Last Updated: Sept. 2, 2025

CoQuí, Correlated Quantum ínterface, is a software project designed for ab initio electronic structure beyond density functional theory (DFT). Starting with a single-particle basis set and a mean-field solution, typically DFT or Hartree-Fock, CoQui employs tensor hypercontraction (THC) decomposition to efficiently process two-electron operators. This sophisticated approach enables CoQui to achieve exceptionally low-scaling algorithms in subsequent many-body calculations.

CoQui utilizes distributed linear algebra to enable high-performance ab-initio calculations applicable to:

• Both k-point (periodic) and molecular systems
• Generic single-particle basis sets, such as Kohn-Sham (KS) orbitals, Gaussian-type orbitals, and their mixtures.

Currently, CoQui interfaces with the following backends (see examples/dft_converter for input preparation):

• Quantum ESPRESSO
• PySCF

Below are some key features of CoQui. For more detailed examples, please visit our examples page.

• THC representation for two-electron Coulomb integrals [ref1, ref2].
• Cholesky decomposition for two-electron Coulomb integrals.

• Hartree-Fock [ref]
• RPA correlation energy [ref]
• GW approximation [ref]
• Second-order exchange (SOX) diagram [ref]
• Self-consistency with quasiparticle approximation
• Self-consistency with full frequency dependence

• Maximally localized Wannier functions via Wannier90 interface
• Constrained RPA to calculate screened interactions [ref]
• Local effective low-energy Hamiltonian for further correlated calculations [ref]

• C++ compiler that supports at least C++20.
• CMake >= 3.2.0.
• MPI Library: openmpi >= 4.
• HDF5 >= 1.8.2 for checkpoint file I/O.
• BLAS Library: OpenBLAS or Intel MKL.
• LAPACK Library: OpenBLAS or Intel MKL.
• SLATE Library for distributed linear algebra.
• Boost >= 1.68.0
• FFTW >= 3.2

CoQui uses CMake to configure the build process. Follow the instructions below step-by-step, and replace the placeholders in square brackets ([]) with your local settings.

# Step 1: Clone the git repository of CoQui
git clone https://github.com/AbInitioQHub/coqui.git coqui.src

# Step 2: Create working directory for CMake to build in
mkdir -p coqui.build && cd coqui.build

# Step 3: Configure with CMake
# Replace `[YOUR_INSTALL_PREFIX]` with the directory where you want CoQui installed.
# Replace `[NCORES]` with the number of cores you want to use for the test processes.
# Replace `[SLATE_INSTALL_PATH]` with your SLATE installation path. 
# Add `COQUI_PYTHON_SUPPORT=ON
export slate_ROOT=[SLATE_INSTALL_PATH]
cmake \
        -DCMAKE_INSTALL_PREFIX=[YOUR_INSTALL_PREFIX] \
        -DCTEST_NPROC=[NCORES] \
        -DCOQUI_PYTHON_SUPPORT=ON \ # Optional: enable Python bindings
        ../coqui.src

# Step 4: Build, test and install
# Replace `[NCORES_MAKE] with the number of cores you want to use for the build processes. 
# The ctests will be executed in parallel using `[NCORES]` processors.
make -j[NCORES_MAKE] && ctest && make install

# Verify: the 'coqui' executable should be in [YOUR_INSTALL_PREFIX]/bin
ls -l [YOUR_INSTALL_PREFIX]/bin/coqui

# Step 5: Set CoQui environment 
# You would need to source this in every new shell, or add 
# this line to your ~/.bashrc or ~/.zshrc to make it persistent.
source [YOUR_INSTALL_PREFIX]/share/coqui/coqui_env.sh

• Quick start: See the step-by-step notebooks in the coqui tutorial.
• Reference inputs: Browse runnable cases in examples.

If you use CoQui in your research, please consider supporting our developers by citing the following papers:

[1] C.-N. Yeh, M. Morales, Low-Scaling Algorithm for the Random Phase Approximation Using Tensor Hypercontraction with k-point Sampling, J. Chem. Theory Comput. 19, 18, 6197–6207 (2023).

[2] C.-N. Yeh, M. Morales, Low-Scaling Algorithms for GW and Constrained Random Phase Approximation Using Symmetry-Adapted Interpolative Separable Density Fitting, J. Chem. Theory Comput. 20, 8, 3184–3198 (2024).