Nonlinear Dynamics: Part 1

Andy Wolski

The Cockcroft Institute, and the University of Liverpool, UK

CAS: Introduction to Accelerator Physics

Budapest, Hungary
October 2016
 Nonlinear Dynamics

In these two lectures on nonlinear dynamics, we shall introduce

a number of topics:

• Mathematical tools for modelling nonlinear dynamics:

– power series (Taylor) maps; symplectic maps.

• Effects of nonlinear perturbations:

– resonances; tune shifts; dynamic aperture.

• Analysis methods:
– normal form analysis; frequency map analysis.

We shall discuss these aspects of the subject in the context of

two types of accelerator system:

1. a bunch compressor (a single-pass system);

2. a storage ring (a multi-turn system).

CAS, Budapest, 2016 1 Nonlinear Dynamics: Part 1

 Nonlinear dynamics: goals of the first lecture

Our aim is to provide an introduction to some of the key

concepts of nonlinear dynamics in particle accelerators.

By the end of the first lecture, you should be able to:

• describe some of the sources of nonlinearities in particle

accelerators;

• outline some of the tools used for modelling nonlinear

dynamics in accelerators;

• explain the significance and potential impact of nonlinear

dynamics in some accelerator systems.

CAS, Budapest, 2016 2 Nonlinear Dynamics: Part 1

 Introduction: from linear to nonlinear dynamics

Particle motion through simple components such as drifts,

dipoles and quadrupoles can be represented by linear transfer
maps.

For example, in a drift space:

x1 = x0 + Lpx0, (1)
px1 = px0, (2)
where x0 and px0 are the horizontal co-ordinate and (scaled)
horizontal momentum at the entrance of the drift space; x1
and px1 are the horizontal co-ordinate and momentum at the
exit of the drift space, and L is the length of the drift space.

Note that:
γmvx dx
px = ≈ , (3)
P0 ds
where γ is the relativistic factor, m is the rest mass of the
particle, vx is the horizontal velocity, and P0 is the reference
momentum.
CAS, Budapest, 2016 3 Nonlinear Dynamics: Part 1
 Introduction: from linear to nonlinear dynamics

Linear transfer maps can be written in terms of matrices.

For example, for a drift space of length L:

! ! !
x1 1 L x0
= . (4)
px1 0 1 px0

In general, a linear transformation can be written:

~
x1 = R ~ ~
x0 + A, (5)
where ~
x0 and ~
x1 are the initial and final phase space vectors,
with components (x0, px0) and (x1, px1), respectively.

~ is a vector.
R is a matrix (the transfer matrix ) and A

~ are constant, i.e. they do not

The components of R and A
depend on ~
x0.

CAS, Budapest, 2016 4 Nonlinear Dynamics: Part 1

 Introduction: from linear to nonlinear dynamics

The transfer matrix for a section of beamline can be found by

multiplying the transfer matrices for the accelerator
components within that section.

For a periodic beamline (i.e. a beamline constructed from a

repeated unit) the transfer matrix for a single period can be
parameterised in terms of the Courant–Snyder parameters
(α, β, γ) and the phase advance, µ:
!
cos(µ) + α sin(µ) β sin(µ)
R= . (6)
−γ sin(µ) cos(µ) − α sin(µ)

CAS, Budapest, 2016 5 Nonlinear Dynamics: Part 1

 Introduction: from linear to nonlinear dynamics

If the beamline is stable, then the characteristics of the particle

motion can be represented by a phase space portrait showing
the co-ordinates and momenta of a particle after an increasing
number of passes through full periods of the beamline.

2 2

1.5 1.5
1 1
p x (10 −3 )

p y (10 −3)
0.5 0.5

0 0
−0.5 −0.5
−1 −1

−1.5 −1.5

−2 −2
−2 −1 0 1 2 −2 −1 0 1 2
x (mm) y (mm)

CAS, Budapest, 2016 6 Nonlinear Dynamics: Part 1

 Introduction: from linear to nonlinear dynamics

If the transfer map for each

period is linear, then the phase
space portrait is an ellipse with
area πJx.

Jx is called the betatron action,

and characterises the amplitude
of the betatron oscillations.

The shape of the ellipse is

described by the
Courant–Snyder parameters.

The rate at which particles

move around the ellipse (phase
advance per period) is
independent of the betatron
action.

CAS, Budapest, 2016 7 Nonlinear Dynamics: Part 1

 Introduction: from linear to nonlinear dynamics

Nonlinearities in the particle dynamics can come from a number

of different sources, including:

• stray fields in drift spaces;

• higher-order multipole components in dipoles and

quadrupoles;

• higher-order multipole magnets (sextupoles, octupoles...)

used to control various properties of the beam;

• effects of fields generated by a bunch of particles on

individual particles within the bunch (space-charge forces,
beam-beam effects...)

The effects of nonlinearities can be varied and quite dramatic.

It is important to have some understanding of nonlinear
dynamics for optimising the design and operation of many
accelerator systems.
CAS, Budapest, 2016 8 Nonlinear Dynamics: Part 1
 Nonlinear transfer map: sextupole magnet

As an example, consider (the vertical component of) the field

in a sextupole magnet:
By 1
= k 2 x2 , (7)
Bρ 2
where Bρ = P0/q is the beam rigidity, and k2 is the normalised
sextupole gradient.

In the “thin lens” approximation, the deflection of a particle on

passing through the sextupole is:
1 1
Z
∆px = − By ds ≈ − k2Lx2, (8)
Bρ 2
where L is the length of the sextupole.

Hence, the transfer map for a sextupole in the thin lens

approximation is:
x1 = x0 , (9)
1
px1 = px0 − k2Lx2. (10)
2
CAS, Budapest, 2016 9 Nonlinear Dynamics: Part 1
 Nonlinear transfer maps: power series representation

A nonlinear transfer map can be represented as a power series:

x1 = A1 + R11x0 + R12px0 + T111x2 2

0 + T112 x0 px0 + T122 px0 + . . .
(11)
px1 = A2 + R21x0 + R22px0 + T211x2 2
0 + T212 x0 px0 + T222 px0 + . . .
(12)

The coefficients Rij are components of the transfer matrix R.

The coefficients of higher-order (nonlinear) terms are

conventionally represented by Tijk (second order), Uijk` (third
order) and so on.

The values of the indices correspond to the components of the

phase space vector, thus:

index value 1 2 3 4 5 6
component x px y py z δ
CAS, Budapest, 2016 10 Nonlinear Dynamics: Part 1
 Effects of nonlinearities

Nonlinearities in a periodic beamline can have a number of

effects:

• the shape of the phase space ellipse can become distorted;

• the phase advance per period can depend on the betatron

amplitude (i.e. depends on the action Jx);

• the motion can be stable for small amplitude, but unstable

at large amplitude;

• features such as “phase space islands” (closed loops around

points away from the origin) can appear in the phase space
portrait...

CAS, Budapest, 2016 11 Nonlinear Dynamics: Part 1

 Effects of nonlinearities

We shall discuss the effects of nonlinearities in periodic

beamlines in the second lecture.

In the remainder of this lecture, we shall look in more detail at

the effects of nonlinearities in a single-pass beamline: a bunch
compressor.

We shall see how nonlinear effects can impact the performance

of a bunch compressor if they are not properly taken into
account in the design of the system.
CAS, Budapest, 2016 12 Nonlinear Dynamics: Part 1
 Nonlinear effects in a bunch compressor

A bunch compressor reduces the length of a bunch, by

performing a rotation in longitudinal phase space.

Bunch compressors are used, for example, in free electron

lasers to increase the peak current.

We shall follow these steps in our analysis:

1. Outline the structure of the bunch compressor.

2. Specify the parameters based on linear dynamics.

3. Perform an analysis of the linear and nonlinear effects.

4. Adjust the parameters to compensate nonlinear effects.

CAS, Budapest, 2016 13 Nonlinear Dynamics: Part 1

 Bunch compressor: structure and operation

CAS, Budapest, 2016 14 Nonlinear Dynamics: Part 1

 Bunch compressor: structure and operation

The rf cavity is designed to “chirp” the bunch, i.e. to provide a

change in energy deviation as a function of longitudinal
position z within the bunch (z > 0 at the head of the bunch).

The energy deviation δ of a particle with energy E is defined as:

E − E0
δ= , (13)
E0
where E0 is the reference energy for the system.

The transfer map for the rf cavity in the bunch compressor is:

z1 = z0 , (14)
eV ωz0
 
δ1 = δ0 − sin , (15)
E0 c
where V is the rf voltage, and ω/2π is the rf frequency.

CAS, Budapest, 2016 15 Nonlinear Dynamics: Part 1

 Bunch compressor: structure and operation

Neglecting synchrotron radiation, the chicane does not change

the energy of the particles. However, the path length L
depends on the energy of the particle.

If we assume that the bending angle in a dipole is small, θ  1:

2L1
L= + L2. (16)
cos θ
The bending angle is a function of the energy of the particle:
θ0
θ= . (17)
1+δ
CAS, Budapest, 2016 16 Nonlinear Dynamics: Part 1
 Bunch compressor: structure and operation

The change in the co-ordinate z is the difference between the

nominal path length, and the length of the path actually taken
by the particle.

Hence, the transfer map for the chicane can be written:

!
1 1
z2 = z1 + 2L1 − , (18)
cos θ0 cos(θ(δ1))
δ2 = δ1 , (19)
where θ0 is the nominal bending angle of each dipole in the
chicane, and θ(δ) is given by (17):
θ0
θ(δ) = .
1+δ

Clearly, the complete transfer map for the bunch compressor is

nonlinear; but how important are the nonlinear terms?

CAS, Budapest, 2016 17 Nonlinear Dynamics: Part 1

 Bunch compressor: linear dynamics

To understand the effects of the nonlinear part of the map, we

shall look at a specific example.

First, we will “design” a bunch compressor using only the linear

part of the map.

The linear part of a transfer map can be obtained by expanding

the map as a Taylor series in the dynamical variables, and
keeping only the first-order terms.

After finding appropriate values for the various parameters

using the linear transfer map, we shall see how our design has
to be modified to take account of the nonlinearities.

CAS, Budapest, 2016 18 Nonlinear Dynamics: Part 1

 Bunch compressor: linear dynamics

To first order in the dynamical variables z and δ, the map for

the rf cavity can be written:

z1 = z0 , (20)
δ1 = δ0 + R65z0, (21)
where:
eV ω
R65 = − . (22)
E0 c

The map for the chicane can be written:

z2 = z1 + R56δ1, (23)
δ 2 = δ1 , (24)
where:
θ0 sin θ0
R56 = 2L1 . (25)
cos2 θ0

CAS, Budapest, 2016 19 Nonlinear Dynamics: Part 1

 Bunch compressor: linear dynamics

As a specific example, consider a bunch compressor for the

International Linear Collider:
q
Initial rms bunch length hz02i 6 mm
q
Initial rms energy spread hδ02i 0.15%
q
Final rms bunch length hz22i 0.3 mm

Two constraints determine the values of R65 and R56:

• The bunch length should be reduced by a factor 20.

• There should be no “chirp” on the bunch at the exit of the

bunch compressor, i.e. hz2δ2i = 0.

With these constraints, we find (see Appendix A):

R65 = −4.9937 m−1, and R56 = 0.19975 m. (26)

CAS, Budapest, 2016 20 Nonlinear Dynamics: Part 1

 Bunch compressor: linear dynamics

We can illustrate the effect of the linearised bunch compressor

map on phase space using an artificial “window frame”
distribution:

The rms bunch length is reduced by a factor of 20 as required,

but the rms energy spread is increased by the same factor.
This is because the transfer map is symplectic, so phase space
areas are conserved under the transformation.

CAS, Budapest, 2016 21 Nonlinear Dynamics: Part 1

 Bunch compressor: nonlinear dynamics

Now let us see what happens when we apply the full nonlinear
map for the bunch compressor.

The full map cannot simply be represented by the two

coefficients R65 and R56: we need to make some assumptions
for the rf voltage and frequency, and the dipole bending angle
and chicane length.

We have to choose all these parameters so that the “linear”

parameters have the appropriate values.

Beam (reference) energy E0 5 GeV

RF frequency frf 1.3 GHz
RF voltage Vrf 916 MV
Dipole bending angle θ0 3◦
Dipole spacing L1 36.3 m

CAS, Budapest, 2016 22 Nonlinear Dynamics: Part 1

 Bunch compressor: nonlinear dynamics

As before, we illustrate the effect of the bunch compressor map

on phase space using a “window frame” distribution:

Although the bunch length has been reduced, there is

significant distortion of the distribution: the rms bunch length
will be significantly longer than we are aiming for.

CAS, Budapest, 2016 23 Nonlinear Dynamics: Part 1

 Bunch compressor: nonlinear dynamics

To reduce the distortion, we first need to understand where it

comes from, which means looking at the map more closely.

Consider a particle entering the bunch compressor with initial

phase space co-ordinates z0 and δ0. We can write the
co-ordinates z1 and δ1 of the particle after the rf cavity to
second order in z0 and δ0:

z1 = z0 , (27)
δ1 = δ0 + R65z0 + T655z02. (28)

Recall the notation for the coefficients in the map: the first
subscript indicates the variable on the left hand side of the
equation, and subsequent subscripts indicate the variables in
the relevant term.

CAS, Budapest, 2016 24 Nonlinear Dynamics: Part 1

 Bunch compressor: nonlinear dynamics

The co-ordinates of the particle after the chicane are then (to
second order):

z2 = z1 + R56δ1 + T566δ12, (29)

δ2 = δ1 . (30)

If we combine the maps for the rf and the chicane, we get:

z2 = (1 + R56R65)z0 + R56δ0
2 T
+(R56T655 + R65 2
566 )z0
+2R65T566z0δ0
+T566δ02, (31)

δ2 = δ0 + R65z0 + T655z02. (32)

CAS, Budapest, 2016 25 Nonlinear Dynamics: Part 1

 Bunch compressor: nonlinear dynamics

The term that gives the strong nonlinear distortion is the term
in z02 in (31). If we can design a system such that the
appropriate coefficients satisfy:
2 T
R56T655 + R65 566 = 0, (33)
then we should be able to reduce the distortion.

The values of R56 and R65 are determined by the required

compression factor.

The value of T566 is determined by the chicane; by expanding

(18) as a Taylor series in δ, we find for θ0  1:
T566 ≈ −3L1θ02. (34)

That leaves us with T655. This is the second-order dependence

of the energy deviation on longitudinal position for a particle
passing through the rf cavity. But if we inspect the full rf map
(15), we find it contains only odd-order terms, unless...
CAS, Budapest, 2016 26 Nonlinear Dynamics: Part 1
 Bunch compressor: nonlinear dynamics

...we operate the rf cavity off-phase. In other words, we have

to modify the rf transfer map to:

z1 = z0 , (35)
eV ωz0
 
δ1 = δ0 − sin + φ0 . (36)
E0 c
The first-order coefficient in the transfer map for δ is then:
eV ω
R65 = − cos φ0. (37)
E0 c
The second-order coefficient is:
1 eV ω 2
 
T655 = sin φ0. (38)
2 E0 c
Note that there is also a zeroth-order term, so the bunch ends
up with a non-zero mean energy deviation hδi after the rf
cavity; but we can take this into account simply by an
appropriate scaling of the field in the chicane.

CAS, Budapest, 2016 27 Nonlinear Dynamics: Part 1

 Bunch compressor: nonlinear dynamics

The linear coefficients are determined by the required

compression factor, and the requirement to have zero final
correlation hzδi. For the ILC bunch compressor:

R65 = −4.9937 m−1, and R56 = 0.19975 m. (39)

The value of T566 is determined by the parameters of the

chicane:
T566 ≈ −3L1θ02 = −0.29963 m. (40)

And the value of T655 is determined by the need to correct the

second-order distortion of the phase space:
2 T
R56T655 + R65 566 = 0 ∴ T655 = 37.406 m−2. (41)

CAS, Budapest, 2016 28 Nonlinear Dynamics: Part 1

 Bunch compressor: nonlinear dynamics

Now, given:
eV ω
R65 = − cos φ0 = −4.9937 m−1, (42)
E0 c
and:
1 eV ω 2
 
T655 = sin φ0 = 37.406 m−2, (43)
2 E0 c
we find, for E0 = 5 GeV and ω = 1.3 GHz:

V = 1, 046 MV, and φ0 = 28.8◦. (44)

Operating with this phase, we are providing over a gigavolt of
rf to decelerate the beam by more than 500 MV.
Because of adiabatic (anti)damping, we will need to reduce the R56 of the
chicane by a factor E1 /E0 , where E0 and E1 are the mean bunch energy
before and after the rf, respectively.

Also, the phase space area occupied by the distribution will be increased by
a factor E0 /E1 .

CAS, Budapest, 2016 29 Nonlinear Dynamics: Part 1

 Bunch compressor: nonlinear dynamics

As before, we illustrate the effect of the bunch compressor on

phase space using a “window frame” distribution. But now we
use the parameters determined above, to try to compress by a
factor 20, while minimising the second-order distortion:

This looks much better: the dominant distortion now appears

to be third-order, and looks small enough that it should not
significantly affect the performance of the machine.

CAS, Budapest, 2016 30 Nonlinear Dynamics: Part 1

 Bunch compressor: some conclusions

We have already learned some useful lessons from this example:

• Nonlinear effects can limit the performance of an

accelerator system. Sometimes the effects are small enough
that they can be ignored; however, in many cases, a system
designed without taking account of nonlinearities will not
achieve the specified performance.

• If we take the trouble to analyse and understand the

nonlinear behaviour of a system, then, if we are fortunate
enough and clever enough, we may be able to devise a
means of compensating any adverse effects.

CAS, Budapest, 2016 31 Nonlinear Dynamics: Part 1

 Nonlinear Dynamics, Part I: Summary

• Nonlinear effects can arise from a number of sources in

accelerators, including stray fields, higher-order multipole
components in magnets, space-charge...

• The transfer map for a nonlinear element (such as a

sextupole) may be represented as a power series in the
initial values of the phase space variables.

• The effects of nonlinearities in accelerator system vary

widely, depending on the type of system in which they
occur (e.g. single-pass, or periodic).

• In some cases, nonlinear effects can limit the performance

of an accelerator system. In such cases, it is important to
take nonlinearities into account in the design of the system.

CAS, Budapest, 2016 32 Nonlinear Dynamics: Part 1

 Nonlinear Dynamics

Appendix

CAS, Budapest, 2016 33 Nonlinear Dynamics: Part 1

 Appendix 1.A: Longitudinal dynamics in a bunch compressor

In a linear approximation, the maps for the rf cavity and the

chicane in a bunch compressor may be represented as matrices:
! !
1 0 1 b
Mrf = , Mch = , (45)
−a 1 0 1
where:
eV ω θ sin θ
a= , and b = 2L1 0 2 0 . (46)
E0 c cos θ0

The matrix representing the total map for the bunch

compressor, Mbc, is then:
! !
1 − ab b R55 R56
Mbc = MchMrf = = . (47)
−a 1 R65 R66

The effect of the map is written:

!
z
z 7→ Mbc~
~ z, where ~
z= . (48)
δ

CAS, Budapest, 2016 34 Nonlinear Dynamics: Part 1

 Appendix 1.A: Longitudinal dynamics in a bunch compressor

Now we proceed to derive expressions for the required values of

the parameters a and b, in terms of the desired initial and final
bunch length and energy spread.

We construct the beam distribution sigma matrix by taking the

outer product of the phase space vector for each particle, then
averaging over all particles in the bunch:
!
hz 2i hzδi
Σ = h~
z~z Ti = . (49)
hzδi hδ 2i

The transformation of Σ under a linear map represented by a

matrix M is given by:

Σ 7→ M · Σ · M T. (50)

CAS, Budapest, 2016 35 Nonlinear Dynamics: Part 1

 Appendix 1.A: Longitudinal dynamics in a bunch compressor

Usually, a bunch compressor is designed so that the correlation

hzδi = 0 at the start and end of the compressor. In that case,
using (47) for the linear map M , and (50) for the
transformation of the sigma matrix, we find that the
parameters a and b must satisfy:
a hδ02i
(1 − ab) = 2 (51)
b hz0 i
where the subscript 0 indicates that the average is taken over
the initial values of the dynamical variables.

Given the constraint (51), the compression factor r is given by:

2 hz12i
r ≡ 2 = 1 − ab, (52)
hz0 i
where the subscript 1 indicates that the average is taken over
the final values of the dynamical variables.

CAS, Budapest, 2016 36 Nonlinear Dynamics: Part 1

 Appendix 1.A: Longitudinal dynamics in a bunch compressor

We note in passing that the linear part of the map is

symplectic. A linear map is symplectic if the matrix M
representing the map is symplectic, i.e. satisfies:

M T · S · M = S, (53)
where, in one degree of freedom (i.e. two dynamical variables),
S is the matrix:
!
0 1
S= . (54)
−1 0

In more degrees of freedom, S is constructed by repeating the

2 × 2 matrix above on the block diagonal, as often as necessary.

In one degree of freedom, it is a necessary and sufficient

condition for a matrix to be symplectic, that it has unit
determinant: but this condition does not generalise to more
degrees of freedom.
CAS, Budapest, 2016 37 Nonlinear Dynamics: Part 1
 Appendix 1.A: Longitudinal dynamics in a bunch compressor

As a specific example, consider a bunch compressor for the

International Linear Collider:
q
Initial rms bunch length hz02i 6 mm
q
Initial rms energy spread hδ02i 0.15%
q
Final rms bunch length hz12i 0.3 mm

Solving equations (51) and (52) with the above values for rms
bunch lengths and energy spread, we find:

a = 4.9937 m−1, and b = 0.19975 m. (55)

CAS, Budapest, 2016 38 Nonlinear Dynamics: Part 1