Figure 1.

Schematic of the multi-fidelity design under uncertainty framework: the quantity of interest Y(x;ξ) is a random field in the design space, realizations of which could originate from information sources of variable fidelity (such as computer codes, experiments, expert opinion, etc.). A response surface S(x) encodes the dependence of a derived quantity of interest f(Y(x;ξ)) on the input design variables x and uncertain parameters ξ. The ability to efficiently construct response surfaces (e.g. $S(\mathbf{\text{x}})=\mathbb{E}[f(\mathbf{\text{Y}}(\mathbf{\text{x}};\xi ))]$) allows one to asses the system's performance as a function of the random inputs and identify optimal configurations based on the desired design criteria (table 1). (Online version in colour.)