CIPSIpy

/ˈsɪpsɪpi/

A simple CIPSI (Configuration Interaction using a Perturbative Selection made Iteratively) implementation in JAX.

Disclaimer: Development of this package made extensive use of LLMs to speed up the process.

Overview

This is a minimal implementation of the CIPSI algorithm for learning purposes:

• Learn and understand CIPSI algorithm and its parallelization
• Gain practical experience with JAX, especially GPU acceleration
• Build a working quantum chemistry tool

Features

• FCIDUMP file reader for molecular integrals from PySCF
• Slater determinant operations with efficient bit manipulation
• Hamiltonian matrix element calculation (Slater-Condon rules)
• Iterative CIPSI algorithm with PT2 selection
• JAX-based implementation for GPU acceleration

Installation

pip install -e .

For development:

pip install -e ".[dev]"

Quick Start

See the examples directory for working examples:

# Run H2 minimal basis example
python examples/h2/run_h2.py

Todo List

See todo.md for project goals, and todo lists separated by project "phases".

License

MIT License - see LICENSE for details