Title: Database creation using Python and Machine‑Learning Interatomic Potential (MACE) Training Using Publicly Available VASP Single‑Point Data

This repository provides a structured, reproducible workflow for creating structures doing single-point training a MACE machine‑learning interatomic potential (MLIP) for crystalline silicon. The project integrates:

• Publicly available Si structures and VASP single‑point outputs
• A clean Python‑based analysis pipeline
• MACE training configuration and evaluation
• Documentation suitable for academic and research use

The goal of this repository is to demonstrate a complete MLIP training workflow without requiring HPC access, by leveraging openly shared datasets and focusing on reproducible analysis, training, and documentation. NOTE: Due to limited computational resources , this project uses publicly released Si structures and VASP single‑point outputs from the following open repository: If you have HPC access we recommend creating rattled structures yourself

TFL600 — Claude Code Tutorial: MLIP Training Dataset Generation with VASP on ARCHER2
GitHub: https://github.com/TFL600/claude-code-tutorial_MLIP-training

The dataset includes:

• 50 rattled + strained Si structures
• Corresponding VASP single‑point outputs
• Parsed extxyz files suitable for MLIP training

All credit for dataset generation, VASP calculations, and workflow design belongs to the original author.

---

2. Ethical and Legal Use of Data

This repository does not redistribute any proprietary VASP files, including POTCAR.
Only publicly available, non‑licensed files from the referenced repository are used.

The dataset is used strictly for:

• Educational purposes
• Benchmarking
• Demonstrating MLIP training workflows
  Proper attribution is provided, and no claim is made that the dataset or VASP calculations were generated by this project.

---

4. Workflow Overview

5. Import publicly available Si structures

6. Inspect and validate dataset

7. Prepare MACE training configuration

8. Train MLIP model on local GPU/CPU resources

9. Evaluate model performance

10. Document workflow and results

Your original contributions include:

• Python analysis scripts
• MACE training configuration
• Documentation and workflow explanation
• Visualization and evaluation notebooks
• Reproducible project structure

5. Dataset Attribution

This project uses publicly available Si structures and VASP single‑point outputs from:

TFL600 — Claude Code Tutorial: MLIP Training Dataset Generation with VASP on ARCHER2
GitHub: https://github.com/TFL600/claude-code-tutorial_MLIP-training

Dataset components used:

• structures_Si.xyz
• train.xyz, val.xyz, test.xyz
• Workflow concepts for rattling, straining, and single‑point parsing

All dataset generation and VASP calculations were performed by the original author.

---

6. License

This repository uses the MIT License, covering only the original code and documentation written here.
External datasets remain under their respective licenses.

For questions or academic collaboration:

Irslan Ullah Ashraf
Computational Physicist — Atomistic Simulation & MLIPs
GitHub: github.com/ullaharslan
Email: ullaharslan@gmail.com