Beyond BGK: Prospects for Learning Surrogate Collision Operators

Kinetic gas dynamics in rarefied regimes exhibit complex behaviors arising from collisional processes,

making accurate descriptions both challenging and computationally expensive.

In this presentation, we explore the potential of employing physics-informed machine learning techniques

to derive novel collision operators for the Lattice Boltzmann Method (LBM).

We present preliminary results in which a neural network is successfully trained

to augment the single relaxation time BGK operator, leveraging on data from

particle-based methods such as molecular dynamics (MD) and direct-simulation Monte Carlo (DSMC).

This work serves as a first step towards the definition of a computationally efficient

framework for studying nonequilibrium gas flows, embedding higher-order effects

within a lower-order representation.