Skip to content

hifirworks/hifir4py

Repository files navigation

Python3 Interface for HIFIR (hifir4py)

image

image

image

image

Welcome to the Python3 interface of HIFIR package--- hifir4py. The Python interface is implemented with Cython.

Detailed documentation of hifir4py can be found at https://hifirworks.github.io/hifir4py.

Installation

One can simply use pip to install hifir4py, i.e.,

pip3 install hifir4py --user

For Windows users, we recommend using Windows Subsystem for Linux (WSL) or Cygwin.

Installation from Source

It is also possible to install from the source. First, down load the most recent stable release

git clone -b release https://github.com/hifirworks/hifir4py.git

You can update to the most recent version via git pull since git clone or last git pull. In addition, you can download the archives at https://github.com/hifirworks/hifir4py/releases.

You need to configure linking against LAPACK by setting the environment variable HIFIR_LAPACK_LIB whose default is -llapack. If you have a specific library path to LAPACK, you then need to set the environment variable HIFIR_LAPACK_LIB_PATH.

To sum up, the following environment variables can be configured

  1. HIFIR_LAPACK_LIB, default is -llapack
  2. HIFIR_LAPACK_LIB_PATH, default is empty
pip3 install . --user

Installation with customized LAPACK/BLAS

Sometimes, it's helpful to have optimized LAPACK package. The following command shows how to link MKL (on Ubuntu).

HIFIR_LAPACK_LIB="-lmkl_intel_lp64 -lmkl_sequential -lmkl_core" HIFIR_LAPACK_LIB_PATH=/opt/intel/mkl/lib/intel64 pip3 install . --user

Copyrights & Licenses

hifir4py is developed and maintained by the NumGeom Research Group at Stony Brook University.

This software suite is released under a dual-license model. For academic users, individual users, or open-source software developers, you can use HIFIR under the AGPLv3+ license free of charge for research and evaluation purpose. For commercial users, separate commercial licenses are available through the Stony Brook University. For inqueries regarding commercial licenses, please contact Prof. Xiangmin Jiao at xiangmin.jiao@stonybrook.edu.

How to Cite hifir4py

If you use HIFIR (including hifir4py) in your research for nonsingular systems, please cite the following paper.

@article{chen2021hilucsi,
    author  = {Chen, Qiao and Ghai, Aditi and Jiao, Xiangmin},
    title   = {{HILUCSI}: Simple, robust, and fast multilevel {ILU} for
               large-scale saddle-point problems from {PDE}s},
    journal = {Numer. Linear Algebra Appl.},
    year    = {2021},
    doi     = {10.1002/nla.2400}
}

If you use our work in solving ill-conditioned and singular systems, we recommend you to cite the following papers.

@article{jiao2020approximate,
    author  = {Xiangmin Jiao and Qiao Chen},
    journal = {arxiv},
    title   = {Approximate generalized inverses with iterative refinement
              for $\epsilon$-accurate preconditioning of singular systems},
    year    = {2020},
    note    = {arXiv:2009.01673}
}

@article{chen2021hifir,
    author  = {Chen, Qiao and Jiao, Xiangmin},
    title   = {{HIFIR}: Hybrid incomplete factorization with iterative
               refinement for preconditioning ill-conditioned and singular
               Systems},
    journal = {arxiv},
    year    = {2021},
    note    = {arXiv:2106.09877}
}

Contacts

  1. Qiao (Chiao) Chen, <qiao.chen@stonybrook.edu>, <benechiao@gmail.com>
  2. Xiangmin Jiao, <xiangmin.jiao@stonybrook.edu>