Answers are grounded in this paper and cite their sources.
We show that the most significant parameter in digital smoothing of spectra by least squares fitting to a cubic polynomial is the length of the smoothing range, and to avoid loss of resolution we recommend that the smoothing range be chosen approximately 0.7 FWHH (full width at half-height) of the narrowest single lines or components. Multiple use of digital smooths is shown not to be equivalent to a least squares fit of the original data but gives nearly the same response curve in smoothing of Gaussian lines.
No takes yet. Share an insight, caveat, or question.
Edwards et al. (1974) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: