The Raman spectrometer was calibrated initially using Ne, Kr and Xe sources. Even though Raman spectrometer was calibrated, Raman shift (Position (cm-1) of particular Raman peak) has an error associated with it. The error basically comes from the pixel size of CCD detector, no of pixels, Laser wavelength accuracy etc. In general, when we record the Raman spectrum for some unknown sample, its peak position are extracted by fitting using a Lorentzian function, which also gives a fitting error.
In this case, how can we determine the total error (which will have contribution from CCD resolution, spectral resolution, Lorentzian fitting error etc.,) in the peak position?
Or in other words, What is the general procedure used for error determination in the Raman position of the recorded spectrum?
I would say that, for the narrowest peaks, the major source of error in determining the peak position is the pixelation, and so I would quote the error of plus/minus half the wavenumber per pixel. However, a Lorenzian fit significantly reduces this error. Limitations of spectral resolution largely effect the peak width.
Dear Hugh J Byrne ,
Thanks for answering the question. We are using Jobin Yvon Triax 550 spectrometer for Raman spectrum collection. The CCD detector details are given below,
1024 X 128 (width x height), no of pixel =1024, 26 micron square pixel.
I have recorded the Raman spectrum of Si sample. It is the narrowest peak (sharp) in nature. I have fitted using Lorentzian function, which gives an error of about 0.01 cm-1.
The step size (cm-1/pixel) of data collection is about 0.5 cm-1.
In this case, what is the value of an error for Raman peak position?
Is it ±0.01 cm-1 or ±0.5 cm-1 or ±0.25 cm-1?
As you indicate, the a dispersion of 0.5 cm-1, enables you to determine the peak position with an accuracy of ±0.25 cm-1. However, with fitting, this can be as good as ±0.01 cm-1.
It really depends on what reason you want to quote it, and for what samples.
For example in publications, I normally quote the dispersion per pixel (cm-1/pixel). We often measure biological or organic materials, which have broad lineshapes.
If, however, you really want to push the precision to the limit, you probably have to check whether there is any calibration drift over the timescale of the measurement.
Dear Rajaji,
The error in the peak position can be determined by the Lorentzian or Gauss-Lorentzian fit to the Raman spectra. Each pixel has an error in spectral resolution given by half of the pixel resolution, in this case 0.25 cm-1, thus you can fit the data adding this error to each point and the error in the peak is the statistical error of the peak fitting. You can repeat this process in 4 spectrum recorded in the same position to take into account the error in the intensity recorded by each pixel. The resultant Raman peak position takes into account both spectral and intensity error of your spectrometer.
Regards
Dear Hugh J Byrne and J. Anaya ,
Thanks for the nice information and discussion.It is really useful. It is inorganic sample (crystal).
Can you please explain me, why an error should be half of the pixel (cm-1/pixel)? Can I get name of the reference book or technical note for this?
The fitted Lorentizan peak will give FWHM and its error value. Is this true error value for FWHM of Raman peak?
Or in other words, how do we find out the error in FWHM (Full Width at Half Maximum) of Raman peak?
I don't have a reference, but "Error" is reallt the degree of accuracy to which you can read something. If you ahve mm gradations on a ruler, the best you can do is measure to the closest mm. In the case of pixels, you can see that the maximum is between two pixels, so the accuracy can be quoted as +/- half a pixel.
In the case of fitting, the programme doesn't know it is Raman pixels, and therefore will five you an error on how well it fits the points to the curve specified, and can be quoted as the error of the fitting of the Raman peak.
Dear Rajaji,
The example of the ruler is a really good one, you don't need any other reference... an easy way to do it is by computing the best fit by least squares of a Lorentzian function to your data with the +-0.25 cm-1 error in Raman shift for each data point of the spectra. You can also evaluate the error of the fitting, which is the error you are looking for. To get a even better evaluation of your accuracy, you can repeat this for 3-4 different spectrum acquired in the same point and evaluating the total error (average and error of the average).
regards
The narrowest possible Raman signal is probably from an isolated (low pesssure) gas molecule, but probably very difficult to measure!!!
- Badges
- Science topic
- Similar topics
- Solid State Physics
- Devices