The decomposition, solution and eigenanalysis of systems of linear equation remain important fundamental problems in scientific computation for which new algorithms and software packages are continually being developed. In order to make meaningful, reproducible quantitative assessments of the value of new algorithmic developments it is helpful to have a common collection of representative problems through which methods can be compared. Of particular recent concern problems characterized by very large, sparse matrices from industrial applications. The Matrix Market seeks to promote test and evaluation of new algorithms and software in this domain by providing a Web-based focal point for the exchange of test corpora of this type.
A prototype of the Matrix Market was released in February 1996 seeded with some 300 matrices from the well-known Harwell-Boeing Collection of sparse matrices. For each matrix we provide a Web page summarizing the properties of the matrix, providing a graphical representation of its structure, and permitting downloading of the matrix data via a mouse click. A search tool allows matrices to be selected by specifying various properties (e.g., real, symmetric, positive definite), by providing minimum or maximum number of rows, columns and nonzeros, by specifying what associated materials are required (e.g. right hand sizes, solution vectors), by application domain, or by contributing institution.
The announcement of the Matrix Market met with an immediate positive response from the research community.
We have been in contact with researchers at the University of Minnesota, the University of Florida and NASA Ames who have additional matrix collections for inclusion in the Matrix Market. We also expect to be able to investigate the use of technologies such as Java to extend the collection from static matrix data to matrix generation software.