Why does patternsearch decrease mesh size without function evaluation?

Dear all, I'm using patternsearch to find the minimum of a user-defined function. However, I think I fail to parametrize the patternsearch algorithm in an appropriate way. Unfortunately, the patternsearch algorithm refines the mesh size to smaller values without calling the objective function. This is the diagostic information I get:

Diagnostic information.
  objective function = @(x)PF_objfun_loading_NEW(x,api,vscs(2:size(vscs,1),:),vscs(1,:),outages,lines)
  1 Inequality constraints
  0 Equality constraints
  1 Total number of linear constraints

  X0 = -500
Modified options:
  options.TolMesh = 0.1
  options.TolCon = 1e-06
  options.TolX = 1e-06
  options.TolFun = 1e-06
  options.TolBind = 0.001
  options.MaxIter = 7000
  options.TimeLimit = Inf
  options.MaxFunEvals = '2000*numberofvariables'
  options.MeshContraction = 0.5
  options.MeshExpansion = 2 
  options.MeshAccelerator = 'off'
  options.MeshRotate = 'on'
  options.InitialMeshSize = 1000
  options.InitialPenalty = 10
  options.PenaltyFactor = 100
  options.ScaleMesh = 'on'
  options.MaxMeshSize = Inf
  options.PollMethod = 'gpspositivebasis2n'
  options.CompletePoll = 'on'
  options.PollingOrder = 'consecutive'
  options.CompleteSearch = 'off'
  options.Display = 'diagnose'
  options.PlotFcns = {  @psplotbestf @psplotmeshsize @psplotfuncount @psplotbestx }
  options.PlotInterval = 1 
  options.Cache = 'on'
  options.CacheSize = 1000000
  options.CacheTol = 2.2204e-16
  options.Vectorized = 'off'
End of diagnostic information.

Iter     f-count          f(x)      MeshSize     Method
    0        1        5.04584          1000      
    1        1        5.04584           500     Refine Mesh
    2        1        5.04584           250     Refine Mesh
    3        1        5.04584           125     Refine Mesh
    4        1        5.04584          62.5     Refine Mesh
    5        1        5.04584         31.25     Refine Mesh
    6        1        5.04584         15.63     Refine Mesh
    7        1        5.04584         7.813     Refine Mesh
    8        1        5.04584         3.906     Refine Mesh
    9        1        5.04584         1.953     Refine Mesh
   10        1        5.04584        0.9766     Refine Mesh
   11        1        5.04584        0.4883     Refine Mesh
   12        2        5.04584        0.2441     Refine Mesh
   13        4        5.04584        0.1221     Refine Mesh
   14        6        5.04522        0.2441     Successful Poll
   15        8        5.04522        0.1221     Refine Mesh
   16       10        5.04522       0.06104     Refine Mesh
Optimization terminated: mesh size less than options.TolMesh.

I'm wondering why the patternsearch algorithm decreases the mesh size within iteration 1 to 13 without evaluating the objective function (f-count remains 1).

Can anybody help?

Kind regards, Sven

Answer 1

[Alan Weiss]

We didn't see your patternsearch call. Perhaps your bounds are well within the rather large initial mesh that you gave, so that patternsearch cannot evaluate a mesh point without stepping out of the bounds. But this is just a guess. For more help, post your call, bounds, and your linear inequality constraint.

Alan Weiss

MATLAB mathematical toolbox documentation