Está en la página 1de 11

J. Chem.

Thermodynamics 105 (2017) 385–395

Contents lists available at ScienceDirect

J. Chem. Thermodynamics
journal homepage: www.elsevier.com/locate/jct

Review

A fresh look at the thermodynamic consistency of vapour-liquid


equilibria data
Jaime Wisniak a,⇑, Juan Ortega b, Luis Fernández b
a
Department of Chemical Engineering, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel
b
Grupo de Ingeniería Térmica, 35071-Parque Científico-Tecnológico, Universidad de Las Palmas de Gran Canaria, Canary Islands, Spain

a r t i c l e i n f o a b s t r a c t

Article history: Design of a separation unit requires real information about the phase equilibrium of the system being
Received 6 October 2016 handled. Accurate equilibrium data allows the best design from a thermodynamic viewpoint and con-
Received in revised form 22 October 2016 tributes to a better knowledge about the behaviour of fluids and their mixtures. The principles behind
Accepted 25 October 2016
the concept of thermodynamic consistency are presented and discussed. The present state of the art
Available online 26 October 2016
shows that no definite test is available for insuring the quality of the measured values. The main available
procedures for testing the consistency of vapour-liquid equilibrium (VLE) data at constant temperature or
Keywords:
pressure are reviewed and analysed and recommendations provided for their proper use, for the presen-
Thermodynamic consistency
VLE
tation of VLE results, and also some possible means for determining their quality. Suitable examples are
Gibbs-Duhem equation provided about the adequate use of the available tests and about their misuse.
Herington test Ó 2016 Elsevier Ltd.
Kojima test
Redlich-Kister test
Van Ness test
Wisniak test
McDermott-Ellis test

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
2. Thermodynamic background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
3. Consistency tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
3.1. Point-to-point test (slope test) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
3.2. Area test. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
3.3. The Redlich-Kister test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
3.3.1. Constant pressure and temperature values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
3.3.2. Isothermal values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
3.3.3. Isobaric values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
3.4. The Herington test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
3.5. Fredenslund’s test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
3.6. The L-W test of Wisniak . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
3.7. The Van Ness point-to-point test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
3.8. Test of Kojima . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
3.9. The Kang test. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
3.10. McDermott-Ellis test for ternary systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
4. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
Note. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395

⇑ Corresponding author.
E-mail address: wisniak@exchange.bgu.ac.il (J. Wisniak).

http://dx.doi.org/10.1016/j.jct.2016.10.038
0021-9614/Ó 2016 Elsevier Ltd.
386 J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395

Nomenclature

A area x mole fraction in liquid phase


A1, A2 area described in Eqs (9)–(12) y mole fraction in vapour phase
A+ positive area, Eq. (15) vE molar excess volume/m3mol1
A negative area, Eq. (15) hE molar excess enthalpy/Jmol1
ak coefficient of the Legendre polynomial, Eq. (22) g Gibbs function of mixing/Jmol1
Ai, Bi, Ci coefficients of Antoine equation gE excess Gibbs function/Jmol1
cal calculated value g Emax maximum value of g E /Jmol1
D coefficient defined by Eq. (12)
maximum value of hE/Jmol1
E
hmax
exp experimental value
fi
o
fugacity of component i in the standard state/ kPa Dh vap
vaporisation enthalpy/Jmol1
I area difference, given by Eq. (20) Dho vaporisation enthalpy of pure component i at operating
J parameter defined by Eq. (21) pressure/Jmol1
Li parameter defined by Eq. (33) Dsoi vaporisation entropy of pure component i at operating
Lk Legendre polynomial of orden k pressure/JK1mol1
NRTL non-random two liquid Dso parameter defined by Eq. (31)/JK1mol1
n number of experimental points or number of compo-
nents in Eq. (54) Greek letters
p pressure/kPa c activity coefficient
poi vapour pressure of pure component i/kPa x weighed volatility, defined by Eq. (30)
R universal gas constant/Jmol1K1 Ui fugacity coefficient of component i in the gas phase/kPa
T absolute temperature/K R total area of the ln(c1/c2) plot
T oi boiling point of pure component i/K h temperature difference/K
VLE vapour-liquid equilibrium
Wi parameter defined by Eq. (34)

1. Introduction exams have survived the test of time, and of the remaining ones,
one, which has been proved to be wrong [17] (and is also wrongly
Separation operations play a central role in the practice of applied [18]) and another, based on partially inadequate principles
chemical engineering. Their purpose is to modify the composition [15], continue to be used. The use of liquid as well as supercritical
of the solutions present in different streams, generating new fluxes gas extraction techniques is increasing and this has generated
having a higher concentration of one or more components. These intensive research on the determination of liquid-liquid (LLE) and
separations are not spontaneous and are usually based on phase solid-gas (SGE) equilibrium data, in order to find the appropriate
equilibrium phenomena. Engineering design of these processes entrainers. Unfortunately, for these operations there has been little
requires equipment that takes into account the thermodynamic progress in the development of methods for determining the ther-
and hydrodynamic behaviour of the pertinent streams. For exam- modynamic quality of this sort of data.
ple, in a distillation process, the first factor defines the height of In this contribution we analyse the different procedures being
the column, and the second, the diameter of the same. In this situ- presently employed to verify the consistency of phase equilibrium
ation, application of thermodynamic principles provides informa- data, their advantages and disadvantages, as well as drawing con-
tion of how the different components will split between one or sequences regarding the state of the art.
more liquid phases and a vapour one, once phase equilibrium
has been achieved at a given temperature and pressure. The appro- 2. Thermodynamic background
priate equilibrium data are obtained on a laboratory scale, using
equipment especially designed for this purpose. The initial reduc- The Gibbs-Duhem equation, in its general and restricted form, is
tion of the information, the experience of the operator, and the the best tool for understanding the complexity of the problem
quality of the equipment (including the pertinent analytical proce- [19]:
dures and techniques), does not insure that the resulting data sat-
isfy the rules of thermodynamics. Tassios [1] has clearly described X vE h
E
xi d ln ci ¼ dp  dT ðgeneral formÞ ð1Þ
the importance of using accurate values of the activity coefficients RT RT 2
in the design of a distillation tower. He has listed the percentage X
changes that take place in the size and cost of this separation unit xi d ln ci ¼ 0; ðform restricted to constant p and TÞ ð2Þ
when making an error of 10% in the calculation of the activity coef-
ficients, for separations of various difficulty levels (very easy, easy, where ci is the activity coefficient of component i in a multicompo-
difficult, and very difficult). For example, when the separation is nent solution of composition xi, and vE and hE are, respectively, the
very easy (relative volatility much larger than 1), the size of the excess volume and enthalpy in the mixing process. Combining
column increases by 3% and the cost of the column by 2%. For very Eq. (1) with
difficult separations (relative volatility close to 1), the pertinent X
g E ¼ RT xi ln ci ð3Þ
changes are both 100%.
Over the years near 20 techniques and tools have been devel- yields an alternative expression for the general Gibbs-Duhem equa-
oped to test the so-called thermodynamic consistency of the phase tion, written as a function of the excess properties:
equilibrium data (mostly for vapour-liquid equilibrium, VLE, see  E
Refs. [2–15], and a previous publication has given a detailed g vE h
E X
d ¼ dp  2 dT þ ln ci dxi ð4Þ
description and analysis of most of them [16]. Very few of these RT RT RT
J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395 387

Applying Eqs. (2) and (4) to a binary solution gives 3.1. Point-to-point test (slope test)
x1 d ln c1 þ x2 d ln c2 ¼ 0 ð5Þ
Eq. (5) may be written

gE vE h
E
c d ln c1 d ln c2
dð Þ¼ dp  2 dT þ ln 1 dx1 ð6Þ x1 þ x2 ¼0 ð8Þ
RT RT RT c2 dx1 dx1
As mentioned before, the Gibbs-Duhem equation is of particular Eq. (8) indicates that if in a plot of the variation of the activity
importance because it represents the basis of most of the so-called coefficients with composition, we draw the tangents to the curves
tests for thermodynamic consistency. What is the meaning of this at any given concentration, measure the corresponding slopes, and
concept? The structure of Eqs. (1) and (2) shows that the Gibbs- calculate the expression given in the left hand side of Eq. (8) then
Duhem equation couples the activity coefficients and decreases the data for that particular point will be consistent (with the rules
by one the number of their degrees of freedom. In other words, if of thermodynamics) if the result is equal to zero. This criterion may
the values of c1 (or c2) for a binary system are known indepen- be applied to check the consistency of every point, hence the name
dently, then the corresponding values of c2 (or c1) are automatically point-to-point test according to Gibbs-Duhem. Although the point-
determined (In the real case both sets carry their own errors, which to-point test is very stringent by its very nature (mathematically
have to be taken into account). Thus, if we have two independent it is enough), it is seldom applied as such because of the difficulty
procedures to determine c1 and c2, we can use Eq. (5) to determine of measuring the slope with enough accuracy. Some researches
the quality of the data. If both sets of data, c1(x1) and c2(x2), fulfil try to obviate this problem by finding an empirical analytic expres-
the Gibbs-Duhem equation we will say that the data satisfy the sion that describes the variation of the activity coefficients with the
rules established by thermodynamics, that is, they are thermody- composition. This procedure is not recommendable because it is
namically consistent. Hence, an important point to be learned here usually applied assuming that all points have the same statistical
is that every paper reporting phase equilibrium data must carry a plot weight, which is not true: the accuracy of the analytical procedures
and a table of the activity coefficients as a function of the composition, used to determine the composition of a phase decreases substan-
as a first symptom of the consistency of the data. These plots and tially as the solution becomes more and more diluted, as seen usu-
Eq. (5) also indicate that both components must deviate in the ally by the higher dispersion presented by the data points in these
same direction (positively or negatively) from Raoult’s law. For sections. This can lead to substantial errors when calculating the
example, this condition is not satisfied by the data reported by number of plates of a column because most of them will be located
Resa et al. [20] for the system {vinyl acetate (1) + butyl acetate in the high dilution sections of the diagram. Although the slope test
(2)} at 101.325 kPa, shown in Fig. 1: is not easily applicable, it can help in the visual detection of pitfalls
As seen, component 1 shows a slight negative deviation from and outliers. For example, Eq. (8) indicates that the slopes of the
Raoult’s law, while component 2 shows an increasing positive devi- two curves ci (xi) are coupled, so that if one of them presents a
ation from the same rule. maximum (or a minimum) the other one must show a minimum
It is also recommended to include the boiling point diagram of (or a maximum) at the same composition. This anomaly occurs
the mixture, avoiding the publication of bad data. This caveat is generally at the diluted end, where the analytical accuracy is
well illustrated in Fig. 2 showing the VLE data for the system lower. A clear example of this phenomenon can be found in the
{dichlorodimethylsilane (1) + heptane (2)} at 101.3 kPa reported data reported by Chandrashekara and Sechadri [22] for the system
by Sacarescu et al. [21]: {methyl ethyl ketone (1) + p-xylene (2)} at 91.3 kPa, reproduced in
Fig. 3:
3. Consistency tests This set of data presents the additional defect that both com-
pounds deviate in opposite direction from ideal behaviour.
The concept will be developed from the simple case described
by Eq. (5) to the general one derived from the integration of 3.2. Area test
Eq. (4) in the full concentration range:
Z  E Z Z Z
g E ðx¼1Þ
g pðx¼1Þ
vE Tðx¼1Þ
h
E x¼1
c1 The Gibbs-Duhem equation may integrated from x1 = 0 to x1 = 1,
d ¼ dp  dT þ ln dx ¼ 0
g ðx¼0Þ
E RT pðx¼0Þ RT Tðx¼0Þ RT
2
x¼0 c2 1 as follows
ð7Þ

Fig. 1. Plot of VLE data of the binary {vinyl acetate (1) + butyl acetate (2)} at 101.32 kPa reported by Resa et al. [20]: (a) T vs x1, y1; (b) ci vs x1. (s) T  x1; (d) T  y1; (4) c1  x1;
(N)c2  x1.
388 J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395

Fig. 2. VLE data of the binary {dichlorodimethylsilane (1) + heptane (2)} at 101.3 kPa. (a) T vs x1, y1 scanned from Sacarescu et al. [21]: upper curve, vapour composition, lower
curve, liquid composition. (b) Experimental values from Sacarescu et al. [21] (c) ci vs x1 according with the publication of Sacarescu et al. [21] (s) T  x1; (d) T  y1; (4)
c1  x1; (N) c2  x1.

Fig. 3. VLE data for the binary system {methyl ethyl ketone (1) + p-xylene (2)} at 91.3 kPa, as reported by Chandrashekara and Sechadri [22] (a) T vs x1,y1. (b) ci vs x1.
Z 0 Z ln c1
2 least, the experimental error (operator, equipment, and analysis).
x1 d ln c1 þ x2 d ln c2 ¼ A1 þ A2 ¼ 0 ð9Þ In addition, there is the error done in plotting the best curves,
ln c1
1
0
extrapolating them to infinite dilution, and in calculating the areas
The mathematical meaning of Eq. (9) is that the areas under the themselves. Since the measurement of both areas is subject to the
two curves (A1 and A2) must be equal. For this reason this test is same of errors, we cannot use one of them as the basis of compar-
called the area test. Since the value of the area is independent of ison and for this reason it is customary to declare the two areas to be
the way it is measured we can write Eq. (9) in the equivalent form equal when the following criteria is met:
Z 1 Z 0 jjA1 j  jA2 jj
ln c1 dx1 þ ln c2 dx2 ¼ A1 þ A2 ¼ 0 ð10Þ 6 0:02 ð11Þ
0 1
jA1 j þ jA2 j
usually written as
which has no thermodynamic meaning but its limits are easier to
understand and quantify. An important question to answer here jjA1 j  jA2 jj
is: what is the actual meaning of equal areas? The curves have been D ¼ 100 62 ð12Þ
jA1 j þ jA2 j
determined using the measured values that carry with them, at
J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395 389

where each area is taken as a positive number. We are then accept- ing process. How large is the latter? To get an estimate let us
ing a 2% tolerance in the value of D; that is, the difference in the par- assume that the volume of mixing is independent of pressure (a
tial areas should be no more that 2% of the total area. In the past, very reasonable assumption since we are dealing with liquid
when the available analytical techniques were not so accurate, solutions). We can then integrate the right-hand side of Eq. (16)
the accepted value of D was 10; it is reasonable to assume that in to get
the near future, with the improvement of analytical techniques, Z  
1
c1  Dv 
the accepted value of D will be less than 2. ln dx   ðpo  po2 Þ ð17Þ
An important point to notice is that the area test is a necessary 0 c2 1  RT 1
but not sufficient condition for thermodynamic consistency. It may Using typical values for the different parameters on the right-
well be that the area under the two curves will be equal but their hand side of Eq. (17) it turns out that the value of this term is order
behaviour will not be according to the Gibbs-Duhem equations. of 106, which is a very small number compared to the value of
It should be understood that the two tests described above are R1
0
ðln c1 =c2 Þdx1 , which can be of the order of 103 to 100. In other
valid only within the limitations of the restricted equation of
words, when analysing isothermal data the effect of the volume of
Gibbs-Duhem, that is only for data obtained under conditions of
mixing can be neglected and the consistency of the data tested using
constant temperature and pressure. Hence, they are not applicable
the simple Redlich-Kister criteria given by Eqs. (14) and (15).
to distillation or extraction data for binary systems (we cannot dis-
till at constant temperature and pressure). Their value resides in
3.3.3. Isobaric values
helping understand the problem of determining thermodynamic
For this case
consistency.
Z x1 ¼1 Z Tðx1 ¼1Þ E
c1 h
ln dx ¼ dT ð18Þ
3.3. The Redlich-Kister test x1 ¼0 c2 1 Tðx1 ¼0Þ RT 2

This test is based on the general expression of the Gibbs-Duhem Again, when considering the difference between the two areas
equation for a binary system. Eq. (7) may be written: A+ and A , we have to consider not only the experimental errors

Z Z Z and the error in determining the areas but also the effect of the heat
x¼1
c1 pðx¼1Þ
vE Tðx¼1Þ
h
E
effect caused by the mixing process. How large is the latter? Similar
ln dx ¼ dp  dT ð13Þ
x¼0 c2 1 pðx¼0Þ RT Tðx¼0Þ RT 2 calculations as above indicate that the numerical value of the
right-hand side of Eq. (19) is about (6  102), which is of the same
Let us now analyse some especial cases of Eq. (13).
order as the permissible value of D, Eq. (15). In other words, for iso-
baric data a correct test for thermodynamic consistency requires tak-
3.3.1. Constant pressure and temperature values
ing into account the heat of mixing effect. This requisite is easier to
In this case
state and almost impossible to fulfil: On the one hand, the heat of
Z x1 ¼1
c1 mixing at various temperatures and compositions is available for a
ln dx ¼ 0 ð14Þ
x1 ¼0 c2 1 very small number of binary systems, and on the other hand, calcu-
lation of the value of the right-hand side of Eq. (18) requires measur-
According to Eq. (14), if we plot ln(c1/c2) against x1, in a semi- ing this property at the bubble point of a set of mixtures having
logarithmic graph (the so called Redlich-Kister plot) [see Fig. 5a], different compositions, a task very hard if not impossible to carry on.
then the areas above (A+) and below (A) the x-axis must be equal. Anyhow, the numerical value of the right-hand side of Eq. (15)
This curve will cross the abscissa at the concentration where suggests that the area test provides a good qualitative picture of
c1 = c2, that is, when g E =RT achieves its maximum (or minimum) the data; if the numerical value of D is larger than 10 it is very
value. Using the same arguments as in the area test, the two areas probable that the data are inconsistent, except for the case when
are declared equal when the mixing process is accompanied by strong thermal effects.
R 
 1 c1   þ  Some applications of the Redlich Kister test are illustrated in
 0 ln c2 dx1  jA j  jA j Fig. 4:
D ¼ 100 R 1 ¼ 100 62 ð15Þ
0
j ln cc1 jdx1 jAþ j þ jA j Fig. 4a represents the VLE data reported by Kohoutová et al. [23]
2
for the system {hexane (1) + benzene (2)} at 101.32 kPa. Measure-
Both areas should be taken as positive numbers. This is a crucial ment of the positive (A+) and negative (A) areas yield D = 1.65,
decision because otherwise, as shown below, the equilibrium val- indicating that the data are consistent according to the Redlich-
ues will be consistent or not consistent depending on which com- Kister test represented by Eq. (15). Fig. 4b illustrates the VLE values
ponent is selected as component 1, an unacceptable proposition. for the binary system {ethanol (1) + water (2)} reported by Paul
The criterion given by Eq. (15) is called the Redlich-Kister test for [24]. In this case D = 5.4, indicating that the data do not satisfy
consistency and, as such, is valid only for equilibrium data taken at the Redlich-Kister test represented by Eq. (15).
constant p and T. We will now describe some of the many attempts that had been
Eq. (13) gives us the possibility of extending the consistency cri- made to develop consistency procedures that do not use Eq. (18).
teria to cases in which only one of the variables p and T is held con-
stant, and thus the opportunity of testing realistic data. 3.4. The Herington test

3.3.2. Isothermal values The first attempt to this approach was probably that of Hering-
According to Eq. (13) we get ton [12]. In 1951, Herington developed an approximation for eval-
Z Z uating an upper limit value for the integral given on the right hand
x1 ¼1
c1 pðx1 ¼1Þ
vE
ln dx ¼  dp ð16Þ side of Eq. (18). Using thermodynamic relations and the experi-
x1 ¼0 c2 1 pðx1 ¼0Þ RT
mental data available at that time Herington developed the follow-
Eq. (16) indicates that when considering the difference between ing inequalities:
the two areas we have to take into account not only the experi- Z T o1 E E
h jhmax j
mental errors and the error in determining the areas but also the dT < ð19Þ
T o2 RT 2
RT oi jhj
effect of the change in volume, positive or negative, caused by the mix-
390 J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395

Fig. 4. Redlich-Kister plot for the binary systems: (a) {hexane (1) + benzene (2)} [23], (b) {ethanol (1) + water (2)} [24].

where hence they declared their data set thermodynamically consistent.


  Had they performed the same calculations for the system {tetrahy-
hE  h   
100jIj  max   drofuran (1) + (1-methylethyl)benzene (2)} they would have
< 50 E   and I ¼ jAþ j  jA j ð20Þ
R g max  T i obtained D = 16.14 and (D  J) = 15.88 and conclude that the same
data were inconsistent. Obviously this is incorrect. The consistency
E
‘‘where jhmax j was the numerically largest value of hE that cannot depend on the numbering of the components. Herington
occurred in the concentration range ð0 6 x1 6 1Þ; jhj the difference [12] was undoubtedly aware of this possible ambiguity and, in
between the highest and the lowest boiling point (in K) of the iso- order to avoid it, he clearly established that the value of D was
baric data, Ti (K) the lowest boiling point, g Emax the maximum value always positive, as shown by the left-hand side of Eq. (21).
of gE in the same concentration range, and R the total area of the ln Marcilla et al. [26] have also reached the conclusion that the
(c1/c2) plot. According to Herington, the function on the right-hand test of Herington is wrong. In their words ‘‘this equation should
side of Eq. (19) was usually several times larger than that on the not be used in any procedure, algorithm, etc. whose purpose is to
left, particularly for systems forming an azeotrope” [12]. evaluate such data sets.”
Herington used the information about the systems known at his
E
time to determine that the function jhmax =g Emax j would seldom 3.5. Fredenslund’s test
exceed the value 3.0. Replacing this value in Eq. (20) yielded the
following relation for determining consistency: Most sets of published VLE data report the composition of the
vapour (yi) and liquid (xi) phases at either constant pressure or
100jIj jhj temperature. The phase rule of Gibbs establishes that the number
jDj ¼ < 150 ¼ jJj ð21Þ
R Ti of independent variables for a binary system is 2, in other words,
The symbol jjhas been added to emphasize the fact that Hering- the reported information is over-dimensioned. Fredenslund et al.
ton defined both D and J as positive numbers. [6] took advantage of this fact to propose a consistency test of
What happens when jDj > J ? Herington stated that if VLE based on ability of the third variable (yi) to predict the experi-
jD  Jj < 10 the data were consistent and if jD  Jj > 10; mental composition of the vapour phase. For these purposes, they
inconsistent. proposed that the set of corresponding values of g Eexp , having only
Herington’s criterion has been applied extensively in the litera- information about the liquid phase, be calculated using Eq. (3)
ture because of its simplicity. Nevertheless, Wisniak [17] examined and then correlated using a series of Legendre polynomials:
it thoroughly and showed it to be incorrect. According to Wisniak,
gE Xn
Herington determined the value of the constant J on the basis of ¼ x1 ð1  x1 Þ ak Lk ðx1 Þ ð22Þ
the very small number of systems (15) for which g E and hE data RT k¼1
were available at his time. The value 10 was selected on the basis
of the accuracy of the analytical methods available. Wisniak used a where
much more extensive database and showed not only that the J Lk ðx1 Þ ¼ ½ð2k  1Þð2x1  1ÞLk1 ðx1 Þ  ðk  1ÞLk2 ðx1 Þ=k ð23Þ
parameter was underestimated and could reach 10 times more
than the one used by Herington, but there was no criterion by with L0(x1) = 1 and L1(x1) = 2x1  1 as the first two Legendre polyno-
which to fix an upper limit for the same. Wisniak also found errors mials. The Legendre polynomials are orthogonal functions so that
in the thermodynamic assumptions made by Herington, as well as their coefficients will be independent of the number of terms taken
pointing out that today the value 10 should be 2 or less. In sum- to represent a particular continuous function (g E in this case).
mary, Wisniak showed that use of the Herington criteria as such The analytical expression given by Eq. (22) allows calculating
will in general declare the data to be consistent when in practice they the pertinent activity coefficient and from there, the predicted val-
probably are not [17]. ues of the vapour composition yi;calc :
It is unfortunate that the Herington test continues to be used
o
today although it should not. Not only this, in too many cases it xi ci f i
yi;cal ¼ ð24Þ
is applied in the wrong manner, as shown by the following exam- pUi
ple: Gill et al. [25] reported VLE data for the binary system
o
{tetrahydrofuran (1) + (1-methylethyl)benzene (2)} at 97.3 kPa where f i is the fugacity of component i at the standard state, and Ui
and examined their consistency using the Herington test. Their its fugacity coefficient in the vapour phase, at the equilibrium pres-
results indicated that D = 16.14 and (D  J) = 15.88 < 10, and sure and temperature of the solution.
J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395 391

According to Fredenslund et al., a set of p  T  x1  y1 data is component i at the boiling temperature of the solution (T), and T oi is
declared consistent if the average absolute deviation of yi,exp and the boiling point of the same at the equilibrium pressure p of the
yi,cal is less than 0.01 units, that is system. The result is
X  
n
jyi; exp  yi;cal j p Dh
o
1 1 Dso
6 0:01 ð25Þ ln ¼ i o  ¼ i ðT oi  TÞ ð28Þ
i¼1
n poi R Ti T RT
o
Normally, three to five terms of the Legendre series are enough where Dhi and Dsoi are the heat and entropy of vaporisation of the
to determine the quality of the data. pure component at the pressure of the solution. Replacing in the
The Fredenslund test should be considered a necessary but not expression for g E and rearranging yields
sufficient condition for consistency because it is based on a global X yi X
statistics and as such it does not consider the possibility of local g E ¼ RT xi ln þ xi Dsoi ðT oi  TÞ ð29Þ
inconsistencies. For this reason, it must always be accompanied xi
by an analysis of the distribution of the residuals ðyi;exp  yi;cal Þ [or Calling
another variable. for example, ðpi;exp  pi;cal Þ around the value zero]. X yi
This distribution must be random, as observed visually, or better, x¼ xi ln ð30Þ
xi
as determined by a statistical test like that of Durbin-Watson [27].
As an illustration, consider the results reported by Ding et al. for the weighted volatility and
the system {isopropyl acetate (1) + DMSO (2)} at 101.325 kPa [28]. X
The authors checked the quality of the data using only Herington’s
Dso ¼ xi Dsoi ð31Þ
test and concluded that they were consistent. Application of Fre- the weighted entropy of vaporisation, we have
denslund method yields the values 0.005 for the average absolute X
deviation of the vapour composition, Eq. (25), for a Legendre series g E ¼ RT x þ xi Dsoi T oi  T Dso ð32Þ
containing 3, 4, and 5 terms. Hence Ding’s data are also consistent
according to Fredenslund method. A somewhat different picture X   E 
Dsoi o g RT
appears when analysing the residual graphs for the vapour compo- Li ¼ xi Ti  T ¼  x ¼ Wi ð33Þ
Ds Dso Dso
sition, Fig. 5(a), and for the pressure, Fig. 5(b):
These graphs show clearly lack of randomness at the dilute end Eq. (33) is valid for any number of components.
of the system for the residuals of the composition of the vapour According to Eq. (33), the data for the particular point are consis-
phase, and lack of randomness for the residuals of the calculated tent when Li = Wi. This is the L–W point-to-point test. For binary
pressure in the full concentration range. Hence, the quality of these systems the test can be converted into an area test by integrating
data is doubtful. Eq. (33) over the complete concentration range to give
Z 1 Z 1
3.6. The L-W test of Wisniak Li dx1 ¼ W i dx1 ð34Þ
0 0

Wisniak [10] developed an additional test for thermodynamic According to Wisniak, the binary data are declared consistent if
consistency based on Eq. (3). Replacing the activity coefficient by parameter D, defined as
ci ¼ yi p=xi poi we write
jL  Wj
X  y p
 D ¼ 100 ð35Þ
g E ¼ RT xi ln i þ ln o ð26Þ LþW
xi pi
is less than 3, when the heat of vaporisation is a reliable experimen-
If the boiling point range is not particularly large, then we can tal value, and less than 5 if the heat of vaporisation has been esti-
use the Clausius-Clapeyron equation in the form mated by an empirical method. In the same manner, a particular
vap experimental point passes the point-to-point test if the condition
d ln poi Dh
¼ ð27Þ
dT RT 2 Li
0:92 < < 1:08 ð36Þ
Wi
to obtain the value of lnðp=poi Þ by integration of Eq. (27) between
the limits (poi ; T) and (p, T oi ) where poi is the vapour pressure of pure is fulfilled.

Fig. 5. Residual plots for the vapour composition, (a) and pressure, (b), for the VLE data for the system {isopropyl acetate (1) + DMSO (2)} at 101.32 kPa, reported by Ding et al.
[28].
392 J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395

Fig. 6. (Li/Wi) coefficients of the Wisniak test for the systems: (a) {water(1) + methanol(2)} by Maripuri et al. [29]. (b) {water (1) + 1-propanol (2)} by Udovenko et al. [30].

  !
The L-W test has the advantage of being simultaneously a dðg E =RTÞ dðg Eexp =RTÞ c1 c1;exp
point-to-point and an area test, and also that it can be applied  ¼ ln  ln
dx1 dx1 c2 c2;exp
without difficulty to multicomponent systems. It has the disadvan-     
tage of using ‘‘imported information” (the heats of vaporisation), d ln c1;exp d ln c2;exp
 x1 þ x2 ð41Þ
which can seriously affect the results of the test. This problem dx1 dx1
becomes more serious when the method is applied to a multicom-
Now, if the errors of the data set are random, the left-hand side
ponent system.
of Eq. (39) becomes zero and Eq. (41) reduces to
Before closing the discussion of the L-W test we add that !
     
according to Trouton’s rule the value of Dsoi is 88 JK1mol1 for c c1;exp d ln c1;exp d ln c2;exp
pure compounds. ln 1  ln ¼ x1 þ x2 ð42Þ
c2 c2;exp dx1 dx1
Two examples on the application of the Wisniak test are shown
in Fig. 6 for the systems {water (1) + alkanol (2)} mixtures: Examination of Eq. (42) shows that if the data are consistent
For the VLE data for the system {water (1) + methanol (2)} sys- then its right-hand side becomes identical with the Gibbs-Duhem
tem, reported by Maripuri et al. [29], the calculated values of the equation. This will occur when the residuals identified by the
coefficient Li =W i satisfy the conditions established by Eqs. (35) left-hand side of the same equation are distributed randomly along
and (36) see Fig. 6(a), with the global parameter D = 2.1, that is, the full concentration range. This is the basic tenet of Van Ness’
the data set is validated by the L-W test. The VLE data for the sys- method.
tem elected, {water (1) + 1-propanol (2)}, reported by Udovenko The implementation of Van Ness’ method to a given set of data
et al. [30], the majority of data show a value of Li =W i < 0.92 out implies calculating the value of lnðc1 =c2 Þ for every point of the set,
of interval defined by Eq. (36), with D = 5.3. The VLE results for this then fitting them using a standard model such as Margules, Wohl,
second system fail the L-W test. NRTL, etc. (Van Ness used the Margules model), and finally,
inspecting the distribution of the residuals given by
3.7. The Van Ness point-to-point test ½lnðc1 =c2 Þ  lnðc1;exp =c2;exp Þ. The set of data is declared consistent
if the latter is random. This is the weak side of this method. It is
In 1995 Van Ness [11,12] introduced a practical consistency test very possible that distribution of the residuals will be random for
based strictly on the Gibbs-Duhem equation. Consider Eq. (3) one or more models and not random for others, an unacceptable
applied to a binary system: ambiguity for testing thermodynamic consistency. As stated, the
Van Ness point-to-point method is more a procedure for deciding
gE if a given model fits the data, than a procedure for testing the ther-
¼ x1 ln c1 þ x2 ln c2 ð37Þ
RT modynamic quality of the same. Both events are not necessarily
Differentiation with respect to x1 yields thermodynamically connected.
The Van Ness method is illustrated in Fig. 7 using two sets of
dðg E =RTÞ d ln c1 d ln c2 c
¼ x1 þ x2 þ ln 1 ð38Þ VLE data reported by Ortega et al. [31,32] for the binary system
dx1 dx1 dx1 c2 {ethyl methanoate (1) + octane system (2)}, measured at
From a thermodynamic viewpoint the sum of the first two 101.32 kPa, which satisfy the Redlich-Kister and Fredenslund test:
terms in the right hand side of Eq. (38) is equal to zero, but this The residuals of lnðc1 =c2 Þ obtained for the first set [31], Fig. 7a,
is not the case when the equation is applied to the experimental show a clear non-random behaviour, indicating that these data do
information, carrying their accumulated error: not satisfy Van Ness’ test. The other set [32], published recently to
correct the previous one, shows a much better distribution of the
dðg E =RTÞ c residuals and confirmation of the Van Ness criteria. The slight devi-
¼ ln 1 ðperfect dataÞ ð39Þ ations are probably a consequence of using an inadequate model to
dx1 c2
represent the data.
!    
dðg Eexp =RTÞ c1;exp d ln c1;exp d ln c2;exp
¼ ln þ x1 þ x2 ð40Þ 3.8. Test of Kojima
dx1 c2;exp dx1 dx1

Subtracting Eq. (40) from Eq. (39) gives This test is based on a combination of three examinations: (a)
point-to-point, based on the relations
J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395 393

Fig. 7. Results of the Van Ness point-to-point test for two different sets of isobaric VLE data for the system: (a) {ethyl methanoate (1) + octane (2)} from Ortega et al. [31] (b)
{ethyl methanoate (1) + heptane (2)} from Ortega et al. [32].

Xn
dðg E =RTÞ c x1 x2
d ¼ 100 jdj j=m d ¼  ln 1 ¼ e ð43Þ DT ¼ T  x1 T o1  x2 T o2 ¼ 0
ð53Þ
dx 1 c2 a0 þ b ðx1  x2 Þ þ c0 ðx1  x2 Þ2 þ   
j¼1

(b) an area test based on where T oi is the boiling point of component i at the operating pres-
Z 1 Z 1 sure and DT is a convenient function of the temperature and
c1
A ¼ 100  jA j A ¼ ln dx þ edx1 ð44Þ composition.
0 c2 1 0 The most serious disadvantage of the Kojima method, besides
(c) an infinite dilution criterion based on being more complicated, is its being based on the highly diluted
region where the data are, in general, more inaccurate due to the
I1 ¼ 100  jI1 j I2 ¼ 100  jI2 j ð45Þ present available analytical procedures. In addition, and as men-
where tioned before, the variation of the heat of mixture with tempera-
"    #, ture and composition, is known for a relatively small number of
g E =RT c c systems. In Kojima’s method, this functionality is assumed to be
I1 ¼  ln 1 ln ð 1 Þ ð46Þ
x1 x2 x1 ¼0 c2 x1 ¼0 c2 x1 ¼0 according to Eq. (51) up to the bubble temperature of the mixture,
a hypothesis hard to justify. In addition, the final result carries the
"    #,   accumulated error of the four regressions, Eqs. (49)–(53). The pres-
g E =RT c2 c
I2 ¼  ln ln 2 ð47Þ ence of the factor x1x2 in the function gE/(RTx1x2) makes it highly
x1 x2 x2 ¼0 c1 x2 ¼0 c1 x2 ¼0 sensitive to errors made in the determination of composition of
the liquid phase. It is highly recommended that the first step be
The activity coefficients at infinite dilution are obtained by
the plotting of a graph that describes simultaneously the variation
extrapolation of the function (gE/RT)/x1x2 to x1 = 0 and x1 = 1:
of both the individual activity coefficients and gE/(RTx1x2) with
 E   
g =RT c the composition. Both sets of curves should predict the same val-
¼ ln 1 ¼ ln c1
1
x1 x2 x1 ¼0 c2 x1 ¼0 ues of the activity coefficients at infinite dilution; otherwise the
 E    data set is most probably inconsistent.
g =RT c
¼ ln 2 ¼ ln c1
2 ð48Þ As an example consider the application of the Kojima dilution
x1 x2 x1 ¼1 c1 x1 ¼1 test to the data published by Iliuta et al. [33] for the binary system
According to Kojima et al. the data set is considered to be con- {1-propanol (1) + water (2)}, Fig. 8a:
sistent when d < 5, A < 3, and I1 and I2 are less than 30. The plots of gE/(RTx1x2) and ln(c1/c2) yield very similar values at
The numerical solution of the different steps requires a previous compositions of infinity dilution and the values of Ii are 14.5 for
correlation of experimental data (not always available). In particu- i = 1 and 5.9 for i = 2, indicating that this set of data are consistent
lar, the functions gE/RT, ln(c1/c2), and hE (excess enthalpy), are according to the Kojima test. Fig. 8b represents the same informa-
assumed to vary according to a Redlich-Kister expansion: tion for the system {methanol (1) + ethyl ethanoate (2)}, as
reported by Jones et al. [34]. In this case the values of gE/(RTx1x2)
g E =RT ¼ x1 x2 ½B þ Cðx1  x2 Þ þ Dðx1  x2 Þ2 þ    ð49Þ and ln(c2/c1) at infinite dilution are significantly different, suggest-
ing that the data are inconsistent. This conclusion is confirmed by
lnðc1 =c2 Þ ¼ a þ bðx1  x2 Þ þ cð6x1 x2  1Þ þ    ð50Þ the fact that I1 ¼ 14:8 and I2 ¼ 78:

E
h ¼ x1 x2 ½k0 þ k1 ðx1  x2 Þ þ k2 ðx1  x2 Þ2
3.9. The Kang test
þ    ðat constant TÞ ð51Þ
The constants from Eqs. (49) and (50) are determined from the This method is a combination of three previous available tests
corresponding experimental values. The coefficients of Eq. (51) are (Herington, point-to-point of Van Ness, and a modification of Koji-
assumed to vary with the temperature as follows: ma’s one), plus the requirement that the value of the activity coef-
ficients of the components at high dilution satisfy known
mi
ki ¼ ni þ ð52Þ thermodynamic relations [15]. The results of these four conditions
T are then assigned a statistical weight (named quality factor, Ftest,
Finally the value of the derivative ðdT=dx1 Þ is determined from varying between 0.025 and 0.25), to take into account that it is dif-
the relation ficult for any particular set of VLE data to pass all the tests
394 J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395

Fig. 8. Representation of the quantities involved in the Kojima’s infinity dilution test for the systems: (a) {1-propanol (1)+water (2)}, measured by Iliuta et al. [33], (b)
{methanol (1)+ethyl ethanoate (2)} measured by Jones et al. [34]. ( ) F = lnðc1 =c2 Þ; (d) F = lnðc2 =c1 Þ; (N) F = g E =x1 x2 RT.

Table 1
Results obtained in application of McDermott-Ellis test [36], modified by Wisniak-Tamir [37] to the ternary system {methyl ethyl ketone (1) + diethyl ketone (2) + methyl isobutyl
ketone (3)} [38], indicating the values obtained for each of terms of Eq. (55).

T/K x1 D Dmax jDj  jDmax j D1max D2max D3max D4max

367.33 0.353 – – – – – – –
367.70 0.415 -0.086 0.091 0.005 0.070 0.001 0.005 0.015
367.93 0.302 0.066 0.101 0.035 0.079 0.001 0.005 0.016
P
D1max ¼ i¼n
i¼1 ðxib þ xia Þðxia þ yia þ xib þ yib ÞDx.
1 1 1 1
P
D2max ¼ 2 ni¼1 j ln cib  ln cia jDx.
P
D3max ¼ ni¼1 ðxib þ xia Þ Dpph. i
P
D4max ¼ ni¼1 ðxib þ xia ÞBi 1
2 þ
1
2 DT.
ðT a þC i Þ ðT b þC i Þ

simultaneously. An equation is provided for determining the over- where Dp, DT, and Dx are the errors in the measurement of the
all VLE data quality factor, QVLE. pressure, temperature, and mole fraction, respectively, and Ai, Bi,
This test has been severely criticized because it uses the Hering- and Ci the pertinent coefficient in the Antoine equation
ton test as one of its elements [35], and because it uses a model ln poi ¼ Ai  Bi =ðT i þ C i Þ. Inspection of Eq. (55) shows that the
(NRTL) to apply the point-to point test of Van Ness [26]. Marcilla expression of Wisniak and Tamir is more general because it takes
et al. went even further, claiming: ‘‘the deviation of the experimen- into account the error of all the variables, not only of the concentra-
tal VLE results with regard to correlation by means of a given tion. Many researchers use a simplified form of the test, leaving only
model should not be used to assess the quality of these data until the first term.
gE models capable of fitting all existing phase equilibrium beha-
Xn  
viour are developed. The consistency of tests based on this idea 1 1 1 1
should not be applied. Therefore, the applicability of two of the five Dmax ¼ ðxib þ xia Þ þ þ þ Dx ð56Þ
i¼1
xia yia xib yib
tests that are combined in the algorithm proposed by Kang et al. is
questioned” [26]. However, a more rigorous use of the method involves the appli-
cation of Eq. (55) with four terms, leaving as a researcher criteria
3.10. McDermott-Ellis test for ternary systems the final evaluation of the quality of data, but always considering
the condition indicated above, D < Dmax. Table 1 illustrates the
According to this method two experimental points, a and b, are application of the method to three points of the ternary system
thermodynamically consistent if D < Dmax where the local devia- reported by Wisniak and Tamir [38]. It contains the numerical
tion D is given by value for each of the terms of Eq. (55); the first term of Eq. (56)
contributes 78% to Dmax; the fourth one is also significant, with a
X
n
D¼ ðxib þ xia Þðln cib  ln cia Þ ð54Þ 16%. Accordingly, it is better to use Eq. (55) and not the simplified
i¼1 form given by Eq. (56). The data values for T=367.70 K would not
where n is the number of components. According to Mc-Dermott- be consistent if only D1max was taken into account. The modified test
Ellis [36], a fixed value is recommended for Dmax. However, Wisniak of McDermott-Ellis indicates that three equilibrium points consid-
and Tamir [37] recommended using a local value of the maximum ered are consistent.
deviation, defined by four terms:
X
n  
1 1 1 1 4. Conclusions
Dmax ¼ ðxib þ xia Þ þ þ þ Dx
i¼1
xia yia xib yib
X
n X n The present state of knowledge indicates that no definite test is
Dp
þ j ln cib  ln cia jDx þ ðxib þ xia Þ available for testing the thermodynamic consistency of VLE data.
i¼1 i¼1
p All the available exams contain empirical concepts that limit their
" #
X n
1 1 capability, and none of them is able to provide a sure answer to the
þ ðxib þ xia ÞBi þ DT ð55Þ question. In the present state of the art of consistency tests, there
i¼1 ðT a þ C i Þ2 ðT b þ C i Þ2
are only two possible answers:
J. Wisniak et al. / J. Chem. Thermodynamics 105 (2017) 385–395 395

(a) Reject total or partial data [15] J.W. Kang, V. Diky, R.D. Chirico, J.W. Magee, C.D. Muzny, I. Abdulagatov, A.F.
Kazakov, M. Frenkel, Quality assessment algorithm for vapour-liquid
(b) Accept all the data
equilibrium data, J. Chem. Eng. Data 55 (2010) 3631–3640.
[16] J. Wisniak, A. Apelblat, H. Segura, An assessment of thermodynamic
Thermodynamic validation of VLE data requires the following consistency tests for vapour-liquid equilibrium data, Phys. Chem. Liq. 36
actions: (1997) 1–58.
[17] J. Wisniak, The Herington test for thermodynamic consistency, Ind. Eng. Chem.
Res. 33 (1994) 177–180.
1. All papers reporting VLE data should always contain a table of [18] J. Wisniak, H. Segura, Comment on ‘‘Experimental vapour-liquid equilibrium
values experimentally obtained directly (p, T, x, y) as well as a data for binary mixtures of cyclic ethers with 1-ethylethylbenzene)”. [B.K. Gill,
V.K. Rattan, S. Kapoor. J. Chem. Eng. Data 53, 2041–2043], J. Chem. Eng. Data 54
semi-logarithm graph of their variation with composition, (2009) (2008) 1165–1167.
extrapolated to x1 = 0, and x1 = 1 [similar to the one shown in [19] J.M. Smith, H.C. Van Ness, M.M. Abbott, Introduction to Chemical Engineering
Fig. 1], together with a curve of the function gE/(RTx1x2) in the Thermodynamics, seventh ed., McGraw-Hill, NY, 2005.
[20] J. Resa, C. González, B. Moradillo, J. Lanz, Vapour + liquid) equilibria for
same range. These should be used properly to determine the (methanol + butyl acetate), (vinyl acetate + butyl acetate), (methanol
coherence of all the information available. + isobutyl acetate), and (vinyl acetate + isobutyl acetate, J. Chem.
2. The quality of the equilibrium data should be checked using Thermodyn. 30 (1998) 1207–1219.
[21] L. Sacarescu, M. Marcu, N. Luchian, G. Sacarescu, Vapour-liquid equilibria for
simultaneously several of the available tests. One of them must dichlorodimethylsilane + heptane, J. Chem. Eng. Data 40 (1995) 71–73.
be the Fredenslund test, accompanied with an examination of [22] M.N. Chandrashekara, D.N. Seshadri, Vapour-liquid equilibria: systems methyl
the both pressure residuals and the corresponding vapour- ethyl ketone-p-xylene and chlorobenzene-p-xylene, J. Chem. Eng. Data 24
(1979) 6–8.
composition residuals. In addition, it is recommended to per-
[23] J. Kohoutová, J. Suska, J.P. Nova, J. Pick, Liquid-vapour equilibrium. XLV. System
form the area test [using a semi-log graph similar to the one methanol 2-propanol water, Collect. Czech. Chem. Commun. 35 (1970) 3210–
shown in Figure (4)] to get a qualitative estimation of the data 3222.
consistency. Additional possibilities include the tests proposed [24] R.N. Paul, Study of liquid-vapour equilibrium in improved equilibrium still, J.
Chem. Eng. Data 21 (1976) 165–169.
by Kojima [9], Wisniak [10], and Van Ness [11,12]. All publica- [25] B.K. Gill, V.K. Rattan, S. Kapoor, Experimental isobaric vapour-liquid
tions should include the necessary information to allow verify- equilibrium data for binary mixtures of cyclic ethers with (1-methylethyl)
ing that the pertinent exam has been properly performed. benzene, J. Chem. Eng. Data 53 (2008) 2041–2043.
[26] A. Marcilla, M.M. Olaya, M.D. Serrano, A.A. Garrido, Pitfalls in the evaluation of
3. As a specific issue of the area tests, the Herington one should the thermodynamic consistency of experimental VLE data sets, Ind. Eng. Chem.
not be used in any form for testing purposes. Res. 52 (2013) 13198–13208.
[27] J. Durbin, J.G.S. Watson, Testing for serial correlation in least squares
regresssion, Biometrika 38 (1951) 159–177.
Note [28] H. Ding, Y. Gao, J. Li, J. Qi, H. Zhou, S. Liu, X. Han, Vapour-liquid equilibria for
ternary mixtures of isopropyl alcohol, isopropyl acetate, and DMSO at
The authors declare no competing financial interest. 101.3 kPa, J. Chem. Eng. Data 61 (2016) 3013–3019.
[29] V.O. Maripuri, G.A. Ratcliff, Measurement of isothermal vapour-liquid
equilibrium for acetone-n-heptane mixtures using modified Gillespie still, J.
References Chem. Eng. Data 17 (1972) 366–369.
[30] V.V. Udovenko, T.F. Mazanko, Izv. Vyssh. Uchebn. Zaved., Khim. Khim.
[1] D. Tassios, Applied Chemical Engineering Thermodynamics, Springer, Berlin, Tekhnol., 15, 1972, 1654.
1993, p. 23. [31] J. Ortega, F. Espiau, R. Dieppa, Measurement and correlation of isobaric
[2] O. Redlich, T. Kister, Thermodynamics of nonelectrolyte solutions, Ind. Eng. vapour–liquid equilibrium data and excess properties of ethyl methanoate
Chem. 40 (1948) 341–345. with alkanes (hexane to decane), Fluid Phase Equilib. 215 (2004) 175–186.
[3] O. Redlich, A.T. Kister, Algebraic representation of thermodynamic properties [32] J. Ortega, L. Fernández, G. Sabater, Solutions of alkyl methanoates and alkanes:
and the classification of solutions, Ind. Eng. Chem. 40 (1948) 345–348. simultaneous modeling of phase equilibria and mixing properties. Estimation
[4] E.F.G. Herington, Tests for the consistency of experimental isobaric vapour- of behavior by UNIFAC with recalculation of parameters, Fluid Phase Equilib.
liquid equilibrium data, J. Inst. Pet. 37 (1951) 457–470. 25 (2015) 38–49.
[5] E. Liebermann, F. Vojtech, Thermodynamic consistency test method, Ind. Eng. [33] M.C. Iliuta, F.C. Thyrion, O.M. Landauer, Effect of calcium chloride on the
Chem. Fundam. 11 (1972) 280–281. isobaric vapour-liquid equilibrium of 1-propanol + water, J. Chem. Eng. Data
[6] A. Fredenslund, J. Gmehling, P. Rasmussen, Vapour-Liquid Equilibria Using 41 (1996) 402–408.
UNIFAC a Group-Contribution Method, first ed., Elsevier, 1977. [34] C.A. Jones, E.M. Schoenborn, A.P. Colburn, Equilibrium still for miscible liquids.
[7] V. Dohnal, D. Fenclová, A new procedure for consistency testing of binary Data on ethylene dichloride-toluene and ethanol-water, Ind. Eng. Chem. 35
vapour-liquid equilibrium data, Fluid Phase Equilib. 21 (1985) 211–235. (1943) 666–672.
[8] K. Kollar-Hunek, S. Kemeny, K. Heberger, P. Angyal, E. Thury, Thermodynamic [35] J. Wisniak., Comment on ‘‘Quality Assessment Algorithm for Vapour-Liquid
consistency test for binary VLE data, Fluid Phase Equilib. 27 (1986) 405–425. Equilibrium Data” (J.W. Kang, V. Diky, R.D. Chirico, J.W. Magee, C.D. Muzny, I.
[9] K. Kojima, H.M. Moon, K. Ochi, Thermodynamic consistency test of vapour- Abdulagatov, A.F. Kazakov, M.J. Frenkel, J. Chem. Eng. Data. 55, 3631–3640), J.
liquid equilibrium data, Fluid Phase Equilib. 56 (1990) 269–284. Chem. Eng. Data 55 (2010) (2010) 5394.
[10] J. Wisniak, A new test for the thermodynamic consistency of vapour-liquid [36] C. McDermott, S.R.M. Ellis, A multicomponent consistency test, Chem. Eng. Sci.
equilibrium, Ind. Eng. Chem. Res. 32 (1993) 1531–1533. 20 (1965) 293–296.
[11] H.C. Van Ness, Thermodynamics in the treatment of vapour/liquid equilibrium [37] J. Wisniak, A. Tamir, Vapour-liquid equilibriums in the ternary systems water-
(VLE) data, Pure Appl. Chem. 67 (1995) 859–872. formic acid-acetic acid and water-acetic acid-propionic acid, J. Chem. Eng.
[12] H.C. Van Ness, Thermodynamics in the treatment of (vapour + liquid) Data 22 (1977) 253–260.
equilibria, J. Chem. Thermodyn. 27 (1995) 113–114. [38] J. Wisniak, A. Tamir, Ternary vapour-liquid equilibrium in the system methyl
[13] P.T. Eubank, B.G. Lamonte, Consistency tests for binary VLE data, J. Chem. Eng. ethyl ketone-diethyl ketone-methyl isobutyl ketone, J. Chem. Eng. Data 21
Data 45 (2000) 1040–1048. (1976) 470–473.
[14] S. Kato, Thermodynamic consistency test for the binary constant-temperature
VLE data using numerically optimized binary parameters, Fluid Phase Equilib.
297 (2010) 192–199. JCT 16-780

También podría gustarte