gproptim

Author:

Rohit Goswami, Maxim Masterov, Satish Kamath

Overview

gproptim is a Gaussian Process Regression library for accelerated saddle point searches on potential energy surfaces. It provides:

• Squared-exponential atom-type (SexpatCF) covariance functions designed for atomic configurations

• Joint energy and gradient observations for data-efficient GP modelling

• Student-t process for robustness to outlier observations

• SCG hyperparameter optimiser + factorized eigendecomp path

• Bayesian model averaging over hyperparameters (FBPMGP)

• Atomic dimer method for transition state searches

• GP-accelerated NEB with climbing image and OIE acquisition

• GP-accelerated geometry minimization

• Random Fourier Features for O(n) approximate GP

• Analytical test potentials (LJ, Muller-Brown, LEPS)

• Highway SIMD + OpenMP + runtime CUDA acceleration

• Python bindings via nanobind with ASE calculator integration

• C API with Fortran module for interop with legacy simulation codes

• Dual Meson + CMake build systems

Tutorial

Tutorial

• Tutorial: GP-Accelerated Saddle Point Search
• Tutorial: Fortran Interop
• MPI-Parallel Saddle Point Search

How-to Guides

How-to

• How to: Use the T-Process for Robust GP Regression
• How to: Run GP-NEB with OIE Acquisition
• How-to: Use the C API from Fortran

User Guide

Guide

• Installation
• EESSI overlay vs. the default pixi environment
• Usage
• Performance Optimisation
• RPC and Serialization

Developer Reference

Reference

• Contributing
• Changelog

Python API Reference

Python API

• API Reference
  • gpr_optim

C++ API Reference

C++ API

• C++ API Reference

License

MIT. See the LICENSE file in the repository root.