A Generalized NMF-Based Method for Analyzing Time-Resolved Spectroscopic Data

Time-resolved spectroscopy is a widely used tool for the investigation of physical and chemical processes. Analysis of the results is often challenging due to the inherent complexity of the data, encoding the chemical nature and time evolution of multiple species involved in the reaction. Many existing analytical methods are unsatisfactory as they introduce bias by relying on unjustified mathematical or mechanistic assumptions about the studied process. Here, we introduce a generalized analytical strategy based on non-negative matrix factorization. The methodology builds on a bottom-up model-free approach, in which physically grounded mathematical constraints can be introduced by active choice, allowing for an unbiased analysis of complex time series of spectroscopic data. The strength of this strategy is demonstrated by successful deconvolution of synthetic data mimicking different types of chemical reactions and typical challenges encountered in time-resolved Raman spectroscopy.