Talk

Speaker

Karsten Wedel Jacobsen

DTU

Title

Accelerating Density Functional Theory calculations with machine learning.

Abstract

Density Functional Theory calculations are increasingly used to identify new materials with optimal properties for a given application. Such material screening studies may involve DFT calculations of thousands of materials and it is therefore a key challenge to speed up the calculations and to improve the search strategy. In the talk I shall describe some recent attempts to use Gaussian processes to accelerate local structure optimization and to apply message passing neural networks to replace DFT calculations of the stability of materials.