Observations of trends for singular values and eigenvectors from library searching ultraviolet-visible spectra with spectral subtraction

A promising procedure for computerized library searching and identification of ultraviolet-visible spectra for one- and two-component mixtures is evaluated. The procedure is based on singular value decomposition and utilizes band position and shape. Searching more than one library spectrum at a time is possible. Library spectra which have similar spectra (collinear) with the sample and each other are indicated by two measures. First, a large condition index must be obtained, and second, two or more large variance-decomposition proportions in the same row need to be associated with the large condition index. The searching procedure has a significant degree of differentiation between the actual sample and potential candidates and is compared with the dot product, Euclidean distance and the correlation coefficient.