×

A numerical search algorithm for experimental design. (English) Zbl 0719.90058

Optimum experimental design is considered as a nonconvex optimization problem. Special properties of this class of problems are derived, two assertions express the meaning of the symmetry of the objective function together with local convexity. Further, the concept of SLM (semi-local minimizers) is introduced which provides a new algorithm. A convergence theorem for a special case based on actual analytic results on D-optimal designs is proved. The summarized outcome of some numerical experiments with several models of practical interest concludes the paper.
Reviewer: K.Frischmuth

MSC:

90C26 Nonconvex programming, global optimization
90C90 Applications of mathematical programming
62K05 Optimal statistical designs
90C30 Nonlinear programming
62P10 Applications of statistics to biology and medical sciences; meta analysis
PDFBibTeX XMLCite