Pure component spectral reconstruction from mixture data using SVD, global entropy minimization, and simulated annealing. Numerical investigations of admissible objective functions using a synthetic 7-species data set.

Abstract

A combination of singular value decomposition, entropy minimization, and simulated annealing was applied to a synthetic 7-species spectroscopic data set with added white noise. The pure spectra were highly overlapping. Global minima for selected objective functions were obtained for the transformation of the first seven right singular vectors. Simple Shannon type entropy functions were used in the objective functions and realistic physical constraints were imposed in the penalties. It was found that good first approximations for the pure component spectra could be obtained without the use of any a priori information. The present method out performed the two widely used routines, namely Simplisma and OPA-ALS, as well as IPCA. These results indicate that a combination of SVD, entropy minimization, and simulated annealing is a potentially powerful tool for spectral reconstructions from large real experimental systems.