Installation

Requirements

The following software libraries and tools are required to compile pCI:

A Fortran compiler: GNU Fortran gfortran v12.2+, Intel Fortran Classic ifort v2020u4+, or LLVM-based Intel Fortran ifx
CMake v3.13+
Python v3.9+
LAPACK and BLAS, or Intel Math Kernel Library (MKL) (optional — required for pconf, ine, and pol)
OpenMPI v4.1+ (optional — required only for MPI-parallel programs)

Older versions may work but the listed versions have been tested.

On HPC clusters

If working on an HPC cluster, environment modules may be available using module load <package> for each required library. Please check your cluster’s documentation for more information.

Obtaining the source code

Users can download the latest version of the pCI code package from our GitHub repository.

Via direct download

From the GitHub repository, click on the green <> Code button towards the top, then click Download ZIP. You can then unzip the downloaded file to obtain the source codes.

Via git

git clone https://github.com/ud-pci/pCI.git

Compiling with CMake

Basic build

$ cd pCI
$ mkdir build && cd build
$ FC=ifort cmake -DMPI_HOME=${OPENMPI_PREFIX} -DCMAKE_INSTALL_PREFIX=$(pwd)/../ ..
$ make
$ make install

For the LLVM-based Intel compiler, replace FC=ifort with FC=ifx.

If LAPACK/MKL is not available, pconf, ine, and pol will be skipped automatically.

If OpenMPI is not available, omit -DMPI_HOME and CMake will automatically skip the MPI-parallel programs:

$ FC=ifort cmake -DCMAKE_INSTALL_PREFIX=$(pwd)/../ ..

Build options

Debug build:

$ FC=ifort cmake -DCMAKE_BUILD_TYPE=Debug -DMPI_HOME=${OPENMPI_PREFIX} -DCMAKE_INSTALL_PREFIX=$(pwd)/../debug ..

Optimized build:

$ FC=ifort cmake -DCMAKE_BUILD_TYPE=Release -DMPI_HOME=${OPENMPI_PREFIX} -DCMAKE_INSTALL_PREFIX=$(pwd)/../opt -DCMAKE_Fortran_FLAGS_RELEASE="-g -O3 -mcmodel=large -xHost -m64" ..

Double-precision two-electron and isotope-shift integrals (increases memory usage but improves numerical accuracy):

$ FC=ifort cmake -DUSE_DP_INTEGRALS=ON -DMPI_HOME=${OPENMPI_PREFIX} -DCMAKE_INSTALL_PREFIX=$(pwd)/../ ..

ScaLAPACK diagonalization in pconf (uses distributed-memory ScaLAPACK instead of serial LAPACK for large matrix diagonalization; requires OpenMPI and LAPACK/MKL):

$ FC=ifort cmake -DUSE_SCALAPACK=ON -DMPI_HOME=${OPENMPI_PREFIX} -DCMAKE_INSTALL_PREFIX=$(pwd)/../ ..

Supplementary programs

The lib/ directory contains supplementary programs that are built separately using Make. These include the all-order (allcore-ci, valsd-ci, sdvw-ci, second-ci), RPA (rpa, rpa_dtm), and QED (qed_rot, qedpot_conf, sgc0) programs. To compile and install all of them to the bin/ directory:

$ cd lib
$ make install

Individual packages can also be built separately:

$ make rpa        # builds rpa and rpa_dtm
$ make qed        # builds qed_rot, qedpot_conf, sgc0
$ make all-order  # builds allcore-ci, valsd-ci, sdvw-ci, second-ci, and others

Troubleshooting

-DMPI_HOME not working: Specifying the executables directly is more robust:

$ FC=ifort cmake \
    -DMPI_Fortran_COMPILER=${OPENMPI_PREFIX}/bin/mpifort \
    -DMPIEXEC_EXECUTABLE=${OPENMPI_PREFIX}/bin/mpiexec \
    -DCMAKE_INSTALL_PREFIX=$(pwd)/../ \
    ..

Intel MPI conflict: If Intel MPI is also installed (e.g. via Intel oneAPI), CMake may detect it instead of OpenMPI. Point CMake to the correct executables explicitly:

$ FC=ifort cmake \
    -DMPI_Fortran_COMPILER=/path/to/openmpi/bin/mpifort \
    -DMPIEXEC_EXECUTABLE=/path/to/openmpi/bin/mpiexec \
    -DCMAKE_INSTALL_PREFIX=$(pwd)/../ \
    ..

If CMake still detects Intel MPI after correcting the paths, clear the CMake cache before reconfiguring:

$ rm -rf build/*