Self-modeling curve resolution of multi-component vibrational spectroscopic data using automatic band-target entropy minimization (AutoBTEM).

Vibrational spectroscopy is being used routinely to measure multi-component samples and often times these data possess spectroscopic non-idealities such as highly overlapping spectral bands, presence of spectral non-linearities, etc. A multivariate curve resolution algorithm coined as automatic band-target entropy minimization (AutoBTEM) was developed to achieve self-modeling curve resolution of pure component spectra from multi-component vibrational spectroscopic data. This AutoBTEM is a variant extension of the band-target entropy minimization (BTEM) that combines a novel automatic band-targeting numerical strategy with exhaustive BTEM curve resolutions and unsupervised hierarchical clustering analysis in an overall blind search approach. It is also found that the number of components or significant factors and the extent of spectral band shifts can be inferred via the automatic band-targeting computations. The AutoBTEM algorithm is demonstrated herein to be successful when tested on two challenging mixture spectral datasets that are ill-conditioned. One is a two-component mid-infrared FTIR dataset containing spectral non-linearities, and the other is a 10-component Raman dataset with highly overlapping bands from its 10 chemical constituent spectra. The resolved pure component spectra correspond well with reference spectra and have an excellent normalized inner product of above 0.95 upon quantitative comparison.