It's the first time for me to ask a question in such an environment and so I hope someone understands my problem and can help me. I try to fit data with a self-defined fit function. In its simplest mode it looks like: y = a * sqrt(R*R - (x-x0)*(x-x0)) for x>n0-R and x<n0+R and y = 0 for all other cases. For the beginning I try to fit data that I produced with this function (so that I can be sure that I have the right model), but I must recognize that the solution of my fit problem strongly depends on the starting values of the fit parameters (especially of R). Could it be that I have forgotten something in defining my function or is this normal (because of the "squareroot"?)? I just hope that ORIGIN could do better if I would do ...? because in "real life" I don't know the values of the parameters and is it not satisfactory to start many fit routines varying the starting values to find the solution that matches best with reality. What should I do? Thanks and many greetings!
One reason for the "ending" values of the parameters being dependent on what the starting/initial values were, is because the fit is not really converging - which could be due to the chi-sq surface being rather flat and so the minimization just wanders around a flat parameter space and ends up in different places depending on where you started and how many times you performed the iteration etc.
The above could happen if, your function is over-parametrized; has too many parameters and could be consolidated/rewritten to have fewer parameters, or if there is too much dependency between some of the parameters etc.
It is hard to reproduce what you are seeing just by trying to create a function with your expression and then trying to recreate your situation. Best would be for you to send an OPJ with data, and your associated FDF file to tech support so they could examine and comment.
non-linear fit with user-defined function
Posted - 06/02/2004 : 08:23:40 AM
