IMX Colloquium - Modelling Materials Failure Processes at the Atomistic and Electronic Structure Scales with Scientific Machine Learning

I will describe recent, rapid progress in the development and application of machine learning interatomic potentials (MLIPs) to 'chemomechanical' problems in structural materials that simultaneously require accurate local chemistry and long-range stress fields - e.g. fracture and plasticity. Examples will include concurrent coupling of quantum mechanics and MLIPs  [1] or even the use of MLIPs standalone [2] to describe plasticity in tungsten. Combining two potentials with different accuracy/cost tradeoff choices in different parts of a large system is also possible [3]. The recent arrival of foundation MLIP models that leverage large datasets and deep learning have produced models capable of describing much of the periodic table with reasonable accuracy. I will critically assess the applicability of the MACE MP0 and MPA models [4] to chemomechanical problems, and present results from fine-tuning them to improve their (already reasonable) out-of-the-box description of these systems [5]. Finally, I will discuss the importance of robust uncertainty estimates when using these surrogate models and report recent efforts in this direction [6].
 
[1] P. Grigorev, A. M. Goryaeva, M.-C. Marinica, J. R. Kermode, and T. D. Swinburne,
Calculation of Dislocation Binding to Helium-Vacancy Defects in Tungsten Using Hybrid Ab Initio-Machine Learning Methods, Acta Mater. 247 118734 (2023) [arXiv:2111.11262]
[2] M. Nutter, J. R. Kermode and A. P. Bartók, Kink-helium interactions in tungsten: Opposing effects of assisted nucleation and hindered migration arXiv:2406.08368 (2024)
[3] F. Birks, T. D. Swinburne and J. R. Kermode, Efficient and Accurate Spatial Mixing of Machine Learned Interatomic Potentials for Materials Science, arXiv:2502.19081 (2025)
[4] I. Batatia et al., A Foundation Model for Atomistic Materials Chemistry arXiv:2401.00096 (2024)
[5] P. Grigorev, F. Birks, T. D. Swinburne and J.R. Kermode, In Prep (2025)
[6] I. R. Best, T. J. Sullivan and J. R. Kermode, Uncertainty Quantification in Atomistic Simulations of Silicon Using Interatomic Potentials, J. Chem. Phys. 161, 064112 (2024)

Bio: Prof. James Kermode, Materials Modelling in the School of Engineering at the University of Warwick, where he is also currently serving as Research Cluster Leader for the Predictive Modelling research cluster. He also directs the EPSRC Centre for Doctoral Training in Modelling of Heterogeneous Systems (HetSys) and the Warwick Centre for Predictive Modelling (WCPM) university research centre. He develops multiscale materials modelling algorithms and the software that implements them, with a particular focus on machine learning and data-driven approaches, and on quantifying the uncertainty in the output of electronic structure and atomistic models. He is also active in applying parameter-free modelling techniques to make quantitative predictions of "chemomechanical" materials failure processes where stress and chemistry are tightly coupled, e.g. near the tip of a propagating crack, where local bond-breaking chemistry is driven by long-range stress fields.