ISSN:
0049-8246
Keywords:
Chemistry
;
Analytical Chemistry and Spectroscopy
Source:
Wiley InterScience Backfile Collection 1832-2000
Topics:
Physics
Notes:
Linear and non-linear least-squares profile-fitting procedures are compared in their ability to resolve severely overlapping peaks in X-ray spectra. With linear methods, errors due to inaccuracies in peak position and width are likely to be greater than statistical errors for a spectrom with 500 K counts unless the spectrometer can maintain a temporal and count are stability of 1 eV or better. While non-linear, iterative, procedures can accomodate changes in position and width of peaks, they are difficult to analyse therotically so simulations have been designed to examine performance. Examples are given where, despite exact knowledge of peak width and separation, the best fit in the ‘chi-square’ sense corresponds to peak heights which are totally incorrect. Estimates of statistical error are virtually meaningless when convergence occurs to such inapprooriate minima but rather precise starting values for the unknown parameters are required to ensure convergence to a relistic solution. In view of these potential difficulties and the lack of any theoritical analysis of the instabilties introduced by statistical noise, errors in background subtraction or unsuitable parameterization of peak models, it is felt that linear method are preferable. Thus, high spectrometer stability still seems to be unavoidable if one is to exploit the availabler statistical accuracy in an X-ray spectrum.
Additional Material:
1 Ill.
Type of Medium:
Electronic Resource
URL:
http://dx.doi.org/10.1002/xrs.1300070304
Permalink