Test Problem Server

Overview

The Texas Test Problem Server is your source for complicated linear systems, to be used in numerical linear algebra, graph theory, or performance evaluation/optimization research. The server hosts a number of applications that can be controlled through a web interface to deliver problems as large, complicated, and varied as you want them.

Features:

• Matrices from high order finite elements
• Irregular domains
• Local refinement
• Unassembled matrices
• Problems with right hand side and analytical solution
• Non-linear and time-evolving problems

Try it out here: tps.tacc.utexas.edu.

The Test Problem Server project is funded by NSF through grant CRI 0751144: An On-demand Test Problem Server.

FAQ

Your most important question is problably "Why is this better than MatrixMarket or the University of Florida collection?"

Here's why. This is not a collection of matrices, but a dynamic set of matrix generators. So if you take a test matrix, you can generate a set of matrices from the same PDE, but with increasing discretizations. Or you can take a test problem and slightly tweak the PDE coefficients to make it a bit harder.

Why don't you just give the user the codes?

Some of our codes are proprietary, and in general we don't want to be responsible for inevitable installation problems.

Why do I have to register on your site?

You don't. Anyone can browse and download archived test problems. However, to be able to generate your own test problems you need to provide a valid email address. By having your register with it, we can offer facilities such as maintaining your history.

People

The following people have contributed to the Texas Test Problem Server:

• Victor Eijkhout, project leader and designer
• Bill Barth, co-designer, and implementer of the MGF generators
• James Kneeland, web site design and implementation
• John Peterson, implementer of the libMesh generator
• Yaakoub El Khamra, implementer of the Defiant generator
• Kyungjoo Kim, implementer of the hp3d generators

Downloads

Our software will be released into the public domain.