Parameter Studies
Parameter studies employ deterministic designs to explore the effect of parametric changes within simulation models, yielding one form of sensitivity analysis. They can help assess simulation characteristics such as smoothness, multi-modality, robustness, and nonlinearity, which affect the choice of algorithms and controls in follow-on optimization and UQ studies.
Dakota’s parameter study methods compute response data sets at a selection of points in the parameter space. These points may be specified as a vector, a list, a set of centered vectors, or a multi-dimensional grid. Capability overviews and examples of the different types of parameter studies are provided in the Users Manual.
With the exception of output verbosity (a setting of silent
will
suppress some parameter study diagnostic output), Dakota’s parameter
study methods do not make use of the method independent
controls. Therefore, the parameter study documentation which follows
is limited to the method dependent controls for the vector, list,
centered, and multidimensional parameter study methods.