Fluid Phase Equilibria 240 (2006) 1–14

Sensitivity of azeotropic states to activity coefficient model

parameters and system variables
Naveed Aslam, Aydin K. Sunol ∗
Department of Chemical Engineering, University of South Florida, 4202 E. Fowler Avenue, Tampa, FL 33620, USA
Received 23 December 2003; received in revised form 5 November 2005; accepted 7 November 2005
Available online 28 December 2005

Abstract
The activity coefficient models, when based on parameters obtained through binary phase equilibrium information result in multiple sets that may
either not be able to predict ternary and higher order azeotropes, or computed values are significantly different from experimental data, particularly at
thermodynamic landmarks. Sensitivity of azeotropic states to model parameters and operating conditions is imbedded into a homotopy continuation
formulation, providing a functionally sound basis for prediction of such states and their corresponding sensitivity. Furthermore, the validity of
parameter set depends on the range of operating variables. The methodology allows systematic variation of model parameters and selection of
a parameter set that allows computation of thermodynamic landmarks, such as azeotropy. The systematic analysis of the effects of parametric
perturbation on the prediction of azeotropes and heteroazeotropes also provide useful insights into capabilities and limitations of phase equilibrium
models. The technique is also useful in developing the effect of system variables on azeotropy conditions. We demonstrated the utility of our
approach through modeling benzene–cyclohexane–n-propanol, ethyl acetate–ethanol–water and water–formic acid–acetic acid system.
© 2005 Elsevier B.V. All rights reserved.

Keywords: Azeotropy; Homotopy continuation; Heteroazeotropes; Activity coefficient model

1. Introduction Reliability of residue curves, as a conceptual design tool,

Recently, conceptual process design has become an impor-
tant area of activity both for industry and academia. The basic
purpose of conceptual process design is to conceive generate,
compare and evaluate the different process alternatives at a very
early stage of design with minimum information available. Con-
ceptual process design allows systematic elimination of large
number of available process alternatives and allows subsequent
detailed analysis of substantially reduced set of alternatives. An
effective conceptual design step requires efficient and reliable
tools that require minimum information. One of the most widely
used conceptual tools is Residue Curve Maps (RCMs) that are
used for conceptual design of non-ideal distillation separation
sequences. Residue curves is the most mature conceptual pro-
cess design tool and is a part of almost all the available design
packages. The methodology is well covered in a recent text Mal-
one and Doherty [9].
∗ Corresponding author. Tel.: +1 813 974 3566; fax: +1 813 974 3651.
E-mail address: sunol@eng.usf.edu (A.K. Sunol).
depends on the accuracy of the model representing the phase
equilibrium and algorithm used for prediction of thermody-
namic landmarks, such as azeotropes. Tolsma and Barton [14,15]
described a homotopy continuation based approach to com-
pute all the possible azeotropes in multicomponent mixtures
with mathematical guarantees. The mathematical treatment of
Tolsma and Barton’s [14,15] approach is independent of the
representation of system non-idealities. However, numerical
tracking of homotopy branches and in turn computation of
azeotropes is highly dependent on phase equilibrium model and
associated parameters. Aslam and Sunol [2] further enhanced
their approach by incorporating both vapor and liquid phase
non-idealities through equations of state, and thus computing
azeotropes for entire pressure range. Tolsma and Barton [14,15]
used activity coefficient models to describe the liquid phase
non-idealities with binary interaction parameters regressed from
binary phase equilibrium data. The activity coefficients, which
are based on parameters obtained from binary information, when
used in modeling the mixtures, with three or more components
may not provide a realistic description of phase non-idealities for
such mixtures mainly due to error associated with extrapolating

0378-3812/$ – see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.fluid.2005.11.019
2 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14

the binary information to describe the phase non-idealities in and

ternary or higher order mixtures. Furthermore, they may not ade-
γi Pisat
quately represent thermodynamic landmarks, such as azeotropy. KiReal = (2.5)
Through homotopy continuation approach, all the azeotropes P
in multicomponent mixtures can be computed by solving for Tolsma and Barton [14,15] further enhanced the Fidkowski et
necessary condition of azeotropy. When liquid phase activity al. [4] approach to isobaric conditions. They developed the fol-
coefficients for multicomponent mixtures are obtained through lowing formulation shown in system of Eq. (2.6) using activity
parameters based upon binary information, their use can lead coefficients to map the liquid phase non-idealities and assuming
to errors in prediction of azeotropy (Englezos and Kalogerakis ideal vapor phase.
[3]). For such a case, the predicted values can be significantly  
different from experimental values, or provide the azeotropic x1 − K10 x1
 . 
values, which are lying outside the physical composition space  
 
where they are of little practical significance, or may not be  . 
 
able to predict the azeotropy. A systematic method is therefore  . 
F (x, T, λ) =   (2.6)
needed to establish the effect of parametric perturbation on the  
 xn − Kn0 xn 
prediction of azeotropes. Since activity coefficient model param-  
 n 
eters can have multiple solutions (Rabinovich [10] and Walas  
[16]), a global solver such as homotopy continuation is needed xi − 1
to establish the sensitivity of model parameters on azeotropy. We i=1
are using a homotopy continuation based approach to develop where Ki0 is the pseudo equilibrium constant and is given as
the sensitivity of model parameters and system variables on follows:
azeotropic states. Our approach is based on exploiting the nec- 
essary and sufficiency condition of azeotropy and imbedding the Ki0 = λKiReal + (1 − λ)KiIdeal (2.7)
sensitivity analysis with respect to model parameters or system
variables. The resulting over determined system of equations is As we set F(x, T, λ) = 0, we have defined a solution space in x,
solved by pseudo arc length continuation algorithm as detailed T, λ for a fixed pressure. The mapped solution of above function
by Allgower and Georg [1] and Speight [11]. We analyzed is called the homogeneous homotopy path and is usually repre-
benzene–cyclohexane–n-propanol, ethylacetete–ethanol–water sented on a temperature and composition bifurcation diagram.
and water–formic acid–acetic acid system. We also demon- Fig. 1 depicts a bifurcation diagram for a hypothetical system
strated that if azeotropy is not predicted under the given set of (ABC). As can be seen in Fig. 1, that two main characteristic
activity model parameters, it could be obtained through selecting points on this bifurcation diagram are bifurcation point appear-
another set of parameters. ing on pure component branches, bifurcation points appearing on
binary and higher order branches and solution points at λ = 1.0.
2. Theory Importance of these characteristic points on a systematic sen-
sitivity analysis of model parameters is described in following
Fidkowski et al. [4] solved the necessary conditions of section.
azeotropy (xi = yi ) using homotopy continuation. The homotopy
map is given as follows: 3. Sensitivity analysis of azeotropic states

hi (x, λ) = λ(xi − yiReal ) + (1 − λ)(xi − yiIdeal ) (2.1) The homogeneous bifurcation diagram has two types of char-
acteristic points as shown in Fig. 1.
where λ is the homotopy parameter, which varies from 0 to 1 and
xi is the liquid composition. The yiIdeal is the vapor composition
• Bifurcation points (appearing) on nth order homotopy
for an ideal gas and ideal solution and yiReal is for a real gas and
branches (n = 1, pure component; n = 2, binary branches and
non-ideal solution. For λ = 0, the xi = yiIdeal and for λ = 1, the
n = 3, ternary branches, etc.).
xi = yi Real .
• Solution points.
yiIdeal = KiIdeal xi (2.2)
Homogeneous homotopy path can be obtained by setting
yiReal = KiReal xi (2.3) Eq. (2.6) equal to zero. However, if the model parameters are
incorporated in Eq. (2.6), the homogeneous homotopy map will
The KiIdeal is the ideal solution equilibrium constant and is become
derived from Raoult’s law while KiReal is a non-ideal equilibrium
constant which maps the liquid phase non-idealities with activity F (x, T, P, λ, {Api }i=1,...,Np ) = 0 (3.1)
coefficients. The vapor phase is assumed ideal.
In Eq. (3.1), Api are the model parameters used in activ-
P sat ity coefficient expression to describe the liquid phase non-
KiIdeal = i (2.4)
P idealtes. For a homogeneous system, we get n + 1 equations
 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14 3

Fig. 1. Hypothetical temperature and mole fraction bifurcation diagram for a three component system (A–C).

in terms of n + Np + 3 variables and parameters. By speci-
fying the Np + 1 variables or parameters and solving n + 1
equation in terms of n + 2 variables with the pseudo arc
length continuation algorithm, we can compute homogeneous
branches. The approach described in Section 2 for comput-
ing the azeotropes, provides the Np activity coefficient model
parameters and temperature or pressure to solve the necessary
condition of azeotropy. This step is followed by a stability
test to distinguish between homogeneous and heterogeneous
branches.
3.1. Bifurcation points appearing on pure component
branches
For a binary mixture of component i and j, the Eq. (2.6) can be
modified to give following equation to calculate the bifurcation
point on pure component branch i.
 perature or pressure can be varied as continuation variables and
1 Pj (Ti )/P
sat b
λi = (3.2)
Kj Pjsat (Tib )/P
For low to moderate pressures, Kj is usually expressed as
a function of infinite dilution activity coefficient, γj∞ . The Eq.
(3.2) results in a bifurcation point on the pure component branch
i. The bifurcation point is actually the point of intersection of
pure component branch i and a binary branch ij. These bifurca-
tion points act as starting points of binary branches. Only the
binary branches which are intersecting with pure component
branches between the bifurcation point values of zero and unity
can lead to physically meaningful solutions at λ = 1.0. For a
binary mixture, a minimum boiling azeotrope bifurcates from
lower boiling component and a maximum boiling azeotrope
bifurcates from higher boiling component. For a given set of
activity coefficient model parameters, infinite dilution activity
coefficient is calculated and a corresponding bifurcation point λij
is computed from Eq. (3.2). Once a bifurcation point is located,
it acts, as the solution in a pseudo arc length continuation algo-
rithm while the activity coefficient model parameters become
continuation parameters. Similarly, system variables like tem-
their effect on bifurcation point identify where azeotrope disap-
pears, λij > 1 or λij < 0.
A bifurcation point on a tracked branch is where a k-ary
branch intersects a (k + 1)-ary branch. While moving along a
k-ary azeotropic branch, a necessary condition for a bifurcation
4 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14

point to occur is, αj (s) = 0 Thus, replacing λ = 1.0 in Eq. (2.6) reduces the homotopy



 continuation map to the necessary condition of azeotropy. Once
γj Pjsat Pjsat the necessary condition of azeotropy is satisfied, any variable
αj = 1 − λ + (1 − λ) (3.3) from a general variable pool {x, T, P, λ, {Api }i=1,...,Np } can be
P P
selected as a continuation variable and with a numerical con-
for some j ∈ {k + 1, . . ., n}. tinuation algorithm, it can be examined that how azeotropic
Thus, a necessary condition for bifurcation point on k-ary conditions are varying with variation in either activity coeffi-
azeotropic branch corresponds to an intersection of a k + 1-ary cient parameters or system variables. Thus, the effect of each of
azeotropic branch with a k-ary branch. The bifurcation point adjustable parameters of activity coefficient model on the pre-
on a k-ary branch is located by constructing an interpolating diction of ternary and higher order azeotropic conditions can be
polynomial, which locates the point where αk+1 = 0 in Eq. (2.6). examined through this approach. A phase stability test is per-
To locate a bifurcation point on a k-ary branch, we set αk+1 (x, formed at each step during continuation.
T, P, λ) = 0.0 and when we incorporate this condition into Eq.
(2.6), we obtain a fully determined system. 5. Numerical continuation
 
x1 − K10 x1 The sensitivity analysis of azeotropic states to model param-
 .  eters and system variables is implemented through a numerical
 
  continuation approach. Let us assume that for a smooth function
 . 
  F(x) = 0, we want to compute a solution curve. The numerical
 . 
  continuation technique traces a desired solution branch through
F (x, T, P, λ) = 

 =0
 (3.4)
 xk − Kk0 xk  predictor–corrector type method. The predictor–corrector type
 k  methods are based on computing the sequence of points xi , i = 1,
  
 xk − 1  2, . . . along the solution branch within a specified tolerance cri-
 
 i=1  terion. The basic operation of tracing a solution branch through
αk+1 (T, P, x, λ) numerical continuation consists of following steps:

For NRTL activity coefficient model, there are three
adjustable parameters and thus variable set becomes
{[xi ]i=1, . . ., k , T, P, λ, {Api }i=1,...,Np }. From the indicated
variable set, we can select any k + 3 variables and perform the
continuation on Eq. (3.4), e.g. we can choose any one of the three
adjustable parameters of NRTL activity coefficient model as a
continuation parameter and perform the numerical continuation
on Eq. (3.4) through a continuation algorithm. Similarly, system
variables like temperature and pressure can also be selected as
continuation variables and can be varied through continuation
algorithm to analyze the effect pressure or temperature has on the
location of bifurcation points on tracked binary or higher order
branches.
In order to develop a sensitivity analysis of homogeneous
azeotropic states, we can add another equation to (3.1) that
defines any of above two indicated points of interests in a bifurca-
tion diagram. The modified system of equations provides some
added flexibility as an additional parameter that was fixed in
system of Eq. (3.1) can be now treated as one of the variables.
The appended system of equations is solved by a numerical con-
tinuation approach.
4. Solution points
On a homotopy branch, a solution point is simply a point that
satisfies the necessary condition of azeotropy. Thus, a solution
point is a point where homotopy branch crosses λ = 1.0 and the
corresponding composition and temperature is the azeotropic
point. Thus, all solution points are either binary or higher order
azeotropes λ = 1.0.
• Prediction of a point along the curve.
• Correction of the predicted point.
Usually a tangent predictor is used for predicting a point
along the curve. For a given point xi on the curve and dx i
ds is
a normalized tangent vector with a step size hi , then the next
predicted point is given as
dxi
Xp = xi + hi (4.1)
ds
The predicted point Xp is close to the curve but not exactly
on the solution curve. To find the next point xi+1 on the curve
we have used a Newton method. Since Newton method can only
be applied to a fully defined system therefore another scalar
equation is added to function F(x).
We developed a Matlab code for numerical tracking of con-
tinuation branches through implementing the following three
essential steps:
(1) Pre-processing step: The first step in a numerical continua-
tion algorithm is the pre-processing of system of equations
before they can be tracked. The pre-processing step includes
automatic system reduction, scaling and degeneracy check.
(2) Path tracking: The second step includes tracking the solution
branch through predictor–corrector type methods.
(3) Post-processing: Once the solution branch is tracked in
post-processing step the tracked branch is analyzed. The
post-processing step includes transformation of solution to
original system, identifying the physical solutions, deciding
about rerunning the solution for tracing other branches.
 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14
Along with above indicated processing steps we have also
appended following features to our code:
(a) Variable step size algorithm: The step size control is an
important feature of any continuation algorithm. Too small
of a step size will lead to error build-up and too big of a
step size result in instability. We have implemented a con-
vergence dependent step size control in above indicated
numerical algorithm. For a given step size of hi and number
of maximum Newton correctors iterations (Ncorr ), the new
step size is selected as follows:
hi+1 = 1.5 hi if converged and n < Ncorr
hi+1 = 0.6 hi if not converged
hi+1 = hi if n = Ncorr
where n is number of Newton iterations for the current step.
(b) Stopping criterion: Usually a continuation path converge
when the continuation parameter attains the value of unity.
However, for the sensitivity analysis reported in this paper,
we are using a stopping criteria based on the maximum
number of points along the curve. The continuation path
can also fail when the step size becomes less then minimum
specified value.
(c) Scaling: The basic purpose of scaling is to minimize the
numerical instability that might occur while tracing the
numerical continuation path. To improve the numerical sta-
bility of path tracing, three techniques can be used.
I. Scaling based on equations.
II. Scaling based on magnitudes of variables.
III. Scaling based on Jacobian matrices.
We used SCLGNP algorithm for variable scaling in
sensitivity analysis (Watson et al. [17]).
(d) Stability of solution: Most of the times along with locat-
ing all the solutions of non-linear equations the information
regarding their stability is also needed. The stability of solu-
tion is determined through computing coefficients of the
characteristic polynomial, the eigenvalues or both of the
Jacobian. The steady state solution is stable if all the eigen-
values have negative real parts. The solution is unstable if
any of the eigenvalue has positive real part. No information
regarding the stability of solution is available if any of the
eigenvalue has a zero real part. We have also analyzed the
variation of stability of azeotropic solution with variations
in system variables and model parameters as described in
next section.
6. Results and discussion
We developed the sensitivity analysis of azeotropic states
with respect to system variables and model parameters for fol-
lowing three systems:
• Benzene–cyclohexane–n-propanol system.
• Ethylacetate–ethanol–water system.
• Water–formic acid–acetic acid system.
5
We selected above three systems mainly because of the avail-
ability of experimental data and presence of ternary azeotropes
in each of them, as we are more interested to analyze the
effect of model parameters and system variables on higher order
azeotropes. Tolsma [13] indicated the possibility of such a sen-
sitivity analysis but did not provide any numerical examples
demonstrating the effect of model parameters and system vari-
ables on points of interest in homogeneous bifurcation diagrams.
We have not only provided the examples regarding the sensitiv-
ity analysis of azeotropic states but has also demonstrated that
how the sensitivity analysis can help in improving the prediction
of higher order azeotropes from parameters based upon binary
data.
6.1. Benzene–cyclohexane–n-propanol system
Fig. 2 contains the bifurcation diagram for benzene
(1)–cyclohexane (2)–n-propanol (3) system at atmospheric pres-
sure. The bifurcation diagram is generated through activity
coefficient parameter set 1 in Table 1. Computed and experi-
mental composition of ternary benzene (1)–cyclohexane (2)–n-
propanol (3) azeotrope are reported in Table 2. It can be seen that
with parameter set 1, one can predict a ternary azeotrope for this
system but the computed values are different from experimen-
tally reported values of Gmehling et al. [5,6]. We systematically
adjusted these parameters through sensitivity analysis described
in Figs. 4–11; these adjusted parameters are listed as set 2
in Table 1. The computed azeotropic composition based upon
parameters set 2 is reported in Table 2 and is much closer to
experimental values of Gmehling et al. [5,6].
Fig. 3 shows the variation of ternary azeotropic composi-
tion with pressure and is based on parameter set 2. Fig. 3
also shows that benzene–cyclohexane–n-propanol azeotrope
is not very sensitive to pressure at higher pressure range as
it is in lower pressure range. This may be due to neglect-
ing the vapor phase non-idealities at higher pressures where
they have significant contribution. We are using pressure here
as a continuation parameter, unlike Knapp and Doherty [8].
Fig. 4 contains the variation of this ternary azeotrope with
Fig. 2. Bifurcation diagram for benzene (1)–cyclohexane (2)–n-propanol (3)
system at atmospheric pressure. (Based on NRTL activity coefficient model
parameters set 1 from Table 1.)
6 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14

Table 1
NRTL activity coefficient parameters for benzene (1)–cyclohexane (2)–n-propanol (3) system
NRTL parameters α12 α13 α23 B12 B21 B13 B31 B23 B32

Set #1 from DECHEMA series 0.300 0.499 0.620 353.02 −87.6 987.2 449.9 −1265 546.3
Set #2 adjusted parameters 0.285 0.525 0.450 353.02 −198.0 943.4 449.9 −1250 485.6

Table 2
Ternary azeotrope of benzene–cyclohexane–n-propanol system at atmospheric pressure (computed and experimental values)
Ternary azeotrope Benzene (1) Cyclohexane (2) n-Propanol(3) Temperature (◦ C)

Experimental data of Fischer 0.305 0.499 0.196 73.70

Computed results based on parameter set 1 0.309 0.472 0.217 73.49
Computed results based on parameter set 2 0.306 0.491 0.201 73.62

non-randomness parameter [α12 ], which is insensitive to ternary
azeotropic composition. Figs. 5 and 6 indicates the effect of non-
randomness parameters [α13 ], [α23 ] on ternary azeotropic com-
position of benzene–cyclohexane–n-propanol azeotrope. Both
of these parameters have significant effect on the azeotropic
composition. Fig. 7 shows that NRTL activity coefficient param-
eters [B12 ] have almost no effect on ternary azeotropic compo-
sition. Fig. 8a shows that for a narrow range the NRTL activ-
ity coefficient parameter [B13 ] has negligible effect on ternary
azeotropic composition. However, for a wider range the ternary
azeotropic composition varies with variation in NRTL activity
coefficient parameter [B13 ] as depicted in Fig. 8b. Figs. 9–11
depicts the effect of NRTL activity coefficient parameters [B21 ],
[B12 ] and [B31 ] on the location of bifurcation point on a binary
branch. Only [B21 ] has significant effect on the location of
bifurcation point on binary azeotropic branch, therefore, it was

Fig. 3. Variation of benzene (1)–cyclohexane (2)–n-propanol (3) ternary Fig. 5. Variation of benzene (1)–cyclohexane (2)–n-propanol (3) ternary
azeotropic composition with pressure. (Based on NRTL activity coefficient azeotropic composition with non-randomness parameter [α13 ]. (Based on NRTL
model parameters set 2 from Table 1.) activity coefficient model all other parameters from set 1 of Table 1.)

Fig. 4. Variation of benzene (1)–cyclohexane (2)–n-propanol (3) ternary Fig. 6. Variation of benzene (1)–cyclohexane (2)–n-propanol (3) ternary
azeotropic composition with non-randomness parameter [α12 ]. (Based on NRTL azeotropic composition with non-randomness parameter [α23 ]. (Based on NRTL
activity coefficient model all other parameters from set 1 of Table 1.) activity coefficient model all other parameters from set 1 of Table 1.)
 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14
Fig. 7. Variation of benzene (1)–cyclohexane (2)–n-propanol (3) ternary
azeotropic composition with NRTL activity model parameter [B12 ]. (Based on
NRTL activity coefficient model all other parameters from set 1 of Table 1.)
selected as an adjustable parameter. Through analyzing the sen-
sitivity of the model parameters and the system variables on the
azeotropic composition and location of bifurcation points, we
can adjust the model parameters and improve the prediction of
azeotropic points. Fig. 12 depicts a bifurcation diagram for ben-
Fig. 8. (a) Variation of benzene (1)–cyclohexane (2)–n-propanol (3) ternary
azeotropic composition with NRTL activity model parameter [B13 ] in a smaller
range. (Based on NRTL activity coefficient model all other parameters from
set 1 of Table 1.) (b) Variation of benzene (1)–cyclohexane (2)–n-propanol (3)
ternary azeotropic composition with NRTL activity model parameter [B13 ] in
a wider range. (Based on NRTL activity coefficient model all other parameters
from set 1 of Table 1.)
7
Fig. 9. Variation of bifurcation point on binary benzene (1)–cyclohexane (2)
branch with NRTL activity coefficient model parameter [B21 ] for benzene
(1)–cyclohexane (2)–n-propanol (3) system (all other parameters from set 1
of Table 1).
Fig. 10. Variation of bifurcation point on binary benzene (1)–cyclohexane (2)
branch with NRTL activity coefficient model parameter [B12 ] for benzene
(1)–cyclohexane (2)–n-propanol (3) system (all other parameters from set 1
of Table 1).
zene (1)–cyclohexane (2) system at atmospheric pressure. The
bifurcation diagram shows two set of binary branches, one is
based on unadjusted parameters (set #1) of Table 1 and other
is based on adjusted parameters (set #2) of Table 1. Fig. 12
Fig. 11. Variation of bifurcation point on binary benzene (1)–cyclohexane (2)
branch with NRTL activity coefficient model parameter [B31 ] for benzene
(1)–cyclohexane (2)–n-propanol (3) system (all other parameters from set 1
of Table 1).
8 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14

Fig. 13. Variation of ethyl acetate (1)–ethanol (2)–water (3) ternary azeotropic
Fig. 12. Bifurcation diagram for benzene (1)–cyclohexane (2) system at atmo- composition with pressure. (Based on NRTL activity coefficient model param-
spheric pressure. (Based on NRTL activity coefficient model parameters set 1 eters set 2 from Table 1.)
and 2 from Table 1.)

indicates that binary branches based on parameters (set #2) of

Table 1 are exactly matching the experimental data of Thornton
and Garner [12] while not compromising the accuracy of the
binary azeotrope prediction.
Table 3 contains the results of a predictor–corrector type
of algorithm used in implementing the numerical continua-
tion of sensitivity analysis for ternary benzene–cyclohexane–n-
propanol azeotrope with system variables and NRTL activ-
ity coefficient model parameters. The ternary benzene–
cyclohexane–n-propanol azeotrope is an unstable node in the
residue curve topology of this mixture. We monitored the varia-
tion in stability characteristics of this ternary azeotrope with sys-
tem variables and NRTL activity coefficient model parameters Fig. 14. Variation of ethyl acetate (1)–ethanol (2)–water (3) ternary azeotropic
by analyzing the eigenvalues. With variation of system variables composition with pressure and disappearance of ternary azeotrope into binary.
and activity coefficient model parameters the stability char- (Based on NRTL activity coefficient model parameters set 2 from Table 1.)
acter of ternary benzene–cyclohexane–n-propanol azeotrope
remained unchanged. Figs. 15 and 16 shows the effect of non-randomness parameter
[α12 ] and [α13 ] on ternary azeotropic composition. Fig. 17a–c
6.2. Ethyl acetate–ethanol–water system depicts the effect of NRTL activity coefficient model parameter
[B23 ] on the ethyl acetate, ethanol and water composition in ethyl
Fig. 13 shows the variation of ethyl acetate (1)–ethanol acetate–ethanol–water ternary azeotrope. Figs. 18–20 shows
(2)–water (3) ternary azeotrope with pressure. Fig. 14 shows the the variation of ternary ethyl acetate–ethanol–water azeotrope
disappearance of ternary ethyl acetate (1)–ethanol (2)–water (3) composition with NRTL activity coefficient model parame-
azeotrope with increase in pressure at around 16.0 atmospheres. ters [B32 ], [B12 ] and [B21 ], respectively. Fig. 21 shows the

Table 3
Implementation results of a predictor–corrector type algorithm for sensitivity analysis of ternary benzene (1)–cyclohexane (2)–n-propanol (3) azeotrope with system
variables and NRTL activity coefficient model parameters
Benzene–cyclohexane–n-propanol Pressure Sensitivity of Sensitivity of Sensitivity of Sensitivity of NRTL
ternary azeotrope sensitivity sensitivity non-randomness non-randomness non-randomness energy parameter
Fig. 3 parameter [α12 ] parameter [α13 ] parameter [α23 ] [B12 ] Fig. 7
Fig. 4 Fig. 5 Fig. 6

Execution time (s) 5.21 0.23 0.422 0.32 0.45

Maximum step size 0.25 0.08 0.08 0.1 0.1
Initial step size 0.01 0.001 0.001 0.01 0.01
Maximum number of Newton 3.0 3.0 3.0 3.0 3.0
iterations
Minimum step size 0.01 0.001 0.001 0.01 0.001
Number of steps 1000.0 40.0 35.0 22.0 43.0
Convergence criteria per step 10−6 10−6 10−6 10−6 10−6
 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14 9

Fig. 15. Variation of ethyl acetate (1)–ethanol (2)–water (3) ternary azeotropic Fig. 16. Variation of ethyl acetate (1)–ethanol (2)–water (3) ternary azeotropic
composition with non-randomness parameter [α12 ]. (Based on NRTL activity composition with non-randomness parameter [α13 ]. (Based on NRTL activity
coefficient model all other parameters from set 1 of Table 1.) coefficient model all other parameters from set 1 of Table 1.)

variation of location of bifurcation point appearing on ethyl
acetate–ethanol binary branch with pressure. Figs. 22–25 shows
the effect of [α13 ], [B12 ], [B23 ] and [B21 ] on the location of
a bifurcation point appearing on ethyl acetate–ethanol binary
branch in ethyl acetate (1)–ethanol (2)–water (3) system. With
the adjusted parameters, the ternary azeotropic composition
for ethyl acetate–ethanol–water system is closer to experimen-
tally reported values of Gmehling et al. [5,6] as shown in
Tables 4 and 5. Table 6 contains the numerical implementa-
tion results of sensitivity analysis of ethyl acetate (1)–ethanol
(2)–water (3) ternary azeotrope with system variables and activ-
ity coefficient model parameters.

Fig. 17. (a) Variation of ethyl acetate (1) mole fraction in ternary azeotropic composition of ethyl acetate–ethanol–water with NRTL activity model parameter [B23 ].
(Based on NRTL activity coefficient model all other parameters from set 1 of Table 1.) (b) Variation of ethanol (2) mole fraction in ternary azeotropic composition of
ethyl acetate–ethanol–water with NRTL activity model parameter [B23 ]. (Based on NRTL activity coefficient model all other parameters from set 1 of Table 1.) (c)
Variation of water (3) mole fraction in ternary azeotropic composition of ethyl acetate–ethanol–water with NRTL activity model parameter [B23 ]. (Based on NRTL
activity coefficient model all other parameters from set 1 of Table 1.)
10 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14

Fig. 21. Variation of bifurcation point on binary ethyl acetate (1)–ethanol (2)
branch with pressure for ethyl acetate (1)–ethanol (2)–water (3) system.
Fig. 18. Variation of ternary azeotropic composition of ethyl acetate–ethanol–
water with NRTL activity model parameter [B32 ]. (Based on NRTL activity
coefficient model all other parameters from set 1 of Table 1.)

Fig. 22. Variation of bifurcation point on binary ethyl acetate (1)–ethanol (2)
branch with non-randomness parameter [α13 ]. For ethyl acetate (1)–ethanol
(2)–water (3) system.

Fig. 19. Variation of ternary azeotropic composition of ethyl acetate–ethanol–

water with NRTL activity model parameter [B12 ]. (Based on NRTL activity
coefficient model all other parameters from set 1 of Table 1.)

Fig. 23. Variation of bifurcation point on binary ethyl acetate (1)–ethanol (2)
branch with NRTL activity coefficient model parameter [B12 ] for ethyl acetate
(1)–ethanol (2)–water (3) system.

Table 4
Ternary azeotrope of ethyl acetate–ethanol–water system at atmospheric pres-
sure (computed and experimental values)
Ethyl acetate (1), ethanol (2), water (3)

Experimental data 0.5789,0.1126,0.3085, T = 343.8 K

Fig. 20. Variation of ternary azeotropic composition of ethyl acetate–ethanol– Without adjustment of parameters 0.5346,0.1336,0.3318, T = 343.0137 K
water with NRTL activity model parameter [B21 ]. (Based on NRTL activity After adjustment of parameters 0.5735,0.1089,0.3172, T = 343.55 K
coefficient model all other parameters from set 1 of Table 1.)
 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14 11

Fig. 24. Variation of bifurcation point on binary ethyl acetate (1)–ethanol (2) Fig. 25. Variation of bifurcation point on binary ethyl acetate (1)–ethanol (2)
branch with NRTL activity coefficient model parameter [B23 ] for ethyl acetate branch with NRTL activity coefficient model parameter [B21 ] for ethyl acetate
(1)–ethanol (2)–water (3) system. (1)–ethanol (2)–water (3) system.

6.3. Water–formic acid–acetic acid system
Wisniak and Tamir [18] reported that water–formic
acid–acetic acid system has a ternary azeotrope. When we
tried to calculate this ternary azeotrope through the NRTL
model parameters available from Gmehling and Onken [7]
[DECHEMA Series 1977] (Table 7 parameter set 1) we
were unable to predict a ternary azeotrope although a binary
water–formic acid azeotrope was calculated. We conducted a
systematic sensitivity analysis and compiled a list of NRTL
parameters (parameter set 2–set 8 Table 7). The parameter
sets described in Table 7 are used to compute the bifurcation

Table 5
NRTL activity coefficient parameters for ethyl acetate (1)–ethanol (2)–water (3) system
NRTL parameters α12 α13 α23 B12 B21 B13 B31 B23 B32

Set #1 based on binary information 0.30 0.35 0.30 282.90 524.40 1286.10 −1705.70 246.10 −586.08
Set #2 adjusted parameters 0.24 0.42 0.32 318.50 575.20 943.40 −1285.50 214.50 −626.80

Table 6
Implementation results of a predictor–corrector type algorithm for sensitivity analysis of ternary ethyl acetate (1)–ethanol (2)–water (3) azeotrope with system
variables and NRTL activity coefficient model parameters
Ethyl acetate–ethanol–water Pressure Sensitivity of Sensitivity of Sensitivity of Sensitivity of
ternary azeotrope sensitivity sensitivity non-randomness non-randomness NRTL energy NRTL energy
Fig. 13 parameter [α12 ] parameter [α13 ] parameter [B12 ] parameter [B32 ]
Fig. 14 Fig. 15 Fig. 16 Fig. 17

Execution time (s) 4.87 0.46 0.58 0.35 0.58

Maximum step size 0.1 0.05 0.09 0.08 0.08
Initial step size 0.01 0.01 0.01 0.005 0.009
Maximum number of Newton 4.0 3.0 3.0 3.0 3.0
iterations
Minimum step size 0.01 0.01 0.01 0.005 0.009
Number of steps 860.0 20.0 25.0 62.0 68.0
Convergence criteria per step 10−6 10−6 10−6 10−6 10−6

Table 7
Parameter sets for water (1)–formic acid (2)–acetic acid (3) system (based upon NRTL model)
NRTL parameter set α12 α13 α23 B12 B21 B13 B31 B23 B32

#1 0.304 0.299 0.302 −297 93.2 646 −20.2 361 −386

#2 0.352 0.335 0.245 −455 65.8 366 −16.8 495 −108
#3 0.250 0.270 0.450 −456 33.2 1206 −10.2 405 −200
#4 0.300 0.200 0.300 −637 73.2 327 −35.2 305 −260
#5 0.300 0.300 0.300 −327 53.2 246 −40.2 495 −38.8
#6 0.250 0.200 0.300 −255 33.2 1046 −18.2 695 −38.3
#7 0.350 0.300 0.300 −487 42.3 866 −95.4 625 −38.3
#8 0.300 0.220 0.350 −197 85.2 476 −48.1 765 −38.3
12 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14

Table 8a
Experimental data of [17] for ternary azeotrope of water (1)–formic acid
(2)–acetic acid (3) system at atmospheric pressure
Water (%) 46.9
Formic acid (%) 36.3
Acetic acid (%) 16.8

Table 8b
Computed values of ternary azeotrope for water (1)–formic acid (2)–acetic acid
(3) system at atmospheric conditions with different parameter set values
NRTL parameter set Water (%) Formic acid (%) Acetic acid (%)

#1 None None None

Fig. 26. Water mole fraction vs. the homotopy parameter for ternary system: #2 30.70 57.20 12.09
water (1)–formic acid (2)–acetic acid (3). #3 32.80 59.45 7.74
#4 35.15 59.83 5.01
#5 30.78 54.49 14.73
#6 34.94 45.17 19.2
#7 32.84 48.64 18.52
#8 37.45 41.66 20.89

data for water–formic acid–acetic acid of Wisniak and Tamir

Fig. 27. Formic acid mole fraction vs. the homotopy parameter for ternary sys-
tem: water (1)–formic acid (2)–acetic acid (3).
diagrams. Fig. 26 contains the bifurcation diagram for
water–formic acid–acetic acid system depicting the variation
of water mole fraction with homotopy parameter for indicated
system under different parameter sets from Table 7. Fig. 27
contains the bifurcation diagram for water–formic acid–acetic
acid and depicts the variation of formic acid mole fraction
with homotopy parameter. For each parameter set of Table 7,
a continuation branch is tracked as shown in Figs. 24 and 25.
Fig. 28 shows the temperature bifurcation diagram for the same
system. Table 8a indicates the experimental ternary azeotropic
[18]. Table 8b contains the computed ternary azeotropic com-
position for water–formic acid–acetic acid system. For each
parameter set of Table 7, there is a difference among computed
values of ternary azeotrope as shown in Table 8b. The parameter
set 6 provides the ternary azeotropic composition, which is
close enough to experimentally reported values of Wisniak and
Tamir [18]. This example provides a basis about systematic
adjustment of phase equilibrium model parameters through
sensitivity analysis to improve the prediction of ternary and
higher order azeotropic points. Table 9 contains the variation
of homotopy parameter values at bifurcation point located
on binary water–formic acid branch for different parameter
sets. For parameter set 1, (Gmehling and Onken [7], vol. I/1)
the bifurcation point is located at homotopy parameter value
of 1.29 ruling out the possibility of any ternary azeotrope.
While for parameter set 6, which provide the computed value
of ternary azeotrope closer to experimentally reported data
has bifurcation point located at homotopy parameter value of
0.59. Figs. 29–31 depict the effect of model parameters on the
location of bifurcation point on pure formic acid branch.

Table 9
Homotopy continuation parameter at bifurcation point on a water–formic acid
binary branch for different set of NRTL parameter sets from Table 7
NRTL parameter set Homotopy continuation parameter at bifurcation
point on a water–formic acid binary branch

#1 1.290
#2 0.653
#3 0.743
#4 0.817
#5 0.602
#6 0.590
#7 0.503
Fig. 28. Temperature bifurcation diagram for ternary system: water (1)–formic #8 0.419
acid (2)–acetic acid (3).
 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14 13

the ternary and higher order azeotropic points is minimized if

not eliminated. The parametric perturbation of azeotropic states
and associated characteristic points in homotopy continuation
maps also provides important insights regarding the limitations
and capabilities of a phase equilibrium model in predicting the
azeotropic states in multicomponent mixtures. The sensitivity
analysis of azeotropic points also provides a basis to judge the
quality of fit. The sensitivity of activity coefficient model param-
eters on azeotropic conditions is examined through homotopy
continuation approach. Effect of system variables on azeotropic
states and associated bifurcation points can also be analyzed
through our approach.

Fig. 29. The variation of bifurcation point on pure formic acid branch with List of symbols
non-randomness parameter [α12 ]. [F] matrix for the necessary condition of azeotropy
[
F] Jacobian matrix of F
KiIdeal equilibrium constant for ideal solution
KiReal equilibrium constant for non-ideal solution
Ki0 pseudo equilibrium constant
Kij∞ infinite dilution equilibrium constant
Kij derivative of equilibrium constant with respect to com-
ponent j
P total pressure of the system
Pisat vapor pressure of component i
s arc length
T temperature
xi mole fraction of the component i in the liquid phase
yiIdeal mole fraction of the component i in the vapor phase for
an ideal gas and ideal solution
Fig. 30. The variation of bifurcation point on pure formic acid branch with yiReal mole fraction of the component i in the vapor phase for
NRTL activity coefficient model parameter [B12 ]. a real gas and non-ideal solution

Greek letters
λ homotopy parameter
λij bifurcation point
γj activity coefficient
γj∞ activity coefficient at infinite dilution

References

[1] E.L. Allgower, K. Georg, Numerical Continuation Methods: An Intro-

Fig. 31. The variation of bifurcation point on pure formic acid branch with
NRTL activity coefficient model parameter [B21 ].
7. Conclusion
This paper highlights a systematic approach to examine the
sensitivity of activity coefficient model parameters and system
variables on the prediction of azeotropes in multicomponent
mixtures. The approach provides a systematic basis to adjust
the parameters, which are based on binary phase equilibrium
information in such a way that error associated with extrapo-
lating the parameters based upon binary information to predict
duction, Springer, Berlin, 1990.
[2] N. Aslam, A.K. Sunol, Reliable computation of binary homogeneous
azeotropes of multi-component mixtures at higher pressures through
equation of states, Chem. Eng. Sci. 59 (2004) 599–609.
[3] P. Englezos, N. Kalogerakis, Applied Parameter Estimation for Chemical
Engineers, Marcel Dekker, New York, 2001.
[4] Z.T. Fidkowski, M.F. Malone, M.F. Doherty, Computing azeotropes in
multi-component mixtures, Comput. Chem. Eng. 17 (1993) 1141–1155.
[5] J. Gmehling, J. Menke, J. Krafczyk, K. Fischer, Azeotropic Data. Part
I, Weinheim VCH, 1994.
[6] J. Gmehling, J. Menke, J. Krafczyk, K. Fischer, Azeotropic Data. Part
II, Weinheim VCH, 1994.
[7] J. Gmehling, U. Onken, Vapor–Liquid Equilibrium Data collection, vol.
I/1. DECHEMA Chemistry Data Series, 1977.
[8] J.P. Knapp, M.F. Doherty, A new pressure swing distillation process for
separating homogeneous azeotropic mixtures, Ind. Eng. Chem. Res. 31
(1992) 346–357.
[9] M.F. Malone, M.F. Doherty, Conceptual Design of Distillation Systems,
McGraw Hill, NewYork, 2001.
14 N. Aslam, A.K. Sunol / Fluid Phase Equilibria 240 (2006) 1–14
[10] M.L. Rabinovich, Mathematical Modeling of Chemical Processes, Mir
Publishers, Moscow, 1992.
[11] A. Speight, Using homotopy methods to solve non-smooth equations,
M.S. Thesis, University of Colorado at Denver, 1999.
[12] J.D. Thornton, F.H. Garner, J. Appl. Chem. 1 (Suppl. 1) (1951)
561.
[13] J.E. Tolsma, Simulation and analysis of heteroazeotropic systems, Ph.D
Thesis, Massachusetts Institute of Technology, Cambridge, MA, June
1999.
[14] J.E. Tolsma, P.I. Barton, Computation of heteroazeotropes, Part I: theory,
Chem. Eng. Sci. 55 (2000) 3817–3834.
[15] J.E. Tolsma, P.I. Barton, Computation of heteroazeotropes, Part II: effi-
cient calculation of changes in phase equilibrium structure, Chem. Eng.
Sci. 55 (2000) 3835–3853.
[16] M.S. Walas, Phase Equilibria in Chemical Engineering, Butterworth Pub-
lishers, Boston, 1985.
[17] L.T. Watson, R.C. Melville, A.P. Morgan, H.F. Walker, A suite of FOR-
TRAN 90 codes for globally convergent homotopy algorithms, ACM
Trans. Math. Software 23 (1997) 514–549.
[18] J. Wisniak, A. Tamir, Vapor–liquid equilibria in the ternary systems
water–formic acid–acetic acid and water–acetic acid–propionic acid, J.
Chem. Eng. Data 22 (1977) 253–260.