Search

Abstract: A stochastic global optimization method based on Multistart is presented. In this, the local search is conditionally applied with a probability that takes in account the topology of the objective function at the detail offered by the current status of exploration. As a result, the number of unnecessary local searches is drastically limited, yielding an efficient method. Results of its application on a set of common test functions are reported, along with a performance comparison against other established methods of similar nature. [Copyright &y& Elsevier]

Abstract: We present three new stopping rules for Multistart based methods. The first uses a device that enables the determination of the coverage of the bounded search domain. The second is based on the comparison of asymptotic expectation values of observable quantities to the actually measured ones. The third offers a probabilistic estimate for the number of local minima inside the search domain. Their performance is tested and compared to that of other widely used rules on a host of test problems in the framework of Multistart. [Copyright &y& Elsevier]

Highlights: [•] A global optimization framework for SMPs and multicore clusters is presented. [•] It exploits hierarchal and dynamic task parallelism of the Multistart method. [•] Gradient/Hessian calculations and Newton’s optimization method are parallelized. [•] Several task distribution schemes are studied and evaluated. [•] Our framework is applied successfully to the protein conformation problem. [Copyright &y& Elsevier]