Geometry optimisation based on machine learning

Home » Geometry optimisation based on machine learning

Geometry optimisation based on machine learning

Aug 31, 2021

Johannes Kästner’s group at the University of Stuttgart have extended DL-FIND to use Gaussian-process regression (GPR) to search for minima, transition states, and reaction paths.

Energies and gradients during the optimisation trajectory enter the GPR algorithms, which build up a surrogate surface that aids the optimisation. The resulting algorithm requires fewer steps than traditional approaches, especially for transition states and reaction paths. Moreover, Cartesian coordinates and, recently, also internal coordinates can be used.

The GPR-enabled version of DL-FIND is freely available.

Citation:

A. Denzel and J. Kästner, “Gaussian process regression for geometry optimization” J. Chem. Phys., 2018, 148, 094114.

A. Denzel and J. Kästner, “Gaussian Process Regression for Transition State Search” J. Chem. Theory Comput., 2018, 14, 5777-5786.

D. Born and J. Kästner, “Geometry Optimization in Internal Coordinates Based on Gaussian Process Regression: Comparison of Two Approaches” J. Chem. Theory Comput., 2021. DOI: 10.1021/acs.jctc.1c00517