ncsu_direct

DIviding RECTangles method

Topics

global_optimization_methods

Specification

• Alias: None

• Arguments: None

Child Keywords:

Required/Optional	Description of Group	Dakota Keyword	Dakota Keyword Description
Optional		solution_target	Specifies a globally optimal value toward which the optimizer should track
Optional		min_boxsize_limit	Stopping Criterion based on shortest edge of hyperrectangle
Optional		volume_boxsize_limit	Stopping criterion based on volume of search space
Optional		convergence_tolerance	Stopping criterion based on objective function or statistics convergence
Optional		max_iterations	Number of iterations allowed for optimizers and adaptive UQ methods
Optional		max_function_evaluations	Number of function evaluations allowed for optimizers
Optional		scaling	Turn on scaling for variables, responses, and constraints
Optional		model_pointer	Identifier for model block to be used by a method

Description

North Carolina State University (NCSU) has an implementation of the DIRECT algorithm (DIviding RECTangles algorithm that is outlined in the SCOLIB method section above). This version is documented in [Gab01] We have found that the NCSU DIRECT implementation works better and is more robust for some problems than coliny_direct. Currently, we maintain both versions of DIRECT in Dakota; in the future, we may deprecate one.

The NCSU DIRECT method is selected with ncsu_direct. We have tried to maintain consistency between the keywords in SCOLIB and NCSU implementation of DIRECT, but the algorithms have different parameters, so the keywords sometimes have slightly different meaning.

Stopping Criteria

The algorithm stops based on:

• method-ncsu_direct-max_iterations - number of iterations

• method-ncsu_direct-max_function_evaluations - number of function evaluations

• method-ncsu_direct-solution_target and method-ncsu_direct-convergence_tolerance

• method-ncsu_direct-min_boxsize_limit

• method-ncsu_direct-volume_boxsize_limit

This method will always strictly respect the number of iterations, but may slightly exceed the number of function evaluations, as it will always explore all sub-rectangles at the current level.

Expected HDF5 Output

If Dakota was built with HDF5 support and run with the environment-results_output-hdf5 keyword, this method writes the following results to HDF5:

• hdf5_results-best_params

• hdf5_results-best_obj_fncs (when responses-objective_functions) are specified)

• hdf5_results-best_constraints

• hdf5_results-calibration (when responses-calibration_terms are specified)