pyscf/gpu4pyscf

What it solves

It provides GPU acceleration for PySCF, a Python-based framework for simulations of chemistry. This allows computationally expensive quantum chemistry calculations—such as Hartree-Fock and Density Functional Theory (DFT)—to run significantly faster by offloading the heavy mathematical workloads to NVIDIA GPUs.

How it works

It acts as a plugin for PySCF, allowing users to move PySCF objects to the GPU using a .to_gpu() method. It leverages CUDA-enabled hardware and libraries like CuPy and cuTENSOR to accelerate tensor contractions and other linear algebra operations essential for electronic structure calculations.

Who it’s for

Computational chemists and researchers performing electronic structure simulations who need to reduce the time required for energy calculations, analytical gradients, and Hessian matrices.

Highlights

• Broad Functional Support: Supports LDA, GGA, mGGA, hybrid, and range-separated functionals via libXC.
• Comprehensive Chemistry Tools: Includes support for analytical gradients, Hessians, geometry optimization, and transition state searches.
• Advanced Models: Implements PCM and SMD solvent models, as well as dispersion corrections (DFTD3/DFTD4).
• High Performance: Demonstrates significant speedups over CPU-based alternatives for various molecular systems.