Abstract: In a computer experiment, the data are produced by a computer program that models a physical system. The experiment consists of a set of model runs; the design of the experiment specifies the choice of program inputs for each run. This paper demonstrates two applications of a Bayesian method for the design and analysis of computer experiments to predict model output corresponding to input values for which the model has not been run. When the original code is long-running, the fast predictor produced by this method can serve as an efficient, though approximate, substitute. The models used in the two examples are (i) a computer model for the combustion of methane and (ii) a computer model that simulates the compression molding of sheet molding compound in the manufacture of an automobile hood.
Key words and phrases: Bayesian prediction, computer model, combustion, compression molding, interpolation, optimal design, stochastic process.