You are on page 1of 11

Fluid Phase Equilibria 226 (2004) 149159

A versatile thermodynamic consistency test for incomplete phase


equilibrium data of high-pressure gasliquid mixtures
Jose O. Valderramaa,b, , Vctor H. Alvarezb
a

Mechanical Engineering Department, Faculty of Engineering, University of La Serena, Casilla 554, La Serena-Chile
b Centro de Informaci
on Tecnologica, Casilla 724, La Serena, Chile
Received 1 March 2004; received in revised form 3 June 2004; accepted 13 July 2004
Available online 5 November 2004

Abstract
A new method to test the thermodynamic consistency of phase equilibrium data in binary mixtures containing a liquid solute and a
supercritical fluid is presented. For the systems of interest, mixtures containing a liquid solute and a supercritical fluid, not only the PTxy
data are not available for the whole concentration range, but also the solute concentration in the gas phase is low (mole fractions from 101
down to 104 ). For these cases, the classical differential or integral methods described in standard books, are not applicable. The proposed
method is specially designed for treating incomplete PTxy data. That is, data that do not cover the whole range of concentration of the
components in the mixture, as those usually found in supercritical fluid mixtures. The method is based on the GibbsDuhem equation, on the
fundamental equation of phase equilibrium and on an appropriate combination between equations of state, mixing rules and combining rules.
The new method is applied to eleven isothermal sets of data for high-pressure binary mixtures and consistency criteria are defined.
2004 Elsevier B.V. All rights reserved.
Keywords: Thermodynamic consistency; Phase equilibrium; GibbsDuhem equation; High-pressure mixtures; Supercritical fluids

1. Introduction
The inaccuracies that arise in measuring experimental
phase equilibrium properties has made it necessary to come
up with methods to test inherent inaccuracies of such data.
Although it is difficult to be absolutely certain about the correctness of a given set of experimental data, it is possible to
check whether such data satisfy certain thermodynamic relationships, establishing that the data are thermodynamically
consistent or inconsistent. The thermodynamic relationship
that is frequently used to analyze thermodynamic consistency of experimental phase equilibrium data is the fundamental GibbsDuhem equation. The GibbsDuhem equation
Abbreviations: EoS, Equation of state; Max, Maximum value; NFC, Not
fully consistent; PR, PengRobinson EoS; Ref, Reference; TC, Thermodynamically consistent; TDM, Try a different model; TI, Thermodynamically
inconsistent; WS, WongSandler
Corresponding author. Tel.: +56 51 551158; fax: +56 51 551158.
E-mail address: citchile@entelchile.net (J.O. Valderrama).
0378-3812/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.fluid.2004.07.002

interrelates the activity coefficients, the partial Gibbs free energy, or the fugacity coefficients of all components in a given
mixture. Depending on the way in which the GibbsDuhem
equation is handled, different consistency tests have been derived. Among these are the Slope Test, the Integral Test, the
Differential Test and the Tangent-Intercept Test [13]. If the
GibbsDuhem equation is not obeyed then the data are inconsistent and can be considered as incorrect. If the equation
is obeyed, the data are thermodynamically consistent but not
necessarily correct.
Similar to the Van NessByerGibbs test [4], the consistency method proposed in this work can be considered as a
modeling procedure. This because a thermodynamic model
that can accurately fit the experimental data must be used
to apply the consistency test. The novelty of the proposed
method is that the empirical interaction parameters of the
model (kij in an equation of state, for instance), are first determined using the GibbsDuhem equation. With the optimum
parameter found using an optimization routine, the funda-

150

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159

mental equation of phase equilibrium, expressed through the


equality of fugacity of each component in all phases, is expected to be fulfilled.
The GibbsDuhem equation in terms of residual properties applied to the gas phase mixture is [3]:



HR
VR
R
yi d[G
]
=

dT
+
dP
i
RT
RT 2

(1)

HR
VR
dT
+
dP
RT
RT 2
For high-pressure vaporliquid equilibrium, especially in
systems containing a supercritical solvent, there are some
problems for testing experimental data: (i) the vapor phase
non-idealities are important and a good model to evaluate the
fugacity coefficients i in Eq. (1) is needed; (ii) for isothermal
data the term involving the residual enthalpy (HR ) vanishes,
but the term involving the residual volume (VR ) cannot be
ignored as done at low pressure; (iii) the data available do
not cover the whole concentration range for both, the liquid
and the gas phase; and (iv) the concentration of the solute
in the supercritical solvent y2 is usually low (mole fractions
from 101 down to 104 ).
Some tests for treating high-pressure VLE data have been
presented in the literature. Chueh et al. [5], developed an
equal-area test, based as usual on the GibbsDuhem equation. The method requires a model for calculating the fugacity coefficients but some problems arise for evaluating the
areas in the zero concentration limit. Won and Prausnitz [6]
presented a method to analyze isothermal data. The method is
complex and requires the definition of arbitrary functions to
represent the variation of an activity coefficient with concentration and the molar volume of the mixture with the pressure.
Christiansen and Fredenslund [7] presented a method to test
high-pressure isobaric or isothermal data. This method includes the calculation of several thermodynamic properties
such as the standard state fugacity, excess enthalpy, fugacity
coefficient, activity coefficient, excess Gibbs free energy, all
which make the method complex to be applied. Muhlbauer
[8] presented a test based only on the concentration of the vapor phase, method that is essentially similar to that of Chueh
et al. [5]. Since this method does not use the liquid phase
concentration, an important part of the available experimental data is left out. Also, Jackson and Wilsak [9] analyzed
several consistency tests, mainly for complete high-pressure
vaporliquid equilibrium data, meaning data for the whole
concentration range in both phases. More details and discussion on all these methods are given by Raal and Muhlbauer
[10] and Poling et al. [11].
We believe that a good consistency test method to analyze
high-pressure vaporliquid equilibrium data should fulfill the
following ten basic requirements: (i) use the GibbsDuhem
equation; (ii) use the fundamental equation of phase equilibrium, that is the equality of fugacities of a component in
all phases; (iii) use for testing, all the experimental PTxy
data available; (iv) does not necessarily require experimental
yi RT d[d ln i ] =

data for the whole concentration range and be applicable for


data in any range of concentration; (v) be able to correlate
the data within acceptable limits of deviations, deviations that
must be evenly distributed; (vi) requires few calculated properties; vii) be able to detect erroneous experimental points;
(viii) makes appropriate use of necessary statistical parameters; (ix) be simple to be applied, considering the complexity
of the problem to be solved; and (x) be able to conclude about
consistency if the defined criteria are not fulfilled.
The method proposed here fulfills these basic requirements and can conclusively determine the consistency or inconsistency of data in most cases, as demonstrated in this
work. For those situations in which the model chosen for
the calculation of the fugacity coefficients in Eq. (1) is not
accurate enough for correlating the pressure (P) and the solute concentration in the gas phase (y2 ), it is concluded that
another model should be tried (TDM). If the data is well
correlated, the test can conclude that the data is thermodynamically consistent (TC), thermodynamically inconsistent
(TI) or not fully consistent (NFC). All these situations are
detailed in another section.

2. Development of equations
Bertucco et al. [12], expressed the GibbsDuhem equation
for a binary mixture at constant temperature T in terms of
the fugacity coefficients 1 and 2 of the components in the
R = RT ln i and V R = RT (Z 1)/P for
mixture. Using G
i
a binary mixture at constant T, Eq. (1) becomes [3]:


(Z 1)
(2)
dP = y1 d(ln 1 ) + y2 d(ln 2 ).
P
For the method proposed in this work, this equation can be
more conveniently written in terms of the solute concentration in a gas phase mixture. If the solute is component 2 in
the binary mixture, the above equation becomes:
y2 dLn 2
1 dP
(1 y2 ) dLn 1
=
+
.
P dy2
(Z 1) dy2
(Z 1) dy2

(3)

This equation can be conveniently expressed in integral


form, as follows:



1
(1 y2 )
1
dP =
d 2 +
d 1 . (4)
Py2
(Z 1)2
y2 (Z 1)1
In this equation P is the system pressure, y2 is the gas phase
solute mole fraction, 1 and 2 are the fugacity coefficients
of components 1 and 2 in the gas phase mixture, and Z is the
compressibility factor of the gas mixture, calculated using an
EoS. In Eq. (4) the left hand side is designated by AP and the
right hand side by A , as follows:

1
dP
(5)
AP =
Py2
A = A1 + A2

(6)

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159


A1 =

(1 y2 )
d 1
y2 (Z 1)1

A2 =

1
d 2 (7)
(Z 1)2

Thus, if a set of data is considered to be consistent AP


should be equal to A within acceptable defined deviations.
To set the margins of errors an individual percent area deviation %Ai between experimental and calculated values is
defined as:
%Ai = 100(A AP )/AP

(8)

In Eq. (5) AP is determined using the Py data of the experimental data set PTxy, while a thermodynamic model,
such as an equation of state, is employed for A in Eq. (7).
If the data are adequately correlated, meaning that the deviations in the calculated pressure and solute concentration in
the gas phase are within acceptable margins of deviations and
the individual area deviation %Ai are within defined margins of errors, the data set is considered to be consistent. A
detailed description about the acceptable errors in predicting
these VLE properties and the areas AP and A are given in
the Appendix A.
The deviations in the calculated pressure and solute concentration in the gas phase for each point i are defined as:
exp

exp

%Pi = 100(Pical Pi )/Pi


exp

(9)

exp

cal
%y2i = 100(y2i
y2i )/y2i .

To evaluate the integrals A and A the following must be


defined: (i) an equation of state; (ii) a set of mixing rules; and
(iii) a set of combining rules. In principle, any appropriate
equation of state and any mixing and combining rules can be
used to evaluate the pressure and the solute concentration in
the gas phase. However, with most models used to correlate
equilibrium data in mixtures containing a supercritical component the error in y2 increases as the error in P decreases
[13]. Thus, the errors in these two variables must be low and
must give low average deviations to accept the model for the
proposed thermodynamic consistency test.
It is proposed here to use the PengRobinson EoS with the
WongSandlerUNIQUAC mixing rules with one interaction
parameter k12 as the default thermodynamic model, to evaluate the fugacity coefficients and the variables Py, for given
values of Tx in the integrals that appear in Eq. (7). Despite
some limitations of the WongSandler mixing rule pointed
out in the literature [14], several works and our own findings
have clearly demonstrated that the WongSandler mixing rule
has the accuracy and necessary flexibility to correlate phase
equilibrium variables in high-pressure systems containing a
supercritical fluid [1518,13]. Therefore we have chosen this
mixing rule as the default model for the proposed consistency
test.
The PengRobinson equation and the WS mixing rules
used as a default model can be expressed as follows [19]:
P=

RT
a
+
V b V (V + b) + b(V b)

(10)

151

where:
a = 0.457235(R2 Tc 2 /Pc )(Tr )

b = 0.077796(RTc /Pc )

(Tr )0.5 = [1 + F (1 Tr0.5 )]

(11)

F = 0.37646 + 1.54226 0.26992 2


For mixtures:
P=

RT
am
+
V bm
V (V + bm ) + bm (V bm )

(12)

In this equation am and bm are the equation of state constants to be calculated using defined mixing rules. For the PR
equation the WS mixing rule can be summarized as follows
[20]:
bm =

xi xj (b a/RT )ij

1 xi ai /bi RT AE
(x)/RT

ai aj
1
= [bi + bj ]
(1 kij )
2
RT
  

ai
AE
(x)
xi
+
am = bm
bi

a 
RT ij

(13)

In these equations am and bm are the equation of state constants,  = 0.34657 for the PR equation, and AE
(x) is
calculated using the UNIQUAC model and assuming that
E
E
E
AE
(x) Ao (x) Go (x), being Go (x) the excess Gibbs
free energy at low pressure [21]. The UNIQUAC model includes two empirical parameters A12 and A21 , besides some
molecular parameters as detailed in the literature [11]. Therefore, for a binary mixture this model includes three adjustable
parameters: A12 and A21 in the UNIQUAC model and one k12
parameter in the combining rule for (ba/RT)12 .
The equations for Z, 1 and 2 using the PengRobinson
equation and the WS mixing rules are given by Orbey and
Sandler [16]. The variables required to evaluate A in Eq.
(7), that is P, y2 , 1 and 2 , are determined by applying
the fundamental equation of phase equilibrium: at a given
temperature and liquid concentration, the fugacity of each
component in the gas phase must be equal to the fugacity of
the same component in the liquid phase. That is:
liq

gas

f1 (T, P, x1 ; A12 , A21 , k12 ) = f1 (T, P, y1 ; A12 , A21 , k12 )


(14)
and
liq

gas

f2 (T, P, x2 ; A12 , A21 , k12 ) = f2 (T, P, y2 ; A12 , A21 , k12 )


(15)
The calculated pressure and gas phase solute concentration
are those determined by the above equations, with parameters
A12 , A21 and k12 determined from experimental data. An optimization routine based on the Marquardt method was used,
as discussed in another section.

152

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159

3. Consistency criteria
To define the criteria for consistency and inconsistency it
is first required that the model be able to correlate the data
within acceptable deviations. The model is accepted if the
deviations defined by Eq. (9) are within 10% to +10%. for
%P and 20% to 20% for %y2 . After the model is found
appropriate, we require that the deviations in the individual
areas defined by Eq. (7) are all within the limits 20% to
+20% to declare the data as being thermodynamically consistent. All these criteria are summarized in Table 1. Also,
our results indicate that six is the minimum number of points
required for the average deviations defined as consistency criteria to have some reasonable meaning. Details on the reasons
for defining these limits are given in Appendix A.
The method proposed here is based on the hypothesis that
true experimental data should be randomly distributed and
therefore random distribution of errors or low errors between
calculated and experimental values data should be found.
Therefore, if an EoS model produces randomly distributed errors of pressure and gas phase solute concentration and these
values are within the defined acceptable limits (Table 1), the
proposed area test represents a reasonable criterion to accept
or reject a set of data from the thermodynamic consistency
point of view. Based on the criteria defined in Table 1, the
following decision rules are applied:
(1) If percentage y2 and percentage P are outside the
defined margins of errors (10 to +10) for P and
(20 to +20) for y2 , a different model (other than
PR+WS+UNIQUAC must be used. This case is designated as TDM (try a different model).
(2) If the model acceptably correlates the data and the area
test is fulfilled for all points in the data set, the proposed
method is conclusive and the data are considered to be
thermodynamically consistent (TC).
(3) If the model acceptably correlates the data and the area
test is not fulfilled for most of the points in the data set
(more than 75% of the areas), the proposed method is
conclusive and the data are considered to be thermodynamically inconsistent (TI).
(4) If the model acceptably correlates the data and some of
the area deviations (equal or less than 25% of the areas)
are outside the limits defined in Table 1, the proposed
method declares the data as being not fully consistent
(NFC).

(5) Not fully consistent data could be further analyzed to


check, if after eliminating some points, the remaining
data fulfill the criteria defined in Table 1 and these remaining data are consistent or inconsistent.
The Fig. 1, shows a flow diagram, clarifying the different
situations that can be found when the proposed method is
applied.
Before applying the consistency method, some aspects related to the objective function, to the optimization method, to
the regression program, to the multiplicity of solutions and
to the trapezoidal integration method used to evaluate the
integrals AP and A , need to be discussed. In the proposed
consistency test method the objective function is W="(AP
A )2 /(N1). That is, the square of the difference between
experimental and calculated areas for the (N1) areas determined using two consecutive points of the data set, as explained before here. The programs developed for this study
explores for multiple acceptable solutions for the adjustable
parameters, all which are stored and reported in our programs.
The multiple solutions are found by starting the iterative procedure from different values between pre-defined ranges that
are considered meaningful and acceptable for the different
parameters. This implies more computer time, but guarantees
the finding of all acceptable solutions in the defined ranges.
The final solution is that which give low deviation of the established objective function and show parameters with some
physical meaning. The definition of low deviation is discussed in the next section.
The non-linear regression program used for this study
considers the use of a modified steepest descent Marquardt
method, a non-linear least-squares procedure commonly
used as optimization routine [22]. As known, one of the
difficulties with Marquardts method (and for that matter any
steepest descent method) is that the solution can converge to
a local minimum rather than to the desired global minimum.
However, we have found that this apparent disadvantage
is very useful for the proposed consistency test method
since in correlating vaporliquid equilibrium data the global
minimum is not necessarily the optimum solution. This
because one expects that the adjusted parameters not only accurately correlate the experimental data but also have certain
physical meaning, meaning that mathematics only cannot
provide.
As known, multiplicity of solutions in EoS models may
not be simply analyzed by finding the solution belonging

Table 1
Deviations for the pressure, the solute concentration in the gas phase and the individual areas defined for the proposed consistency test method
Parameter

Formula

Criterion

Individual percent area deviation


Individual deviation in the system pressure
Individual deviation in the gas phase solute concentration (y2 )
Average deviation in the system pressure
Average deviation in the gas phase solute concentration (y2 )

%Ai = 100(A AP )/AP


exp
exp
%Pi = 100(Pical Pi )/Pi
exp
exp
cal
%y2i = 100(y2i
y2i )/y2i
cal
exp
exp
(Pi Pi )/Pi
%Pav = 100
N
exp
exp
100 cal
(yi yi )/yi
%yav = N

[20% to +20%]
[10% to +10%]
[20% to +20%]
[10% to +10%]
[20% to +20%]

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159

153

Fig. 1. Decision rules for the proposed consistency test method.

to the smallest deviation for a defined objective function.


Also, thermodynamic stability and existence of phases must
be considered [23,24]. The optimization methods used in
this work have been shown to be reliable in many applications and allow identification of local and global minima
[2527].
To evaluate the integral in Eq. (5) for a set of N experimental points, two consecutive data points are used, obtaining
N1 values of the integrals. The trapezoidal rule was used
to evaluate these integrals. The Fig. 2 shows the integrand
functions for AP , A1 and A2 for the mixture CO2 /n-butane
at 344 K. As shown in the Figure, the integration method is

justified since a straight line between two consecutive points


can be assumed without much error. Although the figure is
shown for just one case (CO2 +n-butane at 344 K), we have
found similar behavior for the other sets of data. To check
the validity of the integration method used, a third degree
polynomial was fitted to the integrand functions in Eq. (7)
and the integrals A1 and A were calculated using the fitted polynomial. The deviations between the areas calculated
using the trapezoidal rule and those determined using the
polynomial functions for these two integrals were below 2%.
These deviations are considered acceptable for consistency
analysis.

Table 2
Details for the experimental data used in this study. The values of temperature have been rounded to the closest integer
System

Solvent (1)/solute (2)

Range P (atm)

T (K)

Range x2

Range y2

Ref.

1
2

CO2 /n-butane
Ethylene/1-butene
CO2 /1octanol

8.580.6
2.654.3
69153
40168
40188
64183
64188
3175
4183
883
1197

344
293
313
328
348
403
453
314
324
314
323

0.287/0.998
0.069/1.000
0.186/0.508
0.759/0.138
0.795/0.131
0.366/0.745
0.405/0.769
0.025/0.432
0.047/0.499
0.113/0.927
0.092/0.925

0.2870/0.9700
0.0340/1.0000
0.0004/0.0804
0.0004/0.0808
0.0006/0.0672
0.0055/0.0270
0.0216/0.0527
0.0074/0.0182
0.0065/0.0130
0.0005/0.0079
0.0009/0.0150

[33]
[34]
[35]

18
15
6
7
7
9
7
15
12
8
11

CO2 /limonene

[28]
[29]
[30]

154

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159
Table 3
Properties of the pure components included in the mixtures studied. The
data for carbon dioxide were taken from Prausnitz [1], while for all the other
substances were taken from Daubert et al. [37]

Fig. 2. Integrand functions in Eqs. (5) and (7) for the system CO2 -butane at
344 K. (b) Integrand functions in Eqs. (5) and (6) for the system CO2 -butane
at 344 K.

4. Applications of the proposed method


Eleven isotherms for four binary mixtures were chosen to
show the application of the proposed thermodynamic consistency test. The mixtures were carefully selected so that
various features of the test could be emphasized. Details on
the systems are shown in Table 2. The necessary properties of the pure substances (molecular mass M, critical temperature Tc , critical pressure Pc , and acentric factor ) involved in the mixtures selected for study are presented in
Table 3.
The mixtures CO2 /n-butane and ethylene/1-butene were
already examined by Bertucco et al. [12], and have been used
here to compare the results of the new proposed method.
The mixtures CO2 /1-octanol and co2 /limonene have not previously checked for consistency. However, in the literature
sources from where the data for these two systems were ob-

Components

Tc (K)

Pc (atm)

Carbon dioxide
n-butane
Ethylene
1-Butene
1-Octanol
Limonene

44.0
58.1
28.1
56.1
130.2
136.2

304.2
425.2
282.4
419.6
652.5
660.0

72.83
37.50
49.66
39.67
28.23
27.14

0.225
0.201
0.087
0.191
0.594
0.312

tained, some information is given about the accuracy of the


data provided.
Table 4 presents results for the mixtures CO2 /n-butane,
ethylene/1-butene, CO2 /1-octanol and CO2 /limonene. The
data set for the systems CO2 /n-butane and ethylene/1-butene
were found to be not fully consistent, meaning that one or
more points in the data set give area deviations outside the
defined range for P and y2 . In fact, for the mixture CO2 /nbutane only one point give high area deviation (58.7%),
while for the mixture ethylene/1-butene three points give
high area deviations (39.8, 30.5 and 29.6%). Since
these three points represent 20% of the original data set
of 15 points, the data set is also declared to be NFC. Further analysis of the remaining data is explained in the next
section.
The system CO2 /1-octanol at 313 K is found to be
thermodynamically consistent (TC), meaning that deviations
in P and y2 are within the defined ranges and all points in the
data set give area deviations within 20% to 20%. At 328 K
and 348 K deviations in P and y2 are high and two of the
seven original data points show high deviations. In this case,
the conclusion is that another model should be tried (TDM).
If the two points are eliminated the data set is reduced to only
five points, less than the defined minimum of six points to
apply the test. At 403 and 453 K the data of Weng et al. [28]
were found to be NFC, meaning that some few areas give
high deviations and only one point was not well correlated.
Further analysis was done on these data and detail on this is
presented in the next section. For the system CO2 /Limonene,
the data of Gamse and Marr [29] and of Chang and Chen [30]
are not acceptable correlated and therefore a different model
should be used (TDM). In this last case the comparison
of different experimental data presented in the literature
for the mixture CO2 /limonene shows important differences
between the data reported. In addition, the model used
fit the data with unusual high deviations and the data
could also be declared as thermodynamically inconsistent
(TI)

5. Not fully consistent data


The works of Bertucco et al. [12] and of Jackson and
Wilsak [9] considered the uncertainty of the experimental

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159

155

Table 4
Details of the consistency test for the systems studied using the PR+WS/UNIQUAC model. In the Table, C stands for consistent, TI stands for thermodynamically
inconsistent, TDF for try a different model and NFC for not fully consistent. Also k12 , A12 , A21 are parameters in the WS/UNIQUAC mixing rule
Solvent (1)/solute (2)

Ref.

T (K)

CO2 /n-butane
Ethylene/1-butene
CO2 /1-octanol

[33]
[34]
[36]

344
293
313

18
15
6

328
348
403
453
314
324
314
323

[28]
CO2 /limonene

[29]

k12

%P

Maximum
%P

%y2

41.75
860.74
667.96

0.8
3.1
0.2

4.35
4.1
1.7

2.6
0.3
6.3

239.7
217.07
360.13
213.16

527.4
510.50
381.66
501.54

3.8
12.1
10.0
9.8

10.2
46.8
20.0
14.3

1845.59
436.79
400.87
134.31

60.79
749.56
1456.01
666.98

15.5
23.5
7.1
1.3

60.5
74.9
21.5
18.9

A12

A21

0.1305
0.1618
0.1458

1120.18
496.93
194.39

7
7
9
7

0.1210
0.1954
0.1539
0.1553

15
12
8
11

0.2741
0.0314
0.2063
0.1381

Maximum
%y2

Maximum
%Ai

Result

12.1
15.4
17.6

58.7
39.1
13.4

NFC
NFC
TC

2.3
6.6
9.3
1.6

35.1
31.7
18.3
14.3

35.4
79.5
34.9
47.4

TDM
TDM
NFC
NFC

45.8
18.6
20.6
32.5

70.9
35.6
154.2
94.6

606.2
99.8
1121.8
235.7

TDM
TDM
TDM
TDM

Table 5
Detailed results for the system CO2 /n-butane found to be not fully consistent in Table 5
% Ai

Pexp

Pcal

%P

y2 exp

y2 cal

%y2

CO2 + n-butane(2) (T = 344 K; 18 Data points; k12 = 0.1305; A12 = 1120.18; A21 = 41.75)
0.20
0.21
1.21
8.51
8.42
0.20
0.19
5.33
10.21
10.13
0.20
0.19
4.86
11.91
11.73
0.38
0.38
0.28
13.61
13.35
0.37
0.37
1.13
17.01
16.72
0.36
0.36
1.52
20.41
20.13
0.35
0.36
5.30
23.82
23.56
0.34
0.36
4.79
27.22
27.14
0.33
0.34
4.16
30.62
30.74
0.64
0.67
4.49
34.02
34.35
0.60
0.63
4.19
40.83
41.71
0.56
0.58
3.76
47.63
49.14
0.52
0.52
1.48
54.44
56.65
0.48
0.45
6.45
61.24
63.90
0.44
0.40
8.14
68.05
70.97
0.19
0.31
58.68
74.85
77.35
0.11
0.11
2.07
78.26
79.93

80.57
80.57
% P = 1.9, % y1 = 2.7, % y2 = 3.1, % Aav = 6.9

1.11
0.80
1.47
1.92
1.69
1.38
1.07
0.28
0.40
0.97
2.17
3.18
4.05
4.35
4.29
3.34
2.13
0.01

0.970
0.827
0.723
0.645
0.538
0.464
0.408
0.365
0.332
0.306
0.268
0.246
0.230
0.220
0.216
0.222
0.242
0.287

0.977
0.832
0.733
0.656
0.542
0.465
0.410
0.368
0.335
0.310
0.274
0.251
0.237
0.231
0.233
0.249
0.267
0.275

0.67
0.59
1.32
1.67
0.74
0.24
0.50
0.73
1.00
1.35
2.19
1.94
2.98
4.99
8.03
12.11
10.35
4.28

CO2 + n-butane(2) (T = 344 K; 17 Data points; k12 = 0.1402; A12 = 818.95; A21 = 18.53)
0.20
0.21
0.67
8.51
8.41
0.20
0.19
5.90
10.21
10.12
0.20
0.19
5.48
11.91
11.71
0.38
0.38
0.94
13.61
13.32
0.37
0.37
0.45
17.01
16.66
0.36
0.36
0.83
20.41
20.03
0.35
0.36
4.68
23.82
23.43
0.34
0.35
4.24
27.22
26.97
0.33
0.34
3.71
30.62
30.53
0.64
0.66
4.09
34.02
34.09
0.60
0.63
4.30
40.83
41.34
0.56
0.59
4.38
47.63
48.66
0.52
0.53
0.72
54.44
56.04
0.48
0.47
2.64
61.24
63.25
0.62
0.65
4.87
68.05
70.35
0.11
0.11
3.21
78.26
79.85

80.57
80.57
% P = 1.6, % y1 = 2.6, % y2 = 2.0, % Aav = 3.2

1.13
0.90
1.65
2.17
2.05
1.84
1.63
0.91
0.31
0.20
1.25
2.16
2.94
3.28
3.38
2.03
0.01

0.970
0.827
0.723
0.645
0.538
0.464
0.408
0.365
0.332
0.306
0.268
0.246
0.230
0.220
0.216
0.242
0.287

0.977
0.833
0.733
0.657
0.543
0.466
0.410
0.368
0.335
0.310
0.273
0.249
0.234
0.226
0.226
0.252
0.260

0.69
0.66
1.43
1.81
0.89
0.37
0.58
0.74
0.93
1.17
1.70
1.03
1.52
2.71
4.60
4.16
9.59

AP

Of the 18 original points, the datum in bold face characters on the first block of this Table is eliminated. The numbers on the bottom block of the Table are the
results with the remaining 17 points.

156

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159

data to show thermodynamic consistency. Thus, to apply


these methods you need to know the accuracy of the experimental data to be tested, to then see if they pass a defined
thermodynamic test. The method proposed here is independent of such information. If the data have high experimental
errors the different criteria defined in Table 1 will consider
such errors and will find that the data set is inconsistent.
If some data are erroneous, then the method will give high
deviations for some of the individual areas and the set will be
declared as not fully consistent (NFC). These high individual
area deviations are found even in cases in which the solute
concentration in the gas phase and the pressure are correlated
with deviations within the limits given in Table 1. It should
be also noticed that when the areas AP and A are evaluated
using two consecutive points, each datum point influences
the area values (below and above). Thus, if the value of the
area deviation in one range is excessive, two datum points

Fig. 4. Residuals of pressure and solute concentration in the gas phase for
the system CO2 /n-octanol at 403 K.

Fig. 3. Deviations in the individual areas, the system pressure and the gas
phase solute concentration for the CO2 + n-butane(2) at 344.3 K.

are suspect. In addition, if one point is wrong, two areas are


influenced.
The Fig. 3 shows the deviations in the individual areas,
the system pressure and the gas phase solute concentration
for the mixture CO2 + n-butane(2) at 344.3 K. The Fig. 4
presents the residuals for the seven point set of data for the
system CO2 /n-octanol at 403K. This case is representative of
all cases declared as TC and NFC.
In all these cases we found that the best option is to first
eliminate the point giving the higher deviation in the gas
phase solute concentration and apply the method again. This
procedure is repeated with all points showing individual area
deviations (%Ai ), outside the range 20% to + 20%. If the
eliminated data points are more than 25% of the points in
the initial data set, the whole set is declared to be thermodynamically inconsistent (TI). This percentage has a frontier
character, since the explanatory character of the eliminated
data depends on the system. Also, it should be noticed that
we have defined that the minimum number of data points is
six to make a meaningful decision about consistency or inconsistency of the data set. Thus if a data set has less than six
points, the method cannot be applied. Also, if six points only
are available, the alternative not fully consistent (NFC) is
not possible because no point can be eliminated in this case,
to keep the number of points to the minimum six. Of course
if more points are available, better statistics can be done and
better conclusions about consistency can be drawn.

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159

157

Table 6
Detailed results for the system CO2 /1-octanol at 403 K found to be not fully consistent in Table 4, using the PR+WS/UNIQUAC MODEL
% Ai

Pexp

Pcal

%P

y2 exp

y2 cal

CO2 + 1-octanol(2) (T = 403 K; 9 Points; k12 = 0.1539; A12 = 360.13; A21 = 381.66)
38.02
35.21
7.39
64.15
64.72
29.20
28.21
3.41
78.95
80.60
26.67
27.32
2.42
93.76
97.74
13.25
15.09
13.95
113.50
121.70
9.31
10.08
8.27
128.30
141.55
6.68
7.67
14.82
143.10
160.40
3.36
4.53
34.86
157.91
180.70
3.79
4.61
21.48
167.78
197.30

182.58
220.74
% P = 10.0, % y1 = 0.1, % y2 = 9.3, % Aav = 13.3

0.88
2.09
4.24
7.22
10.33
12.09
14.43
17.59
20.90

0.0055
0.0055
0.0065
0.0083
0.0107
0.0132
0.0170
0.0193
0.0269

0.0061
0.0065
0.0073
0.0091
0.0113
0.0140
0.0177
0.0215
0.0283

10.61
18.31
12.77
9.93
5.26
5.68
4.08
11.50
5.20

CO2 + 1-octanol(2) (T = 403 K; 7 Points; k12 = 0.1780; A12 = 133.74; A21 = 567.41)
38.02
35.01
7.92
64.15
64.81
29.20
28.06
3.91
78.95
80.07
26.67
27.30
2.35
93.76
96.26
13.25
15.24
15.07
113.50
118.37
9.31
10.31
10.73
128.30
136.22
6.68
7.96
19.11
143.10
152.77

157.91
170.15
%P = 4.3, %y1 =0.07, %y2 = 7.6, %Aav = 9.8

1.03
1.42
2.67
4.29
6.17
6.76
7.75

0.0055
0.0055
0.0065
0.0083
0.0107
0.0132
0.0170

0.0059
0.0063
0.0070
0.0085
0.0102
0.0123
0.0152

7.54
14.11
7.42
2.22
4.54
6.68
10.81

AP

The Fig. 3 shows the deviations in the individual areas,


the system pressure and the gas phase solute concentration
for the mixture CO2 + n-butane(2) at 344.3 K. The Fig. 4
presents the residuals for the seven point set of data for the
system CO2 /n-octanol at 403 K. This case is representative
of all cases declared as TC and NFC.
Table 5 presents detailed results for the CO2 /n-butane at
344 K, declared to be not fully consistent because some few
areas gave deviations outside the established limits. When
one point from this data set is eliminated (the one given the
highest area deviation, in bold face in Table 5), the deviations
for the remaining 17 points are within the defined limits of
20% to +20%. Therefore, the original set of 18 data points
Table 7
Summary of the final results for all systems and all isotherms studied
System

Solvent (1)/Solute (2)

T (K)

Final
result

Ref.

CO2 /n-butane

344

18

NFC

Olds et al.
(1949)

17

TC

Ethylene/1-butene

293

15
12

NFC
TC

[34]

CO2 / 1-octanol

313
328
348
403

6
7
7
9
7
7
6

TC
TDM
TDM
NFC
TC
NFC
TC

[36]

TDM/TI
TDM/TI
TDM/TI
TDM/TI

[29]

453
4

CO2 /limonene

314
324
314
323

15
12
8
11

[28]

[30]

%y2

is declared to be NFC, but the remaining set with 17 points


as thermodynamically consistent (TC).
The situation is similar for the ethylene/1-butene mixture,
set of data that was declared in Table 5 to be NFC. For this
case three of the 15 data points were eliminated and the test
applied to the remaining 12 points, finding then to be TC.
Therefore, the original set of 15 data points is declared to
be NFC, but the remaining set with 12 points as thermodynamically consistent. Similar results are found for the system
CO2 /1-Octanol at 403 and 453K. At 403K (see Table 6), the
original set of 9 points is declared to be NFC, but the remaining set with 7 points is found to be thermodynamically
consistent. Similarly, at 453K, the original set of 7 data points
is declared to be NFC, but the remaining set with six points
is found to be thermodynamically consistent. Table 7 shows
a summary of the final results for all isotherms for the four
mixtures.

6. Conclusions
Based on the study presented in this work, the following conclusions can be drawn: (i) a new method to test
the thermodynamic consistency of incomplete high-pressure
phase equilibrium data in binary mixtures, has been presented; (ii) the proposed consistency test method fulfills all
the ten basic requirements defined in this paper for a good
thermodynamic consistency test; (iii) the proposed consistency test method allows to analyze individual data in a
data set and therefore to eliminate doubtful points; and (iv)
the method gives an answer about consistency or inconsistency of a set of experimental PTxy data for most
cases.

158

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159

List of Symbols:
a
ac

force constant in the PR equation of state


force constant in the PR equation of state at the critical point
aij
cross force constant in the EOS mixing rule
am
force constant for a mixture

%Aav average percent area deviation, %Ai /(N 1)
A12 , A21 parameters in the UNIQUAC model
AE
(x) Helmholtz free energy at infinite pressure
AE
o (x) Helmholtz free energy at low pressure
Ap
integral for point y2i to y2i+1 using Py experimental
data
A
integral for point y2i to y2i+1 using a thermodynamic
model
%Ai individual percent area deviation
b
volume constant in the PR equation of state
bm
volume constant for a mixture
bij
cross volume constant in the EOS mixing rule
d
derivative operator
EA
error in the area determined by error propagation
%EA percent error in the calculated area A
F
acentric factor parameter for the PR equation of state
liq
gas
f1 , f1 solvent fugacity in the liquid and gas phases
liq
gas
f2 , f2 solute fugacity in the liquid and gas phases
E
Go (x) Gibbs free energy at low pressure
R
G
residual Gibbs free energy
i
HR
residual enthalpy
kij
binary interaction parameters for the force constant
in an EoS
k12
interaction parameters for the force constant in an
EoS for a binary mixture
Ln
natural logarithm
M
molecular weight
N
number of data points in a data set
P
pressure
Pi
pressure for a point i in the data set
%Pi percent pressure deviation for a point i
Pc
critical pressure
R
ideal gas constant
T
temperature
Tc
critical temperature
Tr
reduce temperature (Tr = T/Tc )
V
volume
VR
residual volume
W
objective function
xi , xj liquid phase mole fraction of components i and j
x1 , x2 liquid phase mole fraction of components 1 and 2
yi , yj gas phase mole fraction of components i and j
y1 , y2 gas phase mole fraction of components 1 and 2
%y2 deviation in the gas phase solute concentration
Z
compressibility factor (Z = PV/RT)
Greek letters

temperature function for the PR equation of state


partial derivative operator

deviation
fugacity coefficient
acentric Factor
constant in the WS mixing rule
Super/subscripts

cal
exp
gas
liq

calculated
experimental
gas phase
liquid phase

Acknowledgements
The authors thank the support of the National Commission for Scientific and Technological Research (CONICYTChile), through the research grant FONDECYT 1040285, the
Direction of Research of the University of La Serena-Chile
for permanent support through several research grants and of
the Center for Technological Information (CIT, La SerenaChile), for computer and library support.

Appendix A
As described in the text, the proposed consistency test
method requires that the model be able to correlate the pressure and the solute concentration in the gas phase as well as to
give deviations between the areas AP and A within defined
maximum limits. The maximum deviation for %Ai were
defined in the range 20% to +20%. Also, the deviation in
the gas phase solute concentration %y2 and the deviation
in the pressure %P, should be both within 10% to +10%.
All these criteria are summarized in Table 1 and justified in
what follows.
The percentages defined to accept the model as accurate
for correlation of the data is based on information presented
in the literature [1,10] and summarized by one of the author
in a recent review and other publications [2527]. To confirm
these acceptable deviation ranges, we have performed calculations of error propagation on the measured experimental
data. This was done using the general equation of error propagation [31], being the temperature T and the liquid phase
concentration x the independent measured variables. The calculated individual area A (evaluated using two consecutive
points), is the dependent variable of interest. As explained before, in the proposed method bubble pressure calculations are
performed. Thus, the system pressure (P) and the gas phase
concentration (y), are calculated for given values of the experimental temperature (T) and the experimental liquid phase
concentration (x). I this way the calculated area A includes
the experimental values of T and x and calculated values of
P and y. The equation for the error EA and for the percent

J.O. Valderrama, V.H. Alvarez / Fluid Phase Equilibria 226 (2004) 149159

error %EA in the calculated area are:






A
A
T +
x
EA =
T
x
%EA =

100 EA
A

(A.1)
(A.2)

Following information given in the literature [12,32], we


have assumed that the temperature and the liquid phase concentration are less affected by experimental errors than the
solute concentration in the gas phase. We assumed maximum
errors of 0.05 K for the experimental temperature and 0.001
for the experimental liquid phase mole fraction. The partial
derivatives in Eq. (A.1) were numerically calculated for several mixtures, giving average estimated percent errors %EA
of 20% with %y2 and %P within the limits defined in
Table 1. Therefore, the range 20% is established as the maximum acceptable error for the individual areas %Ai defined
by Eq. (8).

References
[1] J.M. Prausnitz, Molecular Thermodynamics of Fluid Phase Equilibria, Prentice Hall. Englewood Cliffs, NJ, USA, 1969.
[2] H.C. Van Ness, M.M. Abbott, Classical Thermodynamics of Nonelectrolyte Solutions, McGraw-Hill, New York, 1982.
[3] J.M. Smith, M.M. Abbott, H.C. Van Ness, Introduction to Chemical
Engineering Thermodynamics, fifth, McGraw-Hill, New York, USA,
2001.
[4] H.C. Van Ness, S.M. Byer, R.E. Gibbs, Vaporliquid equilibrium:
part 1. An appraisal of data reduction methods, AIChE J. 19 (2)
(1973) 238244.
[5] P.L. Chueh, N.K. Muirbrook, J.M. Prausnitz, Part II thermodynamic
analysis, AIChE J. 11 (1965) 10971102.
[6] K.W. Won, J.M. Prausnitz, High-Pressure vaporliquid equilibria.
Calculation of partial pressures from total pressure data. Thermodynamic consistency, Ind. Eng. Chem. Fundam. 12 (4) (1973) 459
463.
[7] L.J. Christiansen, A. Fredenslund, Thermodynamic consistency using
orthogonal collocation or computation of equilibrium vapor compositions at high-pressures, AIChE J. 21 (1975) 4957.
[8] A.L. Muhlbauer, measurement and thermodynamic interpretation of
high-pressure vapor, Ph.D. Thesis, University of Natal, South Africa,
1991.
[9] P.L. Jackson, R.A. Wilsak, Thermodynamic consistency tests based
on the GibbsDuhem equation applied to isothermal, binary
vaporliquid equilibrium data: data evaluation and model testing,
Fluid Phase Equil. 103 (1995) 155197.
[10] J.D. Raal, A.I. Muhlbauer, Phase Equilibria Measurement and Computation, Taylor & Francis, UK, 1998.
[11] B.E. Poling, J.M. Prausnitz, J.P. OConnell, The Properties of Gases
and Liquids, fifth, McGraw-Hill Book Co., New York USA, 2001.
[12] A. Bertucco, M. Barolo, N. Elvassore, Thermodynamic consistency
of vaporliquid equilibrium data at high-pressure, AIChE J. 43 (2)
(1997) 547554.
[13] J.O. Valderrama, The state of the cubic equations of state, Ind. Eng.
Chem. Research. 42 (7) (2003) 16031618.
[14] P.D. Coutsikos, N.S. Kalospiros, D.P. Tassios, Capabilities and limitations of the WongSandler mixing rule, Fluid Phase Equil. 108
(1995) 5978.

159

[15] P. Kolar, K. Kojima, Determination of the binary cross virial coefficient in the WongSandler Mixing rule for cubic equation of state,
J. Chem. Eng. Japan 27 (4) (1994) 460465.
[16] H. Orbey, S.I. Sandler, Modeling vaporliquid equilibria. Cubic
Equations of state and their mixing rules, Cambridge University
Press, USA, 1998.
[17] T. Yang, G.J. Chen, W. Chan, T.M. Guo, Extension of the
WongSandler mixing rule to the three parameter PatelTeja equation of state application up to the near critical region, Chem. Eng.
J. 67 (1997) 2733.
[18] F. Brandani, S. Brandani, V. Brandani, The WongSandler mixing
rules and EOS which are thermodynamically consistent at infinite
pressure, Chem. Eng. Sci. 53 (4) (1998) 853856.
[19] D.Y. Peng, D.B. Robinson, A new two-constant equation of state,
Ind. Eng. Chem. Fundam. 15 (1) (1976) 5964.
[20] D.S. Wong, S.I. Sandler, A theoretically correct mixing rule for cubic
equations of state, AIChE J. 38 (1992) 671680.
[21] S.I. Sandler, (editor), Models for Thermodynamic Phase Equilibria
Calculations, Marcel Dekker Inc., New York USA, 1994 p. 109.
[22] M. Reilly, Computer Programs for Chemical Engineering Education,
Sterling Swift Pub. Texas USA 2 (1972) 276295.
[23] G. Hytoft, R. Gani, IVC-SEP Program Package, Tech. Rep. SEP
8623, Institut for Kemiteknik, Danmarks Tekniske Universitet: Lyngby, Denmark, 1996.
[24] J.Z. Hua, J.F. Brennecke, M.A. Stadtherr, Enhanced interval analysis
for phase stability: cubic equation of state models, Indus. Eng. Chem.
Res. 37 (1998) 15191527.
[25] J.O. Valderrama, L.E. Marambio y, A.A. Silva, Mixing rules in cubic
equations of state applied to mixtures containing R134A, J. Phase
Equil. 23 (6) (2002) 495501.
[26] J.O. Valderrama, N. Gonzalez, V. Alvarez, Gassolid equilibrium
in mixtures containing supercritical CO2 using a modified regular
solution model, Ind. Eng. Chem. Res. 42 (16) (2003) 38573864.
[27] J.O. Valderrama, V.H. Alvarez, Vaporliquid equilibrium in mixtures
containing supercritical CO2 using a new modified KwakMansoori
mixing rule, AIChE J. 50 (2) (2004) 480488.
[28] W.L. Weng, J.T. Chen, M.J. Lee, High-pressure vaporliquid equilibria for mixtures containing a supercritical fluid, Ind. Eng. Chem.
Research. 33 (1994) 19551961.
[29] T. Gamse, M. Rolf, High-pressure phase equilibria of the binary
systems carvone-carbon dioxide and limonene-carbon dioxide at 30,
40 and 50 C, Fluid Phase Equil. 171 (2000) 165174.
[30] C.M. Chang, C.C. Chen, High-pressure densities and PTxy diagrams for carbon dioxide + linalool and carbon dioxide + limonene,
Fluid Phase Equil. 163 (1999) 119126.
[31] H.S. Mickley, T.K. Sherwood, C.E. Reed, Applied Mathematics in
Chemical Engineering, McGraw Hill, New York, USA, 1957.
[32] J.A. Barker, Determination of activity coefficient from total pressure
measurements, Aust. J. Chem. 6 (1953) 207210.
[33] R.H. Olds, H.H. Reamer, B.H. Sage, W.N. Lacey, Phase equilibria in
hydrocarbon systems, Ind. Eng. Chem. Res. 41 (3) (1949) 475481.
[34] H.K. Bae, K. Nagahama, M. Hirata, Measurement and correlation
of high-pressure vaporliquid equilibria for the systems ethylene-1butene and ethylene-propylene, J. Chem. Eng. Japan. 14 (1) (1981)
16.
[35] D.S.H. Wong, H. Orbey, S.I. Sandler, Equation of state mixing rule
for nonideal mixtures using available activity coefficient model parameters and that allows extrapolation over large ranges of temperature and pressure, IECR 31 (8) (1992) 20332039.
[36] W.L. Weng, M.J. Lee, Phase equilibrium measurements for the binary mixtures of 1-octanol plus CO2 , C2 H6 and C2 H4 , Fluid Phase
Equil. 73 (1992) 117127.
[37] T.E. Daubert, R.P. Danner, H.M. Sibul, C.C. Stebbins, Physical and
Thermodynamic Properties of Pure Chemicals. Data Compilation,
Taylor & Francis, London, UK, 1996.

You might also like