**Today's Topics:**

- Preregistration Deadline for ICIAM 91
- Alexandre Chorin Elected to NAS
- Prof. Dr. Hansjoerg Wacker
- Advice on Scientific Workstations Sought
- Non-separable Elliptic PDE Solver
- Very Fast Poisson Solvers
- Vectorized Spherical Harmonics Routine
- Iterative Method for Sparse Least Squares Problems
- Computational Experiments for Numerical Analysis Instruction
- Fox Prize, 1991
- Parallel Numerical Analysis Workshop

From: SIAM Publications Department <SIAMPUBS@WILMA.WHARTON.UPENN.EDU>

Date: Fri, 17 May 91 14:31 EDT

ATTENTION ICIAM 91 SPEAKERS AND PARTICIPANTS

You can still save $50 on the conference registration fee if your

payment is received on or before June 14, 1991. If you've misplaced

your preliminary program, please contact SIAM. The conference and

hotel registration forms will be sent immediately upon your request.

Contact: iciam@wharton.upenn.edu

or phone: 215-382-9800

fax: 215-386-7999

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

From: Paul Concus <concus@csr.lbl.gov>

Date: Fri, 17 May 91 11:39:56 PDT

Alexandre Chorin has been elected to the U. S. National Academy of Sciences.

This honor is in recognition of his work in numerical methods, computational

fluid mechanics, and turbulence theory. Congratulations, Alexandre!

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

From: Heinz W. Engl <K310773%AEARN@pucc.PRINCETON.EDU>

Date: Tue, 14 May 91 15:25:02 SDT

On May 14, 1991, Prof. Dr. Hansjoerg Wacker of the Johannes-Kepler-

Universitaet Linz (Austria) has died.

Prof. Wacker got his PhD from the Technical University of Munich.

He held the Chair for Numerical Analysis here since 1974 and worked

mainly on Continuation Methods and Nonlinear Optimization.

In recent years, his main effort was to cooperate with industry on

various research projects, thus contributing a lot to the building up

of Industrial Mathematics in Austria.

Prof. Wacker has written several books and about 80 scientific papers.

He was on the Editorial Boards of Zeitschrift fuer Operations Research,

Computing, Mathematical Engineering in Industry, and Surveys on Mathematics

for Industry.

His death is a great loss for mathematics in Austria and for Industrial and

Applied Mathematics in general.

Heinz W. Engl, Linz

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

From: Fred Hickernell <FRED%BC750.BITNET@YALEVM.YCC.Yale.Edu>

Date: Wed, 15 May 91 17:15 +08:00

My institution is considering purchasing some workstations but have not yet

decided on the which brand (Sun, HP, IBM, DEC, etc.) or configuration.

Although we are well-endowed with microcomputers and have some old VAXes,

we do not yet have any workstations. Thus, I would be interested in

receiving any advice you might have.

We intend to use the workstations to support faculty and graduate student

research as well as upper level undergraduate teaching and project work.

The work we are doing (or want to do) includes numerical PDEs, numerical

linear algebra, scientific visualization, image processing and

computational statistics. We will use packages (e.g. MATLAB, SAS, NAG,

IMSL) as well as languages (e.g. FORTRAN, C). Virtually all computers on

campus are networked through ethernet. Our budget is modest, but we should

be able to buy something an order of magnitude better than a high-end

microcomputer (i486, Mac IIfx).

What is your idea of a starter system that could grow as budget allows?

Thanks,

Fred J. Hickernell fred@bc750.hkbc.hk (internet)

Department of Mathematics fred@bc750.bitnet (bitnet)

Hong Kong Baptist College (852) 339-7015 (office phone)

224 Waterloo Road (852) 338-8014 (fax)

Kowloon, HONG KONG (852) 698-6744 (home phone)

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

From: Michael Page <map@monu1.cc.monash.edu.au>

Date: Thu, 16 May 91 11:32:06 +1000

Before I reinvent the wheel, I am seeking a public-domain subroutine which

solves a non-separable two-dimensional elliptic p.d.e. on a non-uniform grid

in a rectangular domain with Dirichlet boundary conditions. My current

application has no cross-derivative terms but this is not crucial. I would

prefer a Fortran routine but (beggers can't be choosers) either C or Pascal

would be fine.

Previously I have used the BLKTRI routine (from FISHPACK), but now I need

a routine for a non-separable equation.

Judging from what I have heard, a multigrid method seems to be the most

efficient way to do this (but I am open to other suggestions). For example,

an article in the proceedings of the `Multigrid Methods' conference in 1981

(Lect Notes in Math, 960) mentions the release of a suitable code called

MGOO by Foerster and Witsch. If this code is freely distributable I wondered

if someone could either send me a copy or direct me towards an anonymous FTP

site. (I have looked in our local shadow of netlib.) Recommendations

(and locations) of other suitable codes would also be appreciated.

Thanks in advance,

Michael Page (na.page or map@monu1.cc.monash.edu.au)

Mathematics Department

Monash University

Melbourne, Australia

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

From: John McCalpin <mccalpin@perelandra.cms.udel.edu>

Date: Sat, 18 May 91 19:28:56 EDT

I am researching the finite-difference convergence of the time and

space integrated statistical properties of a set of equations that

approximate large-scale oceanic and atmospheric flows. I find myself

in need of a ridiculously fast solver for a Poisson-like equation that

arises at each time step. The equation is the straighforward 5-point

finite difference approximation to:

d^2 d^2

( --- + --- - g^2 ) h(x,y) = f(x,y)

dx^2 dy^2

subject to:

h(0,y) = h(Lx,y) = h(x,0) = h(x,Ly) = 0.

with a uniform Cartesian finite-difference grid of a few hundreds

of grid points on a side.

(1) Does anyone out there have an extremely fast direct (FACR?)

or multigrid solver that is applicable to this problem? (I am

currently using HWSCRT from FISHPAK).

(2) Are any "real" numerical analysts working on problems related

to the convergence of statistical measures of nonlinear hyberbolic

or nearly hyperbolic systems?

Thanks for any help....

John D. McCalpin mccalpin@perelandra.cms.udel.edu

Assistant Professor mccalpin@brahms.udel.edu

College of Marine Studies, U. Del. J.MCCALPIN/OMNET

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

From: Philip Sterne <sterne@dublin.llnl.gov>

Date: 16 May 91 18:55:22 GMT

Can anyone point me to a vectorized Fortran routine for calculating

spherical harmonics? (Preferably Condon-Shortley phase, but that doesn't

matter so much). I have a routine which computes the Ylm's using

recursion to get the associated Legendre polynomials, but the recursion

slows it down considerably on vector machines.

Thanks,

Phil

Philip Sterne sterne@dublin.llnl.gov

Lawrence Livermore National Laboratory Phone (415) 422-2510

Livermore, CA 94550 Fax (415) 422-7300

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

From: Bernard Danloy <danloy@anma.ucl.ac.be>

Date: Tue, 14 May 91 16:43:22 +0200

Several recent notes on NA-NET were concerned by sparse least squares problems;

the last one is the interesting summary of Tim Monks.

I am rather surprised that apparently nobody refers to a possible iterative

solution of the problem (at least, if the rectangular matrix has full rank).

I have myself current research under progress and would be very grateful to

learn more about recent advances. My personal work has been mainly theoretical

up to now but numerical experiments will be made during the next summer (some

test problems from the "real life" are very welcome) and a paper will be

published as soon as possible. Presently, the following result have been

obtained :

1. It is quite possible to apply a "classical" conjugate-gradient method

to the normal equations within an implicit approach (without computing

the product of the matrix by its transposed) : the process converges

mathematically in a finite number of steps.

2. The above approach is numerically unstable but the instability has its

origin in the implementation of the conjugate-gradient algorithm : the

classical approach does minimize the number of arithmetic operations but

is unstable if the residue does not converge to zero (this is obviously

the case in the least squares problems). A new implementation has been

developed which removes the numerical instability.

3. New research is made in the direction of an algorithm which minimizes

the error itself : in other words, if u is the unique exact solution

of Ax=b ( if A is rectangular, u denotes the least squares solution ),

we let e=x-u and we try to minimize (e|e) where ( | ) denotes the usual

scalar product. Let us remember that, if A is square, symmetric and

positive definite, the classical conjugate-gradient approach minimizes

(e|Ae). My research is concerned by a so-called "minimal error" or

"orthogonal directions" algorithm : some papers have been published

that claim that such an approach is unstable and does not converge ;

I did identify a basic error in the paper of Fletcher (Proc. Dundee

1975) and in other papers refering to it. A new correct implementation

is almost ready for testing. Remarks and suggestions about this are

welcome.

Bernard DANLOY

Institut de Mathematique

University of Louvain-la-Neuve

Belgium

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

From: Dave Hill <V5250E%TEMPLEVM@pucc.PRINCETON.EDU>

Date: Mon, 13 May 91 10:45:53 EST

Second Notice for Minisymposium

A minisymposium, "Computational Experiments for Numerical Analysis

Instruction", is being organized as part of the American Mathematical Society's

Eastern Section Meeting in Philadelphia. The meeting will be held at Temple

University Center City Campus on Saturday and Sunday, October 12 and 13, 1991.

The objective is to share instructional computational experiments that help

illuminate a topic in numerical analysis. The experiment can be a classroom

demonstration or a student laboratory assignment. The presentation could focus

on a particular technique, algorithm, or class of problems. Your discussion

should indicate the intended audience, the background in the course preceding

the experiment, tools (computer and/or software) required, options for

extensions, and relationships to other topics in your course.

Please interpret numerical analysis in a broad sense.

For consideration for participation in the minisymposium send a short

description of the experiment including the software and/or hardware required

to:

Dr. David R. Hill

Mathematics Department

Temple University

Philadelphia, Pa. 19122

Email: V5250E@TEMPLEVM.BITNET

Phone: 215-787-1654

Abstracts should be received by June 15, 1991. (No funds are available for

support of minisymposium participants. Overhead projectors will be available,

but computer projection capabilities are unknown at this time.)

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

From: Iain Duff <ISD%IBM-B.RUTHERFORD.AC.UK@pucc.PRINCETON.EDU>

Date: Tue, 14 May 91 09:41:26 BST

LESLIE FOX PRIZE 1991

One Day Meeting - Monday 24 June 1991

The University of Dundee, Scotland

The Adjudicating Committee (J C Mason, N K Nichols and C M Elliott) have

selected seven papers, from the twenty one entries that were submitted for the

1991 Leslie Fox Prize, to be presented by their authors at a meeting that will

be held in the Tower Building, University of Dundee on Monday June 24 starting

at 09.10. The submitted papers were once again of a very high standard and the

selection process was as difficult as ever.

You are cordially invited to attend this meeting at which you will hear some

of the best young numerical analysts talking about their research. The

selected authors, in the order of speaking (chosen randomly) are:

C J Budd (University of Bristol)

H Zha (Stanford University)

J Levesley (Coventry Polytechnic)

J F B M Kraaijevanger (University of Leiden)

B F Smith (Argonne National Laboratory)

J Xu (Pennsylvania State University)

P D Loach (University of Bristol)

A detailed programme is available from the Chairman of the Adjudicating

Committee, Professor J C Mason.

Through the hospitality of the Dundee group, no charges

will be made for tea or coffee, but attendees will be required to pay $6.00

for lunch - which will be arranged nearby in the University (bookings for

lunch will be needed by Monday June 17).

Many of those who plan to attend will also be staying on at the University of

Dundee for the Biennial Conference on Numerical Analysis (June 25 - 28). In

that case please inform the Conference organisers that you plan to attend

the Leslie Fox Prize Meeting first.

If you wish to attend the Leslie Fox Prize Meeting (and are not attending the

subsequent Conference) please contact Dr D F Griffiths, Department of

Mathematics and Computer Science, University of Dundee, Dundee, DD1 4HN,

(dfg@mcs.dund.ac.uk). Those only attending the Fox Prize meeting will

normally be expected to arrange their own accommodation, but some accommodation

may be available at the University by booking through Dr D F Griffiths.

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

From: Frank Plab <fp@castle.edinburgh.ac.uk>

Date: Tue, 14 May 91 10:04:33 BST

CALL FOR PARTICIPATION

One-Day Workshop

on

PARALLEL NUMERICAL ANALYSIS

sponsored by

Edinburgh Parallel Computing Centre

University of Edinburgh

21 June 1991

A one-day workshop on Parallel Computing in Numerical Analysis will be

held at the Edinburgh Parallel Computing Centre (EPCC) on Friday,

21 June 1991. The workshop should give numerical analysts and

other interested researchers an opportunity to exchange experiences in

parallelising numerical algorithms.

Invited presentations:

Prof Iain Duff (Rutherford Appleton Laboratory)

"Exploitation of Parallelism in the Solution of Sparse Systems"

Dr Peter Mayes (Numerical Algorithms Group)

"Parallelism and the NAG library"

If you interested to give a talk and/or to attend please contact

Frank Plab

Edinburgh Parallel Computing Centre

James Clerk Maxwell Building

University of Edinburgh

Mayfield Road

Edinburgh EH9 3JZ

Tel.: 031 - 650 5021

Fax: 031 - 662 4712

e-mail: F.Plab@ed.ac.uk

The Edinburgh Parallel Computing Centre (EPCC) is a multi-disciplinary

institution engaged in research and commercial development in parallel

computing. It is home to a wide range of parallel computing equipment,

including: a Computing Surface with more than 400 T800 transputers and

over 300 users; the UK Grand Challenge machine, which contains 64

i860/T800 hybrid nodes; and both a 64x64 and a 32x32 AMT DAP. EPCC has

a full-time staff of 26, and a large number of associates in computer

science, physical and mathematical sciences.

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

End of NA Digest

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

-------