You are on page 1of 12

Fluid Phase Equilibria 236 (2005) 6677

Prediction of Henrys law constants of small gas molecules in liquid


ethylene oxide and ethanol using force field methods
Chuanjie Wu a,b , Xiaofeng Li a , Jianxing Dai a , Huai Sun a,
a

School of Chemistry and Chemical Technology, Shanghai Jiao Tong University, Shanghai 200240, China
b College of Pharmaceuticals and Biotechnology, Tianjin University, Tianjin 300072, China
Received 17 January 2005; received in revised form 15 July 2005; accepted 1 August 2005

Abstract
Widom insertion method coupled with canonical ensemble (NVT) molecular simulations based on different force fields was utilized to
calculate Henrys law constants (HLCs) of several common gas molecules, N2 , O2 , CH4 , and CO2 , in liquid ethylene oxide and ethanol.
Several simulation options, including insertion grid densities, liquid densities, force field models, and simulation methods were investigated.
The results indicate that reasonably good agreements between the calculated and experimental values were obtained using the united-atom
(UA) model. However, systematically overestimated values are obtained with the all-atom (AA) force field model. The causes of the problems
were investigated comparatively. The data indicate that hydrogen atoms in the AA model are mainly responsible for these problems, but the
underlying sampling algorithm in Widom insertions, which may be the real source of problems, needs further investigation.
2005 Elsevier B.V. All rights reserved.
Keywords: Henrys law constant; Force field; Widom insertion; Monte Carlo; Molecular dynamics simulation

1. Introduction
Henrys law constant (HLC), as a measurement of the solubility of gas molecules in solvent, is a fundamental property
in chemical engineering processes. Traditionally, HLC values are measured experimentally, which is a difficult task
especially under extreme conditions [17]. Accurate prediction using computational methods is highly desirable. Several
methods, such as QSPR [8], empirical methods based on thermodynamics principles [9], or based on solvation models of
quantum chemistry calculations [10], and molecular simulation methods [1117] have been proposed. With the advance
of computer technologies, molecular simulation methods
based on force filed technologies are regarded as promising
alternatives for predicting HLC values. In addition to the significances in applied chemistry, the calculation of the HLC
values using force field methods requires accurate evaluation

Corresponding author. Tel.: +86 21 5474 8987; fax: +86 21 5474 1297.
E-mail address: huaisun@sjtu.edu.cn (H. Sun).

0378-3812/$ see front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.fluid.2005.08.008

of excess chemical potentials, which has a profound impact


in many thermodynamic aspects.
The HLC of solute 2 in solvent 1 (H2,1 ) at constant temperature can be expressed as [18]
f2
x2 0 x2

H2,1 = lim

(1)

where x2 is the mole fraction of the solute in the solvent, f2


is the fugacity of the solute. Using thermodynamic relations,
Eq. (1) can be written as [19]
H2,1 = 1 RT exp(ex,
)
2

(2)

where 1 is the molar density of the solvent, = 1/(kB T), and


ex,
is the excess chemical potential of the solute at infinite
2
dilution. Eq. (2) is the basis for predicting HLC in molecular
simulation methods. The central issue is to accurately calculate the excess chemical potential. Several approaches have
been developed [2027]. The Widom insertion method [20] is
perhaps the default method [28]. However, like many simple Monte Carlo methods, this method has efficiency problem

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

for high-density systems. Most calculations published were


based on simplified united atom (UA) models. On the other
hand, great amount of efforts have been devoted to the development of general AA force fields such as AMBER/OPLS
[29,30], COMPASS [31]. It has been demonstrated that broad
range of physical properties, from gas to condensed phases
can be accurately predicted with high quality parameters
[3032]. Therefore, it is of great interest to evaluate if a general AA force field can be used to predict the HLC accurately.
This project was a response to the Second Industrial Fluid
Property Simulation Challenge (2nd IFPSC) [34]. According
to the requirements of the competition, four gas molecules,
N2 , CO2 , CH4 , and O2 in liquid ethanol at different temperatures were predicted. We started with a goal to predict these
HLC values using an AA force field model. The question
was: if the force field has been derived and validated using
the quantum mechanics and liquid data (densities and heats
of vaporization) of pure substances, can we use it to reliably
predict the HLC that is a measure of gas molecular solubility in solvent? In the process of this project, we encountered
difficulties in getting good agreements with the experimental data. Therefore, we started to search for a solution to the
problems by extending calculations with various options. In
particular, we applied both AA and UA force fields and the
mixtures of the force field models in the calculations. This
paper summarizes what we have learned so far. Although we
cannot make a definite conclusion yet, our results serve to
narrow down the possibilities. Hopefully, they form a good
starting point for further studies.

2. Theory and methods


2.1. Force eld
The AA force field parameters were derived using quantum mechanics ab initio and liquid data for pure substances
[3032]. The result parameters were applied in mixtures without any further adjustment, which means the valence and
charge parameters were transferred, nonbonded LennardJones parameters were constructed based on the LorentzBerthelot combination rules:

ij = i j
(3)
ri0 + rj0
rij0 =
.
2
The parameters, together with corresponding molecules, are
part of the TEAM force field [33].
The TEAM force field uses functional forms similar to
other AA force fields such as CFF [35], COMPASS [31],
AMBER/OPLS [29,30]. The difference however is that it
is a fragment-based force field database containing unique
fragments and associated parameters. A fragment is a complete molecule, which is either a small molecule itself (such
as H2 O, CH4 , N2 ) or a fragment cut from a large molecule

67

and saturated with hydrogen atoms at the ends. Parameters


are transferred and used based on identification of the unique
fragments. Details about the methodology are published elsewhere [32,33].
The force field has the following functional forms [32]:

 
7
2
3
4
kb (b b0 ) 2(b b0 ) + (b b0 )
E=
3
bond

ka [( 0 )2 + c( 0 )3 + d( 0 )4 ]
+
angle

[kt (1 cos(n 0 ))] +

torsion

kbb (b b0 )(b b0 )

kba (b b0 )( 0 ) +

bondangle


i<j

 
 0 6
12
rij0
rij

ij
2
rij
rij

where:
2( 20 )
1
c=
,d =
30 ( 0 )
20 ( 0 )
c = d = 0.0

k0 ( 0 )2

out-of-plane

bondbond

 q i qj
i<j

rij
(4)

(0 = 0, )

(0 = 0, ).

qi and qj are the partial charges located at nuclei. The


partial charges are calculated from bond dipole increments
ij , which represents charge separation between two valence
atoms. Then net partial charges for atom qi is the summation
of all charge bond-increments related to this atom:

qi =
ij .
(5)
j

In order to derive the valence and electrostatic parameters,


quantum mechanics calculations were carried out at the level
of B3LYP/6-311G** using Gaussian 98 software package
[36]. The calculated energies, first and second derivatives
of the energy were subsequently used to fit the force field
parameters. Lennard-Jones parameters were optimized using
experimental densities and vaporization enthalpies [31,32]
of liquids. The parameterizations were conducted using DFF
software package [37]. All the force field parameters (including the force constants initiated with the letter k, the equilibrium molecular structure parameters with the subscript of
0, the charge bond-increments ij , and the two LennardJones 12-6 parameters) of molecules involved in this article
are available in the supplementary materials of this article.
The parameters of the UA force fields came from different
sources. For ethanol and methane molecules, the parameters
were taken from the TraPPE force field [38]. For nitrogen and
oxygen molecules, we derived the parameters by fitting their
liquid densities and heats of vaporization [37]. For carbon

68

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

dioxide, the parameters were taken from the literature [39].


The UA force fields use the same LJ-12-6 function and the
combination law as explained above.
2.2. Excess chemical potential equation
Excess chemical potential (ex,
) of molecular system
2
of canonical ensemble can be evaluated using the following
expression [21]:


ex,
= kT ln exp( Etest 
(6)
2
where Etest is the total potential energy between the test
molecule and the solvent molecules, and can be determined
with Widom insertion method [20];   denotes the ensemble
average.
2.3. Simulation details
The liquid phase space was sampled with NVT ensemble
Monte Carlo (MC) and molecular dynamics (MD) methods. Nonbond interaction was evaluated with charge-group
based cut-off method [31] and tail corrections. The initial
liquid models were prepared by randomly packing solvent
molecules into cubic boxes with periodic boundary condition. Two hundred molecules were placed in the simulation boxes for ethylene oxide; 300 were used for the
ethanol.
2.3.1. Monte Carlo simulation
A general Metropolis Monte Carlo program was implemented. At each trial move of our program, a molecule
is randomly selected and then subject to four basic
movestranslation, rigid rotation, vibration and internal
rotation, performed in sequence. These moves are amplitudecontrolled random processes, with a criterion of about 50%
acceptance ratio. The typical maximum displacements of the
for the translation, 0.20 rad for
basic moves are ca. 0.20 A

the rigid rotation, 0.015 A for the vibration and 0.080 rad for
the internal dihedral rotation, respectively. If one of the basic
moves is rejected, the entire trial is done, and a new trail is
started. Since the molecule is selected randomly, and each
of trail moves is controlled by the Metropolis method, the
procedure meets the detailed balance condition.
We grouped 200300 success Monte Carlo trial moves
(which roughly equal the number of molecules in the simulation box) as one simulation cycle. The configuration space
was normally equilibrated for 5 104 cycles, then sampled in
an interval of 100 cycles for 105 cycles in total for computing
the HLC values.
2.3.2. Molecular dynamics simulation
The molecular dynamics (MD) simulations were carried
out using the Verlet velocity integrator [40]. The temperature
was controlled using stochastic collision method proposed by
Andrea et al. [41]. The time step was 1 fs. The models were

equilibrated for 20 ps, and the configuration was sampled in


an interval of 20 fs from a total trajectory of 20 ps.
2.3.3. Widom insertion
Uniform insertion of the test particle in the configuration
space is a central issue in basic Widom method. The insertion can be randomly placed, which requires a good random
number generator. It has been reported that using grid points
for insertion can enhance the efficiency by 20% [21]. In this
work, we implemented the grid method. We tested the optimal
density of grid points for accuracy and efficiency by varying
the number of grid points from 103 to 503 , which corresponds
3 . The grid density of
to grid density of ca. 0.053.0 grids/A
3

about 0.5 grids/A was found to be necessary, and adopted


in this work.
2.4. Uncertainties
The uncertainty ( H ) of the calculated HLC was estimated using the standard deviation ( B ) of the quantity of
exp( E) in Eq. (3), which was calculated with block average method [42]. Five blocks were applied in the evaluation.
In order to estimate uncertainties, we differentiate Eq. (6):
d(HLC)
= 1 RT
=

exp( Etest )2


d (exp( Etest ))

HLC
d (exp( Etest )) .
exp( Etest )

(7)

Therefore, the uncertainties of HLC can be calculated by the


following equation:
H =

HLC
B .
|exp( E)|

(8)

3. Results and discussion


3.1. Validation results of the force eld parameters
The optimized TEAM force field parameters for nitrogen,
oxygen, carbon dioxide, methane, ethane, ethylene oxide and
ethanol molecules were validated by calculating properties
of molecules in isolation and in liquid phases. Table 1 lists
intramolecular properties of small gas molecules. Note that
the vibrational frequencies as well as structural parameters
are listed for comparison. Ab initio, force field and experimental [43,44] data are listed for comparison. The structural
data for ethane, ethylene oxide and ethanol [45,46] are given
in Table 2, and the comparisons of its vibration frequencies
are illustrated in Fig. 1. The results demonstrate that the force
field parameters are capable of predicting the intramolecular
(gas phase) properties.
A critical aspect of a force field for condensed phase
simulation is its ability in representing the intermolecular

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

69

Table 1
Comparison of experimental, ab initio and TEAM force field calculated intramolecular properties
Molecule

Property

TEAM

B3LYP/6-311G**

Experimental [43,44]

N2

B0 (A)
(cm1 )

1.095
2358.0

1.095
2447.5

1.098
2358.0

O2

B0 (A)
(cm1 )

1.206
1580.0

1.206
1640.9

1.207
1580.0

CO2

B0 (A)
( )
1 (cm1 )
2 (cm1 )
3 (cm1 )
4 (cm1 )

1.160
180.0
2398.5
1392.6
667.0
667.0

1.160
180.0
2436.1
1375.6
666.5
666.5

1.160
180.0
2349.0
1333.0
667.0
667.0

CH4

B0 (A)
( )
1 (cm)1
2 (cm)1
3 (cm1 )
4 (cm1 )

1.091
109.5
1341.6
1561.5
3025.8
3131.9

1.090
109.5
1341.2
1561.1
3028.5
3134.8

1.094
109.5
1306.0
1534.0
2917.0
3019.0

potentials. The charge parameters were derived from ab initio electrostatic potentials. The VDW parameters were optimized using liquid densities and vaporization enthalpies of
the molecules considered, as given in Table 3. The simulated
results using the optimized force field parameters are listed
in the same table for comparison.
To further validate the force field, we applied the obtained
parameters in NPT simulations to predict the PVT equations of state for the liquids. The results are plotted in Fig. 2.
The results indicate that excellent agreements between calculation and experimental data [47] are obtained in a wide
range of temperatures.

The UA force field parameters of N2 and O2 were obtained


using the same procedure and empirical data as those used
for deriving the AA force field parameters. The empirical and
validation results are also given in Table 3, and Fig. 2b and
d.
Without any further adjustment, the parameters were used
in the prediction of HLC parameters in this work.
3.2. HLC of Ar, N2 , CH4 , and C2 H6 in ethylene oxide
Before predicting the HLC values of gas molecules in
ethanol as required by the 2nd IFPSC [34], we tested the pro-

Table 2
Comparison of experimental, ab initio, and molecular mechanics (TEAM force field) calculated geometric properties of ethanol
Molecule

Property

TEAM

B3LYP/6-311G**

Experimental [46,47]

CH3 CH3 (ethane)

C C (A)

C H (A)
CCH ( )
HCH ( )
HCCH ( )

1.537
1.093
111.5
107.4
122.4

1.531
1.094
111.4
107.5
122.2

1.535
1.094
111.2

Cb H3 Ca H2 OH (ethanol)

C C (A)

C O (A)

O H (A)

Ca H (A)

Cb H (A)
CCO ( )
COH ( )
Cb Ca H ( )
Ca Cb H ( )
HCa H ( )
HCb H ( )
Cb Ca OH ( )

1.532
1.429
0.963
1.096
1.095
114.1
106.7
110.1
111.7
106.7
108.4
59.9

1.525
1.425
0.962
1.100
1.093
112.7
107.7
110.2
111.0
107.2
107.6
61.5

1.530
1.425
0.945
1.094
1.094
107.3
108.5
110.3

109.1
108.6

C C (A)

C O (A)

C H (A)
HCH ( )
HCHC ( )

1.468
1.430
1.087
115.499
26.249

1.468
1.429
1.087
115.382
26.184

1.466
1.431
1.085
116.600
22.000

(ethylene oxide)

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

70

Fig. 1. Comparison of experimental and calculated vibrational frequencies. (


) TEAM force field calculation results; (
results. (a) Ethane; (b) ethanol; (c) ethylene oxide.

gram by calculating the HLC values of gas molecules Ar, N2 ,


CH4 , and C2 H6 in liquid ethylene oxide. As we explained in
the previous section, the force field densities agree well with
the experimental data for ethylene oxide in the temperature
range interested in these calculations, we used the experimental densities for liquid ethylene oxide.
Table 4 lists calculated results for argon. Three temperatures and five different grid densities were investigated. It is

) density function theory calculation

clear that the density of grid points should be greater than


3 in order to obtain consistent results. Using MC
0.5 grids/A
or MD method to prepare the solvent samples does not yield
significant differences in the HLC values. The experimental
data [48] are listed in Table 4 for comparison. At 323 K, the
predicted values agree reasonably well with the experimental data, but large deviations are found at lower temperatures.
Our results show flat temperature dependence in the range,

Table 3
Experimental data and NVT MD simulation results for the liquid models
Liquid model

Nitrogen
Nitrogen (UAa )
Oxygen
Oxygen (UA)
Carbon dioxide
Methane
Ethane
Ethanol
Ethylene oxide
a

Experimental [47]

Calculated
(g/cm3 )

T (K)

77.35
77.35
90.17
90.17
216.60
111.66
184.55
320.00
298.20

0.8078
0.8078
1.1420
1.1420
1.1790
0.4240
0.5447
0.7665
0.8624

Hvap (kcal/mol)

Hvap (kcal/mol)

Err. of Hvap (%)

1.3571
1.3571
1.6104
1.6104
3.6500
1.9497
3.5083
9.5414
5.9537

1.3557
1.3510
1.6111
1.6070
3.6549
1.9473
3.5044
9.5431
5.9518

0.1023
0.4495
0.04347
0.2111
0.1343
0.1231
0.1112
0.01782
0.06957

UA = united-atom force field. Except the molecules denoted as UA, the other molecules were all simulated with TEAM all-atom force field.

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

but the experimental values show monotonic decreases as the


temperature increases.
Using the optimized grid density, we performed calculations for N2 , CH4 and C2 H6 in liquid ethylene oxide and the
results are listed in Table 5. Again, experimental data [48] are

71

listed for comparison. Using the TEAM force field, the predictions are generally overestimated; especially for CH4 , the
calculated values are about three times larger than the experimental data. We performed the calculations with UA model
(TraPPE), the results are in much better agreements with the

Fig. 2. Comparisons of calculated () and experimental (solid line) densities under saturated pressure. (a) Nitrogen; (b) nitrogen (UA); (c) oxygen; (d) oxygen
(UA); (e) carbon dioxide; (f) methane; (g) ethane; (h) ethanol; (i) ethylene oxide. Note: UA = united-atom force field. Except the molecules denoted as UA, the
other molecules were all simulated with TEAM all-atom force field.

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

72

Fig. 2. (Continued).

Table 4
Henrys law constant (in MPa) calculated for argon in liquid ethylene oxide
T (K)
273.15

Da [47] (g/cm3 )

Gridb

Grid density (1/3 )

MDc

0.8974

103

MCd

0.061
0.491
1.656
3.926
7.668

125.5
123.2
125.3
125.0
124.8

21.3
7.4
10.1
10.2
10.5

117.3
145.1
147.8
148.1
148.0

40.5
21.5
20.8
19.7
19.6

167.0

203
303
403
503

Experimental [48]

298.15

0.8624

103
203
303
403
503

0.059
0.472
1.592
3.773
7.369

131.9
133.4
131.9
132.3
132.3

6.9
5.1
5.6
5.7
5.7

132.0
133.6
134.7
135.1
135.1

8.1
21.8
18.9
19.2
19.0

142.0

323.15

0.8252

103
203
303
403
503

0.056
0.451
1.523
3.610
7.051

125.8
128.7
127.8
128.2
128.2

3.4
3.2
3.2
2.7
1.7

141.4
135.4
134.9
135.0
135.0

9.9
10.5
10.3
10.5
10.6

127.0

a
b
c
d

The density of liquid ethylene oxide.


The number of the lattice points.
The liquid models were sampled with molecular dynamics simulation.
The liquid models were prepared with Monte Carlo method.

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

73

Table 5
Henrys law constant (in MPa) of N2 /CH4 /C2 H6 in ethylene oxide
Methoda

Potential modelb

Temperature (K) of
273.15

298.15

323.15

(a) N2
MD
MC
MD
MC
Experimental HLC [48]

AA/AA
AA/AA
AA/UA
AA/UA

249.6 21.4
344.7 47.5
162.2 15.3
208.5 44.2
280

256.6 14.9
247.3 50.7
174.1 10.1
174.2 36.3
218

224.5 6.4
230.2 31.5
159.8 6.1
169.5 18.1
182

(b) CH4
MD
MC
MD
MC
Experimental HLC [48]

AA/AA
AA/AA
AA/UA
AA/UA

124.0 19.1
178.9 82.1
36.6 4.8
49.7 12.0
61.3

150.4 12.4
147.1 31.0
50.0 3.5
48.6 12.6
61.4

144.3 6.6
148.8 18.4
53.4 2.4
55.6 6.9
59.5

(c) C2 H6
MD
MC
MD
MC
Experimental HLC [48]

AA/AA
AA/AA
AA/UA
AA/UA

17.0 10.8
17.1 30.5
5.7 2.4
6.6 6.3
8.4

34.7 10.7
15.8 5.9
11.3 3.0
7.9 4.2
10.9

33.6 8.1
34.0 12.2
11.0 1.4
9.9 4.3
12.9

a
b

The method for generating liquid samples.


The force field for liquid ethylene oxide and gas molecules respectively. AA and UA represent all-atom and united-atom force fields, respectively.

experimental data. The MD and MC generated configurations yield similar results, certainly within the uncertainties,
as given in the table.
3.3. Prediction of HLC of N2 , O2 , CO2 , and CH4 in
liquid ethanol
The calculated HLC values of N2 , O2 , CH4 and CO2 in
liquid ethanol at two specific temperatures are compiled in
Table 7, together with the experimental data [49] published
by the IFPSC committee for comparison. It is well known
that the liquid density has strong impact on the calculated
HLC, hence we carried out the calculations using four liquid
densitiesthe experimental densities, the GEMC simulated
densities [38] using the UA model, the NPT MD simulated
densities using the UA model, and NPT MD simulated densities using the AA model. The densities are summarized in
Table 6. The density models used in the prediction are listed
in the first column of Table 7. The second column lists the
simulation methods used for generating the configurations of
Table 6
Calculated and experimental densities (in g/cm3 ) of liquid ethanol
Temperature (K)

GEMC [38]

TraPPE

TEAM

Experimental [47]

275
300
323
325
373
375
425
475

0.804
0.782
0.754
0.756
0.704
0.705
0.635
0.524

0.800
0.780
0.764
0.761
0.715
0.703
0.647
0.552

0.813
0.789
0.772
0.769
0.728
0.725
0.676
0.628

0.808
0.786
0.764
0.762
0.710
0.708
0.643
0.553

the solvent (ethanol); MD and MC simulations, both in the


canonical (NVT) ensemble, were utilized. The third column
lists the force field models used for the simulations. Different AA and UA force field combinations were used for the
solvent (ethanol) and gas molecules. The rest columns of the
table are the predicted HLC values and their estimated standard deviations for the four gas molecules.
Several points can be made based on the data given in
Table 7. First of all, since all density models yield very similar values as given in Table 6, the calculated HLC values
do not show significant deviations from one density model
to another. Secondly, the MD and MC generated configurations are essentially equivalent. As given in the table, when
the liquid model and the force field model are the same, the
calculated HLC values are essentially same (within the estimated standard deviation) between those calculated using the
MD and MC methods. Thirdly, different force field models
yield very different results. With AA force field for solvent,
the predictions are failed by as high as 570% overestimated
in comparison with the experimental values with very large
standard deviations. A general observation is, as long as the
solvent is modeled using the UA model, the predicted results
are close to the experimental data. The force field model
used for the solute molecules makes some, but not significant differences. Overall, the best results were obtained by
the UA/AA model, which means UA for the solvent (ethanol)
and AA for the gas molecules were with AA (TEAM) force
field. Finally, comparing the calculated values for each of the
gas molecules with the available experimental data, we see
that reasonably good results were obtained for N2 , O2 and
CH4 , either use UA/AA or UA/UA model.

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

74

Table 7
Henry law constants (in MPa) of N2 , O2 , CH4 and CO2 gas molecules in ethanol at 323 and 373 K
Densitya

Potential (ethanol/gas)c

N2

(a) 323 K
Expt.
MD
Expt.
MD
Expt.
MD
Expt.
MD
Expt.
MC
Expt.
MC
GEMC
MD
GEMC
MD
GEMC
MC
GEMC
MC
NPT
MD
NPT
MD
Experimental HLC (MPa)[48]

AA/AA
AA/UA
UA/UA
UA/AA
UA/UA
UA/AA
UA/UA
UA/AA
UA/UA
UA/AA
UA/UA
UA/AA

980.0
618.2
223.8
236.7
232.5
260.3
213.4
233.8
200.9
221.7
194.6
217.9
253.6

441.6
250.2
26.3
31.0
10.3
11.0
19.2
32.5
9.1
11.3
9.3
11.7
7.6

396.5
253.4
132.5
140.4
136.9
152.9
123.5
135.1
120.0
132.2
121.9
135.3
174.8

137.0
73.4
12.5
16.5
5.0
4.5
6.2
13.1
3.4
4.3
4.7
6.2
8.7

545.7
179.5
100.6
128.9
104.9
135.0
97.5
122.8
91.3
117.7
85.0
105.3
81.5

187.7
80.2
14.1
8.9
4.4
4.2
10.7
12.4
5.0
6.6
6.2
10.5
2.4

69.8
73.4
50.9
29.3
58.7
29.2
54.2
27.7
44.7
30.4
35.8
22.0
21.1

80.5
122.7
9.4
6.7
5.7
4.4
10.3
10.6
4.9
2.5
10.0
6.8
1.1

(b) 373 K
Expt.
MD
Expt.
MD
Expt.
MD
Expt.
MD
Expt.
MC
Expt.
MC
GEMC
MD
GEMC
MD
GEMC
MC
GEMC
MC
NPT
MD
NPT
MD
Experimental HLC (MPa) [48]

AA/AA
AA/UA
UA/UA
UA/AA
UA/UA
UA/AA
UA/UA
UA/AA
UA/UA
UA/AA
UA/UA
UA/AA

460.6
327.7
207.2
221.8
200.7
218.6
189.1
204.2
188.5
206.7
218.9
235.8
221.3

58.7
28.9
14.7
13.7
7.6
7.0
7.6
10.6
6.3
6.2
16.5
18.4
13.3

252.2
177.0
135.7
145.8
133.4
145.2
126.7
136.9
126.0
138.4
142.7
153.5
159.9

23.3
9.8
6.9
6.4
3.8
3.5
4.7
7.1
2.7
2.8
8.5
9.6
12.8

283.7
117.9
111.1
143.9
106.7
135.7
100.3
124.8
100.9
127.4
116.1
145.7
83.4

15.1
13.3
9.0
12.2
4.7
5.7
4.9
6.2
3.7
4.8
9.6
10.9
5.0

31.3
45.2
66.1
36.6
63.6
37.0
57.7
29.6
59.1
34.8
67.2
40.1
29.7

15.9
15.8
8.4
6.9
4.8
5.0
3.6
9.8
1.9
3.2
12.2
8.0
3.3

a
b
c

Methodb

O2

CH4

CO2

The liquid ethanol densities, as given in Table 6.


The method for preparing liquid ethanol.
The force field for liquid ethanol and gas molecules respectively. AA and UA represent all-atom and united-atom force fields, respectively.

A comparison of Table 7a with Table 7b shows the temperature dependence of the calculated HLC values. To further
explore the temperature dependence of the calculated HLC
values, we extended the calculations using the most successful model (UA/AA force fields, experimental densities, and
MD simulated configurations) over a temperature range of
275475 K. The results are presented in Fig. 3. It is clearly
that each of the predicted HLC curves has a maximum, which
qualitatively agree with existing HLC experimental data
[50].
In order to find out the causes of the large errors in calculated HLC using the AA force field model, we analyzed
the temperature dependence of data calculated using the AA
force field model. As given in Table 7, when the temperature
changes from 373 to 323 K, the calculated HLC values for
N2 increases about 520 MPa. In comparison, the calculated
values (ca. 145 MPa) of O2 are significantly smaller. Both
molecules are non-polar, linear; the simulation conditions
were the same except the force field parameters. Close examination of the force field parameters as given in the supplementary material of this article indicates that the Lennard-Jones
= 0.0738 kcal/mol)
(LJ) parameters for N2 (r0 = 3.6989 A,

are obviously different from the values of O2 (r0 = 3.3679 A,


= 0.09545 kcal/mol).

Fig. 3. Henrys law constants of N2 , O2 , CH4 and CO2 in ethanol with experimental density. Hollow points and the solid lines represent the calculated
HLC with Widom method. Solid points are the experimental HLC values.
( , ) N2 ; (, ) O2 ; (,
) CH4 ; (, ) CO2 .

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

75

Table 8
HLC values of a hypothetic diatomic molecule in ethanol
No.

1
2
3
4
5
6
7
a

Lennard-Jones parameters

HLC (MPa) in ethanol at the temperature of

r0 (A)

(kcal/mol)

323 K

3.8
3.7
3.6
3.6
3.6
3.6
3.6

0.0734
0.0734
0.0734
0.0634
0.0838
0.0938
0.1000

570.1
513.1
461.3
690.8
309.7
214.9
172.6

373 K

67.3
53.8
42.2
61.7
29.0
20.5
16.7

288.3
268.4
249.9
342.5
183.9
137.5
115.8

13.9
11.2
9.0
12.1
7.1
5.1
4.3

281.8
244.7
211.4
348.3
125.8
77.4
56.8

= HLC (323 K) HLC (373 K).

Table 9
Henrys law constants (in MPa) of argon and helium in liquid ethanol
Temperature (K)

Ethanol density (g/cm3 )

Ethanol potential modela

Argon

323
373
323
373

0.7635
0.7103
0.7635
0.7103

AA
AA
UA
UA

323.3
203.7
147.8
148.0

Helium

17.3
12.9
14.3
7.9

1614.0
856.1
455.3
346.2

79.2
29.9
8.2
5.3

AA = TEAM all-atom force field; UA = TraPPE united-atom force field.

We further performed calculations using a hypothetic


diatomic molecule with different LJ parameters. In Table 8,
we listed a comparison of calculated HLC values at 323 and
373 K. In the last column of the table, we listed the difference in HLC values between the two temperatures. Close
examination of the data in this table reveals that the HLC
is more sensitive to the value than to the r0 value as
the temperature decreases. As increases from 0.0634 to
0.0934 kcal/mol, the differences in the calculated HLC values between the two temperatures decrease from 348.3 to
77.4 MPa.
This phenomenon is repeated in the monatomic molecule
cases. Using helium and argon in ethanol as model systems,
we predicted the temperature dependence of the HLC, the
data are listed in Table 9. Two force field models, TEAM
and TraPPE, are used for ethanol. First, we examine the
data obtained using the TEAM force field. The calculated
HLC values of helium in liquid ethanol are 1614.0 MPa at
323 K and 856.1 MPa at 373 K, a decrease of about 800 MPa
in the HLC with an increase of 50 K in the temperature.
For argon, the change of HLC in the same temperature
range is much smaller, about 120 MPa. The only difference
between these two sets of data is the force field parameters for the test particles. The LJ parameters, r0 and , of
and 0.02029 kcal/mol, respectively [39];
helium are 2.8960 A
and
the corresponding parameters for argon are 3.8152 A
0.2301 kcal/mol (as given in the supplementary document).
When the force field model for the solvent is switched to
UA model, the temperature dependence changes dramatically. For argon, two values at different temperatures are
almost the same; for helium, the difference is reduced to about
110 MPa.

4. Conclusion
We have investigated the feasibility of using all-atom (AA)
force field model with Widom insertion method to predict
Henrys Law Constant in this project. Using ab initio and
liquid data, we developed a set of TEAM force field parameters, and validated these parameters by calculating molecular
structures, vibrational frequencies, heats of vaporization and
PVT equations of states of liquids. The agreements with
experimental data are generally satisfactory. We then studied several options in the calculations of HLC values. We
utilized a grid-based insertion scheme, optimized the den 3 was
sity of the grid points. A minimum value of 0.5 grids/A
found to be required for the accuracy and convergence of the
HLC calculation results. We implemented a simple MC simulation engine, and compared the configuration space made
by the MC simulations with that made by MD simulations;
the results are in close agreement. By comparing our calculation results with the experimental data and calculated
data in the literature [42], we validated our Widom insertion
code.
Our main problem however remains unsolved yet. Using
the AA force field model, the calculated HLC values are systematically overestimated. We have investigated this problem
by comparing different molecular systems and different force
field models. Especially, we focus on the temperature dependence of the calculated HLC values using different models.
The obtained results are consistent so that we can make the
following statements based on our observations.
(1) The problem becomes more serious at low temperature
when the density of solvent is large.

76

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677

(2) The problem is associated with the AA model, either


used for the solute or solvent. The worst case is that both
solute and solvent are modeled by the AA model.
(3) The problem is enlarged when weak VDW parameters
are used in the simulation.
(4) The problem is worsened when the percentage of hydrogen atoms in the simulation system increases.
That the large errors are associated with weak LJ parameters and number of hydrogen atoms indicates the hydrogen
atoms not in the UA model may cause the problems. However, underlying there must be a physical reason for the failure
because the potential functions are independent of the simulations and they cannot be changed. From Eqs. (2) and (6),
one knows that the overestimated HLC are due to too much
overlap between the test molecule and solvent molecules. The
correlations of the overestimated HLC values with high densities indicate that there may be a sampling efficiency issue.
Kofke [51] recently suggested that the phase space relationship between the reference and target systems must be considered carefully when free energy perturbation techniques
such as Widom insertion are used. For Widom insertions,
if the insertions do not sample all regions of configuration
space naturally sampled by the N + 1 system, the results
are likely to be flawed. To further explore the solutions, we
are continuing this work by looking into advanced simulation
techniques such as staged insertion [24,25].
Acknowledgements
The authors thank the National Science Foundation of
China (NSFC, No. 20473052) for the financial support on
the development of TEAM force field. They also thank the
IFPSC committees for organizing the competition and providing valuable experimental data and information regarding
the researches topics.
Appendix A. Supplementary data
Supplementary data associated with this article can
be found, in the online version, at doi:10.1016/j.fluid.
2005.08.008.
References
[1] S.Y. Sheikheldin, T.J. Cardwell, R.W. Cattrall, M.D. Luque de Castro, S.D. Kolev, Environ. Sci. Technol. 35 (2001) 178.
[2] T. Shimotori, W.A. Arnold, J. Chem. Eng. Data 48 (2003) 253.
[3] M. Takenouchi, R. Kato, H. Nishiumi, J. Chem. Eng. Data 46 (2001)
746.
[4] T. Shimotori, W.A. Arnold, J. Chem. Eng. Data 47 (2002) 183.
[5] M.E. Miller, J.D. Stuart, Anal. Chem. 72 (2000) 622.
[6] K. Bezanehtak, G.B. Combes, F. Dehghani, N.R. Foster, D.L.
Tomasko, J. Chem. Eng. Data 47 (2002) 161.
[7] R. Kato, H. Nishiumi, J. Chem. Eng. Data 47 (2002) 1140.
[8] N.J. English, D.G. Carroll, J. Chem. Inf. Comput. Sci. 41 (2001)
1150.

[9] E.J. Delgado, J. Alderete, J. Chem. Inf. Comput. Sci. 42 (2002) 559.
[10] E.J. Delgado, J.B. Alderete, J. Chem. Inf. Comput. Sci. 43 (2003)
1226.
[11] S. Murad, S. Gupta, Chem. Phys. Lett. 319 (2000) 60.
[12] S. Murad, S. Gupta, Fluid Phase Equilib. 187188 (2001) 29.
[13] R.J. Sadus, J. Phys. Chem. B 101 (1997) 3834.
[14] M. Jorge, C. Schumacher, N.A. Seaton, Langmuir 18 (2002) 9296.
[15] I.G. Economou, Fluid Phase Equilib. 183184 (2001) 259.
[16] G.C. Boulougouris, J.R. Errington, I.G. Economou, A.Z. Panagiotopoulos, D.N. Theodorou, J. Phys. Chem. B 104 (2000) 4958.
[17] G.C. Boulougouris, E.C. Voutsas, I.G. Economou, D.N. Theodorou,
D.P. Tassios, J. Phys. Chem. B 105 (2001) 7792.
[18] J.M. Prausnitz, R.N. Lichtenthaler, E.G.d. Azevedo, Molecular Thermodynamics of Fluid-Phase Equilib, PTR Prentice Hall, Englewood
Cliffs, NJ, 1986.
[19] K.S. Shing, K.E. Gubbins, K. Lucas, Mol. Phys. 65 (1988) 1235.
[20] B. Widom, J. Chem. Phys. 39 (1963) 2808.
[21] M.P. Allen, D.J. Tildsley, Computer Simulation of Liquids, Oxford
University Press, Oxford, 1989.
[22] K.S. Shing, K.E. Gubbins, Mol. Phys. 46 (1982) 1109.
[23] K.S. Shing, K.E. Gubbins, Mol. Phys. 49 (1983) 1121.
[24] D.A. Kofke, P.T. Cummings, Mol. Phys. 92 (1997) 973.
[25] D.A. Kofke, P.T. Cummings, Fluid Phase Equilib. 150151 (1998)
41.
[26] G.C. Boulougouris, I.G. Economou, D.N. Theodorou, Mol. Phys. 96
(1999) 905.
[27] G.C. Boulougouris, I.G. Economou, D.N. Theodorou, J. Chem. Phys.
115 (2001) 8231.
[28] D. Frenkel, B. Smit, Understanding Molecular Simulation, Academic
Press, New York, 1996.
[29] (a) W.D. Cornell, P. Cieplak, C.I. Bayly, I.R. Gould, K.M. Merz,
D.M. Ferguson, D.C. Spellmeyer, T. Fox, J.W. Caldwell, P.A. Kollman, J. Am. Chem. Soc. 117 (1995) 5179;
(b) W.D. Cornell, P. Ciueplak, C.I. Bayly, I.R. Gould, K.M. Merz,
D.M. Ferguson, D.C. Spellmeyer, T. Fox, J.W. Caldwell, P.A. Kollman, J. Am. Chem. Soc. 118 (1996) 2309.
[30] W.L. Jorgensen, D.S. Maxwell, J. Tirado-Rives, J. Am. Chem. Soc.
118 (1996) 11225.
[31] H. Sun, J. Phys. Chem. 102 (1998) 7338.
[32] H. Sun, Fluid Phase Equilib. 217 (2004) 59.
[33] H. Sun, et al., to be published.
[34] The second Industrial Fluid Properties Simulation Challenge
(IFPSC) is organized by AIChE, ACS, NIST and some companies in 2004. More information is available at http://www.cstl.nist.
gov/FluidSimulationChallenge/problems/problem2.htm.
[35] J.R. Maple, M.J. Hwang, T.P. Stockfisch, U. Dinur, M. Waldman,
C.S. Ewig, A.T. Hagler, J. Comput. Chem. 15 (1994) 162.
[36] Gaussian 98 is licensed by Gaussian, Inc., Pittsburgh, PA, 1998.
[37] DFF (Direct Force FieldTM ) is a software package that can be used
to derive parameters from ab initio and empirical data. It is developed and licensed by Aeon Technology, Inc. San Diego, CA, USA,
2004.
[38] B. Chen, J.J. Potoff, J.I. Siepmann, J. Phys. Chem. B 105 (2001)
3093.
[39] J.O. Hirschfelder, C.F. Curtiss, R.B. Bird, Molecular Theory of Gases
and Liquids, Wiley, New York, 1954.
[40] L. Verlet, Phys. Rev. 159 (1967) 98.
[41] T.A. Andrea, W.C. Swope, H.C. Andersen, J. Chem. Phys. 79 (1983)
4576.
[42] J.R. Errington, G.C. Boulougouris, I.G. Economou, A.Z. Panagiotopoulos, D.N. Theodorous, J. Phys. Chem. B 102 (1998) 8865.
[43] D.R. Lide, CRC Handbook of Chemistry and Physics, CRC Press,
Boca Raton, 1995.
[44] R.T. Jacobsen, R.B. Stewart, M. Jahangiri, J. Phys. Chem. Ref. Data
15 (1986) 735.
[45] N.L. Allinger, K.-H. Chen, J.-H. Lii, K.A. Durkin, J. Comput. Chem.
24 (2003) 1447.

C. Wu et al. / Fluid Phase Equilibria 236 (2005) 6677


[46] NIST Standard Reference Database is available at http://webbook.
nist.gov/chemistry.
[47] C.L. Yaws, Chemical Properties Handbook, The McGraw-Hill Companies, Inc, New York, 1999.
[48] J.D. Olson, J. Chem. Eng. Data 22 (1977) 326.
[49] J.D. Olson, L.C. Wilson, Recommendations for Henrys Law Constants of Nitrogen, Oxygen, Methane, and Carbon Dioxide in

77

Ethanol. Summary of Recommended Values for Second Industrial Fluid Properties Simulation Challenge. Oct. 2004. The data
are available at http://www.cstl.nist.gov/FluidSimulationChallenge/
results2.htm.
[50] K. Fischer, M. Wilken, J. Chem. Thermodyn. 33 (2001)
1285.
[51] D.A. Kofke, Mol. Phys. 102 (2004) 405.

You might also like