Dakota Reference Manual  Version 6.2
Large-Scale Engineering Optimization and Uncertainty Analysis
 All Pages

Estimate order of convergence of a response as model fidelity increases


Alias: none

Argument(s): none

Required/Optional Description of Group Dakota Keyword Dakota Keyword Description
(Choose One)
Group 1 estimate_order Compute the best estimate of the convergence order from three points
converge_order Refine until the estimated covergence order converges
converge_qoi Refine until the response converges
Optional refinement_rate Rate at which the state variables are refined
Optional model_pointer

Identifier for model block to be used by a method


Solution verification procedures estimate the order of convergence of the simulation response data during the course of a refinement study. This branch of methods is new and currently only contains one algorithm: Richardson extrapolation.

Refinement of the model

The model fidelity must be parameterized by one or more continuous state variable(s).

The refinement path is determined from the initial_state of the continuous_state variables specification in combination with the refinement_rate, where each of the state variables is treated as an independent refinement factor and each of the initial state values is repeatedly divided by the refinement rate value to define new discretization states.


Three algorithm options are currently provided:

  1. estimate_order
  2. converge_order
  3. converge_qoi

Stopping Criteria

The method employs the max_iterations and convergence_tolerance method independent controls as stopping criteria.


In each of these cases, convergence order for a response quantity of interest (QoI) is estimated from

\[p = ln\left(\frac{QoI_3 - QoI_2}{QoI_2 - QoI_1}\right)/ln(r)\]

where $r$ is the uniform refinement rate specified by refinement_rate.