Efficient Monte Carlo simulation of streamer discharges with deep-learning denoising models

Electric breakdown in non-conducting gases is a complex process that in its first stages is characterized by filamentary discharges called streamers. Streamer dynamics are inherently nonlinear and span broad temporal and spatial scales, making numerical simulation challenging. Although Monte Carlo methods are intuitive and they model the full electron energy distribution without a priori prescriptions, they suffer from artificial sampling noise which, combined with the non-linearity of streamers, distorts their evolution. Here we investigate the use of deep-learning techniques to mitigate the noise introduced by Monte Carlo sampling. We observe that traditional techniques for noise reduction in images are not satisfactory because they do not impose strict conservation of electric charge. Then we present a charge-conserving denoising filter to improve the efficiency of Monte Carlo simulations of streamers.

• View at Publisher

Free Full Text ( Final Version , 2mb )