During the de-convolution of peaks from a given X-ray photoelectron spectroscopy (XPS) spectra and subsequent curve-fitting in CasaXPS the GL(m) function is used. GL(m) denotes the product of a Gaussian and Lorentzian function and the parameter "m" denotes the percentage of Gaussian character.
In CasaXPS, the standard value of "m" is taken as 30 for deconvolution of XPS spectra into its component peaks, meaning the peaks have a 30 % Gaussian character.
Are there any constraints for setting the value of the parameter "m" ? For instance, I came across literature that for the case of Carbon 1s signals from Polymers, the value of "m" is set as 90.
I came currently dealing with the XPS spectra of vanadium oxide. Is there any thumb-rule or constraint for selecting the value of "m" for fitting Vanadium 2p and Oxygen 1s spectra ?
The outcome of a fit also depends on how you model the inelastic background. If you do this very well, and have high quality data (-> very good specimen that is stable over the entire time of data taking) then the Lorentzian and Gaussian contributions to the data will flow from the fit rather than being an input.
It all depends on what you wnat to do with the fit results. If you want to do physics with the parameters that determine the line shape (i.e. with Gaussian and Lorentzian widths) then excellent state of the art data taking is mandatory.
If you just want to have a "nice looking fit" then see how you can achieve that and always take the same choices with the same kind of specimens to stay consistent and obtain results that you can meaningfully compare.
If you want to determine peak areas then again background modeling is of utmost importance. And don't expect accuracies better than, say, 15% or so - unless you know what to do to get it right. there is ample literature out there.
Dear Abhisek, I do not think that there is such rules for selecting the Gaussian contribution or Lorenzian contribution. It depends on the system and how many peaks you are fitting under a High-resolution spectrum.
In Vanadium I do not know, but for O 1s, if you want to fit two peaks, you can use 50% Gaussian and 50% Lorenzian. Still I will say that there is not such rules for choosing these value. This is basically chosen by us to represent the best fit, which is supporting the research data.
Agreed, m is largely empirical. The only guideline is that since Gaussian is a random distribution, and Lorentzian is tighter, that a signal from a crystal might be expected to have smaller Gaussian component and amorphous material might be expected to have larger Gaussian component. At least that's the rule of thumb I've come across when fitting FTIR spectra.
The outcome of a fit also depends on how you model the inelastic background. If you do this very well, and have high quality data (-> very good specimen that is stable over the entire time of data taking) then the Lorentzian and Gaussian contributions to the data will flow from the fit rather than being an input.
It all depends on what you wnat to do with the fit results. If you want to do physics with the parameters that determine the line shape (i.e. with Gaussian and Lorentzian widths) then excellent state of the art data taking is mandatory.
If you just want to have a "nice looking fit" then see how you can achieve that and always take the same choices with the same kind of specimens to stay consistent and obtain results that you can meaningfully compare.
If you want to determine peak areas then again background modeling is of utmost importance. And don't expect accuracies better than, say, 15% or so - unless you know what to do to get it right. there is ample literature out there.
- Badges
- Science topic