I have been fascinated with sparse direct solvers ever since I encountered them whilst learning finite element analysis back during my undergraduate days. One particularly clever trick that stood out was the multi-frontal method developed by Bruce M. Irons [Irons1970] back in the 1960s that performed Gaussian elimination on the stiffness matrix while it was being assembled in the usual element-by-element fashion.

As a graduate student, I was fortunate enough to re-learn finite element methods from Prof. K-J Bathe. One of the intriguing aspects of sparse solvers he mentioned in his class was the use of methods based on graph theory that had surpassed classical skyline solvers such as COLSOL described in his book [Bathe1995]. This was the first time I had heard about the connection between finite element methods and graph theory - two very distinct worlds that are deeply connected at a rather profound level. Although my curiosity was piqued, I did not dig deeper into sparse solvers until 2006 when I noticed an announcement in SIAM News about an upcoming book by Tim Davis on sparse direct solvers [Davis2006] and immediately preordered it.

I started reading Davis' book in earnest during Fall 2006 and quickly realized that the only way I was going to benefit from the book was to digest the (rather terse) algorithms and reimplement them myself along with tackling all the exercises. Unfortunately, this has been a rather painstaking process and progress was glacially slow until 2011 when I resolved to finally dedicate a part of my weekends to understand sparse solvers, matrix reordering algorithms and the graph theoretical foundations of sparse linear solvers.

CppSparse has been the result of my work so far. It is a sparse matrix library based on the algorithms and exercises in Tim Davis' book. It is implemented in C++ and extensively uses the containers and algorithms in the standard library. This allows me to elide details such as checking for out-of-memory conditions or managing the lifetime of my temporary buffers. The templated nature of CppSparse allows it to be used in both 32- and 64-bit libraries and the same code base supports both real and complex-valued sparse matrices.

CppSparse is distributed as a self-contained header-only library. It has very few dependencies besides a half-decent C++ compiler. I primarily use CppSparse through its Python bindings and aim for it to be a fully-featured sparse matrix library for Python.

CppSparse currently reliably builds on Windows using Microsoft Visual Studio 2010 and MingW GCC 4.5 (64-bit) and is tested with 64-bit Python 2.7 and NumPy. I also occassionally build it on my Mac OS Snow Leopard and back-port the changes to Windows.

Some of the routines in CppSparse use C++11 features such as lambda expressions that are not supported out-of-the-box on the older GCC installed on Snow Leopard; I will eventually remove these features once I have completed a few more chapters in Davis' book. I also had move constructors on the matrix classes that I disabled because SWIG seemed to choke on them.

• SWIG version 2.0.1 or better for generating the Python bindings
• Microsoft Windows 7 SDK (64-bit) that contains the 64-bit version of the Visual C++ compiler, or,
• MingW64 with GCC 4.5 or better for compiling the generated bindings
• Python 64-bit version 2.7 or better (I have not tested CppSparse with Python 3.x)
• NumPy compatible with your Python installation. You can get a 64-bit installer for Windows from Ch. Golke 's website.

* GraphViz (optional). Some of the functions generate graphs in the GraphViz format; you can use tools like dot to post-process these graphs and create pretty pictures. .. _SWIG: http://www.swig.org .. _MingW64: http://tdm-gcc.tdragon.net/ .. _Microsoft Windows 7 SDK (64-bit): http://www.microsoft.com/download/en/details.aspx?id=8279 .. _Python: http://www.python.org .. _NumPy: http://www.numpy.org .. _Ch. Golke: http://www.lfd.uci.edu/~gohlke/pythonlibs .. _GraphViz: http://www.graphviz.org

• Ensure that SWIG is in your path.
• Start up a 64-bit Visual C++ command prompt, navigate to the source directory and run buildswig.bat.

• Ensure that SWIG and GCC are in your path.
• Navigate to the source directory and run make -f Makefile.mingw

• Finish Cholesky factorization
  • Numerical factorization and
  • Column Cholesky factorization are still pending
• Work on QR factorization
• Work on LU factorization
• Work on reordering techniques (phew!)
• Make all return types be std::shared_ptr so they don't get copied constantly in SWIG.
• Add more PyUnit tests
• Finish matrix arithmetic operations
• ...

CppSparse is licensed under the MIT/X11 license:

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.