NA Digest Sunday, July 5, 1992 Volume 92 : Issue 26

Today's Editor:

Cleve Moler
The MathWorks, Inc.

Submissions for NA Digest:

Mail to

Information about NA-NET:

Mail to


From: Zvika Bar-Deroma <AER7101%TECHNION.bitnet@TAUNIVM.TAU.AC.IL>
Date: Sun, 05 Jul 92 22:28:49 IST
Subject: Seeking a Temporary Position in Rhode Island Area

I will spend a year starting the coming September in Providence, RI, and
I am looking for a position in Providence or in the vicinity. I would
like to work in CFD or numerical analysis, as computer professional or
as a teaching assistant. My experience includes work in CFD (viscous
subsonic), computer work (system management in UNIX, VAX/VMS and work as
a consultant) and teaching experience (fluid dynamics and programming).
I will send further information upon request.

I may be contacted at the following address:

Zvi Bar-Deroma
Dept. of Aerospace Eng.
Technion City
Haifa 32000

Fax: +972 4 231848
Phone: +972 4 292706


From: Pamela Burrage <>
Date: Mon, 29 Jun 92 09:56:31 +1000
Subject: MasPar Software

At the University of Queensland (Australia) we have a MasPar MP-1, with
4096 processors. On behalf of the other users here, I would like to
collate information on any software available for the MasPar (public
domain or otherwise). If anyone has such software details, could they
please send them to me.

Thank you.

Pamela Burrage, Centre for Information Technology Research, U of Q.


From: Van Snyder <vsnyder@math.Jpl.Nasa.Gov>
Date: Mon, 29 Jun 92 10:31:06 PDT
Subject: From the _Mathematical_Enquirer_

I found the following in rec.humor on the internet news. It was
produced by several individuals. I don't deserve any credit.
-- Van


You favourite power series reveals your personality!

My life with Moebius: TV shows only one side.

Fermat speaks from the grave: "I missed a factor of two --
hope it's been no inconvenience!"

Was Ramanujan an ET????

Strange attractors stole my wife!

Top psychics predict: Riemann hypothesis, NSF funding, more!

Catastrophe in the royal family: Is Lady Di a butterfly or a cusp?

Trapped in Klein Bottle, Man Swallows Own Mouth and Lives!

Astonishing photos of Ellipse with 3 foci discovered!

Chebyshev can't spell own name!

I computed cohomology groups for the CIA and KGB!


From: W. Govaerts <>
Date: Tue, 30 Jun 92 09:38 N
Subject: Software for Bordered Matrices

In many computational problems linear systems with
coefficient matrix in block form
( A B )
M = ( )
( C^T D )

arise naturally. Here A is n by n, B and C are n by m
and D is m by m; C^T is the transpose of C. If m is much
smaller than n, M is called a bordered matrix. In this
case A often has a special structure (symmetric, banded,
block banded, sparse...) that is not present in M. This
structure of A may make it relatively easy to construct
a solver for A. One then wants to use this solver to
solve linear systems with M. This can be done by block
elimination methods in several ways; unfortunately
stability may be lost in passing from a solver for A to
a solver for M. This is the case especially if A is
ill - conditioned while M is reasonably well- conditioned.
(In the computation of bifurcation points in numerical
continuation and bifurcation A is often theoretically
singular; roundoff and truncation only make A slightly
We want to advertize here a collection of Fortran subroutines
to help remedy this situation. The code was developed by
W. Govaerts (Gent,Belgium) and J. D. Pryce (RMCS,England)
and called BEMW (Block Elimination Mixed for Wider Borders).
It is a reverse communication code with modest size:
less than 16k including an example
program, full documentation and 3 BLAS core subroutines.
It is available now in Fortran single precision.
The calling program has to provide solvers for both A and
A^T but the use of the subroutrines requires no reference
to the elements of A or of the solution procedure. The
matrices B, C, D have to be stored in a specific but
natural way. The first (factorization) part of an application
of BEMW requires m solves with A and m solves with A^T.
The second (solution) part takes as many applications of the
solver with A (respectively A^T) as there are right hand side
vectors to be solved with M (respectively M^T). (In
typical applications solutions with M^T as well as with
M are necessary).
Theory and numerical examples are given by W. Govaerts
(Stable solvers and block elimination for bordered systems,
SIAM J. Matrix Anal. Appl. 12 (1991) 469 - 483) and
W. Govaerts and J. D. Pryce (Mixed block elimination for
linear system with wider borders, to appear in IMA J.
numer. anal.). For practical purposes, it is enough to know
that we expect that BEMW will solve linear systems with
M and M^T in a (backward) stable way provided that the
solvers for A and A^T are themselves stable. In particular,
systems with a well-conditioned M are solved accurately
even if A is arbitrarily ill-conditioned.
We had very satisfactory results about this case in
numerical continuation and bifurcation problems.
People interested in BEMW should send a short note to We shall of course be happy
to hear about their experiences.


From: Susan Ostrouchov <>
Date: Thu, 2 Jul 92 16:29:32 -0400
Subject: LAPACK Update


We have decided on the following proposed maintenance schedule for LAPACK:

LAPACK will be revised through "minor" and "major" updates.
The "minor" updates will occur either bimonthly or quarterly and will
consist of code and commenting changes. The first such update occurred
June 30, 1992. These minor updates will be denoted by a change in the
date and version number of an affected routine. The version number will
be changed from 1.0 to 1.0a and so on.

The "major" updates will occur yearly to introduce new routines
and whatever wasn't caught by the "minor" updates. For these "major"
updates the date on the affected routine will be changed as well
as the version number being incremented from 1.0 to 1.1 and so on.

NAG will receive a new tape yearly at the time of each "major" update.

Updating the version numbers from 1.0 to 2.0 and so on will be reserved
to correspond with new versions of the LAPACK Users' Guide. In other
words, a couple of years down the road.

The source for LAPACK will be updated on netlib at the time of the
"minor" and "major" updates. In addition, via xnetlib, there will be a
compressed/tar file of only the most recent changes that you can obtain
and simply untar on top of your existing LAPACK directory. The entire
package in lapack.tar.Z will always be the most up-to-date, and include
all revisions.

We are also maintaining ERRATA lists for the LAPACK Users' Guide and the
LAPACK source code. These ERRATA lists are contained in the release_notes
for LAPACK. This file may be obtained by sending email to
and in the message type: send release_notes from lapack.

LAPACK, VERSION 1.0a, June 30, 1992


Our first such "minor" update occurred June 30, 1992. All affected
routines have a version number of 1.0a and a date of June 30, 1992.
This update encompassed many minor corrections to documentation,
argument checking and so on. In addition, a few significant errors
have been corrected in certain routines (CHEGST, CHSEQR, SSTEBZ and
their double precision equivalents). One routine changed its name
(CLANST --> CLANHT) to avoid confusion in notation. And finally,
the makefile in LAPACK/TIMING will now do SMALL or LARGE data set
timings. By default, typing 'make' will do the SMALL timings, or if
you type 'make large' it will do only the LARGE timings. Check the
errata lists in the release_notes for more information.

Via xnetlib, there is a tar file of changes that you can obtain and
simply untar on top of your existing LAPACK directory. The file is
called revisions1.0a.tar.Z. The release_notes file contains details of
the changes made to specific routines. The entire package available
in lapack.tar.Z includes all revisions made in release 1.0a.

Additional features introduced with version 1.0a of LAPACK:

-- on-line manpages for LAPACK driver and computational routines

-- on-line LINPACK and EISPACK to LAPACK equivalence information

The manpages are available as a compressed/tar file of troff files
and the equivalence program and may be obtained via xnetlib. A README
file with instructions is included with the files.

If you have any questions/comments please contact us at

Jack Dongarra
Susan Ostrouchov


From: Edson Costa de Barros Carvalho Filho <>
Date: Wed, 1 Jul 92 12:39:26 EST
Subject: Braziliam Symposium on Artificial Intelligence

Rio de Janeiro, 5th - 8th October 1992

DEADLINE 31st July 1992
Final Call for Papers

The ninth Brazilian Symposium on Artificial Intelligence, SBIA'92, will
be held in Rio de Janeiro on October 1992, in conjunction with the
congress of the Brazilian Society for Computing Science (SBC).

This symposium is a promotion of the SBC and its goal is to communicate
the national and international scientific production on artificial
intelligence, and to provide an environment to exchange ideias and
experiences among professional, researchers, and students.


* Technical Sessions
* Poster Sessions
* Invited Talks
* New Product Demonstrations


Knowledge Acquisition and Representation
Philosophy and Artificial Intelligence
Expert Systems and Applications
Perception and Machine Vision
Natural Language Processing
Logic and Formal Systems
Intelligent Interfaces
Neural Networks


Sociedade Brasileira de Computacao - SBIA'92
Av. Venceslau Bras, 71 fundos casa 27
22290 Rio de Janeiro - RJ
Phone (021) 2954846 or (021) 2954442
Fax (021) 5415342


Technical Programme Chairman:

Prof. Edson C de Barros Carvalho Filho
Universidade Federal de Pernambuco
Departamento de Informatica
50739 - Recife - PE
Phone (081) 2718430 or (081) 2713052
email: or


Gentil Lucena, UNB
Guilherme Bittencourt, INPE
Marcos Mota Costa, EMBRAPA-UFPE
Mario Benevides, UFRJ-COPPE
Newton Viera, UFMG
Rosa Viccari, UFRGS
Tarcisio Pequeno, UFCE


From: G.A. Anastassiou <>
Date: 30 Jun 92 15:28:44 CDT
Subject: Contents: Journal of Computational and Applied Mathematics

Journal of Computational And Applied Mathematics
Applied Analysis and Stochastics
Guest Editors: G.A. Anastassiou & S.T. Rachev


G.A. Anastassiou and S.T. Rachev

B.C. Arnold
Logistic and semi-logistic processes [CAM 1212]

J. Boos and H. Tietz
Convexity theorems for the circle methods of summability [CAM 1215]

M. Bramson and C. Neuhauser
A catalytic surface reaction model [CAM 1211]

A. Gross and J.B. Robertson
Ergodic properties of stationary Poisson sequences [CAM 1210]

G. Hinsen
Explicit irregular sampling formulas [CAM 1213]

S. Jansche and R.L. Stens
Best weighted polynomial approximation on the real line; a
functional-analytic approach [CAM 1233]

P.R. Masani
The measure-theoretic aspects of entropy, Part I [CAM 1208]

M.I. Taksar
Skorohod problems with nonsmooth boundary conditions [CAM 1237]


End of NA Digest