Available online at www.sciencedirect.

com

Fluid Phase Equilibria 265 (2008) 139–154

Vapour–liquid equilibria of aroma compounds in hydroalcoholic

solutions: Measurements with a recirculation method and
modelling with the NRTL and COSMO-SAC approaches
Violaine Athès a,∗ , Patrice Paricaud b , Mohammed Ellaite b ,
Isabelle Souchon a , Walter Fürst b
a Unité Mixte de Recherche UMR 782 Génie et Microbiologie des Procédés Alimentaires,
AgroParisTech/INRA, F-78850 Thiverval-Grignon, France
b Unité d’enseignement et de recherche de Chimie et Procédés (UCP),

École Nationale Supérieure de Techniques Avancées (ENSTA), 75739 Paris, Cedex 15, France
Received 16 July 2007; received in revised form 11 January 2008; accepted 14 January 2008
Available online 20 January 2008

Abstract
Aqueous solutions of aroma compounds are commonly encountered in the food and biofuel industries, and the knowledge of their thermodynamic
properties is very important for process design. Here, we report the measurements of the vapour–liquid equilibria of water–ethanol solutions
containing 13 aroma compounds (alcohols, aldehydes and esters), for temperatures between 351 and 366 K, and ethanol concentration ranging
from 10 to 90% (v/v). The new data were obtained with a recirculation apparatus and are represented with the non-random two liquid (NRTL)
activity coefficient model. It is observed that the partition coefficients of the aroma compounds can decrease of several orders of magnitude by
increasing the proportion of ethanol. The COSMO-SAC (conductor-like screening model-segment activity coefficient) model provides reasonable
predictions of the new data for alcohols and aldehydes, but underestimates the partition coefficients of the long ethyl esters. It is observed that the
COSMO-SAC predictions depend a lot on the molecular conformations used to generate the sigma profiles, in agreement with previous studies.
The COSMO-SAC predictions can be empirically improved by changing the real space cut-off distance employed in the ab initio calculations. The
COSMO-SAC model is then reliable not only at infinite dilution but also over a whole composition range. The model is used to define a limit of
validity for the infinite dilution approximation.
© 2008 Elsevier B.V. All rights reserved.

Keywords: Vapour–liquid equilibria; NRTL; COSMO-SAC; Henry’s constant; Aroma compounds; Activity coefficients at finite dilution

1. Introduction
The knowledge of the vapour–liquid equilibria (VLE) in sys-
tems involving aroma molecules is of crucial importance for
many applications in the food industry, such as the design of dis-
tillation processes in alcoholic beverage production. The flavour
perception and the aromatic character of food products are gov-
erned by the partitioning of aromatic compounds between the
vapour and liquid phases. This partitioning is related to the ther-
modynamic properties of the mixtures, and is characterized by
the partition coefficients also called absolute volatilities. Aroma
∗ Corresponding author. Tel.: +33 130815383; fax: +33 130815597.
E-mail address: vathes@grignon.inra.fr (V. Athès).
molecules such as furfural and ethyl decanoate are also encoun-
tered in the production of biofuels. Therefore, the knowledge
of their thermodynamic properties is very important for the
design of reactive and separation processes related to the biofuel
industry.
Aroma compounds are often found at very low concentrations
in alcoholic beverages: their mole fractions are usually lower
than 10−4 [3]. For such low concentrations it can be assumed
that the partition coefficient Ki of aroma i is proportional to the
infinite dilution activity coefficient γi∞ (or to Henry’s constant
Hi ). Such an assumption should be considered with some care,
as we will see later in this paper. Measuring the partition coeffi-
cients of aroma compounds at infinite dilution in hydroalcoholic
solutions is a difficult task, mainly due to the small quantities of
aromas that have to be measured and to the large proportion of

0378-3812/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.fluid.2008.01.012
140 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154
ethanol in the vapour phase, which interferes in the chromato-
graphic analysis. The development of efficient and reproducible
experimental methods that truly achieve phase equilibrium is
also a major concern [4]. Among the available techniques to
measure activity coefficients at infinite dilution, one can find
static methods such as the equilibrium partitioning in closed
systems (EPICS) [5] and phase ratio variation (PRV) [6,7] meth-
ods, and dynamic methods such as recirculation (Gillespie-like
still) [8,9] and inter gas-stripping (exponential dilutor) methods
[10–12]. The reader is directed to the excellent reviews of Raal
et al. [13–15] and Sandler [16] for a detailed discussion about
the measurement techniques and engineering applications of the
infinite dilution activity coefficients. In a previous work [17], we
compared the performances of three different static headspace
methods to determine the vapour–liquid partition coefficients of
aroma compounds in hydroalcoholic solutions at the temperature
T = 298.15 K. We found that the PRV method [6,7] was the most
reliable method for measuring absolute volatilities. However,
other experimental techniques are required to accurately mea-
sure the partition coefficients at higher temperatures. Following
the recommendations of Raal and Mühlbauer [13], we decided
to use here a dynamic method with a Gillespie-like recirculat-
ing still [8,9] to measure the vapour–liquid partition coefficients
of the aroma compounds. The system is fully described in the
experimental part of this paper. It has already been shown that
this procedure is feasible for measuring the partition coefficients
of compounds infinitely diluted in hydroalcoholic solutions [18].
The design of a distillation process involving aroma com-
pounds requires not only some experimental data but also a
fast and accurate thermodynamic model for the determination
of the vapour–liquid equilibria. The thermodynamic models
that are applicable to aroma systems can be classified into two
main categories: the equations of state for polar and associating
compounds [19–22], and the excess Gibbs energy (or activity
coefficient) models [23–28]. One should mention the hydration
Gibbs energy models [29,30] that were developed for the spe-
cific case of organic compounds at infinite dilution in water.
Molecular simulation can also be very useful to predict the PVT
properties of aroma compounds [31] or Henry’s constants via
the Widom test particle insertion method [32]. Macroscopic
approaches and molecular simulation can be complementary,
for example, to understand highly non-ideal phase behaviours
such as the evolution of the solubility of hydrocarbons in water
with respect to temperature and alkyl chain length [33]. Equi-
librium data for systems involving aroma compounds are scare;
thus, predictive models that do not require any adjustment of
parameters on experimental data are of crucial importance. Pre-
dictive models can be developed from a group contribution (GC)
method, which consists in empirically linking the parameters
of the thermodynamic model to the numbers of the various
chemical groups constituting the molecules. GC methods have
been developed both for equations of state [34–37] and activ-
ity coefficient models [38–41]. Faùndez et al. [28] recently
employed the PSRK [35] equation of state and the UNIFAC
activity coefficient model [38] to predict the phase equilib-
ria of aromas + ethanol + water mixtures. The main advantages
of GC methods are their simplicity and efficiency. However,
GC approaches are limited to the types of molecules that were
involved in the adjustment of the GC parameters, and they may
be inaccurate for conditions ranges that were not considered in
the fitting procedure.
A second way to develop predictive macroscopic models
is to somehow involve quantum-chemical calculations [42,43].
One possible approach is to use an equation of state based
on an intermolecular potential fitted to ab initio data. Such
an approach is very similar to the methodology employed in
classical simulation to develop new forcefields, and can pro-
vide very accurate predictions for spherical molecules [44].
However, this approach is still limited to simple cases as the
development of general theories that would take all kinds of
interactions (electrostatic, intramolecular, site–site and polariz-
ability) into account is very difficult. Another strategy to involve
first principle calculations consists in relating the parameters of
a macroscopic thermodynamic model to ab initio calculations
performed on the studied molecules. These parameters are the
same for all molecules and they are determined by adjusting
a large set of vapour–liquid and liquid–liquid (LLE) equilib-
ria data. The only remaining inputs that characterize a given
mixture are the ab initio data. The COSMO-RS activity coef-
ficient model and its later versions [25–27] were developed
based on the above idea. A review on continuum solvation mod-
els that use quantum-mechanical calculations has been recently
proposed by Tomasi et al. [45]. A detailed description of the
COSMO-RS model and a review of all the possible applica-
tions can be found in a book by Klamt [46]. In the COSMO
approach molecular interactions are represented by interactions
between surface segments. These surface segments are deter-
mined by the way surface charges are distributed over the cavity
surface surrounding each molecule in the mixture. Each com-
pound is characterized by ab initio calculations performed on
one molecule solvated in a conductor. These calculations pro-
vide the molecular volume, the molecular surface area, and the
so-called “sigma profile” that corresponds to the distribution of
surface charges on the cavity surface. Lin and Sandler [27,47]
proposed the COSMO-SAC (conductor-like screening model-
segment activity coefficient) model, and an extensive database of
ab initio calculations corresponding to this model was developed
(the VT-2005 database [48]). A comparison between COSMO-
like approaches and the UNIFAC group contribution model was
recently made by Grensemann and Gmehling [49]. Mu et al.
[50] performed an extensive analysis of the performances of
the COSMO-RS and COSMO-SAC approaches, using different
ab initio calculation packages for the generation of the sigma
profiles.
Since we are interested only in phase equilibria at low pres-
sures, we decided to use two activity coefficient approaches:
the non-random two liquid (NRTL) model [23], which is very
simple, fast, and implemented in most process simulation soft-
wares, and the predictive COSMO-SAC model [27,47]. This
paper is organised as follows: we first describe the exper-
imental apparatus and present the new VLE measurements.
We then use the NRTL model to represent the new experi-
mental data, and investigate the prediction capabilities of the
COSMO-SAC model on our experimental data and on other
 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154
Fig. 1. Flowchart of the vapour–liquid equilibrium apparatus Labodest® (1)
electrical immersion heater; (2) mixing chamber; (3) contact path; (4) separation
chamber; (5) solenoid valves; (6) vapour phase; (7) liquid phase; (8 and 9)
circulation steams; (10) sampling port.
data from literature, for extended pressure and temperature con-
ditions.
2. Experimental
2.1. Materials
The hydroalcoholic solutions were prepared by weighting
de-ionized water (Milli-Q system, Millipore Waters, France)
and pure ethanol (ethanol 99.9%, RPE, Carlo Erba) and mixing
them to obtain the adequate %v/v concentrations according to
the alcoholometric tables [51]. The 13 aroma compounds stud-
ied in this work are presented in Table 1. These molecules are
known for their role in the sensory quality of brandies. We give in
Table 1 their CAS number and some of their physical properties
(molecular weight (MW), boiling point (Tb ), and hydrophobicity
constant at 298.15 K (Kow )).
2.2. Apparatus and experimental procedures
Vapour–liquid equilibrium measurements were performed in
a recirculation still (Fig. 1). During the measurements the vapour
generated from the boiling liquid is conveyed to a separate
141
Fig. 2. Vapour–liquid equilibria of the binary mixture ethanol + water at pres-
sure P = 0.1013 MPa. (a) Temperature–composition and (b) y–x diagrams. The
circles represent the experimental data reported in the DECHEMA databank
[58]. The asterisks denote the new experimental data (Tables 6 and 7), and the
lines correspond to the NRTL model.
condensing surface, and then returns to the pot. After some cir-
culation time, the system reaches equilibrium and small samples
of the liquid and condensed vapour are collected for analysis.
Evaporation is carried out by means of an electrical immersion
heater (1) made of quartz glass, which is arranged concen-
trically in a flow heater. An even circulation of both phases
with simultaneous magnetic stirring of the reflowing circula-
tion streams in a special mixing chamber (2) ensures a quick
equilibrium adjustment. Before entering the separation cham-
ber, the vapour–liquid mixture goes through a lengthened contact
path (3) that maximizes the mass transfer between both phases.
The construction of the separation chamber (4) safely prevents
the carrying of liquid particles into the vapour phase. Sample
take-off from the vapour (6) and liquid (7) phases into remov-
able receivers is performed via manually controlled solenoid
valves (5). Both samples may also be directly extracted from the
142 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154

Table 1
Physical properties of the aroma compounds studied in this work: molecular weight (MW), boiling point (Tb ), log10 of the octanol/water partition coefficient at
T = 298.15 K (log10 Kow )

Family CAS Name Supplier Purity (%) MW Tb (K)a log10 Kow b

78-83-1 Isobutanol Sigma 99.5 74.12 380.8 0.76

60-12-8 2-Phenylethanol Aldrich ≥99 112.17 491.4 1.36
Alcohols 123-51-3 Isopentanol Sigma ≥99 88.15 404.3 1.16
928-96-1 cis-3-Hexenol Sigma ≥99 100.16 429.5 1.61
78-70-6 Linalool Fluka ≥99 154.25 471.8 2.97
98-01-1 Furfural Fluka ≥99 96.09 434.7 0.41
Adehydes
78-84-2 Isobutanal Aldrich ≥99 72.11 336.0 0.74
141-78-6 Ethyl acetate Aldrich 99.8 88.11 350.2 0.73
123-92-2 Isopentyl acetate Aldrich ≥99 130.19 414.0 2.25
103-45-7 2-Phenylethyl acetate Aldrich 99 164.21 511.7 2.30
Esters
123-66-0 Ethyl hexanoate Aldrich ≥99 144.21 441.2 2.83
106-32-1 Ethyl octanoate Aldrich ≥99 172.27 480.2 3.81
110-38-3 Ethyl decanoate Aldrich ≥99 200.32 518.2 4.79
a Ref: NIST webbook [55] and PhysProp database [56].
b Ref: PhysProp database [56].

circulation streams (8 and 9) by means of a syringe. A direct
gaseous sampling of the vapour phase is possible by means
of another sampling port (10). The apparatus is designed for
a charge of approximately 35 mL and can be operated at con-
trolled atmospheric pressure. Equilibrium is reached after 5 h
of recirculation. The time needed to reach equilibrium has been
checked by following the time evolution of the temperature of the
vapour phase, and by sampling the circulation streams (8) and (9)
for the analysis of ethanol and aroma compound compositions.
The measurements of the compositions of the volatile com-
pounds in the coexistent liquid and condensed-vapour phases
were done by gas chromatography (GC) through a direct injec-
tion in a CPG Agilent 6890 equipped with a splitless injector
at 220 ◦ C, an auto sampler (Combipal, Varian) and a flame
ionisation detector (250 ◦ C; H2 : 40 mL/min; air: 450 mL/min;
makeup gas He: 33 mL/min). A 19095N-123 INNOWAX col-
umn (30 m × 0.53 mm × 1 ␮m from J&W scientists) was used
for the analysis. The temperature program was as follows: an
initial temperature of 50 ◦ C, raised at 4 ◦ C/min to 70 ◦ C, then
raised at 5 ◦ C/min to 170 ◦ C, then raised at 8 ◦ C/min to 220 ◦ C,
and finally held for 6 min at 220 ◦ C, giving a total running
time of 37.6 min. A sample volume of 1.5 ␮L was injected
in the column. The chromatographic data were obtained with
the Hewlett-Packard Chemstation software from Agilent. The
volatile compounds were quantified with reference to a calibra-
tion table established with pure standard compounds diluted in
ethanol at the concentration range suitable for the analysis of
the samples. The quantification of ethanol was done through
high-performance liquid chromatography (HPLC) analysis on
a P680 HPLC Pump, DIONEX equipment. Twenty microliters
of the samples at 4 ◦ C are injected in an ion exchange column
(300 mm × 7.8 mm, AMINEX HPX-87H, Bio-rad) maintained
at 35 ◦ C, with a flow rate of the mobile phase (H2 SO4 0.01N) set
0.6 mL/min. The HPLC equipment is coupled to a refractometer
detector (410 Differential Refractometer, WatersTM ), with pro-
pionic acid as the internal standard. Using a correlation between
the concentration in mol/L of ethanol and the molar volume of
the sample, which was established from molar volume tables
of the ethanol–water mixture (we neglected the influence of the
aromas on the molar volumes of the samples), and knowing the
volume of the samples and the amount of ethanol from HPLC
measurements, we could determine the number of moles of the
samples and the ethanol mole fraction. Knowing the amount
of each aroma compound from GC measurements, we could
determine their mole fraction. The mole fraction of water was
obtained from subtraction. The measurements were repeated
three times for each sample. Coefficients of variation (CV) are
reported.
3. Thermodynamic models
Since the pressure of the studied systems is low (atmospheric
pressure), and the temperature always lower than the critical tem-
peratures of the pure compounds, one can represent the vapour
phase as an ideal gas mixture, and assume that the activity coeffi-
cients in the liquid phase do not depend on pressure. Two activity
coefficient models are employed in the current work: the widely
used non-random two liquid model [23], and the fully predic-
tive COSMO-SAC approach. The vapour–liquid equilibrium is
determined from the equality of the fugacities in both phases,
i.e.
yi P = xi Pi0 γi , (1)
where P is the pressure of the system, xi and yi the mole frac-
tions of component i in the liquid and vapour phases, and yi
is the activity coefficient of component i in the liquid phase,
which represents the deviation from the ideal solution. Through-
out this paper ethanol is component 1, water component 2, and
the components i = 3, . . ., 15 are the aroma compounds reported
in Table 1. The vapour pressure Pi0 of pure component i is given
in units of MPa by the empirical relation
 
0.1013 Bi Ei
Pi =
0
exp Ai + + Ci ln T + Di T , (2)
760 T
 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154 143

Table 2
Constants used in Eq. (2) for computing the pure compound vapour pressures in MPa at a given temperature T in K
Component Ai Bi Ci Di Ei

Isobutanol 182.89723 −12955 −24.285 1.4262 × 10−5 2

2-Phenylethanol 60.18323 −8985.1 −5.7034 4.6087 × 10−18 6
Isopentanol 102.12723 −10237 −11.695 6.8003 × 10−18 6
cis-3-Hexenol 22.06249 −6629.9145 0 0
Linalool 125.57 −12294 −15.084 0
Furfural 89.67723 −8372.1 −11.13 0.008815 1
Isobutanal 60.51523 −5951.2 −6.2321 2.227 × 10−17 6
Ethyl acetate 61.93123 −6227.6 −6.41 1.7914 × 10−17 6
Isopentyl acetate 77.022868 −8124.2596 −8.43261 0
2-Phenylethyl acetate 20.72652 −7211.494 0 0
Ethyl hexanoate 50.2368 −7232.1865 −4.46056 0
Ethyl octanoate 77.8530 −9296.36 −8.52379 3.40699 × 10−6 2
Ethyl decanoate 89.8632 −10764.853 −10.1679 4.3498 × 10−6 2
Water 68.75623 −7258.2 −7.3037 4.1653 × 10−6 2
Ethanol 69.58223 −7164.3 −7.327 3.134 × 10−6 2

where T (in K) is the temperature of the system. The constants Ai ,
Bi , Ci , Di , and Ei are given in Table 2. These constants were taken
from the literature [52,53] for most compounds. For linalool
we fitted the constants Ai , Bi , Ci , Di , to the experimental data
available in the literature [54]. The vapour pressures of linalool
are represented by Eq. (2) with an average absolute deviation
(AAD) of 2.8%, which is consistent with experimental deviation.
For cis-3-hexenol and 2-phenylethyl acetate no experimental
vapour pressure data are available apart from the boiling point
Tb and the vapour pressure at ambient temperature. The boiling
point Tb is available in the literature [55] for all the studied
compounds. If the experimental vapour pressure at T = 298.15 K
of an aroma compound is not available, we estimate it using a
group contribution method available in the Physprop database
[56,57]. One can then determined Ai and Bi with these two state
points to approximate the evolution of the vapour pressure with
T via Eq. (2).
3.1. Non-random two liquid model
The NRTL model [23] is based on a local composition
approach. The NRTL activity coefficient yi of component i in
mixture of n compounds is given by
n
j=1 τji φji xj
ln γi = n +
k=1 φki xk
n  n  number of aroma + water binary systems (water + furfural,
 xj φij
n k=1 xk τkj φkj
τij −  n , (3)
j=1 k=1 φkj xk k=1 φkj xk
where τ ij characterizes the interactions between compounds i
and j, φij = (−αij τ ij ); αij are called the non-randomness binary
parameters. It is well known that the NRTL model is very flexible
in terms of data adjustments, but that the NRTL binary param-
eters are highly correlated. Thus, it is necessary to reduce the
total number of parameters as much as possible before adjusting
them on the new experimental data. For the sake of simplic-
ity and following the recommendations of Renon and Prausnitz
[23], we first assume that αij = αji = 0.3 for all pair interactions
i, j. Second, we fix τ ij = τ ji = 0 when components i and j are both
aroma molecules, as these parameters do not have any influence
on the Ki at infinite dilution. Note that the NRTL parameters
between the aroma compounds could be determined by match-
ing the NRTL model to the predictions of the COSMO-SAC
model. We assume the following temperature dependence for
the non-zero τ ij
aij + bij
τij = (4)
T + cij T
where aij is a reduced parameter, bij and T are expressed in K
and cij in K−1 . We suppose that cij = cji = 0 when component
i is an aroma compound and component j a solvent molecule
(water or ethanol) in order to limit the total number of adjusted
parameters. The six NRTL parameters (a12 , a21 , b12 , b21 , c12 and
c21 ) characterizing the ethanol–water interactions are adjusted
to a large variety of thermodynamic properties: vapour–liquid
data [58], excess enthalpy [59], azeotropic compositions [58]
and activity coefficients of ethanol at infinite dilution in pure
water [60].
The binary parameters aij and bij for the aroma–solvent
interactions are fitted to the following thermodynamic prop-
erties: new Ki experimental data, Henry’s constants in
pure water [17,53,55,61] (partition coefficients at infinite
dilution in water at T = 298.15 K), VLE data [58] for a
water + isobutanol, water + isopentanol, water + ethyl acetate,
and water + isobutanal), and pseudo-experimental data gener-
ated by COSMO-SAC. Some COSMO-SAC predictions were
indeed involved in the adjustment of the NRTL parame-
ters for the aroma–solvent interactions, as the amount of
experimental data was not large enough: the generated data
are activity coefficient data for the binary mixtures aroma
i + water and aroma i + ethanol, for the temperature range
283.15 < T < 373.15. This methodology that consists in adjust-
ing the parameters of an activity coefficient model such as
NRTL to COSMO predictions was suggested by Klamt [46]. The
144 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154

Table 3 to Lin and Sandler’ paper [27] for further details about the the-
NRTL binary parameters for the ethanol (component 1) + water (component 2) ory. In the COSMO-SAC approach, the activity coefficient γ i of
mixture
component i in the solution S is given by
a12 −4.46256 
b12 930.8739 ln γi = ln γiSG + ni pi (σm )[ln ΓS (σm ) − ln Γi (σm )]. (5)
a21 19.27804 σm
b21 −3376.469
c12 4.7291 × 10−3 The sum in Eq. (5) is the restoring free energy contribu-
c21 −2.2083 × 10−2 tion, and ln γiSG the Staverman–Guggenheim combinatorial term
The parameters aij are adimensional; the parameters bij are given in K and the
[1,2] given by
parameters cij in K−1 . n
ζi z θi ζi 
ln γiSG = ln + qi ln + li − xj lj , (6)
xi 2 ζi xi
NRTL parameters were optimised with the following three-step j=1

procedure: n
where ζi = xi ri / j=1 xj rj is the volume fraction occupied by

component i in the mixture, θi = xi qi / nj=1 xj qj the surface
• Step 1. The four NRTL parameters a1i , b1i , ai1 , bi1 for each fraction of component i, li = z(ri − qi )/2 + 1 − ri , and z = 10. The
ethanol–aroma i binary mixture are optimised on activity parameters ri = vmolec /vref and qi = Amolec /Aref are the nor-
i i
coefficient data generated by COSMO-SAC over the whole malized volume and surface area of molecule i, vmolec and
i
composition range and for 283.15 < T < 373.15. Amolec are the volume and surface area of molecule i, determined
i
• Step 2. The four NRTL parameters a2i , b2i , ai2 , bi2 for 3
from COSMO ab initio calculations, and vref = 66.69 Å ,
each water–aroma i binary mixture are optimised on activ-
Aref = 79.53 Å2 .
ity coefficient data generated by COSMO-SAC over the full
The restoring free energy contribution is the difference
composition range and for 283.15 < T < 373.15. Steps 1 and 2
between the restoring free energies in the liquid mixture and in
provide some good initial guesses for step 3.
pure liquid i; ni = Amoleci /Aeff is the number of surface segment
• Step 3. The eight NRTL parameters (a1i , b1i , ai1 , bi1 ,
a2i , b2i , ai2 and bi2 ) are reoptimised simultaneously by defined with an effective surface area Aeff = 7.5 Å2 ; Γ s (σ m ) and
adjusting all the available data (experimental VLE data Γ i (σ m ) are the “segment activity coefficients” of segment m of
and COSMO-SAC predictions) for the binary mixtures charge density σ m in the liquid mixture S and in the pure liquid i,
ethanol–aroma i and water–aroma i, and for the ternary sys- respectively. Each molecule i is characterized by its sigma profile
tem ethanol–water–aroma i. Since we are more interested pi (σ m ), which is the normalized distribution of charge densities.
in the representation of the infinite dilution region, a higher The distribution pi is given by pi (σm ) = Ai (σm )/Amolec i , where
weight was used for the deviations from the new Ki data in Ai (σ m ) is the
 total area of the segments of charge density σ m , and

the calculation of the error function. Amolec

i = A
σm i m(σ ). The A (σ
i m ) are histograms determined
over the charge density range −0.025 E Å−2 ≤ σ ≤ 0.025 E Å−2
Our procedure aims at generating NRTL parameters that from the COSMO ab initio calculations. The method that we
allow for a good representation of the new data as well as some use here to determine these histograms is similar to the origi-
reasonable predictions far from the infinite dilution region, and nal method
√ used by Lin and Sandler [27,47] (with the constant
outside the temperature ranges of the new data. The optimised reff = Aeff /π ≈ 1.545 Å in Eq. (15) of Ref [27]); thus, the
NRTL binary parameters are reported in Tables 3–5. method is slightly different from the method proposed by Old-
land [62] and Mullins et al. [48], where reff = 0.81764 Å. The
segment activity coefficients Γ i (σ m ) and Γ s (σ m ) are given by
3.2. COSMO-SAC model 
  
W(σm , σn )
The COSMO-SAC model was developed by Lin and Sandler ln Γi (σm ) = −ln pi (σn )Γi (σn ) exp − ,
σn
RT
[27]. It is based on the COSMO-RS (COSMO for real solvents)
(7)
model developed by Klamt [25]. COSMO-SAC and COSMO-
RS solvation models are both fully predictive as the only inputs where the index i stands for all components i and for the mixture
that characterise the studied systems are ab initio data. These S. W(σ m , σ n ) is called the exchange energy. The normalized
models are, like the NRTL model, excess Gibbs free energy distribution of surface charge densities in the mixture, ps (σ m ),
models that provide pressure independent activity coefficients in is given by
liquid phases as a function of temperature T and composition. In n
xi Ai (σm )
this work, we decided to use the COSMO-SAC model, as a broad pS (σm ) = i=1n . (8)
i=1 xi Ai
molec
database of ab initio inputs is available (VT-2005 database [48]).
We also calculated the ab initio input data for some aroma com- The exchange energy W is given by
pounds, by following the recommendations given in Ref [48].
These input data were obtained from the DMol3 package and α 2
W(σm , σn ) = (σ + σn2 ) + chb max(0, σacc − σhb )
saved as “COSMO” files. We provide here only the main work- 2 m
ing equations of the COSMO-SAC model. The reader is directed × min(0, σdon + σhb ), (9)
 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154 145

Table 4
NRTL binary parameters (a1k , b1k , ak1 and bk1 ) between each aroma compound k and ethanol (component 1)
Component k a1k b1k ak1 bk1

Isobutanol 0.29009 −48.6777 −0.14528 −0.34519

2-Phenylethanol −0.80752 0.571528 1.26227 −90.2087
Isopentanol 1.23797 1.3002 × 10−3 −0.87069 −10.0883
cis-3-Hexenol 2.2319 −32.3306 −1.72789 7.35639
Linalool 1.49749 11.9525 −0.89923 −17.8563
Furfural −0.46010 333.133 0.664283 −199.642
Isobutanal −0.45975 3.11884 0.677631 211.745
Ethyl acetate 0.717856 −116.724 −0.58208 260.883
Isopentyl acetate 0.932778 110.101 −0.42258 3.29846
2-Phenylethyl acetate 6.336 × 10−2 789.711 9.8503 × 10−2 −387.255
Ethyl hexanoate 0.416565 950.249 −0.45549 −318.425
Ethyl octanoate 6.41828 −935.268 −2.79184 446.008
Ethyl decanoate 8.10392 −1032.33 −2.17787 95.9156

The parameters aij are adimensional; the parameters bij are given in K.

where α = 16465.9809 and chb = 85,580, both in units of
kcal mol−1 Å4 E−2 , σ hb = 0.0084 E A−2 , σ acc = max(σ m , σ n ),
and σ don = min(σ m , σ n ). The system of Eq. (7) is solved iter-
atively with the Picard substitution procedure, and a damping
factor is used to help convergence. The activity coefficients γ i
are obtained from Eq. (5), and phase equilibrium is determined
from Eq. (1).
4. Results and discussions
4.1. Ethanol–water binary mixture
The new measurements of the temperature and compositions
of the liquid and vapour phases are given in Tables 6 and 7,
and compared to the vapour–liquid experimental data [58] for
the ethanol + water binary mixture at P = 0.1013 MPa. Since the
aroma compounds are at very low concentration, their influence
on the phase equilibria of the ethanol + water systems is neg-
ligible: in Fig. 2, one can observe that the new measurements
of liquid compositions and saturated temperature are in good
agreement with the literature data [58]. The deviations between
the new measurements of the ethanol gas phase composition
and the literature data [58] are lower than 18%. The maximum
temperature deviation at fixed liquid composition is of about
1.5 K. The deviations could be due to some problems inherent
to the recirculation method (partial condensation, pressure and
temperature fluctuations) or due to the uncertainty in the anal-
ysis of the ethanol composition. As shown in Section 4.2 the
new experimental partition coefficients Ki at infinite dilution
are consistent with the VLE experimental data reported in the
DECHEMA database [58], and are in good agreement with the
COSMO-SAC predictions. This confirms the reliability of the
new Ki data.
We adjusted the NRTL binary parameters for the
ethanol–water mixture to a large variety of experimen-
tal data: vapour–liquid data over the temperature range
283 K ≤ T ≤ 409 K [58], excess enthalpy [59], azeotropic com-
positions [58] and activity coefficients of ethanol at infinite

Table 5
NRTL binary parameters (a2k , b2k , ak2 and bk2 ) between each aroma compound k and water (component 2), and deviations between the NRTL calculations and the
new experimental data (Ki values)
Component k a2k b2k ak2 bk2 AAD%

Isobutanol 4.73887 −348.691 −1.3234 × 10−2 115.444 11.2

2-Phenylethanol 7.17463 −870.089 −0.41407 332.749 16.6
Isopentanol 3.04081 839.302 4.24 × 10−3 −62.0967 14.8
cis-3-Hexenol 1.55152 1308.53 −8.359 × 10−2 310.344 21.5
Linalool 8.17274 −222.044 1.19173 352.438 24.7
Furfural 3.62651 −33.4298 0.151363 187.715 12.3
Isobutanal 2.74545 24.3192 0.565063 320.153 21.1
Ethyl acetate 4.99657 −514.015 −6.8057 × 10−3 376.487 13.3
Isopentyl acetate 3.06969 1125.41 −0.1277 1227.96 19.7
2-Phenylethyl acetate 2.19621 2030.37 −2.09879 2068.22 18.2
Ethyl hexanoate 4.33504 1242.08 0.417001 1249.57 28.1
Ethyl octanoate 0.9029 2997.17 6.68 × 10−2 1253.61 28.7
Ethyl decanoate 7.65251 1560.77 3.1600 193.001 27.2

The parameters aij are adimensional; the parameters bij are given in K.The last column provides the absolute average deviations from the new Ki data. AAD: average

absolute deviation, at fixed T and liquid composition: AAD = (1/N) i
|Ki,cal − Ki,exp |/Ki,exp × 100.
146 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154

Table 6
Experimental mole fractions of all the compounds in the liquid phase at equilibrium, and experimental temperatures T of the system
Components Temperatures (K)

366.41 362.06 357.24 356.98 355.92 354.81 353.98 352.88 351.90

Liquid phase mole fractions xi

Ethanol 0.0207 0.0623 0.1444 0.1553 0.1952 0.2817 0.4114 0.4739 0.6326
Water 0.9793 0.9377 0.8556 0.8447 0.8048 0.7183 0.5886 0.5261 0.3674
Isobutanol 1.76 × 10−5 2.79 × 10−5 4.18 × 10−5 4.22 × 10−5 4.36 × 10−5 5.55 × 10−5 6.04 × 10−5 6.46 × 10−5 7.73 × 10−5
2-Phenylethanol 6.15 × 10−6 8.75 × 10−6 1.04 × 10−5 1.03 × 10−4 9.51 × 10−6 1.16 × 10−5 1.24 × 10−5 1.41 × 10−5 1.67 × 10−5
Isopentanol 2.68 × 10−5 4.50 × 10−5 6.70 × 10−5 6.82 × 10−5 6.94 × 10−5 8.81 × 10−5 1.02 × 10−4 1.02 × 10−4 1.23 × 10−4
cis-3-Hexenol 8.30 × 10−8 1.25 × 10−7 1.61 × 10−7 1.64 × 10−7 1.62 × 10−7 2.06 × 10−7 2.21 × 10−7 2.48 × 10−7 2.85 × 10−7
Linalool 3.27 × 10−7 1.24 × 10−6 2.48 × 10−6 2.48 × 10−6 2.57 × 10−6 3.26 × 10−6 3.52 × 10−6 3.74 × 10−6 4.51 × 10−6
Furfural 1.88 × 10−6 2.49 × 10−6 3.25 × 10−6 3.29 × 10−6 3.24 × 10−6 3.91 × 10−6 4.29 × 10−6 4.36 × 10−6 5.06 × 10−6
Isobutanal 6.45 × 10−7 8.69 × 10−7 2.94 × 10−6 2.23 × 10−6 2.61 × 10−6 3.89 × 10−6 4.03 × 10−6 6.72 × 10−6 6.33 × 10−6
Ethyl acetate 2.01 × 10−6 3.63 × 10−6 1.10 × 10−5 1.17 × 10−5 1.24 × 10−5 2.03 × 10−5 2.32 × 10−5 2.62 × 10−5 2.90 × 10−5
Isopentyl acetate 2.03 × 10−7 5.45 × 10−7 2.27 × 10−6 2.40 × 10−6 2.89 × 10−6 4.29 × 10−6 5.30 × 10−6 5.22 × 10−6 6.44 × 10−6
2-Phenylethyl acetate 7.62 × 10−7 1.90 × 10−6 2.04 × 10−6 2.43 × 10−6 2.28 × 10−6 2.79 × 10−6 2.99 × 10−6 3.24 × 10−6 3.90 × 10−6
Ethyl hexanoate 5.69 × 10−8 2.99 × 10−7 1.62 × 10−6 1.70 × 10−6 2.07 × 10−6 2.70 × 10−6 1.00 × 10−6 3.72 × 10−6 4.32 × 10−6
Ethyl octanoate 5.28 × 10−9 4.16 × 10−8 3.02 × 10−7 2.96 × 10−7 3.38 × 10−7 4.48 × 10−7 4.76 × 10−7 4.98 × 10−7 5.93 × 10−7
Ethyl decanoate 1.00 × 10−9 4.86 × 10−8 4.18 × 10−7 3.12 × 10−7 3.37 × 10−7 4.09 × 10−7 3.91 × 10−7 4.46 × 10−7 5.33 × 10−7

Table 7
Experimental mole fractions of all the compounds in the vapour phase at equilibrium
Components Temperatures (K)

366.41 362.06 357.24 356.98 355.92 354.81 353.98 352.88 351.90

Vapour phase mole fractions yi

Ethanol 0.2776 0.4232 0.5384 0.553 0.5479 0.6033 0.6407 0.6122 0.707
Water 0.7224 0.5768 0.4616 0.447 0.4521 0.3967 0.3593 0.3878 0.293
Isobutanol 4.58 × 10−4 4.00 × 10−4 2.02 × 10−4 1.75 × 10−4 1.23 × 10−4 8.61 × 10−5 5.81 × 10−5 2.80 × 10−5 3.62 × 10−5
2-Phenylethanol 6.40 × 10−6 3.31 × 10−6 7.48 × 10−7 6.17 × 10−7 3.04 × 10−7 1.94 × 10−7 1.39 × 10−7 2.08 × 10−7 1.46 × 10−7
Isopentanol 7.16 × 10−4 6.81 × 10−4 2.29 × 10−4 2.00 × 10−4 1.19 × 10−4 7.11 × 10−5 4.43 × 10−5 1.88 × 10−5 2.46 × 10−5
cis-3-Hexenol 1.18 × 10−6 8.65 × 10−7 2.85 × 10−7 2.43 × 10−7 1.38 × 10−7 7.77 × 10−8 3.87 × 10−8 1.09 × 10−8 –
Linalool 2.23 × 10−5 2.66 × 10−5 5.96 × 10−6 4.84 × 10−6 1.99 × 10−6 7.63 × 10−7 3.83 × 10−7 1.22 × 10−7 1.54 × 10−7
Furfural 9.84 × 10−6 7.26 × 10−6 4.01 × 10−6 3.59 × 10−6 2.64 × 10−6 1.83 × 10−6 1.36 × 10−6 7.18 × 10−7 1.14 × 10−6
Isobutanal 4.97 × 10−5 6.25 × 10−5 5.71 × 10−5 5.72 × 10−5 5.16 × 10−5 4.32 × 10−5 3.23 × 10−5 2.15 × 10−5 2.54 × 10−5
Ethyl acetate 2.11 × 10−4 3.05 × 10−4 2.67 × 10−4 2.64 × 10−4 2.06 × 10−4 1.79 × 10−4 1.42 × 10−4 7.47 × 10−5 9.12 × 10−5
Isopentyl acetate 4.32 × 10−5 7.12 × 10−5 4.79 × 10−5 4.64 × 10−5 2.56 × 10−5 1.50 × 10−5 9.02 × 10−6 3.52 × 10−6 3.68 × 10−6
2-Phenylethyl acetate 1.14 × 10−5 7.85 × 10−6 1.59 × 10−6 1.29 × 10−6 5.53 × 10−7 2.36 × 10−7 1.50 × 10−7 6.29 × 10−8 9.50 × 10−8
Ethyl hexanoate 2.16 × 10−5 4.37 × 10−6 3.47 × 10−5 3.02 × 10−5 1.60 × 10−5 7.74 × 10−6 4.11 × 10−6 1.41 × 10−4 1.38 × 10−6
Ethyl octanoate 1.47 × 10−6 9.64 × 10−6 4.10 × 10−6 3.15 × 10−6 1.30 × 10−6 4.64 × 10−7 2.19 × 10−7 6.84 × 10−8 6.98 × 10−8
Ethyl decanoate 6.17 × 10−7 1.19 × 10−5 3.41 × 10−6 2.15 × 10−6 6.82 × 10−7 1.76 × 10−7 8.49 × 10−8 3.36 × 10−8 2.76 × 10−8

Table 8
Coefficients of variation (CV%) of the partition coefficients Ki due to the gas chromatography (GC) analysis (determination of CV from three GC analysis per
sample)
Components Temperatures (K)

366.41 362.06 357.24 356.98 355.92 354.81 353.98 352.88 351.90

CV%
Isobutanol 0.6 0.6 1.8 0.3 0.2 3.0 1.5 0.9 2.3
2-Phenylethanol 2.7 3.5 11.0 2.9 4.6 4.4 0.8 4.0 0.6
Isopentanol 0.6 0.6 1.8 0.3 0.2 3.0 1.5 0.9 2.3
cis-3-Hexenol 0.5 1.1 2.7 0.4 0.5 0.5 2.0 1.0
Linalool 0.6 2.2 2.3 1.6 0.4 1.0 1.4 1.7 0.4
Furfural 1.4 3.3 3.4 0.8 1.5 2.4 0.3 10.3 2.2
Isobutanal 3.7 5.1 3.9 2.1 1.0 0.7 6.7 5.7 5.7
Ethyl acetate 0.6 0.7 0.7 0.8 3.4 3.3 2.0 1.3 2.6
Isopentyl acetate 6.1 1.5 1.2 4.6 3.0 1.3 1.8 1.4 1.5
2-Phenylethyl acetate 1.2 2.9 4.8 1.4 5.3 5.1 0.4 2.0 2.8
Ethyl hexanoate 2.1 0.6 0.3 1.9 2.0 1.4 2.6 1.3 2.2
Ethyl octanoate 5.7 5.4 4.5 1.3 4.3 2.2 4.4 3.4 2.9
Ethyl decanoate 2.4 6.7 1.0 4.8 5.4 3.4 2.2 3.7 4.2
 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154 147

Table 9
Henry’s constants Hi of the aroma compounds in pure water at T = 298.15 K, expressed in MPa with respect to the aroma mole fraction: Hi is related to the partition
coefficient Ki of aroma i at infinite dilution as Hi = PKi , where P is the vapour pressure of water at T = 298.15 K
Compounds COSMO-SAC COSMO file NRTL Exp. or Est.a Ref. data

Isobutanol 0.128 VT-2005 0.0691 0.0680 [53]

2-Phenylethanol 0.00226 VT-2005 0.00144 0.00144 [56]
Isopentanol 0.113 new, rc = 3.7 Å 0.120 0.121 [17]
cis-3-Hexenol 0.0449 new, rc = 3.0 Å 0.0873 0.0873* [56]
Linalool 0.308 new, rc = 3.7 Å 0.144 0.120 [61]
Furfural 0.0149 new, rc = 3.0 Å 0.0187 0.0186 [53]
Isobutanal 1.12 VT-2005 0.969 0.969 [53]
Ethyl acetate 0.680 VT-2005 0.776 0.777, 1.08 [53,66]
Isopentyl acetate 1.42 VT-2005 2.55 2.55 [53]
2-Phenylethyl acetate 0.0217 new, rc = 3.0 Å 0.106 0.106* [56]
Ethyl hexanoate 2.94 new, rc = 3.0 Å 3.67 3.67, 4.02 [17,66]
Ethyl octanoate 2.97 new, rc = 3.0 Å 5.10 5.10, 7.15* [56,66]
Ethyl decanoate 2.92 new, rc = 3.0 Å 4.02 4.02, 12.7* [56,66]
AAD 44.9% 1.9%

The predictions of the COSMO-SAC and NRTL models are compared to the experimental data available from the literature [17,53,55,61], or to Henry’s constants
(PhysProp database [56]) estimated with the Meylan and Howard method [67] (data denoted with the symbol *). The COSMO ab initio calculation (COSMO files)
rc . The average absolute deviation
were taken from the VT-2005 database, or determined in this work with different values of the global cut-off (AAD) were calculated
for both models (the data with the symbol * were not included in the AAD calculation. AAD = (1/N) i
|Hi,cal − Hi,exp |/Hi,exp × 100.
a Estimated values with the Meylan and Howard method [67] (PhysProp database [56]).

dilution in pure water [60]. The resulting NRTL parameters are
reported in Table 3. We assumed that the ideal gas approxima-
tion for the vapour phase remains valid for all the adjusted state
points. The NRTL model represents well the VLE data [58] over
the entire temperature range, as shown in Figs. 2 and 3: at fixed
temperature and fixed liquid composition, the optimised NRTL
parameters given in Table 3 provides an absolute average devi-
ation of 0.77% for the pressures (200 data) and of 1.5% for the
ethanol mole fractions in the gas phase (180 data), over the range
283 K ≤ T ≤ 409 K. The predictions of the activity coefficient of
ethanol over the temperature range 270 < T < 370 K is also very
satisfactory (Fig. 4(b)). The NRTL model provides a reasonably
good representation of the experimental azeotropic composi-
tions [58] for the ethanol–water mixture, as shown in Fig. 3(b).
The representation of the excess enthalpy Hex with NRTL is
however less satisfactory, although the NRTL model predicts
well the sign and magnitude of Hex at T = 298 K (Fig. 4(a)).
We evaluated the performances of the COSMO-SAC
approach [27] combined with the VT-2005 database [48]
for the predictions of the thermodynamic properties of the
ethanol–water mixture. Note that we did not use the sigma pro-
files reported in the VT-2005 database, but we recalculated them
by using the VT-2005 COSMO-files and√employing the Lin
and Sandler methodology [27] with reff = Aeff /π (see Section
3.2). We first observed that the VLE predictions of the COSMO-
SAC + VT-2005 COSMO files model were unsatisfactory for
the water + ethanol mixture. Mullins from Liu’s research group
that developed the VT-2005 database [48] indicated us that this
problem is due to an insufficient optimisation of the molecular
geometry of ethanol, and provided us improved new ab initio
calculations for ethanol. Thus, we kept the ab initio calcula-
tion for water from the VT-2005 database, and used the new
ab initio data for ethanol. The new ab initio data provide much
better VLE predictions, as shown in Fig. 5(a). One can see that
the new optimised geometry of the ethanol molecule is differ-
ent from the original one from VT-2005 (Fig. 5(b)): the ethanol
geometry from the VT-2005 database corresponds to a local
minimum of the energy, while the new geometry corresponds to
the global minimum. The influence of the molecular geometry
on the COSMO-SAC predictions were previously observed by
Mullins et al. [48]. It should be recalled that the parameters of
the COSMO-SAC approach were originally adjusted by using
the gas phase optimised geometry of the molecules. It is hence
reasonable that the new sigma profile of ethanol, which corre-
sponds to a more optimised geometry, gives rise to better VLE
predictions. The COSMO-SAC predictions of the azeotropic
compositions are also very satisfactory, as shown in Fig. 3(b).
However, the COSMO-SAC model does not correctly predict
either the excess enthalpy (Fig. 4(a)) or the activity coefficient
of ethanol infinitely diluted in pure water (Fig. 4(b)).

Table 10
Properties of ethyl esters at T = 298 K: vapour pressure (Pi0 ) predicted with Eq. (2), activity coefficient at infinite dilution (γi∞ ) and Henry constant Hi = Pi0 γi∞
(MPa) in pure water at 298.15 K predicted with COSMO-SAC

Components Pi0 (MPa) COSMO-SAC, γi∞ COSMO-SAC, Hi Exp., Hi

Ethyl hexanoate 2.35 × 10−4 1.25 × 104 2.94 4.0

Ethyl octanoate 2.72 × 10−5 1.09 × 105 2.97 5.1
Ethyl decanoate 3.02 × 10−6 9.66 × 105 2.92 4.0

The Henry constant experimental data from Aprea et al. [66] are reported in the last column of the table.
148 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154
Fig. 3. Vapour–liquid equilibria of the ethanol–water mixture. (a) Isotherms.
The symbols represent the vapour–liquid experimental data [58]. Sev-
eral isotherms were considered: (
) T = 283.15 K, (*) T = 293.15 K, (×) Fig. 4. Thermodynamic properties of the ethanol + water mixture. (a) Excess
T = 303.15 K, (䊉) T = 323.15 K and (♦) T = 343.15 K. (b) Azeotropic compo-
sitions as a function of temperature. The asterisks denote the experimental data
[58]. In (a) and (b) the solid lines correspond to the predictions of the COSMO-
SAC model and the dotted lines represent the curves calculated with the NRTL
model.
4.2. Vapour–liquid equilibria of the aroma compounds in
hydroalcoholic solutions
The vapour–liquid equilibria of the multicomponent
aroma + ethanol + water mixtures were measured with the
apparatus described in Section 2.2, and the mole fractions
of all compounds in the coexistent phases are reported in
Tables 6 and 7. In Table 8 we also report the experimental errors
(coefficients of variation, CV = standard deviation/average
value) on the partition coefficients coming from the gas chro-
matography analysis. The coefficients of variations, defined as
the ratio between the standard deviation and the average value,
were calculated based on three GC analysis per sample. The CV
that we report do not represent the global experimental error
that also comes from pressure, temperature and concentration
fluctuations. The evolutions of the partition coefficients of some
aroma compounds as a function of the ethanol amount in the
liquid phase are displayed in Fig. 6: the ethanol composition
is here expressed in terms of the alcoholic strength %v/v. The
enthalpy at 298.15 K. The circles correspond to a polynomial interpolation of
the experimental data [59]. (b) Activity coefficient at infinite dilution of ethanol
in pure water. The diamonds denote the experimental data [60]. In (a) and (b) the
dotted lines represent the NRTL calculations and the continuous lines represent
the predictions of the COSMO-SAC model.
conversion from mole fraction to alcoholic strength %v/v units
was performed via tables [51]. One can observe that the increase
of the ethanol proportion in the liquid phase decreases the abso-
lute volatility (Ki ) of all aroma compounds of several orders of
magnitude.
We adjusted the NRTL binary parameters between the aroma
compounds and the solvent molecules (water and ethanol), by
following the procedure described in Section 3.1. Experimental
data (new Ki data, VLE data, Henry’s constant) as well as pseudo
data generated with COSMO-SAC (activity coefficients) were
used to adjust the NRTL parameters. We recall that the Henry
constant Hi of solute i is defined as Hi = lim fi /xi , where fi is
xi →0
the fugacity of solute i. At equilibrium the fugacities are equal in
both phases, and the fugacity of the vapour phase can be approx-
imated by the partial pressure Pi = Pyi when the pressure P of
the system is low. As a result, the Henry constant of aroma i at
low pressure is related to the vapour–liquid partition coefficient
Ki at infinite dilution as Hi = PKi . Eq. (1) leads to Hi = Pi0 γi∞ .
 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154
Fig. 5. Effect of the molecular geometry of ethanol on the COSMO-SAC pre-
dictions. (a) Vapour–liquid equilibria of the binary mixture ethanol + water
at pressure P = 0.1013 MPa. The circles represent the experimental data from
the DECHEMA database [58]. The lines correspond to the predictions of the
COSMO-SAC model: the dotted lines correspond to the molecular geometry
(local minimum in energy) proposed in the VT-2005 database (b); the con-
tinuous lines correspond to the COSMO predictions with the new molecular
geometry (global minimum in energy) of ethanol (c), which was provided to us
by Prof. Y.A. Liu’s group.
Thus, the deviations between the theoretical predictions from
the experimental data can come either from the estimations of
Pi0 or from the predictions of γi∞ . To make easier the optimisa-
tion we adjusted the NRTL parameters for 1 aroma compound
Fig. 6. Vapour–liquid partition coefficients (Ki = yi /xi ) of aroma compounds
infinitely diluted in hydroalcoholic solution, as a function of the ethanol com-
position in the liquid phase. The ethanol composition is expressed in terms of
the alcoholic strength %v/v. The symbols represent the new experimental data
for (♦) isopentanol, () cis-3-hexenol, (*) linalool, () 2-phenylethanol, ()
isobutanal, () isopentyl acetate, (䊉) ethyl octanoate and (
) ethyl decanoate. Tables 4 and 5.
149
Fig. 7. Activity coefficients at T = 298.15 K predicted with the COSMO-SAC
and NRTL models for the binary mixtures (a) ethanol + ethyl acetate and (b)
water + ethyl acetate. The symbols denote the COSMO-SAC predictions: the
circles correspond to the predictions of the activity coefficient of ethyl acetate and
the crosses correspond to the predictions of the activity coefficient of the solvent
molecule (ethanol in (a) and water in (b)). The lines represent the corresponding
NRTL calculations.
at a time by considering ternary mixtures of water + ethanol + 1
aroma molecule and ignoring the presence of the other aroma
compounds. We can do this approximation because the concen-
trations of the aroma compounds in the studied solutions are
very low and do not have any influence on the Ki . This approx-
imation is justified at the end of this paper. Although we are
interested mainly in the infinite dilution region and in the temper-
ature range 353–373 K, we adjusted the NRTL parameters also
on pseudo-experimental data (activity coefficient data predicted
by COSMO-SAC) for temperatures between 283 and 373 K, in
order to make the NRTL model more reliable for compositions
far from the infinite dilution region and over a large temperature
range. As shown in Fig. 7 for ethyl acetate, the NRTL model
predicts reasonable values for the activity coefficients outside
the temperature range of the new data and for the whole com-
position range. The optimised NRTL parameters are reported in
150 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154

Fig. 8. Vapour–liquid partition coefficients of aroma compounds infinitely diluted in hydroalcoholic solution, as a function of the ethanol mole fraction composition
in the liquid phase. The symbols represent the new experimental data: (a) (
) isobutanol, (*) cis-3-hexenol and (䊉) 2-phenylethanol; (b) (×) isopentanol and ()
linalool; (c) (♦) isobutanal and () furfural; (d) () ethyl acetate, () isopentyl acetate and (+) 2-phenylethyl acetate. The dotted lines correspond to the representation
of the NRTL model (bubble point calculation at P = 0.1013 MPa), and the solid lines correspond the predictions of the COSMO-SAC model. In (c), the long dashed
line represents the COSMO-SAC predictions obtained with the cis-conformation of the furfural molecule, while the solid line represents the predictions obtained
with the trans-conformation.

A good agreement between the new experimental data and
the NRTL model is obtained (Fig. 8). The deviations are more
important for the ethanol rich region, as the experimental error
seems larger in that region (Fig. 8): we do not observe a sys-
tematic deviation, but some experimental points significantly
deviate from the calculated curves. The NRTL representation of
Henry’s constants in pure water at T = 298.15 K is also very sat-
isfactory (Table 9). Vapour–liquid equilibria data (DECHEMA
data [58]) far from the infinite dilution region were involved
in the parameter adjustments for the following binary mix-
tures: water + furfural, water + isobutanol, water + isopentanol,
water + ethyl acetate, and water + isobutanal. The NRTL can rep-
resent very well these VLE data [58] over the full composition
range (Figs. 9 and 10).
We predicted the same equilibria data with the COSMO-
SAC model. The ab initio calculations (COSMO files) are not
available for all the studied aroma compounds in the VT-2005
database. As a result, we determined the lacking COSMO files
by using the DMol3 ab initio package and by following the
recommendations of the authors of the VT-2005 database. The
COSMO-SAC model provides accurate predictions of the Ki
values for the least hydrophobic aroma compounds (alcohols
and aldehydes). Although the average deviations obtained with
COSMO-SAC are larger than with NRTL, the prediction of the
evolution of the Ki with the ethanol concentration is very satis-
factory (Fig. 8), and COSMO-SAC can predict at least the right
order of magnitude of the Ki . However, more significant devia-
tions are obtained for cis-3-hexenol, furfural and ethyl esters.
For cis-3-hexenol, the deviations between the COSMO-SAC
predictions and the new data could result from the uncertainty
of the vapour pressure correlation for cis-3-hexenol, as this
correlation was determined on only two state points with an
estimated value for the vapour pressure at T = 298.15 K. The
COSMO file of furfural is available in the VT-2005 database:
we observed that the molecular geometry of furfural proposed
in the VT-2005 database corresponds to the cis-conformation,
which is not the most stable conformation in the gas phase. The
most stable geometry in the gas phase is the trans-conformation
[63], while the cis-conformation is the most stable conforma-
tion in water as it gives rise to a larger dipole moment [64,65]
(both conformations are represented in Fig. 11(b) and (c)). As
shown in Fig. 11(a), the sigma profiles corresponding to these
configurations are significantly different, and as a result, the
COSMO-SAC VLE predictions are different (Fig. 8(c)). Since
the COSMO-SAC model was developed by always using the
optimal gas phase geometry of the molecules, it is preferable
to use the sigma profile obtained with the trans-conformation
of furfural. As shown in Fig. 8(c) and 9(b), the sigma profile
of the trans-conformation gives rise to the best COSMO-SAC
predictions in the infinite dilution region. We also examine the
 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154
Fig. 9. Vapour–liquid and liquid–liquid equilibria of the water + furfural binary
mixture at P = 0.1013 MPa. The symbols (
and *) represent the experimental in pure water. Values of rc below 3 Å are not recommended,
data [58]. The thick and solid lines correspond to the COSMO-SAC predictions
obtained from the trans-conformation of furfural (new ab initio calculation with
rc = 3 Å). The dotted lines correspond to the COSMO-SAC predictions obtained
from the cis-conformation of furfural (VT-2005 database [48]). The thin and
solid lines correspond to the NRTL calculations with binary parameters given
in Table 5. Fig (b) is an enlargement of (a) around the water rich region.
binary mixture furfural + water over the full composition range,
to check if the models are predictive far away from the infi-
nite dilution region. The sigma profile of the cis-conformation
gives rise to better COSMO-SAC predictions in the furfural rich
region (Fig. 9(a)), but the predictions are unsatisfactory in the
water rich region, especially in the infinite dilution region. No
liquid–liquid demixing is predicted with the sigma profile of
the cis-conformation. The use of the trans-conformation in the
COSMO-SAC calculations provides much better predictions in
the water rich region. The simultaneous good COSMO-SAC
predictions of the new data and of the VLE data in the water
rich region confirm the reliability of the new data. We note that
the representations of the NRTL model are satisfactory over the
full composition range.
The predictions capabilities of both thermodynamic models
were checked for other water + aroma binary mixtures (Fig. 10).
The COSMO-SAC predictions for these binary mixtures are also
satisfactory. Thus, the COSMO-SAC model seems reliable over
the full composition range, at least for VLE predictions, and
could be used to generate pseudo-experimental data and adjust
the NRTL parameters where not experimental data are available.
For example, this model can be useful to adjust the interac-
tion parameters between the aroma compounds. The NRTL and
151
COSMO-SAC predictions of the liquid–liquid equilibria should
however be considered with great care, as it is usually very diffi-
cult to predict VLE and LLE data with the same thermodynamic
model and the same binary parameters.
The COSMO-SAC predictions for the ethyl esters were first
unsatisfactory as the Ki were underestimated, although we made
sure that the molecular geometry in the gas phase was fully
optimised. The esters that we studied are very hydrophobic,
and it was previously observed that the COSMO-SAC model
is less accurate for very hydrophobic molecules [48]. We then
used an empirical methodology to improve the COSMO-SAC
predictions in the infinite dilution domain, which consists in
changing the real space cut-off distance rc (“cut-off global”
keyword in DMol3) between 3 and 5.5 Å for the COSMO ab
initio calculations. The aim of this approach is to change one
parameter (the cut-off rc ) that characterises the numerical solv-
ing of the ab initio method (cut-off) to generate slightly different
sigma profiles and empirically improve the predictions of the Ki .
This cut-off distance specifies the cut-off radius of the basis set
functions defined around each atom; i.e. it is the radius of the
sphere surrounding each atom, over which integrals are numer-
ically computed in the ab initio calculations. It is clear that
decreasing the cut-off distance make the ab initio calculations
less precise, but it may lead to better COSMO-SAC predictions
via some error compensations. Decreasing the cut-off distance
may make the molecular model less polar (more hydropho-
bic). As a result, using a smaller cut-off may increase the Ki
otherwise the ab initio calculations would be too inaccurate,
while values of rc above 4 do not significantly change the
results.
We then used this method for furfural, cis-3-hexenol, and
the linear ester molecules, and set-up the cut-off to 3 Å for the
COSMO ab initio calculations. The results are improved but the
COSMO-SAC model still underestimates the Ki and Henry’s
constants (Table 9). For furfural we found that a cut-off dis-
tance rc = 3 lead to an estimated value of Hi = 0.0149 MPa for
the Henry constant in pure water, which compares well with
the experimental value [53] (0.0186), while a rc = 5.5 Å lead
to Hi = 0.0116 MPa. The predicted value of Hi does not sig-
nificantly change for rc > 3.7 Å. It is interesting to note that the
COSMO-SAC model predicts almost the same Henry’s constant
for the three ethyl esters, and that the Henry constant of ethyl
decanoate is lower that that of ethyl octanoate (Tables 9 and 10).
This behaviour is also observed experimentally [66]. In Table 10
we reported the values of the vapour pressures Pi0 and predicted
activity coefficients γi∞ at infinite dilution of the ethyl esters
in pure water at 298.15 K. The COSMO-SAC model correctly
predicts that γi∞ has a higher value when the hydrophobicity
of the compound is higher. The fact that the evolution of the
Henry constant with chain length from ethyl hexanoate to ethyl
decanoate is not regular is explained by the fact that Hi is product
of γi∞ and Pi0 that evolve with opposite trends.
We also evaluated the activity coefficients γi∞ of some
aroma compounds at infinite dilution in hydroalcoholic solutions
(Fig. 12(a)), by using the relation γi∞ = Pexp Ki,exp /Pi0 (Texp ),
where Ki,exp is the experimental partition coefficient of the aroma
152 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154

Fig. 10. Vapour–liquid and liquid–liquid of binary mixtures of water + 1 aroma. (a) Water + isobutanol at T = 308.15 K; (b) water + isopentanol at P = 0.1013 MPa;
(c) water + isobutanal at P = 0.102 MPa; (d) water + ethyl acetate at P = 0.1013 MPa. The circles represent the experimental data [58]. The thick lines correspond to
the predictions of the COSMO-SAC model, and the thin lines correspond to the NRTL calculations.

i, Pexp = 0.1013 MPa, and Pi0 (Texp ) the vapour pressure of the
aroma i, calculated at the experimental temperature Texp with
Eq. (2). We observe that γi∞ is very large for hydrophobic com-
pounds like ethyl octanoate and ethyl decanoate, and decreases
of several orders of magnitude (up to five orders) when the
Fig. 11. (a) Normalized sigma profiles of furfural calculated from the cis (dotted
line) and trans (solid lines) conformations. (b) cis and (c) trans conformations
of the furfural molecule.
proportion of ethanol is increased. The COSMO-SAC model
predicts fairly well the evolution of the γi∞ with the ethanol
composition, while the data are well represented with NRTL.
We also checked the infinite dilution approximation by exam-
ining the evolution of the activity coefficients of the aromas as
a function of the total aroma composition in the system. We
used the COSMO-SAC model to predict the influence of the
total proportion of aromas in the mixture on the activity coef-
ficients of ethyl hexanoate in aqueous solutions at 298.15 K
(Fig. 12(b)). We assumed that all the 13 aroma compounds
had the same mole fractions. We found that the deviations of
the γ i from γ ∞
i become significant when the total mole frac-
tion xaroma = aromas xi becomes larger than about 10−3 , which
means an average mole fraction xi for each aroma as low as
8 × 10−5 . This limit corresponds to a deviation of 5% from
γi∞ for isobutanol, 10% for ethyl hexanoate, and 14% for ethyl
decanoate. For xaroma > 0.01, the activity coefficients γ i dramat-
ically decrease (Fig. 12(b)). The decrease of limit γ i is more
important for the most hydrophobic molecules. The limit is
higher in ethanol than in water, as the aroma molecules are more
compatible with ethanol and the effect of the aroma concentra-
tion is less important.
These results show that special care should be taken when one
employs the “infinite dilution” approximation either in exper-
imental studies or data modelling: the activity coefficient at
infinite dilution of a hydrophobic molecule in water is a very
sensitive property that can be influenced by the presence of
impurities. Moreover, the activity coefficients of the aroma com-
 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154 153

dilution approximation for aqueous solutions containing many

aroma compounds, as the total concentrations of aroma com-
pounds may not be negligible and have a large effect of the
activity coefficients, even if the concentration of each aroma is
very low.

List of symbols
aij , bij , cij NRTL binary parameters
Aeff area of a standard surface segment used in the restoring
energy contribution (Aeff = 7.5 Å2 )
Ai (σ) total area of the surface segments of molecule i having
a charge density σ
Ai , Bi , . . ., Ei constants used in the correlations of the pure
compound vapour pressures
Aref standard area used in the Staverman–Guggenheim con-
tribution [1,2] (Aref = 79.53 Å2 )
H Henry constant (MPa)
kB Boltzmann constant (1.3806504 × 10−23 J K−1 )
Ki vapour–liquid partition coefficient: Ki = yi /xi
ni number of surface segments on molecule i (COSMO-
SAC model)
p normalized sigma profile
P pressure (MPa)
qi surface parameter of component i used in the in the
Staverman–Guggenheim contribution [1,2]
rc real space cut-off (Å): cut-off distance that specify the
limit of integration around the atoms, for the com-
Fig. 12. (a) Activity coefficients at infinite dilution of aroma compounds in
putation of integrals (charge density, functionals) that
hydroalcoholic solutions, at the mixture boiling temperature at P = 0.1013 MPa
and for different ethanol mole fractions in the liquid phase. The solid lines are used in the ab initio calculations (option “Cut-
represent the COSMO-SAC predictions, and the dotted lines correspond to the off Global” in Dmol3)
NRTL calculations. The symbols denote the new experimental data for (
) ethyl reff radius of a standard segment
√ in the calculation of the
decanoate, () isopentyl acetate and (*) furfural. (b) Predictions with COSMO- sigma profiles (reff = Aeff /π)
SAC of the activity coefficients of ethyl hexanoate (solid line) and isobutanol
ri volume parameter of component i used in the
(dotted line) in mixtures of water and 13 aroma compounds at T = 298.15 K,
as a function of the total aroma mole fraction xaroma = 1 − xwater . Each aroma Staverman–Guggenheim contribution [1,2]
compound is assumed to have the same mole fraction xaroma /13. R ideal gas constant = 8.314472 J mol−1 K−1
T temperature (K)
vref standard volume used in the Staverman–Guggenheim
pounds at very low concentrations in solutions containing many contribution [1,2] (=66.69 Å3 )
different molecules also at very low concentrations (such as real vimolec volume of molecule i (Å3 )
systems) may significantly deviate from the infinite dilution val- W exchange energy (COSMO-SAC model)
ues due to synergy effects. In our experimental measurements, x liquid phase composition (mole fraction)
the purity of ethanol is high (99.9% mass) and the total mole frac- y gas phase composition (mole fraction)
tion of aroma compounds xaroma is always lower than 3 × 10−4 z coordination number in the Staverman–Guggenheim
in the liquid phases (Table 6), thus, the infinite dilution approx- contribution [1,2]
imation can be used.
Greek letters
5. Conclusion αij non-randomness binary parameter (NRTL model)
γi activity coefficient of component i in the liquid phase
Vapour–liquid equilibria of 13 aroma compounds in hydroal- Γi segment activity coefficient in pure liquid component i
coholic solutions were measured with a recirculation apparatus. Γs segment activity coefficient in the liquid mixture S
The presence of ethanol has a very large effect on the parti- ζ volume fraction in the Staverman–Guggenheim contri-
tion coefficients of the aroma compounds. The new data are bution [1,2]
represented by the NRTL model and predicted with the COSMO- θ surface fraction in the Staverman–Guggenheim contri-
SAC model. The COSMO-SAC model is reliable over the full bution [1,2]
composition range, and can at least predict the right order σ surface charge density (E Å−2 )
of magnitude of the partition coefficients at infinite dilution. τ variable used in the NRTL model
Special care should be taken when one employs the infinite φ variable used in the NRTL model
154 V. Athès et al. / Fluid Phase Equilibria 265 (2008) 139–154
Subscripts
exp experimental
cal calculated
i, j component i, j
S liquid mixture
∞ infinite dilution
Superscripts
0 pure component vapour pressure
SG Staverman–Guggenheim
molec molecular
Acknowledgments
We would like to thank Prof. Lin for useful discussions and for
having provided us the sigma profiles of water and methyl acetate
as test examples. We are also grateful to E. Mullins and Y.A. Liu
and for having provided us the improved ab initio calculations
for ethanol.
References
[1] A.J. Staverman, Recl. Trav. Chim. Pays.-Bas 69 (1950) 163–174.
[2] E.A. Guggenheim, Mixtures, Oxford University Press, Oxford, 1952.
[3] P. Alessi, M. Fermeglia, I. Kikic, Fluid Phase Equilib. 70 (1991) 239–
250.
[4] S.D. Barnicki, Chem. Eng. Prog. 98 (2002) 58–67.
[5] A.H. Lincoff, J.M. Gosset, in: W. Brutsaert, G.H. Jirka (Eds.), Gas Transfert
at Water Surfaces, Reidel, Dordrecht, 1984, pp. 17–25.
[6] L.S. Ettre, B. Kolb, Chromatographia 32 (1991) 5–12.
[7] L.S. Ettre, C. Welter, B. Kolb, Chromatographia 35 (1993) 73–84.
[8] D.T.C. Gillespie, Ind. Eng. Chem. Anal. Ed. 18 (1946) 575–577.
[9] J. Gmehling, U. Onken, H.W. Schulte, J. Chem. Eng. Data 25 (1980) 29–32.
[10] P. Duhem, J. Vidal, Fluid Phase Equilib. 2 (1978) 231–235.
[11] F. Sorrentino, A. Voilley, D. Richon, AIChE J. 32 (1986) 1988–
1993.
[12] C. Coquelet, D. Richon, J. Chem. Eng. Data 50 (2005) 2053–2057.
[13] J.D. Raal, A.L. Mühlbauer, in: G.F. Hewitt, C.L. Tien (Eds.), Series in
Chemical and Mechanical Engineering, Taylor and Francis, Washington,
1998.
[14] J.D. Raal, D. Ramjugernath, in: R.D. Weir, T.W. deLoos (Eds.), Experi-
mental Thermodyanmics, Elsevier, Amsterdam, 2005, pp. 71–86.
[15] J.D. Raal, D. Ramjugernath, in: R.D. Weir, T.W. de Loos (Eds.), Experi-
mental Thermodynamics, Elsevier, Amsterdam, 2005, pp. 340–357.
[16] S.I. Sandler, Fluid Phase Equilib. 116 (1996) 343–353.
[17] V. Athes, M.P.I. Lillo, C. Bernard, R. Perrez-Correa, I. Souchon, J. Agric.
Food Chem. 52 (2004) 2021–2027.
[18] S.P. Christensen, Fluid Phase Equilib. 150 (1998) 763–773.
[19] G.M. Kontogeorgis, E.C. Voutsas, I.V. Yakoumis, D.P. Tassios, Ind. Eng.
Chem. Res. 35 (1996) 4310–4318.
[20] W.G. Chapman, K.E. Gubbins, G. Jackson, M. Radosz, Ind. Eng. Chem.
Res. 29 (1990) 1709–1721.
[21] A. GilVillegas, A. Galindo, P.J. Whitehead, S.J. Mills, G. Jackson, A.N.
Burgess, J. Chem. Phys. 106 (1997) 4168–4186.
[22] J. Gross, G. Sadowski, Ind. Eng. Chem. Res. 40 (2001) 1244–1260.
[23] H. Renon, J.M. Prausnitz, AIChE J. 14 (1968) 135–144.
[24] D.S. Abrams, J.M. Prausnitz, AIChE J. 21 (1975) 116–128.
[25] A. Klamt, J. Phys. Chem. 99 (1995) 2224–2235.
[26] A. Klamt, F. Eckert, Fluid Phase Equilib. 172 (2000) 43–72.
[27] S.-T. Lin, S.I. Sandler, Ind. Eng. Chem. Res. 41 (2002) 899–913.
[28] C.A. Faùndez, V.H. Alvarez, J.O. Valderrama, Thermochim. Acta 450
(2006) 110–117.
[29] E.M. Yezdimer, J. Sedlbauer, R.H. Wood, Chem. Geol. 164 (2000)
259–280.
[30] J. Sedlbauer, G. Bergin, V. Majer, AIChE J. 48 (2002) 2936–2959.
[31] S. Aparicio, R. Alcalde, M.J. Davila, B. Garcia, J.M. Leal, J. Phys. Chem.
B 111 (2007) 4417–4431.
[32] B. Widom, J. Chem. Phys. 39 (1963) 2808–2812.
[33] C. McCabe, A. Galindo, P.T. Cummings, J. Phys. Chem. B 107 (2003)
12307–12314.
[34] S. Skjoldjorgensen, Ind. Eng. Chem. Res. 27 (1988) 110–118.
[35] T. Holderbaum, J. Gmehling, Fluid Phase Equilib. 70 (1991) 251–265.
[36] H.P. Gros, S. Bottini, E.A. Brignole, Fluid Phase Equilib. 116 (1996)
537–544.
[37] C. Le Thi, S. Tamouza, J.P. Passarello, P. Tobaly, J.C. de Hemptinne, Ind.
Eng. Chem. Res. 45 (2006) 6803–6810.
[38] A. Fredenslund, R.L. Jones, J.M. Prausnitz, AIChE J. 21 (1975) 1086–1099.
[39] B.L. Larsen, P. Rasmussen, A. Fredenslund, Ind. Eng. Chem. Res. 26 (1987)
2274–2286.
[40] J. Gmehling, J.D. LI, M. Schiller, Ind. Eng. Chem. Res. 32 (1993) 178–193.
[41] P. Tsavas, E. Voutsas, K. Magoulas, D. Tassios, Ind. Eng. Chem. Res. 43
(2004) 8391–8399.
[42] J.M. Prausnitz, F.W. Tavares, AIChE J. 50 (2004) 739–761.
[43] W. Arlt, O. Spuhl, A. Klamt, Chem. Eng. Process. 43 (2004) 221–238.
[44] P. Paricaud, J. Chem. Phys. 124 (2006) 154505.
[45] J. Tomasi, B. Mennucci, R. Cammi, Chem. Rev. 105 (2005) 2999–3093.
[46] A. Klamt, COSMO-RS: From Quantum Chemistry to Fluid Phase Ther-
modynamics and Drug Design, Elsevier, Amsterdam, 2005.
[47] S.-T. Lin, S.I. Sandler, Ind. Eng. Chem. Res. 43 (2004) 1322–11322.
[48] E. Mullins, R. Oldland, Y.A. Liu, S. Wang, S.I. Sandler, C.C. Chen, M.
Zwolak, K.C. Seavey, Ind. Eng. Chem. Res. 45 (2006) 4389–4415.
[49] H. Grensemann, J. Gmehling, Ind. Eng. Chem. Res. 44 (2005) 1610–1624.
[50] T. Mu, J. Rarey, J. Gmehling, Ind. Eng. Chem. Res. 46 (2007) 6612–
6629.
[51] R.H. Perry, D.W. Green, Perry’s Chemical Engineers’ Handbook, McGraw-
Hill, New York, 1984.
[52] Prosim, Component Plus Software, v. 3.4, 2005.
[53] C.L. Yaws, Yaws’s Handbook of Thermodynamic and Physical Properties
of Chemical Compounds, Knovel, 2003.
[54] M.A.E. Dýaz, T. Guetachew, P. Landy, J. Jose, A. Voilley, Fluid Phase
Equilib. 157 (1999) 257–270.
[55] NIST, Chemistry webbook. http://webbook.nist.gov/chemistry/, 2005.
[56] S.R.C. (SRC), PhysProp database, www.syrres.com/esc/physdemo.htm.
U.S. Environmental Protection Agency.
[57] W.B. Neely, G.E. Blau, Environmental Exposure from Chemicals, 2nd ed.,
CRS Press, Boca Raton, 1985.
[58] J. Gmehling, U. Onken, W. Arlt, Vapour–Liquid Equilibria Data Collection,
DECHEMA, Dortmund, 1981.
[59] J.A. Larkin, J. Chem. Thermodyn. 7 (1975) 137–148.
[60] P. Vrbka, D. Fenclová, V. Laštovka, V. Dohnal, Fluid Phase Equilib. 237
(2005) 123–129.
[61] J. Altschuh, R. Bruggemann, H. Santl, G. Eichinger, O.G. Piringer, Chemo-
sphere 39 (1999) 1871–1887.
[62] R.J. Oldland, Predicting Phase Equilibria Using the Cosmo-based Thermo-
dynamic Models and the VT-2004 Sigma-profile Database, Master Thesis,
Virginia Polytechnic Institute, Blacksburg, 2004.
[63] T.S. Little, J. Qiu, J.R. Durig, Spectrochim. Acta Part A 45 (1989) 789–794.
[64] R. Rivelino, K. Coutinho, S. Canuto, J. Phys. Chem. B 106 (2002)
12317–12322.
[65] R. Rivelino, S. Canuto, K. Countinho, Br. J. Phys. 34 (2004) 84–89.
[66] E. Aprea, F. Biasioli, T.D. Märk, F. Gasperi, Int. J. Mass Spectrom. 262
(2007) 114–121.
[67] W.M. Meylan, P.H. Howard, Environ. Toxicol. Chem. 10 (1991)
1283–1293.