TorchPhysics

Solution of the heat equation on a time dependent domain, solved using TorchPhysics

Welcome to TorchPhysics, a Python library of deep learning methods for solving differential equations. Currently, TorchPhysics implements methods like PINN [1] and DeepRitz [2] which enable the user to

  • solve ordinary and partial differential equations

  • train a neural network to approximate solutions for different parameters

  • solve inverse problems and interpolate external data via the above methods

TorchPhysics can also be used in other deep learning approaches for differential equations since it is built in a modular way. For example, TorchPhysics offers a way to sample points in arbitrary, easy-to-define, domains flexibly.

Guide

All kind of information (features, installation, etc.) to TorchPhysics can be found under the Overview tab.

As an introduction to TorchPhysics a Tutorial exists. There we will present and explain the most important aspects and structure of this library. Under the Examples tab additional applications, in form of Jupyter Notebooks, can be found.

API Reference

Information for all classes, functions and methods can be found in the following documentation:

Bibliography

Indices and tables