efficient model-free deconvolution of measured ultrafast kinetic data
DESCRIPTION
Purpose of this work. Relevance of model-free deconvolution. A direct model-free deconvolution method has been implemented , using a genetic algorithm . The method has been thoroughly tested on femtosecond pump-probe transient absorption and fluorescence upconversion data. - PowerPoint PPT PresentationTRANSCRIPT
Efficient Model-free Deconvolutionof Measured Ultrafast Kinetic DataErnő Keszei and Péter Pataki
Department of Physical Chemistry, and Reaction Kinetics Laboratory, Eötvös University (ELTE)H-1518 Budapest, P.O. Box 32, Hungary; e-mail: [email protected]
Purpose of this workA direct model-free deconvolution method has been implemented,
using a genetic algorithm. The method has been thoroughly tested on femtosecond pump-probe transient absorption and
fluorescence upconversion data.
What is deconvolution?To get the undistorted kinetic function o(t) from the detected (convolved) signal i(t), the spread function s(t) should be known and the integral equation (Eq. C) should be solved.
The procedure yielding an estimate of the original o(t) function is called
deconvolution.
Pros and contras of direct deconvolution
Advantages:2
• Does not require any prior knowledge of the kinetic mechanism
• After deconvolving the detected signal,it is much easier to find the appropriate
mechanism
• Instrumental response parameters can be determinedwithout correlation with kinetic and photochemical
parameters
Difficulties of „classical” signal processing methods:2,3,4
• Possible appearence of mathematically acceptable, but physically nonsense solutions as artefacts
• Unavoidable low-frequency wavy oscillations and high frequency noise
Artefacts can largegely be reduced using a genetic algorithm
for deconvolution, due to the highly flexible genetic operators
1 E. Keszei: J. Chemomet., (sent for publication)
2 Á. Bányász, E. Keszei, J. Phys Chem. A 110, 6192 (2006)
Convolution in ultrafast laser chemistryExperiment: femtosecond pump-probe transient absorption
femtosecond fluorescence upconversion
Limitation: due to uncertainty relation: 100 fs ≤ pulse width
Problem: characteristic times of the studied reactions and the temporal width of the laser pulse are
comparableResult of the measurement: a distorted curve (image,
i );the convolution of the kinetic response function (object,
o)and the instrumental distortion function (spread, s) ttstosoi d==
(Eq. C)
Mostly used method to evaluate ultrafast kinetic data:
Reconvolution = least-squares fitting of a suitable model function
o(t) convolved with the distortion function s(t)
Problem: reconvolution requires a particular kinetic model, which is
usually not known prior to kinetic inference.
Model-free deconvolution enables to get undistorted kinetic data without presupposing any particular kinetic model.
Additional advantage: instrumental distortion parameterscan be determined without any correlation with
kinetic and photochemical parameters, as there is no need for
an additional adjustable „zero time” parameter.
Relevance of model-free deconvolution
References
Conclusion and Perspectives
Thomas Gustavsson and Ákos Bányász for detailed experimental data Balaton exchange project 11038YM
OTKA project T 048 725
Acknowledgement
Model-free deconvolution using a genetic algorithm
Tests on simulated kinetic data 1
Kinetic mechanism used to test transient absorption: consecutive two steps reaction:
τ1= 200 fs, τ2= 500 fs; transient absorption with residual bleaching: A = 5, B = 30, C = – 10
CBA 21
Spread: 255 fs fwhm Gaussian
Experimental error: random noise with a normal distribution of 2 % variance of the maximum amplitude.
Test on real-life experimental data 1
Tests were also performed on experimental fluorescence decay data of adenosine monophosphate in aqueous solution obtained by femtosecond fluorescence upconversion (excited at 267 nm, observed at 310 nm; T. Gustavsson and Á. Bányász, unpublished data).
Contraryly to model-free deconvolution via time-domain iterative methods2 and inverse filtering in the frequency domain2-4, use of genetic algorithms results in a distortion-free deconvolved kinetic signal that
• does not have low-frequency wavy behaviour
• correctly reproduces sudden steplike features of kinetic functions
• efficiently damps experimental error without signal distortion
• fully recovers the whole frequency spectrum of the undistorted kinetic function
Experience shows that there is less systematic distortion if nonparametric (model-free) deconvolutionis applied, even in the case if an established photophysical and kinetic model is known and used to perform statistical inference.
The procedure can efficiently be applied to both synthetic and real-life experimental data.
Further work concentrates on improving the quality of deconvolution by applying a genetic algorithm to create the initial population, deconvolving more real-life experimental data, and developing a user-friendly graphical interface to perform the deconvolution.
time domain results
3 Á. Bányász, E. Mátyus, E. Keszei, Rad. Phys. Chem.
72, 235 (2005)4
Á. Bányász, G. Dancs, E. Keszei, Rad. Phys. Chem.
74, 139 (2005)
frequency domain results
temporally widened signal
loweredamplitude
shallower rise and descent
smooth
ened
step
like
jum
ps
=
convolution results in:
To get the undistorted signal, these effects should be inversed
This is achieved by:1) creation of an initial population (whose members represent fairly good solutions)2) fine-tuning of the population members (to best reproduce the detected signal when convolved with the instrumental response)
rise anddecaysteepening
”cutoff” of the first few data
amplitudeincrease
temporalcompression
creation operators to generate one individualof the initial population
randomlygenerated initialpopulation
evolution of the population by crossover random mutation natural selectiongenetic operators
bestdeconvolved (”winner”)
1)
2)
Model function used to test transient fluorescence: biexponential decay with τ1= 100 fs, τ2= 500
fs spread: 310 fs fwhm Gaussian
- - signal processing
frequency domain results
time domain results
Implementation: a package of user defined Matlab functions and
scriptsInput:a project descriptor text file with parameters of the creation of initial population and evolution operators + files containing measured dataOutput:all input parameters in the same format as the project descriptor, measured input data and all
results in a matrix format as a text file+ a four-panel graphical window
10 20 30 40 50 60
0.0
0.5
1.0
1.5
2.0
ampl
itude
channel
– winner
image
– reconvolved
· residuals
10 20 30
0.1
1
10
image ––
spec
tral
am
plitu
de
channel
– winner
– reconvolved
0 20 40 60 80 100 120
-10
-5
0
5
10
15
ampl
itude
a)
channel
o object
– winner
image– reconvolved
· residuals
filterd resultssignal processing results
frequency domain results
time domain results
Code of the deconvolution procedure via genetic algorithm 1
Initial conditions: [A] = 1, [B] = [C] = 0 at t = 0.
Kinetic response function (F):
21
12
21
221
211 1
tt
C
tt
B
t
A
eeeee
lOD
Kinetic response function (F):
21 )1(fluoresc t
1
t
1 ee
I
Code available at http://keszei.chem.elte.hu/GA