Speaker
Description
Machine Learning Interatomic Potentials (MLIAPs) mix the accuracy of ab initio methods such as Density Functional Theory (DFT), with the scalability of classical interatomic potentials. Fusion structural materials, employed in the construction of future fusion reactors, experiment high neutron radiation doses. This radiation disturbs the crystal structure, generating defects in the lattice. To better understand the formation and dynamics of these defects in the crystal, we present a MLIAP based on the equivariant neural network MACE [1]. Using this potential, we are able to characterize the formation and binding energies of the most common defects in an iron crystal using a reduced database with only some simple defects. The trained potential accurately reproduces the dumbbell defect energy drop in the <110> direction, characteristic of ferromagnetic bcc iron. We also include a top layer with a Ziegler-Biersack-Littmark (ZBL) screened nuclear repulsion potential in order to perform Primary Knock-on Atom simulations (PKA), giving the potential the capability of study neutron collision cascades.
References
[1] I. Batatia, D. P. Kovacs, G. N. C. Simm, C. Ortner and G. Csanyi. MACE: Higher order equivariant message passing neural networks for fast and accurate force fields, Advances in Neural Information Processing Systems, (2022)
[2] Barouh, Caroline, et al. "Interaction between vacancies and interstitial solutes (C, N, and O) in α− Fe: From electronic structure to thermodynamics." Physical Review B 90.5 (2014)
[3] Fu, Chu-Chun, François Willaime, and Pablo Ordejón. "Stability and mobility of mono-and di-interstitials in α-Fe." Physical review letters 92.17 (2004)
abstract | Talk |
---|