### Today's Editor:

- Cleve Moler
- The MathWorks, Inc.
- moler@mathworks.com

- LAPACK is Now Available
- Second Release of XNETLIB
- New Address for Bob Voight
- Butcher's Birthday Problem for Gene
- Looking for E-Mail Addresses
- High Performance Computing and Communications Initiative
- Conference at Kent State on Numerical Linear Algebra
- Intelligent Scientific Computation at AAAI Meeting
- One Day MATLAB Short Course in UK
- Copper Mountain Progran and Information
- Householder Symposium XII
- Appalachian Numerical Analysis Day
- Conference in Italy on Innovative Methods in Numerical Analysis
- Report on the Tenth Parallel Circus
- Contents: Numerical Algorithms

-------------------------------------------------------

From: Jack Dongarra <dongarra@cs.utk.edu>

Date: Fri, 6 Mar 92 12:36:44 -0500

**Subject: LAPACK is Now Available**

( L A P A C K )

( L -A P -A C -K )

( L A P A -C -K )

( L -A P -A -C K )

( L A -P -A C K )

( L -A -P A C -K )

LAPACK is a transportable library of Fortran 77 subroutines for

solving the most common problems in numerical linear algebra: systems

of linear equations, linear least squares problems, eigenvalue problems,

and singular value problems. It has been designed to be efficient

on a wide range of modern high-performance computers.

LAPACK is intended to be the successor to LINPACK and EISPACK.

It extends the functionality of these packages by including driver

routines, iterative refinement and error bounds for linear systems, the

capability for finding selected eigenvalues and invariant subspaces, and

condition estimation for the eigenproblem. LAPACK improves on the accuracy

of standard algorithms for linear systems, for finding singular values and

singular vectors of bidiagonal matrices, and for finding eigenvalues and

eigenvectors of tridiagonal matrices. The algorithms and software in the

package have been restructured to achieve high efficiency on vector processors,

high-performance ``superscalar'' workstations, and shared memory

multiprocessors. In addition to the LAPACK routines, comprehensive

testing and timing suite is provided along with the LAPACK software.

The LAPACK Users' Guide will be available from SIAM in or around May, 1992.

The Users' Guide gives an informal introduction to the design of the algorithms

and software, summarizes the contents of the package, describes conventions

used in the software and documentation, and includes complete specifications

for calling the routines.

The LAPACK routines are available from netlib for users who require only

selected pieces of the package.

For a description of the contents of LAPACK, send email to

netlib@ornl.gov, and in the mail message type: send index from lapack.

A set of working notes are available from netlib as well and details

can be found in the index to LAPACK.

The complete LAPACK package can be obtained on magnetic media from NAG for a

nominal handling charge. For further details, contact NAG at one of these

addresses:

NAG Inc NAG Ltd NAG GmbH

1400 Opus Place Wilkinson House Schleissheimerstrasse 5

Suite 200 Jordan Hill Road W-8046 Garching bei Munchen

Downers Grove, IL 60515-5702 Oxford OX2 8DR Germany

USA England

Tel: +1 708 971 2337 Tel: +44 865 511245 Tel: +49 89 3207395

Fax: +1 708 971 2706 Fax: +44 865 310139 Fax: +49 89 3207396

LAPACK has been funded in part by NSF, DOE, and DARPA, with

developmental support from NAG Ltd., Cray Research, and many friends

and colleagues around the world.

Ed Anderson, Zhao-jun Bai, Chris Bischof, Jim Demmel, Jack Dongarra,

Jeremy Du Croz, Anne Greenbaum, Sven Hammarling, Alan McKenney,

Susan Ostrouchov, and Danny Sorensen

( l l l l )

( a -a a -a )

1/4 * ( p p -p -p )

( a -a -a a )

( c c -c -c )

( k -k -k k )

------------------------------

From: Jack Dongarra <dongarra@cs.utk.edu>

Date: Fri, 6 Mar 92 17:21:23 -0500

**Subject: Second Release of XNETLIB**

We have a second release version of XNETLIB.

XNETLIB is a new version of NETLIB recently developed at the University

of Tennessee and Oak Ridge National Laboratory. Unlike NETLIB, which

uses electronic mail to process requests for mathematical software,

XNETLIB uses an X Window graphical user interface and a socket-based

connection between the user's machine and the XNETLIB server machine to

process software requests.

To receive a copy of XNETLIB send the message "send xnetlib.shar from

xnetlib" to netlib@ornl.gov.

When you receive the shar file, remove the mail header, save it to a

file, type 'sh filename' and follow the instructions in the README

file.

Send comments to xnetlib@cs.utk.edu.

Jack Dongarra

Tom Rowan

Reed Wade

------------------------------

From: Robert G. Voigt <rgv@icase.edu>

Date: Mon, 2 Mar 92 17:20:46 -0500

**Subject: New Address for Bob Voight**

Position Change

Effective immediately my address is

Robert G. Voigt

Room 417

National Science Foundation

1800 G Street, N.W.

Washington, DC

20550

Phone: (202) 357-7727

email: rvoigt@nsf.gov

I am Program Manager for the New Technologies program in the Division

of Advanced Scientific Computing. I look forward to continued

interaction with the scientific computing community.

------------------------------

From: Walter Gander <gander@inf.ethz.ch>

Date: Tue, 3 Mar 92 11:22:55 +0100

**Subject: Butcher's Birthday Problem for Gene**

There were some requests for the problem that John Butcher asked me to

give as birthday present to Gene on occasion of his 60th birthday.

Here it is.

- Walter Gander

\centerline{\bf A birthday problem for Gene Golub}

\bigskip

\centerline{\bf composed by John Butcher}

\bigskip

Let $n$ be a positive integer and $t$ a real number. Define

$a_0$, $a_1$, $a_2$, \dots, $a_n$, and $b_0$, $b_1$, $b_2$, \dots, $b_n$, by

$$\halign{\qquad$#$\hfil&$#$\hfil\qquad&$#$\hfil&$#$\hfil\cr

a_0 & = 1, & b_0 & = 1,\cr

a_1 & = t-1, & b_1 & = 1-t,\cr

a_k & = {(-1)^{k-1}\over k+1}L^\prime_{k+1}((k+1)t),\quad k\ge2,

& b_k & = {(-1)^{k-1}\over k}tL^\prime_{k}((k-1)t),\quad k\ge2, \cr

}

$$

where

$$L_n(t) = 1-{n\choose 1}t + {1\over 2!}{n\choose2}t^2-\cdots+(-1)^n{1\over n!}t^n$$

denotes a Laguerre polynomial.

\bigskip

\noindent{\bf Problem 1.} Prove that

$$

a_0 b_n + a_1 b_{n-1} + a_2 b_{n-2} + \cdots + a_n b_0 = 0.

$$

\bigskip

\noindent{\bf Problem 2.} Find an application to numerical ordinary

differential equations.

\bye

------------------------------

From: Mustafa Pinar <pinar@desci.wharton.upenn.edu<

Date: Thu, 5 Mar 92 12:04:43 EST

**Subject: Looking for E-Mail Addresses**

I am trying to get in touch with the following people

as soon as possible through e-mail. I would appreciate it

if I could get their e-mail addresses from the NANET users

or if someone could direct me to a directory. The names are:

Susana Gomez, Instituto Investigaciones en Mathematicas

Applicadas y en Sistemas-Universidad Nacional Autonoma

de Mexico, Mexico.

Gianni DiPillo, Universita di Roma " La Sapienza", Italy.

Luigi Grippo , same address as DiPillo.

Mei-Qin Chen, The Citadel.

Thank you very much.

And also thanks to all users who responded to my earlier

paper request from a Polish Journal.

Mustafa Pinar

------------------------------

From: Judith Sunley <jsunley@nsf.gov>

Date: Thu, 5 Mar 92 11:28:07 PST

**Subject: High Performance Computing and Communications Initiative**

NOTICE RE: HPCC (High Performance Computing and Communications) Initiative,

and Mathematical Scientists

It has come to our attention that there are some misunderstandings with

regard to NSF activities under the HPCC rubric.

It is essential that all be aware and understand that the Grand Challenge

Application Groups activity, of which you have been informed recently, is

NOT the only avenue for members of the Mathematical Sciences community to

participate in the HPCC initiative.

Individual and group research activities which fit within the guidelines of

HPCC can apply directly to the most appropriate programs within NSF,

including in particular any of the programs in the Division of Mathematical

Sciences, as well as programs in the Computer and Information Sciences

Directorate at NSF, and the recently-announced Computational Approaches to

Real Materials activity. There is no deadline nor pre-proposal letter of

intent required for such activities.

Program directors may be contacted directly, by phone or e-mail, to discuss

ideas and plans for proposals.

Judith Sunley, Director, Division of Mathematical Sciences, NSF

------------------------------

From: Lothar Reichel <reichel@mcs.kent.edu>

Date: Wed, 4 Mar 92 23:56:24 -0500

**Subject: Conference at Kent State on Numerical Linear Algebra**

KENT CONFERENCE

on

NUMERICAL LINEAR ALGEBRA

and

SCIENTIFIC COMPUTATION

at

Kent State University

at

Kent, Ohio

on

March 13-14, 1992

FRIDAY, MARCH 13, 9am - 5:30pm

M.H. Gutknecht, ETH

Changing the norm in conjugate gradient type algorithms.

Y. Saad, University of Minnesota

Multi-color parallel preconditioners.

W.J. Layton, University of Pittsburg

Massively parallel iterative methods for highly nonsymmetric

problems.

R.W. Freund, RIACS, NASA

Formally biorthogonal polynomials and a look-ahead Levinson

algorithm for general Toeplitz systems.

N.M. Nachtigal, RIACS, NASA

An implementation of the QMR method based on coupled two-term

recurrences.

G. Starke, University of Karlruhe

Alternating direction preconditioning for nonsymmetric systems

of linear equations.

D. Calvetti, Stevens Institute of Technology

An adaptive Chebyshev iterative method for nonsymmetric linear

systems based on modified moments.

V. Simoncini, University of Illinois at Urbana-Champaign

A memory-conserving hybrid method for solving linear systems

with multiple right hand sides.

M. Eiermann, University of Karlsruhe

Error estimates for semi-iterative methods.

SATURDAY, MARCH 14, 9am - 5:30pm

G.W. Stewart, University of Maryland

Updatable, rank-revealing matrix decomposition.

P. Van Dooren, University of Illinois at Urbana-Champaign:

Generalized QR and SVD decompositions. Algorithms and

applications.

V. Fernando, University of California at Berkeley

Accurate singular values and the qd algorithm.

W.B. Gragg, Naval Postgraduate School

Algorithms for updating QRP' factorizations.

J.D. Gardiner, Ohio State University

Stability of interval matrices.

R. Nabben, University of Bielefeld

Numerical methods for nonsymmetric block structured matrices

arising in the solution of Euler equations.

M. Hanke, University of Karlruhe

Iterative methods for discrete ill-posed problems.

X.-C. Cai, University of Kentucky

Domain decomposition preconditioned iterative methods for

nonsymmetric problems.

G.S. Ammar, Northern Illinois University

Updating and downdating Szego polynomials.

For questions about the conference, please, contact the organizers:

L. Reichel e-mail: reichel@mcs.kent.edu phone: (216) 672-2547

A. Ruttan e-mail: ruttan@mcs.kent.edu phone: (216) 672-2073

R.S. Varga e-mail: varga@mcs.kent.edu phone: (216) 672-2145

Institute for Computational Mathematics

Kent State University

Kent, OH 44242

------------------------------

From: stanly@crunch.unm.edu (Stanly Steinberg)

Date: Mon, 2 Mar 92 08:25:44 MST

**Subject: Intelligent Scientific Computation at AAAI Meeting**

I am helping organize a symposium,

Intelligent Scientific Computation,

at the AAAI Fall Symposium Series, October 23, 24, & 25, 1992, Cambridge,

Massachusetts. For more information contact:

Stan Steinberg, stanly@math.unm.edu

Elaine Kant, kant@slcs.slb.com

Richard Keller, keller@ptolemy.arc.nasa.gov

or

American Association for Artificial Intelligence

445 Burgess Drive, Menlo Park, CA 94025

(415) 328-3123

fss@aaai.org

Stan Steinberg

------------------------------

From: Venkat Sastry <SASTRY@rmcs.cranfield.ac.uk>

Date: Wed, 4 Mar 92 15:23 GMT

**Subject: One Day MATLAB Short Course in UK**

One day Course

MATLAB: a tutorial introduction

27 May 1992

Applied and Computational Mathematics Group

Royal Military College of Science.

The course is specifically designed for scientists, engineers and

lecturers who either use routine numerical/matrix calculations in

their specific application area or teach the subject at various

levels. No specialist knowledge of programming or computer science is

required, but participants are expected to have basic knowledge of

computing and to be educated to HNC or degree level.

The course provides hands-on experience of the MATLAB package, and

will be followed by an in-depth presentation of case studies which

deal with specialist topics from Signal Processing, Control System

Design, Numerical Computations, Splines and Optimization. Most of the

afternoon is devoted to tutorials and the participants will have ample

opportunity to work on a specialist topic of their choice.

The course lectures will be given by the teaching and research staff

of the Applied and Computational Mathematics Group under the direction

of Dr. Venkat V.S.S. Sastry.

For more information please write to Mrs. P. M. Moore, Applied and

Computational Mathematics Group, RMCS, Shrivenham, Swindon, Wilts.,

SN6 8LA or Tel: (0793) 785317 or e-mail: sastry@uk.ac.cran.rmcs.

------------------------------

From: Tom Manteuffel <tmanteuf@copper.Denver.Colorado.EDU>

Date: Tue, 3 Mar 92 14:53:10 -0700

**Subject: Copper Mountain Progran and Information**

Copper Mountain Conference

on

Iterative Methods

April 9-14, 1992

Preliminary Program:

The Conference Program is now available. There are 115 talks scheduled

in two parallel sessions. The talks begin Friday, April 10 at 8:00 am.

The Program will be mailed to speakers and those who register along with

registration materials. It is also available by email. Any message to the

address cm-info@teuf.denver.colorado.edu will result in the return of the

program along with general information and registration details. If you

should experience any difficulties or require a personalized response

please send a message to cmcim92@copper.denver.colorado.edu.

Lodging:

Lodging is always the most difficult aspect of the meeting because it is not

within our control. Lodging can be arranged by calling Copper Mountain Resort

at 800-458-8386 or 303-968-2882. At this time there are over 40 hotel rooms

and 20 one bedroom condos still available from the block of rooms reserved

for our conference. Please mention the name of the conference. Another source

of lodging at Copper Mountain is the Resort Association. They handle

the booking of a large assortment of private condos within the resort.

Their number is 303-968-6477 (ask for Marie at Extension 17).

Local Transportation:

Limousine service to Copper Mountain from Stapleton Airport is available

through Resort Express (800-334-7433). They offer a group rate

of $22 each way. You may make reservations directly with them. Please

mention the name of the conference.

Tutorial Program:

The Program for the Tutorial on Polynomial Iterative Methods to be held

April 7-8 in Denver is also available. It will attached to the information

accompanying the Conference Program.

------------------------------

From: Tony Chan <chan@math.ucla.edu>

Date: Wed, 4 Mar 92 18:16:02 -0800

**Subject: Householder Symposium XII**

THE HOUSEHOLDER SYMPOSIUM XII

MEETING ON NUMERICAL ALGEBRA

The Householder Symposium XII on numerical algebra will be held

during the week of June 13-18, 1993 at the UCLA Conference Center, Lake

Arrowhead, California, about a 2 hour drive from Los Angeles. This

meeting is the twelveth in a series, previously called the Gatlinburg

Symposia. It has been agreed to rename all subsequent Gatlinburg Symposia

to honor Alston S. Householder, one of the pioneers in Numerical Linear

Algebra and organizer of the first four Gatlinburg meetings. The meeting

is an international conference of experts in the field of Numerical

Algebra. The format of the meeting is a sequence of invited papers during

the day and special workshops organized by the participants in the

evening. There is no formal program, but traditionally a few topics are

emphasized.

The meeting is being organized by the Householder committee, in

cooperation with the SIAM Activity Group on Linear Algebra. The local

organizers are Tony F. Chan of UCLA and Gene Golub of Stanford University.

The venue is a first class rustic alpine resort with modern facilities.

The room charge will include room and board. It is hoped that we will

have the whole conference center to ourselves.

The traditional format of the Householder Symposia requires that the

attendence be limited. The organizing committee invites all qualified

persons to apply to attend. For planning purposes, please let us know

of your intention to attend as soon as possible, preferably before

December 1, 1992. We'll maintain a mailing list and keep you informed.

The actual application should consist of a vita and an extended abstract

(about two pages) of a paper you would present if invited to speak.

The latter will be used by the committee in planning the program.

Material should be sent to us before February 1, 1993 to:

Householder 93

c/o Babette Dalton

Department of Mathematics

University of California, Los Angeles

405 Hilgard Avenue

Los Angeles, California 90024-1555

e-mail: householder93@math.ucla.edu

------------------------------

From: Max Gunzburger <GUNZBURGER%VTCC1.bitnet@VTVM1.CC.VT.EDU>

Date: Thu, 5 Mar 92 14:55 EST

**Subject: Appalachian Numerical Analysis Day**

APPALACHIAN NUMERICAL ANALYSIS DAY

April 4, 1992 - 8:30-5:00

Virginia Tech - McBryde 126

An informal get-together of numerical analysts andcomputational

scientists in the Appalachian region will occur on the above date

and room. The day will be organized as follows. We will convene at

8:30 A.M. at which time all will introduce themselves. Talks will

take place from 9:00-12:00 and from 1:30-5:00. Anyone wishing to

give a talk will put their name into a hat; then, names will be

drawn to determine the order of the talks. The length of the talks

will be determined by the number of persons volunteering to talk.

The time from 12:00-1:30 will be devoted to lunch.

A block of rooms has been reserved (for participants in this event)

at the Best Western Red Lion Inn of Blacksburg (703-552-7770) for

the night of Friday, April 3; the rate is $45.00 per night per room

(up to four people). These rooms will be held until March 21; please

refer to "Virginia Tech Center for Applied Math" when making

reservations. This motel is within 5 minutes of the Virginia Tech

campus. Rooms are also reserved for the night of Saturday, April 4

for those who wish to drive back home on Sunday.

To get to the campus area from Interstate 81, get off at exit 118 and

follow US460 West to Blacksburg. Make sure you take both the

Christiansburg and Blacksburg bypasses (these are US 460) and not

the business routes. Get off US460 at the Prices Fork Road

interchange. The Red Lion Inn is in the direction of Prices Fork,

while McBryde is in the downtown direction. For those coming via

Interstate 77, get off at exit 9 and take US460 East to the Prices

Fork Road interchange; thereafter, follow the above directions.

A map (showing the locations of the Red Lion Inn, McBryde,

and parking areas) is available upon request.

We would appreciate it if recipients of this message wouldpublicize

this event to their colleagues, students, etc.

Further details may be obtained from:

Max Gunzburger

Interdisciplinary Center for Applied Mathematics

Virginia Tech

Blacksburg, VA 24061-0531

703-231-7667

Fax: 703-231-7079

e-mail: gunzburger@vtcc1.bitnet

or gunzburger@vtcc1.cc.vt.edu

------------------------------

From: Maria Morandi Cecchi <MCECCHI@PDMAT1.UNIPD.IT>

Date: Sat, 7 Mar 1992 10:44 N

**Subject: Conference in Italy on Innovative Methods in Numerical Analysis**

INTERNATIONAL CONFERENCE ON

INNOVATIVE METHODS IN NUMERICAL ANALYSIS

BRESSANONE - ITALIA

September 7-11, 1992

The Scientific Committee

C. Brezinski (Lille)

F. Brezzi (Pavia)

I. Galligani (Bologna)

G.H. Golub (Stanford)

M. Morandi Cecchi (Padova)

R. Temam (Paris)

Accommodation and conferences will take place at Bressanone, a summer

extension of Padova University, located into the Alps near Bolzano.

Bressanone can be reached by train through Verona and by car through the

Brennero highway. The closest international airports are Milano airports

of Malpensa and Linate (also Verona has a small airport).

The aim of this conference is to gather mathematicians of new numerical

techniques of several developments in the field of dynamical systems related

to non linear problems.

All the aspects of the subject are intended to be covered.

The proceedings will be published.

Some invited lectures are planned.

Participants are invited to arrive on Sunday September 6 and to leave on

Saturday September 12.

The registration fees (including the social events and a copy of the

proceedings) are 400,000 liras (180,000 liras for accompanying persons) and

they must be sent (in liras and free of charge for us) to

account # 31001/T

beneficiary: CISE (Centro Informatica Studi Editoriali)

Via Prati 19, Padova

bank: Banca Antoniana

Sede - Via VIII febbraio -- Padova, ITALY

Deadline for registration, sending the abstract (1 page) and a proof of

payment is

JUNE 10, 1992

An additional fee of 50,000 liras will be charged for late registration.

The complete papers should be brought at the congress.

If you intend to participate and give a talk, please write as soon as

possible to

Maria Morandi Cecchi

Dipartimento di Matematica Pura ed Applicata

Universita` degli Studi di Padova

Via Belzoni 7

35131 Padova -- Italy

e-mail: adtmora@ipduniv.bitnet

A second announcement, with details about the accommodation and the

proceedings, will be sent in July.

The Organizing Committee

E. Centenaro, S. De Marchi, L. Paccagnella, M. Morandi Cecchi,

M. Redivo Zaglia.

------------------------------

From: Faisal Saied <saied@cs.uiuc.edu>

Date: Fri, 6 Mar 92 15:09:56 -0600

**Subject: Report on the Tenth Parallel Circus**

The Tenth Parallel Circus was held at Oak Ridge on October 25-26.

This series of meetings continues a tradition that was started at Yale

in 1986. The Parallel Circus has evolved into a forum that allows

researchers in parallel computing to present and discuss new

developments in an ``interactive'' environment that is often not

possible at the larger conferences. An attractive feature of the

Parallel Circus is the deliberate attempt to include as many graduate

students in the list of speakers as possible.

The meeting was organized by Gene Golub, and the Math/CS group

at Oak Ridge. In particular, Michael Leuze and Esmond Ng worked

hard to ensure its success. Gene, along with Apostolos Gerasoulis

and Ahmed Sameh, arranged for NSF support for a number of graduate

students to attend, which was well appreciated. The Parallel Circus

well attended, with 32 speakers and over 100 participants. In fact

the large turnout necessitated a change of venue from the hotel to the

American Museum of Science and Energy at Oak Ridge.

The topics ranged over most areas of parallel numerical computing.

Talks on direct parallel methods included Susan Ostrouchov, who

described her contribution to LAPACK, the BLAS and the LACS, the

(basic) Linear Algebra Communications Subroutines. Susan presented

performance statistics for dense triangular solves on the Intel

Touchstone (Delta) machine. Mark Jones from Argonne spoke about using

parallel graph coloring heuristics to construct efficient parallel

preconditioners for large sparse matrices and described performance on

the Delta as well. This approach can avoid the severe problems that

point ILU preconditioners face on massively parallel computers. Tim

Davis from the University of Florida described joint work with Ian

Duff on unsymmetric pattern multifrontal methods for sparse Gaussian

elimination, that go beyond using the pattern of (A + A-transp). The

goal is to detect certain patterns in the sparse matrix and use dense

matrix kernels on the appropriate blocks. Lie Wang discussed the

spectral nested dissection ordering for sparse matrices, using results

from Alex Pothen and Horst Simon that use the second eigenvector to

construct separators for the underlying graph. For parallel

efficiency, separators that lead to higher fill-in may be acceptable

provided the elimination tree is balanced and not to deep. Sesh

Vinugopal from Rutgers addressed another issue that arises in parallel

sparse GE, namely how the computations in the factor phase can be

efficiently scheduled on a parallel computer. In particular, Sesh

described how a DAG of data dependencies between blocks could be

constructed, given the symbolic factorization of the sparse matrix. Mo

Mu from Rice also discussed scheduling for unstructured problems

arising from PDEs, and how ideas from domain decomposition could be

used for enhancing parallelism. Jianping Zhu talked about solving

least squares problems using regularization (A-trans * A + alpha * I)

x = A-trans * b on the iPSC/860, with a hybrid approach that switched

between Householder and Givens transformations. Gene used the

reference to Householder to remark on Householder's affiliation with

the Oak Ridge National Lab.

Yan-Ming (Mark) Chang from Yale spoke about block preconditioners for

very large sparse matrices, in particular on the use of block diagonal

preconditioners. Using an experimental extension of PCGPAK for

parallel computers, Mark discussed the performance of such

preconditioners with inexact solves, in conjunction with CG and GMRES

on various hypercubes. Other `iterative'' talks included two GMRES

related topics: Dan Hu described joint work with Lothar Reichel,

on using GMRES-like methods for the Sylvester equation, AX - XB = C.

Applying the Arnoldi process, results in a smaller Sylvester equation

that can be solved more economically. Gene Wachspress, who was among the

attendees, pointed out that the Sylvester equations arise in certain

control problems. Daniela Calvetti discussed a variant of GMRES that

replaces the usual monomial basis by a Newton basis using Leja points.

Yi-ling Chiang described how good convex hull estimates could be

obtained for the Chebyshev method by considering (A + \mu * I), and

constructing the Manteufel ellipses for the shifted matrix.

Jesse Barlow presented error bounds for Cuppen's divide and conquer

method for tridiagonal matrices that showed that this parallel

algorithm achieved accuracy comparable to the QR method for the

eigenvalues. Cal Ribbens from Virginia Tech gave an entertaining talk

on using Cuppen's method for the generalized eigenvalue problem. The

Schroedinger equation discretized by a HODIE method was cited as one

source of such problems. Results from the Sequent were given.

Deflation turns out to involve a trade-off: the total work can be

reduced, in return for sacrificing load balance. Mike Berry from the

University of Tennessee at Knoxville talked about joint work with Gene

Golub on sparse SVD computations for very large problems. These algorithms,

which use ideas like block Lanczos, subspace iteration and modified

moments, are implemented in SVDPack that Mike has written.

A number of talks dealt with parallel algorithms for partial

differential equations. Pat Worley spoke about parallelizing across

time, in contrast to the more common approach of parallelizing the

linear algebra problems that arise at each time step in traditional

schemes. Waveform relaxation is an important idea in this context, and

Pat discussed recent work that uses multigrid ideas within waveform

relaxation. Another interesting idea presented was the use of cyclic

reduction to solve bidiagonal systems that arise when certain

multistep methods are parallelized in time. Tad Janik described a

parallel p-version of the (mixed) finite element method on the Sequent

using tensor product basis functions. Tad commented that he had more

experience with the Finite Element Circus, than the Parallel Circus.

John van Rosendale talked about robust parallel multigrid methods for

CFD problems in 3D. He discussed a scheme that used a 3D Chebyshev

grid, with plane relaxation and semi-coarsening. Bernard Bialecki

described orthogonal spline collocation methods for separable elliptic

problems, which result in matrices that are amenable to domain

decomposition techniques. Kumar Mahinthakumar spoke about his work on

modeling groundwater contamination on the CM2, where the linear

systems were solved by PCG. Kumar observed that for their problem on the

CM2, diagonal scaling outperformed other preconditioners like ILU, and

attributed this to the very high performance of the matrix vector

products on the CM, relative to ILU. Abdul Khaliq, describing joint

work with Dave Voss, spoke about parallel parabolic methods that

use partial fractions. Partial fractions are an attractive way of

increasing parallelism and have been used by Gallopoulos and Saad,

and Sweet. Khaliq and Voss have extended this approach to handle problems

with discontinuities between the initial values and boundary

values, by constructing L-stable methods with simple partial fraction

expansions. Nikos Chrisochoides of Purdue talked on mapping pde computations

onto distributed memory machines. Nikos, who is contributing to Parallel

Ellpack, spoke of extending the idea of decompositions based on

geometry to decompositions based on phenomena and gave a glimpse of

the X-window style user interface being developed for Ellpack. Thaib

Taha has developed parallel solvers for some higher Korteweg-de Vries

equations, which have strong nonlinearities. The matrices arising in

this method are essentially four-diagonal, and the linear solves are

parallelized by a partitioning method that requires the solution of a

reduced system. Steve Cote described implementations of FFT-based

fast Poisson solvers on the 128 processor iPSC/860 at Oak Ridge. For

large enough problems (255^3), he was able to get over 200 Megaflops

in FORTRAN. Steve pointed out that these results need to be seen in

the context of the rather disappointing performance of the compiler

for the i860, which restricts the 128 node machine to a peak

performance of around 500 Mflops, for compiled code. Yue Zhuge, from

Stony Brook spoke on parallel methods based on Lax-Wendroff, for

hyperbolic problems with interfaces.

There were two talks that served to remind the audience that parallel

computing is not the exclusive domain of ``number crunchers''. Albert

Choy talked about parallel heuristic search methods, which have potential

applications in robot path planning, and Hongwei Du discussed parallel

algorithms for finding the center of a polytope.

In addition to talks related to parallel algorithms, several

speakers addressed the software problem. Michael Frumkin described a

library of systolic algorithms he has developed at Scripps for a Unix

environment. Apostolos Gerasoulis unveiled the PYRROS system that

handles task scheduling for message passing architectures, that grew

out of the older Schedule system. PYRROS pays particular attention to

the important issue of granularity and this capability was illustrated

by examples involving the Gauss-Jordan algorithm. Future plans include

applying PYRROS to sparse matrix computations. Adam Beguelin described

work on HeNCE, the Heterogeneous Network Computing Environment being

developed at Oak Ridge, that builds upon the well known PVM (Parallel

Virtual Machine) system. PVM is available through netlib (``send index

from pvm''), HeNCE will be made available in the near future. HeNCE

allows the user to specify parallelism in her/his program graphically.

Doreen Chang described her work evaluating interactive tools for

parallelizing codes on multiprocessor Crays, such as FORGE and fpp

from Cray. The codes used in the benchmarks were drawn from the

Livermore loops, NAS applications and the Perfect benchmarks. Terry

Pratt from Virginia raised the audience's hopes about the prospects

for parallelizing ``dusty decks'', but warned that these hopes should

not get too high.

Jack Dongarra gave a sneak preview of xnetlib at the beginning of the

conference and in a separate talk, described the present state of LAPACK,

the Linear Algebra package that combines and substantially enhances

LINPACK and EISPACK.

The Chinese banquet has become a tradition at the Parallel Circus,

and the Tenth was no exception. We look forward to the next Parallel

Circus to be held in Minneapolis, April 24--25.

------------------------------

From: J. C> Baltzer <neabbs!baltzer@relay.nluug.nl>

Date: Thu Mar 5 14:23:52 1992

**Subject: Contents: Numerical Algorithms**

Contents

Numerical Algorithms

Volume 2, no. 1

S. Dubuc and A. Malik

Convex hull of powers of a complex number, trinomial equations and the

Farey sequence

M.-J. Lai

Fortran subroutines for B-nets of box splines on three- and four-directional

meshes

C. Rabut

Elementary m-harmonic cardinal B-splines

C. Rabut: High level m-harmonic cardinal B-splines

A. Bultheel, P. Gonzalez-Vera, E. Hendriksen and O. Njastad

The computation of orthogonal rational functions and their interpolating

properties.

Please submit your papers to one of the Editors of NUMERICAL ALGORITHMS.

Volumes 2 and 3 to be published in 1992. Requests for sample copies and

orders are to be sent to J.C. Baltzer AG, Scientific Publishing,

fax: +41-61-692 42 62, e-mail: na.baltzer@na-net.ornl.gov

------------------------------

End of NA Digest

**************************

-------