We develop Machine Learning models of atomic forces for arbitrary chemical systems. The models are trained on electronic structure results obtained for representative and relevant samples of chemical space. The underlying goal is to increasingly replace the ab initio calculation of forces by successively trained machine learning models.
We have developed such models for carbon and hydrogen atoms in organic materials. Other materials will follow.
"Machine Learning for Quantum Mechanical Properties of Atoms in Molecules", M. Rupp, R. Ramakrishnan, O. A. von Lilienfeld,
J. Phys. Chem. Lett. 6 3309 (2015). arxiv.org/abs/1505.00350
Contributor of the following OpenKIM content:
Developer of the following OpenKIM content:
Implementer of the following OpenKIM content: