NA Digest Sunday, April 2, 1989 Volume 89 : Issue 13

Today's Editor: Cleve Moler

Today's Topics:


From: Art Werschulz <>
Date: 30 Mar 89 15:00:45 GMT
Subject: Random Algorithms for Elliptic PDE

Does anybody out there have pointers to information on random
algorithms for *elliptic* partial differential equations? This would
include (but not necessarily be limited to) random walk algorithms,
Monte Carlo algorithms, etc.


Art Werschulz
Columbia University Computer Science Department

USEnet: ...!rutgers!columbia!cs!agw

ATTnet: Columbia University (212) 854-8642 854-2736
Fordham University (212) 841-5323 841-5396


From: Tom Coleman <>
Date: Fri, 31 Mar 89 13:59:17 -0500
Subject: Position at Cornell

Cornell University
Computer Science Department

Research Associate: Entry-level. Conduct research
in the Cornell Computational Optimization Project
(CCOP) with particular emphasis on large-scale
continuous (and piece-wise continuous) problems.
Ph.D. in computer science with specialization in
numerical optimization. Post-doctoral experience
(minimum of six months). Expert knowledge of methods
for piece-wise continuous minimization, with a solid
research tract record in this area. Working knowledge
of MATLAB, Fortran and some parallel computing experience.
Proven ability to analyze and establish mathematical
convergence properties of minimization algorithms.
Salary: $30,000 - $35,000 annually. Send curriculum vitae with
three references to Thomas Coleman, Department of
Computer Science, Cornell University, Ithaca, NY
14853-7501. Cornell University is an equal opportunity
employer and welcomes applications from women and
ethnic minorities.


From: Ed Plum <>
Date: Sat, Apr 1 1989 20:09:64
Subject: Perfect Universal Numerical Kernel

News Release
April 1, 1989
Palo Alto, California

Mathematicians at Palo Alto's Universal Numerical Techniques
Studio (NUTS) today announced an unprecedented breakthrough in
mathematical software. Their Perfect Universal Numerical Kernel
(PUNK) is expected to replace all previously developed mathematical
software. The new algorithm solves all possible numerical problems
and achieves all the goals of quality mathematical software:

* Perfectly Portable -- runs on all computers.

* Perfectly Stable -- introduces no roundoff error.

* Perfectly Parallel -- linear speedup for any number of processors.

* Perfectly Patentable -- of course.

The foundation of the technique is a data base of important, useful
and frequently occurring numbers. The PUNK Preprocessor expands this
into a list of all the floating point numbers for any particular computer.
The user of the system specifies a problem by providing a function
F(X) which evaluates the problem specific residual. PUNK then
generates a sequence of approximate solutions, X, until one is found
for which F(X) = 0. Separate processors on a parallel computer are
assigned to individual components of the solution vector, and so
perfect parallel efficiency is obtained on large problems.
Since all possible floating point numbers are tried until an exact
solution is found, no additional rounding errors are introduced
by PUNK.

The system is now in beta test, and will be available by the
end of the quarter.


End of NA Digest