NA Digest Sunday, March 31, 1991 Volume 91 : Issue 13

Today's Editor: Cleve Moler

Today's Topics:


From: Michael Mascagni <>
Date: Tue, 26 Mar 91 11:27:08 EST
Subject: Fast N-dody Code Wanted

Dear na-netters:
I am looking for a code for evaluating a Newtonian-like potential on all
N-bodies in a simulation. I already have a vectorizing version of the O(N^2)
algorithm for my XMP as well as a parallel version for my CM-2. What I really
would like is a vectorizing/parallel version of one of the new tree-based
algorithms that are O(N log N). If you have one, or know of somewbody else
how has one, let me know.--Michael Mascagni (na.mascagni)


From: Rich Kerswell <>
Date: Fri, 29 Mar 91 10:22:26 EST
Subject: Bessel Functions with Imaginary Index

Nick Trefethen suggested I send a message here regarding available
software. I am looking for a code to evaluate bessel functions
with imaginary INDEX i.e. J_nu(x) and Y_nu(x) with x real but nu
purely maginary. Any suggestions?

Many thanks
Rich Kerswell


From: James W. Demmel <demmel@wsparc.Berkeley.EDU>
Date: Sat, 30 Mar 91 22:45:04 PST
Subject: High Precision Arithmetic

There are several architectural enhancements to support high
precision arithmetic in high-speed microprocessors that have
either appeared or are currently being considered. This is
a request for input regarding what the numerical analysis
community would like in terms of system support for high
precision (128-bit) arithmetic.

Here are some examples of the architectural changes available
or being considered. The IBM RS/6000 has a multiply-add instruction
"a := b+c*d" which computes a correctly rounded value of a.
Another possibility being considered is quad format (128 bits),
and yet another is "double-double", where a 128 bit quantity
is represented by two 64-bit floating point numbers (one for
the high order bits, and the other for the low order bits).
These approaches tradeoff functionality, performance, and
cost in different ways. For example, a 64-bit multiply-add
instruction is least expensive, but it lets one simulate
128-bit arithmetic fairly inexpensively at the cost of somewhat
messy software. Full 128-bit arithmetic is easiest to use, but
most expensive to implement, and may adversely effect the
performance of other instructions.

How important would 128-bit arithmetic be in your computations?
For example, are there some slightly unstable calculations
it would let you do which you cannot now do? Would you need
all operations in 128-bits or just some?

As an enticement to respond, I will summarize and distribute
the responses I get (unless someone requests otherwise).

Jim Demmel
Computer Science Division and Mathematics Dept.
U.C. Berkeley


Date: Wed, 27 Mar 91 13:03:46 MEZ
Subject: ICS'91 Advance Program

June 17 - 21, 1991
Queens Hotel, Cologne, Germany



The fifth International Conference on Supercomputing, sponsored by
ACM-SIGARCH, covers a variety of topics dealing with current research
results in the development and use of supercomputer systems and
their implications for future supercomputer development. The sessions
have been scheduled around the areas of architectural design, algorithms,
applications, performance analysis, and software systems support.


Edward S. Davidson Yoichi Muraoka
University of Michigan Waseda University
Ann Arbor, U.S.A. Tokyo, Japan

Friedel Hossfeld
Research Center Juelich (KFA)


U. Trottenberg (Chair) A. Lichnewski
L.M. Delves P. Mueller-Stoy
I. Duff T. Papatheodorou
W. Giloi P. Sguazzero
J.R. Gurd H. Wijshoff
G. Hoffmann H. Zima
W. Jalby

E. Houstis (Chair) D. Gannon
D. Bailey M. Heath
F. Darema J. McGraw
D. DeGroot P. Messina
G. Fox C. Polychronopoulos
E. Gallopoulos J. Rice

T. Yuba (Chair) S. Nagashima
T. Hoshino H. Tanaka
Y. Kanada Y. Tanakura
H. Kashiwagi H. Terada
N. Koike K. Toda


ACM-SIGARCH with support from KFA in association with AICA, BCS-PPG, GI,


Charles Brownstein, NSF, U.S.A.
Presidential Initiative in High Performance Computing and Communications

Jean-Francois Omnes, EC Brussels, Belgium
The talk will cover European High-Performance Computing Strategies

John Riganati, Supercomputing Research Center, U.S.A.
Ideas in Supercomputing: Philosophy and Pragmatism

Klaus Schulten, University of Illinois, U.S.A.
Molecules to Networks - Biological Computing and Parallel Machines

James E. Smith, W. R. Taylor, Cray Research Inc., U.S.A.
Accurate Modelling of Interconnection Networks in Vector Supercomputers

Koichiro Tamura, ETL, Japan
Research Results of the Project "High Speed Computing System for
Scientific and Technological Uses"

Harry Wijshoff, University of Utrecht, Netherlands
Data Organization in Supercomputers: Practical Implications

Karl-Heinz A. Winkler, Los Alamos National Laboratory, U.S.A.
Simulation and Visualization of Supersonic Flows


David H. Bailey, NASA Ames, U.S.A.
Experience with Parallel Computers at NASA Ames

Hans-Juergen Herrmann, HLRZ, Germany
Stochastic Growth Models

Geerd R. Hoffmann, ECMWF, U.K.
Title to be announced

More Information:

For complete information about the conference, esp. the contributed
program and registration information, please contact :

Ruediger Esser
Research Center Juelich -ZAM-
P.O. Box 1913
D-5170 Juelich
Phone: +49-2461-61-6588
Fax: +49-2461-61-6656
E-mail: zdv003@djukfa11.bitnet


From: Danny Hershkowitz <MAR23AA%TECHNION@TAUNIVM.TAU.AC.IL>
Date: Fri, 29 Mar 91 10:13:22 IST
Subject: ILAS 108 - Workshop

E-mail Address: MAR23AA @ TECHNION (bitnet)
Edited by Danny Hershkowitz

29 March 1991

ILAS-NET Message No. 108

FROM: Patrick Dewilde
SUBJECT: Workshop on Advanced Algorithms

Bonas, France, July 28th-August 10th
1. Description
A two week workshop and course on advanced numerical algorithms and
their realization will be held in Bonas (Southern France) under the
auspices of the Network Theory Section of Delft University of Techno-
logy. It will be largely informal with courses in the morning and
workshop sessions in the afternoon, giving ample time for communication
between participants. Courses will treat: new methods and algorithms
in system theory, circuit theory for time-varying and non-linear com-
putations, problems in computational algebra, the synthesis of dedi-
cated array processors and applications in image processing, inverse
imaging, operations research and computer graphics. Lecturers are
presently being recruited and will be top researchers in the various
fields (they will include H. Dym, S. and R. Nandy, J. Zarzycki, E.
Deprettere, A.-J. van der Veen, M. Verhaegen, P. Dewilde).

2. Cost and Registration
The participation fee will be Dfl. (Dutch Guilders) 2600 for a single
person and Dfl. 4500 for a couple (present rate 1Dfl = 0.52$). The
fee basically covers lodging and full pension in the beautiful and
comfortable Chateau de Bonas located in the foothills of the Pyrenees.
The course itself, including course materials is free and covered by
a grant from JEMNA.

3.Preliminary program and topics
* Advanced system theory, models, mathematical properties, transforms
(including the new W-transform), computational models, embedding,
scattering and inverse scattering, optimal realization theory,
philosophical problems of a system-theoretical nature.
* Non-linear systems, their representation, Volterra series, their
* Modern computational methods in system and signal identification.
* Interpolation theory and its relation to system theory.
* Model theory for ODE's and PDE's, finite element theory, optimal model
reduction theory for large systems, examples using Maxwell's equations.
* Generalized AAK theory.
* Algorithmic description for parallel processing architectures:
parsing to dependence graphs, single assignment descriptions, generation
of distributed control, clustering and partitioning, division of tasks
between software and hardware, links to silicon compilation.
* Parallel processing and the design of parallel algorithms, examples:
Synthetic Aperture Radar, Direction of Arrival, Acoustic Camera, Computer
Graphics, System identification and Robotics, solving systems of equations,
eigenvalue and singular value problems.
The workshops will concentrate on research problems arising in the course
of the lectures and on research problems proposed by participants. They
will be divided in two sets: algorithmic problems and synthesis problems
which will be intertwined.

4. Workshop format
The workshop will be kept informal in all respects. The purpose will
not be to put pressure on the participants, but to create an atmosphere
of creative thinking and discourse. To smoot the procedures of the
workshop, the day organizer will collect propositions for problem speci-
fication and discussion. A schedule will be made on a day to day basis.
Insight and solutions will be reached by suggestions and discussions,
with ample participation of the audience. The final program will also
be constituted in this way: problem rather than topic oriented.

5. Who can attend?
The number of participants is limited to 30 people actively engaged in
research in the area of interest. Participation is by invitation only,
the aim being to constitute a reasonably homogeneous group of lecturers
and students capable of interacting at research level. Prospective
participants are kindly asked to send or fax application data as
detailed under point 6. A final program and further information will be
sent to selected invitees. If necessary, please include a Curriculum
Vitae and/or a letter of reference. Applications will be processed on a
first come, first served basis!

For more information, contact the workshop secretary: Dr. Michel Verhaegen,
Faculty of Electrical Engineering, POB 5031, 2600GA, DELFT, the Netherlands,
FAX: 31-15-623671.


From: John M. Gary <>
Date: Fri, 29 Mar 91 08:32:08 MST
Subject: Position at NIST in Boulder, CO

Position at NIST in Boulder, Colo.

We have a numerical analysis position open at the Boulder, Colorado
laboratory of the National Institute of Standards and Technology. The
position is within the Computational and Applied Mathematics Laboratory of
NIST and is appropriate for a PhD in applied mathematics or computer science
who has worked on the numerical solution of partial differential equations.
We would prefer an individual with a few years experience after the PhD,
but this is not required. The position involves collaboration and
consultation with scientists and engineers on numerical problems which
arise in the various divisions of the laboratory as well as independent
research in numerical analysis. The Boulder laboratory conducts research
in a wide range of disciplines which may involve mathematical models
including electromagnetics, fiber optics, elasticity, fluid flow, and
chemical reactions.

The available computing facilities range from a variety of workstations
(Sun, IBM, SGI) up to a two-processor Cray Y-MP. The Boulder laboratory
employs around 450 persons. The larger part of NIST is located in
Gaithersburg, Maryland. This is a civil service position which requires
US citizenship.

For additional information contact John Gary by phone at
(303)497-3369, E-mail at, or mail to
John Gary
NIST, mail stop 881
325 Broadway
Boulder, CO 80303 .


From: Bob Russell <>
Date: 29 Mar 91 14:19 -0800
Subject: Job Opening at Simon Fraser University

The following job opening has just come up. If you'd like to
inquire informally over e-mail about it, I'd be pleased to
answer any questions. Thanks, Bob Russell (na.brussell@...)


The Department of Mathematics and Statistics of Simon
Fraser University invites applications for a tenure track
position in Applied Mathematics at the Assistant Professor
level starting September 1, 1991 or as soon thereafter as
feasible (this position is subject to final budget appro-
val). Applicants will be expected to have completed a Ph.D.
degree at the time of appointment and to have demonstrated a
strong teaching and research potential. The Department is
particularly interested in applicants whose research
interests are in the areas of Solid or Fluid Mechanics or
Partial Differential Equations, preferably with involvement
in the computational aspects of these areas. However,
applicants in other areas of Applied and Computational
Mathematics are also encouraged to apply.

Applications, including curriculum vitae, should be sent to:

Dr. A.R. Freedman, Chair
Department of Mathematics and Statistics
Simon Fraser University
Burnaby, BC V5A 1S6

Please arrange for three letters of reference to be
sent directly from the referees.

The successful applicant will join a strong, active
group of twelve applied mathematicians working in a wide
range of applied areas including Continuum Mechanics, Dif-
ferential Equations, Elasticity, Fluid Dynamics, General
Relativity, Nonlinear Dynamics and Numerical Analysis. The
Department offers a Bachelor of Science Program in Applied
Mathematics as well as a structured graduate program in
Applied and Computational Mathematics leading to the Master
of Science and Doctor of Philosophy degrees. The Department
has excellent distributed computing facilities.

The University is situated on a mountaintop in Burnaby,
British Columbia, from which it commands striking vistas of
mountains and Burrard Inlet to the north, the metropolis of
Vancouver to the west and the Fraser River delta and valley
to the south and east.

Simon Fraser University is committed to the principle
of equity in employment and offers equal employment oppor-
tunities to all qualified applicants. In accordance with
Canadian immigration requirements, this advertisement is
directed to those who are eligible at the time of the appli-
cation for employment in Canada.


End of NA Digest