Machine learning potentials with Iterative Boltzmann Inversion: training to experiment

Sakib Matin,Alice Allen,Justin S. Smith,Nicholas Lubbers,Ryan B. Jadrich,Richard A. Messerly,Benjamin T. Nebgen,Ying Wai Li,Sergei Tretiak,Kipton Barros

2023-07-11

Abstract:Methodologies for training machine learning potentials (MLPs) to quantum-mechanical simulation data have recently seen tremendous progress. Experimental data has a very different character than simulated data, and most MLP training procedures cannot be easily adapted to incorporate both types of data into the training process. We investigate a training procedure based on Iterative Boltzmann Inversion that produces a pair potential correction to an existing MLP, using equilibrium radial distribution function data. By applying these corrections to a MLP for pure aluminum based on Density Functional Theory, we observe that the resulting model largely addresses previous overstructuring in the melt phase. Interestingly, the corrected MLP also exhibits improved performance in predicting experimental diffusion constants, which are not included in the training procedure. The presented method does not require auto-differentiating through a molecular dynamics solver, and does not make assumptions about the MLP architecture. The results suggest a practical framework of incorporating experimental data into machine learning models to improve accuracy of molecular dynamics simulations.

Applied Physics

What problem does this paper attempt to address?

This paper discusses how to improve the training method for quantum mechanical simulation data using machine learning potentials (MLPs), especially by incorporating experimental data into the training process. Traditional training of machine learning potentials primarily relies on quantum mechanical simulation data, but experimental data has different characteristics and is difficult to directly integrate. The study proposes a training procedure based on Iterative Boltzmann Inversion (IBI), which generates a pairing potential correction for the existing MLP to match experimental data based on the equilibrium radial distribution function (RDF). By applying this method to a pure aluminum MLP, the study found that the modified model significantly improves RDF predictions in the molten phase and also enhances the accuracy of predicting experimental diffusion constants, which were not included in the training data. The paper also points out that although existing MLPs perform well on a large amount of high-fidelity quantum mechanical calculation data, training using experimental data is still relatively limited. Because experimental data typically involves averaged information and may exhibit sparsity and uncertainty, integrating experimental data into the training process poses a challenge. The proposed IBI method does not require automatic differentiation through molecular dynamics solvers and does not make specific assumptions about MLP architecture, providing a practical framework for incorporating experimental data and improving the accuracy of molecular dynamics simulations. In summary, this paper aims to address how to improve training strategies to better incorporate experimental data into MLPs, thus enhancing the accuracy of material property predictions and the credibility of simulations, particularly in the liquid phase.

Machine learning potentials with Iterative Boltzmann Inversion: training to experiment

Machine Learning Potentials with the Iterative Boltzmann Inversion: Training to Experiment

Learning together: Towards foundation models for machine learning interatomic potentials with meta-learning

Learning Together: Towards foundational models for machine learning interatomic potentials with meta-learning

Considerations in the use of ML interaction potentials for free energy calculations

Transferability of datasets between Machine-Learning Interaction Potentials

A dual-cutoff machine-learned potential for condensed organic systems obtained via uncertainty-guided active learning

Learning Interatomic Potentials at Multiple Scales

Refining Potential Energy Surface through Dynamical Properties via Differentiable Molecular Simulation

Lifelong Machine Learning Potentials

Stable and Accurate Atomistic Simulations of Flexible Molecules using Conformationally Generalisable Machine Learned Potentials

Learning QM/MM Potential using Equivariant Multiscale Model

Active learning of Boltzmann samplers and potential energies with quantum mechanical accuracy

Towards a Generalizable Machine-Learned Potential for Metal-Organic Frameworks

chemtrain: Learning Deep Potential Models via Automatic Differentiation and Statistical Physics

Machine Learning in QM/MM Molecular Dynamics Simulations of Condensed-Phase Systems

Tutorial: How to Train a Neural Network Potential

Active Sparse Bayesian Committee Machine Potential for Isothermal-Isobaric Molecular Dynamics Simulations

Synthetic pre-training for neural-network interatomic potentials

Molecular Dynamics with Neural-Network Potentials

Machine Learning Nonadiabatic Dynamics: Eliminating Phase Freedom of Nonadiabatic Couplings with the State-Intraction State-Averaged Spin-Restricted Ensemble-Referenced Kohn-Sham Approach