Contribute Media
A thank you to everyone who makes this possible: Read More

Efficient Parametrizations of Potential Energy Surfaces

Description

Finite temperature atomistic simulations require high accuracy dynamical models with much better performance than direct ab initio calculations. Two of those are high-order polynomial expansions and neural network force fields. We discuss high-performance implementations of both using JAX and point out the intrinsic limitations of the former. We illustrate this with a Lennard-Jones solid and a successful application to the phase diagram of hafnia, a complex and interesting oxide.

Details

Improve this page