Concurrent search algorithms for unconstrained optimization

Abstract

Parallel processing enables performing concurrent operations. The execution time of a given computational task is expected to be decreased by distributing its computations among parallel processors. The majority of the works in the parallel nonlinear optimization literature focuses on speeding up some well-known sequential algorithms by parallel execution of their computationally intensive tasks. In this study, on the other hand, we discuss an alternative perspective that leads to new nonlinear programming algorithms, which are inherently parallel by design. To support our discussion with a successful application, we introduce a concurrent search framework for unconstrained nonlinear optimization and provide extensive numerical results with two algorithms that fall into the proposed framework