Stability of the Modulator in a Plasma-Modulated Plasma Accelerator

J. J. van de Wetering,1, ∗ S. M. Hooker,1 and R. Walczak1, 2

1
John Adams Institute for Accelerator Science and Department of Physics,
University of Oxford, Denys Wilkinson Building,
Keble Road, Oxford OX1 3RH, United Kingdom
2
Somerville College, Woodstock Road, Oxford OX2 6HD, United Kingdom
(Dated: March 27, 2023)
We explore the regime of operation of the modulator stage of a recently proposed laser-plasma
accelerator scheme [Phys. Rev. Lett. 127, 184801 (2021)], dubbed the Plasma-Modulated Plasma
arXiv:2303.14032v1 [physics.plasm-ph] 24 Mar 2023

Accelerator (P-MoPA). The P-MoPA scheme offers a potential route to high-repetition-rate, GeV-
scale plasma accelerators driven by picosecond-duration laser pulses from, for example, kilohertz
thin-disk lasers. The first stage of the P-MoPA scheme is a plasma modulator in which a long,
high-energy ‘drive’ pulse is spectrally modulated by co-propagating in a plasma channel with the
low-amplitude plasma wave driven by a short, low-energy ‘seed’ pulse. The spectrally modulated
drive pulse is converted to a train of short pulses, by introducing dispersion, which can resonantly
drive a large wakefield in a subsequent accelerator stage with the same on-axis plasma density
as the modulator. In this paper we derive the 3D analytic theory for the evolution of the drive
pulse in the plasma modulator and show that the spectral modulation is independent of transverse
coordinate, which is ideal for compression into a pulse train. We then identify a transverse mode
instability (TMI), similar to the TMI observed in optical fiber lasers, which sets limits on the energy
of the drive pulse for a given set of laser-plasma parameters. We compare this analytic theory with
particle-in-cell (PIC) simulations and find that even higher energy drive pulses can be modulated
than those demonstrated in the original proposal.

INTRODUCTION of applications for which LPAs offer an advantage over

conventional, radio-frequency particle accelerators.
In a laser-plasma accelerator (LPA), plasma oscilla- It is important, therefore, to consider alternative laser
tions are driven by pushing free electrons away from an systems and/or develop novel approaches for driving
ultrashort laser pulse via the ponderomotive force. The LPAs. Contemporary thin-disk lasers are efficient and
heavier ions remain approximately stationary relative to can already provide joule-scale pulses at kHz repetition
the electrons, thus the electron-ion charge separation col- rates [5–7]. However, they cannot drive a LPA directly
lectively forms a strong electric field which can be used since the small bandwidth of their gain media limits the
to accelerate charged particles. The plasma wave driven duration of the pulses they generate to [8–10] τ & 1 ps,
in this way will have a phase velocity set by the group ve- which is much longer than the plasma period. We note
locity of the laser pulse, which is well suited for the accel- that pulses from thin-disk lasers have been compressed
eration of relativistic charged particles. The accelerating to a duration below 100 fs, following spectral broadening
gradients achievable by LPA are set by the wavebreaking via self-phase modulation in a gas [11]. However, to date
field E0 = me ωp c/e and can be on the order of 100 GV/m this approach has been limited to pulse energies below
[1], three orders of magnitude larger than those possible 120 mJ.
in radio-frequency cavities. Here the plasma frequency With the objective of applying the desirable features
ωp = (ne e2 /me 0 )1/2 , where ne is the electron density. of thin-disk lasers to LPAs, some of the present au-
Efficient excitation of the plasma wave by a single laser thors recently proposed a scheme, illustrated in Fig.
pulse requires that the duration of the pulse is less than 1, for converting picosecond-duration pulses to a train
half the plasma period Tp = 2π/ωp . For plasma densities of shorter pulses that could be used to resonantly ex-
of interest Tp is in the 100 fs range, and hence single- cite a plasma wave [12]. In this scheme, which we call
pulse LPAs first became practical with the development the Plasma-Modulated Plasma Accelerator (P-MoPA), a
of chirped pulse amplification (CPA) [2], which allowed high-energy, picosecond-duration ‘drive’ pulse is modu-
joule-scale pulses to be compressed to sub-picosecond du- lated spectrally by co-propagating it in a plasma channel
rations. Ever since, most experimental demonstrations with a low amplitude (∼ 1%) plasma wave driven by a
of LPAs have used high intensity ultrashort laser pulses low-energy, short ‘seed’ pulse. To first order, the spectral
from Ti:sapphire CPA laser systems. However, these sys- modulation takes the form of a set of sidebands of angu-
tems suffer from low (∼ 0.1-10 Hz) repetition rates [3] lar frequencies ωm = ωL + mωp0 , where ωL is the angular
and poor (< 0.1%) electrical-to-optical energy efficien- frequency of the input drive pulse, m = ±1, ±2, ±3, . . . is
cies [4]. Despite the advantages gained by being much the sideband order, and ωp0 is the plasma frequency on
more compact, the low efficiency and repetition rate of the axis of the plasma channel. The spectrally-modulated
the laser drivers used today severely limit the number drive pulse is converted into a temporal modulation by
 2

passing it through a dispersive optical system that re- duces Eq. (1) to a linear paraxial wave equation
moves the relative spectral phase, ψm = −|m|π/2, of  
each sideband. This forms a train of short pulses, spaced i ∂ c2 ωp2  
+ 2 ∆ ⊥ a = 2 δn0 (r) + δn(r, ξ; |as |2 ) a
temporally by Tp0 , which can resonantly drive a large am- ωL ∂τ 2ωL 2ωL n0
plitude plasma wave in a plasma accelerator stage with (2)
the same axial density as the modulator.
where as denotes the seed pulse envelope, whose intensity
In our earlier work a one-dimensional (1D) fluid model,
we assume to be unchanging relative to the modulating
and 2D particle-in-cell (PIC) simulations were used to
drive pulse envelope a. We also demand that the chan-
demonstrate the operation of the plasma modulator and
nel is matched to the spot size w0 of the seed and drive
accelerator stages, and to show that GeV-scale energy
pulses. A matched parabolic channel and its respective
gains could be obtained from existing thin-disk laser sys-
unperturbed Gaussian drive pulse envelope take the form
tems. In this paper we derive a full 3D theory of seeded
[24–26]
spectral modulation and we use this to establish the use-
ful operating regime for the modulator stage in the P- n0 (r) = n00 + ∆n(r/w0 )2
MoPA. We find that the range of operation of the mod-  
r2 2c2
ulator is determined by the onset of the transverse mode a(r, ξ, τ ) = a0 f (ξ) exp − 2 − iωL τ 2 2 (3)
w0 ωL w0
instability (TMI), similar to the TMI observed in high
power fiber laser systems [13–17]. This analysis is used where ∆n ≡ (πre w02 )−1 is the channel depth parame-
to establish the regime of parameter space for which the ter, re is the classical electron radius and 0 ≤ f (ξ) ≤ 1
modulator can be operated successfully. The results of is the slowly-varying longitudinal envelope of the drive
the 3D analytic theory are compared with particle-in-cell pulse. Assuming the seed wake is small relative to the
(PIC) simulations, and are found to be in good agree- channel depth parameter |δn(r, ξ; |as |2 )|  ∆n, we can
ment. We find that even higher energy drive pulses can apply time-independent perturbation theory to Eq. (2),
be modulated than those considered in the original pro- yielding the following modulation to the total phase of
posal. the laser pulse
   
2c2 τ δn(r, ξ; |as |2 )
Φ(ξ, τ ) = kL ξ − 1 + (4)
THE PLASMA MODULATOR ωL w02 ∆n ⊥

where kL = ηωL /c is the laser wavenumber with the on-

Seeded Spectral Modulation in Plasma Channels axis plasma index of refraction η = (1 − ωp0 2
/ωL 2 1/2
) and
2
R∞ 2 2
h(. . .)i⊥ = (4/w0 ) 0 (. . .) exp(−2r /w0 )rdr denotes the
Propagation of the envelope of a laser pulse in an ax- intensity-weighted transverse average. Using this expres-
isymmetric plasma channel of electron density n0 (r) = sion we can also retrieve the shift in instantaneous fre-
n00 + δn0 (r) with a small wake δn(r, ξ; |a|2 ) can be ap- quency
proximately described by the paraxial wave equation [18–  
21] (see Supplemental Material [22] for its derivation) ∆ω(ξ; Lmod ) 2c2 ∂ δn(r, ξ; |as |2 )
= −Lmod 2 2 (5)
  ωL ωL w0 ∂ξ ∆n ⊥
i ∂ c2
+ 2 ∆⊥ a = where Lmod = vg τ is the modulator length. This predicts
ωL ∂τ 2ωL
that the spectral modulation amounts to a radial averag-
ωp2   ing of the longitudinal gradient of the wake weighted by
2 δn0 (r) + δn(r, ξ; |a|2 ) − n0 (r)|a|2 /4 a (1)
2ωL n0 the transverse intensity profile of the drive pulse. This
independence of the spectral modulation of radial posi-
where a(r, θ, ξ, τ ) is the envelope of the pulse’s normalized tion is desirable as it allows the spectral modulation to
vector potential, ωL is the laser frequency and the propa- be converted into a temporal one by applying the same
gation is described in co-moving coordinates ξ = z − vg t, chromatic dispersion across the entire cross-section of
2 2 1/2
τ = t, with vg /c = (1 − ωp0 /ωL ) defined as the the modulated pulse. Figure 2 shows the results of a
group velocity of electromagnetic plane waves in uni- 2D PIC simulation which demonstrates that the spectral
form plasma of density n00 , corresponding to the on-axis modulation is indeed independent of the transverse co-
plasma channel frequency ωp0 = ωp (r = 0). This group ordinate, despite the fact that the amplitude and phase
velocity may differ from the group velocity of a tightly of the plasma wave does depend on the radial coordi-
focused laser pulse [23, 24]. Nonlinearities come from nate, as predicted by Eq. (5). Figure 2 also shows sup-
weakly relativistic effects as well as from interaction be- pression of the spectral modulation towards the tail of
tween the pulse and its own excited wake. the drive pulse, which is most apparent in the plot of
Successful seeded spectral modulation requires rela- the retrieved instantaneous frequency. This is also ex-
tivistic and self-wake effects to be negligible, which re- pected from Eq. (5), since the large curvature of the
 3

FIG. 1. [Color online]. Outline of the P-MoPA scheme from the original proposal [12]. A short, low energy seed pulse excites
a small wake in the modulator stage which spectrally modulates a long, high energy drive pulse into interleaving redshifted
(Stokes) and blueshifted (anti-Stokes) pulse trains whilst maintaining a smooth envelope. Chromatic dispersion is then applied
to the spectrally modulated drive pulse to compress it into a multipulse train, which can then be used to resonantly drive a
wakefield in the accelerator stage with the same density as the modulator.

wavefronts of the plasma wave towards the tail of the 0.02

0.06
drive pulse leads to a reduction in the spectral modula-

∆ω/ωL
tion when the longitudinal gradient of the wave is aver-
|a|2
0.03 0.00
aged radially. Note that in the Supplemental Material
[22] we compare the results of PIC simulations in 2D −0.02
0.02 −5.0 −2.5 0.0 2.5 5.0
and cylindrical geometry. All PIC simulations presented
∆ω/ωL

in this paper were performed with axial plasma density 0.00

n00 = 2.5 × 1017 cm−3 , seed and drive laser wavelength
λL = 1030 nm, seed pulse FWHM duration τseed = 40 fs −0.02
0.1
and modulator length Lmod = 110 mm. A complete list 25 −5.0 −2.5 0.0 2.5 5.0
x (µm)

δn/n00
of simulation parameters is included in the Supplemental
0 0.0
Material [22].
−25
−0.1
−5.0 −2.5 0.0 2.5 5.0
Channel Suppression of Spectral Modulation ξ/λp0

As observed in Fig. 2, Eq. (5) implies that the spec- FIG. 2. [Color online]. 2D PIC simulation of the modulator
tral modulation of the drive pulse is limited by wave-front stage in a matched parabolic plasma channel of matched spot
curvature of the plasma wave. For low amplitude wakes size w0 = 30 µm with Wseed = 50 mJ and Wdrive = 0.6 J,
this curvature is dominated by the transverse profile of τdrive = 1 ps. The top panel shows the on-axis longitudinal
the plasma channel. For square channels the wake has intensity profiles |a|2 for the seed and drive pulses. The middle
panel plots the on-axis instantaneous frequency calculated by
flat phase fronts over most of the transverse profile of the a Hilbert transform. The bottom panel displays the full 2D
lowest-order mode of the channel. However, for chan- distributions of the relative amplitude δn/n00 of the plasma
nels with a curved transverse profile, such as parabolic wave and the relative frequency modulation ∆ω/ωL .
channels, the wave-fronts of the plasma wave are curved,
which can strongly suppress the spectral modulation.
Figure 3 shows the results of 2D PIC simulations that curvature increases towards the tail of the drive pulse.
compare the performance of a modulator with square and As a consequence, the generated pulse train does not ex-
parabolic plasma channels. It can be seen that for the hibit complete intensity modulation near its tail, which
square channel the wave-fronts of the plasma wave are would reduce the amplitude of the plasma wave it could
flat across most of the channel width, and as a conse- drive in the accelerator stage. As shown in Fig. 3, in-
quence the pulse train generated by removal of the side- creasing the matched spot size of the parabolic channel
band spectral phase exhibits strong temporal modulation to 50 µm reduces the wake curvature, which leads to im-
over the entire duration of the pulse train. In contrast, proved modulation of the generated pulse train. We note
the wave-fronts of the wake driven in a parabolic channel that the deleterious effects of wave-front curvature would
of 30 µm matched spot size are strongly curved, and this be even more pronounced in 3D geometry (see Supple-
 4

square, 30 µm δn/n00 |a| transverse modes, whereas the anti-Stokes-shifted light

0.1
50 remains relatively well focused. This asymmetry between
x (µm)

0 0.0 0.2
−50 the Stokes and anti-Stokes light leads to the formation
−0.1 0.0
0.0 2.5 −5 0 5 −5 0 5 of low-contrast pulse trains in the modulator, with the
n0/n00 ξ/λp0 ξ/λp0 Stokes-shifted radiation forming a train off-axis, and the
parabolic, 30 µm δn/n00 |a| anti-Stokes light forming a train on-axis. This effect be-
0.1
50 comes worse with increasing drive pulse energy, and can
x (µm)

0 0.0 0.2
−50 be seen to be especially bad for the 2.4 J pulse, which
−0.1 0.0
0.0 2.5 −5 0 5 −5 0 5 has undergone severe transverse break-up and has been
n0/n00 ξ/λp0 ξ/λp0 strongly redshifted. This difference in behaviour between
parabolic, 50 µm δn/n00 |a| the Stokes and anti-Stokes sidebands has previously been
100 0.1
observed as the result of relativistic effects [28]. As shown
x (µm)

0 0.0 0.2
in the Supplemental material [22], a similar effect is pre-
−100 −0.1 0.0 dicted in the non-relativistic regime when non-paraxial
0 1 −5 0 5 −5 0 5
n0/n00 ξ/λp0 ξ/λp0 effects are accounted for.

FIG. 3. [Color online]. Comparison of the performance of

modulators with plasma channels of different transverse pro- Envelope Self-Modulation and Raman Forward
files each with wall thicknesses of 20 µm (see Supplemental Scattering
Material for their parameterizations [22]). The left panels
show the transverse electron density (blue) and normalized
guided intensity (black) profiles of the channels for: top, a We now consider the stability of picosecond-scale
square channel of diameter 30 µm; middle, a parabolic chan- pulses propagating in long unperturbed plasma channels,
nel of matched spot size 30 µm; bottom, a parabolic channel i.e. in the absence of a wake driven by a seed pulse. As
of matched spot size 50 µm (with (50/30)2 × more seed and discussed by Mori [29], as long as 1/kL w0  ωp2 /ωL 2
is
drive pulse energy to account for the larger spot size). The
satisfied, the following parameter determines whether a
middle panels show the relative wake amplitudes δn/n00 at
the end of the modulator, calculated by 2D PIC simulations. laser pulse will be dominated by Raman forward scatter-
The right panels show the on-axis pulse envelopes |a| at the ing (RFS) or envelope self-modulation (SM) instabilities:
end of the modulator before (dashed blue) and after (solid  5/2  4
orange) the expected [27] sideband spectral phase ψm was P τL ne λL
removed (see Supplemental Material [22]). Γ≡ · · ·
1 TW 1 ps 1 × 1019 cm−3 1 µm
(6)
mental Material [22]). where P is the peak laser power and τL is the laser du-
ration. When Γ ≥ 3, RFS dominates, whereas SM dom-
inates when Γ ≤ 0.4. Substituting laser-plasma parame-
STABILITY OF THE PLASMA MODULATOR ters used by Jakobsson et al [12], we find Γ = 6.3 × 10−5
and 1/kL w0 = 55 × 10−4  ωp2 /ωL 2
= 2.4 × 10−4 , mean-
It is important to understand the extent to which in- ing that we are well in the SM-dominated regime. The
stabilities will arise in the modulator, and the range of maximum growth rate and e-folding number of envelope
laser and plasma parameters for which any deleterious SM in a uniform plasma is given by [30, 31]
effects arising from them can be avoided. Since we do
not want to waste any of the drive pulse energy within 1 2 ωp2
γSM = a (1 + a20 )−3/2 , Ne,SM = γSM Tint (7)
the modulator stage, we would like its envelope to re- 8 0 ωL
main smooth as it propagates. Although the specific set
of parameters used in the original P-MoPA proposal [12] where Tint = Lmod /c is the interaction time.The param-
were shown to work in simulation, for the scheme to be eters used by Jakobsson et al [12] yield ∼ 1.3 e-foldings,
practical we want a large, well-defined parameter space and hence SM could become problematic, especially if we
where it is stable. try to include more energy in the drive pulse.
Increasing the drive pulse energy Wdrive eventually
disrupts the plasma modulator with a TMI via a self-
modulation mechanism. This instability drives time- Transverse Mode Stability Condition
varying wakes which excite higher order transverse chan-
nel modes. The results of PIC simulations demonstrating For a pulse propagating in a waveguide, the growth
this phenomenon are given in Fig. 4. It can be seen that of SM is complicated by oscillations in the spot size of
the Stokes sidebands become more defocused towards the the mode, which arise from excitation of more than one
trailing end of the pulse, thereby exciting higher order waveguide mode. This becomes relevant when the SM
 5

Stokes Central Anti-Stokes Stokes Central Anti-Stokes Stokes Central Anti-Stokes

25 25 25
0 0 0
−25 −25 −25
0.04
25 0.024 25 0.06 25
0.03
x (µm)

x (µm)

x (µm)
|a|2

|a|2

|a|2
0 0.016 0 0.04 0 0.02
−25 0.008 −25 0.02 −25 0.01
0.000 0.00 0.00
−5 0 5 −5 0 5 −20 0 20
ξ/λp0 ξ/λp0 ξ/λp0

(a) τdrive = 1 ps, Wdrive = 0.6 J (b) τdrive = 1 ps, Wdrive = 1.2 J (c) τdrive = 4 ps, Wdrive = 2.4 J

FIG. 4. [Color online]. 2D PIC simulations of the intensity profiles, |a|2 , of drive pulses at the exit of the modulator with a
channel of square cross-section with w0 = 30 µm and Wseed = 50 mJ, for various energies and durations of the drive pulse. The
top row displays the intensity profiles decomposed into its redshifted Stokes (ω − ωL ) < −ωp0 /2, central |ω − ωL | < ωp0 /2, and
blueshifted anti-Stokes (ω − ωL ) > ωp0 /2 components; the bottom panel displays the full intensity profile of the drive pulse.

growth rate is slower than the spot size oscillation fre- define the plasma modulator stability condition, which
quency [32] γSM < ωw = 4c2 /ωL w02 . In this section sets bounds on the laser-plasma parameters to prevent
we consider the effects of the plasma channel on self- nonlinear self-modulation from exciting transverse mode
modulation. transitions.
Consider coupling a slightly unmatched drive pulse We will work in the shallow channel limit ∆n  n00 ,
into a parabolic plasma channel so it undergoes small which allows us to neglect the effects of the channel on
spot size oscillations, and assume that no centroid oscil- wake structure [33]. This gives a seed wake of the form
lations are present, so that only radial Laguerre-Gaussian [18]
modes LGp0 (r) are excited (see Supplemental Mate-
rial [22]). Neglecting relativistic effects, applying time- δn(r, ξ) = δns cos(kp0 ξ)LG200 (r) (10)
dependent perturbation theory to Eq. (1), the coeffi- where kp0 = ωp0 /vg and δns denotes the on-axis seed
cients αp (ξ, τ ) of each radial Laguerre-Gaussian mode p wake amplitude. As the drive pulse should remain pri-
at longitudinal coordinate ξ are found to evolve according marily in the fundamental mode, we can approximate Eq.
to: (8) as a two-level system comprising the LG00 and LG10
i ∂αp (ξ, τ ) modes with transitions between them driven by the seed
= wake
ωc ∂τ
X D  δn + δn  E
 NL  ∂α0

αn (ξ, τ ) LGp0  LGn0 ei(p−n)ωw τ , i = 2Ωs cos(kp0 ξ) α0 + 21 α1 e−iωw τ ,
n
∆n ∂τ
X ∂α1

a(r, ξ, τ ) = e−iωc τ αp (ξ, τ )LGp0 (r)e−ipωw τ (8) i = 2Ωs cos(kp0 ξ) 21 α0 eiωw τ + 12 α1 (11)
∂τ
p
2
where Ωs = (ωp0 /8ωL )(δns /n00 ) is the rate of spectral
where δn is the fixed seed wake, δnNL is the self-wake
modulation parameter. Note that since we are using the
of the drive pulse, and ωw and ωc = ωw /2 are the spot
paraxial description, we are implicitly assuming symme-
size and centroid oscillation frequencies respectively [32].
try between the Stokes and anti-Stokes sidebands (see
Coupling the drive pulse into a slightly unmatched chan-
Supplemental Material [22] for the non-paraxial descrip-
nel corresponds to following set of initial conditions
tion). We can already see from these expressions that
α0 = a0 f (ξ), α1 = w a0 f (ξ), αp6=0,1 ≈ 0 (9) the first radial mode spectrally modulates half as fast
as the fundamental, as it is more sensitive to the ra-
where w = −δw/w0  1 is the channel spot size mis- dial drop-off of the seed wake amplitude ∼ LG200 (r).
match parameter. In order to solve Eq. (8), the self-wake This asymmetry, coupled to spot size oscillations, is one
δnNL must be known. We estimate the self-wake as fol- of two effects contributing to plasma-resonant modula-
lows. We first assume that the self-wake can be neglected, tions of the drive pulse intensity which excite a self-
and calculate the intensity modulation of the drive pulse wake. Since the spot size oscillation frequency is nec-
caused by the seed wake only. We then use this inten- essarily much higher than the spectral modulation rate,
sity modulation to calculate the self-wake it would ex- i.e. ωw  Ωs , as a consequence of the seed wake being
cite. This estimate of the self-wake can then be used to small relative to the channel depth parameter, we can
 6

(0) (1) (1)

integrate Eq. (11) by assuming most of the light remains α0 = α0 (ξ, τ ) + α0 (ξ, τ ), α1 = α1 (ξ, τ ). This yields
in the fundamental mode to find a first order solution the following intensity modulation

   
|a(r, ξ, τ )|2 Ωs cos(kp0 ξ)
1 − 2 w +
= (cos[Ωs τ cos(kp0 ξ)] − 1) LG200 (r)
|a0 |2 f 2 (ξ) ωw
 
2Ωs cos(kp0 ξ)
+ 2w cos [Ωs τ cos(kp0 ξ) − ωw τ ] + (cos [Ωs τ cos(kp0 ξ) − ωw τ ] − 1) LG00 (r)LG10 (r) . (12)
ωw

We see from Eq. (12) that the drive pulse is modulated nally in ξ and temporally in τ .
radially, longitudinally, and temporally. This modulation To estimate the self-wake δnNL that would be excited
can be physically understood by splitting it into three ef- by this intensity modulation, we are only interested in
fects which can be isolated by setting certain terms to keeping terms resonant with the plasma ∼ cos(kp0 ξ + φ)
zero. The first is a longitudinally uniform spot size oscil- which excite the largest amplitude self-wake. We also
lation of amplitude δw, which can be recovered by setting only consider propagation times up to Ωs τmod ∼ 1/2, as
Ωs = 0. The second comes from coupling between the δw this provides sufficient spectral modulation for compres-
spot size oscillations and the seed wake spectral modu- sion into a pulse train which roughly coincides with the
lation due to the effect mentioned earlier where higher minimum modulation required to reach the accelerator
order radial modes spectrally modulate slower than the stage wake amplitude plateau discussed by Jakobsson et
fundamental mode in the ∼ LG200 (r) seed wake. The al [12]. With both of these considerations in mind, we can
third depends purely on the seed wake, which can be Taylor expand the cos [Ωs τ cos(kp0 ξ) − ωw τ ] terms in Eq.
seen when setting w = 0. The seed wake introduces lo- (12) to first order in Ωs τ . This gives the plasma-resonant
cal variations in the matched spot size as it perturbs the part of the intensity modulation and the approximate
channel plasma density, varying the spot size longitudi- self-wake it would excite [34]

 
2Ωs [cos(ωw τ ) − 1]
|a(r, ξ, τ )|2res = |a0 |2 f 2 (ξ) 2w Ωs τ sin(ωw τ ) + cos(kp0 ξ)LG00 (r)LG10 (r) ,
ωw
 
δnNL (r, ξ, τ ) e2 ωp0 Wdrive (ξ)λ2L 2Ωs [cos(ωw τ ) − 1]
= 2 Ω
w s τ sin(ωw τ ) + sin(kp0 ξ)LG00 (r)LG10 (r) (13)
n00 8π 2 m2e 0 c5 πw02 ωw

where Wdrive (ξ) indicates the total energy of the drive bility condition
pulse contained between its head and coordinate ξ. To  
prevent self-modulation driving transverse mode transi-  δw/w 2
kp0 w02  ωp0 Wdrive λ2L
 0
 +   Pmod ,
tions (and to make this calculation self-consistent), we  δns /n00 8  πw02
require that the self-wake effect on the LG00 → LG10
transition must be negligible throughout the full propa- 32π 2 m2e 0 c5
Pmod = ≈ 220 GW . (15)
gation in the modulator, resulting in the constraint e2
This sets a limit on the total energy of the drive pulse
Wdrive for a given plasma density, spot size, laser wave-
Z τmod
length, seed wake and channel spot size mismatch. For
dτ
  
   the laser-plasma parameters used by Jakobsson et al [12],
LG10 δnNL eiωw τ LG00  LG10 δnLG00 .
0 τmod taking δns /n00 ∼ 2.5% and δw/w0 ∼ 12% from the PIC
(14) simulations therein, the requirement of Eq. (15) becomes
37 GW  220 GW, which is satisfied. Hence, in this
regime we do not expect the self-modulation effects to
be debilitating to the plasma modulator. In PIC simu-
Substituting Ωs τmod = 1/2 and Eq. (13) into this con- lations, we have found that even letting the LHS of Eq.
straint yields the plasma modulator transverse mode sta- (15) go up to ∼ 70 GW remains stable enough for com-
 7

tions will be determined by the mismatch between the

25 25
transverse amplitude profile of the lowest-order chan-
0.30
0.16
nel mode, and that of the drive pulse at the channel
0 0
entrance. However, even in the limit of a perfectly
−25 −25
matched channel, small spot size oscillations can arise
x (µm)

x (µm)
|a|2

|a|2
25
0.15
25
from other sources, such as ponderomotive and relativis-
0.08 2
tic self-focusing [35, 36]. In addition, the kp0 w02 /8 term
0 0
in Eq. (15), which comes from the seed wake forming a
−25 −25
plasma-resonant variation in matched spot size, ensures
0.00 0.00
−1 0
ξ/λp0
1 −5 0
ξ/λp0
5 that there will always be an upper limit on the drive pulse
energy.
(a) τdrive = 250 fs (b) τdrive = 1 ps
There are two ways that excitation of higher order
transverse modes can be mitigated, other than ensur-
25 25 ing that Eq. (15) is satisfied. However, each of these
0 0.08 0 0.04 comes at a cost. First, the treatment above assumed
−25 −25 propagation in a shallow channel. For deeper channels,
x (µm)

x (µm)

i.e. ∆n ∼ n00 , the self-wake will be partially suppressed

|a|2

|a|2
25 0.04 25 0.02 by the off-resonant plasma wave, and the difference in
0 0 spectral modulation rate of the higher-order mode and
−25 −25
the fundamental will be decreased due to the radial com-
0.00 0.00
ponent of the seed wake. This allows for more energy
−10 0 10 −20 0 20
ξ/λp0 ξ/λp0 to propagate in the modulator without transitions to
(c) τdrive = 2 ps (d) τdrive = 4 ps higher order modes. However, channels of this form
also suppress the spectral modulation towards the tail
FIG. 5. Calculated intensity profiles, |a|2 , of drive pulses of the pulse, as shown in Fig. 3, which is detrimental to
at the exit of the modulator before (top) and after (bot- pulse compression. Another option would be to use a
tom) compression into a pulse train for drive pulses of energy leaky channel that leaks higher order modes faster than
Wdrive = 1.2 J and FWHM duration: (a) 0.25 ps, (b) 1 ps, (c)
the fundamental mode [37, 38]. However, this approach
2 ps and (d) 4 ps. For these 2D PIC simulations the modula-
tor was taken to have a square cross-section with w0 = 30 µm, would have reduced efficiency, since drive pulse energy
and the seed pulse to have an energy of Wseed = 50 mJ. transferred to higher order modes would be lost.

pression into a pulse train despite the excitation of the CONCLUSION

first radial mode being non-negligible. Figure 5 shows the
calculated intensity profiles, |a|2 , of drive pulses at the We have derived a full 3D analytic theory of seeded
exit of the modulator before and after compression into a spectral modulation, and have used this to establish the
pulse train (by removing the expected [27] sideband spec- useful operating regime for the modulator stage in the
tral phase ψm of a pulse without TMI (see Supplemental P-MoPA. This model is found to be in very good agree-
Material [22])) for drive pulses of energy Wdrive = 1.2 J ment with those obtained from 2D PIC simulations of
and various pulse durations. It can be seen that, despite the modulator.
the pulse duration varying by a factor of 16, they each The analytic theory leads to several important conclu-
undergo the same amount of spot size oscillation driven sions. First, the spectral modulation of the drive pulse
transitions to higher order transverse modes. This is in is independent of radial distance from the axis of the
agreement with Eq. (15), which is independent of the modulator channel. This ensures that, after leaving the
drive pulse duration. As a consequence, for drive pulses modulator, the entire spectrally-modulated pulse can be
of duration 1, 2 and 4 ps, the transverse structure of the compressed by a simple optical system that removes the
pulse at the end of the modulator is identical. For drive spectral phase accumulated in the modulator. Second,
pulses of 250 fs duration, the structure is also similar, but curvature of the seed-pulse-driven plasma waves is shown
in this case the pulse duration is only approximately two to reduce the degree of spectral modulation, and hence
plasma periods long, and hence the assumption of small the modulation of the pulse train that is generated after
bandwidth breaks down. We note also that Fig. 5 shows compression. This finding establishes limits on the shape
that in each case the spectrally-modulated drive pulse of the channel used in a P-MoPA modulator.
can still be compressed into a well-defined pulse train We also explored limits to the operating parameters
suitable for the accelerator stage, despite the transverse of a seeded modulator set by the self-modulation of
structure that it has developed. the drive pulse and excitation of higher-order transverse
For most practical applications, the spot size oscilla- channel modes. We found that the operation of the mod-
 8

ulator is limited by the onset of the transverse mode average power lasers for future particle accelera-
instability (TMI), similar to the TMI observed in high tors, AIP Conference Proceedings 1507, 147 (2012),
power fiber laser systems. An analysis of the excitation https://aip.scitation.org/doi/pdf/10.1063/1.4773687.
of higher-order modes allowed the identification of a con- [4] B. Hidding, S. Hooker, S. Jamison, B. Muratori, C. Mur-
phy, Z. Najmudin, R. Pattathil, G. Sarri, M. Streeter,
dition on the energy of the drive pulse, the relative am- C. Welsch, M. Wing, and G. Xia, Plasma wakefield ac-
plitude of oscillations in its spot size, and the relative celerator research 2019–2040: A community-driven uk
amplitude of the seed-pulse-driven wake, that must be roadmap compiled by the plasma wakefield accelerator
satisfied for stable operation. steering committee (pwasc), PWASC (2019).
Finally we emphasize that the results presented here [5] C. Herkommer, P. Krötz, R. Jung, S. Klingebiel,
show that the modulator in a P-MoPA can exhibit stable C. Wandt, R. Bessing, P. Walch, T. Produit, K. Michel,
D. Bauer, R. Kienberger, and T. Metzger, Ultrafast thin-
operation over a much broader range of operating param-
disk multipass amplifier with 720 mJ operating at kilo-
eters than considered in the original proposal [12]. This hertz repetition rate for applications in atmospheric re-
includes operation at higher drive pulse energies, which search, Opt. Express 28, 30164 (2020).
bodes well for the development of high-repetition-rate, [6] S. Nagel, B. Metzger, D. Bauer, J. Dominik, T. Gottwald,
GeV-scale P-MoPAs. V. Kuhn, A. Killi, T. Dekorsy, and S.-S. Schad, Thin-disk
This work was supported by the UK Engineering and laser system operating above 10 kW at near fundamental
Physical Sciences Research Council (EPSRC) (Grant No. mode beam quality, Opt. Lett. 46, 965 (2021).
[7] Produit, Thomas, Walch, Pierre, Herkommer, Clemens,
EP/V006797/1), the UK Science and Technologies Facil-
Mostajabi, Amirhossein, Moret, Michel, Andral, Ugo,
ities Council (Grant No. ST/V001655/1]), InnovateUK Sunjerga, Antonio, Azadifar, Mohammad, André, Yves-
(Grant No. 10059294), UKRI funding (ARCHER2 Pi- Bernard, Mahieu, Benoı̂t, Haas, Walter, Esmiller, Bruno,
oneer Projects) and the Ken and Veronica Tregidgo Fournier, Gilles, Krötz, Peter, Metzger, Thomas, Michel,
Scholarship in Atomic and Laser Physics. This pub- Knut, Mysyrowicz, André, Rubinstein, Marcos, Rachidi,
lication arises from research funded by the John Fell Farhad, Kasparian, Jérôme, Wolf, Jean-Pierre, and
Oxford University Press Research Fund. This work Houard, Aurélien, The laser lightning rod project, Eur.
Phys. J. Appl. Phys. 93, 10504 (2021).
used the ARCHER2 UK National Supercomputing Ser-
[8] R. Paschotta, J. Aus der Au, G. Spühler, S. Erhard,
vice (ARCHER2 PR17125) https://www.archer2.ac. A. Giesen, and U. Keller, Passive mode locking of thin-
uk. This research used the open-source particle-in- disk lasers: effects of spatial hole burning, Appl. Phys. B
cell code WarpX [39] https://github.com/ECP-WarpX/ 72, 267 (2001).
WarpX, primarily funded by the US DOE Exascale Com- [9] T. Südmeyer, C. Kränkel, C. Baer, O. Heckl, C. Sara-
puting Project. Primary WarpX contributors are with ceno, M. Golling, R. Peters, K. Petermann, G. Huber,
LBNL, LLNL, CEA-LIDYL, SLAC, DESY, CERN, and and U. Keller, High-power ultrafast thin disk laser oscil-
lators and their potential for sub-100-femtosecond pulse
Modern Electron. We acknowledge all WarpX contribu-
generation, Appl. Phys. B 97, 281 (2009).
tors. [10] C. R. E. Baer, C. Kränkel, C. J. Saraceno, O. H. Heckl,
This research was funded in whole, or in part, by EP- M. Golling, R. Peters, K. Petermann, T. Südmeyer,
SRC and STFC, which are Plan S funders. For the G. Huber, and U. Keller, Femtosecond thin-disk laser
purpose of Open Access, the author has applied a CC with 141 W of average power, Opt. Lett. 35, 2302 (2010).
BY public copyright licence to any Author Accepted [11] M. Kaumanns, D. Kormin, T. Nubbemeyer, V. Pervak,
Manuscript version arising from this submission. and S. Karsch, Spectral broadening of 112 mJ, 1.3 ps
pulses at 5 kHz in a LG10 multipass cell with compress-
The input decks used for the PIC simulations presented
ibility to 37 fs, Opt. Lett. 46, 929 (2021).
in this paper are available at http://doi.org/10.5281/ [12] O. Jakobsson, S. M. Hooker, and R. Walczak, Gev-scale
zenodo.7734261 accelerators driven by plasma-modulated pulses from
kilohertz lasers, Phys. Rev. Lett. 127, 184801 (2021).
[13] T. Eidam, S. Hanf, E. Seise, T. V. Andersen, T. Gabler,
C. Wirth, T. Schreiber, J. Limpert, and A. Tünnermann,
Femtosecond fiber CPA system emitting 830 W average
∗
[email protected] output power, Opt. Lett. 35, 94 (2010).
[1] T. Tajima and J. M. Dawson, Laser electron accelerator, [14] T. Eidam, C. Wirth, C. Jauregui, F. Stutzki, F. Jansen,
Phys. Rev. Lett. 43, 267 (1979). H.-J. Otto, O. Schmidt, T. Schreiber, J. Limpert,
[2] D. Strickland and G. Mourou, Compression of amplified and A. Tünnermann, Experimental observations of the
chirped optical pulses, Optics Communications 56, 219 threshold-like onset of mode instabilities in high power
(1985). fiber amplifiers, Opt. Express 19, 13218 (2011).
[3] J. W. Dawson, J. K. Crane, M. J. Messerly, M. A. [15] C. Jauregui, T. Eidam, J. Limpert, and A. Tünnermann,
Prantil, P. H. Pax, A. K. Sridharan, G. S. Allen, Impact of modal interference on the beam quality of high-
D. R. Drachenberg, H. H. Phan, J. E. Heebner, power fiber amplifiers, Opt. Express 19, 3258 (2011).
C. A. Ebbers, R. J. Beach, E. P. Hartouni, C. W. [16] A. V. Smith and J. J. Smith, Mode instability in high
Siders, T. M. Spinka, C. P. J. Barty, A. J. Bayra- power fiber amplifiers, Opt. Express 19, 10180 (2011).
mian, L. C. Haefner, F. Albert, W. H. Lowder- [17] C. Jauregui, C. Stihler, and J. Limpert, Transverse mode
milk, A. M. Rubenchik, and R. E. Bonanno, High instability, Adv. Opt. Photon. 12, 429 (2020).
 9

[18] E. Esarey, P. Sprangle, J. Krall, A. Ting, and focusing and guiding of short laser pulses in ionizing gases
G. Joyce, Optically guided laser wake-field acceleration*, and plasmas, IEEE Journal of Quantum Electronics 33,
Physics of Fluids B: Plasma Physics 5, 2690 (1993), 1879 (1997).
https://doi.org/10.1063/1.860707. [32] A. J. Gonsalves, K. Nakamura, C. Lin, J. Os-
[19] N. E. Andreev, L. M. Gorbunov, V. I. Kirsanov, A. A. terhoff, S. Shiraishi, C. B. Schroeder, C. G. R.
Pogosova, and R. R. Ramazashvili, The theory of laser Geddes, C. Tóth, E. Esarey, and W. P. Leemans,
self-resonant wake field excitation, Physica Scripta 49, Plasma channel diagnostic based on laser centroid
101 (1994). oscillations, Physics of Plasmas 17, 056706 (2010),
[20] P. Mora and T. M. Antonsen, Jr., Kinetic mod- https://doi.org/10.1063/1.3357175.
eling of intense, short laser pulses propagating in [33] N. E. Andreev, L. M. Gorbunov, V. I. Kirsanov, K. Naka-
tenuous plasmas, Physics of Plasmas 4, 217 (1997), jima, and A. Ogata, Structure of the wake field in
https://doi.org/10.1063/1.872134. plasma channels, Physics of Plasmas 4, 1145 (1997),
[21] W. Zhu, J. P. Palastro, and T. M. Antonsen, Stud- https://doi.org/10.1063/1.872186.
ies of spectral modification and limitations of the [34] M. N. Rosenbluth and C. S. Liu, Excitation of plasma
modified paraxial equation in laser wakefield sim- waves by two laser beams, Phys. Rev. Lett. 29, 701
ulations, Physics of Plasmas 19, 033105 (2012), (1972).
https://doi.org/10.1063/1.3691837. [35] M. Siegrist, Self-focusing in a plasma due to pondero-
[22] URL_will_be_inserted_by_publisher. motive forces and relativistic effects, Optics Communi-
[23] E. Esarey, P. Sprangle, M. Pilloff, and J. Krall, The- cations 16, 402 (1976).
ory and group velocity of ultrashort, tightly focused laser [36] H. S. Brandi, C. Manus, G. Mainfray, T. Lehner,
pulses, J. Opt. Soc. Am. B 12, 1695 (1995). and G. Bonnaud, Relativistic and ponderomotive
[24] E. Esarey and W. P. Leemans, Nonparaxial propagation self-focusing of a laser beam in a radially in-
of ultrashort laser pulses in plasma channels, Phys. Rev. homogeneous plasma. i. paraxial approximation,
E 59, 1082 (1999). Physics of Fluids B: Plasma Physics 5, 3539 (1993),
[25] C. Kurtz and W. Streifer, Guided waves in inhomo- https://doi.org/10.1063/1.860828.
geneous focusing media part i: Formulation, solution [37] B. Z. Djordjević, C. Benedetti, C. B. Schroeder,
for quadratic inhomogeneity, IEEE Transactions on Mi- E. Esarey, and W. P. Leemans, Laser mode
crowave Theory and Techniques 17, 11 (1969). control using leaky plasma channels, AIP
[26] W. Firth, Propagation of laser beams through inhomo- Conference Proceedings 1812, 040013 (2017),
geneous media, Optics Communications 22, 226 (1977). https://aip.scitation.org/doi/pdf/10.1063/1.4975860.
[27] O. Jakobsson, S. M. Hooker, and R. Walczak, [38] B. Z. Djordjević, C. Benedetti, C. B. Schroeder,
Supplemental material for gev-scale accelerators E. Esarey, and W. P. Leemans, Filtering higher-
driven by plasma-modulated pulses from kilohertz order laser modes using leaky plasma chan-
lasers, http://link.aps.org/supplemental/10.1103/ nels, Physics of Plasmas 25, 013103 (2018),
PhysRevLett.127.184801 (2021). https://doi.org/10.1063/1.5006198.
[28] P. Gibbon, The self-trapping of light waves by beat-wave [39] L. Fedeli, A. Huebl, F. Boillod-Cerneux, T. Clark,
excitation., Physics of Fluids B: Plasma Physics 2, 2196 K. Gott, C. Hillairet, S. Jaure, A. Leblanc, R. Lehe,
(1990), https://doi.org/10.1063/1.859401. A. Myers, C. Piechurski, M. Sato, N. Zaim, W. Zhang,
[29] W. Mori, The physics of the nonlinear optics of plas- J. Vay, and H. Vincenti, Pushing the frontier in the design
mas at relativistic intensities for short-pulse lasers, IEEE of laser-based electron accelerators with groundbreaking
Journal of Quantum Electronics 33, 1942 (1997). mesh-refined particle-in-cell simulations on exascale-class
[30] C. E. Max, J. Arons, and A. B. Langdon, Self-modulation supercomputers, in 2022 SC22: International Conference
and self-focusing of electromagnetic waves in plasmas, for High Performance Computing, Networking, Storage
Phys. Rev. Lett. 33, 209 (1974). and Analysis (SC) (SC) (IEEE Computer Society, Los
[31] E. Esarey, P. Sprangle, J. Krall, and A. Ting, Self- Alamitos, CA, USA, 2022) pp. 25–36.
 Supplemental Material for “Stability of the Modulator in a Plasma-Modulated
Plasma Accelerator”

J. J. van de Wetering,1, ∗ S. M. Hooker,1 and R. Walczak1, 2

1
John Adams Institute for Accelerator Science and Department of Physics,
University of Oxford, Denys Wilkinson Building,
Keble Road, Oxford OX1 3RH, United Kingdom
2
Somerville College, Woodstock Road, Oxford OX2 6HD, United Kingdom
(Dated: March 27, 2023)
arXiv:2303.14032v1 [physics.plasm-ph] 24 Mar 2023

PARAXIAL DESCRIPTION OF PULSES IN where ∆⊥ ≡ (1/r)(∂/∂r)(r∂/∂r) + (1/r2 )∂ 2 /∂θ2 and

PLASMA CHANNELS δn0 (r) = n0 (r) − n00 . This PDE, coupled to a self-
consistent wake solution for δn, fully describes the evo-
The Envelope Model lution of the laser envelope in the axisymmetric, weakly
relativistic, quasi-static linear wakefield regime.
Considering only the high frequency fields associated
with the laser pulse, from Ampère’s law we find that the
normalized vector potential a = eA/me c evolves accord-
The Paraxial Approximation
ing to the wave equation
" #
∂2 2
ωp2 To further simplify Eq. (SM–3) to the paraxial ap-
−c ∆+ (n0 + δn) a = 0 (SM–1)
∂t2 n0 γ proximation, we assume that the discrepancy between
the group velocity and speed of light in vacuum is negli-
where γ = (1 + p2 + a2 )1/2 , p = γme vhf and vhf rep- gible (vg ≈ c) and that the pulse envelope has a large
resents the high frequency quiver velocity of electrons enough longitudinal extent that the term ∂ 2 /∂ξ 2 can
which ignores the low frequency bulk fluid velocity from be neglected. We also assume that the envelope evo-
a wakefield response, n0 is the unperturbed pre-formed lution is slow relative to the carrier frequency so that the
plasma channel density and δn represents the change in terms vg ∂ 2 /∂ξ∂τ and ∂ 2 /∂τ 2 time derivative terms can
density due to the wake driven by the laser pulse. We be dropped. This yields a generalized nonlinear parax-
have also assumed that the electrostatic response is small ial wave equation in an axisymmetric plasma channel
compared to the transverse current. To construct an en- n0 (r) = n00 + δn0 (r)
velope model for the laser evolution, consider a pulse of
the form  
i ∂ c2
+ 2 ∆⊥ a =
a = êL a(r, θ, z, t) exp[i(kL z − ωL t)] (SM–2) ωL ∂τ 2ωL
ωp2  
where the laser frequency and wavenumber in free space
2 δn0 (r) + δn(r, ξ; |a|2 ) − n0 (r)|a|2 /4 a (SM–4)
ωL , kL are constants and êL is the polarization of the 2ωL n0
laser, which we will treat as linearly polarized throughout
this paper. It is convenient to shift from the lab frame where the nonlinearities come from relativistic effects and
axial coordinate z and time t to co-moving coordinates the interaction between the pulse and its own excited
ξ = z − vg t, and τ = t, where vg /c = (1 − ωp0 2 2 1/2
/ωL ) wake.
is defined as the group velocity of electromagnetic plane
waves in uniform plasma at the on-axis plasma channel
density n00 = n0 (r = 0), ωp0 = ωp (r = 0), which may
Matched Plasma Channels
differ from the group velocity of a tightly focused laser
pulse [1]. Substituting a pulse of this form into Eq. (SM–
1) in the weakly relativistic and linear wake limit yields Using Eq. (SM–4) we can derive the form of the
the following expression for the evolution of the envelope matched plasma channel, which confines a Gaussian
a(r, θ, ξ, τ ) beam with a constant spot size as it propagates. Assum-
h ing no contributions from relativistic self-focusing nor
∂ ∂2 ∂2 from plasma wakes, this yields the linear paraxial wave
− 2iωL − 2vg − c2 ∆⊥ + 2
∂τ ∂ξ∂τ ∂τ equation in a channel
∂ 2 ω 2
p
i
− (c2 − vg2 ) 2 + δn0 + δn − n0 |a|2 /4 a = 0  
∂ξ n0 i ∂ c2 ωp2 δn0 (r)
+ 2 ∆ ⊥ a = 2n a. (SM–5)
(SM–3) ωL ∂τ 2ωL 2ωL 0
 2

Substituting a Gaussian pulse with a fixed spot size w0 where integer l ≥ 0 is the transverse index and the
yields the following stable solution Hl functions are the (physicist’s) Hermite polynomials.
  Note that the laser centroid and spot size oscillation fre-
r2 2c2
a(r, ξ, τ ) = a0 f (ξ) exp − 2 − iωL τ 2 2 , quencies are the same in both 2D slab and 3D cylin-
w0 ωL w 0
drical geometry, so we would expect processes tied to
n0 (r) = n00 + ∆n(r/w0 )2 , ∆n = (πre w02 )−1 (SM–6) spot/centroid oscillations to behave similarly in 2D and
where the longitudinal profile of the pulse, 0 ≤ f (ξ) ≤ 1, 3D. However, the rate of spectral modulation does de-
is assumed to be sufficiently slowly varying for the parax- pend on the dimensionality. For example, consider the
ial approximation to hold, n00 is an arbitrary on-axis shallow channel limit ∆n  n00 where the seed wake can
density and re is the classical electron radius. This re- be written in the form

sult for the matched channel can also be shown to confine 
1 1D
2 2
all Laguerre-Gaussian modes with envelope solutions of δn = δns cos(kp0 ξ) e−2x /w0 2D (SM–9)
the form 
 −2r2 /w02
e 3D
apm (r, θ, ξ, τ ) = where δns denotes the on-axis seed wake amplitude. As-
 
2c2 suming that the modulating drive pulse remains in the
αpm (ξ) exp −iωL τ (2p + |m| + 1) 2 2 LGpm ,
ωL w0 fundamental channel mode, the spectral modulation rate
LGpm (r, θ) = parameter Ωs is given by

s √ !|m|    2 1 1D
p! 2r r2 2r 1 ωp0 δns 
2 √
exp − 2 + imθ Lp|m|
2 2
, Ωs = −2x /w
hHG0 |e 0 |HG0 i = 1/ 2 2D
(p + |m|)! w0 w0 w02 4 ωL n00  2 2
hLG00 |e−2r /w0 |LG00 i = 1/2 3D
hLGp0 m0 |(. . .)|LGpm i ≡ (SM–10)
Z 2π Z ∞
2
dθ rdrLG∗p0 m0 (. . .)LGpm (SM–7) hence 2D PIC √ simulations are expected to spectrally
πw02 0 0 modulate ∼ 2 times√faster than predicted by 3D cylin-
where the integers p ≥ 0 and m are the radial and az- drical theory (and ∼ 2 times slower than 1D theory).
|m| The suppression of spectral modulation by wake
imuthal indexes respectively and the Lp functions are
the generalized Laguerre polynomials. Note that it is phase-front curvature caused by the plasma chan-
the interference between the fundamental and first az- nel, as described in the paper, also changes depend-
imuthal and radial modes that set the laser centroid and ing on the dimensionality. This is again primarily
spot size oscillation frequencies ωc = ωw /2 = 2c2 /ωL w02 caused by the difference between hHG0 |δn(x, ξ)|HG0 i
respectively. and hLG00 |δn(r, ξ)|LG00 i, resulting in the suppression
towards the pulse tail being more pronounced in the 3D
cylindrical case. There is also a smaller effect due to
2D SLAB VS 3D CYLINDRICAL GEOMETRY differences in the wake structure itself between 2D and
3D.
To justify the use of 2D PIC simulations to study the
stability of the plasma modulator, we outline the pulse
SPECTRAL PHASE FOR PULSE TRAIN
propagation theory in 2D slab geometry. The physical
FORMATION
description of seeded spectral modulation in 2D slab and
3D cylindrical are similar, but they have some key differ-
ences which come from the transverse Laplacian opera- As derived previously in 1D by Jakobsson et al [2],
tor ∆⊥ taking a different form. The matched parabolic to first order the spectral modulation takes the form of
channel for a Gaussian beam with spot size w0 still takes a set of sidebands of angular frequencies ωm = ωL +
the same form n0 (x) = n00 + ∆n(x/w0 )2 , but the con- mωp0 , where m = ±1, ±2, ±3, . . . is the sideband order.
fined modes are instead described by Hermite-Gaussian These sidebands will each have a relative spectral phase
functions ψm = −|m|π/2. With a more detailed analysis, which
  can be derived from Eq. (SM–26) assuming a narrow
2c2 −1
bandwidth τdrive  ωp0 and that light only remains in
al (x, ξ, τ ) = αl (ξ) exp −iωL τ (l + 1) 2 2 HGl ,
ωL w0 the fundamental channel mode, the full relative spectral
r   √ !
1 x2 2x phase of the modulated drive pulse before compression
HGl (x) = exp − 2 Hl , into a pulse train can be approximately described by the
2l l! w0 w0
s following nearest integer staircase function
Z ∞   
2  ω − ωL  π
hHGl0 |(...)|HGl i ≡ dxHG∗l0 (...)HGl (SM–8) 
ψ(ω) = − nint
πw02 −∞ ωp0 2. (SM–11)
 3
√
To form a pulse train, we wish to remove this spectral where we have included a 1/ 2π normalization in the
phase from the pulse. However, it is not practical to re- definition of the convolution for notational convenience.
move a spectral phase of this form with a dispersive optic. Ignoring relativistic effects, we can split kp2 (r, ξ, η) into
Instead, we can take advantage of the narrow bandwidth contributions from a pre-formed axisymmetric plasma
of each of the sidebands to approximately remove this channel n0 (r) and a plasma wake δn(r, ξ, η)
spectral phase by applying a continuous dispersion func-
 
tion of the form 1 ∂ 1 2 n0 (r)
  i(1 + k/kL ) + 2 ∆⊥ − 2 2 ak =
 ω − ωL  π kL ∂η 2kL kL w0 ∆n
opt
ψ (ω) = +    . (SM–12)  
ωp0  2 2 δn(r, ξ, η)
2 w2 ∗ ak (SM–17)
kL ∆n
This form of ψ opt (ω) was used in the main paper to eval- 0 k

uate the pulse trains that can be formed from the spec-
trally modulated drive pulses. where ∆n ≡ (πre w02 )−1 is the channel depth parameter.
Ignoring the wake for now, a matched parabolic channel
n0 (r) = n00 + ∆n(r/w0 )2 can guide any linear combina-
NON-PARAXIAL DESCRIPTION OF SEEDED tion of Laguerre-Gaussian modes of the form
SPECTRAL MODULATION IN PLASMA
CHANNELS
apm pm pm
k (r, θ, k, η) = αk (k) exp[−ik̃k (k)η]LGpm (r, θ) ,
2 2p + |m| + 1 + n00 /∆n
Unlike the paraxial equation, this description will in- k̃kpm (k) = ,
clude group velocity dispersion as well as asymmetries kL w02 1 + k/kL
between the dynamics of the generated Stokes and anti- LGpm (r, θ) =
Stokes sidebands in the plasma modulator. We will start s √ !|m|    2
by following the procedure outlined by [1] for deriving the p! 2r r2 2r
exp − 2 + imθ Lp|m| .
non-paraxial description of pulses in fully ionized plasma, (p + |m|)! w0 w0 w02
but here we will include contributions from a parabolic (SM–18)
plasma channel and a seed wake. We begin with the full
3D wave equation for a laser pulse propagating in a fully Assuming that k/kL  1 remains valid for the majority
ionized plasma of the pulse ak , (i.e. for pulse durations that are not
 
1 ∂2 too short relative to the laser cycle period), we can also
∆ − 2 2 − kp2 a = 0 (SM–13) write the Laguerre-Gaussian mode solutions in real space
c ∂t
to first order in the form
where a = eA/me c is the normalized vector potential
 
and we have neglected higher order plasma source terms apm (r, θ, ξ, η) = αpm (ξ) ∗ exp[−ik̃kpm (k)η] LGpm (r, θ)
from having a finite electrostatic potential [3]. We now ξ
switch to new coordinates ξ = z − ct, η = (z + ct)/2, ≈ αpm (ξ + k̃0pm η/kL ) exp(−ik̃0pm η)LGpm (r, θ) ,
which yields 2
ωp0 2
  k̃0pm /kL = + (2p + |m| + 1) 2 2 (SM–19)
∂2 2
2kL c2 kL w0
2 + ∆⊥ − kp2 a = 0 . (SM–14)
∂ξ∂η
We seek envelope solutions in the form a = where αpm (ξ) is the inverse Fourier transform of αkpm (k).
[a exp(ikL ξ) + c.c.] êL /2 where kL is a constant. This This describes the first order group velocity dispersion
yields the envelope PDE and wavenumbers of Laguerre-Gaussian modes due to
    the on-axis plasma density and finite spot size effects.
∂ ∂ 2 We can see that for a pulse primarily in the fundamen-
2 ikL + + ∆⊥ − kp a(r, ξ, η) = 0 . (SM–15)
∂ξ ∂η tal mode, after every spot size oscillation the first ra-
dial mode will fall behind the fundamental mode by a
We then take the Fourier transform in the variable ξ and
laser wavelength (and similarly for centroid oscillations).
apply the convolution theorem
  Hence we eventually need to take this group velocity dis-
∂
persion into account if the pulse propagates over many
2i (kL + k) + ∆⊥ ak = kp2 k ∗ ak ,
∂η spot size oscillations in a long plasma channel.
Z ∞
1 If we can treat the wake contribution as a small pertur-
ak (r, k, η) = √ dξe−ikξ a(r, ξ, η) , bation to the matched parabolic plasma channel, we can
2π −∞
Z ∞ use time-dependent perturbation theory to calculate the
1
(f ∗ g)(k) := √ f (k 0 )g(k − k 0 )dk 0 (SM–16) transitions between the channel modes with the following
2π −∞ expression
 4

∂αkpm (k, η) XD   δn(r, ξ, η)   0 0  E

 0 0 
i(1 + k/kL ) exp(−ik̃kpm η) = kc LGpm  ∗ αkp m (k, η) exp(−ik̃kp m η) LGp0 m0
∂η 0 0
∆n k
pm
(SM–20)

where kc = kw /2 = 2/kL w02 are the centroid and spot We now choose to work in the shallow channel limit
size oscillation wavenumbers respectively. ∆n  n00 to ignore the non-separable transverse wake
Assume that the short seed pulse has the same wave- structure introduced by the channel [4]. Note that using
length as the drive pulse, has many laser cycles in its a square-like channel would achieve a similar effect, but
duration, is in the fundamental mode and does not ap- would have Bessel modes rather than Laguerre-Gaussian
preciably deplete. The seed pulse will then have a group modes. The wake excited by the seed pulse considering
velocity of both plasma and finite spot size effects on its group ve-
locity in this limit takes the form
2
ωp0 2
vg,s /c = 1 − k̃000 /kL = 1 − 2 − 2 2. (SM–21)
2kL c 2 kL w0 h  i
δn(r, ξ, η) = δns LG200 (r) cos kp0 ξ + k̃000 η/kL .
Note that the group velocity is slowed by both plasma
(SM–23)
and finite spot size effects. This means that in general
the seed wake will be in the form
Substituting this seed wake into Eq. (SM–20) yields the
δn(r, ξ, η) = δn(r, ξ + k̃000 η/kL ) . (SM–22) non-paraxial plasma modulator equation

1 + k/kL ∂αkpm 2
1 ωp0 δns X p0 m0 h  0 0  i
pm pm
i = 2 c2 n αk±kp0 exp −i k̃ k±kp0 − k̃ k ± (k p0 /k L )k̃ 00
0 η hLGpm |LG200 |LGp0 m0 i (SM–24)
kL ∂η 4 kL 00 0 0
pm

where we can now clearly see that the seed wake modulating a pulse with an initially short bandwidth will generate
Stokes and anti-Stokes sidebands in k-space separated by integer multiples of kp0 . Assuming that k, kp0  kL , the
wavenumber shifts approximate to
 0 0 
pm pm
k̃k±k p0
− k̃ k ± (k p0 /k L )k̃ 00
0 /kc ≈ [2(p0 − p) + (|m0 | − |m|)] (1 − k/kL ) ∓ (2p0 + |m0 |) (kp0 /kL ) . (SM–25)

Using this expression and assuming that most of the light remains in the fundamental mode and that no azimuthal
modes are present, we can approximate the plasma modulator equation as a two-level system of the fundamental and
first radial modes
1 + k/kL ∂αk00 2
1 ωp0 δns  00 10

i = 2 2
αk±kp0 + 12 αk±k exp [−ikw (1 − k/kL ∓ kp0 /kL ) η] ,
kL ∂η 8 kL c n00 p0

1 + k/kL ∂αk10 2
1 ωp0 δns  1 00 1 10

i = 2 c2 n 2 αk±k exp [ikw (1 − k/k L )η] + 2 αk±k exp [±ik w (k p0 /k L ) η] (SM–26)
kL ∂η 8 kL 00
p0 p0

which to zeroth order in k/kL and kp0 /kL gives the same transverse mode transitions than the anti-Stokes. This
result given by the paraxial equation used in the paper asymmetry in the dynamics between the Stokes and anti-
(apart from the slightly different propagation variable η). Stokes sidebands is necessary to explain the transverse
However, to first order we see symmetry-breaking be- separation of Stokes and anti-Stokes light observed in
tween the Stokes and anti-Stokes sidebands which was PIC simulations in the regime where the self-wake of the
not captured by the paraxial equation. We see here that modulating drive pulse is no longer negligible.
the Stokes sidebands are generated faster by the seed
wake and also undergo faster spot size oscillations and
 5

PARTICLE-IN-CELL SIMULATIONS were parameterized in the following form

nsquare
0 (r) − n00
=
∆n


 (r/w0 )10 0 ≤ r < 1.2w0

1.210 1.2w0 ≤ r < 1.2w0 + d


Two-dimensional simulations were performed with the 0 −d


 1.210 1 − r−1.2wd 1.2w0 + d ≤ r < 1.2w0 + 2d
PIC code WarpX (version 22.07) [5]. Results from eight 
0 r ≥ 1.2w0 + 2d
simulations are included in the paper with laser-plasma
parameters and respective figures outlined in Table I. All nparabolic
0 (r) − n00
=
of these eight simulations were performed at an on-axis ∆n

density of n00 = 2.5 × 1017 cm−3 in a modulator of length  (r/w0 )2 0 ≤ r < 2w0


Lmod = 110 mm with seed and drive pulses with the same 4 2w0 ≤ r < 2w0 + d
wavelength λL = 1030 nm and spot size w0 = 30 or 50 r−2w0 −d



 4 1 − 2w0 + d ≤ r < 2w0 + 2d
µm. The seed and drive pulses were polarized out of 
 d
and in the plane of simulation respectively. All simula- 0 r ≥ 2w0 + 2d
tions had a longitudinal resolution of ∆z = λL /50 and (SM–27)
transverse resolution of ∆x = λL /2.5 using a second or-
der Yee field solver and perfectly matched layer (PML) where d = 10 µm. Note that all simulations were initial-
transverse boundary conditions. The transverse window ized with a laser pulse with a gaussian transverse profile
size for all simulations was at least 2.67w0 away from of spot size w0 , which differs slightly from the fundamen-
the axis. The “square” and “parabolic” plasma channels tal guiding mode of the square channel.

PIC Simulation Parameters

Simulation Figures w0 (µm) Wseed (mJ) Wdrive (J) τseed (fs) τdrive (ps) Channel
(i) 2, 3 30 50 0.6 40 1.0 parabolic
(ii) 3, 4a 30 50 0.6 40 1.0 square
(iii) 3 50 139 1.67 40 1.0 parabolic
(iv) 4b, 5b 30 50 1.2 40 1.0 square
(v) 4c 30 50 2.4 40 4.0 square
(vi) 5a 30 50 1.2 40 0.25 square
(vii) 5c 30 50 1.2 40 2.0 square
(viii) 5d 30 50 1.2 40 4.0 square

TABLE I

This research was funded in whole, or in part, by EP- hertz lasers, Phys. Rev. Lett. 127, 184801 (2021).
SRC and STFC, which are Plan S funders. For the [3] E. Esarey and W. P. Leemans, Nonparaxial propagation
purpose of Open Access, the author has applied a CC of ultrashort laser pulses in plasma channels, Phys. Rev.
BY public copyright licence to any Author Accepted E 59, 1082 (1999).
Manuscript version arising from this submission. [4] N. E. Andreev et al., Structure of the wake field in plasma
channels, Phys. Plasmas 4, 1145 (1996).
[5] L. Fedeli, A. Huebl, F. Boillod-Cerneux, T. Clark,
K. Gott, C. Hillairet, S. Jaure, A. Leblanc, R. Lehe,
A. Myers, C. Piechurski, M. Sato, N. Zaim, W. Zhang,
∗
J. Vay, and H. Vincenti, Pushing the frontier in the design
[email protected] of laser-based electron accelerators with groundbreaking
[1] E. Esarey, P. Sprangle, M. Pilloff, and J. Krall, Theory and mesh-refined particle-in-cell simulations on exascale-class
group velocity of ultrashort, tightly focused laser pulses, supercomputers, in 2022 SC22: International Conference
J. Opt. Soc. Am. B 12, 1695 (1995). for High Performance Computing, Networking, Storage
[2] O. Jakobsson, S. M. Hooker, and R. Walczak, Gev-scale and Analysis (SC) (SC) (IEEE Computer Society, Los
accelerators driven by plasma-modulated pulses from kilo- Alamitos, CA, USA, 2022) pp. 25–36.