| Author |
 Topic  |
|
|
th0
2 Posts |
 Posted - 07/09/2019 : 05:03:30 AM
|
Origin Ver. and Service Release (Select Help-->About Origin): b9.5.0.193
Operating System: win10
Hi there,
I am using the NLFit to fit stretched exponential data. In the past, this always worked and on an older Origin version the presented data is not problematic.
What I have: A copy of the ExpDec1 that includes an arbitrary stretched exponential parameter:
y = A1 * exp((-x/t1)^b1) + y0
I force b1 to be 1 in the beginning. This fitting function works nice on an Origin from 2015 or so. It is also used in this forum: https://my.originlab.com/forum/topic.asp?TOPIC_ID=40795 (though I do not understand what is his problem there).
Origin tells me that the fit does converge without performing any fitting - it takes the calculated initial parameters and claims that the fit converged after the first iteration:
---
(49) ----------Levenberg-Marquardt-----------
Reduced Chi-sqr = 0.0021043241167
COD(R^2) = 0.97233115725141
Iterations Performed = 1
Total Iterations in Session = 1
(50) Fit converged. Chi-Sqr tolerance value of 1E-9 was reached.
---
However, the fit obviously did not converge. If I set the starting value of b1 to anything but 1, Origin errors after 0 iterations that the fit does not converge:
---
(51) ----------Levenberg-Marquardt-----------
Reduced Chi-sqr = 0.100196807768
Iterations Performed = 0
Total Iterations in Session = 1
(52) Fit did not converge - reason unknown.
---
If I change the fitting function to look like this:
y = A1 * exp((-x^b1/t1)) + y0
everything works fine:
---
(56) ----------Levenberg-Marquardt-----------
Reduced Chi-sqr = 5.79336207538E-5
COD(R^2) = 0.99923825601305
Iterations Performed = 78
Total Iterations in Session = 78
(57) Fit converged. Chi-Sqr tolerance value of 1E-9 was reached.
---
but then in order to get the correct t1 I have to manually calculate t1^b1, which is of course possible, but not convenient.
Any leads?
|
|
|
AmandaLu
439 Posts |
Posted - 07/09/2019 : 05:39:25 AM
|
Hi,
The fitting function is wrong, please use this one:
y = A1 * exp(-(x/t1)^b1) + y0
BTW, this function is included in Fitting Function Library App. You are welcome to upgrade to Origin 2019 and try this App:
https://www.originlab.com/fileExchange/details.aspx?fid=490
Thanks,
Amanda
OrigjinLab Technical Service
|
 |
|
|
th0
2 Posts |
Posted - 07/09/2019 : 05:47:03 AM
|
Hi!
Thanks for your quick reply and the solution. I was focused on the fact that the fit converges and did not consider an error in the fitting function, but at second glance it's rather obvious.
I cannot update to 2019 as I'm on an institutional license and we just have what we have. But thanks for hinting to the library, when we update next year I will certainly use it.
th0 |
|
|
| |
Topic |
|
|
|
NLFit converges weirdly