NA Digest Saturday, August 31, 1991 Volume 91 : Issue 35

Today's Editor: Cleve Moler

Today's Topics:

Submissions for NA Digest:

Mail to

Information about NA-NET:

Mail to


From: David Goldberg <>
Date: Mon, 26 Aug 1991 12:21:10 PDT
Subject: SLI vs IEEE arithmetic

Lozier, Olver and Turner gave two references explaining the
advantages of their SLI system. It seems only fair to mention
a paper that discusses some potential probems with SLI:

"On Error Analysis in Arithmetic with Varying Relative Precision"
by Jim Demmel, Proc 8th Symposium on Computer Arithmetic, pp 148-152.

David Goldberg,


From: Daniel Woods <>
Date: Wed, 28 Aug 91 16:43:13 PDT
Subject: C Language Interface to the BLAS

I am looking for a C language interface to the BLAS (all levels),
preferably one which would call the Fortran BLAS. Failing this,
I welcome any advice on this issue.

At first glance it appears to be a straightforward implementation
until you realize that C does not have complex and double complex
data types or arithmetic... Also the function CDOT returns a
complex number. Do I need to restrict C users to complex numbers
being stored as a 2-element array (to match the structure used in
Fortran implementations)? Do all Fortrans store complex numbers
as a pair of floating point numbers? As you can see, there are

If there is an interest, and I get sufficient response, I will post
a summary.

Daniel Woods


Date: Tue, 27 Aug 91 14:53:02 EST
Subject: Restoration of Address for Cecilia Jea

I have enjoyed my one year leave to Center for Numerical Analysis at U.
T. Austin. Now, I am back in Taiwan:

Cecilia Jea (or Kang C. Jea)
Department of Mathematics
Fu Jen University
Taipei, Taiwan 24205
R. O. C.

e-mail: fjut006@twnmoe10.bitnet
tel. : (886-2) 903-1111 ext. 2445
fax : (886-2) 904-4509


From: Jorg Peters <>
Date: Tue, 27 Aug 91 20:22:20 EDT
Subject: Change of Address for Jorg Peters

I moved from IBM to RPI.
My new address is

Jorg Peters
Dept. of Mathematics
Rensselaer Polytechnic Institute
Troy, NY, 12180-3590
(518) 276-6896 (FAX: 276-4824)


From: Lothar Reichel <>
Date: Wed, 28 Aug 91 22:25:40 EDT
Subject: New Address for Lothar Reichel

Dear Friends and Colleagues:

I have moved to Kent State. My new coordinates are

Lothar Reichel
Kent State University
Department of Mathematics and Computer Science
Kent, OH 44242

e-mail: or
phone: (216) 672-2547
fax: (216) 672-7824


From: Richard Brualdi <>
Date: Mon, 26 Aug 91 10:31:27 cdt
Subject: Change of Address for Richard Brualdi

On September 1, 1991 I will move up to the Institute for Mathematics
and its Applications at the University of Minnesota and will be
there during the Applied Linear Algebra Program. My address
there will be:

Richard A. Brualdi
514 Vincent Hall TEL: 612-624-7073
IMA/Univ. of Minnesota EMAIL:
206 Church Street S.E. FAX: 612-626-7370
Minneapolis, MN 55455

My email here in Madison will be forwarded so that the address
< > will still be valid as will
< >.

While I am at the IMA I will continue to perform my duties as
I would prefer that correspondence (new submissions, referee's
reports, inquiries etc.) be send to my permanent address:

Richard A. Brualdi
Math. Dept./Univ. of Wisconsin
480 Lincoln Drive
Madison, WI 53706 (USA).

My (very capable) assistant Mary Kirk (
will be in Madison, and through EMAIL, FAX, PHONE, SMAIL,
and occasional visits to Madison, I will take care of
my part of LAA.


From: Richard Brualdi <>
Date: Mon, 26 Aug 91 10:31:27 cdt
Subject: IMA Applied Linear Algebra Year Begins

The IMA's 1991-1992 program on Applied Linear Algebra officially
begins on September 3, 1991. From September 3 to September 10, the
IMA Tutorial on Applied Linear Algebra will be given. The program
from Sept. 3 to Sept. 9 involves four series of lectures:

Ake Bjorck, Algorithms for linear least square problems
Mitchell Luskin, Iterative Methods for large linear systems
Shmuel Friedland, Estimates of the first and second eigenvalues of graphs
Richard A. Brualdi, Combinatorial and qualitative matrix theory

The program on Sept. 10 includes:

Joseph W.H. Liu, Advances in direct sparse methods
Paul Van Dooren, Numerical linear algebra in control theory
George Cybenko, Signal processing applications of linear algebra

The SIAM Conference on Applied Linear Algebra takes place at the
Radisson Hotel Metrodome (adjacent to the University of Minnesota
campus in Minneapolis) from September 11 to September 14.

The fall workshops at the IMA for the Applied Linear Algebra program are:

October 14-18: Sparse matrix computations: graph theory issues and algorithms.

November 11-15: Combinatorial and graph-theoretic problems in linear algebra.

We encourage readers of this announcement on the net to participate
in these and the other IMA workshops. There are no registration fees.


From: Richard Brualdi <>
Date: Mon, 26 Aug 91 12:20:30 cdt
Subject: Special Issue of LAA Honoring Ingram Olkin

Ingram Olkin will celebrate his seventieth birthday in 1994. A
special issue of the journal will honor him for his many
important contributions to linear algebra and its applications to
statistics. The purpose of this announcement
is to solicit papers for this special issue.

Anyone may submit a paper for this issue. Contributions should be
appropriate for publication in Linear Algebra and its Applications
and will be subject to the usual review process.
All papers that fall within the scope of the journal, including
the use of linear algebra in statistics, can be submitted for this special
issue. The deadline for submissions is August 31, 1992.

Papers should be submitted to one of the special editors of the issue:

Friedrich Pukelsheim
Institut fur Mathematik
Universitat Augsburg
Memminger Strasse 6
D-8900 Augsburg, Germany

George P. H. Styan
Department of Mathematics and Statistics
McGill University, Burnside Hall 1240
805 ouest, rue Sherbrooke Street West
Montreal, Quebec, Canada H3A 2K6

Henry Wolkowicz
Department of Combinatorics and Optimization
Faculty of Mathematics
University of Waterloo
Waterloo, Ontario, Canada N2L 3G1

Ion Zaballa
Departamento de Matematica Aplicada
Universidad del Pais Vasco
Apdo. Correos 450
01080 Vitoria-Gasteiz, Spain.

Papers can also be submitted through the editor-in-chief. Publication
of the special issue is planned for the fall of 1993.
We hope that you will join in this tribute to our esteemed colleague.


From: Alan Karp <>
Date: Mon, 26 Aug 91 09:32:38 PDT
Subject: Rules for Bell Prize

The 1992 Gordon Bell Prizes

The Gordon Bell Prizes recognize achievements in large-scale
scientific computing. In the past, the winners of the Prizes were
announced in February at ComCon. Henceforth, the presentations will
be made at the IEEE sponsored Supercomputing meetings held in
November. Further, the article describing the winners will now be
published in IEEE Computer in December instead of in the May issue
of IEEE Software. The next Prize, which would have been called the
1991 Prize, will be renamed to reflect the change.

Moving the awards ceremony means that the schedule and procedures
will change. Entries for the next Prize are due on 1 May 1992, and
finalists will be announced by 30 June 1992. Pending approval by the
Supercomputing '92 program committee, finalists will be invited to
present their work at a special session of that meeting in November
1992. Winners and honorable mentions will be announced following the

The 1992 prizes will be given in two of three categories:

1. Performance: The entrant will be expected to convince the judges
that the submitted program is running faster than any other
comparable engineering or scientific application. Suitable evidence
will be the megaflop rate based on actual operation counts or the
solution of the same problem with a properly tuned code on a machine
of known performance, such as a Cray Y-MP. If neither of these
measurements can be made, the submitter should document the
performance claims as well as possible.

2. Price/performance: The entrant must show that the performance of
the application divided by the list price of the smallest system
needed to achieve the reported performance is better than that of
any other entry. Performance measurements will be evaluated as for
the performance prize. Only the cost of the CPUs, memory, and any
peripherals critical to the application need be included in the
price. For example, if the job can be run on diskless compute
servers, the cost of disks, keyboards, and displays need not be

3. Compiler parallelization: The combination of compiler and
application that generates the most speed-up will be the winner.
Speed-up will be measured by dividing the wall clock time of the
parallel run by that of a good serial implementation of the same
job. These may be the same program if the entrant can convince the
judges that the serial code is a good choice for a uniprocessor.
Compiler directives and new languages are permitted. However, anyone
submitting an entry in other than a standard, sequential language
will have to convince the judges that the parallelism was detected
by the compiler, not by the programmer.

There are some general conditions:

1. The submitted program must have utility; it must solve a problem
that is considered a routine production run, such as making daily
weather predictions or solving an important engineering or
scientific problem. It should not be a contrived or experimental
problem that is intended just to show high speed-up.

2. Entrants in the price/performance category must demonstrate that
the machine they used has real utility. (No fair picking up a few
used Z-80s for $1 each.) Only list prices of components should be
used. If the machine is not on the market, the entry is probably not
eligible although the judges will consider any reasonable estimate
of the price.

3. One criterion the judges will use for all categories is how much
the entry advances the state of the art of some field. For example,
an entry that runs at 15 Gflops but solves a problem in a day that
previously took a year might win over an entry that runs at 20
Gflops solving a more mundane problem. Entrants who believe their
submission meets this criterion are advised to document their claims

4. In all cases the burden of proof is on the contestants. The
judges will make an honest effort to compare the results of
different programs solving different problems running on different
machines, but they will depend primarily on the submitted material.

Contestants should send a three or four page executive summary to
Marilyn Potes, IEEE Computer Society, 10662 Los Vaqueros Circle, Los
Alamitos, CA 90720-2578 before 1 May 1992.


From: Baker Kearfott <>
Date: Mon, 26 Aug 91 10:31:08 CDT
Subject: New Journal -- Interval Mathematics

Note: Since my last communication, this publication has changed from a
"Soviet Periodical" to an "International Journal". Also, the first
issue has come out, and I have a specific order form ("attached") and
subscription instructions. There are new members of the editorial
board, and some of the addresses have changed. Finally, we continue to
welcome high quality submissi

A New International Journal


A new refereed international journal, "Interval Computations" is
available. It contains papers related to the following subjects:

* Theoretical research in interval mathematics
* Interval methods and algorithms and their implementation
* Interval software

Items published are in the form of

* original papers
* surveys
* reports on new software and hardware
* reviews of new monographs
* abstracts of papers
* letters to the editor
* information about symposiums, conferences, etc.
* advertisements of software

Languages of publication are English and Russian. All articles have
abstracts in both l4anguages.

The editor in chief is:

V. M. Nesterov
Leningrad Institute for Information and Automation
(Laboratory for Applied Informatics)
V.O. 14 line, 39
Academy of Sciences of the USSR
Leningrad 199178, USSR

The editorial board includes:

G. Alefeld (Karlsruhe)
B. S. Dobronets (Krasnojarsk)
R. B. Kearfott (Lafayette, LA, USA)
email: (Internet)
S. M. Markov (Sofia)
E. A. Musaev (Leningrad)
H. Ratschek (Duesseldorf)
S. M. Rump (Hamburg)
A. L. Semenov (Moscow)
A. O. Slisenko (Leningrad)
A. G. Yakovlev (Pereslavl-Zalessky)
V. S. Zjuzin (Saratov)

Information regarding submission of items for publication may be
obtained from the editor in chief, from me (B. Kearfott), or from the
first issue of the journal.

There will be nominally four issues per year, but this may vary.
Each issue will contain approximately 100 pages.

If you wish to subscribe to this periodical in the United States or
Canada, please contact me. The price is $32 per year.


Interval Computations - Leningrad - #1, 1991 - 117 P.

A.G.YAKOVLEV Interval computations - subject of research and useful tool
J.HERZBERGER, D.BETHKE On two algorithms for bounding the inverses of an
interval matrix
B.S.DOBRONETS, V.I.SENASHOV On interval extension of some classes of functions
R.B.KEARFOTT, CHENYI HU, MANUEL NOVOA III A review of preconditioners
for the interval Gauss-Seidel method
E.A.MUSAEV Narrowing of intervals by partial derivatives
S.P.SHARY On compatibility of linear tolerance problem
B.V.PALUH, B.V.VASILYOV, V.L.PEROV Application of interval mathematics for
solving technical diagnostics tasks of non-stop manufacture in chemical

V.S.Zjuzin Review of the book: H.Bauch, K.-U.Jahn, D.Oelschlagel, H.Susse,
V.Wiebigke "Interval Mathematics (Theory and Application)"
E.A.MUSAEV Interval Methods for Numerical Computation
C.G.IFHSQ YII Dctcj.pyjt cjdtofybt gj bynthdfkmyjq vfntvfnbrt
S.P.SHARY 7-th All-Union Conference on Interval Mathematics

Baker Kearfott
Department of Mathematics
University of Southwestern Louisiana
U.S.L. Box 4-1010
Lafayette, LA 70504-1010

Office: (318) 231-5270
Home: (318) 981-9744
Email: (Internet)


From: Biswa Datta <>
Date: Wed, 28 Aug 91 15:02:11 CDT
Subject: IMACS Symposium in Bangalore, India

Special Sessions On
"Large-Scale and Parallel Matrix Computations"
"Numerical Methods in Signal, Systems and Control"

IMACS - International Symposium on
Scientific Computing and Mathematical Modeling
December 7-11, 1992
Bangalore, India

Please send three copies of the extended abstracts of the
contributed talks to:

Professor Biswa Nath Datta
Department of Mathematical Sciences
DeKalb, IL 60115
FAX: (815)753-0198
E-MAIL: or

The sessions will have both contributed and invited talks. All
contributed papers will be refereed.

The accepted papers will be published in Proceedings of the
conference. Deadline: October 15, 1991

For further information on registration, local arrangements and
other details of the scientific program of the conference, please

Professor S.K. Dey (chairman)
Department of Mathematics
Eastern Illinois University
Charleston, IL 61920


From: Herman J. Woltring <UGDIST@NICI.KUN.NL>
Date: Thu, 29 Aug 91 00:11 MET
Subject: Cubic Roots Formula

Dear NA-net readers,

Following my request about cubic and quartic roots in NA-net 91(34), I have
received a great number of reactions, including two FORTRAN codes from both
sides of the Atlantic. Thanks to all -- I'm still sorting out which procedure
is the most appropriate for my purpose.

One other subscriber (at a .com account) asked me for copies of any code
received. After some email exchanges on the (non)commercial nature of
their work, consent was obtained. Perhaps it would be a good idea that
a `fair use' copyright clearance be always provided when software is sent
out privately in reply for Calls for Help?

When I posted my comment that I could not get the cubic equation's algorithm
in section 3.8.2. of Abramowitz & Stegun going, I left open whether my FORTRAN
code and/or the published algorithm were wrong. Dan Lozier of the National
Institute of Standards and Technology in Washington DC asked me to provide
supporting evidence on my posting, since he maintains a list of corrections
to the Abramovitz & Stegun book; thus, I sent my code to him. Below is an
extract from his reply which provides the explanation for my problem (i.c.,
real roots of the third-degree real-coefficients polynomial were properly
assessed, conjugate complex roots were not), and which I am posting with his

> Date: Wed, 28 Aug 91 16:26:20 EDT
> From:
> Subject: Section 3.8.2
> To: ugdist@NICI.KUN.NL
> My colleague Marjorie McClain and I have examined Section 3.8.2 and we
> can explain why results computed by your program are sometimes incorrect.
> The branch of the cube root that is taken is very important. It should be
> the principal branch except for a cube root of a negative number, in which
> case the negative real cube root should be taken. The Fortran expression
> C**(1.0/3.0) where C is of type COMPLEX always produces the principal
> branch. Your program works when modified to use arithmetic of type REAL
> for the computation of cube roots when the discriminant is nonnegative.

> In conclusion, we think we can say the formulas in Section 3.8.2 are
> correct but for algorithmic usage they are slightly ambiguous in that
> the precise determination of the cube root is not given.

> Dan Lozier

Perhaps it might be a good idea that this ambiguity be specified in future
printings of the by now almost classical A&S book!

Herman J. Woltring, Eindhoven, The Netherlands


From: Max Gunzburger <GUNZBURGER%VTCC1.bitnet@VTVM1.CC.VT.EDU>
Date: Tue, 27 Aug 91 08:56 EDT
Subject: Position at Virginia Tech

Virginia Polytechnic Institute and State University
Department of Mathematics
Position in Computational Mathematics/Numerical Analysis

The Department of Mathematics at Virginia Tech is seeking
qualified applicants for a tenure-track position at the rank of
Assistant Professor in the area of Computational
Mathematics/Numerical Analysis. The position should be
available beginning with the 1992-1993 academic year. A
Ph.D. and strong research potential are required.

The Department currently includes a large group of Applied
Mathematicians, including a number of numerical analysts and
computational scientists. Also, through the Interdisciplinary
Center for Applied Mathematics, members of the Department
have developed close collaborations with numerous scientists
and engineers throughout the University. We seek applicants
who will be able to interact with the present staff, and aid in the
further development of the Computational
Mathematics/Numerical Analysis program in the Department.
The position involves teaching duties of approximately six
hours per week; thus, there should be strong indications that
an applicant is or will become an effective teacher.

The deadline for receiving applications is March 15, 1992; after
that date, applications will be entertained only if the position
remains unfilled. A curriculum vitae, a description of research
interests, including a synopsis of the dissertation, three letters
of recommendation, and any other supporting materials should
be sent to Max D. Gunzburger, Department of Mathematics,
Virginia Tech, Blacksburg, VA, 24061-0123.

Virginia Tech is an Equal Opportunity, Affirmative Action
Employer. In keeping with Virginia TechYs commitment to
providing equal opportunity for all, women, minorities, and the
disabled are especially encouraged to apply.


From: Ian Sloan <sloan@hydra.maths.unsw.OZ.AU>
Date: Thu, 29 Aug 91 13:20:37 +1000
Subject: Postions at New South Wales

Positions at the University of New South Wales, Australia

Two positions at the University of New South Wales, in Sydney, Australia, have
recently been advertised, with a closing date of October 11. I will pass on
any applications sent to me by e-mail ( or fax
(61-2-6626445 - care of Ian Sloan).

Ian Sloan



(1 Tenured/Tenurable appointment, 1 Fixed term appointment)

Appointments in the areas of Pure Mathematics or Applied Mathematics are

Applicants should have a PhD or equivalent qualification, and proven research
achievement in an area that reinforces and extends existing strengths in the
School. The School currently has research strengths in many areas of pure
and applied mathematics and statistics, and is committed to further developing
its involvement in mathematical computer science. Although an appointment at
the level of lecturer is preferred, appointment at the senior lecturer level
may be considerrewd for an applicant with exceptional qualifications.

The positions will be available from February 1992. Appointment to one
position will be either with tenure, or on the basis of a contract with
provision for conversion to tenure. Appointnment to the other position will
be for a fixed term of three years.

Further information from Professor R.H.J.Grimshaw, Head of School,

Applications close October 11 1991.

Salary: Lecturer $38,500 - $46,000 ($47500 from July 92)
Senior Lecturer $47,500 - $55,000 ($56,500 from July 92)

Commencing salary according to qualifications and experience. Applicants should
forward two copies of their applications, including curriculum vitae, telephone
numbers (home and business), transcripts of academic record and the names and
addresses of two referees to the Staff Office, University of New South Wales,
Kensington, NSW 2033, Australia.


End of NA Digest