 |
Center for Nonlinear Studies |
In this talk, I will present our recent efforts on using machine learning (ML) methods to enable multi-scale dynamical modeling of functional electron materials, and in particular correlated electron systems. I first discuss the ML force field models for large-scale dynamical simulations on two canonical examples of correlated electron systems: the double-exchange and the Falicov-Kimball models. The central idea is to develop deep-learning neural-network models that can efficiently and accurately predict generalized forces required for dynamical evolutions based on local environment. The large-scale simulations enabled by the ML method also reveal new phase-ordering dynamics in these correlated electron systems that is beyond conventional empirical theories. We will also discuss preliminary results of similar ML models for the more difficult Hubbard-type models. In the second part, generalization of the ML framework to represent nonconservative forces of out-of-equilibrium systems is discussed. In particular, we present a novel ML structure for modeling spin transfer torques that play a crucial role in spintronics.
Host: Shizeng Lin, T-4