why patternsearch result is dependent to the initial values?

Hi I have a time dependent measured data set and want to fit a 4 parameter nonlinear model to it. I have used patternsearch to minimize sum square differences between the model values and original data. The optimization process ends in exitflag 3 and the resulted parameters are dependent to the initial values, i.e. when i double the initial values, two parameters change largely and the other two parameters not. I have replaced the resulted parameters in the model and plotted model results together with original data. There is an excellent fit but i do not know how to resolve dependency to the original values.

thanks a lot

Answer 1

[Alan Weiss]

If patternsearch is giving you a good fit with widely different solutions, then your problem might have multiple solutions, and patternsearch finds one that is nearby to its starting point.

Alan Weiss

MATLAB mathematical toolbox documentation

[Effat]

Thanks for the answer. I had guessed it before so i tried globalsearch instead of patternsearch to find its global minimum but the fit was worse than that of patternsearch and again dependent to the initial values.I do not now how to constrain the objective function to get the unique result together with a good fit. I also tried the fining the function tolerance but got no result.