Combined algorithm to compute D-optimal designs

摘要

Algorithmic techniques for computing optimal designs continue being a need in the optimal experimental design field. The increasing interest in finding the optimal experimental conditions makes that new methods are demanded for more complex frameworks showing more realistic situations. Numerical techniques are often the unique viable option for computing these designs since to tackle analytically the problem results impracticable in most of cases. Wynn–Fedorov algorithm, multiplicative algorithm and their modifications are the more frequently used methods in the literature for computing D-optimal designs. However, they are not always suitable and efficient to compute optimal designs since their procedures are very limited by the computational requirements. A new algorithm to obtain D-optimal designs is proposed in this paper. It is based on combining suitable strategies followed by the traditional algorithms. A proof of the convergence is provided and several numerical examples are presented to illustrate its improved results.