refinement_metric

Metric used for guiding adaptive refinement during UQ.

Specification

• Alias: None

• Arguments: None

• Default: requested statistics

Child Keywords:

Required/Optional	Description of Group	Dakota Keyword	Dakota Keyword Description
Required (Choose One)	type of refinement metric	level_mappings	Utilize the level mappings metric for guiding adaptive refinement during UQ.
		covariance	Utilize the response covariance metric for guiding adaptive refinement during UQ.

Description

The refinement metric for each (greedy) refinement candidate defaults to the change induced by the candidate across the set of level mappings (L2 norm in the results for response levels, probability levels, reliability levels, and generalized reliability levels) whenever level mappings are specified. This supports goal orientation during the refinement process. When no level mappings are specified, the default metric is the norm of the change in the response covariance matrix for the QoI targets.

This specification option allows the user to override the default, in particular activating the response covariance option despite the presence of level mappings (see example below).

Examples

method,
    multifidelity_polynomial_chaos
      p_refinement uniform
        expansion_order_sequence = 0 cross_validation
        collocation_ratio = 0.75
        seed = 8674132
      allocation_control greedy
        refinement_metric covariance  # override default of level mappings
        convergence_tolerance = 1.e-6
      samples_on_emulator = 250000
      num_response_levels = 41 41 41
      response_levels = -100:5:100 -100:5:100 -100:5:100