Machine learning improves force fields for molecular simulations
Learning consistent molecular mechanics force fields from first principles
Machine Learning
Summary
Simulating how molecules behave requires models called force fields that estimate energy from atomic interactions. Traditional force fields simplify these interactions but have limitations in accuracy and adaptability. The authors developed a new method that uses machine learning to jointly learn both bonded and nonbonded parameters of these force fields from high-quality quantum data. This approach leads to better predictions of molecular shapes and behaviors, without relying on manual parameter assignments.
What this means in practice
- •For computational chemists: Use the learned force fields to simulate molecular structures and dynamics more accurately without manual parameter tuning.
- •For materials scientists: Model materials at the atomic level with improved energy predictions derived consistently from first-principles data.
Authors
Berkay Günes, Leif Seute, Jigyasa Nigam, Frauke Gräter
Abstract
Classical force fields (FFs) remain the workhorse for large-scale simulations even as machine-learned interatomic potentials (MLIPs) approach ab initio accuracy. They decompose total configuration energies into simple effective interactions whose parameters are traditionally assigned based on atom or bond types, enabling efficient simulations but also limiting their ability to adapt across configurations. Recent machine learning approaches have improved the accuracy and transferability of bonded parameters in these FFs by inferring them as functions of local atomic environments, but still rely on empirical nonbonded parameters for practical simulations. In this work, we introduce a unified approach, \texttt{grappa-fullFF}, which learns both bonded and nonbonded parameters \emph{consistently} and simultaneously from ab initio reference data. By incorporating physically inspired regularization via supervision of the electrostatic potential and an architecture that facilitates charge equilibration, our model recovers accurate electric response properties, achieves state-of-the-art accuracy on geometry optimization benchmarks, and reproduces the conformational sampling of both classical and existing machine-learned FFs, without relying on externally assigned nonbonded parameters.