week ending

PRL 106, 184801 (2011) PHYSICAL REVIEW LETTERS 6 MAY 2011

Nonlinear Longitudinal Space Charge Oscillations in Relativistic Electron Beams

P. Musumeci, R. K. Li, and A. Marinelli
Department of Physics and Astronomy, UCLA, Los Angeles, California, 90095, USA
(Received 16 February 2011; published 4 May 2011)
In this Letter we study the evolution of an initial periodic modulation in the temporal profile of a
relativistic electron beam under the effect of longitudinal space-charge forces. Linear theory predicts a
periodic exchange of the modulation between the density and the energy profiles at the beam plasma
frequency. For large enough initial modulations, wave breaking occurs after 1=2 period of plasma
oscillation leading to the formation of short current spikes. We confirm this effect by direct measurements
on a ps-modulated electron beam from an rf photoinjector. These results are useful for the generation of
intense electron pulse trains for advanced accelerator applications.

DOI: 10.1103/PhysRevLett.106.184801 PACS numbers: 29.27.Bd, 41.75.Ht, 52.35.
g

Nonlinear Langmuir waves are of fundamental interest
in plasma and accelerator physics [1]. Large amplitude
plasma waves can be excited either by the passage of a
high power electron or laser beam in a plasma or by
preparing the system with appropriate initial conditions
far from equilibrium and letting it evolve. A relativistic
electron beam is a particular case where one has a chance
to directly observe the dynamics of the large amplitude
plasma waves associated with density modulations in the
longitudinal beam profile. The study of such systems has
direct relevance and application to the generation and
control of intense bunch trains with high peak current
and fixed periodicity. This is important for applications
such as superradiant production of radiation [2], control of
the microbunching instability [3], and generation of power-
ful driver beams for resonant excitation of dielectric and/or
plasma wakefields in advanced accelerators [4,5].
In order to create a high frequency bunch train structure
in relativistic beams, several techniques have been studied.
These include the use of a transverse mask in a dispersive
region in combination with an energy chirp [5], or before a
longitudinal to transverse exchange beam line [6]. Many
authors have proposed to generate the beam modulation
directly at the cathode [7,8]. Boscolo et al. [9] have looked
into using rf velocity bunching as a way to compensate for
the longitudinal space-charge effects and recover the initial
distribution. In all these cases, the problem arises on how to
preserve the modulation for large enough beam currents
when longitudinal space-charge effects might contribute to
smearing out any structure in the temporal beam profile
[10] thus limiting the attainable peak currents in bunch
train mode.
As opposed to trying to contrast the action of the longi-
tudinal space-charge forces, we explore here what happens
to the beam temporal profile when the effect of the self-
forces is magnified by increasing the charge density (either
by using larger beam charges, or by tighter transverse
focusing). For a small initial modulation, the linear theory
of plasma oscillations predicts a periodic evolution of the
beam density with period set by the relativistic plasma
frequency. The density modulation reaches a minimum
after a propagation distance corresponding to a
=2
plasma phase advance, concurrently with a maximum in
energy modulation. In the absence of damping, which can
be caused by large uncorrelated energy spread or emit-
tance, such energy modulation will be reconverted back
into density modulation after another 1=4 period of plasma
oscillation and so on.
In this Letter we study theoretically and experimentally
the evolution of a large initial density modulation in the
temporal profile of a relativistic electron beam. For large
enough initial modulation, the linear theory breaks down
and the dynamics is more complicate due to the appearance
of harmonics of the modulation frequency along the beam
propagation. The harmonics interfere constructively after
1=2 period of plasma oscillation leading to the formation of
high peak current spikes. This suggests that in order to
obtain very high peak current bunch trains, one could
simply increase the beam charge density and let a premo-
dulated beam propagate for a distance equivalent to 1=2
plasma oscillation period under the action of its own space-
charge forces. The physics of the problem is well described
by a simple 1D cold fluid model which is very similar to the
one employed in the study of large amplitude oscillations
in plasma physics by Davidson et al. [11], with the proper
modifications due to the beam being a non-neutral relativ-
istic plasma. The current spike formation is associated to
the wave-breaking phenomenon when particles overtake
the wave in longitudinal phase space. Fully 3D particle
tracking simulations are employed to evaluate the effects
of finite beam length, beam energy spread, finite transverse
beam size and particle transverse motion, and show a very
good agreement with the 1D model prediction. The validity
of the scheme is demonstrated by the results of an experi-
ment performed at the UCLA Pegasus photoinjector,
where we launch on the cathode a THz-modulated beam
and directly observe the evolution of the temporal profile
with an rf deflector.

0031-9007=11=106(18)=184801(4) 184801-1 Ó 2011 American Physical Society

 week ending
PRL 106, 184801 (2011) PHYSICAL REVIEW LETTERS 6 MAY 2011

To better understand the evolution of the density of a b = 0.01 b = 0.24 1st

Bunching factor
1.10 3rd
9 0.2
9th
1.08
relativistic beam with an initial modulation, one can start 1.06
Initial
After Tp /4 8
After Tp /2 7 0.1
from a 1D cold fluid model. The system can be written as 1.04
6

Density
1.02

Density

 5 0.0
eE 1.00 0.00 0.25 0.50

@t
 þ c @z
 ¼

t/ Tp
4
; (1) 0.98
3 mc 0.96 3
2
0.94
0.92 1
en 0.90 0
@z E ¼
; (2) 0 1 2 3 4 5 6 0 1 2 3 4 5 6
0 kz / 2π kz / 2π

FIG. 1 (color online). Beam density distribution for two differ-


@t n þ c@z n ¼ 0; (3)
3
where the three unknown functions nðz; tÞ,
ðz; tÞ and
Eðz; tÞ describe the electronic density, the deviation from
the reference energy and the longitudinal space-charge
field, respectively. z is the coordinate along the bunch,
e and m are the electron charge and mass and  is the
beam relativistic factor. We assume
  .
Equations (1)–(3) represent a coupled system of non-
linear partial differential equations. For an initial periodic
modulation with period
P  ¼ 2
=k, we can use the solution
ansatz Aðz; tÞ ¼ m Am ðtÞeimkz and, collecting the terms
which have equal orders of m in the exponent, we reduce
the problem to a system of ordinary differential equations.
Equation (2) reduces to Em ¼
enm =imk0 . By scaling
the equations to dimensionless variables, we obtain the
following system of equations
X PARTICLE TRACER (GPT) code [12] are used to validate
@ Gm ¼
nGn Gm
n þ Rm ; (4)
n
X density to facilitate the comparison with the theory.
@ Rm ¼
m Gn Rm
n ; (5)
n We consider a 15 MeV beam with charge 200 pC,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where  ¼ !p t and !p ¼ e2 n0 =m0 3 is the rela-
tivistic plasma frequency. Rm ¼ nm =n0 and Gm ¼
ikc
m =!p 3 . As no parameters appear in this dimen-
sionless formulation of the problem, the evolution of the
system is fully determined by the initial conditions.
Further, this shows how one could equivalently study the
dynamics of the system by looking at different propagation
times, or by varying !p .
For a sinusoidal modulation, the initial charge density
can be written as nðz; 0Þ ¼ n0 ð1 þ 2b cosðkzÞÞ, where
b ¼ heikz i is the bunching factor. In normalized variables,
this can be expressed as R0 ð0Þ ¼ 1 and R1 ð0Þ ¼ b. In
Fig. 1 we show the solutions of the system for different
values of the initial condition b. For small initial modula-
tion (b ¼ 0:01), the results of the linear theory are recov-
ered. After Tp =4, where Tp ¼ 2
=!p is the plasma
oscillation period, the temporal profile is uniform. After
Tp =2 the initial modulation is completely recovered with a

phase shift. As the amplitude of the initial bunching is
increased the dynamics change and harmonics of the fun-
damental wavelength appear. For b ¼ 0:24 the beam den-
sity profile shows current spikes much greater than the
initial peak current at Tp =2. The ratio of the peak current
184801-2
ent initial conditions at the beginning, after Tp =4 and after Tp =2.
The system of Eqs. (4) and (5) has been solved numerically
including contributions up to the 64th harmonic. In the inset we
show the evolution of the bunching factor for the 1st, 3rd, and
9th harmonic.
and the average beam current is larger than 8, with a
significant increase with respect to its initial value of
1.48. We note that even in the nonlinear case the beam
density is always uniform at 1=4 plasma period. In the inset
in Fig. 1 we also show the bunching factors for the 1st, 3rd,
and 9th harmonic. At Tp =2, the first harmonic bunching
returns to a level comparable with its initial value.
Bunching is significant even at high harmonic number.
For b > 0:25, the wave-breaking condition is achieved
and the cold fluid results are no longer valid[11].
Three dimensional simulations using the GENERAL
this model. As an example, we study the ideal case of a
coasting beam with no acceleration and a constant charge
FWHM bunch length 200 m, and rms transverse size
x ¼ 100 m with an initial modulation with period
 ¼ 20 m and amplitude b ¼ 0:25. The transverse
beam emittance is 0.1 mm-mrad and the uncorrelated
energy spread is 0.1%. The beam transverse size is main-
tained constant along the propagation by a focusing sole-
noid channel and no longitudinal focusing is applied.
As expected, 3D effects related to the aspect ratio of the
space-charge field distribution modify the absolute value of
the plasma frequency. In particular, if D ¼ kx = < 1, the
longitudinal component of the space-charge forces and
hence the frequency of the oscillations decrease. From
simulations, we find a scaling for the effective 3D plasma
2D
frequency !p3D ¼ 1þ2D !p , in agreement with recent theo-
retical predictions [13]. These effects only contribute to a
relative slow down of the dynamics and do not alter the
characteristic features of the wave evolution.
The beam longitudinal phase spaces and temporal pro-
files at two different locations along the beam line are
shown in Fig. 2. In these simulations, as well as in any
practical case where the beam has a finite bunch length, the
plasma oscillations only occur in the central core of the
beam, while at the beam head and tail the dynamics is
significantly different due to the lack of a restoring force.

PRL 106, 184801 (2011) PHYSICAL REVIEW LETTERS
FIG. 2 (color online). Longitudinal phase space from the
GPT 3D simulation of a coasting beam case. The particles are
color coded with their radial positions. Particles at higher radius
experience a lower longitudinal space-charge field and advance
slower in the plasma dynamics.
In agreement with the predictions, at z ¼ 0:21 m corre-
sponding to a
=2 phase advance of the 3D plasma fre-
quency, the energy modulation is maximum and the
density modulation is zero. At z ¼ 0:42 m the initial
modulation appears again with an increased harmonic
content and the longitudinal phase space distribution shows
the steepened wave and the particles piling up at the wave-
breaking points. The current spikes in these simulations
reach 1.5 kAmps, 6 times larger than the initial ave-
rage beam current. This current enhancement is slightly
smaller than the 1D model prediction due to the uncorre-
lated energy spread and emittance smearing effects.
Nevertheless, the agreement is remarkable and shows
how the simple dimensionless 1D model essentially cap-
tures the physics of the wave evolution.
The increase in the harmonic content of the modulation
and the large peak current enhancement suggest an inter-
esting possibility of taking advantage of collective beam
forces. In order to generate intense electron bunch trains,
one could start from a large enough initial modulation and
adopt the counterintuitive solution of increasing the charge
in the beam and wait for a propagation distance corre-
sponding to 1=2 plasma oscillation. We performed an
experiment at the UCLA Pegasus Laboratory to test these
ideas. The cathode in a 1.6 cell rf photoinjector is illumi-
nated by a train of 8 equally spaced laser pulses created
using 3 -BBO birefringent crystals [14] of thicknesses 1,
2 and 4 mm. The induced periodicity has frequency 1 THz
and wavelength  ¼ 300 m.
The total charge in the beam is limited by the available
laser energy after the crystals to 40 pC. In order to accel-
erate the plasma oscillation dynamics with this relatively
low charge, the gun solenoid is used to focus the beam to a
tight spot at about 80 cm from the cathode. The beam is
then refocused by a quadrupole and streaked using a
9.6 GHz 500 kV rf deflecting cavity located at a distance
of 1.8 m from the cathode. We vary the plasma frequency
by controlling the laser energy on the cathode and hence
the photoemitted charge. The solenoid field is kept
184801-3
week ending
6 MAY 2011
1.2 1.6 pC
3.9 pC 0.50
1.0 15 pC
RMS size (mm)
0.8
Phase
0.6
0.25
0.4
0.2
0.0 0.00
0.0 0.5 1.0 1.5 2.0
z (m)
FIG. 3 (color online). GPT simulation of the Pegasus beam
line. Evolution of the rms transverse beam size for three different
beam charges (solid lines). The corresponding evolution of the
plasma phase advance calculated using the beam density from
the simulations is also shown.
constant during the measurements. The quadrupole focus-
ing is slightly adjusted each time the charge is changed to
optimize the width of the unstreaked spot on the detection
screen downstream of the deflector.
In Fig. 3 we show the GPT simulation of the evolution of
the rms beam transverse size up to the deflector. From the
simulation, it is possible
R !p3Dto retrieve the beam density and
perform the integral c dz to calculate the plasma phase
advance for different beam charges. When doing this, it is
important to keep track of the plasma frequency reduction
due to the 3D geometric effects in the region close to the
beam focus. As seen in Fig. 3, most of the plasma phase
advance occurs in the region close to the cathode and at the
beam focus.
The images obtained on the screen after the rf deflector
are shown in Fig. 4. The rf gun phase controls the overall
microbunch spacing and in order to preserve the 1-ps
periodicity was set at 35. The residual rf curvature is
responsible for the different spacing of the bunches at
the beam head and tail. For very low charge (1.6 pC)
the temporal beam profile is a good approximation of the
initial modulation. As the charge is increased to 3.9 pC,
corresponding to a
=2 plasma phase advance, the profile
becomes uniform. For larger charges, the modulation
reappears. Note that only 7 peaks are visible. This is ex-
pected, since, due the
phase shift in plasma oscillation,
the spikes arise from the valleys of the original distribution.
The temporal resolution of the measurement is limited
by the unstreaked beam size and by the spatial resolution of
the screen. In order to demonstrate the increased harmonic
content in the spikes, we repeat the measurement at a
downstream screen. In this case, due to the finite transverse
dimensions of the screen, it is not possible to capture the
full length of the beam in a single shot. Zooming in on a
small temporal region covering only few of the spikes we
obtain the results shown in Fig. 5. The current spikes in the
16 pC case (right inset) are significantly narrower than the
1.6 pC image (left inset). In the graph showing the streak
 week ending
PRL 106, 184801 (2011) PHYSICAL REVIEW LETTERS 6 MAY 2011

1.6 pC 1.6 pC 16 pC
a) 1.6 pC 2 mm b) 3.9 pC 16 pC
140 GPT simulation
Convolution of simulation
120 and screen resolution

100

Arb. units
80

60

40

20
c) 9.0 pC d) 22 pC
0
0.0 0.5 1.0 1.5 2.0 2.5
Time (ps)

FIG. 5 (color online). Close-up view of the spikes on the high

resolution screen. The lineouts also show the simulation results
and the convolution of the simulations with the screen spatial
resolution.

20 1.6 pC e) content. These results demonstrate the possibility to con-

intensity ( arb. units)

3.9 pC
9.0 pC
15
22 pC
10
5 dielectric wakefield acceleration.
0 UCLA Pegasus Laboratory for their help. This work was
0 50 100 150 200
y (pixel)
FIG. 4 (color online). Bunch train streak-images for four dif-
ferent charges. The lineouts of the beam temporal profiles are
also shown. The FWHM width of the unstreaked spot size on this
screen is 17 pixels.
lineouts, the vertical scales are normalized to yield a
constant average beam current. In the GPT simulation,
the density spike has a peak current almost twice larger
than the measurement, suggesting that even in these con-
ditions the temporal resolution is not enough to resolve the
peak width. If we convolve the simulation result with
the unstreaked spot dimensions we retrieve almost exactly
the measured profile.
In conclusion, in this Letter we explored theoretically
and experimentally the physics of nonlinear longitudinal
space-charge oscillations. The nonlinear theory of plasma
waves has been applied to describe the evolution of large
initial modulations on the density of a relativistic electron
beam. We demonstrated experimentally the validity of our
theoretical model by imposing a ps-initial modulation on
the beam from the cathode and directly observing its
evolution with an rf streak camera with subps temporal
resolution. The plasma phase advance is controlled by
tuning the beam charge and focusing. As predicted in
shot-noise reduction schemes [15], after 1=4 oscillation
the initial modulation is completely washed out and one
obtains a uniform temporal profile. After 1=2 oscillation
the modulation reappears, with enhanced harmonic
trol nonlinear longitudinal space-charge oscillations and
open the way to optimization of this scheme for the gen-
eration of high peak current bunch trains for applications in
the field of coherent radiation generation, and plasma and
The authors want to acknowledge all the members of the
250 300
supported by DOE grants DE-FG02-92ER40693 and DE-
FG02-07ER46272 and ONR grant N000140711174.
[1] J. M. Dawson, Phys. Rev. 113, 383 (1959).
[2] A. Gover, Phys. Rev. ST Accel. Beams 8, 030701 (2005).
[3] E. Saldin et al., Nucl. Instrum. Methods Phys. Res., Sect.
A 528, 355 (2004); Z. Huang et al., Phys. Rev. ST Accel.
Beams 7, 074401 (2004).
[4] C. Jing et al., Phys. Rev. Lett. 98, 144801 (2007).
[5] P. Muggli, V. Yakimenko, M. Babzien, E. Kallos, and K. P.
Kusche. Phys. Rev. Lett. 101, 054801 (2008).
[6] Y. E. Sun, P. Piot, A. Johnson, A. H. Lumpkin, T. J.
Maxwell, J. Ruan, and R. Thurman-Keup, Phys. Rev.
Lett. 105, 234801 (2010).
[7] Y. Li and K. J. Kim, Appl. Phys. Lett. 92, 014101 (2008).
[8] J. G. Neumann, R. B. Fiorito, P. G. O’Shea, H. Loos, B.
Sheehy, Y. Shen, and Z. Wu, J. Appl. Phys. 105, 053304
(2009).
[9] M. Boscolo, M. Ferrario, I. Boscolo, F. Castelli, and S.
Cialdi, Nucl. Instrum. Methods Phys. Res., Sect. A 577,
409 (2007).
[10] J. R. Harris, J. G. Neumann, K. Tian, and P. G. O’Shea,
Phys. Rev. E 76, 026402 (2007).
[11] R. C. Davidson. in Methods in Nonlinear Plasma Theory
(Academic Press, NY, 1972).
[12] http://www.pulsar.nl/gpt.
[13] A. Marinelli et al., Phys. Rev. E (to be published).
[14] P. Musumeci et al., J. Appl. Phys. 108, 114513 (2010).
[15] A. Gover and E. Dyunin, Phys. Rev. Lett. 102, 154801
(2009).

184801-4