AIChE Journal - 2004 - Potoff - Vapor Liquid Equilibria of Mixtures Containing Alkanes Carbon Dioxide and Nitrogen
AIChE Journal - 2004 - Potoff - Vapor Liquid Equilibria of Mixtures Containing Alkanes Carbon Dioxide and Nitrogen
AIChE Journal - 2004 - Potoff - Vapor Liquid Equilibria of Mixtures Containing Alkanes Carbon Dioxide and Nitrogen
New force fields for carboii dioxide and nitrogen are introduced that quantitatively
reproduce the vupor - liquid equilibiia (VLE) of the neat systems and their mixtures
with alkanes. [n addition to the usual VLE calculations for p~ireCO, and N,, calculu-
tions of the binary mixtures with propane were used in the force-field deeueiopment to
achieve a good balance between dispersive and electrostatic (quudrupole- quadrupole)
interactions. The transferability of the force fields was then assessed fiom calciilations
of the VLE for the binary mixtures with n-hexane, the binary mixture of CO,/N,,
- - and
the ternary mixture of C02/N2/propune. The VLE calculations were carried out using
configiirational-bias Monte Carlo simulations in either the grand canonical ensemble
with histogram- reweighting or in the Gibbs ensemble.
Introduction
The design of separation processes requires accurate to limit the predictive capability of EOS models to systems
knowledge of the vapor-liquid equilibria for each of the pure where experimental data are readily available.
components and the multicomponent mixture. Of particular Another promising alternative to EOS models for VLE
interest is the phase behavior of mixtures containing alkanes, calculations is molecular simulation. Over the last 10 years,
CO, and N,, which are typically found in the petrochemical the introduction of new simulation methodologies, such as
industry. One example is the flooding of petroleum reservoirs Gibbs ensemble Monte Carlo (GEMC) (Panagiotopoulos,
with CO, or N, to enhance oil recovery. Furthermore, signif- 1987; Panagiotopoulos et al., 1988), a recent molecular dy-
icant effort has been expended designing chemical processes namics variant of GEMC (Baranyai and Cummings. ZOOO),
where supercritical CO, is used as a “green”so1vent (Brunner, histogram-reweighting techniques (Ferrenberg and Swend-
1994; Mishima et al., 2000). This is due to the fact that CO, sen. 1988. 19891, and Gibbs-Duhem integration (Kofie, 1992).
has nearly ideal physical properties for such applications, in- have made the calculation of phase equilibria from simula-
cluding a low critical temperature of 304.3 K and moderate tion a routine task. If an accurate and transferable force field
critical pressure of 73.8 bar (Rainwater, 1991). is available, then molecular simulations have an advantage
Calculations of vapor-liquid equilibria (VLE) in chemical over EOS calculations in terms of their predictive capability,
engineering have traditionally been performed with an equa- which comes at the expense of a significantly higher expendi-
tion of state (EOS) (Reid et al., 1977). These EOS models ture of computer time necessary to determine a coexistence
are based on pure-phase properties, are easy to use, and re- curve.
quire little computational effort, but often yield only qualita- In this work, we demonstrate how molecular simulation
tive predictions of binary and ternary mixture VLE. For a techniques can be used to develop transferable force fields
quantitative representation of the VLE, however, it is usually and to predict the VLE of binary and ternary mixtures of
necessary to introduce empirically derived binary interaction substances relevant to common industrial processes. We fo-
parameters. This need for binary interaction parameters tends cus our attention on mixtures of nonpolar and (slightly) polar
molecules. The accurate prediction of VLE for these systems
has until now been possible only by fitting system-specific bi-
C‘orrcspondencr concerning this a r t i c l e s h o u l d he addressed to J . 1. Siupmonn. nary interaction parameters (Liu and Beck, 1998) or by using
Permanent a d d r e x of J. J . Potoft: Dept. of Chemical Engineermg and Materials
Science, Wayne State University. 5050 Anthony Wayne Drive. Drtroit. MI 48202. special combining rules for unlike-pair interactions on differ-
the same system is not entirely consistent and limits the N (in N,) 36.0 3.31 - 0.482
Nitrogen
where r , , . E , , , and a,, are the separation, LJ well depth, and As in the case of CO,, a number of intermolecular force
LJ size. respectively, for the pair of atoms i and j. The LJ fields for N, can be found in the literature (Murthy et al.,
parameters for interactions between like atoms are listed in 1983; Galassi and Tildesley, 1994). None of these force fields
Table 1. The parameters for unlike-pair interactions are cal- were optimized for VLE calculations, which is the motivation
culated with the Lorentz-Berthelot combining rules that are behind the development of the TraPPE N, force field. This
commonly used in molecular simulations (Allen and Tildes- force field consists of three sites. Each nitrogen atom is mod-
ley, 1987) eled by a Lennard-Jo2es site separated by the experimental
bond length of 1.10 A. The Lennard-Jones parameters for
these sites are listed in Table 1. The gas-phase quadrupole
moment of N, ( Q = - 1.4X esu) (Gray and Gubbins,
5, = 4% E,, ' (3)
1984) was reproduced by placing point charges of - 0.482 e
on each Lennard-Jones site. To maintain charge neutrality, a
point charge of +O.964 e was placed at the center of mass
C-C sites are separated by 1.54 A, a value close to the exper- (COM) of the N, molecule.
AIChE Journal
July 2001 Vol. 47, No. 7 1677
15475905, 2001, 7, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.690470719 by Central Library, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Simulation details tions were truncated at 10 A with the appropriate long-range
Pure component and binary mixtures corrections applied, and the electrostatic interactions were
calculated via an Ewald summation technique with the same
Pure-component and binary-mixture phase diagrams were
parameters.
calculated from grand canonical histogram-reweighting
Monte Carlo simulations. The application of histogram-re-
weighting methods to binary mixture VLE calculations is de- Results
scribed in detail in the original references and is not re- Pure components
peated here (Potoff and Panagiotopoulos, 1998; Potoff et al.. We present two models for COz. The first is a minor varia-
1999). The Monte Carlo program employs rigid molecule tion of the EMP2 force field where parameters for the LJ
translations and rotations. conformational changes for the well depth and collision diameter were shifted slightly to
semiflexible alkanes, and molecule insertions and removals. compensate for a change in the C-0 bond length. In the
The coupled-decoupled configurational-bias Monte Carlo original model byoHarris and Yung (1995), an effective bond
method (Martin and Siepmann, 1999) was used for the latter length of 1.1490A was used, while the experimental bond
three types f: moves. Lennard-Jones interactions were trun- length of 1.16 A was used in this work to maintain consis-
cated at 10 A, and analytical long-range corrections were ap- tency between this model and the TraPPE models for the
plied to the energy, pressure, and chemical potential (Allen n-alkanes and nitrogen. The second model for CO, is an al-
and Tildesley, 1987). Electrostatic interactions were calcu- ternate parameterization derived by us (TraPPE CO, ). In
lated with the Ewald summation technique using tinfoil particular, the Lennard-Jones parameters and partial charges
boundary conditions, an Ewald screening parameter K X I, = were optimized to reproduce the vapor-liquid phase diagram
5, and K,,, = 5 for the upper bound of the reciprocal space of the binary mixture of CO,/propane. Thus, propane is es-
summation (Allen and Tildesley, 1987). sentially used as a “probe” molecule to find a suitable bal-
Pure-component simulations used a box length of L = 20 ance of the Lennard-Jones (overlap and dispersive) and
A (or 25 A near the critical region), while L = 25 A was used Coulombic (first-order electrostatic and induction) interac-
for the binary-mixture calculations. Critical points for the bi- tions. It should be noted, that this approach treats the inter-
nary mixtures were determined via a mixed-field analysis actions as effective interactions. that is, the Lennard-Jones
(Bruce and Wilding, 1992; Poioff and Panagiotopoulos, 1998) r p hpart implicitly includes many-body dispersive interactions
using a system size of L 25 A, except for the CO,/n-hexane and the additional contributions arising from instantaneous
mixture, where L = 30 A. The critical parameters reported dipole-quadrupole ( r - ‘ ) and quadrupole-quadrupole ( r - ‘“1
~
here are estimates for a single system size. To obtain more interactions. In a similar way, the partial charges mimic both
precise values of the critical parameters would require addi- the first-order electrostatic and second-order induction
tional and longer simulations for larger system sizes (Wilding, forces, and the TraPPE CO, quadrupole moment ( Q =
1995; Potoff and Panagiotopoulos, 1998). The final configura- - 4 . 5 2 ~lO-”esu) is slightly enhanced compared to the ex-
tion of a previous run was used as the initial configuration perimental value for an isolated CO, molecule ( Q = - 4.4 X
for another simulation at a different state point. Equilibra- 10p”esu) (Gray and Gubbins, 1984), but remains within the
tion periods of more than lo6 Monte Carlo steps (MCS) were statistical uncertainty of the experimental measurement.
used. Away from the critical region, vapor-phase simulations
were run for 5 x 10‘ MCS, while simulations in the liquid
phase were run for 1 to 2 X lo7 MCS. The production periods
were lengthened to 5 X lo7 MCS in the critical region. Simu-
lation data were stored in the form of lists containing the
number of molecules of each species and the energy of the
system. Samples were written to this list every 250 or 500
MCS.
Ternary mixtures
z
h
m 2
-
I $2
~
s
3 -
1 -
2.5 -
I
T 7 000 (I/K) 8 9 70 11 72
7 P 7 0 0 0 (7/K)
Figure 2. Clausius-Clapeyron plot of the saturated va-
por pressure for carbon dioxide vs. the in- Figure 4. Clausius-Clapeyron plot of the saturated va-
verse temperature. por pressure for nitrogen vs. the inverse tem-
f-xpcrimental data and critical point (solid line and asterisk) perature.
(NI5T 2000), EMP2 model (squares), and TraPPE model Experimental data and critical point (solid line and asterisk)
(ill'le\) (NIST, 2000). G T model (squares), and TraPPE model
icirclcs).
80
0
F 200 400
kg/m3
600
I
800
Mixtures
As a further test of our models for CO, and N,, we per-
formed calculations for a number of binary mixtures. The first
of these mixtures was CO,/propane at T = 294.15 K, which
was also used in the parametrization of the TraPPE CO,
model (see earlier). In Figure 5, the results of the simulations
for the TraPPE force field are shown. For comparison, we
Figure 3. Vapor-liquid coexistence curve for nitrogen.
also include results for simulations of the EMP2 CO, model
Experimental coexistence data and critical point (solid line
and asterisk) ( N E T , ZOOO), G T model (squares), and TraPPE (Harris and Yung, 1995) with the TraPPE-EH propane model
model (circles). and the experimental data of Reamer et al. (1951). As an
I
1 I
I ‘50
60
100
2 >
cu (ZI
2 40 0
Q
50
20
0
0 0.2 0.4 0.6 0.8 1 0 02 04 06 08 I
XC@*’ Yc**
Xm2p Ym*
Figure 6. Pressure-composition diagram for CO,/
Figure 5. Pressure-composition diagram for C O J
n-hexane at 353.15 K.
propane at 294.15 K.
txperiniental ddtli (ciicles) (1 I e t a l . 1981) m d FrdPPE
E x p e r i m e n t a l d a t a (circles) ( K e a r n e r e t al.. 1951), CO,/TraPPE-EH u-hesanc (solid line ,>lid d S t c r i \ k )
EMP?/TraPPE-EH propane (dashed line), TraPPE
COJTraPPE-EH propane (solid line).
AIChE Journal
1680 July 2001 Vol. 47, No. 7
15475905, 2001, 7, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.690470719 by Central Library, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
200 7 1
’OX
0 100
100 75 50 25 0
I
0.2 0.4 0.6 0.8 1 Propane
for CO, and N,. These models were found to reproduce ex-
perimental results for the pure components to a high accu-
ment with experiment. This is a significant advance over other racy. For mixtures containing C 0 2 , our new model was found
recent studies of mixture-phase behavior via molecular simu- to yield results that are in quantitative agreement with exper-
lation, which have used special combining rules for unlike-pair iment, while previous simulations for CO,/alkane mixtures
interactions belonging to interaction sites on molecules of were not able to predict the correct phase behavior when
different types (Potoff et al., 1999) or have fitted system- Lorentz-Berthelot combining rules were consistently used.
specific binary interaction parameters (Vrabec and Fischer, Our CO, model was shown to be transferable, in the sense
1997; Liu and Beck, 1998). that the phase behavior of CO,/propane and CO,/n-hexane
are equally well reproduced. Simulations of mixtures of
Conclusions CO,,”,, however, showed small, but noticeable deviations
from experiment. As a result, further refinements might be
In this work we have developed the TraPPE force fields necessary for the N, force field, the most likely being the
addition of a third Lennard-Jones site at the bond center to
better tune the dispersive interactions and the repulsive shape
300 of the model. Finally, with the calculation of the ternary phase
diagram for CO,/N,/propane, we have shown that molecu-
lar simulation can be used to predict the phase behavior of
complex, multicomponent systems to a high accuracy without
requiring special binary interaction parameters.
200
> Acknowledgments
i/
m
s
a.
Financial support from the National Science Foundation, an Al-
fred P. Sloan Research Fellowship, and an MSI Research Scholar-
ship (J.J.P.) is gratefully acknowledged. Part of the computer re-
7 00 sources were provided by the Minnesota Supercomputing Institute
(MSI). We thank Bin Chen for stimulating discussions.
Literature Cited
Allen, M. P., and D. J. Tildesley, Computer- Sinzulution of Liquids,
0 Clarendon Press, Oxford (1987).
0 0.2 0.4 0.6 0.8 Baranyai, A,, and P. T. Cummings, “Liquid-Vapor Coexistence by
Molecular Dynamics Simulation,” J. Chem. Phys., 112, 3516 (2000).
Bruce, A. D., and N. B. Wilding, “Scaling Fields and Universality of
the Liquid-Gas Critical Point,” Phys. Reu. Lett., 68, 193 (1992).
Figure 8. Pressure-composition diagram for Ndpropane Brunner, G., Gus Extraction, Springer, New York (1991).
at 270 K. Chen, B., and J. I. Siepmann, “Transferable Potentials for Phase
Expcrimental data (circles) (Yucelen and Kidnay. 1999) and Equilibria: 3. Explicit-Hydrogen Description of Normal Alkanes,”
TraPPE C O f l r a P P E NZ (solid line and asterisk). J. Phys. Chem. B , 103,5370 (1999).