NA Digest Sunday, June 25, 1989 Volume 89 : Issue 25

Today's Editor: Cleve Moler

Today's Topics:


From: Hans Munthe-Kaas <>
Date: 19 Jun 89 10:19 +0200
Subject: Professorship in Trondheim

The Norwegian Institute of Technology
Division of Mathematical Sciences

Full Professor of Mathematics (Numerical Analysis)

The Norwegian Institute of Technology invites applications for a tenured
position as Full Professor of Mathematics (Numerical Analysis). The Division of
Mathematical Sciences is part of the Department of Physics and Mathematics,
and currently has 23 faculty positions, 11 of which are full professorships.
The Division offers a program in Industrial Mathematics leading to the
engineering degree of (at M.S. level), and has a doctoral program
leading to the degree. The program in Industrial Mathematics consists
of courses in mathematics, numerical analysis and statistics, with a general
emphasis on mathematical modelling. Research interests of the current faculty
in numerical analysis include numerical quadrature, spline approximations,
numerical solution of ordinary differential equations and numerical linear

Special emphasis will be placed on qualifications of candidates in areas of
numerical analysis with relevance to numerical modelling using vector or
prallel computing algorithms; in particular qualifications within numerical
linear algebra, numerical solution of differential equations, and numerical

Applicants are sought who show significant research accomplishments as well as
serious concern for teaching, and commitment to initiating and promoting

The closing date for receipts of applications is September 1, 1989.

The Professor is engaged on the condition that he/she at
any time participates in teaching and examination work in accordance with the
relevant program of study, and furthermore he/she must accept, without
compensation, any revisions made by statutory law or royal decree to curricula,
pension arrangements and retirement age.

A letter of application including a curriculum vitae and a list of
publications should be addressed to The King and sent to:

The University of Trondheim
The Norwegian Institute of Technology
Personell Section
N-7034 Trondheim

In addition, reprints and preprints should be sent, in quadruplicate, to the
same address, no later than one month after the closing date. Work in progress,
referred to in the letter of application, must arrive within three months
after the closing date.

Further information is available from the above address.


From: Rich Sincovec <>
Date: Wed, 21 Jun 89 17:48:10 -0700
Subject: Freund Wins Award

Roland Freund was recently informed that he is the recipient of the
"Heinz- Meier-Leibniz Award". It is awarded by the German Secretary
of Education and Science. It is a one time award related to the
anniversary of some important event in science (possibly Leibniz's
birthday). It is awarded in three different fields: Public Law,
Applied Math, and some engineering discipline. The award honors
important contributions of junior scientists not older than 33 years.
Roland will receive the Applied Math Award which also includes a
monetary award. Roland did not know that he was nominated for this
award until he received notice that he was the recipient of the award.
A committee selected Roland based on the quality and the importance of
his research papers.

Roland will receive the award this week in a ceremony held in an old
castle which is part of the University in Muenster.

Congratulations, Roland!


From: AlBert DeKnuydt <prlb2!kulcs!kulesat!deknuydt@uunet.UU.NET>
Date: Thu, 22 Jun 89 18:11:27 GMT
Subject: Mininization Problem

This MAIL contains a description of a problem on numerical
minimisation and numerical differentiation. The domain of
the problem is image processing.

1) Function description :

The rather complicated function to minimize depends on 5
variables. These 5 variables define an image of which a
histogram is calculated. The function result is the following
operation on this histogram. (In fact the entropy)

N(i) = number of occurrences of value i
N = total number of occurrences

function_result = - SUM [N(i)/N] * blog [(N(i)/N)]
all i with
N(i) <> 0

One single function evaluation takes about 3 minutes CPU on a
VAX 8530.

2) What we tried up to now :

We tried the following NAG routines.

E04JAF : Quasi Newton algorithm using function values only,
easy-to-use version.

Problem : accurracy of result nor function
evaluation not controllable. Routine doesn't use
large enough steps towards the minimum.

E04JBF : Quasi Newton algorithm using function values only,
comprehensive (= not "easy-to-use" ?) version.

Problem : steps too small. Routine E04HBF, used
to determine initial step length for making
difference approximations to the partial
derivatives of the target function, expects target
function to be of machine accuracy. Because our
function is essentially discrete, this is

E04VCF : Sequential QP (Quadratic Programming) method,
using first derivatives.

Problem is now with the routine to calculate these
derivatives. It almost always returns with a
diagnostic complaining about accuracy (forward and
central difference estimates don't agree to half a
decimal place). When we check these derivatives
with E04ZCF, it says it doubts about the
correctness of the derivatives. Probably, the
cause is the slightly discrete character of the
target function.

(We use MARK 12 release of the NAG library, if this is of any


Anybody has an idea :

What the reason of the failure of the NAG routines
might be ? And how to solve this ?

Of another way to minimize this kind of function ?

eMAIL deknuydt@kulesat.uucp UUCP
deknuydt%kulesat.uucp@blekul60 BITNET

B. DeKnuydt & J. Smolders
K.U.Leuven ESAT/MI2
Kardinaal Mercierlaan 94
B-3030 Heverlee-Leuven


Date: Fri, 23 Jun 89 16:02:31 EMT
Subject: Positions at Bergen Scientific Centre


Reservoir Modelling-Numerical Algorithms-Software Development

Bergen Scientific Centre, IBM, is building up its research activity based on
numerical oil reservoir modelling. This work involves the development of
adaptive multigrid solvers for Navier Stokes flows and the development of
multigrid solvers for reservoir simulation. Current areas of interest
include : automatic error estimation and grid refinement, parallelization on
shared/distributed memory architectures and the treatment of complex
geometries. The orientation of this work is with a view to using
parallel/vector computer architectures.

We are looking for scientist(s) interested in either of the following



as related to oil reservoir simulation.

Appointments in these areas will be for a period of one or two years.

Bergen provides a pleasant environment in the heart of the Western
Fjords of Norway with excellent opportunities for outdoor pursuits
especially skiing, and water-sports. The centre is English speaking
and provides a friendly and flexible environment for scientists of
many different backgrounds.

Contact Chris Thompson at either of the electronic mail addresses:


Aladin Kamel:

Patrick Gaffney:


From: John R. Rice <>
Date: Fri, 23 Jun 89 12:31:24 EST
Subject: Vector/Parallel Polynomial Arithmetic

Vector/Parallel Polynomial Arithmetic

Does anyone know of software or algorithm implementations of
polynomial arithmetic for vector or parallel machines? We are
interested in

Univariate & Sparse Multivariate Polynomials

and the operations:

Multiplication Division
Evaluation at multiple points Interpolation
Residue computation

Reply to John R. Rice jrr& 317-494-6003
C. Bajaj 317-494-6531


From: Jon Claerbout <agate!shelby!portia!hanauma!jon@ucbvax.Berkeley.EDU>
Date: 25 Jun 89 02:01:53 GMT
Subject: Scientific programming in C++

Scientific programming in C++ ?

I converted some simple scientific Fortran/Ratfor programs into C++
to see if they would look suitable for a textbook
such as my last book "Imaging the Earth's Interior".
I conclude C++ is about as good as fortran-ratfor.
Unfortunately, mixing Fortran with C++
ranges from undocumented to impossible.

The new Lippman C++ book looks like a replacement for the Stroustrup book
since it fully describes the new AT&T version 2.0.
Pedagogically it is a big improvement too.
Since both Stroustrup and Lippman describe both C and C++
I infer they mean eventually to replace C by C++
(else why the 100+ extra pages to explain C
which K&R already do beautifully)?
A section called "linkage to other languages" mentions C but not Fortran.

We couldn't link gnu C++ mains to fortran subroutines.
I converted some simple fortran scientific programs to a C++ style
designed to please fortran users
and I posted them to [the UNIX network newsgroup] comp.lang.fortran.

Jon Claerbout
Dept. of Geophysics
Stanford University


From: Cleve Moler <>
Date: Sun Jun 25 12:43:27 PDT 1989
Subject: No NA News Digest Next Week

I'll be out of town next weekend, and things are pretty slow
anyway, so it will be two weeks until the next NA News Digest.
Happy Fourth of July.


End of NA Digest