NA Digest Sunday, May 7, 1989 Volume 89 : Issue 18

Today's Editor: Cleve Moler

Today's Topics:


From: Izzy Nelken <>
Date: Sun, 30 Apr 89 20:32:41 EDT
Subject: Fifth Parallel Circus

Review of the Fifth Parallel Circus
Izzy Nelken, Department of Computer Science, Rutgers University

The fifth Parallel Circus was held at Rensselaer Polytechnic Institute (RPI)
on Friday and Saturday, April 28th. and 29th. RPI is located in
Troy, NY, overlooking the Hudson river in beautiful Rensselaer county.
The Parallel Circus is an informal forum for presentations of up
to date research related to parallel computing. The Circus usually
takes place twice a year, each time at a different location. The first circus
was hosted by Martin Schultz at Yale University, the second took place
at Cornell University and was organized by Charles Van Loan. The
third and fourth meetings were held at IBM Kingston and at Rutgers University
and were organized by Vijay Sonnad and Apostolos Gerasoulis respectively.

This meeting was chaired by Joseph Flaherty and Bolek Szymansky of RPI
and Gene Golub of Stanford University. There were some forty attendees
and more than twenty talks. The atmosphere was informal and
participants were encouraged to ask questions and make comments
which sometimes led to lively discussions. The talks were wide
ranging in their subject matters. People spoke about hardware
considerations, implementations of various algorithms, theory of parallel
computation and compiling and scheduling issues.

P. Procacci, D. Frye and D. Schneider of IBM described
implementations of Chemistry and Physics applications. F. Saied
of Yale University talked about parallelizing the solution of the
Schrodinger equation via variants of the ADI method. Interesting
issues related to compiling were covered by J. Bruno from RPI
and J. Li and R. Mirchandaney from Yale University. W. Miranker
of IBM gave a talk about the solution of differential equations
using an associative memory computing paradigm. M. Johnson
of Bennet Laboratories showed video tapes of simulations of the
visual process in the brain.

There were several presentations of work done at RPI. E. Kaltofen,
T. Spencer and M. Goldberg gave theoretical talks about the
computations of matrix functions, time work tradeoffs and the
construction of an independent set of a graph. W.R. Franklin described
geometric algorithms used in determining line intersections and
P. Baehmann spoke about Transputers and Quadtrees. S. Hammond compared
static and dynamic scheduling techniques for the ICCG algorithm,
G. Shroff talked about the eigenproblem for a general complex
matrix and C. Norton described several neural network models.

A. Elster from Cornell University talked about fault tolerance issues and
I. Nelken of Rutgers University described several results concerning
the Gaussian elimination and Gauss Jordan algorithms. S. Foresti of
IBM spoke of multilevel finite elements. Last but not least on this list is
A. Greenbaum of the Courant Institute at New York University
who has graciously volunteered to host the next parallel circus.
She described her Ultracomputer implementation of a multipole method
which is related to the Trummer problem. Trummer's problem was posed by
Gene Golub and solved by Apostolos Gerasoulis.

The more social activities included a dinner at one of the
best local restaurants and a guided tour of the beautiful RPI
campus. We were shown the computer facilities which are located
in the renovated Voorhees Chapel. Apparently, the local sentiment
was against the destruction of the Chapel to make way for a computer
center so a compromise was found. One can walk into the Chapel, sit down
and work on a computer terminal.

Most of the participants were happy with the meeting and enjoyed the
talks. There were a few ideas about extending the Circus and meeting
also in the west coast. The conferences give exposure to students
and provide a meeting place for scientists interested in parallel computing.


From: Walter Gander <>
Date: 28 Apr 89 09:26:38+0200
Subject: Public Domain Software for Karmarkar Algorithm

Niklaus Wirth received the following letter, in which a PC
implementation of Karmarkar's Algorithm is sought. Since this
company wants to develop public domain software it seems to me
that we should help them and put their request in the na-digest.

-- Walter Gander

PC Implementation of Karmarkar's Algorithm

We are preparing to develop an implementation of Karmarkar's algorithm
on the IBM AT platform, to solve a specific formulation of large but
sparse LP problems. We are hoping to locate someone who already has
suitable pseudo or source code which we could interface with an
existing matrix generator and report writer system. Presently this
system interfaces with a Simplex LP solution, but it is far too
inefficient for the size of problem we are dealing with. The maximum
problem size we would like to address is 1000 rows by 8000 columns,
with a non-zero element density of about 3%. Shadow-price information
is not important for this application.

The software we will develop will be available in source form in the
public domain. We will fully credit the sources of code we may
incorporate, and will provide full details of solution strategies and
algorithms to anyone who has provided us with source code.

We would very much appreciate any contacts you may provide us in this
regard. FAX transmission may be sent to us by telephoning first at
(604)-361-9306. Otherwise please reply by express mail.

Yours truly

David Ormerod, Manager
303-535 Yates Street
Victoria, B.C.
Canada V8W 1K7
(604) 361-3290/9306


From: Martin Steenhuis <mcvax!hp4nl!tnoibbc!>
Date: 2 May 89 16:18:36 GMT
Subject: C Simplex Source Sought

Does anyone have the C-source to an implementation of a large-scale simplex-
routine based on the Bartels/Golub algoritm?

Any help is appreciated. Please reply by mail.

Thanks in advance. Martin Steenhuis

Martin Steenhuis : TNO - IBBC USENET : ms@tnoibbc
Expert System Group : PO-box 49 UUCP : ..!hp4nl!tnoibbc!ms
: 2600 AA Delft
: the Netherlands VOICE : +31 15 606293


From: Michelle Jones <>
Date: Wed, 3 May 89 11:49 EDT
Subject: SIAM Conference on Applied Probability


CONFERENCE DATE: March 5-7, 1990

TITLE: SIAM Conference on Applied Probability
in Science and Engineering
Cosponsored by the ORSA/TIMS Applied
Probability Group

ORGANIZERS: Bernard Matkowsky, Northwestern University
James McKenna, Bellcore
George Shanthikumar, University of California, Berkeley

ABSTRACT DEADLINE: September 27, 1989

CONTACT: SIAM Conference Coordinator
117 S. 17th Street, 14th Floor
Philadelphia, PA 19103-5052
(215) 564-2929
(FAX) (215) 564-4174

PLACE: New Orleans, LA

TOPICS: Probabilistic Analysis of Algorithms;
Applications of Probability in Biology;
Applications of Probability in Physics;
Queueing Theory; Stochastic Optimization;
Probability Theory and Classical Applied Mathematics;
Stochastic Scheduling; Applied Probability;
Stochastic Models in Manufacturing

SPEAKERS: David Aldous, University of California, Berkeley
Robert Eisenberg, Rush Medical College, Chicago
Richard Karp, University of California, Berkeley
Thomas Liggett, University of California, Berkeley
Bernard Matkowsky, Northwestern University
Richard Weber, Combridge University
George Weiss, National Institute of Health
Ward Whitt, AT&T Bell Labs.


From: Bob Chojnacki <>
Date: Wed, 3 May 89 08:22 CST
Subject: Public Domain IMSL?

I recently saw a posting on a different email list that requested a public
domain version of IMSL. Any of you who have IMSL or looked into in "renting"
it know the price of IMSL. Is there such a beast (public domain IMSL).
Another question, (since we will probably be getting a numeric library of one
sort or another), is how does IMSL compare to NAG? I realize libraries of this
calibre cost a bit and perhaps justifiably so. However, it is part of my job
to look into alternatives.

Bob Chojnacki
Center for Manufacturing Engineering/Department of Mechanical Engineering
Northwestern University Evanston, IL 60208 USA


From: John Lewis <>
Date: Thu, 4 May 89 08:39:34 PDT
Subject: Sparse Matrix Symposium -- Driving to Salishan

The following information was accidentally omitted from the final
mailing to participants. For those who choose not to drive
themselves, the SIAM buses are scheduled to leave the Portland
airport at 3:00pm, 5:00pm, 7:30pm and 9:30pm on Sunday, May 21.


The instructions given on the map supplied by Salishan Lodge describe
two routes, one using Highway 99W through Tigard, the alternative
using Highway 22 near Salem. The first of these is apparently more
scenic, offering a tour through the wine country of Oregon (and a
chance to stop and sample). From the map and the resort's
description, the second route appears to be distinctly a second choice
-- longer, with more freeway driving. In fact, the first route is 112
miles long and takes about 2.5 hours while the second is around 95
miles and takes about 2 hours. I was only able to drive the first
route, but some of my comments apply to the second route as well.

Note at the beginning that the large scale map has the Portland
Airport in the wrong location! As indicated in the more detailed
map, it is accessed from I(nterstate) 205, not I5. The drive from
the airport to the junction of I205 with I5 is about 23 miles. At
this point follow the Tigard exit (I5 North) for the first route, the
Salem exit (I5 South) for the second. For the first route, go north
on I5 just under 3 miles to Exit 292 (marked variously Sunset
Highway, Ocean Beaches, Tigard and Beaverton, and Highway 217W).
After two miles on this road, take the exit for Tigard/McMinnville. A
left turn at the end of the exit ramp takes you in the correct
direction, with about four miles of urban driving before you enter
the countryside. The mileages in the table below are total distances
from the airport.

Tigard/McMinnville exit (99W): 28.0
Newberg: 41.1
Dundee 45.0
Highway 18 (McMinnville bypass) 49.3
Sheridan 67.7
Valley Junction (Highway 22) 78.5
Highway 101 (South to Lincoln City) 101.5
enter Lincoln City 103.1

Highway 101 passes through Lincoln City for about six miles.
Salishan Lodge is about three miles further south, 112 miles from the
airport. The entrance to the resort is easy to find: A left turn at
an isolated traffic light, well marked by several quite large and
lighted signs on the left side of the highway.


From: Iain Duff <>
Date: Fri, 5 May 89 05:01:13 CDT
Subject: Positions at CERFACS

Recruitment of PhDs and post-docs.


The European Centre for Research and Advanced Training in
Scientific Computation (CERFACS) in Toulouse, France is continuing
to expand. A call for candidates to join CERFACS at or around
October 1989 at either the post-doctorate or post-graduate level
has just been made. The PhD grant (initially for two years) is
10590 FFr per month, and the post-doctorate grant (annually
renewable) is for 14290 FFr per month. The cost of living in the
Toulouse region is fairly low and, depending on one's tastes, it
should be possible to live fairly well on these salaries. It is
planned to award up to ten grants. There are presently around
forty research workers at the Centre.
There is no set application form. Applications should be made to
CERFACS, 42 Ave G Coriolis, 31057 Toulouse Cedex, France. (Tel
+33-61-07-96-96, FAX +33-61-07-96-13, email cros@frtls12.bitnet)
by June 15th 1989, preferably including a CV and an indication of
the main interests of the candidate. e-mail can also be sent to me
(at although I would prefer if you
simultaneously sent the same communication directly to CERFACS. I
am happy to send further information on CERFACS to any interested
There are four teams at CERFACS, in parallel algorithm
development and numerical computation, instability and turbulence,
aerodynamic flows, and visualization methods for computational
fluid mechanics.
Currently the main machines in the Centre are an ETA-10P and two
Alliant FX/80s, an Encore Multimax, a Gould NP1, and an IBM 9370.
There is good access to an IBM 3090/VF (currently 4 processor ..
to be upgraded to six), a 4-processor CRAY-2, and a CRAY X-MP. It
is planned to install a more massively parallel machine in the
near future. The working environment is based on workstations
(mostly Suns).
There are also possibilities of funding visits from more
senior researchers. It is easiest if interested parties contact
me directly.

Iain Duff
Harwell and CERFACS


From: W. A. Light <>
Date: Fri, 5 May 89 14:07:39 -0100
Subject: Positions at University of Lancaster

Applicants should have research interests in some area of numerical analysis.
The department has recently become part of a new school of Computing,
Engineering and Matheamtical Sciences and will particularly welcome applications
from candidates with interests in an area relevant to the other departments in
the School.

Applicants should possess a strong research record in numerical analysis, and
should also have ability in teaching.

Closing date.....21st June 1989.

E-mail W. A. Light or write Department of Mathematics, University of Lancaster,
Lancaster, La1 4FY, England for further information about the post and the


End of NA Digest