Transfer learning and multi-fidelity modeling of laser-driven ion acceleration

Modeling of intense, laser-driven ion acceleration requires expensive particle-in-cell (PIC) simulations that may struggle to capture all the multi-scale, multi-dimensional physics involved at reasonable costs. Explored here is an approach to ameliorate this deficiency using a multi-fidelity model that can incorporate physical trends and phenomena at different levels. As the base framework for this study, an ensemble of approximately 10,000 1D PIC simulations was generated to buttress separate ensembles of hundreds of higher fidelity 1D and 2D simulations. Using transfer learning with deep neural networks, one can reproduce the results of more complex physics at a much smaller cost. The networks trained in this fashion can in turn act as surrogate models for the simulations themselves, allowing for quick and efficient exploration of the parameter space of interest. Standard figures-of-merit were used such as the hot electron temperature, peak ion energy, conversion efficiency, etc. These surrogate models are also useful for incorporating more complex particle acceleration schemes, such as laser pulse shaping where the simulation input parameter space is greatly expanded and standard parameterization of laser pulses (pulse length, intensity, etc.) are no longer descriptive. We can rapidly identify and explore under what conditions dimensionality becomes an important effect and search for optima in feature space. A description of the ensemble simulation and machine learning methodology will be presented along with multi-dimensional parameter space maps and optimizations for short-pulse, laser-driven particle sources found through this work.