Authors:

Stephen Brice Rider

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Due to the Schrödinger equation being impossible to solve analytically for multi-particle systems, various techniques have been developed to approximate the electronic properties of these systems. One such method is Density Functional Theory (DFT), which relies on the idea that the electron density that minimizes the system's energy corresponds to the solution to the Schrödinger equation. However, the computational efficiency of even DFT prohibits simulations of systems with atoms in O(1000). To overcome such a challenge, machine-learning models have been trained to predict the atomistic forces with orders of magnitude improved computational efficiency. This research aims to expand the applications of force fields to ceramics at extreme conditions using an automated, open-access pipeline utilizing nanoHUB infrastructure. The pipeline consists of a tool that uses Quantum Espresso, a coded implementation of DFT, to acquire the inputs to a machine- learned force field created with Spectral Neighbor Analysis Potential (SNAP). From there, molecular dynamics via the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) is used to improve the model iteratively. The result is a tool that uses machine learning to deliver quantum-accurate interatomic potentials to the user for any ceramic and semiconductor. This tool was designed to investigate known ceramics to better inform the design of military protective gear, but it can be used to generate interatomic potentials for general applications of any material class. Keywords: Neural Network; Ceramics; Cyberinfrastructure; DFT; Force Field

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Stephen Brice Rider