convergence_tolerance

Stopping criterion based on relative error reduction

Topics

method_independent_controls

Specification

• Alias: None

• Arguments: REAL

• Default: 1.e-4

Child Keywords:

Required/Optional	Description of Group	Dakota Keyword	Dakota Keyword Description
Optional (Choose One)	Convergence tolerance type	relative	Assess convergence in adaptive UQ methods using statistical metrics that are relative to a benchmark
		absolute	Use absolute statistical metrics for assessing convergence in adaptive UQ methods

Description

The convergence_tolerance specification provides a real value for controlling the termination of iteration through satisfaction of a specified accuracy at minimum total cost.

For multilevel and multifidelity sampling methods, it is a relative convergence tolerance on the mean squared error (MSE), where the MSE is typically comprised of known estimator variance (stochastic error), neglecting unknown residual bias (deterministic error).

The reference estimator variance is then defined as follows:

• Multilevel approaches ( multilevel_sampling and multilevel_multifidelity_sampling): tolerance relative to initial multilevel Monte Carlo (MLMC) error after pilot sample

• Control variate approaches ( multifidelity_sampling and approximate_control_variate): relative to Monte Carlo (MC) error from pilot sample where these are all initial estimator variances fot the algorithm, recognizing that control variate approaches can extract no benefit from low-fidelity pilot samples prior to the definition of corresponding sample increments.

Permissible values are between 0 and 1, non-inclusive. Default value is 1.e-4.