PNNL

Abstract

The ability of two different approaches in retrieving thermodynamic and spectroscopic properties of chemical species, using Beer’s law, was investigated on a suite of synthetic uv-vis spectra generated at various degrees of random error. Analytical and numerical solutions to direct solutions to Beer’s law were compared to solutions starting from orthogonalised absorbance matrices filtered from random noise by the method of Singular Value Decomposition. Solutions making use of absorbance matrices filtered from random noise were the most successful at retrieving properties of chemical species at larger errors. Repeated automated simulation however indicated such solutions to display a number of occurrences associated to poor fits to the synthetic data while direct solutions to Beer’s law produced more consistent sets of values. A bootstrap statistical analysis of the repeated simulations showed the median quality of the fit (e.g. sum-of-squares of the deviations) to be nonetheless far superior for solutions making use of error/noise filtration than those of direct solutions to Beer’s law. These former solutions are therefore recommended for the extraction of spectroscopic and thermodynamic/kinetic properties of chemical species from spectroscopic data, however not without a rigorous validation of the model.