌫ = 0. Diver-
tors have been operated in both the high and the low
back-pressure limit. The advantage of a high back pres-
sure is that more of the energy moving down the di-
vertor channel can be radiated away primarily through
atomic radiation 共Sec. VI.A兲, which reduces the peak
heat loads. The advantage of a sonic flow is that the
plasma is efficiently swept into the divertor chamber.
An important issue is the ability of materials to with-
stand the high energy fluxes in the energy of the imping-
ing particles. The energy of the impinging particles is
greatly modified by the sheath potential between a
plasma and a material surface. The flow of a plasma into
a wall is roughly at the speed of sound. However, the
electrons would naturally flow into the wall at the elec-
tron thermal speed, which is a factor 冑mi / me faster. To
preserve the quasineutrality of the plasma, a jump in the
electrostatic potential occurs on a Debye length scale
between a plasma and a wall with the potential holding
back the electrons. This means 冑Te / me exp共e⌬⌽兲 ⬇ Cs,
or ⌬⌽ / eTe ⬇ ln共冑mi / me兲 ⬇ 4. The ions impinge on the
wall not only with their thermal energy but also with the
kinetic energy obtained from the sheath potential q⌬⌽.
The required divertor flux is ⌫ = n̄␲a2 / 共␶p␦兲, where n̄ is
the average density in the plasma, ␶p is the confinement
time for particles, and ␦ is the width of the divertor out-
flow channel.
Stellarators use an island chain about a rational sur-
face, ␫ equal to a rational number, at the plasma surface
to divert the plasma into the divertor chamber 共Renner
et al., 2002兲. This idea is analogous to that used in toka-
maks where the separatrix that lies between field lines
that encircle the plasma and those that go to the divertor
chamber is at the ␫ = 0 rational surface, which is the only
rational surface on which an island can form in axisym-
metry.
The variation in the electrostatic potential across the
divertor region has important implications for the over-
all confinement of the plasma. The ions can sense the
presence of open magnetic-field lines once they are
within a banana orbit width of the plasma edge. Because
of their higher thermal speed, electrons tend to leave as
soon as they cross onto open field lines, while the ions
penetrate deeper into the open field-line region. Conse-
quently, it is natural for a strong shear in the E ⫻ B flow
to occur near the plasma edge, which can stabilize the
microtubulence. A region of greatly reduced transport
at the plasma edge is called a high, or H, mode of con-
finement and was the first transport barrier observed
共Wagner et al., 1982兲.
ACKNOWLEDGMENTS
The author would like to thank many colleagues and
friends who have answered questions, read drafts, and
made helpful suggestions. Three physicists, who work
with me at Columbia, deserve special thanks: Dmitry
Maslovsky, who drew or placed in appropriate form all
of the figures, David Lazanja, and Ronald Schmitt. Fig-
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
ure 5 was provided by the National Institute for Fusion
Science in Japan, Fig. 6 by the Max-Planck-Institut für
Plasmaphysik in Germany, Figs. 7 and 8 by the Princeton
Plasma Physics Laboratory in the United States, and
Fig. 9 by Edward Lazarus. This work was supported by
Grant Nos. DE-FG02-95ER54333 and DE-FG02-
03ER5496 from the U.S. Department of Energy.
APPENDIX: GENERAL COORDINATE SYSTEMS
Clever choices of coordinates are an important part of
classical physics. Any three quantities, which are con-
ventionally denoted by 共x1 , x2 , x3兲, can be used as coor-
dinates if they are well-behaved functions of the Carte-
sian coordinates and if positions in Cartesian
coordinates are well-behaved functions of 共x1 , x2 , x3兲.
The superscripts on the x’s number the coordinates and
are not powers. The position in space associated with
each coordinate point is defined by the transformation
function xជ 共x1 , x2 , x3兲. The transformation function can be
given in Cartesian coordinates as
xជ = x共x1,x2,x3兲x̂ + y共x1,x2,x3兲ŷ + z共x1,x2,x3兲ẑ. 共A1兲
For example, cylindrical coordinates 共R , ␸ , Z兲 are de-
fined by xជ 共R , ␸ , Z兲 = R cos ␸x̂ + R sin ␸ŷ + Zẑ.
Most people find it surprising that a coordinate trans-
formation is defined by giving xជ 共x1 , x2 , x3兲 rather that
xi共xជ 兲 with the index i going from 1 to 3. The reason
xជ 共x1 , x2 , x3兲 is given is that we want to convert functions
of position, like the temperature T共xជ 兲, into functions of
共x1 , x2 , x3兲. Given xជ 共x1 , x2 , x3兲 one has T共x1 , x2 , x3兲
= T„xជ 共x1 , x2 , x3兲….
The quantities 共x1 , x2 , x3兲 are valid coordinates only if
the Cartesian coordinates 共x , y , z兲 are well-behaved
functions of 共x1 , x2 , x3兲. This implies the Jacobian
J⬅
⳵xជ
⳵x 1 · 冉
⳵xជ
⳵x 2 ⫻
⳵xជ
⳵x3
冊 共A2兲
cannot be infinite. The condition that the quantities
共x1 , x2 , x3兲 be well-behaved functions of the Cartesian
coordinates implies that the Jacobian cannot be zero.
The gradients of the three coordinates ⵜ ជ xi with the
index i = 1,2,3, and the three tangent vectors ⳵xជ / ⳵xi are
related by the orthogonality relation,
⳵xជ
ⵜជ xi · j = ␦ij .
共A3兲
⳵x
This relation, which is fundamental to the whole theory
of general coordinates, can be proven using the chain
rule. First consider each coordinate to be a function of
the Cartesian coordinates xi共x , y , z兲, so ⵜជ xi = ⳵xi / ⳵xជ
= 共⳵xi / ⳵x兲x̂ + ¯. The derivatives of the transformation
equations, called tangent vectors, are ⳵xជ / ⳵xj = 共⳵x / ⳵xj兲x̂
+ 共⳵y / ⳵xj兲ŷ + 共⳵z / ⳵xj兲ẑ. When one takes a dot product be-
tween the gradient of a coordinate and a tangent vector,
ជ xi · ⳵xជ / ⳵xj = ⳵xi / ⳵xj. If
one finds using the chain rule that ⵜ
i = 1 and j = 2, one has ⵜ ជ x1 · ⳵xជ / ⳵x2 = 共⳵x1 / ⳵x2兲 1 3, which is
x ,x
 Allen H. Boozer: Physics of magnetically confined plasmas 1137

zero. In words, the derivative of x1 with respect to x2 The covariant and contravariant representation of
while holding x1 constant is zero. If one lets i = 1 and j vectors have distinct roles in both differential and inte-
= 1, one has ⵜជ x1 · ⳵xជ / ⳵x1 = 共⳵x1 / ⳵x1兲 2 3, which is one. gral vector calculus. First consider the gradient of a sca-
x ,x
The use of general coordinates is illustrated by the lar,
derivation of the equations for field lines. Field lines are ⳵f
defined by dxជ / d␶ = B ជ 共xជ 兲 where ␶ is a parameter that de- ⵜជ f = 兺 ⳵xi ⵜជ xi , 共A8兲
fines positions along the line. We want the trajectory of
the field lines in general coordinates, which means we which is a covariant vector.
Next consider the curl of a vector. This is easy for a
want the functions xi共␶兲. Now dxជ / d␶ = 兺共⳵xជ / ⳵xi兲dxi / d␶.
vector written in the covariant representation, which is
Dotting both sides of the defining equation for field lines
the expansion using the coordinate gradients,
by the various coordinate gradients and using the or-
thogonality relation, one finds that ជ=
ⵜជ ⫻ B 兺 ⵜជ Bi ⫻ ⵜជ xi . 共A9兲
i
dx ជ · ⵜជ xi .
=B 共A4兲 One can expand ⵜ ជ B in terms of the coordinate gradi-
d␶ i
ents and use the dual relations to show that
ជ ·ⵜ
If B ជ x3 is nonzero, one can use x3 as the parameter
ជ=1 ⳵xជ
that defines positions along the line. The chain rule im- ⵜជ ⫻ B
J
兺 ⑀ijk⳵iBj ⳵xk , 共A10兲
ជ ·ⵜ
plies dx1 / dx3 = B ជ x1 / B
ជ ·ⵜ
ជ x 3.
The orthogonality relation allows one to write the gra- where
dients of the coordinates in terms of the tangent vectors, ⳵
for example, ⳵i ⬅ 共A11兲
⳵xi
1 ⳵xជ ⳵xជ
ⵜជ x1 = ⫻ . 共A5兲 and ⑀ijk is the contravariant fully antisymmetric tensor.
J ⳵x2 ⳵x3 This tensor is defined by ⑀1,2,3 = 1 with the sign changing
This relation is called a dual relation and is important in if the order of two adjacent indices are changed. For
practical calculations, for otherwise the calculation of a example, ⑀2,1,3 = −1. If two indices are identical, the fully
coordinate gradient would require a function inversion antisymmetric tensor is zero, ⑀1,1,3 = 0. The curl of a co-
of xជ 共x1 , x2 , x3兲 to obtain x1 as a function of the Cartesian variant vector is a contravariant vector.
coordinates 共x , y , z兲. A vector in three dimensions can be The divergence can be calculated using the contravari-
expanded using any three independent vectors, so ant vector. One uses the dual relations to write

⳵xជ ⳵xជ ⳵xជ ⳵xជ ⳵xជ ⳵xជ J

ⵜជ x1 = a1 2 ⫻ 3 + a2 3 ⫻ 1 + a3 1 ⫻ 2 . 共A6兲
ជ=
B
2
兺 ⑀ijkBiⵜជ xj ⫻ ⵜជ xk , 共A12兲
⳵x ⳵x ⳵x ⳵x ⳵x ⳵x
Dotting this equation with ⳵xជ / ⳵x1, one finds that a1 where the components of the covariant antisymmetric
= 1 / J. Dotting the equation with ⳵xជ / ⳵x2, one finds a2 tensor ⑀ijk have identical values to the components of the
= 0. Similarly, one finds a3 = 0. contravariant tensor ⑀ijk. The divergence is simple be-
Dual relations also exist that give the tangent vectors cause the divergence of cross gradients is zero, and
in terms of the gradients, for example, ជ · 共fvជ 兲 = vជ · ⵜ
ⵜ ជ f + fⵜ
ជ · vជ . Writing the Jacobian as 1 / J
ជ 1 ជ 2 ជ
= ⵜx · 共ⵜx ⫻ ⵜx3兲, which can be proven using the dual
⳵xជ
= Jⵜជ x2 ⫻ ⵜជ x3 . 共A7兲 relations, one finds
⳵x1
⳵JBi
These dual relations are derived in an analogous man- ជ=1
ⵜជ · B 兺 . 共A13兲
ner. J ⳵xi
Given a transformation function xជ 共x1 , x2 , x3兲, one can The curl can be taken only if a vector is in covariant
expand any vector using the tangent vectors ⳵xជ / ⳵xi as form and the divergence can be taken only if a vector is
ជ = 兺Bi共⳵xជ / ⳵xi兲, is called
basis vectors. This expansion, B in contravariant form. How can one convert a vector
ជ . The or- from one form to another? The covariant components
the contravariant representation of the vector B
ជ · 共⳵xជ / ⳵xi兲, which if B
are given by B = B ជ is known in con-
thogonality relation implies the expansion coefficients i
ជ ·ⵜ
ជ x i. travariant form yields Bi = 兺共⳵xជ / ⳵xi兲 · 共⳵xជ / ⳵xj兲Bj. The
are given by Bi = B
metric tensor is defined as
An arbitrary vector can also be expanded using the
gradients of the coordinates as the expansion vectors, ⳵xជ ⳵xជ
ជ = 兺B ⵜ
ជ xi, which is called the covariant representation. gij ⬅ · , 共A14兲
B i ⳵xi ⳵xj
ជ · ⳵xជ / ⳵xi.
The expansion coefficients are given by Bi = B so Bi = 兺gijBj.
The dot product of two vectors is given by A ជ ·Bជ The name metric tensor comes from its role in deter-
= 兺AiBi = 兺AiBi. mining distances between two coordinate points. The

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

1138 Allen H. Boozer: Physics of magnetically confined plasmas

vector between two adjacent coordinate points is ␦xជ ជ ␺ , while the derivative of ␺
the form 共⳵␺t / ⳵t兲xជ = −uជ · ⵜ t p
= 兺共⳵xជ / ⳵xi兲␦xi. The square of the distance is 共␦xជ 兲2 ជ ␺ . Let B ជ ⬅ⵜជ ⫻A
ជ,
has the form 共⳵␺p / ⳵t兲xជ = 共⳵␺p / ⳵t兲c − uជ · ⵜ
= 兺gij␦xi␦xj. p
ជ = 关uជ · ⵜ
then uជ ⫻ B ជ 共␪ / 2␲兲兴ⵜជ ␺ − 共uជ · ⵜ
ជ ␺ 兲ⵜ
ជ 共␪ / 2␲兲 + ¯, and
A vector in covariant form can be rewritten in contra- t t
ជ xi · ⵜ
ជ xj. Using ជ can be written
the time derivative of A
variant form using Bi = 兺gijB where gij ⬅ ⵜ

冉 冊 冉 冊
j
the dual relations one can show that gij is the matrix ជ
⳵A ⳵␺p ជ ␸ ជ + ⵜជ s.
inverse of gij. =− ⵜ + uជ ⫻ B 共A19兲
There are three types of integrals: line, area, and vol- ⳵t xជ ⳵t c 2␲
ume. A line integral is performed along a curve, which ជ ␪ = −共⳵␪ / ⳵t兲 ជ and uជ · ⵜ
ជ ␸ = −共⳵␸ / ⳵t兲 ជ
means one of the three coordinates is varied while two The expressions uជ · ⵜ x x

are held constant. Let us assume the first coordinate is plus Eq. 共A17兲 imply
varied, then dxជ = 共⳵xជ / ⳵x1兲dx1 and 兰B ជ · dxជ = 兰Bជ · 共⳵xជ / ជ · uជ = − s + 共⳵g/⳵t兲 .
A 共A20兲
c
⳵x 兲dx , which can also be written as 兰B
1 1 ជ · dxជ = 兰B dx1. In
1
other words, a line integral is an ordinary integral of a
covariant coefficient. REFERENCES
The only difficulty in area, or surface, integrals is be-
Albajar, F., M. Bornatici, and F. Engelmann, 2002, Nucl.
coming comfortable with the definition of the area ele-
Fusion 42, 670.
ment. A surface is defined by holding one coordinate Alfvèn, H., 1940, Ark. Mat., Astron. Fys. 27A, 22.
constant, say x1, and varying the other two. The area Allen, P. B., S. Aubin, and R. B. Doak, 2000, Am. J. Phys. 68,
element is then 228.
Antonsen, T. M., and B. Lane, 1980, Phys. Fluids 23, 1205.
⳵xជ ⳵xជ 2 3 ជ 1
daជ 1 ⬅ ⫻ dx dx = ⵜ x Jdx2dx3 , 共A15兲 Arnold, V. I., 1963, Russ. Math. Surveys 18 共5兲, 9.
⳵x2 ⳵x3 Aymar, R., 2000, Plasma Phys. Controlled Fusion 42, B385.
Baker, C. C., R. W. Conn, F. Najmabadi, and M. S. Tillack,
where a dual relation was used to obtain the second
1998, Energy 23, 649.
form. An area integral is then the double integral
Bateman, G., 1978, MHD Instabilities 共MIT Press, Cambridge,
兰B · daជ = 兰B1Jdx2dx3, where B1 = B ជ ·ⵜជ x1 is the contravari- MA兲.
ant coefficient. Bauer, F., Betancourt, O., and P. Garabedian, 1984, Magneto-
The volume element is obtained by dotting the area hydrodynamic Equilibrium and Stability of Stellarators
element with the distance across the surface, d3x 共Springer, New York兲.
⬅ 共⳵xជ / ⳵x1兲dx1 · daជ 1, which can be written as d3x Baumjohann, W., and R. A. Treumann, 1996, Basic Space
= Jdx1dx2dx3. In other words, the integral of a function f Plasma Physics 共Imperial College, London兲.
Beer, M. A., S. C. Cowley, and G. W. Hammett, 1995, Phys.
over a volume is the triple integral 兰fJdx1dx2dx3.
Plasmas 2, 2687.
The expression for the time derivative of a general
Beidler, C., G. Geiger, F. Hernegger, E. Harmeyer, J.
covariant vector, namely, the vector potential, is impor-
Kisslinger, W. Lotz, H. Maassberg, P. Merkel, J. Nührenberg,
tant for the theory of the evolution of magnetic fields. F. Raw, et al., 1990, Fusion Technol. 17, 148.
Let A ជ =␺ ⵜ ជ 共␪ / 2␲兲 − ␺ ⵜ
ជ 共␸ / 2␲兲 + ⵜ
ជ g. The time derivative Bellan, P. M., 2000, Spheromaks 共Imperial College, London兲.
t p
at a fixed spatial point xជ is Bender, C. M., and S. A. Orszag, 1978, Advanced Mathemati-

冉 冊 冉 冊
cal Methods for Scientists and Engineers 共McGraw-Hill, New
⳵Aជ
⳵t xជ
=
⳵␺t ជ ␪
ⵜ
⳵t xជ 2␲
−
⳵␪/2␲
⳵t
冉 冊 xជ
ⵜជ ␺t + ¯ + ⵜជ s,
York兲, p. 560.
Berger, M. A., 1984, Geophys. Astrophys. Fluid Dyn. 30, 79.
Bernstein, I. B., 1983, in Basic Plasma Physics I, Handbook of
共A16兲 Plasma Physics, No. 1, edited by A. A. Galeev and R. N.
Sudan 共North-Holland, Amsterdam兲, p. 421.
where the ¯ stands for terms involving ␺p and ␸, which Bernstein, I. B., E. A. Frieman, M. D. Kruskal, and R. M.
are of the same form as those for ␺t and ␪. The quantity Kulsrud, 1958, Proc. R. Soc. London, Ser. A 244, 17.

s⬅ 冉 冊 冉 冊 冉 冊
⳵g
⳵t xជ
+ ␺t
⳵␪/2␲
⳵t xជ
− ␺p
⳵␸/2␲
⳵t xជ
. 共A17兲
Bessenrodt-Weberpals, M., F. Wagner, O. Gehre, et al., 1993,
Nucl. Fusion 33, 1205.
Betti, R., and J. P. Freidberg, 1995, Phys. Rev. Lett. 74, 2949.
Bhattacharjee, A., Z. W. Ma, and X. Wang, 2001, Phys.
Time derivatives holding the coordinates 共␺t , ␪ , ␸兲 fixed, Plasmas 8, 1829.
which are denoted by a subscript c, are related to time Bialek, J., A. H. Boozer, M. E. Mauel, and G. A. Navratil,
derivatives holding the spatial position xជ fixed by the 2001, Phys. Plasmas 8, 2170.
chain rule, Bickerton, R. J., J. W. Connor, and J. B. Taylor, 1971, Nature

冉冊 冉冊 冉 冊
共London兲, Phys. Sci. 229, 110.
⳵f ⳵f ⳵xជ
= + · ⵜជ f. 共A18兲 Biglari, H., P. H. Diamond, and P. W. Terry, 1990, Phys. Fluids
⳵t c ⳵t xជ ⳵t c B 2, 1.
Bishop, C. M., 1989, Plasma Phys. Controlled Fusion 31, 1179.
The velocity of the 共␺t , ␪ , ␸兲 coordinates through space is Bisnovatyi-Kogan, G. S., and R. V. E. Lovelace, 2001, New
uជ ⬅ 共⳵xជ / ⳵t兲c, so the time derivative of the coordinates has Astron. Rev. 45, 663.

Rev. Mod. Phys., Vol. 76, No. 4, October 2004

 Allen H. Boozer: Physics of magnetically confined plasmas
Bondeson, A., Y. Q. Liu, D. Gregoratto, C. M. Fransson, B.
Lennartson, C. Breitholtz, Y. Gribov, and V. D. Pustovitov,
2002, Phys. Plasmas 9, 2044.
Boozer, A. H., 1981, Phys. Fluids 24, 1999.
Boozer, A. H., 1983, Phys. Fluids 26, 1288.
Boozer, A. H., 1984a, Phys. Fluids 27, 2055.
Boozer, A. H., 1984b, Phys. Fluids 27, 2441.
Boozer, A. H., 1986, J. Plasma Phys. 35, 133.
Boozer, A. H., 1988, Phys. Fluids 31, 591.
Boozer, A. H., 1992, Astrophys. J. 394, 357.
Boozer, A. H., 1999, in Magnetic Helicity in Space and Labo-
ratory Plasmas, edited by M. R. Brown, R. C. Canfield, and
A. A. Pevtsov 共American Geophysical Union, Washington,
D.C.兲, p. 11.
Boozer, A. H., 2002, Phys. Rev. Lett. 88, 215005.
Boozer, A. H., 2003, Phys. Plasmas 10, 1458.
Boozer, A. H., T. K. Chu, R. L. Dewar, H. P. Furth, J. A.
Goree, J. L. Johnson, R. M. Kulsrud, D. A. Monticello, G.
Kuo-Petravic, G. V. Sheffield, S. Yoshikawa, and O. Betan-
court, 1983, in Plasma Physics and Controlled Nuclear Fusion
Research 1982, Proceedings of the International Conference
on Plasma Physics and Controlted Nuclear Fusion 共Interna-
tional Atomic Energy Agency, Vienna兲, p. 129.
Boozer, A. H., and G. Kuo-Petravic, 1981, Phys. Fluids 24, 851.
Brambilla, M., 1998, Kinetic Theory of Plasma Waves: Homo-
geneous Plasmas 共Clarendon, Oxford/Oxford University, New
York兲.
Cary, J. R., and R. G. Littlejohn, 1983, Ann. Phys. 共N.Y.兲 151,
1.
Cary, J. R., and S. G. Shasharina, 1997, Fiz. Plazmy 23, 552
关Plasma Phys. Rep. 23, 509 共1997兲兴
Chan, V. S., C. M. Greenfield, L. L. Lao, T. C. Luce, C. C.
Pettu, G. M. Staebler, and DIII-D Team, 2000, Nucl. Fusion
40, 1137.
Chance, M. S., M. S. Chu, M. Okabayashi, and A. D. Turnbull,
2002, Nucl. Fusion 42, 295.
Cheng, C. Z., and M. S. Chance, 1986, Phys. Fluids 29, 3695.
Cheng, C. Z., L. Chen, and M. S. Chance, 1985, Ann. Phys.
共N.Y.兲 161, 21.
Chirikov, B. V., 1979, Phys. Rep. 52, 263.
Clebsch, A., 1859, J. Reine Angew. Math. 56, 1.
Connor, J. W., 1984, Plasma Phys. Controlled Fusion 26, 1419.
Connor, J. W., and R. J. Hastie, 1975, Nucl. Fusion 15, 415.
Connor, J. W., R. J. Hastie, and J. B. Taylor, 1978, Phys. Rev.
Lett. 40, 396.
Diamond, P. H., and Y.-B. Kim, 1991, Phys. Fluids B 3, 1626.
Dorland, W., F. Jenko, M. Kotschenreuther, and B. N. Rogers,
2000, Phys. Rev. Lett. 85, 5579.
Dreicer, H., 1960, Phys. Rev. 117, 329.
Drummond, W. E., and D. Pines, 1962, Nucl. Fusion Suppl. 3,
1049.
Ferrari, A., 1998, Annu. Rev. Astron. Astrophys. 36, 539.
Fisch, N. J., 1978, Phys. Rev. Lett. 41, 873.
Fisch, N. J., 1987, Rev. Mod. Phys. 59, 175.
Fisch, N. J., and A. H. Boozer, 1980, Phys. Rev. Lett. 45, 720.
Fisch, N. J., and M. C. Herrmann, 1994, Nucl. Fusion 34, 1541.
Fitzpatrick, R., 2001, Phys. Plasmas 8, 871.
Fitzpatrick, R., and T. C. Hender, 1991, Phys. Fluids B 3, 644.
Freidberg, J. P., 1987, Ideal Magnetohydrodynamics 共Plenum,
New York兲.
Furth, H. P., J. Killeen, and M. N. Rosenbluth, 1963, Phys.
Fluids 6, 459.
Furth, H. P., J. Killeen, M. N. Rosenbluth, and B. Coppi, 1965
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1139
in Plasma Physics and Controlled Fusion Research 1964 共In-
ternational Atomic Energy Agency, Vienna兲, Vol. 1, p. 103.
Galeev, A. A., and R. Z. Sagdeev, 1968, Sov. Phys. JETP 26,
233.
Gardner, C. S., 1963, Phys. Fluids 6, 839.
Garnier, D. T., J. Kesner, and M. E. Mauel, 1999, Phys.
Plasmas 6, 3431.
Garren, D. A., and A. H. Boozer, 1991a, Phys. Fluids B 3,
2805.
Garren, D. A., and A. H. Boozer, 1991b, Phys. Fluids B 3,
2822.
Gentle, K. W., 1995, Rev. Mod. Phys. 67, 809.
Glasser, A. H., J. M. Greene, and J. L. Johnson, 1975, Phys.
Fluids 18, 875.
Goedbloed, J. P., 1998, Phys. Plasmas 5, 3143.
Goldston, R. J., and P. H. Rutherford, 1995, Introduction to
Plasma Physics 共Institute of Physics, Philadelphia兲.
Goldston, R. J., R. B. White, and A. H. Boozer, 1981, Phys.
Rev. Lett. 47, 647.
Grad, H., 1973, Proc. Natl. Acad. Sci. U.S.A. 70, 3277.
Grad, H., and H. Rubin, 1959, in Proceedings of the 2nd United
Nations International Conference on the Peaceful Uses of
Atomic Energy, Geneva 1958 共Columbia University, New
York兲, Vol. 31, p. 190.
Greene, J. M., 1979, J. Math. Phys. 20, 1183.
Hamada, S., 1962, Nucl. Fusion 2, 23.
Hastie, R. J., 1997, Astrophys. Space Sci. 256, 177.
Hawryluk, R. J., 1998, Rev. Mod. Phys. 70, 537.
Hawryluk, R. J., S. Bartha, W. Blanchard, M. Beer, M. G. Bell,
R. E. Bell, H. Berk, S. Bernabei, M. Bitter, B. Breizman, N.
L. Bretz, R. Budney, C. E. Bush, J. Callen, R. Camp, et al.,
1998, Phys. Plasmas 5, 1577.
Helander, P., and D. J. Sigmar, 2002, Collisional Transport in
Magnetized Plasmas 共Cambridge University, Cambridge,
England兲.
Hinton, F. L., and R. D. Hazeltine, 1976, Rev. Mod. Phys. 48,
239.
Hirshman, S. P., W. I. van Rij, and P. Merkel, 1986, Comput.
Phys. Commun. 43, 143.
Horton, W., 1999, Rev. Mod. Phys. 71, 735.
Hudson, S. R., D. A. Monticello, A. H. Reiman, A. H. Boozer,
D. J. Strickler, S. P. Hirshman, and M. C. Zarnstorff, 2002,
Phys. Rev. Lett. 89, 275003.
Huebener, R. P., 1979, Magnetic Flux Structures in Supercon-
ductors 共Springer, Berlin兲, pp. 121–130.
Hutchinson, I. H., 2002, Principles of Plasma Diagnostics, 2nd
ed. 共Cambridge University, Cambridge, England兲.
ITER Physics Basis Editors, ITER Physics Expert Group
Chairs and Co-Chairs, and ITER Joint Central Team and
Physics Integration Unit 1999, Nucl. Fusion 39, 2137.
ITER Physics Expert Group on Confinement and Transport,
ITER Physics Expert Group on Confinement Modelling and
Database, and ITER Physics Basis Editors, 1999, Nucl.
Fusion 39, 2175.
ITER Physics Expert Group on Diagnostics and ITER Physics
Basis Editors, 1999, Nucl. Fusion 39, 2541.
ITER Physics Expert Group on Disruptions, Plasma Control,
and MHD, and ITER Physics Basis Editors, 1999, Nucl.
Fusion 39, 2251.
ITER Physics Expert Group on Divertor, ITER Physics Ex-
pert Group on Divertor Modelling and Database, and ITER
Physics Basis Editors, 1999, Nucl. Fusion 39, 2391.
ITER Physics Expert Group on Energetic Particles, Heating
1140 Allen H. Boozer: Physics of magnetically confined plasmas
and Current Drive, and ITER Physics Basis Editors, 1999a,
Nucl. Fusion 39, 2471.
ITER Physics Expert Group on Energetic Particles, Heating
and Current Drive, and ITER Physics Basis Editors, 1999b,
Nucl. Fusion 39, 2495.
Jackson, J. D., 1999, Classical Electrodynamics, 3rd ed. 共Wiley,
New York兲, p. 192.
Jarboe, T. R., I. Henins, A. R. Sherwood, C. W. Barnes, and H.
W. Hoida, 1983, Phys. Rev. Lett. 51, 39.
Jenko, F., W. Dorland, M. Kotschenreuther, and B. N. Rogers,
2000, Phys. Plasmas 7, 1904.
Kadomtsev, B. B., and O. P. Pogutse, 1970, in Reviews of
Plasma Physics, edited by M. A. Leontovich 共Consultants Bu-
reau, New York兲, Vol. 5, p. 303.
Kamada, Y., and the JT-60 Team, 1999, Plasma Phys. Con-
trolled Fusion 41, B77.
Karzas, W. J., and R. Latter, 1961, Astrophys. J., Suppl. 6, 167.
Keilhacker, M., A. Gibson, C. Gormezano, and P. H. Rebut,
2001, Nucl. Fusion 41, 1925.
Keilhacker, M., and the JET team, 1999, Plasma Phys. Con-
trolled Fusion 41, B1.
Kerst, D. W., 1964, J. Nucl. Energy, Part C 4, 253.
Kesner, J., A. N. Simakov, D. T. Garnier, P. J. Catto, R. J.
Hastie, S. I. Krasheninnikov, M. E. Mauel, T. S. Pedersen,
and J. J. Ramos, 2001, Nucl. Fusion 41, 301.
Kolmogorov, A. N., 1954, Dokl. Akad. Nauk SSSR 98 共4兲, 527.
Kovrizhnikh, L. M., 1969, Sov. Phys. JETP 29, 475.
Kruskal, M. D., 1962, J. Math. Phys. 3, 806.
Kruskal, M. D., and R. M. Kulsrud, 1958, Phys. Fluids 1, 265.
Landau, L. D., 1946, J. Phys. 共Moscow兲 10, 25.
Lazarus, E. A., J. B. Lister, and G. H. Neilson, 1990, Nucl.
Fusion 30, 111.
Lichtenberg, A. J., and M. A. Lieberman, 1992, Regular and
Chaotic Dynamics, 2nd ed. 共Springer, New York兲.
Lieberman, M. A., and A. J. Lichtenberg, 1994, Principles of
Plasma Discharges and Materials Processing 共Wiley-
Interscience, New York兲.
Lin, Z., W. M. Tang, and W. W. Lee, 1995, Phys. Plasmas 2,
2975.
Lindl, J. D., 1998, Inertial Confinement Fusion 共AIP/Springer,
New York兲.
Littlejohn, R. G., 1981, Phys. Fluids 24, 1730.
Loarte, A., 2001, Plasma Phys. Controlled Fusion 43, R183.
Lundquist, S., 1951, Phys. Rev. 83, 307.
Lüst, R., and E. Martensen, 1960, Z. Naturforsch. A 15A, 706.
Lüst, R., and A. Schlüter, 1957, Z. Naturforsch. A 12A, 850.
Mathews, J., and R. L. Walker, 1964, Mathematical Methods of
Physics 共Benjamin, New York兲, p. 380.
McCormick, K., P. Gurgill, R. Burhenn, et al., 2003, J. Nucl.
Mater. 313-316, 1131.
Mercier, C., 1964, Nucl. Fusion 4, 213.
Merkel, P., 1987, Nucl. Fusion 27, 867.
Merkel, P., 1988, in Theory of Fusion Plasmas 1987, edited by
A. Bondeson, E. Sindoni, and F. Tryon 共Editrice, Bologna兲, p.
25.
Migliuolo, S., 1993, Nucl. Fusion 33, 1721.
Miyamoto, K., 1997, Fundamentals of Plasma Physics and
Controlled Fusion 共Iwanami Book Service Center, Tokyo兲.
Morozov, A. I., and L. S. Solov’ev, 1966, in Reviews of Plasma
Physics, edited by M. A. Leontovich 共Consultants Bureau,
New York兲, Vol. 2, p. 201.
Moser, J., 1967, Math. Ann. 169, 136.
Mynick, H. E., T. K. Chu, and A. H. Boozer, 1982, Phys. Rev.
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
Lett. 48, 322.
Neilson, G. H., A. H. Reiman, M. C. Zarnstorff, A. Brooks,
G.-Y. Fu, R. J. Goldston, L.-P. Ku, Z. Lin, R. Majeski, D. A.
Monticello, et al., 2000, Phys. Plasmas 7, 1911.
Nevins, W. M., 1998, J. Fusion Energy 17, 25.
Newcomb, W. A., 1959, Phys. Fluids 2, 362.
Newcomb, W. A., 1960, Ann. Phys. 共N.Y.兲 10, 232.
Northrop, T. G., 1963, The Adiabatic Motion of Charged Par-
ticles 共Interscience, New York兲.
Nührenberg, C., and A. H. Boozer, 2003, Phys. Plasmas 10,
2840.
Nührenberg, J., W. Lotz, P. Merkel, C. Nührenberg, U.
Schwenn, E. Strumberger, and T. Hayashi, 1995, Fusion Tech-
nol. 27, 71.
Ohkawa, T., 1970, Nucl. Fusion 10, 185.
Onsager, L., 1931, Phys. Rev. 38, 2265.
Ortolani, S., and D. D. Schnack, 1993, Magnetohydrodynamics
of Plasma Relaxation 共World Scientific, Singapore兲.
Parks, G. K., 1991, Physics of Space Plasmas 共Addison-Wesley,
Redwood City, CA兲.
Peng, Y.-K. M., 2000, Phys. Plasmas 7, 1681.
Petty, C. C., T. C. Luce, P. A. Politzer, M. R. Wade, S. L. Allen,
M. E. Austin, B. Bray, K. H. Burrell, T. A. Casper, M. S. Chu,
et al., 2000, Plasma Phys. Controlled Fusion 42, B75.
Pfirsch, D., and A. Schlüter, 1962, MPI/PA/7/62 共Max-Planck
Institut für Physik und Astrophysik, München兲.
Post, D. E., R. V. Jensen, C. B. Tarter, W. H. Grasberger, and
W. A. Lokke, 1977, At. Data Nucl. Data Tables 20, 397.
Prager, S. C., 1999, Plasma Phys. Controlled Fusion 41, A129.
Priest, E., and T. Forbes, 2000, Magnetic Reconnection: MHD
Theory and Applications 共Cambridge University, Cambridge,
England兲.
Raman, R., T. R. Jarboe, B. A. Nelson, V. A. Izzo, R. G.
O’Neill, A. J. Redd, and R. J. Smith, 2003, Phys. Rev. Lett.
90, 075005.
Renner, H., J. Boscary, H. Greuner, H. Grote, F. W. Hoff-
mann, J. Kisslinger, E. Strumberger, and B. Mendelevitch,
2002, Plasma Phys. Controlled Fusion 44, 1005.
Rogers, B. N., R. E. Denton, J. F. Drake, and M. A. Shay,
2001, Phys. Rev. Lett. 87, 195004.
Rosenberg, A. L., D. A. Gates, A. Pletzer, J. E. Menard, S. E.
Kruger, C. C. Hegna, F. Paoletti, and S. Sabbagh, 2002, Phys.
Plasmas 9, 4567.
Rosenbluth, M. N., and P. H. Rutherford, 1981, in Fusion, Vol.
I, Magnetic Confinement, edited by E. Teller 共Academic, New
York兲, p. 31.
Rutherford, P. H., 1973, Phys. Fluids 16, 1903.
Rutherford, P. H., and E. A. Frieman, 1968, Phys. Fluids 11,
569.
Ryutov, D. D., M. S. Derzon, and M. K. Matzen, 2000, Rev.
Mod. Phys. 72, 167.
Sasinowski, M., and A. H. Boozer, 1995, Phys. Plasmas 2, 610.
Shafranov, V. D., 1958, Sov. Phys. JETP 6, 545.
Sheffield, J., 1994, Rev. Mod. Phys. 66, 1015.
Spitzer, L., Jr., 1958, Phys. Fluids 1, 253.
Stern, D. P., 1970, Am. J. Phys. 38, 494.
Stix, T. H., 1992, Waves in Plasmas 共AIP, New York兲.
Strickler, D. J., L. A. Berry, and S. P. Hirshman, 2002, Fusion
Sci. Technol. 41, 107.
Swanson, D. G., 2003, Plasma Waves, 2nd ed. 共Academic, Bos-
ton兲.
Sykes, A., 2001, Plasma Phys. Controlled Fusion 43, A127.
 Allen H. Boozer: Physics of magnetically confined plasmas
Tajima, T., and K. Shibata, 1997, Plasma Astrophysics
共Addision-Wesley, Reading, MA兲.
Talmadge, J. N., V. Sakaguchi, F. S. B. Anderson, D. T. Ander-
son, and A. F. Almagri, 2001, Phys. Plasmas 8, 5165.
Tang, W. M., 2002, Phys. Plasmas 9, 1856.
Tang, X. Z., and A. H. Boozer, 2004, Phys. Plasmas 11, 171.
Tataronis, J., and W. Grossmann, 1973, Z. Phys. 261, 203.
Taylor, J. B., 1964, Phys. Fluids 7, 767.
Taylor, J. B., 1974, Phys. Rev. Lett. 33, 1139.
Taylor, J. B., and R. J. Hastie, 1968, Plasma Phys. 10, 479.
Terry, P. W., 2000, Rev. Mod. Phys. 72, 109.
Thiffeault, J.-L., and A. H. Boozer, 2003, Phys. Plasmas 10,
259.
Todd, A. M. M., M. S. Chance, J. M. Greene, R. C. Grimm, J.
L. Johnson, and J. Manickam, 1977, Phys. Rev. Lett. 38, 826.
Troyon, F., and R. Gruber, 1985, Phys. Lett. 110, 29.
van Rij, W. I., and S. P. Hirshman, 1989, Phys. Fluids B 1, 563.
Vedenov, A. A., E. P. Velikhov, and R. Z. Sagdeev, 1961, Nucl.
Fusion 1, 82.
Rev. Mod. Phys., Vol. 76, No. 4, October 2004
1141
Wagner, F., G. Becker, K. Behringer, et al., 1982, Phys. Rev.
Lett. 49, 1408.
Wesson, J., 2004, Tokamaks, 3rd ed. 共Oxford University, New
York兲.
White, R. B., 2001, The Theory of Toroidally Confined Plas-
mas, 2nd ed. 共Imperial College, London兲.
White, R. B., and M. S. Chance, 1984, Phys. Fluids 27, 2455.
Whiteman, K. J., 1977, Rep. Prog. Phys. 40, 1033.
Wolf, R. C., 2003, Plasma Phys. Controlled Fusion 45, R1.
Woltjer, L., 1958, Proc. Natl. Acad. Sci. U.S.A. 44, 489.
Wong, K.-L., 1999, Plasma Phys. Controlled Fusion 41, R1.
Yamada, H., A. Komori, N. Ohyabu, O. Kaneko, K. Kawa-
hata, K. Y. Watanabe, S. Sakakibara, S. Murakami, et al.,
2001, Plasma Phys. Controlled Fusion 43, A55.
Yoshizawa, A., S.-I. Itoh, K. Itoh, and N. Yokoi, 2001, Plasma
Phys. Controlled Fusion 43, R1.
Zarnstorff, M. C., L. A. Berry, A. Brooks, et al., 2001, Plasma
Phys. Controlled Fusion 43, A237.