[ont.events] Domain Decomposition and Iterative Refinement Algorithms for Finite Element Problems.

ylfink@water.waterloo.edu (ylfink) (04/04/88)

DEPARTMENT OF COMPUTER SCIENCE
UNIVERSITY OF WATERLOO
SEMINAR ACTIVITIES

SCIENTIFIC COMPUTATION SEMINAR

                    - Thursday, April 7, 1988

Professor  Olof  Widlund  of  New York University, will
speak   on   ``Domain   Decomposition   and   Iterative
Refinement Algorithms for Finite Element Problems''.

TIME:                4:00 PM

ROOM:              MC 5097

ABSTRACT

In  the  last  few years, there has been a considerable
development   of  so  called  iterative  substructuring
methods,  in which the domain of an elliptic problem is
partitioned   into   non  overlapping  subregions.   In
particular,  a  number of algorithms have been designed
for  which  the  rate of convergence varies very slowly
with  the  number of degrees of freedom as well as with
the  number  of  subregions.   In each of the iteration
steps,  problems on each subregions are solved, and the
rate  of  convergence  is  often  enhanced  by  using a
conjugate gradient method.

The  oldest  of  all  domain  decomposition  methods is
probably the alternating method due to Schwarz.  It was
introduced   already  in  1870.   In  that  method  the
subdomains  overlap.   In  this  talk,  we  will  first
demonstrate   that  there  is  a  similar  ``additive''
algorithm,   which   shows   considerable  promise  for
parallel   computers.   A  general  framework  for  the
analysis   of   this   method   will   be   introduced.
Applications  are then given which show how fast domain
decomposition    methods   and   iterative   refinement
algorithms  can  be  designed.   It  is also shown that
Yserentant's  hierarchical basis function method can be
fit  into  this  framework.   We  note that a number of
algorithmic   ideas   and  technical  tools  have  been
borrowed   from   the   earlier   work   on   iterative
substructuring methods.