Towards exact molecular dynamics simulations with machine-learned force fields

Stefan Chmiela,Huziel E Sauceda,Alexandre Tkatchenko,Klaus-Robert Müller

doi:10.1038/s41467-018-06169-2

Abstract

Molecular dynamics (MD) simulations employing classical force fields constitute the cornerstone of contemporary atomistic modeling in chemistry, biology, and materials science. However, the predictive power of these simulations is only as good as the underlying interatomic potential. Classical potentials often fail to faithfully capture key quantum effects in molecules and materials. Here we enable the direct construction of flexible molecular force fields from high-level ab initio calculations by incorporating spatial and temporal physical symmetries into a gradient-domain machine learning (sGDML) model in an automatic data-driven way. The developed sGDML approach faithfully reproduces global force fields at quantum-chemical CCSD(T) level of accuracy and allows converged molecular dynamics simulations with fully quantized electrons and nuclei. We present MD simulations, for flexible molecules with up to a few dozen atoms and provide insights into the dynamical behavior of these molecules. Our approach provides the key missing ingredient for achieving spectroscopic accuracy in molecular simulations.

Highlights

Molecular dynamics (MD) simulations employing classical force fields constitute the cornerstone of contemporary atomistic modeling in chemistry, biology, and materials science
To solve this accuracy and molecular size dilemma and to enable converged ab initio molecular dynamics (AIMD) simulations close to the exact solution of the Schrödinger equation, here we develop an alternative approach using symmetrized gradient-domain machine learning to construct force fields with the accuracy of high-level ab initio calculations
While methods for identifying molecular point groups for polyatomic rigid molecules are readily available[48], LonguetHiggins[49] has pointed out that non-rigid molecules have extra symmetries. These dynamical symmetries arise upon functionalgroup rotations or torsional displacements and they are usually not incorporated in traditional force fields and electronicstructure calculations

Summary

Introduction

Molecular dynamics (MD) simulations employing classical force fields constitute the cornerstone of contemporary atomistic modeling in chemistry, biology, and materials science. An alternative is a direct fit of the potential-energy surface (PES) from a large number of CCSD(T) calculations, this is only practically achievable for rather small and rigid molecules[3,4,5] To solve this accuracy and molecular size dilemma and to enable converged AIMD simulations close to the exact solution of the Schrödinger equation, here we develop an alternative approach using symmetrized gradient-domain machine learning (sGDML) to construct force fields with the accuracy of high-level ab initio calculations. Our approach contributes the key missing ingredient for achieving spectroscopic accuracy and rigorous dynamical insights in molecular simulations

Objectives

Methods

Results

Conclusion