NA Digest Sunday, September 11, 1988 Volume 88 : Issue 36

Today's Editor: Cleve Moler

Today's Topics:


From: Dennis Gannon <>
Date: Tue, 6 Sep 88 22:02:27 EST
Subject: Int. Conf. on Supercomputing 1989

June 5-9, Crete, Greece

Conference Co-Chairmen
George Paul, IBM USA T. Papatheodorou, CTI Greece

Program Committee Directors
D. Gannon, Indiana Co-Chairman
E. N. Houstis, Purdue Co-Chairman
F. Hossfeld, KFA Chairman Europe and Africa
Y. Muraoka, Waseda Chairman Japan and Far East
J. Sopka, DEC Chairman North and South America

The International Conference on Supercomputing is now soliciting papers
for its third year. The proceedings, published by Springer-Verlag in
1987 and ACM in 1988 will again be published by ACM in 1989. Papers
are solicited in the following areas.

Applications of Supercomputing including:
Computational Mechanics, Fluid Dynamics and Astronomy,
Simulation and Modeling from Physics, Chemistry and Biology,
Artificial Intelligence and Symbolic Computation,
Graphics and Visualization,
Mathematical Software, Numerical Algorithms and Theoretical Studies.
Software Systems including:
Operating Systems,
Parallel Languages, Compilers and Automatic Parallelization Tools,
Performance Evaluation Tools, Methods and Modeling,
Programming Environments and Hight Level Problem Solving Systems.
MIMD, SIMD and Data Flow Systems Architectures,
Memory System Design (Distributed, Shared or Hierarchial),
Bus, Network and Communication Systems,
Instruction Architecture (RISC, CISC, etc.) and Performance Studies.

Authors should send five (5) copies of the manuscript to the program
chairman of their region. The deadline for submissions is January 10,1989.
Authors will be notified of acceptance by March 10. The addresses for
submissions are:

Europe and Africa: North and South America: Japan and Far East:
Dr. Friedel Hossfeld John. R. Sopka Dr. Yoichi Muraoka
KFA Julich Digital Equipment Corp. Dept. of Electrical Eng.
ZAM 110 Spit Brook Road Waseda University
Postfach 1913 Nashua, New Hampshire 3-4-1 Okubo, Shinjuku-ku
D-5170 Julich zip: 03062-2698 Tokyo, Japan
Fed. Rep. of Germany USA

A full list of invited speakers and program committee members will
follow on a later posting.


From: Regis Bossut <RPARBS%FRFUPL11.BITNET@Forsythe.Stanford.EDU>
Date: Wed, 07 Sep 88 10:28:43 EDT
Subject: Reciprocal of Complementary Error Function


I am currently looking for an implementation of the reciprocal of complimentary
error function ERFC, in FORTRAN, C or IBM370 Assembler. I have been looking
for it in the NANET libraries, in ABRAMOVICZ's book, and in GRADSTEIN's book
without any response. So, I have been implementing it by the NEWTON iterative
algorithm on the ERFC function, but this is slow and not optimized. Can anybody
help me ? Many thanks in advance.



From: Kuo Long Wu <>
Date: Fri, 9 Sep 88 17:31:59 CDT
Subject: Sensitivity of Householder Tridiagonalization

Dear friends :

I have the following problems related to the routines TRED1 and TRED2 in
the EISPACK. The original routines in TRED1 and TRED2 did the Symmetric matrix
vector operation based on Inner Product calculation. But the new BLAS2 and
BLAS3 routines SSYMV and SSYMM did the operations based on Saxpy operations.
The problems are :

1. The inner product operation and Saxpy operation on the vector
architecture machine perform different results. I used the uniform
random matrix to test the Householder transformation to reduce
the symmetric matrice to tridiagonal. The results are very different
even the matrice size is 70, it will be greater than 1.0e-7. When the
matrice size is 500, it will greater than 1.0e-3. Is anyone can help
to make the result getting better ?

2. If this is true, then the BLAS2 and BLAS3 will introduce the similar
problems as this one. Is anyone can estimate the error between the
original BLAS approach and BLAS2-BLAS3 ?

Please send your suggestions to me, it will be greatly appreciated. My address
is :
E-mail will be :

[Comment from the editor:

This question illustrates a common misunderstanding about the
"accuracy" of Householder, and other reductions, to tridiagonal
and Hessenberg forms. In the symmetric case, the EISPACK routines
TRED1/2/3 start with a real symmetric matrix, A. They produce an
orthogonal matrix H, which is the product of n-2 Householder
reflections, so that

T = H'*A*H

is tridiagonal. The roundoff error analysis guarantees that
the computed H is orthogonal to machine accuracy and that T
is exactly similar to some machine accuracy sized perturbation
of A . This implies that the eigenvalues and transformed
eigenvectors of T are accurate approximations to those of A.
But there are lots of T's and H's for which this is true.
There is no guarantee that the T's and H's computed by different
implementations or on different machines are close to each other.

If the Householder reduction is followed by an eigenvalue
routine, such as TQL2, the result is an nearly orthogonal X
so that

D = X'*A*X

is diagonal. The diagonal elements of D are the eigenvalues and
the columns of X are the corresponding eigenvectors. Again,
D and X are not unique. Different implementations or different
floating point arithmetic may produce very different, but still
satisfactory, D's and X's. The elements of D may come in
different orders; the signs of the columns of X may be different;
if there are multiple eigenvalues, the corresponding columns of
X may be linear combinations of each other.

So, it is not "fair" to expect a Householder tridiagonalization
routine to produce something close to one fixed "right answer".
The only reasonable tests are to check that

H is close to orthogonal
H*T is close to A*H.

Better yet, complete the eigenvalue/vector computation and check that

X is close to orthogonal
X*D is close to A*X.

(These checks should be made relative to some norm of A.)

-- Cleve Moler ]


From: David Kincaid <>
Date: Fri, 9 Sep 88 22:53:40 CDT
Subject: Conference on Iterative Methods for Large Linear Systems


Don't wait until October --- the conference hotel is filling up fast!
For those staying over, the conference hotel seems to
be booked for the weekend after this conference.
However, there are many available hotels in Austin.
Information on some nearby hotels is given at the end of this announcement.

Registration Special: As a one time special, the registration rate will be
$95 until September 15, 1988.

Banquet Anecdotes: At the banquet, we will be asking those present if they
care to stand and share an anecdote or two related to their past
association with David Young. Since this will serve in place of an after
dinner speech, please give some thought to possible remarks you might make.
We hope to hear a number of short recollections of past events both
humorous, serious and heartfelt.

Cards & Letters: At the banquet, messages of best wishes (telegrams/
mailgrams/cards/letter of best-wishes) will be read from anyone who cannot
attend but wants to send in one beforehand.

(Almost) Final Program (as of 9/10/88)

October 19-21, 1988
Center for Numerical Analysis
The University of Texas at Austin

All conference events (meetings and social) will be held at the Austin
Marriott at the Capitol, 701 E. 11th St., Austin, Texas 78701.
The hotel is located in downtown Austin at the corner of 11th Stret
and Interstate 35.

Co-sponsored by
Society for Industrial and Applied Mathematics
Special Interest Groups for Linear Algebra and Supercomputing

Celebrating the Sixty-fifth Birthday of
David M. Young, Jr.

Tuesday (October 18, 1988)

5:30p-7:00p Pre-Conference Social (On-your-own) ---
Calypso Bar / Second Level

5:30p-7:00p Pre-Conference Registration --- Third Level

Wednesday (October 19, 1988)

8:00a-5:00p Registration / Information --- Third Level

8:20a-8:30a Opening Remarks --- Salon D / Third Level

8:30a-9:30a Garrett Birkhoff * speaker (Harvard University)
Robert E. Lynch (Purdue University)
``ELLPACK and ITPACK as Research Tools for
Solving Elliptic Problems''

9:30a-9:50a Robert E. Lynch* (Purdue University)
``New Finite Difference Approximations for Boundary Conditons''

9:50a-10:20a Coffee Break

10:20a-11:00a David M. Young* (University of Texas at Austin)
Tsun-zee Mai (University of Alabama)
``The Search for Omega''

11:00a-11:40a Owe Axelsson* (University of Nijmegen, The Netherlands)
``Some Optimal Order Preconditioning Methods for
Diffusion Problems Based on
Algebraic Decompositions''

11:40a-12:00n John Whiteman* (Brunel University, U.K.)
``Finite Element Treatment of Singularities in
Elliptic Boundary Value Problems''

12:00n-1:30p Lunch (On-your-own)

1:30p-2:10p Mary F. Wheeler* (University of Houston)
``Domain Decomposition --- Multigrid Algorithms for Mixed
Finite Element Methods for Elliptic PDE's''

2:10p-2:50p Olof Widlund* (New York University)
``Domain Decomposition Algorithms for Elliptic Problems''

2:50p-3:00p Stretch Break with Celeste Hamman, fitness consultant

3:00p-3:20p Coffee Break

Parallel Session:

3:20p-3:40p Paul J. Lanzkron, Donald J. Rose* Daniel B. Szyld
(Duke University)
`Convergence of Nested Iterative Methods for Linear Systems''

3:40p-4:00p David J. Evans* C. Li (Loughborough University of
Technology, U.K.)
``D 1/2 -Norms of the SOR and Related Method for a
Class of Nonsymmetric Matrices''

4:20p-4:40p Martin Hanke* (Universitat Karlsruhe, W. Germany)
``On Kaczmarz' Method for Inconsistent Linear Systems''

4:40p-5:00p Steve F. Ashby (Lawrence Livermore National Labortory)
``Polynomial Preconditioning for Conjugate Gradient Methods''

5:00p-5:20p Randall Bramley (University of Illinois at Urbana-Champaign)
``A Projection Method for Large Sparse Linear Systems''

Parallel Session:

3:20p-3:40p Robert E. Wyatt* (University of Texas at Austin)
``Iterative Methods in Molecular Collision Theory''

4:00p-4:20p M. Dryja (University of Warsaw, Poland)
W. Proskurowski* (University of Southern California)
`Composition Method for Solving Elliptic Problems''

4:20p-4:40p Seungsoo Lee, George S. Dulikravich*
Daniel J. Dorney (Pennsylvania State University)
``Distributed Minimal Residual (DMR) Method for
Explicit Algorithms Applied to Nonlinear Systems''

4:40p-5:00p M. G. Petkov* (Academy of Sciences, Bulgaria)
``On the Matrix Geometric Progression and the Jordan Canonial Form''

5:00p-5:20p David V. Anderson* & Alice Koniges (Lawrence Livermore National
Laboratory) ``to be announced''

5:20p-7:00p Reception (Light Hors D'oeuvres)---Calypso Terrace / Second Level

7:00p Dinner (On-your-own)

8:00p-10:00p Tennis Doubles-Mixer
Penick-Allison Tennis Center --- corner of Trinity and MLK Blvd

Thursday, October 20, 1988

8:00a-4:00p Registration / Information --- Third Level

8:20a-8:30a Second Day Remarks --- Salon D / Third Level

8:30a-9:10a D. L. Harrar James M. Ortega* (University of Virginia)
``Solution of Three-Dimensional Generalized Poisson
Equations on Vector Computers''

9:10a-9:50a Paul Saylor* (University of Illinois)
``Iterative Methods for Complex Linear Algebraic Equations''

9:50a-10:20a Coffee Break

10:20a-11:00a Richard S. Varga* (Kent State University)
``Remarks on -Step Iterative Methods''

11:00a-11:40a Louis W. Ehrlich* (John Hopkins University)
``A Local Relaxation Scheme (Ad-Hoc SOR) Applied to
Nine Point and Block Difference Equations''

11:40a-12:00n Paul Concus* (Lawrence Berkeley Laboratory)
Paul Saylor (University of Illinois)
``Preconditioned Iterative Methods for Indefinite
Symmetric Toeplitz Matrices''

12:00n-1:30p Luncheon --- Salon E / Third Level

1:30p-2:10p Howard C. Elman* (University of Maryland)
``Uses of Reordering, Partial Elimination and Fourier
Methods for Sparse Iterative Solvers''

2:10p-2:50p Loyce M. Adams* (University of Washington)
``Fourier Analysis of Two-Level Hierarchical Basis Preconditioners''

2:50p-3:10p C.-C. Jay Kuo Tony F. Chan*
(University of California, Los Angeles)
``Two-color Fourier Analysis of Iterative
Methods for Elliptic Problems with Red-Black Ordering''

3:10p-3:30p Coffee Break

3:30p-3:40p Stretch Break with Celeste Hamman, fitness consultant

3:40p-4:00p Kang C. Jea* (Fu Jen University, Taiwan)
David M. Young (University of Texas at Austin)
``On The Effectiveness of Adaptive Chebyshev Acceleration
for Solving Systems of Linear Equations''

4:00p-4:20p Anne Greenbaum* (New York University)
``Predicting the Behavior of Finite Precision Lanczos and
Conjugate Gradient Computations''

4:20p-4:40p Tsun-zee Mai* (University of Alabama)
David M. Young (University of Texas at Austin)
``On the Adaptive Determination of Iteration Parameters''

4:40p-5:00p David R. Kincaid* (University of Texas at Austin)
``A Status Report on the ITPACK Project''

5:00p-7:00p Conference Social (Cash Bar) --- Foyer / Fourth Level

7:00p Banquet --- Ballroom (Salon A -- D) / Third Level

Friday (October 21, 1988)

8:00a-4:00p Registration/Information --- Third Level

8:20a-8:30a Final Day Remarks --- Salon D / Third Level

8:30a-9:10a Gene Golub* (Stanford University)
John de Pillis (University of California, Riverside)
``Toward an Effective Two-parameter SOR Method''

9:10a-9:50a Eugene L. Wachspress* (University of Tennessee)
``The ADI Minimax Problem for Complex Spectra''

9:50a-10:20a Coffee Break

10:20a-11:00a Thomas Manteuffel* (University of Colorado at Denver
Los Alamos National Laboratories)
Wayne D. Joubert (University of Texas at Austin)
``Iterative Methods for Nonsymmetric Linear Systems''

11:00a-11:40a Louis A. Hageman* (Westinghouse --- Bettis Laboratory)
``Relaxation Parameters for the IQE Iterative Procedure for
Solving Semi-Implicit Navier-Stokes Difference Equations''

11:40a-12:00n C. Douglas, J. Mandel W. Miranker* (IBM Watson
Research Center, Yorktown Heights)
``Fast Hybrid Solution of Algebraic Systems''

12:00n-1:10p Luncheon --- Salon E / Third Level

1:10p-1:50p Dan C. Marinescu John R. Rice* (Purdue University)
``Multilevel Asynchronous Iterations for PDE's''

1:50p-2:00p Stretch Break with Celeste Hamman, fitness consultant

2:00p-2:20p Coffee Break

Parallel Session:

2:20p-2:40p Avi Lin* (Temple University)
``Asynchronous Parallel Iterative Methods''

2:40p-3:00p Thomas C. Oppe (University of Texas at Austin)
``Experiments with a Parallel Iterative Package''

3:00p-3:20 I.M. Navon* H.-I. Lu (Florida State University)
``A Benchmark Comparison of the ITPACK Package on ETA-10 and
Cyber-205 Supercomputers''

3:20p-3:40p Hungwen Li, Ming-Cheng Sheng & Anne C. Elster
(IBM Watson Research Center, Yorktown Heights, & Cornell University)
``Sparse Matrix-Vector Multiplication on the Polymorphic Torus''

3:40p-4:00p Jerome Dancis* (University of Maryland)
``Diagonalizing the Adaptive SOR Iteration Method''

Parallel Session:

2:20p-2:40p S. Galanis, Apostolos Hadjidimos* D. Noutsos (University of
Ioannina, Greece, Purdue University)
``On an SSOR Matrix Relationship and Its Consequences''

2:40p-3:00p A. Haegemans J. Verbeke* (Katholieke Universiteit
Leuven, Belgium)
``The Symmetric Generalized Accelerated Overrelaxation (GSAOR) Method''

3:00p-3:20p Apostolos Hadjidimos (Purdue University University of
Ioannina, Greece)
Michael Neumann* (Universtiy of Connecticut)
``Convergence Domains and Inequalities for the Symmetric SOR Method''

3:20p-3:40p Ezio Venturion (University of Iowa)
``to be announced''

3:40p-4:00p Kaibing Hwang* Jinru Chen (Nanjing Normal University,P.R. China)
``A New Class of Methods for Solving Nonsymmetric Systems of Linear
Equations --- Constructing and Realizing Symmetrizable Iterative Methods''

Conference Adjourns

This conference will be dedicated to providing an overview of the state of
the art in the use of iterative methods for solving sparse linear systems
with an eye to contributions of the past, present, and future.
The emphasis will be placed upon identifying current and future
research directions in the mainstream of modern scientific computing.

Recently, the use of iterative methods for solving linear systems
has experienced a resurgence of activity as scientists attack extremely
complicated three dimensional problems using vector and parallel
Many research advances in the development of iterative methods for
high-speed computers over the past forty years will be reviewed
as well as focusing on current research.

The conference is organized by
D. Kincaid, L. Hayes, G. Carey and W. Cheney, who are members of
the host organization --- the Center for Numerical Analysis (CNA)
of The University of Texas at Austin.
This meeting is being co-sponsored by the Special Interest Groups for
Linear Algebra and Supercomputing of the
Society for Industrial and Applied Mathematics.
Support for this conference is provided, in part, by
the Office of Naval Research,
the Department of Energy,
the National Science Foundation,
the Air Force Office of Scientific Research,
and The University of Texas at Austin.

To make reservations call the Austin Marriott at the Capitol,
[(512) 478-1111 or (800) 228-9290] for rooms at
the special conference rate of 55 for single or double rooms.
All reservations are handled on a first-come-first-served basis.
Reservations must be received by October 4, 1988.
Reservations made after this date are subject to guest room availability.

Many major airlines fly into Austin via Dallas or Houston with some
direct flights from other locations.
The primary air-carriers serving Austin are
American, American-West, Continental, Delta, Northwest, Pan Am,
Southwest, TWA, United, and USAir.
Since airlines give discount rates for those staying over a Saturday night,
plan to stay and enjoy the weekend in Austin!
An information desk operated by the City of Austin is located
in the airport and is a good source for free material on events and
sights of interest
in and around Austin. (Just ask for a packet of information.)

A hotel courtesy-van is available
for transportation between the airport and the hotel
(a short 15 minute trip).
[Regular hours of operation are on the hour and half-hour from
6:00a-12:00n and by request 12:00n-12:00m with frequent trips
Tuesday evening before the conference and Friday evening after
the conference.]
The hotel also operates a free shuttle bus to The University of
Texas at Austin which is approximately ten blocks north.
[See the posted schedule in the hotel.]
Please contact the Bell Station to confirm transportation.
The City operates the ``catch a'dillo bus'' (short for armadillo)
for transportation around the downtown area.

On Wednesday evening, a tennis doubles-mixer will be arranged.
So bring your tennis gear!

Celeste Hamman, fitness consultant, will lead daily stretch breaks
each day during the conference to energize us for the late afternoon sessions.
For those interested, Celeste will lead a supervised jog
around the Capitol/University area during the Wednesday
lunch break (leaving at 12:15pm from the front of the Marriott
and finishing at 1:00pm).
Also, she will be available during the conference for individual
instruction on subjects such as conditioning, race-walking, etc.

Austin is the capitol of Texas with several points of interest
including the State Capitol Building, Governor's Mansion, Lyndon B.
Johnson Presidential Library and Museum, Zilker Park and Barton
Springs (spring-fed natural swimming pool always ), and many
Near to Austin are several scenic lakes, such as Lake Travis,
and the ``Texas hill country.''
Austin is in the center of Texas with historic San Antonio and
the Alamo only 70 miles south,
Dallas/Ft. Worth 200 miles north, and Houston 200 miles Southeast.
The weather is usually quite pleasant in Austin in October
but is known to change rapidly (October averages: high, low).
Many restaurants and night-spots are located on
``6th Street'' five blocks south of the hotel.
South of 1st Street is ``Town Lake'' with the popular ``hike-`n-bike'' trail
for jogging, speed-walking, or an enjoyable stroll.
Austin also offers a host of other activities that
participants can individually arrange.
Some information will be available at the conference table.

The long papers from the conference will be published by Academic
Press in book form and will appear in 1989.
Pre-publication orders can be placed during the conference.

There are several activities on The University of Texas at
Austin campus on the Monday and Tuesday precedings this conference.
The following events may be of interest:
(1) a ``Workshop on Parallel and Vector Computing'' hosted by
the Texas Institute of Computational Mechanics
[contact: Dr. Graham Carey, WRW 305G, TICOM, UT Austin, Austin, TX
78712 or call Pat Bozman (512) 471-4676]
(2) the fifth annual ``Computing Science Research Review'' hosted by
the Department of Computer Sciences [contact: Joanne Click
(512) 471-9729,]
(3) the ``Robert Todd Gregory Memorial Lecture'' by
Professor Garrett Birkhoff on late Tuesday afternoon in the Department
of Mathematics.

Conference advanced registration fee is $95, if received by September 15, 1988.
If paid after this date or at the conference, the registration fee is $125.
This fee includes morning and afternoon coffee breaks,
two luncheons, a reception, and a banquet honoring Professor Young.
The student registration fee is $18, which allows admission to the technical
sessions and coffee breaks only.
The evening social activities of the conference are available to companions of
conference participants at the following rates:
$10.75 reception (Wednesday night), $15 each luncheon (Thursday or Friday),
$27.50 banquet (Thursday night),
no charge for conference socials (cash bar).
To register, detach and mail the registration form below.
For additional information, contact the CNA at the address below or at
Tel: (512) 471-1242 ; Arpanet:;
Bitnet: sheri@uta3081.

(detach and mail)





Number of guest banquet tickets ( 27.50 per person):

Amount Enclosed:
(Make checks payable to The University of Texas at Austin.)

[]Yes, I plan to attend. Enclosed please find my registration fee of

[]Yes, please sign me up to play in the tennis doubles-mixer.

Iterative Conference Hotel & Some Nearby Hotels

1. Austin Marriott at the Capitol $55.00 (single or double)
701 E. 11th St.
Austin, TX 78701
(512) 478-1111

2. Rodeway Inn-Capitol $28.95 (single) $34.95 (double)
1201 N. IH-35
Austin, TX
(512) 472-8331
(3-4 blocks east; across freeway bridge)

3. Ramada Inn-Capitol $39.00 (single or double)
300 E. 11th
Austin, TX
(4-5 blocks west; down and up hill)

4. Waller Creek Plaza Hotel $54.00 (single or double)
500 IH 35N
Austin, TX
(6 blocks South, down freeway access road)


End of NA Digest