Vous êtes sur la page 1sur 18

Fluid Phase Equilibria, 13 (1983) 91-108 91

Ekvier Science Publishers B.V., Amsterdam -Printed in The Netherlands



Paul M. Mathias and Thomas W. Copeman

Air Products & Chemicals, Inc.
Box 538, Allentown, PA 18105

Density-dependent local composition (DDLC) mixing rules (Whiting and
Prausnitz, 1981, 1982 and Mollerup, 1981) offer great promise to extend
equations of state to highly nonideal mixtures. In this work we investigate
various aspects of the DOLC concept and offer two useful forms. All
development is done with the Peng-Robinson (1976) equation of state, but the
results have general applicability.

Practical models to describe phase equilibrium can be divided into two
broad categories: equations of state and activity-coefficient models.
Equations of state have been successfully applied to mixtures containing
,nonpolar and slightly polar components at all conditions of engineering
interest, including the critical and retrograde regions. On the other hand,
the mathematical flexibility of activity coefficient models has been
considered necessary to model systems which exhibit high liquid-phase
nonideality. The activity-coefficient approach works well at low reduced
temperatures where the liquid phase is relatively incompressible and up to
moderate pressures. Serious problems arise when these bounds are violated.
,Further, the use of different models for the various phases precludes the
correct description of mixture critical points. Additional problems arise for
supercritical components (Abrams, et al., 1975). The equation-of-state
approach does not inherently suffer from these limitations and thus it would
be very valuable to develop equations of state with the mathematical
flexibility to describe complex mixture behavior.
Recent research in equations of state has focussed on the mixing rules as a
promising way to achieve mathematical flexibility through a semi-theoretical
development. Vidal (1978) has shown that the standard mixing rules used with
the Soave (1972) and Peng-Robinson (1976) equations of state are very similar
to regular solution theory and thus inadequate for "chemical" systems.
Further, Huron and Vidal (1979) have suggested improvements to the standard
mixing rules by applying the ideas of NRTL model (Renon and Prausnltz, 1968)

0378-3812/83/$03.00 0 1983 Elsevier Science Publishers B.V.


to the excess Gibbs energy of the Soave equation. The results are promising
but the Huron and Vidal model does not meet the important low-density 1 imit.
This has been pointed out by Mollerup (1981) and Whiting and Prausnitz (1981,
1982); these investigators have separately proposed a physically appealing
model to account for mixture nonideality: density-dependent local composition
(DDLC) theory.
In this work we have investigated density-dependent local-composition
theory as applied to the popular Peng-Robinson (PR) equation of state. The
standard Peng-Robinson equation provides a very good description of
asymmetric, nonpolar systems but the DDLC-PR model proposed by Mollerup (1981)
.can be inaccurate by up to 3-4 orders of magnitude for highly asymmetric
systems. Noting this, we have adopted a modlficatlon of an idea proposed by
Dimitrelis and Prausnitz (1982). The new model enables improved correlations
of a wide variety of complex mixtures. Our proposed DOLC model (along with
the other recently proposed forms) could prove to be too expensive in computer
time for systems with a large number of components. Thus we propose a
truncated version of the theory which, however, retains several attractive


According to the Peng-Robinson (1976) equation of state the pressure of a
pure fluid is given as a function of the temperature and molar volume as

P=RT_ (1)
v-b v(v+b) + b(v-b)

a and b are component dependent parameters. b is independent of temperature

and a is temperature dependent such that the vapor pressure of the pure fluid
i s accurately carrel ated. Thus,

b = 0.07780 R Tc/Pc

a = ace(T)

a = 0.45724 R=T,=/P,

B = [ 1 + CI(1 - fiR) + c2(1 - FR)Z + c3(l - j-YR)l ] 2

Equation (5) Is slightly different from that origlnally proposed by Peng

and Robinson (1976). The original PR model is obtained when c2=c3=0. The
additional parameters are necessary to correlate the vapor pressure of highly
polar substances 11 ke water and methanol. In this work we wish to focus on
the development of mixing rules. Inaccurate pure component vapor pressure

representation artificially distorts the excess Helmholtz free energy and

obscures the analysis of mixture effects (Van Ness, et al., 1973).

For mixtures, Peng and Robinson assumed that eqn. (1) remains unchanged
but the parameters a and b are calculated as the mole fraction averages shown


a = I E xixjaji (6)

b = Z xibi (7)

a.. is given by the usual combining rules which include the pair dependent
biiary parameter kji

aji = (1 - kji) (aiaj)'fl

Equations (1) and (6-8) deflne the "standard" Peng-Robinson model. We

refer to this form as the van der Waals l-Fluid (VOW-l) or "conformal" model.
It should be noted that the mixing rule defined by eqns. (6) and (7) do not
imply that the mixture is random (all pair radial distribution functions are
identical) but rather that the various pair radial distribution functions can
be superimposed on a reduced basis (Mansoori and Leland, 1972).


Mollerup (1981) derived his version of the density-dependent local
composition Peng-Robinson equation of state by assuming that local
compositions arise from differences fn the attractive parameter "a" per unit
surface area (q).

Lx a /q E
1 j j ji ji ji
zxq (9)
v-b v(v+b) + b(v-b) i i i 1 x. E..
J J'

where, Eji = exp ~~~~~ (10)

(I 1
= exp - aji/qji ~ en (11)
RT 2-&b

and, aji/qj, = (1 - kji) (aiaj/qiqj)lB (12)

Actually, Mollerup did not employ the universal "degree-of-randomness"
parameter a (Whiting and Prausnitz, 1981), however, it is included here for
The attractive Helmholtz energy corresponding to eqn. (9) is given by

Aa = - - Z xiqian BXjEji (13)
II. i j

Useful insight into the nature of the DDLC theory is obtained by an

expansion of the logarithm and exponential terms in eqn. (13) as follows:

A = - Zxiqi Exjaji/qji
i j

r 1 Q
- 2xiqi Exjafji/qZji - (Exjaj /qji)= - Fv2 - .._._ (14)
i j j 2RT
\ r'

In eqn. (14) we have used the simp ifying notation,

1 (1 + $ 2)b
Fv e - en (IS)
2$b Iv + (I -4)b

It is interesting to note that the first term of eqn. (14) has exactly the
same functional dependence as the standard or VDW-1 Peng-Robinson equation of
state. (A minor difference arises in the i-j combining rule when qi # q., but
this could be absorbed by an appropriately chosen binary parameter, kji. ; To
fix ideas, we define the first term of eqn. (14) as the "equivalent" VDW-1
contribution and lump all the remaining terms into the "asymmetric" or
"nonconformal" contribution.
It is illustrative to compare the fugacity coefficients of eqn. (13) with
those of its equivalent VDW-1 form. Due to the complexity of the equations,
we only show the comparison for a binary with component 1 at infinite

= -9 2 1) - (16)

a12 a22 1
where, X =

The right

[ 912
- -

qz d
side of eqn. (16)


is always negative and thus the fugacity


coefficient is reduced from its VDW-1 value. (The opposite effect is obtained
for negative values of 0~. However, only positive values of (I are
physically meaningful.)
In general, it can be shown that the local composition effect reduces the
Helmholtz energy of a system at fixed T, V, g. This is a qualitatively
correct result since the DDLC configuration should be more favorable than its
equivalent VDW-1 form.


The VOW-1 PR-EOS provides a good description of most nonpolar systems,
including those which are highly asymmetric. For example, the methane-decane
binary is accurately correlated over a temperature range of 311-583 K, even
with the default interaction parameter, k2I=O; the average absolute errors in
the K-values of methane and decane are 3.5% and 8.8%, respectively. It is
extremely important that the DDLC mixing rules retain this desirable
capability. Therefore we have conducted an investigation of the model
proposed by Mollerup (eqn. 9) by using the methane-decane binary as a test
Figures 1, 2, and 3 present the results of the investigation. We have
somewhat arbitrarily chosen to show the liquid-phase fugacity coefficient of
methane as a function of composition at 310 K and 270 atm. In general, the
DDLC mixing rules cause a decrease in the fugacity coefficient. Figure 1
shows that this decreasecan be enormous, giving fugacity coefficients too low
by as much as 3-4 orders of magnitude when the q of decane is unity. Figure 1
also presents the fugacity coefficients with the size parameter from Bondi
(1968), q=6, and two higher values. The local composition effect is still too
large. A non-randomness parameter (o) of 0.2 along with the same decane
size parameters is shown in figure 2. Figure 3 illustrates the additional
effect of setting the binary parameter to 0.5. These comparisons indicate
that the Mollerup model requires large and unreasonable parameters to
correlate this relatively simple system.


The results of the last section show that the "asymmetric" contribution iS

greatly over-predicted by the Mollerup version of the DDLC-PR EOS for

asymmetric, nonpolar systems. A similar conclusion has already been reached
P 1.0
kz,- 0.0

q-+-y-- o.b

0.0 0.2
x(t&aM) OS6

Fig 1. Effect of LC model (eqn. 9). Liquid-phase of methane-decane binary at

310 K and 270 atm.


1 VOW-l.9'1 (OATAwlnuw 10x)


s 0.1:
- -xiq=ro___---
_I All lx%

u. 40

I 1 I
0.0 0.2 0.4 0.8 0.8

Fig 2. Effect of LC model (eqn. 9). Liquid-phase of methane-decane binary at

310 K and 270 atm.

i VOW-~. q=l (DATAWITNIN 10%)

-1 -__-
_--- LC.qrlO _____c
u. q=s __-- ---
I l.c. q*o All u
,,i _

0.0 0:2 0:s 0.k

Fig 3. Effect of LC model (eqn. 9). Liquid-phase of methane-decane binary at

310 K and 270 atm.

Fig 4. Computertime for calculation of fugacity coeffkients for n-component


by Dimitrelis and Prausnitz (1982) who further point out that models analogous
to eqn. (9) make the rather stringent assumption that local compositions
Inevitably occur In the j-l pair interaction whenever ai/qi # aj/qj. They
have proposed that local composition effects are not caused when ji
interactions are different from ii but rather when ji are different from some

"ideal" combination of ii and jj interactions, referred to as ji".

Dimitrelis and Prausnitz define jiO as an arithmetic mean. Since the standard
Peng-Robinson generally produces satisfactory results for nonpolar systems
with kji = 0. the geometric mean should be a very good first approximation for
We assume that the total molar internal energy of the ji pair in a
multi-component fluid is divideb into two parts: ideal and excess. The
"ideal" part is entirely conformal as in the standard Peng-Robinson model.
However, the "excess" contribution causes non-conformal local compositions
described by the DDLC theory. \

-a -a0 -aEX
U U + u..
ji = ji 31

-a0 a FV
U (1 - kji) - (19)
ji =_._._
al/T T I VP!.!

-aEX a FV
= ~ (-dji) - (20)
al/T T I V,N

Equation (19) corresponds to the standard Peng-Robinson model. The usual

binary parameter kji is retained. However, in the data analysis we have
generally set kji = 0. The parameter dji determines that part of the internal
energy that produces local composition effects. Note that 4 is not symmetric
(i.e., d jf # dfj). The total mixture molar internal energy is obtained by
summing the ideal and excess ji contributions using the VDW-1 and DDLC mixing
rules, respectively.

-a0 -a0
U = E Z xjxj uji (21)

-aEX -aEX
U = B Xrqi x ji j< (22)

i j

where = x. - (23)
'ji J
X 'k Eki

and Eji = exp (-dji)Fv

1 (24)

The Helmholtz energy may be obtained by integrating eqns. (18-24) with

respect to l/T. Thus,

1 (25)
_a RT
A = - L E xixj (aiaj)'f2 (1 - kji)Fv - - X xiqi rn I: xj Eji
ij cri [j

It should be noted that the integration to obtain Helmholtz energy requires

an expression for the athermal limit. Here we assume that the repulsive term
of the standard Peng-Robinson equation of state yields a reasonable athermal
contribution to the total Helmholtz free energy. Whiting and Prausnitr (1982)
have shown that the Peng-Robinson repulsive term resembles the Flory-Huggins
athermal term. Also we note that highly asymmetric nonpolar systems (e.g.,
methane-decane) are well described by the standard Peng-Robinson model.
Expressions for all configurational properties can easily be obtained from
eqn. (25). However no derived properties are shown in order to save space.
The above derivation is only valid from a phenomenological viewpoint. It
does however state the assumptions that have been made. The new model meets a
very important limit in addition to the ones at high temperature and low
density noted by Whiting and Prausnitz (1981) and Mollerup (1981): that is,
the nonconformal effect is not necessarily large (or even nonzero) for an
asymmetric system. This feature allows the model to retain the good results
of the standard Peng-Robinson model and facilitates accurate correlation of
the phase equilibria of highly nonideal polar systems.


The local-composition Peng-Robinson model defined by eqn. (25) produces
greatly improved correlations over the standard Peng-Robinson model, as shown
in the next section. However, the computational load might prove to be
unacceptably high for practical applications with a large number of
components. This arises from the bulk of computer time being consumed in mole
fraction summation calculations. (Also see Boston and Mathias, 1980.) In the
model defined by eqn. (25), these summations are density dependent and thus
must be redone at each iteration for the mixture molar volume.

Figure 4 shows a comparison of the average computer time required by the

standard Peng-Robinson model and DDLC version (eqn. 25) for a single
calculation of the fugacity coefficients in an n-component system. In our
opinion, a reasonable increase in computer time (say, a factor of two) is
justified by a superior model, but an order of magnitude increase is
impractical (with current computer capabilities for process simulation) when
the number of components in the mixture is greater than about 15. Figure 4
also includes a comparison for the approximate or truncated model we
propose. We conclude that the truncated model may be used for large systems
without incurring a prohibitive computational burden.

According to the truncated DDLC model, the attractive Helmholtz energy of a

mixture is given by

A = - a Fv - - Fvz (26)

where a is given by the standard Peng-Robinson form (eqns. 6 and 8) and a ,
the NonConformal term, is given by

aNc = H xi+ aci E xj tji (27)

1 j

The parameters tii are the lumped contribution of all the nonconformal
effects. Note thatt is not symmetric (tji # tij) and thus the model has 3
parameters per binary pair; the third parameter is kji.

The truncated model defined by eqns. (6, 8, 26, and 27) may be derived by
expanding eqn. (25) analogously to eqn. (14) and retaining only the first

nonconformal term. Denoting this term as A we obtain,

-aNC 1 2 P z
A =-- Zxiaci xj dji - (z xj dji) F, (28)
2RT i j

Notice that 01 and the qs and as have been absorbed in the ds, This does
not affect the composition dependence since only the first nonconformal term
Is used. Further, the temperature dependence of a (eqns. 3 and 5) has been
removed. This has facilitated good representation of data over a wide
temperature range with temperature-independent parameters (e.g.,
water-hydrocarbon systems).

The temperature dependence of eqn. (28) may also be justified on a

theoretical basis. Mansoori and Leland (1972) have shown that the first
nonconformal contribution to the compressibility factor is of order 1/T2.
This is consistent with eqn. (28).
The composition dependence chosen for eqn. (27) is somewhat arbitrary. It
is suggested by writing eqn. (28) for a binary,

-aNC 1 rzz 221

A = - - x1 x2 x1 acl d21 + x2 ac2 d12 (29)
2RT 1 1

Equations (28) and (29) always give a negative contribution to the total
Helmholtz energy. This is physically correct but it could restrict the

mathematical flexibility of the model. Therefore, we have chosen the

relatively empirical form of eqn. (27). However, we should note that if the
regressed value of t is large and negative, it suggests that the physical
meaning is lost.
In this work only binary data were analyzed and thus it was not possible to
evaluate whether the more correct composition dependence of eqn. (29) is
necessary for multicomponent mixtures. Further, for the systems we have
analyzed the t's have been positive and thus eqns. (27) and (28) are
An extremely appealing aspect of the model is that it reduces exactly to
the standard Peng-Robinson model when all the t's are equal to zero. Thus it
is especially applicable to multicomponent systems whfch are largely nonpolar
but have a few important polar components (e.g., water in petroleum or coal
conversion processes).


Vapor-liquid equilibrium was correlated for various systems using van der
Waals one-fluid and local-composition mixing rules, eqn. (25). One parameter
(k12) was fit with the van der Waals mfxing rules and two parameters (d12 and
d21) were fit with the local composition rules. The non-randomness parameter
(a) was set to unity and size parameters (q) were derived from Bondi (1968).
Table 1 presents maximum and average K-value deviations for these two methods.
In each system, local composition mixing rules enabled an improved fit to the
data. The methanol-benzene system was correlated to a much higher accuracy,
while improvement was less, but St.111 significant, for the other systems.
Figures 5 and 6 present plots of K-value deviations from data for the
methanol-benzene system. We should note the symmetric form of eqn. 25
= d21) did not show significant improvement over the VDW-1 form.

Correlation of Vapor-Liquid Equilibrium Data

Ref. Temp. Pressure Mix Rules Parameters %Kl max* XK2 max %Kl avg %K2 a"g

(OC) (Atml k d d
12 21 12

Benzene(2) Naga ta 60.8-76.7 1.0 vow-1 0.10 -57.0 62.1 17.1
1969 LC 0.61 -0.17 -7.0 -14.3 3.0

Carbon Dioxide(P) Ohgak i 25-40 5.7-80.0 VDW-1 0.06 14.2 27.0 14.8
1976 LC 0.23 -0.37 -4.1 -22.5 2":: 8.0

Water(2) Griswold 100 1.1-3.6 VDW-1 -0.23 -49.6 22.6
1952 LC -0.35 -0.02 :::: -29.7 10.6

Methanol(2) Churkin 50 3.1-6.1 VDW-1 0.04 33.0 16.9 23.9
1976 LC 0.0 0.59 -1.6 2.5 8.5

* Average absolute deviation in K-value of component 1.





x (Methanol)

Fig 5. Comparison of new LC model (eqn. 25) with experimental data (Nagata,

Fig 6. Comparison of new LC model (eqn. 25) and VI@1 form with experimental
data (Nagata, 1969).

Correlation of Liquid-Liquid Equilibrium Data

1 2 3
System REP Temp Mix Rules Parameter %l avg w avg
c k k d d d
12a 12b 21a 12a 12b

Water(e) Hill 5-80 VDW-1 -0.18 49.1 26.7
1926 LC -0.1 -0.05 -0.05 0.14 1.0 0.7

LC(tr)4 0.07 0.12 0.45 1.2 2.6

Benzene (l)-
Water(e) Tsonopoulos 10-200 ZW-' -0.08 132.1 2691.1

1982 LC(tr) 8::: -0.15 0.04 0.74

1.08 0.06 ::: :::

Water(Z) @clpoulos 10-200 VDW-1 -0.24 "X >10000
LC 0.15 -0.24 2.39 0.15 12.0
LCItr) 0.48 0.0 1.35 16:7 12.9


Water(P) 1982
Brady 37-277 ZW-' -:z -0.16 1.16 0.21 23.2
1036.0 '"Z
i_C(tr) 0:31 0.01 1.36 3.5 618

d =d + d
12 12a T 12b

Average absolute deviation in calculated mole fraction of component 1 in liquid phase 2.

Average absolute deviation in calculated mole fraction of component 2 in liquid phase 1.

For LC(tr), read "de as et".

Fig 7. Correlation of benzene-water mutual solubility data (Tsonopoulos and

Wilson, 1982) with truncated LC model.






Fig 8. z.Zrap;lation of I-methylnapthalene-water mutual solubility data (Brady

., 1982) with truncated LC model.

Liquid-liquid equilibrium was correlated for a few selected systems using

van der Waals one-fluid and local composition (including the truncated) mixing
rules. Table 2 presents average absolute deviations in calculated mole
fractions for these methods. One parameter was fit to the van der Waals
mixing rule, four parameters (to include temperature dependence) were fit to
the local composition and three to the truncated local composition mixing
rules. Although the overall improvement is substantial with four local
composition parameters, it should be noted that temperature dependence in the
and d12 parameters must be correlated for an accurate fit. In this
regard, the local composition equation of state is analogous to the current
local composition activity coefficient models. It should be noted that even
with four binary parameters (also typically used with the UNIQUAC activity
coefficient model) the local composition method provides a significant
improvement beyond methods like Peng and Robinson (1980) for correlation of
liquid-liquid equilibria. These methods use separate interactlon parameters
(temperature dependent) for each phase. No ambiguity or discontinuity arises
with the local composition method at a critical end point where (any) two
phases merge.
The results with the truncated local composition model are particularly
encouraging since the (three) interaction parameters are not temperature
dependent. Figures 7 and 8 present a comparison of model calculations against
experimental data. Note that tZII for water in the hydrocarbons are nearly
zero. A two-parameter fit would yield about the same accuracy.

In this work we have investigated the density-dependent local composition
theory as applied to the Peng-Robinson equation of state. We propose two
useful models to account for nonconformal or asymmetric effects. The approach
is also applicable to any other equation of state derived from the generalized
van der Waals partition function.

Finally, the results of this work demonstrate that an equation of state

model which employs DDLC mixing rules and thus meets the required limits at
high temperature and low density is promising as a substitute for activity
coefficient models in practical applications.

The authors wish to thank Professors J. M. Prausnitz and F. P. Stein and
D. Dimitrelis for helpful discussions.
The work reported in this paper was sponsored in part by the International
Coal Refining Company and the United States Department of Energy under
Contract DE-AC05-780R03054.

List of Symbols

Helmholtz Energy
Peng-Robinson parameter, eqn. 1
Second Virial Coefficient
Peng-Robinson parameter, eqn. I
Binary Interaction Parameter
Binary Interaction Parameter
Surface Area Size Parameter
Number of Moles
Gas Constant
Binary Interaction Parameter
Molar Volume
Mole Fraction
Non Randomness Parameter
Temperature Dependent Pure Component Parameter, eqn. 5
Fugacity Coefficient


a Attractive Contribution
Infinite Dilution


i Component i
i Comoonent 5
i ComPonent k
C Critical

Abrams, 0. S. and Prausnitz, J. M., 1975, AIChE J. 21:116-128.

Abrams, 0. S., Seneci, F., Chueh, P. L., and Prausnitz, J. M., 1975, Ind. Eng.
Chem., Fundam., 14:52-54.

Bondi, A., 1968. Physical Properties of Molecular Crystals, Liquids, and

Glasses, John Waley and Sons, New York.

Boston, J. F. and Mathias, P. M., 1980, Phase Equiilibria and Fluid Properties
in the Chemical Industry - 2nd Internation Conference, Berlin. March

Brady, C. J., Cunningham, J. R., and Wilson, G. M., 1982, Gas Processors
Association, RR-62.

Churkin, V. N., Gonshov, V. A., and Pavlov, S. Y., 1978. Zh Fiz Khim. 52:488.

Dimitrelis, D. and Prausnitz, 3. M., 1982. "Thermodynamic Properties of

Strongly Nonideal Fluid Mixtures from an Extended Quasi-Chemical Theory,"
paper presented at Fall Annual Meeting. American Institute of Chemical
Engineering, Los Angeles, CA, Nov 14-19, 1982.

Gmehling, J. and Onken, U., 1981. Vapor-Liquid Equilibria Data Collection,

Dechema, Frankfurt.
Griswold, J. and Wong, S. Y.. 1952, Chem. Eng. Prog. Symp. Ser. 48:18.

Hill, A. E., and Malisoff, W. M. 1926, J. Am. Chem. Sot. 48:918.

Huron, M. J. and Vidal, J., 1979, Fluid Phase Equilibria, 3:255-271.

Mansoori, G. A. and Leland, Jr., T. W., 1972, J. Chem. Sot., Faraday

Trans. II, 68:320-344.

Mollerup, J., 1981, Fluid Phase Equilibria, 7:121-138.

Nagata, I., 1969, J. Chem. Eng. Data, 14:418.

Ohgaki , K. and Takashi, K., 1976, J Chem. Eng. Data 21:53.

Peng, D. Y., and Robinson, D. 8.. 1976. Ind. Eng. Fundam., 15:59-64.

Peng, 0. Y., and Robinson, D. B., 1980, in ACS Symposium Series 133, Ed:
Newman, S. A., Washington, DC, 393-414.

Reamer, H. H., Olds, R. H.. Sage, 8. H. and Lacey, W. N.. 1942, Ind. Eng.
Chem., 36:1526.

Soave. G., 1972, Chem. Eng. Sci., 27:1197.

Tsonopoulos, C. and Wilson, G. M., 1982. High Temperature Mutual

Solubilities of Hydrocarbons and Water. I. Benzene, Cyclohexane, and
n-Hexane. accepted for publication in AIChE Journal.

Van Ness, H. C., Byer, and S. M., Gibbs, R. E., 1973. AIChE Journal

Vidal, J., 1978, Chem. Eng. Sci., 33:787-791.

Whiting, W. B. and Prausnitz, J. M., 1981. Equations of State for Strongly

Nonideal Fluid Mixtures: Application of The Local-Composition Concept, paper
presented at Spring National Meeting. American Institute Chemical
Englneering, Houston, Texas, 5-g April 1981.

Whiting, W. B. and Prausnitz, J. M., 1982, Fluid Phase Equilibria, 9:119-147.