Vous êtes sur la page 1sur 23

The Validity of the Potential Model inPredictingthe Structural, Dynamical, Thermodynamic Properties of the Unary and

Binary Mixture of Water-Alcohol: Methanol-Water Case

Abdalla Obeidat1 and Hind Abu-Ghazleh

1
Dept. of Physics, Jordan University of Science and Technology, Irbid, 22110, Jordan

ABSTRACT
Two intermolecular potential models of methanol (TraPPE-UA and OPLS-AA)have been used in order to examine their validity in
reproducing theselected structural, dynamical, and thermodynamic properties in the unary and binary systems.These two modelsare
combined with two water models (SPC/E and TIP4P).The temperature dependence of density, surface tension, diffusion and
structural properties for the unary system have been computed over specific range of temperatures (200-300K).The very good
performance of the TraPPE-UA potential model in predicting the surface tension, diffusion, structure, and density of theunary
system led us to examine its accuracy and performance in its aqueous solution with water. In the binary system the same properties
were examined,using different mole fractions of methanol.The TraPPE-UA model combined with TIP4P-water shows a very good
agreement with the experimental resultsfor the density and surface tension properties; whereas the OPLS-AA combined with
SPCE-water shows a very agreement with experimental results regarding the diffusion coefficients. Two different approaches have
been used in calculating the diffusion coefficients in the mixture, namely the Einstein equation (EE) and Green-Kubo (GK)
method. Our results show the advantageous of applying GK over EE in reproducing the experimental results and in saving
computer time.

Keywords:TraPPE-UA and OPLS-AA force fields,Surface tension, liquid density, diffusion coefficients, Einstein relation, Green-
Kubo, radial distribution function

Introduction

The study of alcohol-water mixtures has receivedhigh theoretical and experimental attention in the physical and chemical sciences
over the past several years1-26.It is also commonplace in numerous engineering, industrial, medical and biological applications27,
28
.Methanol has been applied in a wide spectrum of applications; they are widely used in industry and everyday life as well as a
chemical reagent, solvent, anti-freeze, pesticide and cleaning agent8.
The cornerstone of any molecular dynamic simulations is the empirical potential model used to describe the intermolecular
interactions. Many computational investigationshave been undertakentoexamine the dynamic and structural properties of the pure
methanol23, 29-33
and their aqueous solutions, and they seemto depend strongly on the molecular potentialmodel used in the
simulation. To validate the potential model, many thermodynamic,structural, mixing and transport properties should be reproduced
quantitatively or qualitatively when compared to experimental published data.Each potential modelwas constructed and
parameterizedfor different purposes. One of the most common force fields used to describe the intermolecular interaction for
several types ofmoleculesis the OPLS model34.It was developed byWilliam L. Jorgensenet al for liquid simulations.In the OPLS-
AA model, each atom has its own parametersincluding all the hydrogenatoms in the methyl group35.Another common force field
used to describe the interaction between molecules is theTraPPE-UA force field which treats the hydrogen atoms in the methyl
group asadead load36.

1
Complementary experimental and theoretical studies on liquid water and its mixture with alcohol have been recentlyundertaken.
Among them, Soetens and Bopp checked the ability of two flexible models (PHH (methanol) and BJH (water)) in producing
mixing properties of water-methanol mixture which are already satisfactory for the pure liquid methanol and water18.Many
potential models have been invented to reproduce specific properties of methanol. Monica and Berend showed qualitative
differences between four models of methanol (i.e. J1,J2, H1 ,H2 and van Leeuwen and Smit33. L Bianchi et al studied the structural
properties of methanol, using six-site APR6 potential model and compared it with the extended three site H1 model and with the
corresponding experimental values29. Edgar et al applied molecular dynamic simulations to study the thermodynamic, dynamical
and dielectric properties of methanol-water mixture, using the OPLS-AA potential model combined with two different water
models(SPCE and Tip4P)6.In our (forthcoming) paper wecompared the performance of nine models of methanol (J1, J2, H1, H2,
B3, OPLS-AA,TraPPE-UA, van Leeuwen and GROMOS) in reproducing the different properties of pure methanol such as surface
tension, diffusion, density, and the radial distribution functions. All the mentioned properties were carried out at 11 temperature
points between 200 and 300K with step of 10K.We found that both the TraPPE-UA and OPLS–AA potential models show good
performance in predicting the mentioned properties. Consequently,our interest in this work is to comparethe performance of these
two models in predicting the above listed propertiesin the binary mixtures. All simulationswere performed with fixed system size
of 400 molecules in the unary system and a total of 1,000 molecules in the binary system. In the current work, our first goal is to
check the performance of the TraPPE-UA and OPLS-AA force fields through testing two thermodynamic properties, i.e., the
density and the surface tension, in the unary and in the binary systems.The dynamical property is also calculated to check the
accuracy and performance of the two latter force fields in the same systems, using either the Einstein equation EE or the Green-
Kubo GK method37, 38.For the pure systems the diffusion coefficientswere computed using the Einstein equation (EE).Our results
are slightly different from those of ref6, 10. The conflicting and high fluctuatingresults for the diffusion coefficient obtained from
Einstein equation (EE) in the binary system led us to use the Green-Kubo (GK) approach. In addition to calculating the above
properties, we examined the validity of TraPPE-UA and OPLS-AA models by calculating the structural properties, specifically the
radial distribution function of methanol and its aqueous solutions as a function of temperature for the unary system and as a
function of methanol mole fraction in the binary system.
The paper is organized as follows: in section II the simulation methodologyis introduced and details on the intermolecular potential
models are given. In sectionIII and IV we present and discuss the simulation results for the thermodynamic properties and
dynamical properties of unary and binary systems of methanol. The study has been completed by calculating the structural
properties of the studied systems.The last section is devoted to conclusions and directions to future work.
Simulation Details
All simulations were performed using GROMACS package39.The density map of the binary water-methanol however was confined
to a sphere of radius of 4nm.The molecules are confined first inside a box of about 3x3x10 nm 3(see Figure 1) then the box is
inserted in the middle of a bigger box of 3x3x30 nm3 to assure vacuum in both sides for the binary mixture with 1000 molecules.
For the unary systems, the width of the slap was 3x3x3 nm3 while the whole simulation box is of 3x3x12 nm3 with a total number
of molecules of 400. Our systems were equilibrated under pressure for 1ns using Berendsen algorithm40 at 1 bar, followed by 4ns
equilibration under a constant volume using Nose-Hoover thermostat41.The data were collected after 6 ns.The appropriate cutoff
radius of interaction was taken from our previous work on methanol42, which is 1.3 nm for OPLS-AA and 1.4 nm for TraPPE-UA
for the Leonard Jones potential and for the PME long range interaction.

2
Figure1. Simulation box of the water-methanol mixtures

While analysing the data to compute the diffusion coefficients of binary systems, we found that the EE method did not converge
toa straight line even when we increased the time of equilibration and time of collecting data to 30 ns. Hence, we were obliged to
use the Green-Kubo method to calculate the diffusion coefficient, which is valid for relatively short time ~100ps, but we had to
save the data at every step of 2 fs, which requires a very high hard disk space. In all of our simulations, periodic boundary
conditions were applied in all directions.
Potential Model
In this work, we chose the best two potential models of methanol based on our unpublished work; the Transferable Potentials for
Phase Equilibria unite-atom (TraPPE-UA) and the Optimized Potentials for Liquid Simulations all-atoms (OPLSS-AA) force
fields. Regarding the water potential models, we chose the TIP4P-water and the SPC/E-water modelsdue to their validity of
reproducing most of the real water properties. In TraPPE-UA model, the methanol molecules are represented by three interacting
sites located on the methyl group (CH3 ) and on the hydrogen and oxygen of the hydroxyl group (OH). In OPLS-AA model, each
atom in methanolis represented by one interacting site.In both OPLS-AA and TraPPE-UA force fields methanol is considered to be
rigid. Thetotal pair intermolecular potential between atomsi and j is the sum of Lennard-Jones (LJ) and Coulomb potentials
12 6
𝑞𝑖 𝑞𝑗 𝜎𝑖𝑗 𝜎𝑖𝑗
𝑉𝑖𝑗 (𝑟𝑖𝑗 ) = + 4𝜖𝑖𝑗 (( ) −( ) )
4𝜋𝜖0 𝑟𝑖𝑗 𝑟𝑖𝑗 𝑟𝑖𝑗

Where 𝜖𝑖𝑗 and 𝜎𝑖𝑗 are the Lennard-Jones (LJ) parameters for interaction between atoms of different types.
The Lorentz-Berthelot rules are used for the interaction between the unlikeatoms:𝜎𝑖𝑗 = (𝜎𝑖 + 𝜎𝑗 )/2 and 𝜖𝑖𝑗 = √𝜖𝑖 𝜖𝑗
Results
We have computeda number of thermodynamic, transport and structural properties of liquid methanol and its binary mixture with
water for two different potential models of methanol and of water.
1. Density

The most important quantity when one advising a new model for soft-matter is the vapor-liquid phase diagram, usually most
published model data start from room temperature up to estimated critical temperature. At low temperatures, the only rigorous

3
method of estimating the vapor density may be through scaled model43. We followed the usual method of estimating the liquid
density by inserting all the molecules in a slab within a larger empty box and fitted the profile density to a hyperbolic tangent
function after assigning zero to the vapor density

1 1 𝑧 − 𝑧0 (1)
𝜌(𝑧) = (𝜌𝑙 + 𝜌𝑣 ) − (𝜌𝑙 − 𝜌𝑣 )tanh⁡( )
2 2 𝑑
Where𝜌𝑙 , 𝜌𝑣 are the bulk densities of the liquid and vapour respectively, 𝑧0 is the place of the Gibbs dividing surface and 𝑑 is the
width of the interface. The density of the pure methanol for both the TraPPE-UA and OPLS-AA is shown in Figure 2.The results
indicate that the TraPPE-UA potential model shows better agreement with the experimental value 44 compared to the OPLS-AA.
This result is not surprising since TraPPE-UA was modeled originally to give the best vapor-liquid phase diagram.

Figure2.Thedensity (𝜌) of TraPPE-UA and OPLS-AA force fields of methanol as a function of temperature (T) compared to
experimental data

The very good performance of the TraPPE-UA potential model in predicting the density of pure system of methanol led us to
examine its accuracy and performance in its aqueous solution with water. Experimental results45and MD simulations of the total
and partial densitiesfrommethanol-water binary mixtureare displayed in Figure 3 including the densities of the pure systems which
are shown at the end points. It iswell-known that the density of methanol is lower than that of water; consequently, the density of
the mixtures must decrease as the mole fraction of methanol increases regardless of the model. Both the TraPPE-UA and OPLS-
AA potential models of methanol with the TIP4P-water give the trend in total density similar to the experimental value but the
TraPPE potential model shows a better agreement with the experimental results as shown in Figure 4. The total densities of the
mixture werealso compared with the relation of mixing rules, i.e., 𝜌(𝑥) = (1 − 𝑥)𝜌𝑙𝑤+ 𝑥𝜌𝑙𝑀 where 𝜌𝑙𝑤 ,⁡𝜌𝑙𝑀 are the densities of pure
liquid water and pure liquid methanol respectively,andxis the methanol mole fraction. From the results, one can deduce that it is not
necessary to estimate the total density as a function of mole fraction, but rather to estimate them from the mixing rules and the
calculated end points.

4
Figure3. Partial densities of methanol (OPLS_AA) and water(TIP4P) and total density compared to experimental data and to the
relation of mixing rules

Figure4. Partial densities of methanol (TraPPE-UA) and water(TIP4P) and total density compared to experimental data and to the
relation of mixing rules
At mole fraction of 0.3 for methanol, we were unable to estimate the liquid density of both water and methanol due to their peculiar
profile shapes.Refer to the right panel of figure 5. From the figure we can predict that the structure at 𝑥 = 0.1 is a core-shell (CS),
where the methanol profile has peaks (like horns) near the borders of the slap, but by increasing the mole fraction over 50%, the
profiles are smooth. One concludes that methanol is totally homogeneous inside the water and vice versa.The given structure is
called a well-mixed (WM) structure. Near 𝑥 = 0.3, the structure is not clear.For this reason, we did another experiment by placing
the molecules inside a sphere of radius of 4nm with simulation time of about 10 ns; then the density map was plotted to figure out
the structure of the mixture as shown in the left and middle panels of figure 5. From the left panel, it is clear that the structure of
methanol at 𝑥⁡ < ⁡0.3 is core-shell and for 𝑥 > 0.5 the structure is well-mixed, and at 𝑥 = 0.3 specifically, the structure is neither.
It is clear that both solutions start to segregate from each other; the new structure looks like Russian-doll (RS).

5
The number density contour map and the density profile of methanol around the symmetry axis (𝑧 = 0) were computed at different
mole fractions 𝑥𝑀𝐸𝑂𝐻 =10, 30, 50 and 90% of methanol at T=300K.

Figure5. Left and middle panels are the number density contour map of the methanol and water of OPLS-AA/TIP4P
mixtures respectively. The right panel is the density profile of the molecules inside the slap as a function of distance.From
top to down, the sub-figures are at mole fraction of methanol of 0.1, 0.3, 0.5, and 0.7 respectively.

6
From Figure5(a-d) we found that the density profiles at all mole fractions of methanol seems to be approximately symmetric about
the midpoint 𝐿𝑧 /2.This gives an indication that the mixture (i.e. inhomogeneous system) is properly equilibrated. We also observed
that the preferred structure at low mole fractions of methanol is the core-shell; this structure is converted to a well-mixed structure
by increasing the mole fraction of methanol.At𝑥𝑀𝐸𝑂𝐻 =0.3, atthe first sight the distribution is somewhatsimilar to the phase
separated structure (RDS).This led us to examine whether this structure is regular or not at different temperatures. In the current
work, we chose different temperatures 210, 240, 270 and 300 and checked the distribution of methanol into water throughout
density contour map plot at 𝑥𝑀𝐸𝑂𝐻 =0.3, 0.5, 0.7, and 0.9. We found that the RDS structure appearsagain at T=270K at 𝑥𝑀𝐸𝑂𝐻 =0.3
and0.7. This confirms that this structuredid not regularly happen at all temperaturesand mole fractions of methanol. Figure 6
illustrates this unusual RDS structure at the latter concentrations of methanol at T=270K. In our forthcoming papers in binary
mixtures of ethanol and propanol with water using the same force fields studied in this paper, we will show the effect of carbene
(CH2) on the structure of mixtures as a function of mole fraction and as a function of temperature. The Russian-Doll structure was
clearly demonstrated in recent work of Hrahsheh et al46 on binary pentanol-water mixture. More details about the distribution of
different type of molecules in the mixture are discussed in the supplementary reading.

Figure 6. Left and right panels are the number density contour map of the methanol and water of OPLS-AA/TIP4P mixtures
respectively at T=270.

Surface Tension
Surface tension isthethermodynamic property, which plays an essential role in many daily and industrial aspects. Due to high
fluctuation of the pressure, we ought to equilibrate our systems for a long time, sometimes for 10 ns, and then collect the data for
another 10 ns. The surface tension has been calculated using the following relation:
1 ∞ (2)
𝐿𝑧 𝑃𝑥𝑥 + 𝑃𝑦𝑦 𝑟𝑠 3𝑠 3 − 𝑠
𝛾 = (𝑃𝑧𝑧 − ) + 12𝜋𝜖𝜎 6 (𝜌𝑙 − 𝜌𝑣 )2 ∫ 𝑑𝑠 ∫ coth ( ) ⁡𝑑𝑠
2 2 𝑑 𝑟3
0 𝑟𝑐

7
Where 𝑃𝛼𝛼 is the 𝛼𝛼 component of the pressure tensor, 𝐿𝑧 is the box length on 𝑧 direction, 𝜖 and 𝜎 are the Lennard-Jones
parameters, and 𝑟𝑐 is the cutoff radius.

The effect of temperature on the surface tension was examined for the unary system of methanol using TraPPE-UA and OPLS-AA
potential models. It is clear from Figure 7 that the surface tension of methanol decreases with temperature in both models, which
agrees with the results of other authors42, 43.From Figure 7, we notice that the TraPPE-UA model showsa better agreement with the
experiment at relatively high temperatures; whereas OPLS-AA model showsa better agreement with the experimental results at
very low temperatures.

Figure 7.Temperature dependence of the surface tension liquid methanol. The MD results are shown for TraPPE-UA(▼) and
OPLS-AA (■) force fields compered to experimental data(●)47

The surface tension was also calculated at different mole fractions of methanol from 𝑥𝑀𝐸𝑂𝐻 =0.02 to 0.1 by step of 0.02 followed
by step of 0.1 up to 𝑥 = 1.The comparison between TraPPE-UA and OPLS-AA force fields in reproducing the surface tension in
the binary mixture is shown in Figure 8.

Figure 8.Surface tension of methanol-Water Mixture at different mole fraction of methanol. The MD results are shown for
TraPPE-UA (▼) and OPLS-AA (■) force fields compared to experimental data (●)48

8
From the figure, it is evident that both theTraPPE-UA and OPLS-AA potential models fit the experimental data very well when
they are mixed with TIP4P-water, but the TraPPE-UA model showsa better overall agreement especially at a very low
concentration of methanol. It is worth mentioning that the fast drop in the surface tension when increasing the methanol
composition from 0 to 20% is mostly due to the methyl group (CH3). The contour maps density of methanol in the mixture shows
that the structure is a core-shell, where most of the molecules on the outer-shell are mainly methanol. On the other hand, theatoms
distribution inside the methanol at the surface shows that the hydroxyl group directs inward the surface, whereas the methyl group
directs outward the surface. Even though by increasing the methanol concentration the structure changes to a well-mixed, but the
outer most shell of the sphere mainly consists of methanol, please refer to the supplementary reading.

Diffusion Coefficient
It is well-known that the diffusion coefficient is a sensitive quantity to the size of the box and time of the simulated system. In our
systems, the diffusion coefficients have been strongly affected by the period of simulation time. In principle, the Einstein relation
and the Green-Kubo methods should produce the same results. As we will see later, even though the Einstein relation was perfect
in reproducing the diffusion coefficients in the unary systems, it fails badly to estimate the coefficients for the binary systems even
with simulation time up to 30 ns.

Einstein equation (EE)


The self-diffusion coefficients were calculated from the time dependence of the mean square displacement (MSD) using Einstein
equation (EE)
𝑀𝑆𝐷 = 6𝐷|< 𝑟(𝑡)2 − 𝑟(0)2 >| (3)
The Einstein equation (EE) has been used to compute the diffusion coefficient in both the unary and binary mixture of methanol
with water. For the pure methanol the self-diffusion coefficients werecalculated in the temperature range from 200 to 300 K.All
mean square displacement versus time plotswere linear; this means that our results for diffusion coefficient converge and the time
of simulation is sufficient.The MDS for the two potential models were plotted versus |< 𝑟(𝑡)2 − 𝑟(0)2 >|which yields a straight
line with the slope equals to 6D. Figure 9 shows the diffusion coefficients as a function of temperature.It is clear from the figure
that the TraPPE-UA and OPLS-AA potential models follow the experimental trends.Both models predict higher values than the
experimental results, but the TraPPE-UA model shows a better agreement over a wide range of temperatures up to 280K. The
above conclusion is confirmed when estimating the activation energy of diffusion through Arrhenius plot. Figure 10 shows
𝐸

Arrhenius plot of diffusion by assuming that the diffusion is proportional to ~𝑒 𝑘𝐵 𝑇 where 𝐸 is the activation energy of diffusion.It
is clear that the TraPPE-UA model not only predicts the experimental data, but also gives the closest activation energy of diffusion.

9
Figure9. Temperature dependence of the self-diffusion coefficient of pure liquid methanol. The MD results are shown for TraPPE-
UA(▼) and OPLS-AA (■) force fields compered to experimental data (●)49

Figure 10. Arrhenius plot for the logarithm of self-diffusion (D) as a function of temperature (T) inverse compared to experimental
data (●).The MD results are shown for TraPPE-UA (▼) and OPLS-AA (■) force fields

Regarding the diffusion coefficients of species in binary mixtures, theEinstein equation fails to reproduce the experimental data.
Figure 11 shows the variation of MDS as a function of time up to 6ns.From the figure, we see that the MSD is linear, but in three
different regions, the slope changes from relatively high value to lower value then back again to higher value. The data should be
collected from the last region since EE is valid for a long time.For this reason, we increased the time to 20ns, but the linearity of
MSD as a function of time did not converge to a one linear function. After collecting the data, the diffusion is plotted as a function
of mole fraction of methanol(Figure 12), and the results were completely random.

10
Figure11.Mean square value(MSD) for (OPLS-AA)/TIP4P at 𝑥𝑚 =30%

Figure12.Diffusion coefficient of methanol/water mixtures as a function of methanol mole fraction compered to experimental
results (●)2

Instead of repeating all our MD simulations over a long time (maybe be > 50ns) to get a sensible value of diffusion coefficients, we
used Green-Kubo (GK) method which requires a short simulation time. Even though the simulation time required to calculate the
diffusion coefficients using GK is short, which is an advantage of applying GK, we ought to save every frame at 2fs up to 100 ps
which requires a very large disk space.

Green-Kubo (GK) Method


The Green-Kubo formula (GK) is the method of choice for calculating the diffusion coefficientin short time simulations, which
isgiven by:

11
1 ∞ (4)
𝐷𝑖 = ∫ 𝑑𝑡 〈𝑉𝑘 (t). 𝑉𝑘 (0)〉
3𝑁𝑖 0

Where𝑉𝑘 (t) is the center of mass velocity vector of molecules at some time t, 𝑉𝑘 (0) is the same but at time 𝑡 = 0,〈… . . 〉 denotes
the ensemble average, and 𝑁𝑖 is the total molecules of type 𝑖. To calculate the diffusion coefficients we applied the criteria of
Zlenko50. Figure 13 shows the self-diffusion coefficient of methanol and water. It is evident that the MD simulated diffusion
coefficient produced from the OPLS-AA/SPCE-water adequately fits the experimental data over the entire range of composition.
This indicates that combining methanol with SPCE-water is more valid in assessing methanol diffusion coefficients.

Figure13.Self-diffusion coefficients of methanol with TIP4P-water (left panel) and with SPCE-water (right panel) compared to
experimental data

Radial Distribution Function (RDF)


We now proceed to investigate how the molecular structuralfeatures of the puremethanolchanges as the temperature increases for
both models throughout the position of the first peak, the coordination number, the structure function and the distinct radial
distribution function. All the radial distribution functions RDFs(i.e.𝑔𝑜𝑜 (𝑟),𝑔𝑜𝐻 (r)𝑔𝐶𝑂 (r),𝑔𝐶𝐶 (r), ,𝑔𝐻𝐶 (r), ,𝑔𝐻𝐻 (r))were calculated for
the OPLS-AA and TraPPE-UA potential models. Figures 13 and 14 show the radial distribution function of O-O for TraPPE-UA
12
and OPLS-AA respectively. From the figures we see that the position of the first maximum does not change with temperature and
it is about 2.7𝐴° as the experiment proposes51 and the height of the first maximum peak decreases as the temperature increases.This
behaviour is expected since by decreasing the temperature, the molecules will come closer to each other, and the network becomes
more structured, and as a consequence the coordination number will increase as well.The first peak 𝑔𝑜𝑜 (𝑟)shows a reduction in
magnitude functions as the temperature increases as a result of breaking the hydrogen bonds in the liquid methanol for both
potential models considered for methanol.The coordination number is of about 2 at 300K and it increases slowly as the temperature
decreases. But generally speaking the value is slightly lower than 2, which suggests that the network of the methanol even at low
temperature is a chain network. The distinct radial distribution function will be explained later when we discuss the mixture of
methanol-water, but most of the PDF contributions come from 𝑔𝑂𝑂 and 𝑔𝐶𝐶 as will be explained in the following paragraphs. For
the other radial distribution functions, please refer to the supplementary reading of this manuscript. We noticed from the figures
that the TraPPE-UA potential is more structured, and this is not unexpected since the force field contains three atoms, while the
OPLS-AA has 6 atoms.

Figure 14. Radial distribution functions for (O-O) liquid phase methanol TraPPE-UAas function of temperature

13
Figure 15. Radial distribution functions for (O-O) liquid phase methanol OPLS-AA as function of temperature

Figure 17 shows the evolution of the first peak with the temperature increase.The position of the first peak in the OPLS_AA and
TraPPE-UA models is nearly around (2.6-2.9)𝐴° . From the figure we notice that the height of the first peak changes linearly with
temperature, but the variation becomes high for the TraPPE-UA force fields, starting from about a value of 5.3 at 𝑇 = 200𝐾 to a
value of 4 at 𝑇 = 300, while the value of the peak in OPLS-AA changes from 3 to 2.5 as the temperature changes from 200 to
300K.

Figure 16.First peak in the Oxygen-Oxygen radial distribution function obtained from MD results forTraPPE-UA (▼)and OPLS-
AA (●)force fields

In the light of the above results, we can say that the 𝑔𝑜𝑜 (𝑟)is more structured in the TraPPE-UA models as the temperature
decrease(i.e. the interaction between methanol molecules is stronger in this potential model).
For the binary mixtures, it is worth showing the RDF of O-O of all force fields used in this work at 300K to understand what is
going on when increasing the concentration of methanol. Figure 17 shows such behavior at 𝑥 = 0 (pure SPCE and TIP4P-water)
and at 𝑥 = 1 (pure methanol of TraPPE-UA and OPLS-AA).The corresponding coordination numbers will be shown later, but it is
about 4 for water which is tetrahedral as suggested in the literature.

14
Figure 17.RDF of oxygen-oxygen of all models (top for water models and bottom for methanol models)
The corresponding distinct RDF is shown in Figure 18; on the other hand, the distinct RDF is calculated as follows:

1 sin(𝑘𝑟) (5)
𝐺𝑑 (𝑟) = 1 + 2
∫ 𝑘 2 𝐻𝑑 (𝑘) 𝑑𝑘
2𝜋 𝜌𝑚 0 𝑘𝑟

Where 𝜌𝑚 is the molecular number density and 𝐻𝑑 (𝑘) is the intermolecular contribution as a function of wavenumber 𝑘 which is
given as:
∑𝛼 ∑𝛽 𝑓𝛼 (𝑘)𝑓𝛽 (𝑘) ℎ̂𝛼𝛽 (𝑘) (6)
𝐻(𝑘) = 2
[∑𝛼 𝑓𝛼 (𝑘)]

Where 𝑓𝛼 is the X-ray form factor approximated to be equal to the total number of electrons in each group, which is a valid

approximation for low values of 𝑘, and ℎ̂𝛼𝛽 is defined as:

∞ sin⁡(𝑘𝑟)
ℎ̂𝛼𝛽 (𝑘) = 4𝜋𝜌𝑚 ∫0 𝑟 2 (𝑔𝛼𝛽 (𝑟) − 1) 𝑑𝑟 (7)
𝑘𝑟

We see that both potentials reproduce the experimental curve for water models. For methanol models, both models reproduce the
experimental data but with advantage for the OPLS-AA regarding the first minima, while the TraPPE-UA works better for
predicting first maxima. We notice from the figure that the first peak occurs about 2.6A, which means that the O-O PDF is
responsible for the appearance of the first peak, while the C-C RDF is responsible for the second and the third peak; see the
positions of maximum points of the 𝑔𝐶𝐶 in the supplementary reading.Even though all models reproduce the experimental data
quite well, the small deviation from experimental data might be due to the approximating the form factor to be the same number as
that of electrons in the atom. Even though such approximation is valid only for small values of 𝑘, and in calculating the distinct
15
RDF, the integration is taken over the whole k-space. The effect of this approximation will be shown later when we discuss in
details the total structure factor.

Figure 18. The distinct RDF of water models (top) and methanol models (bottom) as a function of distance compared to
experimental data7
Figure 19 shows the oxygen-oxygen PDF of binary mixtures at a methanol mole fraction of 0.6.It is clear that when the oxygen
water is involved, the results do not depend on the model chosen for methanol since the RDF of pure water is almost the same for
both SPCE and TIP4P-water. For O-O PDF of methanol, we notice that the first maximum peak depends greatly on the model
chosen, and the first maximum peak is higher for TraPPE-UA than the OPLS-AA as for pure systems, but the height decreases
sharply by increasing water concentration.

16
Figure 19. RDF of oxygen-oxygen as a function of distance, the left top is the O-O RDF of methanol, while the right top and
bottom graphs when water oxygen is involved

Figure 20 shows the variation of oxygen-oxygen RDF functions as the mole fraction of methanol increases.
The figure shows only the OPLS-AA methanol mixed with TIP4P-water (left panel), and mixed with SPCE-water (right panel).For
TraPPE-UA, please refer to the supplementary reading. These figures are important to calculate the coordination number which
might be thought as the number of hydrogen bonds which will be discussed in detail later. As an example regarding the
coordination number, if the methanol behaves as solute the OM-OW at 𝑥 = 0.2 indicates that the hydroxyl oxygen has 2.6 water
molecules within a 3.34𝐴° radius, and the carbon is surrounded by 14 water molecule as can be figured out from the CM-OW. On
the other hand if the methanol behaves as a solvent (𝑥 = 0.8 as an example), the OM-OW indicates that the hydroxyl oxygen has 1
water molecules within a (min number) radius, and the carbon (from C M-OW) is surrounded by (number) water molecules within
radius of3.48𝐴° . It is also clear that by increasing the methanol concentration, the first maximum grows substantially while the
second maximum remains almost unchanged. One might notice that by adding more methanol, the first minimum becomes lower,
which is an indication that the first shell is separated from the second shell.

17
Figure 20. RDF of oxygen-oxygen as a function of distance at different methanol mole fraction, the left-half figure shows the RDF
of OPLS-AA mixed with SPCE-water while right-half figure shows the OPLS-AA mixed with TIP4P-water

Figure 21 shows how the maximum value of the first peak of oxygen-oxygen RDF changes with mole fraction of methanol. It can
be seen that the relation between the value of the first maxima of the O-O RDF is linearly proportional to the increase of methanol
concentration up to x=0.7 regardless of the type of oxygen.This does not necessarily mean the coordination number should
increase as we will see later.

Figure 21. Maximum value of the oxygen-oxygen RDF as a function of mole fraction of methanol for the two methanol
modelsmixed with TIP4P-water

Figure 22 shows the corresponding distinct RDF of the above figures but with TraPPE-UA methanol compared to experimental
data at the end points (𝑥 = 0 or 1). The figure shows how the structure of water (𝑥 = 0) changes slowly by increasing the
concentration of methanol till it reaches the pure structure of methanol (𝑥 = 1). By increasing 𝑥, the second and third peaks start to
appear, and this is a clear indication that the second and third peaks come from the C-C RDF.

18
Figure 22. The Distinct RDF of the mixtures, left panel for TIP4P-water mixed with the two methanol models while the right panel
when SPCE-water replaces TIP4P-water

The total structure factor S(K) obtained from the MD simulation for the Mixture of methanol-water at 𝑇 = 300𝐾is shown in
Figure 23 and compared with the structure factors obtained fromX-ray and neutron diffraction 7. The structural change is obvious
in the mixture as the mole fraction of methanol increases. From the figure we notice that the 𝑆(𝐾) for pure water is almost the
same as the experimental data beyond the third maximum, and the positions of the first three maxima are predicted well by the
models. As the methanol concentration increases, the structure function coincides better regarding the first maxima, but the second
maxima deviate to a lower value till 𝑥 = 0.61. We also see that the phase of the third maxima starts to deviate outward. This means
that the position of the C-C RDF also deviates to the right compared with the experimental data. The most awkward result is at 𝑥 =
0.8. We notice that all mixed models almost coincide with each other, but the height of the first maxima is higher than the
experimental data, and the second maximum is lower. At 𝑥 = 1 for pure methanol, the results are almost perfect when compared to
experimental data, except for the phase shift of the first minima. The total scattering structure function of experimental pure water
shows a double-peak first maximum, which is something SPCE and TIP4P-water could not recover in contrast to results of Galicia-
Andres’ et al7. Upon adding methanol, the double-peak first maximum occurred at 𝑥 = 0.2, and by increasing the concentration of
methanol, the first maxima increase slowly and the second maxima decrease till it becomes a shoulder. Same behavior also
occurred regarding the fourth maxima. For the pure methanol, the TraPPE-UA shows a better agreement over the OPLS-AA
regarding the appearance of the shoulders, but both models show a shift in the first minima inward. The changes in the total
scattering structure function is due to the change of the network of molecules from tetrahedral for pure water to short chain-like for
methanol passing through cross-linked network at 𝑥~0.7 at which the simulated results show the biggest discrepancy with
experimental data. The main reason for not recovering the results of Galicia-Andres’ et al is due to the X-ray form factor. In our

19
calculations we approximated the form factor to be equals to the number of atoms in each group, which is valid, as explained
earlier, for low values of wavenumber (𝑘); while the form factor in Galicia-Andres’ et al work was calculated using a sum of five
Gaussians with coefficients taken from ref52 as:
(𝛼) (𝛼)
𝑓𝛼 (𝑘) = ∑5𝑖=1 𝑎𝑖 exp(−𝑏𝑖 𝑘 2 ) + 𝑐_𝛼 (8)

Figure 23. The total structure factor of water/methanol mixtures of all mixed-models at methanol mole fraction of 0, 0.2, 0.4, 0.6,
0.8, and 1.
Coordination number
The complexity in the structure of methanol-water mixture isdue to the fact that the hydrophilic head in methanol
structure.Thismakes it capable of forming hydrogen bonds with itself and with other polar molecules like water.The number of
hydrogen bonds can be approximated through the coordination number, which is defined as the area under the RDF curve up to the
first minimum of the 𝑔𝑂𝑂 , as:
𝑟 2
𝑛𝛼𝛽 (𝑟) = 4𝜋𝜌 ∫0 min 𝑟 ′ 𝑔𝛼𝛽 (𝑟 ′ )𝑑𝑟 ′ (9)
Where ρ is the number density andthe integration starts from 𝑟 = 0 and ends at first minimum of the site RDF 𝑔𝛼𝛽 (𝑟 ′ ) .The studied
coordination numbers are(OW-OW), (OM-OM),(OW-OM) and (OM-OW), where the OW and OM correspond to the water
oxygen and methanol oxygen respectively.
Figure 24 shows the number of hydrogen bonds between the OW-OW and OM-OM as a function of mole fraction of methanol. Figure
24(a) shows the effect of water type using OPLS-AA for methanol, while Figure 24(b) shows the effect of methanol type when
mixed with SPCE-water. From the figures we noticed that the type of water or methanol does not change the hydrogen bond, and
that the variation of hydrogen bonds with increasing the mole fraction of methanol is linear. One might deduce that the network
structure is tetrahedral when 𝑥 = 0 (pure water) and the network structure starts to change slowly by increasing x till we get a
chain network when 𝑥 = 1 (pure methanol). Even though the total network structure depends highly on the number of hydrogen

20
bonds, one might deduce from the figures by ignoring the mixing hydrogen bonds that the network structure at 𝑥~0.7 is cross-
linked (𝑛𝐻𝐵 = 3), where each type of molecules share roughly the same contribution of 1.5 coordinated molecules.

Figure 24. Number of hydrogen bonds per molecule as a function of methanol mole fraction, (a) H-bonds per molecule of OPLS-
AA with water models, (b) H-bonds per molecule of SPCE-water with methanol models

Figure 25 shows all mixing hydrogen bonds as a function of methanol mole fraction.We notice that both coordination numbers
calculated from OW-OW and OM-OW RDF’s decrease as 𝑥 increases.While for OW-OM and OM-OM, the hydrogen bonds increase as
𝑥⁡increases. Since the variation is linear, one might deduce that the relation between the number of hydrogen bonds of O W-OW and
𝑡𝑆 𝑡𝑆
OM-OM follows the mixing rules as 𝑛𝐻𝐵 = 𝑥𝑛𝑀 + (1 − 𝑥)𝑛𝑊 , where 𝑛𝐻𝐵 is the total number of hydrogen bonds of the same type
(MM or WW), 𝑛𝑀 is the number of methanol hydrogen bonds and 𝑛𝑊 is the number of water hydrogen bonds, see Figure 24(b).
𝑡𝑀 ′
The same is true for the total mixing hydrogen bonds 𝑛𝐻𝐵 which can be written in terms of water-methanol hydrogen bonds 𝑛𝑊
′ 𝑡𝑀 ′ ′
and methanol-water hydrogen bonds 𝑛𝑀 as 𝑛𝐻𝐵 = 𝑥𝑛𝑊 + (1 − 𝑥)𝑛𝑀 . One might notice that for pure water, the molecules form
four hydrogen bonds. Adding methanol i.e., starting from left to right, the four coordinated molecules decrease linearly with
increasing the concentration of methanol, while the two and three coordinated molecules increase with increasing 𝑥𝑀𝐸𝑂𝐻 . If one
starts from right to left, i.e., adding water to methanol, number of hydrogen bonds decreases linearly, and at the same time increase
in the three coordinated molecules species (sees OM-OW).

Figure 25. Number of H-bonds per molecule as a function of mole fraction of OPLS-AA/SPCE-water mixtures

21
Conclusion
A series of molecular dynamic simulationsfor pure liquid methanol and its mixture with water have been performed using two
different Methanol models (TraPPE-UA and OPLS-AA) with two different water (SPCE and TIP4P) potential models. The results
are compared with the experimental thermodynamic, dynamical and structural data in order to examine the validity of presented
potential models in producing the structural, dynamical, and thermo-dynamical properties in both the unary and binary mixture of
methanol-water. The density and surface tension have been computed in order to reveal the performance of the above mentioned
models in the pure and mixed methanol with water.
The overall conclusion is that the TraPPE-UA potential model works well in reproducing the surface tension, density and diffusion
in both the unary and binary mixture of methanol with water.Unfortunately, the results of diffusion based on Einstein equation
(EE) randomly fluctuated over an entire range of methanol mole fractions regardless of potential model we used. The results
obtained by Green-Kubo (GK) method give the trend in diffusion coefficient similar to the experimental results in all systems(i.e.
the unary and mixed system).Furthermore,the OPLS-AA model of methanol combined with SPCE wateris more valid in producing
methanol diffusion coefficient.
Finally, it is worth mentioning that the validity of potential model in predicting the desired properties in unary system does not
mean that the model is valid in predicting the same propertiesin the binary mixture.

References

1. F. Biscay, A. Ghoufi and P. Malfreyt, The Journal of chemical physics 134 (4), 044709 (2011).
2. Z. Derlacki, A. Easteal, A. Edge, L. Woolf and Z. Roksandic, The Journal of Physical Chemistry 89 (24), 5318-
5322 (1985).
3. S. Dixit, A. Soper, J. L. Finney and J. Crain, EPL (Europhysics Letters) 59 (3), 377 (2002).
4. A. J. Easteal and L. A. Woolf, The Journal of Physical Chemistry 89 (7), 1066-1069 (1985).
5. L. C. G. Freitas, Journal of Molecular Structure: THEOCHEM 282 (1-2), 151-158 (1993).
6. E. Galicia-Andrés, H. Dominguez, L. Pusztai and O. Pizio, Journal of Molecular Liquids 212, 70-78 (2015).
7. E. Galicia-Andrés, L. Pusztai, L. Temleitner and O. Pizio, Journal of Molecular Liquids 209, 586-595 (2015).
8. O. Gereben and L. s. Pusztai, The Journal of Physical Chemistry B 119 (7), 3070-3084 (2015).
9. X. Gong, A. Bandis, A. Tao, G. Meresi, Y. Wang, P. Inglefield, A. Jones and W.-Y. Wen, Polymer 42 (15), 6485-
6492 (2001).
10. G. Guevara-Carrion, J. Vrabec and H. Hasse, The Journal of chemical physics 134 (7), 074508 (2011).
11. J.-H. Guo, Y. Luo, A. Augustsson, S. Kashtanov, J.-E. Rubensson, D. K. Shuh, H. Ågren and J. Nordgren, Physical
review letters 91 (15), 157401 (2003).
12. J. Kida and H. Uedaira, Journal of Magnetic Resonance (1969) 27 (2), 253-259 (1977).
13. C. A. Koh, H. Tanaka, J. M. Walsh, K. E. Gubbins and J. A. Zollweg, Fluid phase equilibria 83, 51-58 (1993).
14. M. A. Matthews and A. Akgerman, Journal of Chemical and Engineering Data 33 (2), 122-123 (1988).
15. S. Y. Noskov, G. Lamoureux and B. Roux, The Journal of Physical Chemistry B 109 (14), 6705-6713 (2005).
16. G. Pálinkás, I. Bakó, K. Heinzinger and P. Bopp, Molecular Physics 73 (4), 897-915 (1991).
17. M. S. Skaf and B. M. Ladanyi, The Journal of Physical Chemistry 100 (46), 18258-18268 (1996).
18. J.-C. Soetens and P. A. Bopp, The Journal of Physical Chemistry B 119 (27), 8593-8599 (2015).
19. T. Takamuku, K. Saisho, S. Nozawa and T. Yamaguchi, Journal of Molecular Liquids 119 (1), 133-146 (2005).
20. T. Takamuku, T. Yamaguchia, M. Asato, M. Matsumoto and N. Nishi, Zeitschrift für Naturforschung A 55 (5),
513-525 (2000).
21. D. S. Venables and C. A. Schmuttenmaer, The Journal of chemical physics 113 (24), 11222-11236 (2000).

22
22. E. J. Wensink, A. C. Hoffmann, P. J. van Maaren and D. van der Spoel, The Journal of chemical physics 119
(14), 7308-7317 (2003).
23. H. Yu, D. P. Geerke, H. Liu and W. F. van Gunsteren, Journal of computational chemistry 27 (13), 1494-1504
(2006).
24. N. Zhang, Z. Shen, C. Chen, G. He and C. Hao, Journal of Molecular Liquids 203, 90-97 (2015).
25. Y. Zhong, G. L. Warren and S. Patel, Journal of computational chemistry 29 (7), 1142-1152 (2008).
26. I. Bakó, L. Pusztai and L. Temleitner, Scientific Reports 7 (2017).
27. H. Ghahremani, et al., Chem. Sin 2 (6), 212-221 (2011).
28. D. a. W. G. C. Ballal, The Journal of chemical physics 139 (11), 114706 (2013).
29. L. Bianchi, A. Adya, O. Kalugin and C. Wormald, Journal of Physics: Condensed Matter 11 (47), 9151 (1999).
30. L. Bianchi, O. Kalugin, A. Adya and C. Wormald, Molecular Simulation 25 (5), 321-338 (2000).
31. M. Haughney, M. Ferrario and I. R. McDonald, Journal of Physical Chemistry 91 (19), 4934-4940 (1987).
32. G. Palinkas, E. Hawlicka and K. Heinzinger, Journal of Physical Chemistry 91 (16), 4334-4341 (1987).
33. M. E. van Leeuwen and B. Smit, The Journal of Physical Chemistry 99 (7), 1831-1833 (1995).
34. W. L. Jorgensen, D. S. Maxwell and J. Tirado-Rives, J. Am. Chem. Soc 118 (45), 11225-11236 (1996).
35. K. Kahn and T. C. Bruice, Journal of computational chemistry 23 (10), 977-996 (2002).
36. B. Chen, J. J. Potoff and J. I. Siepmann, The Journal of Physical Chemistry B 105 (15), 3093-3104 (2001).
37. M. S. Green, The Journal of Chemical Physics 22, 398 (1954).
38. R. Kubo, Journal of the Physical Society of Japan 12 (6), 570-586 (1957).
39. H. J. Berendsen, D. van der Spoel and R. van Drunen, Computer Physics Communications 91 (1-3), 43-56
(1995).
40. H. J. Berendsen, J. v. Postma, W. F. van Gunsteren, A. DiNola and J. Haak, The Journal of chemical physics 81
(8), 3684-3690 (1984).
41. S. Nosé, The Journal of chemical physics 81 (1), 511-519 (1984).
42. A. Obeidat, A. Jaradat, B. Hamdan and H. Abu-Ghazleh, Physica A: Statistical Mechanics and its Applications
(2018).
43. A. Obeidat, unpublished (2018).
44. J. Jonas and J. Akai, The Journal of chemical physics 66 (11), 4946-4950 (1977).
45. S. Mikhail and W. Kimel, Journal of Chemical and Engineering Data 6 (4), 533-537 (1961).
46. F. Hrahsheh, Y. S. Wudil and G. Wilemski, Physical Chemistry Chemical Physics 19 (39), 26839-26845 (2017).
47. J. J. Jasper, Journal of physical and chemical reference data 1 (4), 841-1010 (1972).
48. H. Ghahremani, A. Moradi, J. Abedini-Torghabeh and S. Hassani, Der Chemica Sinica 2 (6), 212-221 (2011).
49. N. Farhadian and M. Shariaty-Niassar, Iranian Journal of Chemical Engineering 6 (4), 63 (2009).
50. D. Zlenko, Biophysics 57 (2), 127-132 (2012).
51. A. Narten and A. Habenschuss, The Journal of chemical physics 80 (7), 3387-3391 (1984).
52. D. Waasmaier and A. Kirfel, Acta Crystallographica Section A: Foundations of Crystallography 51 (3), 416-431
(1995).

23

Vous aimerez peut-être aussi