Understanding Asynchronous Parallel Pattern Search

Abstract

Asynchronous parallel pattern search (APPS) is a nonlinear optimization algorithm that dynamically initiates actions in response to events, rather than cycling through a fixed set of search directions, as is the case for synchronous pattern search. This gives us a versatile concurrent strategy that allows us to effectively balance the computational load across all available processors. However, the semi-autonomous nature of the search complicates the analysis. We concentrate on elucidating the concepts and notation required to track the iterates produced by APPS across all participating processes. To do so, we consider APPS and its synchronous counterpart (PPS) applied to a simple problem. This allows us both to introduce the bookkeeping we found necessary for the analysis and to highlight some of the fundamental differences between APPS and PPS.

Articlenote

This research was sponsored by the Mathematical, Information, and Computational Sciences Division at the United States Department of Energy and by Sandia National Laboratories, a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy under contract DE-AC04-94AL85000.

This research was funded by the Computer Science Research Institute at Sandia National Laboratories and by the National Science Foundation under Grant CCR-9734044.