Ultra-fast interpretable machine-learning potentials

Abstract

All-atom dynamics simulations are an indispensable quantitative tool in physics, chemistry, and materials science, but large systems and long simulation times remain challenging due to the trade-off between computational efficiency and predictive accuracy. To address this challenge, we combine effective two-and three-body potentials in a cubic B-spline basis with regularized linear regression to obtain machine-learning potentials that are physically interpretable, sufficiently accurate for applications, as fast as the fastest traditional empirical potentials, and two to four orders of magnitude faster than state-of-the-art machine-learning potentials. For data from empirical potentials, we demonstrate exact retrieval of the potential. For data from density functional theory, the predicted energies, forces, and derived properties, including phonon spectra, elastic constants, and melting points, closely match those of the reference method. The introduced potentials might contribute towards accurate all-atom dynamics simulations of large atomistic systems over long time scales.

High pressure hydrogen by machine learning and quantum Monte Carlo

“…Moreover we believe that further substantial improvements will be possible by using baseline potentials (e.g. the recent ultrafast MLP [39]) much more efficient than DFT ones. Indeed, we expect that our approach will enable the study of systems that have been so far impossible to simulate, due to the almost intractable computational cost required by advanced many-body techniques, such as QMC.…”

Multibody Terms in Protein Coarse-Grained Models: A Top-Down Perspective

“…The initial values of c n are set to reproduce the optimized potential given by eqs and . The spline representation (eq ) is then further optimized by adjusting the c n coefficients following the ODEM procedure (see Figure ).…”

Science‐Driven Atomistic Machine Learning

“…[84,87] In a similar vein, models like UF3 or ChiMES use explicit body-order expansions of the energy in terms of products of two-body functions. [88,89] To illustrate the benefits of inductive biases such as equivariance and high body-order for science-driven ML, it is instructive to consider the MACE approach of Batatia et al [82] In MACE, both the equivariance and the bodyorder of the model can be controlled via hyperparameters. As shown in Figure 6, both of these factors lead to improved predictive accuracy.…”