MOIST is a modular library for modern implicit solvation in molecular quantum chemistry, usable from Fortran, C, and Python, and implemented with performance and differentiability in mind.
Note
MOIST is in pre-release: the Fortran, C, and Python APIs may still change between releases. Feedback, feature requests and contributions are very welcome; please open an issue.
- Cavities: smooth, fully differentiable Discretization via Reference-Onto-Surface Projection (DROP) cavities: Smooth van der Waals (SvdW-DROP) and electron-isodensity (ρ-DROP) surfaces, the vdW improved Switching Gaussian (iSwiG) cavity, and marching tetrahedra; Numerical Surface Area (NumSA) and marching cubes for reference areas and volumes
- Polarizable Continuum Models: Conductor-like PCM (CPCM), Conductor-like Screening Model (COSMO), Integral Equation Formalism PCM (IEF-PCM), Surface and Simulation of Volume Polarization for Electrostatics (SS(V)PE), Dielectric PCM (DPCM), Solvation-Layer Interface Condition (SLIC), Linearized Poisson-Boltzmann (LPB), and nonlocal electrostatics
- Pressure Models: classical Pressure-Volume (PV) and Gaussians On Surface Tesserae Simulating HYdrostatic Pressure (GOSTSHYP) pressure components
- Reference Interaction Site Models: 1D and 3D RISM (in development) with a modular Lennard-Jones potential library (UFF, DREIDING, GAFF2, OPLS-AA, SPC/E, and transition-metal parameters)
- Interfaces: Fortran, C and Python APIs, a PySCF module (HF and DFT energies and nuclear gradients), and the
moistCLI; ORCA and TURBOMOLE implementation will be available with their next major releases
To build this project from the source code in this repository you need to have
- a Fortran compiler supporting Fortran 2018
- a C11 compiler and a C++17 compiler (e.g. gcc/g++ with gfortran, icx/icpx with ifx or ifort, or clang/clang++ with gfortran)
- one of the supported build systems:
- a LAPACK / BLAS provider, like MKL, OpenBLAS, or Accelerate on macOS
See Tested platforms for the compilers and platforms covered by CI.
Optional dependencies are
- FORD to build the developer documentation
- asciidoctor to build the man page
- Python 3.10 or newer with the CFFI and setuptools packages to build the Python API
Setup a default build with
meson setup buildYou can select the Fortran compiler by the FC environment variable. To compile and run the projects testsuite use
meson test -C build --print-errorlogsIf the testsuite passes you can install with
meson configure build --prefix=/path/to/install
meson install -C buildBuild options
The build is configured through meson options passed to meson setup:
| Option | Default | Effect |
|---|---|---|
-Dlapack=... |
auto |
BLAS/LAPACK backend: auto, mkl, mkl-rt, openblas, netlib, accelerate, custom |
-Dcustom_libraries=... |
[] |
Libraries for the custom backend: paths, names, or a leading -L<dir> |
-Dilp64=true |
false |
Use 64-bit-integer (ILP64) BLAS/LAPACK; needs an ILP64 BLAS/LAPACK |
-Dopenmp=false |
true |
OpenMP parallelisation |
-Dapi=false |
true |
C API (include/moist.h) |
-Dpython=true |
false |
Python extension module, see Python API |
-Ddocs=true |
false |
Man page (needs asciidoctor) |
-Ddev_tests=false |
true |
The dev_tester binary with long-running diagnostic tests |
The cmake build exposes the same toggles as -DMOIST_* options, see below.
Configure and compile a default build with
cmake -B build -G Ninja
cmake --build buildThe Fortran compiler is selected through the FC environment variable. To install, set the prefix at configure time and install the built tree
cmake -B build -G Ninja -DCMAKE_INSTALL_PREFIX=/path/to/install
cmake --build build
cmake --install buildThe unit-test suite is built by default (the moist-tester target; disable with -DMOIST_TESTS=OFF). Run it with
ctest --test-dir build --output-on-failureBuild options
The optional numerical features map onto -DMOIST_<FEATURE> cache variables, with defaults matching the meson options:
| cmake option | meson equivalent | Default |
|---|---|---|
-DMOIST_ILP64=ON |
-Dilp64=true |
OFF |
-DMOIST_OPENMP=OFF |
-Dopenmp=false |
ON |
-DMOIST_API=OFF |
-Dapi=false |
ON |
-DMOIST_LAPACK=... |
-Dlapack=... |
auto |
-DMOIST_TESTS=OFF |
— | ON |
-DMOIST_LAPACK accepts the same backends as meson (auto, mkl, mkl-rt, openblas, netlib, accelerate, custom); custom takes its libraries from -DMOIST_CUSTOM_LIBRARIES=.... For example, a build against ILP64 BLAS/LAPACK:
cmake -B build -G Ninja -DMOIST_ILP64=ON
cmake --build buildThe moist command line tool is organised into subcommands. The general form is
moist <subcommand> [options] <input>Run moist --help to list all subcommands, or moist <subcommand> --help (for example moist cavity drop svdw --help) for the options of a specific command. moist --version, moist --citation, and moist --license print the version, the relevant literature references, and the license notice.
To build a molecular cavity with the SvdW-DROP scheme from an XYZ coordinate file:
moist cavity drop svdw coord.xyzHere cavity selects the cavity-only workflow, drop the DROP cavity construction, and svdw the smooth van der Waals level set (the alternative level set is cfc). Common options for this command are --radii {cpcm,smd,d3,cosmo,bondi} to choose the atomic radii set, --nleb <N> to set the Lebedev points per atom. The other cavity constructors, NumSA and iSwiG, are available as moist cavity {numsa,iswig} <coord>, and moist cavity mc {svdw,cfc} <coord> integrates the level-set isosurface with marching cubes.
moist solvent <name>reports the tabulated properties of a solvent by name or alias.
moist provides first class API support for Fortran, C and Python; see the API reference for an overview. Other programming languages should try to interface with moist via one of those three APIs. To provide first class API support for a new language the interface specification should be available as meson build files.
The recommended way to access the Fortran module API is by using moist as a meson subproject.
The complete API is available from moist module, the individual modules are available to the user as well but are not part of the public API and therefore not guaranteed to remain stable. ABI compatibility is only guaranteed for the same minor version.
The communication with the Fortran API uses the error_type and structure_type of the modular computation tool chain library (mctc-lib) to handle errors and represent geometries, respectively.
The Python API is disabled by default and can be built in-tree or out-of-tree. The in-tree build is mainly meant for end users and packages. To build the Python API with the normal project set the python option in the configuration step with
meson setup build -Dpython=true -Dpython_version=$(which python3)The Python version can be used to select a different Python version, it defaults to 'python3'. Python 2 is not supported with this project, the Python version key is meant to select between several local Python 3 versions.
Proceed with the build as described before. The package can be used directly from the build tree with PYTHONPATH=build/python, or installed to make it available in the selected prefix. With -Dpython=true the test suite also runs the Python tests and requires numpy, pytest, and PySCF.
For the out-of-tree build see the instructions in the python directory.
Every combination below builds and passes the full test suite in CI. The oldest tested compilers are GCC 14, Intel ifx 2025.1 and classic Intel ifort 2021.12; older versions may work but are not tested.
Platform and compiler matrix
| Platform | GCC 15 | conda-forge GCC 15 | Intel ifx 2025.1 | Intel ifort 2021.12 | Python |
|---|---|---|---|---|---|
| Linux x86_64 (Ubuntu 24.04) | meson, CMake | meson, CMake | meson, CMake | meson, CMake | 3.10, 3.13 |
| Linux arm64 (Ubuntu 24.04) | meson, CMake | meson, CMake | – | – | 3.13 |
| macOS arm64 (macOS 15) | meson, CMake | meson, CMake1 | – | – | 3.13 |
| Windows x86_64 | meson, CMake2 | – | meson, CMake3 | – | – |
1) With clang 19 and libc++ as the C/C++ compilers.
2) MSYS2 UCRT64, which ships only its current GCC, at present 16.x.
3) Native toolchain with icx and Visual Studio 2022; static library only, meson with -Ddefault_library=static.
- BLAS/LAPACK: reference LAPACK (Linux, GCC), MKL (ifx, ifort), OpenBLAS (conda-forge, MSYS2), Accelerate (macOS)
- ILP64: 64-bit integer BLAS/LAPACK with GCC 15 and OpenBLAS on Linux x86_64
- Debug: GCC 14 debug build with runtime checks (
-fcheck=bounds,do,mem,pointer,_GLIBCXX_ASSERTIONS) on Linux x86_64 - Python: the bindings are tested against PySCF, both from a meson build and as a pip-installed package
Contributions are welcome; see CONTRIBUTING.md for the code style, linting, testing, and documentation guidelines.
If you use moist, please cite the software itself (see CITATION.cff; the DOI badge above links to the latest release):
- L. Wittmann, MOIST: The Modular and Open-source Implicit Solvation Toolkit [Computer software], Zenodo 2026. https://doi.org/10.5281/zenodo.23473416
together with the references for the cavities, models, and solvers you used. For the DROP cavities, these include
- L. Wittmann, A. Pausch, A Smooth and Fully Differentiable Molecular Cavity Based on Discretization via Reference-Onto-Surface Projection, ChemRxiv 2026. https://doi.org/10.26434/chemrxiv.15003893/v2
For the isodensity cavity (ρ-DROP), please also cite
- L. Wittmann, From Density to Boundary and Back: A Fully Differentiable, Self-Consistent Isodensity Cavity, ChemRxiv 2026. https://doi.org/10.26434/chemrxiv.15007095/v1
At the end of a run, moist prints the citations for the cavities, models, and solvers it used; the full list is printed by
moist --citationmoist is free software released under the MPL-2.0 license (see LICENSE).
It also bundles third-party code under separate licenses. The full inventory, with origins and license texts, is in THIRD_PARTY_NOTICES.md.
We gratefully acknowledge the people and projects moist builds on:
- Jaś Kachnowicz for help with the initial version of the solvent properties and geometries.
- Jacob Williams for modern Fortran implementations of the SLSQP, L-BFGS-B, nlesolver, fmin, LSQR, LSMR, and LUSOL solvers.
- The FINUFFT developers at the Flatiron Institute, Martin Reinecke for ducc0, and the xsimd developers for the FFT and SIMD code.
- The authors of the force-field parameter sources: Peter Boyd for lammps_interface; the openmmforcefields developers; Leela S. Dodda and the Jorgensen Lab for OPLSAA-DB; Gajanan Kanagalingam, Sebastian Schmitt, Florian Fleckenstein, and Simon Stephan for TUK-FFDat; and the VeloxChem developers.
- The developers of the mctc-lib, mstore, toml-f, jonquil, test-drive, and fclap projects.
These components retain their original copyright notices and licenses; the full texts accompany each component in the source tree and are catalogued in THIRD_PARTY_NOTICES.md.