RESUMEN
The use of Principal Components plots in the exploratory investigation of reactions monitored by UV-Vis spectroscopy is described. The effects of different types of pre-processing (raw, mean-centred, and standardised) are illustrated. Four types of plot (scores versus time, loadings versus wavelength, scores versus scores, and loadings versus loadings) are considered. The approach is used to investigate the reaction between phenylhydrazine and benzophenone to give a hydrazone. Observable deviations from ideal behaviour indicate differential crystallisation of the product, and the presence of small quantities of an intermediate during the reaction. Additional information about the reaction is gained by comparing selected components from several batches. PCA is easily performed at- or on-line, and the information gained can be used to help decide upon a suitable harder model for further analysis.
RESUMEN
Several methods are described for determining rate constants for second order reactions of the form U + V --> W using chemometrics and hard modelling to analyse UV absorption spectroscopic data, where all species absorb with comparable concentrations and extinctions. An interesting feature of this type of reaction is that the number of steps in the reaction is less than the number of absorbing species, resulting in a rank-deficient response matrix. This can cause problems when using some of the methods described in the literature. The approaches discussed in the paper depend, in part, on what knowledge is available about the system, including the spectra of the reactants and product, the initial concentrations and the exact kinetics. Sometimes some of this information may not be available or may be hard to estimate. Five groups of methods are discussed, namely use of multiple linear regression to obtain concentration profiles and fit kinetics information, rank augmentation using multiple batch runs, difference spectra based approaches, mixed spectral approaches which treat the reaction as two independent pseudospecies, and principal components regression. Two datasets are simulated, one where the spectra are quite different and the other where the spectrum of one reactant and the product share a high degree of overlap. Three sources of error are considered, namely sampling error, instrumental noise and errors in initial concentrations. The relative merits of each method are discussed.