I have a constrained optimization problem which is not differentiable (because of some random errors), but it should be solvable with patternsearch.
However, I have a nonlinear equality constraint, and it seems the algorithm is making many inefficient steps. To give you some idea:
The optimization is for two variables. The optimum is around [4,0] (where lower bounds are [0,0], so that lower bound is binding for the second argument)
Initial value: [0.3,0.05]; Ceq = 1.8;
Then it evaluates [8.3,0.05], where the equality constraint is -1.2
And then in the first iteration pattern search starts to evaluate [8.3,x] where x will vary between 0.04 & 0.52, and includes more than 10 points, where any increase in x just worsens the equality constraint (gets more negative).
So my question is: is there a way to ensure that the patternsearch takes more efficient steps? I already set 'Cache' to 'on'.