Unconstrained Nonlinear Optimization in Parallel via
Multisplitting (Master's defense)
Abstract
Large-scale nonlinear programming problems (NLP),
and in particular, unconstrained nonlinear optimization problems,
push the limits of the available computational
resources of today's serial computers.
Parallel computing has two salient features that address
the increasing size and complexity of large-scale problems:
(i) the breakdown of large-scale problems too large to be
solved currently in the serial environment and
(ii) the speed advantage of running several processors at once as opposed
to running just one processor.
The degree to which nonlinear optimization algorithms are amenable
to parallelization by problem decomposition or {\it multisplitting}
forms the basis of our investigation. Nonlinear optimization multisplitting
(NOMS) is the replacement of a large-scale
NLP by a set of nonlinear subproblems in which each subproblem can be
solved locally and independently in parallel by using existing
sequential algorithms.
Convergence of the unconstrained NOMS algorithm will be discussed
and numerical results will be presented.
For further information please contact:
mittelmann@asu.edu
