Berlin 2018 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 47: Poster Session II

MM 47.26: Poster

Mittwoch, 14. März 2018, 18:30–19:45, Poster C

Uncertainty quantification strategies for atomistic simulations with classical effective potentials — •Peter Brommer and Sarah Longbottom — University of Warwick, Coventry, UK

Molecular dynamics (MD) simulations with classical effective potentials account for a significant fraction of current HPC usage. In those simulations, the energy of the system (and thus the forces on the atoms) depends only on the positions of the atoms; all electronic degrees of freedoms are accounted for in the interaction potential or force field. In contrast to ab-initio methods, effective potentials require the user to make some assumptions and approximations. Unfortunately, the bias, approximations and uncertainties that enter the effective potential creation process at every step (choice of interaction terms, functional form and parameter values) are badly controlled, and in most cases the confidence of the final result of a MD simulation is unknown a priori. The selection of interaction terms and functional form are often motivated by intuition and prior experience of the researcher and then validated by comparing the results in a certain situation after the fact, which is then used to justify using a potential even when such a validation is no longer feasible. However, for simulations to be truly predictive, it is essential to quantify the uncertainty one incurs by using a certain representation of the interactions between atoms. Here, I present an overview of various strategies to combat this deficit of effective potentials and the computational impact of such approaches.