You are on page 1of 123

Transverse Beam Break-Up in Linear

Electron Accelerators
(undergraduate thesis)

by

Gil A. Travish *

Collider Physics Group


Accelerator and Fusion Research Division
Lawrence Berkeley Lab
University of California
Berkeley, California 94720

January, 1990

*Work supported by the Office of Energy Research, U.S. Dept of


Energy, under Contract No. DE-AC03-76SF00098
Acknowledgements

Dr. Andrew Sessler served as a teacher, motivator and oracle. To him

I owe much more than just the opportunity to write this thesis. His

guidance will serve me well beyond this work. I learned from him

what “being a Physicist” means.

Professor David Judd was invaluable as my campus thesis advisor. He

allowed me total freedom in choosing the content and form of my

thesis. He also proofread and commented on various drafts of this

work. Any “quality” found in this thesis is due in large part to his

diligent efforts.

I have also had the pleasure of receiving many explanations from Don

Hopkins. Without him my understanding of accelerators would have

been greatly diminished and largely incorrect. He spent many hours

of his time explaining to me nearly every topic found in Chapter 2 as

well as many others.

David Whittum taught me the analytic methods I needed to complete

this thesis. He also taught me how to write scientific papers (although

any of my stylistic shortcomings are no fault of his). He provided many

calculations which served as the basis for Chapters 3, 4 and 5. He

read early versions of this manuscript and made many helpful

comments. He also assisted in the development of the computer code

BBUNS. His knowledge is only surpassed by his kindness.

i
I also wish to acknowledge Jarvis Leung and Simi Turek for their help

in proofreading this text and listening to my convoluted explanations.

To be sure, there are many others who assisted me. I thank them

deeply and apologize for not including their names. Finally, I would

like to dedicate this to my parents who made all of this, including

myself, possible.

ii
Chapter 1
Introduction

Some historical background on accelerators


and beam break-up is given first. Then, beam
break-up is described qualitatively.
Throughout, an overview of the thesis is given.

1.1 Organizational Overview.

The organization of this work is outlined in the table of contents. This

chapter will cover the qualitative descriptions of accelerators and the

problem of beam break-up (BBU). Assumptions to be used throughout

this work will be clearly outlined in Section 1.5. A synopsis of the

analytic and numerical work is also given.

The second chapter will describe the components of an accelerator

which are relevant to beam break-up. The third chapter, along with

the first two Appendices, gives a detailed description of the beam

break-up problem. Included is a derivation of the BBU equation and

the definition of a wakefield. The notation to be used for the

remainder of the paper is also introduced. The fourth chapter

1
presents a number of analytic solutions to the equation derived in

chapter three. The fifth chapter discuses methods to reduce BBU.

The sixth chapter deals with numerical studies: two cases are

considered. The final chapter is a summary.

The various chapters and sub-sections are numbered in ascending

order. All equations are numbered using the following convention: c.n,

where c is the chapter number and n is the equation number within

that chapter. Equations which are referenced later in the work have

bold numbers. Important equations are emphasized with a * in the

left margin. Terminology is given in bold when first used. As usual,

acronyms are first given in parenthesis after the full phrases. For
typographical purposes the symbol “ι” (a Greek iota) is used to denote

the imaginary “i”. Also, vectors are written in bold letters, while

scalars are in plain type. Finally, full titles are included for most

references in the hope that readers will find this useful when

searching for literature.

1.2 Some Historical Background on Accelerators.

In 1928, Wideröe described the first successful accelerator. And, ever

since then there have been problems! Wideröe based his design

loosely on the ideas put forth four years earlier by Ising.1 The machine

was a linear ion accelerator. The design relied on alternating voltage

applied to sections of tube of ever increasing length. 2 The potential

differences in the gaps (between adjacent tubes) could accelerate the

2
ions. The “trick” was to have the frequency of the applied voltage,

and the lengths of the tubes, just right so that the ions would not be

de-accelerated. A schematic of Wideröe’s design is given in Figure 1.1.

Ions

Figure 1.1. Wideröe’s accelerator was designed to


accelerate ions. A potential difference in the gaps between
the drift tubes was used to accelerate the particles. The
entire structure was under a vacuum.

Wideröe’s design motivated Sloan and Lawrence to invent and build

the cyclotron. There device, along with advanced versions of Van de

Graaff and Cockcroft-Walton accelerators, launched the fields of

accelerator and experimental high energy physics.3 Since the late

twenties, many schemes have been devised to accelerate just about any

kind of charged particles. Circular machines (the betatron, developed

by Kerst in the forties, and the cyclotron) quickly provided the

desired high energy particle sources. Linear accelerators took a back

seat to the more economical circular machines until the technical

advances of the period during, and immediately following, World War

II. These achievements provided the necessary technology to make

linear accelerators feasible and economical. Accelerator scientists

3
were fortunate that the requirements for radars overlapped those for

accelerators. The post-war period also saw the advancement of the

klystron (in part by the Varian brothers): an ideal power source for

linear accelerators.

The accelerator field quickly separated into three categories: electron,

ion and proton machines. The electron and ion machines were linear

designs by necessity4,5 while the proton machines were both circular

and linear.

Hansen’s group at Stanford lead the way in the development of linear

accelerators (linacs) for electrons. By 1947 they had their first

machine working, and by 1955 Stanford had a 40 MeV prototype for a

1 GeV machine. Other groups around the world followed with various

designs.

1.3 A Very Brief History of Beam Break-Up.

Beam break-up (also called beam blow-up) was probably first observed

in the mid-fifties. However, it was not until the mid-sixties that the

phenomenon was understood. In fact, one might say that before that

time, BBU was not recognized as a serious problem, in part because of

the relatively low beam currents used.6 The first observations which

were actually identified as BBU were of beam current reductions.

During the testing and development of accelerators, it was observed

that if the beam was injected with a high enough current, the beam

4
would begin to lose current as it travelled down the accelerator. Later

it was observed that the beam’s transverse dimensions grew.

The type of beam break-up observed in these early experiments is now

called regenerative BBU. The beam is disrupted by fields within a

single section (cavity) of the accelerator (see Figure 1.2). As the name

implies, the fields grow exponentially through feedback (i.e. the fields

“regenerate”).

Regenerative BBU

Amplification

Feedback

Figure 1.2. Regenerative BBU is caused by the


amplification of fields within a section where a feedback
mechanism exists.

As beam currents became higher and accelerators longer, another type

of beam break-up was observed. The more sections that were added

to the accelerator, the worse the problem became. This “new”

problem was termed cumulative BBU. It began to harshly limit the

performance of various machines. Unlike other more subtle problems

which limited beam quality, BBU put a definite limit on how high beam

current could be and drastically reduced the beam quality.

Multisection BBU, as cumulative BBU is sometimes called, results from

progressively larger fields being excited in subsequent sections of the

5
accelerator (see Figure 1.3). In other words, cumulative BBU is,

logically enough, an effect which becomes more pronounced as the

beam goes further down the accelerator: the instability accumulates

along the accelerator.

External RF

Beam Fields

Figure 1.3. Cumulative BBU is causes by larger and larger


fields being excited in subsequent cavities. Here field
strength is represented by shading.

The growing fields act on the beam, causing a section of the beam to

be displaced further from its initial position. In turn, the further off-

axis a beam section is, the higher the field it will excite. The coupling

between the beam and the fields excited in the accelerator leads to

the BBU instability (see Figure 1.4).

Many “fixes and cures” were devised to limit the problems caused by

BBU; however, BBU remained a serious problem, and remains one in

many contemporary accelerators. The problem has not been

eliminated and BBU has to be taken into account in all future designs.7

6
Cavity Cavity Cavity

Cut-away Cut-away Cut-away

Figure 1.4. The growth of the transverse dimensions of


the beam due to BBU is illustrated by a short “pulse”. The
beam is injected slightly off axis (left). As it travels
through successive cavities, the displacement grows.
Eventually, the pulse becomes large enough to hit the
beam pipe and it will “break-up”.

Thus, the problem is this: a beam traversing an accelerator is subject

to an instability, called beam break-up, which must be controlled if the

performance of the accelerator is to be realized. This thesis is

dedicated to analytically defining and understanding this problem.

1.4 BBU Imposed Limitations.

As was mentioned in Section 1.3, BBU places a limit on beam current,

pulse length and injection energy. These limitations are problems for

the accelerator scientist. Trouble occurs if a part of a beam “scrapes”

the pipe wall. Namely, current will be lost, the beam will break up or

become shorter--pulse shortening (see Figure 1.5). Beam break-up

can be sufficiently severe that beam propagation is impossible because

the entire beam is lost (from hitting the walls).

7
Current
z=0
0
z=0.5zmax
0
z=z max
0

Time

Figure 1.5. Beam current measurements showing pulse


shortening due to BBU at three points along the
accelerator. This is a drawing of a typical oscilloscope
signal.8,9

Ultimately, though, it is the end product, the accelerated particles,

that is of concern. Whereas the procedure used to accelerate particles

may be of little interest to an accelerator user, the beam’s final

characteristics are of great importance. It is at this level that BBU

must be taken into account. It is not sufficient to accelerate a beam

from one end of a machine to another. The quality of the beam must

conform to the needs of the users.

To understand the effect BBU has on beam quality, we need to quantify

the concept of beam quality. There are several “figures of merit” for

beams depending on the application. Fortunately, two related

parameters are sufficient for understanding the effects of BBU.10

Luminosity of two colliding beams is defined as the interaction rate

per unit cross section. Alternatively, the luminosity can be thought of

as the density of phase space at the two beams’ interaction point. In

8
terms of the repetition rate, ƒ, the number of particles, N the

luminosity may be expressed as


2
L=ƒ N
4πσxσy (1.1)

where σx and σy are the transverse beam’s dimensions.11 The higher

the luminosity, the more interactions. (Luminosities are often given in

cm -2sec -1 with a typical figure around 10 31-10 32 for electron

colliders.) Unfortunately, the larger the beam’s cross sectional area,

the lower the luminosity. Thus, beam break-up lowers the luminosity

by increasing the transverse dimensions of the beam.

Often it is the area of phase space that is used as a measure of beam

quality. The emittance of a beam is related to the phase space area

(see Figure 1.6). A gross measure of the area is obtained using the rms

normalized emittance, εx =
rms x2 p x2 - xp x 2 (with a corresponding

equation for y).

The higher the emittance, the lower the luminosity ( L~ε-1/2). BBU

acts to increase a beam’s transverse emittance. For colliders, this

reduces the useful information that can be obtained from the beams.

An increased phase space area may render other devices inoperable.

In the case of free electron lasers, the gain is reduced. It is for these

reasons that BBU must be controlled to a greater extent than would be

acceptable for simply propagating a beam through an accelerator.

9
px px

x x

ε is sum of εrms is area


areas. "around"
(a) (b)

Figure 1.6. The emittance can be thought of as the sum of


the phase space areas as shown in (a). Alternatively, and
more usefully, it can be defined as the area enclosed by a
boundary as in (b).

It is the growth of the size of the beam from the start to the end of the

accelerator that is the critical measure of beam quality degradation

due to BBU. A quantity useful for comparing the growth under various

parameters is the distance over which the beam size grows by one e-

fold. This length is called the growth length.

1.5 The Assumptions.

In chapter 3 an equation (“the BBU equation”) is derived which

describes a beam’s transverse growth due to BBU. It is done so under

a number of assumptions which allow for a tractable derivation. These

assumptions will be outlined here, will apply to the remainder of the

paper, and will serve to define the scope of this work.

10
This thesis will only concern itself with cumulative, transverse beam

break-up in linear, electron accelerators. Regenerative effects are

ignored; longitudinal motion due to such effects as synchrotron

motion and longitudinal BBU are also ignored. Some of the analysis

does carry over to other devices, such as proton linacs, but this will

not be considered. In addition, many of the calculations assume a

relativistic beam, and do not apply at low energies. The only source of

BBU which will be discussed is wakefields. In addition, the entire

theory is linear in the beam’s transverse displacement: higher order

effects are ignored (or assumed small). In fact, the radius of the beam

is neglected: only the transverse motion of the beam’s centroid is

analyzed. Several of the calculations assume that the beam parameters

(i.e. energy) vary slowly over a betatron wavelength (see Chapter 2 for a

definition). In turn, it is often assumed that the betatron wavelength


is less than a growth length (i.e. λ β<L g): this is the strong focusing

regime. Finally, the acceleration cells along the accelerator are

assumed to be electromagnetically uncoupled: crosstalk is ignored.

1.6 Analytic Work.

The beam break-up equation is given by


s
d ξ dlnγ dξ
2
+ k 2β ξ = - 1 ⌠ I(s′)
ξ(s′,z) ds′
* dz 2
+
dz dz Dγ

W⊥(s-s′)
IA
0 , (1.1)

where the terms and notation are explained in chapter 3.

11
Even under the approximations used to obtain the BBU equation, it is

still not possible to solve the equation in full generality. Despite this,

several special cases and approximations lead to simple, useful models.

In Chapter 4, the following cases are considered:

1) No wake
2) Constant wake
3) Linear wake
4) General wake for a coasting beam
5) The two-particle model
6) General wake including acceleration

These solutions permit us to better understand how BBU affects the

beam. This knowledge allows us to find ways to counteract BBU.

1.7 Fixes and Cures.

Only one method to counter BBU is explored in this work: the BNS

effect.12 The BNS effect is caused by a spread in energy along the

entire beam. It is investigated in Chapter 5.

To be sure, there are many other remedies for BBU: cavity design

changes, RF phase adjustments, stronger focusing elements, etc..

However, these will not be considered here.13 In other words, we will

only examine how certain changes to the beam can reduce BBU, rather

than changes to the accelerator.

12
References and Notes
(Chapter 1)

1
P. M. Lapostolle and A. L. Septier, ed., Linear Accelerators (Amsterdam: North Holland
Publishing Company, 1970), pp. 1-18.
2
The length of the tubes had to increase along the accelerator to compensate for the increased
speed of the ions because all the tubes were connected to the same AC source (i.e. a fixed
frequency).
3
The Van de Graaff accelerator is just an electrostatic device where particles are moved
mechanically against a potential difference. The Cockcroft-Walton accelerator uses a series of
capacitors, charged in parallel and discharged in series to produce a large potential difference
(up to 1MV) which accelerates particles.
4
Circular designs could only be used to accelerate electron to tens or hundreds of MeV. Beyond
that energy range, radiative losses begin to dominate for electrons (radiation power α γ 4) and
linear machines become much more efficient. Modern circular electron machines can achieve
~100 GeV.
5
Historically, the magnetic fields needed to bend and focus the heavy-ions in a circular orbit
made high energy heavy-ion circular machines impractical. Advancements in magnet designs
have raised some new possibilities for circular electron machines.
6
See J. Haimson, “High Current Traveling Wave Electron Linear Accelerators,” IEEE
Transactions on Nuclear Science NS-12 3 (1965) for a discussion of early BBU observations.
7
Even a cursory review of the literature will reveal several papers which discuss the
limitation imposed by BBU. For a consideration of BBU in a “next generation” linac design see
F. Selph and A. Sessler, “Transverse Wakefield Effects in the Two-Beam Accelerator,”
Lawrence Berkeley Lab Preprint LBL-20083 (1985).
8
For actual photographs see Fig. 1 of R. H. Helm and G. A. Loew, “Beam Breakup” in Linear
Accelerators, ed. by M. Lapostolle and A. L. Septier (Amsterdam: North Holland Publishing
Company, 1970), p. 175.
9
Current reduction and pulse shortening are shown in Fig. 4 of R. J. Briggs, et a l . , “Beam
Dynamics in the ETA and ATA 10 kA Linear Induction Accelerators: Observations and Issues,”
IEEE Transactions on Nuclear Science 28, 3 (1981).
10
P. B. Wilson, “Linear Accelerators for TeV Colliders,” in Laser Acceleration of Particles , AIP
Conference Proceedings No. 130 (1985).
11
This expression for luminosity assumes identical beams with gaussian profiles (i.e. ρ~exp[-
x2/(2σx2)] ).
12
BNS is an acronym for V. Balakin, A. Novokhatsky, and V. Smirnov three Soviet scientist
who proposed a damping scheme for VLEPP. See V.E. Balakin, e t a l . , “VLEPP: Transverse
Beam Dynamics,” Proceedings of the 12th International Conference on High-Energy
Accelerators, ed. by F.T. Cole and R. Donaldson (Batavia, Illinois: Fermi National Accelerator
Laboratory, 1984), pp. 119-120.
13
For a treatment of other BBU remedies in multibunch linacs see K. A. Thompson and R. D.
Ruth, “Controlling Transverse Multibunch Instabilities in Linacs of High Energy Linear
Colliders,” Stanford Linear Accelerator Center Preprint SLAC-PUB-4801 (1989).

13
Chapter 2
Linear Accelerators

The various components which make up a


linear accelerator are described. The uses and
effects of the devices are discussed, and a
model for an accelerator is proposed. Figures
and photographs of components are given at
the end of the chapter.

2.1 The Basic Components.

The purpose of an accelerator is, as its name implies, to bring matter

to high velocity. Faced with the problem of accelerating charged

particles to very high velocities, one might be tempted to use a device

analogous to a sling-shot. Indeed one could set up a potential

difference between two points and accelerate a particle in this

manner. Problems arise and the complexities begin when one is

required to do this for many particles, and very high energies, with

great consistency and precise control over all parameters. In fact, the

very first problem is obtaining the particles to be accelerated.

1
2.1.a Injectors.

For an electron linear accelerator, there is little problem in obtaining

electrons. Various electron guns have been designed that produce

streams of electrons. Most guns work on the same principle as the

electron gun in television picture tubes.1 A cathode is heated to

induce electrons to leave its surface. The electrons accelerate towards

the anode. Once the electrons “overshoot” the anode, they are

focused, accelerated to some injection energy, and sent to the

accelerator. This entire assembly is referred to as the injector.

Injectors can be much more sophisticated than described above (see

Figure 2.11 at the end of this chapter). Nevertheless, two basic

requirements are common to all injectors: 1) particle production and

2) particle “pre-acceleration”. If the particles to be accelerated are

not readily available, then a scheme must be devised to generate the

particles. Assuming this is done, the particles must still be pre-

accelerated to the injection energy. The purpose for this is quite

simple: acceleration techniques work most efficiently in certain

specific energy ranges. The injection energy is picked to be a low one.

Methods, different from those used in the main accelerator, which are

more efficient for lower energies can be used to pre-accelerate the

particles (see Table 2.1). Also, different current regimes require

different guns and pre-acceleration techniques: the requirements on

a gun producing ~1 Amp (peak current) are very different from one

that produces ~10 kA.

2
Table 2.1. A comparison of four different accelerators;
their injection energies, final energies and peak currents.
Note the wide variation, and hence the radically different
injector designs.

Name SLC (SLAC) SSC ATA (LLNL) ETAII (LLNL)


Particles Used e+ / e- p e- e-
Injection Energy 1.21 GeV 1 TeV 2.5 MeV 2.5 MeV
Final Energy 50 GeV 20 TeV 50 MeV 6 MeV
Peak Current 1.2 kA 70 mA 10 kA 1.5 kA

In linear accelerators, injectors also play a critical role in beam quality.

Unlike circular machines, where beam quality can be improved in

storage rings, linacs are limited by the overall quality of the injected

beam. Thus, an improvement in the injector will most probably

improve the final beam quality.

2.1.b Cavities.

During World War II many methods were devised to produce high

power at microwave frequencies (~ 3-30 GHz). 2 These methods were

soon put to use in linear accelerators. At first, the availability of

microwave power sources from surplus military radars was exploited.

Later, custom power sources were built. It was well known at the time

that a conducting cavity could be “excited” by externally supplied

power.3 In fact, the idea of using cavities as resonators to accelerate

particles had its origins before World War II. The idea was simple:

the ability to excite large fields with modest amounts of RF4 power

could be exploited to accelerate particles.

3
Cavities are often made of very good conductors, copper being the

most commonly used.5 The acceleration cavity is constructed with a

minimum of three considerations:

1) The beam must be able to enter and exit.


2) There must be an input for externally supplied RF.
3) Appropriate fields have to excited so as to accelerate the
beam.

A cavity is often represented by a pillbox. Cavities are often designed

to be cylindrical since this geometry facilitates engineering,

construction and analytic work. Figure 2.1 is a representation of a

cavity with the first two requirements identified. Also, Figure 2.12 at

the end of the chapter is an artists conception of a special cavity

design.

RF Feed

RF
Cavity Port

Beam Line

Beam
Port

Figure 2.1. A pillbox model of a cavity showing the beam


and RF ports as well as a section of beam pipe to be
attached at one beam port.

The third consideration requires that there be an axial electric field.

There are many (even infinitely many) modes which satisfy this.

However, it is usually desirable not to excite certain modes, such as

BBU modes with transverse electric fields. In addition, the lower

4
modes often are easier to excite and provide greater acceleration.

This topic is an entire field in itself. For the purposes of studying

BBU, it is sufficient to be aware of the simple model for a cavity and

that fields can be and are excited in them. Cumulative beam break-up

concerns itself not with the fields excited by the external RF, but

rather with the fields induced by the beam. This will be analyzed in

the following chapters. However, there are additional structures to be

considered before we continue.

The single cavity is usually not sufficient to accelerate the beam to the

desired energy. There is a finite limit on the strength of the fields in

a cavity.6 And, there is an energy range where a particular cavity is

most efficient. Thus, several cavities are often used (see Figure 2.2).

Power Power Power Power


supply supply supply supply

RF Feed RF Feed RF Feed RF Feed


Cavity Cavity Cavity Cavity
Port Port Port Port

Beam Line

Cut-away

Figure 2.2. A series of cavities is often required to


accelerate a beam to the desired energy. Each cavity will
have its own RF feed.

From a manufacturing standpoint, if one cavity can be made, several

cavities can be made for a much lower cost per unit. However,

“stringing” cavities together requires sophisticated power supplies

and switching circuits. The RF power must be supplied at the

5
appropriate moment and at the correct phase (see Section 2.2). From

a beam break-up standpoint, a string of cavities complicates matters

because fields from a cavity can excite adjacent cavities: this

phenomenon is termed cross-talk. However, as stated in the

introduction, we will assume that cross-talk is negligible. In other

words, the fields are cut off in the beam pipe which adjoins cavities.

To alleviate some of the difficulties with multiple cavity structures,

another structure was devised: the disk-loaded structure. It is

presented here simply for completeness (see Figure 2.3).

Instead of a series of cavities, the disk-loaded structure has a series of

disks with irises in them. Adjacent disks form a cavity-like enclosure.

In this way, RF input can be fed in from one end and extracted out the

other. Disk-loaded structures can be designed in standing or traveling

wave configurations. Calculating beam break-up in such a structure

requires either considering the entire structure to be one unit, or

analyzing the cross-talk.7

6
RF Input

Iris

Figure 2.3. A cut-away section of a disk-loaded structure.

Cavities can be used for a number of other purposes. Cavities are used

to bunch or de-bunch a beam8: parts of the beam are made to

aggregate. Another primary use is in klystrons. Klystrons are devices

which are commonly used to produce RF in a process which is

opposite to that in an accelerator: energy is extracted from an

electron beam and is used to excite fields. Klystrons are but one of

several devices used to produce the RF needed to drive accelerators.

These will not be considered in this thesis since they do not directly

pertain to beam break-up. Nevertheless, a photograph of such a device

is included at the end of this section (see Figure 2.13).

Next, we examine another method of particle acceleration.

7
2.1.c Induction Cells.

The induction cell and the linear induction accelerator (LIA) were

developed at the Lawrence Livermore National Lab (LLNL) based on

concepts put forth by Christofilos. The idea is straightforward;

however, the engineering is not. The LIA cells work by sending a

short pulse into an area where a portion of the energy propagates

through a ferrite core while most of the energy travels down a narrow

channel. The channel ends in a tapered gap. The wave produces a

pulsed field across the gap (and hence a potential difference). The

field exists for the length of time it takes the wave to propagate

through the ferrite. This field accelerates the particles passing

between the gap. In simpler terms, the ferrite insulates (by providing

a transient inductance) the two sides of the cavity. Figure 2.4 is a

diagram of the basic LIA module.

Beam
Ferrite Focusing coil

Pulse
input

Vacuum RF reflector
Accelerating gap

Figure 2.4. A LIA module or induction cell.

An induction module can also be understood in terms of a simple

equivalent circuit (see Figure 2.5).9

8
Z
t

ZL I
B

Figure 2.5. An equivalent circuit for an induction cell. The


grey section on the left represents the power
transmission: Zt is the equivalent impedance of the
transmission cables. The compensation load, ZL, is
designed to match the beam load: this represents the
ferrite core. Thus, the drive current is divided, equally,
between the beam and the load.

Although the fields in a LIA module may be quite different from those

in a resonant cavity, the BBU equations are much the same. Further,

there are models which allow LIA modules to be treated in the same

manner as a pillbox cavity. 10 The numerical examples given in Chapter

6 are of LIA’s.

2.1.d Magnets.

The need for transverse focusing in an accelerator is apparent if one

considers all the destabilizing forces acting on the beam: space

charge, wakefields, RF in cavities, etc. This need is primarily satisfied

by magnets,11 although in theory electric fields could be used to

modify the beam motion. In practice, magnetic fields exert much

higher forces on a high velocity beam.12

9
Perhaps the simplest magnetic field which is of any use is the

solenoidal field. If we contemplate a particle travelling inside a

solenoidal field, we see that the Lorentz force will tend to couple the

transverse directions of the particle motion. To understand this more

quantitatively, we will solve for the equation of motion of a particle

moving through a solenoidal field.

e1
e3 V B

e2
I

Figure 2.6. Solenoidal focusing. A particle of velocity v


traverses a region where a solenoidal magnetic field is
used to “focus” the particle.

The basic geometry is given in Figure 2.6; the unit vectors are e1, e2

and e3 (corresponding to generalized directions 1, 2 and 3). We take

the particle’s charge to be q, its velocity to be v with v z>>v ⊥ and the

magnetic field is given by B=Be3. Working in cgs-Gaussian units

(where charge is in statcoulomb), the Lorentz force on the particle is

given by

q qB
F = c v × B = c v × e3
qB
= c (v2e3 - v1e2) . (2.1)

For a relativistic particle,

dv⊥ qB
= (v e - v1e2)
dt γmc 2 1 (2.2)

10
where m is the particle’s mass, γ is the Lorentz factor, and the

particle’s energy is assumed to be constant. We see that this can be


separated into two coupled first order equations (one in e1 and one in

e2). Thus, the particle will tend to move in a helix. The frequency of

oscillations is given by

qB
ωc=
* γmc . (2.3)

This is known as the cyclotron (or gyro) frequency. The solenoid

provides a guiding field: the particle will follow the field lines. Of

course, if the particle's transverse velocity becomes too great, the

magnetic field will not be sufficiently strong to deflect the particle.

A more careful analysis must be done to obtain the focusing effect for a

particle entering and then exiting a solenoid. In this case, the particle

will be influenced by the radial magnetic fields at the solenoid edges.

In addition, the axial field will ramp up (down) at the entrance (exit)

of the solenoid. The radial field crossed into the particle’s axial

velocity produces a “kick” in the azimuthal direction. In turn, the

azimuthal velocity crossed into the axial magnetic field produces a

“kick” towards the axis and in the radial direction. The equation of

motion for the particle is


q
mγdv = c (v×B) - mγ ×( ×r) - 2mγ( ×vr)
dt (2.4)

where again the Lorentz factor is assumed constant and cylindrical


coordinates have been used. Here ω=(vθ/z) in the z-direction.

Separating this into coordinates, we have three equations:

11
dvr q v2θ
r: mγ = c vθBr + mγ r
dt
dvθ q vv
θ: mγ = c vzBr - 2mγ rr θ
dt
q
z: mγdvz = - c vθvr
dt . (2.5)

We can assume that the total angular motion is small throughout the

solenoid and that r is nearly constant. Then, the Coriolis force in the
θ equation can be neglected and vz≈constant. Next, the paraxial

approximation is used: Br ≈(-r/2)(∂Bz/∂z)| r=0 . Thus,

qBz
vθ + r=0
2γmc . (2.6)

The particle’s azimuthal velocity will be the same upon exiting the

solenoid as it was upon entry. This small azimuthal velocity provides

the focusing effect via the Lorentz force. The particle will rotate about

the axis at a frequency

qB
ωβ = 1
* 2 γmc , (2.7)

called the betatron frequency,13 which is one half the cyclotron

frequency. This frequency is independent of the particles distance

from the axis. Thus, a particle traversing several closely spaced

solenoids will undergo focusing while oscillating at the betatron

frequency. Notice that to maintain a constant focusing force, the


magnetic field must increase in proportion to γ. For longer solenoids,

the cyclotron oscillations will dominate over the betatron motion.

Another type of magnet, commonly found in circular accelerators, is

the dipole magnet. This magnet provides an angular deflection of the

12
beam in one dimension. The dipole magnet is used for bending

(steering), not focusing. Thus, dipoles are not useful in conventional

linacs. However, higher order magnets, namely quadrupoles, are used

extensively (see Figure 2.14 at the end of the chapter).

A quadrupole magnet can focus only one transverse direction, and it

does so at the expense of the other direction: one direction is

focused, the other is defocused. Thus, a pair of quadrupoles is

required to provide symmetric focusing.

The transverse motion of particles through a lattice of quadrupoles is

often calculated using transfer matrices. A focusing section comprised

of a focusing (F) and a defocusing (D) with drift spaces (O) between

them is often called a FODO “lens” (see Figure 2.7 for the geometry).

The orbit of a particle through a series of FODO lenses is calculated by

multiplying together the matrices of each lattice element. 14 The

result, when the magnets are periodically spaced,15 and the thin lens

approximation is used, is that the frequency of oscillations is inversely

proportional to the particle’s energy. Thus, particle orbits under both

solenoidal and quadrupole focusing can be described by the same

equation, simply with different frequencies of oscillation.

13
X-axis F O D O F O D

x
z
y
Y-axis D O F O D O F

Figure 2.7. A lattice of equally spaced quadrupole lenses.


Each quadrupole focuses (F) one transverse direction
while defocusing (D) the other. The drift spaces (O) are all
assumed equal. Then lattice is made up of equal FODO
“lenses”.

2.1.e Monitors.

As with any other sophisticated device, it is important to have

diagnostics and sensors in accelerators. Sensors are required to

determine where the beam is in the beam pipe as well as to determine

the condition of the beam. Accelerator operators as well as physicists

need to know various beam parameters such as current, transverse

velocity, frequency of betatron oscillations, etc. We will briefly discuss

two types of sensors (sometime called beam-bugs), as they are used to

measure beam break-up. The first sensor employs capacitive electric

field pickup plates; the second is the b-dot loop. These colorful

names are rather descriptive.

14
Beam Monitor

Beam Pipe

Cut away

Figure 2.9. A diagram of a beam-bug showing the two sets


of parallel plates. Each set monitors one of the transverse
directions.

The first type can be thought of as sets of parallel plates which detect

the beam position (see Figure 2.8).

The potential difference induced between the plates is proportional to

the position of the transverse location of the beam. By using two sets

of plates, one for the x direction and one for the y direction, the

location of the beam’s center of charge at that position in the

accelerator can be known. In practice, the beam-bug is often made of

cylindrical plates which are sectioned off into four arcs. By

performing the proper signal analysis, both transverse positions can be

determined. As the beam goes by the monitor, a response voltage is

displayed versus time on an oscilloscope. This enables the operator to

know the position the beam within the beam pipe.

The other type of sensor we consider is the b-dot loop (see Figure

2.8). The name refers to a device which measures the magnetic flux

15
through a loop of wire. The magnetic flux in the azimuthal direction is

proportional to the beam current passing through at a given point

along the accelerator. Namely, in cgs units,

ε = 1c dΦ = 1 d
dt c dt ∫
loop
B⋅ dA

(2.8)

where ε is the EMF induced in the loop: the change in the magnetic

field induces an electric field according to Faraday’s law. The

magnetic field of the beam is proportional to the beam current. Thus,

the voltage induced in the wire loop is proportional to the rate of

change of current, or dB/dt. The name of this device arises from the

common usage of B (“B dot”) to indicate the differentiation of B with

respect to time.

Beam Monitor

Beam
⊗ B⊗

Cut away B-dot loop

Figure 2.9. A cut away view of a b-dot loop. The wire loop
is positioned with its plane parallel to the direction of
beam propagation so that the azimuthal magnetic field will
link to the loop.

16
2.2 Longitudinal Motion.

We have already discussed the transverse motion a single particle

executes when it is traversing a magnetic field. An accelerated beam

will also have a longitudinal oscillation. This is associated with

synchrotron motion and is caused by a phenomenon know as phase

stability.

Phase stability is a simple concept to grasp. Each cavity has a RF

phase associated with it.16 The phase is measured with respect to the

trough of the accelerating field. In an RF linac, the cavity phases are

tuned so that an ideal beam will arrive somewhere ahead of the crest,

at the threshold energy (see Figure 2.10). The phase-energy wells

form regions of stability.17,18 These regions allow for a range of stable

phases.

Energy
Late
Threshold

Early

φs RF Phase

Figure 2.10. A model of phase stability. Particles arriving


ahead (behind) of φs receive less (more) acceleration than
those arriving at φs.

A particle may arrive too early at a cavity if its velocity is too great.

However, since it arrived ahead of the threshold energy, it will receive

17
less acceleration than those arriving at the correct time. The opposite

is true for a particle arriving too late. Thus, particles will oscillate

about a common velocity. These oscillations in velocity generate

oscillations in the particle positions: synchrotron motion of the beam.

Phase stability enables a beam to be accelerated in a RF linac with a

static relation between the cavity phases.

Longitudinal beam motion is negligible in highly relativistic beams.

Synchrotron motion is minimal because the beam’s velocity is almost

at the speed of light. In other words, for a realistic spread in energy,


the velocity spread, ∆vz, of the beam is minimal.

Consider a beam with an average energy of mc2γ, and with a spread of

2∆γ. Then,

1 1/2 1 1/2
∆vz = c 1 - -c1-
(γ + ∆γ )2
(γ - ∆γ ) .
2
(2.9)

For ∆γ<<γ, ∆vz≈2c(∆γ/γ3). As a concrete example, take ∆γ/γ=5% (quite

large) and γ=5→100 (≈2.5→50 MeV) then ∆vz/c≈3 10 -3→5 10 -8.19 For

the purposes of studying beam break-up, this motion may be ignored.

In addition to synchrotron oscillations, there are other effects which

perturb a beam’s longitudinal distribution, namely, the well

documented effect of longitudinal BBU.20 It is analogous to transverse

BBU. As was stated in Chapter 1, these effects are neglected here.

18
2.3 Photographs of Devices.

Figure 2.11. A top view of an injector used at the


accelerator research center (ARC) at the Lawrence
Livermore National Lab (LLNL). The injector is used for
the sub-harmonic drive relativistic klystron (SHARK). 21
The cables feeding in the external RF power to the first
two cells are visible on the right.

19
Figure 2.12. An artist's conception of a relativistic
klystron.cavity It is designed to extract RF energy from a
relativistic beam passing through its axis.

20
Figure 2.13. A conventional klystron used at SLAC on the
Stanford linear collider (SLC). The device is held
vertically with the electron gun at the bottom. It can
produce a peak power of 67 MW and an average power of
42 KW.

21
Figure 2.14. A typical quadrupole magnet. The beam would
travel through the center, going into the page.

22
References and Notes
(Chapter 2)

1
More accurately, most electron guns in televisions work in a manner similar to those in electron
accelerators, since cathod ray tubes have been in existence longer than televisions.
2
The range of frequencies defined as “microwaves” varies in the literature.
3
The term “excited” is used to indicate that electromagnetic fields are produced in a cavity, and
sustain for a finite length of time.
4
The term RF (from radio frequency) is used in this paper to connote power sources operating in
the microwave region.
5
Copper not only has the excellent conductivity (second only to silver among the elements at
room temperature), it is also economical and has excellent thermal properties, is easy to
machine, electroplate, etc.
6
Field strength in a cavity is usually limited by the amount of RF power that can be generated
and delivered to it. More fundamentally, the field strength is limited by multipactoring
(multiple electron collisions). In other words, a resonant growth of electron current (from
secondary emission) can occur. This current puts an upper limit on the fields in the cavity
because all the power is absorbed by the current. Another limit on the field strength is field
emission. If the local fields in a cavity wall are sufficiently high, electrons can be liberated.
Eventually, the electrons spark across to another surface.
7
In cases where the “BBU mode” cannot propagate through the irises in the disks, the cross-
talk can be ignored.
8
Bunching or phase compression can be induced in a suitable structure whose phase velocity is
less than the speed of light. See R. H. Helm and R. Miller’s article, “Particle Dynamics,” in
Linear Accelerators, op. cit ., p.119.
9
This model for an induction cell, along with the figure, where derived from ones given by R. J.
Briggs, et al., “Beam Dynamics in the ATA 10 kA Linear Induction Accelerators: Observations
and Issues,” IEEE Transactions on Nuclear Science, NS-28 3 (1981).
10
R. J. Briggs, “Deflection of Beam Electrons From Symmetric Excitations of an Accelerating
Gap,” Lawrence Livermore National Lab ATA Note 204 (1983).
11
The use of magnets in linacs may become less prevalent if ion-channel focusing and other
“next-generation” schemes prevail.
12
In the injector, it is often undesirable to have magnetic fields because the beam will be
defocused (see S. Humphries, Principles of Charged Particle Acceleration, pp.156-157 for a
brief discussion). Thus, electric fields are used to focus and steer. In circular machines, where
small changes in a beam’s orbit can be more useful, electric fields are also used. Parallel plates,
called kickers, are employed to make minor orbit corrections, and to stochastically cool beams.
13
Betatron oscillations derive their name from the betatron accelerator where they were first
observed. However, the actual frequency differs from that in a typical betatron.
14
See S. Humphries, Principles of Charged Particle Acceleration (New York: John Wiley &
Sons, 1986), Ch. 6 and 8.
15
Fortunately, this is often the case in linacs.
16
In non-relativistic LIA’s phase stability is ensured by proper timing of the induction cells, and
ramping the applied voltage rather than using a square pulse. At relativistic velocities, this
becomes unnecessary: longitudinal spread of the beam is negligible.
17
A bucket refers to the region of phase stability. A bunch is an area in phase space filled with
particles.
18
It is often helpful to think of relativistic phase stability as analogous to a surfer riding a
wave. If the surfer’s initial speed is close to that of the wave, the surfer will be able to ride
the wave to the beach. This, of course, is only an analogy.
19
See S. Humphries, op. cit , pp.430-436.

23
20
For a consideration of longitudinal wake effects in a linac see R. B. Palmer, “The
Interdependence of Parameters for TeV Linear Colliders,” Stanford Linear Accelerator Center
PUB-4295 (1987).
21
A brief review article on relativistic klystron research at LLNL can be found in Energy &
Technology Review, LLNL (July, 1988), pp.58-59.

24
Chapter 3
The BBU Problem

The problem of beam break-up is formulated.


The concept of a wake is introduced, and the
BBU equation is derived.

3.1 A Qualitative Explanation of BBU.

Perhaps the most fundamental limit placed on beam propagation is the

Alfvén current limit (the “space-charge limiting” current):1 the

highest current possible for a beam propagating in free space. At this

current, the beam’s “self” field will deflect the beam in the direction
opposite to propagation.2 For reference, the Alfvén current is, I A =

(mc 3/e) ≈ 5.1x1013 statamps (≈17.1 kA), where m and e are the mass

and charge of the electron and c is the speed of light in cgs (mks)

units.3

Propagating a beam in free space does not allow for much control over

the beam. Therefore, beams are usually sent through evacuated pipes

(see chapter 2 for more details). Maxwell’s equations afford us a

1
solution to the fields of a particle in a cylindrically symmetric pipe.4 If

the walls were made from perfect conductors, then this geometry

would not be much more interesting than free space (as long as the

particle does not strike the pipe wall). Since, typically, beam pipes

are not made from superconductors, the walls have a finite

conductivity and problems in propagation exist.5 If the particle comes

near the wall, even further difficulties arise; and as soon as the particle

traverses other beamline components still more complications ensue.

When a beam of particles is considered these same problems exist and

new ones are created. Any of these difficulties can lead to beam break-

up.

This thesis only concerns itself with beam break-up due to wakefields:

those problems caused by traversing beamline components.6

RF Feed
Cavity Port

Beam Line

Particle
Cut-away

Figure 3.1. An off-axis particle traversing a cavity can


excite fields. In addition, external RF can be used to
excite the cavity.

A particle traversing a cavity can excite fields in the cavity. In turn,

these fields can affect the particle. However, for a relativistic particle,

2
the transit time through the cavity is sufficiently short that any effects

are negligible. In other words, the fields require a finite time to be

established. And, by causality, the effect of the fields cannot travel

faster than the speed of light, so the signal cannot “catch-up” to the

particle. A cavity can also be driven by an external field (see Figure

3.1). This field certainly can, and does, affect the particle. However,

we shall assume that the engineers have done their job and that this

effect has been minimized. Now, had a second particle followed the

first one within the time that the fields were still excited, the second

particle could have been deflected by the fields excited by the first

particle (see Figure 3.2). In addition, the second particle would excite

the cavity further. This is the simple two-particle model of

cumulative7 beam break-up due to wakefields. It is solved analytically

in section 4 of this chapter.

3
Cavity
Head
Particle
Beam Line

Test
Wake Field
Particle
Cut-away

Figure 3.2. The two-particle model description; an off-axis


particle traverses a cavity and excites wakefields. A
second particle following the first one receives an impulse
from the vxB force of the cavity fields excited through the
J⋅ E coupling of the first.

If we now consider a beam of particles traversing several cavities, a

more complete picture of wakefield BBU in induction linacs can be

ascertained. The beam may be thought of as a stream of slices. Each

slice will excite fields in the structure, coupling through the dipole

moment of the axial current density to the axial electric field of the
modes (J⋅ E). The transverse B-field excited by the first n slices will

deflect the (n+1)th slice via the Lorentz (vxB) force. The further off-

axis a slice is, the greater its dipole-moment-induced fields in

downstream cavities will be. These larger fields cause larger impulses

on follow-on particles and an instability ensues (see Figure 3.3).

4
Cavity Cavity
Head
Slice
Beam Line

Cut-away Cut-away

Figure 3.3. Wakefield BBU for a beam. A beam can be


considered as a stream of slices. The first slice excites a
field which deflects the n-1 slices which follow. The
second slice excites a field which deflects the n-2 slice to
its rear, etc..

3.2 Notation

Most of the notation used in this treatise adheres to convention and

tradition. Unfortunately, different authors use different notation. By

necessity, the units are a mixture of cgs and mks (and sometimes even

inches). This necessity arises because theorists prefer cgs while

experimentalists prefer mks. These preferences are not arbitrary

choices. Cgs (Gaussian) units are appealing in theoretical work

because of the equal footing of the electric and magnetic fields (among

other reasons). However, using cgs units for experimental

measurements would not be practical: test equipment, manufacturing

tools, etc. are all in mks or English units. These are functional

traditions. In keeping with this, this work uses mks units for

physically significant quantities such as current and length, but most

5
analytical calculations are in cgs. Problems associated with convention

are overcome with explicit definitions.

The geometry of an accelerator can be considerably simplified for our

purposes. The beams transverse displacement from the beamline axis


will be modeled by one dimension, ξ. This requires two assumptions.

The first is that the displacement of a cross section of the beam can be

modeled by the displacement of the cross section center of mass.

This assumption is well justified for BBU because we are only

concerned with the mean motion of the beam. The second

assumption is that two transverse dimensions of the beam, i.e. (x,y) or


(r,φ), can be approximated by a single one, i.e. ξ. This second

assumption is valid if the transverse coordinates are not coupled.

Recall that in the case of solenoidal focusing, bending magnets,

asymmetric (flat) beams, etc., the transverse motion is coupled. For

cylindrically symmetric structures and beams with no coupling

between coordinates, the one dimensional approximation is quite

suitable.

The next figure illustrates the relevant coordinates and lengths:

6
S Beam
Smax Beam Head
Tail

Z Beamline R Zmax Beamline


Start End

Magnification
of Beam

Figure 3.4. A sample section of the beam pipe showing the


use of the s and z coordinates. The z coordinate measures
the distance down the accelerator. The s coordinate
measures the distance along the beam starting from the
beam head. The insert shows the ξ coordinate.

The notation used in the above figure and for the remainder of this

thesis is presented in Table 3.1.

7
Table 3.1. A list of notation used in this paper. MKS units
are listed in the third column.

s beam (pulse) coordinate in units of time [sec]


smax pulse length in units of time [sec]
Z beamline coordinate [m]
Z max total beamline (accelerator) length [m]
ξ transverse beam displacement from pipe axis. [m]
rb beam radius [m]
L cavity length [m]
D cavity spacing in units of time (D=spacing/c) [sec]
R cavity radius [m]

The beam coordinate, s, is a convenient variable used in place of time.

The relation between time, the beam and the beamline coordinate is

given by,

t = vz - s ≈ cz - s
z (3.1)

where v z is taken to be constant and ≈ c. This definition will be re-

introduced during the derivations in the next section. For

completeness, the next table gives some additional notation used in

this paper.

Table 3.2. Further notation used in this paper. MKS units


are listed in the third column.

c speed of light in vacuum [m/sec]


e the charge of an electron (a positive value) [C]
m the electron mass [kg]
γ beam energy (E=γ mc 2) [-]
ρ charge density (charge/unit length) [C/m]
W⊥ transverse wakefield [V/C/m]
kβ betatron wavenumber, kβ = ω β/c [m -1]
λβ betatron wavelength, λ β=2π/kβ [m]

Although definitions of the transverse wakefield vary,8 this paper uses

a wakefield with units cm-2sec -1 in cgs (Volts/Coulomb/meter in mks).

8
3.3 Derivation of the BBU Equation.

The derivation of the beam break-up equation requires three steps. In

step one we calculate the impulse imparted to a particle crossing a

cavity due to the fields in the cavity. Step two requires calculating the

fields in a cavity due to the moment of an off-axis particle. Step three

involves combining the results of steps one and two by use of a force

equation.

3.3.a Step One: The Impulse

Let us begin by assuming that we know the fields in an excited cavity.

We now wish to calculate the transverse impulse imparted to a particle

traversing such a cavity. The Lorentz force integrated along the cavity

gives the change in momentum,9


t
∆p⊥ = e E⊥ + 1
c ( v × B) ⊥ dt′
t0 , (3.2)

where t 0 and t are the time of entry and exit through the cavity,

respectively. It is a simple matter to convert this into an equation

written in terms of the vector potential,10


L
A⊥ + (v × (∇ × A))⊥ dz
e ∂
∆p⊥ = cv -
z ∂t
0 , (3.3)

where dz=vz dt, E=-∂A/c∂t, and B=∇xA. This equation can be

simplified by use of the vector relation

9
v × (∇ × A) = ∇( v⋅ A) - A × (∇ × v) - (v⋅ ∇ ) A - (A⋅ ∇ ) v, (3.4)

and by the assumption that the beam velocity in the transverse


direction, v ⊥ , is constant. This amounts to an assumption that the

transverse velocity is not altered in the cavity (an acceleration, of

course, can still be imparted). In other words, we are working to first

order in A. Then,

v × (∇ × A) ⊥ = ∇ ⊥( v⋅ A) - (v⋅ ∇ ) A⊥ (3.5)

and,


L
e
∆p⊥ = - cv  ∂ + v⋅ ∇  A - ∇ (v⋅ A) dz
 ⊥ ⊥
z  ∂t
0 (3.6)

The left term in the integrand is the convective derivative,

 ∂ + v⋅ ∇  A = dA⊥
 ∂t  ⊥ dt . (3.7)

Hence,

 L L 
e 
∆p⊥ = - cv
z  0 ∫
dA⊥ dz -
dt 0 ∫
∇ ⊥(v⋅ A) dz 
 (3.8)

Assuming that v⊥ <<v z, we may take v≈vz (where z is the unit vector in

the direction of propagation, z). Then,

∆ p⊥ = e e
c (A⊥(z=0,t 0) - A⊥ (z=L,t) ) + c ∫ 0
∇ ⊥ Az dz
. (3.9)

The time dependence is explicitly given for the vector potential

because of integration over the convective derivative. In other words,

even if the vector potential is uniform in space, the field might change

10
during the particle’s finite transit time. If the vector potential had

e ιωt time dependence, then A(t)=A(0)e ιωL/c. These first terms

represent the fringe fields of a cavity due to ports or other non smooth

features. Under suitable geometry, this term may be neglected.

Specifically, if the cavity ends are perpendicular to the walls and the

ports are small compared to the cavity radius, then the fringe fields

are negligible.

The second term can be evaluated by considering two orthonormal

modes of excitation: the transverse electric (TE) and the transverse

magnetic (TM) modes. For TE modes, the longitudinal electric fields


vanish, Ez=0, so Az=0 and the integrand vanishes. Hence,

∆p⊥ =0 for TE modes. (3.10)

For TM modes we have11

* ∆ p⊥ = e
c ∫ 0
∇ ⊥ Az dz
(3.11)

which can be written as

∆ p⊥ = - ιe
ωc ∫ 0
∇ ⊥ E z dz
(3.12)

if we consider the vector potential (and hence the fields) to have e-ιωt

time dependence. Equation (3.12) constitutes a solution to step one:

the impulse due to the fields in a cavity is now known (to first order).
These results were obtained by neglecting the transverse velocity, v⊥ .

Had this not been done, there would have been a contribution from

the transverse components of the vector potential. Thus, the TE

11
modes would, in fact, impart some impulse to the particle. This

contribution is negligible for relativistic beams. As such, we shall only

consider TM modes for beam break-up.

Now we are required to perform step 2. A linearized solution for the

fields due to an off-axis particle can be obtained using any one of

several methods. The derivation presented here follows from a

“dominant mode” calculation of the fields in an ideal cavity.12 Another

derivation is presented in Appendix 2. The method used in the

Appendix is due to W. Panofsky13 and is one of the earlier treatments

of the theory of beam break-up. The approach relies on energy

conservation and physical insight which only physicists of Panfosky’s

caliber seem to possess. The Panofsky derivation, although very

intuitive, is not included in the main text because it arrives at a less

useful model of BBU.

3.3.b Step Two: The Field.

We consider the electric field as a sum over orthogonal modes of the

electric fields. In other words,14

E( r,t) = ∑ p λ(t)E λ( r) + m λ(t)Fλ( r)


λ (3.13)

where λ =(m,n,p) is the mode number triplet, E and F are the time-

independent, divergent-free and curl-free electric fields,

respectively.15 (Do not confuse F for field with F for force.) The vector

potential can be written in the Lorentz gauge as

12
A( r,t) = ∑ p λ(t)Eλ( r)
λ . (3.14)

(Note, here p is merely used to differentiate it from p, not to indicate

a complex variable.) We also require the magnetic field,

B( r,t) = ∑ hλ(t)Bλ( r)
λ . (3.15)

In order to solve for the fields in a cavity excited by an off-axis particle

or beam, we employ Maxwell’s equations. After obtaining a suitable

second order differential equation, we take into account the losses in a

cavity due to ohmic heating. As it will be shown, it is sufficient to


consider λ=(1,n,0) modes; in other words, dipole modes with no z

variation. Next, we solve the equation to first order in the particle’s

displacement.

Let us begin by applying Faraday’s law which yields

pλ ≡ ωλ dhλ Bλ ≡ - 1 (∇ × E λ)
dt kλ , (3.16)

where ωλ =k λ c as usual. Similarly, Ampere’s law, equated to the curl of

Faraday’s law, produces the following relation:

hλ ∇ 2E = 1 dp λ E + dm λ F + 4π J
λ c dt λ λ c
kλ dt (3.17)

where J is the current density, and the orthogonality of the modes has

allowed the summation to be suppressed (i.e. the equation holds for all
λ). In fact, we can now consider a single mode to dominate. If we

now examine the equation by components (E and F), differentiate with

respect to time, and simplify, we arrive at

13
ωλp λ 2 d 2p λ
∇ Eλ - 1 E λ = 4π dJ
kλ c dt 2 c dt (3.18)

where J is understood to be the z (curl-free) component of the total

current density.16 Next, we require the wave equation,


2
∇ E λ + k 2λ E λ = 0 , (3.19)

in order to obtain the relation

 ω2 + d 2  p = - 4π d
 λ dt 2  λ dt ∫
Cavity
(J⋅ Eλ)dV
. (3.20)

Equation (3.20) holds for an arbitrary cavity geometry under the

assumption that the fields are cut off in the beam pipe. (This

assumption allows us to neglect the “beam-port loading” and “beam-

wall” loading terms.) In terms of the vector potential coefficient,

 ω2 + d 2  p λ = - 4π
 λ dt 2  c ∫ (J⋅ Eλ)dV
Cavity . (3.21)

We have arrived at an equation for the rate of energy flow into a cavity

due to a beam with current density J.17 Unfortunately, this relation

does not take into account the losses of a cavity due to a finite Q

value.18 The best way to introduce these losses is by adding a damping

term 19 to give,

 ω2 + ωλ d + d 2  p λ = - 4π
 λ Qλ dt dt 2  c ∫
Cavity
(J⋅ Eλ)dV
. (3.22)

This can be inverted to give,

14
t

⌠ sin ωλ(t-t′)  4π
∫ (J⋅ Eλ)dV  dt′
ωλ
pλ = -  exp - (t-t′)
ωλ  c

2Qλ
0  Cavity 
(3.23)

with t=0 defined as the time at which the beam enters the cavity.

This ungainly equation can be greatly simplified. We begin with the

current density (assumed uniform) induced by a charge distribution

with a step radial profile,

ρ = ρ 0 θ( r b - r - ξr ) (3.24)

where θ is the step function and ξ is the displacement of the beam

centroid. Pursuant to the assumptions stated in chapter 1, ξ<<r.

Thus, we can obtain an expansion to first order in ξ for the charge

distribution,

ρ ≈ ρ 0 θ( r b - r ) + ρ 0 δ( r b - r ) ξcosφ (3.25)

where ρ 0=I/(πr b2c) and I is the beam current. This is the monopole

and dipole terms in the multipole expansion of ρ. Notice that the

higher order poles involve higher powers of ξ. Hence, we can limit

our discussion to m=0,±1 modes (i.e. λ=(0,±1,n,p)). Now, J = c ρz in

the limit that the transverse velocity of the beam is negligible.


Possessing the current density, we now only require the field, E z.

The exact form of the field will depend on the particular cavity

geometry. Fortunately, a pillbox cavity (a cylidrical cavity with

perpendicular ends) serves as a useful model and allows for a simple

solution. We will proceed with this special case, and generalize the

result later. Applying Maxwell’s equations, one obtains the result20

15
E z = E 0 Jm xmnr
R ( )e ±ιmφ cos 
pπz 
L  (3.26)

where J m is the Bessel function (of the first kind) of order m, x mn is

the nth root of J m , and the time dependence is suppressed. Notice

that for m=±1, the field is odd in φ. Hence, the monopole term in the

current density expansion does not contribute to the integral of J.E.

This is not true for m=0, or when the current density is not uniform
over the cavity.21 The constant E0 is calculated from the normalization

requirement,


Cavity
E λ⋅ E *λdV = 1

, (3.27)

with “ 1 ” having units to be chosen later. Since the normalized E 0

factor will only depend on the geometry of the cavity and the mode of

excitation, we will not solve for it.22 Having assembled the necessary

pieces, the solution is at hand:

∫ ∫
L
(J⋅ Eλ)dV = E0 Jm( r
xmn b
R ) 1
rb Iξ cos 
pπz 
L 
dz
0
Cavity (3.28)

where a factor of π comes from the φ integration. The integral term

involving ξ can be approximated by using integration by parts,

discarding the term involving the derivative of ξ and assuming the

current does not vary appreciably over the cavity.23 Then for p≠0,


L
pπz  pπz  L
ξ cos   dz ≈ ξ L sin 
L pπ L  0
0 . (3.29)

16
Such approximations are not necessary if we consider the lowest axial
mode, p=0. Then, the integral term simply becomes L<ξ>, where

brackets connote averaging over the cavity. We proceed with the p=0

mode, keeping in mind that other modes could easily be analyzed if

necessary. Finally, Eq. (3.23) can be expressed as

p λ = - A Ec0 ⌠ ωλ sin ωλ(t-t′)


(I ξ ) dt′
* D

exp -
2Qλ
(t-t′)
ωλ
0 (3.30)

where A is a unit-less constant which depends on the cavity geometry,

and D is the cavity spacing, as before.24 This concludes step two: we

have obtained the fields in a cavity due to the off-axis displacement (or

moment) of a beam.

The third and final step in our derivation will obtain a force equation

based on this result and the result obtained from step one.

3.3.c Step Three: The Force Equation.

We desire to find the force impressed upon a particle traversing an

excited cavity. The force is obtained from the time derivative of the

impulse:

∫ ∫
L L
∆p⊥
F⊥ ≈ = e ∇ ⊥Az dz = e ∇ ⊥Az dz
∆t ∆tc D
0 0 (3.31)

where ∆t is the particle transit time from one cavity to another and
the relation ∆t = D/vz ≈ D/c was used. Further,

17
 dξ 
F = m d  γ 
dt dt  . (3.32)

We immediately obtain the formula:

d 2ξ dlnγ dξ
= F - ω2β ξ -
* dt 2 mγ dt dt (3.33)

where the focusing term, (ωβ) 2ξ, was obtained in Chapter 2 (see Eq.

2.7). Finally, the equation of motion (3.33), driving force (3.31) along

with the field equation (3.30) can be combined with the above results

to yield,

d 2ξ dlnγ dξ
+ + ω2β ξ =
dt 2 dt dt
t
E20
- AB e c 1 ⌠ exp -
ωλ
(t-t′)
sin ωλ(t-t′)
(I ξ ) dt′
D mγ
⌡ 0
2Qλ ωλ
(3.34)

where B is a unit-less geometrical factor resulting from the averaging

of the gradient of the vector potential along the beam cross section.

Once again, it is desirable to make a change of variables from (t,z) to

(s,z). In doing so, the above equation can be expressed as25


s
d ξ dlnγ dξ
2
+ k 2β ξ = 1 ⌠ I(s′)
ξ(s′,z) ds′
* dz 2
+
dz dz Dγ

W⊥(s-s′)
IA
0 (3.34)

where I A = (mc 3/e) is the Alfvén current and W⊥ (s) is called the

transverse wake function. For the special case of a pillbox cavity,

ωλ sin ωλs
W⊥(s) = - AB ∑ E20,λ exp - 2Q s
ωλ
λ λ . (3.35)

(This result is, of course, only valid for the λ=(m=±1,n,p=0) T M

modes.) Many authors include the 1/D term into the definition of the

18
wake; however, by not including it, the wake describes the response of

the cavity alone. Thus, the arrangement of the cavities along the

accelerator can be taken into account by making D a function of z.

With this definition, the wake has units of 1/(cm-sec) in cgs.26

3.4 Understanding the Wake

The wake, or wakefield, has already been described as the fields

excited by off-axis particles which persist long enough to affect follow-

on particles. The transverse wakefield calculated in section 3.3.c was

derived using several approximations. Our model uses a wakefield

which is proportional to the displacement of the exciting charge but

independent of the test charge’s displacement. The advantage of this

result is that it does not require us to explicitly know the fields in the

cavity. In general, the wakefield can easily be expressed in terms of

the fields27:


L
W⊥ 1
= ec ( E + v × B) ⊥dz
0 . (3.36)

This is simply the Lorentz force integrated over the cavity28 and

divided by the (test) particle charge. This form is especially useful for

computer codes which solve for the fields in arbitrary geometries.

Analytic results for complicated geometries are often intractable. It

also gives some additional insight into the nature of the wake. The

Lorentz force on a particle is quite familiar. The model presented in

section 3.4 used the second term (the dipole term) in the moment

19
expansion of the wake. (The monopole term takes into account DC

effects. Higher order terms taken into account the test particles

displacement as well as non-linear contributions to BBU.)

Alternatively, the wake can be considered as an average over the cavity

of the force on a particle (especially if one were to include the 1/D

factor). Thus, our model uses this averaged force as a measure of the

force on the beam weighted by the beam displacement off-axis. Then,

the left hand side of Eq. (3.34) is the driving force.

It is quite revealing to study the Fourier transform of the wake,29


+∞
W⊥( ω) = AB ds′ W⊥(s′) e ιωs′
2π 0 (3.37)

which yields the following result for a pillbox cavity:


2
W⊥( ω) = ABE 0 2 1
2π Ω + ω2λ (3.38)

where Ω=(-ωλ/(2Q) + ιω) 30. In and of itself this might not seem an

interesting result; however, this is, in fact, the effective impedance of

the cavity (to within some constant factors). Convention requires that

we write

Z⊥ = 2π ιL
c W⊥ (3.39)

as the impedance (units: sec/cm in cgs; Ω/m in mks). The impedance

is quite familiar to engineers. Being a frequency-dependent quantity,

rather than a time-dependent one, means that impedance can easily

be measured in the lab.

20
BBU can then be understood in terms of a simple model (much like an

equivalent circuit). A cavity can be modeled by a single impedance

boundary condition. This is often referred to as the Briggs model (see

Figure 3.5).31

Z
s

Rp

Figure 3.5. The Briggs model of a cavity: an impedance


terminated idealized pillbox. Here, Zs is the surface
impedance, R and L are the cavity radius and length,
respectively, and Rp is the pipe radius.

Although this model can be quite useful for obtaining approximations

to a cell’s impedance (or wake), we will not consider it further.

Instead, we turn our efforts to understanding and solving the BBU

equation.

21
References and Notes
(Chapter 3)

1
Alfvén’s original paper was concerned with cosmic rays, not electron beams; however, the
concepts are equivalent. See H. Alfvén, “On the Motion of Cosmic Rays in Interstellar Space,”
Phys. Rev. 55, 5 (1939).
2
For a relativistic beam, the electric and magnetic fields negate one another to within a factor
of 1/γ 2. At the Alfvén current, the net force from the fields is sufficient to stop the beam.
3
For a further discussion, see R. B. Miller, An Introduction to the Physics of Intense Charged
Particle Beams (New York: Plenum Press, 1985), pp. 84-87.
4
For a partial derivation of the fields see A. W. Chao, “Coherent Instabilities of a Relativistic
Bunched Beam” (Stanford, California: SLAC, 1982), SLAC-PUB-2946, pp. 6-10.
5
The resistive wall instability is one such problem. For example, see V. K. Neil and D. H.
Whittum, “Transverse Resistive Wall Instability in a High-power Short-Wavelength Free-
Electron Laser,” Lawrence Livermore National Lab UCRL-96712 (1987).
6
For the remainder of this paper, the term beam break-up (BBU) will be used to mean
wakefield beam break-up.
7
For an explanation of the differences between cumulative and regenerative BBU see Ref. 3.
8
For an explanation of the various formalisms used see, G. Craig and J. DeFord, “The Connection
Between SLAC-DESY and LLNL Wakefield Formalisms,” Lawrence Livermore National Lab
Beam Research Memo 89-7.
9
The following derivation follows that given by Panofsky and Wenzel in 1956. See 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).
10
By our initial assumptions (stated in chapter 1), the integration performed from 0 to L should
have been evaluated from -∞ to +∞. Recall that we are assuming that the fields evanese in the
beam pipes and that they do not affect the fields in adjacent cavities. However, no information
is lost using finite limit since the fields die-off exponentially outside of the cavities.
11
Although the Panofsky-Wenzel theorem is now used to refer to another result, the result
given here is their original one. See Ref. 7.
12
A search of the literature seems to indicate that such calculations are deemed too
straightforward to publish. However, the author would like to credit D. H. Whittum for the
use of his notes which provided the basis for the derivation in this paper.
13
W. K. H. Panofsky, “Transient Behavior of Beam Break-Up,” SLAC TN-66-27 (1966).
14
This is a standard technique. See J. D. Jackson, Classical Electrodynamics (New York: John
Wiley & Sons, 1975), ch.8. Additionally, see J. C. Slater, Microwave Electronics (Princeton,
New Jersey: D. Van Nostrand Company: 1950), pp.57-67.
15
Recall that any vector field can be represented as a sum of a divergent-free vector and a curl-
free vector. This formulation could equivalently be done with E=Ez and F=E⊥ : Ez is divergent-
free because there cannot be a potential difference along the cavity walls, and E⊥ is curl-free
because the walls are conducting.
16
The assumption that v≈vz implies that J≈Jz, so that this is not an additional approximation.
17
Actually, the result is much more general than suggested. The equation holds for any ideal
cavity with no ports or wall losses. See Slater, Ref. 17.
18
The Q, or quality, of a cavity is defined as the ratio of the energy stored to the power lost.
See Appendix 2 for more details.
19
It is a non-trivial task to arrive at a simple loss term involving Q directly from Maxwell’s
equations.
20
The derivation of the fields in a cavity is given in Jackson, op. cit ., pp. 353-356.

22
21
For m=0, there is a contribution which does not depend on the beam’s displacement. Thus, this
contribution does not lead to an instability. The case of a non uniform current distribution over
the cavity length will not be considered here.
22
It can be shown that
E 0,λ = 2 1 1
πLR 2 J (x
1 mn) pπR  2
1 + 
 xmn L  .
23
This assumption is valid, in general, as long as ξ does not change appreciably over the
distance of the cavity. For the p=0 mode, no such approximation is needed to solve this term.
24
The factor A=4πLJm(x mnrb/R)/rb is simply a collection of constants pertaining to the cavity.
For particular cases, A can be simplified; however, we are primarily concerned with the units
involved.
25
In light of the approximations used, it is to be assumed heretofore that ξ does not vary
appreciably over a cavity length, and so ξ≈<ξ>.
26
With these units, E02 has units of 1/(cm-sec2).
27
Just as the transverse wakefield is defined here, the longitudinal wakefield is defined as the
integral over the longitudinal (z component) of the Lorentz force. However, since the transverse
velocity is considered to be negligible, the magnetic field contribution to the Lorentz force is
often neglected. Then,


L
1
Wz = ec E zdz
0 .
The general relation between the transverse and longitudinal wakes is given by,
∂W⊥
= ∇ ⊥Wz
∂s .
28
Again, the integral should be performed from -∞ to +∞, but in the limit that the fields are cut-
off in the beam pipes, then this is a suitable approximation.
29
The integration is performed from a lower limit of zero because the wake, W⊥ , is zero for s<0
(by causality).
30
For clarity, the summations over the field modes have not been performed. In other words,
the results given hold for a single mode, λ.
31
R. J. Briggs, “Deflection of Beam Electrons From Symmetric Excitations of an Accelerating
Gap,” Lawrence Livermore National Lab ATA Note 204 (1983).

23
Chapter 4
Analytic Solutions

Various special-case analytic solutions to the


beam break-up equation are derived. The two-
particle model is explored to gain a better
understanding of the BBU problem. Finally, a
“general” solution to the BBU equation is
derived.

4.1 Special Case Solutions.

Analytic solutions to the BBU equation exist under certain

approximations. As with other problems, it is instructive to consider

certain special cases before attempting a general solution of the BBU

equation. These special-case solutions serve many uses: verification of

more general solutions, confirmation of numerical solutions, and a

means of better understanding the equation.

The first two special cases, the “no wake” and “constant wake”,

examine the simplest transverse oscillations. The linear wake case,

treated in Section 4.1.c, is the first practical solution. This case

outlines some of the methods that are used to derive the general

1
solution of Section 2.3. The final special case considered is that of a

“coasting” beam for a general wake. This case also presents

techniques useful for obtaining the general solution.

4.1.a No Wake.

For a zero wake (i.e. W⊥ =0), the BBU equation reduces to

d 2ξ dlnγ dξ
+ + k 2β ξ = 0
dz 2 dz dz (4.1)

A zero wake might be achieved in a perfectly smooth, perfectly

conducting beam-pipe. However, because of relativistic effects, even

realistic beam-pipes can be considered to have zero wake.1

Clearly, if the energy does not vary along the accelerator, then this

reduces to a simple harmonic oscillator, the betatron-motion equation

of Chapter 2. If, however, the energy does vary along the accelerator

(as it would in any accelerator), then we have a more general oscillator

equation.

4.1.b Constant Wake.

A constant wake can be used to model effects from externally excited

cavity fields. For a beam with constant current distribution, and a

constant wake geometry,

2
d 2ξ dlnγ dξ
+ + k 2β - W⊥ I smax ξ = 0
dz 2 dz dz Dγ I A . (4.2)

Thus, the frequency of transverse oscillations is altered by the

wakefield; however, the amplitude is not. This indicates that there is

no instability for this model. There can still be beam growth, but not

in an unstable fashion.2 Hence, our assumption in chapter 3 that, “the

engineers did their job,” (i.e. that the externally excited fields do not

contribute to BBU) was a good one.

4.1.c Linear Wake.

The simplest, non-trivial wakefield is a linear one: W⊥ (s)=W 0s.3 (No

generality is lost by not considering W⊥ (s)=W 0s+W1). This form of the

wake is useful because it can be used to model wakes under certain

conditions.4 Specifically, if the wake changes slowly over the beam’s

length then the linear approximation is useful.

Assuming a linear wake and slow variations in γ and kβ (over a betatron

wavelength), the BBU equation can be solved in terms of an asymptotic

series. An outline of the derivation is sufficient to reveal some of the

steps needed to obtain the more general solution of section 2.4 (see

Appendix 3 for a complete derivation). We begin by rewriting the BBU

equation under the above assumptions:


s
d 2ξ 1 dγ dξ
+ + k 2β ξ = W0 I ξ (s-s′) ds′
dz 2 γ dz dz Dγ I A 0 (4.3)

3
The first step to obtaining a solution involves making a “WKB”

approximation with,


 
z
γ(0) k β(0)
ξ(s,z) = Re χ exp  ι k β(z′)dz′
γ(z) k β(z)  
0 (4.4)

where χ is a complex variable. The variable χ can be solved for in

terms of a power series, and the explicit dependences on γ(z) and

k β(z) are eliminated by making a change of variables from z to


z

z= ⌠ γ(0) k β(0)
dz′
⌡ 0
γ(z′) k β(z′)
. (4.5)

Finally, an expression for the transverse displacement is obtained in

terms of a “parameter of largeness”:

∞ 

 
z

(ιη)n γ (0) k β(0)


ξ(s,z) = Re ∑
 n=0 n! (2n)!  γ (z) k β(z)
exp  ι k β(z′)dz′
 0  . (4.6)

where

η = W0 I zs
2
D I A 2γ0k 0 . (4.7)

This can be expressed asymptotically, for |η|>>1, as,5

ξ(s,z)= 1

γ (0) k β(0)
γ (z) k β (z) 4 (
2η -1/6exp 3 3 2η 1/3 )
(
× cos k β(z)z - 3 2η
4
1/3
- π
12 .) (4.8)

When γ=γ 0 and k β=k 0 are constant, and the current distribution is

taken to be a constant over the beam, the solution is exact. Notice

that there is exponential growth proportional to (z) 1/3(s) 2/3. In other

words, the growth length is given by

4
1 = 3 3 W0I s2
* Lg 4 γ0k 0I A D (4.9)

where ξ~exp[(z/Lg) 1/3]. This result can most easily be evaluated

numerically (see Figure 4.1).

Figure 4.1. Growth predicted by the asymptotic formula


for BBU. The envelope is proportional to z1/3. The
parameters used are similar to those in Table 4.1 for the
“ATA”; however, modifications were made to make η large.
The horizontal axis is the distance down the accelerator in
meters. The vertical axis is the normalized displacement,
ξ/ξ0.

Although the linear wake can be exploited further, we will continue in

out pursuit of a more general solution. Since we have already obtained

solutions for special wakes, we next turn to a solution for a general

wake; but under the assumption that the beam is not being

accelerated.

5
4.1.c No Acceleration.

For a “coasting” beam, with γ(z)=γ0, a solution to the BBU equation is

derivable for an arbitrary wake. We consider the case of a constant

current profile along the beam. Then, the BBU equation reduces to


s
d 2ξ(s,z)
+ k 2β(z) ξ(s,z) = K(z) ξ(s′,z) W⊥(s-s′) ds′
dz 2
0 , (4.10)

where K(z)=-I(z)/[D(z)γI A ] and the assumed dependencies (on s and z)

have been shown explicitly.6 Laplace transforming in s results in7

d 2ξ(p,z)
+ k 2β(z) - K(z) W⊥(p) ξ(p,z) = 0
dz 2 . (4.11)

This is the same form as Eq. (11) obtained in Section 2.1.a under the

assumption of a constant wake. The Laplace transform allows the case

of a more general wake to be considered. If we were to further


assume that K and kβ did not vary in z, then an exact solution would be
trivial:8 ξ(p,z)=C 1exp{ιC2z[k β2-KW⊥ (p)] 1/2}, where C1 and C2 are

(complex) constants. Then, an inverse Laplace transform recovers


ξ(s,z):

+ι∞
ξ(s,z) = Re C1 dp exp ps + ιC2z k 2β - K W⊥(p)
2πι -ι∞
. (4.12)

Had we not assumed K and kβ to be independent of z, then a “WKB”

approximation (as in the previous section) could have been used to

obtain an approximate solution. Carrying this out, one finds that

6
 k 2β(0) - K(0) W⊥ (p)  1/2
χ 0 2  →
 k β(z) - K(z) W⊥ (p) 
ξ(p,z) =Re


z
→ exp ι dz′ k 2β(z′) - K(z′) W⊥(p)
0 , (4.13)

where χ 0 is a complex constant and


W⊥(p) =
∫ 0
ds W⊥(s)e -ps
(4.14)

Again, an inverse Laplace transform would recover ξ(s,z).

4.2 The Two-Particle Model.

The two-particle model is the simplest representation of BBU. We will

use the general results found in chapter 3 to quickly derive the

simplified equations of motion. As before, we begin by considering a

charged particle traversing a cavity and exciting some fields. A second

charged particle follows a time S (or, equivalently, a distance cS)

behind the first particle. As a first case, we assume that the particles

are not accelerated. Both particles are assumed to have an initial


transverse displacement of ξ0 (before having entered the cavity). The

first particle is subject only to betatron motion, and satisfies,

d 2ξ1
+ k 2β ξ1 = 0
dz2 . (4.15)

The second particle will experience an impulse due to the fields

excited by the first particle. Thus,

7
d 2ξ2
+ k 2β ξ2 = CW ξ1
dz 2 (4.16)

where CW=NeW ⊥ (S)/(I A γD), Ne is the total charge of the two particles,

and the betatron wavelengths of the two particles are assumed equal.
The first equation is easily solved; we take ξ1=ξ0exp(ιk βz) with the

understanding that the displacements, ξ, are now complex variables

and that only the real part is of interest. Then, the second particle’s

motion can be given by

ξ2 = ξ0 CW ze ιkβz + ξ1
2 ιkβ . (4.17)

Thus, the growth in the transverse displacement of particle two will

be proportional to the distance traveled down the accelerator (see

Figure 4.2).

Figure 4.2. A plot of the displacement of the tail particle.


The horizontal axis is the distance down the accelerator in
meters. The vertical axis is the normalized displacement,
ξ2/ξ0. The parameters used are those given in Table 4.1
for the “SLC”.

The two-particle model can be interpreted as follows: the first

particle represents the head of the beam with, say, a charge of Ne/2;

8
the second particle represents the tail of the beam also with charge

Ne/2. The particles are then considered as macro-particles which can

crudely model the beam. We now have a simple growth estimate for

the tail of a beam.

The usefulness of two particle model can be extended by taking into

account the difference in betatron wavelength that the two particles


might have. Take k 1 and k 2=k 1+∆k to be the first and second

particles’ betatron wavenumbers, respectively. For ∆k<<k 1, the result

is given by9

ξ2 = ξ0 1- CW  2ι sin  ∆kz  exp ι k 1+ ∆k  z + ξ1


 4k 1∆k  2 2 . (4.18)

Notice that the growth is no longer linearly proportional to z. Instead,

the growth is oscillatory (see Figure 4.3). The differing envelopes of

Figures 4.2 and 4.3(b) make the contrast clear.

9
(a) 10 (b)

Figures 4.3(a,b). Two plots of the displacement of the tail


particle with a one percent spread in the betatron
wavelength (i.e. ∆k=.01k 1). The parameters used are those
given in Table 4.1 (a) for the “TBA” and (b) for the “SLC”
(only one third of the length is plotted). The oscillatory
growth is partially visible in (a) and not at all in (b) because
the frequency of oscillation is so large. The horizontal axis
is the distance down the accelerator in meters. The
vertical axis is the normalized displacement, ξ2/ξ0.

The coefficient in front of Eq. (4.18) suggests that the growth of the

second particle’s displacement can be minimized by a judicious choice


of ∆k. Namely, if

W⊥(S)
∆k = CW = Ne
4 k1 4 k1I A γ D (4.19)

then the growth will be a minimum (zero). Figure 4.4 shows a 3-D

plot of the displacement as a function of the spread in the betatron


wavelength, ∆k, and the distance down the accelerator, z. The region

of minimum growth is clearly visible.

10
Figure 4.4. A 3-D plot of displacement versus ∆k and z.
For the parameters used, a ∆k of around 2.5% yields a
minimum growth. Note, betatron motion (i.e. the ξ1 term)
has been ignored.

A spread in the betatron wavenumber can be achieved in one of two


ways. The first is a variation of the energy (the relation ∆k/k=-∆E/E

holds). This suggests that a beam with a higher energy in the head

than in the tail might experience less growth due to BBU. This

concept is explored in detail in chapter 5. A second way to achieve

the spread is by a variation of the magnetic field. This might be

induced by a time varying quadrupole magnet.11

11
Table 4.1. A list of parameters used for the two particle
model. Two machines are given: the “simplified SLC”12
and the “simplified ATA”.

Name Simplified Simplified Simplified Simplified


SLC (mks) SLC (cgs) ATA (mks) ATA (cgs)
Number of Particles 5 1010 5 105 3 1015 3 1015
Final Energy 50 GeV γ = 9.8 104 50 MeV γ =100
Betatron Wavelength 100 m 104 cm 1m 100 cm
Bunch or Pulse Length 3.3 psec 3.3 psec 50 nsec 50 nsec
W⊥/D 2.4 1015 8 109 7.9 1013 2.6 108
(V/C/m2 ) (1/cm2 -sec)
Machine Length 3000m 3 105 cm 100 m 104 cm
Growth with no ∆k. 6.9 ≈6000
∆k/k for min. growth 0.024 growth reduced by a factor
of 3 at ∆k/k=0.024
∆E (equivalent energy) 1.2 GeV 1.2 MeV

The examples given in Table 4.1 illustrate the power of the two-
particle model. The examples suggest that a spread in energy (or kβ)

might be useful for the SLC (even though a 1.2 GeV is quite large),

while for the ATA growth is only reduced by a factor of 3 for a 2.4%
spread in k β.13 In fact, a spread in energy is used in the SLC (and

elsewhere) to control BBU. Much of the success of the SLC can be

attributed to the increased luminosity brought about by the control of

BBU through such methods.

4.2 General Solution by Steepest Descents.

The special cases considered above have outlined methods useful in

obtaining a more general solution to the BBU problem. In this section

we seek such a solution. The derivation is done in detail to

compensate for the brevity of the special cases. Only three

12
assumptions (beyond those used to obtain the BBU equation) are

necessary. The first assumption is that the beam current, I, does not

vary along the beam. The second is that the beam energy, or
equivalently γ, also does not vary along the beam. (In chapter 5, the

second assumption is replaced by assuming that γ does not vary along

the accelerator.) The third assumption is that γ and k β vary slowly

over a betatron wavelength, λ β. For this derivation, no restrictions are

placed on the distribution of acceleration cells along the accelerator.

Under the present assumptions, the transverse beam displacement, ξ,

satisfies


s
d 2ξ(s,z) 1 dγ dξ(s,z) + k 2ξ(s,z) = K(z)
+ β ξ(s′,z) W⊥(s-s′) ds′
dz2 γ dz dz 0
(4.20)

where K(z)=I(z)/[I A γ(z)D(z)]. As was done for the special cases, we

begin by making a WKB approximation. Whence,


 
z
γ(0) k β(0)
ξ(s,z) = Re χ exp  ι k β(z′)dz′
γ(z) k β(z)  
0 (4.21)

so that χ must satisfy


s
dχ K(z)
= χ W⊥(s-s′) ds′
dz 2ιk β(z) 0 . (4.22)

A judicious change of variables allows us to remove the implicit z

dependances. Define
z

z= ⌠ dz′
γ0k 0
K(z′)
⌡ 0
k β(z′)
, (4.23)

13
where γ0=γ(0) and k 0=k β(0). (Notice that this z is differs from the

previous definition: no factor of K(z) was present.) Then, Eq. (4.18)

becomes


s
dχ(s,z) 1
= χ(s′,z) W⊥(s-s′) ds′
dz 2ιγ0k 0 0 . (4.24)

This equation yields a solution after Laplace transforming in s (to

deconvolve the integral). Performing the transformation gives

dχ(p,z) 1
= χ(p,z) W⊥(p)
dz 2ιγ0k 0 (4.25)

where the transformed variables are indicated by their p dependence

(rather than the often used tilde). The solution is immediate:

W⊥(p)z
χ(p,z) = χ(p,0) exp -ι
2γ0k 0 . (4.26)

Now, an inverse Laplace transform yields


+ι∞
χ(s,z) = 1 dp χ(p,0) e θ(p)
2πι -ι∞
(4.27)

where

W⊥(p)z
θ(p) = ps - ι
2γ0k 0 . (4.28)

A steepest descents calculation can now be employed to ascertain the

asymptotic nature of the transverse displacement. The exponent,


θ(p), is differentiated with respect to p and set equal to zero to find an
expression for the stationary point, p0:14

z dW⊥( p 0) = 0
* s-ι
2γ0k 0 dp . (4.29)

14
Then, the result is

χ( p 0,0) W⊥( p 0) z
χ(s,z) = exp p 0s - ι -δ
* 2π θ″( p 0) 2γ0k 0
(4.30)

with
2
θ″( p 0) = -ι z d W⊥( p 0)
2γ0k 0 dp 2 (4.31)

and δ=2arg[θ´´(p 0)], where prime indicates differentiation with respect

to p (i.e. ´≡d/dp). Recalling that


W⊥(p) =
∫ 0
ds e -ps W⊥(s)
, (4.32)

and the expression for ξ in terms of χ from Eq. (4.21), we have

assembled the general solution.

As a verification of this derivation, we consider a linear wake:


W⊥ (s)=W 0s (so W⊥ (p)=W 0/p2). For simplicity, we take z=zK(z). Solving

Eq. (4.29) yields


1
p 0 =  - ι W0K z  3
 γ0k 0 s
(4.33)

for the stationary point. Substituting this into Eq.(4.30) we may find
the exponential growth of χ (and, hence, of ξ):
1
χ(s,z) ~ exp 3 3  W0K  3 s2/3z1/3
4  γ0k 0  . (4.34)

We see at once that the growth length, for adiabatic variation of γ, k β

and K, is given by

1 = 3 3 W0I s2
* Lg 4 γ0k 0I A D (4.35)

15
where ξ~exp[(z/Lg) 1/3]. This is in exact agreement with the growth

length found in Eq. (4.9) of Section 4.1.c. Having established some

confidence in the general result, it is instructive and useful to examine

a more general wake.

We can make the assumption that the wake is an analytic function:


often the wake may be expressed as a sum: W⊥ (s)=Σan sn . In chapter 3

the “dominant mode” approximation was explicitly made. This

implies that one term of the wake will dominate the growth, and the
others may be neglected. We shall consider W⊥ (s)=W 0 n! sn-1 so that

W⊥ (p)=W 0/pn . Then, from Eq. (4.29) we have that

1
p 0 = -ι n W0 z n+1
2γ0k 0s (4.36)

which we write as p 0~exp(ιφ) η1/(n+1) where η=nW 0z/(2γ0k 0 s) and φ

satisfies the relation exp[i(n+1)φ]=-ι. Then, the exponent can be

expressed as
n/(n+1) 1/(n+1)
θ = A η1/(n+1) s z (4.37)

where A=exp[ιφ](n+1)/n. In other words, the growth length is given

by

1 = W0 sn(cos φ)n+1 (n+1)


n+1

* Lg 2γ0k 0 nn
, (4.38)

and ξ~Exp[(z/Lg) 1/(n+1)]. This case reduces to the linear wake for n=2.

It is straightforward to calculate the other factors involved in the


expression for ξ; however, the exponential behavior of the beam’s

transverse displacement is the primary concern.

16
Having obtained a rather general solution to the BBU equation, we

proceed with analyzing methods to control BBU.

17
References and Notes
(Chapter 4)

1
There are other effects (namely resistive wall instabilities) which can lead to beam blow-up
even in slightly resistive pipes. See P. L. Morton, V. K. Neil and A. M. Sessler, “Wake Fields
of a Pulse of Charge Moving in a Highly Conducting Pipe of Circular Cross Section,” Journal of
Applied Physics 37, 10 (1966).
2
Amplitude growth could occur if kβ of W⊥ had a z dependence.
3
For consistency, one should define W⊥ =W 0(s/swake ) so that W 0 can have the same units as
the wake. However, here we have absorbed swake into W 0. Thus, W0 has units of (cm-sec2) -1 .
The factor W0/swake represents the slope of the linear wake.
4
See K. L. F. Bane, "Wakefield Effects in a Linear Collider," SLAC Pub No. 4169, December,
1986.
5
A. W. Chao, B. Richter and C. Yao, “Beam Emittance Growth Caused by Transverse Deflecting
Fields in a Linear Accelerator” (Amsterdam, Holland: North-Holland Physics Publishing),
Nuclear Instruments & Methods in Physics Research, 178 (1980).
6
Note that although γ has been assumed to be a constant, kβ may vary with z. This is true
because the focusing B-field can have a z dependence.
7
It is possible to consider a general current distribution; however, the Laplace transform of the
product Iξ would then have to be involved.
8
A constant current along the accelerator is not a bad assumption since the current would be
primarily lost due to BBU. Thus, until the beam growth was large, the current profile along the
accelerator would be nearly constant. The assumption that the acceleration cells are equally
spaced, and hence that D(z) is a constant, is also a reasonable one. Most accelerators have
equal spacing for at least groups of cells. Finally, a constant focusing field (for a constant
energy) might also be reasonable for sections of an accelerator.
9
K. A. Thompson and R. D. Ruth, “Controlling Multibunch Beam Breakup in TeV Linear
Colliders,” SLAC-Pub-4537 (1989).
10
The straight lines appear on this graph due to an insufficient number of sampling points on the
plot. The numerical calculation was done with a sufficient number of points, simply the plot
was not.
11
Consider a quadrupole with peak field of 10kG. This is equivalent to about 0.4 J/cm3. For a
typical magnet, this might amount to a few Joules; a manageable amount of energy. However, if
we require a 3% variation of the field during a 10 psec pulse, then we have a peak power of
36MW/cm3. This is an enormous amount of power. Conventional magnets could not vary the
field on such a short time scale. However, it might be possible to achieve this with RF
quadrupoles: cavities excited to produce quadrupole fields.
12
See Ref. 4. The simplified SLC is based on the Stanford Linear Accelerator at SLAC. The
simplified ATA is based on the Advanced Test Accelerator at LLNL. The former is an RF
accelerator, the latter is a linear induction accelerator.
13
Indeed, putting a spread in energy along the bunch for the SLC has been proposed in several
papers as part of a cure to BBU (see Ref. 9). On the other hand, more extensive numerical
simulations indicate that this would be of little help in the ATA. See Chapter 6 for more
details.
14
There may, of course, be several stationary points. Here it is assumed that there exists a
single one which dominates any others which might exist. This is identical to the “dominant
mode” assumption used in chapter 3 to derive the BBU equation. A “spike” in W ⊥ (p)
corresponds to this mode.

18
Chapter 5
The BNS Effect

An effect first described by Balakin,


Novokhatsky, and Smirnov is discussed. Its
use as a method to damp BBU is explored
analytically.

5.1 An Explanation of the BNS Effect.

The results from the two-particle model of Chapter 4 suggest a

mechanism for controlling BBU. As was demonstrated, a spread in

energy between the first particle (the head of the beam) and the

second particle (the tail of the beam) negated BBU. In this chapter we

extend this concept to a many-particle beam. This technique is

attributed to Balakin, Novokhatsky, and Smirnov (hence the acronym

“BNS”) who first proposed it for the VLEPP accelerator (located at

Novosibirsk, USSR).1 Recently, BNS and Landau damping have been

recognized as physically distinct effects; however, earlier works often

used the term Landau damping to refer to both phenomenon.

1
A spread in energy along the beam produces a spread in betatron

wavelengths (or, equivalently, frequencies). This distribution of

wavelengths implies that a resonance leading to BBU is less likely to

occur. More accurately, the distribution will reduce the amplitude of

transverse oscillations: the fields set up by lead particles will couple

less effectively to trailing particles whose betatron wavelengths differ.

Different wavelength for each particle will reduce the coupling along

the entire beam. This is precisely what can be exploited to counteract

BBU. In fact, BBU “relies” on a resonant effect to disrupt the beam.

In the two-particle model BBU was entirely negated by the appropriate

choice of the second particle’s energy. This can only be approximately

done for a beam because there is no one appropriate energy. It seems

clear, though, that certain energy spreads will minimize BBU.2 We

seek to uncover these next.

5.2 Analytic BNS Derivation.3

For a coasting beam the Lorentz factor, γ, is assumed constant in z, and

the BBU equation may be written as


s
d 2ξ(s,z)
+ k 2β(z) ξ(s,z) = k 2β M(z) ξ(s′,z) W⊥(s-s′) ds′
dz 2
0 (5.1)

where M(z)=(z)I(z)/[I A γD(z)k β2].5 A spread in energy (or, equivalently,

in γ) along the beam can be treated by a complementary spread in kβ.

2
Namely, k β=k 0+∆k(s/smax ) models a small (∆k<<k 0) linear spread in

the beam’s energy. As before, a WKB approximation is used with

ξ(s,z) = Re χ(s,z) exp (ιk β(s)z ) (5.2)

resulting in a relation for χ in the strong focusing limit:


s

=
M(z)k 0 ⌠ s-s′
χ(s′,z) W⊥(s-s′) exp -ι∆kz s ds′
dz 2ι
⌡ 0
max
. (5.3)

A Laplace transform in s yields

dχ(p,z)
χ(p,z) W⊥ p + ι s∆kz 
M(z)k 0
=
dz 2ι max . (5.4)

This equation is easily integrated. With the initial condition that


χ(s,0)=1, 6 an inverse Laplace transform results in


+ι∞
χ(s,z) = dp 1 θ(p)
p e
-ι∞
(5.5)

where the exponent is given by


z

θ(p) = ps +
M(z)k 0 ⌠  ∆kz′ 
W⊥ p + ι s dz′
max

⌡ 0 . (5.6)

In order to solve this integral, we perform an expansion of θ. Then,

the stationary point (i.e. where θ′(p)=0) for ∆k=0 is obtained. Next,

the “shifted” stationary point, for ∆k≠0 is calculated. And finally, the

integral is calculated using steepest descents.

The expansion is performed in a parameter ε=ι∆k/smax which is

assumed small. In other words, it is required that |p|>>|εz|. Then,

3
M(z)k 0 ′ ″
z W⊥(p) +W⊥(p) εz +W⊥(p) ε z +…
2 2
θ(p) = ps +
2ι 2! 3! , (5.7)

where prime denotes differentiation with respect to p (i.e. ´≡d/dp).


Let p 0 denote the stationary point for ∆k=0 (i.e. θ′(p 0)=0). Then, p 0

obeys

M(z)k 0 ′
s+ zW⊥( p 0) = 0
2ι . (5.8)

It is now possible to write the shifted stationary point, ps, in terms of

p 0 and an expansion in ε. The result is that

(3)
W ( p 0) ε2z2
p s = p 0 - εz - ⊥ +…
2! ″ 4!
W⊥( p 0) . (5.9)

(The notation W ⊥ (n) is used to denote the nth derivative with respect
to p for n>2). The first three terms in the expansion of p s are

sufficient as long as |εzW⊥ (3)/W⊥ ″|<<1. Combining Eq. (5.7) for the

exponent, θ, with the above expression for the shifted stationary point

yields

k 0M(z) ′ ′ ″
z W⊥( p 0)-W⊥( p 0) p 0+ εz W⊥( p 0)+ ε z W⊥( p 0)+…
2 2
θ( p s) =
2ι 2! 4! .
(5.10)

Now it is possible to see that p0 should be the root of Eq. (5.8) which

maximizes the imaginary part of the term of zero order in ε in the


above equation (i.e. Im[W⊥ (p 0)-W⊥ ′(p 0)p 0] is to be maximized). Finally,

performing a steepest descents calculation on Eq. (5.5) results in

χ(s,z) = 1 1 θ + ιδ
ps e
2π θ″ , (5.11)

where δ=2arg(θ″(p s) and

4

 W(3)( p 0) 2
 2 2
k 0M(z)
+ W⊥ ( p 0)  ε z +…
(4)
θ″( p s) = z W⊥( p 0) -  ⊥
2ι ″ 4!
 W⊥( p 0)  . (5.12)

The above equations provide a complete recipe for calculating

transverse beam growth including BNS effects. The results are valid

asymptotically (i.e. for large z) and assume strong focusing holds.

As in Chapter 4, we can consider a particular wake. Before doing this,

it is possible to make some general comments on the BNS effect.

Since only the asymptotic growth and damping are of interest here,
the expression for θ, Eq. (5.10), is of primary concern. Specifically, if
Im[W ⊥ ″(p 0)]>0 then there will be exponential damping (for large z). 7

Notice that exponential damping can happen with either a positive or


a negative spread in energy. However, for Im(p0) ∆k>0 there will be a

region (in z) in which there can still be (algebraic) growth.


Nevertheless, a sufficiently large spread in kβ can ensure damping.

With these observations in mind, we consider W⊥ (p)=W 0/pn as was

done in Chapter 3. Then, the unshifted stationary point satisfies


1
nM(z)W0 n+1
p0 = k0 z
2ι s . (5.13)

Let p 0~exp(ιφ) η1/(n+1) where η=nk 0K(z)W 0z/(2s) and φ satisfies the

relation exp[i(n+1)φ]=-ι.8 Equation (5.10) gives the expression for the

exponent:

θ = n+1 1 n+1 s 2 2
n p 0s - 2! εzs + 4! p 0 ε z +…, (5.14)

where ε=ι∆k/smax as before. Substituting for p0, ε and grouping terms

yields

5
2
Re(θ) = n+1 1/(n+1)s 1 - n  ∆kz  η-2/(n+1) +…
n (cos φ) η 4!  smax . (5.15)

For ∆k=0, this reduces to

Re(θ) = n+1
n (cos φ) nk 0M(z)W0z
1/(n+1) n/(n+1)
s . (5.16)

Thus, ξ~exp[(z/Lg) 1/(n+1)] with the growth length equivalent to that

found in chapter 4 (for constant γ). For ∆k≠0, we desire the damping

length: the distance over which the transverse displacement


decreases by one e-fold. Thus we seek LBNS where ξ~exp[(-z/LBNS ) m ]

and m is some rational number. For large z, the exponent, θ from Eq.

(5.15), contains a term


2 1/(n+1) n+2 2n+1
Re(θ D) = - n+1 (cos φ) s∆k 2 s n+1 z n+1
4! max nk 0M(z)W0 . (5.17)

Setting m=(2n+1)/(n+1) and rearranging terms, we have an


expression for LBNS in terms of Lg:

2n+2
1  (n+1) ∆ks (cos φ)  2n+1
* LBNS
= L1/(2n+1)
g 
 (4!n) 1/2 smax

 . (5.18)

5.3 Landau Damping.

Landau damping refers to an effect caused by a spread in energy within

each beam cross-section (beam “slice”) which can be used to

counteract BBU. It is distinct from BNS damping although the two are

often confused in the literature. Landau damping is a well known

phenomenon in plasma physics.9 The effect was first predicted and

derived theoretically by L. Landau in 1946.10 Experimental

6
confirmation in plasmas came much later. The phenomenon often

reduces growth due to BBU because beam particles have a distribution

of energy.

Analytic solutions with Landau damping are not considered here, and

are difficult to find in the literature. Numerical analysis is also difficult

due to the many (~30) particles per slice needed to obtain valid

statistics. The code described in the next chapter is capable of

performing there calculations; however, testing has not been

completed at this time.

7
References and Notes
(Chapter 5)

1
V.E. Balakin, et a l . , “VLEPP: Transverse Beam Dynamics,” Proceedings of the 12th
International Conference on High-Energy Accelerators, ed. by F.T. Cole and R. Donaldson
(Batavia, Illinois: Fermi National Accelerator Laboratory, 1984), pp. 119-120.
2
Because only certain energy spreads are effective (as will be shown in Section 5.2), BNS
damping is often called the BNS effect. In other words, it does not always lead to damping
(this was made clear in the authors’ original work).
3
This derivation follows one kindly furnished by D. H. Whittum, “BNS Damping of
Instabilities of a Coasting Beam,” (Not yet published).
4
This derivation follows one kindly furnished by D. H. Whittum, “BNS Damping of
Instabilities of a Coasting Beam,” (Not yet published).
5
Note, the factor K(z) used in previous chapters is related to M(z) by M(z)=kβ2(z)K(z). M(z) is
used in this chapter for algebraic reasons.
6
This initial condition amounts to assuming that the entire beam is injected off axis by one unit.
For other injection profiles, χ(s,z=0),

7
χ(p,0) =
∫ 0
ds χ(s,0) e ps
.
There is an additional condition for exponential damping. Namely, that the term of second
order in ε be of a higher power in z than the term of zero order in ε.

8
Equation (5.9) gives the expression for the shifted stationary point:
2
p s = e ιφη1/(n+1) - ι ∆kz -  s∆kz  n+2 e -ιφη-1/(n+1) + …
2smax max 4! .
9
For example, see F. F. Chen, Plasma Physics and Controlled Fusion (Plenum Press: New York,
1984), Sec. 7.4-7.5.
10
L. Landau, “On the Vibrations of the Electronic Plasma,” J. Phys. U.S.S.R. 10, 25 (1946).

8
Chapter 6
Numerical Analysis of BBU

This chapter describes a program which


numerically solves the BBU equation due to an
arbitrary transverse wakefield. The methods
used to arrive at a solution to the BBU equation
are explored. Finally, two applications of the
code are discused.

6.1 The BBUNS Code.

The previous chapters have dealt with analytic solutions to the BBU

equation. In all cases, some assumptions about the beam parameters

or wake function had to be made to yield tractable solutions. An

alternative approach to solving the BBU equation is by numerical

methods. The advantages are clear: few assumptions need be made.

A computer program can solve for BBU dynamics with realistic current

and energy variations. The disadvantages are that several parameter

sets must be solved if any pattern is to be found.

To allow for the solution of BBU dynamics under realistic parameters,

a program has been designed and written to numerically calculate the

1
BBU dynamics of a beam.1 The Beam Break-Up Numerical Simulator

(BBUNS) can accommodate effects such as Landau and BNS damping,

as well as realistic current variations along the macro-pulse. The wake

must be pre-specified either analytically or numerically. Often, a field

solver is employed to numerically calculate the wakefields in a realistic

structure. 2 The program then integrates the BBU equation at discrete

steps along the beam and along the accelerator. The results are

presented in tables and graphs. Parameters such as growth length and

maximum displacement are also calculated.

BBUNS allows each of the parameters in the BBU equation to vary in s

and z. Thus, variation of energy along the pulse may be used to study

BNS damping. Similarly, variation in the current along the beam can

model pre-bunching of the beam. In addition, each parameter carries

a macro-particle index, i (see Figure 6.1). In this way, energy variation

within a slice may be used to study Landau damping.

nth beam slice

ith macro-particle

Figure 6.1. A sample beam cross-section, or “slice”,


indicating the distribution of macro-particles within a
slice.

In order to solve the equation for an arbitrary wakefield, it is necessary

to divide the beam into slices. These slices, or macro-particles, are

the mesh of the s dimension. A mesh is also applied to the z

2
dimension. For simplicity, the program does not employ adaptive step

sizes since this would require a great deal of interpolation.

BBUNS begins by executing the main loops required to solve the

diffeo-integral equation at all the points along the beam mesh and

accelerator mesh. There are two main steps necessary to accomplish

this. The first step is computing the integral,


s
I(s′)
H(s) = ds′ W⊥(s′-s)ξ(s′,z)
IA
0 . (6.1)

where ξ is averaged over all the particles within a slice,


N
ξ(s′,z) = ∑ ξi(s′,z)
i=1 , (6.2)

to obtain the force on the beam slice centroid at s. This integration is

performed using a Gaussian quadrature algorithm.

The second step requires solving the second order differential

equation,

d 2ξ(s,z) e H(s) dξ(s,z) dlnγ(s,z)


= - k β2ξ(s,z) -
dz 2
mc γ(s,z) D(z)
2 dz dz (6.3)

or equivalently, the two first order equations,

dξ df(s,z) d 2ξ
f(s,z) = g(s,z) = =
dz dz dz2 . (6.4)

This is accomplished by use of a fourth order Runge-Kutta algorithm.

The Gaussian integration scheme considers a function ƒ to be

integrated on [a,b]. Gauss’ formula (without remainder) is,

3
b n
ƒ(y) dy = b-a
2
∑ w iƒ(yi)
a i=1 (6.5)

with,

yi = b-a xi + b+a
2 2 (6.6)

where x i is the ith root of the Legendre polynomial, Pn (x), and,

2
wi = 2 P ′n( xi)
2
(1-xi ) (6.7)

is called the ith “weight” 3. The algorithm can be of arbitrary, nth,

order. The implementation of this method is separated into three

parts. The first algorithm calculates the roots and weights. The

second interpolates the function, ƒ, between [a,b] for the n sample

points necessary. The interpolation is performed using a four-point

Lagrangian method.4 In order to solve for H(s), three functions must


be interpolated separately: I(s), ξ(s,z) and W ⊥ (s). Finally, the third

part performs the sum and returns the value of the integral.

The Runge-Kutta algorithm for advance in z is,5,6

fn+1 = fn + 1 [k 1 + 2(k 2) + 2(k 3) + k 4]


6
g n+1 = g n + 1 [l 1 + 2(l 2) + 2(l 3) + l 4]
6 (6.8)

with the coefficients given by

k 1 = hfn(x,y,z) l 1 = hg n(x,y,z)

k 2 = hfn(x+ h ,y+ k 1,z+ l 1) l 2 = hg n(x+ h ,y+ k 1,z+ l 1)


2 2 2 2 2 2
k 3 = hfn(x+ h ,y+ k 2,z+ l 2) l 3 = hfn(x+ h ,y+ k 2,z+ l 2)
2 2 2 2 2 2
k 4 = hfn(x+h,y+k 3,z+l 3) l 4 = hfn(x+h,y+k 3,z+l 3)
(6.9)

4
and where h is a mesh size (in this case ∆z).

This method solves for the (n+1)th step (in z), given the nth step.

The following formulas are the two first order equations obtained from

the second order equation.


dξn
fn =
dz

g ′n = dfn =
dz
e H(s) dξ(s,z) dlnγ(s,z)
-k β
2
γ(s,z)ξ(s,z) -
mc 2γ(s,z) D(z) dz dz . (6.10)

Computation time is a concern when solving problems numerically.

Beyond the inherent efficiency of the Runge-Kutta and Gaussian

quadrature algorithms there are a number of features of the code

which were designed to minimize run time. The first feature allows

the user to specify how far down the beam the effects of the wakefield

will be calculated. This truncation length allows a user to include the

physics while excluding much of the numerically intensive

calculations. This is especially useful for long wakefields which decay

rapidly or have a high frequency, low amplitude “ring” 7. By

appropriately setting the truncation length, the user can have the code

ignore the small contributions from the “tail” of the wakefield.8 This

enables the code to run much faster when the wakefield data contains

more points than necessary for numerical accuracy. In the event that

the user does not want to shorten the effect of the wakefield, the

truncation length simply becomes the lesser of either the length to

which the wakefield is defined or the pulse length. Figure 6.2 shows

the s and z coordinates along with the truncation length.

5
nth Effective Beam
macro-particle Length of Head
ξ Force
Truncation Length

Figure 6.2. In addition to the s and z coordinates


(previously described), there is a truncation length which
limits the effective length over which the wakefield
induced forces act.

The most time saving feature of the code is also of physical

significance. The integral, H(s), represents a driving force. This

driving force is only exerted on the beam slice while in, or very near,

the accelerating gap.9 On the scale of the accelerating structure the

gap spacing is very small (see Figure 6.3). This enables, even requires,

that the driving force only be applied to the beam for a very short

distance in the structure. 10 This physical requirement is fortunate

from a numerical stand point. It requires the code to calculate the

integral only a few times per slice for each accelerating gap. 11 This

discrete kick method reduces the code execution time and yields a

more physical result.

6
Gap

Beam Slice

∆z
Typical

Figure 6.3. The typical spacing (mesh) in z may be larger


than the gap size of LIA’s. A discrete kick is applied to the
beam only when the beam is within a ∆z of the gap. This
produces a more physical result and saves computing time.

6.2 Examples.

To illustrate the usefulness and power of the numerical simulation, we

consider two examples. The first is linear induction accelerator in

which we will study the effect of the beam current on the growth

length. The second example is of a future acceleration concept in

which a periodic energy variation along the accelerator is simulated.

The problem would be quite complicated to solve analytically, but

numerically poses no difficulty.

7
6.2.a The Advanced Test Accelerator.

The advanced test accelerator (ATA) was built at the Lawrence

Livermore National Lab in the early 1980’s.12 It was based on an

earlier induction accelerator, the experimental test accelerator (ETA).

The design called for a long (~100 meter), high current (~10kA)

induction accelerator reaching a final beam energy of ~50 MeV with a

pulse length ~40-70 ns (see Table 6.1). From the onset, the machine

was plagued with problems. Simply propagating the high current

beam through the accelerator proved impossible. The problem of BBU

was not unknown to the designers; however, it was thought that

proper corrections were made to compensate for the large wakefields

measured in the ETA cells.13

Table 6.1. The ATA and ETA nominal parameters. The


ATA represented a “technological leap” from previous
induction accelerators.

Parameter ETA ATA


Beam Energy (MeV) 4.5 50
Pulse Length (ns) 40 40-70
Magnetic Focusing (kG) 0.8kG (max) 2-3 (max)
Number of cells 8 ~190
Current (kA) 10 10

Once operation of the accelerator began, it was found that BBU was

responsible for ~11 e-fold increase in the beam’s transverse size. At

the time, a solenoidal magnetic focusing system with a 3 kG field was

used. This growth was unacceptably large (at the desired 10 kA beam

current). Reducing the beam current could reduce the BBU

substantially, but then the usefulness of the ATA would be lessened.

8
To confirm the effect of current reduction on BBU growth, analytic

and numerical models can be utilized. A simulation was run with

BBUNS to determine the growth length for beam currents ranging

from 3 to 10 kA. The parameters for the runs are those given in Table

6.1 with the exception of the beam current. In reality, the ATA has a

pulse length ~50 ns with only ~30-40 ns of the pulse in the “flat top”.

In other words, the pulse is not nearly square. To model this in

BBUNS, the following current profile was used:

I(z) = I 0(1 - 0.5 e -s/r) (6.11)

where r=5 ns is the rise time and smax =40 ns.14 No current variations

were used along the accelerator (i.e. pulse shortening due to BBU was

not taken into account). Further, the magnetic focusing was assumed

to be a constant 2 kG where as the ATA actually uses a field which is

ramped up during the first several cells to its maximum value. It was

assumed that the entire beam was displaced by one unit from the axis.

The extent of BBU growth in the ATA is attributable to the wakefields

of its induction cells. To calculate the wake the AMOS code was

used.15 The design of the cells, including material specifications, was

entered into the code, and the field solver returned the wake shown

in Figure 6.4. The large value of the main peak is responsible for most

of the BBU.16

9
The Realistic ATA Wake
12
x10

1.5

1.0
V/C/m/m

0.5

0.0

-0.5
0 1 2 3 4 5 6 7 8 9
s [meters]

Figure 6.4. The wake in an induction cell of the ATA was


calculated using the AMOS codes. Realistic boundary
conditions are taken into account. Note the long, low
amplitude “ringing” which persists for ~20 meters (not
shown).

The wake was then used by BBUNS to obtain the results summarized

in Table 6.2.

10
Table 6.2. ATA growth parameters for various beam
currents. The maximum displacement is normalized to
the initial beam offset.

Current (kA) Lg e-folds Maximum


3 133 <1 <2
4 25 3.1 23
5 13 6.1 444
6 9 8.8 6621
7 7 11.5 94,000
8 5.7 13.9 1 x 106
9 4.9 16.3 1.2 x 107
10 4.3 18.5 1.2 x 108

A clearer picture emerges if we plot the reciprocal of the growth

length versus the current (see Figure 6.5). As predicted by the

formulas of Chapter 4, the relation is nearly linear.

Inverse Growth Length vs. Current

0.20
1/Lg (1/meters)

0.15

0.10

0.05

3 4 5 6 7 8 9 10
Current (kA)

Figure 6.5. The nearly linear relations between current


and (Lg) -1 is clearly exhibited in this plot.

Clearly, the growth due to BBU is substantial in the ATA at all but the

lowest currents. A method of controlling this growth was devised by

11
the Livermore Beam Research Group: laser guiding.17 The method was

effective at controlling BBU growth but introduced other problems.

Unfortunately, the funding for the ATA has been cut so that

experimental and numerical comparisons will no longer be possible.

6.2.b The Two-Beam Accelerator.

The two-beam accelerator (TBA) is a “next generation” machine

designed to achieve TeV energies.18,19 Figure 6.6 is an artist’s

conception of one possible configuration of a TBA. In this

arrangement, called the free electron laser (FEL) TBA, energy from a

low power, high current beam, the drive beam, is extracted to drive a

high power, lower current beam. A small fraction (~1%) of the drive

beam’s energy is extracted in each FEL section. The energy is

replenished by induction cells, and the process repeats.

Figure 6.6. An artist’s drawing of the TBA-FEL.

The high gradient structure is designed to achieve gradients of

hundreds of MeV per meter. Thus, a TeV accelerator would only be a

few kilometers long. Many issues regarding the feasibility of the TBA

12
have been discussed, yet several issues remain unresolved. One such

issue is beam break-up in the drive beam.

Since the TBA would have to be many times the length of current

designs, BBU is an important factor. Additionally, the FEL’s require a

low emittance to maintain the required gain. Thus, the FEL-TBA is

more sensitive to effects from BBU than present designs. Previous

work has shown a growth length of ~100 meters for present

parameters (see Table 6.3). In fact, with improved LIA modules and

more liberal estimates, the growth length may be many times greater

than this.

Table 6.3. A set of TBA parameters used for BBU


evaluation by BBUNS. Note (*), the module spacing is
incorporated in the D factor (i.e. D=δ(z-*)).

Parameter Symbol Value


Current (DC) I 1.5 kA
Lorentz factor (average) γ 27.4
Pulse length smax 50 nS
Module spacing (*) 100 m
Betatron wavelength λβ 150 cm

Whereas previous simulations have neglected the energy variations

(due to extraction and re-acceleration),20 the present analysis includes

the change. Specifically, a square wave variation with a nominal value


of γ0=27.4 was used to model the energy profile. (In reality, the energy

varies differently; however, the magnitude of the fluctuations should

suffice to model the effects on BBU.) To uncover any effects this could
have on BBU, a large change, ∆γ=3, was used (i.e. γmax =28.9). Further,

the wake of the ATA cell (Section 6.2.a) was used. The results were as

13
follows. With a constant energy profile the growth length exceeded

1.2 x 10 5 meters. The growth length for the square wave profile was

found to be 160 meters: a drastic decrease in the growth length. This

seems to indicate that abrupt energy fluctuations can induce a fair

amount of BBU. Fortunately, the 160 meter growth length is

substantially larger than the 100 meter sections being proposed.21

14
6.3 Figures.

On the next few pages are the results from BBUNS calculations of

beam displacement (normalized to the initial displacement) near the

end of the ATA at various current values. The current profile is

featured below:

Figure 6.7. The charge density profile, ρ, is plotted versus


s. BBUNS uses the ρ rather than current. The two are
related by the speed of light: ρ=I/c. The vertical scale
varies with the maximum current.

Note that the scales on the vertical axes (displacement) change.

15
Figure 6.8. Current = 3 kA.

16
Figure 6.9. Current = 4 kA.

Figure 6.10. Current = 5 kA.

Figure 6.11. Current = 6 kA.

Figure 6.12. Current = 7 kA.

Figure 6.13. Current = 8 kA.

17
Figure 6.14. Current = 9 kA.

Figure 6.15. Current = 10 kA.

Figure 6.16. The maximum beam displacement is plotted


along the accelerator for 5 kA. Note the two distinct
growth regimes. The first is the region of linear growth,
the second is the rapid, exponential growth. The onset of
exponential growth indicates that the instability has
established a resonant condition.

18
References and Notes
(Chapter 6)

1
For a complete description of the code see G. A. Travish, “The Beam Break-Up Numerical
Simulator,” Lawrence Berkeley Lab LBL-28029 (1989).
2
. DeFord, G. Craig, R. McLeod, “The AMOS (Azimuthal Mode Simulation) Code,” IEEE
Particle Accelerator Conference Proceedings (Chicago, Illinois, 1989).
3
M. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions (Toronto, Ontario:
Dover, 1972), p.887.
4
Ibid., p.878
5
Ibid., p.897
6
Note: the prime indicates differentiation with respect to z.
7
G. Craig and J. Deford, “Diagonal Resonances and Trapped TE Modes,” Lawrence Livermore
National Lab Beam Research Program Research Memo 88-44 (Livermore, California, 1988).
8
It has been shown that at least in the ATA wakefield, the “trapped” ringing mode does not
contribute significantly to beam break-up. See D.H. Whittum,, “Beam Break-Up in the Two
Beam Accelerator,” IEEE Particle Accelerator Conference Proceedings (Chicago, Illinois, 1989).
9
See note 18.
10
It is possible to consider the force as averaged over the entire structure. However, this is less
physical and much more CPU intensive. Yet, averaging is usually done for analytic work. For
an exception to this, see Ref. 3.
11
Actually, most LIA gaps are so much smaller than the distances between the successive gaps,
that the kick need only be applied once: typical gap size < 3cm, typical cell spacing ≈50cm.
12
“Generating Intense Electron Beams for Military Applications,” Energy & Technology Review
(Lawrence Livermore National Lab: 1981).
13
G. J. Caporaso, e t a l . , “Beam Breakup (BBU) Instability Experiments on the Experimental
Test Accelerator (ETA) and Predictions for the Advanced Test Accelerator (ATA),” IEEE
Transactions on Nuclear Science NS-30 4 (1983).
14
B. B. Godfrey, “Computer Code Benchmark for the Beam Breakup Instability,” Mission
Research Corporation MRC/ABQ-N-410 (1989).
15
See Ref. 2.
16
Several BBUNS calculations have shown that the wake can be truncated to ~2.5 meters
without severely affecting the extent of beam growth.
17
G. J. Caporoso, et al., “Laser Guiding of Electron Beams in the Advanced Test Accelerator,”
Physical Review Letters 57 13 (1986).
18
A. M. Sessler, “New Particle Acceleration Techniques,” Physics Today (January, 1988).
19
D. B. Hopkins and A. M. Sessler, “Status of LBL/LLNL FEL Research for Two Beam
Accelerator Applications,” Lawrence Berkeley Lab LBL-26253 (1988).
20
D. H. Whittum, et al, “Beam Break-Up in the Two Beam Accelerator,” op. cit.
21
Many other considerations require that the TBA not be longer than 100 meters before a new
low-energy beam be used. See A. M. Sessler, D. H. Whittum and J. S. Wurtele, “A Study of
Phase Control in the FEL Two-Beam Accelerator,” LBL-27765 (1989).

19
Chapter 7
Conclusion

A summary of the work is presented and


possible future topics are suggested.

In the pursuit of higher energy, higher luminosity, and lower

emittance beams, the quality of the beam has become paramount.

Even small BBU growth is unwelcome since it increases the beam’s

emittance. An increased understanding of BBU enables solutions to

this instability to be devised. This thesis has presented a primarily

pedagogical explanation of beam break-up.

A 1D theory for the problem of transverse beam break-up in linear

electron accelerators has been formulated. The problems caused by

this instability and various analytic solutions to the BBU equation have

been briefly discussed. Most notably, an expression for the growth

length due to a general wake has been given in Chapter 4.

BBU has also been investigated using the two particle model. The

simplicity of this model allows for an intuitive understanding of BBU.

1
Fortunately, the model is sufficiently accurate that it physically

applicable. Using this model, concepts for damping BBU were

explored.

One method to counter growth due to BBU was discussed in detail in

Chapter 5. The BNS effect holds promise in future accelerators where

a spread in energy along the beam can be incorporated into the

design.

The analytic results derived in this work are useful under idealized

beam and accelerator parameters. For a more realistic model,

numerical solutions are required. The BBUNS code described in

Chapter 6 allows for the solution of the BBU equation for an arbitrary

wake and with realistic parameters. The two examples considered in

this work require additional investigation. For the ATA, Landau

damping needs to be taken into account. Further, the laser guiding

technique could be simulated by BBUNS. Additionally, the ATA has a

more complex geometry than has been used in past numerical studies:

the LIA modules are clustered into groups of nine and ten cells each.

The TBA example also needs to be explored more carefully. The

energy variations due to periodic extractions and reaccelerations must

be modeled realistically: a square wave is a crude approximation.

2
Appendix 1
Physical Interpretation of the Impulse

A physical interpretation is given of the effect


of the impulse imbued to a particle traversing
an excited cavity. A force equation, similar to
that given in chapter 3, is derived.

There is a simple manner in which to view the effect the impulse ∆p⊥

imposes on a particle traversing a cavity. Consider the following

model of an accelerator: a series of N cavities each of width L and

separated by a distance D from one another. (See Figure A1.1).

L D

••• n-1 n n+1 •••

Figure A1.1. A simple model of an accelerator: a series of


cavities, each length L, equally spaced, a distance D apart.

In going from the nth to the (n+1)th cavity a particle receives a


momentum impulse, |∆p⊥ |=p n+1-p n , which results in a displacement,

1
ξ, of the particle. In the highly relativistic limit, the ratio of the

transverse to the longitudinal energy yields

p n+1 - p n ξn+1 - ξn
=
mγ D (A1.1)

where D is in units of time. Then the net displacement of the particle

is simply

∆p n→n+1
∆ξn→n+1 = D
mγ . (A1.2)

We can now approximate this relation by a differential equation,

d γ dξ = D dp = D ∆p
m dn m ⊥
dn dn . (A1.3)

For pedagogical reasons we expand this equation to obtain

d 2ξ dlnγ dξ
= D ∆p ⊥ -
dn 2 mγ dn dn . (A1.4)

This formulation is rather direct; however, it requires interpreting the

cavity index, n, as a continuous variable. The more common

formulation simply solves F=ma and uses time as the continuous

dependent variable. This is done in Chapter 3.

2
Appendix 2
The Panofsky Derivation

The Panofsky derivation of the BBU equation is


given.

We begin by identifying the “field integral”, F(t), as

F(t) = ⌠ ∂E z(t,z)
dz
⌡ 0
∂ξ
, (A2.1)

where ξ is the 1D transverse displacement of the beam from the beam

pipe axis. Notice that F(t)=ι( ωc/e)∆p ⊥ . (Do not confuse the field

integral, F(t), with force, F, as used in the previous section.) Thus, in

obtaining an expression for F(t), we will be able to linearize Eq. 3.33.

We solve for F(t) by considering the relation


∂U
= ( J ⋅ E) dV - ωU
∂t Q
Cavity (A2.2)

which describes the rate of energy change in a cavity excited at a


frequency ω (near resonance) due to 1) the flow of energy into the

1
cavity from a beam of current density J and 2) the loss of energy in

the cavity walls from Ohmic heating. The integral is averaged to obtain

a mean energy gain rate. The quality, Q, of the cavity has the standard

definition:

Stored enegry
* Q =ω0
Power loss , (A2.3)

where ω0 is the resonant frequency of the cavity and the quantities are

assumed to be time averaged. (Note: Q is dimensionless.) Next, note

that for an axial symmetric field (with the field vanishing on axis) the

following approximation is valid:

⌠ ∂E
∫ E ⋅ dz ~ ξ
⌡ ∂ξ
z
dz = ξF(t)
. (A2.4)

In other words, the rate of energy flow into the cavity is proportional

to the beam’s transverse displacement off axis. Substituting the result

of Eq. (A2.4) into Eq. (A2.2) we obtain,

∂U
+ ωU = I ξF(t)
∂t Q
(A2.5)

where I is the beam current (assumed constant). The energy, U, is

typically related quadratically to the field integral: U=KF(t)2.1 After

some simplification, we obtain

∂F(t) Iξ
+ ω F(t) =
∂t 2Q 4K
(A2.6)

where an additional factor of 1/2 on the right hand side comes from
the phase relation between F(t) and ξ.2 A solution for F(t) is

2

t
F(t) = I ξ(t′) exp - ω (t-t′) dt′
4K 2Q
-∞
. (A2.7)

The current, I, was assumed constant and thus placed outside the

integral. This was done for purposes of clarity, and later such an

assumption will not be made. We can now substitute the field integral

into the equation of motion (Eq. 3.33) to obtain


t


d 2ξ dlnγ dξ
= e I 1 ξ(t′) exp - ω (t-t′) dt′ - ω2β ξ -
dt 2 ιωD 4K mγ 2Q dt dt
-∞
.(A2.8)

One final simplification can be made by defining C=e2/(ωD4Kmγ) and

J=I/e. Then C is a dimensionless parameter and J is the number of

electrons/unit length. And,3


t


d 2ξ dlnγ dξ
= CJ ξ(t′) exp - ω (t-t′) dt′ - ω2β ξ -
dt 2 ι 2Q dt dt
-∞
. (A2.9)

These equations are simply Eq. (3.33) with the field integral linearized
and written in terms of inherent cavity properties (Q,ω) and the beam

displacement. We see at once that the “driving” force, the integral

term, is proportional to the beam displacement. Thus, the further

displaced the forward sections of a beam are, the greater the force on

the rear sections. This is the mechanism by which an instability

propagates down and along the beam.

At this point one might be tempted to conclude that the problem is

well defined and a solution should be sought. This is nearly correct.

Our formulation of BBU requires improvement in two respects. The

first is the use of time as the dependent variable. Time may be a

3
natural choice in many problems, but it is an awkward coordinate in a

linear accelerator. Just what does a solution for the beam


displacement at a time t 1 mean? We can interpret such a solution to

mean that at a particular position, z, in the accelerator, a particular


part of the beam, s, has displacement ξ(t). Recall that time is related

to s and z by Eq. (3.1), so that

d = c d
dt dz (A2.10)

where the differentiation with respect to z is assumed to be at a fixed

s. Applying the above result along with Eq. (3.1) and Eq. (A2.9), the

following result is realized:


d 2ξ dlnγ dξ
= CJ2 ξ(s′) exp - ω (s-s′) ds′ - k 2β ξ -
dz2 ιc 2Q dz dz
s-s′≥0 (A2.11)

where k β=ωβ/c. Notice that the limits of integration are from s´=0 to

s´=s (s´<0 is physically meaningless). Now a solution, ξ(s,z), is quite

easy to interpret: it is the displacement of the beam slice a distance s

from the beam head, at a point z down the accelerator. We have now

corrected one of the shortcomings of our earlier BBU equation.

The second shortcoming to our model of BBU arises as a result of the

approximations used to obtain a solution for the field integral, F(t). To

see this inadequacy, consider a super-cavity with infinite Q. Under

this assumption, the integrand reduces to a simple integral over the


displacement, ξ. In other words, the field (or, the field integral)

would be a constant, say, after the beam left the cavity. But Maxwell’s

equations tell us that a static field cannot exist in an enclosed cavity

4
when the field is excited by a non-static source (the beam). In fact,

our model should include some oscillating term for the driving force.

To remedy this flaw in our model, we shall turn to our second

derivation. In addition to a better model of BBU, we shall acquire a

different perspective on the origins of the driving term.

5
References and Notes
(Appendix 2)

1
An estimate for K can be obtained as follows:

∫ ( ) πR L = 8L
2 4
U= 1 E 2maxdV ~ 1 F2 R R F
2 2
8π 8π L
where Emax is the maximum field (i.e. when the magnetic field is zero) and R and L are the
cavity radius and length, respectively. Then, K≈R 4/(8L).
2
See Panofsky’s note for details about the phase relation.
3
Notice that a simple change of variables also yields
t


d 2ξ C′J dlnγ dξ
= ξ(t′) exp - ω (t-t′) dt′ - D2k 2β ξ -
dn 2 ι 2Q dn dn
-∞

in accordance with Eq. (f2). Here C´ = e2D/(ω 4Kmc2γ ) and kβ=ω β/c.

6
Appendix 3
The Linear Wake Solution

A solution to the BBU equation is derived for


the case of a linear wake and adiabatically
varying γ and k β.

Consider a wake defined by W ⊥ (s)=W 0 s where W 0 is some constant. 1

Assume a constant current profile for the beam (i.e. a square charge
distribution). And, assume that the beam energy, γ, and betatron wave-

number, kβ, vary slowly over a betatron period (i.e. one λ β). Then, the

BBU equation reduces to


s
d 2ξ 1 dγ dξ
+ + k 2β ξ = - W0 I ξ (s-s′) ds′
dz 2 γ dz dz Dγ I A 0 (A3.1)

The solution which follows is exact in the case of a “coasting” beam


(i.e. γ(z)=γ0). The solution is obtained by first finding a “WKB”

approximation for the transverse displacement,


 
z
γ(0) k β(0)
ξ(s,z) = Re χ exp  ι k β(z′)dz′
γ(z) k β(z)  
0 (A3.2)

1
where χ is a complex variable. Working in complex variables, with the

understanding that the real part of the solutions is what we desire,


will be more efficient. The complex variable χ must satisfy the

following relation (ignoring terms α k β´´),

′  k′  2


s
γ′ kβ
+ χ 3   + χ′2ιk β = - W0 I
β
χ″ - χ′ + χ′ χ (s-s′) ds′
γ kβ 4  kβ  Dγ I A 0
(A3.3)

where k β´(0)=0. Further, by the WKB approximation, we may ignore

the higher order terms resulting in


s
dχ ι
= W0 I χ (s-s′) ds′
dz D I A 2γ(z)k β(z) 0 . (A3.4)

We can make a change of variables to eliminate the explicit


dependence on γ(z) and k β(z). Let
z

z= ⌠ γ(0) k β(0)
dz′
⌡ 0
γ(z′) k β(z′)
(A3.5)

so that

d = γ(0) k β(0) d
dz γ(z) k β(z) dz . (A3.6)

Our equation reduces to


s

= ιK χ(s′,z) (s-s′) ds′
dz 0 . (A3.7)

where

K = W0 I 1
D I A 2γ(0)k β(0) . (A3.8)

2
This is most easily solved by a power series. Using

(zs2)n
χ= ∑ an
n!
n=0 (A3.8)

we arrive at

∞ ∞


s
(zs2)n s2 = ιK a (z)n
∑ ∑ n 2n
an+1 s′ (s-s′) ds′
n=0 n! n=0 n! 0 . (A3.9)

In turn, this yield a recursive relation for the coefficients. Namely,

an+1 = ιK 1 - 1 an
2n+1 2n+2 . (A3.10)

Performing the recursion and simplifying yields,


n-1
an = (ιK ) n
∏ 4j2 + 16j + 2
j=1 (A3.11)

where a 0=1. The finite product is simply [(2n)!] -1. Thus, Eq. (A3.8)

can be written as


(ιK)n (zs )
2 n
χ= ∑ n! (2n)! ,
n=0 (A3.12)

and from Eq. (A3.2) we can arrive at the result for ξ. Following

convention, we define a parameter η=(Kzs2) and write

∞ 

 
z

(ιη)n γ (0) k β(0)


ξ(s,z) = Re ∑
 n=0 n! (2n)!  γ (z) k β(z)
exp  ι k β(z′)dz′
 0  .
(A3.13)

The series can be summed for |η|>>1. The asymptotic result is2

ξ(s,z)= 1

γ (0) k β(0)
γ (z) k β (z)
2η -1/6exp 3 3 2η 1/3
4 ( )
* (
× cos k β(z)z - 3 2η 1/3 - π )
4 12 . (A3.14)

3
4
References and Notes
(Appendix 3)

1
Note, here we have absorbed a scaling factor swake into W 0. Thus, W0 has units of (cm-sec2) -
1. For consistency, one should define W⊥ =W 0(s/swake ) so that W 0 can have the same units as
the wake.
2
A. W. Chao, B. Richter and C. Yao, “Beam Emittance Growth Caused by Transverse Deflecting
Fields in a Linear Accelerator” (Amsterdam, Holland: North-Holland Physics Publishing),
Nuclear Instruments & Methods in Physics Research, 178 (1980).

You might also like