b'Machine learningModeling that integrates quantum mechanics and machine interatomic potentialslearning reveals irradiation-induced atomic scale defect clusters for radiation damage andthat are invisible to transmission electron microscopy.physical properties inW hen subjected to the irradiation of high-energy particles, point defects and their atomic scale clusters will be generated in crystalline model fluorite systems materials in large amounts. Although these small defects are largely hidden under high-resolution transmission electron microscopy, they can have profound influence on material properties. For nuclear fuels in particular, their thermal conductivities can be significantly degraded due to the scattering of phonons by atomic scale defects. Using fluorite-structured thorium dioxide as an exemplar, this project demonstrated the synergistic combination of density functional theory calculations and machine learning interatomic potential as a powerful tool that enables exhaustive PROJECT NUMBER:exploration of the large configuration spaces of small point defect clusters. Unraveling 21A1050-078FP the most energetically favorable ground state configurations of these defects is an TOTAL APPROVED AMOUNT:important step toward establishing the quantitative structure-property relationship for $1,000,000 over 3 years irradiated materials. By learning directly from large density functional theory data sets without assuming any fixed functional forms, a machine learning interatomic potential PRINCIPAL INVESTIGATOR:can reproduce density functional theory potential energy surfaces far more accurately Chao Jiang than an empirical potential. This increased level of fidelity is crucial because empirical CO-INVESTIGATORS: potential-driven ground state searches have been known to yield results that do not Koushik Araseethota Manjunatha, INL agree with density functional theory. Additionally, a machine learning interatomic Miaomiao Jin, INL potential is many orders of magnitude faster than density functional theory since it Tiankai Yao, INL does not treat electrons explicitly, which makes ground state searches computationally Zilong Hua, INL efficient. The combined density functional theory plus machine learning interatomic potential approach led to several unexpected discoveries, including ground state COLLABORATOR: polymorphism and ground state structures that defy physical intuition. Finally, the Columbia University atomistic configurations of small interstitial and vacancy clusters as revealed in this project were used as inputs for non-equilibrium molecular dynamics simulations to assess their impact on thermal energy transport in thorium dioxide. These results provide the physics basis for predicting the in-reactor performance of thorium dioxide as an alternative nuclear fuel.TALENT PIPELINE:Miaomiao Jin, postdoc at INLPUBLICATION:C. Jiang, C.A. Marianetti, M. Khafizov, et al. Machine learning potential assisted exploration of complex defect potential energy surfaces, npj Comput Mater 10, 21 (2024).The impact of atomic scale defects on the thermal conductivity of thorium dioxide.39'