Evaluating the performance of equivariant neural network force fields on point defects in solids
ORAL
Abstract
Point defects are a universal feature of crystalline materials, whose identification is often addressed by combining experimental measurements with theoretical models. Recently, the defect modelling community has made steps toward adopting machine learning approaches for defect simulations, to avoid the limits imposed by costly DFT supercell relaxations.1 However, many challenges remain to be tackled before these ML methods can be adopted by the wider defects community, such as energy accuracies beyond coarse screening and the ability to predict forces (i.e. interatomic potentials). Indeed, many of the limitations of state-of-the-art ML force-fields (MLFFs) for modelling defects, and their underlying root causes, are not yet well established.
In this work, we evaluate the performance of state-of-the-art equivariant graph neural network models2,3 on a variety of relevant defect case studies, including defects of the same type in a disordered host structure, and defects of the same type across many different compositions (using a large dataset of oxygen vacancy supercell relaxations4 across hundreds of metal oxide compounds, initialised using the ShakeNBreak defect structure-searching approach5,6). We identify key limitations in the standard architectures of ML potentials for the investigation of localised perturbations (e.g. defects, polarons, interfaces) within large systems, along with strategies to reduce these shortcomings.
1. M. D. Witman, A. Goyal, T. Ogitsu, A. H. McDaniel and S. Lany, Nat Comput Sci, 2023, 3, 675–686.
2. S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth, N. Molinari, T. E. Smidt and B. Kozinsky, Nat Commun, 2022, 13, 2453.
3. A. Musaelian, S. Batzner, A. Johansson, L. Sun, C. J. Owen, M. Kornbluth and B. Kozinsky, Nat Commun, 2023, 14, 579.
4. Y. Kumagai, N. Tsunoda, A. Takahashi and F. Oba, Phys. Rev. Materials, 2021, 5, 123803.
5. I. Mosquera-Lois, S. R. Kavanagh, A. Walsh and D. O. Scanlon, Journal of Open Source Software, 2022, 7, 4817.
6. I. Mosquera-Lois, S. R. Kavanagh, A. Walsh and D. O. Scanlon, npj Comput Mater, 2023, 9, 1–11.
In this work, we evaluate the performance of state-of-the-art equivariant graph neural network models2,3 on a variety of relevant defect case studies, including defects of the same type in a disordered host structure, and defects of the same type across many different compositions (using a large dataset of oxygen vacancy supercell relaxations4 across hundreds of metal oxide compounds, initialised using the ShakeNBreak defect structure-searching approach5,6). We identify key limitations in the standard architectures of ML potentials for the investigation of localised perturbations (e.g. defects, polarons, interfaces) within large systems, along with strategies to reduce these shortcomings.
1. M. D. Witman, A. Goyal, T. Ogitsu, A. H. McDaniel and S. Lany, Nat Comput Sci, 2023, 3, 675–686.
2. S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth, N. Molinari, T. E. Smidt and B. Kozinsky, Nat Commun, 2022, 13, 2453.
3. A. Musaelian, S. Batzner, A. Johansson, L. Sun, C. J. Owen, M. Kornbluth and B. Kozinsky, Nat Commun, 2023, 14, 579.
4. Y. Kumagai, N. Tsunoda, A. Takahashi and F. Oba, Phys. Rev. Materials, 2021, 5, 123803.
5. I. Mosquera-Lois, S. R. Kavanagh, A. Walsh and D. O. Scanlon, Journal of Open Source Software, 2022, 7, 4817.
6. I. Mosquera-Lois, S. R. Kavanagh, A. Walsh and D. O. Scanlon, npj Comput Mater, 2023, 9, 1–11.
–
Presenters
-
Seán R Kavanagh
Harvard University
Authors
-
Seán R Kavanagh
Harvard University
-
Boris Kozinsky
Harvard University