Models of DG bands in Raman spectra: effects of basis functions and initial values on the uniqueness and S/N of fits obtained for atmospheric aerosol using automated methods

Soot, coal, graphite and similar black-carbon materials contain layered sheets of sp2-bonded carbon in regions larger than about 4-nm. Such materials typically exhibit Raman spectra (RS) with a peak near 1580 cm^-1 (G band) and, except in the case of graphene or highly ordered graphite, a peak in the range 1300 cm^-1 - 1350 cm^-1 (D band). Here such materials are termed DG-carbon (DGC). To better characterize DGC, researchers often represent the measured RS in sets of multiple D and G bands. Comparisons between ‘best-fit’ D and G bands reported by different researchers are often difficult, partly because initialization and optimization parameters and background removal procedures are inadequately described. We have developed fitting code using the Levenberg-Marquardt (LM) algorithm, as implemented by the LMFIT python package. To illustrate the use of the code, fifteen DGC spectra, and four non-DGC spectra were fitted. RS fits with two Lorentzian functions converged to a single optimized fit for +/- 20 cm^-1 variability of initial peak positions. For five-band fits, in some cases: 1) the final fit varied with the initial values for the functions, and 2) constraints on the peak center locations were required to keep them within expected ranges. When the RS were weaker or there appeared to be large luminescence, the variability in final fits was greater. The code is provided here to facilitate more-quantitative comparisons of results obtained by different researchers. Automated fitting methods are needed for characterizing the large numbers of RS of atmospheric DGC particles measured using Raman hyperspectral imagers.