Generating exact D-optimal designs for polynomial models

ABSTRACT

This paper compares several optimization algorithms that can be used to generate exact D-optimal designs (i.e., designs for a specified number of runs) for any polynomial model. The merits and limitations of each algorithm are demonstrated on several low-order polynomial models, with numerical results verified against analytical results. The efficiencies -- with respect to estimating model parameters --of the D-optimal designs are also compared to the efficiencies of one commonly used class of experimental designs: fractional factorial designs. In the examples discussed, D-optimal designs are significantly more efficient than fractional factorial designs when the number of runs is close to the number of parameters in the model.

REFERENCES

• 1: Atkinson, A. C. and Donev, A. N. (1992). Optimum Experimental Designs. Oxford University Press, Oxford.
• 2: Box, M. J. and Draper, N. R. (1971). "Factorial Designs, the |X^T X| Criterion, and Some Related Matters." Technometrics Vol. 13 Issue 4. 731--742.
• 3: Goel, T., Haftka, R. T., Papila, M., Shyy, W. (2006). "Generalized Pointwise Bias Error Bounds for Response Surface Approximations." International Journal for Numerical Methods in Engineering Vol. 65 Is. 12. 2035--2059.
• 4: Goos, P., Kobilinsky A., O'Brien T. E., and Vandebroek, M. (2005). "Model-Robust and Model-Sensitive Designs." Computational Statistics & Data Analysis Vol. 49. 201--216.
• 5: James C. Spall, Introduction to Stochastic Search and Optimization, John Wiley & Sons, Inc., New York, NY, 2003
• 6: Spall, J. C., Hill, S. D., and Stark, D. R. (2006). "Theoretical Framework for Comparing Several Stochastic Optimization Approaches," in Probabilistic and Randomized Methods for Design under Uncertainty (G. Calafiore and F. Dabbene, eds.). Springer-Verlag, London, Chapter 3 (pp. 99--117).

INDEX TERMS

Keywords:
general linear regression, mathematical optimization, optimal experimental design