Modeling the effect of spectral phase shaping of ultra-short laser pulses for laser-driven ion acceleration on ensemble scales

In this work we performed 1D and 2D particle-in-cell modeling of ion acceleration by spectrally shaped laser pulses in a multi-fidelity framework that covered several orders of magnitude of parameter space. Spectral shaping by means of a dazzler allows us to modify the temporal shape of a short-pulse laser on femtosecond to picosecond time scales. An ensemble of several thousand 1D PIC simulations was used as support for several hundred 2D PIC simulations. The 1D dataset is used to inform a neural-network surrogate model which is then elevated to 2D fidelities via ad-hoc transfer learning, whereby the trends embedded in the 1D data are used to inform the 2D surrogate model at higher accuracies. More complex network architectures allow us to synthesize non-congruent datasets in ways otherwise not possible and may allow for the integration of more realistic data in future work. Pulse shaping suggests that we can not only achieve higher ion energies than otherwise accessible with just a Gaussian pulse but also that typical quantities of interest may be partially tuned independently of one another to a limited degree.