Abstract

Quantum Mechanical/ Molecular Modelling (QM/MM) is a method that has been used historically to great success to accelerate atomistic simulations, with applications including fracture and dislocation motion. This technique relies on the idea that a system of interest can be partitioned into a local region which needs to be modelled accurately (QM) and a remainder which can be modelled cheaply (MM), with a coupling scheme between the two [1]. This idea naturally extends to modern machine-learned interatomic potentials (MLIPs) – where expensive and cheap models can replace the QM and MM regions respectively. A benefit of using these flexible potentials is that with the correct constraints during fitting, properties like elastic constants can be perfectly matched between regions. This talk will introduce ML-MIX, a LAMMPS package created to track defects and mix arbitrary potentials in an easy to use and efficient way. This method is demonstrated through case-studies on atomic cluster expansion (ACE) and ultra-fast (UF3) potentials [2,3] where dynamic quantities of interest are reproduced. [1] Kermode, J., Albaret, T., Sherman, D., Bernstein, N., Gumbsch, P., Payne, M. C., Csányi, G. and De Vita, A. (2008), Low-speed fracture instabilities in a brittle crystal. Nature 455, 1224–1227. [2] Drautz, R. Atomic Cluster Expansion for Accurate and Transferable Interatomic Potentials. Phys. Rev. B 2019 , 99 (1), 014104. [3] Xie, S.R., Rupp, M. & Hennig, R.G. Ultra-fast interpretable machine-learning potentials. npj Comput Mater 9, 162 (2023).