Parameter space exploration of short-pulse laser-driven ion acceleration via ensemble simulations and neural networks

Application of deep learning to shaped, short-pulse laser-driven ion acceleration. Using a neural network as a universal approximator function, i.e., a surrogate model, we can map out large areas of parameter space. The neural network is informed by a large dataset of about 1,000, mid-fidelity particle-in-cell simulations modeling instances of Target-Normal Sheath Acceleration. The neural-network-based function allows us to rapidly explore regions of interest in search of optimal input parameters and features of interest.