**Today's Topics:**

- Software for Mixed Integer Programming Problem
- Fluid Dynamics Conference at Oxford
- MATLAB User's Group
- Generating Eigenvalues in a Particular Order
- Out of Core Solvers
- Multi-dimensional Quadrature Software
- Replacement of EISPACK RSP
- Position at Argonne
- Positions at Boeing Computer Services
- Position at NC State

From: Steve Horne <horne@alcvax.pfc.mit.edu>

Date: Mon, 25 Sep 89 09:20 EDT

I have the following problem in Mixed Integer Linear Programming.

S_0 is a point (vector) in Euclidean n-space En

A is an n x m matrix containing m vectors in n-space, m>n.

d is a vector of m distances from S_0 to each column in A

x is an unknown vector of m weights, n+1 of which are gt 0, the rest

zero

y is an unknown vector whose components contain (n+1) ones

and (m-n-1) zeros.

e is a vector with m components, all ones.

The Problem:

minimise d.y

with constraints

A.x = S_0

e.x = 1

x le y

y is (0,1)

x ge 0

In words, select from m points the n+1 points whose convex hull contains

S_0, and the sum of whose distances from S_0 is a minimum.

Typical dimensions -- S has 5-7 components; m is about 30.

I need a subroutine which solves the above, preferably in Fortran or C.

Ideally, the routine would not be a general mixed integer routine, but would

take advantage of some of the special structure. However, I'll take anything I

can get. The application is approximation of a function whose value is

known in scattered points in n-space. The subroutine(s) would become part of a

package to be used in programming the feedback control computer for Alcator

C-Mod, a tokamak under construction at MIT. Any contribution used in the

package will be carefully and gratefully acknowledged.

Thanks for your help

Steve Horne

(617) 253-8663

Horne@alcvax.pfc.mit.edu

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

From: Bette Byrne <bette%na.oxford.ac.uk@nsfnet-relay.ac.uk>

Date: Mon, 25 Sep 89 15:07:21 BST

12TH INTERNATIONAL CONFERENCE ON

NUMERICAL METHODS IN FLUID DYNAMICS

University of Oxford

9-13 July 1990

The conference will cover all areas of Computational Fluid Dynamics with

particular emphasis on:

Algorithm Development Parallel Computing

Hypersonic Flows Transition and Turbulence

Environmental Flows Propulsion Systems

It will be held in the Lecture Theatre of the Zoology and Psychology Building

and hosted by the Computing Laboratory, Department of Engineering Science

and the Mathematical Institute. Accommodation has been arranged in a number

of Oxford colleges and the Randolph Hotel for the duration of the Conference.

Details of this and other accommodation in Oxford or its surrounding

countryside will be made available upon request.

There will be six invited speakers as well as contributed papers. Two page

abstracts (including sample figures) of contributed papers should be submitted

before December 8, 1989: five copies are required. Notifications of acceptance

will be given by March 12, 1990. Camera ready copies of the final manuscript

will be due at the conference for publication in the proceedings. Abstracts

should be submitted according to the author's home country as follows:

USA USSR and Eastern Europe

Professor M Holt Professor V Rusanov

Dept of Mechanical Engineering Keldysh Inst Appl Mathematics

University of California Miusskaya Pl.4

Berkeley,CA 94720 125047 Moscow A-47

India, Asia, Pacific Rim Canada, Western Europe, Israel

Professor K Oshima (and all other countries)

Inst Space & Astro Science Professor R Temam

3-1-1 Yoshinodai Laboratoire D'Analyse Numerique

Sagamihara Universite Paris Sud/Bat.425

Kanagawa 229, Japan 91405 Orsay, France

A limited number of bursaries will be made available to young potential

contributors born after 9 July 1960. Applications with abstracts should be

sent to Prof. P. J. Zandbergen, Department of Applied Mathematics,

Twente University Technology, PO Box 217, 7500 AE Enschede, The Netherlands.

For further details contact:

Mrs Bette Byrne

Institute for Computational Fluid Dynamics

Oxford University Computing Laboratory

8-11 Keble Road

Oxford OX1 3QD

Tel.+44-865-273883

Fax.+44-865-273839

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

From: Chris Bischof <bischof@antares.mcs.anl.gov>

Date: Tue, 26 Sep 89 09:27:53 CDT

In response to an initiative by Howard Wilson from the University

of Alabama, there is now a MATLAB User group mailing list and

software repository. If you are interested in joining and are not

on the mailing list yet (i.e. you have not yet received the message

describing the setup of the user group and library), send a message

to

matlab-users-request@mcs.anl.gov

and I will add you to the mailing list and send you information

about the mailing list and library.

-- Chris Bischof

Mathematics and Computer Science Division

Argonne National Laboratory

Argonne, IL 60439

(312) 972-8875

bischof@mcs.anl.gov

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

From: Farid Alizadeh <alizadeh@UMN-CS.CS.UMN.EDU>

Date: 26 Sep 89 19:44:54 GMT

Here is a problem I have come across in the course of some optimization

problem.

Let A(x) be a real, symmetric n x n matrix with some or all of its entries

variables. (the vector x consists of variables from the first row to the last

and in each row from the first column to the last. Also, since the matrix

is symmetric the variable at ij entry is the same as the variable at ji.)

It is well-known that the eigenvalues of A(x) are real valued continuous

and smooth functions of x. At a given point x_0 compute the eigenvalues

and impose an arbitrary order on them; so we will get l_1(x_0), l_2(x_0),

..., l_n(x_0). Now compute the eigenvalues at new points x_1, x_2, ...,

x_k, ... The problem is that at each new point x_k I also need to compute a

permutation that reorders eigenvalues so that all eigenvalue

functions l_r are continuous and smooth. That is

l_r(x_0), l_r(x_1), ..., l_r(x_k), ...

are the values of the SAME continuous and SMOOTH function. For a simple

example consider the matrix with one variable:

1 x

x 1

Then, the eigenvalues of this matrix are:

l_1(x)=1+x

l_2(x)=1-x.

Now, for various values of x, say, QR method sometimes produces l_1(x) as the

first eigenvalue and sometimes l_2(x). This is true of just about any other

algorithm that I am aware of. (Notice that eigenvalues generated in increasing

or decreasing order do not represent smooth functions.)

The problem is this: Assuming that we can easily compute the eigenvalues

with enough precision, how can I reorder the eigenvalues at each x_k

so that the r'th eigenvalue in my list is the value of the continuous and

smooth function l_r.

Does anyone know how to do this or know of any reference?

Farid Alizadeh

CSci Dept., University of Minnesota, Mpls.

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

From: Jeff Simon <simon@ncsa.uiuc.edu>

Date: Thu, 28 Sep 89 07:22:21 CDT

I am seeking information on software for the solution of non-symmetric

linear systems implemented for out of core operation. The matrix is

banded and solver may be direct or iterative. I greatly appreciate

information forwarded and replies may be sent to:

simon.ncsa.uiuc.edu

Thank you,

Jeff Simon

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

From: George Corliss <georgec@marque.mu.edu>

Date: 28 Sep 89 03:06:58 GMT

I would appreciate pointers to public domain software (Fortran

preferred) for multi-dimensional quadrature. I checked netlib, but

did not see anything promising. Did I miss something? The

application in question is 4-dimensional. The integrand is

moderately smooth, of no special form.

Thanks in advance.

George Corliss, Marquette University, Milwaukee, WI 53233

georgec@marque.mu.edu, ...!uwvax!marque!georgec, 6591CORL@MUCSD.BITNET

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

From: Jerzy Wasniewski <mfci!wasniews@uunet.UU.NET>

Date: Thu, 28 Sep 89 13:23:44 EDT

What we need is an (IEEE 64-bit) accurate, more efficient replacement of the

RSP EISPACK routine which calculates the eigenvalues and eigenvectors of a

dense, real symmetric (packed) matrix. The algorithms used by RSP are Fortran

translations of the ALGOL TRED3, TQL2, and TRBAK3 procedures described in

Num. Math. 11 (1968) authored by Bowdler, Martin, Reinsch, and Wilkinson.

Jerzy Wasniewski

c/o Multiflow Computer, Inc.,

31 Business Park Drive,

Branford, CT 06405, U.S.A

Tel. office: 203 488 6090

Tel. home (temporary): 203 387 0171

Email address:

wasniewski@multiflow.com

or

na.wasniewski@na-net.stanford.edu

[Editor's comments: What's wrong with RSP? How much more efficient,

or more accurate, do you want or expect? RSP is pretty hard to beat.

Some speedup is available by replacing the QR accumulation of

transformations by inverse iteration using the EISPACK path TRED3,

IMTQLV, TINVIT, TRBAK3. Humberto Madrid wrote a Ph. D. thesis at

the University of New Mexico four years ago where he investigated

"perfect shifts" and "fast Givens" transformations. Jack Dongarra

and Danny Sorensen at Argonne have been touting a divide and

conquer approach for several years. Some recent contributions

by Peter Tang at Argonne and by W. Kahan at Berkeley appear to

guarantee orthogonal of eigenvectors. This approach should find

its was into LAPACK, now under development at Argonne, NYU and NAG.

I guess all this may lead to a 20 to 40% improvement in execution

time, and about the same accuracy, as RSP. That's certainly

worthwhile, but you'll have to wait a little while to have portable,

robust software comparable to that in EISPACK. --Cleve Moler.]

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

From: Jorge More <more@antares.mcs.anl.gov>

Date: Wed, 27 Sep 89 14:45:16 CDT

ARGONNE NATIONAL LABORATORY

MATHEMATICS AND COMPUTER SCIENCE DIVISION

Advanced Scientific Computing

The Mathematics and Computer Science (MCS) Division of Argonne

National Laboratory invites applications for a regular staff

position in the area of advanced scientific computing and

parallel architectures, with emphasis on numerical linear

algebra, optimization, or partial differential equations.

Qualified candidates will also be considered for the position of

Scientific Director of the Advanced Computing Research Facility.

Applicants with a Ph.D. in (applied) mathematics or computer

science will be given preference; however, outstanding candidates

with degrees from other disciplines will be considered. The

position requires extensive knowledge of methods of computational

mathematics, intimate knowledge of advanced computer

architectures, and familiarity with modern visualization

techniques. Applicants must have an established record of

research accomplishments, as evidenced by publications in

refereed journals and conference proceedings.

The MCS Division offers a stimulating environment for basic

research. Current research programs cover areas of applied

analysis, computational mathematics, and software engineering,

with emphasis on advanced scientific computing. The division

operates the Advanced Computing Research Facility (ACRF), which

comprises a network of advanced-architecture computers, ranging

from an 8-processor Alliant FX/8 to a 16,384-processor Connection

Machine CM-2. A network of Sun and NeXT workstations supports

the general computing needs of the division. Argonne's central

computing facilities include a CRAY X/MP-14; additional access to

supercomputers is provided through the major networks.

The Scientific Director of the ACRF is responsible for keeping

abreast of current developments in advanced scientific computing

and maintaining the facility and the research program that it

supports in the forefront of computer science research. The

Director is assisted by a Deputy Scientific Director. The day-

to-day operation of the facility is the responsibility of the

Manager of the MCS Computing Facilities.

Argonne is a multipurpose national laboratory operated by The

University of Chicago for the U.S. Department of Energy. It is

located 25 miles southwest of Chicago.

Applicants are requested to send a detailed resume to Rosalie L.

Bottino, Employment and Placement, Box J-MCS-37017-83, Argonne

National Laboratory, 9700 S. Cass Avenue, Argonne, IL 60439.

Further information about the position can be obtained from Dr.

Hans G. Kaper, Director, MCS Division (kaper@mcs.anl.gov),

telephone 312-972-7162. Argonne is an equal opportunity/

affirmative action employer. Women and minorities are especially

encouraged to apply.

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

From: Roger Grimes <rgrimes@atc.boeing.com>

Date: Thu, 28 Sep 89 06:45:06 PDT

Applied Mathematics

Boeing Computer Services

The Applied Mathematics group in Boeing Computer Services,

located in Seattle, Washington, anticipates openings in early

1990 for well qualified individuals in applied numerical

analysis, especially numerical linear algebra. We are looking

for entry level or experienced PhDs with either a dissertation or

post-graduate experience in numerical analysis and an interest in

applying this experience to challenging real world problems.

Problems will be oriented towards applications of numerical analysis

but will include parallel computing and the environment for large

scale computation, with concerns for distribution of tasks and

visualization.

Our group consists of about 40 mathematicians performing

consulting and research work for other parts of The Boeing

Company, for commercial customers, and for governmental agencies.

We require individuals capable of working independently in a

dynamic, multidisciplinary environment on complex algorithm,

analysis, and application software problems. Permanent U.S.

residency is required; U.S. citizenship is perferred.

Qualified applicants should send a resume to:

Roger G. Grimes

Manager, Computational Mathematics

Boeing Computer Services

P.O. Box 24346, M/S 7L-21

Seattle, WA 98124-0346

USA

Questions regarding the position may be sent, via e-mail, to

Roger Grimes at na.grimes@na-net.stanford.edu or

rgrimes@atc.boeing.com.

Boeing is an Equal Opportunity Employer.

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

From: Tim Kelley <ctk@matctk.ncsu.edu>

Date: Sun, 1 Oct 89 14:03:17 EDT

NORTH CAROLINA STATE UNIVERSITY

Department of Mathematics

Our department intends to make a senior level appointment in PDE-related

applied mathematics, beginning in the Fall of 1990. Mathematical analysts

working on optimization or control problems and related computational

questions are especially encouraged to apply. Candidates must

have outstanding research credentials and a demonstrated competence in

teaching. Send a vita and arrange to have at least three letters of

recommendation sent to: J.C.Dunn, Search Committee Chairman, Department of

Mathematics, Box 8205, North Carolina State University, Raleigh, NC 27695-

8205. Address electronic mail inquiries to jcd@ncsumath.bitnet.

The closing date for applications is January 26, 1990. North Carolina

State University is an equal opportunity / affirmative action employer.

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

End of NA Digest

**************************

-------