Skip to content

ashwani12-cmd/BPN-MLIP-Dataset-and-Models

Repository files navigation

BPN-MLIP-Dataset-and-Models

This repository contains the datasets, trained machine-learned interatomic potential, and analysis workflows used in the study of anisotropic fracture, elastic properties, and thermal transport in the biphenylene network (BPN).


Repository Structure

  • full_data/
    Complete set of DFT-generated atomic configurations used during dataset construction.

  • train/
    Curated training datasets used for SNAP model fitting, including strained, shear-deformed, and perturbed configurations.

  • test/
    Independent test datasets used for validation of the trained SNAP potential.

  • models/
    Trained SNAP regression models obtained after weighted optimization.

  • potential_files/
    SNAP parameter and coefficient files used directly in LAMMPS simulations.

  • elastic_constant/
    Input files, scripts, and simulation data used for calculating elastic constants from molecular dynamics simulations using the trained SNAP potential.

  • crack/
    Atomic configurations containing pre-defined cracks aligned along the X and Y directions in the biphenylene network (BPN). The folder is organized into X/ and Y/ subdirectories, each containing LAMMPS trajectory files with crack geometries used to study direction-dependent crack stability and fracture response using the trained SNAP potential.

  • scripts/
    Python and Jupyter notebooks used for training, optimization, and post-processing of the SNAP potential. This includes workflows for weighted fitting using DFT energies, forces, and stresses, evaluation of second-order force constants and elastic constants, computation of mean squared errors (MSE) between SNAP and DFT results, and generation of validation plots.

  • qe_input_file_BPN_unitcell/
    Quantum ESPRESSO input files for DFT calculations of the BPN unit cell used to generate reference energies, forces, stresses, and elastic properties.


Scripts Description

  • scripts/trainning.ipynb
    Implements the training and optimization workflow for the SNAP interatomic potential using diverse atomic configurations (uniaxial strain along X and Y, shear strain, and random perturbations). Weighted fitting is performed using DFT energies, forces, and stresses. The notebook also computes second-order force constants and elastic constants using SNAP and evaluates MSE between SNAP and DFT, which is combined with energy, force, and stress MSEs to assess overall potential quality.

  • scripts/plot_graph_after_optimization.ipynb
    Performs post-optimization analysis and validation of the trained SNAP potential. This includes comparison of SNAP and DFT results for energies, forces, stresses, second-order force constants, and elastic constants, along with graphical visualization and summary plots.


Usage Notes

This repository is intended for academic and research purposes. Users are advised to validate the workflows before applying them to other material systems.


Citation

If you use this repository or any part of the code or data, please cite the associated publication or acknowledge this GitHub repository. Citation details will be updated once the paper is published.


Disclaimer

All scripts and data are provided as-is and were developed for specific simulation setups and material systems. No guarantee is made regarding direct applicability to other systems without modification.

About

This repository contains the datasets, trained machine-learned interatomic potential, and analysis files used in the study of anisotropic fracture and thermal transport in the biphenylene network (BPN). It includes the full set of DFT-generated configurations, curated training and test datasets, and optimized SNAP model files.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages