NA Digest Thursday, July 11, 1996 Volume 96 : Issue 26

Today's Editor:
Cleve Moler
The MathWorks, Inc.

Submissions for NA Digest:

Mail to

Information about NA-NET:

Mail to

URL for the World Wide Web: -------------------------------------------------------

From: P. Knijnenburg <peterk@CS.LeidenUniv.NL>
Date: Tue, 2 Jul 1996 17:06:46 +0200 (METDST)
Subject: Subroutine On Demand Service

Subroutine On Demand-service

The Computer and Software Systems group of the Department of Computer Science
at Leiden University proudly presents a new WWW service.

In the mathematical software community, it has been widely accepted
that adopting a library of basic routines for problems in linear
algebra can help in improving the clarity, portability, modularity,
maintenance, robustness, and even the efficiency of mathematical
software. Examples of such libraries include Level 1, 2, 3 BLAS,
EISPACK, LINPACK, LAPACK etc. These libraries mainly deal with
dense matrices. A similar approach can be taken for mathematical
software that deals with general sparse matrices. The package
SPARSKIT provides many primitives for working with sparse matrices.

Developing, optimizing, and maintaining sparse software by hand
is an error-prone and complex task. In an attempt to resolve
these problems, the group at Leiden has proposed an alternative approach
to the development of sparse software. Rather than dealing with
the sparsity of matrices at programming level, the sparsity is dealt with
at compilation level. A special kind of restructuring compiler,
referred to as a sparse compiler, automatically transforms a dense
program into efficient sparse code that operates on sparse storage schemes.

This experimental service provides you with an interface to our
sparse compiler and allows you to obtain sparse versions of
Level 2 BLAS and Level 3 BLAS subroutines. The compiler transforms
the program into a sparse version which is then sent back to you.

The Subroutine On Demand-service can be accessed through the following URL:

We hope to welcome you to our service,

Best regards,

Aart J.C. Bik
Peter J.H. Brinkhaus
Peter M.W. Knijnenburg
Harry A.G. Wijshoff


From: Nick Higham <>
Date: Tue, 2 Jul 1996 12:00:33 +0100
Subject: Bibliography on Numerical Algorithms

The bibliography of the book

author = "Nicholas J. Higham",
title = "Accuracy and Stability of Numerical Algorithms",
publisher = "Society for Industrial and Applied Mathematics",
address = "Philadelphia, PA, USA",
year = 1996,
pages = "xxviii+688",
URL = "",
isbn = "0-89871-355-2"

is now available as BibTeX database acc-stab-num-alg.bib
from my home page at
I hope this file will also be added to the Bibnet archives in the
near future.

Nick Higham


From: Karin Remington <>
Date: Sun, 7 Jul 96 13:49:30 EDT
Subject: NIST Sparse BLAS

NIST Sparse BLAS (v. 0.9)

As part of the ongoing standardization efforts in the BLAS Technical Forum[1],
we are releasing beta software for the proposed Sparse BLAS[2,3].
We hope this allows the testing of such kernels in applications and
helps motivate discussion with the numerical user-community regarding
interface, functionality, and performance issues.

The NIST Sparse BLAS implementation provides computational kernels for
low-level sparse matrix multiply and triangular solve operations, and
includes most of the functionality of the proposed Sparse BLAS Toolkit [2]
interface as well as a separate "lite" interface with minimal overhead.

The Sparse BLAS routines are written in ANSI C and have both C and Fortran 77
bindings. Source code and documentation, as well as related references,
can be found at

Karin Remington
Roldan Pozo
Applied and Computational Mathematics Division
National Institute of Standards and Technology (NIST)

[1] BLAS Technical Forum,

[2] "A Revised Proposal for the Sparse BLAS Toolkit", S. Carney,
M. Heroux, G. Li, R. Pozo, K. Remington, K. Wu,

[3] "A Proposal for User Level Sparse BLAS", Duff, M. Marrone, G. Radicati,
Technical Report TR/PA/92/85, CERFACS, December 1992


From: Brad Bell <>
Date: Fri, 5 Jul 1996 09:30:31 -0700 (PDT)
Subject: IEEE arithemetic and F2C

An unexpected property of IEEE arithmetic / a bug in F2C

As I understand IEEE standard arithmetic, the values -0 and
0 are equal but the values -Inf and +Inf are not equal. For example,
if I run the following C-code:

# include <stdio.h>
int main()
double x = 0.;
double y = -0.;
double rx = 1. / x;
double ry = 1. / y;

printf("x = %f\n", x);
printf("y = %f\n", y);
printf("rx = %f\n", rx);
printf("ry = %f\n", ry);
printf("x == y = %d\n", x == y);
printf("rx == ry = %d\n", rx == ry);


I should get the following result:

x = 0.000000
y = -0.000000
rx = Inf
ry = -Inf
x == y = 1
rx == ry = 0

It is unexpected that taking the reciprocal of two equal values
can have different results. Properties like this can cause very subtitle
bugs in software. For example, take the following macro that is part of
the f2c distribution:

#define abs(x) ((x) >= 0 ? (x) : -(x))

If x has the value -0, this code will return the value -0 as its absolute
value. This lead to the value -Inf for the following expression:

1. / abs(-0.)

which is unexpected. I found this problem while taking the condition number of
a matrix that had minus zero as one of its singular values. The resulting
condition number was -Inf which was less than the bound I was checking against.
This lead to the conclusion that the matrix was not singular.


From: Roland Sweet <>
Date: Thu, 11 Jul 1996 11:08:50 -0600
Subject: Double Precision Version of FISHPAK

I have had two requests recently for a double precision version of FISHPAK.
I know that such a version was created many years ago, but have no idea if
it is still around. Does anyone know if it is available?

Roland Sweet


From: Helge Holden <>
Date: Fri, 5 Jul 1996 15:18:34 +0200
Subject: WWW Server for Preprints on Conservation Laws

Dear colleague,

This is to inform you that we have set up an electronic preprint server
on World Wide Web (WWW) for preprints on conservation laws. This
server was announced at "The Sixth International Conference on
Hyperbolic Problems: Theory, Numerics, Applications" in Hong Kong this
June, and was well received. The server has URL
and accepts papers containing new and recent results concerning
conservation laws and related areas. The main focus is on mathematical
analysis and numerical methods, but papers concerning applications are
also welcome. Preprints submitted should be in postscript format. The
server was released this June, and does not contain many preprints yet.
However, we hope that you will use this server to make it into a
convenient tool for the community working on conservations laws. If
you have questions, do not hesitate to contact us using, e.g., the
email address

Best regards,

Harald Hanche-Olsen Helge Holden Knut-Andreas Lie

Department of Mathematical Sciences
Norwegian University of Science and Technology
N-7034 Trondheim

email addresses:


From: Wai Sun Don <>
Date: Mon, 8 Jul 96 17:44:41 -0400
Subject: Numerical Differentiation by Pseudospectral Methods

Hello, this message announces the first version of Pseudopack.

Pseudopack is a software package to do numerical differentiation by
pseudospectral methods. Currently, only Chebyshev methods are supported,
i.e., polynomial interpolation at grid points which are the extrema of a
Chebyshev polynomial.

Brief description of version 1.0 features:

1. Differentiation of one or several vectors can be computed by a single
subroutine call.
Example : 1-D PDE with two or more dependent variables.

2. Differentiation can be done on either the first or the second index of
a two dimensional array
Example : 2-D PDE with differentiations in both x and y directions.

3. Several algorithms are provided:
1. Matrix-Matrix Multiply Algorithm
2. Even-Odd Decomposition Algorithm
3. Cosine Transform-Recursion Algorithm

4. Minimized roundoff error for the Chebyshev differentiation.

5. Further reduction of roundoff error and matrix condition number by the
use of a coordinate mapping.

Its aims are to provide minimum roundoff error and good efficiency on
several computational platforms. They are SGI, Sun, IBM R6000 and Cray.

The software package is written in Fortran 77 with the C preprocessor. We
have used a few Fortran 90 extensions, namely long function and variable
names, and use of the enddo keyword.

It is freely available, at least to all non-commercial users.

You can read the description and/or download the tarfile from our Web site,

Questions, comments, or suggestions to

Wai Sun Don and Alex Solomonoff


From: Alan Harvey <harvey@csli.Stanford.EDU>
Date: Tue, 9 Jul 1996 10:57:53 -0700
Subject: New Book, Domain Decomposition

Domain Decomposition:
Parallel Multilevel Methods for
Elliptic Partial Differential Equations

Barry Smith, Petter Bjorstad, and William Gropp

published by Cambridge University Press

The emergence of parallel computers and their potential for the numerical
solution of Grand Challenge problems has led to extensive research in
domain decomposition methods. These are general, flexible methods for the
solution of linear or nonlinear systems of equations arising from the
discretization of partial differential equations that use underlying
properties of the PDE to obtain fast solutions. The broad family of domain
decomposition methods can be said to include many multigrid and multilevel

This book presents an easy-to-read discussion of domain decomposition
algorithms, their implementation and analysis. The relationship between
domain decomposition and multigrid methods is carefully explained at an
elementary level, and discussions of the implementation of domain
decomposition methods on massively parallel super computers are also
included. All algorithms are fully described and explained, and a
mathematical framework for the analysis and complete understanding of the
methods is also carefully developed. In addition, numerous numerical
examples are included to demonstrate the behavior of this important class
of numerical methods.

This book is ideal for graduate students about to embark on a career in
computational science. It will also be a valuable resource for all those
interested in parallel computing and numerical computational methods.

1. One level algorithms
2. Two level algorithms
3. Multilevel algorithms
4. Substructuring methods
5. A convergence theory
Appendix 1: Preconditioners and Accelerators
Appendix 2: Software for Numerical Parallel Computing

Published June 1996 258 pp. 46 line diagrams 2 tables
ISBN 0-521-49589-X (Hardback)

For more information, consult the online catalogs at and, the authors' homepage at, or send enquiries to


From: Steve Leon <>
Date: Wed, 10 Jul 1996 17:19:50 -0500 (EST)
Subject: New Book, ATLAST Computer Exercises for Linear Algebra

Book announcement

ATLAST Computer Exercises for Linear Algebra
Edited by Steven Leon, Eugene Herman, and Richard Faulkenberry
Prentice-Hall, August 1996

The ATLAST collection of computer exercises represents the best creative
efforts of the more than 350 faculty members who participated in the thirteen
ATLAST workshops offered between 1992 and 1995. Workshop participants
designed computer exercises and projects suitable for use in undergraduate
linear algebra courses. From the entire ATLAST database of materials, the
editors have selected a comprehensive set of exercises covering all aspects
of the first course in linear algebra. Each chapter is divided
into two sections. The first section consists of shorter exercises
and the second section consists of longer projects.

The computer exercises are all based on MATLAB. A unique feature
of MATLAB is that it is the only major mathematical software package
that is based almost entirely on matrices. A collection of MATLAB
routines (M-files) has been developed to accompany this book. Many of
these routines are designed to give visual illustrations of important
linear algebra concepts such as coordinate systems, linear transformations,
and eigenvalues. Other M-files illustrate visual applications such as using
linear transformations for computer animations or using matrix factorizations
for digital imaging. Still other M-files can be used to generate
special structured matrices. Students are then challenged to discover
properties of the special matrices.

The entire collection of ATLAST M-files can be obtained either from the
ATLAST Web page:
or by anonymous ftp from (directory: pub/atlast)
These files are required for many of the exercises and projects in the
book. The collection of files will be updated and expanded a few times a
year; so check the web page or ftp site periodically for the latest versions.
Of the sixty M-files currently in the collection some of the more interesting
files to preview are:
1. Transfor - a utility for visualizing linear transformations
2. Cogame - a utility for visualizing linear combinations and
coordinate systems in the plane
3. Eigshow - a utility for visualizing eigenvalues and eigenvectors
of 2x2 matrices
4. Pyr - a utility for studying rotations in three space by examining
the pitch, yaw, and roll of an airplane
5. Svdshow - a utility for visualizing the singular value decomposition
of a 2x2 matrix
The ATLAST book has computer lab projects for each of these utilities.

To take full advantage of the ATLAST M-files they should be used in
conjunction with the exercises and projects in the ATLAST book.
The book will be published as an inexpensive paperback manual that
can be used in conjunction with any of the standard linear algebra
textbooks. Anticipated list price should be about $12 or $13 depending
on the markup at your bookstore. (Super bargain: Prentice-Hall is offering
a 50% discount if the manual is adopted along with one of the Prentice-Hall
linear algebra texts.)

Unlike other manuals of computer exercises, the ATLAST collection is a
massive collaboration representing a wide variety of views.
Publication is scheduled early in August, 1996; so it is possible
to preview the book and order it for this coming Fall. Those needing
to make a decision before August can obtain a copy of the
book manuscript by sending an email message to:

ATLAST is an NSF sponsored project.

Augmenting the
Teaching of
Algebra through the use of


From: Chaoqun Liu <>
Date: Mon, 1 Jul 1996 12:49:00 -0500 (CDT)
Subject: Materials from Computational Fluid Dynamics Short Course


Chaoqun Liu and Zhining Liu (Editors)

As the world turns to the twenty-first century, most CFD people are facing
very crucial challenges: what we should do and what we can do? In this
two-volume book, we collected the materials presented in the ACFD
short course, June 24 - 28, 1996, Louisiana Tech University.
A number of worldwide well-known scientists in CFD and Applied Mathematics
presented their works and gave preview in this short course.
To CFD people, this book will provide a good source for advanced CFD knowledge.

The contents in this book include:

o CFD: State of the Art and Air Force Needs (L. Sakell)
o Multistage Central Difference Schemes for the Euler and Navier-Stokes
Equations (E. Turkel)
o A Brief Introduction to High-Resolution Schemes (P. Roe)
o Upwinding Without Dissipation (P. Roe)
o Multidimensional Upwinding-Motivation and Concepts (P. Roe)
o High Resolution Schemes for Complex Hyperbolic Conservation Laws (P. Roe)
o Computational Modeling of Turbulent Flows (T-H Shih, W. Liou, and J. Zhu)
o Principles in Multilevel Adaptive Methods (C. Liu)
o Multigrid Methods in Fluid Dynamics (A. Brandt)
o First-Order System Least Squares -- FOSLS (S. McCormick)
o DNS for Flow Transition Around Complex Geometry (C. Liu and Z. Liu)
o Direct Numerical Simulation for the Receptivity and the Whole Process
of Transition Around Joukowsky Airfoils (Z. Liu, G. Xiong, and C. Liu)
o Implementation of Multigrid Methods in Prediction of Turbulent
Combustion with Detailed Chemistry (C. Liao, X. Zheng, Z. Liu,
C.Liu, and S. McCormick)
o Multigrid Methods for Two-Equation Turbulent Models (X. Zheng,
C. Liao, C.Liu, C.H. Sung, and T.T Huang)
o Multilevel Local Refinement and Multigrid Methods for 3-D Turbulent Flows
(C.H. Sung, C. Liao, and X. Zheng)

All the presentations in the short course have been well recorded, which
run about 36 hours (in six VHS video tapes).

During the short course, the Numerical Simulation Group, led by
Prof. Chaoqun Liu, also released three codes:

o DFPT (Direct numerical simulation for Flat Plate boundary layer Transition)
o FINS (Full-multigrid Incompressible N-S solver)
o MTC3D (Multigrid Turbulence Combustion -- 3D)

The above codes are released for non-commercial purpose. Anyone
who is willing to get the codes needs to sign an Agreement Form
before obtaining the code.

If you are interested in the above materials, please contact

Margaret Dunn
The Graduate School
Louisiana Tech University
Ruston, LA 71272, USA
Tel: (318) 257-2924


Dr. Chaoqun Liu, Associate Professor
ACFD Short Course Chair
Department of Mathematics and Statistics
Louisiana Tech University
P.O. Box 3189, Ruston, LA 71272-0001
Tel : (318) 257-2538
Fax : (318)257-3935
email :


From: Bette Byrne <>
Date: Tue, 2 Jul 96 10:12:13 GMT
Subject: Special Meeting Honoring Bill Mortin

17-18 APRIL 1997

This is a special meeting to honour Bill Morton in the year of his
retirement from the chair of Numerical Analysis at Oxford. Four speakers,
Professors Morawetz, Parlett, and Roe as well as Bill himself will give
major talks while the rest of the presentations will be given by some of
Bill's former students and colleagues.

The topics will naturally be determined by Bill's interests over the years in
numerical analysis and CFD as well as, no doubt, anecdotal evidence of
his influence on the life of the communities and the people that he has
worked with.

The meeting will take place in the Computing Laboratory and will begin
mid-morning on Thursday, 17th and finish during the late afternoon on
Friday, 18th. Accommodation for Thursday night will be available at St
John's College. There will also be a Special Dinner in St John's College
on Thursday evening.

If you require any further information please contact: Bette Byrne, Oxford
University Computing Laboratory, Wolfson Building, Parks Road, Oxford,
OX1 3QD, Tel: +44 1865-273883, Fax: +44 1865-273839, Email:


From: Raymond Chan <>
Date: Tue, 2 Jul 1996 21:21:33 +0800 (EAT)
Subject: Banquet for Gene Golub's 65th Birthday

The announcement of our Workshop in the last NA-Digest unfortunately
left out the Special Event about the Banquet for Prof. Golub's 65th

Banquet for Gene Golub's 65th Birthday
at the
Workshop on Scientific Computing 97
Hong Kong
March 10-12, 1997

There will be a a banquet on March 12, 1997 to celebrate the
65th birthday of Professor Gene Golub.

Dr. Raymond Chan, Department of Mathematics,
The Chinese University of Hong Kong, Shatin, Hong Kong.
Tel: (852) 2609-7970, Fax: (852) 2603-5154.
E-mail:, WWW:


From: Manfred Trummer <>
Date: Wed, 3 Jul 1996 07:08:56 -0700
Subject: Pacific Northwest Numerical Analysis Seminar

The 1996 Pacific Northwest Numerical Analysis Seminar (PNWNAS) is hosted by
Simon Fraser University, and will be held in Vancouver, British Columbia, on

Saturday, September 21, 1996.

The deadline for submissions to the student paper competition is July 31, 1996.

For information on the meeting and the conference program please
visit the conference web site


From: Uwe Zimmermann <>
Date: Thu, 4 Jul 1996 16:21:53 +0200
Subject: Symposium on Operations Research

Technical University Braunschweig, September 4-6, 1996

The preliminary programm is already in print and will be published in

DGOR Bulletin and in GM"OOR Newsletter

next week.

The online version is already available at URL

The preliminary program contains the actual schedule and can be downloaded
in compact PostScript format via ftp

The schedule is ordered by section number and contains location, time,
chairman, speaker, and title of each talk. We emphasize that this is only the
preliminary version of the program. The final version will be distributed with
the conference material at the registration desk of SOR96. Please, send
comments and corrections to

A summary of the respective personal data, including current information on
financial issues will be sent to each participant, speaker and chairman via
electronic mail during the next weeks.

Uwe Zimmermann Braunschweig, July 4, 1996
chairman SOR96


From: A. Bellen <BELLEN@univ.triest>
Date: Mon, 8 Jul 1996 13:22:18 +0200 (MET-DST)
Subject: Scientific Computing and Differential Equations

Grado (Italy) September 15-19, 1997.


The meeting will include plenary talks, minisymposia, sessions of
selected communications, new talent discovery programme and an award
of the Dahlquist prize. A conference banquet and a half-day excursion
are also planned, as well as social activities for accompanying

PLENARY SPEAKERS (with provisional titles):
o Chris Budd: "The recursive projection method for semilinear
problems with bifurcation"
o Peter Crouch: "Numerical integration on manifolds: the role of
differential geometry"
o Ernst Hairer: "Backward error analysis of integration methods"
o Marlis Hochbruck: "Krylov approximation to the matrix exponential and
its applications to differential equations"
o Karel in 't Hout: "Analysis of numerical methods for functional
differential equations"
o Roswitha Maerz: "DAE solutions on infinite intervals and stability"
o Olavi Nevanlinna: "Waveform relaxation and Krylov subspace iteration"
o Stan Osher: "Subscale capturing and its numerous applications"
o Linda Petzold: "Numerical solution of highly oscillatory constrained
multibody dynamical systems"
o Jesus Maria Sanz-Serna (to be announced)
o Daniel Stoffer: "On the qualitative behaviour of symplectic
o Andrew Stuart (to be announced)

MINISYMPOSIA are scheduled on the following topics:

Boundary Value Problems (Organizer: U. Ascher)
Differential Algebraic Equations (S. Campbell)
Delay Differential Equations (R. Vermiglio)
Parallel ODE methods (P. van der Houwen)
Applications of ODEs 1 (B. Leimkuhler and S. Reich)
Applications of ODEs 2 (R. Skeel)
Waveform Relaxation methods (S. Vandewalle)
Partial Differential Equations (J. Verwer)
ODE software (P. Thomsen)
Generalizations of Runge Kutta methods (Z.Jackiewicz)
Continuous ODE methods (B. Owren)
Nonlinear stability (A. Ostermann)
Hamiltonians (M.P. Calvo)
Numerical methods on manifolds (H. Munthe-Kaas)
Krylov space methods for ODEs. (C. Lubich)
Stochastic ODEs (K. Burrage)

The Organizing Committee decided to reserve two plenary talks for
young researchers, under 30 years of age. Our intention is to provide
an opportunity for young colleagues, whether research students or
postdoctoral workers, to present innovative and important research at a
plenary forum. The speakers will be chosen by the Organising Committee
(A. Bellen, A. Iserles, S.P. Norsett, M. Zennaro) on the basis of
extended abstracts (4-6 pages).

LaTex or Tex files should be sent to
by February 28, 1997.

A number of sessions are planned for selected papers.

Everybody wanting to attend the Conference or interested in receiving
the next announcements is encouraged to pre-register by e-mailing name,
institution, full address, e-mail address, communication (yes/no) by
December 31, 1996, to the following address:

More information on the programme, fees, deadlines, accomodation,
social events etc. will feature in the next announcement and made
available on www:

Dipartimento di Scienze Matematiche
Universita' di Trieste
34100 Trieste, Italy


From: Rob Schreiber <>
Date: Mon, 8 Jul 1996 10:51:13 -0700
Subject: Principles and Practice of Parallel Programming

Principles and Practice of Parallel Programming
Las Vegas, Nevada, June 18th-21st, 1997

The goal of the PPoPP Symposia is to provide a forum for papers on the
principles and foundations of parallel programming, on
tools and techniques for parallel programming, and on experiences in using
parallel programming to solve applications problems.

Topics of interest include:

* design and implementation of parallel programming languages and systems;
* programming experience with parallel programming languages and systems;
* restructuring compilers and program manipulation systems for parallel
* environments, debuggers, monitoring tools and operating system support
for parallel programming;
* the relationship between parallel programming languages, compilers and
machine architecture;
* performance aspects of parallel programming systems; and
* experiences in using parallel programming to solve applications problems.

Papers should report on original research in any of these areas of
parallel programming, and should contain enough background material to
make them accessible to the entire parallel programming research
community. This year, we are particularly soliciting papers that
describe experiences in using parallel programming for specific
applications problems. To be relevant to this conference, such papers
should describe experiences that highlight the strengths and weaknesses
of parallel tools and platforms used in the experiment. For example, a
paper that simply gave a high-level description of a parallel
implementation of a structural mechanics application would not be
appropriate for this conference. On the other hand, if this experience
highlighted some weakness in the compiler or runtime system of the
parallel platform, a paper describing this experience and drawing
relevant conclusions would be appropriate for PPoPP. In general, papers
reporting on experience should indicate how the experiments illustrate
general principles; papers oriented towards foundations should indicate
how the work illuminates or influences practice.

Online Conference Information

The complete Call for Papers and additional information about the
conference can be obtained at the following URL:

Program Committee

Chris Bischof, Argonne National Labs, USA
Guy Blelloch, Carnegie-Mellon University, USA
David Callahan, Tera Computers, USA
Sid Chatterjee, University of North Carolina, Chapel Hill, USA
Paul Feautrier, Universite de Versailles St-Quentin, France
Dennis Gannon, Indiana University, USA
John Gilbert, Xerox PARC, USA
Susan Graham, University of California, Berkeley, USA
Mike Heath, University of Illinois, Urbana-Champaign,USA
Bruce Hendrickson, Sandia National Labs, USA
Ken Kennedy, Rice University, USA
Keshav Pingali, Cornell University, USA
Lori Pollock, University of Delaware, USA

Program Chair

Keshav Pingali
Cornell University
Department of Computer Science
Upson Hall
Cornell University
Ithaca, NY 14853
(607) 255-7203
fax: (607) 255-4428

General Chair

Rob Schreiber
Hewlett-Packard Laboratories Palo Alto, CA
HP Labs 3L-5
1501 Page Mill Road
Palo Alto, CA 94304-1126
fax: (415) 857-8508


From: Craig Douglas <>
Date: Sun, 7 Jul 1996 19:55:40 +0100
Subject: Report from CERFACS Iterative Methods Workshop

International Linear Algebra Year - Workshop on Iterative Methods,
UNESCO Centre, Toulouse, France (June 10-13, 1996).

The ILAY Workshop on Iterative Methods was held at the UNESCO Centre in
Toulouse, France, on Monday, June 10 through Thursday, June 13, 1996. The
meeting was organized by the Parallel Algorithm Project at CERFACS (Centre
Europeen de Recherche et Formation Avancee en Calcul Scientifique). It was
sponsored by CNES, DGA, EDF, EU, INRIA, U.S. Army, Mairie de Toulouse, and
the Region Midi Pyrenees Conseil General Haute-Garonne.
The workshop was divided into four distinct topics by day: an industrial
day, a Krylov method day, a domain decomposition day, and a multigrid day.
This was the last of four workshops held during the year and the fifth in two
years, beginning in St. Girons in July 1994.

Industrial Day

The talks during the industrial day were somewhat tutorial and were of
interest to the industrial members of the workshop audience. The four day
poster session began with 5 minute summaries by the contributors.
Frederic Lafon (Thomson-CSF) described some large-scale numerical
simulations in microelectronics performed on clusters of workstations using a
domain decomposition approach, for which high-level dynamic load balancing
strategies had to be developed.
Pieter Hemker (CWI, Amsterdam) gave a survey of adaptive finite-volume
multigrid methods for nonlinear problems, including aerodynamics and
semiconductor device simulations.
Francois-Xavier Roux (ONERA) spoke about the dual Schur complement method
for structural analysis problems. Roux illustrated the effectiveness of this
approach for a structure problem based on a submarine shaped object with
362,000 degrees of freedom, which was analyzed on an Intel PARAGON with 98
processors in roughly five minutes.
Gene Golub made the point that actual computing times should be presented
in more talks, rather than mere condition numbers or iteration counts.
Gerard Meurant (CEA) spoke about methods (developed jointly with Golub)
for obtaining bounds for the error in the CG algorithm. Numerous examples
showed how sharp the bounds are. The error norms using CG information were
Ray Tuminaro (Sandia National Laboratory) spoke about AZTEC, a parallel
linear algebra library for the solution of linear systems.
At the end of the day, cocktails were provided at the Conseil General of
the Midi Pyrenees region, a supporter of science and technology.

Krylov Methods Day

Henk van der Vorst (University of Utrecht) spoke about hybrid iteration
methods based on combining BiCG-type methods with one or more GMRES steps.
One point made was that allowing a local increase of the residual may be
necessary to avoid stagnation and in the end to achieve convergence.
Andy Wathen (Oxford) spoke about the convergence of minimum residual
methods for the symmetric indefinite augmented systems arising in CFD
applications as well as in other areas.
Anne Greenbaum (Courant Institute, NYU) spoke about the symmetric Lanczos
process for solving linear systems and computing the action of the exponential
of a matrix on a given vector, and presented a new residual bound for the case
of a symmetric indefinite matrix.
Mario Arioli (IAN-CNR, Pavia) spoke about stopping criteria for iterative
methods used to solve linear systems arising from PDEs. He outlined a method
for prescribing criteria which guarantee that the computed solution of Ax=b
is the exact solution of a problem which is a mild perturbation of the
original PDE.
Gene Golub (Stanford) spoke about the use of Gauss quadrature rules for
bounding and estimating quadratic and bilinear forms using a Lanczos
Algorithm. In combination with Monte Carlo methods, estimates of the trace of
certain matrix functions can be obtained.
Tony Chan (UCLA) spoke about sparse approximate inverse preconditioners
and their use in combination with fast wavelet transforms. Sparse approximate
inverses can give good convergence rates and are in principle highly parallel,
but their applicability is limited to matrices whose inverse has entries which
decay rapidly away from the main diagonal. Chan showed how this difficulty
can be removed sometimes by transforming the matrix to a suitable wavelet
basis. Gene Golub mentioned that it is an open question whether it is
possible to express the inverse of a general sparse matrix in a compact
fashion similar to the characterization of the inverse of a symmetric
tridiagonal matrix.
David Silvester (University of Manchester) described stabilization methods
for FEM discretizations of Stokes and advection-diffusion problems.
Howard Elman (University of Maryland) introduced a new preconditioner for
linearized Navier-Stokes equations. Under suitable assumptions, one can show
that the preconditioned system has a condition number which is independent of
viscosity and is only slightly dependent on the mesh spacing.
In the late afternoon, cocktails were held at the Mairie, where the
conference participants had also the opportunity to admire art objects, such
as the famous paintings by Henri Martin and a number of interesting statues,
including one of Pierre de Fermat who was for many years a member of the
Toulouse parliament.

Domain Decomposition Day

Jim Douglas, Jr. (Purdue University) presented a domain decomposition
approach based on Robin boundary conditions for solving the nonlinear PDEs
describing two-phase, immiscible flow in a porous medium. For constant
coefficient model problems, the error along the interface can be eliminated
with a procedure similar to the elimination of error components in the
classical ADI method.
Alfio Quarteroni (CRS4 and TU Milan) gave an extensive overview of domain
decomposition methods and applications developed at his institutions. One of
the many attractive features of DD methods is that using different domains,
quite different aspects of the physical phenomena being modeled can be
captured. This was illustrated with a number of examples including shallow
water models, transonic aerodynamics calculations, seismic data processing and
many others.
Yvon Maday (ASCI and Paris VI) spoke about iterative substructuring
preconditioners for the mortar finite element method. The mortar method is
useful when different discretizations are used on different subdomains or when
the underlying grids do not match. A new algorithm for solving the linear
algebraic systems arising from the mortar method was presented.
Yuri A. Kuznetsov (Russian Academy of Sciences, Moscow) also spoke about
iterative methods for finite element problems with non-matching grids. Using
mortar elements results in a symmetric indefinite system for which a suitable
preconditioner is developed. The preconditioner is spectrally equivalent to
the coefficient matrix and has "optimal" arithmetic complexity.
Patrick Le Tallec (INRIA) spoke about a generalized Neumann-Neumann
preconditioner in computational mechanics problems with complicated geometries
and very fine local grids. This technique has been applied to stress
prediction near junctions of oil platforms, to nonlinear shell problems, and
to models of the human trachea.
David Keyes (Old Dominion University and ICASE) talked about multi-domain
and multi-model aspects of Newton-Krylov-Schwarz methods for the solution of
nonlinear boundary value problems in transonic aerodynamics. He discussed the
preconditioning of a higher order discrete operator with a lower order
discrete operator, and the role of pseudo-transient continuation in the
pursuit of steady states.
After a visit to CERFACS, the conference participants were taken to
Verfeil, a small village outside Toulouse, where a banquet was held. The
restaurant (La Promenade) is situated by a scenic castle with imposing walls.

Multigrid Day

Wolfgang Hackbusch (Kiel) spoke about composite finite element methods for
resolving small details near the boundary. Hackbusch showed how to move grid
points in order to have a less fine mesh near the boundary while retaining
good convergence and constructing a nested set of grids that are highly
suitable for multigrid.
Susanne Brenner (University of South Carolina) spoke about multigrid
methods for computing the stress intensity factor in singular function
representation of solutions to BVPs on domains with re-entrant angles or
cracks. Using a novel full multigrid algorithm, improved theoretical results
of optimal order for the convergence rates were derived. A matrix was even
Craig Douglas (IBM and Yale University) spoke about an additive multigrid
preconditioner based on structured grids when the original problem is on an
unstructured grid. Results show that unless great care is taken, severe
problems can occur due to interpolation errors that would not normally occur
with standard multigrid techniques.
Peter Deuflhard (Konrad-Zuse-Zentrum and Freie Universitat Berlin) talked
about cascadic (or one way) multigrid with applications to the
Bio-Heat-Transfer Equation, a 3D parabolic PDE arising in problems of cancer
determination and treatment. Other applications of this methodology to
industrial problems were shown.
Tom Manteuffel (Unversity of Colorado-Boulder) spoke about First Order
Systems Least Squares (FOSLS). FOSLS recasts a mathematical model to yield a
systematic approach for solving PDEs which results in well-posed, self-adjoint
problems for which efficient solution techniques are available. This is done
by augmenting the original problem with a few extra variables, adding a few
constraints, and constructing a suitable quadratic functional whose unique
minimizer is the solution to the original problem.
Ulrich Ruede (University of Augsburg) spoke about parallel multigrid
adaptive iteration. All aspects of a problem can be adaptively set or changed
(multilevel algorithm, relaxation scheme, h-p formulation of problem between
grids, etc.). Grids can be considered virtual. They are only substantiated
on demand of the numerical procedures. With suitable data structures,
adaptive iterations can be implemented with an overhead that is proportional
to the numerical work performed.

Concluding Remarks

The talks in the last two days were somewhat different from what is normally
found in a "pure" iterative methods workshop. On first glance it might have
been described as an invasion by the PDE hordes from both the East and West
(which was fitting for a town like Toulouse, which historically has been
invaded from a number of directions). Actually, this was a set of talks on
applied linear algebra. The numerical solution of partial differential
equations is one of the main sources of large sparse linear systems, which are
usually solved by some combination of direct and iterative methods. Various
new developments in solution techniques were described at the workshop and a
constructive interaction between experts in numerical linear algebra and
numerical PDEs was fostered.

Michele Benzi Craig Douglas
Parallel Algorithms Project IBM and Yale University
Toulouse, France


From: Jerry Young <R1GWY@VM1.CC.UAKRON.EDU>
Date: Wed, 03 Jul 96 10:38:22 EDT
Subject: Position at University of Akron

Department of Mathematical Sciences

A visiting assistant professor postition is available for the 1996/97 academic
year. Primary responsibilities include teaching three courses per year and
offering a seminar in the applicant's area of research. Applicants should
possess a Ph.D. in applied mathematics or mathematics with an area of
expertise in differential equations, applied analysis, modeling, fluid
dynamics, optimization, wave propagation, or scientific computing.
The University of Akron is the third largest state university in Ohio. The
department offers Bachelor and Master degrees in applied mathematics,
mathematics, statistics, and computer science. An engineering applied
mathematics doctoral program is offered cooperatively with the college of
All materials (application letter, curriculum vitae, unofficial copy of
graduate transcripts, and three letters of recommendation) should be sent to:

Dr. D.C. Buchthal
Departmentof Mathematical Sciences
The University of Akron
Akron, OH 44325-4002

Review of completed applications will begin July 25, 1996, and continue until
the position is filled.
Women and minorities are encouraged to apply. The University of Akron is an
equal education and employment institution.


From: Stefan Funken <>
Date: Thu, 04 Jul 1996 13:03:50 +0200
Subject: Position at the University of Kiel

Am Mathematischen Seminar (Bereich II) der Christian-Albrechts-
Universit"at zu Kiel ist ab 1. September 1996 eine Stelle als

Wissenschaftliche Mitarbeiterin/Wissenschaftlicher Mitarbeiter

f"ur zun"achst zwei Jahre zu besetzen. Die Verg"utung erfolgt nach
BAT IIa (3/4 der regelm"a"sigen Arbeitszeit bei einem Promotionsvertrag
oder Vollbesch"aftigung bei erfolgter Promotion).

Gesucht wird eine Mitarbeiterin oder ein Mitarbeiter, die/der bereits
in Angewandter Mathematik gearbeitet hat und vertiefte
Analysis-Kenntnisse im Bereich der Approximationstheorie und/oder der
partiellen Differentialgleichungen mitbringt. Voraussetzung sind
Kenntnisse im wissenschaftlichen Programmieren sowie die Bereitschaft,
Probleme aus Industrie und Wissenschaft in Algorithmen, Programme und
Simulationen umzusetzen und zu visualisieren.

Diplom oder Staatsexamen bzw. qualifizierte Promotion in Mathematik.

Die Universit"at ist bestrebt, den Anteil von Wissenschaftlerinnen zu
erh"ohen und fordert deshalb entsprechend qualifizierte Frauen
nachdr"ucklich auf, sich zu bewerben. Frauen werden bei gleichwertiger
Eignung, Bef"ahigung und fachlichen Leistungen vorrangig

Bewerbungen Schwerbehinderter werden bei entsprechender Eignung

Bewerbungen mit den "ublichen Unterlagen (Lebenslauf, Lichtbild,
Zeugnisse, Diplom, evtl. Dissertation) sind bis zum 26.7.96 zu richten

Prof. Dr. Carsten Carstensen
Mathematisches Seminar II
Ludewig-Meyn-Str. 4
24098 Kiel
Weitere Informationen unter


From: Norbert Henze <>
Date: Wed, 10 Jul 96 17:43:15 CST
Subject: Positions at University of Karlsruhe

In der
Fakult{t f}r Mathematik der Universit{t Karlsruhe (TH)
sind eine
Professur (C4) und eine Professur (C3)
Mathematische Modellbildung und Wissenschaftliches
Rechnen in den Ingenieurwissenschaften
zu besetzen.

Dabei ist insbesondere bei der Professur (C4) an eine Pers|nlichkeit
mit einem Forschungsschwerpunkt gedacht, der eng mit den
Ingenieurwissenschaften verbunden ist (z.B. ein Teilgebiet der Mechanik). Zu
den Lehraufgaben beider Professuren geh|rt auch die Durchf}hrung von Vorlesungen
im Servicebereich.

Erwartet wird die Mitwirkung bei Aufbau und --- im Falle der
Professur (C4) --- Leitung eines Instituts f}r
Wissenschaftliches Rechnen und Mathematische Modellbildung sowie Erfahrung in
der Zusammenarbeit mit den
ingenieurwissenschaftlichen Fakult{ten oder anderen Partnern innerhalb und
au~erhalb des Hochschulbereichs.

Die Hochschule ist bestrebt, den Anteil von Professorinnen zu erh|hen und
begr}~t deshalb die Bewerbung entsprechend qualifizierter Frauen.
Schwerbehinderte werden bei entsprechender Eignung bevorzugt

Bewerbungen mit den }blichen Unterlagen, einer kurzen Darstellung der
wichtigsten Forschungsergebnisse und geplanter Forschungsvorhaben, der
betreuten Promotionen und durchgef}hrten Projekte sowie einer Liste der abgehaltenen
Vorlesungen werden erbeten bis zum

1.Oktober 1996

an den Dekan der Fakult{t f}r Mathematik der
Universit{t Karlsruhe, 76128 Karlsruhe.


From: Jin Hun Yuan <>
Date: Wed, 3 Jul 1996 20:49:58 -0400
Subject: Position at Universidade Federal do Parana, Brazil

Research Position at Universidade Federal do Parana, Brazil

Several research positions are possible to be available at
Universidade Federal do
Parana, Curitiba, Brazil, to work in Department of Mathematics in the
area of Optimization and Numerical Analysis. Curitiba is a very pleasant
city , like european city, in Brazil.

The work may start in Feb. or Mar., 1997. The position is for one year
with possibility of extensions to a second, third and even forth year.
The position includes a (free tax) fellowship R$2,465.00-2,925.00 (about
US$2,400.00-2,850.00) per month (one year is 12 months fellowship), return
tickets for successful candidate and his (her) spouse, and settle expenses.

To apply submit a detailed CV with publication list, research
plan, certified copies of degree/diplomas, reprints of important publications
(also arrange three recommendation letters directly for recent doctor) to

Professor Jin Yun Yuan
Departamento de Matematica - UFPR
Centro Politecnico, CP: 19.081
CEP: 81.531-990, Curitiba, PR, Brazil

The closing date is October 18, 1996. Late applications will however be
considered until positions are filled. There are also some short time
(for example, three months) positions opened for researchers. The salary
is competitive and dependent on experience.


From: Keith Geddes <>
Date: Thu, 11 Jul 1996 14:30:25 -0400 (EDT)
Subject: Positions at the University of Waterloo

Two Research Positions
The University of Waterloo

Hybrid Computing Position:

We are interested in research areas where symbolic computing has
the potential to yield improvements in the numerical approximation
of engineering (or applied science) problems.

This position requires substantial breadth in computational
engineering, a solid mathematical foundation, and some experience
with computer algebra systems.

Numerical Special Function Position:

We are interested in research areas where the rapid, high-precision
evaluation of special functions yields novel, or significantly improved,
numerical approximation of applied science (or engineering) problems.

This position requires substantial depth in applied mathematics, a
solid programming background, and some familiarity with modern
software testing techniques. Some of the development effort of the
person holding this position will be devoted to extending the Maple
fixed-precision and arbitrary-precision special function libraries.

Either or both positions available January 1, 1997:

- at the level of Research Associate or Postdoctoral Fellow
- one-year renewable positions, dependent on funding
- candidate must have a Ph.D. (or equivalent) in computer science,
engineering, or mathematics, with a research emphasis in computational

Appointments will be through the Symbolic Computation Group (SCG) in the
Department of Computer Science, University of Waterloo. SCG has been a
center of research on algorithms and systems for computer algebra since
1980. The commercial computer algebra system, Maple, was originated at
SCG and the group continues to carry out collaborative research and
development with Waterloo Maple Inc.

Send CV, and arrange to have at least two letters of reference sent,
by October 1, 1996 to either:

Prof. Keith Geddes Prof. George Labahn
Symbolic Computation Group Symbolic Computation Group
Department of Computer Science Department of Computer Science
University of Waterloo University of Waterloo
Waterloo ON N2L 3G1 Waterloo ON N2L 3G1
Canada Canada

Keith Geddes (
George Labahn (

Co-directors, Symbolic Computation Group


From: SIAM <>
Date: Tue, 09 Jul 96 15:24:32 EST
Subject: Contents, SIAM Optimization

SIAM Journal on Optimization
August 1996, Volume 6, Number 3

Ill-Posedness and the Complexity of Deciding Existence of Solutions to Linear
Jorge R. Vera

An Asymptotical O(nL)-Iteration Path-Following Linear Programming Algorithm That
Uses Wide Neighborhoods
Pi-Fang Hung and Yinyu Ye

Two Infeasible Interior-Point Predictor-Corrector Algorithms for Linear
Jianming Miao

A New Finite Continuation Algorithm for Linear Programming
Kaj Madsen, Hans Bruun Nielsen, and Mustafa C. Pinar

A Conjugate Duality Scheme Generating a New Class of Differentiable Duals
Aharon Ben-Tal and Marc Teboulle

Asymptotic Convergence Analysis of Some Inexact Proximal Point Algorithms for
Ciyou Zhu

Complexity Analysis of an Interior Cutting Plane Method for Convex Feasibility
Jean-Louis Goffin, Zhi-Quan Luo, and Yinyu Ye

Tensor Methods for Equality Constrained Optimization
Dan Feng and Robert B. Schnabel

Convergence Properties of an Augmented Lagrangian Algorithm for Optimization
With a Combination of General Equality and Linear Constraints
A. R. Conn, N. Gould, A. Sartenaer, and Ph. L. Toint

A New Merit Function and a Successive Quadratic Programming Algorithm for
Variational Inequality Problems
Kouichi Taji and Masao Fukushima

Co-Coercivity and Its Role in the Convergence of Iterative Schemes for Solving
Variational Inequalities
D. L. Zhu and P. Marcotte

Computing the Moore-Penrose Inverse for the Covariance Matrix in Constrained
Nonlinear Estimation
Wolfgang M. Hartmann and Robert E. Hartwig

The Effective Energy Transformation Scheme as a Special Continuation Approach to
Global Optimization With Application to Molecular Conformation
Zhijun Wu

Viscosity Solutions of Minimization Problems
Hedy Attouch

Incremental Least Squares Methods and the Extended Kalman Filter
Dimitri P. Bertsekas

Network Design Using Cut Inequalities
Francisco Barahona

A Bilevel Programming Method for Pipe Network Optimization
Jianzhong Zhang and Detong Zhu

Solving Multiple Knapsack Problems by Cutting Planes
C. E. Ferreira, A. Martin, and R. Weismantel


From: SIAM <>
Date: Thu, 11 Jul 96 14:39:04 EST
Subject: Contents, SIAM Control and Optimization

Control and Optimization
SEPTEMBER 1996 Volume 34, Number 5

A Uniqueness Result for the Linear System of Elasticity and Its Control
Theoretical Consequences
Enrique Zuazua

New Necessary Conditions for the Generalized Problem of Bolza
P. D. Loewen and R. T. Rockafellar

Boundary Exact Controllability of Interface Problems with Singularities I:
Addition of the Coefficients of Singularities
Serge Nicaise

Approximate Feedback Linearization: A Homotopy Operator Approach
Andrzej Banaszuk and John Hauser

Perturbed Optimization in Banach Spaces III: Semi-infinite Optimization
J. Frederic Bonnans and Robert Cominetti

Viscosity Solutions and Viscosity Subderivatives in Smooth Banach Spaces with
Applications to Metric Regularity
Jonathan M. Borwein and Qiji J. Zhu

Optimal Relaxed Controls for Infinite-Dimensional Stochastic Systems of Zakai
N. U. Ahmed

Stabilization by Constrained Controls
Georgi V. Smirnov

Minimax Rendezvous on the Line
Wei Shi Lim and Steve Alpern

Polynomial Filtering for Nonlinear Discrete time Non-Gaussian Systems
Francesco Carravetta, Alfredo Germani, and Massimo Raimondi

An Entropy Formula for Time-Varying Discrete-Time Control Systems
Pablo A. Iglesias

On Markovian Fragments of COCOLOG for Logic Control Systems
Yuanjun Wei and Peter E. Caines

Model Simplification and Optimal Control of Stochastic Singularly Perturbed
Systems Under Exponentiated Quadratic Cost
Zigang Pan and Tamer Basar

Analysis and Optimization of Feedback Control Algorithms for Data Transfers in
High-Speed Networks
Rauf Izmailov

Heavy Traffic Analysis of a Controlled Multiclass Queueing Network via Weak
Convergence Methods
Harold J. Kushner and L. Felipe Martins

Stabilizability Does Not Imply Homogeneous Stabilizability for Controllable
Homogeneous Systems
Rodolphe Sepulchre and Dirk Aeyels

Modified Projection-Type Methods for Monotone Variational Inequalities
Michael V. Solodov and Paul Tseng


From: Oleg Burdakov <>
Date: Thu, 4 Jul 1996 18:20:44 +0100
Subject: Contents. Optimization Methods and Software

Table of Contents
Optimization Methods and Software (OMS)
Volume 7, Number 1 (July, 1996)

Chr.H. Bischof, P.M. Khademi, A. Bouaricha, and A. Carle
Efficient computation of gradients and Jacobians by dynamic exploitation

C.G. Broyden
A breakdown of the block-CG method

G.E. Stavroulakis and L.N. Polyakova
Difference convex optimization techniques in nonsmooth computational


From: E. B. Saff <>
Date: Wed, 10 Jul 1996 16:21:58 -0400
Subject: Contents, Constructive Approximation

Table of Contents: Const. Approx., Vol. 12, No. 3, 1996

307 B. Beckermann and C. Carstensen
QD-Type Algorithms for the Nonnormal Newton-Pade
Approximation Table

331 R. Schaback
Approximation by Radial Basis Functions with Finitely
Many Centers

341 J. J. Guadalupe, M. Perez, F. J. Ruiz, and J. L. Varona
Weighted Norm Inequalities for Polynomial Expansions
Associated to Some Measures with Mass Points

361 F. Deutsch, V. A. Ubhaya, J. D. Ward, and Yuesheng Xu
Constrained Best Approximation in Hilbert Space III.
Applications to n-Convex Functions

385 P. J. Berry
de Boor-Fix Dual Functionals and Algorithms for
Tchebycheffian B-Spline Curves

409 R. A. DeVore, Y. K. Hu, and D. Leviatan
Convex Polynomial and Spline Approximation in
Lp, 0 < p < infinity

423 L. Baratchart, J. Leblond, and J. R. Partington
Hardy Approximation to L^infinity Functions on Subsets
of the Circle

437 A. L. Levin and E. B. Saff
Erratum: Optimal Ray Sequences of Rational Functions
Connected with the Zolotarev Problem

439 P. Borwein and T. Erdelyi
Questions About Polynomials with {0, -1, +1} Coefficients:
Research Problem 96-3


End of NA Digest