General Options

Basic Control Parameters

Maximum number of search steps [maxseq=20]: The maximum number of search steps to perform. SEQOPT will stop searching if maxseq is exceeded.

Maximum number of function evaluations [maxtrus=200]: The maximum number of times to execute your analysis code. The maximum value for this option is 4000. SEQOPT will stop searching if maxseq is exceeded.

Tolerance on maximum relative constraint violation [loostol=0.001]: Designs with a relative constraint tolerance less than loostol will be considered feasible.

Initial increment for pattern search [ngrdpll=16]: Controls the size of the initial increment for the the pattern search.

The initial pattern increment for design variable i is:

Δx_i^initial = ( upperBound_i - lowerBound_i ) / ( ngrdpll )

Pattern search refinement multiplier [ngrdmlt=4]: Controls how the pattern search increment is refined.

The new pattern increment is:

Δx_i^new = Δx_i^old / ngrdmlt

Final increment for pattern search [mxgrd=64]: Controls the size of the final increment for the the pattern search.

The final pattern increment for design variable i is:

Δx_i^final = ( upperBound_i - lowerBound_i ) / ( mxgrd )

Surrogate Model Refinement

Maximum number of function evaluations for improving the surrogate model fit [nglobal=3]: Controls the maximum number of times per search step that your analysis will be executed to improve the overall quality of the surrogate models. Put another way, this parameter controls the maximum number of model improvement points that will be generated at each search step. See Task 5

Maximum number of function evaluations for searching near promising infeasible designs [nctrgl=3]: Controls the maximum number of times per search step that your analysis will be executed to search near promising infeasible design points. See Task 5

Random Number Seed

Seed for random number generator [rand=0]: Each time SEQOPT is run (even for the exact same design problem) it will likely take a different path to the solution. This is because there is some built-in randomness in the generation of the initial orthogonal array and in the generation of the model improvement point cloud. This parameter (rand) gives you control over this behavior.

If rand=0, the system clock will be used to seed the random number generator and subsequent optimization runs will (in general) all be different from one another. If you set rand to any other non- zero integer, all subsequent optimization runs will be identical (assuming that all of the other option parameters are the same).

Parallel Computing

Number of available parallel processors Defines the number of processors leveraged for running the optimization. Mutlicore machines can reduce runtimes by increasing this number.