NA Digest Sunday, September 30, 1990 Volume 90 : Issue 34
Today's Editor: Cleve Moler
Today's Topics:
-------------------------------------------------------
From: The NA Net <na@na-net.stanford.edu>
Date: Sun Sep 30 20:08:03 PDT 1990
Subject: Updated NA Net Mailing List
Mark Kent has put a lot of effort into bringing the NA Net mailing
list up to date. It now has 1558 entries. The list is used both
for this NA Net News Digest and for the NA Net e-mail forwarding
service. The later is accessed by sending mail to
na.lastname@na-net.stanford.edu
If there is an unambiguous match to 'lastname' in the mailing list,
the mail will automatically be forwarded. If there is more than one
match, you will be sent a message with the possible messages,
together with the initials which must be prepended to distinguish
among them. If there is no match, you will be sent a message saying so.
Thanks, Mark, for updating the list.
-- Cleve
------------------------------
From: Nick Trefethen <nick@hydra.maths.unsw.OZ.AU>
Date: Fri, 21 Sep 90 22:31:47 +1000
Subject: Examples of "Pseudo-eigenvalues"
Dear colleagues:
I am in the process of writing a review of applications in numerical
analysis of the theory of what I like to call "pseudo-eigenvalues" of
non-normal matrices. If A is a matrix and eps is a positive
real number, then a complex number z is defined to be an
"eps-pseudo-eigenvalue of A " if it satisfies either of the
following equivalent conditions:
(i) z is an eigenvalue of A+E for some perturbation
matrix E with norm(E) <= eps ; or
(ii) The resolvent satisfies norm(inv(zI-A)) >= 1/eps .
Many topics seem to be illuminated by this notion, such as the
power-boundedness of families of matrices, the convergence of matrix
iterations, the behavior of certain adaptive algorithms, and the
stability of discretizations of p.d.e.'s.
Can any of you help me by pointing out references in this area
that I may have missed? For example, the definitions above were
used by Varah in 1979; have other authors used them too, perhaps earlier?
A plot of numerically computed pseudo-spectra (i.e., level curves
of the norm of the resolvent) appears in a 1987 paper of Demmel;
have any other such plots been published? More generally, I am
always on the lookout for interesting examples of matrices whose
eigenvalues are highly sensitive to perturbations, and have found
matrices of this kind in the literature due to Kahan, Ruhe, and Higham,
among others. Are there other good examples I may be unaware of?
I would be grateful for any comments or references on these topics.
Nick Trefethen
School of Mathematics, University of New South Wales
P.O. Box 1, Kensington, NSW 2033 Australia
nick@hydra.maths.unsw.oz.au
------------------------------
From: Youcef Saad <saad@riacs.edu>
Date: Sat, 29 Sep 90 14:18:05 PDT
Subject: Changes of Address for Youcef Saad
Colleagues:
From Oct. 1 to Oct 30, my address will be
Youcef Saad, National Tsing Hua University
Institute of Applied Math.,
Hsinchu, Taiwan 30043, Republic of China.
Starting Nov. 1st 1990, my permanent address will be:
Youcef Saad
Computer Science Dept.
University of Minnesota
4-192 EE/CSci Building
200 Union Street S.E.
Minneapolis, MN 55455-0159
Phone: (612) 624-7804 (Office)
(612) 625-4002 (CS dept)
(612) 624-9802 (Office at the
Minnesota Supercomp. Inst.)
e-mail: saad@cs.umn.edu, na.saad@na-net.stanford.edu
------------------------------
From: Martin Berzins <martin@dcs.leeds.ac.uk>
Date: Mon, 24 Sep 90 08:15:40 BST
Subject: Textbook Recommendation Sought
Introductory texts in mathematical software & scientific computation.
Starting in January 1991, I have to teach a 20 lecture + practicals
course to provide an introduction to numerical methods with a strong
slant towards mathematical software and scientific computation.
Does anyone know of a suitable text, preferably with examples in
Pascal? The best I've managed to do so far are two out-of-print books
- An Introduction to Numerical Computation in Pascal, Dew and James,
Macmillan and
-Numerical Methods A Software Approach, R.L. Johnston, Wiley.
Both are out of print now.
The book by Ortega and Poole on differential equations is very
good and has some
of the material but is too advanced for first-year students.
Any suggestions?
-- Martin Berzins (na.berzins)
------------------------------
From: Stephen Boyd <boyd@ISL.Stanford.EDU>
Date: Mon, 24 Sep 90 08:58:35 MDT
Subject: New Book on Linear Control Design
New book available
Linear Controller Design: Limits of Performance
by Stephen Boyd and Craig Barratt (Stanford EE Dept.)
is now available.
The main topic of the book is closed-loop design and the computation
of performance limits using convexity. The book introduces a standard
framework for the control design problem and describes many practical
design specifications in this framework. It is shown that many of these
specifications are closed-loop convex; the corresponding control design
problems can therefore be solved using the convex optimization algorithms
described in the latter part of the book.
Some special points of interest for NA people:
* The whole point of the book is to apply more numerical computing to
the problem of controller design than is commonly done. This is done
by casting many controller design problems as infinite-dimensional
nondifferentiable convex optimization problems.
* Two standalone chapters cover norms of signals and systems,
and their computation.
* Two standalone chapters cover convex analysis and nondifferentiable
convex optimization, emphasizing ellipsoid and cutting-plane methods.
Many numerical examples illustrate the typical behavior of these
algorithms.
* Over 27000 lines of matlab source were written to generate the numerical
examples and 216 plots and figures. To the authors' knowledge, the extent
of the numerical computation in this book is unprecedented in a book on
control systems.
* Title: Linear Controller Design: Limits of Performance
(428 pages, 216 figures, bibliography, index).
* Publisher: Prentice-Hall Inc.
* ISBN: 0-13-538687-X, Title Code: 538-686.
-- Stephen Boyd & Craig Barratt
------------------------------
From: SIAM Publications Department <SIAMPUBS@wharton.upenn.edu>
Date: Wed, 26 Sep 90 11:01 EDT
Subject: First Issue of SIAM Journal on Optimization
SIAM JOURNAL ON OPTIMIZATION
February 1991 Volume 1, Number 1
CONTENTS
William C. Davidon
Variable Metric Method for Minimization
Roger Fletcher
A New Variational Result for Quasi-Newton Formulae
Kurt M. Anstreicher
On the Performance of Karmarkar's Algorithm over a Sequence of Iterations
Guangye Li
Successive Element Correction Algorithms for Sparse Unconstrained Optimization
V. Jeyakumar
Composite Nonsmooth Programming with Gateaux Differentiability
John R. Engels and Hector J. Martinez
Local and Superlinear Convergence for Partially Known Quasi-Newton Methods
Jong-Shi Pang, Shih-Ping Han, and Narayan Rangaraj
Minimization of Locally Lipschitzian Functions
Y. Dai, G. van der Laan, A. J. J. Talman, and Y. Yamamoto
A Simplicial Algorithm for the Nonlinear Stationary Point Problem on an
Unbounded Polyhedron
Jiu Ding and Tien-Yien Li
A Polynomial-Time Predictor-Corrector Algorithm for a Class of Linear
Complementarity Problems
C. T. Kelley and Ekkehard W. Sachs
A New Proof of Superlinear Convergence for Broyden's Method in Hilbert Space
Jorge J. More and Gerardo Toraldo
On the Solution of Large Quadratic Programming Problems with Bound Constraints
O. L. Mangasarian
Convergence of Iterates of an Inexact Matrix Splitting Algorithm for the
Symmetric Monotone Linear Complementarity Problem
Virginia Torczon
On the Convergence of the Multidirectional Search Algorithm
EDITORIAL BOARD
Kurt M. Austreicher
Francisco O. Barahona
Dimitri P. Bertsekas
Frank H. Clarke
Andrew R. Conn
John E. Dennis
Donald Goldfarb
Clovis Gonzaga
C. T. Kelley
Masakazu Kojima
O. L. Mangasarian
Jose Mario Martinez
Jorge J. More
George Nemhauser
Jorge Nocedal
Michael L. Overton
R. T. Rockafellar
Roger W. H. Sargent
Michael A. Saunders
Robert B. Schnabel
Alexander Schrijver
Richard A. Tapia
Eva Tardos
PH. L. Toint
Layne T. Watson
Roger J.B. Wets
Margaret H. Wright
Yinyu Ye
Stavros A. Zenios
Francesco Zirilli
Jochem Zowe
For additional information, please contact
SIAM
3600 University City Science Center
Philadelphia, PA 19104-2688
telephone: 215-382-9800
fax: 215-386-7999
e-mail: siam@wharton.upenn.edu.
------------------------------
From: SIAM Publications Department <SIAMPUBS@wharton.upenn.edu>
Date: Fri, 28 Sep 90 14:30 EDT
Subject: Contents: SIAM J. Control and Optimization
Table of Contents
SIAM J. Control and Optimization
Vol. 29, No. 1, January 1991
J. S. Gibson and A. Adamian
Approximation Theory for Linear-Quadratic-Gaussian Control of
Flexible Structures
B. Charlet, J. Levine, and R. Marino
Sufficient Conditions for Dynamic State Feedback Linearization
Manlio Gaudioso and Maria Flavia Monaco
Quadratic Approximations in Convex Nondifferential Optimization
D. S. Gilliam, B. A. Mair, and C. F. Martin
An Inverse Convolution Method for Regular Parabolic Equations
P. Acquistapace, F. Flandoli, and B. Terreni
Initial Boundary Value Problems and Optimal Control for
Nonautonomous Parabolic Systems
Paul Tseng
Applications of a Splitting Algorithm to Decomposition in Convex
Programming and Variational Inequalities
R. H. Kwong
On the Linear Quadratic Gaussian Problem with Correlated Noise
and its Relation to Minimum Variance Control
U. Faigle and W. Kearn
Note on the Convergence of Simulated Annealing Algorithms
A. A. Stoorvogel
The Singular H Control of Problem with Dynamic Measurement Feedback
Henry Hermes
Asymptotically Stabilizing Feedback Controls and the Nonlinear
Regulator Problem
Vilmos Komornik
Rapid Boundary Stabilization of the Wave Equation
G. da Prato, A. J. Pritchard, and J. Zabcyzk
On Minimum Energy Problems
Michael Pedersen
Pseudodifferential Perturbations and Stabilization of Distributed
Parameter Systems: Dirichlet Control Problems
For additional information regarding the SIAM J. Control and Optimization,
please contact
Vickie Kearn, Publisher
SIAM
3600 University City Science Center
Philadelphia, PA 19104-2688
telephone: 215-382-9800
fax: 215-386-7999
e-mail: siampubs@wharton.upenn.edu
------------------------------
From: Joel Saltz <saltz-joel@CS.YALE.EDU>
Date: Mon, 24 Sep 90 00:48:51 EDT
Subject: ICASE Meeting on Scalable Multiprocessors
PRELIMINARY AGENDA
UNSTRUCTURED SCIENTIFIC COMPUTATION ON SCALABLE MULTIPROCESSORS
Ramada Inn, Kill Devil Hills, NC
October 29-31, 1990
Sponsored by ICASE
Monday, October 29, 1990
Manuel Salas, NASA Langley Research Center
An Assessment of Unstructured-Grid Methods for the
Analysis of Aircraft Fields
Harold Trease, Los Alamos National Laboratory
The Free-Lagrange Method on the Connection Machine
Steven Hammond, Robert Schreiber, NASA Ames Research Center
Mapping Unstructured Grid Problems to the Connection Machine
Steven G. Kratzer, Supercomputing Research Center
Unstructured Sparse QR Factorization on SIMD Computers
Kapil K. Mathur, Thinking Machines Corporation
Data Parallel Algorithms for the Finite Element Method
Joel Saltz, ICASE, NASA Langley Research Center
Runtime Compilation for Multiprocessors
Ting-Fook Ngai, Stephen Lundstrom and Michael Flynn,Stanford University
Run-Time Scheduling of Unstructured Scientific
Computation on Scalable Multiprocessors
Charles Koelbel, Rice University, Piyush Mehrotra, ICASE,
Compiler Support for Unstructured Scientific Computations
Steve Roy, Princeton University
Parallelization of Adaptive Fast Multipole Algorithms
Jack B. Dennis, Dataflow Computer Corporation
Compiling Irregular Codes for Data Parallel Execution
Tuesday, October 30, 1990
Roy Williams, California Institute of Technology
Parallel Load Balancing for Parallel Unstructured Meshes
Geoffrey Fox, Syracuse University
Parallel Software and Algorithmic Issues for Fast Nbody Solvers
John A. Board, Jr., James F. Leathrum, Jr., Duke University
Mapping the Fast Multipole Algorithm onto MIMD Systems
Shahid Bokhari, University of Engineering & Technology, Lahore
On-line Parallel Binary Dissection
Rainald Lohner, The George Washington University,
Jose Camberos, Marshal Merriam, NASA Ames Research Center
Parallel Unstructured Grid Generation
David Keyes, Yale University and ICASE,
Adaptive Refinement in Domain Decomposition Methods
Calvin J. Ribbens, Virginia Polytechnic Institute and State University
A Moving Mesh Scheme for Dynamically Adaptive Domain Decomposition
R. Henderson, G. Karniadakis, Princeton University
Hybrid Spectral Element-Finite Difference Techniques on
Multiprocessing Environments
Sukumar R. Chakravarthy, Sampath Palaniswamy
Rockwell International Science Center
Some Aspects of Single-Zone Structured-Grid CFD For Hypercube
MIMD Computer
Charbel Farhat, University of Colorado at Boulder
L. Fezoui, S. Lanteri, INRIA, France
Unstructured Compressible Flow Computations on the Connection Machine
Wednesday, October 31, 1990
David Nicol, The College of William and Mary
Algorithmic Issues in Mapping Workload
Kirk R. Pruhs, Taieb F. Znati, Rami G. Melhem, University of Pittsburg
Dynamic Mapping of Adaptive Finite Element Computations onto
Linear Arrays
Sharon L. Smith and Robert B. Schnabel, University of Colorado at Boulder
Centralized and Distributed Dynamic Scheduling for Adaptive,
Parallel Algorithms
William Celmaster, BBN Advanced Computers Inc.
Random-Access Bandwidth and Grid-Based Algorithms on Massively
Parallel Computers
Mo Mu, John Rice, Purdue University
Performance of PDE Sparse Solvers on Hypercubes
Edoardo S. Biagioni, Jan F. Prins, University of North Carolina
Scan Directed Load Balancing for Highly-Parallel Mesh-Connected
Computers
Please contact Emily Todd by October 8 1990 if you are interested
in attending.
E-Mail emily@icase.edu
US Mail
Emily Todd
MS 132C ICASE
NASA Langley Research Center
Hampton VA
23665
------------------------------
From: Ruediger Esser <ZDV003%DJUKFA11.BITNET@Forsythe.Stanford.EDU>
Date: Mon, 24 Sep 90 08:52:29 MES
Subject: ICS'91 Call for Papers
CALL FOR PAPERS
1991 ACM International Conference on
Supercomputing
June 17-21, Cologne, Germany
Conference Co-Chairmen
Edward S. Davidson, University of Michigan, Ann Arbor, USA
Friedel Hossfeld, Research Center Juelich (KFA), Germany
Program Director
Yoichi Muraoka, Waseda University, Tokyo, Japan
Program Committee
EUROPE AND AFRICA NORTH AND SOUTH AMERICA JAPAN AND FAR EAST
U. Trottenberg (Chair) E. Houstis (Chair) T. Yuba (Chair)
L.M. Delves D. Bailey T. Hoshino
I. Duff F. Darema Y. Kanada
W. Giloi D. DeGroot H. Kashiwagi
J.R. Gurd G. Fox N. Koike
G. Hoffmann E. Gallopoulos S. Nagashima
W. Jalby D. Gannon H. Tanaka
A. Lichnewski M. Heath Y. Tanakura
P. Mueller-Stoy J. McGraw H. Terada
T. Papatheodorou P. Messina K. Toda
P. Sguazzero C. Polychronopoulos
H. Wijshoff J. Rice
H. Zima
The fifth International Conference on Supercomputing is soliciting papers
on significant new research results in the development and use of super-
computer systems. Contributions should emphasize the novel aspects of the
work being reported and should discuss their implications for future
supercomputer development. Papers are welcome on topics including the
following areas:
ARCHITECTURAL DESIGN OF SUPERCOMPUTER SYSTEMS
SOFTWARE SYSTEMS SUPPORT FOR SUPERCOMPUTING
APPLICATIONS OF SUPERCOMPUTING
SUPERCOMPUTING ALGORITHMS AND PERFORMANCE ANALYSIS
Submissions
Authors are invited to send five copies of the full manuscript in English
to the program chairman of their region. Their addresses are:
EUROPE AND AFRICA NORTH AND SOUTH AMERICA JAPAN AND FAR EAST
Ulrich Trottenberg Elias Houstis Toshitugu Yuba
GMD/F1 Dept. Computer Science ETL 1-1-4 Umesono
Schloss Birlinghoven Purdue University Tukuba, Ibaraki
D-5205 Sankt Augustin W. Lafayette, IN 47907 305 Japan
Germany USA
The deadline for submissions is DECEMBER 1, 1990. Authors will be
notified of acceptance by February 20, 1991. Final, camera-ready versions
of accepted submissions will be due by March 20, 1991. The Conference
Proceedings will be published by ACM.
SPONSORS: ACM-SIGARCH in association with AICA, BCS-PPG, CSRD, CTI, CWI,
GI, GMD, INRIA, IPSJ, KFA, SBMAC, and SIAM-SIAGS.
For further details, please contact: Ruediger Esser
KFA-ZAM
D-5170 Juelich, Germany
Phone: +49-2461-61-6588
Fax: +49-2461-61-6656
E-mail: zdv003@djukfa11.bitnet
------------------------------
From: Steve McCormick <smccormi@copper.Denver.Colorado.EDU>
Date: Fri, 28 Sep 90 07:54:40 -0600
Subject: 5th Copper Mountain Conference on Multigrid Methods
*** Announcement and Call for Papers ***
5th COPPER MOUNTAIN CONFERENCE
on
MULTIGRID METHODS
Copper Mountain, Colorado
March 31-April 5, 1991
Organized by
University of Colorado at Denver
Front Range Scientific Computations, Inc.
Society for Industrial and Applied Mathematics
Sponsored by
U.S. Air Force Office of Scientific Research
National Science Foundation
IBM
Special Features
Multigrid Circus
Student Paper Competition
Code/Algorithm Bake-Off
Chairmen: Tom Manteuffel & Steve McCormick,
University of Colorado at Denver
Program Chairman: Jan Mandel, University of Colorado at Denver
Program Committee: Joel Dendy, Los Alamos National Laboratory
Craig Douglas, IBM
Guy Lonsdale, Gesellschaft f. Math. u. Datenverarbeitung
Seymour Parter, University of Wisconsin
Joseph Pasciak, Brookhaven National Laboratory
John Ruge, University of Colorado at Denver
Klaus Stueben, Gesellschaft f. Math. u. Datenverarbeitung
James Thomas, NASA Langley
Pieter Wesseling, Delft University
Workshop Chairman: Paul Frederickson, RIACS
This conference traditionally provides a forum for interaction
among basic researchers and practitioners. To promote this, morning
and evening seminar sessions will be held and afternoons will be open
for informal workshops, discussions, and special sessions. Anyone
interested in organizing such a session should contact one of the
chairmen.
The conference will be held at the Copper Mountain Resort, about
75 miles west of Denver. Deluxe condominium accommodations and
adjacent meeting rooms are in the Village Square Lodge at the base of
the Copper Mountain F lift.
PAPERS: Please send title, short abstract, mailing address, phone
number, and e-mail address (if available) BY DECEMBER 1, 1990 to:
Division of Extended Studies
University of Colorado at Denver
Campus Box 164
P.O.Box 173364
Denver, CO 80217-3364, USA
Complete manuscripts for inclusion in the preliminary conference
proceedings will be due MARCH 1, 1991. Selected papers will be
refereed, with authors' consent, for publication in a special
issue of the SIAM Journal on Numerical Analysis.
REGISTRATION: The conference fee ($195 for SIAM members and $215 for
non-members by March 1, $235 and $260 thereafter, $45 any time for
students) should be paid with registration directly to the above
address for the Division of Extended Studies. Please contact one of
the conference chairmen for necessary forms. Conference fee includes
preliminary proceedings, a reception, coffee breaks, and a banquet.
MULTIGRID CIRCUS: Inspired by the finite element circus, we are
including a circus for informal lectures and presentation of late-
breaking research results. Anyone who attends the conference is
welcome to lecture during these sessions, which are tenatively
planned for late afternoon on the first four days.
STUDENT PAPER COMPETITION: Travel assistance will be awarded to a few
students who submit papers that we judge to be the best research papers.
Deadline for submission of student papers is JANUARY 1, 1991. Please
submit your paper with a brief statement of your background (institution,
advisor, expected graduation date) to the Division of Extended Studies
at the above address. Clearly mark it "Student Paper Competition."
MULTIGRID CODE/ALGORITHM BAKE-OFF: Entrants must submit ONE test problem that
has not already been submitted (Poisson's equation on a cube is already taken,
sorry) with a code to generate the matrices, initial guesses, and right-hand
sides in a known language (no copyright notices, please). This will be
distributed to all involved. You will be responsible for timing your own
program on some machine(s). You must also be willing to time other
participants codes on your machine(s). You are not obligated to debug other
people's codes, however. Scalar, superscalar, vector, parallel, and
combination machines are welcome. Craig Douglas will oversee this mess, and
will edit a joint paper with the participants. For more information, and the
real rules, send e-mail to douglas-craig@cs.yale.edu or bells@yktvmv.bitnet.
Deadline for entering is JANUARY 1, 1991.
LODGING should be reserved directly with the Copper Mountain Resort
by calling 1-800-458-8386 in the USA outside of Colorado, 1-800-332-3828
in Colorado, and 303-968-2882 overseas. (FAX: 303-968-2711.) The rates are
$70 for a lodge room (one or two queen beds) and $142 for a two-bedroom
condominium. Limousine and bus service is available from Stapleton
International Airport in Denver. Detailed information is available
on registration forms.
SHORT COURSE: The conference will be preceeded by a week-long multigrid
short course in Denver, March 25-29, 1991. For more information, please
contact:
Chaoqun Liu
Computational Math Group, Campus Box 170
University of Colorado at Denver
1200 Larimer Street
Denver, CO 80204
(303) 556-4807
e-mail: cliu@copper.colorado.edu
FOR REGISTRATION FORMS/FURTHER INFORMATION:
Tom Manteuffel or Steve McCormick
Computational Math Group, Campus Box 170
University of Colorado at Denver
1200 Larimer Street
Denver, CO 80204
(303)556-4807
e-mail: tmanteuf@copper.colorado.edu
smccormi@copper.colorado.edu
------------------------------
From: Steve McCormick <smccormi@copper.Denver.Colorado.EDU>
Date: Fri, 28 Sep 90 07:55:48 -0600
Subject: Multigrid Short Course
*** MULTIGRID SHORT COURSE ***
*** Information & Schedule ***
University of Colorado at Denver
March 25-29, 1991
Principal Lecturer: Achi Brandt
Supporting Lecturers: John Adams, William Briggs, Chaoqun Liu,
Steve McCormick, John Ruge
Shortcourse Chairman: Chaoqun Liu
Purpose: To provide an understanding of the principles and
procedures for multilevel methods, especially for partial
differential equations, including new multilevel approaches
in computational fluid dynamics.
Registration Fee: $500 (regular), $150 (student). Includes books
and other course materials, refreshments, and
computer lab access.
Topics:
. Basic tutorial
. Advanced methods for PDE's (including
multigrid procedures for general systems, nonlinearity,
ellipticity/nonellipticity, time dependence, inverse
problems, indefiniteness, discontinuities, singularities,
performance prediction/analysis, constrained optimization)
. Adaptive techniques (MLAT, FAC)
. Algebraic methods (AMG)
. Computational fluid dynamics (including finite volume element
methods, high Reynolds number flow, steady and unsteady
Navier-Stokes equations, Steady and unsteady Euler equations,
flow transition, and two-phase porous flow)
. Introduction to non-PDE multilevel techniques (including
integral and integro-differential equations, fast
dense matrix multiplication, many-body interactions,
Direct solvers, large determinants, global discrete
highly-nonlinear optimization, and multilevel Monte Carlo
in statistical physics)
Emphasis will be placed on concepts and basic principles, illustrated by
concrete examples and results. Certain codes will be made available to
participants for laboratory experimentation during and after the short
course. Adjustments will be made to the schedule based upon the expressed
interests of the participants.
For Registration and Further Information, Please Contact:
Chaoqun Liu
Computational Mathematics Group
University of Colorado at Denver
1200 Larimer Street, Campus Box 170
Telephone: (303) 556-4807 or (303) 556-4886
e-mail: cliu@copper.colorado.edu
------------------------------
From: Cleve Moler <moler@mathworks.com>
Date: Sun Sep 30 20:55:07 PDT 1990
Subject: Is There a "Public Domain MATLAB?"
Is There a "Public Domain" MATLAB?
>From time to time, I see references to, or get requests for, the
"public domain" version of MATLAB. As the original author of
MATLAB, and one of the founders of The MathWorks, I would like
to explain how I regard "public domain" MATLAB.
There are two versions of MATLAB. I wrote the first, which we now
refer to as "classic" MATLAB, over the period from 1977 to 1984,
while I was on the faculty at the University of New Mexico. It is
an interactive matrix laboratory, written in Fortran, which uses
some of the subroutines from LINPACK and EISPACK. I distributed
a few hundred copies of the source code, usually charging a
small service charge, and including a letter requesting that the
code not be redistributed. I never used the term "public domain".
The second version, written in C by Steve Bangert and John Little,
is the basis for a family of products from The MathWorks, Inc.,
a company which Bangert, Little and I founded in 1985. These
products are called PC-MATLAB, Mac-MATLAB, Pro-MATLAB, etc.
I obviously recommend that anyone interested in using MATLAB
acquire the MathWorks version appropriate for his or her machine.
In addition to my commercial interest, I believe the MathWorks
versions are preferable scientifically, educationally, and, in the
long run, economically. The MathWorks versions:
* Are faster in execution,
* Have much better storage management,
* Include powerful graphics,
* Are extensible and programmable,
* Can be expanded with sophisticated "toolboxes",
* Are supported by scientific software professionals.
The only feature of classic MATLAB that is not present in modern
MATLAB is the "chop" function which allows the simulation of
shorter precision arithmetic. It is an interesting curiosity,
but it is no substitute for roundoff error analysis and it makes
execution very slow, even when it isn't used.
I know of several serious bugs in classic MATLAB, particularly
in logical and looping operations, but I don't intend to fix them.
In fact, there have been no fixes made to the code since about 1982.
I stopped distributing any copies myself 4 or 5 years ago.
The number of computers for which MathWorks MATLAB is not available
is declining as old machines are retired and new machine versions
are announced.
A few other commercial systems, for example SCT's CTRL-C,
are based on classic MATLAB. That's OK. CTRL-C, was done with
my permission and it helped establish MATLAB in control and
systems engineering. Now the company is a worthy competitor.
I realize that classic MATLAB is available on a few bulletin boards
and through some "freeware" services. In some cases, unauthorized
statements about public domain software are included. I have
regarded this as a mixed blessing. It certainly gives the MATLAB
approach to computing valuable exposure, but I am afraid that some
users of classic MATLAB do not realize how inferior it is to the
MathWorks products.
In summary, here is my position:
* There is no such thing as "public domain" MATLAB.
* I no longer distribute "classic" MATLAB.
* The code and documentation of classic MATLAB cannot
be used for commercial purposes without my permission.
* I would appreciate it if anybody making a copy of
classic MATLAB for personal or educational use
would include this statement with it.
-- Cleve Moler
moler@mathworks.com, or,
na.moler@na-net.stanford.edu
------------------------------
End of NA Digest
**************************
-------