A Meta-Analysis of Laser-Ion Acceleration Experiments

Machine learning can help uncover new patterns in large datasets. In this study we apply machine learning to laser-ion acceleration data gathered from decades of previous experimental work, since the low-repetition-rates of intense laser systems have limited the size of datasets. This growing dataset consists of hundreds of distinct data points from tens of existing experimental campaigns on a variety of laser systems. From these experiments, we extract their parameters including intensity, pulse duration, wavelength, target thickness, and maximum proton energy. Our meta-analysis of laser-proton acceleration evaluates how existing theoretical/empirical models perform over this large parameter space. Then we apply machine learning methods including neural networks to find patterns in the data and help identify factors that can optimize laser-ion acceleration in future experiments. We plan to release this dataset in an open format that allows corrections and contributions of new data to encourage future collaboration among the community.