| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Composite Materials for Nuclear Applications III
|
| Presentation Title |
Equivariant neural network force fields for 11-cation chloride molten salts system |
| Author(s) |
Chen Shen, Siamak Attarian, Mark Asta, Izabela Szlufarska, Dane Morgan |
| On-Site Speaker (Planned) |
Chen Shen |
| Abstract Scope |
Molten salts are promising ionic liquids for clean energy applications like nuclear and solar energy. However, evaluating their properties from a microscopic perspective is challenging. Machine learning interatomic potentials (MLIPs) offer near ab initio accuracy with the efficiency of classical force fields. Standard MLIPs typically fit a few species, while universal potentials fit 50 or more. We propose a middle ground, treating 10-20 elements to achieve near ab initio accuracy for a large composition space. Our 'SuperSalts' potential targets 11-cation chloride melts (e.g., LiCl, NaCl, KCl) and predicts properties like density, specific heat, and viscosity with near-DFT accuracy. Using ~70,000 ab initio configurations, SuperSalts potential is more efficient than fitting numerous suballoys separately and serves as a validated resource, suggesting a shift towards machine learning approaches in molten salt modeling. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Nuclear Materials, Machine Learning, Modeling and Simulation |