Unconstrained derivative-free optimization by successive approximation (Q953370)

The authors present an algorithmic framework for unconstrained derivative-free optimization based on successive approximation. The framework relies on the grid restrainment and search space partitioning to ensure that a subsequence of iterates exists that converges to a local minimizer of a continuously differentiable function. Its convergence is established under mild assumptions. A variant of the Nelder-Mead algorithm conforming to the framework is described. The algorithm is tested on the Moré-Garbow-Hillstrom test suite and on some multi-dimensional test problems with dimension ranging up to 100. The results showed that the proposed algorithm outperforms two previously known convergent simplex variants.