Research data for "Intermediates of Forming Transition Metal Dichalcogenides Heterostructures Revealed by Machine Learning Simulations"

This dataset supports the paper "Intermediates of Forming Transition Metal Dichalcogenides Heterostructures Revealed by Machine Learning Simulations". Included Files: ocp_active.zip: Modified version of ocp (https://github.com/Open-Catalyst-Project/ocp) tailored for active learning applications. deployed.pth: Pre-trained model used in the experiments. chemiscopy_run.py: Script integrating the chemiscopy and nequip modules, designed for dataset visualization. new_energy.py: Modified version of the nequip module, featuring a repulsive potential function. test_datasets.extxyz & train_datasets.extxyz: The test and training datasets in extxyz format. How to use the modified version of the nequip module: To train this version of the potential function, it is recommended to use nequip<=0.5.6 (on Linux). The NequIP training files need to be updated as follows: model_builders: - new_energy.EnergyModel - StressForceOutput min_bond_len: 1.8 Then run: export PYTHONPATH=${PYTHONPATH}:$PWDnequip-train config.yml # Train the potential functionnequip-deploy build --train-dir nequipresultsdir build.pth # Deploy the trained model