PolymorphGen-MLPKD Framework

A comprehensive computational framework for polymorphic material structure generation, machine learning potential (MLP) training, and knowledge distillation (KD) for efficient property prediction.

Overview

This framework integrates multiple computational methods to:

• Analyze atomic configurations through entropy-symmetry landscape for thermodynamic dimensionality reduction
• Generate diverse atomic configurations through genetic algorithms in entropy-symmetry space
• Implement knowledge distillation from complex MPNN models to efficient DNN models
• Automate DFT task distribution and dataset generation

Features

• Polymorph Structure Analysis: Physical dimensionality reduction of thermodynamic configurations using entropy-symmetry landscape
• Polymorph Structure Generation: Genetic algorithm-based mutation in entropy-symmetry landscape
• Multi-scale Sampling: Combines ML-AIMD with targeted genetic mutations
• Automated DFT Workflow: High-throughput calculations with Auto-DFT platform
• Machine Learning Potential: Integration of MPNN and DNN for accurate force fields
• Knowledge Distillation: Transfer learning from MPNN to DNN models

Prerequisites

• Python 3.7 or higher
• NumPy
• SciPy
• Matplotlib
• ASE (Atomic Simulation Environment)
• DeepMD-kit v3.0.0
• DeepMD-GNN