pattern search does not perform function calculation between iterations.

Hi all,

I wrote a function and try to find an optimal number by using pattern search. From the message below, the function has no problem. However, as it shows that the number of Func-count does not change between iterations, which is very weired. Anyone has an idea to solve it, please advise.

Iter Func-count f(x) MeshSize Method

0 1 -78.1881 16

1 1 -78.1881 8 Refine Mesh

2 1 -78.1881 4 Refine Mesh

3 1 -78.1881 2 Refine Mesh

4 1 -78.1881 1 Refine Mesh

Answer 1

[Alan Weiss]

I suggest that you turn on iterative display. I suspect that most of the values patternsearch samples evaluate to NaN or Inf or complex, but cannot say without seeing more details.

Alan Weiss

MATLAB mathematical toolbox documentation

[Wei Ting]

Hi Alan,

Thanks for your reply. I looked into the problem and find the following. As you can see the 1st input variable was changed by 16 from 458 to 442 by pattern search, which does not agree to the value of 0.0625 shown in displayed iteration. Weirly, it actually corresponds to 1/0.0625 = 16. Any idea on the root cause?

% the first eight numbers are input variables to the function and the 9th one is figure of merit

458.0000 158.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -81.3368

22 89 -93.1464 0.0625 Refine Mesh

442.0000 158.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -92.3996

Regards,

Wei Ting

[Wei Ting]

Hi Alan,

well, it does not follow 1/(mesh size). I past more data for your reference. Really weird that the change of variable does not follow mesh size. The relevant calculations from an iteration to the next is highlited in bold.

Iter Func-count f(x) MeshSize Method

20 71 -91.8847 0.0625 Successful Poll

442.0000 166.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -82.6993

426.0000 174.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -89.3162

426.0000 166.0000 466.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -83.1097

426.0000 166.0000 434.0000 38.0000 750.0000 22.0000 200.0000 224.0000 -91.4110

426.0000 166.0000 434.0000 22.0000 750.0000 38.0000 200.0000 224.0000 -91.2354

426.0000 166.0000 434.0000 22.0000 750.0000 22.0000 216.0000 224.0000 -91.8847

426.0000 166.0000 434.0000 22.0000 750.0000 22.0000 200.0000 240.0000 -91.8847

410.0000 166.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -85.2580

426.0000 158.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -93.1464

21 80 -93.1464 0.125 Successful Poll

458.0000 158.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -81.3368

426.0000 174.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -89.3162

426.0000 158.0000 434.0000 54.0000 750.0000 22.0000 200.0000 224.0000 -92.8723

426.0000 158.0000 434.0000 22.0000 750.0000 54.0000 200.0000 224.0000 -91.1870

426.0000 158.0000 434.0000 22.0000 750.0000 22.0000 232.0000 224.0000 -93.1464

394.0000 158.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -82.7859

426.0000 142.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -92.3996

426.0000 158.0000 434.0000 22.0000 750.0000 22.0000 168.0000 224.0000 -93.1464

426.0000 158.0000 434.0000 22.0000 750.0000 22.0000 200.0000 192.0000 -93.1464

22 89 -93.1464 0.0625 Refine Mesh

442.0000 158.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -92.3996

426.0000 166.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -91.8847

426.0000 158.0000 466.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -88.7299

426.0000 158.0000 434.0000 38.0000 750.0000 22.0000 200.0000 224.0000 -92.9372

426.0000 158.0000 434.0000 22.0000 750.0000 38.0000 200.0000 224.0000 -92.4792

426.0000 158.0000 434.0000 22.0000 750.0000 22.0000 216.0000 224.0000 -93.1464

426.0000 158.0000 434.0000 22.0000 750.0000 22.0000 200.0000 240.0000 -93.1464

410.0000 158.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -89.3162

426.0000 150.0000 434.0000 22.0000 750.0000 22.0000 200.0000 224.0000 -82.6993

426.0000 158.0000 434.0000 6.0000 750.0000 22.0000 200.0000 224.0000 -92.9917

426.0000 158.0000 434.0000 22.0000 750.0000 6.0000 200.0000 224.0000 -93.4117

23 100 -93.4117 0.125 Successful Poll

458.0000 158.0000 434.0000 22.0000 750.0000 6.0000 200.0000 224.0000 -91.2467

426.0000 174.0000 434.0000 22.0000 750.0000 6.0000 200.0000 224.0000 -89.6189

426.0000 158.0000 434.0000 54.0000 750.0000 6.0000 200.0000 224.0000 -93.1212

[Alan Weiss]

I am no longer sure what your question is. Clearly, the iterative display shows that the function count increases each iteration. Does that solve your problem? Or is there something else? Perhaps you are confused about patternsearch scaling. You can always set the ScaleMesh option to false to have patternsearch not scale the mesh.

Alan Weiss

MATLAB mathematical toolbox documentation