A Simple and Efficient Non-DFT-Based Machine Learning Interatomic Potential to Simulate Titanium MXenes

TL;DR

This paper introduces a new machine learning interatomic potential that accurately simulates titanium MXenes and related materials, offering a computationally efficient alternative to density functional theory methods.

Contribution

A simple, non-DFT-based machine learning potential for titanium MXenes that achieves DFT-level accuracy in structure and elastic property predictions.

Findings

Accurately reproduces structure and elastic properties of titanium MXenes.

Demonstrates efficiency over traditional DFT calculations.

Identifies limitations and suggests improvements for the MLIP.

Abstract

Titanium MXenes are two-dimensional inorganic structures composed of titanium and carbon or nitrogen elements, with distinctive electronic, thermal and mechanical properties. Despite the extensive experimental investigation, there is a paucity of computational studies at the level of classical molecular dynamics (MD). As demonstrated in a preceding study, known MD potentials are not capable of fully reproducing the structure and elastic properties of every titanium MXene. In this study, we present a simply trained, but yet efficient, non-density functional theory-based machine learning interatomic potential (MLIP) capable of simulating the structure and elastic properties of titanium MXenes and bulk titanium carbide and nitride with precision comparable to DFT calculations. The training process for the MLIP is delineated herein, in conjunction with a series of dynamical tests. Limitations of the MLIP and steps towards improving its efficacy to simulate titanium MXenes are discussed.