Abstract: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.
What problem does this paper attempt to address?
The paper is primarily dedicated to addressing the problem of constructing efficient and accurate Environment-Adaptive Machine Learning Potentials (EAML) in Molecular Dynamics (MD) simulations. Specifically, the research focuses on the following aspects:
1. **Challenges and Background**: Developing interaction potentials that can accurately capture a wide range of physical phenomena and diverse environments is a significant challenge. These environments include various structural forms (such as crystalline, amorphous, defects, interfaces, etc.), different phases (solid, liquid, gas, plasma), and varying external conditions (temperature, pressure, chemical environment, etc.). Traditional methods like Empirical Interaction Potentials (EIPs) and Machine Learning Interaction Potentials (MLIPs) have made some progress, but there is still room for improvement in terms of accuracy, efficiency, and transferability.
2. **Method Introduction**: The authors propose a new method to construct environment-adaptive interaction potentials. First, all atomic environments of interest are clustered into several groups, each representing a unique local environment, and the corresponding local potentials are defined accordingly. To ensure global continuity of the entire potential energy surface, the authors introduce a Multi-Body Multi-Potential (MBMP) extension, which smoothly blends these local potentials by calculating probability functions.
3. **Key Techniques**: The paper details how to use orthogonal decomposition methods to construct invariant descriptors in multi-element systems (i.e., "orthogonal proper descriptors"), which can effectively characterize different local atomic environments. Additionally, it presents the key techniques for combining these local potentials to form a globally continuous potential energy surface.
4. **Application Validation**: Finally, the authors validate the effectiveness of the proposed environment-adaptive machine learning potential by predicting the observable properties of Tantalum (Ta) and Indium Phosphide (InP) compounds and comparing the results with Density Functional Theory (DFT) calculations.
In summary, the main goal of this paper is to address how to construct machine learning potentials that are both accurate and efficient, capable of adapting to various complex local atomic environments, and maintaining good performance across different environments.