ANALYSIS OF RESIDUALS IN FITTING SPECTROSCOPIC DATA

We show that a detailed study of the residuals of a fitting to data may bring some statistical information about the numerical pertinence of the theoretical model and about the quality of its optimization. It also provides an interesting way to obtain the pure experimental error with a slightly wrong model. The theory of the method is presented in the framework of least-squares fittings of linear (or almost linear) models. Next, the analysis of results from fittings to a set of spectroscopic measurements on niobium nitride, NbN, gives rise to a discussion about the interest of the method and about its limits.