GElib

GElib is a C++ library with a CUDA backend for implementing SO(3)-equivariant neural networks. GElib is developed by Risi Kondor and released under a combination of the Mozilla public license v.2.0 and a custom noncommercial license bundled with the source code. The library can be downloaded from https://github.com/risi-kondor/GElib.

This document provides documentation for GElib’s PyTorch interface. The PyTorch interface only provides access to a subset of the C++ library’s functionality.

GElib is currently in alpha. Some features described below may still be incomplete or untested.

Contents:

• Introduction
  • Features
  • Installation
  • Configuration
  • Troubleshooting
  • Known issues
  • Usage
• Mathematical background
  • Equivariant neural networks
  • The Clebsch–Gordan product
• SO3part
  • Spherical harmonics
  • Group action
  • Clebsch-Gordan products
  • Diagonal Clebsch-Gordan products
  • Fourier parts
  • GPU operations
• SO3vec
  • Fourier vectors
  • Clebsch-Gordan products
  • Diagonal Clebsch-Gordan products
  • Fproduct and Fmodsq
  • GPU operations
• SO3partArr
  • Gather
• SO3vecArr
  • Gather

Indices and tables

• Index

• Module Index

• Search Page