You are on page 1of 164


Los Angeles

Plasma Wake-field Acceleration in the Blowout Regime

A dissertation submitted in partial satisfaction of the

requirements for the degree Doctor of Philosophy in Physics


Nikolai Barov

The dissertation of Nikolai Barov is approved.


David B. Cline


Claudio Pellegrini


Eli Yablonovitch


James B. Rosenzweig, Committee Chair

University of California, Los Angeles


This dissertation is dedicated to my parents.


List of Figures.......................................................................................................... vi

Vita.......................................................................................................................... xi

Abstract.................................................................................................................... xii

Chapter 1 Introduction to Plasma Wake-field Acceleration.................................... 1

1.1 Plasma Accelerators............................................................................... 2

1.2 Introduction to the blow-out regime PWFA.......................................... 6

1.3 Scaling with plasma density.................................................................. 17

Chapter 2 Numerical Simulations of PWFA Dynamics........................................ 21

2.1 NOVO and Modifications..................................................................... 21

2.2 Incorporating a solenoidal field in the model........................................ 27

2.3 PIC code................................................................................................. 28

Chapter 3 Drive Beam Radial Dynamics................................................................ 36

3.1 Analytical model of rarefaction............................................................. 38

3.2 Maxwell-Vlasov model of the drive beam........................................... 46

3.3 Matching............................................................................................... 54

3.3.1 A near-ideal match beam conditioner..................................... 56

3.3.2: Sinusoidal ramps................................................................... 58

3.3.3 No matching section............................................................... 63

Chapter 4 Applications of PWFA Technology....................................................... 65

4.1. A design for a 2.5 TeV collider............................................................ 65

Chapter 5 The AWA Facility.................................................................................. 71

5.1 Introduction........................................................................................... 71

5.2 Simulations of low-charge performance................................................ 74

5.3 Diagnostics............................................................................................ 84

5.3.1 Transverse profile diagnostics................................................ 84

5.3.2 ICT’s and Faraday cups.......................................................... 85

5.3.3 Prompt radiators and Streak camera measurements............... 87

5.3.4 Pepper-pot............................................................................... 89

5.4 AWA drive linac operation.................................................................... 99

5.5 Creation of a witness beam in the same cavity as the drive beam......... 104

Chapter 6 Drive Beam Focusing Experiment in the Blow-out Regime................. 109

6.1 The plasma source and modifications................................................... 111

6.2 Experimental Results............................................................................. 114

Chapter 7 Electron Acceleration Measurements in a Nonlinear Plasma Wake-field


7.1 High-resolution, broadband spectrometer.............................................. 128

7.2 Experimental results............................................................................... 133

7.3 Future experiments................................................................................. 137

Appendix A The Control and Data Taking System................................................. 144


Figure 1. A trailing beam accelerated in a series of plasma sections........................ 6

Figure 2. The drive and trailing beams in an underdense plasma, showing only the
first oscillation.[F-PWFA cartoon]............................................................................ 7

Figure 3. A configuration space plot of plasma electron positions from a particle in

cell (PIC) simulation.................................................................................................. 7

Figure 4. The longitudinal wake and initial bunch profile for a 20 nC beam in a
3 ×1013 plasma.......................................................................................................... 25

Figure 5. The beam longitudinal phase space after interaction with the plasma...... 26

Figure 6. The integration contours used in discretizing the Maxwell equations...... 29

Figure 7. On-axis and r = 75 mm longitudinal field from PIC simulation............. 33

Figure 8. Longitudinal phase space from PIC simulation....................................... 34

Figure 9. Configuration space plot of plasma electron position in PIC simulation 35

Figure 10. Evolution of the pinch point as a function of S , the propagation distance
in units of the betatron wavelength......................................................................... 45

Figure 11 Normalized beam spot area for a number of bunch charges.................. 50

Figure 12. Normalized beam spot area for a number of different emittances........ 51

Figure 13 Normalized beam spot area for a number of different plasma densities 52

Figure 14 Comparison of Maxwell-Vlasov equilibrium prediction with time-

dependent beam model (at z=10 and 20 cm into the plasma)................................ 53

Figure 15 Use of a plasma lens to match into a long plasma column.................... 57

Figure 16. Matching of the beam body with a ramp of length 1.63 cm
(k r = .964 cm − 1 ), corresponding to = /2 ........................................................ 61

Figure 17. Matching of the beam body with a ramp of length 3.33 cm
(k r = 0.472 cm −1 ), corresponding to = .......................................................... 62

Figure 18. Schematic of a - collider using a hardware transformer scheme... 69

Figure 19. The AWA drive linac............................................................................ 72

Figure 20. The AWA drive and witness beamlines (simplified view)................... 73

Figure 21. AWA RF system................................................................................... 73

Figure 22. AWA laser system block diagram........................................................ 74

Figure 23. Radial cross-section and electric field pattern of -mode 2 12 cell gun. 77

Figure 24. Standard case simulation output............................................................ 82

Figure 25. Longitudinal phase space after the linac............................................... 83

Figure 26. Test setup for the calibration of the ICT.............................................. 87

Figure 27. An electron beam intercepted by a Cerenkov radiator......................... 89

Figure 28. The emittance mask design.................................................................. 92

Figure 29. Background subtracted pepper-pot image of a 20 nC beam................. 94

Figure 30. Projected image from one row of beamlets and (smoothed) background
sampled between the rows....................................................................................... 96

Figure 31. The grid used for analyzing the emittance data..................................... 98

Figure 32. Laser transverse profile on cathode...................................................... 101

Figure 33. Longitudinal phase space of a drive and witness beam created by delaying
emission by 40 psec.................................................................................................. 106

Figure 34. Streak camera projected image of drive and witness beam.................... 107

Figure 35. The modified plasma source, shown without beam diagnostics............. 112

Figure 36. Diagnostics beam line and plasma cell................................................... 115

Figure 37. Cerenkov streak image showing nb ( x,t) , with earlier times (beam head) on
top............................................................................................................................ 122

Figure 38. Transverse beam profile after collimation, from digitized CCD camera
image of Cerenkov light........................................................................................... 123

Figure 39. Transmission data and simulations for an initially nearly -matched
case.......................................................................................................................... 124

Figure 40. Time-slice dependence of beam intensity and half-widths, form experiment
(exp.) and PIC simulation (sim.)............................................................................. 125

Figure 41. Dependence of distribution measures (peak, 80 th percentile, and mean)

as a function of n0 for -mismatched case.............................................................. 126

Figure 42. The spectrometer system, including the slit at the plasma anode,
quadrupole, bend magnet, ICT and phosphor screen................................................ 130

Figure 43. A simulation of trajectories in the spectrometer...................................... 131

Figure 44. The final position of a particle vs. ....................................................... 132

Figure 45. A simulated view of the spectrometer screen showing a y -focus for high
energy particles........................................................................................................ 133

Figure 46. Images of the energy spectrom taken at spectrometer focal plane. Image c)
is a no-plasma background....................................................................................... 136

Figure 47. Spectrometer phosphor image with the gray scale adjusted to accentuate

the high energy tail................................................................................................... 137

Figure 48. The laser splitter used to control the delay between the drive and witness
beams....................................................................................................................... 142


There are many people who helped make this experiment a success. First and

foremost among these is my advisor, Jamie Rosenzweig, who encouraged me to go ahead

with the plans which ultimately lead to the Argonne experiment. Along the way, he has

been a plentiful source of ideas and inspiration.

When initially starting out in the accelerator field, I received a lot of help from a

fellow graduate student, Gil Travish. I greatly enjoyed woking with him during the early

days of the Saturnus project.

After starting to work on the Argonne experiment, I owe much of my knowledge

about photoinjectors to Eric Colby. It has been a great experiemce to work with him.

I am greatful for the generosity of Jim Simpson, and later Wei Gai, who were

very accomodating with the scheduling accelerator time for the plasma experiment, and

with resources needed along the way. Wei has provided much coaching in the practical

aspects of conducting experiments. He has always encouraged me to follow my judgment

in the course of this work.

I am also thankful to the other members of the Argonne group that have helped

with the experiment, including Manoel Conde, John Power, George Cox, Richard

Konecny, and Paul Schoessow. From UCLA, Alex Murokh’s assistance with the

spectrometer design work is also appreciated


June 25, 1965 Born, Sofia, Bulgaria

1989 B.S., Physics, Minor in Mathematics

University of California, San Diego

1992-1998 Research assistant

Department of Physics and Astronomy
University of California, Los Angeles


N. Barov, M. E. Conde, W. Gai, and J. B. Rosenzweig, “Propagation of Short Electron

Pulses in a Plasma Channel”, Physical Review Letters, Vol. 80 (1), Page 81 (1998).

N. Barov, M. E. Conde, W. Gai, and J. B. Rosenzwei, “The UCLA/ANL Plasma

Wakefield Experiment”, presented at the Joint ICFA/JAERI-Kansai International
Workshop, Kyoto, Japan, July 1997.

N. Barov, M. E. Conde, J. B. Rosenzweig, G. Cox, W. Gai, R. Konecny, J. Power, P.

Schoessow, J. Simpson, “Beam Driven Plasma Wake-field Measurements at ANL”,
presented at Workshop on Second Generation Plasma Accelerators, Kardamyli, Greece,
June 1995.

N. Barov, M. E. Conde, J. B. Rosenzweig, P Schoessow, G. Cox, W. Gai, R. Konecny, J.

Power, J. Simpson, “Measurements of Plasma Wake-fields in the Blow-out Regime”, in
Proceedings of the 1995 IEEE Particle Accelerator Conference.

N. Barov and J. B. Rosenzweig, “Propagation of Short Electron Pulses in Underdense

Plasmas”, Phyical. Review E, Vol. 49, Page 4407 (1994).

C. Pellegrini, J. Sandweiss, and N. Barov, “Use of an Inverse Free Electron Laser in a

Linear Collider B Factory”, AIP proceedings, Advanced Accelerator Concepts, Fifth
Workshop, AIP conference proceedings (1992).


Plasma Wake-field Acceleration in the Blowout Regime


Nikolai Barov

Doctor of Philosophy in Physics

University of California, Los Angeles, 1998

Professor James B. Rosenzweig, Chair

The blowout regime is a limiting case of a nonlinearly excited plasma wave that

overcomes problems identified in past linear regime studies of the plasma wake-field

accelerator. Here, it is studied with the aid of simulations and by conducting experiments.

Fluid model and particle-in-cell simulations have been used to study the propagation of

drive pulses short enough to couple to the plasma mode, over distances comparable to the

energy depletion length. Evolution of the beam radial dynamics in the simulations, and

the associated wake-field phase shifts, outline phenomena important for achieving the

maximum energy gain for the accelerated electrons. An experiment to demonstrate long-term

drive beam propagation of a short electron pulse in an underdense plasma shows good

agreement with simulations, with minor differences thought to originate from the limitation

of simulating only a two-dimensional system. An experiment to measure electron

acceleration is part of an ongoing effort, and has shown a 2.6 MeV energy gain and a 22

MeV/m acceleration gradient.

Chapter 1. Introduction to Plasma Wake-field Acceleration

Particle accelerator technology based on microwave cavities has come to a highly

refined state after some fifty years of development. These devices are limited in peak

acceleration gradient due to material damage thresholds and ability to trap dark current,

both effects improving at higher frequencies. The acceleration gradient sets the overall

size of a linear collider, dominating the construction cost of the facility, while power

losses determine the operating expenses. While the use of higher operating frequencies

may yield a higher gradient, the variety of proposals for the next generation of linear

collider, spanning a range from 1.3 to 30 GHz, suggests that the advantages of this

approach are still unclear. The energy scale of interest for these machines is in the 0.5 to

2 TeV range, where electroweak symmetry breaking occurs. Additionally, as the history

of high energy physics shows, achievement of higher center-of-mass energy collisions is

always accompanied by new and unexpected discoveries. However, the cost of each of

these colliders is high enough to make funding by a single nation uncertain. This may

force a large international collaboration as the only setting in which to build such a

machine. In light of this conflict between future generation linear collider requirements

and the capabilities of the conventional technology, it has been argued that the field needs

an entirely new technology on which to base future machines. Research into advanced

and exotic particle acceleration techniques was begun with the goal of inventing just such

a technology.

Particle acceleration methods pursued by the advanced accelerator community

include laser, electron beam, and microwave-driven schemes based on plasmas, dielectrics,

metallic structures, and far-field (or vacuum) acceleration, such as reflecting a laser close

to an intense focus, and the inverse FEL (IFEL) accelerator. Already, some promising

applications emerging from this field have been in the area of electron sources, with a

transmission line-based pulsed system reported in [1] and a very promising plasma-based,

or cathodeless injection, electron source under investigation[2]. It will not be surprising to

see acceleration methods now termed exotic emerge as compact sources of 10-20 MeV

electrons for research, medical or industrial applications. Although the acceleration gradient

for some schemes is many orders of magnitude higher than the ~20 MeV/m conventional

linacs are operated in, the record having been set at 100 GeV/m with the laser-wake-field

accelerator, many more factors are involved in developing a collider-compatible device.

The additional requirements in these applications on beam quality preservation, mechanical

reliability, power efficiency and repetition rate translate to a much longer development

cycle for the new technology, perhaps on the scale of 20 years or more. The plasma

wake-field accelerator discussed in the rest of this Thesis, because it relies on a source of

high energy electrons for the driver, is probably limited only to collider applications. The

research into these devices usually proceeds in several stages, with 1st generation experiments

establishing proof of principle, 2nd generation demonstrating high gradient, and 3rd generation

concerned with beam injection and emittance preservation. The current state of development

of the plasma wake-field accelerator (PWFA) discussed in this Thesis, places it somewhere

around the 2nd generation mark, leaving many questions in terms of its ultimate suitability

in such an application unanswered. Although some of these areas are in desperate need of

attention, as is the practice in much of advanced accelerator research, potential problems

are only briefly summarized in the present work.

1.1 Plasma Accelerators

Plasmas first attracted interest as acceleration devices in the context of laser-driven

accelerators. With the help of a plasma, the very high peak powers at the focus of a high

power laser could drive longitudinal electric fields having the correct phase velocity for

particle acceleration. Examples of this are the plasma beat wave accelerator (PBWA),

and the laser wake-field accelerator (LWFA) proposed by Tajima and Dawson[3]. A

figure of merit for the electric field a plasma can sustain is given by the wavebreaking

limit, E0 = 4 n0 mec 2 , which is the roughly the field from a 100% sinusoidal modulation

in plasma electron density at the plasma wavelength. It is also a non-relativistic limit on

the peak electric field for trapping of plasma electrons, seen as a steepening and eventual

breaking of the wave in the longitudinal phase space. This field scales as the square root

of the plasma density, with 0.96 and 30 GeV/m for a 1014 and 1017 cm−3 plasma,

respectively. Fields of this magnitude can be achieved without the usual concerns about

material damage thresholds, because the plasma medium can be restored for each new


The concept of wake-field acceleration (as in LWFA) was borrowed from a parallel

line of research, that of trying to harness the energy a bunched beam loses in passage

through a microwave cavity. These losses are usually undesirable in a linac application,

and much effort is spent in trying to minimize them. When a significant amount of power

from one beam (the drive beam) can be coupled into the cavity’s fundamental mode, it

can result in acceleration for a second beam (the trailing beam) appropriately delayed in

time. In search of better transformer ratios, the ratio between the peak accelerating and

peak decelerating field, this idea has ultimately been replaced by the two-beam accelerator

(TBA) where the power is coupled to an off-axis resonant structure, such as the dielectric

step-up transformer being developed at the AWA[4], and the metallic-based CLIC

accelerator [5].

An electron bunch will also drive a wake-field response in a plasma - the idea

behind the plasma wake-field accelerator. Originally, the PWFA was construed in the

linear regime, where the plasma electron density perturbation is small. Chen et al.[6]

considered wake fields from a point-like electron beam, or the limiting case of a small

radius drive beam: r << k p−1 . The quantity k p−1 = 4 re n0 is called the collisionless skin

depth (re and n0 are the classical electron radius and unperturbed plasma density,

respectively), and corresponds to one spatial radian of a plasma oscillation at phase

velocity v ph = c . Later work generalized the theory to include finite in radius and length

bunches [7], leading to consideration of disk-like drive bunches ( r << k p−1 )[8]. Where the

plasma electron motion for pointlike bunches is mostly radial, it becomes mostly longitudinal

in the opposite, disk-like bunch limit. In both cases, the wake field Ez is proportional to a

convolution integral over the drive beam current I( ) over the longitudinal coordinate

moving with the beam = z − ct ,

Ez ∝ ∫ I( ′ )cos(kp ( − ′ ))d ′ , (1.1)

which is also of same form for the time-averaged wake-fields in microwave cavities and

dielectrics. For symmetric in time bunches, a theorem (the Wake-Field Theorem) limits

the transformer ratio to no higher than 2.0. Pursuit of higher transformer ratios in both the

linear regime and for the nonlinear plasma dynamics considered later in this chapter

usually takes the form of using triangular bunches, with a slow ramp-up in the current

followed by a sharp cut-off at the tail.

According to reference [26], plasma wake-fields were first demonstrated by Berezin

et al. [9] with a 0.4 nC pulse train generating a 0.25 MeV/m acceleration gradient in a

n0 = 1011 cm −3 plasma. Bunch train excitation of a wake field, as is intuitive from the

form of Eqn. 1.1, is most efficient when the bunches are spaced apart by one plasma

period p . The single drive bunch experiment by Rosenzweig et al. [10] saw a 1.6

MeV/m wake-field, ultimately driving the plasma outside of the linear regime [11].

Experiments were also carried out in Japan by Nakajima et al. [12], demonstrating a 30

MeV/m acceleration gradient, and Nakanishi et al. [13], using bunches from two identical

linacs for the drive and trailing beams. At the time of this writing, several experiments

aimed at achieving higher gradients and a better understanding of the PWFA mechanism

are planned around the world. One proposal uses the 46 GeV beam of the Stanford Linear

collider to drive wake-fields in a 0.2 −1× 1015 cm−3 plasma[14], with the aim of measuring

a 1 GeV/m acceleration gradient. Additional experiments have been proposed at UCLA[15],

Fermilab[16], and LANL[17], with further experiments planned in Japan, as well as a

bunch train-driven experiment in Novosibirsk[18].

In addition to supporting very high fields, an accelerator based on plasma technology

must be scalable to very high final energies. As in conventional accelerator based

colliders, a wake-field based collider will consist of a large number of acceleration

sections, as shown in Figure 1, the trailing beam receiving an energy gain as it passes

through each one. A plasma accelerator is inherently a high frequency, short wavelength

device,which makes it harder to synchronize the phase fronts in the individual acceleration

sections. Current technology for drive beam creation translate to a plasma density centered

around 1014 cm−3 (wavelength p = 3 mm ), as opposed to the ≈ 1017 cm −3 densities of the

laser-driven schemes. At these longer wavelengths, the accelerating wave fronts in

individual PWFA sections are easier to synchronize in time.

Figure 1. A trailing beam accelerated in a series of plasma sections.

1.2 Introduction to the blow-out regime PWFA

In 1991 Rosenzweig, Breizman, Katsouleas and Su [19] suggested an operating

regime, the “blow-out” regime, in which the drive beam is focused well enough ( r << k p−1 )

and dense enough that the plasma electrons are blown out of a region encompassing the

drive beam radius, and an ion channel is formed in the plasma, as in Fig 2. For beams that

are not too short, this tends to happen very close to the beam density needed to make the

plasma underdense (nb > n0 ), so that one can talk about an underdense PWFA and the

blow-out regime almost interchangeably. With even higher nb / n0 values, allowed by

advances in photoinjector technology, it is possible to exceed the wavebreaking limit, E0 ,

by almost a factor of 2 in some simulations. This is not surprising, since this limit comes

from a 1-D treatment of a nonlinear electrostatic wave. In the underdense, as well as in

the overdense regime, the plasma response generates strong focusing forces which affect

the drive and trailing beams. For the underdense case, this focusing is of a type much

better suited to so-called 3nd generation PWFA experiments, where preservation of the

quality of the trailing beam is important.

Figure 2. The drive and trailing beams in an underdense plasma, showing only the first

Figure 3. A configuration space plot of plasma electron positions from a particle in cell
(PIC) simulation. Parameters are the same as in the Chapter 2 example, except for using
twice as much drive beam charge, 40 nC.

Although the plasma electron dynamics are very nonlinear, the blow-out regime

presents the trailing beam with wake-fields that have a greater amount of linearity. This

behavior can easily be illustrated with the expression for the radial force acting on the

beam (and assuming that the ions are immobile),

4 e2 r
Fr = ∫ r(n (1− z,e ) − ni )dr , (1.2)
r 0 e

where ne and ni are the plasma electron and ion densities, and c z,e is the average

plasma electron longitudinal velocity. This formula is only exact for an ultrarelativistic

rigid beam so that the solution is a function of only the longitudinal position within the

bunch, = z − ct , and has no explicit time dependence. For an ion channel, this force can

be very strongly focusing; this is called ion channel focusing [30][20][21]. Integrating

this equation up to the channel radius rchan yields the focusing gradient in this regime,

Fr = −2 nie 2r (r < rchan ), (1.3)

which has the desired linear form. This linearity is not guaranteed in the overdense

regime because variations in ne and z ,e within the trailing beam volume will add a

time-dependence to the focusing force as well as sextupole and higher order terms,

adversely affecting the emittance. One further consequence of linearity becomes apparent

on applying the Panofsky Wenzel theorem [22] governing the longitudinal and radial


Fr Fz
= . (1.4)

Unlike the linear regime which has a changing with focusing gradient, this dependence

is not present in the blowout regime, hence an acceleration gradient independent of the

radius. Again, this is not guaranteed in the linear regime, with particles at larger betatron

orbits gaining less energy.

For the blowout regime, the only remaining variation of the wake-field in the

channel is a dependence of the acceleration gradient on , also present in conventional

linacs, and one which can be reversed.

Ion channel focusing is very strong because it is a first-order effect, unlike alternate-

gradient quadrupole focusing. It is this property that plays an important role in maintaining

the drive beam density in both underdense and overdense schemes. In attempting to

match a beam to the ion channel, the beta-function, eq = eq

/ at equilibrium, where

the beam envelope will not have any betatron oscillations is given by eq = (Fr r) mc 2 ,


eq = (2 ren0 / ) −1/2 . (1.5)

For n0 = 1014 cm −3 and = 40 , the -function is 4.7 mm, so a matched beam size will

increase by a factor of 2 over such a length in the absence of focusing. This matching

condition will only be true for the later parts of the beam, where the channel is fully

formed, leaving beam particles closer to the leading edge with less of a focusing force.

This leads to a dynamic process of radial spreading or erosion near the beam head, a topic

discussed in Chapter 3, which deals more generally with underdense regime propagation

of short, symmetric pulses. The conclusion of this work is that effective beam propagation

will happen only when the beam density exceeds the background plasma density by a

factor in the region of nb n0 = 2-4, assuming the beam radius to be given by the matching

condition of Eqn. 1.5. This topic is felt to be quite important for effective, as well as

predictable, plasma wake-field creation, and is picked up again in the experimental part

of the Thesis. Chapter 6 presents the results of an experiment devised to test this model

of drive beam propagation.

Ion Motion

For the conditions achievable in the experiments discussed in the later chapters,

the plasma ion motion is small enough to be of negligible effect. For this reason, ion

motion is neglected for most of the remainder of this work. However, one can easily

envision cases where ion motion is large enough to significantly modify the plasma and

beam dynamics. Since the force on the ions is toward the axis, the tendency will be

toward increased ion density, along with a nonlinear radial profile, with the same

consequences for the focusing force as when the electrons remain within the channel. The

equation governing the ion motion, (analogous to Eqn. 3.5 to be covered in Chapter 3),

r ′′ + k2 r = 0 , has,

2 e 2nb
k2 = (1.6)
mi c2

While the emittance of the drive beam can always be artificially inflated to keep its

density from being too large, these dynamics have perhaps the most serious consequences

for a high energy trailing beam, whose matched density can easily exceed 1 ×1018 cm −3

for 1 TeV and 1 nC. For hydrogen ions, this can easily lead to density perturbations of

order unity within the trailing beam volume, an obviously unacceptable situation. In this

case, ion density at the back of the bunch will be given by,

ni ,0
ni ≅ (1.7)
 m N r n 1/2

cos2   8re e b e o  
  mi 2 n  

In minimizing this density perturbation, the ratio Ne / n ought to be considered a constant

set by electron source dynamics: if less particles per bunch are used it is usually desirable

to make use of the smaller available emittance. The ion density perturbation can also be

reduced by using a heavier species, and by decreasing n0 or z .

Beam Loading

A calculations of the maximum number of particles that can be accelerated in the

wake-field of an overdense regime-plasma was performed by Katsouleas, et al. [23]. In

this regime it’s possible to extract nearly 100% of the wave energy, but at the expense of

an equally large energy spread in the trailing beam, or no accelerating force for the very

last particles. To minimize the energy spread with smaller amounts of beam loading, they

suggest using a reverse-triangular trailing beam, with a diminishing intensity toward the

back. This paper also examines the emittance dilution due to off-axis variations in the

loaded wake, prescribing use of a wide drive beam and narrow witness beam as the best

way to overcome these limitations. This really amounts to calculating the effect of the

trailing beam’s self-wake on its emittance and energy spread.

For the blowout regime, the radial effects are nonexistent, and only the longitudinal

wake profile is of interest. For a Gaussian in time beam, this self-wake will be a nonlinear

function of longitude, and is thus undesirable in terms of longitudinal emittance preservation.

A short, constant intensity beam will have a self-wake much closer to being linear; the

beam charge can further be adjusted for this to cancel the slope in the plasma wave.

Although a more careful study of these effects ought to be done for the underdense case,

beam loading at the 10% level is probably feasible.

3-D effects and scattering

While the discussion has until now been limited to cylindrically symmetric systems,

treating the full 3-D system exposes a richer set of effects. Because 3-D capable computer

simulations have a greater associated operating costs and complexity, these phenomena

have not yet been studied exhaustively.

Interaction with a wake-inducing device, whether it is a linac cavity or a plasma,

can couple power not only into the fundamental mode but also into deflection modes,

typically leading to unstable beam propagation. For an overdense plasma, this comes in

the form of the two-stream instability, which is partially slowed because the nonlinear

focusing acts to decohere a transverse impulse on the beam. In the underdense case the

phenomenon is called the electron hose instability [24], with the unstable behavior being

absolute, or growing exponentially at a fixed position in the beam. The implications for

a plasma accelerator were studied by Krall and Joyce [25], simulating a long (~ 3 p )

beam, thus chosen to create a large transformer ratio. They report visible deflections at

the tail of the 20 MeV drive beam after only 5 cm in the plasma, with the tail becoming

completely destroyed in 10 cm. They conclude that the electron hose mechanism rules

out the possibility of using such a long beam as a PWFA driver.

It is now interesting to write down an estimate for the amount of hosing for a

short, symmetric beam depositing most of its energy in a long plasma column. Eventual

uses of this technology make it interesting to consider a ≥ 1 GeV drive beam in a ≥ 1 m

long plasma. The effective length for the bunch, 2/ p, can be split up into two halves.

The beam head and pinch region can be assumed to be in the two-stream regime, and the

tail in the electron hose regime. To avoid calculating the more complicated two-stream

interaction, a lower limit for the amount of hosing at the tail can be accomplished by

assuming the leading half of the beam not to deflect. The deflection amplitude of the

remaining half can be evaluated using the formula from [25] for the number of e -foldings

Ne ,

Ne = (3/2 )3 / 2 ( 1/ )1 / 6 [(ct + ) / ] (− )
1/3 2/3
p / p , (1.8)

For a GeV beam propagating 1.0 meter in a 1 ×1014 plasma, the number of e-foldings is

Ne =5.85, meaning a response which exceeds the linear growth rate due to transverse

beam deconfinement. The limit on the permissible amount of hosing, as will be shown

below, is set by synchrotron losses of the trailing beam, which forces very tight tolerances

on the wake-field structure there. To be successful, plasma wake-field based collider

technology must incorporate a suppression scheme for this instability. Whittum et al. [24]

suggest several methods which might be of some help. The equations leading to the

instability include two frequencies, the betatron frequency ck and the plasma dipole

mode frequency 0 = p / 2 . The growth rate can be slowed by detuning either k or

0 , with 0 being the more effective method. One suggestion offered in this reference is

to vary the radius of the conducting wall surrounding the plasma, leading to a modified

frequency ′0 = 0 (1+ b 2 /2 R2 ) where b is the channel radius and R is the wall radius.

To be effective, any such detuning must happen on a length scale shorter than the growth

rate of the instability. The topic of electron hose suppression obviously requires further

study and will most likely modify the current set of ideas about how to build such a

plasma based accelerator. It should be noted that the treatment above ignores the role of

the return current, and the fact that the channel radius is increasing close to the beam

head, both of which may have an influence on the actual growth rate, and possibly

diminishing it. Unlike the case of a beam traversing a set of microwave cavities, this

system has a growth rate independent of the charge, due to the change in coupling with a

larger channel radius. The stability analysis should also include the detuning effect brought

on by the rapid variation in beam energy as it is deposited into the plasma wave. The

relevant parameter is the fractional energy loss on the betatron length scale, leading to

damping when K 2 ′ < 1.

The electron-hose instability can also be diminished by use of a much shorter

( z << k p−1 ) drive beam. In fact, a review article by Esarey, Sprangle, Krall and Ting [26]

hints that given short enough pulses, the instability entirely vanishes. The disadvantages

of this approach include a decreased transformer ratio and greater difficulty in creating

the ion channel quickly enough to aid in beam propagation.

Synchrotron losses

When there is a high energy (~TeV) beam accelerated in this system, any transverse

fields will cause it to synchrotron radiate. These fields may be the result of electron-hosing,

a misalignment between the trailing beam and plasma channel, or, to a lesser extent, the

finite radial size of the trailing beam. At high energies, the losses can become very large

as in the result of Montague and Schnell [27] for a 1 TeV beam misaligned by 100

microns for =1 m (1 ×1014 plasma), which suffers a betatron orbit averaged energy loss

of 76 MeV/m (152 MeV/m peak), or about an order of magnitude less than the wave-breaking

field at this density. Misalignments can result from errors in the drive beam’s initial state,

either in position or momentum. Further, a trailing beam which has begun to oscillate in

one plasma section (as in Fig 1) will keep oscillating in the remaining sections even

without further drive beam errors. In this way, the steering errors accumulated throughout

the process of acceleration will add stochastically, possibly leading to an unmanageably

large final error from the point of view of the collider interaction point requirements.

The electric and magnetic fields that cause the ∆E = 152 MeV/m deceleration are

comparable to fields that cause beamstrahlung in the process of colliding e+/e- bunches.

At very high energies, one cause for energy spread in collision point dynamics calculations

is the fluctuation in the number of photons emitted, N , given in reference [28],

5 ∆s
N = (1.9)
2 3

where is the fine structure constant, is the bend radius and ∆s is the length of the

orbit in the magnetic field. For the above 1 TeV case, N = 5.7 after passing through a 1

m of plasma, assuming a paraxial ray displaced by 100 microns. The approximate expression

for the energy spread E is given by,

 4.3

≅ ∆E   (1.10)
N 

and has a value of E = 132 MeV, indicating that some electrons lose much more than

the mean because the energy spread is governed by high energy photons which are

relatively few in number. The implication is that, to avoid large energy spread in the

trailing beam, any transverse misalignments would have to be very tightly controlled, to

better than about 6 µm in order to keep the final energy spread below 200 MeV.

A certain amount of synchrotron radiation will also result from an on-axis beam’s

finite transverse size[27], but these effects are much smaller and can usually be neglected.

Emittance growth through scattering

Surprisingly, the emittance increase through Coulomb scattering can be very small,

even as the trailing beam propagates through as much as 1 kilometer of plasma. The

calculation by Montague and Schnell [27] for a PBWA case closely follows the treatment

of J. D. Jackson [29] for the scattering of fast particles by atoms. Scattering calculations

have a logarithmic dependence on the maximum distance over which the Coulomb force

can act. For scattering by atoms, this distance is the atomic radius. In a plasma, the

calculation uses the screening distance, d . The case of an ion channel is slightly different,

since there are no electrons in the channel to perform the screening, and the force from an

ion can still act on a distance scale comparable to the channel radius, rchan . The expression

for the mean square angle (in one plane) after many collisions in a distance dz is:

 2e 2   L 
〈 x ′ 〉 = n0
ln , (1.11)
dz scatt  pc   R 

where R is the effective radius of the proton and L is the maximum length over which

the Coulomb force can act. The form of this expression masks the quantum mechanical

arguments used in deriving it: for a high energy beam, the minimum scattering angle is

governed by the uncertainty principle and the maximum is limited by considerations of

the DeBroigle wavelength of the electron.

The evolution of the emittance will depend on the amount of angular spread

already present in the beam, and the orientation of the transverse phase ellipse, through

the factor of beta in the formula [27]:

d n d
= 〈 x ′ 2 〉. (1.12)
dz 2 dz scatt

Assuming that the beam is -matched, this equation can be integrated between the initial

and final relativistic factor, i and f ,

( ) ln(r
re 2 ren0
∆ = − R) . (1.13)

n f i chan

For a 1 TeV accelerator with n0 = 1 ×1014 and an acceleration gradient of 1 GeV/m

( ′ = 20 cm −1 ), and assuming rchan ≅ 100 µm gives,

∆ n = 7 × 10 −9 cm , (1.14)

which is several orders of magnitude below the beams now achievable with photoinjector

technology with 1 nC of charge, 0.5 ×10 − 4 cm . However, the physics requirements at the

interaction point of future colliders may dictate even lower emittances. It may also be

desirable to accelerate a beam to even higher energies, where the emittance increase will

be greater due to the square root dependence of the result on f .

Where the above calculation has been performed for Hydrogen, plasmas using a

gas with higher Z can be formed with only partial ionization. In the case of Ar++, the

Coulomb force near the nucleus will be Z times larger than assumed in equation 1.11.

Here, the dominant effect can be approximated by the formula for scattering off neutral

atoms, given by multiplying the earlier expression by ≈ Z 2 (and using ni = n0 /2 ),

 2Ze 2   a 
′ 2 〉 = ni 
〈 xscatt ln , (1.15)
dz  pc   R 

where L has been set to a ( ≅ 1.4Z −1/3 h2 / me 2 ), the atomic radius. The same 1 TeV

collider example will then have ∆ n = 2.5 × 10 −6 cm, which may be too large for some

applications. This formula agrees with the result of Buchanan [30], which uses similar

assumptions, but with the refinement that particles scattered at large angles are considered

to be lost from the beam, and do not contribute to the emittance.

1.3 Scaling with plasma density

The equations governing the plasma dynamics allow a simple scaling to transform

a solution at one n0 value to a different set of conditions. This transformation is very

useful when evaluating the effect of changing one or several of the parameters of the

beam-plasma system. Scaling arguments allow a quick evaluation of the larger parameter

space involved with choosing the configuration of the 2.5 TeV collider in Chapter 4, and

are also helpful to motivate the discussion in Chapter 5 about the operating point for the

AWA linac, pointing out the importance of shorter bunches.

For a rigid driver, the solution for the plasma dynamics can be obtained by

solving the equations of motion for the plasma electrons (the ion motion is neglected)
r r 1 r r r
t c
( )
= e  E (ri ) + vi × B( ri )  , (1.16)

where the subscript i is over all the particles. The solution will also involve the Maxwell

equations which can be rewritten in dimensionless units, following reference [32], with

r ′ = kp r , E ′ = −E / r n0 mc2 , H ′ = −H / r n0 mc2 , t ′ = pt , and j ′ = − j /(en 0c) , (where

j = ev (r − ri )). This set of equations then become,

pi′ r r r
= E ′ + vi′ × B ′ (1.17)

∇′ × E ′ = − (1.18)

∇′ × H ′ = j ′ + . (1.19)

The solution to these equations will yield the plasma electron density in units of the

initial density n0 , and fields E ′ and B′ in units of the wave-breaking field E0 . When

translating back to the original coordinates, there is a freedom in the choice for n0 , so that

for every solution in the dimensionless variables there exists a family of solutions in the

un-primed coordinates. This allows the scaling of a known solution at one n0 to a

different density, provided that they correspond to the same fundamental dynamics in the

dimensionless units.

A scaling of this type can be used to explore the trade-offs of a plasma accelerator

on different length scales. From the drive beam’s point of view, the interaction in the

primed variables can be kept the same if the beam dimensions are kept the same in units

of k p−1 and the ratio nb / n0 is preserved. The drive beam charge must scale like a density

over a volume (Q ∝ nb / L3 ) or,

Q ∝ n0−1 / 2 . (1.20)

Higher plasma densities, therefore require less charge. From the scaling for the fields, the

electric field per unit drive beam charge obeys the following scaling,

Ez / Q ∝ n0 (1.21)

a result which strongly favors higher plasma densities and, therefore, short wavelengths

for the driver. Noting that the inverse square of the bunch length obeys the same scaling,
z ∝ n0 , reducing the bunch length by a factor of two will generate an acceleration field

four times higher. The emittance to charge ratio required to match the beam to the same

radius in units of k p−1 at different values of n0 remains constant as a function of scaling.

n Q = const. (1.22)

Since this ratio is often used as a figure of merit for photoinjector beams, higher plasma

densities do not require unreasonable values for the emittance. This expression should not

be too surprising, given the fact that it can be derived from treating the beam as simply

another plasma species.


This chapter has discussed much of the physics topics impacting the eventual use

of this technology as a particle accelerator. The successful evolution of this idea will have

to operate in a regime that addresses beam hosing, synchrotron losses, emittance growth,

and ion motion. Most of these topics do not play a very large role in the current set of

experiments aimed at showing that blowout-like conditions are achievable with present

technology and can achieve the predicted acceleration gradients. Except for some difficult

to control three-dimensional effects, these experiments ought to closely follow the

predictions of the two-dimensional simulation methods presented in the following chapter.

Examining the results of these simulations will reveal more of the character of the

blowout regime plasma accelerator.

Chapter 2. Numerical Simulations of PWFA Dynamics

The underdense PWFA has been numerically simulated with a variety of codes.

Rosenzweig, et al.[19] made use of NOVO and ISIS. Katsouleas[31] used MAGIC. Krall

and Joyce[25] studied radial equilibrium and electron-hose using FRIEZR, and a related

3-D code, ELBA. Additional codes capable of treating this problem include ARGUS and


The present work initially relied on the fluid code NOVO [32], adding modifications

to allow for different models for the drive beam. However, the fluid equations used to

model the plasma electrons no longer treat these dynamics accurately given the rapid

variations in density and velocity at the channel radius. Despite these inaccuracies, this

code runs much faster than a particle-in-cell (PIC) code and is valuable for quickly

running a large number of cases.

To improve on the accuracy of NOVO, as well as to gain the capability of a

time-dependent solution, where the electron beam encounters a plasma of a finite size, a

PIC code was developed.

2.1 NOVO and Modifications

NOVO was developed by Breizman and Chebotaev [32], and named after the

University of Novosibirsk. This code treats the plasma electrons as a cold fluid; the ions

are assumed to be stationary. These electrons are described by a velocity and a density

field, defined on a rectangular mesh. The simulation variables in this code are unitless,

arrived at by transforming the physical variables into dimensionless units according to:

r ′ = kp r , E ′ = −E / r n0mc 2 , H ′ = −H / r n0 mc2 , t ′ = pt In the new units, the fluid

p r r r 1 r r
+ ( v ⋅∇ ) p = −e E + (v × H) , (2.1)
t c

becomes (dropping the primes for the new variables),

+ Vr r = 1 + 2Vz − Vr 2 [(1+Vz − Vr 2 )(Er − H) + H + Vr E z ] ,

Vr V
t r

Vz V
+ Vr z = 1 + 2Vz − Vr 2 [(1+ 2Vz )E z + Vr H + Vr Vz (Er − H)] , (2.3)
t r

where Vr = r /(1− z ), and Vz = z /(1− z ). An additional quantity N = ne (1− z ) is

defined for solving the continuity equation:

N 1
+ (rNVr ) = 0 . (2.4)
t r r

For a rigid, ultrarelativistic driver, all field quantities can be assumed to depend

only on = z − ct , or that d / dt = 0. With this simplifying assumption, the transformed

Maxwell equations become,

(Er − H) = −NVr (2.5)


Ez = NVr (2.6)

(rH) = (NVr ) + (NVz ) + j b . (2.7)
rr r t r r

The solution progresses from the front of the bunch backwards, as a function of decreasing

, where the discretized equations link the solution at (r, ) and (r, − ∆ ), where ∆ is

the longitudinal mesh size.

The original version of the code employed a rigid drive beam, having a Gaussian

space and time profile. For a more sophisticated treatment of the drive beam, the transverse

distribution at each longitudinal slice can be assumed to be in equilibrium with the radial

potential, as governed by the time-independent Vlasov equation. This model is discussed

in the next chapter, but the solution is still time-independent - it does not predict how

quickly this radial equilibrium is established. For solving the initial value problem such

as a Gaussian beam propagating in a finite-length plasma, a third model for the drive

beam has been developed, using a super-particle representation of the drive beam which

now evolves in a time-dependent way.

Strictly speaking, allowing the drive beam to have a time-dependent behavior

violates the assumption of time independence (d/dt = 0 ) used to derive the fluid and field

equations in NOVO. The loss of accuracy due to this will be small if the length scale on

which the beam distribution changes, the beta function , is much longer than k p −1 , as is

usually the case with high-energy beams. To keep numerical noise to a minimum, a large

number of super-particles, about 5000, are used to represent the beam, each having a

form factor to distribute the charge over several neighboring mesh points in r and z . The

particle orbits are then integrated under the influence of the plasma fields, in self-consistent

fashion. There are two options for evaluating the focusing force for a particle with z ≅1,

by using Fr ≅ e(Er − H ), or by using the value of ne (r, ) and equation 1.2. These two

methods ought to give equivalent results, but were seen to differ by up to a few percent

because of inaccuracies accumulated in solving the Maxwell and cold fluid equations.

The second method is used for evaluating the focusing force because it reproduces the

correct ion channel focusing gradient whenever ne = 0 . Because the plasma fields change

slowly, the particle integration can be sub-cycled, or propagated over many time steps

until a significant difference in the averaged quantity nb accumulates.

The blow-out regime plasma dynamics can now be illustrated with a real example,

using parameters very similar to those found in the AWA experiments. With the PIC

simulations later in the chapter as a reference point, the validity of the fluid model can

then be evaluated for the stated conditions. In this simulation, a 20 nC, n = 200 , z = 1.94

mm (= 2k p −1 ) Gaussian beam with an initial energy of 20 MeV is beta-matched into a 10

cm long, n0 = 3 ×1013 cm −3 plasma column. The initial rms transverse beam size obeying

the matching condition is x = 208 µm, which makes the plasma underdense by a factor

of nb / n0 = 3.1, where nb is the peak beam density. Figure 4 is a plot of the longitudinal

electric field profile near the end of the plasma column, showing the characteristic saw-tooth

wake-field shape of the underdense regime, with a peak acceleration field of 130 MeV/m.

The field changes sign at =-0.54 cm vs. -0.53 cm for the PIC simulation of Fig. 7, and

thus the two simulations have almost the same period. Cases with higher charge have

been known to exhibit some unphysical period broadeneing due to the violation of the

fluid assumptions at the channel radius. In all cases where blowout occurs, the return

current is concentrated in an narrow region just beyond the channel radius, showing up as

a spike in Vz . The plasma dynamics at the channel radius are too intricate to be faithfully

modeled by the fluid treatment.

Figure 5 shows the resulting longitudinal phase space. The tail of this beam

extends far enough back to significantly beam-load the wave. This fact suggests that such

a beam is not the best choice for a driver, with a slightly longer but more sharply cut-off

beam capable of enhanced performance. In fact, the Gaussian tail extends far enough to

overlap with the optimum accelerating phase of the wave, with some particles accelerated

to ~30 MeV.

Figure 4. The longitudinal wake and initial bunch profile for a 20 nC beam in a 3 ×1013
plasma, with = 0 corresponding to the bunch center.

Figure 5. The beam longitudinal phase space after interaction with the plasma.

2.2 Incorporating a solenoidal field in the model.

Whenever a static longitudinal magnetic field must be included in the problem, as

in the model of Chapter 3, the NOVO equations can be modified to include the additional

TE-like field components driven by the azimuthal currents, Hr , Hz , E . The additional

field components evolve according to,

rH = − NV
rr r r t

= − NV ,

with E trivially related to Hr ( E =− Hr ). In this geometry, the TM fields along with Vz

and Vr completely decouple from the TE-like solution, and equations 2.5 through 2.7 do

not have to be modified. The fluid equations are in fact coupled, and become,

+ Vr r = G (1+ Vz − Vr2 )(Er − H + Fc ) + H + Vr Ez + V Hz
] (2.8)

+ Vr z = G (1+ 2Vz )E z + Vr H + VrVz (Er + Fc − H ) − V Hr
] (2.9)

[ ]
+ Vr = G V Ez + Vr V (H − Er − Fc ) − Vr H z − Hr (2.10)
t r


Fc = . (2.11)
r(1− Vz ) 1− Vr + 2Vz

G = 1+ 2Vz − Vr2 . (2.12)

The most visible change in plasma dynamics when a solenoidal field is added is a

transverse component to the return current.

2.3 PIC code

A number of limitations became apparent with using the code NOVO as the

principal simulation tool for this work. Because it is not a time-dependent code, it cannot

estimate the transient effects of the beam's initial entry into the plasma. Also not adequately

treated are the time-dependent electromagnetic effects of a mismatched beam executing

radial oscillations in the plasma. Finally, because NOVO is a fluid code, the validity of

the solution requires that the plasma electron density be a smoothly varying function.

This condition is almost always violated when the plasma electrons rush back to the axis,

as a velocity shear in r causes this profile to steepen and the wave to 'break'. The

consequence of this is an artificial slowing of the rarefaction radius motion, or a lengthening

of the oscillation period. Despite these concerns, NOVO is still fairly accurate for simulating

the focusing of the drive beam. The decision to write a new (as yet unnamed) code came

with the realization that such a code is not as difficult to create as was usually thought,

and some encouragement from Paul Schoessow. The information provided in a class on

computational physics relevant to accelerators, taught by R. Cooper, was also of great

help in originally becoming interested in beam simulations, and also containing information

about the present topic.

Figure 6. The integration contours used in discretizing the Maxwell equations.

For the time-advance of the Maxwell equations, the prescription of Birdsall and

Langdon [33] was followed. A staggered mesh with uniform spacing is created, as in Fig.

6, and the finite difference equations are derived by integrating the Maxwell equations

around the appropriate contours. These difference equations are advanced in time according

to a leapfrog integration scheme, with the ∇× E equation used to advance B from the

k −1/2 th time step to the k +1/2 th time step and the ∇× B used to advance E from the

k th to the k +1th step. With bulk polarization and magnetization effects set to the vacuum

values, these equations become:

Er (i + 12 , j) c
[ ]
B (i + 12 , j − 12 ) − B (i + 12 , j + 12 ) − 4 Jr (i + 12 , j) (2.13)

Ez (i, j + 12 ) c 1
∆z i [ ]
(i + 12 )B (i + 12 , j + 12 ) − (i − 12 )B (i − 12 , j + 12 )
− 4 J r(i, j + 12 )

B (i + 12 , j + 12 ) c
= [E (i + 12 , j) − Er (i + 12 , j +1)]
∆z r (2.15)

∆z z
[E (i +1,j + 12 ) − Ez (i, j + 12 )]

Again, the half-integer spatial indices refer to the field quantity at Dr times the value of

the index. The PIC simulation also consists of a number of super-particles, of order

2 ×105 in most cases. The forces are computed on these particles, which in turn influence

the development of the fields via the jr and jz terms in equations 2.13 and 2.14, again

using leapfrog integration to compute the orbits. The momenta, like the B-field, are

defined on half-integer time steps, and the positions are defined on the integer steps.

The solution is then propagated for one full cycle in the following order:

1) Propagate B from time step k −1/2 to k +1/2 ; ( E at step k is already


2) Advance the momenta p from step k −1/2 to k +1/2 using Ek and the

average of the magnetic fields in the last step.

3) Advance the positions x from step k to k +1 using these momenta.

4) advance E from step k to k +1 using the current density j(x) = evk +1 2 (x − 12 ( xk′ + x ′k +1)) ,

with the velocity v k +12 derived from the momentum, and is the Kronecker delta


The particles’ positiona and momenta are advanced using the relativistically

correct equations of motion:

pk +1/2 = pk −1/2 + ∆t ⋅(eEk (r, z) + ev × (Bk−1/2 (r, z) + Bk+1/2 (r,z)) / (2c)) , (2.16)

xk +1 = xk + c∆t ( pk +1/2 mc)2 − 1 (2.17)


jk +1/2 ( x ′) = +ec ( pk +1/2 mc)2 −1 ⋅ ( x ′ − 12 (x k + x x +1) (2.18)

Where the field on the particles, B(r,z) and E(r,z) are computed by linearly interpolating

fields between the four closest mesh points. The currents are assigned to the mesh

according to the same interpolation, done in reverse (gather-scatter). This interpolation is

also called area overlap weighing, discussed in [33].

The fields were initialized to zero, while the plasma electrons's initial positions

were assigned using the Hammersley's sequence [34], and distributed evenly in r and z

but given variable weights to create a uniform plasma. Accuracy is improved when the

number of particles loaded per mesh cell is large (~10). Fortunately, the plasma ions do

not have to be included in the model, as long as it can be assumed that they are immobile.

This is because the field equations have only a j source term and not a term.

A few comments ought to be stated about the correctness of this algorithm with

respect to adequately simulating PWFA performance. In the PWFA case, it's only necessary

for the algorithm to be accurate for about one plasma period, which diminishes the

importance of all the possible sources of error. One source of error for this particular

method of solving Maxwell’s equations is the inaccuracy when approaching the Nyquist

wavenumber for the mesh, leading to a dispersion equation for plane waves propagating

in the x -direction,

 ∆t  c∆t  k∆x 
sin = sin , (2.19)
2  ∆x  2 

where ∆x and ∆t are the space and time increments. This makes the phase velocity

smaller than the speed of light for these modes when constrained by the Courant condition

in two dimensions, c∆t < ∆x / 2 for ∆x = ∆y . To overcome this, some codes use a

frequency-domain (Fourier transform) approach to decompose the modes at each time

step and advance each mode with the correct phase velocity. The errors which accumulate

due to the dispersive behavior are limited to the higher frequencies. No attempt has been

made to evaluate the effect this has on the overall solution, such as the acceleration

gradient or other global quantities of interest.

One future improvement planned for this code is to add a correction which maintains

∇⋅ E = 4 . In the absence of currents, the field solver will identically preserve ∇⋅ E

within the volume. There is no such guarantee with the particle integrator and charge

assignment algorithm used, which will result in minute inconsistencies with the continuity

equation, ∇⋅ j + / t = 0. Unless corrected, these errors will result in corresponding

errors in ∇⋅ E and contribute to additional high-frequency noise[35].

Simulating the same conditions as the NOVO run presented in the last section

yields an electric field profile as shown in Figure 7. This curve is not as smooth as in the

fluid code case due to the presence of high frequency noise. The cumulative effect

washes out, making the final longitudinal phase space, Figure 8, closely resemble that of

the NOVO output. The highest energies obtained with these two simulation methods are

very similar. One difference the PIC results exhibit is the lack of curvature in energy vs.

delay when approaching the highest generated energies. This is a consequence of the

sharper peaks in Ez evident in Fig 7, as opposed to the rounded-off curve in the NOVO

solution of Fig. 4.

Fig. 9 is a configuration space snapshot of the plasma electrons in the region of

the drive beam. In this ficure, it is difficult to determine the remaining electron density in

the channel from inspection alone. Recall that the super-particles originating close to the

axis have negligibly smaller weights in comparison to those at larger radii. The highest

acceleration gradient is just ahead of the point where the channel once again collapses.

Figure 7. On-axis and r = 75 µm longitudinal field from PIC simulation. The center of
the bunch is at = 0 .

Figure 8. Longitudinal phase space from PIC simulation.



r (cm)




5.8 6 6.2 6.4 6.6 6.8 7
z (cm)

Figure 9. Configuration space plot of plasma electron position in PIC simulation.

Chapter 3. Drive Beam Radial Dynamics

A beam initially introduced into a plasma column at the equilibrium radius, will

begin to evolve in such a way as to more closely match the plasma focusing radial

potential well. The weaker focusing near the head of the pulse, where the plasma has had

less time to respond, will allow the beam slices at this position to radially expand. If the

front-most slice expands so much that it’s density is negligible for the purpose of plasma

channel formation, the interaction effectively starts at the second slice—the first slice has

eroded away. This erosion, called emittance driven erosion, can proceed to subsequent

slices until the beam is deconfined, but the asymptotic erosion rate from a 1-D model[36],

( )(rb rc )
2(1 + f )
0 − p = 0.181 f 0.9 , (3.1)

where rb and rc are the beam and plasma channel radii, and f is the ratio of the line

densities of channel ions to beam electrons. This expression means that beam head

erosion can be small over a length scale of energy depletion,

mc3 0
Le ≅ . (3.2)

In 2-D particle-in-cell simulations, Krall, Nguyen, and Joyce [36] report a much diminished

erosion rate, almost resembling a near-equilibrium state. They attribute this to the beam’s

forming radial wings, which cannot be accommodated in the 1-D formalism. Such an

inherently 2-D effect can only happen if the focusing potential acting on the beam is

nonlinear, causing the beam’s transverse phase space to filament. Such radially nonlinear

behavior is always present in the transition near the pinch region from a small perturbation

in the plasma electron density to the point where the channel is fully rarefied.

A slightly different type of equilibration has been witnessed in an experiment

performed by Rosenzweig et al. [37] in the overdense regime. An initially large radius

beam was observed to self-focus and phase-mix in a 30 cm plasma column, eventually

evolving into a Bennett-like radial solution, nb (r) = nb /[1 + (r / b)2 ]2 . Time-resolved

transverse imaging revealed that the Bennett-like equilibrium solution to be representative

of the beam as a whole.

These results suggest that a beam model incorporating the assumptions used in

deriving the Bennett pinch solution, namely that after phase mixing, the beam obeys the

time-independent Vlasov equation. The simulation results discussed in Section 3.2

incorporate the solution of this equation at each longitudinal beam slice self-consistently

into the code NOVO, as described in Chapter 2. In order to stay away from the trivial

solution to these equations, where the beam is totally deconfined, the beam must be

forced to have some initial density at the leading edge. This is done by incorporating a

solenoidal field in the model, which is not uncommon for such a beam-plasma system, as

many plasma devices rely on a solenoidal field to radially confine of the plasma column.

Finally, the validity of the time-independent model is established by directly comparing it

to a time-dependent treatment where the Vlasov solution is replaced by a super-particle

representation for the beam electrons.

A discussion of the time-independent case necessarily brings up the question of

what role initial conditions play in more rapidly achieving the equilibrium state—how to

match the beam into the plasma channel. Section 3.3 discusses this matching starting

from the simplest case of initially focusing the beam to the equilibrium radius. The

matching problem is greatly facilitated in the case of longitudinal tailoring of the plasma

density on a length scale of , so that preferential focusing of the beam body brings a

closer resemblance to the Maxwell-Vlasov equilibrium case. In all cases, the quality of

the initial match has implications for the plasma accelerator, ultimately affecting the

phase stability of the wake-fields.

3.1 Analytical model of rarefaction

Long term propagation in the blowout regime is achievable only when there is a

sufficiently strong plasma focusing gradient and a large enough beam density. The beam

density is responsible for the rate of channel formation, as well as the ultimate degree of

rarefaction of the plasma electrons. In turn, the focusing gradient this generates over

most of the beam feeds back on the peak maintainable beam density for the given set of

conditions. This section develops a model of these dynamics which gives a set of minimal

requirements on the beam quality necessary to achieve the desired propagation.

The beam matched to the ion channel equilibrium beta function derived earlier,

eq = 2 ren0 , will have a transverse beam size given by,

 
= n
. (3.3)
 2 re n0 

To simplify the mathematics, it will initially be assumed that the entire beam is subject to

the same focusing gradient as seen at the beam’s longitudinal center, or that the beam

head suffers no additional radial expansion. The correct treatment of this expansion

appears in the next section using the Maxwell-Vlasov treatment, where the results are

compared to the present mode. Using this equation for a Gaussian beam with Nb electrons,

the peak beam density is given by,

Nb Nb ren0
nb = 3/2 2 = . (3.4)
(2 ) z r 2 n z

Even for a very long beam, this beam density must be greater than the plasma density, on

which it explicitly depends. Shorter beams, of the kind optimally suited for a plasma

wake-field accelerator will also require evaluation of the time needed to create the channel.

As the plasma electron motion is mainly driven by the beam’s radial electric field, it can

be approximated as a uniform radial expansion, neglecting any longitudinal plasma currents.

Near the beam axis (where the space-charge force is linear), the equation of motion for

the plasma electrons radius as a function of = z − ct is given by,

r ′′ − k2 r = 0 (3.5)

where the prime indicates a derivative with respect to , and k 2 is treated as constant by

assuming that the beam density is constant and equal to the maximum value, and is given

by the expression:

k2p  nb 
k 2 = 2 renb = . (3.6)
2  n0 

Note that this expression also neglects the restoring forces due to the charge imbalance in

the plasma, and thus is approximately correct only when nb >> n0 .

The solution of equation 3.6 is, taking a plasma electron to be initially stationary

at radius r0 ,

r = r0 cosh(k ) (3.7)

Since the plasma electron density is proportional to the distance between electrons,

expansion of these distances by a uniform factor (ignoring longitudinal plasma motion)

lowers the plasma density by the square of that factor, and

n(k ) = . (3.8)
[ cosh(k )]2

Requiring that the beam channel be sufficiently rarefied by the beam’s longitudinal

center, or that n(k )/ n0 ≤ 0.01 (99% rarefied) happens approximately at k = 3. Although

this result was derived assuming a constant current profile, it can be used to estimate the

effect of having a Gaussian beam, by using the effective half-length, 2 z 2. The

condition on k z which corresponds to sufficient rarefaction by the center of a Gaussian

beam is given by (corrected from the original publication [38] by supplying a missing

factor of 2),

k z > . (3.9)

Squaring equation this expression and substituting into equation 3.6 gives an expression

for the self-consistent beam parameters needed for the major portion of the beam to

propagate in an ion channel,

nb 36
> . (3.10)
n0 (k p z ) 2

Combining with equation 3.4 yields simply,

36 n
Nb > . (3.11)
re (k p z )

Incorporating, also, the limit on the bunch length in order for the wake field not to be

diminished by the oscillatory response of the plasma, k p z < 2,

Nb > n

This inequality can most easily be satisfied by photoinjector-derived beams. It is satisfied

by more than an order of magnitude for low-charge beams intended for FEL applications,

with Nb = 6.25 ×10 9 , n = 1 mm-mrad and = 40 . For an example at higher charge, the

TTF design parameters, Nb = 5 ×1010 , n = 20 , satisfy the inequality by nearly a factor of

5. The safety margin is somewhat narrower using the original AWA design numbers with

n = 750 mm-mrad and Nb = 6.25 ×1011 , which just barely satisfies the inequality by a

factor of 1.5.

This analysis has so far neglected the radial return forces arising from the plasma

oscillatory response. A more accurate estimate comes from including the return force on

the plasma electrons, which was ignored in equation 3.5 by assuming nb / n0 >>1 . With

the additional terms, this equation becomes,

  r2 
r ′′ − 2 re  nb − n0 1 − 02  = 0. (3.13)
  r 

Without attempting an exact solution to equation 3.13, the additional term is no larger

than n0 in the limit of large r , where it corresponds to the electrostatic contribution of the

ion column. The solution to this equation will have r at least as large as the old solution

(Equation 3.10) if nb / n0 is incremented by one. With this change, equation 3.10 becomes,

nb 36
> +1 , (3.14)
n0 (k p z ) 2

or, keeping k p z = 2.0 ,

≥ 3.86 . (3.15)

An exact solution of Equation 3.13 modifies this numerical factor only slightly, yielding,

≥ 3.45 . (3.16)

In assuming no beam head expansion, this number will be seen to be an underestimate.

Allowing the beam head to expand, as in the one-dimensional model discussed below,

has the effect of moving the pinch point backwards. In order to keep it at the same

location, the beam density must be increased by a proportionately much larger amount.

One-dimensional model of beam head expansion

The same beam expansion dynamics can be solved with the aid of one-dimensional

theory using the same assumptions as were made above, but with the correct model for

the beam expansion as a function of . The solution thus obtained will check the validity

of the assumption that the entire beam is exposed to the maximum ion channel focusing

gradient, with no additional beam head expansion. Allowing all quantities to depend on

the new variable, Z = kp (ct − z), the equations governing the plasma electron expulsion

and the beam envelope evolution can be written in dimensionless form for a long beam

with constant current:

X  1 1
2 + 1 − 2  X = , (3.17)
S R X3
R 1  nb 1 
= X +1 − 2  , (3.18)
2  n0 R 

where S is the propagation distance in units of the betatron wavenumber

(S = k ct = ct 2 re n0 ), R is the plasma electron radius in units of its initial value, and

X is the beam radius in units of the equilibrium radius in an ion channel. To aid in

understanding the dynamics represented in these equations, it should be noted that Eqn.
3.17 is simply the beam envelope equation with the usual 1/ r emittance term. Setting

S = 0 in this equation recovers the solution for the equilibrium radius ( X = 1) in the

limit that R is large. Likewise, the second equation is that of an oscillator with period of

2 , or p in real units, driven by the nb term, or just the dimensionless form of Equation

3.13. The derivation for Eqn. 3.18 uses the approximation Z ≅ kp c t , which is only

valid when the bunch length is short compared to the betatron period. Including the extra

convective terms needed to make this an equality introduces an additional time scale into

the problem and would not allow writing this equation in dimensionless form. Numerical

integration of these equations on a grid including 1000 longitudinal beam and plasma

slices reveals the time evolution of the pinch point location (defined as the position where

1 −1 R2 = 0.99 ), as shown in Figure 10 using nb n0 = 8.4 . The distance of the pinch

point behind the head of the beam Zpinch is, at first, quickly changing, but reaches an

asymptotic value after about S ≅ 2 , or approximately one betatron wavelength. By

choosing nb n0 appropriately in this case, the final value of Zpinch (≅ 2.5) has been

intentionally made to be equal to 2 , in agreement with the earlier criteria for channel

rarefaction by half the effective width (the center) of a Gaussian beam with z = 2k p−1.

These criteria, therefore, require,

≥ 8.4 . (3.19)

The extra beam head expansion allowed in this analysis changes the requirement for

nb n0 from the old value of 3.45 to the much higher value used in this case of nb n0 = 8.4 .

Comparisons of this solution to the full axisymmetric fluid code shows agreement of the

two methods over a wide range of the remaining variables in the problem, namely the

matched beam radius and the energy.

It is interesting that the dependence on in the definition of S means that the

asymptotic value for the pinch point is reached faster in units of the energy depletion

length, which rises linearly with . It is, therefore, not possible to stay on the rising edge

of the slope, thus enabling use of a smaller nb n0 , by going to a higher energy drive

pulse. Actual cases will behave even worse in this respect since this model neglects any

beam energy losses due to wake-fields, as this will modify the expression for the matched

radius over the propagation distance and lower the beam density.

The fact that satisfactory performance is only achieved at fairly high values for

nb n0 merely indicates that the original criteria of rarefaction by the center of the bunch

is much too strict. Effective drive beam propagation for the initial set of experiments can

be achieved with 90% rarefaction by the beam’s falling edge, or nb n0 ≅ 2 . This value of

nb n0 is sufficient to assure effective propagation of the drive beam in the plasms. The

remaining electrons in the channel must also be sufficiently rarefied by the arrival of the

trailing beam. The outward momentum for these particles after the passage of the drive

beam helps in keeping this density down. The remaining electrons will respond to the

trailing beam’s space-charge forces, and can lead to emittance increases due to a

longitudinally varying, as well as a radially nonlinear focusing force.

Figure 10. Evolution of the pinch point as a function of S , the propagation distance in

units of the betatron wavelength.

3.2 Maxwell-Vlasov model of the drive beam

In the systems of interest here, a large part of the beam propagates in an ion

channel, where linear focusing allows a very straightforward beam model to be applied.

The head and transition region, where the rarefaction is not complete, are influenced by

the nonlinear radial forces found in such channels. The fact that the plasma electron

rarefaction there is more pronounced near the axis leads to a betatron oscillation period

that is longer for beam electrons having radial orbits going far away from the axis. The

work of Bennett [39] on self-focused streams considers the mixing from this spread in

frequencies as one of the mechanisms leading to a nearly steady-state beam solution; the

same sort of dynamics were considered by Rosenzweig et al.[40] in the analysis of a self

pinched beam in the overdense regime. Taking, for the moment, the plasma focusing

potential V(r) ≡− ∫ F(r ′ )dr ′ as a given in the problem, the beam distribution will obey

the Vlasov equation governing the distribution function f (r, pr ) ,

pr f f
+ Vr (3.20)
m r pr

having dropped the f t term by invoking the above mixing arguments. The solution to

this equation can be separated in the two variables,

f (r, pr ) = R(r)P(pr ) (3.21)

The solution for R(r) makes use of the fact that near the axis, the potential is approximately

linear and no mixing can occur. The phase-space density near the axis, f (0,0) is, by

Liouville’s theorem, a constant of the motion. With this constraint, Equation 3.20 can

now be solved while also keeping the number of particles constant, with the transverse

temperature as a free parameter. The solution is of the form,

R(r) = b [1+ (r / b) ] ,
2 2

where b is the Bennett radius. This radial density distribution is characterized by tails

much longer than for a Gaussian, and can account for a sizable fraction of the total


In the dynamic case, when the plasma electron distribution is quickly evolving as

a function of , the NOVO solution for the plasma fields supplies Vr at each longitudinal

slice in the problem,

mc2eBz 2
Vr = ∫ (Er − B )dr + r (3.23)

where the second term is the focusing contribution from the solenoid. This leads to a

solution of the form,

Nb  −ap2r 
f (r, pr ) = exp[ −aW(r)] exp  (3.24)
(2 ) 3/2 z pz  2 m 

the spatial part of which is again used as a driving term in the plasma dynamics equations

for computing the response at the next -slice.

Computational results

The Vlasov equilibrium of the last section and the NOVO equations, modified to

be consistent with a longitudinal magnetic field from the solenoid, as described in Chapter

2, can now be used to simulate the behavior of the beam-plasma system with different

choices for the drive beam charge, emittance, and plasma density. These simulations are

then compared to the result from the last section, where the drive beam was assumed not

to undergo any radial spreading.

The simulations require one additional parameter choice, the strength of the solenoid,

which yields an equilibrium beam radius for the leading edge of the beam of,
 2m c2 
= e n
 eBz 

The initial expulsion of the plasma electrons is much delayed for k much larger than
sol p

unity due to the return current flowing inside the beam, thus making the plasma electron

motion longitudinal. For this reason, the condition sol pk < 2 has been maintained. The

resulting value of the magnetic field makes the cyclotron frequency much lower than

other time scales in the problem, p >> c, and therefore only slightly modifies the

plasma dynamics.

The effects of varying the charge while keeping all other parameters constant is

summarized in Fig. 11, where n = 400 mm-mrad , k p z = 2, z = 1 mm , n0 = 1014 cm−3 ,

and = 300 . Here, the plasma begins to be nearly totally rarefied at the beam’s longitudinal

center for all bunch charges over Q = 20 nC, for which nb, eq / n0 = 4.0 , but only appears

to reach 99% rarefaction at about Q = 35 nC, in agreement with the one-dimensional


Figure 12 displays the results of varying the emittance while holding all other

beam parameters constant ( Nb = 6 ×1010 , k p z = 2, = 300 and z = 1 mm). According

to equation 3.19, rarefaction at the middle of the beam should be achieved for all emittances

smaller than 110 mm-mrad, while in Fig. 12 the rarefaction has already slightly deviated

from completion at 100 mm-mrad.

Figure 13 shows the parametric dependence of the rarefaction condition on plasma

density in order to reveal the effects of varying the quantity k p . The normalized beam

spot size is plotted as a function of for a beam with parameters Nb = 6 ×1010 , = 300 ,

z = 1 mm . Only the case with n0 = 1 ×1014 rarefies well, as has been anticipated. for

smaller plasma densities, the focusing is not strong enough to focus the beam to much

higher density that the plasma density, and in turn, the plasma electrons are not expelled

as quickly by the beam density. For much larger densities, the plasma has time to respond

to the beam charge and the beam is well focused, but simply not denser than the plasma.

In this case, electron expulsion is further delayed by the greater amount of charge

imbalance at the higher plasma density under conditions of a partially formed channel.

Results from the radial equilibrium treatement discussed up to now can be compared

with a time-dependent analysis using an initially Gaussian beam. The computational

technique, as described in Chapter 2, represents the beam with a collection of superparticles,

4000 in this case, which evolve self-consistently with the computed plasma electric and

magnetic fields. Such initial conditions do not guarantee a final beam shape which accurately

matches the Maxwell-Vlasov equilibrium. This is mainly due to the lack of damping of

beam electrons at the extreme leading edge of the beam. These particles are subjected

mainly to the static solenoid focusing field, and will collectively oscillate in radius over

long distances, in a way not consistent with the equilibrium analysis. For distances much

shorter than these oscillations, the head of the beam will initially expand to sizes much

larger than the solenoid equilibrium radius sol for the particular beam emittance. Figure

14 shows a comparison between the two computational techniques when the beam has

propagated for 10 and 20 cm in the time-dependent analysis. At the longer distance, the

beam head has undergone a greater amount of expansion, but the longitudinal position

where beam density reaches an acceptably high level remains unaffected. Finally, these

propagation distances are long in comparison to the transient erosion effects, such as

were found in the one-dimensional analysis. According to the earlier result, this reorientation

ought to be mostly complete after the first 3 cm of propagation.

εn 400 mm-mrad 15 nC
kpσz 2 20 nC

25 nC
n b,eq/nb (kp ξ)

10 30 nC

35 nC

-6 -4 -2 0 2 4
kp ξ

Figure 11 Normalized beam spot area (defined as a function of longitudinal position by

[ ]
neq nb (kp ,r = 0) , with neq = Nb exp (− 2 2 z2 ) (2 )3 / 2 eq2 z , for a number of different
bunch charges with other parameters held constant ( n = 400 mm-mrad, k p z = 2 , = 300 ,
and z = 1 mm ).

N 30 mm-mrad
6 x 1010
100 mm-mrad
kpσz 2
150 mm-mrad
n b,eq/nb (kp ξ)

200 mm-mrad
300 mm-mrad


-6 -4 -2 0 2 4
kp ξ

Figure 12. Normalized beam spot area, as defined in Fig. 11, for a number of different
emittances with other parameters held constant ( Nb = 6 ×1010 , k p z = 2 , = 300 , and
z = 1 mm )..

N 10
6 x 10

1000 ε 100 mm-mrad


σ 1 mm
n b,eq/nb

100 1E12

10 1E14

-0.3 -0.2 -0.1 -0.0 0.1 0.2 0.3
ξ (cm)
Figure 13 Normalized beam spot area as a function of , as defined in Fig. ?, for a
number of different plasma densities with other parameters held constant ( Nb = 6 ×1010 ,
n = 100 mm-mrad , = 300 , and z = 1 mm )..

z=20 cm
Maxwell-Vlasov equilibrium
z=10 cm
n b,eq/nb (kp ξ)


0.0 0.1 0.2 0.3 0.4 0.5

ξ (cm)

Figure 14 Comparison of Maxwell-Vlasov equilibrium prediction with time-dependent

beam model (at z=10 and 20 cm into the plasma), with the on-axis beam density as a
function of normalized to that obtained from ion channel equilibrium. The beam-plasma
system has the following parameters: Nb = 6 ×1010 , k p z = 2 , = 300 , k p sol = 2 , and
z = 1 mm .

3.3 Matching

Effective long-term propagation in the plasma, as well as optimal generation of

accelerator-like wake-fields, is heavily influenced by the initial conditions, the matching

of the drive beam to the plasma's ion channel. An attempt to match the body of the beam

to the focusing gradient of a fully formed ion channel only matches those parts of the

beam which do, in fact, propagate in a channel. This leaves the beam head, the transition

region, and part of the beam body in an unmatched state, resulting, in the short term, in

betatron oscillations for those beam slices. The effect of an unmatched beam head also

leads to a several eq -long reorientation process during which the beam distribution is

rapidly changing, ultimately resembling the radial equilibrium of the previous section,

with a greatly expanded beam head and a Bennett-like radial profile in the transition

region. This section begins with a definition for ideal matching, which, as it will be seen,

is very difficult, or even impossible to attain. This is followed by a discussion of the

consequences of various types of mismatch: greater beam expansion and reduced PWFA

performance. The bulk of the Section explores the idea of using a 'matching section' at

the beginning of the plasma column in an effort to more faithfully emulate the ideal

matching state. This matching section has, in general, a plasma density different from that

found in the rest of the column; its special lensing properties act as a beam conditioner to

manipulate the beam distribution to more closely resemble the ideally matched state.

An ideally matched beam will, shortly after being introduced into the plasma,

propagate with no betatron oscillations for each longitudinal slice. This condition is quite

restrictive, and may not be achievable in practice. In order for a beam-plasma system to

qualify as ideally, the beam would, initially, have to resemble the final state, having an

expanded beam head and a Bennett-like transition region. Such a beam is not at all

commonly available from accelerators; the most common model for accelerator-derived

beams has a Gaussian distribution in six dimensions. Starting with such a beam, which

has no correlation between z and orientation of the transverse phase space ellipse, it is

very difficult to manipulate this phase space as a function of z in the precise way

required for an ideal match. Given that the introduction of the beam into the plasma will

almost always result in some amount of mismatch, it is instructive to explore the negative

consequences associated with this, particularly in the case of the PWFA.

The rapid initial expansion associated with a mismatched beam head, as was

stated earlier leads to the rapid (≅ 6 eq long) reorientation of the head and transition

region, resembling the process of emittance-driven erosion. During this expansion, the

effective beam centroid for wake-field generation moves backwards in the beam frame,

since the beam density at the head, and hence the wavenumber for the radial expulsion of

the plasma electrons k 2 = 2 ren0 , tends to diminish with time. This causes a phase shift

in the longitudinal wake-field, which, if uncompensated in some way renders the final

optimum accelerating phase initially on the wrong side of the ion channel collapse. The

region where the ion channel collapses is characterized by plasma electron densities

much higher than the background, sometimes as high as 30n0 , and mildly relativistic

motion which causes highly nonlinear fields due to decoherence of the plasma oscillations.

Trailing beam electrons caught on the wrong side of this point would be severely disrupted

and lost in the course of crossing over. Even for electrons starting on the correct (accelerating)

phase, the maximum available radius for linear focusing is diminished, since, closer to

the point of collapse, the channel radius is smaller.

Betatron oscillations at the transition region can result in the same type of phase

shift. In this case, the phase will oscillate with time, making this situation harder to

correct with the plasma density tapering technique proposed for erosion-caused phase


For portions of the drive beam which always propagate in an ion channel, some

amount of mismatch is tolerable. For a relativistic beam, the forces imparted on the

plasma electrons will be independent of the beam radius, as prescribed by Gauss' law.

3.3.1 A near-ideal match beam conditioner

The idea developed in this section it to use a plasma lens as a beam conditioner in

such a way that it can manipulate the different parts of the beam to approximate the

desired matching state. The most basic requirements for such a scheme are that the beam

head is matched to an external focusing channel, as with the solenoids in Figure 15, while

the body is matched to the ion channel focusing. Additionally, the changeover between

these two modes should happen on the appropriate length scale, or roughly at the pinch

point. Beam slices behind this point point, ought to propagate identically in the system.

This requirement argues for having a matching lens with the same longitudinal wavenumber

for the ion channel formation, k 2 = 2 renb , as is found in the plasma column, or that the

beam density in the matching lens and the plasma column be approximately the same.

This is only a requirement on the beam density and radius, and not a requirement on the

plasma density in the matching section, as long as it is underdense. The idea behind this

scheme is for the beam to ‘sample’ the later plasma focusing in the much shorter matching

section. With the help of appropriate external focusing, these time varying radial kicks

are then translated to differential beam sizes at the start of the column.

Figure 15 is a prototype of such a conditioner, starting with a beam focus at the

upstream end of the matching lens, with a peak beam density that is close to the target

density in the long plasma column. By using a higher plasma density in the matching lens

the beam body can be focused this beam even further toward the lens exit. To simplify

the discussion, a thick lens with phase advance of = /4 will have the beam exiting

the lens also at a waist, but one where the beam density is higher than before. The action

of the second set of quads is to demagnify this beam and bring about another waist at the

start of the column. If the plasma lens is short compared to the distance to the plasma

column, the beam head will also be demagnified by approximately the same amount.

Figure 15 Use of a plasma lens to match into a long plasma column.

The idea of a matching lens can best be described with a real example. Using a

1 GeV beam with n = 50 mm-mrad into a n0 = 1 ×1014 cm −3 plasma for the column, the

matched beta function is eq = 3.35 cm. Inputting a beam radius at the start of the lens
that is twice the equilibrium (2 eq ) will result in a beam core radius that is roughly 2 eq

toward the end of the lens. In this example, this can be accomplished with a 5.7 cm long

plasma lens with n0 = 1 ×1014 cm −3 . The drift length from the lens exit to the second set

of quads is 0.5 meter, along with a 2.0 meter drift from the quads to the plasma column.

Using point-to point imaging, these quads will demagnify the beam by a factor of four, so

that the beam body is 28 microns and the beam head is 112 microns. The head of this

beam would need a very strong solenoid, having a longitudinal field of 60 kG, impractical

for this application. This example does show a mechanism by which the beam head and

body can be focused preferentially, in a way corresponding to the requirements of a

near-ideal match. Wake-fields in the matching lens will cause a certain amount of beam

energy modulation, on the order of 2.5% in this case, but not enough to cause achromatic

focusing in the quads.

A more realistic system exhibiting the same dynamics can be optimized in a

number of ways. The choice of a thick lens, simplifying this example, can be replaced

with a thin lens, so that the waist can be after the lens exit. This can bring about a much

larger decrease in the beam radius without increasing the peak beam density in the


3.3.2: Sinusoidal ramps.

Beam conditioning similar to the type discussed in the last Sub-section can also

be accomplished with a smoothly varying ramp at the start of the plasma column, although

not as effectively. Some finite in longitude rising edge of the plasma density will always

be present in real devices, and must be accounted for in focusing and matching calculations.

This section considers a sin 2(kr z) functional form for such a rising edge, examining

phase advance ≅ /2 and solutions for the beam body in this matching region. The

effect of this ramp on respective parts of the beam can be explored with an analytical

model for ion focusing using the expression for the focusing experienced by the beam as

given simply by the ion density,

k(z) = 2 re n0 (z)/ (3.26)

The plasma density profile is, then, specified by the following expression:

n0 (z) = 0 (z < 0)
= nm sin2 (k rz) (0 < z < 2kr ) (3.27)
= nm (z > 2kr )

In the plasma boundary region (the ramp, 0 < z < 2kr ), the equation of motion for the

beam particles in either transverse dimension is of the form,

+ K0 sin 2(kr z)x = 0, (3.28)
dz 2

where the constant K0 = 2 re nm / . Note that Eq. 3.28 is formally the Mathieu equation

[41] in the region of rising density. Use of the Mathieu form guides the discussion of the

matching problem. For example, we may look for solutions that match a waist (that is

d x dz = 0 ) at z = 0 to a waist at the end of the boundary region z = 2k r . For this to

be true, any ray which is initially parallel must either be parallel or cross the x = 0 plane

at z = 2k r . In short, this requires simultaneous odd and even periodic solutions of the

Mathieu equation. However, the theory of Mathieu equations explicitly denies this

possibility, since the odd and even solutions have distinct frequencies (that is, K0 must

be different for the two types of solutions) [41]. Thus the beam strictly must not be at a

waist at the beginning of a ramp, but in fact must be converging to achieve a match at

z= /2 kr .

In practice, the best way to derive the initial conditions at the beginning of the

ramp is to start with the desired match at z = /2 kr , and solve the envelope equation for

the rms beam size,

d2 2

2 + K0 sin (kr z) =
3 , (3.29)

backwards in z . In keeping with the previous discussion, we can classify the ramped

profiles according to the phase advance,

/2 kr / 2 kr
≡ ∫ (z)
= ∫ 2
dz (3.30)
0 0

which is approximately n /2 , with n equal to a positive integer. For example, the

shortest ramp has an approximate phase advance of / 2, corresponding to a quarter of a

betatron oscillation; an initially parallel ray approximately comes to a focus at z = /2 kr .

The time dependent model of the plasma fluid response can now be applied to

smoe real cases, after stating some additional requirements for the system. The first is the

the plasma conditions do not change appreciably over the distance from the head to the

tail of the beam, which can be quantified by requiring k r z << 1. This is satisfied in all

the calculations given below, where the beam length is typically 1 mm, while the rise

length is several centimeters. In addition, there is a less relevant requirement mainly

concerning calculation of the wake fields behind the beam, k p << kr , which is satisfied
for the denser regions of the plasma (which are the cases of interest, since k p ≈ z ).

The matching of the beam body, which is here defined to be the last half of the

beam, is illustrated in Figure 16. In this case, the ramp is 1.63 cm long (k r = 0.964 cm −1 ),

corresponding to ≈ 2 . The beam-plasma system has the same parameters as in

Figure 14. It should be noted that the beam has a virtual waist (that which is obtained if

the plasma is removed) of 123 microns, which is not much larger than the ion channel

equilibrium. Thus the matching section in this case does little to suppress beam head

oscillation and loss. A matching section which allows the initial external focusing to be

eased (since the virtual waist is larger), must use a longer ramp, in this case with a phase

advance ≈ . The virtual waist for these conditions is 149 microns, and so the depth of

the initial focus is over twice eq . The evolution of the rms transverse size of the beam

body is shown in Figure 17. The collisionless damping of the beam size oscillations

indicates the presence of some deviation from constant, linear focusing in the bulk of the





σ (µm)





0.0 2.0 4.0 6.0 8.0 10.0
z (cm)

Figure 16. Matching of the beam body with a ramp of length 1.63 cm (k r = .964 cm − 1 ),
corresponding to = /2 . The beam-plasma system has the same parameters as in
Figure 14.


σr (µm)



0.0 2.0 4.0 6.0 8.0 10.0
z (cm)

Figure 17. Matching of the beam body with a ramp of length 3.33 cm (k r = 0.472 cm −1 ),
corresponding to = . The beam-plasma system has the same parameters as in Figure

3.3.3 No matching section

If neither a separate matching plasma lens nor a sin 2(kr z) ramp can be used for a

given PWFA design, the system must be designed in such a way as to mitigate the initial

wake-field dephasing and betatron oscillation effects. The transient beam erosion-caused

dephasing can be cured simply by tapering the plasma density in this region so a lower-

density plasma at the start of the column compensates the phase difference by virtue of a

longer p . The peak acceleration field in this region is diminished at the expense of

maximizing the stably accelerating phase space region—the accelerator bucket size—by

delaying the channel collapse region during times that the beam shape is still changing.

Even with such a correction in place, the beam-plasma system is still susceptible to a

periodic modulation of the wake field when the body and transition region are mismatched.

More study is required to pinpoint the exact cause of this effect, but it is plausible that a

combination of beam initial conditions and plasma density modulations can be found that

cancel it. Solutions to these phase stability problems are obviously needed for any actual

accelerator design, but will increasingly play a role in the ongoing PWFA-related



This chapter has covered many aspects of beam propagation of short pulses in the

underdense regime. The one-dimensional equations introduce the idea that the most

important parameter for the plasma wake-field accelerator is the normalized matched

beam density, nb n0 . Depending on the specific goals, this number should be anywhere

from 2.0 to 8.5. Two-dimensional simulations give a much more detailed picture of beam

and plasma dynamics. Use of a beam model that incorporates the Vlasov equation, as is

done for the case of a Bennett pinch, illustrates the effect of the nonlinear dynamics

present in the transition region, near the beam head. This model is a good predictor for

the amount of beam head expansion during the initial phases of propagation. Without

resorting to lengthier time-dependent calculations, this model can give a quick estimate

of wake-field and focusing performance of a drive beam in a long plasma column.

The beam-plasma system also has some important time-dependent effects. In

order to maintain high quality acceleration and focusing fields for the trailing beam, the

effects of drive beam betatron oscillations and emittance-driven erosion on the plasma

dynamics must also be considered. These conditions can be optimized either by using the

conditioning properties of thin and thick plasma lenses on the drive beam, or by modifying

the plasma density as a function of length in order to compensate for phase differences.

Chapter 4. Applications of PWFA Technology.

Future research into the mechanism of plasma wake-field acceleration will bring a

clearer picture of how to apply it to an eventual collider application. So long as this idea

is competitive with other advanced acceleration methods, theoretical and experimental

studies will address the final use of this technology in an increasingly detailed manner.

One milestone would be the construction of a test accelerator, possibly using a 0.4-1 GeV

drive beam deposited into several plasma sections and accelerating a >10 GeV trailing

beam, in a setting geared to test extending this acceleration to much longer distances. The

design of such a test facility would have to tackle many of the issues of a final collider

configuration, and is often the precursor to building an actual collider.

There does not, at present, exist a clear enough picture of the final form of a

plasma-based collider facility. Continuing progress may identify new problems, or reveal

new ways of optimizing the setup to better meet a specific set of goals. In the meantime,

it would be very helpful to visualize the future by initiating a collider facility design

which is much less detailed, and ignores many of the topics outlined in Chapter 1. Such a

design would reveal likely choices for quantities such as the power efficiency, the repetition

rate, the accelerated charge and the facility size and physical layout. In this way, plasma

technology can be compared to other methods of particle acceleration. A design such as

this was presented by Rosenzweig, calling it a “straw-man” design.

4.1. A design for a 2.5 TeV collider

When considering plasma wake excitations operated in the blowout regime for a

collider application, it ought to be stated at the outset that ion focusing only works for

accelerating a negatively charged species. The ion channel, whose radial force is focusing

for electrons will defocus and radially eject any positrons. Regions where positrons

would be focused, and are confined to the fairly small volume of the channel collapse

region, also characterized by large plasma electron densities and velocities that would be

detrimental to beam quality. The blowout regime is only suitable as a basis for e − − ,

- or e − − e − colliders.

Any collider design based on wake-field technology must also consider the

transformer ratio to be used in operation. Early wake-field studies placed much importance

on this number. It was argued that if one uses a 1 GeV beam to impart only a 2 GeV gain

for the trailing beam, the advantage of high gradient is lost in the amount of space needed

to create the driver. A solution to this was pointed out by W. Gai[42], who suggested

creating a bunch train in one work-horse linac and distributing these among a set of wake

modules, thus creating a ‘hardware’ transformer. Limitations on the drive bunch length

due to hosing instability problems require just this sort of a solution if one is to keep the

overall size of the collider facility to reasonable dimensions.

A collider facility based on this hardware transformer principle, to be described in

detail below, was proposed by J. Rosenzweig[43] at the 1996 Snowmass workshop. This

design uses a counter-propagating distribution network for the drive bunches, which

necessarily fixes the spacing between bunches in the drive bunch train. To synchronize

the arrival time of a drive bunch at its respective plasma module to the passage of the

trailing beam, the drive bunches must be separated by twice the plasma cell spacing. This

calls for either a high bunch train repetition frequency or very long plasma sections. In

this example, each plasma section is 5.7 m long, and the space between modules is 2.66

m, which translates to a 55 ns periodicity within the bunch train. Although the subject of

multi-bunch effects in the linac must be examined with much greater care in the context

of such an application, the 55 ns spacing seems adequate for suppression of the unwanted

deflection modes. At the same time, these bunches are spaced closely enough together to

significantly beam-load the linac, a way of increasing the efficiency.

Additional considerations when choosing the beam and plasma wake parameters

in the modules include the plasma wavelength and the maximum deceleration gradient.

Longer plasma wavelengths ease the problem of time synchronization between the drive

and trailing beams. The beam and plasma parameters for a single module, listed in Table

1. The chosen plasma density, 2 ×1014 cm -3 , is a good compromise between a large

acceleration gradient, 1 GeV/m in this case, and longer wavelengths, 2.2 mm in this case.

The 3 GeV initial driver energy assures that some of this energy remains at the end, and

the beam does not completely decelerate in the plasma. The projected wake-fields are

excited by 20 nC in the drive bunch at an rms bunch length of z = 0.8 mm, and a

normalized emittance of 50 mm-mrad. These parameters are similar to those simulated

for TTF, and are in line with performance capabilities of state-of-the art photoinjectors

using bunch compression. As the maximum accelerating wake is below the wavebreaking

limit for the chosen plasma density, the design is also flexible in terms of generating

improved performance given an improved set of bunch characteristics.

Table 2 shows the parameters of a heavily beam-loaded linac based on a TESLA-like

1300 MHz structure. For the set of parameters considered, the majority of the input RF

power , 78%, is coupled to the beam, with a relatively smaller percentage for wall losses.

As stated above, this requires careful thought about removing power coupled into the

higher order modes in the form of wake-fields. Both deflection (m=1), and axisymmetric

(m=0) wakes must be considered, as they can affect the transverse position and final

energy of buches as a function of relative delay from the start of the bunch train. The

accelerating gradient is chosen to be relatively low (6 MeV/m), in order to mitigate the rf

power dissipation and related issues. A peak power of 5.9 MW during the 14.5 µs bunch

length and 5% duty cycle dissipates an average 66 kW per section, not a very high value

in terms of cooling requirements considering the 1.1 m length.

Parameters for evaluating the entire facility as a whole are listed in Table 3.

Assuming that the plasma wake-field mechanism can be 20% efficient (20% beam loading),

this implies that 2 nC can be accelerated in the system. A combination of beam loading

and off-crest operation makes the final energy 1.25 TeV, which assumes a more conservative

0.87 GeV/m acceleration gradient instead of the 1.0 GeV/m listed in Table 1. The overall

size of the facility will be 4.3 km, plus additional space needed for the gamma converter

and detector. Collisions would occur in this facility at an average rate of 3.5 kHz. This

would require a total wallplug power of 335 MW, or an overall efficiency of 7.6 %, a

quite attractive number considering that power coupling through the plasma medium

represents an additional power conversion step.

Thus far, nothing has been said of the method of switching the drive bunches to

their respective modules, and then combining the drive and trailing beams so they are

coaxial. First, the short spacing between bunches, as well as their high energy, forces this

switching to be performed at a subharmonic of the RF frequency. The 250 bunches would

be differentiated in a binary RF splitting scheme, to be directed into a beam pipe leading

to the 180 degree bends at the start of each module. The final stage of this bending is to

be performed without the use of a magnet; considering the very high energies of the

trailing beam, a strong magnetic field would cause cause too much synchrotron radiation

and must be avoided. Instead this drive/trailing beam switching would have to be performed

with very high frequency rf kickers, in the region of 50 GHz. The stability of the RF

splitting and combining systems is of course critical in determining the performance of

the collider.

Figure 18. Schematic of a - collider using a hardware transformer scheme.

Beam Energy 3 GeV

Beam Charge 20 nC
Stored Energy/Bunch 60 J
Bunch Length 0.8 mm
Norm. Emittance 50 mm-mrad
Plasma Density 2X1014 cm-3
Plasma Wavelength 2.2 mm
Decelerating Wake 500 MeV/m
Accelerating Wake 1 GeV/m
Wake Module Length 5.7 m
Intermodule Drift 2.66 m

Table 1. Drive beam and accelerating module parameters for the plasma wake-field based
collider depicted in Figure 18.

Avg. accel gradient 6 MeV/m
Shunt impedance ZT2 30 MW/m
Active length 500 m
Cavity length 1.1 m
Peak RF power (cavity) 5.9 MW
Number of bunches 2 x 250
Beam current (in fill) 690 mA
RF flat top 14.5 msec
Duty cycle 5%
Avg. bunch rep. rate 865 kHz
power/cavity 66 kW
power 30 MW
Total avg. beam power 104 MW

Table 2. Design parameters of heavily beam-loaded 1300 MHz drive linac.

Accelerated charge 2 nC
Wake efficiency 20%
Length of Collider 1 x 2.16 km
Accel. beam energy 1.25 TeV
Avg. collision rate 3.5 kHz
Drive linac/wall effic. 40%
Total wall power 335 MW
Total efficiency 7.6 %

Table 3. Accelerated beam and overall system collider performance.

Chapter 5. The AWA Facility
The collider design presented in the last chapter made use of photoinjector technology

for drive pulse generation. The beam quality (emittance) and high charge per bunch Q

these devices are capable of make it difficult for other technologies, such as the thermionic

gun, to compete. The success of the AWA plasma experiments has, likewise, depended

on a detailed understanding of photoinjector dynamics. One challenge has been to develop

diagnostics that not only address the usual difficulties arising in the measurement of a

photoinjector beam, but also work with the high amount of charge present in this system.

In parallel with this, numerical simulations and laboratory experience have suggested

strategies for optimizing the beam parameters needed for high gradient plasma wake-field

generation. The plasma experiment also required the addition of a pulse splitter in the

laser path introduce a witness pulse with a variable delay behind the main pulse.

The second part of the chapter discusses the beam diagnostics developed for use

with the plasma experiment and elsewhere in the beamline. The placement of this topic in

the present chapter is meant to emphasize the fact that diagnostics development has

strongly been governed by the theoretical and practical considerations outlined in the

first part of the chapter.

5.1 Introduction

In the original simulations for the design of this system, Ho and Schoessow[44]

showed that the gun could produce 100 nC, a number previously thought to be very

difficult to attain. Originally designed to produce high wake-fields in dielectric-loaded

waveguides, emittance was considered to be a less important parameter than the bunch

length. These initial gun and linac dynamics studies yielded an rms normalized emittance

of n = 750 mm-mrad, and an rms bunch length of z = 2.87 mm[48].

The AWA drive linac consists of a single cavity gun followed by two standing

wave linac sections operating at L-band (1300 MHz), as shown in Fig. X. The UV (248

nm) laser is brought in slightly off-axis and travels the length of the linac before striking

the 1.6 cm diameter magnesium photocathode. After gaining about 1.5 MeV in the gun

cavity, the electrons are focused by the gun solenoid and further accelerated by 8 MeV

(design) in each linac section. One additional solenoid between the two linac tanks is

used for matching the beam into the smaller diameter beam pipe and quad triplet following

the linac (aperture clearances in the linac sections are 4 inches in diameter). During

operation, the RF system described in Figure 21 allows control of the power and relative

phase of the RF into each cavity; crystal diode detectors monitor the forward and reverse

component in each waveguide feed, as displayed by an oscilloscope. The laser is mode-

locked to a sub-harmonic of the RF, with a phase shifter to control the relative timing of

the two systems.

Linac tank 1 Linac tank 2 Quad triplet

Gun solenoid
Linac solenoid
Drive gun UV laser

Figure 19. The AWA drive linac.

Figure 20 shows the layout of the drive gun in relation ot the 4.5 MeV witness

gun. The two beamlines are combined with a dipole chicane so that a properly delayed

witness beam can probe wake-fields in a dielectric structure or a plasma. Figure 22 is a

block diagram of the laser system for driving photoemission. This system is based on a

dye oscillator and uses the Nd:YAG oscillator simply to pump the dye laser, as opposed

to chirped pulse amplification (CPA) systems in which the YAG is hte primary oscillator.

Figure 20. The AWA drive and witness beamlines (simplified view).

Variable coupler.
To linacs Phase shifter
φ To Gun

KW amp
81.25 MHz To laser system
φ 75X synthesizer Nd:YAG mode locker

HV RF vs. laser phase shifter

Figure 21. AWA RF system.

Figure 22. AWA laser system block diagram.

5.2 Simulations of low-charge performance

This section considers a different operating point for the AWA linac than the

100 nC case originally envisioned. The most compelling reason for this comes from the

scaling laws for plasma wake-field generation. The largest contribution to bunch lengthening

in a photoinjector is the longitudinal wake-field, which varies linearly with Q . The

wake-field Ez strongly depends on the drive pulse length. Wavelength scaling of the

beam-plasma interaction results in a quadratic dependence,

Ez / Q ∝ z (∝ n0−1 ), (5.1)

of the wake-field per unit charge on the bunch length. The same scaling also prescribes

an emittance,

n / Q = const., (5.2)

needed to achieve the same beam radius in units of k p −1 . This means that any effort to

decrease z by lowering Q will be successful as long as it stays ahead of the linear

relationship between emittance and charge from this equation. The rapid dependence of

Ez on z made it tempting to explore the parameter space of a photoinjector to optimize

for short bunches, at the expense of emittance and charge. The remainder of the motivation

for conducting these simulations came from the need to account for experimental conditions

different than those assumed in the initial proposals for the AWA project, including a

smaller diameter photocathode, no sagitta for the laser pulse, small apertures downstream

of the accelerator (the drive-witness beam combiner chicane has a 2.2 cm vertical aperture),

and lower than anticipated cavity fields. One additional consideration in favor of lower

charge is the difficulty in estimating the errors arising from the electrostatic approximation

of the beam space-charge fields used in PARMELA. This code has been compared with

an electromagnetic code ITACA [45] for the TTF case [46] (8 nC, t ,laser = 20 psec at

L-band) by Serafini and Colby [47], but it may cease to be accurate given an order of

magnitude more charge. Finally, the low-Q simulations have been extremely valuable in

guiding the operation of the machine, as well as devising tests of machine performance

based on the available diagnostics.

Obtaining an estimate for the plasma response given the original AWA drive linac

design results from Ho’s 100 nC study [48], with z = 2.9 mm and n = 750 mm-mrad,

the maximum wake-field is initially 250 MeV/m, assuming a Gaussian in time pulse.

Emittance-caused bunch degradation decreases the gradient to an average of 180 MeV/m

as seen by a v ≈ c test particle in the 10 cm long n0 = 2 × 1013 cm−3 plasma. Ho’s work

did not include information about the longitudinal profile of the bunches, with more

recent simulations revealing a steeper than Gaussian temporal falling edge which allows

raising the plasma density to achieve a 750 MeV/m maximum wake. The precise shape of

the beam’s tail is, therefore, crucial to estimating the amount of beam loading this causes

at a given n0 , and therefore the maximum achievable wake-field. The amount of beam

loading is important both in thinking about how to optimize the accelerated bunch parameters

and in measuring the bunch profile actually going into the plasma.

Although the AWA design is fundamentally very different from the present

conception of a photoinjector, the discussion of the beam dynamics present in this machine

will greatly benefit from a brief review of the concepts used to design the majority of

photoinjectors. Many of the current photoinjector designs are intended as drivers for an

FEL, placing much attention on the emittance and energy spread of the beam. The most

important contributions to the final emittance are the thermal, space-charge, and RF

components. The thermal emittance results from the electrons being emitted from the

cathode with some finite random velocity, as happens when the laser photon energy

exceeds the work function of the cathode material. This term is usually a small fraction of

the overall emittance; the rest of the design exercise takes the form of trying to preserve

this original emittance by minimizing the RF and space-charge contributions. Many

theoretical discussions make use of a pure-fundamental mode N+1/2 cavity structure, as

pictured in Figure 23. The transverse and longitudinal field patterns of this mode are,

Ez = E 0 cos(kz)sin(wt + 0) (5.3)

Er = 12 krE0 sin(kz)sin(wt + 0) (5.4)

B = 12 krE0 cos(kz)cos( wt + 0 ), (5.5)

Figure 23. Radial cross-section and electric field pattern of -mode 2 12 cell gun.

but a more realistic field profile will include higher-order spatial components which

become more prominent closer to the irises and thus cause an emittance increase. Even in

the absence of these higher-order terms, Kim’s treatment [49] points out that the time-

dependent nature of the radial RF forces acting on the bunch may result in different

amounts of focusing head-to-tail, or the opening of a ‘bow-tie’ in the transverse phase

space. This effect will be minimized for a launch angle 0 which results in the bunch

exiting the last cell at 90 degrees, which is approximately 0 = 81 degrees for the present

case. The next, and usually the most important effect is the space-charge component of

the emittance. This derives in part from radial nonlinearities in the space-charge defocusing

force, as in the tails of a Gaussian transverse profile. More importantly, a non-constant

current profile will also cause a longitudinal variation in this force, resulting in yet

another bow-tie, usually with the higher current center of the beam at a different betatron

phase than the head and tail. A method of overcoming this effect, as developed by

Carlsten [50], involves finding betatron orbits which are insensitive to small changes in

the beam current. This is called space-charge compensation of transverse emittance and,

in practice, is accomplished with a careful choice of focusing solenoid strength, position,

and drift length between the gun and later accelerating cavities.

The evolution of the longitudinal phase space, often spoken in terms of the

‘longitudinal emittance’, is important to understand in order to minimize the energy

spread for FEL applications, and to enhance the compressibility in magnetic bunch

compression schemes. The longitudinal dynamics will primarily be influenced by RF

compression, space-charge and wake-field effects. Launch phases 0 less than 90 degrees

will tend to shorten the pulse, while the longitudinal component of the space-charge will

lengthen it. For a PWFA driver not using magnetic bunch compression, longitudinal

emittance considerations simply reduce to obtaining the desired current profile with an

energy spread low enough not to cause chromatic effects in the focusing elements that

match the beam into the plasma, with less of a concern about the correlation between

these two variables.

Because of the very high charge and current output of the AWA, and the fact the

initial emittance requirements were not very strict, the design chosen for this machine

greatly differs from convention. Magnetic compression for such a beam may not be

feasible because of the emittance dilution associated with coherent synchrotron radiation

in the compressor bends, a problem which lacks a complete and well-accepted theoretical

treatment at the moment. The AWA design has never been subjected to an emittance

compensation analysis. In both emittance compensation and bunch compression, the

manipulation of the electron distribution is carried out as a function of longitude. In the

AWA drive linac the initial laser spot on the cathode is very large, and subsequent

corrections are all a function of the radius, a fundamentally different approach. Examples

of corrections include laser pulse shaping, space-charge forces, spherical aberrations in

the solenoids and path length differences, all of which are capable of linear and nonlinear

radial corrections on the electron population. The cumulative effect is for the final state

of an off-axis particle to be close to the phase space orientation of a particle born close to

the axis.

The laser pulse delivered to the photocathode has a cupped shape, or sagitta, with

larger radius electrons being emitted before those at smaller radius. This is accomplished

with a custom designed pulse shaper by optically delaying the laser pulse in a discrete set

of radial rings, with the delay thus introduced being much larger than the original laser

pulse length. This has several effects on the electron beam dynamics within the gun

cavity. For a cavity operated in bunching phase, advancing the launch of the off-axis

particles results in less acceleration in the cavity. The space-charge forces for a cupped

electron pulse are also very different than for a disk-like one, having a longitudinal

component which causes an energy spread in the opposite direction. This beam is allowed

to expand to a very large radius (almost 5 cm for some electrons) before being focused

with the nonlinear solenoid, having a spherical aberration with a shorter focal length at

larger radii. The off-axis electrons also travel a longer path, partially reversing the original

sagitta. After undergoing the corrections of longitude, energy, and transverse momentum

as a function of radius, the bunch distribution is ‘frozen out’ by further acceleration in the

linac sections.

The work to optimize AWA operation at lower charge was begun with the version

of PARMELA modified by Ho[48]. As this early version of PARMELA did not include a

treatment of the image charge forces when the bunch is close to the cathode, it was

desirable to switch to a later version of PARMELA which included this capability (Ho’s

work used the code TBCI-SF to correctly model these early electromagnetic effects). The

simulations presented in this section were all performed with a PARMELA modified by

Eric Colby [51], who also ‘set up’ the AWA linac problem by using the code SUPERFISH[52]

to solve for the cavity RF modes and POISSON for the solenoid fields, to be used as an

input to PARMELA.

The features of the AWA designed manipulate the beam as a function of radius

proved to work against achieving the best results at lower Q . The biggest problem is the

long drift length after the gun to the first focusing element (~12 cm) and to the first linac

cavity (~24 cm). Even at Q =25 nC, the space-charge induced energy spread the beam

accumulates in the drifting state causes chromatic focusing and emittance dilution in the

solenoid, as well as bunch lengthening before further acceleration in the linac.

After carrying out simulations with a variety of starting conditions, the operating

point shown in Figure 24 appears to be a good compromise between bunch length, charge

and emittance. As an aid in finding the best possible parameters, sets of simulations were

automatically run while varying one of the input parameters such as the initial laser

radius, the solenoid strengths, and the injection phase. Additional simulations were

performed while scanning a two-dimensional space in the input parameters, performing

as many as 64 simulations to cover an 8 × 8 grid. To achieve a reasonable emittance, this

simulation uses a much smaller launch phase, 0 = 37 degrees, than the 81 degrees called

for in Kim’s theory. The smaller 0 causes a bow-tie that partially cancels the chromatic

effects in the solenoid, which are already diminished because of the reduced energy

spread at lower 0. The emittance in Figure 24 escalates to an unphysically high value in

the solenoids around z = 15 cm and z = 150 cm , because the emittance estimate does not

take into account the twisting motion of the beam in the solenoid, which is a function of

radius due to the lens nonlinearity. The fact that n appears to become smaller in the drift

space after emerging from the second linac tank may be a sign of emittance compensation

in the beam system. A plot of the longitudinal phase space, shown in Figure 25, reveals

that a negative correlation between energy and z is possible by off crest operation in the

linac. This solution does not take account of the energy lost to wake fields in the linacs,

which takes away much of this correlation.

Q 25 nC

0 37 deg.
E gun 80 MeV/m
E linac 16 MeV/m

n 350 mm-mrad

z 2.2 mm

Table 4. Parameters of optimum Low-Q case.

Figure 24. Standard case simulation output.

Figure 25. Longitudinal phase space after the linac.

5.3 Diagnostics

In the course of this work, the beamline after the accelerator, as well as the layout

of the plasma experiments described in Chapters 6 and 7 have undergone many changes.

The diagnostics described in this section have been used as building blocks to provide the

necessary beam measurements in each experimental configuration.

5.3.1 Transverse profile diagnostics

High resolution measurements of the transverse beam profile have been essential

for many stages of the experiment. The pepper pot emittance diagnostic described at the

end of this Chapter, in which the resolution should be comparable to the hole diameters

(~300 microns), is one such application. The focused electron beam spot size inside the

plasma chamber, with a beam radius as low as r = 250 microns, is one of the pieces of

data used to estimate the peak beam density, a crucial parameter for the plasma experiments.

In the course of the experiments, beam imaging based on phosphorescent screens, optical

transition radiation (OTR), and Cerenkov radiation have been developed.

Optical transition radiation (OTR) can be used in several ways to reveal detail

about the beam distribution under test. Because this radiation is peaked at an angle

=1 from the beam direction, the radiation pattern from a finite divergence beam will

reveal information about its angular spread. If an OTR surface is to be used simply as a

transverse profile monitor, the light signal from a single electron will appear to come

from a point source if the CCD camera is focused onto the OTR surface, different form

the case of angular divergence measurements, where the camera must be focused at

infinity. The resolution of this method at optical wavelengths can, in theory, be comparable

to the wavelength of the light, according to reference[53]. This reference also gives an

estimate for the resolution in an experiment, which achieved 14 µm resolution for light

centered on the visible region. The method of OTR imaging was chosen to diagnose the

focused beam spot inside the plasma chamber under plasma-off conditions because it has

the advantage of not saturating under intense beam conditions. A 30 nC beam mostly

contained within a circular region with a 1 mm diameter may saturate the phosphor

coating. The surface generating the OTR used in the ANL experiments was simply a

front surface silvered mirror mounted at 45 degrees to the beam direction. One disadvantage

of using OTR is a relatively low intensity light output. This forces the use of a wide-open

camera iris, also needed to collect light from large angles, and some image amplification.

The video signal, therefore, contains a larger noise level than for a typical phosphor

screen image.

5.3.2 ICT’s and Faraday cups

An accurate charge measurement can be performed nondistructively with one of

several ICT’s used in the system. Since the decay time of the signal from this device is

short (about 20 ns), the electronics processing the signal can be made to reject the any

dark current contribution. The ICT, therefore has an advantage over the Faraday cup,

which is an inherently destructive measurement, is sensitive to the dark current, and can

suffer losses due to secondary emission and arcing.

The ICT’s used are commercially available from Bergoz, and are claimed to have

a linear response with bunch rise times of a few picoseconds[54]. Since it is difficult to

find an electronic source capable of such short pulses other than a photoinjector, this

claim is difficult to test on the bench. ICT’s have become an accepted way to perform

this measurement and are used in a number of photoinjector facilities having shorter

pulses than AWA. The ICT’s used in the present work have been calibrated with pulses

as short as 1 ns, in the arrangement depicted in Figure 26. The ICT acts as a pulse

stretcher, and, therefore, does not reproduce the original pulse shape. It does, however,

reproduce the pulse area, as given by the equation given for the voltage Vout into a 50

ohm load, with N being the turns ratio,

∫ Voutdt = ∫ Ibeam dt (5.6)

With a 100 ns integration window on the oscilloscope, the measured value for N was

40.1, where the manufacturer’s value was 40.0. The measured value of N did not

significantly change upon using a longer ~100 ns excitation pulses. The difference between

the measurement and the quoted value is within the error of the measurement method, so

the quoted value has been retained for use as the instrument’s calibration value. In the

actual beamline application, there is a significant (~30 m) length of cable between the

ICT and the signal processing electronics. The test in Figure 26 was repeated using

different cable lengths between the ICT output and the oscilloscope, up to 120 m. The

dispersive properties of the RG-58C/U cable used significantly alter the pulse shape, but

the integrated signal suffers little loss, 4% in the cable used for the experiment. This

amount of loss is comparable to the DC loss of a cable terminated at 50 ohm.

Figure 26. Test setup for the calibration of the ICT.

Faraday cups used in the beamline belong to one of several types. The most

common of these combines a lead beam stop with a phosphor screen profile disgnostic.

To allow the camera to view the phosphor surface, the face of this diagnostic is inclined

45 degrees to the beam direction. This geometry degrades the accuracy of the measurement

by allowing a fraction of the particles to scatter in the metal and emerge at >45 degrees

without being stopped, amounting to probably a 15% signal loss when using lead. Since it

was necessary to measure the amount of charge entering the plasma chamber, this also

required the use of a Faraday cup, this time mounted at normal incidence to the beam.

5.3.3 Prompt radiators and Streak camera measurements

One method of gaining information about the temporal characteristics of the bunch

is by intercepting the beam with a prompt radiator and directing the light to a streak

camera. Cerenkov radiation and optical transition radiation (OTR) are both feasible sources

of optical radiation, with a Cerenkov plate generally giving a stronger signal. The geometry

in Figure 27 is useful for a solid radiator, such as quartz. The Cerenkov angle for quartz

(index of refraction n = 1.46 ) is ~45 degrees. The radiated phase-fronts form a cone in

the material, but the collection optics can select only a small part of this cone, in this case

the radiation whose azimuthal angle is very close to the direction of the plate inclination.

The diagnostic can be designed so that the radiation exits the dielectric at the Brewster

angle, which suppresses internal reflections for the polarization of the photons of interest.

Multiple reflections can cause an apparent lengthening of the pulse, and were further

suppressed by roughening the upstream surface of the dielectric. An idea for further

suppressing these reflections in future experiments is to paint this surface with an optical


The light emerging from the diagnostic must then be collected and sent to the

streak camera, a Hamamatsu model C1587 for the present experiments. At the light input

of this instrument, the radiation is collimated with a slit, with an opening of between 10

and 40 microns appropriately set for obtaining the highest resolutions for this instrument

(approaching 1 psec). Beam trajectories parallel to the reference in Figure 27 but higher

or lower on the page will take a different amount of time to arrive at the streak camera.

To avoid this time spreading, the optics from this diagnostic to the streak camera must be

able to map the signal from two electrons that differ only in y (out of the page) onto the

slit. Whenever the experimental geometry requires an image rotation between the source

and the streak camera, this can be accomplished with a dove prism (see reference[12]), or

by a similar choice in relay mirror geometry, as was done at AWA.

Applying this diagnostic to the experiment to determine the extent of beam self-pinch

in the plasma, it was necessary to obtain both temporal and transverse information about

the beam distribution. The imaging optics for relaying the light to the streak camera

consisted of two high-quality achromatic lenses (model AAP 1000.0-76.2 from CVI

Laser, Inc.) with a 3 inch diameter and 1 meter focal length. The use of achromats

enables the shorter wavelength Cerenkov light to be transmitted with very similar optical

properties to the incandescent light used to align the system. The lenses were mounted 1

meter away from the Cerenkov radiator and the streak camera, with a 2 meter space

between them. This enabled an undistorted viewing area at the source with a diameter

larger than the diagnostic. This feature is important, as an improper choice of optics will

truncate the signal at some radius, and also result in aberrations close to this radial limit.

The resolution can further be degraded when the beam multiple-scatters in the

diagnostic, both within the quartz plate, and at the copper foil used as an air-vacuum

transition in the focusing experiment. The transverse resolution of this diagnostic was

estimated to be 80 µm, which is negligibly small when added in quadrature with the

≅ 250 m beam sizes measured in this experiment.

Figure 27. An electron beam intercepted by a Cerenkov radiator.

5.3.4 Pepper-pot

The emittance of a space-charge dominated beam is very difficult to measure

using the quadrupole scan technique often used for GeV-energy beams. A 'pepper-pot'

measurement[55] overcomes this limitation because most of the beam charge is scraped

away. This device is simply a metallic plate with a pattern of holes going through it. It

acts as a beam mask, by either stopping or significantly scattering the electrons crossing

the bulk of the material. The electrons selected by the hole regions never come into

contact with the material, and emerge as a series of beamlets. The dynamics of the

surviving electrons can be made, to a good approximation, free of space-charge influences

over significant drift lengths. After drifting, the transverse spreading of these beamlets

from their original size, as well as the location of the beamlet centroids, is measured

using a phosphor screen. For low enough charge, this is a measurement of the angular

distribution of the electrons going through each hole. The data can be analyzed to yield

not only the emittance but also the beam divergences in both planes, which is a valuable

measurement that is difficult to accomplish through other means.

The pepper pot design appropriate for measuring the AWA beam's emittance was

influenced in part by the large amount of charge per bunch as well as the relatively large

emittances anticipated. For the first step in this design process, the ratio Rb of the

space-charge to emittance terms in the envelope equation for each beamlet, following the

treatment of Rosenzweig and Travish [56], gives quick estimate of the amount of space-

charge forces still present in the drift after the pepper pot, but adapting the result to a

mask with round holes instead of slits:

I  r 

Rb = , (5.7)
I0  n 

with I being the peak current prior to collimation, I0 = 17 kA , n the normalized emittance

and r the radius of the holes. The pepper pot should also be designed with a hole size

much smaller than the anticipated beamlet size on the phosphor screen. Anticipating the

large angular spread of the AWA beam, it was important to have as large as possible an

angular acceptance for the beamlets, defined by the ratio of the hole diameter to the plate

thickness. Electrons having comparable angles will effectively see a smaller hole and

contribute a smaller signal. The same geometric effect will guarantee that a significant

number of such electrons will encounter the inner surface of the hole and scatter. Small-angle

scatterings will appear as a halo around the main beamlet image on the phosphor and

distort the measurement, particularly because outlier particles in the transverse phase

space contribute a disproportionate amount to the final emittance measurement. For this

reason, much care was taken to measure any true outliers accurately. The final design

criterion was, then, to have a hole spacing equal to six times b, the anticipated rms

beamlet size on the phosphor screen. Given a slightly irregular image, this also helps in

identifying from which hole a given part of the image may have originated.

The final choice for the emittance mask design, shown in Fig. 28 has 26 holes of

320 µm diameter stacked in a honeycomb pattern on a 1 mm thick 85% tungsten alloy.

The spacing between the holes is 3 mm. The ram-electron discharge machining (EDM)

process used to fabricate the mask resulted in slightly tapered holes, with a diameter

approaching 400 µm toward the downstream side. This taper greatly diminishes the

possibility of an electron actually hitting the side of the hole, even in the case of a slight

misalignment with respect to the beam axis, and thus represents an advantage for accurately

measuring the emittance.

Figure 28. The emittance mask design.


The design criteria outlined in the previous paragraph can now be evaluated,

using some assumptions about bunch characteristics. The most important resolution limit

in this measurement will be due to the finite hole size. A beam with no angular divergence

will have an image that reproduces the mask pattern on the screen. The transverse rms

size in of a uniformly illuminated circular region is r 2 , where r is the radius. The

beamlets of this perfect emittance beam will have a size x = 80 µm, which will seem

like a 0.9 mrad divergence, or an emittance of n = 28 mm-mrad assuming a 5 mm rms

spot size and 15 MeV. This number is the resolution of the measurement system due to

finite hole size, and is much smaller when added in squares to the actual beam emittance.

A similar resolution limit results from the finite space-charge force within a single beamlet.

Using the same numbers, Q = 20 nC, and a beam current which is constant within its

0.75 cm (25 ps) length, the beam radius will evolve according to r ′′ + k2 r = 0 (keeping

only the space-charge term), with k 2 = 2 renb 3 3

, with the appropriate factor of 1 for

the self-force of a relativistic beam. The beam density in this example is 1.06 ×1011 ,

making kz = 0.1 . The x = 80 µm spot size computed above would increase by another

0.4%, making the space-charge limit to the resolution negligible in comparison to the

resolution limit due to finite size holes. To say this another way, the ratio of the emittance

to space charge terms is equal to Rb = 10−3 for typical beam emittances. One additional

limit to the resolution comes from the fact that the bulk of the charge does not actually

stop within the mask, and is, therefore, present after the beamlets have been formed. The

space-charge force from these scattered electrons can interact with the beamlets and,

likewise, limit the measurement resolution. The large angular spread of the scattered

electrons, which are uniformly distributed within a solid angle approaching 2 , means

this effect is only significant in about the first 5 cm after the mask. The nonlinear kicks

will be confined to an area where the scattered electrons’ radial profile closely matches

the original beam size, or within ~1 cm from the mask. The error in centroid position for

a beamlet near the core can be up to 0.7 µm, with a smaller error at larger radii. Treating

this like a random error in beamlet centroid position, the effect on resolution is still much

smaller than that from finite hole sizes. It should be noted that the linear forces, although

they will not affect the emittance measurement because the beamlet pattern will expand

uniformly, will contribute to errors in measuring beam divergence.

Emittance Measurement

Figure 29 shows a CCD image of a 20 nC beam generated during the first trial of

this pepper pot design at ANL. This data has already had a dark current image subtracted

from it; the dark current image is taken at a different time but remains relatively constant

on the order of minutes. The photo-current signal is characterized by a set of intense

central peaks, but secondary peaks and tails are also present. Even though the peaks are

very narrow, the tails sometimes extend more than half way to the adjacent peak, as in the

lower right hand corner of the image. This can complicate the process of data analysis

because it may not be possible to decide from which hole a given part of the signal has

originated. At the same time, the tails are intense enough above background to be discernible

from the background by visual inspection alone, a feature very helpful at the analysis


Figure 29. Background subtracted pepper-pot image of a 20 nC beam.

A more useful way of viewing this data is to project a row of beamlets onto the

horizontal axis. The projection of a rectangle drawn around all of the signal belonging to

the second (from the top) row of beamlets is shown in Fig. 30. The signal has two sources

of background in it: the noise associated with the CCD camera having a value of 14.5

(full video is 255), and the background associated with electrons scattered by the full 1.5

mm width of the tungsten plate, with a value of 0.35. The first source of background can

most easily be minimized by using a wider camera aperture (and less gain in the CCD

camera's internal amplifier), and further reduced by using a cooled CCD array. The

second source of background is small enough and constant enough to be easily subtracted

away. The background in the signal data of can be compared to the a (smoothed)

background sampled by moving the integrating rectangle between the second and third

rows of beamlets. The presence of tails, as found to the right of the first and second peaks

in this data, causes the signal to rise above the background level. This data also shows

that the original design criteria, of sufficiently separating the peaks, has been met. The

only possible exception is the presence of tails long enough to approach the adjacent

beamlet. The solution to this problem—increasing the spacing between holes—must be

rejected because it would result in the beam to be sampled in too few places. The

remainder of this section describes the methods used for the analysis of this data.


26 signal

projected intensity






-50 0 50 100 150 200 250 300
position (pixels)
Figure 30. Projected image from one row of beamlets and (smoothed) background sampled
between the rows.

Data Analysis

One strategy for analyzing pepper pot data is to assume a functional form for the

data, such as a set of Gaussians, and make a fit to the data. If a Gaussian is not the correct
function, this will lead to a large and make it difficult to estimate the error due to the

choice in fit function. The “bow-tie” phase space profile commonly seen in photoinjector

simulations has a much sharper than Gaussian peak, and a functional form which changes

depending on the orientation of the phase space ellipse. Thus, the choice for a fit function

is a difficult one, further complicated by any inherently 3-D effects in the real system.

The present analysis uses an alternative approach. Since the pepper pot data is a

measurement of the angular distribution as sampled at each hole, the data can directly be

translated to a phase space distribution. The data can then be treated with a linear

transformation on the phase space capable of removing the x - x ′ correlation, which also

has the property to preserve the actual and measured emittance. When the data is projected

onto the x ′ axis, it more closely resembles a Gaussian, since the noise and fluctuations in

the individual peaks are washed out. After these steps, it is much easier to interpret the

data by appropriate choice in fit function, or by directly computing the rms width of the

distribution, or the 95% rms width, a number much less senstive to noise.

The first step in the data analysis is to define a set of active areas within the image

associated with each beamlet. A set of rectangles serve this purpose (see Fig. 31), as well

as to define the origin of the local coordinates for each beamlet. Depending on whether

the beam under test is converging or diverging in x and y , the peaks in the data will have

a separation different than that of the holes in the mask: if (xi, yi ) are the set of vectors

for the mask holes, then the peaks on the phosphor screen will be at (ax i,byi ), and

fi (x, y) is the data of the i th rectangle (beamlet) in its local coordinates. At this stage in

the analysis, a and b (the amount that the rectangles are stretched in either direction) are

set by the user and can be varied around the optimum value. Finally, quantity ∑ ∫ fi(x, y)dy
is used for computing the emittance in the x-direction, but this must first be justified in

terms of the definition for the emittance chosen for this measurement.

Figure 31. The grid used for analyzing the emittance data.

This analysis makes use of the statistical definition for the emittance of a particle

distribution in one transverse dimension:

=〈 x 2 〉〈x' 2 〉 −〈 xx' 〉2 ,

where the brackets are statistical averages. The method is simplified by the fact that is

invariant upon a linear transformation of the variables, thus if x' is replaced by x ′ + Kx

(K is a constant), one obtains the same result:

= 〈x2 〉〈(x' +Kx) 2 〉− 〈x(x' +Kx)〉 2 = 〈x 2 〉〈x' 2 〉− 〈xx' 〉 2 .


For a given value of k, the second (correlation) term will vanish. Because the difference

of the two terms is a constant, this value of K also minimizes the first term. The

emittance, therefore, can be measured by evaluating only the first term and minimizing

with respect to K .

Evaluating the first term, 1 , in the case of the actual emittance data yields:

1, x = x∑∫ dxdyx' 2 fi ((x' − axi ), y' ), (5.10)


where x is the RMS beam size at the mask and xi is the x-location of the ith hole. This is

the same type of linear transformation as proposed above, therefore, 1, x equals the

emittance upon minimizing with respect to a.

For the data presented (Q = 20 nC), x is .45 cm and x′ = 2.0 mrad, resulting in

an emittance of n,x = 280 mm-mrad, assuming a 15 MeV beam. This data was taken

without the benefit of simultaneous spectrometer measurements, and cannot be used in

predicting the normalized emittance due to the uncertainty in beam energy. It is possible

that the energy may have been as low as the 10-12 MeV range during this particular day

of operation. This would correspond to a lower emittance value, approaching 200 mm-mrad,

and in better agreement with the later measurements at slightly lower charge (Q = 7 −14

nC) where the emittance was n = 130 and 149 mm-mrad, respectively. These numbers

are slightly better than the PARMELA prediction, in part because the beam collimation at

several apertures in the experiment is not accounted for in the simulations.

5.4 AWA drive linac operation

Any direct comparison between drive linac performance and simulations requires

detailed knowledge about all the variables in the system. The inputs to a simulation

include the cavity modes, amplitudes and phases, the laser spatio-temporal distribution on

the cathode, the launch phase, and the magnet settings. For the real system, the description

of the measurements performed to establish the working parameters will reveal that

enough pieces of information are missing to make a direct comparison with simulation as

yet unfeasible. This is unfortunate, as a direct comparison to simulations has the capability

of either spotting some error in the theory or pointing out a weak component in the


The electron beam dynamics in later parts of the linac system are, to a large part,

governed by the emission from the cathode. Included in this are the transverse profile and

temporal extent of the laser on the cathode, the bulk quantum efficiency of the cathode

and any spatial variations, the electric field in the cavity (Schottky enhancement effect),

and the relative timing between the laser and the RF field.

The transverse laser profile is measured by splitting a part of the laser light before

it strikes the cathode, and allowing it to propagate in air an equivalent distance where the

profile is measured with a CCD camera focused onto a UV to visible downconverter (a

specially processed YAG wafer supplied by Star Tech Instruments Inc.). A typical laser

spot, displayed in Figure 32, contains some degree of clustering of the laser energy, or

hot spots. The impact of such a laser spot on the subsequent electron beam dynamics

cannot be predicted with only a 2-D model, and thus must remain a project for the future.

The temporal extent of the laser is monitored with an autocorrelator (before amplification

and frequency doubling), and confirmed with a streak camera.

Figure 32. Laser transverse profile on cathode.

Measurements of the average quantum efficiency of the polished magnesium

cathode have given a result as high as 1.5 ×10 − 4 , somewhat lower than the 5 ×10 − 4

achieved at Brookhaven’s ATF. No attempt has been made to quantify any variations of

this quantum efficiency as a function of position on the cathode, but prolonged use brings

the appearance of darkened spots on the cathode surface. The most likely cause for these

is contamination from pump oil. These dark spots are probably associated with a drop in

the quantum efficiency.

The launch phase is not an easy parameter to measure for any photoinjector

application. A variety of techniques can be used, with varying amounts of accuracy.

ATF’s solution is to process the signal from a stripline beam position monitor and mix it

with the RF. This gives the final phase of the electron bunch; the initial phase can be

inferred by using a model for the accelerating field in the gun to solve for the amount of

slippage. A less direct method involves measuring the beam divergence after the gun exit

and fitting this to a model of the gun’s performance. For the AWA system, a reference

point is established by changing the phase to maximize the charge (measured at the first

ICT). The Shcottky enhancement effect is known to boost the quantum efficiency at

0 = 90 degrees, when the electric field is maximum, although not as sharply as at other

facilities due to the higher photon energy used at AWA. For performing a quantum

efficiency measurement, the amount of charge scraped from the beam before it arrives at

the first ICT depends on the solenoid and linac settings. The apparent quantum efficiency

peak will, therefore, depend on the charge lost up to that point. One additional attempt to

establish 0 involves changing the phase in the direction of 0 = 0. For small amounts of

charge (below 4 nC), simulations show that the beam current has a sharp cutoff, with no

beam emitted for 0 < 0 , and a stably propagating beam for 0 > 0 . An attempt to

observe this phenomenon in the laboratory produced ambiguous results, the two methods

used for the measurements showing a 20 degree difference among them. By adjusting 0

in the direction of zero, the beam spot at the spectrometer screen disappeared much

sooner than the ICT signal.

Attempts to establish the field strength in the gun cavity have been the least

precise, and have thus raised the most questions. The drive gun has neither a coupling

probe to measure the field, nor enough space after it for a beam energy measurement. The

net input power, measured to be as high as 2.0 MW, would result in a 98 MeV/m peak

on-axis acceleration gradient in the absence of other losses (the design value is 1.5 MW

and 92 MeV/m). One possible loss mechanism is dark current loading of the cavity,

which is difficult to quantify because of the inability to measure the total amount of dark

current, including electrons that never emerge from the cavity. An additional method of

obtaining cavity field information is through the response of the beam to the gun solenoid

- too strong a setting will cause overfocusing and beam loss. When compared to the

PARMELA results, with the gun field as a free parameter, this data is most consistent

with a 55 MeV/m [57]. This number seems too pessimistic; considering the high amount

of charge produced, a value around 65-70 MeV/m would be more credible.

Operation of the linac requires a choice for the RF input power and phase to the

cavities, the magnet settings and the laser profile and timing. Except for the laser timing,

it ought to be possible to reset all the other parameters to some interesting operating point

from a past run. The laser timing can then be set in reference to the delay which maximizes

the ICT signal. The original operating point is established in reference to the cavity phase

shifts which maximize the dark current energy peak on the spectrometer; the cavity RF

inputs are set to the maximum value where arcing is not a problem. Upon setting the

focusing lattice to achieve good beam transmission through the machine while avoiding

gross overfocusing, the beam energy, energy spread, and bunch length can then be

measured. The fairly accurate beam measurements required for the focusing experiment

of Chapter 6 yielded the results summarized in Table 5. Since this experiment intended
to measure radial dynamics and not wake-fields, it was not necessary to expend a great

deal of effort in optimizing for short bunches. It is hoped that more effort in this regard

will eventually produce even shorter pulses, resulting in higher acceleration gradients in a


Q 7-14 nC
Bunch length 25 +/- 3 psec FWHM

n (Q=7 nC) 130 mm-mrad

n (Q=7 nC) 149 mm-mrad

Eb 14.5 MeV

Table 5: Beam parameters for the focusing experiment.

5.5 Creation of a witness beam in the same cavity as the drive beam

For the PWFA experiments it was necessary to probe the wake fields left behind

by the drive bunch with a separate witness beam. One possible choice was to use the

beam from the witness gun, having 200 pC and 4.5 MeV. The biggest drawback of this

approach was the problem of having to focus both beams, which have different energies,

with the same matching solenoid. For this to work, the witness beam must come to a

waist in the middle of the solenoid and to another waist at the beginning of the plasma,

while the drive beam has only the second waist. The witness beam is also susceptible to

the space-charge force of the drive beam, with large steering errors possible at the point

where the two beams are combined. At the final waist, these steering errors would have

to be measured with both beams on, a difficult measurement to do in the presence of the

two orders of magnitude more intense drive beam. In view of all this, it was decided to

pursue a different method of generating a witness beam for the PWFA experiment, that of

generating the witness beam in the same cavity as used for the drive beam.

If two laser pulses are incident on the photocathode in the same RF cycle, but

separated by some time, the result will be two distinct electron bunches. Several factors

influence the relative timing between these bunches. Launching the bunches off-crest in

the gun will tend to move them closer together (RF compression, about a 20% effect).

Longitudinal space-charge will push them apart. The two beams also have very similar

final energies. The energy of the witness pulse can be increased by off-crest operation of

the linacs. In simulations, this energy increases countered by longitudinal wake fields in

the linacs, which will lead to an additional 0.47 MeV energy loss for the witness beam in

this case. A witness beam with a slightly higher energy than the driver was desirable

because we wanted to measure an energy increase in the witness. It was hoped that the

witness beam energy spectrum on the spectrometer could always be distinct from the

drive beam, including the no-plasma background. For the purpose of matching both

beams into the plasma, it was also necessary for their radial dynamics to be very similar.

To equalize the radial space-charge forces of each bunch, the two laser pulses must have

the same power density, but with a smaller radius witness bunch. The optics system to

create these two laser pulses consisted of two coaxial mirrors oriented in the same

direction. The front mirror has a 1.6 mm hole in its center, allowing this part of the laser

pulse to reach the back mirror and be delayed in time relative to the rest of the pulse,

mostly reflected by the front mirror. Inserting a card between the two pulse-shaping

mirrors blocks the trailing pulse and terminates the witness beam.

The effect of the pulse splitter was simulated with PARMELA by delaying

photocathode emission for all electrons within a given radius. The final longitudinal

phase space from this simulation (Figure 33) does not include the effect of wake fields in

the linac on the witness beam energy. Computing these losses with the code ABCI[58]

results in a 0.47 MeV less energy for the witness, making its energy distribution blend

with that of the main beam.

Figure 34. is a streak camera picture of a 16 nC drive beam and a 4 nC witness

beam and a 6 mm separation between the pulse shaping mirrors. This witness pulse is less

than optimal because of the high amount of charge and a bunch length greater than a

plasma half-period.

Figure 33. Longitudinal phase space of a drive and witness beam created by delaying
emission by 40 psec. Wake-fields would lower the energy of the witness by an additional
0.47 MeV, causing it to blend with the drive beam’s energy distribution.


Current (arb. units)




0.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0
t (psec)

Figure 34. Streak camera projected image of drive and witness beam.


This Chapter has examined the AWA photoinjector dynamics from the point of

view of performing beam-plasma experiments in the underdense regime. This work

identifies a beam in the range of 20-30 nC as the most suitable as a plasma wake-field

driver. The simulations performed at this operating point address all final beam quantities

of interest: emittance, energy, energy spread, bunch length, and bunch longitudinal profile.

For the actual operation of the machine, the initial simulations have provided only a good

starting point. Even if somewhat loosely coupled to actual laboratory practice, the

simulations have provided valuable insight into the underlying beam dynamics. The

process of conducting these simulations has generated many ideas about constructing

tests of beam and accelerator conditions, as well as how to streamline machine tuning

when seeking to obtain a specific set of conditions.

Also covered were the diagnostic techniques necessary to characterize both the

initial and final beam distributions in the plasma experiment. All initial beam measurements

are of the time-integrated sort and are not sensitive to time dependent effects such as the

bow-tie shape in the phase space. A time-dependent emittance measurement was originally

considered, but abandoned primarily due to an anticipated lack of signal after emittance

mask collimation. Apart from the time-dependent considerations, the beam instrumentation

developed—charge measurement, emittance, bunch length, and energy—gives an accurate

picture of the actual beam conditions in the plasma experiments.

Chapter 6. Drive Beam Focusing Experiment in the Blow-
out Regime.

From the drive beam perspective, an experimental demonstration of high-gradient

plasma wake-fields in the blow-out regime can be divided into two separate problems.

When radial spreading of the drive beam is not significant, such as in the case of very low

emittance, the given drive beam must be shown to excite plasma radial and longitudinal

wake-fields in accordance with simulation results. For larger emittance beams, the radial

confinement, emittance-driven erosion, betatron oscillations and associated wake dephasing

effects must also be adequately understood. Since each of these areas is a difficult

research topic on its own right, it was a natural choice to attempt an understanding of the

radial dynamics first.

Past underdense propagation experiments have used beams much longer than a

plasma wavelength, typically the ~1ns long beams from an induction linac. These

experiments have quantified the long-term, or asymptotic, erosion rate due to a combination

of inductive and emittance mechanisms, but do not include measurements of the transient

erosion behavior, important for predicting the propagation characteristics of short (< p )

beams. These experiments have also been concerned with a finite in radius plasma, as

would result from either pre-ionization with a laser beam or self-ionization by the electron

beam. The remainder of experiments with a short plasmas (thin and thick plasma lenses)

[59]and long plasmas [37] have been in the overdense regime, where the focusing force

depends on nb and not n0 , and the radial dynamics are generally very different. Only one

experiment [60] has made use of a photoinjector-derived beam; such beams are often

characterized with a slice emittance much smaller than the projected emittance, a concern

which will be revisited toward the end of the chapter. The present experiment is, therefore,

the first measurement of the focusing properties of a short beam in an underdense


A separate goal for this experiment is not only to obtain a predictable response

from the beam-plasma system, but to observe beam radial confinement which is adequate

in terms of PWFA performance with the beam on hand. It was not obvious that the proper

conditions for this would be met from the start, as the initial emittance estimates were in

fact larger than those encountered in the experiment. It was uncertain that the combination

of emittance, charge, and bunch length would satisfy the requirement from Chapter 3,

nb, eq n0 ≥ 2.0 , to make long term propagation in the underdense regime possible. Upon

meeting this requirement, there is also a choice about the length of the plasma which is

sufficient to probe the long term propagation characteristics. The transient part of the

erosion takes place in the first ≈ 4 eq , so the length chosen for the plasma, Lp = 12 cm is

over twice as long. Extending the plasma to lengths much beyond this, where the drive

beam core is allowed to lose most of its energy, is undesirable because, in an experiment

where the energy spectrum of the spent beam is not diagnosed, the fractional energy loss

can become a large uncontrolled parameter. For the n0 ≅ 1 ×1013 cm −3 plasma used, the

fractional energy loss is below 15 %, and the effect on beam propagation will be small in

case of a discrepancy between the actual and simulated energy loss. The choice of

plasma length was also dictated by the amount of position and pointing jitter of the

incident electron beam. A self-guided beam will propagate in the direction of the initial

centroid motion; a 10 mrad initial error becomes a 1 cm error in the beam centroid x at

the end of a meter long plasma. The jitter encountered with only a 12 cm plasma length

confirmed that measurements would have been rendered far more difficult if using a

longer plasma.

As this Chapter initiates the discussion of the plasma generation hardware, the

first Section is devoted to a description of the plasma cell used in the experiments and the

modifications from its original form.

6.1 The plasma source and modifications.

The plasma source used for the focusing and acceleration experiments was first

developed by Rosenzweig[10][11][37] for overdense-regime experiments of plasma wake-

fields at the AATF. This device is a DC hollow cathode arc discharge (HCA) using

10 −4 −10 −3 Torr of Argon (an excellent review of HCA’s can be found in reference [61]).

This application called for a nearly constant-density plasma of variable length up to 30

cm. The cathode consists of two concentric Tantalum tubes. Unlike previous such HCA's,

the gas was fed into the device in the annular region between the tubes in order to isolate

the incoming electron beam from the higher gas pressure, as well as from the irregular

plasma profile which occurs in the cathode region of these devices. The plasma column is

confined in the radial direction with an axial magnetic field in the range of several

hundred Gauss. The plasma density can be varied by adjusting the gas flow, the arc

current, or the confinement field.

In planning for the present work, the peak acceleration field anticipated was

around 250 MeV/m, with a deceleration field of around half this number. With the

original plasma length, it seemed possible to completely deplete the energy for a large

portion of the drive beam. This is undesirable because at the point of depletion the bunch

profile changes drastically, uncontrollably altering the radial and longitudinal wakes. A

new length was chosen for the plasma column, 15 cm. To shorten the column it was

necessary to alter the confinement field pattern and to bring the two electrodes closer to

the center of the device. New pole piece inserts were fabricated for the confinement

solenoid end-plates to achieve the correct geometry for the magnetic field in the new

configuration. After operation of the now more compact plasma source, excessive heat

buildup became apparent on glass insulator and O-ring seals closest to the cathode. This

problem was solved with a bigger multi-layer heat shield between the cathode and the

O-ring, use of 500° F rated vacuum O-ring grease, and forced air cooling of the glass.

The modified plasma source is shown in Figure 35.

Figure 35. The modified plasma source, shown without beam diagnostics.

An ion channel matched beta-function of 0.5 cm for a 1 ×1014 cm -3 plasma

introduced another set of constraints on the experimental device. The beam would have to

be focused with a strong lens close to the cathode. A solenoid was chosen for this

because the focal length could be altered by changing a single control. Achieving this

beta function also meant a beam convergence of 32 mrad rms, or an unobstructed cone

with an opening angle of 7 degrees to keep beam scraping at the 3 percent level. This

requirement is at odds with the recommendation that an HCA cathode be constructed

with a tube length much greater than its diameter, since the plasma density in our device

becomes large near the tip and the beam must travel the length of the tube with no

focusing. The first modification involved a new design for a shortened cathode holder

cooling channel and input flange, and cutting 0.5 inches from the tip of the cathode tube.

During the first plasma wake-field experiments, 3 nC out of 15 nC were scraped from the

drive beam, owing mainly to the fact that the beam energy was lower than expected. The

entire cathode assembly was redesigned for the focusing experiment using larger diameter

inner and outer tubes of 5/8" and 3/4" outer diameters. This improvement greatly minimized

the amount of scraping, but still required careful alignment of the beam to the aperture


The plasma density and profile were measured with a Langmuir probe and also

with a 140 GHz interferometry technique. The probe used during beam experiments was

constructed from 0.32 mm thick Tantalum wire, and a Boron Nitride insulator. This probe

could be remotely actuated to measure the plasma density at any time during the experiment.

Using a slightly different arrangement, the radial dependence of the density can be

integrated and directly compared to the mm-wave interferometer data, which measures

the same quantity by measuring the phase shift of the waves as they pass transversely to

the plasma column. The two techniques disagreed by 20%, with the probe registering the

lower density. The interferometry technique is considered the more accurate one, therefore

this measurement serves as a calibration of the probe method.

In order for the plasma to be coupled to the linac, it was necessary to bridge the

very different vacuum requirements of the two devices. One solution is to separate the

two vacuum chambers with a foil, thin enough to keep the multiple scattering emittance

increase under control. Keeping this emittance increase low usually requires placement of

the foil near a beam waist. As heat radiation prevented placing the foil near the plasma

start, and a separate waist upstream would have required the complexity of having additional

quadrupoles, the idea was ultimately rejected. Another possibility for isolation is by using

differential pumping, the method chosen here. A 1500 l/s Sargent Welch turbomolecular

pump was attached to the vacuum chamber upstream of the cathode, reducing the pressure

by a factor of 400. The AWA beam line also contains two additional 150 l/s turbo pumps

between the plasma experiment and the drive gun. The pressure in the gun cavity typically

increases by about 2 ×10− 7 Torr during plasma operation. This extra load on the gun is

primarily due to argon atoms, so the risk of contamination is minimal. The added gas

pressure has not been observed to increase the incidence of arcing in the gun or linac

cavities, except at the highest gas flow rates.

6.2 Experimental Results

The plasma chamber used in the experiments is shown in Fig. 36, with electron

pulses derived from the AWA beam entering from the left. The beam energy used for

these experiments was E = 14.5 MeV. Beam diagnostics immediately upstream of the

plasma chamber include an energy spectrometer, an emittance measurement system, an

integrating current transformer (ICT) and Faraday cup to measure Q , a phosphor screen

and, an optical transition radiation (OTR) screen at the focal point of the -matching

solenoid to obtain transverse beam profile images. The bunch length was measured by

refocusing the beam onto the Cerenkov radiation diagnostic at the end of the plasma, as

described below.

Figure 36. Diagnostics beam line and plasma cell (shown without the plasma radial
confinement solenoid). The anode diagnostics include (a) tungsten collimator with (b) 1
mm wide slit, (c) 500 µm thick quartz Cerenkov plate, and (d) mirror and outgoing light.

The plasma density, which is in the region of n0 = 1.15 ×1013 cm -3 , is mapped

with a Langmuir probe, and found to be within 10% of the peak value over the nominal

plasma length, with a steep initial ramp near the cathode tip. The half-maximum point,

located 9 mm downstream of the cathode, is Lp ≅ 12 .25 cm away from the anode. It is

important to note that the arc current density in the beam region is four orders of magnitude

smaller than the beam current density, and thus does not significantly affect the beam's

transverse motion. The plasma density is typically measured at the start and finish of a 1

hour experimental run.

The bunch parameters Q, r, and at the beginning of the plasma cannot be

measured while the plasma is on, and thus are not simultaneously measured along with

the beam's final state under the influence of plasma focusing. These quantities are also

measured before and after the plasma run, to eliminate the possibility of a drift in accelerator

conditions. The plasma focusing experiments demand that Q be measured for every shot

using the upstream nondestructive ICT, at which point the charge is larger than that

propagating in the plasma chamber, due to a small amount of scraping by the cathode

assembly aperture. This scraping fraction is quantified under plasma-off conditions by

simultaneous measurements with the ICT and FC1, the Faraday cup following the cathode,

with the charge measurements performed at these two locations being well correlated.

The initial focal spot size near the waist of the -matching solenoid was measured

with plasma off at the OTR screen located at FC1. Some of these profiles deviated

significantly from the cylindrical symmetry assumed in the simulations, having an

asymmetric shape with an aspect ratio as high as 4:1. The 25% of the shots having an

asymmetry larger than 2:1 were rejected for the purpose of this analysis. The remaining

images tested for peak intensity as well as for fractional integrated intensity inside a test

radius of 0.28 mm. A symmetric Gaussian with x= y ≡ r, having the same integral

defines an effective radius, which was found to be nearly independent of Q , with mean

value of r =284 ±24 µm. The presence of mild asymmetries causes use of the effective

r to underestimate the peak beam density, predicting nb / n0 =2.0 at 14 nC (assuming

Gaussian, a 25 psec FWHM longitudinal bunch profile, as described below). The actual

value from the peak intensity in the image data is nb / n0 =2.0 and 2.5 at Q=7 and 14 nC,

respectively, and the plasma is well underdense at beam input for all experimental conditions.

The beam divergence x′ and bunch length were measured with plasma off using

the anode diagnostics, including a 1-mm wide (4 mm deep) tungsten slit aperture followed

by a 0.5 mm thick quartz plate inclined 12 degrees to the beam direction. A portion of the

radially polarized outgoing Cerenkov light from this diagnostic (transverse resolution is

80 µm) exits the plate at the Brewster angle to suppress internal reflections and is sent to

either a CCD camera for time-integrated imaging or to the 2 psec resolution streak

camera. The beam charge after collimation by the slit aperture is recorded on the Faraday

cup following the anode, FC2. For measuring x′, the beam is allowed to drift from the

initial waist to the anode. The divergence is calculated from the width of a Gaussian fit to

the profile along the slit, and the drift length from the initial waist (the error in the waist

location can be up to 6 mm, causing a 5% error in x ′ ). The beam divergence is 18.5

mrad at 14 nC and decreases linearly to 15.5 mrad at 7 nC, yielding n ≅ = 149 and

130 mm-mrad and initial i =1.52 and 1.74 cm, for high and low charge, respectively.

These emittance values also agree with upstream pepper-pot measurements. Note that

the input beam -function is slightly mismatched in these cases to the ion focusing

eq = 1.25 cm.

The input bunch length is measured by refocusing the beam on the anode, with the

Cerenkov light analyzed with a Hamamatsu C1587 streak camera, giving a pulse length

of 25± 3 psec FWHM. Interestingly, the bunch length was nearly constant over a wide

range in Q . Given the beam distribution, the measurement method used may have

resulted in a slight underestimate in the bunch length.

After the beam is collimated, the transmitted charge Qtrans is recorded on the

downstream Faraday cup FC2, and the value multiplied by a factor of 1.25 to correct for

the charge scraped by the anode wall after multiple scattering in the 130 µm thick Cu exit

foil and quartz plate. With plasma focusing, the fraction of the initial charge transmitted

through the slit, = Qtrans / Q, serves as an independent measure of the beam radius.

With the plasma on, the time-integrated Cerenkov profile shown in Fig. 38 displays

a narrow, intense peak as a result of ion focusing. The x -projected FWHM in this case is

0.9 mm, 40% broader than at the plasma start, but consistent with the beam head spreading

predicted by simulations. The projected peak intensity is ten times greater than in the no

plasma case, and is accompanied by a reduction in intensity at large x , away from the

peak. As the beam transverse centroid jitter is not small compared to the slit aperture, a

large shot-to-shot spread in due to increased scraping appears. To select only shots

centered on the slit, a shot is rejected if its y -peak (defined by the 0.8-maximum points)

is within 100 µm of one, but not both, apertures. This selection rejects images in which

the scraping is mainly on one side but does not reject a beam wide enough to scrape on

both sides. Values of selected in this way for an initially matched ( z0 = 0 ) case are

displayed in Fig. 3, along with the results of simulations. Varying the value of x′ in the

simulations within expected errors produces values 3% higher and 7% lower than

nominal; even so, the measured includes points falling far below this range, especially

for lower Q .

The spread in the data of Figure 39 is notably large, with events extending as low as

the no-plasma case. The low shots did not appear to be transversely wider than the rest,

as would happen if the lack of transmission were due to diminished beam focusing. The

data in these shots stayed well away from either collimator edge, having transverse

dimensions even smaller that the rest of the data for the shots with the lowest . The data

for these shots shared another aspect with the high-n shots, that of having less signal at

large x than in the plasma off case. This is an indication that the large initial beam

divergence sending particles at large x in the no plasma case is somehow countered by

the effect of the plasma. One explanation for this data is that the peak observed is not the

primary peak, but is the result of irregular initial conditions which allow a satellite beam

to break away from the main beam. Another possibility involves the focusing properties

of a transversely non-symmetric beam, a topic that is presently not fully understood and

requires further study using three-dimensional simulations. Finally, the AWA accelerator

has been observed to operate in a mode where electron emission from the gun persists for

several RF buckets, a phenomenon termed explosive mode emission [62]. While this is

ordinarily associated with much higher-Q operation, the existence of multiple pulses

would mean that one of these pulses can act as the main pulse with the present diagnostics.

None of the above possibilities offer a particularly satisfactory explanation for the observed

phenomena. In any case, it can be stated that a large ensures a vertically well slit-centered

distribution which is approximately symmetric.

The simulation results given in Fig. 39 use the plasma fluid code NOVO[32],

modified to include a super-particle representation of the beam electrons, as descirbed in

Chapter 3. Results of this code have been benchmarked against the electromagnetic

particle-in-cell (PIC) code for the conditions in this experiment; the calculation of is

nearly identical (smaller in NOVO by 4%), and so NOVO was used in this analysis

because of its speed. The calculations use 8000 beam particles, initialized to a thermal

distribution derived from the measured beam size and divergence.

A more detailed view of the underlying dynamics is offered by the time-resolved

streak camera images which, however, contain information only about longitudinal and

horizontal profiles, as measured in a 20 µm vertical strip centered in the slit aperture.

Adequate beam symmetry and vertical centering in the slit were insured in these

measurements by requiring ≥ 0.45. A streak image satisfying this test is shown in Fig.

4, a false color image of the distribution nb ( x, y = 0,t ) . The predicted flaring of the beam

head can be observed in this image. No streak image of the plasma-off case is shown,

since beam expansion causes the intensity to drop below useful levels.

Each image that was analyzed further was sliced into short (5.7 psec) t-slices; for

every t-slice the integrated (in x) intensity, as well as the beam width, defined as the

region containing half the integrated intensity (−x1/2 ≤ x ≤ x1/2 ), were determined. As

the streak images at this resolution are inherently noisy, the results of this analysis for 10

images within a narrow charge window (10.2 < Q <11.8 nC) were summed to produce a

composite picture of the beam distribution. This is shown in Fig. 40, along with the rms

error bars and PIC simulation results, analyzed to give beam width as a function of

t-slice. The experimental data and PIC simulations display a profile flared at the beam

head due to radial expansion, as well as a beam core which is nearly matched and slightly

larger, r ≅ 370 m , due to adiabatic anti-damping of the emittance, than at input. The

simulation's agreement with the data is quite good, with a notable deviation in that the

experimental results generally show less pronounced beam-head expansion, except for

one leading simulation point — this value of x1/2 is obtained from very few beam

simulation particles, and is therefore subject to unphysical statistical fluctuations. The

larger expansion in the simulated beam-head behavior may be due to the assumption of

an input beam with a thermal transverse phase space. This is not accurate for photoinjector

beams, where the emittance of a t-slice is smaller than that of the full beam, with much of

the total emittance arising from a correlation between t and orientation of transverse

phase space[63]. We note finally that the time-resolved measurements show no significant

evidence for onset of electron-hose instability.

The above data was taken with beams nearly matched to the focusing channel

strength. In order to explore dynamics of mismatched beams further, data runs were taken

with the input beam focused to a waist 2.5 cm upstream of the plasma column start, so

that it was strongly mismatched in both and d / dz . This was partly tried as a test for

system performance under less than optimal conditions. This case, with n0 unchanged,

displayed a smaller spread in than seen in Fig. 39, with values ranging from .27 to .59;

the average was =.42 +/-.01 (with plasma-off, =0.1), while simulation results were in

the range of =.45 to .52. Further, for these mismatched beams was observed for

several different values of n0 , a test which serves to distinguish the difference between

the case nb < n0 , where the focusing has negligibly weak dependence on n0 , and the case

nb > n0 , where the envelope oscillations proceed at twice the ion-channel betatron frequency,

k env = 2k ≡ 2/ eq ~ n0 . Several measures of the distribution in these runs are shown

as a function of n0 in Fig. 41: peak, 80th percentile, and mean. All of these quantities are

sensitive to n0 , with strong suppression of transmission occurring where focusing phase

advance k env Lp gives a local minimum beam size at the plasma exit. This sensitivity is a

signature that nb > n0 over significant lengths of the plasma.

In conclusion, short intense relativistic bunches derived from an rf photoinjector

have been observed, through integrated and time-dependent imaging, as well as collimator

transmission, to self-guide in an underdense plasma for lengths many times the initial .

Picosecond-resolution imaging of transverse beam distribution displays the equilibrium

body and trumpet-shaped head predicted by simulations and analysis. The simulations are

in quantitative agreement with the beam sizes and transmissions obtained for symmetric

beams in the experiment, with the observed deviations likely arising from the approximate

nature of the beam simulation model. The simulation results predict stable beam guiding

which continues over longer distances, as must be the case for their effective use as

underdense PWFA drivers. These results also point to a problem which must be addressed

in future PWFA work: stabilizing the transverse beam centroid and distribution shape.

This will be accomplished through advances in laser (e.g. spatial filtering) and photocathode

(e.g. high quantum efficiency, uniform emitter such as Cs 2Te) technology. Additionally,

for refined predictions of PWFA performance, further development of the beam-plasma

interaction computational model, including three-dimensional effects, would be useful.

Figure 37. Cerenkov streak image showing nb ( x,t) , with earlier times (beam head) on

plasma on
120 plasma off
Projected intensity




0.0 0.20 0.40 0.60 0.80 1.0
Transverse position (cm)

Figure 38. Transverse beam profile after collimation, from digitized CCD camera image
of Cerenkov light.

plasma on
0.7 simulation
0.6 plasma off






0 2 4 6 8 10 12 14
Qi (nC)

Figure 39. Transmission data and simulations for an initially nearly -matched case.

1.4 100
x1 / 2 exp. intensity exp.

Integrated intensity (arb. units)

1.2 x1 / 2 sim. Intensity sim.

0.8 60



0 0
0 10 20 30 40 50 60 70 80
Time (ps)

Figure 40. Time-slice dependence of beam intensity and half-widths, form experiment
(exp.) and PIC simulation (sim.).


0.2 th
80 percentile
0.1 mean
1.10 1.15 1.20 1.25 1.30 1.35 1.40
13 -3
n (10 cm )

Figure 41. Dependence of distribution measures (peak, 80 th percentile, and mean) as a

function of n0 for -mismatched case.

Chapter 7. Electron Acceleration Measurements in a
Nonlinear Plasma Wake-field

The goal of the experimental program at Argonne has been to demonstrate the

highest plasma wake-field that can be driven with this electron source, originally aiming

at a number exceeding 100 MeV/m. The ability to generate a drive/witness beam has

added a second priority, that of demonstrating effective acceleration and transport of the

witness beam in the nonlinear wake-field. This is usually trivial in the case microwave

cavities, with iris dimensions of order centimeters. In a plasma experiment, the witness

beam must be placed in a < 200 µm diameter and < 3mm long volume in order to

effectively trap and accelerate in the wake-field. At the time of this writing the demonstration

of high gradient is only partially complete. The second goal has seen the development of

the necessary techniques and hardware, but has not yet been attempted. This chapter will

detail the equipment and experimental strategy developed for both of these goals, and

present the work completed to date.

The first attempt at measuring the strength of the wake-field using the present

equipment was reported in [64], with the conclusion that the wake-field from a 16 nC

drive beam had a magnitude of at least 4.1 MeV/m. These data were difficult to interpret

because the measured energy increase was smaller than the overall energy spread of the

drive beam, and comparable to the energy jitter. The experience gained in the course of

this work suggested that the measurements would be facilitated by improvements in the

energy measurement system, better understanding and instrumentation of the drive beam

transverse dynamics (the topic of the previous chapter), improvements in drive bunch

characteristics, and a better understanding of drive beam fluctuations. The last point, the

stability of the drive beam, has required advances on two fronts: a search for an accelerator

operating point which affords greater stability and the development of beam measurement

techniques to at least indirectly monitor the beam conditions.

The first topic discussed is the development of the revised spectrometer system.

The latest data taken with this equipment is much more revealing than the earlier (1995)

experiment, so only the more recent results are discussed. This experiment was performed

without the benefit of a witness beam, but instead relied a measurement of the drive beam

self-modulated energy spread to infer the peak wake-field strength. The rest of the Chapter

presents a more detailed view of the future direction of this experiment, including the

planned witness beam propagation studies.

7.1 High-resolution, broadband spectrometer

After interacting with the plasma, the drive beam acquires a very large energy

spread, up to 100% in some simulated cases, while also expanding very quickly in size.

The large shot to shot fluctuations in the photoinjector make it desirable to capture as

much data as possible for every single shot. To this end, it is necessary to simultaneously

observe both the accelerated and decelerated electrons even at the highest projected beam

energy modulation. A second factor driving the spectrometer design is the beam expansion

after exiting the plasma column, which is comparable to, but can exceed the initial

angular spread required to match the beam into the plasma. For the focusing experiment

the initial beam convergence was 18 mrad, which may increase toward the end because

of imperfect matching and the fact that the beam is decelerating. If the size of this

expanding beam is not quickly brought under control, it would force the use of very wide

apertures in the spectrometer in order to avoid beam scraping (loss of signal) or scattering

off the walls (signal showing up at the wrong energy). Additionally, the spectrometer

design was to allow for sufficiently good resolution throughout the energy range, and the

ability to tolerate beam initial conditions that don’t necessarily correspond to those

prescribed by the simulations.

The final configuration of this system is depicted in Fig. 42, including the slit

collimator at the plasma anode, a defocusing in the bend plane quadrupole, followed by

the bend magnet and phosphor screen. The slit collimator is the same as was used in the

focusing experiment, but serves a dual purpose in the present context. Initial collimation

in the dispersive plane is a technique of improving the resolution in spectrometers of all

types. In this case, it eliminates the possibility of initial position jitter altering the measured

energy. The collimator also retains its original function as a diagnostic of the quality of

focusing and self-guiding of the drive beam, with an ICT to record the outgoing charge.

In should be noted that collimating the beam in this way amounts to selecting only a

subset of the particles to be included in the energy measurement, as opposed to measuring

the energy spectrum of the entire beam. The results obtained in this way are intended to

be compared to simulations in which the same collimation is done numerically. An

advantage of the approach is that the rejected particles mainly belong to the beam head,

where the change from the original energy is very small. This improves the signal to

noise ratio for the particles emerging at the highest energies, which may be relatively few

in number. The remainder of the magnetic components in the system are designed to

minimize the -function in the bend plane over a large energy range without having to

readjust magnet strengths. This means that a monoenergetic beam will be imaged point-

to-point from the slit to the phosphor.

The performance of this system was evaluated with a realistic electron distribution

using a FORTRAN program, importing the particle phase space from a beam-plasma

simulation as the input. The particle trajectories are then tracked in space, under the

influence of the magnetic fields as measured by a hall probe. Figure 43 shows a random

subset of these trajectories in relation to other objects in the system. The program calculates

the intersection of the particle trajectories and the screen area, recording all the data about

the particle at time of intersection. The sum of these records is displayed in Figure 44 to

reveal the energy resolution of the system. This plot shows a much better resolution at

high energies than at the lower end of the spectrum. Effective measurement of these

accelerated particles, which may be few in number under actual experimental conditions,

requires that they are focused in y at the screen, as shown in Figure 45 using the same


Figure 42. The spectrometer system, including the slit at the plasma anode, quadrupole,

bend magnet, ICT and phosphor screen.

Figure 43. A simulation of trajectories in the spectrometer.

Figure 44. The final position of a particle vs. . The resolution at high energy is designed
to be better than for lower energies.

Figure 45. A simulated view of the spectrometer screen showing a y -focus for high
energy particles.

7.2 Experimental results

Improvements to the AWA linac prior to these results include the conditioning of

the drive gun cavity to sustain a higher gradient and an enlargement of the laser spot on

the cathode. This resulted in an obvious reduction in the bunch length available from the

machine, but also a change in the longitudinal profile, with the bunches exhibiting a long

tail in the trailing end of the beam. These tails were long enough, in fact, to allow

self-modulation of the beam energy in the plasma to probe the maximum generated

acceleration gradient. This was confirmed on the first try (using still the old spectrometer),

suggesting that by observing this self-modulation effect, changes could be made to the

AWA parameters, the focusing solenoid and the plasma density, in order to optimize the

wake-field performance.

The experimental setup used for these runs is essentially a combination of the

focusing experiment beam line with the energy measurement system. The setup uses two

ICT’s intended for computing the fraction of the transmitted charge in real time, as was

done using the Faraday cup beam dump in the focusing experiment. In this way, the shots

for which the drive beam was insufficiently focused or at the wrong transverse displacement

relative to the slit could be rejected.

When directing the AWA drive beam into the plasma chamber, it was immediately

apparent that the presence of plasma had the effect of greatly increasing the total energy

spread of the beam. These wider spectra were characterized by a large electron population

at lower energies and sometimes a faint high energy tail. Not all of the shots displayed

these features, with the total amount of energy spread greatly varying from shot to shot.

The energy spread also tended to become wider with increased plasma density. It also

had a strong dependence on the gun launch phase.

When attempting to optimize the system for the highest generated wake-field, it

became apparent that changing the linac settings would be constrained by having to keep

the initial energy spread below some nominal value. Certain settings for the gun launch

phase also had the effect of causing excessive energy jitter. Both of these effects are

undesirable in the task of proving that the accelerated electrons actually had a much

lower energy at the start of the plasma. With this parameter search, it was not always

obvious in which direction to proceed because of the large shot-to-shot fluctuations. As

further gains seemed to become more difficult, a data set was recorded at a plasma

density in the region of 1.5 − 2 × 1013 and ICT2 registering as much as 15 nC. The initial

charge was, therefore, in the range of 25-30 nC. Since the shots with large energy spreads

were infrequent, occurring only about 25% of the time, only these shots’ spectrometer

images were saved. Fig. 46 shows two examples where it is apparent that even within this

set of ‘good’ shots there is a large amount of fluctuations. Fig. 46 a) shows a large

deceleration gradient with a high energy tail that seems to be at a different y -position

than the rest of the beam. This would be true if a plasma transverse deflection mode is

excited, and the beam tail is kicked sideways. The no-plasma background, shown in Fig.

46 c) shows that the initial energy spread is much smaller than the plasma wake-field

caused energy modulation.

Figure 47 displays a feature found in only one of the saved images, and not

discovered until the data analysis, and then only by changing the color map to enable

very faint features to stand out. This image shows a high energy tail in which the

transverse kick is small. It also extends to very high energy, or 2.6 MeV higher than the

mean. Over the 12 cm interaction length, this implies an average gradient of 22 MeV/m.

The energy gain is approximately twice as large as the maximum energy loss for this

shot. This ratio can be directly compared to the transformer ratio from 2-D simulations,

thus validating the prediction that the transformer ratio can be as high as 2.0 for roughly

symmetric in time bunches. It should also be stated that many conditions can lead to an

absence of particles at these energies, such as transverse deflections, wake-field dephasing,

or an absence of particles at the correct phase. This can be used to explain the high

incidence of the data resembling the images in Fig. 46 a) and b).

The scarcity of images showing the highest energy accelerated electrons may also

have implications for the relative difficulty of future experiments aimed at witness beam

charge preservation. In this case, the drive beam must create a stable accelerating bucket

while at the same time the witness beam is present at the correct location to be captured

into it. Such capture must happen in the absence of 2-D effects such as phase changes,

and any 3-D effects which may also change the shape of the accelerator bucket.




Figure 46. Images of the energy spectrom taken at spectrometer focal plane. Image c) is a
no-plasma background.

Figure 47. Spectrometer phosphor image with the gray scale adjusted to accentuate the
high energy tail. The initial energy is 13.8 MeV, with an rms energy spread of 140 KeV
and an energy jitter of 92 KeV.

7.3 Future experiments

The results quoted above leave some room for improvement in terms of what can

ultimately be expected from the present laboratory setup. Future results can benefit from

increased stability on the part of the electron beam as well as a more complete set of

measurements of the bunch initial conditions. These experiments will hopefully reveal an

even higher maximum acceleration gradient, and also address the question of what is the

best way to accomplish the witness beam charge preservation experiment. The plans for

both experiments are described below.

Higher gradient acceleration.

The results presented in the last section show the presence of a plasma wave

capable of accelerating electrons with a 22 MeV/m gradient. With more work, it may be

possible to improve this number, but certainly not by an order of magnitude, in the range

predicted by the earlier PARMELA simulations. Apart from seeking record-breaking

gradients, the results can greatly be improved by strengthening the connection between

the beam and plasma initial conditions and the observed energy modulation. To this end,

it will be very beneficial to achieve linac conditions having a greater amount of shot-to-shot


Perhaps the most difficult to account fluctuation, partly because it is not measured

on a shot-by-shot basis, is that of the bunch longitudinal profile. While it may not be

possible to achieve vastly better control over this parameter, it is possible to improve the

measurement of the average bunch length of a large number of shots. The focusing

experiment (Chapter 6) suggested that slice emittance considerations may play a role the

measurement being an underestimate. This could be tolerated for the focusing data because

of the weak dependence of the result on z . Slice emittance effects can be eliminated

from the bunch length measurement by first scattering the beam with a foil. The drift

length from the foil to the Cerenkov emitter will serve to wash out the slice emittance


Trailing beam capture

Two-dimensional simulations of the blowout regime suggest that a short, high

quality trailing beam matched into the plasma at the correct phase will become radially

confined and accelerated. An experiment that demonstrates this would represent a big

step in showing that such a plasma based device is capable of behaving like a true

accelerator. The most difficult aspect of such an experiment is that it depends on having

two electron beams, each of which must separately be focused onto the same point in

space but appropriately separated in time. Two possible ways of creating such a drive/trailing

beam pair include forming them in separate electrons sources, or creating them in the

same photocathode cavity. In the former case, the two beams will be well-distinguished

in energy, enabling a more accurate energy analysis, but the independent jitter from two

electron sources will make it harder to consistently focus both beams to the same spot. A

trailing beam created in the same cavity as the drive beam will experience very similar

conditions during propagation down the beamline. This hopefully means that, although

the drive beam will have some position jitter, the trailing beam will follow it. Disadvantages

of this approach include the inability to introduce steering corrections on the trailing

beam alone, and a greater difficulty in distinguishing the two beams in the downstream

energy measurements.

Each of the two methods of generating and focusing a witness beam into the

plasma chamber were tried in the laboratory for the purpose of determining their relative

merits. When the ~4.5 MeV witness beam encounters the matching solenoid, it can be

focused at a very low value for the magnet current. As the current is increased from that

point, the beam increases in size. At even higher settings, it decreases in size again. This

is the second focus, corresponding to a beam envelope that has a tight focus near the

center of the solenoid, and becomes large but converging near the end of the solenoid to

come to another focus inside the plasma chamber. The first focus was observed at a

relative solenoid setting of 370, while the second focus was at 2080, or slightly higher

than 200 Amps. When compared to a setting of 1800 needed to place the drive beam

waist at the same location, these results indicate that the witness beam energy is was too

high to achieve similar focal properties for both beams. The solution is to energy-tune the

solenoid focal length by decreasing the RF power to the witness gun. Throughout this

test, the focused beam was observed to fluctuate in position over distances comparable to

the spot diameter. This fact suggests that, given the additional (and independent) drive

beam fluctuations, the overlap between the two beams will certainly not be perfect for

every shot. This test was conducted using only the witness beam, without attempting to

also direct the drive beam into the plasma chamber. Co-propagating both beams has the

additional problem that it may not be possible to diagnose the focal spot of the ~500 pC

witness beam in the presence of a 25 nC drive beam. Thus, it may be very difficult to take

account of the drive beam space-charge forces acting on the witness beam, especially

near the combiner dipole (see Fig 20) where the two beams are displaced off-axis. At

very small delays, there forces are enough to steer the witness beam completely away, an

effect which will also vary as a function of the drive beam charge fluctuations. In the face

of these difficulties, it was decided not to pursue the use of the low energy witness beam

for the study of plasma wake-fields.

In order to adopt the method of generating a witness beam in the same cavity as

the drive beam, as covered in Chapter 5, several changes to the hardware were required.

The old coaxial mirrors scheme to generate the witness beam had the disadvantage that

the witness beam charge could fluctuate independently of the drive beam, as would

happen if there is a laser hot spot within the area directed toward the witness beam

mirror. As the two resulting electron beams propagate in the same beamline, there is no

way to measure the charge of the witness beam after creation. A new beam splitter in

which divides the laser beam according to a fixed energy ratio is shown in Figure 48. A

quartz optical wedge is mounted at 45 degrees to the beam direction and inclined half-way

between S and P polarization. This creates a reflection from the forward edge of

approximately 5% of the power; the reflection form the backward edge is easily discarded

because it is at the wrong angle. The main beam loses 20% of its energy in two passes

through the beam splitter. In order for the two laser beams to have a similar power

density at the cathode, this device uses a telescope to reduce the trailing laser beam spot

by a factor of three. The last trailing beam mirror is attached to a remotely actuatable

mount to allow some steering for the trailing electron beam. Since the point of zero delay

is very difficult to deduce from the geometry of this device, it was calibrated with the

streak camera. An added advantage of this device over the older technique is the ability to

place the second beam ahead of the primary beam. This feature can be used to make a

null test of the plasma conditions.

The first test of the new beam splitter to generate electrons produced 1 nC for the

trailing beam at a drive beam charge of 20 nC, verifying the design 5% power split.

Focusing both beams separately into the plasma chamber resulted in a witness beam with

a transverse spot smaller than the drive. However, its position on the OTR mirror used to

perform the spot size analysis was seen to change as a function of the delay, with the two

beams overlapping at zero delay. The most likely reason for this is the change in beam

energy as a function of gun cavity launch phase, combined with a slightly off-axis

trajectory in the solenoid. Better solenoid alignment, as well as use of the remote actuator

used to adjust the transverse displacement of the trailing beam on the cathode ought to

result in both beams focused onto the same spot.

Attempts to measure the bunch length of the trailing beam proved to be unreliable

due to a lack of signal. The bunch length seemed to be close to 20 psec FWHM, a large

amount of beam lengthening for a 1 nC beam. In order for the trailing beam to be

contained in an accelerating phase of the wave, its overall length must be shorter than

about 10 psec. The observed bunch lengthening may be caused by hot spots in the laser

beam, possibly resulting from aberrations caused by the telescope. Such aberrations can

result from the laser crossing the lenses off-axis and at a finite incidence angle. Although

this alignment has been done very carefully, it may not yet be perfect.

Figure 48. The laser splitter used to control the delay between the drive and witness

Except for the excessive bunch lengthening, it appear that the remainder of the

drive/trailing beam initial conditions are satisfactory in terms of being able to capture the

trailing beam into the plasma wave. Given that the remaining problems can be solved, the

next challenge will be to show radial confinement and acceleration for the trailing beam.

A reasonable question to ask a this point is, even if the trailing beam is at the

proper delay and its volume is small enough to effectively be guided by the plasma, how

can this fact be demonstrated in the laboratory? One problem with this may be that the

trailing beam’s spectrometer signal will blend with that of the drive beam’s tail. It is

difficult to understand what set of conditions are responsible for the creation of this tail,

which may approach the actual trailing beam in brighness on the spectrometer screen.

One possible solution for this is the use of slightly different initial conditions for the

trailing beam. Through careful control of the trailing beam emission center on the

photocathode (this is under remote control, by deflecting the laser), the trailing beam will

hopefully arrive at the plasma with some transverse momentum. Upon capture in the ion

channel, the bunch centroid orbit will oscillate about the drive beam axis and exit the

plasma with a different, but hopefully finite transverse component. The witness beam will

thus show up at a different place in the non-bend direction of the spectrometer screen;

this may happen quite naturally, even without a deliberate attempt to control its initial

conditions. After successfully differentiating the witness beam from the drive beam tail

electrons in this way, the surviving witness charge can be obtained by image processing

techniques. This measurement can be compared to the witness beam charge incident on

the plasma, a fraction of the ICT signal.

APPENDIX A. The Control and Data Taking System
Software capable of performing online data analysis can greatly facilitate an

experimental effort. Observing a phosphor screen beam image over time gives very little

information; plotting the beam width against the magnet setting tells a very different tale.

The same sentiment can be expressed about displaying histograms, temporal trends,

curve fits, performing image analysis, and applying a filter function to reject certain data.

This approach is particularly advantageous in the case of photoinjectors, because of the

large shot-to-shot fluctuations present in these systems, and the necessity, due to cost

considerations, of using machine run time wisely.

Because of the large number of analysis functions that can be employed, a danger

exists in adopting a software package with too much functionality. If a great deal of

programming is required to set up a new measurement or display, either it won't be

implemented, or the setup effort will compete for attention with the actual experiment.

As with any amount of programming, there is also the possibility of making an error. A

desirable feature of such a system, therefore, is the ability to configure as many of the

measurements as is practical through the graphical user interface (GUI). This approach

can never be made universal, and so a certain amount of programming is always necessary.

The best approach for this programming is to develop a well-structured subroutine library,

thus modifying the software only at the highest level.


Before an in-depth discussion of this software system, it is useful to examine the

major components making up the system hardware. This system has evolved through a

series of equipment updates, and also through some trial and error. At the heart of the

system is a Macintosh 7300 computer, closely coupled to the 8-bit video frame grabber

expansion board, model LG-3 from Scion Corp., through the internal PCI bus. A second

expansion card is used for communication to GPIB instruments, such as digital oscilloscopes

and the CAMAC chassis, which houses additional ADC and DAC expansion cards. From

the hardware point of view, all data for a given shot can be collected at a 30 Hz rate, with

the exception of oscilloscope measurements.

The LabView System

Although wishing to replace it at times, the LabView application form National

Instruments has been used throughout this project. The LabView package includes a

graphical programming language (a program is a set of pictures instead of an ASCII file),

an environment to define a user interface, and a data acquisition interface library. From

the computer language point of view, this approach has several advantages over writing

the entire control system in C, listed here for the benefit of anyone contemplating the use

of LabView for a similar application:

1) A large selection of GUI controls, including

2) charts and graphs.

3) Varable-length arrays (these are automatically subscripted in loops.)

4) Automatic syntax checking; automatically supplied documentation for

subroutine call parameters. This and other features facilitate debugging.

6) An existing set of libraries for data acquisition.

The disadvantages include the following:

1) LabView uses call-by-value. There are time penalties for passing large

amounts of data through the subroutine hierarchy.

2) Surprisingly, the same time penalties exist for accessing global data.

3) LabView is proprietary, unstandardized, and not fully documented.

4) Maintaining large projects is difficult.

5) Despite the best intensions of the programmer, the wiring diagram (this

contains the code) can become cluttered and difficult to understand. It is this aspect

of the language that prevents easily changing an existing code.

A number of the advantages align with the notion stated earlier, that when a new measurement

requires a programming change, the method for making this change must be straightforward

and leave little room for error.

Also used in this system is the IMAQ library (by National Instruments) for LabView,

containing a set of general purpose image processing routines, compatible with the Graftek

Inc. supplied routines for interfacing to the frame grabber card.

The Control and Data-taking Software

Taking a que from the existing large scale packages for accelerator control, such

as EPIX and Vsystem, it is an advantage if the software can stores machine settings and

measured values in the form of a database. Use of a database structure enables the use of

sophisticated search techniques, allowing the results to be displayed as the data is taken.

In fact, the data filters defined in the CERNLIB package, a set of off-line data analysis

tools for the high energy community represent one such type of search. In CERNLIB, the

user defines an N-tuple by supplying a filter subrouting.

Some of the ideas about how to structure the control system software came from

the documentation for EPIX, a control system package popular with large accelerator

applications, as well as a similar commercial system, Vsystem by Vista Control Systems,

Inc. These systems include a real-time database, to store and process the data streaming

in from the experiment. Further motivation came from the CERNLIB documentation, an

analysis package written for for high energy physics community. One capability within

CERNLIB is for the user to write a filter function acting on the data which can select a

subset of all the events processed, called an N-tuple. These N-tuples can then be handled

as individual entities, compared in plots, etc.

The desired functionality for the plasma experiment software was, then, a system

having a database structure similar to EPIX, but with the capability of selecting subsets

and sub-subsets of data, like in CERNLIB, with the distinction that as much of this

selection ought to be GUI-driven.

The sources of data in the system are the video frame grabber board and the

ADC's. The image data is first processed, and one or more features are extracted. The

reduced data, along with the ADC data is stored into the database; each piece of data, and

hence database column, has a unique string identifier. The next step is to compute

intermediate and final results from the existing columns in the database. A special type of

intermediate result is a column which accepts a 1 or 0 value, a 1 signifying that the shot

in question belongs to a particular subset (as in the N-tuples of CernLib). This prescription

for creating a subset relies on preparing custom code for every new case. A simpler, but

more general approach, is to define subsets based on a set of limits for any given column.

Figure 99 shows the VI for setting these limits by dragging the mouse while viewing a

chart displaying the history of the variable in question. These limits can also be set evenly

spaced over the entire range of the variable.

After the data and results are stored in the database, it can either be saved, along

with the raw image data, or plotted in real time. The plotting function is driven by a

command parser which accepts a command of the form:


where varx is the string identifier for some column, limx:y(varz) includes only shots

whith varz inside limit bands x through y (or only inside x if ":y" is omitted). Similarly, an

entry of the form varx=value excludes shots where varx is not equal to value. Any

number of such comma-separated conditions can be specified, all of which need to be

true in order for the shot to be included in the plot (or-tied).

The control system also features several VI's for the setting of magnet currents.

The VI’s for this function are hidden most of the time, except when the magnet settings

need to be changed. Data taking is disabled during the time that the current is being

ramped to the new setting, and then entered into the database for reference during normal


Although never intended to be a replacement for saving the raw data on disk for

off-line analysis, the control system has been very useful for both making routine

measurement and spotting new trends during the experimental runs. One of the routine

measurements is of the focused spot size while changing the solenoid strength. This

creates a plot that is useful in determine the solenoid strength which places the beam

waist at the OTR screen. The space-charge forces in the linac modify the waist position

somewhat, and so the limits VI can select an active bunch charge range for this measurement.

Experience has shown this method of locating the waist to be very repeatable.

[1] T. Srinivasan-Rao, and J. Smedley, Table top, pulsed, relativistic electron gun with
GV/m gradient, in Proceedings of the Seventh Workshop on Advanced Accelerator
Concepts, Lake Tahoe, California, S. Cattopadhyay, editor, AIP (1996).
[2] E. Dodd, J. K. Kim, and D. Umstadter, “Ultrashort-pulse Relativistic Electron
Gun/Accelerator”, in Advanced Accelerator Concepts, Seventh Workshop (proceedings),
Lake Tahoe, CA 1996, S. Chattopadhyay, ed.
[3] T. Tajima and J. M. Dawson, “Laser electron accelerator”, Phys. Rev. Lett. 43 267
[4] J. Simpson et al., Argonne National Laboratory report No. ANL-HEP-TR-89-81
(April, 1990).
[5] H. Braun, et al. “CLIC- A Compact and Efficient High Energy Linear Collider”, in
Proceedings of the 1995 IEEE Particle Accelerator Conference, p. 716.
[6] P. Chen, et al., “Acceleration of electrons by the interaction of a bunched electron
beam with a plasma”, Phys. Rev. Lett. 54 693 (1985).
[7] R. Keinigs and M. Jones, Physics of Fluids, 30, 252 (1987).
[8] J. B. Rosenzweig, et al. Plasma Wake Field Acceleration: A Proposed Experimental
Test, in Laser Acceleration of Particles (conference proceedings, Malibu), C. Joshi,
editor, AIP (1985).
[9] A. K. Berezin, et al., “Wake Field Excitation in Plasma by a Relativistic Electron
Pulse with a Controlled Number of Short Bunches,” Fizika Plasmy, 20 663 (1994), also,
Plasma Phys. Rep. 20 596 (1994).
[10] J. B. Rosenzweig, et al. Experimental Observation of Plasma Wake-Field Acceleration,
Physical Review Letters 61, 98 (1988).
[11] J. B. Rosenzweig et al. Experimental Measurement of Nonlinear Plasma Wake
Fields, Phys. Rev. A 39, 1586 (1989).
[12] K. Nakajima, et al. Plasma Wake-field Accelerator Experiment at KEK, NIM-A 292,
12, (1990).
[13] H. Nakanishi, et al. Wakefield Accelerator Using Twon Linacs, NIM-A 328, 596
[14] T. Katsouleas, et al. “A proposal for a 1 GeV Plasma-Wakefield Acceleration
Experiment at SLAC”, submitted to the Proceedings of the 1997 Particle Accelerator
Conference, Vancouver, Canada.
[15] S. C. Hartman, et al., “Initial Measurements of the UCLA RF Photoinjector,” NIM-A
340 219 (1994).
[16] J. Rosenzweig, et al. “Proposal for Staged Plasma Wake-field Accelerator Experiment
at the Fermilab Test Facility”, Fermilab proposal P890, 1996.

[17]B. E. Carlsten, et al. “Subpicosecond, Ultra-bright Electron Injector”, ‘95 PAC.
[18] K. V. Lotov, “Simulation of Ultrarelativistic Beam Dynamics in Plasma Wake-field
Accelerator”, in Phys. Fluids 5, Iss. 3, 785 (1998).
[19] J. B. Rosenzweig, B. Breizman, T. Katsouleas, and J. J. Su, “Acceleration and
focusing of electrons in two-dimensional nonlinear plasma wake fields”, Phys. Rev. A
44, 6189 (1991).
[20] W. E. Martin, G. J. Caporaso, W. M Fawley, D. Prosintz, and A. G. Cole, Phys. Ref.
Lett. 54, 685 (1985).
[21] K. Takayama and S. Hiramatsu, Phys. Rev. A 37, 173 (1988).
[22] W. K. H. Panofsky and W. A. Wenzel, “Some considerations concerning the transverse
deflection of charged particles in radio-frequency fields”, Rev. Sci. Instr. 27 967 (1956).
[23] T. Katsouleas, S. Wilks, P. Chen, J. M. Dawson, and J. J. Su, “Beam Loading in
Plasma Accelerators”, Particle Accelerators 22, 81 (1987).
[24] D. H. Whittum, et al., “Electron-hose instbility in the ion-focused regime”, Phys.
Rev. Lett. 67, 991 (1991).
[25] J. Krall and G. Joyce, “Transverse equilibrium and stability of the primary beam in
the plasma wake-field accelerator”, Phys. Plasmas 2 1326 (1995).
[26] E. Esarey, P. Sprangle, J. Krall, A. Ting, Review of plasma-based accelerator concepts,
IEEE Transactions on Plasma Science, 24 (2) 252 (1996).
[27] B. W. Montague and W. Schnell, “Multiple scattering and synchrotron radiation in
the plasma beat-wave accelerator”, in Laser Acceleration of Particles (conference
proceedings, Malibu), C. Joshi, editor, AIP 1985.
[28] K. Yokoya, Quantum Correction to Beamstrahlung due to the Finite Number of
Photons, NIM A 251 1 (1986).
[29] J. D. Jackson, Classical Electrodynamics, Second Edition, p. 643.
[30] H. L. Buchanan, Electron Beam Propagation in the Ion-focusied Regime, Phys.
Fluids 30 (1) 221 (1987).
[31] T. Katsouleas, et al., “A Proposal for a 1 GeV Plasma-Wakefield Acceleration
Experiment at SLAC”, submitted to Proceedings of the 1997 IEEE Particle Accelerator
[32] B. N. Breizman, T. Tajima, D. L. Fisher and P. Z. Chebotaev, “Excitation of nonlinear
wake field in a plasma for particle acceleration”, in Coherent radiation generation and
particle acceleration, A Prokhorov, editor, AIP Press, New York, p. 263 (1992).
[33] C. K. Birdsall and A. B. Langdon, Plasma physics via computer simulation, McGraw-
Hill 1985.
[34] W. H. Press, et al., Numerical Recipes in Fortran, Cambridge Univ. Press, 1992.

[35] W. Mori, private communication.
[36] J. Krall, K. Nguyen, and G. Joyce, Numerical Simulations of Axisymmetric Erosion
Processes in Ion-Focused Regime-Transported Beams, Phys. Fluids B 1, 2099 (1989).
[37] J. B. Rosenzweig, et al., Demonstration of electron beam self-focusing in plasma
wake fields, Phys. Fluids B, 2 1376 (1990).
[38] N. Barov and J. B. Rosenzweig, “Propagation of Short Electron Pulses in Underdense
Plasmas”, Phys. Rev. E, 49 4407 (1994).
[39] W. H. Bennett, Self-Focusing Streams, Physical Review 98, 1584 (1955).
[40] J. B. Rosenzweig, et al., Demonstration of electron beam self-focusing in plasma
wake fields, Phys. Fluids B 2 1376 (1990).
[41] Handbook of Mathematical Functions, edited by M. Abramowitz and I. Stegun
(Dover, New York, 1972), P. 721.
[42] W. Gai, Private communication.
[43] J. Rosenzweig, N. Barov, E. Colby, P. Colestock, “A Linear Collider Based on
Nonlinear Plasma Wake-field Acceleration”, in proceedings of Snowmass ‘96.
[44] P. Schoessow, et al., The Argonne Wakefield Accelerator, in Proceedings of the 2nd
European Particle Accelerator Conference, Nice, June 1990.
[45] L. Serafini and C. Pagani, Proc. 1st ePAC, Rome, June 1988, Ed. World Scientific, p.
[46] E. Colby , et al.,”Design and Construction of High Brightness RF Photoinjectors for
TESLA”, in Proceedings of the 1995 Particle Accelerator Conference, vol. 2, p. 967,
[47] E. Colby, Private comminication.
[48] C.-H. Ho, A High Current, Short Pulse Electron Source for Wakefield accelerators,
Ph.D. Thesis, Univ. California Los Angeles, 1992.
[49] K.-J. Kim, “RF and space-charge effects in laser-driven RF electron guns”, NIM A
275 201 (1989).
[50] B. E. Carlsten, et al., NIM A 285, 313 (1989).
[51] E. Colby, et al. Design and Construction of High Brightness RF Photoinjectors for
TESLA, in Proceedings of the 1995 Particle Accelerator Conference, vol. 2, p. 967,
[52] Los Alamos National Laboratory Report LA-UR-96-1834, 1997.
[53] W. Leemans, “Electron beam Monitoring at High Frequencies and Ultra-fast Time
Scales”, in Advanced Accelerator Concepts, Seventh Workshop, 1996, page 23, S.
Chattopadhyay, ed.

[54] K. B. Unser, “Design and Preliminary Test of a Beam Intensity Monitor ofr LEP”, in
Proceedings of the 1989 IEEE Particled Accelerator Conference, Vol. 1, page 71.
[55] Y. Yamazaki, T. Kurihara, H. Kobayashi, A. Asami, NIM A, 322, 139 (1992).
[56] J. B. Rosenzweig and G. Travish, “Design Considerations for the UCLA PBPL
Slit-based Phase Space Measurement Systems”, PBPL internal note, Feb. 21, 1994.
[57] M. E. Conde, et al. “Generation and Acceleration of Hich Charge Short Electron
Bunches”, submitted to NIM.
[58] Y. H. Chin, CERN LEP-TH/88-3 1988.
[59]H. Nakanishi, et al,. Phys. Rev. Lett. 66, 177 (1991).
[60] G. Hairapetian, et al,. Phys. Rev. Lett. 72, 1244 (1994).
[61] J.-L. Delcroix, and A. R. Trindade, Hollow cathode arcs, in Advances in Electronics
and Electron Physics, Vol. 35, L. Marton, editor, Academic Press (1974).
[62] P. Schoessow et al., The Argonne Wakefield Accelerator High Current Photocathode
Gun and Drive Linac, .in Proceedings of the 1995 Particle Accelerator Conference,
[63] X. Qiu et al., Physical Review Letters 76, 3723 (1996).
[64] N. Barov, et al. Measurements of plasma wake-fields in the blow-out regime, in
Proceedings of the 1995 Particle Accelerator Conference, vol. 1, p. 631, IEEE.