Abstract:Probably the most popular algorithm for unconstrained minimization for problems of moderate dimension is the Nelder-Mead algorithm published in 1965. Despite its age only limited convergence results exist. Several counterexamples can be found in the literature for which the algorithm performs badly or even fails. A convergent variant derived from the original Nelder-Mead algorithm is presented. The proposed algorithm's convergence is based on the principle of grid restrainment and therefore does not require su… Show more
“…(10) guarantees the existence of limit pointB ∞ and assures us that no limit point contains zero vectors. The latter is a consequence of (8) and does not require an additional lower bound λ on b as in [4]. From (11) and (12) it follows that˜ k > α/(2Λ − α) > 0.…”
Section: Definitionmentioning
confidence: 88%
“…The proof again goes along the lines of [4], except that the grid-restrainment error must be applied twice.…”
Section: Definitionmentioning
confidence: 97%
“…(D) > 0 implies that D positively spans R n (for proof see [4]). Now suppose that the setD is obtained from D by restraining the members of D to grid G using partitioning P. Let d min denote the shortest member of D. Then (provided that no member ofD has zero length)…”
We present an algorithmic framework for unconstrained derivative-free optimization based on dividing the search space in regions (partitions). Every partition is assigned a representative point. The representative points form a grid. A piecewiseconstant approximation to the function subject to optimization is constructed using a partitioning and its corresponding grid. The convergence of the framework to a stationary point of a continuously differentiable function is guaranteed under mild assumptions. The proposed framework is appropriate for upgrading heuristics that lack mathematical analysis into algorithms that guarantee convergence to a local minimizer. A convergent variant of the Nelder-Mead algorithm that conforms to the given framework is constructed. The algorithm is compared to two previously published convergent variants of the NM algorithm. The comparison is conducted on the Moré-Garbow-Hillstrom set of test problems and on four variably-dimensional functions with dimension up to 100. The results of the comparison show that the proposed algorithm outperforms both previously published algorithms.
“…(10) guarantees the existence of limit pointB ∞ and assures us that no limit point contains zero vectors. The latter is a consequence of (8) and does not require an additional lower bound λ on b as in [4]. From (11) and (12) it follows that˜ k > α/(2Λ − α) > 0.…”
Section: Definitionmentioning
confidence: 88%
“…The proof again goes along the lines of [4], except that the grid-restrainment error must be applied twice.…”
Section: Definitionmentioning
confidence: 97%
“…(D) > 0 implies that D positively spans R n (for proof see [4]). Now suppose that the setD is obtained from D by restraining the members of D to grid G using partitioning P. Let d min denote the shortest member of D. Then (provided that no member ofD has zero length)…”
We present an algorithmic framework for unconstrained derivative-free optimization based on dividing the search space in regions (partitions). Every partition is assigned a representative point. The representative points form a grid. A piecewiseconstant approximation to the function subject to optimization is constructed using a partitioning and its corresponding grid. The convergence of the framework to a stationary point of a continuously differentiable function is guaranteed under mild assumptions. The proposed framework is appropriate for upgrading heuristics that lack mathematical analysis into algorithms that guarantee convergence to a local minimizer. A convergent variant of the Nelder-Mead algorithm that conforms to the given framework is constructed. The algorithm is compared to two previously published convergent variants of the NM algorithm. The comparison is conducted on the Moré-Garbow-Hillstrom set of test problems and on four variably-dimensional functions with dimension up to 100. The results of the comparison show that the proposed algorithm outperforms both previously published algorithms.
“…According to Burmen et al (2005), the performance of this approach could be shortly summarized as follows.…”
Section: Nelder-mead (Nm) Algorithmmentioning
confidence: 99%
“…First-best levels are always higher than the second-best ones in line with the mainstream of previous literature on 32 Variants of the original Nelder-Mead method are focused on avoiding problems related to this one. For further details, see Burmen et al (2005). 33 The differences in favour of RIM were initially more significant when the problems were straightforward implemented.…”
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.