Machine Learning Interatomic Potentials (MTP) Pipeline

Project Overview

This repository contains a complete workflow for developing Moment Tensor Potentials (MTP)—a class of machine learning interatomic potentials—for Tungsten (W) and Gallium Arsenide (GaAs).

The goal of this project is to bridge the gap between the high accuracy of ab initio Density Functional Theory (DFT) and the computational scalability of classical Molecular Dynamics (MD). By training MTP models on quantum mechanical data, we enable large-scale atomistic simulations with near-quantum accuracy at a fraction of the cost.

Key Features

• Data Processing Pipeline: Custom Python scripts to parse raw Quantum Espresso logs into ML-ready datasets
• BOMD Data Generation: Workflow for generating training data using Born-Oppenheimer Molecular Dynamics
• Model Training: MTP training using the mlip-3 package
• Validation: Integration with LAMMPS to calculate elastic properties

Repository Structure

mlip/
│
├── run_preprocessing.bat            # Batch script to automate the full data processing pipeline
├── requirements.txt                 # Python dependencies
├── README.md                        # Project documentation
│
├── scripts/                         # data processing scripts
│   ├── out2csv.py                   # Parses .out files to structured CSV
│   └── csv2cfg.py                   # Converts CSV to MLIP .cfg training format
│
├── training_config/                 # MTP hyperparameters
│   └── mtp_init.almtp               # Initialization settings for potential training
│
├── validation/                      # Testing scripts
│   └── lammps_elastic.in            # LAMMPS input for calculating elastic constants
│
└── data_samples/                    # Sample data for testing the pipeline
    ├── qe_md.in                     # Sample Quantum Espresso Input (W Supercell)
    ├── qe_md.out                    # Sample Raw Output (Truncated)
    ├── extracted.csv                # Intermediate parsed data
    └── sample_training_set.cfg      # Final formatted training data

Data Collection & Results

Data Generation (BOMD)

Training data was generated using Quantum Espresso.

• System: Tungsten ($3\times3\times3$ Supercell) and GaAs ($3\times3\times3$ Supercell)
• Simulation Type: vc-md (Variable Cell Molecular Dynamics)
• Conditions: Temperatures of 500K, 1500K, and 2500K; Strains of $\pm5%$ and $\pm10%$ (Tensile/Compressive).

Validation Results

The potentials were validated by simulating elastic constants in LAMMPS and comparing them to experimental benchmarks.

Material	Property	Simulation (MTP)	Actual (Ref)
Tungsten	Bulk Modulus	363.84 GPa	310 GPa
Tungsten	Poisson's Ratio	0.20	0.28
GaAs	Bulk Modulus	84.17 GPa	75.5 GPa
GaAs	Poisson's Ratio	0.32	0.31

🚀 Usage Instructions

1. Prerequisites

Ensure you have Python installed. It is recommended to use a virtual environment.

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Note: This project also requires the MLIP Package and LAMMPS with the MLIP interface installed.

2. Data Preprocessing (Python)

We provide scripts to convert raw DFT logs into the .cfg format required by MTP.

A. Run Individual Scripts To process a single file manually:

# Step 1: Parse QE output to CSV (Extracts Forces, Stress, Energy)
python scripts/out2csv.py --input data_samples/qe_md.out --output data_samples/extracted.csv

# Step 2: Format CSV to MTP Configuration (.cfg)
python scripts/csv2cfg.py --input data_samples/extracted.csv --output data_samples/training_set.cfg --mode w

B. Automated Batch Processing Use the included batch file to process multiple .out files automatically. This script iterates through a folder of raw logs, parses them, and merges them into a single training set.

./run_preprocessing.bat

3. Model Training

Train the Moment Tensor Potential using the processed configuration file and the initialization parameters.

# Train for 100 iterations (example)
mlp train training_config/mtp_init.almtp data_samples/sample_training_set.cfg --save_to=trained_potential.mtp

4. Validation (LAMMPS)

Deploy the trained potential to calculate material properties.

# Run elastic constant calculation
lammps -in validation/lammps_elastic.in