Bayesian Optimization of Laser-Wakefield Acceleration via Spectral Pulse Shaping

In this paper, we investigate the effect of spectral pulse shaping of the laser driver on the performance of channel-guided, laser-plasma accelerators. The study was carried out with the assistance of Bayesian optimization using particle-in-cell simulations. We used a realistic plasma profile based on a novel optical-field-ionized channel technique with ionization injection and low on-axis plasma densities to maximize the energy gain of the electron bunch trailing the laser. Spectral shaping allows us to modify the temporal profile of the laser driver while keeping the laser energy constant, affecting the acceleration and injection processes. Given the complexity and breadth of the parameter space in question, we used numerical optimization to identify high performers. In particular, we found laser profiles with additional spectral content that, when used with optimal plasma channel parameters, result in charge content an order of magnitude higher than the baseline Gaussian case while also increasing the mean energy of the electron bunch.

💡 Research Summary

This paper investigates how spectral pulse shaping of the driving laser can be used to improve the performance of channel‑guided laser‑wakefield accelerators (LWFA). Because the interaction depends on many coupled, nonlinear parameters—laser temporal profile, spectral phase, plasma‑channel density, channel taper, and injection dynamics—the authors employ Bayesian optimization (BO) with Gaussian‑process (GP) surrogate models to efficiently explore the high‑dimensional design space while keeping the number of expensive particle‑in‑cell (PIC) simulations manageable.

The authors first describe the physics of spectral shaping. By adjusting the spectral phase φ(ω) up to fourth order (group‑delay dispersion, third‑order dispersion, and fourth‑order dispersion) one can generate a wide variety of temporal pulse shapes—chirped, asymmetric, double‑peaked, or with extended pedestals—without changing the total pulse energy. Analytic considerations show that appropriate GDD can compensate plasma group‑velocity dispersion, allowing the pulse to self‑compress at a chosen propagation distance, thereby extending the effective acceleration length. Higher‑order terms further control the pulse front and pre‑pulse structure, which influences ionization injection and wake formation.

For the numerical study the authors use the WarpX PIC code in a boosted‑frame configuration to reduce computational cost. The laser is injected via the LASY package, which accepts arbitrary spectral amplitude and phase files, and the plasma channel is modeled after a hydro‑dynamic optical‑field‑ionized (HOFI) channel with low on‑axis density (∼10¹⁷ cm⁻³) and a tunable radial density profile. The simulation domain spans –256 µm ≤ z ≤ 0 and 0 ≤ r ≤ 128 µm with a longitudinal resolution of 0.025 µm, providing sufficient detail to resolve the laser wavelength and plasma wave.

The optimization variables include the second‑, third‑, and fourth‑order spectral phase coefficients, the channel radius, the linear density taper, and the overall channel length. The objective function combines two key performance metrics: (i) the total charge of the accelerated electron bunch and (ii) its mean energy, with a modest penalty on energy spread. The BO loop proceeds as follows: (1) an initial random set of ∼20 simulations is performed; (2) a GP model is trained on the observed objective values; (3) the Expected Improvement acquisition function is maximized to select the next point; (4) the new simulation is run and the GP is updated. This cycle is repeated for roughly 120–150 evaluations.

Results show that spectral shaping alone can increase the bunch charge by an order of magnitude compared with a baseline Gaussian pulse of identical energy. The optimal spectral phase adds modest high‑frequency content and a positive GDD that causes the pulse to compress after a few millimeters of propagation, aligning the peak intensity with the region of strongest wakefield. When the plasma‑channel parameters are co‑optimized, the mean electron energy rises by 20–30 % (e.g., from 1.2 GeV to 1.5 GeV) and the relative energy spread is reduced by a few percent. The best channel configuration features a modest linear density gradient (≈0.5 % mm⁻¹) and a radius of ~30 µm, which together preserve laser guiding and enhance the dephasing length.

The authors discuss limitations: the BO approach still requires a few hundred costly PIC runs, and experimental implementation of the exact spectral phase may be constrained by the bandwidth of available acousto‑optic programmable dispersive filters (AOPDFs) or Dazzler devices. Moreover, the study focuses on a single‑objective (charge and energy) formulation; extending the framework to multi‑objective optimization that explicitly includes beam emittance, stability, or laser‑energy loss would be valuable. Finally, the scalability of the identified optimal parameters to higher laser powers or different plasma densities remains an open question.

In conclusion, the paper demonstrates that Bayesian optimization, when coupled with high‑fidelity PIC simulations, can uncover non‑intuitive combinations of laser spectral phase and plasma‑channel design that dramatically improve LWFA performance. The work provides a clear pathway toward data‑driven accelerator design, suggesting that future compact, tunable particle sources could benefit from systematic, machine‑learning‑guided exploration of laser‑plasma parameter spaces.