Parallel algorithms for nonlinear optimization

Official URL: http://192.168.1.20/record=b1378219 (Table of Contents)

Parallel algorithm design is a very active research topic in optimization as parallel computer architectures have recently become easily accessible. This thesis is about an approach for designing parallel nonlinear programming algorithms. The main idea is to benefit from parallelization in designing new algorithms rather than considering direct parallelizations of the existing methods. We give a general framework following our approach, and then, give distinct algorithms that fit into this framework. The example algorithms we have designed either use procedures of existing methods within a multistart scheme, or they are completely new inherently parallel algorithms. In doing so, we try to show how it is possible to achieve parallelism in algorithm structure (at different levels) so that the resulting algorithms have a good solution performance in terms of robustness, quality of steps, and scalability. We complement our discussion with convergence proofs of the proposed algorithms.