Why does pattern search give result worse than initial guess?

Hi,

I use pattern search (R2017a) for a constrained minimization problem.

From a previous run using pattern search, I obtained a set of feasible results. But when I feed these numbers to pattern search as initial values again, I occasionally get worse results.

I noticed there was a similar post regarding fmincon. Could this be a similar issue?

Thanks.

Shi

Answer 1

[Walter Roberson]

It could be a constraint violation issue: the new "worse" values might violate the constraints less .

[Shi Qiu]

Is there a value I can check for the "degree of violation"? I thought both of the results are acceptable to pattern search.