Gaussian graphical models with graph constraints for magnetic moment interaction in high entropy alloys

Xinrui Liu, Yifeng Wu, Douglas L. Irving, Meng LiarXiv 2021 · DOI · arXiv

Abstract

This article is motivated by studying the interaction of magnetic moments in high entropy alloys (HEAs), which plays an important role in guiding HEA designs in materials science. While first principles simulations can capture magnetic moments of individual atoms, explicit models are required to analyze their interactions. This is essentially an inverse covariance matrix estimation problem. Unlike most of the literature on graphical models, the inverse covariance matrix possesses inherent structural constraints encompassing node types, topological distance of nodes, and partially specified conditional dependence patterns. The present article is, to our knowledge, the first to consider such intricate structures in graphical models. In particular, we introduce graph constraints to formulate these structures that arise from domain knowledge and are critical for interpretability, which leads to a Bayesian conditional autoregressive model with graph constraints (CARGO) for structured inverse covariance matrices. The CARGO method enjoys efficient implementation with a modified Gauss-Seidel scheme through proximity operators for closed-form posterior exploration. We establish algorithmic convergence for the proposed algorithm under a verifiable stopping criterion. Simulations show competitive performance of CARGO relative to several other methods and confirm our convergence analysis. In a novel real data application to HEAs, the proposed methods lead to data-driven quantification and interpretation of magnetic moment interactions with high tractability and transferability.

Introduction

High Entropy Alloys (HEAs) are advanced materials composed of five or more elements in near-equal proportions, known for their exceptional mechanical and corrosion-resistant properties. A critical challenge in HEA design is understanding the interaction of magnetic moments, which significantly influences material stability and performance. While first-principles simulations like Density Functional Theory (DFT) can calculate individual atomic magnetic moments, they do not explicitly model the underlying interaction network. This research addresses this gap by framing the problem as an inverse covariance matrix estimation task, introducing a novel Bayesian framework that incorporates physical domain knowledge directly into the statistical model.

Methodology: The CARGO Framework

The researchers developed CARGO (Conditional Autoregressive model with Graph cOnstraints) to estimate structured inverse covariance matrices. Unlike standard graphical models like the graphical Lasso (gLasso), which primarily focus on sparsity, CARGO enforces constraints derived from the physical geometry of the crystal lattice.

Graph Constraints and Topological Distance

The model utilizes Gaussian Markov Random Fields (GMRF) where atoms are nodes and edges represent magnetic interactions. The core innovation lies in the graph constraints ($N$), which account for:

  1. Node Types: Interactions are categorized by the elements involved (e.g., how a Nickel atom affects an Iron atom).
  2. Topological Distance: The model respects the periodic, repeating nature of the crystal, ensuring that interactions at the same relative distance and orientation are treated consistently across the lattice.
Figure 2: Illustration of topological distances and neighbor interactions within a crystal lattice, where numbered nodes represent different shells of neighbors.

Bayesian Estimation and Convergence

CARGO employs a constrained Wishart prior to integrate these structural rules. To solve the resulting optimization problem, the authors implemented a modified Gauss-Seidel algorithm using proximity operators. This approach provides closed-form solutions for posterior exploration, ensuring efficient computation and guaranteed algorithmic convergence even with complex constraints.

Experimental Setup and Datasets

The performance of CARGO was validated using both synthetic and real-world datasets:

  • Synthetic Data: Simulations with varying atom counts ($p = 20, 50, 100$) and three node types were used to benchmark CARGO against standard methods like gLasso and PLT0.
  • Real HEA Data: The study analyzed a NiCrCoFe (Nickel, Chromium, Cobalt, Iron) system across three distinct crystal structures: face-centered cubic (fcc), hexagonal close-packed (hcp), and double hcp (dhcp).
Figure 1: Representative crystal structures (fcc, hcp, and dhcp) of the NiCrCoFe alloys used for real-data validation.

Key Results and Discussion

Performance and Accuracy

In synthetic tests, CARGO consistently outperformed gLasso and PLT0, achieving lower Mean Squared Error (MSE). The advantage of CARGO became more pronounced as the system size ($p$) increased, demonstrating its ability to leverage structural constraints to navigate high-dimensional parameter spaces effectively. The algorithm demonstrated rapid convergence, typically within 25 iterations for a 100-atom system.

Physical Insights: Magnetic Frustration

When applied to the NiCrCoFe alloy, CARGO provided data-driven confirmation of "magnetic frustration." The model identified that Chromium (Cr) consistently exhibits negative interaction coefficients with Ni, Co, and Fe. This implies that the presence of Cr actively suppresses the magnetic moments of its neighbors, a finding highly consistent with established metallurgical literature and physical theory.

Conclusion and Future Directions

The CARGO framework represents a significant advancement in the statistical modeling of materials. By embedding physical constraints into a Bayesian graphical model, it achieves high interpretability and transferability. Future research will explore hierarchical modeling to capture random effects across multiple samples and extend the framework to predict other material properties such as formation energy and elastic moduli, potentially creating a unified statistical framework for alloy design.

Summary updated

Abstract

This article is motivated by studying the interaction of magnetic moments in high entropy alloys (HEAs), which plays an important role in guiding HEA designs in materials science. While first principles simulations can capture magnetic moments of individual atoms, explicit models are required to analyze their interactions. This is essentially an inverse covariance matrix estimation problem. Unlike most of the literature on graphical models, the inverse covariance matrix possesses inherent structural constraints encompassing node types, topological distance of nodes, and partially specified conditional dependence patterns. The present article is, to our knowledge, the first to consider such intricate structures in graphical models. In particular, we introduce graph constraints to formulate these structures that arise from domain knowledge and are critical for interpretability, which leads to a Bayesian conditional autoregressive model with graph constraints (CARGO) for structured inverse covariance matrices. The CARGO method enjoys efficient implementation with a modified Gauss-Seidel scheme through proximity operators for closed-form posterior exploration. We establish algorithmic convergence for the proposed algorithm under a verifiable stopping criterion. Simulations show competitive performance of CARGO relative to several other methods and confirm our convergence analysis. In a novel real data application to HEAs, the proposed methods lead to data-driven quantification and interpretation of magnetic moment interactions with high tractability and transferability.
BibTeX
@misc{liu2021gaussian,
  author = {Xinrui Liu and Yifeng Wu and Douglas L. Irving and Meng Li},
  title = {{Gaussian graphical models with graph constraints for magnetic moment interaction in high entropy alloys}},
  year = {2021},
  eprint = {2105.05280},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2105.05280},
}
Figure