NA Digest Sunday, May 20, 1990 Volume 90 : Issue 20

Today's Editor: Cleve Moler

Today's Topics:


From: Paul Saylor <>
Date: 18 May 90 15:42 +0200
Subject: A Notable Anniversary

Notable Anniversaries:

Forty years ago this month David Young handed in his thesis, a piece
of work described by one of his admirers as a ``great intellectual


From: Germund Dahlquist <>
Date: Wed, 16 May 90 20:02:18 +0200
Subject: Special Issue of BIT on Conjugate Gradient Methods

The Scandinavian journal BIT has recently published a special
issue on "Preconditioned Conjugate Gradient Methods" with Owe
Axelsson as guest editor. It contains 18 contributions from the
whole world and covers about 375 pages. It has also appeared in
book form and is available at US$ 42.00 (42 dollars) from
Box 113
DK-1004 Copenhagen K


From: David H Schultz <>
Date: Thu, 17 May 90 12:01:40 -0500
Subject: Numerical Analysis Tutorial Program

This is in response to Baker Kearfott's query about teaching numerical

We have written a computer based numerical analysis tutorial. It is completed
menu driven and all inputs are thoroughly prompted. Results are presented
numerically and graphically.

Most of the major methods presented in the textbook Applied Numerical Analysis
by Gerald/Wheatley are covered in the tutorial. Each program gives you the
choice of going directly to the method or of first going to an explanation of
the method.

The major advantage of the tutorial is that the students can enter their own
problems without doing any programming. The tutorial prompts the students for
this information and then merges it into the code. The tutorial runs on IBM

If you are interested, I can send you additional information. If you would
like a copy, I have a limited number I could send out.

David H. Schultz
Math. Dept.
University of wis.-Milwaukee
Milwaukee,Wi 53201


From: T. R. Hopkins <>
Date: Thu, 17 May 90 10:39:51 BST
Subject: Machine Readable Bibliographies

David Morse and I have had a remark on Algorithm 620 accepted for
publication in Toms (it was scheduled to come out in the March issue -- I
suspect it will be in the next one). This consists of a database
containing the author, title, reference, keywords and shar index for all
the complete Collected Algorithms of the ACM plus references to all the
remarks, certificates etc. (to the end of 1989)
along with some Fortran programs for generating
a Bibtex database or a cumulative index (as used to appear in earlier
CACMs). An awk script to produce the Bibtex database would be easy to
produce -- it was used as a prototype for the Fortran!

We also have similar databases for the algorithms which have appeared in
Applied Statistics and the Computer Journal (this one hasn't been
completely checked out yet).

The problem at the moment is that due to some small anomolies (we didn't
think about things hard enough at the start of this project :-) ) the same
Fortran codes don't work on all the data sets. An MSc student is just
about to start a project to amalgamate all the available references.

If there is sufficint demand we can certainly generate Bibtex databases
for both the CALGO and AS algorithms and make them available from our
netlib with copies to the US feeds.

Tim Hopkins & David Morse
Computing Laboratory
University of Kent
Canterbury, Kent CT2 7NF


From: Roger Grimes <>
Date: Tue, 15 May 90 13:53:25 PDT
Subject: Extending the BLAS

After being raised on the BLAS by Dick Hanson at Washington State
University in 1974, I am a firm believer in the BLAS. I am glad
that they have been extended formally to include sparse, 2-level,
and 3-level operations. I have been somewhat involved in these
standardized extensions. I am also glad that individuals, companies,
universities, and groups find them so useful that they provide
local extensions. Maybe someday we will found a BLAS standards
committee to standardize some off the local extensions. In the meantime
I have a passionate nit to pick. When someone (e.g., Cleve Moler
in the May 7th issue of this newsletter) proposes a BLAS extension
they should at least follow the BLAS conventions. Back in the dark
history of numerical analysis (1979) the BLAS were published in TOMS.
These conventions were followed for all of the published (and hence
standardized) extensions to the BLAS. These conventions should be
followed when ANYONE proposes a BLAS extension. One of the most
violated convention is the naming convention. BLAS names should
have a four letter root name (e.g., SCAL). That is because the
full BLAS subroutine has a precision prefix which is 1 or 2 characters
and Fortran has a 6 character limit for subroutine names. Instead
of rashly proposing names like SCATTER and GATHER as Cray Research
did in the early 80's or SCALD for scaling by division as Cleve did
2 weeks ago, please, consider the ramifications of what you do on
others and what it means to create a generally useful tool. The
BLAS conventions have survived the test of time (an amazing 15
years of utility in a rapidly changing field) -- they are worth
adhering to.

Roger Grimes


From: Ken Clark <>
Date: Tue, 15 May 90 17:21:55 EDT
Subject: ARO's European Numerical Analysis Activities

The Mathematical Sciences Division of the Army Research Office
currently manages the scientific activities of the Mathematics
and Physics Division of the European Research Office (ERO).
The ERO (located in London) is the European counterpart to ARO,
sponsoring individual investigator research programs, workshops,
travel between Army laboratories and conferences abroad, and maintaining
cognizance of research activities in Britain, Europe, and throughout
the Near East.

As program manager of the ARO program in Numerical Analysis and
Scientific Computing, I am taking a 10 week leave of absence (5/22 to 7/27)
to act as Chief of the Math. and Physics Division at ERO. A large
portion of my activities will involve traveling to conferences and
visiting universities to discuss recent research activities with
investigators currently supported by ERO, but also with any other
relevant (hopefully interested) scientists. I am particularly interested
in meeting members of the European numerical analysis community and will
no doubt contact many of them myself. However, in the likely event that
I do not contact someone who has something interesting they'd like to
discuss, I would welcome hearing from them too. Of course if you are
based in the US but are traveling abroad for the summer and I
haven't spoken with you before, I'd be happy to hear from you as well.
My address in London will be

Dr. Kenneth D. Clark
Mathematics and Physics Division
US Army Research, Development, and Standardization Group (UK)
Edison House
223 Old Marylebone Drive
London NW1, 5th


Wishing you all safe travel and a productive summer,
Ken Clark


From: Peter Arbenz <>
Date: 18 May 90 14:06 +0200
Subject: Joint Conference on Vector and Parallel Processing in Zurich

Second announcement & Preliminary Program


Joint Conference on Vector and Parallel Processing

Swiss Federal Institute of Technology

September 10-13, 1990

Sponsored by
Schweizer Informatiker Gesellschaft (SI)
IEEE Switzerland Section

in cooperation with
Gesellschaft fuer Informatik GI-PARS
British Computer Society BCS-PPSG
Computer Society of the IEEE
Swiss Chapter of the ACM

The past decade has seen the emergence of the two highly
successful series of CONPAR and of VAPP conferences on the
subject of parallel processing. The Vector and Parallel
Processors in Computational Science meetings were held in
Chester (VAPP I, 1981), Oxford (VAPP II, 1984) and
Liverpool (VAPP III, 1987). The International Conferences on
Parallel Processing took place in Erlangen (CONPAR 81),
Aachen (CONPAR 86) and Manchester (CONPAR 88). The format of
the joint meeting will follow the pattern set by its
predecessors. It will not only attract experts in the field
from throughout the world but also people interested in the
practical use of vector and parallel computers for their
future work. It is intended to review hardware and
architecture developments together with languages and
software tools for supporting parallel processing. Another
objective of the conference will be to highlight advances in
algorithms and applications software for vector and parallel


Prof. H. F. Jordan (Univ. of Colorado, USA):
Parallel Computers - Architecture and Systems

Prof. R. H. Perrot (Queen's Univ. Belfast, U.K.):
Parallel Languages and Tools

Keynote Address:

Prof. A. P. Speiser, ABB Corporate Research Baden,
Digital Electronics for 50 Years: No Limits to Growth?

Invited Presentations:

Prof. V. P. Bhatkar, Centre for Development of Advanced
Computing, Pune, India:
Parallel Computing: An Indian Perspective

Dr. G. Cybenko, Center for Supercomputing Research and
Development, University of Illinois at Urbana - Champaign,
The Use and Interpretation of the PERFECT Benchmarks.

Dr. E. Odijk, Philips Research Lab, Eindhoven, The
Parallel Object-oriented Computer for Information Management


Distributed and Shared Memory Systems: Convergence or

Contributed Presentations:

The International Program Committee has selected papers for
the following sessions:

- New Models of Computation
- Logic Programming
- Large-Grain Data Flow
- Interconnection Networks
- Communication Issues
- Reconfigurable and Scalable Systems
- Implementation Issues for Novel Architectures and Languages
- High-Performance Systems and Accelerators
- Performance Prediction, Analysis, and Measurement
- Performance Monitoring and Debugging
- Compile-Time Analysis and Restructurers
- Load Balancing and the Mapping Problem
- Process Partitioning and Work Distribution
- Concurrency Control
- The TOPSYS Tool Environment
- Visualization and Runtime Analysis
- Parallel Linear Algebra
- Algorithms for Matrix Factorization
- Architectures and Algorithms for Image Processing
- Efficient Use of Vector Processors
- Performance Considerations
- Transputer Tools and Applications
- Array Processors and Applications
- Algorithmic Studies for Hypercube-Type Systems
- Cellular/Systolic Architectures and Algorithms

Awards and Computation Race:

The Awards Committee will award one or two prizes and a
Plaque of Recognition in each of four categories:
- Best technical contribution
- Excellent presentation (based on nomination by the
- Best visually animated presentation of concurrency and/or
parallelism (see below)
- The fastest computer solution to a (highly?) parallel
problem (see below)
The committee anticipates announcing the awards and the
results of the competition during the last session of the
conference. At that session, selected animations will be
shown, and participants in the computation race will present
selected descriptions of their projects.

Computation Race:

Participants in the computation race are encouraged to
prepare a brief (< 15 minutes) presentation of their
project: hardware, software, algorithm, experience gained.
For details and submissions (until July 1, 1990) please

Prof. J. Nievergelt
ETH - ZentrumPhone: +41-1-254-7380
InformatikFax: +41-1-262 -3973
CH-8092 Zurichemail:

Visualisation of Concurrent Processes:

The Awards Committee is looking for original research and
development work in the area of Visualization of Concurrent
Processes. Processes are said to be concurrent or
simultaneous or parallel if they co-exist in time.
Contributions should be animated, esthetically pleasing,
provide 'aha-effects', and have their origin in a field of
application such as parallel circuit and architecture
design, parallel geometric algorithms, pattern
recognition, artificial neural nets, particle
motion simulation, cellular automata, etc. The expected
length of the individual contributions should be in the
range of 30 to 240 seconds of running video or real-time
display on a workstation.
Contributions must be submitted by July 1, 1990 to

Prof. P. Stucki
University of ZurichPhone: +41-1-257-4350
Dept. of Computer Sc.Fax: +41-1-257-4343
Multimedia Laboratoryemail:
CH-8057 Zurich

Exhibition, Demonstrations and Posters:

Complementary to the tutorials and regular sessions of the
conference, there will be an exhibition and presentation of
vendor products, current research work at universities,
demonstrations of non-commercial system prototypes, and
poster sessions.
The exhibition will be open to the public. It will be housed
in a convenient and attractive setting of the main building
of ETH just outside of the seminar rooms where the
conference will take place. During the breaks in the morning
and in the afternoon, coffee will be served to the
participants of the conference in this area, so that the
exhibition will get sufficient attention.
The exhibition will have four parts:

Registration and Conference Secretariat:

Please contact the Conference Secretariat in order to get
forms for conference registration and hotel reservation.

Secretariat CONPAR 90 / VAPP IV
Dr. Ernst Rothauser
IBM Research Laboratory Zurich
CH-8803 Ruschlikon, Switzerland
Tel.+41-55-67 2076
Fax:+41-1-724 0141


End of NA Digest