Optimizing accuracy and precision in spectral data

摘要

Parameters derived by computer fitting of Gaussian or Lorentzian curves to spectral data suffer from inaccuracies derived by the data gathering strategy, by instrumental errors and by the inherent statistical fluctuation of the measurements. When measurements of extreme quality are required, minimum uncertainty for peak centroid and linewidth for a given data collection time is obtained by proper choice of the number of 'channels' in the peak. The optimum conditions that are derived analytically are supported by results of fitting spectra with Gaussian curves. Fittings were done with the simplex procedure yielding precise values of the centroid and linewidth. Data used were x-ray fluorescence signals and computer-simulated spectral peaks. Results of the theoretical calculations and of computer fittings to both empirical data and simulated spectra indicate that the optimum number of channels in a peak, for the best quality of measurements of the peak centroid and its linewidth, is independent of the peak integral and has full width at half-maximum in the neighborhood of 2.5 channels. Copyright © 2005 John Wiley & Sons, Ltd.