Machine Learning-Driven Analytical Models for Threshold Displacement Energy Prediction in Materials
Published in arXiv, 2025
This work demonstrates the potential of machine learning, specifically the Sure Independence Screening and Sparsifying Operator (SISSO) method, to develop analytical expressions for predicting threshold displacement energy (Ed) values in materials. The models constructed using SISSO outperform traditional approaches like that used in Konobeyev et al. in terms of accuracy and generalizability, leveraging fundamental material properties. These results provide a valuable tool for estimating Ed in monoatomic materials, reducing the reliance on complex experiments and simulations.
For polyatomic materials, while the SISSO models showed limited success due to the inherent complexity and diversity of the data, incorporating improved datasets and additional features could enhance their predictive capability. The effective Ed formulations for complex materials were previously constrained by the availability of known monoatomic Ed values, considering our monoatomic predictions, this method offers a more robust pragmatic path forward. This study emphasizes the importance of cohesive energy and melting temperature as key contributors to Ed. These parameters are strongly related to the atomic bonding strength within the solid’s structure, reaffirming their role in defect formation dynamics.
Future work should focus on expanding datasets, exploring temperature dependence, and refining models to capture the nuanced behavior of polyatomic materials under varying conditions. This approach represents a significant step toward efficient, data-driven prediction of radiation damage parameters, facilitating advancements in materials science applications in nuclear energy, aerospace, and other radiation-intense environments.
Recommended citation: Duque, Rosty B. Martinez, Arman Duha, and Mario F. Borunda. "Machine Learning-Driven Analytical Models for Threshold Displacement Energy Prediction in Materials." arXiv preprint arXiv:2502.01813 (2025).
Download Paper