| T O P I C R E V I E W |
| Andreas_L |
Posted - 04/13/2002 : 1:00:21 PM
Dear all,
I am doing a mulit-dataset refinement (see also topic: simultaneous fitting 4/10/2002)
with a relatively complex function
a00+(a0+ a1*exp((x-x0)/s)+ a2*exp((x-x0)/(2*s))+ a3*exp((x-x0)/(3*s))) *exp(-Q*(1/5-1/k))
where k is separate for each set of data.
I set all variables to appropriate values and
already a good fit with refining a1, a2, a3, x0, s.
When I want to refine Q (initial value 160),
there is the error 28036, suggesting setting
the parameter fixed.
This also occurs when I set all other parameters fixed.
Then I tried to vary Q over a wide range and checked the chi^2 without refining.
The curves change considerably, and also with
strongy deviating curves, the indicated chi^2
do not change at all.
So, Q strongly influences the curves considerably, and "by eye" I think there should be an optimum of Q around 160, but
refinement fails.
Any ideas/tricks, with which I can settle that problem? Some parameters in the control sections, which may improve that behaviour?
Best regards
Andreas Leineweber
|
| 2 L A T E S T R E P L I E S (Newest First) |
| H.Steen |
Posted - 04/17/2002 : 11:58:01 AM
I have not tested your function, however is it is possible that you are plotting using a logarithmic scale on your Y axis (I do see that you have several exponential functions) ? If this is the case remember that a visible large difference for small values will be ignored due to the fact that the differences for high values are several orders higher even if they by manual inspection looks equal (again at a plot using logarithmic scale). To solve the problem you have to either change the weight for each data point or, if possible, normalize your data before the fit is done. In my case I fitted a function to Log(Y) instead of Y and then recalculated the result to find the fitting function to the dataset Y.
Helge
|
| rtoomey |
Posted - 04/15/2002 : 10:19:15 AM
Andreas,
From the looks of your equation, it might be overparameterized. This may be the reason Q does not converge at around 160. However, I do not have an explanation as to why the Chi^2 value does not change. If possible, could you please send the following items to OriginLab Technical Support:
- Your fitting function definition file (*.FDF) (see Note)
- Sample data
- A list of each parameter's meaning
- At least one set of initial parameter values
- A list of constants (if there are any)
- A set of instructions with which I can try to reproduce the behavior
Note: FDF files are saved to the \Origin\FitFunc folder. To locate your FDF, open Windows Explorer and locate the FitFunc folder. Then, locate any FDF files which are named user##.fdf (where ## is an index number). Your FDF file will most likely be the one which was most recently created (as noted in its Properties).
We look forward to assisting you.
Sincerely,
Ryan Toomey
Technical Support Engineer
OriginLab Corporation
|
|
|
