auto_reorder

Address any model misordering in MFMC numerical solutions by reordering on the fly.

Specification

• Alias: None

• Arguments: None

Description

The ordering of the model hierarchy is continuously redefined as needed to allow for proper evaluation of MFMC estimator variance expressions. The specified model order in the ensemble surrogate model specification is only used as the starting point for the algorithm.

On-the-fly model reordering can result in lower estimator variance, but it can affect solution robustness (due to discontinuous gradients) and complicate comparisons with published MFMC implementations. The use of global-local numerical solver selections can mitigate the first concern to a degree.

Examples

method,
    multifidelity_sampling
    numerical_solve override
      global_local
      model_reordering auto_reorder   # default approach
    solution_mode online_pilot
    pilot_samples = 50 seed = 8674132
    max_function_evaluations = 500