2DCOR-GC: An application of the generalized two-dimensional correlation analysis as a route to optimization of continuous flow supercritical fluid reactions

Article abstract of DOI:10.1021/ac049360l

A new approach for optimization and monitoring of continuous reactions has been developed using 2D correlation methods for the analysis of GC data (2DCOR-GC). 2DCOR-GC maps are obtained following perturbation of the system that allow the effect of changing reaction parameters such as time, temperature, pressure, or concentration to be both monitored and sequenced with regard to changes in the raw GC data. In this paper, we describe the application of the 2DCOR-GC technique to monitoring the reverse water-gas shift reaction in scCO₂. 2DCOR-GC is combined with FT-IR data to validate the methodology. We also report the application of 2DCOR-GC to probe the mechanism of the alkylation of m-cresol with isopropyl alcohol in scCO₂ using Nafion SAC-13 as the catalyst. These results identify coeluting peaks that could easily be missed without exhaustive method development.

Full text of DOI:10.1021/ac049360l

Anal. Chem. 2004, 76, 6197-6206
2 DCOR-GC: An Ap p lic a t io n o f t h e Ge n e ra lize d
Tw o -Dim e n s io n a l Co rre la t io n An a lys is a s a Ro u t e
To Op t im iza t io n o f Co n t in u o u s Flo w S u p e rc rit ic a l
Flu id Re a c t io n s
J a s on R. Hyde ,*,† Ric ha rd A. Bourne ,† Is a o Noda ,‡ Phil Ste phe ns on,† a nd Ma rtyn Polia koff†
School of Chemistry, University of Nottingham, University Park, Nottingham, U.K., NG7 2RD, and
Procter & Gamble Company, 8611 Beckett Road, West Chester, Ohio 45069
tion was previously available. It has the advantage that, even when
there is little understanding of a spectral region, changes can be
correlated with features in better-understood regions of the
spectrum, thus revealing more information about an overall
system. Perturbations have included changes in concentration5
(which is particularly applicable in cases where the Beer-Lambert
law is not obeyed, as a linear response to the perturbation is not
required) and changes in temperature6,7 and pressure,8,9 which
are commonly measured as a function of time. The application of
2D correlation analysis is not limited to different regions of the
same spectral observation. Heterospectral correlation of differing
measurements is relatively commonplace and has found applica-
tions in various combinations of techniques, such as IR-Raman
and IR-NIR. This type of analysis holds the advantage that IR is
well understood, and correlations with the Raman or NIR help
explain and expand the understanding of these other spectral
observations.10-12
A new approach for optimization and monitoring of
continuous reactions has been developed using 2 D cor-
relation methods for the analysis of GC data (2DCOR-GC).
2 DCOR-GC maps are obtained following perturbation of
the system that allow the effect of changing reaction
parameters such as time, temperature, pressure, or
concentration to be both monitored and sequenced with
regard to changes in the raw GC data. In this paper, we
describe the application of the 2 DCOR-GC technique to
monitoring the reverse water-gas shift reaction in scCO2 .
2 DCOR-GC is combined with FT-IR data to validate the
methodology. We also report the application of 2 DCOR-
GC to probe the mechanism of the alkylation of m -cresol
with isopropyl alcohol in scCO2 using Nafion SAC-1 3 as
the catalyst. These results identify coeluting peaks that
could easily be missed without exhaustive method devel-
opment.
In recent years, the 2D correlation analysis has been extended
in its application to include nonspectroscopic analytical techniques.
Two-dimensional correlation gel permeation chromatography (2D
GPC) has recently gained much attention,13-17 where the 2D
correlation analysis has been applied to help understand the
changes in polymer structure during polymerization processes.
For example, the 2D-GPC technique has yielded information about
the growth of SiO2 during a sol-gel production.15
Two-dimensional (2D) correlation spectroscopy was first
introduced into NMR in the late 1960s and has since flourished
as an analytical technique. The application of the correlation
approach has led to the development of familiar experiments such
as COSY, NOSY, and DOSY, which reveal substantially more
information than 1D experiments.1,2
The concept of a generalized 2D correlation was first extended
to IR spectroscopy in 1993 by Noda,3 where it provides further
insight into spectral changes with respect to an external perturba-
tion. The correlation function as described by Noda is based upon
a mathematical background,4 where the only prerequisite is the
subtraction of reference data from a set of spectral data collected
sequentially under an external perturbation. Perturbation-based
2D IR studies have elucidated many systems where little informa-
(5) Marrcott, C.; Noda, I.; Dowrey, A. E. Anal. Chim. Acta 1 9 9 1 , 1, 131-143.
(6) Noda, I.; Liu, C. Y.; Ozaki, Y.; Czarnecki, M. A. J. Phys. Chem. 1 9 9 5 , 99,
3068.
(7) Ozaki, Y.; Liu, C. Y.; Noda, I. Appl. Spectrosc. 1 9 9 7 , 51, 526.
(8) Noda, I.; Story, G. M.; Marcott, C. Vib. Spectrosc. 1 9 9 9 , 19, 461.
(9) Smeller, L.; Heremans, K. Vib. Spectrosc. 1 9 9 9 , 19, 375.
(10) Noda, I.; Liu, C. Y.; Ozaki, Y. J. Phys. Chem. 1 9 9 6 , 100, 8674.
(11) Noda, I. Chemtracts-Macromol. Chem. 1 9 9 0 , 1, 89.
(12) Ozaki, Y.; Noda, I. Two-Dimensional Correlation Spectroscopy; American
Institute of Physics: Melville, NY, 2000.
(13) Izawa, K.; Ogasawara, T.; Masuda, H.; Okabayashi, H.; O’Connor, C. J.; Noda,
I. PCCP Phys. Chem. Chem. Phys. 2 0 0 2 , 4, 1053-1061.
(14) Izawa, K.; Ogasawara, T.; Masuda, H.; Okabayashi, H.; O’Connor, C. J.; Noda,
I. Phys. Chem. Commun. 2 0 0 2 , 12-16.
* To whom correspondence should be addressed. Fax: +44 115 9513058;
Tel: +44 115 9513386; E-mail: jason.hyde@nottingham.ac.uk.
† University of Nottingham.
‡ Procter & Gamble Co.
(1) Ernst, R.; Bodenhausen, G.; Wakaun, A. Principles of NMR in One and Two
Dimensions; Oxford Univeristy Press: Oxford, U.K., 1987.
(2) Bax, A. Two-Dimensional Nuclear Magnetic Resonance in Liquids, Delft
University Press: Boston, 1982.
(3) Noda, I. Appl. Spectrosc. 1 9 9 3 , 47, 1329-1336.
(4) Ozaki, Y.; Sasic, S.; Tanaka, T.; Noda, I. Bull. Chem. Soc. Jpn. 2 0 0 1 , 74,
1-17.
(15) Izawa, K.; Ogasawara, T.; Masuda, H.; Okabayashi, H.; Noda, I. Macromol-
ecules 2 0 0 2 , 35, 92-96.
(16) Izawa, K.; Ogasawara, T.; Masuda, H.; Okabayashi, H.; O’Connor, C. J.; Noda,
I. J. Phys. Chem. B 2 0 0 2 , 106, 2867-2874.
(17) Izawa, K.; Ogasawara, T.; Masuda, H.; Okabayashi, H.; Noda, I. Phys. Chem.
Commun. 2 0 0 1 , article no. 12.
10.1021/ac049360l CCC: $27.50 © 2004 American Chemical Society
Published on Web 09/23/2004
Analytical Chemistry, Vol. 76, No. 21, November 1, 2004 6197

The principal method for displaying 2D correlated chromato-
grams is to specify two orthogonal axes that define the correlation
plane.13,16 In 2D correlation analysis (COR)-GC, the axes represent
the chromatograms collected, where the variable is the retention
or elution time, τ, of each of the individual compounds. The
correlation intensity is plotted on a third axis, normal to this plane,
and is displayed as contours. This diagram is known as a
“correlation map” and simplifies a series of complex chromato-
grams revealing overlapping or coeluting peaks by the extension
into the second dimension and also identifies peaks that are
selectively coupled by the mechanism of a reaction.
To achieve a change in intensity of a chromatographic peak,
a perturbation has to be applied to the studied system. This
perturbation can be any reasonable systematic physical or chemi-
cal change. The chromatograms are collected in the normal
quantitative manner at intervals as the system is perturbed. The
change of intensity of the chromatographic peaks, as a function
of this perturbation, is known as the dynamic chromatogram. (The
term dynamic chromatogram in the context of this paper refers
to the changes in intensity of a chromatographic peak as a function
of the perturbation and should not be confused with a “dynamic”
system, such as the interconversion of stereoisomers in chroma-
tography.) Thus, if a reaction were monitored by GC while the
temperature of the system was increased, the dynamic chromato-
gram represents changes from low to high temperatures. The
corresponding correlation maps also reflect this perturbation, and
correlations between eluting compounds will be related or coupled
by this increase in temperature.
synchronous and asynchronous 2D correlation intensities. Y˜1(ω) and
Y˜2 (ω) are the forward Fourier transform and the conjugate of
*
the Fourier transform of the chromatographic intensity variations
˜y(τ,t) observed at retention times τ1 and τ2, which are calculated
with respect to the variable t. The Fourier angle, ω, is also traced
along the external variable, t.
Fortunately, such complex calculations are not necessary in
most cases, provided that the chromatograms are collected at m
equally spaced intervals, because then the correlation intensities
can be directly calculated by
m
1
Φ(τ1,τ2) )
˜yj(τ1)˜yj(τ2)
∑
m-1
j)1
m
m
1
Ψ(τ1,τ2) )
˜yj(τ1) Njk˜yk(τ2)
m-1 ∑ ∑
j)1
k)1
where Njk is the Hilbert-Noda transform matrix.
0
if j ) k
otherwise
Njk )
{
1/ π(k - j)
These direct calculations can be solved by a number of
mathematical packages such as Mathematica, Maple, or MatLab.
In such cases, the correlation planes are plotted as a series of
contour lines of equal correlation intensity.
Synchronous correlation maps are an indication of the degree
of similarity or coherence between two chromatographic signals
τ1 and τ2, and represent the simultaneous changes of these signals.
Synchronous correlation maps are necessarily symmetric about
the diagonal axis, τ1 ) τ1. Since τ1 will always change simulta-
neously with respect to itself, any change in the dynamic
chromatogram will exhibit a signal at coordinate Φ(τ1, τ1). Such
signals are termed “autopeaks” and are always positive. Autopeaks
show the correlation intensity of an individual signal with respect
to the particular external perturbation. Thus, if chromatographic
peak intensity changes strongly over the course of the perturba-
tion, it will exhibit a strong autopeak. Peaks that do not undergo
strong changes of intensity display weaker autopeaks and, in some
cases, may not exhibit any autopeak at all. This is an advantage
in correlation chromatography, since peaks that do not change
in intensity, never appear on the correlation map. Thus, the
inclusion of an internal standard will not interfere with the
correlation analysis in any way, and it can be used to align data
prior to the calculation of the dynamic chromatograms.
The variation of chromatographic signal intensity, y, can be
thought of quite simply as y(τ,t), where t is the tth step of the
perturbation. This encompasses the perturbation, t, measured
between Tmin and Tmax at each chromatographic index, τ, e.g.,
retention or elution time. In a similar fashion, the dynamic
chromatogram can be defined as ˜y(τ,t), but it is more formally
defined as
for Tmin e t e Tmax
y(τ,t) - jy(τ)
0
˜y(τ,t) )
{
otherwise
where jy(τ), the reference chromatogram, is usually taken to be
the mean-averaged chromatogram collected between Tmin and
T
max, again defined more formally as
Tmax
1
jy(τ) )
y(τ,t) dt
∫
Tmin
Tmax - Tmin
The reference chromatogram can in fact be any single
Cross-peaks occur at coordinates Φ(τ1, τ2) and are due to
simultaneous or coincidental changes at two different retention
or elution times. These signals can be either positive or negative
and represent changes in the same or opposite directions of
intensity, respectively. Thus, if Φ(τ1, τ2) is positive, both τ1 and τ2
are increasing or decreasing at the same time; if Φ(τ1, τ2) is
negative, one of the peaks is increasing while the other is
decreasing simultaneously. Zero correlation between τ1 and τ2
indicates that the two signals are independent, changing randomly,
or that one of the signals does not vary.
chromatogram in the series, for example, the first or last, or it
can even be set to zero, in which case the dynamic chromatograms
are the same as the raw chromatographic data. Once a series of
dynamic chromatograms have been collected, the generalized
cross-correlation function can be applied. This function takes the
form
Φ(τ1,τ2) + iΨ(τ1,τ2) )
∞Y˜ (ω)Y *(ω) dω
1
∫
1
2
0
π(Tmax - Tmin
)
Φ(τ1, τ2) and Ψ(τ1, τ2), the real and imaginary components of
the 2D correlation function, are known respectively as the
Asynchronous correlation maps are antisymmetric about the
τ1 ) τ1 diagonal and show the degree of dissimilarity or out-of-
6198 Analytical Chemistry, Vol. 76, No. 21, November 1, 2004

phase changes between τ1 and τ2. Since an individual signal cannot
change in either of these two manners, asynchronous correlation
maps cannot contain autopeaks. Instead, the asynchronous maps
only show off diagonal peaks, which must be connected via the
τ1 ) τ1 diagonal to construct correlation squares between τ1 and
τ2. Again, these peaks may be positive or negative, but in this
case, a positive peak is obtained if an intensity change in τ1 occurs
before an intensity change in τ2; negative peaks are delayed
responses, and thus an intensity change in τ1 occurs after an
intensity change in τ2. This rule, however, is reversed when the
synchronous correlation intensity at the corresponding coordinate
is negative; i.e., Φ(τ1, τ2) < 0. Collectively, these relations are
known as Noda’s rules.3
In the context of 2DCOR-GC, coeluting compounds exhibit τ1
≈ τ2 and thus may give rise to a single chromatographic peak.
However, provided that the two compounds exhibit different
dynamic behavior, an asynchronous relationship can still arise,
even if the two signals are moving in the same direction. This
point is of great importance to cross-correlation chromatography,
since rarely do two separate compounds chromatographically
behave identically, and thus identification of coelution can be
simplified enormously, as demonstrated in this paper.
In this paper, we describe and validate the application of the
2D correlation analysis as directed toward gas chromatography,
with the aim both to optimize and to understand chemical
reactions performed in supercritical fluids (SCFs). An SCF is
defined as any substance that is above its critical temperature and
pressure (but below the pressure required to form a solid) and
close to its critical density.18 SCFs exhibit a unique combination
of gaslike and liquidlike properties.19 This combination of unique
properties enables an SCF to dissolve both liquids and solids,
where the solubility is generally pressure dependent. This
phenomenon has been exploited to reduce mass transport limita-
tions present during hydrogenation reactions.20,21 More recently
it was used to tune reaction conversion and selectivity in a wide
range of synthetically valuable transformations.22,23 However, the
interrelation of SCF reaction parameters, conversion, and selectiv-
ity are not always simple, and much research is needed to
understand the role of phase behavior and catalyst activity.24 Some
of these research activities involve continuous fixed-bed catalytic
reactors interfaced to on-line GC analysis.25 Such apparatus can
yield so much data that information about the reaction can only
be extracted with difficulty. It is our intent to tackle this
superabundance of data that leads to the development of 2DCOR-
GC.
useful information about a real reaction under SCF conditions.
Therefore, we first validate 2DCOR-GC by analyzing the relatively
simple supercritical reverse water-gas shift reaction (RWGSR),
shown by eq 1, and then apply the same technique to the
CO2 + H2 h CO + H2O
(1)
continuous alkylation of m-cresol over a solid acid catalyst in
supercritical carbon dioxide (scCO2).
The RWGSR is a well-known reversible reaction in equilibrium
among CO, H2O, CO2, and H2. It is particularly relevant to
reactions in SCFs, because it is a potential side reaction in the
hydrogenation of organic compounds in scCO2 where it generates
CO, a toxic gas, which can poison the catalyst. The advantage of
choosing a reaction like the RWGSR from the point of view of
validating 2DCOR-GC is that there are relatively few simple
analytical instruments that can detect all four components simul-
taneously. However, there are many that can detect two or three
of the components. In this paper, we have used three instruments,
FT-IR and two separate micro-GCs fitted with different analytical
columns (M5A and HSA, respectively), each of which can quantify
only three of the components. However, by 2D correlation of the
outputs from these instruments, we can obtain a full analytical
picture of the equilibrium. Furthermore, the simplicity of the
equilibrium means that it is easy to check that the 2D correlation
is producing meaningful results. The RWGSR actually has a more
immediate application in our own research interest area, since it
is a key process in the generation of H2 + CO2 mixtures by the
catalytic decomposition of HCO2H, which we are using as a route
to miniaturizing supercritical hydrogenation “without gases”.26
The second part of this paper involves applying 2DCOR-GC
to continuous Friedel-Crafts alkylation in scCO2. These reactions
still have serious mechanistic and analytical questions,27 which
can be addressed by 2DCOR-GC. Furthermore, the nature of these
experiments makes it very easy to apply appropriate perturbations
that are needed for successful 2DCOR-GC studies.
The Friedel-Crafts alkylation of aromatics is a valuable
synthetic tool that is widely employed in industry. Usually the
aim is to make the monoalkylated product often with the alkylation
occurring at a specific location on the aromatic ring. Traditionally,
this is achieved by using an excess of a Lewis or Brønsted acid
catalyst.28 The chemistry, however, often ends up being environ-
mentally unfriendly, generating an excess of acidic waste that
requires a large amount of cleanup. The reaction in scCO2 offers
a promising route to cleaner alkylation by using solid acid
catalysts, particularly in a continuous process.23,29,30 Performing
this reaction in a continuous reactor gives several advantages
including ease of catalyst and solvent separation, shorter reaction
times, control of reactant concentrations and continuous produc-
tion of product from a small reactor volume.22 Thymol (T, Chart
1) is a commercial product that can be produced by the Friedel-
The primary purpose of this paper is to demonstrate the validity
of our technique and to show that it really can provide chemically
(18) A. D. M. IUPAC Recommendations; Blackwell Science Ltd.: Oxford, U.K.,
1997.
(19) Hyde, J. R.; Leitner, W.; Poliakoff, M. In High-Pressure Chemistry; Eldik, R.
V., Klarner, F., Eds.; Wiley VCH: Weinheim, Germany, 2002; p 371.
(20) Hitzler, M. G.; Poliakoff, M. Chem. Commun. 1 9 9 7 , 1667-1668.
(21) Hitzler, M. G.; Smail, F. R.; Ross, S. K.; Poliakoff, M. Org. Proc. Res. Dev.
1 9 9 8 , 2, 137-146.
(26) Hyde, J. R.; Poliakoff, M. Chem. Commun. 2 0 0 4 , 13, 1482-1484.
(27) Albright, L. F. Ind. Eng. Chem. Res. 1 9 9 8 , 37, 296-297.
(28) Olah, G. A. Friedel-Crafts and Related Reactions; Wiley Interscience:
London, 1963.
(29) Chateauneuf, J. E.; Nie, K. Abstr. Pap. Am. Chem. Soc. 2 0 0 0 , 219, 446-
ORGN.
(22) Hitzler, M. G.; Smail, F. R.; Ross, S. K.; Poliakoff, M. Chem. Commun. 1 9 9 8 ,
359-360.
(23) Gray, W. K.; Smail, F. R.; Hitzler, M. G.; Ross, S. K.; Poliakoff, M. J. Am.
Chem. Soc. 1 9 9 9 , 121, 10711-10718.
(24) Grunwaldt, J. D.; Wandeler, R.; Baiker, A. Catal. Rev.-Sci. Eng. 2 0 0 3 , 45,
1-96.
(30) Funamoto, G.; Tamura, S.; Segawa, K.; Wan, K. T.; Davis, M. E. Res. Chem.
Intermed. 1 9 9 8 , 24, 449-459.
(25) Walsh, B.; Hyde, J. R.; Poliakoff, M. Green Chemistry, submitted.
Analytical Chemistry, Vol. 76, No. 21, November 1, 2004 6199

Crafts alkylation of m-cresol (M-C) with isopropyl alcohol or
propene. Unless optimized, the reaction yields several products
that have similar physical properties and are often difficult to
separate fully by conventional GC. In addition, there are several
possible routes from M-C to a particular product, and it is not
necessarily clear which route predominates under a given set of
experimental conditions. We demonstrate that solving or elucidat-
ing such problems is particularly suitable for tackling with 2DCOR-
GC.
Ta b le 1 . De t e c t io n Ta b le fo r t h e RWGS R Re a c t a n t s
a n d P ro d u c t s
CO2
H2
CO
H2O
µGC HSAa
µGC M5Ab,c
FT-IR
yes
no
yes
coelute
yes
no
coelute
yes
yes
yes
no
yes
a
HSA. porous polymer. b M5A, molecular sieves. c Using He as the
carrier gas, the response to H2 is nonlinear, but this nonlinearity is
not important in the experiment reported here.
EXPERIMENTAL SECTION
increased such that the total concentration remained constant.
Organic samples were collected directly from the BPR. Each
sample was analyzed by standard quantitative GC methods (100
µL org/ 5 mL acetone, 0.1-µL injection volume, Shimadzu 2010,
RTX-5 30 m, 40 °C iso 2 min, 10 °C/ min 200 °C, iso 5 min).
2 D-Correlation Analysis. 2D correlation analysis of both the
GC and FT-IR was performed by converting the AIA (a standard
chromatographic file type) format files into ACSII XY. These files
were imported into MATLAB, and 2D contour maps were
generated by a program based on code similar to that made
available by Ozaki et al. (Matlab toolbox available for download
at http:/ / science.kwansei.ac.jp/ ∼ozaki/ twod.zip) The reference
for each correlation analysis was taken as the mean, jy(τ), in each
case. Chromatograms were baseline normalized by accumulative
summation and differentiation.
As explained above, the 2D correlation analysis produces two
different types of 2D plots, synchronous and asynchronous. The data
from these two plots yield different information, provided that the
process under study is not in an equilibrium state without further
changes, as in the unperturbed RWGSR. The analysis can be
performed on two types of data; a “homo” correlation, where the
two sets of data come from the same analytical instrument (e.g.
FT-IR-FT-IR, or GC1-GC1) and “hetero” correlation, where the
data come from different instruments (e.g., FT-IR-GC, GC1-GC2).
We exploit the full combinations of synchronous, asynchronous,
and homo- and heterocorrelation methods.
The continuous flow scCO2 reactors at Nottingham have been
described previously21 and were used with only minor modifica-
tions for this work. In particular, all reactions were performed
with a Jasco 1580-81 back pressure regulator rather than manual
valves, to simplify product collection and to facilitate expansion
of the SCF in a single stage. The different experiments were
carried out as follows:
Water-Gas Shift Temperature-Based Perturbation. Bottled
CO2 and H2 (1:1 molar ratio) were delivered into a 10-mL catalyst
bed containing a commercially available catalyst (2% Pd supported
on SiO2/ Al2O3) heated to reaction temperature (160 f 420 °C, in
20 °C steps). The system pressure was maintained by an electronic
back pressure regulator set to 100 bar. Effluent gases were passed
into a high-pressure infrared flow cell24 (path length 1 mm, CaF2)
and IR spectra were measured (Nicolet Impact 410, 4-cm-1
resolution, 64 scans). The low-pressure gases were then passed
1
via / 16-in. SS-316 tubing into a glass tee piece with an internal
volume of 2.5 mL. A vent was placed at the third entrance to the
tee piece to avoid overpressurizing the sample line that led to
the micro-GC. Gases were collected into a Varian 4900 micro-GC
with an internal pump that drew gaseous samples into the injection
port. Analysis was performed on two capillary columns (M5A 20
m with back flush (8 s) 120 °C/ 20.0 psi isobaric; HSA 0.4 m; 80
°C/ 20.0 psi isobaric. Both channels used a 50-ms injection time),
and results were converted into the AIA format (a standard
chromatographic file type).
Water-Gas Shift Concentration-Based P erturbation. For-
mic acid (99.9% anhydrous, Aldrich UK) was passed by HPLC
pump over a 10-mL catalyst bed (JM catalyst, 2% Pd/ SiO2) heated
to 450 °C. The pressure of the gases generated was maintained
by an electronic back pressure regulator (Jasco 1580-81) and held
constant at 100 bar. Water was then mixed with HCO2H via HPLC
pumps with a gradient flow such that the total flow rate was
maintained at 1 mL/ min over a period of 3 h. The solution was
passed over the catalyst bed. GC analyses of the gases exiting
from the infrared cell were collected in a manner identical to that
described previously.
Alkylation of m-cresol by isopropyl alcohol (IPA): The design
and use of a supercritical CO2 reactor has been described
previously22 and was used without further modification to the
design. The reactor used was a 12-mm tube ∼16 cm in length
packed with Nafion SAC-13 catalyst. scCO2 and reactants were
delivered as described earlier; m-cresol and IPA (Aldrich, 99%;
3:1 molar ratio) were pumped via HPLC pump into the static mixer
where it was mixed with the scCO2. The catalyst temperature was
monitored by a thermocouple (K-type) from within the catalyst
bed. The flow rates of the organic mixture and scCO2 were
RESULTS AND DISCUSSIONS
2 D Correlation of the Water-Gas Shift by FT-IR and GC.
We have perturbed the RWGSR in two simple ways: first by
changing the reaction temperature, which will alter the equilibrium
constant, or second, by an isothermal change in reactant concen-
trations, which will change the equilibrium position. The experi-
ment involves a 1:1 molar ratio of CO2 and H2 passed over a heated
metal catalyst, which promotes the RWGSR. Both GC and FT-IR
probes are then used to analyze the gases produced. This
experiment provides us with three sets of data, one from FT-IR
and one each from the two GC systems. These data can be used
for traditional IR-IR 2D correlation, which validates the experi-
ment, and IR-GC and GC-GC correlations, which demonstrate
the principle of our new method. The detection capabilities of the
three techniques are shown in Table 1.
The chromatograms were collected using (i) a M5A molecular
sieve column, which is capable of light gas separation with an
elution order of H2, Ar, Ne, O2, N2, CH4, and CO, and (ii) an HSA
porous polymer column, which separates gases heavier than CO
and hydrocarbons. The H2, N2, and O2 elute as a single peak on
the HSA column but are separated on the M5A; see Figure 1.
6200 Analytical Chemistry, Vol. 76, No. 21, November 1, 2004

show an apparent correlation (negative) with CO because, as the
CO concentration increases, there is a correspondingly smaller
volume available for air to be drawn through the sample injection
line.
The heterocorrelation of the M5A and FT-IR data are shown
in Figure 2c, clearly illustrating the relationship between the two
techniques. As before, the CO and CO2 responses dominate the
GC and FT-IR, respectively, and therefore, these signals show the
strongest correlation intensities. The CO GC peak at just above
120 s in the M5A chromatogram strongly correlates with the entire
FT-IR spectra, showing a negative correlation with both the
fundamental and overtone IR bands of CO2 but a positive
correlation with the CO and H2O IR bands. In addition, this series
of correlation peaks can distinguish between the ν(O-H) bands
of H2O and the overtones of CO2, which occur in a similar region
of the spectrum, because the bands show opposite correlation with
the CO peak. H2, O2, and N2 are all IR inactive due to their
symmetry; nevertheless, they still correlate positively to the CO2
2300-cm-1 bands in the FT-IR spectra, because their concentra-
tions are changing in the same direction.
Fig u re 1 . Chromatogram showing the separation of H2, CO, CO2,
and adventitious air by (a) µGC HSA and (b) µGC M5A during the
RWGSR (see Experimental Section.) Note that under the conditions
used here the chromatograph is working in the region of nonlinear
detection of H2, so that the peak, H2, shows only a small variation in
height/area with changing concentration. Nevertheless, even this
modest nonlinear response is quite sufficient to demonstrate the
power of our 2DCOR-GC approach. Also note in (a) the peak due to
H2/air which despite its very low intensity gives a signal in the 2D
correlation.
In summary, the correlations in this system can be seen to
behave as expected both in the homo- and heterocorrelations,
showing that the combination of these two measurements is a
valid analytical technique. Moreover, there is little argument over
the products of this reaction, its mechanism, and equilibrium
position; the GC-GC, GC-FT-IR correlations indeed show the
expected behavior.
2 D Correlation GC of the Water-Gas Shift P erturbed by
H2 O Concentration. It is possible to perturb the RWGSR
equilibrium by changing the concentration of one of the reactants
isothermally. To this end, a mixture of CO, H2O, CO2, and H2
was generated by the catalytic decomposition of HCO2H over a
2% Pd catalyst. It is known that HCO2H can decompose both
catalytically and thermally to produce a mixture of these gases,
the composition of which depends entirely on the pressure,
temperature, and particular catalyst. It is also known that high
temperatures favor the pathway that leads to H2O and CO, but
the presence of H2O is likely to shift the RWGSR equilibrium
toward the left-hand side of eq 1 and to favor CO2 and H2 as the
final products. We have previously employed the decomposition
of HCO2H and H2O to generate high-pressure CO2 and H2
mixtures capable of hydrogenating organic compounds under SCF
conditions. This approach to chemistry in an SCF holds the
practical advantage that high-pressure bottled gases are eliminated
from the laboratory,26 an aspect that is currently under com-
mercialization by HEL Labs, UK.
P erturbation by Temperature. The RWGSR was run at
constant pressure (10 MPa), and the temperature of the catalyst
bed was varied between 160 and 420 °C, while the composition
of the gas emerging from the reactor was analyzed. Figure 2b
shows the correlation of the temperature-dependent IR data. The
CO2 combination band at 3200 cm-1 and the corresponding ν3
fundamental bands at ∼2300 cm-1 can be seen to exhibit a positive
correlation showing that the bands decrease or increase together
as the catalyst bed temperature is changed. By contrast, the CO
and H2O vapor bands, at 2100 and 1500 cm-1, respectively, show
a negative correlation with the strong CO2 ≈ 2300-cm-1 band,
indicating that as the CO2 concentration decreases, the CO and
H2O concentrations both increase. This is the behavior expected
in this system and shows that the RWGSR is indeed occurring.
The apparent absence of autopeaks for the CO and H2O vapor
indicates that under these conditions the absolute changes in
intensity for these bands are small compared to that of CO2.
Figure 2a presents the correlation of the M5A GC data. It can
immediately been seen that the GC data have yielded a meaningful
map. In more detail, the CO response shows a strong autopeak,
which is due to the large change in intensity of the CO peak during
the course of the perturbation; see Figure 1b. On the other hand,
H2 has a much smaller change in intensity due to the nonlinear
response, and the autopeak is too weak to register on this plot.
However, despite the weakness of the H2 autopeak, it is absolutely
clear from the off-diagonal peaks in Figure 2a that the concentra-
tion of H2 decreases synchronously as the concentration of CO
increases. The two other correlation signals are due to O2 and N2
impurities, which arise because a small volume of air is drawn
with the reacting gases from the gas injection device into the GC
(see Experimental Section). Surprisingly, these impurities also
We now investigate the role of H2O by studying the effect of
diluting pure HCO2H with H2O prior to the catalytic decomposition
of the acid (see Experimental Section). The results are displayed
in Figure 3, which shows the homocorrelations of the two GC
chromatograms (plots a and b) and their hetrocorrelation (plot
c). The M5A homocorrelation, Figure 3a, is similar to that in
Figure 2a, although the nature of the perturbation was quite
different in the two cases. The fact that these two plots are so
similar increases confidence in the 2DCOR-GC approach. Figure
3b shows the corresponding homocorrelation of the HSA GC
chromatograms. In this case, due the enhanced leveling of
contours, it is possible to observe four peaks, one due to CO2,
Analytical Chemistry, Vol. 76, No. 21, November 1, 2004 6201

Fig u re 2 . (a) (Top left) Homo M5A and (b) (bottom right) homo-FT-IR produced during the temperature-perturbed RWGSR. (c) (Top right)
The heterocorrelation of the M5A and FT-IR. Note the correlation of H2, O2, and N2, which are IR inactive, and also the small positive correlation
of H2O close to the CO2 overtones ∼3600 cm-1. In all of the 2DCOR-GC plots, red is taken as a positive correlation and blue as a negative
correlation.
one due to CO, and the final two very weak signals corresponding
to H2, N2, and O2, and a small amount of CH4. It can be seen that
the H2/ air peak correlates negatively with CO and positively with
CO2, while the CO2 and CO are negatively correlated. The
heterocorrelation (Figure 3c) corroborates the identity of the
major peaks. In detail, Figure 3 confirms that when the concentra-
tion of CO decreases (in the presence of H2O), the concentration
of CO2 and H2 increase as expected.
Alkylation of m -Cresol with Isopropyl Alcohol in scCO2
over a Nafion SAC-1 3 Catalyst. As a more realistic validation
of the 2D correlation of GC data, we have tested the technique
on a classical chemical reaction, the alkylation of m-cresol,
performed under continuous flow scCO2. The key scientific
question is whether, under these conditions, the reaction proceeds
via initial alkylation of the OH group followed by rearrangement
or via direct attack on the aromatic ring. The technical question
in the context of 2DCOR-GC is whether any useful information
can be derived from a reaction mixture that is far more compli-
cated than that in the RWGSR. In this alkylation reaction, there
are in fact at least eight possible species, which can be separated
relatively easily by GC as shown in Figure 4, where for the
purposes of this analysis we are ignoring any minor products.
The synchronous and asynchronous maps for the alkylation
reaction are shown in their entirety in Figure 5 and in detail in
Figure 6. The perturbation was flow rate, and hence reaction
residence time, achieved by maintaining a constant concentration
profile over a range of differing flow rates of the scCO2 solvent.
We now show that an interesting application of 2DCOR-GC is
the identification of overlapping peaks of coeluting species, which
may not be easily seen from the chromatograms presented in a
6202 Analytical Chemistry, Vol. 76, No. 21, November 1, 2004

Fig u re 3 . (a) (Top left) M5A homocorrelation and (b) (bottom right) the HSA homocorrelation map. Note the small correlations around the
∼0.1 min CO peak; these signals are due to H2, O2, and N2 (prior to CO) and CH4 (post to CO). The small volume of CH4 produced is due to
the increasing residence time within the reactor as the overall gas flow rate decreases. (c) (Top right) the corresponding heterocorrelation plot,
which confirms the identity of the major peaks. In all of the 2DCOR-GC plots, red is taken as a positive correlation and blue as a negative
correlation.
more conventional manner. As mentioned in the introduction, one
of the more interesting products of this reaction is thymol, T in
Figure 4, which gives rise to the peak with a retention time of
∼13 min. Panels a and b of Figure 6 show expanded portions of
the synchronous and asynchronous maps in the region of this
peak. It is immediately clear that the autopeak peak for T is more
complicated than the other correlation peaks in this diagram. In
2D-FT-IR, such patterns in the synchronous map can be attributed
either to a shift in peak position or to two overlapping peaks
changing in opposite directions. This ambiguity is usually resolved
in the asynchronous map where, in this case, the pattern observed
is exactly the pattern that would be expected for two overlapping
peaks if the map were for 2D-FT-IR. However, there is a difficulty;
a single-component peak in 2DCOR-GC can also exhibit this
asynchronous pattern, because an increasing concentration in GC
produces asymmetric peaks which have small differences in the
peak maximums, i.e. anti-Langmuir/ Langmuir isotherm-type
adsorptions. Examination of the raw chromatograms (not il-
lustrated) suggests that there are two components that are moving
in the same direction, which is not suggested by either of the
correlation maps. A solution to this problem can be to align the T
peak to a common maximum and, hence, a constant elution time,
because unlike FT-IR, a distance usually separates the peaks in
the chromatogram, which is large compared to the size of the
peaks themselves. Therefore, a minor correction in the position
of one particular peak will not have any effect on the shape,
intensity, or retention time of the other peaks in the chromato-
gram. Although this procedure will clearly alter the absolute value
of the retention time, it does not affect the qualitative arguments
that we need to apply to these maps.
Panels c and d of Figure 6 show the expanded region of the
synchronous and asynchronous maps using the chromatograms
after the adjustment of the peak assigned to T. The key point is
that the synchronous map, Figure 6c, still shows that the autopeak
of “corrected” T has a shape quite different from those of the the
autopeaks assigned to M and P, respectively. Comparison with
Analytical Chemistry, Vol. 76, No. 21, November 1, 2004 6203

Ch a rt 1 . S e q u e n t ia l Ord e r o f t h e Ch a n g e s in
P ro d u c t Dis t rib u t io n Ob s e rve d a
Fig u re 4 . (Top) Gas chromatogram of the alkylation products of
M-C with isopropyl alcohol. (Bottom) Chemical structures of the
compounds produced during the alkylation of M-C with isopropyl
alcohol.
a The perturbation was reaction residence time. From this
sequence it can be seen that D-P was the last compound to
increase in concentration, as the residence time was decreased,
while E was the first compound to change in concentration during
the course of the perturbation.
the asynchronous map, Figure 6d, now reveals the characteristic
pattern for two overlapping peaks in 2DCOR-GC at ∼13 min, and
the presence of two components changing in the same direction,
is confirmed by the cross-peaks involving T, which show positive
correlations. This preeluting component has been identified as
o-thymol, O′, and by Gaussian deconvolution, a chromatographic
resolution of 0.28 can be determined. Indeed, this overlap can be
confirmed by rerunning the samples with a different column and
program to resolve the peaks.31 The behavior of O′ can be seen
to be closely related to that of T, producing correlation signals of
identical sign to T, the strength of which, however, differs in each
case. The cross-correlation intensities of the three isomers, O′,
T, and M can all be seen to be positive, indicating that their
concentrations are increasing or decreasing together during the
course of the perturbation.
The second application of 2DCOR-GC is to narrow down a
sequence in which the various products are formed. Using these
aligned chromatograms, full correlation maps can be generated.
Figure 5 shows the synchronous and asynchronous maps at high
contour resolution. One can now sequence the event with respect
to all eight major components of this mixture, by using Noda’s
rules. The sequence produced is found to be M-C > E > O′ > T
) P ) D-O > M > D-P; see Chart 1. With a single experiment,
it is not possible to draw conclusions as to the order of events
between T, P, and D-O and because we have tentatively assigned
this sequence based on the strength of the correlation signals.
The sequence and correlation maps suggest that D-O is a primarily
formed from O′, while D-P arises from a further reaction product
of P. The position of M in the sequence suggests that it is formed
via a mechanistic route different from that of the other isomers,
possibly a direct alkylation or a transalkylation. The correlations
of E and M-C with all products suggest that M-C and E are in
rapid equilibrium; thus, it is difficult to conclude if further reaction
products originate from M-C or E. It is most likely a combination
of the two.
Further possible relationships can be drawn from these
correlation maps, and there are several chemically plausible routes
to each product, via either direct alkylation, Fries rearrangements,
or intramolecular transalkylation reactions. It is important to stress
here that this reaction is complex, and one single experiment will
not answer all of the scientific questions posed. Moreover, the
ease of detection of the O′ isomer, which was previously difficult,
even with an extremely detailed examination of the raw data,
shows that the use of 2DCOR-GC is a highly worthwhile exercise.
CONCLUSIONS
We have shown that the generalized 2D correlation approach
described by Noda3 can successfully be applied to GC data
obtained from reactions performed in scCO2. Although the
RWGSR is a simple reaction, the 2D correlation of the GC data
validates the use of the correlation method, thus opening the way
for further experimentation using this technique. In this paper,
the application of FT-IR has been used to relate this technique to
the more traditional 2D-FT-IR, while also revealing information
about each chromatographic component in the IR. We have
applied the 2D correlation method to the GC analysis of super-
critical fluid reactors, as a tool to help understand changes in
(31) Amandi, R. Ph.D. Thesis, Nottingham University, Nottingham, U.K., 2004.
6204 Analytical Chemistry, Vol. 76, No. 21, November 1, 2004

Fig u re 5 . (a) (Top) Full synchronous correlation map generated from the T aligned chromatograms collected during the scCO2 alkylation of
M-C with isopropyl alcohol over a Nafion catalyst. (b) (Bottom) Corresponding asynchronous correlation map.
Analytical Chemistry, Vol. 76, No. 21, November 1, 2004 6205

Fig u re 6 . Synchronous and asynchronous correlation maps of the thymol, T, region of the chromatogram (12.9-13.45 min). (a) (Top left)
Synchronous correlation map of the raw untreated data. (b) (Top right) The corresponding asynchronous correlation map. (c) (Bottom left)
Synchronous correlation map of the T aligned data. (d) (Bottom right) The corresponding asynchronous correlation map of the T aligned data.
product distributions as a result of simple perturbations, such as
a change in system pressure, temperature, residence times, or
organic concentrations. The analysis holds the advantage that it
reduces a large number of GC chromatograms into two informa-
tion-rich maps. Examination of the 2DCOR-GC maps also reveals
relationships that may not be apparent from traditional GC analysis
due to inconsistent conversions. 2DCOR-GC also enhances the
resolution of GC by identification of overlapping responses, which
often becomes apparent within the asynchronous 2DCOR-GC map.
By understanding the changes in selectivity or conversion with
each perturbation in turn, a greater understanding of the chemical
processes occurring in SCF reactors can be achieved, and we
believe this will allow a more rapid optimization of the overall
process.
ACKNOWLEDGMENT
We thank EPSRC (GR/ R41644) and Thomas Swan & Co. Ltd.
for support. We are grateful to M. Guyler, P. Fields, R. Wilson, S.
K. Ross, B. Walsh, P. Licence, R. Amandi, and M. W. George for
their help and advice.
Received for review April 30, 2004. Accepted July 18,
2004.
AC049360L
6206 Analytical Chemistry, Vol. 76, No. 21, November 1, 2004