NA Digest Sunday, May 26, 1991 Volume 91 : Issue 21

Today's Editor: Cleve Moler

Today's Topics:


From: Jack Dongarra <>
Date: Fri, 24 May 91 14:28:25 -0400
Subject: NA Net Whitepages Facility

NA Net Adds Whitepages Facility

The "whitepages" is a new feature which has been added to the NANET.
This facility is available to everyone. However, we would especially
encourage members of the NANET to join. Like any directory service,
it will only work if everybody takes a few moments to join, and then as
necessary updates their entry. Below is a short description of the
whitepages, and how to use it.


1). Querying the Database
To find out information about a person send mail to
"". In the message body specify
their firstname, lastname, or both. See Appendix B for an
example of a query.

2). Joining the Whitepages
To join the whitepages send mail to "".
In the message body specify the two mandatory fields and as many of
the optional fields as you want.


Optional Fields

The fields can be specified in any order. The subject line of your
message is ignored. All fields are entered into the database as
characters. So spaces can be used for readability. In the rare exception
that your first and last name combination is not unique, send mail to
"". We will manually insert your name into
the database despite the duplication. This should not cause
any problems for people querying the database for it is set up to
return ALL people with a given last name, first name, or combination.
A message will be sent back to you confirming the operation was
successful. See Appendix B for an example of joining the database.

3). Removing your entry
To remove your entry from the NA-NET database send mail to
"". In the message body specify the
following two fields.


The values can be specified in any order. Again the subject line of your
message is ignored. For security purpose, a confirmation message will be
sent to both the address requesting the removal, and also to the address
listed in the whitepages database. See Appendix B for an example of
removing your entry.

4). Changing fields
To change the value of a field or add to new field - send mail to
"". In the message body specify the
following two fields plus the fields to be added or changed.


The values can be specified in any order. Again the subject line
of your message is ignored. You cannot change your name. If you
wish to do so, first remove your entry and then rejoin with the new
name. If you wish to clear the value of a field, simply include the
field with no value. A message will be sent back to you confirming
the operation was successful. See Appendix B for an example on how
to change or add a field.

Below are some real examples to give you an idea of how the whitepages works.

1). For example, to find out more about Bill Rosener

mail to:
Subject: anything

Bill Rosener
mail to:
Subject: anything

Rosener Bill
mail to:
Subject: anything

whois Bill Rosener
mail to:
Subject: anything

whois Rosener Bill

A message similar to the following will then be returned.
Last Name: Rosener
First Name: Bill
Middle Name: Joseph
Affiliation: University of Tennessee
Office Address: Department of Computer Science - 107 Ayres Hall
City State Zip: Knoxville, TN 37996-1301
Country: U.S.A.
Office Phone: (615) 974-3647
Research: Highly-interactive interfaces for 3D graphics
Home Address: 1403 White Ave. Knoxville TN, 37916
Fax: 615-974-8296
E-mail Address:
Other: Feel free to call me from 8 am until 11 pm.

2). For example, to find out more information on all people
whose last name is "smith".

mail to:
Subject: anything


3). An example of how Bill Rosener would join the whitepages.

Subject: anything

Last_name: Rosener
First_name: Bill
Middle_name: Joseph
Affiliation: University of Tennessee
Office_address: Department of Computer Science - 107 Ayres Hall
City_state_zip: Knoxville, TN 37996-1301
Country: U.S.A.
Office_phone: (615) 974-3647
Research: Highly-interactive interfaces for 3D graphics
Home_address: 1403 White Ave. Knoxville TN, 37916
Other: Feel free to call me from 8 am until 11 pm.

Note; The optional fields "Fax" and "Home_phone" were not given.
So people querying Bill Rosener will not find this information.

4). An example of how Mike Foobar would remove his entry from the whitepages.

Subject: anything

Last_name: foobar
First_name: mike

5). An example of how Mike Fox would change his E-mail address and
Fax number while also clearing his Home Phone number in the whitepages.

Subject: anything

Last_name: fox
First_name: mike
E_mail_address: fox@new.address
Fax: (764) 285-4239

Jack Dongarra
University of Tennessee and
Oak Ridge National Laboratory


From: Gene Golub <golub@umiacs.UMD.EDU>
Date: Sun, 26 May 91 10:07:51 -0400
Subject: Housing at ICIAM meeting

Many of you are probably shocked by the high cost of the conference hotel
at the ICIAM meeting in Washington ( $99 per night + taxes). There are
several hotels nearby where the summer rate is about $75 per night.
Advertisements for these hotels can be found in the Sunday NY Times.

Gene Golub


From: Robert D. Silverman <
David L. Elliott <delliott@CEC2.WUSTL.EDU>
Jim Roth <>
Date: Thu, 23 May 1991 01:49:49 GMT
Subject: D. H. Lehmer

It is with a very great deal of sadness that I must report that D. H. Lehmer
passed away last night.

I am deeply saddened by this loss as Professor Lehmer's work has been a
source of personal inspiration for me. I consider him to be the father
of computational number theory; my major field of interest.

Bob Silverman
Mitre Corporation

Professor Lehmer worked on (among other things) the solution of sets of linear
Diophantine equations via Chinese Remainder Theorem. This involved building
a special-purpose computer which had many "delay lines" (recirculating loops)
of prime-integer length. When the numbers in all of the loops lined up (as in
a Las Vegas slot machine hitting a jackpot) a bell would ring and the data
would be dumped and printed out, then re-loaded into the loops. The first of
these machines was purely mechanical; later on (1965?) he bought surplus
radar delay lines and worked out a purely electronic number-theory machine
along the above lines, which he talked about at UCLA. He claimed that his
technique got answers hundreds of times faster than an IBM 704 would have.
Does anyone know whether he ever got a special VLSI chip built for
such purposes?

David L. Elliott
Washington University

There is a photograph and explanation of Dr. Lehmer's factoring machine
in the Dover book "Recreations in the Theory of Numbers" by Albert Beiler.
This machine had many gear driven wheels which would be halted when sets of
holes in them lined up, triggered by a photocell. It was called a
photoelectric number sieve.

Apparently it was an amazing, though tempermental, device, and was
inadvertantly "jammed" by a local Ham radio operator whose transmissions
triggered the photocell at times.

The book also mentions Lehmer's first factoring machine - a bicycle
sprocket and chain contraption!

Will future generations look back similarly at our current massively
parallel computers doing quantum chromodynamics calculations?

Jim Roth
Digital Equipment Corporation


From: Gregory Grefenstette <>
Date: 23 May 91 14:13:50 GMT
Subject: Do Approximate Methods for Singular Value Decomposition Exist?

Singular Value Decomposition on large matrices is time-consuming.
If you don't need the exact values of the singular values
produced, can they be approximated with some sort of
confidence in a time faster than O(n^3)?

Gregory Grefenstette
Computer Science Dept., Univ. of Pittsburgh


From: David West <>
Date: 23 May 91 16:25:25 GMT
Subject: Separable Least Squares: Seeking Fastest Method

The subject line contains the essence of this request; the following
is mere background.

I'm aware of the Golub-Pereyra and Kaufman versions of variable
projection, but my prospective application requires extreme speed and
does not benefit from the high precision of the orthogonal
transformation methods, since it is well-conditioned and has
low-precision data.

It therefore seems that I should use Kaufman's method (since it
would appear to require less computation than the Golub-Pereyra
method, other things being equal), but seek a version that can save
the factor of two in operation count by which the normal equations
beat Householder transformations for linear least squares.

This appears to give (at least) two possibilities:

1) a normal-equations version of Kaufman's method;
2) a sqrt-free Givens version of Kaufman's method.

I've never seen either of these in print; can anyone give me
pointers? Code would be even better.

David West
Industrial Technology Institute


From: E. Hairer <>
Date: Wed, 22 May 1991 17:04:00 +0200
Subject: Volume II on Stiff and Differential-Algebraic Problems

We are happy to announce that our book
Solving Ordinary Differential Equations, Volume II
Stiff and Differential-Algebraic Problems
(Springer Series in Comp. Math., Vol. 14)
Springer Verlag 1991, ISBN 3-540-53775-9
will appear in July 1991.
E. Hairer and G. Wanner, Geneva


From: Dave Carlson <>
Date: Mon, 20 May 91 10:26:03 PDT
Subject: ILAS Seed Grants


The International Linear Algebra Society (ILAS) announces its intention
to award two Seed Grants of $400 each for the 1991-1992 year for proposals
to improve the teaching of (primarily undergraduate) linear algebra.
The purpose of these awards is to encourage work in this area by
providing funding to assist individuals in preparing grant proposals
and thereby to indicate to granting agencies the support of the linear
algebra community for these proposals.
PROCEDURES: To apply for a 1991-1992 ILAS Seed Grant, an individual should
submit to

Professor Frank Uhlig
Mathematics Department
120 Math Annex
Auburn University
Auburn, AL 36849-5307
phones: office 205-844-3641, FAX 205-844-5900, home 205-887-8302

a brief (2-5 pages) outline of the grant proposal under preparation, including
(1) activities proposed for support in the overall proposal,
(2) activities proposed for support by the ILAS Seed Grant,
(3) site(s) of the proposed activities,
(4) vita of principal investigator(s), including linear algebra teaching
(5) expected granting agency and program,
(6) approximate budget.

Any individual(s) from any country with experience in teaching
linear algebra may apply, whether ILAS member or not. We encourage
applications from outside the U.S.
The selection of the awardees will be made by the ILAS President upon the
recommendation of the ILAS Education Committee.
The deadline for applications for the first 1991-1992 ILAS Seed Grant will
be July 1, 1991; this award will be announced by August 1, 1991. The dead-
line for the second award will be January 1, 1992; this award will be
announced February 1, 1992.
Each awardee is asked to send to Professor Uhlig a copy of the grant proposal
for which they have received an ILAS Seed Grant. ILAS may use these grant
proposals in preparing lists of possible eduational activities in linear
OTHER INFORMATION: Individuals wishing more information about possible grant
activites and procedures before applying for an ILAS Seed Grant may contact
Professor Uhlig or either of the other members of the ILAS Education

Professor Dave Carlson
Mathematical Sciences Department
San Diego State University
San Diego, CA 92182-0314
phones: office 619-594-6670, FAX 619-594-6746, home 619-488-2864

Professor Steve Leon
Mathematics Department
Southeastern Massachusetts University
North Dartmouth, MA 02747
phones: office 508-999-8320, home 508-992-5757

The U.S. programs which seem most appropriate are
(1) NSF Undergraduate Curriculum Development in Mathematics: Calculus.
This program is headed by John S. (Spud) Bradley; its next deadline
will be early October. (By "calculus" read "calculus and the first two
years". This is the program through which the Williamsburg Workshop was
funded. It is envisioned that, possibly as part of a broader effort, this
program will continue for a long time.)
(2) FIPSE (The Fund for the Improvement of Postsecondary Education, in
the U.S. Department of Education). This program's next deadline is October
16. FIPSE funds both undergraduate and graduate proposals in a broad set
of areas; it "indicates general problems...and invites applicants to address
these problems imaginatively".
(3) NSF Faculty Enhancement. This program is headed by William Haver;
its next deadline will be in April 1992.
Please note that NSF cannot award grants to non-American institiutions,
although non-American scientists may receive NSF funds under grant to U.S.


From: John Mason <>
Date: Tue, 21 MAY 91 13:48:05 BST
Subject: Student Position at Shrivehnam and NPL


A PhD student is needed to work with Prof John Mason at Royal Military
College of Science in Shrivenham (a faculty of Cranfield), in cooperation
with Dr Maurice Cox of National Physical Laboratory, starting as soon as
available (e.g. September/October). An additional payment of #1850 per year
is offered to supplement the 3-year SERC grant, which brings the basic total
annual payment to about #6000 (away from home).
The project involves developing algorithms for approximating structured
data, such as data whose abscissae lie on families of curves. The work will
centre on spline and polynomial methods, convergence and approximation
questions will be considered, and computer software will be developed for use
at National Physical Laboratory.
The Applied and Computational Maths Group at Shrivenham has a strong
academic staff in numerical analysis/approximation and excellent facilities,
including a new network of Sun SPARCstations. Prof Mason and Dr Cox are
co-organisers of the Algorithms for Approximation series of conferences, the
next being in July 1992
Applicants (expecting upper 2nd class), with a background in numerical
maths or approximation and some computing experience, should contact a.s.a.p
Prof J.C.Mason, ACM Group, RMCS, Shrivenham, Swindon SN6 8LA (England).
Tel: 0793 785311 Email: mason@uk.cran.rmcs


From: Heikki Apiola <>
Date: 20 May 91 15:18:44 GMT
Subject: Hensinki Workshop on Symbolic and Numeric Computation

Workshop on Symbolic and Numeric Computation
Helsinki 30. - 31.5. 1991


Helsinki University Computing Centre
The Rolf Nevanlinna-Institute
CSC (Centre for Scientific Computation)


The workshop aims to bring together different computational cultures
represented by the symbol algebra, numerical and graphical software and
matrix-vector language communities. We will try to provide a forum for
exchanging ideas and introducing projects that develop or use some
combination of such tools.


Helsinki University Computing Centre
Teollisuuskatu 23
SF-00510 Helsinki
Tel: +358-0-70851
Fax: +358-0-7084441


Heikki Apiola
Juha Fagerholm
Lasse Holmstr|m
Marko Laine
Esko Valkeila


James Davenport
Jarmo Hietarinta
Lasse Holmstr|m
Ari Lehtonen
Mika Sepp{l{
Victor Kistlerov
Ken Rimey
Norman Thomson
Richard Eller
Ari Lehtonen
Juha Haataja
Norman Thomson
Heikki Apiola
Heikki Haario
Marc Gaetano
Alexei Serebrovski
Esko Valkeila
Eero Korpelainen


Send your name, phone, (e-mail) address and other relevant information by
May 27 (prolonged dead-line) to the following address:

Paula M{ki-V{lkkil{ e-mail:
CSC/VTKK phone: +358-0-4572718
PO-BOX 40 telefax:+358-0-4572302
SF-02101 Espoo


There is no fee for academic participants. For industry the fee is 750 Fmk/day
The Finnish Meteorological Institute and the State Technical Research Centre
(VTT) are treated as academic institutions.


FinnAPL and the Helsinki University Computer Centre will arrange an APL-
afternoon on Wed. 29.5. at 14 - 16.30. The main speaker will be Norman
Thomson from IBM Winchester. He will give a talk on "Mathematical appli-
cations of APL2". There will be demonstrations of IBM's APL2 and STSC's
APL*PLUS on an IBM RS 6000 workstation and "an announcement of STSC's
software grant for some Finnish Universities"


From: John W. Baugh Jr. <taco!!>
Date: 22 May 91 14:04:06 GMT
Subject: Eighth Conference on Computing in Civil Engineering

Plans are underway for ASCE's Eighth Conference on Computing in Civil
Engineering, which will be held June 10-12, 1992 in Dallas, Texas
(500-word abstracts are due July 1, 1991). Although I do not
officially represent ASCE, I would like to encourage participation in
this conference, especially from those of you who are working in
PARALLEL and DISTRIBUTED PROGRAMMING (my own areas of interest). In
addition, I hope that researchers in other disciplines such as
mechanical engineering and numerical mathematics (who may be doing
parallel computing in areas like finite element analysis and numerical
optimization) will participate. Finally, I would be interested in
hearing from any of you who plan to submit papers or are otherwise
interested. See the call for papers below for more details.

Thank you,

John W. Baugh Jr., Assistant Professor Internet:
Department of Civil Engineering BITNET: jwb%cepmax@ncsuvm.bitnet
North Carolina State University UUCP: mcnc!cepmax!jwb
Raleigh, North Carolina 27695-7908 phone: (919) 737-7697

Call for Papers
American Society of Civil Engineers
Technical Council on Computer Practices
Eighth Conference on Computing in Civil Engineering
10-12 June, 1992
Dallas, TX

The eighth in the series of Conferences on Computing in Civil
Engineering will be held 10-12 June, 1992 in Dallas, TX, in
conjunction with A/E/C Systems '92. This series, initiated in 1978,
provides a continuing forum for the advancement of intelligent
computer use in the profession. The conference will bring together
diverse groups within the profession and those affiliated with it to
share knowledge, experience and techniques. The objective is to
disseminate current and projected developments in both software and
hardware and to educate users in the ways in which computers can be
applied to practice, administration and education.

This conference will include a "Symposium on Geographic
Information Analysis," as well as sessions describing computer use in
construction management, water resources, transportation,
geotechnical, structural and other aspects of civil engineering.
Papers are being sought that describe computer applications across all
aspects of Civil Engineering, including education, design, assessment,
and management. Computer applications in large and small engineering
firms as well as government and education will be considered. Papers
describing completed applications will be given priority, but quality
discussions of work-in-progress will also be considered. Papers
describing Geographic Information Analysis are particularly solicited.

Deadlines: 500-word Abstracts Due: July 1, 1991
Notification of Acceptance: October 1, 1991
Submission of Complete Papers: December 1, 1991
Conference: June 10, 1992

Submit Abstracts to:

Symposium on Geographic Eighth Conference on Computing in
Information Analysis Civil Engineering

Prof. Jeff Wright Dr. Barry Goodno
School of Civil Engineering School of Civil Engineering
Purdue University Georgia Tech
West Lafayette, IN 47907 Atlanta, GA 30332
Voice: 317 494 2175 Voice: 404 894 2227
FAX: 317 494 0395 FAX: 404 894 2278


From: Masaaki Shimasaki <>
Date: Thu, 23 May 91 14:44:09 JST
Subject: Supercomputing Symposium in Fukuoka

I n t e r n a t i o n a l S y m p o s i u m
o n
S u p e r c o m p u t i n g
6-8 November 1991, Fukuoka, JAPAN

Sponsored by : Kyushu University

Supported by : Information Processing Society of Japan
The Japan Society of Mechanical Engineers
The Japan Society for Industrial and
Applied Mathematics
(Application is being processed)
Japan Society for Software Science and
(Application is being processed)

Supercomputing is giving a significant influence on computer
architecture, systems software, algorithms and applications.
The aim of this symposium is to promote communications between
researchers of different disciplines, discuss current problems
and get proper insight into future of supercomputing. Papers
presenting original research and practical experiences in
supercomputing are sought. Authors should submit an extended
abstract in about 800 English words to be received by the
program chair no later than June 28, 1991. Authors will be
notified of acceptance or rejection by August 9, 1991. A final
copy of each accepted paper within 10 pages in the symposium
format will be due by September 20, 1991.

Suggested Topics :
Vector and Parallel Supercomputing
Performance Evaluation of Supercomputers and Benchmarking
Compilers and Optimization for Supercomputers
Numerical analysis in Supercomputing Environment
Supercomputing in Science and Engineering
Networking in Supercomputing Environment
Supercomputing in Nonnumeric Applications and Artificial Intelligence

Symposium Chair : Kazuo Ushijima (Kyushu University)

Program Chair : Masaaki Shimasaki
Computer Center, Kyushu University
Fukuoka 812, Japan
Tel: (092) 641-1101 ext.2507
Fax: (092) 631-3196

Program Committee : Masato Abe (Tohoku Univ.)
Jack Dongarra (Univ. of Tennessee
and Oak Ridge National Lab.)
Iain S Duff (Rutherford Appleton Laboratory
Yasuyo Hatano (Chukyo Univ.)
Satosi Hoshino (Kyoto Univ.)
Noriko Iida (NACSIS)
@ 9akumIsC Kanada (Univ. of Tokyo)
Sadatoshi Kumagai (Osaka Univ.)
Katsuhiko Miyoshi (Hokkaido Univ.)
Yoichi Muraoka (Waseda Univ.)
Tooru Nagai (Nagoya Univ.)
Yoshio Oyanagi (Univ. of Tokyo)
Aad van der Steen (ACCU)
Hidehiko Tanaka (Univ. of Tokyo)
Shinji Tomita (Kyoto Univ.)
Takao Tsuda (Kyoto Univ.)

Local Arrangement Chair : Norihiko Yoshida (Kyushu University)
Tetsuya Furukawa (Kyushu University)
Hiroyuki Sato (Kyushu University)

Important Dates : Deadline for submission - June 28, 1991
Notification - August 9, 1991
Final paper due - September 20, 1991
Symposium - November 6-8, 1991


From: David Keyes <keyes-david@CS.YALE.EDU>
Date: Wed, 22 May 91 13:32:53 EDT
Subject: Multigrid/Domain Decomposition Workshop


(submitted by David Keyes, with thanks to Olof Widlund)

Seventy-eight registered attendees from academia, the petroleum and
computer industries, and government research laboratories gathered at
Rice University on March 28 and 29, 1991 for a Workshop on Parallel
Multigrid and Domain Decomposition with Applications to Problems in
Porous Media. The workshop, organized by Professor Mary Wheeler of Rice
and sponsored by a variety of federal, state and industrial partners,
followed the Fifth SIAM Conference on Parallel Processing also held in
Houston. Wheeler had organized four hours of special sessions devoted
to computational oil reservoir modeling at the SIAM conference the
preceding day, creating a link between the two meetings for many

The SIAM speakers were primarily from the oil industry and set the stage
by focusing on the special difficulties of reservoir simulation such as
those caused by order-of-magnitude variability in rock permeabilities in
the Darcy pressure equation and irregularities in computational workload
due to wells and a variable number of phases present from point to point
in the domain of simulation. The workshop speakers were from
universities and government laboratories and focused on multigrid and
domain decomposition as elliptic solvers well-suited to reservoir models
and amenable to parallel implementation. Several speakers in each group
presented parallel performance data, primarily taken on Cray, Intel, and
Thinking Machines platforms. Typical elliptic systems encompassing
millions of unknowns are now solvable to physically reasonable precision
in tens of wall-clock seconds by iterative methods on such machines.

Wheeler expressed the hope that a new set of ``standard'' problems for
comparing algorithms would arise from the workshop. By contemporary
standards the canonical Society of Petroleum Engineers (SPE) test suite
of sparse matrices derived from the Implicit Pressure/Explicit Species
(IMPES) formulation of the reservoir physics is of rather low dimension
--- a few thousand degrees of freedom. Workhorse industrial algorithms
for the implicit pressure equation employ splittings or preconditioners
such as incomplete or nested factorizations whose parallelism is limited
and difficult to detect at compile time, particularly in nonuniformly
refined problems. Operator decompositions, hierarchical function-space
decompositions, and domain decompositions are natural means through
which to seek parallelism, and they have generated a considerable
research activity throughout the 1980's. The Copper Mountain series of
Multigrid conferences and the International Symposia on Domain
Decomposition Methods for PDEs have been focal points for much of this
activity, and the respective communities have begun to overlap. A small
joint workshop with a focus on applications was thus timely.

David Young of the University of Texas-Austin led off the workshop with
a look at iterative methods with operators expressible as rational
functions of matrices. Partial fraction expansions of the rational
functions are used to achieve coarse-grained parallelism.

Oliver McBryan of the University of Colorado-Boulder described Parallel
Superconvergent Multigrid, which is most suited for the opposite extreme
of massive parallelism. PSMG provides extremely fast convergence rates
for smooth problems. Its behavior on problems with extreme spatial
variability in the coefficients is to be investigated.

Semi-coarsening Multigrid was the subject of back-to-back talks by Alan
Weiser of Rice, who featured the Intel iPSC/2, and Joe Dendy of Los
Alamos, who featured the Connection Machine. Semi-coarsening is an
adaptation of multigrid to the numerical anisotropy of ``pancake''
shaped reservoirs in which standard multigrid is used in the vertical
direction only (the direction associated with the largest coefficients).
Full plane problems are solved in the horizontal directions, making the
strong coefficient discontinuities less troublesome. The main drawback
of these methods relative to isotropic multigrid is that the amount of
storage required for successively coarsened grids does not decrease as

Steve McCormick of the University of Colorado-Denver gave an
interpretation of the Asynchronous Fast Adaptive Composite (AFAC) Grid
Method as a nonlinear multilevel projection method and presented a
variety of applications in progress in and outside of PDEs.

Joe Pasciak of Brookhaven analyzed multilevel preconditioners for
elliptic problems utilizing refined meshes without relying on regularity
estimates for the original PDE.

Parallel implementations of mixed finite-element models of
three-dimensional reservoir problems, done jointly with Mary Wheeler,
were presented by Rice graduate student Lawrence Cowsar. In their
scalable hybrid algorithm, multigrid is employed to solve the system of
interface unknowns created by domain decomposition.

Noting that significant approximate information (analytical or
numerical) is often available in limited regions of complex problems, a
``LEGO block'' philosophy to constructing parallel domain-decomposed
preconditioners was espoused by David Keyes of Yale. This is joint work
with Bill Gropp of Argonne who also spoke, describing the MIMD
application of such an adaptive object-oriented technique to a
chemically reacting flow problem which is serving to pace its further

Olof Widlund of the Courant Institute described extensions of the
Schwarz decomposition framework to nonsymmetric and indefinite problems
(joint work with Xiao-Chuan Cai) and to multicomponent shell element
and scalar three-dimensional problems (joint work with Barry Smith).
Near mesh-independent convergence rates have been achieved for
high-Reynolds number transport and for some industrial applications of
linear elasticity.

Craig Douglas of IBM presented non-telescoping parallel multigrid
algorithms relying on symmetry groups of the underlying differential
operators rather than coarsening processes to derive multiple
computational tasks through whose combination the desired solution is

Returning to the theme of strong nonsymmetry due to convective
transport, Randy Bank of UCSD described a hierarchical basis multigrid
algorithm that uses the streamline diffusion discretization. The
streamline diffusion term that stabilizes the discretization also
stabilizes the multigrid iteration. Theoretical convergence results
retain a dependence on the fineness of the coarse grid, however.

Spectral element methods were presented as high-efficiency parallel
solvers for incompressible viscous flows by Paul Fischer of Caltech.
Characteristic-based time-stepping and domain-decomposed preconditioned
iterative solution of the elliptic part form the basis of this widely
applied and ported flow solver.

In a talk supplemented by animated graphics produced on a CM-2, Shi-Yi
Chen of Los Alamos demonstrated that lattice gas automata have found a
niche in the modeling of flows in porous media at the detailed
resolution of the pores themselves, in two and three dimensions.

Clint Dawson of Rice presented domain decomposition procedures for
parabolic problems in which the interfacial degrees of freedom are
advanced explicitly with a stability limit related to the subdomain
diameter, and the subdomains advanced implicitly and independently.
Theoretical estimates dictate a subdomain granularity at which the fine
and coarse grid errors balance.

Closing the two-day conference with a talk that embraced physics,
mathematics, and parallel computation, Herb Keller of Caltech described
the port of a three-dimensional incompressible code used to follow wavy
Taylor vortex bifurcations to the Intel iPSC/2, en route to the
528-processor Delta Machine recently delivered to Caltech.

An interesting addition to the workshop was a session of code
demonstrations in the facilities of the Rice Center for Research on
Parallel Computation (CRPC) on its eve, featuring informal interactive
tutorial sessions by Bank, Gropp, and Weiser.

A recurrent quip making reservoir simulation ``the mother of all
problems'' may be contested by those in other pace-setting application
areas of scientific computing, but it is clear that the combination
of multiple scales, coefficient inhomogeneity, three-dimensionality, and
geometrical complexity will keep theoreticians and practitioners busy
at least until oil and water mix.


End of NA Digest