Skip to content


Folders and files

Last commit message
Last commit date

Latest commit



11 Commits

Repository files navigation

Documentation Status CI codecov Maintainability Gitter Chat


Parallelized Matrix Exponentiation for Numerical Time evolution

Parament is a GPU-accelerated solver for time-dependent linear differential equations. Notably, it solves the Schrödinger's equation with arbitrary time-dependent control terms. Parament is open-source and is released under the Apache Licence.


The official documentation of Parament is available on Read The Docs.

Research paper

If you want to know the nitty-gritty details, have a look at the research paper which got published in Computer Physics Communications:

Konstantin Herb, Pol Welter:
Parallel time integration using Batched BLAS (Basic Linear Algebra Subprograms) routines
Computer Physics Communications 270, 2022, 108181, ISSN 0010-4655

Try it out

You can try Parament for free, in the browser, via Google Colab (no local GPU required!).


Prerequisites (Windows)

  1. Install Microsoft Visual Studio (Community Edition is sufficient)

  2. Install the CUDA toolkit. Make sure the version of CUDA supports your GPU.

  3. Add the MSVC compiler to your PATH. For Visual Studio 2019, it can be found at C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.26.28801\bin\Hostx64\x64.

    You will have to substitute the exact version of the compiler you are using.

Prerequisites (Linux)

CUDA toolkit must be installed.

Building parament

Note: this step is for building the shared library (DLL). If you are only interested in using parament from Python, skip ahead. The python installer will build the library automatically.

Run build.bat or in the tools folder. The working directory must be set to the repository root. The script will create a /build folder, containing the binaries.

Any extra command line arguments are forwarded directly to nvcc. Use this to e.g. to compile for an older GPU architecture, that is not targeted by the default nvcc settings.

Building & installing pyparament

Pyparament is the official Python wrapper. It requires Python >=3.6. You can get parament directly from PyPI.

pip install parament

Alternatively, you can also download the very latest version directly from Github.

pip install git+

Pip will automatically compile the C library, and install it alongside the Python wrapper.

If you need to pass special arguments to the nvcc compiler, you can do so by setting a NVCC_ARGS environment variable. For instance, if you are targeting a Tesla K80 GPU (compute capability 3.7) from CUDA 11+, run (Windows and Linux, respectively):

set NVCC_ARGS=-arch=compute_37
pip install parament
pip install parament

After you have installed pyparament, you can run the test suite (requires pytest to be installed).

pytest --pyargs parament

Source tree structure

This is a rough overview of the most important files and folders in this repository.

├── _docs                       The source code for the documentation.
├── _src                        The root of the parament source.
│   ├── _cuda                   CPP source code.
│   └── _python                 Source code of Python wrapper.
│       └── _test               Built-in test-suite. See above.
├── _tests                      Runner script to execute test suite on Kaggle.     
└── _tools                      Various scripts, mainly related to building.
    └── build.bat               Build script for Windows. See above.
    └──                Build script for Linux. See above.

Various other scripts in /tools are mostly used in our internal automated build system.