Environment-adaptive machine learning potentials

Aug 3, 2026

Environment-adaptive machine learning potentials enable accurate modeling of energies and forces across diverse physical phenomena and varying external conditions. The method constructs local, many-body interatomic potentials by clustering atomic environments, assigning each atom a probability of belonging to each cluster, and using a many-potential expansion to smoothly blend local surfaces with global continuity. Predictions for Ta and InP are benchmarked against density functional theory calculations.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Introduction

Motivation for environment-adaptive interatomic potentials

Background: empirical interatomic potentials and MLIPs

Descriptors and representations of atomic structures

Environment-adaptive machine learning potentials (method overview)

Clustering local atomic environments

Many-body many-potential expansion and blending via probabilities

Benchmarking against density functional theory

arXiv :2405 .00306v2 [ cond-mat .mtrl-sci ] 29 Jul 2024 Environment-adaptive machine learning potentials Ngoc Cuong Nguyen Department of Aeronautics and Astronautics, Massachusetts Institute of Technology 77 Massachusetts Avenue, Cambridge, MA 02139 Dionysios Sema Department of Mechanical Engineering, Massachusetts Institute of Technology 77 Massachusetts Avenue, Cambridge, MA 02139 (Dated: July 31, 2024) The development of interatomic potentials that can accurately capture a wide range of physical phenomena and diverse environments is of significant interest, but it presents a formidable challenge. This challenge arises from the numerous structural forms, multiple phases, complex intramolecular and intermolecular interactions, and varying external conditions. In this paper, we present a method to construct environment-adaptive interatomic potentials by adapting to the local atomic environment of each atom within a system. The collection of atomic environments of interest is partitioned into several clusters of atomic environments. Each cluster represents a distinctive local environment and is used to define a corresponding local potential. We introduce a many-body many-potential expansion to smoothly blend these local potentials to ensure global continuity of the potential energy surface. This is achieved by computing the probability functions that determine the likelihood of an atom belonging to each cluster. We apply the environment-adaptive machine learning potentials to predict observable properties for Ta element and InP compound, and compare them with density functional theory calculations. I. INTRODUCTION Molecular dynamics (MD) simulations require an accurate calculation of energies and forces to analyze the physical movements of atoms. While electronic structure calculations provide accurate energies and forces, they are restricted to analyzing small length scales and short time scales due to their high computational complexity. Interatomic potentials represent the potential energy surface (PES) of an atomic system as a function of atomic positions and thus leave out the detailed electronic structures. They can enable MD simulations of large systems with millions or even billions of atoms over microseconds. Over the years, empirical interatomic potentials (EIPs) such as the Finnis-Sinclair potential [1], embedded atom method (EAM) [2], modified EAM (MEAM) [3], Stillinger-Weber (SW) [4], Tersoff [5], Brenner [6], EDIP [7], COMB [8], ReaxFF [9] have been developed to treat a wide variety of atomic systems with different degrees of complexity. EAM potential has its root from the Finnis-Sinclair potential [1] in which the embedding function is a square root function. The MEAM potential [3] was developed as a generalization of the EAM potential by including angular-dependent interactions in the electron density term. The SW potential takes the form of a three-body potential in which the total energy is expressed as a linear combination of two- and three-body terms. The Tersoff potential is fundamentally different from the SW potential in that the strength of individual pair interactions is affected by the presence of surrounding atoms. The Brenner potential is based directly on the Tersoff potential but has additional terms and parameters which allow it to better describe various chemical environments. EDIP is designed to more accurately represent interatomic interactions by considering the effects of the local atomic environment on these interactions. Because EAM, MEAM, Tersoff, Brenner, EDIP, ReaxFF and COMB potentials dynamically adjust the strength of the bond based on the local environment of each atom, they can describe several different bonding states and complex behaviors of atoms in various states, including defects, phase transitions, surfaces, and interfaces within materials. One of the key features of ReaxFF and COMB is their ability to handle charge equilibration in a manner that inclu

FAQs

Why are environment-adaptive machine learning potentials needed?

Accurate potentials must capture wide-ranging physical phenomena under diverse structural forms, phases, interactions, and external conditions, which is difficult for a single fixed-form potential.

How does the proposed method construct environment-adaptive potentials?

Atomic environments are clustered, each cluster defines a local potential, and a many-body many-potential expansion blends these local potentials using probability functions for cluster membership to ensure global continuity.

How are the environment-adaptive machine learning potentials evaluated?

Observable properties for Ta and the InP compound are predicted and compared with density functional theory calculations to assess accuracy.

Aug 3, 2026

Uploaded by River Wang