NA Digest Monday, December 16, 1996 Volume 96 : Issue 47

Today's Editor:

Cleve Moler

The MathWorks, Inc.

moler@mathworks.com

Today's Topics:
• NA Digest Top Ten List
• Award for Kowalski, Sikorski, and Stenger
• Change of Address for Andy Cleary
• Query on Anti-symmetric, Block Toeplitz Matrices
• Query on Approximation by Continued Fractions
• Query on Minimization Algorithms
• A Problem Involving Positive Matrices
• UK and Republic of Ireland Section of SIAM
• New Book: Inverse Problems in Geopysical Applications
• New Book on Scientific Computing
• New Book: Mathematical Programs with Equilibrium Constraints
• New Book: Theory and Algorithms for Linear Optimization
• Brenner-Scott Book Back in Print
• Multidisciplinary Optimization Test Suite
• 1997 Dundee Conference Update
• Deadline for Conference in Bulgaria
• Conference on Matrix Analytic-Methods in Stochastic Models
• Conference on Numerical Modelling in Continuum Mechanics
• ENUMATH 97 Conference Announcement
• Workshop on Parallel Solution of PDEs
• Conference on Practice and Theory of Automated Timetabling
• London Mathematical Society Meeting on Numerical Analysis of PDEs
• Techniques Nouvelles de Traitement des Matrices Creuses
• Randomization and Approximation Techniques in Computer Science
• The First International ISAAC'97 Congress
• Dean Position at George Mason University
• Faculty Position at Cornell University
• Postdoctoral Position at Lawrence Berkeley Laboratory
• Special Issue of Linear Algebra and its Applications
• Contents, Math. of Control, Signals, and System
• Contents, SIAM on Scientific Computing

Submissions for NA Digest:

Information about NA-NET:

URL for the World Wide Web: http://www.netlib.org/na-net/na_home.html -------------------------------------------------------

From: Paul McMahan <mcmahan@cs.utk.edu>
Date: Fri, 13 Dec 1996 12:56:23 -0500
Subject: NA Digest Top Ten List

NA-Digest Update

The NA-Digest has been in business for the last ten years, providing
news, announcements, a forum for questions, and job postings to
the community. There are now over 6,000 members on the NA-Digest
distribution list in countries all over the world. Check
http://www.netlib.org/na-digest-html/breakdown.html to see the
breakdown of where those members are located.
The top ten list is:
edu US Educational 1897
com US Commercial 601
de Germany 424
uk United Kingdom 398
gov US Government 298
fr France 272
ca Canada 230
it Italy 179
au Australia 145
nl Netherlands 136

The NA-Digest Archives are maintained at Netlib
(http://www.netlib.org/na-digest-html/). Of the 15,000+ hits
per day that Netlib receives, almost 300 of these are to the
NA-Digest archives. We recently added an html browsing
interface as well as a search engine to the NA-Digest Archives
to help in finding information.
Best wishes,
Jack Dongarra and Paul McMahan


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

From: Frank Stenger <stenger@sinc.cs.utah.edu>
Date: Mon, 16 Dec 1996 13:46:08 -0700
Subject: Award for Kowalski, Sikorski, and Stenger

We wish to announce that M. Kowalski, K. Sikorski and F. Stenger were
just (in the fall of 1996) awarded FIRST PRIZE, by the SECRETARY OF
NATIONAL EDUCATION in POLAND, for the research leading to the
publication of the monograph: SELECTED TOPICS IN APPROXIMATION AND
COMPUTATION, which was published by Oxford University Press, in 1995.
This prize represents the most prestigious science research award in
Poland; it is awarded annually to only selected group of researchers.

M. Kowalski, K. Sikorski, and F. Stenger


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

From: Andrew Cleary <cleary@jason.llnl.gov>
Date: Tue, 10 Dec 1996 09:05:11 -0800 (PST)
Subject: Change of Address for Andy Cleary

Please note my new address:

Andy Cleary
Center for Applied Scientific Computing
Lawrence Livermore National Labs
P.O. Box 808, L-561
Livermore, CA 94551

phone: (510) 422-1939
fax: (510) 423-2993

email: cleary@jason.llnl.gov


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

From: Lee-Fen Ko <lfko@comp.uark.edu>
Date: Tue, 10 Dec 1996 15:33:47 -0600 (CST)
Subject: Query on Anti-symmetric, Block Toeplitz Matrices

There does not seem to be program available for computing the
determinant specifically for a real anti-symmetric matrix. So the
choices seem to be either LU decomposition or treat the matrix as
a Hermitian matrix and use the corresponding EISPACK package. Does
anyone know which route is more efficient for very large matrices?

In addition, the matrices are also block Toeplitz. Is there any
efficient way to compute their determinants?

Any information will be much appreciated.

Lee Ko


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

From: Vijay Sonnad <sonnad@VNET.IBM.COM>
Date: Mon, 9 Dec 96 13:52:38 CST
Subject: Query on Approximation by Continued Fractions

I would like to request assistance with a problem in approximation
of a function with continued fractions.

The function that I am trying to approximate is of the form

1/ sqrt( a0 + a1 * x + a2 * x**2)

Here a0, a1 and a2 are constants, and x is a variable; I need to
approximate the above function between x=0 and x=1 with a continued
fraction.

One difficulty is that a0 could be of the order 1.0e-6 (a1 and a2
are of order 1), and hence the function would vary rapidly around
x=0, and that is where the maximum accuracy is needed; it is fine if
if the approximation is not good around x=1 .

Please let me know if there is a reference (text, paper, whatever),
that would show me how to approximate the above function with a
continued fraction. Also, is there public domain software that would
do this for me?

Thank you .
Vijay Sonnad


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

From: Lisa Enz Irvine <lirvine@bme.jhu.edu>
Date: Wed, 11 Dec 1996 09:51:52 -0500
Subject: Query on Minimization Algorithms

Hi!

I am trying to determine 15 parameters of a model by minimizing the
least squared error between model responses and several types of experimental
data. I need a parallelizable algorithm that will minimize a function of at
least 15 variables subject to upper and lower bounds on the variables and that
does NOT require derivatives. I have tried conjugate gradient and quasi-Newton
methods with finite-difference gradients, but without success. I am currently
using a direct search complex algorithm which is converging to a solution but
much too slowly. Any suggestions would be greatly appreciated.

Thanks,
Lisa Irvine
lirvine@bme.jhu.edu


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

From: Avi Vardi <vardi@mipg.upenn.edu>
Date: Fri, 13 Dec 96 12:02:44 EST
Subject: A Problem Involving Positive Matrices

Can somebody help me with this problem which has intrigued me for some time:

Given A, an n by n matrix, a > 0 for all i >< j
i,j

( >< stands for 'not equal to'
>= stands for 'greater or equal to'
<= stands for 'less than or equal to' )

define B(x), an n by n matrix and c(x), an n-dim column vector,
both depend on an n-dim column vector x, as follows:

                 a      if x  <= x  +1              a      if x  >= x  -1
b     = b    =    i,j       i     j                  j,i       i     j
 i,j     j,i                                +   
                  0        otherwise                 0        otherwise


                 ---                    
b      =   -     \      b           
 i,i             /       i,j      
                 ---                    
                 j><i               
                                        


              ---                         ---
              \      a                    \     a
c   =   -     /       i,j         +       /      j,i   
 i            ---                         ---
          j><i:  x  <= x  +1          j><i:  x  >= x  -1
                  i     j                     i     j
  

for all i,j.

                    --- 
Among x satisfying  \  x   = 0, there is a unique solution to  B(x) x = c(x).
                    /   i
                    ---