Using sensitivity information in the construction of kriging models for design optimization

We discuss the effects of including sensitivity information in the construction of surrogates to approximate a scalar response of a set of variables. In particular, we consider whether sensitivity information can be used to improve the accuracy of surrogates and their predictive utility when used in connection with derivative-based optimization algorithms. The particular class of surrogates that we study are the interpolatory kriging models. Originally designed for modeling stochastic phenomena in geology, this class of models has been recently popularized for deterministic phenomena in the context of the design and analysis of computer experiments, or DACE.