Sie sind auf Seite 1von 5

J. Phys. Chem.

B 1998, 102, 2427-2431 2427

Statistical Associating Fluid Theory Equation of State with Lennard-Jones Reference


Applied to Pure and Binary n-Alkane Systems

Chang-keng Chen, Michal Banaszak,† and Maciej Radosz*,‡


Exxon Research and Engineering Company, Annandale, New Jersey 08801, Chemistry Department, UMIST,
Manchester, M60 1QD, U.K., and Department of Chemical Engineering and Macromolecular Studies Group,
Louisiana State UniVersity, Baton Rouge, Louisiana 70803-7303
ReceiVed: August 29, 1997

A prototype equation of state, developed on the basis of thermodynamic perturbation theory of the first order
(TPT1), using the Lennard-Jones potential for the reference fluid, is found to represent pure-fluid and binary
phase equilibria in the systems of small chain molecules, such as normal alkanes.

Introduction energy ares;


The modeling of phase behavior and thermodynamic proper- ares ) aseg + achain + aassoc (1)
ties of model chain molecules, such as n-alkanes, is important
in the chemical and polymer industry. Theory of hard-core The segment Helmholtz energy aseg, per mole of molecules, is
homonuclear linear chains has been worked out analytically by calculated from
many workers. For example, Wertheim1 developed the ther-
modynamic perturbation theory of the first order (TPT1) which aseg ) maseg
0 (2)
gives an analytical equation of states for hard-core homonuclear
chain molecules, as shown by Chapman et al.2 and Chiew3 using where aseg
0 , per mole of molecules, is the residual Helmholtz
the Percus-Yevick approximation, was also able to obtain energy of nonbonded spherical segments, which is our reference
analytically both thermodynamic and structural properties of fluid, and m is the segment number (number of segments per
hard-core homonuclear chain molecules. Similarly, the theories molecule). Since we are concerned with non-associating fluids
of Schweizer and Curro4 and of Dickman and Hall5 have been in this work, we will set the association term aassoc in eq 1 equal
applied to describe the thermodynamics of hard-core chain to zero.
molecules. An equivalent definition of SAFT for nonassociating fluids
There have also been attempts to go beyond the hard-core in terms of the compressibility factor Z is given below:
homonuclear linear chains (e.g., extensions of the generalized
Flory dimer (GFD) theory to square-well potential by Yethiraj 0 + (1 - m)Z0
Z ) mZseg chain
(3)
and Hall6 and extensions of TPT1 to sticky potential and square-
well potential by Banaszak et al.7). Especially pertinent to this where
work are attempts to extend TPT1 to the Lennard-Jones (LJ) ∂
chains (e.g, by Banaszak et al.,8 Ghonasgi and Chapman,9 and Zchain ) 1 + F ln(gseg) (4)
0
∂F
Kraska and Gubbins10).
The goal of this work is to develop a prototype, LJ-based, The F is the segment density and gseg is the radial distribution
equation of state (EOS) within the TPT1 formalism. Such an function of the reference fluid (i.e., the fluid of nonbonded
approach, of deriving analytical equations of state within TPT1 segments). A more detailed description of SAFT is given by
for spherically symmetric potentials, is usually referred to as Huang and Radosz.11,12 Examples of methods for estimating
the statistical associating fluid theory, or SAFT for short. We the reference equation of state (i.e., aseg seg
0 and Z0 and g
seg are
want to test an LJ-based SAFT equation of state on real pure given in Table 1).
n-alkanes and their mixtures. This is a preliminary step in In this work, aseg seg seg are estimated using a LJ
0 , Z0 , and g
developing a theoretically-based equation of state applicable to potential for the reference fluid, which is given below,
large chainlike molecules, such as polymers, including associat-

[(σr ) - (σr ) ]
ing polymers. 12 6
φ(r) ) 4 (5)
LJ-Based SAFT
where  is the LJ well depth and r is the distance. This potential
In general, for any spherically-symmetric potential, SAFT is used to obtain the Helmoltz energy of the LJ segments
can be defined, for example, in terms of the residual Helmholtz aseg
0 ) aljs and their radial distribution function gseg ) gljs,
where the subscript ljs refers to pure LJ spheres.
* To whom correspondence should be addressed: phone, 504-388-1750; The aljs term is estimated using an empirical equation of state
fax, 504-388-1476; email, radosz@che.lsu.edu. fitted to molecular simulation data for pure LJ fluids. This is
† Chemistry Department.
‡ Department of Chemical Engineering and Macromolecular Studies a modified Benedict-Webb-Rubin equation developed by
Group. Johnson et al.13
S1089-5647(97)03181-7 CCC: $15.00 © 1998 American Chemical Society
Published on Web 03/11/1998
2428 J. Phys. Chem. B, Vol. 102, No. 13, 1998 Chen et al.

TABLE 1: Methods for Estimating the Reference Equation Similar to the other versions of SAFT, this EOS also requires
of State, aseg seg
0 and Z0 , and the Reference Radial Distribution three pure component parameters. They are the segment number
Function gseg m, the segment diameter, σ, and the segment energy, /k. These
aseg seg
0 and Z0 gseg ref parameters are obtained for each pure compound by regressing
its vapor pressure and liquid density data.
Lennard-Jones hard sphere 2
argon hard sphere 11
sticky sphere sticky sphere 7 Mixing Rules
square-well square-well 7
Lennard-Jones Lennard-Jones 8 In order to extend the pure-fluid EOS described above to
9
10
mixtures, it is customary to use mixing rules for m, σ, and /k.
One example of mixing rules is that rooted in the van der Waals
TABLE 2: Coefficients for the Radial Distribution Function one-fluid theory (vdW1). In the vdW1 spirit, one postulates
Correlation, Equation 10, for Lennard-Jones Spheres molecular parameters of a hypothetical pure fluid having the
ajk ajk same residual properties as the mixture of interest. Such mixture
a11 0.361 795 622 a33 3.177 855 237 parameters are obtained by averaging the pure component
a12 1.150 780 175 a34 11.254 331 34 parameters. For example, the vdW1 mixing rules for SAFT,
a13 -1.747 104 817 a41 16.332 419 86 proposed by Huang and Radosz11 are as follows:
a14 0.743 036 489 a42 11.053 932 30

∑i ∑j XiXjmij
a21 1.494 834 320 a43 -20.021 430 11
a22 -0.477 275 498 a44 -6.452 129 961 m) (11)
a23 11.740 968 99 a51 -7.919 015 75
a24 -7.691 645 525 a52 -5.691 714 633
a31 -9.315 979 208 a53 13.167 522 72 where
a32 -11.615 971 73 a54 2.228 311 134

The gljs term is developed in this work by fitting the gljs data mij ) 1/2(mi - mj) (12)
obtained in this work from Monte Carlo (MC) simulation and
taken from the literature. It turns out that a good fit to the MC and X’s are the mole fractions of species i and j.
data is a necessary, but not sufficient, condition to obtain good
predictions for real systems. In other words, there are several XiXjmimjσ3ij
models that we found to be capable of fitting such data, but
only some of them are capable of working well for the real ∑i ∑j
systems. We will describe, therefore, only two examples of σ3x ) (13)
the gljs models that not only fit the MC data, but also can
represent the real systems, as it will be discussed in the
[ ∑i Ximi] 2

subsequent sections.
In the first example, gljs is a function of the reduced
parameters r*, F*, and T*. These reduced parameters are
∑i ∑j XiXjmimjσ3ij
defined as follows: x ) (14)

r* ) r/rx ) reduced intermolecular distance (6)


σ3x [ ∑i ximi]2

F* ) Fσ3 ) reduced density (7) where


T* ) kT/ ) reduced temperature (8)
σij ) 1/2(σi - σj) (15)
The complete gljs function, expressed as
and

()
5 5 4
1 j-1
g )1+
ljs
∑ ∑∑aijk (r* - 1)
i)1 j)1 k)1
i-1

T*
(F*) k
(9)
ij ) (1-kij)(ij)1/2 (16)

is applicable in the ranges of 0.9 < r* < 1.1, 0.9 < F* < 0.9, where kij is an optional adjustable parameter.
and 1.0 < T* < 6.0, where aijk’s are the regression coefficients. Another example of mixing rules, also proposed by Huang
In the second example, we make a simplifying assumption and Radosz11 and used in this work, is referred to as the volume-
that r* is equal to one, which leads to the following expression: fraction (VF) mixing rules. The averaging equations for m are

()
5 4 the same as eqs 11 and 12 but the averaging equations for σ
1 j-1
gljs ) 1 + ∑ ∑
j)1 k)1
ajk(F*)k
T*
(10) and /k are different:

Equation 10 is our working equation used in all the calculations


σ3x ) ∑i ∑j XiXjσ3ij (17)
of this work. Table 2 gives the values of the regressed
coefficients ajk for eq 10.
A prototype, LJ-based, SAFT equation of state, therefore, is
x ) ∑i ∑j fifjij (18)

obtained by substituting the aljs and gljs into the corresponding


aseg
0 and g
seg described in eqs 1-4. where σij and ij are the same as defined in eqs 15 and 16; fi is
SAFT Equation of State J. Phys. Chem. B, Vol. 102, No. 13, 1998 2429

TABLE 3: LJ-Based SAFT Parameters


%AAD
n-alkane MM m σ, in A /k in K psat Vl
methane 16.043 0.9934 3.7356 149.95 1.93 1.16
ethane 30.07 1.5626 4.2199 193.25 1.45 1.98
propane 44.097 1.8521 4.653 218.12 1.39 2.13
butane 58.124 2.2010 5.0046 233.28 1.17 2.78
pentane 72.151 2.545 5.2924 243.95 1.56 1.67
hexane 86.178 2.8076 5.576 255.09 1.09 2.04
heptane 100.205 3.1849 5.8109 259.54 0.38 1.86
octane 114.232 3.5262 6.0263 264.26 0.28 1.28
nonane 128.258 3.8878 6.231 267.32 0.33 0.78
decane 142.276 4.2401 6.4201 270.08 0.71 0.88
dodecane 170.34 5.044 6.7697 272.01 1.36 0.95
C14 198.394 5.615 7.1088 278.04 1.00 1.03
C16 225.432 6.1888 7.4076 282.95 0.73 0.88
C20 282.556 7.635 7.9213 285.59 1.15 0.64
C28 394.77 10.4768 9.1015 285.35 1.93 Figure 2. %AAD in n-alkane liquid density for SAFT (filled circles,
C36 506.77 12.7829 10.4008 287.0 1.29 solid line for the average %AAD) and for LJ-based SAFT (open circles,
C44 619.21 14.7996 11.6299 288.0 0.94 dotted line for the average %AAD).
CO2 44.01 2.4526 3.5444 159.80 0.52 1.38
ethene 28.054 1.4737 4.0627 183.96 1.01 1.15
propene 42.081 1.8539 4.5002 214.87 0.71 1.06
butene 56.108 2.007 4.8871 240.79 2.36 1.44
1-hexene 84.156 2.5168 5.5409 265.07 5.08 2.63

Figure 3. LJ-based segment number m as a function of the n-alkane


molar mass.

%AAD is an order of magnitude lower. On average, %AAD


Figure 1. %AAD in n-alkane vapor pressure for SAFT (filled circles, for the vapor pressure is lower by about 40% and %AAD for
solid lines for the average %AAD) and for LJ-based SAFT (open the liquid density (same as in the liquid volume) is lower by
circles, dotted line for the average %AAD). about 60%, than those for the original SAFT. Table 3 also gives
representative results for linear olefins and carbon dioxide.
the volume fraction of species i defined as
The pure-n-alkane parameters regressed in this work are
plotted versus molar mass in the range from methane (16 g mol)
Xi(miσ3i )γ to n-C44 (619 g mol). Figure 3 shows the segment number m
fi ) (19) as a function of molar mass, which is linear. Figure 4 shows
∑j Xi(mjσ3j )γ the segment energy as a function of molar mass. While not
linear, this curve is smooth and continuous. It exhibits a rapid
increase for the segment energy with increasing MM in the range
The exponent γ is a constant. For example, in this work, γ, is of 16-100 g mol, but it reaches an asymptotic value of about
set equal to 1/3. 290 for larger alkanes. Both trends, for m and /k, are similar
This LJ-based SAFT equation of state is applied to binary to those observed for SAFT.
mixtures using a block-algebra simultaneous (BAS) flash This is not the case for the segment diameter σ which is
algorithm described by Chen et al.14 shown as a function of molar mass in Figure 5; σ increases
nonlinearly up to about 100 g mol, then it reaches a linear
Application to Pure Alkanes
relationship at higher MM. For SAFT, that parameter remains
The pure component parameters, m, σ, and /k, are given in constant for large molecules, independent of MM, which is an
Table III, along with the molar mass (MM), percent absolute ideal result for extrapolation (e.g., to polymers). The result
average deviation (%AAD) in vapor pressure (psat) and %AAD presented in Figure 5 means that the size of an effective LJ
in liquid molar volume (Vl). These average absolute deviations segment scales linearly with the molar mass for large alkanes.
are summarized in Figures 1 and 2, and compared with those For extrapolation to even larger alkanes, this is more reliable
regressed by Huang and Radosz11 on the same sets of experi- than a nonlinear relationship, but not as desirable as σ equal to
mental data. Figures 1 and 2 suggest a significant improvement a constant. We attempted, therefore, to make σ constant, instead
in the goodness of fit. For example, for the large alkanes, the of treating it as an adjustable parameter in regression. This
2430 J. Phys. Chem. B, Vol. 102, No. 13, 1998 Chen et al.

Figure 4. LJ-based segment energy /k as a function of the n-alkane Figure 6. Pressure-mole fraction phase diagram for ethane-decane
molar mass. mixture at 377.59 K; circles are the experimental data by Reamer and
Sage,17 the solid curve is predicted by SAFT, the dotted curve is
predicted by the LJ-based SAFT with the VF mixing rules, and the
dashed-dotted curve is predicted by the LJ-based SAFT with the vdW1
mixing rules.

Figure 5. LJ-based segment diameter, σ, as a function of the ν-alkane


molar mass.

approach, however, caused a noticeable decrease in the goodness Figure 7. Pressure-mole fraction phase diagram for ethane-propane
of fit which, as it turns out, is sensitive to σ. Apparently, this mixture at 344.26 K; circles are the experimental data by Maschke
theory accounts for the behavior of real chain molecules if the and Thodos,17 the solid curve is predicted by SAFT, and the dotted
effective LJ chains, which mimic these real molecules, are curve is predicted by the LJ-based SAFT with the VF mixing rules.
allowed to have increasing segments with increasing molar mass.
Parenthetically, there is another consequence of having to make it possible to estimate the three parameters for all n-alkane-
deal with σ that increases with increasing MM. This conse- like molecules on the basis of their molar mass only. Similar
quence concerns estimating σ for asymmetric mixtures, that is, correlations can also be obtained for aromatic and other
mixtures containing molecules differing in size (e.g., solvent hydrocarbon and non-hydrocarbon molecules.
and polymer). In such cases, the reduced distance defined in
eq 6, can easily fall outside the range of applicability of eq 9, Application to Alkane Mixtures
either on the low or the high side. Since there are no simulation We first illustrate the difference between the two types of
data available to extend this range of r*, we do not use eq 9 to mixing rules, vdW1 and VF, with an example of a pressure-
estimate gljs; instead we use the simplified eq 10, which is mole fraction (PX) phase diagram for an ethane-decane mixture
independent of r*, for all pure-component and mixture results shown in Figure 6. The points represent the experimental points,
presented in this work. the solid curves represent the SAFT results, the dotted curves
The parameters shown in Figures 3-5 are correlated as simple represent the LJ-based SAFT with the VF mixing rules, and
functions of molar mass. For example, the segment number the dashed-dotted curves represent the LJ-based SAFT with the
for the alkanes is modeled as a linear function of MM as follows: vdW1 mixing rules. The results presented in Figure 6 suggest
that the vdW1 mixing rules tend to underestimate the decane
m ) 0.892324 + 0.0232526(MM) (11) mole fraction, especially in the liquid phase where the error
can be as high as 20%. By contrast, the VF mixing rules tend
Similarly, the segment size, σ, for alkanes is correlated with
to overestimate the decane mole fraction but result in a better
respect to molar mass as follows:
estimate than that from the vdW1 mixing rules. In fact, the
σ ) 4.9338 + 0.010757(MM) (12) quality of the VF prediction is comparable to, but not as good
as, that obtained from SAFT, which is known to be reliable for
As for the segment energy, /k, it is set equal to 285 K for the predicting binary n-alkane mixtures. In general, the VF mixing
alkane molecules larger than 300 g mol. These correlations rules are consistently more accurate than the vdW1 mixing rules,
SAFT Equation of State J. Phys. Chem. B, Vol. 102, No. 13, 1998 2431

results quite well, at least at lower pressures where we have


experimental data, and the quality of its fit is still comparable
to that of SAFT.
As we continue extending the difference in size by going to
more asymmetric mixtures, for example, ethane-n-C44; as is
shown in Figure 9, the LJ-based SAFT predictions tend to
diverge near the critical region, which is unrealistic. This can
be fixed empirically but it is not pursued in this work. Better
models and better mixing rules for gljs, as it is explored, for
example, by Banaszak et al.,15 seem to be the most promising
areas for improvement.
For the record, all the results presented in this work have
been obtained with the binary coefficient kij set equal to zero
(no fitting), and the SAFT predictions have been made using
the vdW1 mixing rules only.
Figure 8. Pressure-mole fraction phase diagram for ethane-n-C28 Conclusion
mixture at 573.15 K; circles are the experimental data by Huang et An LJ-based SAFT equation of state, developed on the basis
al.,18 the solid curve is predicted by SAFT, and the dotted curve is
predicted by the LJ-based SAFT with the VF mixing rules. of thermodynamic perturbation theory of the first order (TPT1),
is found to represent pure-fluid and binary phase equilibria in
the systems of small chain molecules, such as normal alkanes.
This prototype equation of state is about 50% more accurate
than SAFT in fitting the pure-component vapor pressures and
liquid densities, and it is as accurate as SAFT in estimating
phase diagrams of symmetric n-alkane mixtures.
Acknowledgment. The authors gratefully acknowledge the
assistance of Drs. Marco Duran and Peter Condo. A preliminary
account of this work was presented at the Annual Meeting of
the American Institute of Chemical Engineers, San Francisco,
November 13-18, 1994 (paper 195c). Partial funding was
provided to M.R. by the U.S. Department of Energy, Grant DE-
FG02-96ER12201.
References and Notes
(1) Wertheim, M. S. J. Chem. Phys. 1986, 87, 7323.
(2) Chapman, W. G.; Gubbins, K. E.; Jackson, G.; Radosz, M. Ind.
Figure 9. Pressure-mole fraction phase diagram for ethane-n-C44 Eng. Chem. Res. 1990, 29, 1709.
mixture at 573.15 K; the solid curve is predicted by SAFT, and the (3) Chiew, Y. C. Mol. Phys. 1990, 70, 129.
dotted curve is predicted by the LJ-based SAFT with the VF mixing (4) Schweizer, K. S.; Curro, J. G. J. Chem. Phys. 1989, 91, 5059.
rules. (5) Dickman, R.; Hall, C. K. J. Chem. Phys. 1988, 89, 3168.
(6) Yethiraj, A.; Hall, C. K. J. Chem. Phys. 1991, 95, 8484.
when used with the LJ-based SAFT, for all the n-alkane mixtures (7) Banaszak, M.; Chiew, Y. C.; Radosz, M. Phys. ReV. E 1993, 48,
studied. Therefore, we will discuss only the VF-derived results 3760.
(8) Banaszak, M.; Chiew, Y. C.; O'Lenick, R.; Radosz, M. J. Chem.
for other examples of the n-alkane binaries, and compare them Phys. 1994, 100, 3803.
with those from SAFT. (9) Ghonasgi, D.; Chapman, W. G. AIChE J. 1994, 40, 878.
The next example is for a nearly symmetric system of (10) Kraska, T.; Gubbins, K. E. Ind. Eng. Chem. Res. 1996, 35, 4727.
(11) Huang, S. H.; Radosz, M. Ind. Eng. Chem. Res. 1990, 29, 2284.
ethane-propane. The term symmetric refers to the difference (12) Huang, S. H.; Radosz, M. Ind. Eng. Chem. Res. 1991, 30, 1994.
in size; these components are much more similar in size than (13) Johnson, J.K.; Zollweg, J. A.; Gubbins, K. E. Mol. Phy. 1993, 78,
those discussed above. As is shown in Figure 7, both equations 591.
(14) Chen, C. K.; Duran, M. A.; Radosz, M. Ind. Eng. Chem. Research,
of state predict the experimental data very well. In fact, the 1993, 32, 3123.
LJ-based SAFT is slightly more accurate. (15) Banaszak, M.; Chiew, Y. C.; Radosz, M. Fluid Phase Equilib. 1995,
Figure 8 shows the results for an asymmetric system of 111, 161.
ethane-n-C28. The MM difference is about 360 g mol and the (16) Reamer, H. H.; Sage, B. H. J. Chem. Eng. Data 1962, 7, 16.
(17) Maschke, D. E.; Thodos, G. J. Chem. Eng. Data 1962, 7, 232.
ratio of the LJ diameters, σC28/σC2, is greater than 2. As Figure (18) Huang, S. H.; Lin, H. M.; Chao, K. C. J. Chem. Eng. Data 1988,
8 illustrates, the LJ-based SAFT still predicts the experimental 33, 143.

Das könnte Ihnen auch gefallen