OpenQP 1.2.0

Python bindings for the Open Quantum Platform (Fortran core + Python wrapper)

Open Quantum Platform: OpenQP

Open Quantum Platform (OpenQP) is a quantum chemical platform featuring cutting-edge capabilities like Mixed-Reference Spin-Flip (MRSF)-TDDFT with an emphasis on open-source ecosystem.

Web tools

• Try OpenQP Web - browser app for choosing example workflows, previewing structures, and downloading OpenQP input files. Calculations are prepared locally in the browser and are not executed online yet.
• OpenQP Input Generator - existing browser input generator with a 3D molecular editor.
• OpenqpView - browser-based viewer for OpenQP logs, JSON, Molden, cube, and XYZ data.

Capabilities

Electronic structure

• HF and DFT (LibXC functionals) with RHF, ROHF, and UHF references
• TDHF/TDDFT, SF-TDDFT, and MRSF-TDDFT including the DTCAM-series functionals
• UMRSF-TDDFT excitation energies from a UHF reference (energy-only)
• MRSF-EKT ionization potentials and electron affinities with Dyson orbitals and pole strengths

Properties & spectroscopy

• Analytic gradients; native analytic HF/DFT Hessians covering open-shell (UHF/ROHF) references, ECPs, and range-separated (CAM/LRC) functionals
• Vibrational frequencies, thermochemistry, and native IR and Raman intensities
• NMR chemical shieldings with CGO and GIAO (London-orbital) gauge treatments
• MRSF-TDDFT nonadiabatic couplings (NACME) for dynamics workflows
• Spin-orbit couplings (SOC) between MRSF-TDDFT states from one- and two-electron contributions (Relativistic MRSF-TDDFT)
• Implicit solvation: energy-only ddPCM continuum solvent via the ddX library

Geometry & reaction paths

• Minima, transition states, MECI/MECP, constrained optimization, IRC, and NEB
• Built-in native optimizer (lib=oqp, TRIC/DLC + restricted-step RFO), plus geomeTRIC and SciPy backends

Reliability & performance

• Point-group symmetry: detection, MO/state/mode labels, and petite-list reductions accelerating integrals, XC, gradients, and response
• DFT grids: Lebedev plus SG-0/SG-1/SG-2/SG-3 pruned grids with per-element DE2 radial quadrature; OpenMP-parallel XC kernels
• Native initial guesses (hcore, huckel, modhuckel, minao, sap), JSON restart, OpenTrustRegion (TRAH) SCF stabilization, Davidson auto-restart, and MINRES/AUTO Z-vector fallbacks
• OpenMP/MPI parallelism, ILP64 BLAS/LAPACK, pip installs, and Docker images

Ecosystem

• LibXC, basis_set_exchange, libecpint, DFT-D4, PyRAI2MD, OpenQP Web, the OpenQP Input Generator, and the browser-based OpenqpView viewer

Upcoming Features

• Efficient electrostatic embedding QM/MM by ESPF QM/MM
• Scalar-relativistic (X2C) framework extending the relativistic MRSF-TDDFT treatment

Quickstart

• pip install openqp or
• Ready to Use Docker Image of openqp/openqp or
• Building from Source Files Using the Instructions Below.

Requirements

• GCC, G++, Gfortran: Version >= 8
• CMake: Version >= 3.25
• cffi: Perform pip install cffi
• ninja (optional)
• MPI Library: OpenMPI For MPI Support. Consult detailed documentation for other MPI libraries

Download the Source Files

git clone https://github.com/Open-Quantum-Platform/openqp.git

Pip install

cd openqp
pip install .

This is the recommended source install path. It builds and installs the OpenQP Python package and native library together, so setting OPENQP_ROOT is not required for normal openqp command-line use after installation. Native initial guesses such as huckel, modhuckel, minao, and sap work without external guess-generation dependencies.

or

Detailed Compile

Two representative configurations are shown below; both build the native library and install the Python package. See Common options for the remaining build switches.

Linux (GCC + OpenMP)

cd openqp
cmake -B build -G Ninja \
  -DCMAKE_C_COMPILER=gcc \
  -DCMAKE_CXX_COMPILER=g++ \
  -DCMAKE_Fortran_COMPILER=gfortran \
  -DCMAKE_INSTALL_PREFIX=. \
  -DENABLE_OPENMP=ON
ninja -C build install
cd pyoqp
pip install .

macOS (Homebrew GCC + Accelerate BLAS)

Use the Homebrew GCC toolchain and the native Accelerate BLAS/LAPACK provider (32-bit integer interface). Adjust the -15 suffix to your installed Homebrew GCC.

cd openqp
cmake -B build -G Ninja \
  -DCMAKE_C_COMPILER=/opt/homebrew/bin/gcc-15 \
  -DCMAKE_CXX_COMPILER=/opt/homebrew/bin/g++-15 \
  -DCMAKE_Fortran_COMPILER=/opt/homebrew/bin/gfortran-15 \
  -DCMAKE_INSTALL_PREFIX=. \
  -DENABLE_OPENMP=ON \
  -DLINALG_LIB=auto \
  -DLINALG_LIB_INT64=OFF
ninja -C build install
cd pyoqp
pip install .

Common options

Option	Default	Description
-DENABLE_MPI=ON	OFF	Build with MPI; pair with an MPI Fortran wrapper, e.g. -DCMAKE_Fortran_COMPILER=mpif90.
-DENABLE_OPENMP=ON	ON	OpenMP threading.
-DUSE_LIBINT=ON	OFF	Use libint for ERIs instead of the built-in Rys-quadrature engine.
-DLINALG_LIB=<vendor>	auto	BLAS/LAPACK provider (auto, MKL, OpenBLAS, netlib, …).
-DLINALG_LIB_INT64=ON	ON	Use the default 64-bit-integer BLAS/LAPACK interface.
-DENABLE_OPENTRAH=OFF	ON	Skip the external OpenTrustRegion (OpenTRAH) solver; the TRAH SCF converger then uses the built-in native implementation only.
-DOQP_REUSE_EXTERNALS=OFF	ON	Disable the automatic reusable cache for bundled external dependency builds.

Bundled external dependencies are cached automatically across fresh OpenQP build directories. The cache uses the platform cache directory (~/Library/Caches/openqp/externals on macOS, $XDG_CACHE_HOME/openqp/externals or ~/.cache/openqp/externals elsewhere) and is keyed by operating system, architecture, CMake generator, compiler IDs/versions, compiler flags, build type, linkage mode, BLAS/LAPACK choice, BLAS/LAPACK integer ABI, and bundled external versions, so Fortran modules and libraries are reused only by compatible builds. Advanced users can relocate the cache with -DOQP_EXTERNALS_ROOT=<path>. The cache is not safe for concurrent use: do not run two fresh OpenQP builds with the same configuration at the same time, or configure one of them with -DOQP_REUSE_EXTERNALS=OFF.

LP64 BLAS/LAPACK (-DLINALG_LIB_INT64=OFF) is a macOS-only exception intended for a consistent native Accelerate stack. OpenBLAS and MKL builds should use ILP64 for normal use.

To build without Ninja, drop -G Ninja and replace ninja with make.

Test

openqp --run_tests all     # Run all tests from all folders in examples

Run

For OpenMP or sequential run:

openqp any_example_file.inp

For OpenMP and MPI run:

mpirun -np number_of_mpi openqp any_example_file.inp

Setting the OpenMP thread count

The number of OpenMP threads (per process / MPI rank) can be set directly from the input file or the command line, instead of relying on OMP_NUM_THREADS:

[input]
omp_threads=16        # OpenMP threads per process / MPI rank

openqp any_example_file.inp --omp 16

Precedence (highest first): --omp → input omp_threads → OMP_NUM_THREADS → built-in default. The value is applied before the OpenMP runtime initializes, and is honored on the programmatic Runner(input_dict=...) path as well. If OpenQP was built without OpenMP (-DENABLE_OPENMP=OFF), the request is ignored with a warning and the run is serial. MPI rank count remains launcher-controlled (mpirun -np ...); --nompi disables MPI.

Performance and threading

OpenQP parallelizes the two-electron integral / Fock build with OpenMP. A few defaults and knobs keep it efficient:

• Sequential BLAS inside the integral build (automatic). A threaded BLAS keeps its own worker pool that would oversubscribe the cores against the OpenMP integral threads. OpenQP forces BLAS to a single thread for the duration of the integral build (via the BLAS library's own runtime setter, so it cannot be re-introduced by a stray OPENBLAS_NUM_THREADS/MKL_NUM_THREADS) and restores full BLAS threading afterward for diagonalization and other BLAS-heavy phases. The frontend also sets conservative *_NUM_THREADS=1 and OMP_STACKSIZE defaults (via setdefault, so an explicit environment wins). Measured ~1.5x faster at 24-28 threads on a dedicated node; near-ideal scaling to the performance-core count.

• Low-memory Fock accumulator (automatic for large, sparse systems). The Fock build normally keeps one Fock copy per thread (fockdim x nfocks x nthreads); for large bases on many cores this can reach several GB. OpenQP can instead accumulate into a single shared Fock with atomic updates (~Nthreadsx less memory). It auto-engages only when the replicated buffers would exceed a cap and the density is sparse (so atomic contention is negligible). Controls:

  • OQP_FOCK_MEM_MB — replicated-memory cap in MB before switching (default 4096).
  • OQP_FOCK_SPARSITY — significant shell-pair fraction below which the density is considered sparse (default 0.5).
  • OQP_FOCK_ATOMIC=1/0 — force the shared-atomic / replicated mode regardless.

Detailed Documentation

For more in-depth information, visit:

• OpenQP Documentation

OpenQP Web

Prepare OpenQP inputs in the browser at app.openqp.org. The web app provides example workflows, local structure preview, downloadable .inp and .xyz files, tutorials, and links to the existing OpenQP browser tools. It does not run OpenQP jobs online, require login, or expose API keys in the browser.

Input Generator

Easily create input files for OpenQP using our Web-based Input Generator.

OpenqpView

Inspect OpenQP calculation outputs directly in the browser with OpenqpView. Recent OpenqpView development added a GitHub Pages deployment, full-periodic-table molecule rendering, local file/drop/paste loading, WebGL auto-spin controls, and support for OpenQP log, JSON, Molden, cube, and XYZ data. Files and pasted text are processed locally in the browser and are not uploaded to a server.

Citing OpenQP

If you use OpenQP in your research, please cite the following papers:

• Mironov V, Komarov K, Li J, Gerasimov I, Mazaheri M, Park W, Lashkaripour A, Oh M, Nakata H, Ishimura K, Huix-Rotllant M, Lee S, and Choi CH. "OpenQP: A Quantum Chemical Platform Featuring MRSF-TDDFT with an Emphasis on Open-source Ecosystem" Journal of Chemical Theory and Computation, 2024
• Park W, Komarov K, Lee S, and Choi CH. "Mixed-Reference Spin-Flip Time-Dependent Density Functional Theory: Multireference Advantages with the Practicality of Linear Response Theory." The Journal of Physical Chemistry Letters. 2023 Sep 28;14(39):8896-908.
• Lee S, Filatov M, Lee S, and Choi CH. "Eliminating Spin-Contamination of Spin-Flip Time-Dependent Density Functional Theory Within Linear Response Formalism by the Use of Zeroth-Order Mixed-Reference (MR) Reduced Density Matrix." The Journal of Chemical Physics, vol. 149, no. 10, 2018.
• Lee S, Kim EE, Nakata H, Lee S, and Choi CH. "Efficient Implementations of Analytic Energy Gradient for Mixed-Reference Spin-Flip Time-Dependent Density Functional Theory (MRSF-TDDFT)." The Journal of Chemical Physics, vol. 150, no. 18, 2019.

Contributors

• Cheol Ho Choi, Kyungpook National University, South Korea, [e-mail], https://www.openqp.org
• Seunghoon Lee, Seoul National University, South Korea, [e-mail]
• Vladimir Mironov, [e-mail]
• Konstantin Komarov, [e-mail]
• Jingbai Li, Hoffmann Institute of Advanced Materials, China, [e-mail]
• Igor Gerasimov, [e-mail]
• Hiroya Nakata, Fukui Institute for Fundamental Chemistry, Japan, [e-mail]
• Mohsen Mazaherifar, Kyungpook National University, South Korea, [e-mail]
• Vladimir Makhnev, VladimirMakhnev
• Alireza Lashkaripour, Alireza Lashkaripour

Legal Notice

See the separate LICENSE file.