0

I am using the lmfit library in Python and I get huge uncertainties. I have noticed it happens when the best fit parameter itself is very small. Do you know how the fitting uncertainties are evaluated?

Here I show an example of a Voigt fit result (Voigt is a convolution of a Gaussian and Lorentzian). The program suggests that the Lorentzian width (lwid8) is very, very small (so the profile should be just Gaussian I suppose). The first column is the best fit result and the second column is the error. [gaussian and lorenzian width output errors]

gwid8: 0.06615227 +/- 0.02554561 (38.62%) (init = 0.1)

lwid8: 1.0736e-04 +/- 1.51371359 (1409918.22%) (init = 0.001)

Amyx
  • 17
  • 3
  • Hi, You should always provide informations in text form and not as screenshots. They might be helpful when showing design issues or graph bot not for numbers and code – gerda die gandalfziege Dec 20 '22 at 17:50
  • Thank you! It's my first question ever, I will edit it :) – Amyx Dec 21 '22 at 11:43
  • Please post a minimal, working example of code that shows the problem. By itself, large uncertainties are not a problem, they just tell you which parameters are not having much effect on the fit. As an example (and maybe a guess) if the amplitude of the Lorentzian component was tiny, then the value of its width would probably not be well-defined and so be "very uncertain". – M Newville Dec 21 '22 at 15:21

0 Answers0