[ont.events] Solving Very Large-Simply-Constrained Optimization Problems.

ylfink@water.UUCP (10/28/87)

DEPARTMENT OF COMPUTER SCIENCE
UNIVERSITY OF WATERLOO
SEMINAR ACTIVITIES

SCIENTIFIC COMPUTATION SEMINAR

                    -  Thursday, October 29, 1987

Dr. Ron Dembo, of the University of Toronto, will speak
on ``Solving Very Large-Simply-Constrained Optimization
Problems''.

TIME:                4:00 PM

ROOM:              MC 5097

ABSTRACT

One  of the significant challenges open to optimization
researchers today is the design of efficient algorithms
for  very  large  nonlinearly-constrained  optimization
problems.   One  possible  strategy  is  to reduce such
problems  to  a  sequence  of simply-constrained (i.e.,
box-constrained)  problems  for  which  very successful
algorithms have been developed.  We present some of the
algorithms for box-constrained minimization.  They have
been  used  to  solve  problems  of  up  to  1  million
variables  and  are very well-suited to vector computer
architecture.   Some  experience  with vectorization of
these  algorithms  will also be given.  A framework for
solving  the  general  NLP problem using smoothed exact
penalties will also be discussed.