NA Digest Sunday, July 9, 1989 Volume 89 : Issue 26
Today's Editor: Cleve Moler
Today's Topics:
-------------------------------------------------------
From: Michelle Jones <SIAM@wharton.upenn.edu>
Date: Mon, 26 Jun 89 14:42 EDT
Subject: SIAM Conference on Geometric Design
TO: NA NET
FROM: Michelle Jones, Marketing Manager, SIAM
SUBJECT: Announcement -- SIAM Conference on Geometric Design
DATE: November 6-10, 1989
TITLE: SIAM Conference on Geometric Design
ORGANIZER: Robert E. Barnhill
Arizona State University
PLACE: Sheraton Mission Palms Hotel
Tempe, Arizona
TOPICS: Teleological modeling, computer graphics, parametric curves
and surfaces in CAGD, images of matrices, domain processing
and manipulation, surface fitting and other related subjects.
INVITED SPEAKERS:
Alan Barr, California Institute of Technology, Teleological
modeling: A New Approach for Representing Objects
Philip J. Davis, Brown University, The Decline and Renaissanc
of Geometry
Rida Farouki, IBM, Numerical Stability of Geometric Algorithms
and Representations
David Gossard, Massachusetts Institute of Technology, Geometry
in Conceptual Design
Gerald Farin, Arizona State University, NURBS: Theoretical and
Practical Issues
John Gregory, Brunel University, Parametric Curves and Surfaces
in Computer-Aided Geometric Design
Cleve Moler, Ardent Computer Corporation, Images of Matrices --
Mathematical Visualization
John Rice, Purdue University, Using Domain Processing for
Solid Modeling
Tom Sederberg, Brigham Young University, Algorithms for
Computing Intersections of Parametric Surface
Peter Wilson, Rensselaer Design Research Center, Geometric
Aspects of PDEs
Mike Wozny, Rensselaer Polytechnic Institute, Visualization?
Or Merely Geometry and Computer Graphics.
CONTACT: SIAM Conference Coordinator
117 S. 17th Street, 14th Floor
Philadelphia, PA 19103-5052 USA
215-564-2929
(FAX) 215-564-4174
E-Mail: siam@wharton.upenn.edu
------------------------------
From: Jack Dongarra <dongarra@antares.mcs.anl.gov>
Date: Fri, 30 Jun 89 10:31:20 CDT
Subject: Argonne Summer Institute in Parallel Processing
SUMMER INSTITUTE IN PARALLEL COMPUTING
A Two-Week Institute at the Advanced Computing Research Facility
Mathematics and Computer Science Division
Argonne National Laboratory
September 5-15, 1989
Summer Institute Faculty Computer Facilities
Don Austin, DOE ALLIANT FX/8 (8 processors)
Mani Chandy, CALTECH AMT DAP (1024 processors)
Tom DeFanti, U. OF ILLINOIS, CHICAGO ARDENT Titan (4 processors)
David Gelernter, YALE UNIVERSITY BBN Butterfly GP1000 (96 processors)
John Gurd, UNIV. OF MANCHESTER, U.K. BBN Butterfly II (45 processors)
Ken Kennedy, RICE UNIVERSITY ENCORE MULTIMAX (20 processors)
Alex Nicholau, U.. OF CALIF.,IRVINE INTEL iPSC HYPERCUBE (32 processors)
Burton Smith, TERA COMPUTER INTEL iPSC HYPERCUBE (16 processors,
Guy Steele, THINKING MACHINES with vector capability)
ARGONNE STAFF SEQUENT BALANCE (24 processors)
STELLAR GS1000
THINKING MACHINES CM-2 (16,384 processors)
Eligibility and Selection Criteria:
Institute limited to 25 graduate students and postdoctoral researchers.
Preference given to those likely to advance parallel computing research.
Only one person from the same institution and department accepted.
Applications due July 15, 1989, supported by a letter of recommendation.
Note: Participants will receive free lodging for
September 5-15 and a stipend for meals and incidental
expenses. Travel costs will be reimbursed up to $750.
For further information, write or call:
Teri Huml
Mathematics and Computer Science Division
Argonne National Laboratory
Argonne, Illinois 60439-4844
312-972-7163
huml@mcs.anl.gov
The Institute is supported by the National Science
Foundation Science and Technology Center for Research
on Parallel Computing and by the U.S. Department of
Energy
------------------------------
From: Wlodek Proskurowski <proskuro%castor.usc.edu@usc.edu>
Date: Fri, 30 Jun 1989 12:25:37 PDT
Subject: NAG's Multigrid Routine
This spring I gave the following problem as a part of the take home quiz to
my graduate class in numerical PDEs:
The NAG software package contains a multigrid routine (D03EEF) for
solving general 2nd order elliptic PDEs on rectangular regions. It has two
options for approximating first derivatives by using a) central, or b) forward
differences. The following test example (from p.8 of the Mark 13 Release
NAG Manual) was run on the SUN computer in double precision:
-Du+100(u_x+u_y)=f in [0,1]^2 with
Dirichlet boundary conditions corresponding to the exact solution u=x^2+y^2.
Here D denotes the Laplacian, and u_x partial wrt x. The obtained results were:
# of levels # of iterations ||erorr||_2
a) b) a) b)
3 33 7 5e-11 4.0e-2
4 14 8 3e-12 1.6e-2
5 9 8 7e-13 3.0e-3
Explain the behavior of the rate of convergence and the error as a function of
N=2^(#of levels) in both cases. Why such extreme differences in the results
occur (note that the method is implemented correctly and there are no bugs
in the program)?
Comments and questions to NAG.
1. Only the result for 3 levels is given in the manual. Moreover, given there
are actually squares of the 2-norm (additionally, without normalization), so
the numbers read: a) 1.7e-19 and b) 1.28e-1.
2. Who cursorily looking at these rusults would want to use the routine: one
option is super accurate but extremely expensive, the other fast but gives
practically no useful information (compare with the norm of the solution!)
3. I hope you want NAG to be used not only by numerical analysts who have time
to invest (or students to do the work) to find out that the routine works
well only the test example is ill chosen, especially in the complete absence
of proper explanations.
------------------------------
From: John Conroy <conroy@super.org>
Date: 30 Jun 89 22:43:19 GMT
Subject: max x'Ax+b'x
I am interested in solving the following problem:
max x'Ax+b'x
||x||2=1
where A is a symmetric n by n matrix and || ||2 is the 2-norm.
When b=0, the solution is simply the eigenvector corresponding to the
maximal eigenvalue. If the 2-norm is replaced with linear constraints
and/or equalities, its a quadratic programming problem.
However, as stated above the best I know is to attack it as a
non-linear constrained optimization problem, which seems like overkill to
me. Any pointers or suggestions?
John Conroy
Supercomputing Research Center, Lanham, MD
------------------------------
From: Bernd Fischer <fischer@na-net.stanford.edu>
Date: Wed, 5 Jul 89 12:23:16 PDT
Subject: Colloquium on Applications of Mathematics in Hamburg
First Announcement
International
Colloquium on Applications of Mathematics
on July 6 and 7, 1990
in Hamburg
The Institute of Applied Mathematics of the University of Hamburg
will hold an international
Colloquium on Applications of Mathematics
on the occasion of the 80th birthday of Lothar COLLATZ.
Invitations for a main lecture have been accepted by: Ph. Ciarlet
(Paris), D. Gaier(Giessen) R. B. Guenther(Corvallis),
W. C. Rheinboldt(Pittsburgh).
Short lectures (ca. 15 minutes duration) connected with the topic
of the Colloquium are welcome from everybody interested, subject
to space and time restrictions.
Participants from East and Southeast Europe con possibly be given
some support for local expenses.
Those who may wish to participate in the above mentioned Colloquium
and want to receive further information are requested to send a note
as soon as possible, but not later than
December 15, 1989
to
University of Hamburg
Institute of Applied Mathematics
Bundesstrasse 55
D-2000 Hamburg 13
West Germany
------------------------------
From: Mark Kent <kent@na-net.stanford.edu>
Date: Thu, 6 Jul 89 09:22:26 PDT
Subject: Updates to NA-NET Mailing List
Updates to the NA-NET mailing list are as follows.
Changes:
randolph bank to rbank@ucsd.edu
petter bjorstad to petter@eik.ii.uib.no
theodorus dekker to dirk@fwi.uva.nl
eva edberg to evaedb%folke.se@majestix.ida.liu.se
sylvan elhay to elhay@cs.ua.oz.au
john gilbert to john@eik.ii.uib.no
ivan graham to igg@maths.bath.ac.uk
malvin kalos to kalos@tcgould.tn.cornell.edu
jerry kautsky to j.kautsky@research.cc.flinders.oz.au
richard liu to liu@mssun7.msi.cornell.edu
paul muir to muir@husky1.stmarys.ca
roy nicolaides to rn0m@andrew.cmu.edu
takashi nodera to nodera@math.keio.ac.jp
yoshio oyanagi to oyanagi@is.tsukuba.ac.jp
louise perkins to perkins@lll-crg.llnl.gov
david ryan to dmryan@cs.cornell.edu
alastair spence to as@maths.bath.ac.uk
grace wahba to wahba@stat.wisc.edu
pieter wesseling to piet%dutinfh@uunet.uu.net
hongyuan zha to prlb2!kulcs!kulesat!zha@uunet.uu.net
New entries:
roberto ansaloni acray2%icineca2.bitnet@forsythe.stanford.edu
jarle berntsen jarle@eik.ii.uib.no
margaret cheney cheneym@turing.cs.rpi.edu
shenaz choudhury sc7@vms.cis.pittsburgh.edu
brian coomes coomes%csfsa.cs.umn.edu@umn-cs.cs.umn.edu
tim davis davis@uicsrd.csrd.uiuc.edu
peter derijk actrijp@hutruu0.bitnet@xxx
julio dix jd01%swtexas.bitnet@forsythe.stanford.edu
david dobson dobson@rice.edu
henry ellingworth himen%ecs.oxford.ac.uk
raffael eperego perego%icnucevm.bitnet@forsythe.stanford.edu
stein eriksen stein@eik.ii.uib.no
jesper fabricius unijf%neuvm1.bitnet@forsythe.stanford.edu
anders forsgren andersf@math.kth.se
marco frontini marfro@ipmma1.polimi.it
albert gilg zeus@ztivax.siemens.com
linitial ginitial lg04c7%swtexas.bitnet@forsythe.stanford.edu
walter hoffman walter@fwi.uva.nl
doug james doug@mathel.ncsu.edu
tom kirke u15305%uicvm.bitnet@forsythe.stanford.edu
jian le jian@eeg.com
rob leland leland%na.oxford.ac.uk
alain leroux leroux%frbdx11.bitnet@forsythe.stanford.edu
fx litt litt%bliulg11.bitnet@forsythe.stanford.edu
christian lubich c80427%ainuni01.bitnet@forsythe.stanford.edu
herbert muthsam a8131daa%awiuni11.bitnet@forsythe.stanford.edu
marcus naraidoo ma_mn@cms.bristol.ac.uk
makoto natori natori%gama.is.tsukuba.junet@relay.cc.u-tokyo.ac.jp
lois petherick lmp%myrias.uucp@relay.cs.net
pia pfluger pia@fwi.uva.nl
shirley pomeranz pomeranz@tusun2.knet.utulsa.edu
ekkehard sachs tfb403%dkluni01.bitnet@forsythe.stanford.edu
antonia vecchio iam@areana.na.cnr.it
david watkins watkins%wsumath.bitnet@forsythe.stanford.edu
harry yserentant uma005%ddohrz11.bitnet@forsythe.stanford.edu
- Mark
------------------------------
From: Jack Dongarra <dongarra@antares.mcs.anl.gov>
Date: Thu, 6 Jul 89 17:25:21 CDT
Subject: New Position for Dongarra
After a rewarding 16 year association with Argonne National Laboratory
I have accepted a position at the University of Tennessee and Oak Ridge
National Laboratory as a professor in the Computer Science Department
and a member of the Mathematical Science group at Oak Ridge.
I will move from Illinois to Tennessee during the first week of September.
I look forward to the challenges and opportunities of this new position
as well as my continued involvement with such projects as LAPACK, netlib,
the "LINPACK Benchmark" and others from Tennessee.
Jack Dongarra
------------------------------
From: Hilde Devoghel <hilde@na-net.stanford.edu>
Date: Thu, 6 Jul 89 16:32:39 PDT
Subject: Fellowship at John von Neumann Supercomputer Center
Postdoctoral fellowship for 1989-1990 at John von Neumann National
Supercomputer Center, PO Box 3717, Princeton, NJ 08543, in NSF funded
project involving special functions of mathematical physics,
acceleration techniques for slowly convergent series, numerical
quadrature and multivariate interpolation, in a supercomputing environment.
Contact: Professor Michael P. Barnett at JvNC, bitnet address MBARNETT@JVNCD.
------------------------------
From: Philip Gill <SVEN%vax.num-alg-grp.co.uk@nsfnet-relay.ac.uk>
Date: Fri, 7 Jul 89 15:00 GMT
Subject: NAG Optimization with Discontinuous Derivatives
The last NA-Net distribution contained a message from
DeKnuydt and Smolders concerning the use of the Nag library to
minimize the function:
function_result = - SUM [N(i)/N] * blog [(N(i)/N)]
all i with
N(i) <> 0
where
N(i) = number of occurrences of value i
N = total number of occurrences.
The Nag routines E04JAF, E04JBF and E04VCF discussed by
DeKnuydt and Smolders are designed for smooth nonlinear
optimization---i.e., they can be expected to work only when the
first and second derivatives of the objective function exist and
are continuous. The problem described above does not fall into
this category. The NAG routine E04CCF, based on the Nelder and
Mead polytope method, is intended for problems whose derivatives
are discontinuous.
Philip Gill
Mathematics Department, UCSD.
------------------------------
From: Henry Greenside <romeo!hsg@cs.duke.edu>
Date: 7 Jul 89 18:46:25 GMT
Subject: Scaling of Condition Number with Resolution
Can people suggest any references or proofs about the
relationship between the condition number of a matrix
obtained by discretizing an elliptic pde and the order
of accuracy of the discretization?
It seems commonly known that second-order accurate
finite difference approximations to elliptic operators,
e.g., (d^2 f(x) / Dx^2), lead to matrices with
condition numbers that scale as nx^2, i.e.,
quadratically with the number of uniform mesh points.
Similarly, it seems well known that matrices arising
from Chebyshev spectral expansions of such operators
scale as nx^4, as a fourth power.
Have analyses been made about how general this
is, e.g., for more general operators:
d/dx( a(x) df(x)/dx )
or how boundary conditions affect this scaling?
If there is interest, I will summarize replies
to the net.
------------------------------
From: Petter Bjorstad <petter@eik.II.UIB.NO>
Date: Sat, 8 Jul 89 17:47:05 +0200
Subject: Professorship in Norway
Professor in Computer Science
The Department of Computer Science, University of Bergen
requests applications for
a tenured position as Full Professor of Computer Science,
(Scientific Computing / Optimization)
for immediate consideration.
The department has 12 full time faculty members (6 full professors
and 6 associate positions), 2 adjoint (part-time) professors,
14 research fellowships (Ph.D. students).
and 45 Master degree students.
The department gives courses for undergraduate as well as
graduate studies. There are two main directions of study
at the advanced level, {computer science} and {scientific computing,
(numerical analysis
and optimization)}.
This position will have a special responsibility within the
"Scientific Computing" direction of study.
All faculty members have state of art workstations (SUN-3 or newer), the
computing environment is based on an ethernet network directly
connected to the
international Internet (ARPA-Net). The department has created
a laboratory for parallel processing (Alliant FX/8 and Intel Hypercubes)
jointly with the CMI research institute, and also a laboratory for AI research.
The department moved into a new building (The High-Tech Center
of Bergen) in the spring of 1989. Several other computer science
related research groups, including IBMs Scientific Center are
located in the same building.
The Department conducts research in the following areas:
In computer science:
Analysis of Algorithms, Datacommunication and Coding Theory,
Artificial Intelligence, Programming Development (Languages,
specifications, verifications and environments).
In scientific computing:
Numerical Integration, Numerical solution of Partial Differential
Equations, Accelleration of Convergence, Discrete and
Continuous Optimization.
There is a strong focus on the use of parallel computers
in all areas of research.
The department has both national and international cooperations
with research groups at other institutions
(In particular in the United States and Europe).
Locally, we cooperate with the Christian Michelsen Research
Institute and with the IBM Nergen Scientific Center.
There are also other groups within the university doing
computer science or computer science related work.
(Computer Science in Social Sciences, Computer Lingvistics and
Computer Psychology)
Prospective applicants should be able to teach
in the department,
and must have an outstanding research
record in numerical analysis/optimization.
A documented interest in aspects of such research related to
parallel computer systems will be especially welcomed.
For more information on how to apply, please drop an E-mail
note to: petter@eik.ii.uib.no
or na.bjorstad@na-net.stanford.edu
OR write to: Institutt for Informatikk
Thormohlens gate 55
N-5008 BERGEN
NORWAY
------------------------------
From: Arnold Neumaier <neumaier@math.wisc.edu>
Date: Sat, 8 Jul 89 16:23:40 cdt
Subject: Address Change for Arnold Neumaier
Next Tuesday I'll return to Germany. My new address is
Prof. Dr. Arnold Neumaier
Inst. f. Angewandte Mathematik
Universitaet Freiburg
Hermann-Herder-Str. 10
D-7800 Freiburg
West Germany
My email address is
neum%sun1.ruf.uni-freiburg.dbp.de@relay.cs.net
------------------------------
End of NA Digest
**************************
-------