Beam dynamics basics in RF linacs

N. Pichoff
CEA/DIF/DPTA/SP2A/LFPA, France, [email protected]

Abstract
After a short introduction to the advantages and drawbacks of linacs versus
circular accelerators, the beam acceleration process with RF cavities is
explained. Then individual particle motion with time or longitudinal
coordinate (longitudinal, non-linear; transverse, periodic-linear) and beam
statistical motion (r.m.s. definition, matching, mismatching and emittance
growth) are presented.

1 Introduction
This one-hour lecture has been presented at the CERN Accelerator School dedicated to small
accelerators in Zeegse, the Netherlands. Its aim is to give to CAS students an introduction to the basics
of RF beam dynamics to obtain a good understanding of linacs. Students wanting to learn more about
linacs are advised to read books [1] and [2].
After a short introduction on the advantages/drawbacks of linacs versus circular accelerators,
the two main functions are presented:
– Beam acceleration with RF cavities. Many configurations of electromagnetic fields can oscillate
with their own frequency in a cavity. These modes are excited by the RF source or the beam
itself. In nominal conditions, only one mode, whose electric field is along beam propagation, is
excited by the RF source. The energy gain of a cavity can be obtained from a simple model
using the cavity voltage, an average phase and a correction coefficient called the transit time
factor. Moreover, different types of cavities (sections) are used depending on their acceleration
efficiency at a given energy (given by the effective shunt impedance). Finally, for a proper
acceleration, these cavities have to be synchronized with the beam.

– Beam transport in a periodic confinement. In order to reduce emittance growth along the linac,
the total phase-space swept by the beam has to be as small as possible. This is obtained by
putting cavities (for acceleration and longitudinal focusing) and quadrupoles (for transverse
focusing) in accelerating sections composed of periodic lattices. In each accelerating section,
the beam has to be matched at entrance allowing a periodic motion of its envelope. Between
accelerating sections, dedicated matching sections can be used to re-match the beam because of
the change of lattice.

In order to be coherent with the reader, some notions which were not presented in the short one-
hour lecture are exposed in the paper. However, these notions were introduced in other lectures in this
school, in particular by Alessandra Lombardi in her lecture Transverse dynamics I Beam lines and Joel
Leduff in his lecture Longitudinal beam dynamics and stability.

2 Why RF linacs?
The goal of a particle accelerator is to set a wanted beam within a lower cost. By wanted, one means a
given particle type, with a given intensity, at a given energy within a given emittance (or brightness)

145
 N. P ICHOFF

in a given time structure. The cost should include the construction as well as the operation costs
(including the staff).
The three main competitors fulfilling this goal are synchrotrons, cyclotrons, and RF linear
accelerators (linacs)1.
The main advantages of linacs over other competitors are
– they can handle high-current beams (they are not, or less limited by tune shift);

– they can run in high duty-cycle (the beam passes only once at each position);

– they exhibit low synchrotron-radiation losses (no dipoles).

Their main drawbacks are

– they are very space and cavity consuming;

– the synchrotron radiation damping of light particles (electrons/positrons) cannot be easily used
to reduce the native beam emittance.

That is why the linacs are mainly used

– as low-energy injectors (where the space-charge force is more important and the duty-cycle is
high);

– with high-intensity/power beams (high space-charge level or/and duty cycle);

– in new lepton collider project at very high energy (no radiation losses).

3 Beam acceleration with RF cavities

The role of an RF cavity is to give energy to the beam. As the RF represents generally the major part
of the cost of a linac structure (except the building), the RF structure has to be chosen, studied, and
optimized very carefully. In this lecture, only the principle of an RF cavity is presented. More precise
information can be found in the CAS sections dedicated to RF [3].
The principle of travelling-wave cavities usually used for acceleration of ultra-relativistic beams
is presented in Section 3.1.
The mostly used standing-wave cavity is then presented in Section 3.2. Excitation of resonant
modes is explained (3.2.1). A simple model of energy gain, including the cavity voltage, the particle
average phase and its transit time factor calculation is presented (3.2.2). The effective shunt
impedance quantifying the efficiency of the different cavities helps to choose between different types
of accelerating structures (3.2.3).
Finally, Section 3.3 shows the way to calculate the relative phases between all the cavities to
accelerate the beam as chosen by the linac designers.

3.1 Travelling-wave cavity

Travelling-wave cavities are mostly used to accelerate ultra-relativistic particles. These cavities
generally have two power ports. One from which the power enters, the other, at the other end, through
which the power exits (Fig. 1). The electric field travels through the cavity from the input to the output
port. Its phase velocity is matched to the beam velocity. The field phase is chosen to continuously
accelerate the beam.
1
We should not forget electrostatic machines, suitable for low-current, low-energy beams.

146
B EAM DYNAMICS BASICS IN RF LINACS

Fig. 1: A travelling-wave cavity

The RF phase velocity in empty cavities or wave-guides is usually higher than (or equal to) the
speed of light in vacuum c. As particle velocity cannot exceed c, the RF phase velocity should be
slowed down to reach the synchronism condition. This can be done by introducing some periodic
obstacles into the guide (such as an iris-loaded waveguide). The periodic field can then be expanded in
a Fourier series, with different wave numbers:

+∞
Ez (t , z ) = ∑ ez
n = −∞
n ⋅ exp( j ⋅ (ωt − k n z )) . (1)

Here ezn are the space harmonic amplitude, kn are the space harmonic wave numbers,

2πn
kn = k0 + , (2)
d

d is the obstacle period, and k0 is the guide wave number. The phase velocity vn of space-harmonic
number n is

ω
vn = . (3)
kn

A particle, whose velocity is close to the phase velocity of one space harmonic, exchanges
energy with it. Otherwise, the average effect is null.
A complete calculation of these insertion obstacles as well as a large bibliography can be found
in Ref. [4]. This kind of travelling-wave accelerating structure is mainly used to accelerate ultra
relativistic electrons. These cavities are explained in more detail in Maurizio Vretenar’s lecture
Differences between electron and ion linacs.
The model of travelling-wave acceleration, even when accelerating with standing-wave cavities,
is often used to simplify the calculation of equations of longitudinal motion.

147
 N. P ICHOFF

3.2 Standing-wave cavity

3.2.1 Field calculation

An RF cavity is simply a piece of conductor enclosing an empty volume (generally vacuum).
Solutions of Maxwell’s equations in this volume, taking into account the boundary conditions on the
conductor surface, allow the existence of electromagnetic field configurations in the cavity: the
resonant modes.

G G ρ G G G G G G G G
∇⋅E = , ∇⋅B = 0, n × En = 0 , n ⋅ Bn = 0 ,
ε0
G G
G G ∂B G G G 1 ∂E G G Σ G G G
∇× E = − , ∇ × B = μ0 J + 2 ⋅ . n ⋅ En = , n × Hn = K .
∂t c ∂t ε0
G
μ0 = 4π⋅10 T.m.A : permeability of free space,
–7 –1
n , the normal to the conductor,
ε0 = 1/μ0c2 : permittivity of free space, ΣG(C/m2), the surface charge density,
–1
c = 299,792,458 m.s : speed of light in vacuum. K (A/m), the surface current density.

Maxwell's equations Boundary conditions

Each mode, labelled n, is characterized by an electromagnetic field amplitude space

G G G G
configuration En ( r ) /Bn ( r ) (see DTL example on Fig. 2) oscillating with an RF frequency fn. The
electric field amplitude configuration is the solution of the equation

G G ω2 G G
∇ 2 E n + 2n ⋅ E n = 0 , (4)
c
G G
where E n (r ) should satisfy the boundary conditions, ω n = 2π ⋅ f n is the mode pulsation.
The electric (as well as magnetic) field in the cavity at a given time is a weighted sum of the
contribution of all the modes:
G G G G G G
E (r , t ) = ∑ e n (t ) ⋅ E n (r ) = ∑ a n ⋅ e jωnt ⋅ E n (r ) . (5)

Here an is a complex number, en(t) is the field variation with time, solution of [5]:

ω n2 G G
∫ (E × H )⋅ n ⋅ dS
G
en + ω n2 ⋅ e n = − ⋅ n
εμ S
. (6)
G G G 1 d G G G G
∫( )
1 d
+ H × E n ⋅ n ⋅ dS ′ − J (r , t ) ⋅ E n (r ) ⋅ dV
∫
ε dt S′
ε dt V

G
H is the magnetic induction,
G
J is the current density, of the beam for example,
S is the conductor inside surface,
S' is the open surface of the cavity,
V is the inside volume of the cavity.

148
B EAM DYNAMICS BASICS IN RF LINACS

f = 352 MHz f = 548 MHz f = 952 MHz

Fig. 2: Three RF mode configurations in a four-gap DTL (D. Uriot, CEA/DSM)

– The first term on the right-hand side is an integral over the conductor surface. Because of power
losses by Joule effects on this non-perfect conductor, it can be modelled by a damping term:

ωn
− ⋅ en . (7)
Q0 n

The calculation of Q0n, the quality factor of the mode, can be deduced from power loss
considerations:
Un(0) is the energy stored by the n-mode at time t = 0. For t > 0, no more power is injected in
the cavity. Let us define k(t) as the ratio between the field amplitude at t and this at t = 0:
en (t )
k (t ) = . (8)
en (t = 0 )

The energy lost per unit time is the power dissipated in the conductor Pn:
dU n (t )
= − Pn (t ) . (9)
dt

The average power dissipated in the conductor per cycle is proportional to the square of the
current density (and then the magnetic field) close to the surface:
Rs Rs
Pn = ∫K = ∫H
2 2
n dS n dS , (10)
2 S
2 S

where Rs is the surface resistance defined as:

μ 0π f 0
Rs = , for normal conductors (11)
σ

σ is the conductivity (1/σ = 1.7⋅10–7 Ω.m for copper).

f 02 ( GHz ) ⎛ T ⎞
Rs ≈ Rres + 9 ⋅ 10−5 exp ⎜ −1.92 ⋅ c ⎟ , for superconducting niobium (12)
T (K) ⎝ T ⎠

Here Rres is the residual resistance (10–9–10–8 Ω) depending on the surface imperfections,
T is the working absolute temperature,
Tc = 9.2 K is the critical temperature.

149
 N. P ICHOFF

From equations (8) and (10) can be deduced

Pn ( t ) = k ( t ) ⋅ Pn ( t = 0 ) .
2
(13)

The stored energy is proportional to the square of the field:

ε0 G 2 1 G 2
Un =
2
∫∫∫ En dv =
2 μ0
∫∫∫ Bn dv , (14)

giving:
U n (t ) = k (t ) ⋅ U n (t = 0 ) .
2
(15)

Equation (9) becomes

dk 2 P
= − n 0 ⋅ k 2 = 2 ⋅ k ⋅ k , (16)
dt U n0

giving the time evolution of the electric field amplitude:

den P
= − n 0 ⋅ en , (17)
dt 2 ⋅ U n0

whose solution is
⎛ P ⎞
en ( t ) = en ( 0 ) ⋅ exp ⎜ − n ⋅ t ⎟ . (18)
⎝ 2 ⋅U n ⎠

A comparison with the damping term of a second-order linear differential equation written in
(7):
en + 2 ⋅ α ⋅ en + ωn2 ⋅ en = 0 ,
 (19)

whose solution is

en ( t ) = en ( 0 ) ⋅ exp ( −α ⋅ t ) ⋅ cos ( )
ωn2 − α 2 ⋅ t , (20)

gives
ωn ⋅U n
Q0 n = . (21)
Pn

– In the second term, the integration is performed over the open surfaces S' and represents the
coupling of the cavity with the outside. This coupling can be divided into two contributions:
the injected power coming from the power generator through the coupler,
an additional damping, which can be represented by another quality factor Qexn known as the
external Q, corresponding to power losses through the opened surfaces.

The coupling can be calculated from the coupler geometry with electromagnetic codes. This
second term can then be written:

150
B EAM DYNAMICS BASICS IN RF LINACS

ωn
− ⋅ en + S n ⋅ e j (ω RF t + ϕ0 ) . (22)
Qexn

Here S n ⋅ e j (ω RF t + ϕ0 ) is the RF source filling through the coupler.

– The last term, represents the field excited by the beam, known as the beam loading. It is
proportional to the beam intensity:

k n ⋅ I (t ) . (23)

I (t ) is a complex number (it has an amplitude and a phase) representing the beam current.
Equation (6) can finally be modelled by

d 2 en ωRF den
+ ωn2 ⋅ en = S n ⋅ e ( RF 0 ) + kn ⋅ I ( t ) ,
j ω t +ϕ
+ ⋅ (24)
dt 2 Qn dt

which is the equation of a damped harmonic oscillator in a forced regime.

Qn is the quality factor of the cavity, with
1 1 1
= + , (25)
Qn Q0 n Qexn

Qn
τ = 2⋅
ωRF

is the cavity filling time.

One notes that both the coupler and the beam can excite some RF modes.
Equation (24) is of an RLC circuit which is often used to model the system. A complete study
of this model, out of the scope of this lecture, can be found in Ref. [5].
Among all these modes, one, having an electric field on-axis along the longitudinal direction, is
used to accelerate the beam. The cavity geometry is calculated to match the frequency of this
accelerating mode up to the RF frequency. This mode is excited in the cavity through the power
coupler whose geometry is calculated and adjusted to transfer electromagnetic energy in the cavity to
the beam without reflection (this procedure is called coupler matching).

3.2.2 Energy gain model in a standing-wave cavity

A cavity has a finite length L. s0 is the cavity input abscissa. Ez(s) is the amplitude of the electric field
longitudinal component on axis.
The energy2 gained by a charged particle on-axis in the cavity is
s0 + L
ΔW = ∫ qE z (s ) ⋅ cos(φ (s )) ⋅ ds , (26)
s0

where q is the particle charge,

2
This is actually the longitudinal energy, but we can consider that there is no transverse field on cavity axis.

151
 N. P ICHOFF

φ(s) is the cavity RF phase when the particle is at abscissa s. It is defined as

s0 + s
ω ds ′
φ (s ) = φ 0 + ∫ (27)
c s0
β z (s ′)

where φ0 = φ(s0) is the RF phase when the particle enters the cavity.
Writing φ(s) = φ(s) + (φs − φs), φs being an arbitrary phase and using trigonometric relationships,
one gets for the energy gain:
s0 + L s0 + L

ΔW = cos φ s ⋅ ∫ qEz (s ) ⋅ cos(φ (s ) − φ s ) ⋅ ds − sin φ s ⋅ ∫ qEz(s ) ⋅ sin(φ (s ) − φ ) ⋅ ds.

s (28)
s0 s0

By defining φs as
s0 +L

∫ qEz(s ) ⋅ sin(φ (s ) − φ ) ⋅ ds = 0 ,
s0
s

giving the definition of the average phase φs (also called synchronous phase),

⎛ s0 + L ⎞
⎜ ∫ Ez ( s ) ⋅ sin ⎡⎣φ ( s ) ⎤⎦ ⋅ ds ⎟
φs = arctan ⎜⎜ s +0 L ⎟
s
0 ⎟ , (29)
⎜ Ez ( s ) ⋅ cos ⎣⎡φ ( s ) ⎦⎤ ⋅ ds ⎟
⎜ s∫ ⎟
⎝ 0 ⎠

one finally gets

ΔW = qV0 ⋅ T ⋅ cos φ s , (30)

with
s0 + L

V0 = ∫ Ez(s ) ⋅ ds .
s0
(31)

V0 is the cavity voltage. q⋅V0 represents the maximum energy (in eV) that a particle with charge
q could gain if the field were always maximum.

s0 + L
1
T=
V0 ∫ Ez (s ) ⋅ cos(φ (s ) − φ s ) ⋅ ds . (32)
s0

T is known as the transit time factor. It depends on the particle initial velocity as well as on the
field amplitude. It can be noted that this definition does not make any assumption about the field shape
(no symmetry) resulting from a slightly different average phase definition which can be found in the
literature (which is often taken as the RF phase when the particle reaches the mid-cavity). When the
velocity gain in the cavity is much lower than the input particle velocity, T depends only on the
velocity and can be easily tabulated.

152
B EAM DYNAMICS BASICS IN RF LINACS

The transit time factor T is difficult to calculate using formula (32) as φs has to be known. In
fact, T does not depend on φs when the velocity gain in the cavity is small. The transit time factor is
then the maximum energy that can be gained in the cavity normalized to the cavity voltage. This
maximum energy gain is simply the modulus of the complex number whose real part represents the
energy gain:
1
∫ Ez (s ) ⋅ e
jφ ( s )
T= ⋅ ds . (33)
V0

3.2.3 Shunt impedances

At first order, one considers that only one mode, the accelerating mode, is excited in the cavity. The
transverse component of the electric field is generally null on the axis. An expression of the field z-
component on the axis is then
Ez ( s, t ) = Ez 0 ( s ) ⋅ cos (ωt + ϕ ) . (34)

Here Ez0(s) is the field amplitude.

Let Pd be the power deposition in the cavity. It can be modelled by
V02
Pd = , (35)
2⋅R
where V0 is the cavity voltage.
R is known as the cavity shunt impedance and is very useful in cavity design. For an optimal
acceleration, it has to be the highest possible.
As for the quality factor Q, the shunt impedance R is linked to the energy lost in the cavity.
One can easily show that the ‘R over Q’ factor depends only on the geometry of the cavity:
R V02 2 ⋅ Pd V02
= = , (36)
Q ω ⋅ U Pd 2 ⋅ ω ⋅U

as Eq. (14) shows that U ∝ V02 .

One can easily see that, for the same aspect ratio, doubling the frequency (corresponding to
divided by 2 the cavity length and by 8 its volume), has no effect on the cavity ‘R over Q’.
Generally, because the electric field is changing with time as the particle transits through the
cavity, the maximum energy, q⋅V, which can be gained in the cavity by a particle of charge q is lower
than q⋅V0. One defines the transit time factor T as:
V
T= ≤ 1. (37)
V0

It is a corrective factor on the energy gain taking into account the particle transit time in the
cavity. It obviously depends on the particle velocity.
The effective shunt impedance RT2 is then proportional to the ratio between the square of the
maximum energy ΔUmax that can be gained by the beam and the power lost in the cavity:
ΔU max
2
RT 2 = . (38)
2 Pd

153
 N. P ICHOFF

It is some sort of cavity efficiency and has to be maximized.

The effective shunt impedance is often used to compare the efficiency of different structures at a
given energy. As their geometries are generally different, one extends the preceding definition per unit
length to allow a better comparison.
Let L be the cavity length3. The cavity average electric field E0 is defined as
V0
E0 = . (39)
L

The power deposition per unit length in the cavity P'd is then
E 02
Pd′ = , (40)
2⋅Z

where Z is the cavity shunt impedance per unit length.

The effective shunt impedance per unit length ZT2 is then proportional to the ratio between the
square of the maximum energy ΔU'max that can be gained per unit length by the beam and the power
lost per unit length in the cavity:
′2
ΔU max
ZT 2 = . (41)
2 Pd′

As it depends on the particle velocity, one chooses the structure that maximizes ZT2 at a given
energy. Figure 3 represents the evolution of the effective shunt impedance per metre for two different
proton linac structures (SDTL and CCL) with different apertures φ. The higher the aperture (room for
beam) the lower the effective shunt impedance. SDTL structures are more efficient at lower energy as
CCL structures are more efficient at higher energy. In this case, the optimum transition energy is
around 100 MeV for protons.

19 MeV 45 MeV 85 MeV 234 MeV 375 MeV

Fig. 3: Effective shunt impedance per metre of different structures (TRISPAL, C. Bourrat)

3
On account of the cavity fringe field, L is often arbitrarily defined as the physical length of the cavity.

154
B EAM DYNAMICS BASICS IN RF LINACS

3.3 Notion of synchronism

A linac is designed such that one theoretical particle, called the synchronous particle, enters
successively on axis of RF cavities with a designed RF phase law in order to get a wanted energy gain.
This very important notion of synchronism allows understanding the efficiency as well as the stability
of the linacs.
Particles can be accelerated with travelling waves as well as standing waves.

3.3.1 With travelling waves

Synchronous β z⋅c
particle

ω
k

Fig. 4: Particle accelerated by a travelling wave

The on-axis RF accelerating field in a travelling-wave cavity can be written as

Ez ( z, t ) = E0 ⋅ cos (ω t − kz ) , (42)

where ω is the RF pulsation, and k is the RF wave number (Fig. 4).

The synchronism condition is reached when the particle longitudinal velocity equals the RF
phase velocity:
k
βz c = . (43)
ω

Here c is the speed of light in vacuum, βz is the reduced longitudinal velocity of the
synchronous particle. It can be noted that when the paraxial approximation4 is used, βz is replaced by
β, the reduced total speed of the particle.

3.3.2 With standing waves

In most linacs, the beam is accelerated with RF cavities or gaps operating in standing-wave conditions.
An RF power, produced by one or many RF sources, is introduced through a coupler in a resonant
cavity exciting the wanted standing-wave accelerating mode. The cavity shape has been calculated and
adjusted to match the accelerating mode to the power-supply frequencies and to throw the other mode
frequencies far from the RF one.

2 2
4
As β = β z ⋅ 1 + x′ + y′ , the paraxial approximation occurs when x' << 1 and y' << 1.

155
 N. P ICHOFF

As a first step, let us assume a set of thin independently phased RF cavities along the beam path
(Fig. 5).

RF phase φi-1 φi φi + 1
Particle velocity βsi-1⋅c βsi⋅c
Distances Di-1 Di
Synchronous phase φsi-1 φsi φsi+1
Cavity i-1 i i+1

Fig. 5: A set of independently phased cavities

φi is the absolute RF phase in the ith cavity when t = 0 (the t = 0 instant has been arbitrarily
chosen),

βsi is the synchronous particle reduced velocity at the ith cavity output,
φsi is the RF average phase of the ith cavity of the synchronous particle,
Di is the distance between the ith and the i + 1th cavities.
The synchronism condition is reached when the RF phase difference between two consecutive
cavities, Δφi = φi +1 − φi , equals the phase evolution during synchronous particle time of flight between
both cavities,
Di Di
ω⋅ = 2π ⋅ ,
β si ⋅ c β si λ

plus the difference Δφ si = φ si +1 − φ si between the designed synchronous phase in each cavity:

Di
Δφi + [ 2π ⋅ n ] = Δφsi + 2π ⋅ . (44)
β si λ

λ = c/f is the RF wavelength.

One observes that the synchronism condition does not depend on the RF field amplitude. It has
a non-intuitive consequence: an increase of the accelerating field amplitude in the cavities without
phase change does not induce an increase of the synchronous particle final energy but a change of the
synchronous phase fulfilling the synchronism condition.
Two different kinds of structure exist:
– Coupled cavity structures where the phase between consecutive cavities is fixed. The
synchronism condition is achieved by adjusting the distance between the cavities.

In a Drift Tube Linac (DTL), for example, the phase difference between the cells is fixed
(= 2π). The distance between cells is then calculated to get

⎛ Δφ ⎞
Di = ⎜ si + 1⎟ ⋅ β si λ . (45)
⎝ 2π ⎠

156
B EAM DYNAMICS BASICS IN RF LINACS

– Independent cavity structures where the distance between cavities is fixed. The synchronism
condition is then achieved by adjusting the phase difference between the cavities.

In a Superconducting Cavity Linac (SCL), for example, the distance between cavities is fixed
by the cryogenics mechanics (as small as technologically possible). The phase difference
between cavities is then calculated to have
Di
Δφi = Δφsi + 2π ⋅ + [ 2π n ] . (46)
β si λ

4 Beam transport
A beam is a set of particles. The beam motion can then be expressed as the statistical result of the
motion of individual particles. Many codes are now transporting beam particles individually,
reconstructing beam properties where they are needed.
The first step is to study individual particle motion in the accelerator (Section 4.1). This motion
can be expressed by the variation of the particle coordinates with time (4.1.1) or with longitudinal
position along the linac (4.1.2). The non-linear dynamics along the longitudinal direction (4.1.3) is
traditionally separated from the linear and periodic transverse dynamics (4.1.4), despite a more or less
influent coupling.
The second step consists in constructing a beam as statistical properties of a cloud (or
distribution) of particles, in particular through its emittance (4.2.1) and r.m.s. properties (4.2.2). For a
proper beam transport, these beam properties have to be matched to the linac (4.2.3) to avoid
degradation of the beam optical properties along the transport through the linac.

4.1 Particle motion

4.1.1 General equation of motion with time

Electromagnetic fields can be divided into two contributions:
G
– the electric field: E ,
G
– the magnetic field: B .

The intensity of these contributions depends on the referential where they are expressed.
The equation of motion of a particle of charge q in these fields is:
G G G G G G ⎞
dp ⎛ p
= q ⋅ ⎜⎜ × B(t , r ) + E (t , r )⎟⎟ . (47)
dt ⎝γ ⋅m ⎠
G G
Here p is the momentum of the particle, m its mass, γ its reduced energy, r its position.
Using the reduced momentum,
G G
p
γβ = ,
m⋅c
one gets
G G
dγβ Z ⋅ c ⎛ γβ G G G G ⎞
= ⋅ ⎜ × cB (t , r ) + E (t , r )⎟⎟ . (48)
dt mc 2 e ⎜⎝ γ ⎠

157
 N. P ICHOFF

Here c is the velocity of light in vacuum, mc 2 e is the particle rest mass in eV, Z is the particle
charge number.
It is convenient to multiply the magnetic field by c because it is homogeneous to an electric
field. In that case a magnetic field of 1 T is ‘equivalent’ to an electric field of 300 MV/m.
The particle reduced energy γ can be deduced from its reduced momentum:
G 2
γ = 1 + γβ . (49)

The variation of the position is given by

G G
dr γβ
= ⋅c. (50)
dt γ
G G
The motion of each particle can be described using only two vectors (six coordinates): r , γβ . ( )
The new generation of codes uses time more and more as an independent variable. It is
particularly the case for simulations of RF guns, RFQs or beams with strong space-charge. Moreover,
Maxwell’s equations can be solved with time to compute the evolution of the electromagnetic field
and be coupled with the particle’s motion.
Nevertheless, because of the particularity of accelerators that have a privileged axis, because
transport elements are placed at given positions along the beam path and because diagnostics
measuring beam properties are also placed at given positions, time is replaced in ‘traditional’ beam
dynamics by the longitudinal position in the accelerator as independent variable.

4.1.2 Equation of motion with longitudinal position

4.1.2.1 General equations

Let us call s the longitudinal position (abscissa) in the linac path (rather than z to avoid any confusion
with the particle longitudinal position in the bunch at a given time), the equation of motion (47) can be
rewritten:
G G G G
dp v×B+E
=q⋅ , (51)
ds vz

ds
vz = ,
dt

where vz is the particle longitudinal velocity.

Using reduced momentum and convenient physical parameters, one gets
G G G G
dγβ Z β × cB + E
= ⋅ , (52)
ds mc 2 e βz

c is the velocity of light in vacuum, mc 2 e is the particle rest mass in eV, Z is the particle charge
number.
A projection on a Cartesian frame5 (x, y, z) gives

5
In general, x and y play the same role in a linac (in contrast with that of a circular accelerator).

158
B EAM DYNAMICS BASICS IN RF LINACS

⎧ d γβ x Z ⎛ E ⎞ d γβ z x′
⎪ = ⋅ ⎜ y ′ ⋅ cBz − cBy + x ⎟ =
⎪ ds
2
mc e ⎝ βz ⎠ ds
⎪ d γβ
⎪ Z ⎛ E y ⎞ d γβ z y ′
⋅ ⎜ cBx − x′ ⋅ cBz +
y
⎨ = ⎟= (53)
⎪ ds
2
mc e ⎝ βz ⎠ ds
⎪
⎪ d γβ z = Z ⋅ ⎛⎜ x′ ⋅ cB − y ′ ⋅ cB + Ez ⎞⎟
⎪⎩ ds mc 2 e ⎝
y x
βz ⎠

dx px β x dy p y β y
x′ = = = and y′ = = =
ds pz β z ds pz β z

are the slopes of the particle;

vw
βw = ,
c
the reduced velocity w-component; w being x, y or z; vw being the particle velocity w component;
m and q are, respectively, the rest mass and the charge of the particle; c is the speed of light.
One clearly observes that longitudinal and transverse motions are coupled. However, for an
easier understanding, and because the coupling is often very weak, the longitudinal and the transverse
motions are usually treated as uncoupled, the longitudinal velocity vz variations being considered
apart. To uncouple the transverse and longitudinal motions, the paraxial approximation has to be
done.

4.1.2.2 Paraxial approximation

The paraxial approximation consists in assuming that the velocity of the particles is mainly along the
longitudinal direction, i.e., the slopes of the particle are very small:
x ′ 2 + y ′ 2 << 1 . (54)
Its natural consequence is that the longitudinal velocity or momentum can be assimilated to the
total velocity or momentum:

β z = β ⋅ 1 + x ′2 + y ′ 2 ≈ β . (55)

For example, if x' < 100 mrad and y' < 100 mrad, the error on β (or βz) is lower than 1%.
This approximation is very good at high energy where the beam divergence is small, but is
sometime more difficult to justify at very low energy.

4.1.2.3 Energy gain calculation

From Eqs. (53) and (49), one can easily obtain the energy gain:

dγ ⎛ d γβ x d γβ y d γβ x ⎞
= β z ⎜ x′ ⋅ + y′ ⋅ + ⎟, (56)
ds ⎝ ds ds ds ⎠

giving
dγ
=
q
ds mc 2
(
⋅ x′E x + y ′E y + Ez . ) (57)

One finds the well-known result that the magnetic field does not contribute to energy gain.

159
 N. P ICHOFF

4.1.3 Longitudinal particle dynamics (motion in non-linear force)

4.1.3.1 The longitudinal variables

The variables generally used to describe the longitudinal particle motion, as a function of s, are
– φ, the absolute particle phase, calculated from the RF frequency, with φ = 0 arbitrarily chosen;

– W, the particle kinetic energy6.

The evolution of φ and W with s is given by the equations

⎧ dφ ωRF 2π
⎪ ds = β c =
⎪ z β ⋅ λRF ⋅ 1 − x′2 − y ′2
⎨ . (58)
⎪ dW
⎪⎩ ds = q ⋅ ⎡⎣ x′Ex ( s,φ , r ) + y ′E y ( s,φ , r ) + Ez ( s,φ , r ) ⎤⎦

Applying these equations to the synchronous particle, one gets

⎧ d φs 2π
⎪⎪ ds = β ⋅ λ
⎨
s RF
. (59)
⎪ s = q ⋅ E ( s,φ ,0 )
dW
⎪⎩ ds z s

Let us define the reduced phase and energy variables for each particle:

⎧ϕ = φ − φs
⎨ . (60)
⎩ w = W − Ws

All the particles are now referenced to the synchronous particle. Late particles compared to
synchronous particle have a positive ϕ.
The equations of motion of particles with respect to the synchronous particle (with these new
variables) become
⎧ dϕ 2π ⎛ 1 1 ⎞
⎪ = ⋅⎜ − ⎟
⎪ ds λRF ⎜ β ⋅ 1 − x′2 − y ′2 β s ⎟
⎨ ⎝ ⎠ . (61)
⎪ dw
⎪ = q ⋅ ⎣⎡ x′Ex ( s,φ , r ) + y ′E y ( s, φ , r ) + Ez ( s,φ , r ) − Ez ( s,φs ,0 ) ⎦⎤
⎩ ds

When the beam is accelerated by a standing-wave cavity structure, synchronous particles enters
successive cavities giving it a strong energy gain, separated by long drift spaces where no acceleration
occurs. In order to ease the understanding of the physics, this periodic acceleration scheme can be
replaced by a continuous acceleration one. This scheme consists in assuming that the beam is
accelerated by a travelling wave propagating at the same speed as that of the synchronous particle.
This scheme allows a mathematical resolution of the dynamics equations7 with an electric field that
does not depend on s or t.

6
This is really a ‘longitudinal’ particle property only in paraxial approximation!
7
Equations are smoothed for analytic solutions, but quantified for a numerical solution!

160
B EAM DYNAMICS BASICS IN RF LINACS

4.1.3.2 The continuous electric field model

The electric field, which is generally a function of s, is then chosen constant. The field amplitude of
the travelling wave is E0T (mean electric field) on axis. E0 is defined as the potential gain of one
cavity V0 divided by the distance between the centres of consecutive cavities. The transit time factor T
has been included to take into account the variation with the particle velocity of the acceleration in
standing-wave cavities.
The on-axis electric field longitudinal component becomes

Ez ( s, ϕ , r = 0 ) = E0T ⋅ cos (ϕ + φs 0 ) , (62)

φs0 being the RF average phase of the synchronous particle.

The energy gain per metre of the synchronous particle is then

G = qE 0Ts ⋅ cos φ s 0 . (63)

Here Ts is the transit time factor of the synchronous particle.

Let us assume an axi-symmetric accelerating field, the off-axis electric field longitudinal
component can be written:
Ez ( s, φ , r ) = E0T ⋅ R (r ) ⋅ cos (ϕ + φs 0 ) , (64)

with r being the radial position of the particle, and R(r) expressing the radial evolution of the electric
field longitudinal component. It can usually be written: R (r ) = 1 + O(r 2 ) . Close to the axis, Bessel
function, a solution of Maxwell’s equations in axi-symmetric geometry in vacuum, can be used to
express R(r) [6] [2], but far from the axis, the cavity geometry has a strong influence through the
boundary conditions. The radial position (r) can be replaced by (x, y) if the cavity is not axi-
symmetric. Some authors include the variation of the field with r in the transit time factor: T(r).
G G
Applying Gauss’s law: ∇ ⋅ E = 0 and remarking that the electric field transverse component is
null on axis, one gets the electric field transverse component:

1 r ∂E ( s,ϕ , r )
Er ( s , ϕ , r ) = − ⋅ ∫ z ⋅ r ⋅ dr
r 0 ∂s
(65)
r
1 ET
= − ⋅ 0 ⋅ sin (ϕ + φs 0 ) ⋅ ∫ R ( r ) ⋅ r ⋅ dr .
r βsλ 0

The electric field radial component can be written

⎛r ⎞
Er ( s , ϕ , r ) = −
E0T
βsλ ⎝2
( )
⋅ sin (ϕ + φs 0 ) ⋅ ⎜ + O r 3 ⎟ .
⎠
(66)

4.1.3.3 The simplified equations of motion

Three assumptions should be made to decouple the longitudinal motion from the transverse one:
– Generally, the paraxial assumption occurs, and one considers:

x ′ 2 + y ′ 2 << 1 . (67)

161
 N. P ICHOFF

( )
– In general, one can assume: r/2 + O r 3 << β s λ and (x’, y’) << 1. The contribution of the
transverse electric field to the energy gain can then be neglected in Eq. (61)8:

x ′E x + y ′E y << E z − E zs . (68)

– Finally, one can assume that the longitudinal field does not depend on the radial position r, by
taking

R(r ) ≈ 1 . (69)

Using these assumptions, Eqs. (61) become

⎧ dϕ 2π ⎛ 1 1 ⎞
⎪ =− ⎜⎜ − ⎟,
⎪ ds λ ⎝ β ( s ) β s ( s ) ⎟⎠
⎨ (70)
⎪ dw
⎪ ds = −q ⋅ E0T ⋅ ( cos φs 0 ⋅ (1 − cos ϕ ) + sin φs 0 ⋅ sin ϕ )
⎩

which is in fact the equation of motion of on-axis particles.

Moreover, a small longitudinal velocity dispersion assumption is often made:
1 1 1
− = δβ −1 << . (71)
β βs βs

A first-order development around synchronous velocity gives

w
δβ −1 = − . (72)
(β s γ s )3 ⋅ mc 2
If one considers that the transit-time factor does not depend on the beam particles’ energy:
T (w) = Ts , (73)

Eqs. (70) become

⎧ dϕ w ∂Hφ w
⎪ = −2π ⋅ =
( β sγ s ) ⋅ mc ⋅ λ ∂w
3
⎪ ds
2

⎨ . (74)
⎪ ∂H
⎪ ds = − q ⋅ E0Ts ⋅ ( cos φs 0 ⋅ (1 − cos ϕ ) + sin φs 0 ⋅ sin ϕ ) = − ∂ϕ
dw φw

As ϕ and w are canonical variables with the independent variable s, a Hamiltonian Hϕw has been
used to describe the particle motion:
2π w2
Hϕ w = − ⋅ − q ⋅ E0Ts ⋅ R ( r ) ⋅ ⎡⎣sin φs 0 ⋅ ( cos ϕ − 1) + cos φs 0 ⋅ ( sin ϕ − ϕ ) ⎤⎦ . (75)
( β sγ s ) ⋅ mc 2 ⋅ λ 2
3

8
This assumption is not valid when sinφs tends to zero (no acceleration). However, when φs equals 0, even if the transverse
kick is the main contributor to energy variation, it is small compared to the energy and always null on the axis.

162
B EAM DYNAMICS BASICS IN RF LINACS

In the phase space (ϕ, w), particles are following curves for which Hφw = Cst. They are
represented on Fig. 6 for on-axis particles. On Fig. 6(a), βsγs = Cst, as on Fig. 6(a), an adiabatic
acceleration (βsγs ≠ Cst) is added and the bucket turns into the well-known golf club shape.

(1)
(2)

w
w

a) b)
ϕ ϕ

Fig. 6: Particle trajectories in longitudinal phase space

a) βsγs = Cst,

b) Adiabatic acceleration: the golf club represents the input acceptance [in red, (1)]. In blue (2)
are the trajectories of two particles. They exhibit the damping of the phase oscillation
amplitude with acceleration.
A particle entering the cavity later than the synchronous particle gets a larger energy gain. A
particle entering the cavity earlier gets a smaller energy gain.
The synchronous phase of the synchronous particle is a stable point situated between –π/2
and 09.
The choice of the synchronous phase delimits a phase acceptance:
– The higher limit φ1 is the phase for which a late particle gets the same energy gain as the
synchronous particle:

φ1 = −φ s 0 ⇒ ϕ1 = −2 ⋅ φs 0 . (76)

– At the lower limit φ2, the confinement potential equals the potential at the higher limit (φ1). As
the potential is the integral of the force, φ2 is the phase for which the horizontally hatched
surface on Fig. 7 equals the vertically hatched one. It can be calculated from the Hamiltonian
given in Eq. (75):

Hϕ w (ϕ = φ2 − φs 0 , w = 0 ) = Hϕ w (ϕ = φ1 − φs 0 , w = 0 ) . (77)

φ2 is the solution of
(sin φ 2 − φ 2 cos φ s 0 ) + (sin φ s 0 − φ s 0 cos φ s 0 ) = 0 . (78)

– The choice of the synchronous phase determines also the energy acceptance ΔE corresponding
to the difference between the potential energy of a particle with a phase φ1 and that of the
synchronous particle. It can also be calculated from the Hamiltonian given in Eq. (75):

Hϕ w (ϕ = 0, w = ΔE ) = Hϕ w (ϕ1 = φ1 − φs 0 , w = 0 ) , (79)

9
For a positively charged particle, as for a negatively charged one it depends on convention (is qE0 > 0 or E0 > 0 ?).

163
 N. P ICHOFF

giving
ΔE = 2 ⋅ qE 0 T (φ s 0 cos φ s 0 − sin φ s 0 ) . (80)

1
⎛ (β s γ s )3 ⋅ λ ⋅ mc 2 ⎞2
ΔE = ⎜ ⋅ 2 ⋅ qE 0T (φ s 0 cos φ s 0 − sin φ s 0 )⎟ . (81)
⎜ π ⎟
⎝ ⎠

q⋅E0T

Late particle

βs
Synchronous particle

Early Particle

φ2 φs0 φ1 ϕ+φs0

Fig. 7: Energy gain — synchronous particle

The acceptance area in the phase-energy space is called the bucket, its limit is called the
separatrix. The energy acceptance ΔE and the phase φ2 are represented as a function of the
synchronous phase on Fig. 8.

1 0

0.9 -30

0.8 -60
Phase acceptance limite ϕ2 (°)
Energy acceptance / (2 E0 T)

0.7 -90

0.6 -120

0.5 -150

0.4 -180
Energy acceptance
0.3 φ2 -210

0.2 -240

0.1 -270

0 -300
-90 -80 -70 -60 -50 -40 -30 -20 -10 0
Synchronous phase (°)

Fig. 8: Bucket dimensions as a function of the synchronous phase

164
B EAM DYNAMICS BASICS IN RF LINACS

For small phase amplitude oscillations, Eq. (74) becomes

⎧ dϕ w
⎪ ds = −2π ⋅
( β sγ s ) ⋅ mc 2 ⋅ λ
3
⎪
⎨ (82)
⎪ dw
⎪⎩ ds = q ⋅ E0Ts ⋅ sin φs 0 ⋅ ϕ

giving the second-order differential equation of phase evolution:

d 2ϕ 2 dϕ
+ ⋅ + k z2 ⋅ ϕ = 0 , (83)
ds 2 ς ds

with
2πq ⋅ E 0Ts ⋅ sin (− φ s 0 )
k z2 = . (84)
(β s γ s )3 ⋅ mc 2 ⋅ λ
Here kz is the phase advance per metre of the beam core. In periodic structures of period L,
σ z = k z L is the longitudinal core phase advance per lattice.
2 β sγ s
ς= ⋅ . (85)
3 d (β s γ s ) ds

Here ς is the damping length of the core oscillations.

Both ς and the variation of kz with βsγs contribute to phase oscillation damping with
acceleration. The adiabatic damping of the phase amplitude oscillation ϕa, defined when the
contribution of ς is negligible, can be calculated [7]:
−3/ 4
ϕa ∝ ( β sγ s ) . (86)

The Liouville theorem implies that the energy amplitude oscillation wa variation is

wa ∝ ( β s γ s )
3/ 4
.

The Hamiltonian in linear force then becomes

2π w2 ϕ2
Hϕ w = − ⋅ + q ⋅ E T ⋅ sin φ ⋅ . (87)
( β sγ s ) ⋅ mc 2 ⋅ λ 2
3 0 s s0
2

The curves for which the Hamiltonian is constant are then ellipses.

4.1.4 Motion in linear force

We have seen that the longitudinal particle motion is basically non-linear, but it can be linearized
when the particle phase oscillation amplitude is very small compared to the bucket phase size. The
transverse forces are much more linear than the longitudinal one, and the use of linear focusing force
is very close to reality, and can be solved analytically.

4.1.4.1 Linear transverse forces

In linacs, the main elements used to transport a beam are the cavities and the quadrupoles. Both these
elements induce transverse forces.

165
 N. P ICHOFF

Quadrupoles
In a perfect thick lens model, the quadrupole’s magnetic field is
⎧ Bx = G ⋅ y
⎨ (88)
⎩B y = G ⋅ x

over a distance L known as the quadrupole’s magnetic length, and 0 outside.

G is the quadrupole gradient (in T/m).
With the paraxial approximation and because a magnetic field does not change the particle
energy, the equations of transverse dynamics in quadrupole are then
⎧ dγβ x dx ′ q⋅G
⎪ ds = γβ z ds = − mc x
⎪
⎪ dx βx
′
⎪ ds = x = β
⎪ z
⎨ dγβ . (89)
⎪ dy ′ q ⋅ G
= γβ z
y
= y
⎪ ds ds mc
⎪
⎪ dy ′ = y ′ = β y
⎪⎩ ds βz

The transverse force is linear in a perfect quadrupole.

Actually, fringe fields and non-perfect hyperbolic poles induce non-linear effects which are
generally neglected at first order in linacs.

RF gap
When a particle travels through a cavity, the integration of the effect of the radial electric field and the
azimuthal magnetic field can be modelled by a transverse kick, which is linear at second order. This
kick modifies the particle transverse momentum:
π qE0TL
Δ ( γβ r ) = − 2 ⎣ ( )
⋅ sin φ ⋅ ⎡ r + O r 3 ⎤ = Δ ( γβ z ) ⋅ r ′ + γβ z ⋅ Δr ′ .
⎦
(90)
mc γβ z λ
2

With:
β r2 = β x2 + β y2 .

The term in r' shows that the particle transverse oscillation is damped by acceleration in
accelerating cavities. The dependence on φ shows that the transverse dynamics is naturally coupled at
second order to the longitudinal one by RF cavities.

4.1.4.2 Motion of a particle in periodic linear force

At first order, the motion of a particle can be linearized and the motion along all directions can be
decoupled. The motion equation in the w direction (w being x, y or ϕ) is a solution of a second-order
equation:

166
B EAM DYNAMICS BASICS IN RF LINACS

d 2 w Aw dγβ z dw
+ ⋅ ⋅ + k w (s ) ⋅ w = 0 . (91)
ds 2 γβ z ds ds

Aw being a constant equal to 1 for w = x or y and 3 for w = ϕ.

Now, let us consider that the focusing force is periodic with period S, i.e., k x (s + S ) = k x (s ) .
Generally, the damping term given by the acceleration is very small and can be considered as a
perturbation:

Aw dγβ z dw
⋅ ⋅ << k w (s ) ⋅ w , (92)
γβ z ds ds S
S

a S
giving the average value of quantity a over one lattice period.

With this assumption, the solution of Eq. (91) is given by Floquet’s theorem:

w ( s ) = β wm ( s ) ⋅ I w γβ z ⋅ cos ⎡⎣ψ w ( s − s0 ) + ψ w ( s0 ) ⎤⎦ , (93)

with βwm periodic ( β wm (s + S ) = β wm (s ) ), known as the structure’s beta function, solution of

d 2 β wm 2 ⎡ 1 ⎛ d β ⎞2 ⎤
+ 2 ⋅ k w ( s ) ⋅ β wm − ⋅ ⎢1 + ⋅ ⎜ wm ⎟ ⎥ . (94)
ds 2
β wm ⎢⎣ 4 ⎝ dt ⎠ ⎥⎦

I w γβ z , known as the Courant–Snyder invariant (which is actually invariant with no

acceleration), and ψw the particle phase advance, are defined as:

s
ds
ψ w (s ) = ∫ . (95)
s0
β wm (s )

In (w, w') phase space, particles are turning around periodic ellipses whose equations are

γ wm (s ) ⋅ w 2 + 2 ⋅ α wm (s ) ⋅ w ⋅ w′ + β wm (s ) ⋅ w′ 2 = I w γβ z , (96)

with
1 dβ wm (s )
α wm (s ) = − , (97)
2 ds
and

1 + ⎡⎣α wm ( s ) ⎤⎦
2

γ wm ( s ) = . (98)
β wm ( s )
The surface of the ellipses decreases as 1 γβ z which is close to 1 γβ with the paraxial
approximation.
The phase advance per lattice σw defined as:

σ w = ψ w (s + S ) − ψ w (s ) , (99)

167
 N. P ICHOFF

gives an idea of how fast the particles are turning around the ellipses. 2π /σ w is the number of lattice
periods after which the particle has made one turn around the ellipse. One can notice that, in linear
forces, the phase advance per lattice is the same whatever the particle amplitude.
As an example, let us have a look at a particle motion along one direction in a FODO channel.
On Fig. 9, five FODO lattices have been represented. The particle phase advance per lattice is
360°/5 = 72°. The particle position in 2D phase space is represented by the red point in four different
positions in the lattice. Each line corresponds to one position:
– 1st line: middle of focusing quadrupole,
– 2nd line: between focusing and defocusing quadrupoles,
– 3rd line: middle of defocusing quadrupole,
– 4th line: between defocusing and focusing quadrupoles.
One observes that lattice after lattice, at the same position, the particle turns around an ellipse.
The ellipse shape, whose equation is given by (96), depends on the position within the lattice. This is
very important to understand that these ellipses have nothing to do with the beam (no beam has been
defined here, but just one particle). These ellipses are defined by the transport channel only.
w'

w w w w w w

Fig. 9: Particle transport in a FODO channel

As a conclusion, we should keep in mind that a large number of assumptions have been made to
obtain that result. The opportunity of each assumption has to be studied very carefully in practical

168
B EAM DYNAMICS BASICS IN RF LINACS

cases. Nevertheless, the results presented here give a good understanding of the transverse beam
dynamics in a linac.

4.2 Beam motion

A beam is a set of particles. The beam motion can then be expressed as the statistical consequence of
the individual motion of each particle. Many codes are now transporting beam particles individually,
reconstructing beam properties where they are needed. However, some rules should be applied to
minimize the degradation of beam optical properties.

4.2.1 Emittances

4.2.1.1 Why?
A beam is a set of billions of particles with different characteristics. It would be very convenient to
define and use a simple quantity which would quantify its ‘quality’ and which would be conserved
through a perfect transport and acceleration. How do we define the ‘quality’ of a beam? Something
like: Its ability to perform its goal.
In some accelerators, the goal is to focus the beam on a surface as small as possible, in others, it
is to keep it as parallel as possible over a long distance, and in others it is to do both! All these
qualities can be quantified by a simple quantity known as the beam emittance10.

4.2.1.2 What?
Emittance represents the phase-space volume occupied by beam particles. The phase space can be
either 2D [(x, x'),(y, y'), (phase, energy)], 4D [(x, x', y, y'), (x, x', phase, energy)] or 6D [(x, x', y, y',
phase, energy)]. When the phase space is (x, x') or (y, y'), it is called the geometric emittance.
With this definition, a theorem, known as the Liouville theorem, stipulates that the emittance is
conserved in conservative forces which depend only on the geometrical dimensions of this phase
space (even non-linearly).

4.2.1.3 How?
Let us take a beam in a laminar flow, i.e., all the particles trajectories are parallel to each other
(Fig. 10). At a given position s0, the beam radius is R. All beam particles have a zero slope. In the
(x, x') phase space, this beam is a line. The surface it occupies is null, its emittance is zero. One metre
farther in a drift, the beam size radius is still R. A perfectly laminar beam can be transported without
focusing element over an infinite distance. Let us focus this beam with a perfect lens of focal length f,
i.e., with a focusing force proportional to the distance from its centre. The beam is focused at distance
f to a zero size. A perfectly laminar beam can be focused on an infinitely small surface. In this case,
the beam envelope (dashed line on Fig. 10) follows the trajectory of the outside particle.
Let us take a parallel beam with a non-null emittance whose phase-space shape is an ellipse in
(x, x') phase space (Fig. 11). At a given position s0, the beam radius is R. Extreme particles have a zero
slope as central particles have a slope between –x'0 and +x'0. The phase-space surface it occupies, its
emittance, is non-null. One metre farther in a drift, the beam size radius has increased as the emittance
has increased. Let us focus this beam with a perfect lens of focal length f, i.e., with a focusing force
proportional to the distance from its centre. The beam is focused at distance f to a non-zero size. The
beam size would grow as its emittance grows. The beam envelope is then limited by different
particles.

10
However, it is not unbelievable to define other quantities if the goal is different (like the brightness or the luminosity).

169
 N. P ICHOFF

x' x' x' x'

x x x x

Fig. 10: Particle trajectories in laminar beam

x' x' x' x'

x'0

x x x x

Fig. 11: Particle trajectories in non-zero emittance beam

It is clear that the higher the emittance, the lower the capacity to transport the beam parallel
over a long distance and to focus it to a very small size. Moreover, to maintain the beam within a
given size, the higher the emittance, the more focusing elements per unit length are needed.

170
B EAM DYNAMICS BASICS IN RF LINACS

Nevertheless, let us have a look at the focusing of a laminar beam with a non-linear lens (the
focusing force is not proportional to the distance from its centre, Fig. 12). It is clear that, through a
non-linear lens, the emittance defined as the phase-space volume occupied by the beam is conserved
(still zero, a line). However, because the phase-space shape is no longer a straight line, it will be
impossible to focus this beam on an infinitely small size or to transport this beam parallel over a
infinite distance.
This implies that another definition of emittance is needed to take into account this effect. This
‘new’ emittance should be
– a phase-space surface,
– non-zero when the beam phase-space distribution is curved,
– easy to calculate whatever the particle distribution in the beam (it is difficult to calculate the
phase-space surface occupied by a Gaussian beam which has, per definition, no limit or when
only a few of the particles stand very far from the others),
– linked to quantities that can be measured and which have an interest for application.
This quantity is known as the root mean squared (r.m.s.) emittance, linked to the root mean
squared dimensions of the beam.

x x x x

x x x x

Fig. 12: Particle trajectories in beam focused by non-linear lens

4.2.2 Beam r.m.s. dimensions and Twiss parameters

A bunch has N particles. Its dimensions are defined statistically as follows:
– The beam average position:

1
w =
N
∑ wi , (100)
i =1, N

171
 N. P ICHOFF

which is the position of its centre of gravity (c.o.g.).

– The beam average slope:

1
w′ =
N
∑ wi′ , (101)
i =1, N

which is the slope of its c.o.g.

– The beam r.m.s. size:

(w − ) 1
∑ ( wi − )
2 2
w = σ w = w = w , (102)
N i =1, N

which is the square root of the average square distance of particles from the beam c.o.g.
position.
– The beam r.m.s. divergence:

( w′ − ) 1
∑ ( wi′ − )
2 2
w ′ = σ w′ = w′ = w′ , (103)
N i =1, N

which is the square root of the average square slope of particles from the beam c.o.g. slope.
– The beam 2D phase-space coupling term is:

′ = σ ww′ =
wwrms (w − w ) ⋅ ( w′ − w′ ) , (104)

which gives the coupling of the phase-space distribution.

– The beam r.m.s. emittance:

εw = w 2 w ′2 − ( w − w ) ⋅ ( w′ − w′ )
2
. (105)

Let us meditate a little on the properties of this ‘new’ emittance:

– It has the dimension of a surface in the (w, w') phase space.
– When the coupling term is zero, the r.m.s. emittance is simply the product between r.m.s. size
and r.m.s. divergence. Otherwise, the emittance is smaller.
– If the beam (w, w') phase-space chart is a line whose equation is

k ⋅ w if w ≤ w0
w′ = ,
0 otherwise

whose surface is zero (laminar beam) its r.m.s. emittance is simply

172
B EAM DYNAMICS BASICS IN RF LINACS

~ 2 − k 2 ⋅ (w − w
ε~w = k 2 ⋅ w ) 2 =0,
whatever the value of k or w0.
– When a particle is transported through a linear transformation

⎛ w ⎞ ⎛ M 11 M 12 ⎞ ⎛ w ⎞
⎜⎜ ⎟⎟ = ⎜⎜ ⎟⋅⎜ ⎟ ,
⎝ w′ ⎠1 ⎝ M 21 M 22 ⎟⎠ ⎜⎝ w′ ⎟⎠ 0

it is easy to show that

ε~w1 = M 11 ⋅ M 22 − M 12 ⋅ M 21 ⋅ ε~w0 .

The emittance is conserved when the transfer matrix determinant is one (quadrupoles, drifts,
etc.).
– When a beam is accelerated from a reduced momentum γβ 0 to γβ1 , the transverse transfer
matrix determinant of the accelerating element is not 1 but : γβ 0 /γβ1 . This effect is due to the
increase of longitudinal momentum without variation of the transverse one leading to a
reduction of the particle slope by γβ 0 /γβ1 . The r.m.s. emittance which is conserved even with
acceleration is called the normalized r.m.s. emittance and its expression (in paraxial
approximation) is

ε~wn = βγ ⋅ ε~w . (106)

– Because it is linked to average dimensions of the beam, the r.m.s. emittance is smaller than the
surface of the smallest ellipse enclosing all the beam particles. The ratio between this ‘total’
emittance and the r.m.s. emittance depends on the particle distribution.
The beam Twiss parameters are then deduced from the beam r.m.s. dimensions:

~ ~2
w ~′2
w (w − w ) ⋅ (w′ − w′ )
βw = ~ , γ~w = ~ , α~w = − . (107)
εw εw ε~w

~
γ~w ⋅ w 2 + 2 ⋅ α~w ⋅ w ⋅ w′ + β w ⋅ w′ 2 = n ⋅ ε~w (108)

is the equation of an ellipse in (w, w') phase-space fitting the best the particle distribution.
Generally, at least 90% of the bunched-beam particles occupy an ellipse with n = 511:
Generally, at least 90% of the continuous-beam particles occupy an ellipse with n = 411:
The parameter w can be x, y, z or ϕ. Figure 13 is represented by the phase-space 2D projections
of a beam modelled by ~100 000 macro-particles. Ellipses in red correspond to ellipses calculated with
Eq. (108) with n = 5. They contain, in this example, 92% of the particles.

11
If the beam distribution were a uniform ellipse in real space, 100% of the particles would occupy this ellipse.

173
 N. P ICHOFF

Fig. 13: Phase-space beam distribution and Twiss parameters

4.2.3 Matched/mismatched beam

A beam is matched when its Twiss parameters at a given position s correspond to the transport channel
periodic Courant–Snyder parameters.
Matched beam : Beam’s Twiss parameters = Channel’s Courant–Snyder parameters.
In this condition, the same beam phase-space shape is reproduced period after period. The beam
envelope evolution with s is periodic and the smoothest possible. On Fig. 14 has been represented the
evolution of beams in the same FODO channel as in Fig. 9. In the middle line are plotted the ellipses
representing the beams in the phase space at the focusing quadrupole centre. The dashed black circle
represents a particle trajectory in this phase space. One matched (in continuous red) and two
mismatched (in dashed pink and dotted blue) beams are represented. One particle of each beam has
also been represented.

Fig. 14: Matched and mismatched beam in FODO channel

174
B EAM DYNAMICS BASICS IN RF LINACS

The matched beam ellipse is periodic, as one particle is replaced by an other one. Its envelope
(last line) is periodic with the lattice period L.
The mismatched beam ellipses sweep a bigger area (dashed-black circle) than the beam ellipse
surfaces. Their envelope period is greater than the lattice period. Its oscillation is a combination of two
oscillations with two different periods: one is the lattice period L, the other is 2π σ w ⋅ L , σw being the
channel phase advance per lattice.
– When the force is linear (Fig. 15), all particles oscillate in the phase-space with the same period
(i.e. the same phase advance per lattice). The beam phase-space distribution changes lattice after
lattice, but its emittance is kept constant.

– When the force is non-linear (Fig. 16) [external force or force induced by space-charge
(Coulomb interactions between beam particles)], the particle’s phase-advance per lattice
depends on its oscillation amplitude. Beam particles no longer turn all at the same angular speed
in the phase space, and an apparent emittance growth is observed12. This effect is known as
beam filamentation (Fig. 17). After a long time (many particle betatron periods), the phase
space swept by the beam is completely full of particles. The apparent emittance is higher.

X'
Hamiltonian = Cst
X'

X X

Matched beam Mismatched beam

Fig. 15: Matched (left) and mismatched (right) beam in linear forces

12
Even if the phase-space area occupied by the particle is kept constant (Liouville’s theorem applies).

175
 N. P ICHOFF

Courbes iso-hamiltoniennes
Hamiltonian = Cst X' Faisceau
Initial beam
initial
X'

X X

Faisceau final
Final beam
Faisceau adapté
Matched beam Faisceau désadapté
Mismatched beam

Fig. 16: Matched (left) and mismatched (right) beam in non-linear forces

2 2 2
r'/R'a
r'/R'a

r'/R'a

0 0 0

2 2 2

2 1 0 1 2 2 1 0 1 2 2 1 0 1 2
r/Ra r/Ra r/Ra

2 2 2
r'/R'a
r'/R'a

r'/R'a

0 0 0

2 2 2

2 1 0 1 2 2 1 0 1 2 2 1 0 1 2
r/Ra r/Ra r/Ra
Fig. 17: Filamentation of mismatched beam in non-linear force

176
B EAM DYNAMICS BASICS IN RF LINACS

5 Conclusion
This paper is a short introduction containing the first basic notions useful to lightly touch upon linacs.
A better understanding cannot be obtained without tackling subjects like general beam dynamics,
existing structures, RF control, space-charge effects, or resonances. Motivated students are strongly
advised to read Thomas Wangler’s book [2].

Acknowledgements
Many thanks to Tom Wangler for his wonderful and very useful book on linacs and his
encouragement, and to my colleagues Romuald Duperrier and Didier Uriot for their help.

References
[1] P.M. Lapostolle and A.L. Septier (Eds) Linear Accelerators (North Holland, Amsterdam, 1970).
[2] T.P. Wangler, Principles of RF Linear Accelerators (Wiley, New York, 1998).
[3] RF Engineering for Particle Accelerators, CERN Accelerator School, Oxford, UK, 3–10 April
1991, CERN 92-03, 2 vols.
[4] M. Weiss, Introduction to RF linear Accelerators, CERN Accelerator School, Fifth General
Accelerator Physics Course, Jyväskylä, Finland, 7–18 september 1992, CERN 94-01, 2 vols.
[5] T. Nishikawa, Transients and beam loading effect, Linear Accelerators, P.M. Lapostolle and
A.L. Septier (Eds) (North Holland, Amsterdam, 1970).
[6] A. Carne et al., Numericals methods. Acceleration by a gap, Linear Accelerators
P.M. Lapostolle and A.L. Septier (Eds) (North Holland, Amsterdam, 1970).
[7] J. Le Duff, Dynamics and acceleration in linear structures, CERN Accelerator School, Fifth
General Accelerator Physics Course, Jyväskylä, Finland, 7–18 September 1992, CERN 94-01,
2 vols.

177