Data for the study: "Force and stress calculations with a neural-network wave function for solids"

This dataset contains the raw data and plotting code used in the study titled: "Force and stress calculations with a neural-network wave function for solids". The relevant code is available at https://github.com/bytedance/netobs. Data Files Directory structure deepsolid-force/: Data files for calculating forces. Graphene/{X} contains training curves and forces by different estimators for 1x1x1 graphene cell, with one carbon atom stretched X angstroms. LiH_TRI/{X} contains forces by different estimators for 1x1x1 LiH cell, with the hydrogen atom displaced along the diagonal line for X Bohr. LiH_NU/{X} is similar to LiH_TRI/{X} but uses the NU feature instead of the TRI feature mentioned in the paper. fs means the function f is using the triangular version and f0 means it's using the original version. Same for function g. deepsolid-stress/: Data files for calculating stress LiH/{X} contains pressure and stress data for 1x1x1 LiH cell with lattice constant equals to X angstrom. File format Raw force/stress data are stored in files named netobs_ckpt_xxxxxx.npz. Forces are stored in digest/force fields of the npz file and stresses are stored in values/value field. The training curves are stored in train_stats.csv files. Code for Plotting Figures prep_data.py: Script to process data and create CSVs in digest/. The easiest way to run it is using command uv run prep_data.py plot.ipynb: Jupyter notebook for generating the plots. The easiest way to run it is using command uvx --with numpy,pandas,lmfit,matplotlib,git+https://github.com/AllanChain/acplot jupyter lab plot.ipynb