Hybrid ab initio-machine learning simulations of dislocations

LML_retrain is an advanced coupling scheme to embed small DFT simulations in large-scale MD. To enable this embedding, we retrain (make small parameter adjustments to) linear machine learning potentials, giving seamless coupling between DFT and MD, to significantly extend the scope of hybrid simulation methods. He segregration to screw dislocation in W The method is described in detail in a recent publication: Calculation of dislocation binding to helium-vacancy defects in tungsten using hybrid ab initio-machine learning methods Acta Materialia, 2023: https://doi.org/10.1016/j.actamat.2023.118734 ArXiv preprint: https://arxiv.org/abs/2111.11262 Project funded by ANR JCJC MeMoPas: https://anr.fr/Project-ANR-19-CE46-0006 Team: Petr Grigorev (https://pgrigorev.github.io) and TD Swinburne (PI, https://tomswinburne.github.io), CNRS and CINaM Marseille Requirements Python packages can be easily installed via pip or other package managers; see e.g. anaconda Running retraining scripts requires numpy and scipy Evaluating atomic structures and generating descriptor vectors requires Atomic Simulation Environment on top of the LAMMPS python module. LAMMPS should be compiled with ML-SNAP package which is not included by default, i.e. with a traditional make for LAMMPS cd /path/to/lammps/src make yes-ML-SNAP make yes-[OTHER PACKAGES] make mpi mode=shlib Script for calculating dislocation glide barriers depends on matscipy dislocation module.