Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Comparative Study of Machine Learning Models Predicting Energetics of Interacting Defects

Hao Yu

TL;DR

We address predicting the energetics of interacting surface defects by comparing descriptor-based ML, graph neural networks, and configurational cluster expansion (CE) on a small DFT-derived dataset. CE delivers the most accurate energy predictions with limited data and serves as a low-cost generator of surrogate energies to train DL models such as DimeNet, which approaches MBTR performance when trained with ~5,000 CE configurations. The study demonstrates the practical value of CE for enabling rapid screening of defect configurations on surfaces and clarifies the data requirements for successful DL application in these systems. Collectively, the work provides a preliminary, data-efficient evaluation of ML techniques for imperfect surfaces and defines a workflow for leveraging CE to bootstrap DL models.

Abstract

Interacting defect systems are ubiquitous in materials under realistic scenarios, yet gaining an atomic-level understanding of these systems from a computational perspective is challenging - it often demands substantial resources due to the necessity of employing supercell calculations. While machine learning techniques have shown potential in accelerating materials simulations, their application to systems involving interacting defects remains relatively rare. In this work, we present a comparative study of three different methods to predict the free energy change of systems with interacting defects. We leveraging a limited dataset from Density Functional Theory(DFT) calculations to assess the performance models using materials descriptors, graph neural networks and cluster expansion. Our findings indicate that the cluster expansion model can achieve precise energetics predictions even with this limited dataset. Furthermore, with synthetic data generate from cluster expansion model at near-DFT levels, we obtained enlarged dataset to assess the demands on data for training accurate prediction models using graph neural networks for systems featuring interacting defects. A brief discussion of the computational cost for each method is provided at the end. This research provide a preliminary evaluation of applying machine learning techniques in imperfect surface systems.

A Comparative Study of Machine Learning Models Predicting Energetics of Interacting Defects

TL;DR

We address predicting the energetics of interacting surface defects by comparing descriptor-based ML, graph neural networks, and configurational cluster expansion (CE) on a small DFT-derived dataset. CE delivers the most accurate energy predictions with limited data and serves as a low-cost generator of surrogate energies to train DL models such as DimeNet, which approaches MBTR performance when trained with ~5,000 CE configurations. The study demonstrates the practical value of CE for enabling rapid screening of defect configurations on surfaces and clarifies the data requirements for successful DL application in these systems. Collectively, the work provides a preliminary, data-efficient evaluation of ML techniques for imperfect surfaces and defines a workflow for leveraging CE to bootstrap DL models.

Abstract

Interacting defect systems are ubiquitous in materials under realistic scenarios, yet gaining an atomic-level understanding of these systems from a computational perspective is challenging - it often demands substantial resources due to the necessity of employing supercell calculations. While machine learning techniques have shown potential in accelerating materials simulations, their application to systems involving interacting defects remains relatively rare. In this work, we present a comparative study of three different methods to predict the free energy change of systems with interacting defects. We leveraging a limited dataset from Density Functional Theory(DFT) calculations to assess the performance models using materials descriptors, graph neural networks and cluster expansion. Our findings indicate that the cluster expansion model can achieve precise energetics predictions even with this limited dataset. Furthermore, with synthetic data generate from cluster expansion model at near-DFT levels, we obtained enlarged dataset to assess the demands on data for training accurate prediction models using graph neural networks for systems featuring interacting defects. A brief discussion of the computational cost for each method is provided at the end. This research provide a preliminary evaluation of applying machine learning techniques in imperfect surface systems.
Paper Structure (7 sections, 2 equations, 3 figures, 2 tables)

This paper contains 7 sections, 2 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Atomic Representations
  4. Configurational Cluster Expansion
  5. Results
  6. Conclusion
  7. Disclaimer

Figures (3)

  • Figure 1: Example of lithium slab with defects on surface. Left: Top view is looking at the surface with defects (represented by gray circles). Circles filled with purple represent the surface lithium atoms, while circles with half filled purple represent the lithium atoms in the second layer. Dashed square indicate the unit cell. Right: Side view of a lithium slab. Surface atoms represented with filled purple circles. Atoms at first 2 layers match the ones on the left figure. Dashed square is unit cell box. We tried different representations by keeping atoms with 2/4/6 layers in the structure.
  • Figure 2: Train and Test predictions with initial small data set.
  • Figure 3: Test performance of DimeNet training with more data provided by cluster expansion model