The goal of this project is to improve the environment for computational science through research in fundamental mathematical algorithms and development of well-engineered general-purpose scientific software. This research combines development and analysis of algorithms for particular mathematical problem domains with the application and advancement of underlying methodologies such as computer arithmetic, parallel computing, languages, software design, user interfaces, documentation, testing, performance evaluation, and information dissemination. The customers for our work in algorithms and in the dissemination of software and related information include both NIST scientists and engineers, as well as the science and engineering community at large. Our work in testing and evaluation methodology is of particular interest to the math software research community, as well as to developers of math software products in the commercial sector.

Highlights of this year's activities include the following.

- A new joint project with the NIST Statistical Engineering Division on
*Tools for Evaluating Mathematical and Statistical Software*has been initiated. Particular tasks in the areas of numerical linear algebra, special function evaluation, and statistical software are being undertaken. The Matrix Market, a visual database of test data for large sparse matrix algorithms, is the first visible output of this project. Has already generated positive feedback on its usefulness to the research community; collaborations have begun with users of sparse matrix technology, such as the Boeing Company, to include additional large-scale test data in the collection. Work on the development of a Web-based software testing service for special functions has also begun. - Significant strides applying object-oriented software design to improve portability and reuse of mathematical software have been taken. Several demonstration software packages for core linear algebra operations in C++ have been released are are seeing widespread attention. This work has also spawned interest in the development of a standardized set of basic linear algebra software for elementary sparse matrix operations. This is being undertaken in collaboration with Cray Research.
- We are undertaking several projects to develop algorithms, software and tools for distributed parallel scientific computing. On scalar processors algorithms exemplified by our highly successful MGGHAT package, which solves partial differential equations using high order hierarchical-basis adaptive multigrid methods, have proven extremely effective. Moving computations such as these to distributed parallel achitectures is quite difficult due to complex load balancing and data communications issues. This year, the prototype of a new software package named PHAML was completed which parallelizes such computations based upon a new multigrid-based domain decomposition and refinement-tree-based load balancing strategies. We continue to contribute to the development of software development tools, such as the Parallel Applications Development Environment (PADE), a joint project with the High Performance Systems and Services Division and the NIST Physics Laboratory. In addition we provide specialized support for use of the IBM SP2 parallel computer with educational materials and utilities such as xllcreate.
- The Guide to Available Mathematical Software (GAMS) continues to see high
use by the science and engineering community. With more than 9,000 users
per month, we saw our one millionth and two millionth Web hits this
year. Digital's AltaVista search engine identifies more than 3,000 external
Web pages that link to the GAMS server, and the McKinley Group (a
commercial venture than rates Web pages for presentation and content)
awarded GAMS its 4-star (highest rating). We continue to investigate new
technologies for increasing the usefulness of scientific software
repositories, and this year we released a prototype of
*HotGAMS!*, a Java-enabled GAMS client which provides a new interface which is more capable, portable and efficient.

Generated by boisvert@nist.gov on Mon Aug 19 10:08:42 EDT 1996