Accurate neural-network-based fitting of full-dimensional two-body potential energy surfaces

TL;DR

This paper presents a neural network approach for accurately fitting full-dimensional two-body potential energy surfaces of atmospheric gases, improving both accuracy and computational efficiency over existing methods.

Contribution

The authors develop a permutationally invariant polynomial neural network (PIP-NN) method tailored for full-dimensional two-body potentials, ensuring correct asymptotic behavior and superior performance.

Findings

PIP-NN achieves the highest accuracy among tested models for large datasets.

The method is computationally efficient, comparable to PIP regression, and much faster than other neural network approaches.

Full-dimensional potentials for N₂-Ar and N₂-CH₄ were successfully constructed and validated.

Abstract

We describe the development of machine-learned potentials of atmospheric gases with flexible monomers for molecular simulations. A recently suggested permutationally invariant polynomial neural network (PIP-NN) approach is utilized to represent the full-dimensional two-body component of the dimer energy. To ensure the asymptotic zero-interaction limit, a tailored subset of the full invariant polynomial basis set is utilized and their variables are modified to achieve a better fit of the correct asymptotic behavior at a long range. The new technique is used to build full-dimensional potentials for the two-body N$_2-$Ar and N$_2-$CH$_4$ interactions by fitting databases of ab initio energies calculated at the coupled-cluster level of theory. The second virial coefficients with full account of molecular flexibility effects are then calculated within the classical framework using the PIP-NN potential surfaces. To showcase the advantages of the PIP-NN method, we compare its accuracy and computational efficiency to several kernel-based and neural-network-based approaches using the MD17 database of energies and forces for ethanol. For large training set sizes, the PIP-NN models attain the best accuracy among examined models, and the computation time is shown to be comparable to that of the PIP regression model and several orders of magnitude faster than the quickest alternatives.