XPINN Toolkit

Overview

The XPINN Toolkit is a minimal framework for implementing eXtended Physics-Informed Neural Networks (XPINNs). XPINNs allow the decomposition of complex physical domains into subdomains, enabling parallel and localized learning of partial differential equations (PDEs). This toolkit is designed to provide a lightweight and modular foundation for XPINN research and experimentation.

Features

• Modular design supporting XPINN components:
  • Subdomains (SubDomain)
  • Sub-networks (SubNet)
  • Core PINN/XPINN models
• Utility functions for model management and PDE handling
• Test suite and pretrained model checkpoints
• Example usage for quick experimentation

Project Structure

xpinn-toolkit/
├── poetry.lock                 # Poetry lock file
├── pyproject.toml             # Project metadata and dependencies
├── README.md                  # Project documentation
├── src/
│   └── xpinntoolkit/
│       ├── __init__.py
│       ├── examples/          # Example setups and training scripts
│       ├── models/            # Core XPINN components
│       │   ├── PINN.py
│       │   ├── XPINN.py
│       │   ├── SubNet.py
│       │   ├── SubDomain.py
│       │   ├── utils.py
│       │   └── __init__.py
│       └── __pycache__/
├── tests/                     # Tests and training outputs
│   ├── compare_XPINN_PINN.py
│   ├── train_XPINN_PINN.py
│   ├── xpinn_sequential_model.pth
│   ├── xpinn_sequential_step_*.pth
│   ├── sequential_training_progress.png
│   └── __init__.py

Installation

This project uses Poetry for dependency management.

To install the project in editable mode:

poetry install

To activate the virtual environment:

poetry shell

Usage

Examples demonstrating how to define domains, set up sub-networks, and train XPINNs are located in:

src/xpinntoolkit/examples/

You can also explore the tests/ folder to see training comparisons and saved models from sequential runs.