About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Materials and Chemistry for Molten Salt Systems
|
| Presentation Title |
Decoding the Pair Distribution Function of Molten Uranium Chloride from X Ray Absorption Spectra by Machine Learning |
| Author(s) |
Chamod K P Mallawathantrige, Kaifeng Zheng, Shubhojit Banerjee, Vyacheslav Bryantsev, Anna L. Smith, Anatoly Frenkel |
| On-Site Speaker (Planned) |
Chamod K P Mallawathantrige |
| Abstract Scope |
Understanding local structure of disordered materials is required for studying their properties. Behavior of actinides in molten salts depends on temperature-dependent coordination environment. Most structural techniques, however, are unable to extract the pair distribution function of nearest neighbors in molten salts under in-situ conditions, where bonding disorder is asymmetric. We obtain local structure of molten UCl3 using an “objective neural network-EXAFS” (ONNE) method. The method inverts EXAFS spectra to the pair distribution function (g(r)) of nearest neighbors to U. We construct a theoretical dataset to train a convolutional neural network to predict g(r) from the experimental spectra. We validate the method by predicting the g(r) based on theoretical EXAFS spectra averaged over MD trajectory. Close agreement between the ONNE prediction and MD results for U-Cl coordination numbers and distances was obtained. These findings demonstrate that ONNE provides a reliable route to quantitative local structure determination in disordered actinide melts. |
| Proceedings Inclusion? |
Planned: |