Hauptseminar Porous Media SS 2021/Machine learning in physics
- Machine learning for the construction of force fields
- David Zimmer
- Samuel Tovey
Description: In recent years machine learning methods have been employed to develop non-parametric models for the approximation of the potential energy surface. These models, trained on ab-initio simulation data, deliver the accuracy of these ab-initio methods at the speed and system size of classical approaches. Development of these models requires several important steps, all of which are under active study and for which many approaches exist. Such steps include selecting training data, representing this data, and the choosing machine learning model to use. This talk will address current approaches for developing these models as well as well as some of the results that have been found by using the machine learning methods. The presenter will build upon the foundation work of the previous talks on DFT methods and atomistic methods in order to demonstrate the position of the machine learning approaches as a bridge between them.
- Gaussian process regression and Neural network potentials
- Sampling configuration space for training data
- Representation of molecular environments
- Failures in ML potentials
- Study of chemical reactions on an atomistic scale
Albert P. Bartók, Gábor Csányi.
Gaussian Approximation Potentials: A brief tutorial introduction.
International Journal of Quantum Chemistry , 2015.
[PDF] (308 KB) [DOI]
Tsz Wai Ko, Jonas A. Finkler, Stefan Goedecker, Jörg Behler.
A fourth-generation high-dimensional neural network potential with accurate electrostatics including non-local charge transfer.
Nature Communications 12(1):398, 2021.
Marcel F. Langer, Alex Goeßmann, Matthias Rupp.
Representations of molecules and materials for interpolation of quantum-mechanical simulations via machine learning.
Stefan Chmiela, Huziel E. Sauceda, Klaus-Robert Müller, Alexandre Tkatchenko.
Towards exact molecular dynamics simulations with machine-learned force fields.
Nature Communications 9(1):3887, 2018.
Stefan Chmiela, Huziel E. Sauceda, Igor Poltavsky, Klaus-Robert Müller, Alexandre Tkatchenko.
sGDML: Constructing accurate and data efficient molecular force fields using machine learning.
Computer Physics Communications 240:38-45, 2019.
Machine learning for quantum mechanics in a nutshell.
International Journal of Quantum Chemistry 115(16):1058-1073, 2015.