Accurate Concentration Measurements Using Infrared
Absorption without Determination of the Baselines
P. M ARTEAU * and P. TOBALY
Laboratoire d’ IngeÂnierie des MateÂriaux et des Hautes Pressions, CNRS, Institut GalileÂe,
Av. J. B. CleÂment, 93430 Villetaneuse, France
Infrared absorption spectra have become widely used for in situ
determ ination of concen trations in com plex m ixtures. However, ex-
perim ental accuracy is often limited by lack of knowledge of the
baselines. The m athematical treatment described below allows the
spectra to be superim posed on any background with a shape as
complex as a parabolic one. The concentrations are thus accurately
determ ined whatever the baseline of the m ixture spectrum and even
those of the calibration spectra. A mathematical proof of this prop-
erty is given within the fram ework of the presen tly used m ethod,
and tests of the ef® ciency of the method are presen ted.
Index Headings: Infrared spectra ; Baselines; Concentrations; Phase
equilibria.
INTRODUCTIO N
Infrared spectroscopy has become widely used for in
situ analysis in various industrial ® elds, such as food,
drugs, and petroleum. It is a nonintrusive, nondestructive
method that gives real-tim e results. Furthermore, when
associated with multivariate statistical methods, 1 infrared
spectroscopy enables quantitative measurem ents from
spectra including contributions of a considerable number
of species. However, unknown baselines caused by dif-
ferent optical effects m ay lead to large errors. We have
therefore developed a method allowing an accurate de-
termination of concentrations by elimination of the base-
lines.
We did not use the standard m ethods of multivariate
analysis for different reasons related to our particular
problem, which involved the determination of the phase
equilibria of binary m ixtures of hydrocarbons under high
pressures. 2±4 Such mixtures usually consist of a liquid
phase and a gas phase in equilibrium at given temperature
and pressure. The problem is to determine the concentra-
tion of each species in each of the two phases. [Results
are commonly assembled as plots of coexisting lines (or
phase envelopes) such as those in Fig. 3.]
The standard procedure is to use statistically oriented
analysis, such as classical least-squares (CLS), partial
least-squares (PL S), and principal component regression
(PCR), with a set of calibration samples m ade of known
mixtures with concentrations covering the range of the
target samples. In the present case, the m ixtures with rel-
evant concentrations can be obtained only under high
pressure. Therefore, we used pure sample spectra as the
reference. In addition, owing to the fact that spectral lines
may be broadened by density effects under high pres-
sures, we used integrated absorptions that rem ain pro-
portional to concentrations. Moreover, integration pro-
Received 13 March 1998; accepted 11 Decemb er 1998.
* Author to whom correspondence should be sent.
0003-7028 / 99 / 5304-0396$2.00 / 0
396 Volume 53, Number 4, 1999 q 1999 Society for Applied Spectroscopy
vides a mean of averaging over the noise. Finally, we
found it easier to solve a determined system than to use
a least-squares m ethod because, in this case, the com-
ponents are well de® ned and in reduced number.
Of course, our method is based upon the linear rela-
tionship that exists between integrated absorptions and
molecular concentrations. However, it is well known that
the measurem ent accuracy is strongly dependent on an
accurate determination of the baseline or on its elimina-
tion by ® rst or second derivation. Although the baseline
can be properly determined in the case of sharp lines well
separated by wide frequency domains of zero absorption,
for m ost of the m ixtures, such as hydrocarbon mixtures,
the near-infrared spectrum exhibits broad and overlapping
bands that do not leave any frequency domain on which
the baseline could be even roughly determined. Further-
more, the baseline of the absorption spectrum of a sample
introduced in an optical cell can be strongly shifted away
from its normal position and also distorted for many rea-
sons. For example, if the cell is ® lled with a liquid of
refractive index greater than 1, or even with a gas under
high pressure, as shown in Fig. 1, the re¯ ectivity at the
window/sample interface is lowered and the signal trans-
mitted through the cell can be increased by up to 10%
compared to that transmitted through the empty cell taken
as a reference. 5±7 Second, the presence in the cell of a
sample with a high refractive index can change the beam
focusing so that the energy reaching the detector is also
changed. Finally, in the vicinity of the critical region tur-
bidity can occur, producing light scattering as a function
of frequency. In the latter case the shape of the baseline
can be more complex than a simple straight line.
In order to achieve accurate measurem ents of the con-
centrations in such mixtures, in spite of the lack of
knowledge of the baselines, we have developed a partic-
ular method that allows one to be quite free of any prob-
lem concerning the baselines of the m ixture spectra as
well as those of the calibration spectra. This m ethod al-
lows for the presence of background pro® les as complex
as parabolic ones without loss of information or decrease
in the signal-to-noise ratio, while the second-derivative
method is limited to straight baselines.
The ® rst part of this paper is devoted to the description
of the proposed mathematical treatment along with a
mathematical proof of the independence of concentration
results on the baselines of the calibration spectra. In the
second part, som e details about the experimental setup
and procedure used for determination of the integration
frequency domains are given, as well as tests of the ef-
® ciency of the m ethod.
APPLIED SPECTROSCOPY
 F IG . 1. Variation of the baselines of spectra of pure compressed supercritical propane at 150 8 C as a function of the concentration.

E
M ATH EMATICAL TREATM ENT 1
A ij 5 S i (n ) dn (2)
The molar concentration of a sample can easily be de- c 0i Dj
duced from its infrared absorption spectrum through the where c 0i is the concentration of the pure species i for the
linear relationship that exists between absorbance and reference sample, S i(n ) its absorption spectrum , and l the
concentration. This linearity is usually valid at low den- absorption pathlength.
sity, i.e., for m ost gases. At higher densities, the pro® le The concentrations c 1 , . . ., c n can be accurately de-
of the spectral bands is broadened, and some other effects duced from relation 1 only if the baselines of all the
due to interaction between m olecules m ay change the ab- spectra S mix (n ) and S i (n ) are perfectly known. Such a sit-
sorptivity at given frequencies. The latter effect is not uation is rarely encountered. However, with the following
present or limited in the case of nonpolar or weakly polar generalized method the baselines can be ignored.
compounds8 such as hydrocarbons. Therefore, the present Baseline Treatment. Since the baseline of an infrared
method is limited to those types of compounds for which absorption spectrum can never be determined with an ab-
the m ixture spectrum is merely the sum of the separate solute accuracy, the baseline is considered as an addi-
contributions. As for the broadening effect, it is account- tional unknown quantity de® ned as b n 2 1 g n 1 d where
ed for by using integrated absorptivities that remain ap-
b , g , and d are three unknown param eters. Although a
proximately constant for m ost of the considered spectral higher order polynomial in n could be used as well, as
domains. Of course, this condition must be veri® ed care- required in some situations, 9 the restriction to the second
fully in terms of the reference spectra. The chosen do- order provides simplicity while covering m ost of the ex-
mains were tested as explained in the second part of this perimental situations: constant baseline (positive or neg-
article. ative), linear vs. frequency, or even a parabolic shape.
Although we dealt only with binary mixtures, the The recorded spectrum of the mixture still contains
mathematical treatment can be easily extended to n com- information on the relative concentrations c 1, . . ., c n, but
pounds. We therefore establish the equations for a mix- their determination now requires n 1 3 integrated ab-
ture of n compounds. If the absorption bands are strongly sorption measurem ents, leading to the following system:
overlapping, the concentrations c 1 , . . ., c n are deduced
from the absorption spectrum by m easuring the integrated A 11 c 1 1 ...1 A n1 c n 1 b A n11 1
1 g A n11 2
1 d A n11 3
5 B1
absorptions on n frequency domains D 1 , . . ., D n and solv-
ing the following system: A 1n1 3 c 1 1 ... 1 A nn 1 3 c n 1 b A nn11 1
3 1 g A nn11 2
3 1 d A nn11 3
3 5 B n1 3

E
(3)
A 11 c 1 1 ... 1 A n1 c n 5 S mix (n ) dn where
D1

A 1n c 1 1 ... 1 A nn c n 5
E Dn
S mix (n ) dn (1)
A nj 1 1
5
E Dj
n 2 dn A nj 1 2
5
E Dj
n dn

where S m ix (n ) is the absorption spectrum of the mixture,

S mix (n ) 5 l 2 1 ln[I 0(n )/I (n )]; and the A ji values are the in-
tegrated absorptivities of the pure species i on the fre-
A nj 1 3
5
E Dj
dn Bj 5
E Dj
S mix (n ) dn .

quency domain j, namely In each of the n 1 3 equations of system 3, the measured

APPLIED SPECTROSCOPY 397

integrated intensity is expressed as the sum of the con- absorption coef® cient for species i on the frequency do-
tributions due to each species and to the baseline. When main D j is in fact found to be equal to

E
one is solving the system, the b , g , and d variables are
calculated so that the contributions of each species to the 1
(A ij ) 9 5 [S i (n ) 1 m in 2
1 n in 1 p i ] dn (4)
n 1 3 integration dom ains are in the same proportions as c 0i Dj
for the pure com pound. As a consequence, correct values where S i(n ) is the true but unknown absorption spectrum.
of the concentrations are always found even if the ab- The m easured (A ji ) 9 coef® cient can also be expressed as
sorption spectrum of the m ixture is strongly distorted by
the background of an unknown baseline. m i n1 n i n1 pi n1
Baselines of Calibration Spectra. It is noteworthy (A ij ) 9 5 A ij 1 Aj 1
1 Aj 2
1 Aj 3 (5)
c 0i c 0i c 0i
that knowledge of the baselines of the calibration spectra,
i.e., the spectra of the pure species leading to the deter- where A ji (i 5 1 . . . n) are the true but unknown integrated
mination of the A 1j , . . ., A jn coef® cients, is not necessary absorption coef® cients.
as long as these baselines can also be represented by a Therefore, in relation 3, instead of the determinant D,
second-order polynomial in frequency n . This particular

) )
A 11 ... A n1 A n1 1 1 A n1 1 2 A n11 3
advantage of the method, although already recognized in
similar methods by several authors, 10 ±12 had never been D 5 A 1n ... A nn A nn 1 1
A nn 1 2
A nn1 3

proven as far as we know. It is indeed hard to understand A 1n1 3 ... A nn 1 3 A nn 11 1

3 A nn 11 2
3 A nn11 3
3
from a physical point of view, but it can be explained on
a m athematical basis as follows. The m easured integrated one obtains the determinant D 9 ,

* *
m 1 n1 1 n 1 n1 2 p1 n1 3
A 11 1 A1 1 A1 1 A1 ... A n1 1 ... A n11 1
A n1 1 2
A n1 1 3
c 10 c 01 c 10
m 1 n1 1 n 1 n1 2 p1 n1 3
D9 5 A 1n 1 An 1 An 1 An ... A nn 1 ... A nn1 1 A nn 1 2 A nn 1 3 .
c 10 c 01 c 10
m 1 n1 1 n 1 n1 2 p1 n1
A 1n 1 3 1 An 3 1 An 3 1 An 3 ... A nn1 1 ... A nn11 1 A nn 11 2 A nn 11 3
c 10 1 c 10 1 c 10 1
3 3 3 3 3

However, it is clear that these two determinants are equal.
As a matter of fact, due to the general properties of the
determinants, column 1 of D 9 can be replaced by the
following linear combination:
m1 n1
col. 1 2 col.(n 1 1) 2 col.(n 1
c 01 c 10
p1
2 col.(n 1 3)
c 01
so that this column becomes equal to that of D. Since the
same holds for columns 2 to n, the two determinants are
identical.
The determinants associated with the c 1 , . . ., c n
variables, and hence the concentrations of species 1 to n,
are also unchang ed for th e sam e reaso n. A s a
consequence, the concentrations deduced from Eq. 3 are
always the same whatever the baselines of all these
spectra m ay be.
It m ust be emphasized that the determinants associated
with the variables b , g , and d remain functions of the
baselines o f th e calib ratio n sp ectr a. T heref ore, the
baseline of the m ixture spectrum cannot be determined
by using the calculated values of b , g , and d . In other
words, although the baselines can be of no importance
for the problem, it is clear, once m ore, that they cannot
be determined. This consideration does not matter since
the goal is only to extract the true values of the
concentrations.
EXPERIM ENTAL DETAILS AND TESTS OF TH E
METHO D
The present method is general and in some sense in-
dependent of experimental detail, and, although the m ain
goal of this article was to explain the reason why base-
2)
lines of both sample and calibration spectra do not intro-
duce any more errors, we give below a brief account of
the experiments and test procedures. Full experimental
(6)
detail is given elsewhere. 13,14
Ex p erim en ta l S etu p . T h e high -pr essure cell is
equipped with sapphire windows and is inserted into a
controlled-temperature furnace. Pressure is varied by
moving a piston, and the cell may be rotated around the
optical axis so that both phases may be analyzed in turn.
Some of the studies were conducted with a Jobin Yvon
THR grating m onochromator with a PbS infrared detec-
tor, while for the others a Bomem MB155 Fourier trans-
form interferometer was used. The resolution of the for-
mer was 0.8 nm, while the latter yielded spectra with a
resolution of 4 cm 2 1.
Determination of the Integration Dom ains. A typical
spectrum of a methane±tetralin mixture is shown in Fig.
2 together with the expected spectral contributions due to
the pure components. As in both cases, these contribu-
tions are due to overtones and combinations of funda-
mental vibrations of CH, CH 2, and CH 3 groups; the cor-
responding spectra are strongly overlapping, and the de-
termination of the concentrations needs a system such as

398 Volume 53, Number 4, 1999

F IG . 2. Typical spectrum of a methane±hydrocarbon liquid mixture and the frequency domains used for calculation of the concentrations. Contri-
butions of each component are strongly overlapping.

Eq. 3 to be solved. The choice of the ® ve integration
domains is made by a trial-and-error procedure in three
steps as described below.
Test on the Pure Components. The domains were cho-
sen so that the spectral contributions were signi® cant for
at least n domains (here two domains) for each com-
pound. Other wise, the system could be degenerate (de-
terminant 5 0). The choice is by no means unique, and
once it is done it has to be tested. The calculation was
run on spectra of pure samples (i.e., in the liquid state
for the solute and at various pressures for the solvent).
These spectra were of course not part of the calibration
set. As a m atter of fact, these spectra m ay be very dif-
ferent from the calibration ones, because the density may
be varied by varying pressure. The results are compared
with the known concentrations, and the width and posi-
tion dom ains are adjusted if necessary. This process is
repeated until satisfactory results are found, i.e., until the
predicted values are in good agreement with the real val-
ues, as shown in Table I. In practice, no more than three
or four such adjustments were necessary. In our opinion,
this approach provides the only possible test of the lin-
earity of integrated absorptions on the given domains. A
direct veri® cation of such a linearity would require a cor-
rection of the baselines of the pure sample spectra and
would not be consistent with the spirit of the m ethod.
Infrared measured values and real values are reported in
Table I. They are in good accordance as con® rm ed by a
value of R 2 5 0.99305. The discrepancies observed at
low pressures, up to 2.3%, could be m ainly due to the
inaccuracy of the pressure m easurements, which are par-
ticularly sensitive in a region of rapid variations of the
pressure, volum e, and temperature (PVT) data. As a con-
sequence of the noise, the calculated concentration values
of solute are not exactly zero, but it will be noticed that
these values are very low, typically on the order of 10 2 3
mol/L.
Test on a M ixture of Given Composition. The test of
ef® ciency of the method owing to the determination of
the solute concentrations is performed by analyzing a
``one-phase’ ’ mixture for which the initial amount of sol-

TABLE I. Concentration values obtained from the analysis of

spectra of pure propane at 415.5 K and various pressures by run-
ning the program for the m ixture propane-decalin. Comparison
with real values deduced from PVT data.
Solute
Propane (mol/L)
Pressure (mol/L)
(MPa) Measured IR Measured IR Real values
7.95 0.001 5.046 5.080
6.71 0.003 3.690 3.695
6.0 0.003 2.972 2.967
5.18 0.002 2.284 2.292
4.53 0.002 1.848 1.858
4.04 0.003 1.571 1.574
3.16 0.001 1.126 1.135
2.61 0.0 0.877 0.896
2.2 0.001 0.718 0.732
F IG . 3. Phase diagrams of the m ethane±squalane mixture in the critical
1.87 2 0.001 0.588 0.608
region.

APPLIED SPECTROSCOPY 399

TABLE II. Measured concentrations in a ``one-phase’’ propane-naphthalene mixture at 423 K and various pressures. The results obtained
without baseline treatment are completely wrong.
C (naphthalene)
(m ol/L)
Pressure IR (our
(MPa) Real method)
8.88 0.467 0.464
8.88 0.467 0.466
12.17 0.529 0.528
13.89 0.555 0.556
ute is known. Since the variable internal volum e of the
cell can be deduced from the piston position, the real
concentration values of the solute are also known. As can
be seen from data reported in Table II, m easured values
at various pressures favorably compare to the real ones.
Although the amount of propane initially introduced in
the cell was not known, its concentration has also been
determined by the infrared calculation as well as its m ole
fraction, which rem ains constant, as expected, apart from
the experimental uncertainties.
Comparison w ith the Synthetic Method. In order to def-
initely test the calibration m ade for a methane±decaline
mixture, our infrared result has been compared to that
obtained by using the synthetic method. The synthetic
method consists in determining the discontinuity of (dP /
dV ) T where P is the pressure and V the volume of a
sample of known composition. (For a more detailed de-
scription of this m ethod, see, for example, Ref. 15). This
method was performed inside our infrared cell so that
one point of the phase equilibrium line was determined.
The composition of the sample was determined by weigh-
ing the respective amounts of solute and solvent intro-
duced in the cell. We have also m easured this com posi-
tion by the infrared m ethod in the one-phase state at high
pressure. The mole fractions of methane were found to
be equal to 0.5684 and 0.5716, respectively, for the in-
frared and the synthetic m ethods.
Necessity of Integration and Baseline Treatm ent. To
illustrate two of the advantages of the m ethod, we have
added in Table II the results of calculations performed
with ® ve-point intensities instead of integrated intensities
and calculations without any baseline treatment. As can
be seen from the results, the use of integrated intensities
obviously improves the accuracy of the results, while the
baseline treatment appears m andatory as the results are
in considerable error without this treatment.
Noise Sensitivity. A test of sensitivity of the m ethod
TABLE III. Results obtained from the successive recorded spectra
of a m ethane± squalane m ixture at 370 K and 133 MPa.
Spectrum C squala ne C m e thane Methane
number (mol/L) (m ol/L) mole fraction
1 0.531 15.360
2 0.532 15.340
3 0.533 15.354
4 0.529 15.345
5 0.534 15.319
6 0.534 15.336
400 Volume 53, Number 4, 1999
C (propane) Propane mole
IR (no (mol/L) fraction
IR (point baseline IR (our IR (our
intensities) treatm ent) method) m ethod)
0.556 4.136 6.181 0.9302
0.555 4.129 6.188 0.9300
0.682 4.465 6.999 0.9299
0.729 4.665 7.288 0.9291
against signal-to-noise ratio has been perform ed by com-
paring the results obtained from the analysis of several
spectra of a given mixture at constant temperature and
pressure. Results reported in Table III correspond to a
mixture of squalane and m ethane at a pressure of 133
MPa and 370 K. The spectra were recorded in sequence
so that no spectrometer drift could occur in between. Dis-
crepancies obser ved between concentration values are on
the order of 1% for the solute and 0.25% for the solvent.
Comparison of our Results with Literature Data.
The high-pressure parts of the phase coexistence lines of
a m ethane-squalane m ixture at 323 and 370 K are shown
in Fig. 3. Our results can be compared to those interpo-
lated from the data obtained by Paas et al. 13 with the use
of the synthetic method. A general agreem ent is obser ved
apart from som e dispersion of their data at 323 K. This
result illustrates the dif® culty usually encountered in do-
ing m easurements in the critical region and the ability of
the infrared m ethod to do it.
CONCLUSIO N
The m ethod presented here can be used in m any cases
whenever the baseline cannot be accurately determined.
Apart from the m athematical proof and the ef® ciency
tests given here, several other tests have also been made
by adding various backgrounds to a m ixture spectrum
and also to the calibration spectra. As expected, the con-
centrations c 1 and c 2 derived from Eq. 3 did not change.
This generalized method has been applied for the deter-
mination of the concentrations in the gas and liquid phas-
es of binaries such as heavy hydrocarbon±m ethane and
heavy hydrocarbon±propane.13, 14, 16 Good results were ob-
tained even in the vicinity of the critical region where
turbidity induced strong distortions in the background.
1. H. Martens and T. Nñ s, Multivariate Calibration (John Wiley, New
York, 1991).
2. I. Swaid, G. Nickel, and G. M. Schneider, Fluid Phase Equilibria
21, 95 (1985).
3. U. K. Deiters and G. M. Schneider, Fluid Phase Equilibria 29, 145
(1986).
4. P. Marteau, J. Obriot, and R. Tufeu, J. Supercritical Fluids 8, 20
(1995).
0.9666 5. M. C. Jones, NBS Technical Note 390 (National Bureau of Stan-
dards, Gaithersburg, Mar yland, 1970).
0.9665
0.9664 6. J. Chamberlain, Infrared Phys. 12, 145 (1972).
7. J. Quazza, P. Marteau, and H. Vu, Infrared Phys. 13, 245 (1973).
0.9667
0.9663 8. M. Buback, ``Absorption Spectroscopy in Fluid Phases’ ’ , in Super-
critical Fluids: Fundamentals for Applications, E. Kiran and J. M.
0.9664
H. Levelt Senghers, Eds. (Kluwer, Amsterdam, 1994), pp. 499 ±526.
 9. I. K. Salomaa and J. K. Kauppinen, Appl. Spectrosc. 52, 579,
(1998).
10. D. M. Haaland and R. G. Easterling, Appl. Spectrosc. 36, 665
(1982).
11. D. M. Haaland and R. G. Easterling; Appl. Spectrosc. 34, 539
(1980).
12. D. M. Haaland, in Practical Fourier Transform Infrared Spectros-
copy, J. R. Ferraro and K. Krishnan, Eds. (Academic Press, New
York, 1990), pp. 396±468.
13. P. Marteau, P. Tobaly, V. Ruf® er-Meray, and A. Barreau, Fluid
Phase Equilibria 119, 213 (1996).
14. P. Marteau, J. Obriot, A. Barreau, V. Ruf® er-Meray, and E. Behar,
Fluid Phase Equilibria 129, 285 (1997).
15. R. Paas, Z. Alwani, E. Horvath, and G. M. Schneider, J. Chem.
Thermodynam. 11, 693 (1978).
16. P. Marteau, P. Tobaly, V. Ruf® er-Meray, and J. C. De Hem ptinne,
J. Chem. Eng. Data 43, 362 (1998).
APPLIED SPECTROSCOPY 401