Inhalt des Dokuments
Force fields for molecular dynamics simulations
Molecular dynamics (MD) simulations with classical force fields are the cornerstone of atomistic modeling in biology, chemistry, and material science. These classical potentials, however, are often lacking the quantum effects in molecules and materials. We design ML models that learn flexible force fields from high-level ab initio calculations, allowing to run accurate and fast MD simulations that properly capture quantum effects. In this way, new insights into the dynamical behavior of small- to medium-sized molecules can be gained at reasonable computational cost and time. For example, symmetric gradient domain machine learning (sGDML), a kernel-based ML method, faithfully reproduces global force fields at quantum-chemical CCSD(T) level of accuracy from a very limited amount of training data. It allows converged molecular dynamics simulations with fully quantized electrons and nuclei.
The sGDML (symmetric gradient domain machine learning) is a force field model able to faithfully reproduce detailed global potential energy surfaces (PES) for small- and medium-sized molecules from a limited number of user-provided reference calculations.
Python library available at www.sgdml.org