1. Introduction
Particle accelerators are among the largest and most expensive scientific instruments. Their large footprint is dictated by the modest acceleration gradient (in tens of MeV per meter), limited by the breakdown of metallic accelerating cavities. Accelerating electrons in the fully or partially ionized medium (i.e. a plasma) lifts this limitation, making accelerators thousands of times smaller, literally “tabletop.” Since plasmas are free of the damage limits of conventional accelerators, they may build up TV/m fields within structures propagating at a nearluminal speed. First ideas of harnessing collective plasma fields to actively control the phase space of a highenergy electron beam (ebeam), were brought into the world over 60 years ago [1, 2, 3]. Yet, it was not until the first decade of this century that the accelerator community started witnessing systematic progress in plasma acceleration of electron and positron beams [4, 5, 6, 7, 8].
Competition with conventional linear accelerators in generation of quasimonoenergetic (QME) ebeams requires independently driven nearluminal, highfield plasma structures, such as Langmuir plasma waves [9, 10, 11] or cavities (“bubbles”) of electron density [12, 13, 14, 15]. The accelerating buckets must retain their shape in the course of propagation or change the shape and potentials in a controllable fashion to avoid degradation of the externally injected ebeam. It is equally important for the injection mechanism to ensure subsequent acceleration of the beam without picking up additional unwanted charge (the “dark current”). To this end, control over driver evolution and the plasma density profile is of paramount importance.
Driving the accelerating plasma structures with a radiation pressure of a femtosecond, multiterawatt (TW) laser pulse (hence the term “laser wakefield”) provides abundant opportunities for alloptical control of both injection and acceleration processes [16, 17]. Early demonstrations of QME laserplasma acceleration [4, 5, 6] were a perfect example of this control. It was not until the optical driver closely matched the plasma parameters, to ensure its propagation as a whole, without breaking up longitudinally or transversely, that the long coveted QME electron bunches were realized. The matching [18] made it possible for the laser to produce a “bubble” almost completely devoid of electrons in its immediate wake [19, 20]. The bubble acts at the same time as a nonlinear waveguide for the laser pulse and an accelerating bucket for the electrons. The ponderomotive force of the pulse maintains the bubble shape. It expels all electrons facing the pulse (hence the term “blowout regime”), while the bulk electrons are attracted to the propagation axis. The difference between attractive force due to the charge separation and the repulsive radial ponderomotive force controls the trajectories of electrons making up the bubble shell. The resulting soft channel, approximately replicating the threedimensional (3D) shape of the pulse [19], evolves in a lockstep with the optical driver [15, 16, 20, 21, 22, 23, 24, 25]. In consequence, it traps initially quiescent background electrons, eliminating the need for an external photocathode [16, 21, 22]. Notably, in the regimes featuring production of lowemittance ebeams, only a tiny minority of electrons making up the bubble shell are trapped and subsequently accelerated. Their collective fields, i.e., beam loading [26], contribute very little to the bubble evolution and are unable to change the kinetics of selfinjection [21, 22, 23].
Two fundamental relativistic optical phenomena underpin the matching conditions [18]. The first one is relativistic selffocusing. As electrons oscillate in the field of focused laser beam, the relativistic increase in their mass and, hence, the nonlinear refractive index reach maximum near axis, where the laser intensity is the highest. The plasma thus acts as a focusing fiber, compensating for diffraction. If the pulse power
P
exceeds the critical value
Pcr=16.2nc/n0
GW [27] even by a few percent, the selffocusing will saturate [28] only at the point of full electron blowout [13]. Here,
n0
is the background electron density,
nc=meω02/4πe2
is the critical density for radiation with a frequency
ω0
, and
me
and
−e
are the electron rest mass and charge. Matching the spot size of the incident pulse to the value
balances the force due to the charge separation and the ponderomotive force acting upon the electron at the boundary of the bubble. Here,
kp=ωpe/c=1.88n20
μm^{−1} is the plasma wave number,
ωpe=4πe2n0/me≪ω0
is the Langmuir plasma frequency,
c
is the speed of light in vacuum, and
n20
is the background density in units
1020
cm^{−3}. The matched pulse propagates in a single filament confined to the bubble. Conversely, strong mismatching results in a transverse breakup of the pulse, massive energy loss to the plasma, and disruption of selfguiding [29, 30]. The other key physical phenomenon, which limits electron energy gain, is selfphasemodulation, viz. accumulation of frequency red shift, imparted by the wake, at the selfphasemodulation. As the pulse propagates, this shift reaches a large fraction of
ω0
, while the negative group velocity dispersion (GVD) in the plasma delays these lowfrequency components, etching away the pulse leading edge. In the frame of reference comoving with the bubble, these components start to accumulate around the point where electron density drops to zero
^{1}
, building up an optical shock with a subcycle rising edge [16, 22, 23, 24, 25]. If the bubble were nonevolving, the etching velocity would be its phase velocity, which defines the electron dephasing length,
Ld=2/3nc/n0rm
[18]. In addition, etching velocity defines the pulse energy loss, which also limits electron energy gain. The pulse loses most of its energy and is unable to drive the bubble after a distance
Ldepl=nc/n0cτL
(the depletion length), where
τL
is the duration of the incident pulse. Matching the dephasing and depletion lengths, so that
τL=2rm/3c
, promises to maximize the acceleration efficiency and, possibly, reduce electron energy spread via phase space rotation at the end of acceleration cycle. Under the matching condition, the maximal energy gain scales as [18]
Here,
λ0,μm
is the laser pulse wavelength,
λ0=2πc/ω0
, in microns. To ensure robust selfguiding and preserve selfinjection, the power ratio must be at least
P/Pcr≡κ>10
, or
n20λ0,μm2>1.8×10−4κPPW−1>1.8×10−3PPW−1
. This, in combination with (2), yields a rather discouraging scaling,
According to (3), GeV energy gain in the matched regime requires at least 117 TW laser power, or 3.75 J per 32 fs matched pulse. The matched plasma density is, in this case,
n0=2.4×1018
cm^{−3}, and the dephasing/depletion length is
Ld=6.9
mm. Laboratory experiments with cmlength gas jets from slit nozzles had approached this regime very closely, demonstrating backgroundfree ebeams with the energy up to 900 MeV, yet at the repetition rate below 10 Hz [35]. Conversely, generating these nearGeV ebeams at a kHz repetition rate, for the applications dependent on dosage, would call for a 4 kW averagepower laser amplifier, a technology of the distant future [36, 37]. Evidently, existing sub–50 TW systems are limited to the modest sub450 MeV yields.
Apart from frustrating the production of GeV beams at a high repetition rate, this matching strategy only partly solves the problem of ebeam quality. While aiming to stabilize transverse dynamics and avoid filamentation of the drive pulse, the physical arguments leading to the scaling (2) assume that the pulse selfcompression remains unaltered. Yet this process, apart from limiting the energy gain, destroys ebeam most assuredly if acceleration extends through the pulse depletion. (A plethora of evidence exists to this effect, both in laboratory experiments and numerical simulations [16, 19, 22, 23, 24, 25, 38, 39, 40, 41, 42].) To enable a new generation of compact particle and radiation sources [43, 44], one has to bypass the limitations this of scaling by designing an optical driver resilient to selfphasemodulation and selfcompression. Photon engineering of this kind, aiming to produce ebeams capable to emit quasimonochromatic, highflux γray pulses via Thomson (or inverse Compton) scattering [44], is the focus of this chapter.
Inverse Compton scattering is an emerging radiation generation technique [25, 44, 45, 46, 47, 48, 49, 50, 51, 52], which has already shown its potential for obtaining quasimonochromatic, strongly collimated γray pulses through the collision of a short QME ebeam and a midIR to UV interaction laser pulse (ILP) [53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73]. During the interaction, relativistic electrons, propagating at an angle to the ILP, experience its Lorentzcompressed wave front, the maximum compression occurring along the ebeam direction. As they oscillate in the ILP electromagnetic field, electrons emit radiation, scattering the compressed wave front. An observer in the far field thus detects an angular distribution of highenergy photons, with the energy being the highest for a detector placed in the ebeam direction. For the headon collision, the ILP photon energy is Doppler upshifted by a factor
4γe2
, where
γe
is the electron Lorentz factor. A beam of 900 MeV electrons thus converts 1.5 eV ILP photons into 19 MeV γphotons. As the energy of emitted photons is much lower than the electron energy, the recoil is negligible. This lowenergy semiclassical limit of the general quantummechanical inverse Compton scattering, known as Thomson scattering (TS), is the subject of this chapter. Ebeams from conventional accelerators [53, 54, 55, 56, 57, 58, 59, 60, 61, 62], produce multipicosecond TS γray pulses. These have a high degree of polarization and are thus attractive as ebeam diagnostics [53, 54]. Their other applications are generation of polarized positrons from dense targets [55] and nuclear resonance fluorescence studies [56, 57, 58, 59, 60, 61]. However, the large footprint of conventional accelerators makes such radiation sources scarce and busy user facilities. In addition, the large (cmscale) size of the radiofrequency–powered acceleration cavities makes it difficult to generate and synchronize ebeams (and, hence, TS γray pulses) on a subpicosecond time scale relevant to highenergy density physics [74]. Luckily, a miniature LPA offers an alternative technical solution that permits production of even shorter (viz. femtosecond), yet highcurrent (viz. kA) ebeams [75]. To drive narrowband TS γray sources, these beams have to meet some minimal requirements, such as a combination of a nearGeV energy with a percentscale energy spread, a fivedimensional (5D) brightness above
1016
A/m^{2} [76], and preferably absent lowenergy background. These requirements, in combination with the kHzscale repetition rate dictated by the applications, are clearly conflicting even for the most ambitious laser technology [77, 78]. LPA experiments, guided by the theoretical scaling (2), are presently struggling to reach this level of performance. Typically, acceleration through pulse depletion, carried out in pursuit of ever higher energy, consistently builds up massive energy tails in the ebeams [39, 40, 41, 42]. These beams produce a largebandwidth γray TS signal [63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73], which is incompatible with applications in nuclear photonics and radiography [48, 49, 60]. The current trend is to use the existing lowquality beams and try extending the highenergy tail of the photon distribution beyond 10 MeV, by using higher harmonics of the ILP [72], or by using fewGeV electrons from singleshot petawatt LPA facilities [73], or by employing an ILP of relativistic intensity [68].
Seeking the remedy to this situation, we take advantage of the fact that the LPA ebeams readily lend themselves to alloptical manipulation. Modifying the drive pulse dynamics, through a judicious choice of its phase and shape, alters kinetics of electron selfinjection. This, in turn, introduces modulations to the ebeam current and/or imparts a chirp to its longitudinal momentum. As the ebeam phase space imprints itself onto the spectrum of emitted photons, alloptical control of electron source enables tailoring the TS γray signal [25, 52].
As explained earlier, the plasma response compresses the optical driver of a conventional LPA (i.e. a transformlimited multiTW pulse) into a subcycle relativistic optical shock; this happens long before electron dephasing. The shock snowplows the ambient plasma electrons, causing electron density pileup inside the shock and a multifold increase in the field of charge separation behind it [16, 23]. The resulting uncontrolled elongation of the bubble causes massive continuous injection of electrons from its shell. Because of this dark current, caused by the uncompensated adverse optical process, maximization of the energy gain conflicts with the preservation of ebeam quality. We propose to resolve this conflict by incoherently mixing the pulse at the fundamental frequency with a frequencyupshifted pulse of the same, or lower, energy (on a subJoule scale) [24, 52]. As the photon diffusion rate due to GVD drops as the frequency grows, the blueshifted stack component is resilient to selfcompression. Because of the strong frequency dependence of the diffusion rate (
∼ω−3
) [79], even a modest 25% frequency upshift appears to be sufficient [24]. Simulations show that even the stack of fully overlapping components, in the fashion of Ref. [80], remains resilient to selfcompression (at least on the time scale of electron dephasing). The presence of the almost undeformable blue component does not permit formation of the intensity gradients at the subcycle scale. In the absence of the optical shock, the bubble expansion and, hence the dark current, is insignificant. The particle flux and charge in the energy tail drop multifold in comparison to the reference case of a transformlimited optical driver (the latter complying fully with the scaling (2)). Advancing the blueshifted component by
T∼τL
improves the situation even further. Emulating, in this way, a piecewise, largebandwidth negative chirp, we essentially place a protective screen ahead of the vulnerable unshifted tail. As this “hard hat” plows through the plasma, expelling background electrons, the soft tail maintains the bubble shape, thus defining kinetics of selfinjection. In the regime of our simulations, the dephasing length, defined by the etching of the head, extends by almost 80%, while the electron energy doubles against the reference case, reaching almost 900 MeV over 2.5 mm acceleration distance. Regardless of the time delay, very quiet injection keeps the ebeam brightness above
4×1016
A/m^{2}, favoring the use of these beams in Thomson sources [76]. In the case of a timedelayed stack, extracting the ebeam before dephasing (using, for instance, a gas cell target of variable length [81]), thus changing the ebeam energy in the interval 400−900 MeV, preserves its
1017
A/m^{2} brightness. This permits the tuning of mean energy of the TS γray signal between 4 and 16 MeV, preserving
1.5×106
photons in a microsteradian observation solid angle. Notably, the low energy in the stacked driver (1.4 J) and the ILP (25 mJ) permits maintaining a halfkHz repetition rate while staying below kW average power, a hard yet practical task [77]. A longer ILP would help increase the photon yield by another order of magnitude, without jeopardizing the repetition rate. Overall, this brings an expectation of greater than
109
ph/s yield, which is not as high as
1013
ph/s permitted by large linacs [57], yet sufficient to identify considerable masses of enriched uranium within minutes [61]. From the viewpoint of laboratory practice, computerized manipulations of the phase and shape of the subJoule stack components, using adaptive optics and genetic algorithms [82, 83], should aid greatly in practical realization of the system. This optimization approach is especially effective at a kHzscale repetition rate and low pulse energy.
The structure of the chapter is as follows: Section 2 describes the computational approach and defines parameters of the case studies. These parameters are representative of LPA experiments carried out in numerous laboratories worldwide. The reported case studies may thus serve as a reference for practical realization of the scheme in an existing experimental setting. In Section 3, we demonstrate the efficiency of using the stacks for generation of lowbackground electron beams, doubling their energy in comparison with the predictions of accepted scaling laws. Section 4 explores alloptical control over parameters of QME ebunches, through variation of the time delay between the stack components. A 60% increase in electron energy and a factor 3.5 increase in brightness are demonstrated using the same target and the same energy in the stack. This permits the tuning of the TS γray pulse parameters (mean energy, photon yield, and power) in the broad range of interest for nuclear photonics applications [49]. Section 5 summarizes the results and points out directions for future work.
2. Interaction regimes and simulation methods
Reduced and full 3D particleincell (PIC) simulations shed light on the physical processes essential for ebeam shaping. Quasistatic simulations using the relativistic, cylindrically symmetric, optical cycleaveraged code WAKE [13, 28] associate the massive dark current with the transformation of the optical driver into an optical shock and help develop a strategy for dark current reduction. WAKE computes the complex envelope of the laser vector potential using an extended paraxial solver. The solver preserves GVD in the presence of large frequency shifts and accurately calculates radiation absorption due to wake excitation [32, 84]. The grid
∆ξ≈∆r/3≈λ0/13≈63
nm and time step
∆t≈1.325/ω0
, with 30 macroparticles per radial cell, are sufficient to capture all physics relevant to pulse propagation and evolution of the bubble. Here,
ξ=z−ct
and
r2=x2+y2
. WAKE includes 3D test particle tracking in the full (unaveraged) electromagnetic fields. Testparticle simulations allow the study of the physical process of selfinjection (bubble and driver evolution) in the absence of effects due to beam loading [21, 22, 23]. Simulations using the relativistic, fully explicit, quasicylindrical code CALDERCirc [85] explore manipulations of ebeam phase space, leading to production of clean and tunable beams. The code uses a numerical Cherenkovfree electromagnetic solver [86] and thirdorder splines for the macroparticles. These features, in combination with a fine grid
∆z=0.125c/ω0≈16nm≈∆r/16
, small time step
∆t=0.1244/ω0
, and 45 macroparticles per cell, maintain low sampling noise and negligible numerical dispersion and avoid numerical emittance growth.
We demonstrate the limits of alloptical control fixing the total laser energy at 1.4 J. This energy may be concentrated in a single, transformlimited, linearly polarized Gaussian pulse with a carrier wavelength
λ0=0.805
μm and full width at halfmaximum in intensity
τL=20
fs (the reference case). The plasma begins at
z=0
with a 0.5 mm linear ramp, followed by a uniform section of density
n0=6.5×1018
cm^{−3}. The pulse, propagating toward positive
z
, is focused at the plasma border into a spot
r0=13.6
μm. The electric field in the focal plane is
Here,
ex
is the unit polarization vector, and the normalization factor
meω0c/e=4
TV/m. The single 70 TW pulse (
ℇ0=3.27
) has
τL=2rm/3c
and thus depletes at the point of electron dephasing, promising to maximize acceleration efficiency. Contrary to expectations, this strategy leads to copious dark current and overall low beam quality. Thus we seek to compensate for the red shift imparted by the wake. This task requires a very broadband negative frequency chirp, that is, blue shifting the leading edge by an amount comparable to the carrier frequency [23, 24, 25]. As the required broadbandwidth, kW averagepower laser amplifiers are not going to be available soon [37], we propose to synthesize a largebandwidth, negative piecewise chirp, by optically mixing transform limited, narrowbandwidth blocks of the same or different energy [24]. We demonstrate the emerging opportunities by splitting the 1.4 J energy evenly between two linearly (orthogonally) polarized, transformlimited 20 fs pulses, one of which, as shown in
Figure 1(b.1) and
(c.1), is significantly blueshifted and may be advanced in time. Electric field of this incoherent stack is
Estackz=0=E0+Ehead
, with
Figure 1.
Snapshots of electric field on axis (in units of
meω0c/e=4
TV/m) show that the frequency shift between stack components makes the stack resilient to selfcompression, while the nonzero time delay further increases this resilience. The pulses propagate to the right;
z=ct
is the centroid of the carrierfrequency component (
E0
) in vacuum. Red (thick dark gray):
E0
. Blue (thin gray):
Ehead
(in the simulation,
E0⊥Ehead
). Dashed curve:
E⊥2=E02+Ehead2
, where
⋯
denotes averaging over an optical cycle. Panels (a) correspond to the reference case, (b) to case A (
T=0,Ω=1.5
), and (c) to case B (
T=15fs,Ω=1.5
). Top row shows the fields at the plasma entrance (
z=0
). Bottom row corresponds to (a.2)
z=1.6
mm (point of full compression in the reference case), (b.2) and (c.2)
z=2.15
mm (dephasing in case A).
Here,
ℇhead=ℇ0=2.31
,
ey
is the unit polarization vector, the delay
T
is positive, and
Ω=ωhead/ω0>1
. Changing the frequency ratio and the delay permits accessing a broad range of ebeam parameters. We demonstrate the limits of this range by setting
Ω=1.5
. (Ref. [24] reports on other options, with equally promising results.) We consider the case of full overlap,
T=0
(case A), and then introduce the time delay that maximizes electron energy,
T=3τL/4=15
fs (case B).
Figure 1
shows that both stacks shrink slowly compared to the reference case, while advancing
Ehead
in time further increases rigidity of the stack.
To simulate the TS [51], we extract
Nb
macroparticles from the first and second buckets of the wake and use them to sample the sixdimensional (6D) phase space of the ebeam. We then propagate a corresponding distribution of noninteracting electrons in free space by solving the relativistic equations of motion. In the absence of a laser field, the trajectories are ballistic. The ebeam collides headon with the linearly polarized (in the
x
direction) interaction laser pulse, which we specify analytically in the paraxial approximation. The ILP has a 0.8 μm carrier wavelength (photon energy
Eint=1.55
eV), 250 fs duration corresponding to 0.3% FWHM bandwidth in spectral intensity, and a 16.8 μm waist size (corresponding to a Rayleigh length of 1.1 mm). Timing between the ebeam and the ILP is such that the centroid of the beam and the peak of the ILP intensity arrive at the ILP focal plane simultaneously. Since in all regimes under consideration the ebeams are relativistic and lowdensity,
neγe−3≪1016
cm^{−3}, space charge forces are negligible [46, 47]. As the energy radiated by an electron passing through the ILP is small if compared to the energy of the electron, the recoil and radiation damping are also negligible. The ILP is shorter than 7% of its Rayleigh length and the ebeam spot size is in the submicron range; hence, the interaction occurs in an almost planewave geometry. To avoid broadening the TS spectra [44, 50, 68], a linear interaction regime is chosen, with the ILP normalized vector potential
aint=0.1
(hence the ILP energy 25.5 mJ). Using the computed orbits of individual electrons and taking a weighted average over the ensemble yield the mean energy density radiated per unit frequency
ω
and solid angle
Ω
per electron [87]:
Here,
wi
is the macroparticle weight;
n
is the unit vector in the direction of observation; and
ri
and
βi=vi/c
are the radius vector and normalized velocity of the electron. A beam with a charge
Q
radiates the energy
d2Itot/dωdΩ=Q/ed2Ie/dωdΩ.
In all cases, we show the TS spectra for the emission in the direction of ebeam propagation (onaxis observation).
3. Stacking suppresses dark current
Propagation dynamics of a bicolor stack is entirely different from that of the quasimonochromatic reference pulse. To track changes in the pulse evolution brought about by the stacking, we use the laser vector potential,
a~=arzζe−iω0ζ
, where
ζ=ξ/c
is a retarded time, and
arzζ
is the complex envelope from WAKE simulations.
Figure 2(b)
and
2(c)
show, for the reference pulse and stack B, the radially integrated mean frequency and frequency variance [88],
Figure 2.
Negative chirp synthesized via pulse stacking mitigates frequency red shift and slows down selfcompression (WAKE simulations). The pulses propagate in a uniform plasma toward positive
z
. Evolution of the (a) pulse energy; (b) mean frequency; (c) frequency variance; and (d) pulse length computed from the
ζ
variance of the energy density on axis is shown in the reference case (black) and in case B (red/dark gray). The frequency is in units of
ω0
. The curves terminate as soon as electrons reach dephasing. The negative chirp reduces depletion, red shift, and spectral broadening, thus mitigating contraction of the pulse. Using the stacked driver with a blue component advanced in time (case B) strongly delays dephasing.
Here,
a~rzω=∫−∞+∞a~rzζe−iωζdζ
is the Fourier transform of the laser vector potential and
Az=∫0∞rdr∫0∞a~rzω2dω
.
Figure 2(d)
shows the mean pulse length computed from the
ζ
variance of intensity on axis,
Here,
τz=B−1∫−∞+∞ζa0zζ2dζ
is the position of the pulse centroid and
Bz=∫−∞+∞a0zζ2dζ
.
Figure 3
links the local frequency shift to the longitudinal distortion of the pulse. The frequency shift is extracted from the phase of the normalized vector potential,
a~0zζ=ae−iω0ζ+iϕ
, using two independent methods [84]. First, the Wigner transform
yields variation of the “photon density” in time at a given point
z
on axis. Second, we calculate the instantaneous frequency using the rate of the envelope phase change,
ωζ=ω0−dϕ/dt=ω0−∂ϕ/∂ζ
. Mean frequency, frequency variance, pulse duration, and photon density are experimentally measurable markers of the nonlinear optical processes. They help identify the regimes of pulse propagation and wakefield excitation [89].
Figure 3.
Deformations of the optical driver dictate ebeam quality. Increasing resilience of the driver to selfcompression, via negative piecewise chirp imparted by stacking, suppresses the lowenergy tail. Left column: Reference case. Right column: Case B (
T=15
fs,
Ω=1.5
). (a), (b) local frequency shift, spectra, and longitudinal distortion of the pulse from WAKE simulations. The pulse propagates to the right. (a) the pulse at the plasma entrance (
z=0
) and (b) at the point of electron dephasing ((b.1)
z=2.03
mm, (b.2)
z=3.07
mm). Grayscale is the absolute value of the Wigner transform (9) in arbitrary units; black curves are lineouts of the instantaneous frequency (in units of
ω0
) extracted from the complex pulse envelope. Top inset: normalized intensity on axis. Right inset: radially integrated spectral power,
Sω=∫0∞ω2arzω2rdr
, in arbitrary units. (c) Electron spectra at dephasing from CALDERCirc simulation. Black curves in (c.1) and (c.2) show the reference case spectra (
z=2.03
mm). Red (dark gray) in (c.2): Case B (
z=3.07
mm). Negative chirp of stack B doubles the energy of the QME signal, while suppressing the flux in the tail by more than an order of magnitude.
Figure 3(a.1) and 3(b.1) reveal destruction of the reference pulse at the point of electron dephasing. As the pulse plows through the plasma, it maintains the comoving negative gradient in the nonlinear index of refraction, located at its leading edge [16, 23]. A large local frequency red shift gradually accumulates along the index gradient, eventually exceeding
ω0/2
.
Figure 2(a)–(c) show that, in the process, the energy and mean frequency of the reference pulse drop by 60%, while the mean bandwidth increases 10fold. The negative GVD of the plasma slows down the redshifted radiation components, etching away the pulse leading edge. One can clearly see, in the top inset in
Figure 3(b.1), the resulting cyclelength optical shock of relativistic intensity. What is more, the photons making up the optical shock keep sliding into the bubble, filling it with the midIR radiation at
ω<ω0/4
. Mixing radiation of different frequencies and uncorrelated phases leads to sharp variations in the envelope phase, making the local frequency poorly defined, causing oscillations of the envelope in the tail area. More importantly, the reference pulse fully contracts long before electron dephasing (from 20 to 10 fs, according to
Figure 2(d)
) and then almost explosively elongates as newly generated midIR radiation slides into the bubble. Continuous injection ensues, building up the massive energy tail shown in
Figure 3(c.1), containing threequarters of the charge accelerated above 50 MeV. CALDERCirc simulation corroborates and adds more details to this unfavorable scenario.
Figure 4
links selfinjection to the bubble evolution. The bubble size (shown in (a)) is defined as the length of the accelerating phase on axis (i.e. the length of the region inside the bubble where the longitudinal electric field is negative). Panel (b) tracks accumulation of the charge in the first two buckets, counting only electrons with
E>50
MeV. The collection phase space (longitudinal momenta of electrons at dephasing vs. their initial positions, (c)) and collection volume (initial positions of electrons with
E>50
MeV at dephasing, (e)) parameterize the energy gain of electrons with their initial coordinates. As the reference pulse adjusts for selfguiding, its spot size oscillates at least once.
Figure 4(a)
shows that it is during this early stage (between
z≈0.55
and 1.3 mm) that the bubble expands, injecting electrons, and then stabilizes, forming the QME bunch. Selfcompression of the pulse starts early. As the optical shock builds up, the bubble starts to expand. Explosive expansion after
z≈1.3
mm, with an almost 65% increase in size by the dephasing point (
z=2.03
mm), adds an extra 1.25 nC to the energy tail (in effect, multiplying the tail charge by a factor of 6). Beam loading saturates injection near dephasing, eventually destroying the bubble.
Figure 4.
Stacking reduces expansion of the bubble, suppressing the dark current, avoiding the buildup of a lowenergy tail in electron spectra (CALDERCirc simulations). Black: the reference case. Red/dark gray: case B. (a) Length of the accelerating phase on axis vs. propagation length. (b) Charge accelerated in the first two buckets. (c) Longitudinal collection phase space of electrons from the first two buckets. (d) Energy spectra. These are identical to those in
Figure 3(c.2), yet shown on a logarithmic scale, to evaluate suppression of the tail. (e) Collection volume. Data in plots (c)–(e) correspond to electron dephasing,
z=2.03
mm in the reference case and 3.07 mm in case B. The piecewise negative chirp of the stacked driver suppresses expansion of the bubble, reducing the flux in the energy tail by more than an order of magnitude, while doubling the energy of QME component.
Figure 3(a.2)–
(c.2) shows that the piecewise negative chirp turns the tide, enabling acceleration through dephasing without sacrificing ebeam quality. As a collateral benefit, electron energy doubles against the scalingprescribed limit of the reference case.
Figure 4(c)
shows that, in case B, the injection starts later than in the reference case, while the bucket contracts rather than stabilizes around
z≈1.3
mm, expelling onethird of the earlier injected charge. This reduces the bunch charge by a factor 8 in comparison with the reference case. Yet, from the data in
Table 1
, this reduction comes from clipping the bunch, from 5.5 fs to less than a femtosecond, with the average current preserved (
≈88
kA). As soon as the QME bunch forms, the resilience of the stack to selfcompression keeps it almost backgroundfree (cf.
Figures 3(c.2) and
4(d)
). In summary, stacking changes the system dynamics as follows. First, per
Figure 2(a)
, the pulse energy loss is merely onequarter, in stark contrast with 60% of the reference case. Second, as is seen in
Figure 2(d)
, the stack reaches full compression at the point of electron dephasing rather than halfway through.
Figure 3(b.2) shows that the energy of the stacked driver finally concentrates in a spike 2.5 optical cycles long, with the instantaneous frequency almost uniform along the pulse body. There is no sign of photon phase space rotation, with a midIR tail protruding into the bubble, nor there is a signature of a subcycle rising edge (the key feature of the reference scenario). Hence, the electron density pileup inside the compressed stack is minimal (cf. Figure 6 of Ref. [24]). Thereby, the resulting 7fold reduction in the bubble expansion rate, evaluated from
Figure 4(a)
, reduces the average flux in the tail by a factor of 16 and the charge by a factor of 6. The QME peak dominates the electron spectrum, having the mean energy twice as high, and 5D brightness a factor of 3.3 higher than its reference counterpart. Notably, the boost in energy has little to do with beam loading. As mentioned earlier, the current density in the QME bunches (the key factor defining the effect [26]) is almost the same in both cases. As expected, the WAKE test particle simulations show that the beam loading reduces electron energy by 25% in the reference case and merely by a few percent in case B. Hence, threequarters of the observed energy boost are due to the favorable changes in the driver dynamics and quasistatic plasma response brought about by photon engineering.
Parameter (unit) 
z
(mm) 
Q (pC) 
E
(MeV) 
σE
(MeV) 
στ
(fs) 
σα
(mrad) 
ε⊥n
(mm mrad) 
I
(kA) 
Bn
(A m^{−2}) 
W (mJ) 

QME bunches 
Reference  2.027  493.5  426.5  25.7  5.5  2.93  0.69  89.7  0.38×10^{17}
 210.4 
Case A  2.148  288.8  524.8  26.3  3.8  2.75  0.64  76.2  0.38×10^{17}
 151.5 
Case B  1.473  73.8  442.9  31.8  0.85  2.16  0.3994  87.3  1.1×10^{17}
 32.7 
Case B, dephasing  3.067  73.8  881.9  28.6  0.85  1.35  0.3994  87.3  1.1×10^{17}
 65.1 
Energy tails (
E>50
MeV) 
Reference  2.017  1454  212.9  67.0  11.0  9.0  —  132  —  309.5 
Case A  2.1425  363.3  253.2  113.4  5.5  6.52  —  66.1   91.2 
Case B  1.4735  27.55  114.7  43.25  0.33  4.15  —  82.7  —  3.2 
Case B, dephasing  3.062  329.8  298.4  174.6  8.84  5.16  —  37.3  —  98.4 
Table 1.
Electron beam statistics.
Collection volume presented in
Figure 4(e)
indicates that only electrons with initial radial positions such as those that enter the bubble sheath are trapped and accelerated. There is no sign of transient injection from the nearaxis region. The injection candidates fill a thin cylindrical shell with a radius slightly smaller than the bubble radius, accurately reproducing evolution of the pulse spot size in the crosssection at the highest intensity, which agrees with the matching condition (1). Indeed, for the 70 TW, 20 fs reference pulse, the power ratio
P/Pcr=16.25
yields the matched spot
rm≈9.4
μm. The collection radius in
Figure 4(e)
varies by
±1
μm from this value through 70% of the acceleration distance. The dephasing length, calculated with this
rm
, is
Ld≈1.65
mm, which is within 10% of the value estimated from
Figure 4(e)
(black markers). The estimated energy gain (2) is 430 MeV, nearly the same as the simulated
E≈426.5
MeV. Thus, the QME bunch of the reference scenario complies with the scaling predictions exceptionally well. Yet, accumulation of the lowenergy tail ruins the beam by the end of acceleration.
In case B, slow selfcompression of the rigid head delays dephasing. Applying the scaling formulae to the 35 TW, 20 fs head with the wavelength
λhead=2/3λ0≈0.533
μm (so that
P/Pcr≈3.6
), we find
rm≈7.3
μm. Again, this value of
rm
deviates from the collection radius shown in
Figure 4(e)
(red markers) by less than
±1
μm throughout the entire interaction. The estimated dephasing length,
Ld≈2.95
mm, is 15% longer than that obtained in the CALDERCirc simulation. At the same time, the estimate of the energy gain at dephasing (using a generic formula (5) of Ref. [18]) is only 630 MeV, which is 30% lower than the gain obtained in the CALDERCirc simulation with a bicolor stack. Even though, according to the scaling, using the head alone should boost electron energy effectively, the presence of the unshifted tail is important. As the rigid head of the stack plows through the plasma, driving the wake, the flapping of the slightly mismatched tail inside a soft channel (electron density bubble) controls the bubble radius, thus determining kinetics of selfinjection [25]. The presence of the tail is thus essential to maintain sufficiently high charge (or modulate the beam current [25]). Since the tail rides inside the bubble devoid of electrons, it remains uncompressed. As the head starts experiencing red shift, the GVDdelayed redshifted radiation superimposes onto the smooth profile of the tail. The combination of the two does not permit formation of a subcycle rising edge (compare
Figure 3(b.1) and 3(b.2)), which avoids uncontrollable expansion of the bubble.
Table 1
quantitatively assesses improvements in ebeam quality, showing parameters of QME bunches and energy tails at dephasing. Statistics of case B are complemented with the data taken at the point where the bunch energy matches the energy gain in the reference case (
E≈430
MeV,
z≈1.47
mm).
Table 1
presents the metrics that are essential to evaluate the 5D brightness,
Bn=2Iπε⊥n−2
, the quantity defining a capability of the beam to drive a TSbased γray source [76]. Here,
I=Q/στ
is the mean current;
Q
is the bunch charge;
στ
is the rootmeansquare bunch length; and
ε⊥n=2−1/2εxn2+εyn21/2
, where
εin=mec−1pi2−pi2ri2−ri2−piri−piri21/2
is the rootmeansquare normalized transverse emittance. Statistics for case B show that the QME bunch progresses through dephasing with the normalized transverse emittance conserved, as should be the case for the adiabatically slowly varying structure. Simulating ebeam dynamics, while conserving the emittance better than in the fourth digit, became possible due to elimination of the numerical Cherenkov radiation [86] in CALDERCirc. Any degradation of the Thomson γray signal observed in the simulations must be thus attributed to the physical causes rather than to numerical artifacts. The QME bunch B, apart from 3.2% energy spread, subfs duration, and 400 nm emittance (about half of that of the reference case), has the 5D brightness
1.1×1017
A/m^{2}, preserved throughout acceleration. This is most encouraging for using the beam as a driver of a highflux Thomson source [76].
4. Stackdriven electron beams generate highflux, femtosecond γray pulses via Thomson scattering
The capability of a stackdriven LPA to suppress the lowenergy background and to increase electron energy gain, while preserving 100pC scale charge and 100kA average current, boosting the ebeam brightness beyond
1017
A/m^{2}, is an asset for the design of radiation sources.
4.1. Improving performance of the γray source: suppressing emission of lowenergy photons and boosting the energy of quasimonochromatic signal
A broadband ebeam accelerated with a transformlimited pulse is poorly suited to produce quasimonochromatic γray pulse via the TS mechanism.
Figure 5(a)
demonstrates the phase space of the ebeam at dephasing, with a QME component accompanied with a massive energy tail. We separate the macroparticles making up the QME ebunch and the tail, as shown in
Figure 5(a)
, and carry out two sets of TS simulations with these initial conditions. The partial photon spectra displayed in
Figure 5(c)
reveal modest energy of γphotons emitted by the QME bunch in the direction of its propagation,
Eγ≈3.85
MeV. This is too low to meet the needs of nondestructive inspection systems for special nuclear materials; these require photon energy tunable in the range 5−15 MeV [49].
Figure 5(c)
reveals a rather high ratio of noise to the quasimonochromatic signal, 1:2, on average. Conversely, the virtual lack of the lowenergy background in case B favorably reflects on the TS signal. The latter, from
Figure 5(d)
, has an average noisetosignal ratio twice as low, 1:4, and nearly a factor 4 higher mean photon energy,
Eγ≈16
MeV. Yet, the signal bandwidth is rather high, 15.7%.
Figure 5.
The stackdriven LPA delivers a lowbackground QME electron bunch. Thomson scattering from this bunch produces a quasimonochromatic, subfs γray pulse. Electron beams are extracted from CALDERCirc simulations at dephasing. (a), (c) reference case. (b), (d) case B. (a), (b) longitudinal phase space of the bunch; inset: energy spectrum (in units MeV^{−1}). (c), (d) γray flux in the direction of ebeam propagation (in units
1012
MeV^{−1} sr^{−1}). QME components of ebeams and corresponding quasimonochromatic components of γray pulses are depicted in red (dark gray).
From a singleparticle theory [50], both variation in energy of an electron and a misalignment of its trajectory with the propagation axis of the ILP tend to reduce the photon energy, thus contributing to the photon energy spread. The QME bunch of case B, apart from having a 3.2% energy spread, has a rather high rootmeansquare divergence,
σα=1.35
mrad, or
2.33γe−1
. Here,
γe
is the mean Lorentz factor of the bunch, and
σα=2−1/2σx2α+σy2α1/2
, where
σiα=pz−1pi2−pi21/2
.
TS simulations with the reduced phase space of ebeam help identify the primary contributor to the photon energy spread. First, we plot in
Figure 6
(black in both panels) the spectrum of γphotons emitted by the bunch with a complete 6D phase space (the region of phase space depicted with red/dark gray markers in
Figure 5(b)
). The signal with a 15.7% energy spread is centered at
Eγ=16
MeV. Then, transverse momenta of all macroparticles are set to zero, while their longitudinal momenta,
pz
, are unchanged. This preserves the energy spread while zeroing out the divergence. Lastly,
pz=pz=1725.8mec
is assigned to all electrons, while
px
and
py
are unchanged. This preserves mradscale divergence of the bunch, while nearly zeroing out the energy spread. In the zerodivergence case (spectrum depicted in red/dark gray in
Figure 6(a)
), the mean photon energy increases to 18 MeV, while the energy spread stays at 12.8%. In stark contrast, the case with a nearzero energy spread yields a TS signal with a subpercent energy spread, centered at
Eγ≈4γe2Eint≈18.4
MeV (red/dark gray in
Figure 6(b)
). Thus, the γray signal receives its large (more than 10%) bandwidth almost entirely from a fewpercent electron energy spread. Further steps in optimization of the LPA should aim to reduce the energy spread below 1%. One practical way to do it (at the expense of reduction in charge and, hence, brightness) is to select electrons from narrow energy intervals within the ebeam bandwidth by selective focusing with highly chromatic magnetic quadrupole lenses [90] before the collision of ebeam with the ILP. (At the same time, the magnetic quadrupole will disperse the residual energy tail.)
Figure 6.
Energy spread in the electron bunch determines the energy spread of emitted γphotons. Black: TS spectrum from the simulation using complete phase space of electron bunch (same as red (dark gray) in
Figure 5(c)
and
5(d)
). Red (dark gray): simulations with the reduced phase space of electrons, with (a) zero dispersion of transverse momentum (zero divergence) and (b) zero dispersion of longitudinal momentum (viz. almost vanishing energy spread).
The collimation of highenergy γphotons and the number of photons in the observation cone are important metrics for applications. To evaluate the reduction in photon energy and flux with an increase in the observation angle (viz. to estimate the effective apex angle of the photon emission cone), we select the macroparticles making up the QME ebunch and carry out the TS simulation with these initial conditions. We detect the photons scattered in and out of the ILS polarization plane, in the direction of ebeam propagation (
θ=0
) and at small angles measured from the direction of ebeam propagation, up to the rootmeansquare divergence angle,
θ=σα=2.33γe−1
. As the scattering angle increases to
θ=γe−1
, the mean photon energy drops by 25%, the energy spread staying at the 15–20% level. At the same time, the photon flux drops 10fold. To a good approximation, there are virtually no photons with the energies above 10 MeV outside the observation cone of apex angle
2θ=2γe−1
. Thus, to estimate the number of QME highenergy photons scattered in the direction of ebeam propagation, we choose conservatively the observation solid angle
∆Ωγ=π/2γe−2
, that is, the solid angle of the cone with an apex angle
2θ=2γe−1
(
≈2×0.69
(
2×0.41
) mrad for the beam A(B) at dephasing). We take the photon flux corresponding to the direct backscattering (
θ=0
), integrate it over the energy, and multiply the result by
∆Ωγ
.
Table 2
presents statistics of quasimonochromatic γray signals corresponding to the QME entries in
Table 1
. The entries corresponding to the beam B show the photon yield over
1.5×106
per pulse. This is comparable to the experimental yields with 100MeV scale ebeams,
3×105
to
107
[66, 67, 70], calculated for the entire forward hemisphere, and integrated over the entire broad bandwidth of the γray beam. Yet, our highestenergy photons reach 16 MeV while preserving a 16% energy spread and microsteradian collimation, which is strikingly better than 50–100% spread and millisteradian collimation reported for the subMeV photons [66, 67, 70].
Parameter 
Eγ
(MeV) 
σE
(MeV) 
∆Ωγ
(μsr) 
Nγ106

Wγ
(μJ) 

Reference  3.85  0.72  2.25  8.95  5.5 
Case A  5.67  0.97  1.49  5.08  4.6 
Case B  4.36  0.93  2.09  1.52  1.1 
Case B, dephasing  16.0  2.51  0.53  1.58  4.0 
Table 2.
Statistics of γrays emitted by the QME bunches with parameters from
Table 1.
4.2. Tuning the energy and flux of the quasimonochromatic γray signal
The structure of ebeam phase space is sensitive to variations in the drive pulse evolution brought about by changes in the initial conditions. This responsiveness improves parameters of the beam to make it suitable to drive a quasimonochromatic radiation source based on Thomson scattering, a process demanding exceptional beam quality.
Figure 7
displays the capability of the alloptical Thomson source to produce γray pulses with different characteristics (such as a mean energy and flux) while maintaining high yield and low background. This flexibility is demonstrated without changing the energy of the stack or the frequency ratio between its components. This is an asset to applications, which often need to adjust to new conditions in a timely fashion, avoiding a major upgrade of the laboratory.
Figure 7.
The stack with fully overlapped components (case A, panels (a), (c)) preserves high electron and γray flux similar to that of the reference case, while suppressing the background. The stack with a blue component advanced in time (case B, panels (b), (d)) permits doubling electron energy against the reference case, increasing the photon energy by a factor 3 in comparison to case A, at the expense of reduction in flux. (a), (b) electron energy spectra. (c), (d) spectra of γphotons emitted in the direction of ebeam propagation. Gray: Spectra of the reference case. Black: electron and photon spectra corresponding to ebeams extracted at dephasing (
z≈2.15
mm, case A;
z≈3.07
mm, case B). Dashed curves in (c) and (d): spectra of photons emitted by the electron energy tails. Red (dark gray): Ebeam of case B, extracted prior to dephasing,
z≈1.47
mm.
Changing the ebeam energy, while keeping the brightness fixed, is one option offered by case B. Standard targets, such as printed gas cells of variable length [81], permit the necessary reductions in the plasma length.
Figure 7(b)
shows progress of the ebeam through dephasing, from the energy matching the maximal gain in the reference case (
≈430
MeV) to 882 MeV. Notably, the QME bunch reaches dephasing, while maintaining
Bn=1.1×1017
A/m^{2}. This conserves the number of photons emitted into the μsrscale detection angle,
Nγ≈1.55×106
. Thus, a quasimonochromatic (15.7–21.3% energy variance) 0.85 fs γray pulse with the mean energy tunable between 4 and 16 MeV (1.3–4.7 GW average power) can be produced.
Reducing the time delay between stack components increases photon yield at relatively low energies (
Eγ≤10
MeV), the ultimate example of which is case A (full overlap). The physical difference between cases A and B is remarkable. Stack A is an incoherent mix in the fashion of Ref. [80] rather than a pulse with a negative piecewise chirp. As the stack A components plow through the plasma, they both ride on the downslope of the nonlinear index, such as depicted, e.g. in
Figure 7
of Ref. [16] or
Figure 5
of Ref. [22]. From
Figure 1(b.2), the frequency unshifted component,
E0
, redshifts and compresses to nearly a single cycle, in the same fashion as the reference pulse does (cf.
Figure 1(a.2)). At the same time, the blueshifted
Ehead
remains virtually intact.
Figure 1(b.2) also shows that the stack components stay together, accumulating merely a twocycle delay due to the difference in their group velocities. The stack thus does not break up longitudinally. The superposition of compressed
E0
and almost intact
Ehead
does not allow for the formation of a subcycle rising edge (optical shock) such as develops in the reference case (
Figure 1(a.2)). The undelayed
Ehead
is thus akin to a rigid exoskeleton protecting the vulnerable
E0
.
Deformation of a stack with fully overlapped components, the process that defines the dephasing length [18], is dominated by rapid selfcompression of the least resilient stack component,
E0
. As
E0
selfcompresses (and hence, the stack selfcompresses) at the same rate as the drive pulse in the reference scenario, the dephasing length remains almost unchanged, and the boost in electron energy (by 23%) is unremarkable.
Table 1
shows that the QME bunch of case A is not very different from its reference counterpart (note that their 5D brightness is the same.) Conversely, the electron energy tail in case A is suppressed, with a factor 4 reduction in charge and 7 in average flux. This drastically improves the noisetosignal ratio of emitted γrays, from roughly 1:2 in the reference case to better than 1:5 in case A, as may be evaluated from photon spectra in
Figure 7(c)
.
Table 2
shows that the γray pulse, in case A, contains about
5×106
photons, a factor 3.3 higher yield than in case B. Further, these photons receive 50% boost in energy with respect to the reference case
^{2}
. Changing the time delay between the stack components is thus a proper way to control the TS photon yield and energy.
4.3. Final notes on alloptical control of quasimonochromatic Thomson sources
The results presented above offer an opportunity for tuning the γray energy and flux in a broad range, without changing the total (Jouleperpulse) laser energy in the LPA.
Particularly, the interval of photon energy variation, 4–16 MeV, with the yield
1.5−5×106
photons per pulse, meets the needs of nuclear photonics applications [49]. Raising the photon yield per second is technologically possible. Using a longer (up to 2.5 ps) ILP with the same spot size and amplitude (
aint=0.1
) may increase the yield by up to an order of magnitude. This long ILP is still shorter than its halfRayleigh length. This is sufficient to preserve the nearly planewave character of the interaction with the ebunch. As the energy (0.25 J) of the long ILP is still below the stack energy (1.4 J), their repetition rates may be matched. Then, using a kWscale laser amplifier [77] permits the production of Joule scale pulses (LPA stack and ILP) at a few hundred Hz, increasing the photon yield beyond
109
ph/s. Apart from keeping the photon yield at a competitive level, this repetition rate permits real time optimization of the experiment [82, 83], not possible with one shot per hour PW facilities [73].
The estimated
109
ph/s yield is sufficiently high to make the tunable TS source interesting for the design of a nondestructive inspection system for special nuclear materials. Simulations, based on data from recent detection experiment [61], indicate that the TS γray flux of
106
ph/s, with a 5% signal bandwidth and a 10 Hz repetition rate, is sufficient to identify a nuclear resonance fluorescence peak from a 1 kg of highly enriched uranium within 10 minutes. The higher flux accessible with our source would compensate for its still significant (up to 20%) bandwidth. Alternatively, frequency chirping of the ILP may reduce the photon energy spread [51]. Given the unconventional Ushape of electron momentum chirp in the QME ebunch (cf.
Figure 5(a)
and
5(b)
), this topic deserves special consideration but is beyond the scope of this chapter.
Lastly, ebeam optimization studies, using stacks with different frequency ratios [24], show that, as long as the time delay and energy partition between the stack components are fixed, and
Ω≥1.25
, the electron energy gain is quite insensitive to the frequency ratio. Even though the particle flux
dN/dE
in the QME bunch drops as
Ω→2
, reducing the quasimonochromatic γray yield, the TS signal still has a quality far exceeding that accessible in the reference scenario. This observation permits a considerable technological flexibility in a practical realization of this concept. Frequency shifting on the modest scale (
Ω≤1.5
) can be accomplished with a Raman cell, with subsequent conventional chirpedpulse amplification [91, 92, 93]. Alternatively, energyefficient methods of frequency doubling may be applied.
5. Summary and outlook
In a conventional LPA, a cavity of electron density maintained by the radiation pressure of a single narrowbandwidth laser pulse accelerates electrons selfinjected from the ambient plasma. Deformations of the bucket, which carry on in lockstep with the deformations of the optical driver, determine the structure of the ebeam phase space. Optimizing the nonlinear evolution of the drive pulse is an essential element of LPA design, offering new avenues to coherently control ebeam phase space on the femtosecond scale.
Compact sources of quasimonochromatic γphotons, based on the TS mechanism, are highly sensitive to the quality and phase space structure of the driving GeVscale ebeams. Reaching sufficient ebeam brightness and energy, while maintaining a modest facility footprint and high repetition rate, is a major challenge for a traditional LPA. The first roadblock is the limit on electron energy imposed by dephasing, with unavoidable beam contamination with a lowenergy background, while the second is the low repetition rate of petawattscale lasers (which limits the dosage, frustrating applications). Reducing the energy in the drive pulse to a subJoule level may alleviate the latter, yet aggravating the former. Our simulations show the way to resolve this conflict, by synthesizing the LPA drive pulse by incoherently stacking collinearly propagating 10TWscale pulses of different wavelengths. Stacking introduces a frequency bandwidth sufficient to compensate the red shift imparted by the wake excitation. Unlike a single, transformlimited pulse, the stack is nearly immune to degradation while driving the bubble in a dense plasma (
n0∼1019
cm^{−3
}). Advancing the blueshifted component of the stack in time emulates the negative frequency chirp [16, 23, 24, 25]. This delays dephasing of the electrons, doubling their energy compared to the scaling predictions, using no manipulations of a few mmlength gas target. Importantly, immunity of the stacked driver to selfcompression keeps the lowenergy electron flux so modest as to almost avoid contamination of TS γray pulse with lowenergy photons.
Simulation data presented here show that increasing the delay between the stack components, while keeping the same total laser energy and frequency ratio, permits increasing electron energy from 525 to 900 MeV, boosting the 5D brightness nearly threefold. This puts generation of 10 MeVscale, fewGW, quasimonochromatic, femtosecondlength γray pulses via Thomson scattering within reach of existing laser technology. Energy of these pulses, containing up to
5×106
photons into the microsteradian cone, may be tuned in the range 4–16 MeV and, possibly, beyond, while the lowenergy photon background remains insignificant. Increasing the LPA repetition rate to kHz level, at affordable average power, promises boosting photon yield beyond
109
ph/s, making tunable, alloptical TS γray sources interesting for applications [48, 49, 61]. As a further development, focusing the stack components differently [52], or propagating the stack in a channel [24, 25], enables generation of a train of GeVscale, ultrabright electron bunches with a femtosecond synchronization. These unconventional comblike ebeams may emit polychromatic trains of highflux γray pulses consisting of a few distinct energy bands, in the range 3–17 MeV [52]. The natural mutual synchronization of fslength ebunches and γray pulses may be an asset to nuclear pumpprobe experiments. With a γray beam spectrally resolved, each beamlet may give a “movie frame” on a femtosecond time scale to image ultrafast phenomena in a dense matter [74].