Description
Neural-network techniques for fluid closures
In this talk, we will present neural-network methods to construct fluid closures for Euler type models that enable to extend their physical validity to weakly collisional regime. They consist in learning the heat flux and the pressure tensor with neural networks thanks to simulations of the underlying kinetic dynamics. A detailed numerical study of the stability of the obtained fluid solver is carried out.
This is a join work with L. Bois, E. Franck, V. Vigon.
