17 June 2003

Chair Dr. Ronald F. Boisvert, NIST, Gaithersburg, MD, USA

Vice-Chair Dr. Pieter Hemker, CWI, Amsterdam, The Netherlands

Secretary Professor Wayne Enright, University of Toronto, Canada

The working group currently has 30 full members representing 18 countries. The Working group has an additional 18 affiliate members. (Affiliates are former WG 2.5 full members group members who are interested in working group activities, but are unable to regularly attend meetings or commit large amount of time to working group projects. They maintain email contact, and attend Working Group meetings when possible.)

During our most recent meeting, Dr. Morven Gentleman (Dalhousie University, Canada) was changed to affiliate status at his request.

During our most recent meeting, three new members were nominated to membership, and each was accepted by vote. They are

The working group holds yearly meetings. These are typically comprised of a 1.5 day business meeting and an associated 2-4 day workshop. The workshop provides an opportunity for the working group to interact with researchers in the local region on issues of mutual interest. The workshops are usually informal. Recent and planned meetings are given in the table below. The locations for meetings in 2006 and 2007 are highly tentative at this point.

The last Working Conference of WG 2.5 was held in Ottawa, Canada in September 2000. The topic of the conference was The Architecture of Scientific Software. This was the eighth Working Conference of WG 2.5. Current plans are to hold a Working Conference in association with the 2006 meeting.

Portland 2003. During the business meeting several group projects were reviewed, leading to discussions on the current and future status of Fortran, Java Grande applications, a proposed standard API for discrete Fourier transforms, and a book on accuracy and reliability in scientific computing (see below).

As part of the meeting an open two-day workshop entitled The Mathematics of Mathematical Software was held (see http://math.nist.gov/mmsworkshop/). Thirteen speakers made presentations. They included WG 2.5 members, as well as researchers from Intel, Hewlett-Packard, and the Oregon Graduate Institute. Among the topics discussed were formal verification methods, interval arithmetic, automatic code generation for FFTs, high-performance numerical software for Intel processors, and practical error analysis. Approximately 35 persons attended the workshop.

St. Wolfgang 2004. During the business meeting there were extensive discussions on current Fortran standardization efforts, the status of the Collected Algorithms of the ACM, and the UK MONET (Math on the Net) project.

As part of the meeting, a 3.5-day workshop on Numerical and Symbolic Scientific Computing was held. Twenty-seven papers and 11 posters were presented. The main theme of the workshop was the interplay between symbolic computing and numerical computing, and researchers from both backgrounds participated. Four WG 2.5 members made presentations. Participants represented the University of Linz and the University of Vienna in Austria, as well as German universities in Augsburg, Munich, Graz, Stuttgart, Essen, Berlin, and Kaiserlautern. Other non-WG 2.5 participants came from Switzerland, Hungary, Bulgaria, Russia, and the UK. Also participating were the research institutes CAESAR (Center for Advanced European Studies and Research) and RICAM (Johann Radon Institute for Computational and Applied Mathematics). Approximately 75 persons were in attendance.

1. Fortran

Three WG 2.5 members (Wolfgang Walter, Van Snyder, and John Reid) are participating in the ISO WG5, which is developing a revision to the Fortran standard. Currently known as Fortran 200x, the proposed standard will add a number of data abstraction features that will continue Fortran’s march toward object orientation. Included are parameterized derived types, procedure pointers, (single) inheritance, and polymorphism. Additional features for C interoperability and internationalization are also under consideration. During the WG 2.5 discussion there was concern expressed that Fortran 200x may be overly complex, reaching too far beyond Fortran’s core constituency of high performance computing users. There was also some dismay at the fact that user-defined error handling will remain unsupported. Approval of the proposal by ISO WG5 is expected in April, with final balloting in September 2004.

2. Accuracy and Reliability in Scientific Computing

WG 2.5 is developing a book on this topic. Members, as well as a few invited guests are contributing chapters. The purpose of the book is to raise awareness of the difficulties in producing reliable numerical software, and to provide information on a variety of techniques available to improve numerical software reliability. The target audience is working engineers and scientific programmers who may have limited knowledge of numerical analysis.

The book will have three parts. The first, "What Can Go Wrong", will present examples of well-known failures in scientific software. A discussion of the nature of mathematical and computational models will be presented, followed by a survey of the difficulties found in scientific computation, such as effects of floating-point arithmetic (e.g., cancellation), unstable algorithms, and slow convergence will be discussed. Finally, a discussion of the concepts of accuracy, reliability, stability, verification, and validation will be presented. Part two, "Diagnostic Tools", will present a variety of analytic and computational tools that can be used to assess the accuracy and reliability of existing applications. Finally, part three will present "Technology for Improving Accuracy and Reliability." Included will discussions of software reliability engineering, interval arithmetic, and language issues.

Drafts for approximately three quarters of the chapters have been received. The remaining drafts will be delivered by October 1, 2003, and the final manuscript will be ready by March 2004.

The lead for this project is Bo Einarsson, Sweden.

3. API for Discrete Fourier Transforms