VOLUME 86, NUMBER 6 PHYSICAL REVIEW LETTERS 5 FEBRUARY 2001

Plasma Electron Trapping and Acceleration in a Plasma Wake Field Using a Density Transition
H. Suk, N. Barov, and J. B. Rosenzweig
Department of Physics and Astronomy, University of California at Los Angeles, Los Angeles, California 90095

E. Esarey
Center for Beam Physics, Ernest Orlando Lawrence Berkeley National Laboratory, Berkeley, California 94720
(Received 28 April 2000)
A new scheme for plasma electron injection into an acceleration phase of a plasma wake field is
presented. In this scheme, a single, short electron pulse travels through an underdense plasma with
a sharp, localized, downward density transition. Near this transition, a number of background plasma
electrons are trapped in the plasma wake field, due to the rapid wavelength increase of the induced
wake wave in this region. The viability of this scheme is verified using two-dimensional particle-in-cell
simulations. To investigate the trapping and acceleration mechanisms further, a 1D Hamiltonian analysis,
as well as 1D simulations, has been performed, with the results presented and compared.

DOI: 10.1103/PhysRevLett.86.1011 PACS numbers: 52.40.Mj, 29.17. +w, 29.27.– a, 52.75.Di

Compared to standard radio-frequency linear accelera- cal LWFA are relatively nonuniform and may not give as
tors, advanced accelerators using plasmas can produce high a beam quality as the blowout regime of the PWFA
much higher acceleration gradients, in excess of 1 GeV兾m. [8,9], in which a plasma electron-rarefied region is formed
Hence, extensive research on plasma-based accelerators in the wake of the driving electron beam. Inside of this
[1–4] has been performed in recent years. For plasma- rarefied region, an accelerating electron experiences accel-
based accelerators, short, intense laser [laser wake field eration dependent only on longitudinal position, and focus-
accelerator (LWFA)] or electron beam [plasma wake field ing which is linear in offset from the axis, just as in more
accelerator (PWFA)] pulses are used to drive large am- conventional accelerators. Furthermore, beam electrons in
plitude plasma waves. In these schemes, the maximum this PWFA scheme experience less scatter with background
achievable accelerating gradient scales as the nonrelativis- plasma electrons during acceleration. Thus superior beam
tic plasma frequency vp 苷 共4pn0 e2 兾me 兲1兾2 . Here n0 , e, quality is expected from the PWFA in this regime.
and me denote the plasma density, electron charge, and In this Letter, we propose a new self-injection scenario
electron mass, respectively. Thus high gradient operation for the PWFA in the blowout regime, where the beam den-
implies use of short period waves, and in order to obtain a sity is greater than the plasma density, nb . n0 (under-
beam with small energy spread, an ultrashort (ø1 ps) ac- dense condition). It is known that in the one-dimensional
celerating pulse must be injected into such a system. This (1D) limit of a PWFA with uniform plasma density, self-
requirement, however, is difficult to meet with an external trapping of background electrons by the wave is very diffi-
injector, especially in the case of the LWFA. The chal- cult [10,11]. Further, it is observed in two-dimensional
lenge of injection in the LWFA has led to proposals of (2D) simulations of the PWFA in the blowout regime
all-optical plasma electron injection schemes using two [5] that self-trapping in a uniform plasma is even more dif-
or three laser pulses [6]. However, these optical methods ficult when transverse motion is allowed. To achieve self-
require extremely accurate laser spatial and temporal over- trapping in the PWFA, we propose to introduce a sharp,
lap, which again leads to technical difficulties. localized density gradient. In this scheme, a single short
To avoid these problems, Bulanov et al. [7] proposed a electron beam pulse is sent through an underdense plasma
self-injection scheme using a single laser pulse propagat- with a sharp downward density transition with kp Ls , 1,
ing in an inhomogeneous plasma. In their scheme, trap- marking the boundary between a dense upstream region
ping of background plasma electrons occurs from wave (I) and a less dense downstream region (II). When the
breaking induced by a gentle density decline, in which beam passes a sharp downward plasma density transition,
the density scale length Ls 苷 n0 兾jdn0 兾dzj is much larger the wavelength of the plasma wave changes rapidly. In
than the plasma skin depth kp21 苷 yb 兾vp , where yb ⬵ c this situation, the plasma electrons that originate just in-
is the driving pulse’s group velocity. As the laser pulse side region II spend much of their oscillation in region I
propagates down the density gradient, the phase velocity before returning to near their initial position in z, advanced
of the wake gradually decreases until it becomes equal to in wave phase compared to the nominal (uniform plasma)
the plasma fluid oscillation velocity, which results in con- region II oscillation. At this position, normally (for a uni-
ventional wave breaking. Bulanov’s scheme is much sim- form plasma) the electron is phased in the wake such that
pler than other plasma injection methods but leads to an the electric field is zero, but in the case we now consider,
injected beam pulse with a relatively large phase spread. the faster oscillation of the electron in region I allows the
In addition, the accelerating and focusing fields in a typi- electron entering region II to remain in an accelerating

0031-9007兾01兾86(6)兾1011(4)$15.00 © 2001 The American Physical Society 1011

phase. This proposed dephasing mechanism, which is The longitudinal phase space of the plasma electrons
justified in qualitative and quantitative detail below, al- at the two times corresponding to Figs. 1(a) and 1(b) are
lows plasma electrons to be trapped and accelerated in shown in Fig. 2. Figure 2(a) shows that a significant
region II. This phenomenon is shown in Fig. 1, which amount of plasma electrons are injected into the acceler-
displays a 2D particle-in-cell (PIC) simulation using the ating wave. Figure 2(b) shows that the trapped plasma
code MAGIC. This injection mechanism is fundamentally electron population (in the range of 5 15 MeV兾c) has
different from Bulanov’s gentle density gradient scheme attained excellent separation in momentum from the back-
for the LWFA. In Bulanov’s case, plasma electrons are ground plasma. The population of trapped particles occu-
trapped in the second plasma oscillation period, but they pies a longitudinal phase extent of df兾2p ⬵ 9%, with a
arise from wave breaking in the first period. On the other large charge of about 0.5 nC for these parameters. Care-
hand, in the present scheme, plasma electron trapping oc- ful comparison of Figs. 2(a) and 2(b) indicates very little
curs in the first rarefied cavity, due to localized nonlaminar phase slippage of the trapped electrons during accelera-
motion near the sharp density transition, and at wake am- tion. With such a small beam phase extent, it would be
plitudes well below conventional wave breaking. possible to simply compensate the momentum spread by
We have explored density transition-induced particle slowly lowering the plasma density further, to rephase the
trapping using 2D PIC simulations (both MAGIC, and trapped electrons to forward in the wave.
a UCLA-specific code NOVO-PIC) in order to illumi- In systematic simulation studies of this scheme, we var-
nate the physical mechanisms relevant to the trapping ied the characteristics of the density transition: the sign
process. The MAGIC results shown in Fig. 1 were ob- of density change, the amplitude of the density difference
tained with an ambient plasma density of n0I 苷 5 3 across the transition, and the density transition scale length.
1013 cm23 for kpII z , 11.2 and n0II 苷 3.5 3 1013 cm23 In all simulations, the driving beam and plasma character-
for kpII z $ 11.2, plasma electron temperature kTe 苷 3 eV, istics were kept similar, in which blowout is complete and
and stationary ions. Here we parametrize lengths in the plasma electron motion is both nonlinear and mod-
terms of the plasma skin depth in region II, kpII 苷
p
erately relativistic. In these studies we observed trapping
4pn0II e2 兾me c2 . The ultrarelativistic (16 MeV) drive only in the case of a downward (in the direction of the beam
beam density distribution employed in the simulation travel) density transition. In addition, it was found that the
was chosen as a bi-Gaussian nb 共r, j兲 ~ e2r 兾2sr e2j 兾2sz
2 2 2 2
number of trapped particles increased as the amplitude of
(j 苷 z 2 nb t), with peak density nb 苷 2.4n0I 苷 3.4n0II , density decrease was made larger. Finally, if the length
and dimensions kpII sz 苷 1, kpII sr 苷 0.56. In Fig. 1(a) over which the density was linearly decreased from its ini-
we observe the trapped population of electrons just as they tial to final value becomes larger than a plasma skin depth
return to near their initial position (beam at kpII z 苷 18.6), kp21 苷 c兾vp , then the trapping disappears completely.
and the associated structure of the wake, which is some- A further, definitive clue as to the trapping mechanism
what complicated at this point. Two features are clearly is that the trapped electrons all initially dwell, as stated
shown, however: a clear difference in wave frequencies above, in region II. These observations have led to the
between regions I and II and a local breaking of the wave development of our trapping model, in which the initial
near the boundary. In Fig. 1(b), it is seen that, after motion of the plasma electrons is in the negative z direction
traversing an additional few plasma wavelengths past the (as well as the positive radial direction) under the forces
boundary (beam at kpII z 苷 35.2), the trapped plasma elec- induced by the drive beam. Upon entry into region I,
tron bunch is loaded into a well-behaved blowout regime the electrons experience wave fields which have a shorter
wave. Furthermore, this injected electron population is oscillation wavelength and can be dephased and trapped.
transversely controlled by uniform ion focusing in the This model also qualitatively explains why we must have
blowout region. kp Ls , 1. If it is that the scale length is longer than the
oscillation distance in z, then the plasma electrons cannot

FIG. 1. Configuration space 共r, z兲 distributions of the plasma FIG. 2. Longitudinal momentum versus z of the plasma elec-
electrons from the MAGIC code 2D PIC simulation. trons for Figs. 1(a) and 1(b).

1012
VOLUME 86, NUMBER 6 PHYSICAL REVIEW LETTERS 5 FEBRUARY 2001

access the region of significantly higher plasma electron moving with the wave, the Hamiltonian H is constant,
density.
At this point, we place our model on firmer analytical H 苷 2ef 1 gme c2 2 bgme c2
ground. As a useful 2D analytical model of plasma elec- 苷 me c2 共1 2 xf 兲 1 me c2 xe . (3)
tron motion in the blowout regime does not exist, we pro-
ceed to develop a 1D, fully relativistic analysis of plasma Here the quantity xf corresponds to the field (and the
motion in both regions and propose a scenario in which plasma fluid) properties, as outlined above, while xe ⬅
p
the motion of the plasma electrons is not strongly affected 共1 2 be 兲兾共1 1 be 兲 describes the dynamical state of any
by the existence of the transition. A Hamiltonian analysis electron, fluid, or otherwise injected into the system. As H
of the phase space trajectories of trapped electrons is thus is invariant, the difference Dx 苷 xf 2 xe is also constant.
made possible. This analytical model is then compared Knowledge of the value of H and the function xf thus al-
to the results of 1D PIC simulations, in order to verify as- lows one to map the electron trajectories in phase space and
pects of the model and also point to its limitations. Without easily determine which trajectories are trapped. With H
discussing any model details, some limitations are imme- as written above, electrons with Dx . xf,min are trapped,
diately apparent. First, in the blowout regime, the plasma defined such that as xf ! xf,min , xe approaches zero and
electron density tends to vanish in the wake of the drive be ! 1. With this model in mind, we can now dis-
beam, while in 1D nonlinear wave theory, the plasma den- cuss the approximate effects of the density discontinuity.
sity does not drop below one-half of the initial density n0 . In this analysis we assume that the density discontinuity
In addition, a 1D wave driven by an electron beam is some- can be approximately modeled by using the Hamiltonian
what of a practical improbability, as it implies total beam analysis in both regions on either side of a sharp density
charge in excess of what is presently found. Nonlinear transition. The fields on both sides of the moving (in the t
1D plasma waves may be driven by ultrashort laser pulses, frame) density boundary are assumed to be correctly given
however, and so the present analysis may have some direct by the wave analysis. Note that at the boundary, the value
application to density transitions in the LWFA. Note that of xf is discontinuous, which implies that H changes sud-
a large amplitude, nonlinear wave is necessary to allow the denly when the electron crosses the boundary. This discon-
trapping process we wish to describe, since for waves with tinuity in H mathematically allows the trapping of initially
amplitudes well below the wave breaking limit, the plasma cold plasma electrons (xe ⬵ 0) in an ultrarelativistic phase
electrons do not move appreciably from their initial posi- velocity wave.
tion and do not attain relativistic oscillation velocities. We also note that, because the system is 1D, at the den-
We begin the analysis by reviewing nonlinear 1D plasma sity transition boundary (which moves backward in the
wake field theory, which has its basis in the fundamen- wave frame at 2yb ⬵ 2c) there is a self-consistent dis-
tal work of Akhiezer and Polovin [12–15]. In this the- continuity in the electric field of DE 苷 4ps. Here s is
ory, with the wave ansatz assumed [all system spatial the net surface charge density “absorbed” by the bound-
and temporal dependences can be expressed using t 苷 ary due to displaced electrons from the upstream side and
vp 共t 2 z兾yb 兲], the differential equation governing rela- “emitted” by those that would be displaced to the down-
tivistic cold∑fluid takes ∏the following
∑ form: ∏ stream side. The field near the boundary is illustrated in
d 2 1 2 bb b b nb Fig. 3, which displays E共z兲 found by theory and simula-
p 苷 bb 2
1 , (1) tion. While the peak field in the two regions is the same,
dt 2 1 2 b2 bb 2 b n0
where nb is the beam density, bb 苷 yb 兾c is the normal- the field in the downstream, high-density region changes
ized beam (as well as wave phase) velocity, and b 苷 y兾c more quickly, since the plasma frequency is higher.
is the normalized plasma electron velocity. Even though the electric field discontinuity at the density
Since the plasma wave is impulsively excited by the transition is self-consistent in terms of displaced charge, it
driving beam, we concentrate on the plasma dynamics in is of course unphysical in the sense that the fluid electrons
the region behind the beam. There, for a very relativistic are not emitted or absorbed by the boundary between the
beam (bb ! 1),∑ Eq. (1) ∏can be rewritten two regions. The fictional absorption at the boundary
s as
1 12b is also conceptually at odds with the physical model of
1
x 00 苷 2 2 2 1 , with x 苷 , (2) trapping, in which the plasma particles move significantly
x 11b past the boundary. Nevertheless, we find some agreement
where the prime denotes d兾dt. Equation (2) is equiva- with this model and 1D PIC simulations, as shown in
lent both to the Poisson equation and the plasma elec- Fig. 3. Here we also display the simulation-derived
tron equation of motion, so we can write the field and electric field associated with the same physical parame-
n
plasma fluid quantities, i.e., density, n0 苷 12 共1 1 x12 兲, po- ters (beam surface charge density of 0.36 mC兾cm2 ,
tential, 2ef 苷 me c2 共1 2 x兲, and electric field, 2eE 苷 n0 苷 3.5 3 1013 cm23 for z . 0.5 cm, and n0 苷 5 3
kp me c2 x 0 ; as well as the dynamical quantities g 苷 12 共x 1 1013 cm23 for z , 0.5 cm) as used in the analytical
12x 2
x 兲, b 苷 11x 2 , and bg 苷 2 共 x 2 x兲, using only x.
1 1 1
model. The agreement is quite good for the fields calcu-
With these results, we can perform a Hamiltonian analy- lated in the region z , 0.5 cm, which is the important
sis of the plasma electron motion. In the Galilean frame zone for electron trapping dynamics, as we shall see.
1013
VOLUME 86, NUMBER 6 PHYSICAL REVIEW LETTERS 5 FEBRUARY 2001

FIG. 3. Plot of the wake electric field in the two regions (n0 苷
3.5 3 1013 cm23 for z . 0.5 cm and n0 苷 5 3 1013 cm23 for FIG. 4. Phase space trajectory of a trapped electron originating
z , 0.5 cm). The solid line indicates the field in the region 0.01 cm downstream of the plasma density transition, for the
calculated from the analytical model, while the dashed line indi- simulation and parameters of Fig. 1.
cates the model-derived field in the conjugate region. The thick
dotted line is the field calculated from a 1D PIC simulation. Note that even though the exact phase space trajectory in
the transition region is not accurately predicted, the final
phase of the electron after trapping is well predicted by the
In order to determine how to connect the distinct values
analytical model.
of H in the two regions of interest, the position of the elec-
The PWFA-based self-injection scheme we have
trons as a function of t must be known, from which we
proposed in this paper demands further theoretical and
find the time of boundary crossing. This can be accom-
numerical investigations of detailed issues such as beam
plished by integrating the velocity,
quality of the trapped plasma electrons, dephasing, drive
Z t2 µ ∂
c Z t2 b beam quality degradation in the plasma, etc., and this will
z苷c b dt 苷 dt be done in the near future. To verify this new injection
t1 vp t1 b 2 1
µ ∂ scheme, we are planning an experiment with an existing
1 Z t2 1 2 xe2
苷 dt . (4) argon plasma source [16], which was originally developed
kp t 1 2xe2 for the underdense plasma lens experiment at the Neptune
Advanced Accelerator Laboratory of UCLA.
R case of the fluid electrons, xf 苷 xe , we have
In the special
This work was supported by the U.S. Department of
kp z 苷 12 关n共t兲 2 n0 兴 dt, a result that emphasizes that
Energy, under Contracts No. DE-FG03-92ER40693 and
the wave fluid motion is initially in the negative direction
No. DE-AC-03-76SF0098.
and arrives back at the starting position after one period.
Using the formalism we have developed, we can com-
pare the dynamics of a trapped electron from the 1D PIC
simulation to that derived from the analytical model. The [1] C. E. Clayton et al., Phys. Rev. Lett. 70, 37 (1993).
electron we follow in the simulation is initially located [2] K. Nakajima et al., Phys. Rev. Lett. 74, 37 (1995).
at z 苷 5.1 cm (1 mm downstream of the density transi- [3] C. W. Siders, Phys. Rev. Lett. 76, 3570 (1996).
[4] F. Amiranoff et al., Phys. Rev. Lett. 81, 995 (1998).
tion). To employ the theoretical model, we must follow the
[5] D. Umstadter, J. Kim, and E. Dodd, Phys. Rev. Lett. 76,
electron initially in the low-density (downstream) region, 2073 (1996).
calculate its position and momentum crossing the transi- [6] E. Esarey et al., Phys. Rev. Lett. 79, 2682 (1997); E. Esarey
tion using Eqs. (3) and (4), and connect the values of H et al., Phys. Plasmas 6, 2262 (1999).
across the boundary. This calculation is then repeated in [7] S. Bulanov et al., Phys. Rev. E 58, R5257 (1998).
the high-density (upstream) region, the final value of H in [8] J. B. Rosenzweig et al., Phys. Rev. A 44, R6189 (1991).
the downstream region is calculated, and the motion exam- [9] N. Barov et al., Phys. Rev. Lett. 80, 81 (1998); N. Barov
ined for trapping. The results of this exercise are shown et al., Phys. Rev. ST Accel. Beams 3, 011301 (2000).
in Fig. 4, which displays the phase space trajectory of the [10] T. Katsouleas and W. B. Mori, Phys. Rev. Lett. 61, 90
electron in both the model and simulation. Good agree- (1988).
ment on the predicted trajectory is obtained in the region [11] J. B. Rosenzweig, Phys. Rev. A 38, 3634 (1988).
[12] A. I. Akhiezer and R. V. Polovin, Sov. Phys. JETP 3, 696
of negative momentum, as expected from the agreement
(1956).
of the fields in the upstream region (Fig. 3), while notable [13] R. D. Ruth and A. W. Chao, in Laser Acceleration of Par-
disagreement occurs when the electron reapproaches the ticles, edited by P. J. Channel, AIP Conf. Proc. No. 91 (AIP,
density transition from the upstream side. This disagree- New York, 1982).
ment arises from the inconsistency of the discontinuous [14] J. B. Rosenzweig, Phys. Rev. Lett. 58, 555 (1987).
field model we have developed, because of non-negligible [15] C. B. Schroeder et al., Phys. Rev. E 59, 6037 (1999).
charges passing through (not stopping at) the transition. [16] H. Suk et al., IEEE Trans. Plasma Sci. 28, 271 (2000).

1014