Advanced Options

Advanced Control Parameters

Num surrogate based optimization runs at each search step [nruns=20]: Controls the number of gradient based optimization runs that will be performed (at each search step) using the surrogate models (Task 3). Each gradient based optimization run will be started from a random initial starting point.

Increment for optimization grid [ngrdopt=512]: Each design variable's domain is split into a grid having ngrdopt intervals. SEQOPT automatically moves each design point to the nearest point on this grid. It is recommended that ngrdopt be greater than or equal to mxgrd.

Num iterations w/o improvement before starting pattern search [mxstall=max(n/2, 3)]: If mxstall consecutive search steps fail to either find a new best design or a new least infeasible point, a local pattern search will be initiated. The default value for mxstall is the maximum of 3 or n/2, where n is the number of design variables.

Approx pt. cloud size when generating model improvement points [ncloud0=5000]: The maximum number of points in the initial model improvement point cloud.

SQP Algorithm

The following parameters apply to the gradient based optimizer when it is executed on the surrogate model problem. (In SEQOPT, the gradient based optimizer is never executed directly on your analysis). These parameters are applied separately to each of the nruns optimization runs

Max number of function evaluations on surrogate model [maxnfe=2000]: The maximum number of function evaluations that the gradient based optimizer will perform when optimizing on the surrogate model problem.

Max number of iterations [nitmax=75]: The maximum number of optimization iterations that the gradient based optimizer will execute when optimizing on the surrogate model problem. This applies separately to each of the nruns optimization runs.

Finite difference step size for calculating gradients [prtsiz=1.0e-7]: Relative perturbation size for forward difference derivatives.

Tolerance on objective function [objtol=1.0e-5]: Convergence tolerance on objective function value.

Tolerance on constraint feasibility [contol=1.0e-5]: Constraints with a relative constraint violation less than control will be considered feasible.

Tolerance on projected gradient [pgdtol=1.0e-5]: Convergence tolerance on the projected gradient.