co2-potential 0.6.4

A Python package interfacing with the CO2CO2 shared library.

CO₂–CO₂ Potential Energy Library

High-performance C++ routines for computing CO₂ dimer potentials (1-body, 2-body, SAPT-S), exposed to Python via a ctypes wrapper.

Repository Layout

CO2CO2_AUTODIFF/                 # Project root
├── src/                         # All C++ source + Makefile
│   ├── mbCO2CO2.cpp             # C-API exports (energies, grads, Hessians)
│   ├── mbCO2CO2.h               # Function declarations
│   ├── x1b.cpp/.h               # 1-body poly routines
│   ├── x2b.cpp/.h               # 2-body poly routines
│   ├── sapt-s.cpp               # SAPT-S routines
│   ├── poly-*.cpp/.h            # polynomial basis eval
│   ├── autodiff/                # bundled autodiff headers
│   ├── Eigen/                   # bundled Eigen headers
│   └── Makefile                 # builds libCO2CO2.so
├── co2_potential/               # Python package
│   ├── __init__.py
│   ├── libCO2CO2.so             # copied in by Makefile
│   ├── wrapper.py               # ctypes wrapper + convenience functions
│   └── benchmark.py             # benchmarking script
├── .github/
│   └── workflows/
│       └── wheels.yml           # CI: build wheels + upload to PyPI
├── setup.py                     # pip install / build-ext hook
├── pyproject.toml               # PEP 517 build config
├── MANIFEST.in                  # include shared lib & .py files
├── README.md
└── LICENSE

Installation

Linux (x86_64) and macOS (Apple Silicon)

pip install co2_potential

macOS (Intel x86_64)

Pre-built wheels are not available for Intel Macs. Install from source:

pip install co2_potential --no-binary co2_potential

This requires a C++ compiler (clang/g++) and may take some time to compile.

Building from Source

git clone https://github.com/sodelab/CO2CO2_AUTODIFF.git
cd CO2CO2_AUTODIFF
cd src && make clean && make && cd ..
pip install -e .

On macOS, to build a universal binary (x86_64 + arm64):

cd src && make clean && make UNIVERSAL=1 && cd ..

Usage

import numpy as np
from co2_potential.wrapper import p1b, p2b_4, p2b_5, sapt

# Example: 6-atom dimer → 18 coordinates
xyz = np.array([
    0.0,   0.0,  0.000,    # monomer A: C, O, O
    0.0,   0.0, -1.162,
    0.0,   0.0,  1.162,
    3.75,  0.0,  0.000,    # monomer B: C, O, O
    3.75,  0.0, -1.162,
    3.75,  0.0,  1.162
], dtype=np.double)

E1 = p1b(xyz)        # 1-body energy
E2_4 = p2b_4(xyz)    # 2-body 4th-order
E2_5 = p2b_5(xyz)    # 2-body 5th-order
Es = sapt(xyz)       # SAPT-S energy

print(f"E1 = {E1:.6f}, E2(5) = {E2_5:.6f}, SAPT = {Es:.6f}")

Functionality

The library provides the following Python functions via co2_potential.wrapper:

Dimension and Version Getters

• get_p1b_dim(): Returns the number of coordinates for monomer (should be 9).
• get_p2b_dim(): Returns the number of coordinates for dimer (should be 18).
• get_p2b_4_dim(): Returns the number of coordinates for 2-body 4th-order.
• get_p2b_5_dim(): Returns the number of coordinates for 2-body 5th-order.
• get_sapt_dim(): Returns the number of coordinates for SAPT-S.
• get_version(): Returns the version string of the underlying C++ library.

1B (Monomer) Functions

• p1b(xyz): Monomer energy. xyz is a numpy array of shape (9,).
• p1b_gradient(xyz): Monomer gradient. Returns numpy array of shape (9,).
• p1b_hessian_rev(xyz): Monomer Hessian (reverse-mode autodiff). Returns numpy array of shape (9, 9).
• p1b_hessian_fwd(xyz): Monomer Hessian (forward-mode autodiff). Returns numpy array of shape (9, 9).

2B (Dimer) Functions – 4th Order

• p2b_4(xyz): Dimer 2-body 4th-order energy. xyz is a numpy array of shape (18,).
• p2b_gradient_4(xyz): Dimer 2-body 4th-order gradient. Returns numpy array of shape (18,).
• p2b_hessian_4_rev(xyz): Dimer 2-body 4th-order Hessian (reverse-mode). Returns numpy array of shape (18, 18).
• p2b_hessian_4_fwd(xyz): Dimer 2-body 4th-order Hessian (forward-mode). Returns numpy array of shape (18, 18).

2B (Dimer) Functions – 5th Order

• p2b_5(xyz): Dimer 2-body 5th-order energy. xyz is a numpy array of shape (18,).
• p2b_gradient_5(xyz): Dimer 2-body 5th-order gradient. Returns numpy array of shape (18,).
• p2b_hessian_5_rev(xyz): Dimer 2-body 5th-order Hessian (reverse-mode). Returns numpy array of shape (18, 18).
• p2b_hessian_5_fwd(xyz): Dimer 2-body 5th-order Hessian (forward-mode). Returns numpy array of shape (18, 18).

SAPT-S Dimer Functions

• sapt(xyz): SAPT-S dimer energy. xyz is a numpy array of shape (18,).
• sapt_gradient(xyz): SAPT-S dimer gradient. Returns numpy array of shape (18,).
• sapt_hessian_rev(xyz): SAPT-S dimer Hessian (reverse-mode). Returns numpy array of shape (18, 18).
• sapt_hessian_fwd(xyz): SAPT-S dimer Hessian (forward-mode). Returns numpy array of shape (18, 18).

The core computations are implemented in C++ for performance and exposed to Python via a ctypes wrapper.

Benchmarking

This package includes a benchmarking script which tests the accuracy and performance of the CO₂ potential functions against reference values:

python -m co2_potential.benchmark --all         # Test all (default)
python -m co2_potential.benchmark --energies    # Test only energies
python -m co2_potential.benchmark --gradients   # Test only gradients
python -m co2_potential.benchmark --hessians    # Test only hessians
python -m co2_potential.benchmark --p1b         # Test only p1b functions
python -m co2_potential.benchmark --p2b_4       # Test only p2b_4 functions
python -m co2_potential.benchmark --p2b_5       # Test only p2b_5 functions
python -m co2_potential.benchmark --sapt        # Test only sapt functions

The benchmark compares computed results to hard-coded reference values (with units: energies in kcal/mol, gradients in kcal/mol/angstrom, and Hessians in kcal/mol/angstrom²) and reports pass/fail status for each test.

Supported Platforms

Platform	Architecture	Wheel Available
Linux	x86_64	Yes
macOS	arm64 (Apple Silicon)	Yes
macOS	x86_64 (Intel)	Build from source

Contributing

Contributions are welcome! If you have suggestions for improvements or new features, please open an issue or submit a pull request.

License

This project is licensed under the MIT License. See the LICENSE file for details.