Neural-network based collision operators for the Boltzmann equation

Sean T. Miller, Nathan V. Roberts, Stephen D. Bond, Eric C. Cyr

Abstract: Highlights•We present a neural-network-based model of a collision operator.•Our model is orders of magnitude more efficient to evaluate than direct simulation.•We demonstrate good agreement with the direct simulation model used for training.•The direct simulation model used for training is a stochastic (DSMC) model.•Our model has embedded support for mass, momentum, and energy conservation.