Efficient model‐free deconvolution of measured femtosecond kinetic data using a genetic algorithm

Abstract

Due to the uncertainty relation between the temporal and spectral widths of a laser pulse, sufficient selectivity in excitation and detection energy does not allow much shorter pulses in a femtosecond pump-probe experiment than about 100 fs. Many ultrafast chemical processes have comparable characteristic times, so the results of these experiments are severely distorted by convolution of the kinetic response function with the pulses used. If we do not know the underlying photochemical and kinetic model, the only way to overcome the limitation in time resolution due to convolution is to perform a model-free deconvolution. Most existing deconvolution methods-even after specifically adapted to femtochemical experimental data (Bá nyá sz Á , Keszei E. Nonparametric deconvolution of femtosecond kinetic data. J. Phys. Chem. A 2006; 110: 6192-6207)-do not provide a smooth deconvolved data set that could be used for unbiased statistical inference. Here, we report an efficient model-free deconvolution method that enhances temporal resolution and improves statistical inference from measured pump-probe data, using a genetic algorithm (GA). The proposed algorithm enables to create a fairly good initial population and uses highly efficient population dynamics to result in individuals who represent excellent solutions of the deconvolution problem without noise amplification, even in the case of a sharp initial steplike rise of the signal. The treatments of both synthetic and experimental data are supporting the outstanding applicability of the proposed deconvolution method.

Deconvolution of femtosecond time‐resolved spectroscopy data in multivariate curve resolution. Application to the characterization of ultrafast photo‐induced intramolecular proton transfer

“…As time resolution is usually a few hundreds of femtoseconds, any transient phenomena occurring faster, or on the same timescale, are affected by the convolution of the true kinetics with the IRF. Direct deconvolution of the kinetic traces can be performed by fitting the model function [6], by model-free methods using inverse filtering based on Fourier transformations [19][20][21] or by methods based on genetic algorithms [22]. We propose here to tackle the problem of deconvolution in the frame of MCR by fitting the time-dependent concentration profiles extracted from the MCR resolution into pure components with a function including the description of the IRF.…”