Atomistic Study of U₃Si₂

Uranium silicide (U₃Si₂) has gained significant attention as an advanced fuel material for nuclear reactors, particularly in research reactors, due to its superior thermal conductivity and higher uranium density compared to traditional uranium dioxide. However, U₃Si₂ is known to amorphize under certain conditions, which poses challenges for its stability and performance. Understanding the mechanisms behind this amorphization is crucial for predicting its behavior in reactor environments and optimizing its use as a fuel material.

In this study, we employed DFT+U simulations to achieve a detailed understanding of U₃Si₂, focusing on properties that are crucial for its practical application. To accelerate the heavy calculations and sample sufficient configurations, we utilized Machine Learning Assisted Canonical Sampling (MLACS) [1], allowing us to efficiently compute the necessary data. With these configurations, we applied the aTDEP method to calculate elastic moduli, forces, and thermal conductivity, comparing our results with experimental data. These findings highlight the sensitivity of thermodynamic properties to structural adjustments and underscore the need for precise modeling in predicting fuel behavior.

A key extension of this research is the introduction of aluminum into the U₃Si₂ matrix. This modification is essential for its use as a dispersion fuel in research reactors, where it will be embedded within an aluminum matrix. We are investigating the impact of aluminum on the elastic and thermal properties of the alloy, as well as its phase-change behavior. Understanding these interactions is critical for predicting the performance and stability of U₃Si₂-Al under reactor conditions.

Looking ahead, we aim to scale up our study by training a machine learning potential based on these DFT+U results. This approach will enable a broader exploration of U₃Si₂ and U₃Si₂-Al properties, paving the way for more comprehensive assessments of their suitability in nuclear fuel applications.


References
Castellano, Aloïs, et al. "A b initio canonical sampling based on variational inference." Physical Review B 106.16 (2022): L161110.
J. Bouchet, et al., Uranium Science. 2021.