About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Advancements in Molten Salt/Metal Technology in Energy Applications: From Atoms to Plants
|
| Presentation Title |
Accurate Free Energy Simulations in Molten Salts with Machine Learning Potentials and High-Level Quantum Chemical Methods |
| Author(s) |
Vyacheslav Bryantsev, Luke Gibson, Rajni Chahal |
| On-Site Speaker (Planned) |
Vyacheslav Bryantsev |
| Abstract Scope |
Understanding the thermodynamic properties of molten salts is essential for advancing next-generation nuclear technologies. We introduce a computational framework that enables accurate predictions of chemical potentials and redox free energies by integrating machine learning interatomic potentials (MLIPs) with electronic structure calculations. Using a thermodynamic integration scheme, we compute free energy differences from MLIP-driven molecular dynamics simulations, capturing both liquid and solid phase behavior with DFT-level accuracy and beyond. To improve the accuracy further, we train our MLIPs on data from hybrid DFT, RPA, and MP2 using a delta learning scheme. This allows the MLIP to approximate high-level accuracy more efficiently, using only a fraction of the data required for direct training. We present preliminary results applying this framework to molten BeF₂. This methodology offers a scalable and efficient pathway to model more complex systems, such as FLiBe, and to probe the behavior of minority components in molten salts. |