Analytica Chimica Acta xxx (xxxx) xxx

Contents lists available at ScienceDirect

Analytica Chimica Acta

journal homepage: www.elsevier.com/locate/aca

Resolution of intermediate surface species by combining modulated

infrared spectroscopy and chemometrics

ctor C. Goicoechea a, c, María J. Culzoni a, c, *,
Mirta R. Alcaraz a, c, Alejo Aguirre a, b, He

n E. Collins a, b, **
Sebastia
a
Consejo Nacional de Investigaciones Científicas y T
ecnicas (CONICET), Godoy Cruz 2290 CABA, C1425FQB, Argentina
b
gico para la Industria Química (INTEC), Universidad Nacional del Litoral, CONICET, Güemes 3450, Santa Fe, S3000GLN,
Instituto de Desarrollo Tecnolo
Argentina
c
Laboratorio de Desarrollo Analítico y Quimiometría (LADAQ), Ca
tedra de Química Analítica I, Facultad de Bioquímica y Ciencias Biolo

gicas, Universidad
Nacional del Litoral, Ciudad Universitaria, Santa Fe, S3000GLN, Argentina

h i g h l i g h t s g r a p h i c a l a b s t r a c t

Novel proceedings for the analysis of

modulation excitation spectroscopy
coupled to phase sensitive detection
data.

Chemometric data modeling of
simulated and experimental systems
by MCR-ALS.

Application to the resolution of
adsorption-desorption dynamics of
oxalic acid on titanium dioxide
system.

a r t i c l e i n f o a b s t r a c t

Article history: In this work, aiming at exploiting the essential particularities of modulation excitation spectroscopy
Received 10 May 2018 (MES) coupled to phase sensitive detection (PSD) and chemometrics, a data analysis procedure was
Received in revised form implemented to analyze phase-resolved infrared (IR) data. The fundamental principle of the proceedings
12 October 2018
is the application of successive multivariate curve resolution-alternating least squares (MCR-ALS) reso-
Accepted 24 October 2018
Available online xxx
lutions to MES-PSD IR data. The applicability of the strategy was evaluated in several cases of simulated
data considering the effect of spectral band overlapping and presence of noise. Outcomes related to data-
processing are depicted in detail. As a proof of concept, the data resolution approach was validated by
Keywords:
Modulation excitation spectroscopy (MES)
resolving an experimental real system related to the adsorption-desorption dynamic of oxalic acid on
Phase sensitive detection (PSD) titanium dioxide by in situ IR spectroscopy in attenuated total reflection (ATR) mode. After data reso-
Multivariate curve resolution-alternating lution, different oxalate species were assigned to each spectral band and information about the kinetics
least squares (MCR-ALS) in terms of phase lag was obtained. In the light of the obtained results, this approach is rather appealing
in other research fields, very helpful for the non-chemometric community and foresees manifold ap-
plications, specially, in catalytic system investigations.
© 2018 Elsevier B.V. All rights reserved.

* Corresponding author. Consejo Nacional de Investigaciones Científicas y Te
cnicas (CONICET), Godoy Cruz 2290 CABA, C1425FQB, Argentina.
** Corresponding author. Consejo Nacional de Investigaciones Científicas y Te
cnicas (CONICET), Godoy Cruz 2290 CABA, C1425FQB, Argentina.
E-mail addresses: mculzoni@fbcb.unl.edu.ar (M.J. Culzoni), scollins@santafe-conicet.gov.ar (S.E. Collins).

https://doi.org/10.1016/j.aca.2018.10.052
0003-2670/© 2018 Elsevier B.V. All rights reserved.

Please cite this article as: M.R. Alcaraz et al., Resolution of intermediate surface species by combining modulated infrared spectroscopy and
chemometrics, Analytica Chimica Acta, https://doi.org/10.1016/j.aca.2018.10.052
2 M.R. Alcaraz et al. / Analytica Chimica Acta xxx (xxxx) xxx
1. Introduction
Understanding of reaction mechanisms and structure-activity
relationships under working conditions is of central importance
to perform a rational design of new and optimized catalytic systems
[1,2]. Rigorous operando experimentation requires the simulta-
neous measurement of conversion and selectivity, alongside with
the acquisition of spectroscopy data for the catalyst under steady-
state conditions. Thus, increasing efforts to implement cells/
microreactors to spectroscopically assess the reactant/catalyst
interface, while simultaneously measuring activity and selectivity,
have been carried out by several research groups [3]. However,
when investigating reaction mechanisms under steady-state, it is
virtually impossible to selectively extract signals of the true in-
termediates, spectator species and/or the catalyst itself, e.g., the
background, due to the static nature of the measurements. More-
over, the static signals are usually much more intense than those
from the intermediates, making very difficult the identification and
tracking of intermediates. Then, transient experiments have been
developed and increasingly used for the analysis of reaction in-
termediates by perturbing a catalytic system under working con-
ditions. Thus, accurate background compensation is carried out by
difference spectroscopy. In this matter, two distinct alternatives
have come out for external stimulation, which additionally can
supply kinetic information of the reaction: i) relaxation technique,
e.g., parameter jump, and, more recently, ii) modulation excitation,
e.g., periodic parameter change. The latter can only be applied
when the system responds reversibly to a periodic external exci-
tation, for instance, in the absence of catalyst deactivation.
In modulation excitation spectroscopy (MES), a periodic
perturbation (modulation) of an external parameter is applied to
stimulate the system around the steady-state while time-resolved
spectra are collected [4e6]. Consequently, if a modulated signal is
observed, it can be unambiguously assigned to an active part of the
system that was affected by modulated excitation. Moreover, the
stationary signals will not change. Thus, all the species in the sys-
tem that are affected by the external parameter will change peri-
odically with the same frequency of the perturbation, but with a
phase lag, which implicitly contains the kinetic constant of an
elemental reaction step [4e6]. When small perturbations around
the steady-state are performed, the tracking of the surface reaction
intermediates can be made under real reaction conditions, i.e.,
operando conditions, and quantitative kinetic information can be
obtained. Thus, the MES-operando approach is a powerful tool that
allows the sensitive and selective detection of species directly
involved in a reaction, and the monitoring of their dynamic be-
haviors. Moreover, after the introduction of demodulation or phase
sensitive detection (PSD), which was firstly developed by Baurecht
and Fringeli [4], further insights into spectral details of MES data
could be obtained. PSD analysis transforms spectra from the time-
to the phase-domain and leads to some advantageous properties
that have been extensively discussed in the literature and applied
to diverse experimental systems [4,6e8].
Applications of MES-PSD in heterogeneous catalysis have been
demonstrated for in situ attenuated total reflection-infrared spec-
troscopy (ATR-IR) [9,10], diffuse reflectance-IR Fourier transformed
spectroscopy (DRIFTS) [11e13], X-ray absorption spectroscopy
(XAS) [14,15] and X-ray diffraction (XRD) [8,16]. Additionally, it has
been widely employed to investigate enzymatic reactions [4,5,17]. It
is known that spectroscopic data gathered from different surface
species or reaction sites show partially or highly overlapped bands,
which represent a challenge when individual target signals must be
identified. Then, the combination of MES-operando and signal
processing by PSD with chemometric analysis could improve the
capability to obtain valuable quantitative information [13,15,18].
Please cite this article as: M.R. Alcaraz et al., Resolution of intermediate surface species by combining modulated infrared spectroscopy and
chemometrics, Analytica Chimica Acta, https://doi.org/10.1016/j.aca.2018.10.052
Automated data processing procedures, like least-square fitting of
data with linear combination (LC) of reference spectra and principal
component analysis (PCA) are powerful tools, yet have several
limitations. More precisely, LC requires standards and PCA often
provides mixed components that are not easily interpretable. In
this context, multivariate curve resolution-alternating least squares
(MCR-ALS) is a widespread iterative soft-modeling method that
performs the bilinear decomposition of a data matrix into two
submatrices that comprise chemically meaningful information for
the compounds involved in the system [19,20], i.e., pure spectral
profiles and the varying abundance throughout the process.
To our knowledge, there is only one work reported in the liter-
ature that combines MES-PSD experimentation and data analysis
with MCR-ALS [16] to process in situ X-ray absorption spectroscopy
experiments. This approach allowed extracting detailed informa-
tion about the reversible reductioneoxidation reaction of a
rhenium promoted cobalt FischereTropsch synthesis catalyst.
However, a systematic approach has not been developed yet.
In the present work, MCR-ALS was applied to model MES-PSD
data. Firstly, with the aim of having an insight into the chemo-
metric resolution possibilities based on the properties of the data,
simulated spectra were modeled mimicking an A / B / C kinetic
system. The implemented strategy consists in conducting succes-
sive MCR-ALS analyses to discriminate the individual contributions
of each component in terms of both kinetics (concentrations as a
function of the phase angles) and characteristic spectral bands.
Finally, to prove its capability to solve a complex experimental
system, the developed methodology was applied to the resolution
of the adsorption-desorption kinetics of oxalic acid on titanium
dioxide, monitored by ATR-FTIR spectroscopy.
2. Experimental section
2.1. Simulated datasets
Simulated datasets based on the following consecutive reactions
were created:
k1 k2
A / B* / C (1)
where A is the reactant, B* is an intermediate and C is the product of
the reaction. The simulations were done considering the following
assumptions: a continuous stirred tank reactor (CSTR) with a resi-
dence time of 10 s and reaction constants k1 and k2 equal to 101
s1 and 102 s1, respectively.
The phase lags for this system are:
u
tgð4A  4Ae Þ ¼ (2)
k1 þ t1
tgð4B  4A Þ ¼ u=k (3)
2
tgð4C  4B Þ ¼ u t (4)
where u is the stimulation frequency, ð4A  4Ae Þ is the phase lag of
the A concentration in the reactor with respect to the inlet, (4Be4A)
is the phase lag of B and (4Ce4B) is the phase lag of C. Considering a
squareewave inlet stimulation of the reactant concentration with a
u of 8.3 mHz, the calculated phase lags were 14.6 , 79.2
and 27.6 , for A, B and C concentration, respectively. After
obtaining the response of the species as a function of time, the
corresponding spectra were simulated.
First, a set of simulated time-resolved IR spectra were generated
by applying the modulated stimulation to the bands corresponding
 M.R. Alcaraz et al. / Analytica Chimica Acta xxx (xxxx) xxx 3

to the species A, B and C, i.e., simulated MES experiment. Then,
time-resolved spectra were processed using the PSD method in
order to obtain the phase-resolved spectra (for further details see
Supplementary material, Figs. S1eS4). Therefore, 9 datasets of
simulated spectra were proposed in order to evaluate the capability
of MCR-ALS to model this kind of data (see Fig. 1). In this sense, a
wide variety of cases ranging from the simplest case with no
overlapping bands to the most complex case with highly over-
lapped bands were considered. Moreover, the influence of several
noise levels on the results was studied by adding noise to the
simulated time-resolved IR spectra.
2.2. Apparatus
ATR-FTIR spectra were recorded using a home-built stainless
steel flow-through cell described in detail elsewhere [21], with a
total cell volume of 60 mL. The internal reflection element (IRE) is a
45 trapezoidal ZnSe crystal whose dimensions are
80  10  4 mm, providing 10 internal reflections at the liquid/
crystal interface. A Nicolet 8700 FTIR spectrometer (Thermo-Elec-
tron Scientific Instruments Corp., Madison, USA) with a cryogenic/
liquid-nitrogen cooled MCT detector was employed. Time-resolved
ATR-FTIR spectra were recorded in kinetic and rapid-scan mode at a
resolution of 4 cm1. Liquids, acquired from two separate bottles,
were flowed through the cell using a pulse-free peristaltic pump
(Ismatec ICP4, Wertheim, Germany) placed at the end of the cell. A
pneumatically actuated three-way valve controlled by computer
program allowed switching the stream of the solutions.
2.3. Materials and reagents
Oxalic acid (99.9%, H2C2O4) and potassium chloride ACS (KCl)
were purchased from Carlo Erba RSE (Barcelona, Spain). Hydro-
chloric acid ACS (37%, HCl) and sodium hydroxide (99.9%, NaOH)
were purchased from Merck (Darmstadt, Germany), and titanium
dioxide P-25 (TiO2, 21 nm-particle size) was acquired from Degussa
(Essen, Germany). Double deionized water (18 MU) was used for
the preparation of the solution.
2.4. Experimental dataset: H2C2O4 adsorption on TiO2
Aiming at performing the TiO2 P-25 deposition over a ZnSe
crystal, a suspension of 1.06 mg mL1 TiO2 in water was prepared
and dispersed during 2 h in ultrasonic bath. An aliquot of 1.0 mL
was then dropped over the crystal and dried in oven at 60  C,
during 2 h and using a heating rate of 3  C min1. A homogenous
layer of 2 mm thickness was obtained and characterized by scanning
electron microscopy (SEM) (Fig. S-5, Supplementary material).
Then, the crystal with the deposited layer was placed in the ATR
flow-through cell. Before starting the experiments, deionized water
was flowed over the TiO2 layer to eliminate impurities. Then, a
blank solution, pH ¼ 3.7 and ionic strength (I) ¼ 0.01 mol L1, was

Fig. 1. Simulated phase-resolved spectra datasets based on the consecutive reactions (A / B* / C), spanning the cases from noiseless and no overlapped spectral bands (1) to
highly overlapped spectral band (6) with high noise level (7e9). Three different noise levels (7*, 8* and 9*) were added to the same dataset (6).

Please cite this article as: M.R. Alcaraz et al., Resolution of intermediate surface species by combining modulated infrared spectroscopy and
chemometrics, Analytica Chimica Acta, https://doi.org/10.1016/j.aca.2018.10.052
4 M.R. Alcaraz et al. / Analytica Chimica Acta xxx (xxxx) xxx
flowed during 2 h to stabilize the film. The pH and I were adjusted
with HCl (or NaOH, as appropriate) and KCl, respectively. Once the
film was stabilized, the blank solution was changed to 2 mmol L1
H2C2O4 solution and the temporal spectra evolution was recorded.
After reaching the steady-state, the MES experiments were per-
formed at 20  C by periodically switching the H2C2O4 to blank so-
lutions at a frequency of 8.3 mHz using a flow rate of 2 mL min1.
2.5. Algorithms and software
MCR-ALS algorithms are available online at http://www.mcrals.
info/and were implemented in MATLAB 9.2 (R2017a) [22]. The PSD
method was implemented in GNU Octave, which can be down-
loaded from http://www.gnu.org/software/octave/. Thermo Scien-
tific™ OMNIC™ Specta Software was used to collect and process
the data.
3. Results and discussion
3.1. MCR-ALS resolution of simulated MES data
In a first attempt to resolve the system, the simulated MES data
composed of three species, which have only one spectral maximum
each and vary in time, and the static background contribution was
processed by MCR-ALS in its original form (Fig. 2A, see Figs. S1eS4,
Supplementary material for more details). As suspected, the
impossibility of appropriately modeling the non-negligible contri-
bution of both the background and the data noise deepened the
unsatisfactory resolution of the system. As a result, three different
signals, i.e., two for the three analytes, one of them having two
Fig. 2. Simulated MES data in the time domain (A), spectral (B.1) and temporal (B.2)
profiles reached after applying MCR-ALS to the simulated MES data, and spectral (C.1)
and temporal (C.2) profiles reached by applying MCR-ALS to the simulated MES data
after subtracting the background contribution.
Please cite this article as: M.R. Alcaraz et al., Resolution of intermediate surface species by combining modulated infrared spectroscopy and
chemometrics, Analytica Chimica Acta, https://doi.org/10.1016/j.aca.2018.10.052
spectral maxima, and one for the background, were obtained
(Fig. 2B1). It is pertinent to mention that the application of principal
component analysis (PCA) to the original MES data suggested the
presence of three components, instead of the four considered in the
simulations. The following approach consisted in processing the
MES data free of background. This pre-treatment was achieved by
subtracting the first spectrum, obtained at the initial time, to the
whole data. Although in this case the MCR-ALS resolution provided
three different spectral bands, two of them were unresolved com-
binations of two components (Fig. 2C1). Besides, it is notable how
the noise contribution was magnified due to the background sub-
traction, clearly affecting the spectral profiles. In the following
discussion, the advantages of applying PSD to MES data before
MCR-ALS resolution will be commented. As an advance, the correct
estimation of the components of the system, the elimination of the
contribution of static signals as well as the enhancement of the
signal to noise ratio (SNR) without resorting to data pre-treatments
such as baseline corrections or smoothing, among others, can be
mentioned.
3.2. Data-processing considerations
In the framework of multi-way modeling, it is mandatory to
consider the effect of the noise and the structure of the data. It has
been demonstrated that noise plays an important role in the per-
formance of the data resolution and, in consequence, in the reli-
ability of the extracted information [23]. Going into details related
to noise, it is important to consider not only its level but also its
characteristics. Generally, the noise present in the dataset is
introduced by instrumental interferences or by stochastic varia-
tions of the system. However, when original signal is transformed,
e.g., by derivative filtering, Fourier transform, PSD, among others,
the noise adopts a non-random structure that significantly affects
the multi-way modeling and the proper resolution.
Concurrently, the inner structure of the dataset should be
known in advance, since it would determine the right procedure to
follow for data analysis. As can be seen in Fig. 1, datasets obtained
from MES-PSD contain variables with high degree of collinearity
both in the spectral and the phase lag modes. From the mathe-
matical standpoint, this situation represents a troublesome chal-
lenge in the chemometric resolution due to the fact that, in the
presence of high collinearity, the matrices are close to be singular
and, in consequence, the inverse of (CTC) or (STS) cannot be
computed (see below Eqs. (5) and (6)). Thus, some strategies are
herein proposed in order to manage the collinearity and enable the
chemometric resolution.
Besides, when a qualitative analysis is performed, spectral and
concentration evolution information is generally demanded, which
means achieving results with physicochemical meaning. To tackle
this situation, several algorithms able to obtain profiles with
chemical interpretation can be used. However, they must be care-
fully selected on the basis of the characteristics of the systems. For
example, in the present case, spectral data are combined with
phase lag measurements; therefore, the data may become linearly
dependent [24,25]. Moreover, as stated before, the selected algo-
rithm ought to be capable of facing the challenges of high collin-
earity and non-random noise.
MCR-ALS is a widespread chemometric algorithm that has been
successfully applied in a broad range of spectroscopic in-
vestigations of protein systems, including structural transitions and
evolutionary systems [26e28]. In spite of being widely described in
the literature, some issues about the MCR-ALS resolution will be
emphasized to facilitate the comprehension of the procedure fol-
lowed to data analysis.
The first step in MCR-ALS analysis is the estimation of the
 M.R. Alcaraz et al. / Analytica Chimica Acta xxx (xxxx) xxx
number of spectrally active components that are present in the data
matrix (D), i.e., the rank of the data. Later, initial estimates, either C-
type or S-type, are built to initiate the iterative alternating least
square (ALS) optimization, which stops when convergence is
reached. During the ALS optimization, mathematical or chemical
constraints can be implemented in order to diminish the number of
possible solutions, i.e., to reduce the rotational ambiguity and
assure the uniqueness of the solution, and also to confer physico-
chemical meaning to the obtained profiles [29]. Constraints that are
usually applied are non-negativity and unimodality, among others.
In this step, considering C-type estimates, the optimization is done
by following the equations:
 1  þ
C ¼ DS ST S ¼ D ST (5)
 1
S T ¼ CT C CT D ¼ Cþ D (6)
whereþsuperscript represents the pseudoinverse of the corre-
sponding matrix. Finally, the mathematical solution is given as the
product of the constrained profile matrices, i.e., CST, obtained after
convergence. In the context of this work, it is important to
emphasize that on the contrary to the common usage, C and S
matrices comprise the profiles related to the spectral bands and the
evolution of concentration in phase-domain after PSD treatment
(which implicitly contains kinetic information of the chemical
system) of the components involved in the reaction, respectively.
3.3. Data-processing procedure
Some outcomes related to the chemometric resolution were
found when MES-PSD data was analyzed in its original structure.
First, neither singular value decomposition (SVD) nor PCA were
able to correctly determine the rank of the data, providing a
number of factors lower than the real number of components
present in the system. As it was stressed above, the initialization of
the iterative ALS optimization requires either phase lag or spectral
initial estimates. Even though it is reported that the information
provided as initial estimates may not significantly affect the final
results, it is known that the closer the profiles to the true infor-
mation, the better the chemometric resolution [29]. Thus, in case
no prior information about the system is available, there are
different techniques to aid the estimation of the profiles. Here,
SIMPLISMA and SIMPLISMA-based methodologies could not be
applied since original MES-PSD data contain negative values [30].
Also, when profiles retrieved by evolving factor analysis (EFA) [31]
were used, unsatisfactory results were obtained.
Under this complex scenario, the following procedure, pictori-
ally shown in Fig. 3, was proposed in an attempt to accomplish
chemometric resolution. At this stage, it should be reiterated that
both simulated and experimental data matrices comprise phase lag,
with both positive and negative values, and spectral information.
Hence, negative values were first converted into positive by means
of computing its absolute value (Fig. 3.1b, top). This approach had
two purposes: to turn all the values into positive and to diminish
the linear dependence of the variables, in mathematical terms, by
disruption of the harmonic feature of the phase lag profiles. Then,
each matrix was individually decomposed by MCR-ALS.
It is important to mention that when the absolute value was
used, SVD and PCA were able to provide the right number of factors
(Fig. 3.1b, bottom). Even though this is not a crucial step for the
resolution, when unknown samples are evaluated, such informa-
tion sheds light on the complexity and the composition of the
system. Eventually, two consecutive MCR-ALS analyses were
Please cite this article as: M.R. Alcaraz et al., Resolution of intermediate surface species by combining modulated infrared spectroscopy and
chemometrics, Analytica Chimica Acta, https://doi.org/10.1016/j.aca.2018.10.052
5
performed.
For the first resolution, the initial estimates of the profiles in the
phase lag mode were obtained by estimation of the so-called purest
variables [30] and used to start the ALS optimization (Fig. 3.2a,
bottom), in which non-negativity (in the two modes) and
normalization (in the phase lag mode) constraints were applied.
Once the first resolution was accomplished, the optimized obtained
profiles could be attributed to the spectral bands and the phase lag
behavior of each individual component (Fig. 3.2b). However, the
phase lag profiles did not follow the proper features, because the
data is in absolute value and the harmonic function was not obeyed
(Fig. 3.2b, bottom). With the aim of managing this troublesome
situation, a second MCR-ALS was proposed using the optimized
spectral profiles both as initial estimation and selectivity constraint
in the ALS optimization (Fig. 3.3a, bottom), besides applying
normalization and non-negativity constraints in the phase lag and
spectral mode, respectively. Therefore, as can be seen in Fig. 3.3b,
profiles with physicochemical interpretation were successfully
achieved.
3.4. Application to simulated dataset
A 9-second-order dataset was created considering different
degrees of spectral band overlapping and noise level, starting from
noiseless non-overlapped bands. For the most complex case, i.e.,
highly overlapped spectral bands, random numbers taken from a
Gaussian distribution were added to the original time-resolved
simulated spectra, before applying PSD algorithm. The standard
deviation of the Gaussian noise was considered as 5%, 10% and 50%
of the maximum intensity of the corresponding data. Then, PSD
transform was performed on the second-order data. Each second-
order matrix consisted in 101  361 data points for the first and
the second dimension aiming at mimicking the phase angle and the
wavenumbers, respectively.
Generally, in the MCR-ALS resolution there are many variables
that can be modified, mainly related to the flexibility in building the
initial estimates and the application of a wide variety of constraints,
to drive the algorithm to the final solution. Although this feature
may be advantageous, in other cases could lead to laborious and
non-standardized data analysis procedures. On the contrary to this
common situation, the herein developed approach was applied to
each simulated dataset, going from the simplest to the most com-
plex case, without the necessity of introducing modifications. In
other words, the strategy can be seen as a group of simple and
correlated steps to systematically resolve MES-PSD data. Thus, this
approach is rather appealing in other research fields and very
helpful for the non-chemometric community, in particular for
catalytic and physicochemical researchers. Furthermore, it could be
considered as a substitute for the frequently used deconvolution
methodology, precluding possible ambiguous interpretations.
Fig. 4 depicts the MCR-ALS results for the simulated data
considering three species with different degree of spectral over-
lapping. Besides, a case in which the spectrum of one of the
involved species has two spectral bands was considered (Fig. 4A2).
In all cases, the procedure allowed successfully resolving the vari-
ety of emulated chemical systems.
A similar behavior was observed when the datasets comprising
highly overlapped spectral bands with different degrees of noise
were analyzed. Visual inspection of Fig. 5 indicates that the MCR-
ALS resolutions were satisfactory in all cases. The minimal shifts
among the phase lag profiles, which can be attributed to the dif-
ferences in the noise level, were evaluated in terms of the criterion
of similarity (s12). To compute it, the phase lag profiles retrieved by
MCR-ALS for the cases having 5%, 10% and 50% of noise were
compared with the one obtained in noiseless condition by means of
6 M.R. Alcaraz et al. / Analytica Chimica Acta xxx (xxxx) xxx

Fig. 3. General data-processing proceedings. First, the original MES-PSD data (1.a) is transformed by means of computing the absolute value (1.b). PCA loadings are shown for
original (1.a, bottom) and transformed (1.b, bottom) MES-PSD data. The first MCR-ALS resolution was performed in the transformed MES-PSD data (2.a) by using purest variables as
initial estimates (2.a, bottom) obtaining the spectral (2.b, bottom) and phase lag (2.b, top) profiles. Second MCR-ALS was performed in the original MES-PSD data (3.a) using the
previously optimized spectral profiles (3.a, bottom) to achieve spectral (3.b, bottom) and phase lag (3.b, top) physically recognizable profiles.

the following expression [32]: improved up to 3-order of magnitude [4].

sT1 s2 3.5. Analysis of a real system

s12 ¼
s1 s2
in which s1 and s2 are the phase lag profiles retrieved by MCR-ALS
for each dataset without and with noise addition, respectively. The
value of s12 ranges from zero to one, corresponding to the extreme
situations of no and complete overlapping, respectively. The
calculated s12 figures (ranging from 0.9996 to 1.0000) support the
conclusions previously extracted.
It is worth noticing that PSD is a narrow band technique leading
to dramatic improvement of the signal-to-noise ratio (SNR) by
elimination of most background SNR. Moreover, co-addition of
spectra over multiple periods is usually performed in order to
further enhance the SNR. In this regards, it has been proved that the
SNR in the signals after PSD, i.e., in phase domain, could be
(7)
As a proof of concept, the developed chemometric approach was
applied to an experimental system obtained by studying the
adsorption-desorption dynamic of H2C2O4 on TiO2 by IR spectros-
copy in ATR-FTIR mode.
Fig. 6A and 6 B show the ATR spectra in the time-domain and
phase-domain after applying PSD analysis, respectively. Fig. 6C
shows the resolved bands and the corresponding phase lag profiles
retrieved by MCR-ALS. After chemometric decomposition, different
oxalate species could be tentatively assigned to each spectral pro-
file. In the carbonyl spectral region (1750-1690 cm1), a broad band
constituted by signals from different oxalate species adsorbed on
the TiO2 surface can be identified. In this region, one of the com-
ponents features a maximum band at 1728 cm1, with a D4 ¼ 14 ,

Please cite this article as: M.R. Alcaraz et al., Resolution of intermediate surface species by combining modulated infrared spectroscopy and
chemometrics, Analytica Chimica Acta, https://doi.org/10.1016/j.aca.2018.10.052
 M.R. Alcaraz et al. / Analytica Chimica Acta xxx (xxxx) xxx
Fig. 4. Spectral (A) and phase lag (B) profiles achieved after application of the che-
mometric procedure to the simulated noiseless phase-resolved spectra, spanning the
cases from noiseless and no overlapped spectral bands (1) to highly overlapped
spectral band (6).
and is assigned to species I, which represents monodentate and
monoprotonated oxalate. A second component shows a strong
band at 1726 cm1 and a weaker band at 1703 cm1, with a phase
lag of 42 , which is associated to species II, i.e., bidentate oxalate
with the CeC bond perpendicular to the surface. A third profile that
presents a maximum band at 1716 cm1 and an incipient shoulder
at 1699 cm1 (D4 ¼ 45 ) is attributed to species III, whose rep-
resents monodentate and monoprotonated oxalate with partici-
pation of hydrogen bond. Interestingly, a fourth band with a small
contribution at 1714 cm1 (D4 ¼ 70 ) was obtained and identified
as species IV, which represents bidentate with the CeC bond par-
allel to the surface oxalate species. All these observations are in
accordance with previously reported works, where ATR-FTIR in-
vestigations of H2C2O4 adsorption on Degussa P-25 TiO2 were
performed [33,34].
On the other hand, the second spectral region, between
1500 cm1 and 1700 cm1, shows a band arising from the asym-
metric stretching modes of CO 1
2 [ya(C-O)] (1610 cm ), which
overlaps with HOH-bending band [d(OH)] of the water
(~1645 cm1). Consequently, the assignment and identification of
species is rather challenging due to the strong and broad band of
the water. However, the resolved MCR-ALS profiles can be tenta-
tively attributed to changes in the surface hydration and/or
Please cite this article as: M.R. Alcaraz et al., Resolution of intermediate surface species by combining modulated infrared spectroscopy and
chemometrics, Analytica Chimica Acta, https://doi.org/10.1016/j.aca.2018.10.052
7
Fig. 5. Spectral (A) and phase lag (B) profiles achieved after application of the che-
mometric procedure to the simulated phase-resolved spectra with increasing noise
level. Different tones of green were used to show the slight differences between phase
lag profiles of the same species. (For interpretation of the references to colour in this
figure legend, the reader is referred to the Web version of this article.)
intensity changes due to hydrogen bonding on the charged surface
(for detailed information see Fig. S-6, Supplementary material)
[33]. Eventually, it is important to highlight that MCR-ALS also
provides profiles with kinetic information of the individual spe-
cies involved in the reaction in terms of the phase lag. In conse-
quence, for further evaluation, the specific kinetic figures can be
computed using the information comprised in the profiles.
However, in light of obtaining unequivocal results, additional
experiments need to be performed; for instance, MES experi-
ments at different frequencies of perturbation.
4. Conclusions
A novel and systematic chemometric procedure was imple-
mented to model MES-PSD IR data. The strategy comprised suc-
cessive MCR-ALS steps and allowed extracting spectral and kinetic
information of active species. The proceedings was developed and
optimized on simulated data, spanning the cases from noiseless
and non-overlapped spectral bands to highly overlapped spectral
bands with high noise level. One of the important accomplishments
reached from this analysis was that, even though MCR-ALS tends to
be operator-dependent, the presented strategy can be applied
without the necessity of introducing modifications. Favorable re-
sults were achieved and drawbacks of presence of non-random
noise and high-collinearity were successfully overcome.
8 M.R. Alcaraz et al. / Analytica Chimica Acta xxx (xxxx) xxx
Fig. 6. ATR-FTIR spectra (A) in the time-domain and (B) phase-domain after PSD for
the adsorption-desorption reaction of H2C2O4 on TiO2. Phase lag (C.3) and spectral
profiles reached after applying the chemometric procedure to the region 1 (C.1), where
species I (pink line, 1728 cm1, D4 ¼ 14 ), species II (green line, 1726/1703 cm1,
D4 ¼ 42 ), species III (orange line, 1716/1699 cm1, D4 ¼ 45 ) and species IV (red
line, 1714 cm1, D4 ¼ 70 ) are observable, and the region 2 (C.2) with the contribu-
tion of HOH-bending band of the water (blue line, ~1645 cm1) and C  O asymmetric
stretching mode of CO 1
2 (dark grey line, 1610 cm ). (For interpretation of the refer-
ences to colour in this figure legend, the reader is referred to the Web version of this
article.)
To prove the efficiency of the proposed strategy, experimental
data gathered from a study of the adsorption-desorption dynamic
of oxalic acid on titanium dioxide by in situ IR spectroscopy in ATR
mode was resolved. By application of the introduced procedure, the
chemometric resolution allowed identifying the individual species
that are involved in the system, as well as obtaining kinetic infor-
mation in terms of phase lag, which could be used to compute the
specific kinetic figures of each active species. In this sense, four
spectral bands were recognized in the carbonyl spectral region
(1750-1690 cm1) and were attributed to different oxalate species
adsorbed on the TiO2 surface. Besides, two bands could be identi-
fied in the region between 1500 cm1 and 1700 cm1, which could
be tentatively attributed to changes in the surface hydration and/or
intensity changes due to hydrogen bond on the charged surface.
The obtained results demonstrated that the presented data
resolution approach is an effective, robust and powerful tool to
obtain relevant information of a complex catalytic system.
Please cite this article as: M.R. Alcaraz et al., Resolution of intermediate surface species by combining modulated infrared spectroscopy and
chemometrics, Analytica Chimica Acta, https://doi.org/10.1016/j.aca.2018.10.052
Additionally, it needs to be emphasized that the depicted pro-
ceedings can be considered as a recipe to follow, aiding to expand
its application to the non-chemometric community. Hence, this
approach is rather appealing for applications in other research
fields, potentially very helpful for the scientific community and
foresees manifold applications, specially, in catalytic system
investigations.
Acknowledgment
The authors are grateful to Universidad Nacional del Litoral
(Project CAI þ D 2016-50120150100110LI), CONICET (Consejo
Nacional de Investigaciones Científicas y Te
cnicas, Project PIP-2014
Nº 086 and PIP-2015 Nº 0111) and ANPCyT (Agencia Nacional de
Promocio
n Científica y Tecnolo

gica, Projects PICT 2014-0347 and
PICT-2014-0497) for financial support. MRA and AA thank CONICET
for their fellowships.
Appendix A. Supplementary data
Supplementary data to this article can be found online at
https://doi.org/10.1016/j.aca.2018.10.052.
References
[1] M.A. Ban ~ ares, M.O. Guerrero-Perez, J.L.G. Fierro, G.G. Cortez, Raman spec-
troscopy during catalytic operations with on-line activity measurement
(operando spectroscopy): a method for understanding the active centres of
cations supported on porous materials, J. Mater. Chem. 12 (2002) 3337e3342.
[2] F.C. Meunier, The design and testing of kinetically-appropriate operando
spectroscopic cells for investigating heterogeneous catalytic reactions, Chem.
Soc. Rev. 39 (2010) 4602e4614.
[3] Y. Denkwitz, B. Schumacher, G. Ku
, R.J. Behm, Activity, stability, and
cerova
deactivation behavior of supported Au/TiO2 catalysts in the CO oxidation and
preferential CO oxidation reaction at elevated temperatures, J. Catal. 267
(2009) 78e88.
[4] D. Baurecht, U.P. Fringeli, Quantitative modulated excitation Fourier transform
infrared spectroscopy, Rev. Sci. Instrum. 72 (2001) 3782e3792.
[5] D. Baurecht, I. Porth, U.P. Fringeli, A new method of phase sensitive detection
in modulation spectroscopy applied to temperature induced folding and
unfolding of RNase A, Vib. Spectrosc. 30 (2002) 85e92.
[6] A. Urakawa, T. Bürgi, A. Baiker, Sensitivity enhancement and dynamic
behavior analysis by modulation excitation spectroscopy: principle and
application in heterogeneous catalysis, Chem. Eng. Sci. 63 (2008) 4902e4909.
[7] A. Urakawa, T. Bürgi, A. Baiker, Kinetic analysis using square-wave stimulation
in modulation excitation spectroscopy: mixing property of a flow-through
PM-IRRAS cell, Chem. Phys. 324 (2006) 653e658.
[8] V. Marchionni, D. Ferri, O. Kro € cher, A. Wokaun, Increasing the sensitivity to
short-lived species in a modulated excitation experiment, Anal. Chem. 89
(2017) 5801e5809.
[9] P. Müller, I. Hermans, Applications of modulation excitation spectroscopy in
heterogeneous catalysis, Ind. Eng. Chem. Res. 56 (2017) 1123e1136.
[10] A. Aguirre, A.L. Bonivardi, S.R. Matkovic, L.E. Briand, S.E. Collins, ATR-FTIR
study of the decomposition of acetic anhydride on fosfotungstic
wellsedawson heteropoly acid using concentration-modulation excitation
spectroscopy, Top. Catal. 54 (2011) 229e235.
[11] A. Aguirre, S.E. Collins, Selective detection of reaction intermediates using
concentration-modulation excitation DRIFT spectroscopy, Catal. Today 205
(2013) 34e40.
[12] J. Vecchietti, A. Bonivardi, W. Xu, D. Stacchiola, J.J. Delgado, M. Calatayud,
S.E. Collins, Understanding the role of oxygen vacancies in the water gas shift
reaction on ceria-supported platinum catalysts, ACS Catal. 4 (2014)
2088e2096.
[13] L. Lemaitre, A. Berliet, S. Maury, M. Rivallan, Surface modifications of cobalt
Fischer Tropsch catalyst followed by operando DRIFT and chemometrics,
Catal. Today 283 (2017) 172e175.
[14] D. Ferri, M.S. Kumar, R. Wirz, A. Eyssler, O. Korsak, P. Hug, A. Weidenkaff,
M.A. Newton, First steps in combining modulation excitation spectroscopy
with synchronous dispersive EXAFS/DRIFTS/mass spectrometry for in situ
time resolved study of heterogeneous catalysts, Phys. Chem. Chem. Phys. 12
(2010) 5634e5646.
€nig, J.A. van Bokhoven, T.J. Schildhauer, M. Nachtegaal, Quantitative
[15] C.F.J. Ko
analysis of modulated excitation X-ray absorption spectra: enhanced preci-
sion of EXAFS fitting, J. Phys. Chem. C 116 (2012) 19857e19866.
[16] A. Voronov, A. Urakawa, W.v. Beek, N.E. Tsakoumis, H. Emerich, M. Rønning,
Multivariate curve resolution applied to in situ X-ray absorption spectroscopy
data: an efficient tool for data processing and analysis, Anal. Chim. Acta 840
 M.R. Alcaraz et al. / Analytica Chimica Acta xxx (xxxx) xxx
(2014) 20e27.
[17] M.V. Toledo, C.R. Llerena Suster, M.L. Ferreira, S.E. Collins, L.E. Briand, Mo-
lecular recognition of an acyl-enzyme intermediate on the lipase B from
Candida Antarctica, Catal. Sci. Technol. 7 (2017) 1953e1964.
[18] M. Garland, Combining operando spectroscopy with experimental design,
signal processing and advanced chemometrics: state-of-the-art and a glimpse
of the future, Catal. Today 155 (2010) 266e270.
[19] R. Tauler, Multivariate curve resolution applied to secon order data, Chemo-
metr. Intell. Lab. Syst. 30 (1995) 133e146.
[20] C. Ruckebusch, L. Blanchet, Multivariate curve resolution: a review of
advanced and tailored applications and challenges, Anal. Chim. Acta 765
(2013) 28e36.
[21] A. Aguirre, C.L.A. Berli, S.E. Collins, ATR-FTIR spectrokinetic analysis of the CO
adsorption and oxidation at water/platinum interface, Catal. Today 283 (2017)
127e133.
[22] Matlab, The MathWorks Inc., Natick, MA, USA, 2017.
[23] M. Ahmadvand, H. Parastar, H. Sereshti, A. Olivieri, R. Tauler, A systematic
study on the effect of noise and shift on multivariate figures of merit of
second-order calibration algorithms, Anal. Chim. Acta 952 (2017) 18e31.
[24] A.C. Olivieri, G.M. Escandar, Practical Three-way Calibration, Elsevier, Wal-
tham, USA, 2014.
[25] G.M. Escandar, A.C. Olivieri, N.M. Faber, H.C. Goicoechea, A. Mun ~ oz de la Pen
~ a,
R.J. Poppi, Second- and third-order multivariate calibration: data, algorithms
and applications, TrAC Trends Anal. Chem. (Reference Ed.) 26 (2007)
752e765.
[26] M.R. Alcara
z, A. Schwaighofer, H. Goicoechea, B. Lendl, Application of MCR-
ALS to reveal intermediate conformations in the thermally induced a-b
Please cite this article as: M.R. Alcaraz et al., Resolution of intermediate surface species by combining modulated infrared spectroscopy and
chemometrics, Analytica Chimica Acta, https://doi.org/10.1016/j.aca.2018.10.052
9
transition of poly-l-lysine monitored by FT-IR spectroscopy, Spectrochim. Acta
Mol. Biomol. Spectrosc. 185 (2017) 304e309.
[27] A. Borges, R. Tauler, A.d. Juan, Application of multivariate curve resolution to
the temperature-induced unfolding of a-chymotrypsin, Anal. Chim. Acta 544
(2005) 159e166.
[28] S. Navea, A. de Juan, R. Tauler, Modeling temperature-dependent protein
structural transitions by combined near-IR and mid-IR spectroscopies and
multivariate curve resolition, Anal. Chem. 75 (2003) 5592e5601.
[29] A. de Juan, S.C. Rutan, R. Tauler, 2.19 - two-way data analysis: multivariate
curve resolution e iterative resolution methods, in: S.D. Brown, R. Tauler,
B. Walczak (Eds.), Comprehensive Chemometrics, Elsevier, Oxford, 2009,
pp. 325e344.
[30] W. Windig, J. Guilment, Interactive self-modeling mixture analysis, Anal.
Chem. 63 (1991) 1425e1432.
[31] M. Maeder, A. de Juan, 2.16 - two-way data analysis: evolving factor Analysis,
in: S.D. Brown, R. Tauler, B. Walczak (Eds.), Comprehensive Chemometrics,
Elsevier, Oxford, 2009, pp. 261e274.
[32] V. Go
mez, M. Miro
, M.P. Callao, V. Cerda, Coupling of sequential injection
chromatography with multivariate curve resolution-alternating least-squares
for enhancement of peak capacity, Anal. Chem. 79 (2007) 7767e7774.
[33] S.J. Hug, B. Sulzberger, In situ fourier transform infrared spectroscopic evi-
dence for the formation of several different surface complexes of oxalate on
TiO2 in the aqueous phase, Langmuir 10 (1994) 3587e3597.
[34] A.D. Weisz, L. Garcıa Rodenas, P.J. Morando, A.E. Regazzoni, M.A. Blesa, FTIR
study of the adsorption of single pollutants and mixtures of pollutants onto
titanium dioxide in water: oxalic and salicylic acids, Catal. Today 76 (2002)
103e112.