Maximin designs for computer experiments

Decision processes are nowadays often facilitated by simulation tools. In the field of engineering, for example, such tools are used to simulate the behavior of products and processes. Simulation runs, however, are often very time-consuming, and, hence, the number of simulation runs allowed is limited in practice. The problem then is to determine which simulation runs to perform such that the maximal amount of information about the product or process is obtained. This problem is addressed in the first part of the thesis. It is proposed to use so-called maximin Latin hypercube designs and many new results for this class of designs are obtained. In the second part, the case of multiple interrelated simulation tools is considered and a framework to deal with such tools is introduced. Important steps in this framework are the construction and the use of coordination methods and of nested designs in order to control the dependencies present between the various simulation tools