Abstract:van der Waals dispersion-repulsion interactions, commonly represented in atomistic force fields by the Lennard-Jones (LJ) potential, play an important role in the accuracy of molecular simulations. Training the force field parameters used in the LJ potential is challenging, generally requiring adjustment based on simulations of macroscopic physical properties. The large computational expense of these simulations, especially if many parameters are trained simultaneously, limits the size of training data set and number of optimization steps that can be taken, often requiring modelers to perform optimizations within a local parameter region. To allow for more global LJ parameter optimization against large training sets, we introduce a multi-fidelity optimization technique which uses Gaussian process surrogate modeling to build inexpensive models of physical properties as a function of LJ parameters. This allows for fast evaluation of objective functions, greatly accelerating searches over parameter space and enabling the use of global optimization algorithms. We use an iterative framework which performs optimization with differential evolution at the surrogate level, followed by validation at the simulation level and surrogate refinement. Using this technique on two previously studied training sets, containing up to 195 physical property targets, we refit a subset of the LJ parameters for the OpenFF 1.0.0 (Parsley) force field. We demonstrate that this multi-fidelity technique can find improved parameter sets compared to a purely simulation-based optimization by searching more broadly and escaping local minima. In particular, this technique often finds significantly different parameter minima that have comparably accurate performance. In most cases, these parameter sets are transferable to other similar molecules in a test set. This multi-fidelity technique provides a platform for fast optimization against physical properties that can be refined and applied in multiple ways to the development of molecular models.

Machine Learning Directed Optimization of Classical Molecular Modeling Force Fields

Utilizing Machine Learning for Efficient Parameterization of Coarse Grained Molecular Force Fields

Efficient Machine Learning Force Field for Large-Scale Molecular Simulations of Organic Systems

Using physical property surrogate models to perform accelerated multi-fidelity optimization of force field parameters

Force field optimization by end-to-end differentiable atomistic simulation

Machine Learning of Accurate Energy-Conserving Molecular Force Fields

PhyNEO: A Neural-Network-Enhanced Physics-Driven Force Field Development Workflow for Bulk Organic Molecule and Polymer Simulations.

Empirical optimization of molecular simulation force fields by Bayesian inference

Machine Learning Force Fields: Construction, Validation, and Outlook

Hybrid Classical/Machine-Learning Force Fields for the Accurate Description of Molecular Condensed-Phase Systems

Towards Exact Molecular Dynamics Simulations with Machine-Learned Force Fields

Fine-Tuning Property Domain Weighting Factors and the Objective Function in Force Field Parameter Optimization

Improving machine learning force fields for molecular dynamics simulations with fine-grained force metrics

Forces are not Enough: Benchmark and Critical Evaluation for Machine Learning Force Fields with Molecular Simulations

MACE-OFF23: Transferable Machine Learning Force Fields for Organic Molecules

On the design space between molecular mechanics and machine learning force fields

Parametrization of Non-Bonded Force Field Terms for Metal-Organic Frameworks Using Machine Learning Approach

Accurate interatomic force field for molecular dynamics simulation by hybridizing classical and machine learning potentials

The emergence of machine learning force fields in drug design

On-the-fly machine learning force field generation: Application to melting points

An Initial Design-enhanced Deep Learning-based Optimization Framework to Parameterize Multicomponent ReaxFF Force Fields