Difference spectra for the reduction of the feasible bands of response profiles of multivariate curve resolution: Comparison of the feasible bands of difference spectra and ordinary data

Abstract

In multivariate curve resolution (MCR) methods, a bilinear decomposition of the experimental data matrix (D) is performed to estimate, for example, the concentration profiles (C) and the spectra (S) using the model equation, D=CST+E, where E is the error or noise matrix. The main limitation of all soft-modeling methods is the fact that there is often no unique solution and sometimes a range of feasible solutions that fit the data equally well (while fulfilling the applied constraint) is obtained. This fact is known as rotational ambiguity. In the present work, the effect of differencing the spectra as a preprocessing method on the feasible bands of response profiles of the MCR results is investigated. The feasible band of several simulated ordinary and difference data sets of pH titration of a diprotic acid as a three-component system was analyzed and compared. Also, a real experimental data were studied by the proposed method. The results showed that by using this pretreatment strategy, the feasible bands of MCR results are reduced. Actually, the reason is the fact of reducing the number of compounds in one and, hence, reducing the compounds overlapping. For analyzing the data with high degree of overlapping, this effect is very obvious. © 2013 John Wiley & Sons, Ltd.