See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/231683808

Application of molecular simulation to derive

phase diagrams of binary mixtures

Article in Macromolecules · July 1992

DOI: 10.1021/ma00040a010

CITATIONS READS

144 133

4 authors, including:

Barry Olafson Mario Blanco

California Institute of Technology NanoPEC
17 PUBLICATIONS 13,638 CITATIONS 68 PUBLICATIONS 2,281 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Nono-porous Anodic Aluminum Ceramic Membranes View project

Multiscale modeling of Polymeric membranes View project

All content following this page was uploaded by Mario Blanco on 08 May 2015.

The user has requested enhancement of the downloaded file.

 Macromolecules 1992,25, 3667-3676 3667

Application of Molecular Simulation To Derive Phase Diagrams of

Binary Mixtures
Cun Feng Fan,' Barry D. Olafson, and Mario Blanco
Molecular Simulations Inc., 796 North Pastoria Avenue, Sunnyvale, California 94086
Shaw Ling Hsu
Polymer Science and Engineering Department, University of Massachusetts,
Amherst, Massachusetts 01003
Received January 24,1992; Revised Manuscript Received April 2, 1992

ABSTRACT The miscibility behavior of three binary mixtures, solvent with solvent, polymer with solvent,
and polymer with polymer, was studied by use of a combinationof the Flory-Huggins theory and molecular
simulation techniques. Fundamental parameters in the Flory-Huggins theory, including the heat of mixing
associatedwith pairwise interactions(Aw12)and the number of possible interactionpartners, i.e., coordination
number, z, are calculated from molecular simulations. The pair energies ( ~ 1 1 ,w22, ~ 1 2 )are obtained by
averaging a large number of c o n f i a t i o n sgenerated by a Monte Carlo approach which includes the constraints
associated with excluded volume. The temperature dependence of the interaction parameter x is obtained
with the formalism developed in this study. In all cases, the calculated upper critical solution temperatures
compare favorably with experimental values. This approach provides an opportunity to test the Flory-
Huggins theory for a number of model binary systems and to characterize their miscibility behavior. This
combined approach also facilitatesstudy of the thermodynamicbehavior of a binary mixture without possessing
specific knowledge or experimental data of the system under investigation.

Introduction interaction energy terms required for the thermodynamic

The miscibility behavior of binary mixtures, both small expressions.
molecular systems and macromolecular systems, has Miscibility behavior can best be illustrated by calculating
proven to be of interest from theoretical and practical the free energy of mixing as a function of composition at
points of view. Some important applications directly different temperatures. A schematic drawing of such a
related to this subject include selection of solvent for a
particular solute, definition of solvent mixing behavior, calculation for a system exhibiting the upper critical
inclusion of plasticizers in polymers, definition of polymer solution temperature (UCST) is shown in Figure 1as an
dissolution and swellingprocesses, and derivation of phase example. At temperatures above the critical temperature
diagrams for molecular and polymeric alloys. Although T,,the AG (Gibbsfree energy of mixing) versus 42 (volume
the solution thermodynamicsand phase equilibrium theory fraction of second component) has only one minimum.
have been well established to perform such calculations Two components are miscible for any composition. For
for binary systems, difficulties are often encountered temperatures below T,,, the two points in contact with the
during actual implementation due to the absence of straight line define two binodal points, A and D (for a
detailed information regarding each component. Many symmetric case there are two minima with dAG/a4 = 0).
parameters characteristic of a binary mixture are thus Two inflection points with a2AG/a42 = 0 define spinodal
obtained by fitting a theoretical model with some exper- points B and C. In the region of 42 < 4~ and 42 > 4~ two
imental data. Prediction of the thermodynamicbehavior components are miscible. In the region between 4~ and
for a system which is not well-known remains a difficult 4c, the system is unstable, separating into two phases with
task.
compositions equal to 4~ and 4 ~ When . $2 is between 4~
This situation has recently been improved by advances
in the area of atomistic and molecular simulation tech- and 4~ or 4~ and 4 ~the , system is metastable; Le., the
niques using powerful workstations. Accurate force fields system is stable for small fluctuations in composition and
can usually be established by defining parameters in in- becomes unstable for larger fluctuations, phase separating
teratomic potentials using structural and spectral data. by the spinodal mechanism into two phases with com-
Molecular simulations can then be accomplished for a positions 4~ and $D. At the critical temperature, T,,,A-
variety of systems, including well-characterized ones in D merge into a single point defined by azAG/aq? = a3AGl
order to gain a better fundamentalunderstanding of atomic ad3 = 0. By connecting the binodal points at different
level interactions and statistics and ill-characterized temperatures, the coexistence curve designated by the
systems in order to predict useful physical properties. dotted line shown in Figure 1can be obtained. This curve,
Many factors govern mixing processes, including the also known as the phase diagram, describes the phase
chemical nature of individual componentsas well as tem- change as a function of temperature and composition.
perature. For polymers, additional factors may include
chain packing, the degree of crystallinity, molecular weight, The actual calculation of the AG of mixing depends on
and chain flexibility. Although all these factors cannot the specific theory or model used. The Flory-Huggins
be addressed by the molecular simulation technique, it is lattice theory is probably the simplest and most widely
possible in selected cases to obtain structural data for used theory to calculate the free energy of mixing, AG,
individual components in the mixture as well as the and to construct the phase diagram of a binary system.'
A general expression for the AG of a binary system, be it
* To whom correspondence should be addressed. solvent-solvent, solvent-polymer, or polymer-polymer,
0 1992 American Chemical Society
3668 Fan et al. Macromolecules, Vol. 25, No. 14, 1992
the single most important and essential parameter gov-
erning the accuracyof many simulationtechniques. These
include energy minimization using molecular mechanics,
molecular dynamics, and Monte Carlo approaches.’ The
energy of a system, which can be a single molecule, a pair
of molecules, or even a condensed state assembled with a
large number of molecules, is defined by different energy
terms. Covalent interactions may be described by terms
such as bond, valence angle, torsion, and hybridization
terms; terms describing nonbonded interactions include
van der Waals, electrostatic, and hydrogen-bonding in-
I
teractions. The specific force field used in our study has
0, been previously defined.8 The software POLYGRAF is from
Figure 1. Schematic diagram illustrating the free energy of Molecular Simulations, Inc. The necessity for incorpo-
mixing change as a function of composition at different tem- rating partial charges depends on the specific force fields
peratures. The dashed line is the coexistence curve (phase used. Earlier studies have shown that incorporation of
diagram) constructed from the free energy plot.
charges appeared unnecessary.+l1 Some contributions
can be expressed as2 from charges have been incorporated in the force field. In
AG - 41 42 this study, we found that, for some systems, ignoring
- - -1n 41+ -1n
RT x1
4J2 + xd14J2
x2
(1) electrostatic effects will lead to unfavorable resulb when
compared to experimental data. In these cases, partial
where AG is the free energy of mixing per mole of lattice charges can be calculated utilizing the charge equilibrium
site, 4 is the volume fraction, and x is the chain length method.I2 We have generally used experimental data as
with each repeating unit defined as occupying a lattice a reference.
site. The x parameter is defined as Awl2 used in eq 2 can be obtained in a straightforward
x = zAw12/RT (2) fashion simplyby calculatingthe energiesof three different
pairs as defined by eq 4. A question arises as to whether
where z is the coordination number of the model lattice the calculated Awl2 from simple pairwise interactions is
and Awl2 is the energy of formation of an unlike pair from sufficientlyaccurate to represent interactions in the actual
the stoichiometric equation condensed state. To improve the reliability of the
1 1 calculation of Aw12, we chose to take into account and
-[1,11
2
+ 5[2,21 = [I121 (3) properly weigh a large number of relative orientations of
the two molecules (like or unlike). Simple energy min-
and is defined as imization using several selected configurations is not
1 representativeof the interaction energyfor binary mixtures
Aw12= ~ 1 2 - 5 ( ~ 1 1~ +2 2 ) (4) exhibiting normal Boltzmann distributions. Molecular
dynamics represents an improved samplingtechnique. At
where wi, is the energy of a particular ij pair. low temperatures, relative movementof the two molecules
When eqs 1and 2 are used to calculate the free energy forming a pair is quite small. Molecular dynamics samples
of mixing, it is necessary to compute z and Aw12. In this only localized configurational space which can lead to
study, the procedure to calculate these parameters using incorrect sampling using reasonable amounts of computer
a molecular simulation technique will be discussed. In
addition, the phase behavior associated with three classic time. At high temperatures, two molecules may have
textbook examples, a solvent-solvent mixture, a polymer- sufficient kinetic energy to drift apart, an event which
solvent solution, and a polymer-polymer mixture, will be cannot occur in the actual condensed state. A manifes-
analyzed. Polymer solutionsor blends represent extremely tation of these problems using straightforward molecular
exciting areas of research. Sophisticated theories such as dynamics simulations is that the Awl2 is strongly depend-
the reference interaction site model (RISM) and lattice ent on the initial configurations chosen.
cluster theory (LCT)have been recently developed to treat A critical step in the reliable calculation of Awl2 is to
the problems beyond Flory-Huggins original approxima- establish an efficient algorithm for sampling relative
tions.3~~ In a recent series of papers, Dudowicz et al. orientations of a pair of molecules. In this study, an
provided a very detailed discussionconcerningdeficiencies approach employing the Monte Carlo method which
in the original Flory-Huggins t h e ~ r y .From
~ those and includes constraints arising from excluded volume is
other studies using Monte Carlo methods6 it is evident deve10ped.l~The excluded-volumeconstraint method can
that the Flory-Huggins theory is somewhat inadequate in be applied in a variety of situations to sample energetics
many aspects. It is not our intention, however, to analyze of molecules embedded in simple or complex topological
the validity of the Flory-Huggins theory itself but only to environments, i.e., molecular packing or s01vation.l~The
provide a methodology for use of the theory incorporating procedure used in this study, as illustrated in Figure 2,
molecular simulationtechniques to study mixing in binary includes the following steps:
systems. Due to its clarity in concept and simplicity in 1. Construct proper structures for the molecules of
application, Flory-Huggins theory will always remain a interest, either the whole molecule in the case of solvent
good reference regardless of other developments in this or the monomer repeat unit in the case of polymer. The
area. With the methodology developed in this study, it structures are optimizedusing molecular mechanics. The
is possible to apply Flory-Huggins theory in a new and overall shape of each molecule is represented by its van
useful manner. der Waals surface.
Simulation Technique 2. Position the geometric centers of both molecules (1
The classical force-field characteristic of interatomic and 2) being considered near the origin of the Cartesian
interactions, whether intramolecular or intermolecular, is coordinate frame.
Macromolecules, Vol. 25,No.14, 1992 Phase Diagrams for Binary Mixtures 3669

2 7 7 2
zI
X X
step 2 step 3

Figure 3. Schematic drawing of a random direction n pointing

to the surface of a unit sphere.
orientation of molecule 2 with respect to molecule 1can
be specified with five angular variables. The first three
variables are the Euler angles (CY, 8, y) which specify the
orientation of molecule 2 in the defined coordinate system.
The function of step 3 is to randomly choose three Euler
X X
angles to generate a particular orientation of molecule 2.
To ensure uniform samplingin the configurational space,
step 4 step 5
the three Euler angles are determined by
(Y = 27r(rand,)

cos 8 = l-2(rand2)

y = h(rand3) (5)

where randl, randz, and rand3 are random numbersranging

between 0 and 1. The random numbers used in the three
X equations are generated independently. The new coor-
step 6 dinates of the atom j in molecule 2, rj(xJ,Z), after
Figure 2. Consecutive steps to perform a calculation of the pair application of step 3 are
energy of a single configuration. The shape of the molecules is
represented by their van der Waals surfaces. In step 1(notshown rj(x,y,z) = Ar:(xy,z) (6)
here) the optimum structuresof the two molecules are generated.
Step 2 positions two molecules near the origin; the coordinates where rjo(x,y,z) is the initial coordinate of atom j in
of molecule 1 (unshaded) will be unchanged throughout the molecule 2 before rotation and A is the transformation
calculations. Step 3 randomly rotates molecule 2. Step4 chooses matrix defined by15 eq 7.
a random direction n in the space. Step 5 translate molecule 2
along n with a minimum distance to void overlap between two
molecules. Step 6 shows the final position of two molecules for cos y cos @ cos a - COS y cos @ sin a + -cos y sin @
which the energy of the configuration is calculated. sin y sin a sin y cos a
-sin y cos @ cos a - -sin y cos @ sin a+ sin y sin @
3. Choose a particular orientation of molecule 2 by cos y sin a cos y cos a
randomly choosing the three Euler angles. sin @ cos a sin @ sin a cos @
4. Randomly choose a vector pointing from the origin
to the surface of a unit sphere.
5. Translate molecule 2, the orientation of which is The last two angular variables are the polar angles 4
determined by step 3, along the vector determined by step and 8 as shown in Figure 3. The unit vector n defined by
4, until the van der Waals surfaces of each molecule just these two angles is
touch each other.
6. Calculate the pairwise interaction energy of this n = (sin 8 cos 4 sin 8 sin 4 cos 8 ) (8)
specific configuration and use the Metropolis transition which will determine the relative position of molecule 2
matrix to determine whether to accept this configuration. to molecule 1 when molecule 2 is translated along this
7. Repeat steps 3-6. Calculate Wij by averaging all the direction in step 5. This step is necessary to ensure that
accepted configurations. Monitor the convergence of wij all specific interactions are considered. A uniform sam-
as a function of the number of Monte Carlo trials. The pling scheme for 4 and 8 can be achieved usingls
resultant wij can then be used to calculate Awl2 using eq
4. q5 = 27r(rand,)
The fmt two steps are quite simple and straightforward.
Steps 3 and 4 involve using a procedureto randomlyrotate
a molecule and generate a vector in a three-dimensional COS 8 = l-2(rand5) (9)
space. Similar routines to perform these two operations
can be found in molecular simulation texts.'4 A detailed where rand4 and rands are random numbers ranging
discussion concerning steps 3-6 follows. between 0 and 1. Figure 4 shows the projections of the
Relative Orientation of the Two Molecules. If the end of the vectors generated through eq 9 onto three
position of molecule 1is fixed near the origin, the relative mutually perpendicular planes. It demonstrates the
3670 Fan et al. Macromolecules, Vol. 25, No.14, 1992

-I 0 I
X

W
i

(b)
Figure 5. Calculation of translation vector m (n is determined
in step 4) for a two-atom system: (a) Geometric relationship
between different positions before and after the translation. The
large sphere (i) represents the fixed atom and the small sphere
(j)represents the movable atom. The shaded sphere is the
position of j after the translation. (b) Two possible solutions for
the translation vector rn, r+n and r-n, and two corresponding
new positions of atom j, j + and j-.
two atoms, located near the origin, overlap. The positions
of the two atoms are defined by vectors ri and rj,
respectively. Suppose atom j is moved in the n direction
(chosen in step 4) to avoid any overlap between the two;
the new position of atom j, represented by the shaded
sphere, is at the distance determined by vector rn. The
geometric relationships between vectors defined in Figure
5a yield the equations
rij= rj - r,

rn + rij= R

1rn + rijf = (R,+ R~)'

.1 0 I
2 r2+ 2(rijm)r+ lrij2- (R,+ R~)'= o
(C)
where Ri and Rj are the van der Waals radii of atoms i and
Figure 4. Projections of the end position of a random vector n j, respectively. The minimum displacement r, therefore,
pointing to the surface of a unit sphere after loo0 sampling: (a)
on the x-y plane; (b) on the y-z plane; ( c ) on the z-x plane. is the solution of the following equation:

uniformity of the distribution of these vectors in all r* = -(rij-n)f [(rij.nl2- lrii12+ (R,+ R ~ ) ~ (10)
I ~ / ~
directions (the density becomes higher outward due to The two roots indicate that atom j can be translated in
the direction of projection). This uniform samplingmake two ways. In this specific case, r+ represents an atom
it possible for molecule 2 to access any position on the van being moved in the positive direction. r- is associated
der Waals surface of molecule 1if topologicallypermissible. with the atom being moved in the negative direction as
These data are obtained for a sampling number of 1OOO. shown in Figure 5b. If the number of atoms in molecules
Translation of Molecule 2 to the van der Waals 1and 2 are ml and m2, respectively, it is then necessary
Surface of Molecule 1. The purpose of step 5 is to to calculate all the rij's (mlX m2)between atoms of these
perform the minimum translation in the direction n, two molecules using eq 10. The maximum and minimum
determined a t step 4, necessary for molecule 2 to avoid an rijwill be the two correctsolutionsfor the translationvector
overlap of van der Waals volume with molecule 1. To rn. All the other rij's will cause some overlap. In this
demonstrate this method, we use a simplified case con- study, the maximum rij is chosen to determine the
taining only two atoms as shown in Figure 5. Initially the displacement of molecule 2.
Macromolecules, Vol. 25, No. 14, 1992 Phase Diagrams for Binary Mixtures 3671

21 28 19 M

E (kcal/mol) Figure 7. Statistical error due to finite sampling as a function

of sampling time N.
Figure 6. Pair energy distributions of polyisoprene and poly-
styrene segmenta: (0) the original data of all the configurations
generated; (+) after the original data were sampled using the truly average pair energy (E) be obtained. Limited
Metropolis algorithm at 200 K and ( X ) at 400 K. sampling statistics, Le., finite N , will cause ( E ) ~ toscillate
o
about (E). The uncertainty can be analyzed by dividing
Calculation of the Pair Energy. Generation of the N configurations into M subdivisions, with each
relative orientations using the above procedure leads to division containing L (L = N/M)configurations. The
configurations of varying energetics. The configuration average energy of each subdivision is ( E ) L(I~= 1,M). The
of the two molecules often gives very favorable energies. statistical error due to a finite number, N , is given by17
Many actually locate a local minimum on the potential
energy surface. We considered three sampling methods
t o weigh t h e appropriateness of t h e generated
configurations: (1)The energy of each configuration can
be lowered by molecular mechanics, and all are accepted. The error monotonically decreases as N increases. Dif-
This approach excludes temperature effects and does not ferent values of (E)Nfor different L (L= 1, 10,100,lOOO)
lead to a Boltzmann distrubition. (2) Accept all the have been calculated. For example, if N is 30 OOO, the
configurations generated but weigh each with the Boltz- statisticaluncertaintyfor A(E)Nfor differentL converges
mann factor, e-E/kt.(3) Use a Metropolis Monte Carlo tO0.0016. Consideringthesmallest Aw12to be -0.01 (poly-
sampling algorithm. If we define u, as the probability isoprene/polystyrene system) for the systems study here,
that if the system is in state v making a transition to state sampling 100 OOO configurations (about 7% error) should
v', then u, is given
provide reasonable accuracy. The statistical errors for
the other two systems studied here are much smaller due
AE,SO to larger Aw12S. As expected, there is a linear relationship
between A(E)Nand N-1/2as shown in Figure 7 (L= 1).
The statistical error at different N , therefore, can be
estimated from this plot. Generating 100 OOO pair con-
where A E d = E t - E,. The configuration is accepted if figurations with the total atom number of a pair being
ita energy is less than or equal to the energy of the previous -50 and calculating their energies required -3 h on a
configuration. If the energy is higher, a random number SGI 4D35 workstation. The computation time has not
between 0 and 1 is generated. The configuration is proven to be a major limitation in this study.
accepted only if exp(-AE,/RT) is larger than or equal to Calculation of Coordination Number z. In the
the value of the random number. The third method is the Flory-Huggins scheme, x, defined in eq 2, is perhaps the
more rigorous samplingtechnique and is used throughout single most important parameter required to describe the
this study. miscibility behavior of a binary system. None of the free
In practice, the energies of all generated configurations energy expressions contains explicit dependence on the
are stored and the Metropolis algorithm is then used to coordination number, i.e., how many neighbors of inter-
process the stored data at different temperatures. Figure action need be considered. In this study, however, the
6 shows the pair energy distribution of polyisoprene and coordination number of a system has to be computed
polystyrene segments of the original data for 50000 directly. A single coordination number, as in eq 2, has a
configurationsas well as the distributions at two different definite physical significance only when the two compo-
temperatures, 200 and 400 K, obtained by Metropolis nents, either molecules or polymer segments,have a similar
sampling. The average pair energy for all the configu- volume or surface area. The difficulty in defining a
rations is 29.501 kcal/mol and for temperatures at 400 and coordination number for a system in which two components
200 K is 29.464 and 29.419 kcal/mol, respectively. Each are not compatible in size clearly limits the applicability
distribution is somewhat asymmetrical. There is a long, of the approach developed here, and an independent study
low-energytail. The number of configurationswith energy is necessary.
higher than the peak value decreases dramatically. This An algorithm to approximate the possible number of
phenomenon clearly indicates the merit of this sampling the nearest neighbors in contact with the center molecule
technique; i.e., very few configurations with high energy can be provided. For a binary system, at least four different
or bad van der Waals contacts are generated. combinationsare possible; i.e., the center molecule, as well
If N configurations are generated, the average energy as the surrounding molecule, can be either species 1or 2.
( E ) Nis calculated. Only when N goes to infinity will a This leads to four coordination numbers. A more complex
3672 Fan et al. Macromolecules, Vol. 25, No.14, 1992
isobutyl ketone and polyisobutylene. The phase diagram
with different molecular weights of polymer has been
presented earlier.’ The polymerpolymer system studied
is oligomer mixtures of polyisoprene and polystyrene, one
of the few systems with a clearly defined UCST and an
availability of the entire phase diagram.
Results and Discussion
Solvent-Solvent System. For simplicity, a well-
known solvent-solvent system is chosen. This study also
serves as a useful reference for comparison to polymer
results. The phase diagram of n-hexane (CHdCHz)&H3)
as component 1 and nitrobenzene (PhN02) as component
2 is used as a typical example of partially miscible binary
mixtures in many physical chemistry textbooks.18 The
application of the approach discussed above to a solvent-
solvent system is straightforward. The structures of both
molecules are optimized with the Dreiding force field.*
(C) (d) Tworoutesareavailabletocalculatetheenergyofanunlike
Figure 8. Calculation of coordination number z by packing pair, w12. It is possible to fix molecule 1 (n-hexane) and
nearest neighbors around the center molecule (unshaded). The sample different configurations of molecule 2 (nitrohen-
darkly shaded molecule represents the one being packed, and zene) or its converse. These two methods provide exactly
the lightlyshadedmoleculesrepresent existingnearestneighbors. the same distribution of w12, an indication of the self-
The shapeofthemoleculearepresents tbeirvanderWaalssurface: consistency of the methods. The calculated Awl2 as a
(a) with no existing neighbor; (b)with one existing neighbor; (e)
with two existing neighbors; (d) with three existing neighbors.
function of temperature is shown in Figure 9. For the
After d no more space around the center molecule is available convenience of calculation of the critical temperature we
to pack one more nearest neighbor. The coordination number have fitted Awl2 versus ?‘of this system with the following
obtained in this particular search is 4. functional form:
situation is that the surroundingmolecule can be a mixture AwI2 = -7.66 X + 1.05 X 104/p (13)
of 1 and 2. We find that the four types of coordination
numbers are quite similar when the molecular sizes of the The original Flory-Huggins theory considers only the
2 species are similar. As shown in Figure 8, when the lattice configuration entropy. By using the expression of
position of the first nearest neighbor is determined by the Awlz or x as a function of temperature, it is possible
steps 2-5, placement of the second nearest neighbor needs to account for other entropic contributions due to specific
to avoid overlap with both the center molecule and the interactionsbetweenthe twocomponentsand their relative
first nearestneighbor molecule. In addition, it also needs orientations. The molecular origin of the temperature
be in contact with the center molecule to qualify as a dependence of x is that the different configurations of
nearest neighbor. Additional constraints need therefore pairs have different energies and their contribution to the
be introduced to pack molecules around the center Awlz is temperature dependent. With the availability of
molecule. These constraints dictate that the maximum x ( T ) , the entropy and heat of mixing can he calculated by
displacement r+ has to be determined by the atoms of the the following equations:’
reference molecule and the atoms of the molecules being
added. By repeating this calculation, one obtains an
average coordination number for the two molecules
considered. The coordination number obtained in this
fashion depends on the number of sampling or packing
undertaken. The initial packing scheme has proven to be
very effective. After a certain number of sampling, little
space remains available for additional nearest neighbors. The calculated coordination number for this system is
After a few neighbors are introduced, it becomes difficult 7.8. The phase diagram of n-hexane and nitrobenzene is
to introduce additional molecules about the center mol- shown in Figure 10. The calculated critical temperature
ecule. The number of nearest neighbors then increase is 268 K while the value in the literature is 293 K.18 It is
very slowly with an increase of the packing number. In clear from eq 2 that the x and, therefore, TCrdepend
these calculations, we have taken the value at the onset strongly on the coordination number z. Figure 11 shows
of a plateau of the plot of nearest neighbor against packing the criticaltemperature as afunction of assumed different
number. The coordination number obtained is about 7. z values. The calculated critical temperature would be
The critical temperature is usually strongly dependent on matched with the experimental value 293 Kif an assumed
the coordination number. z of 11.5 were used. This z of 11.5 actually occurs in this
It is clear that a method can be developed to apply the simulation after an average of 10 OOO tries of consecutive
molecular simulation technique in conjunction with the packing. When the coordination number is not well
Flory-Huggins theory for analyzing binary mixtures. In defined or easily calculated, it might be possible to fit the
this study, three systems have been chosen to validate the calculated critical temperature with the experimental one
methodology outlined above. The examplesare relatively to obtain z and acquire certain information concerning
simple, and relevant experimental data are available for microstructure or packing.
comparison with calculation. The first is the n-hexane Solvent-Polymer System. A diisobutyl ketone (com-
and nitrobenzene solvent-solvent system. The phase ponent 1) and polyisobutylene (component 2) mixture is
diagram of thismixture appears in most physicalcbemistry one of interesting examples of a solvent-polymer system.
textbooks.18 The solvent-polymer system chosen is di- The polymer’s molecular weight dependenceof the critical
Macromolecules, Vol. 25, No. 14, 1992 Phase Diagrams for Binary Mixtures 3673
0.8

0.6
D
2
3
2
0.4

3
Q

0.2

0.0
1W 200 300 400 500 600
T (K)
Figure 9. A w , ~of n-hexane and nitrobenzene as a function of
temperature.
(b)
Figure 12. Constraints in the calculation of the pair energy for
a systemcontainingpolymer. (a)A solvent-solvent system. There
are noconstraints. Allfourconfigurationsofmolecule 2 (shaded)
are possible. (b) A solvent-polymer system. The unshaded
spheres represent two dummy atoms introducedat the head and
tail positions of the polymer segment. Because of chain con-
nectivity these two positions are inaccessible to the solvent
molecules. The configurationsof molecule 2 (darklyshaded) in
contact with the dummv atoms will be reiected in thecalculation
of the pair energy. The radii of the dummy atoms are their van
der Waals radii.

significantlylarger than the polymer monomer repeatunit.

Themostconvenient waytodefine thesizeof alatticesite
is for it to he sufficientlylarge to hold one solvent molecule
or one polymer segment containing two monomer repeat
*, units. A polymer segment then becomes -C(CH&-
Figure 10. Calculated coexistence (0)and spinodal (e)CUNB CH2C(CH3)zCH2-. With this definition, the relative sizes
of n-hexane and nitrobenzene binary mixtures. ofthesolventmoleculeandthepolymersegmentarealmwt
identical, within 5% in either the area of the molecule’s
van der Waals surface or the volume encompassed by this
surface. The molecular surface area and volume are
calculated using the van der Waals radii of the atoms in
the molecule. Such an approach to define the size of a
lattice site may, however, not be generally applicable.
Effects of polymer connectivity also require consider-
ation. Any configuration generated in the two-solvent
mixture is physicallyacceptable. For a long-chainpolymer,
the situation is quite different. The ends of a polymer
segment are normally connected to other segments,making
some positions of the polymer segment inaccessible to
solvent molecules as shown in Figure 12. Chain connec-
tivity must be considered in the solvent-polymer system,
and configurations violating this constraint should be
rejected. It is possible to introduce two dummy atoms at
.”” the head and tail positions of the polymer segment as the
0 5 10 I5 D two unshaded spheres drawn schematically in Figure 12b.
e The polymer segment now becomes C’C(CH3)zCHzC-
Figure 11. Upper criticalsolutiontemperature of the n-hexane (CH3)2CH2Cf.The two C’atoms at theends of the polymer
and nitrobenzene system as a function of coordination number segment are dummy atoms. To obtain the optimum
2. structure of the redefined polymer segment, these two
temperature and the phase diagram of this binary mixture dummy atoms C’s represent -CH3 in the united atom
has been presented in Flory’s classid book.’ This system approximation. The structure of the segment is then
contains typical questions in need of consideration when optimized.
applying the methodology developed here to a polymer Upon generation of a configuration, contacts between
system. Thechemicalstructureof the solvent and polymer atoms in the solvent molecule and these C‘ atoms are
repeat unit are (CH3)3CCOC(CH3)3and -C(CH3)2CHz-, examined. Any configuration containing such contads is
respectively. The first question to be addressed is the rejected as shown in Figure 12b. In the calculation of pair
definition of the lattice size because the solvent size is energy w12, the energy terms (bond, angle, torsion, and
3674 Fan et al. Macromolecules, Vol. 25, No. 14, 1992
330-

14 am

.
. I

0 o5.m
0 0 .
. O 0 . * "6
310' * 0 0 .
0 0 .
0 0 .
0 0
T (K) 0 0

290'

0.01! ' ' ' ' ' ' ' ' ' 1 270
0.0 0.1 0.2
100 200 300 400 500 600

T (K) 92
Figure 13. Awl2of the diisobutyl ketone and polyisobutylene Figure 14. Phase diagrams (coexistence curve) of diisobutyl
segment as a function of temperature. ketone and polyisobutylenewith three different molecular weight
polyisobutylene: ( X I 22 700, (0)285 OOO, and (+) 6 OOO OOO.
van der Waals energy) associated with the two C' atoms Table I
are not included. The sole function of these C' a t o m in Critical Temperatures of Diisobutyl Ketone and
the calculation of w12 is the use of their van der Waals Polyisobutylene
radii to reject improperconfigurationswhich violate chain
connectivity. The same approach is used to calculate w22, ~~

MW exptl calc ( I = 5) calc ( z = 6.5) cdc ( z = 8)

the pair energy between polymer segments. One may
argue, usingthis criteria, that certain generated polymeric 22700 292 225 295 360
confiiations, e.g., when a segment is connectedto another 285000 319 240 315 390
6000000 329 245 320 395
segment, are excluded. Them configurations do not affect
the mixing process, since they should have no contribution
to the heat of mixing. The competition between like and Polymer-Polymer System. Most polymer pairs are
unlike species only occurs for those configurations acces- immiscible, as easily understood from eq 1. With high
sible to both the solvent molecule and other polymer molecular weight polymers (large x in the equation), a
segments. Therefore, a similar constraint for calculating slightly positive value of Awl2 willlead to phase separation.
w22 is justifiable. An oligomericpolyisoprene (component 1)and polystyrene
(component2)mixture is perhaps one of the best studied
The calculated Awl2(T), plotted in Figure 13,can be binary blends exhibiting an upper critical solution
fitted with following equation: The monomer repeat units of polyiso-
Awlz = 8.20 X 10" + 5.45JT (16) prene and polystyrene are -H~C(CH~)C=CHCHZ- and
-PhCHCH2-, respectively. The double bond in polyiso-
The argument for taking chain connectivity into account prene has a cis configuration. The isoprene segment can
in the calculation of w12 and w z also holds in the calculation have different conformations, with the optimum structure
of z. We define zo as the maximum coordination number obtained after several cycles of energy minimization and
of the system and z as the coordination number calculated molecular dynamics. The calculated results were not found
with the constraint associated with chain connectivity. to be very sensitive to the initial conformation of the
The precise number of neighboring lattice sites used in eq polymer segment, mainly because the sampling method
2 for a polymer segment should be zo - 2.' When the assures that nearly every part of the molecule is touched
constraint discussed above is applied to a solvent-polymer by another segment. A problem with the polystyrene
system, the calculated result should be zo for pair [1,13, segment defined above is that the tacticity has not been
-
zo 2 for pair [2,21,and zo - 1 for the average of the pairs taken into account. A more complicated segment needs
[1,21and [2,11. Thus,when the constraint is imposed, be defined if this factor is to be considered. The choice
the number of possible nearest neighbors for a particular made in this study is applicable to an atactic polymer.
pair is automatically obtained and the approximation cited Very high molecular weight (MW > 20 OOO) polyisoprene
in the original derivation eliminated. A single z can be and polystyrene are known to be immiscible. Accurate
obtained by averaging different pairs which is about 6.5 experimentaldata exist for systems with molecular weights
for the system discussed here. ranging from 10o0 to 3000.
The phasediagramfor three different molecular weights The discussion in the previous sections regarding the
of the polymer is shown in Figure 14. The numbers of effects of chain connectivity should also be applicable to
repeat units, x 2 used in eq 1, of three polymers are 203, the mixture involving two polymers. The calculated Awl2
2544, and 53 571. These numbers are only half of the real as a function of temperature is shown in Figure 15. The
degrees of polymerization because the polymer segment data, after fitting, yield the following equation:
is defined as twice as large as the real monomer repeat
unit. Table I presents a comparison of experimental data AwI2 = 6.49 X (17) + 1.66lT
with the calculated results for several different 2's. The The calculated coordination number is about 5. As
agreement between the calculated results of z = 6.5 and discussed previously, the coordination number should be
the experimental data is excellent. The critical temper- zo - 2. We find that ZO, the coordination number without
atures calculated for three different molecular weight the constraint of the chain effect, is quite consistent for
polymeraolvent systems are all within a few degrees of all three systems, i.e., -7. This consistency indicates that
the experimental data. using a single dummy atom at both ends of the polymer
Macromolecules, Vol. 25, No.14, 1992 Phase Diagrams for Binary Mixtures 3675
this type of calculation is related to the size of a lattice
site. No two molecules or polymer segments or a com-
bination of the two have exactly the same van der Waals
0.020 I\ volume or surface area. If the size difference of the two
components is small, one can use x , the chain length, to
adjust the lattice size, even for a small-molecule system.
For example,the van der Waals volumes of n-hexane and
nitrobenzeneare 180and 147A3,respectively. If we define
the size of the lattice as the volume of nitrobenzene, then
x1 in eq 1 is 1.22 while x2 is 1. However, the physical
meaning of pair energies then becomes ill-defined. For
example, w11 is the pair energy of two 1.22 111. w12 is one
1.22 [l]and one 121. The w22 remains the pair energy of
two [2]. A scale down of the w11 and w12 values of this
100 200 300 400 500 600 system according to the volume ratio (1.22) was unsuc-
T (K) cessful. The calculated critical temperature changed
Figure 15. Awl2 of polyisoprene and polystyrene segments as
drastically with the volume ratio. No easily definable
a function of temperature. procedure exists to reasonably correct for both x and w.
This factor was thus ignored in this study. When the size
Table I1 difference of two molecules is large, this approximation
Critical Temperature of Polyisoprene (pip) and will introduce serious problems. Another problem caused
~~
Polystyrene (PSI by the size mismatch, as mentioned previously, is that the
MW Tcr (K) coordination number z becomes ill-defined. One needs to
PiP PS exptl calc (z = 4) calc ( z = 5) exercise judgement about the approximations applicable
for each system. Nevertheless, this approach can be
lo00 lo00 243 230 267 applied to many different systems especially when only
2000 2700 329 397 473
2700 2100 408 410 495 qualitative results or relative comparisons are required.
2700 2700 448 450 550 The validity of using the molecular simulationtechnique
to predict polymer miscibility depends on the theoretical
segment in this approach is justifiable. Instead, if one model it is based upon and the simulation technique
introduces a monomer unit at the ends of the polymer applied. The Flory-Huggins theory has certain deficien-
segment,the excluded-volumeeffect of this bulky dummy cies. It cannot predict the existence of the lower critical
group will reject most of the configurations in the solution temperature; it fails for the composition depend-
calculation of both Aw12 and z. The z will be too small ence of x , and it fails to recognize additional entropic
compared to zo to be reasonable. This characteristic of contributions other than a purely mixing component. The
our approach has its root in the step of determining the approach used in this study also possesses some of the
translation vector. deficiencies of this theory. Advancement is achieved,
Experimental results for polyisoprene and polystyrene however, by providing the molecular basis of the tem-
have been obtained from two sources.1921 The data are perature dependence of x as well as the capability to
listed in Table 11, together with calculated results, If a calculate this dependence. Some studies suggest that the
coordination number of 5 is used, calculated results differ Flory-Huggins theory strongly overestimatesvalues of the
significantly from the experimental values. Except for
the case of relatively low molecular weight (MW 10oO),
the discrepancy between the experimental values and the
- critical temperature,22and this tendency is also evident
in our polymer calculations. We should, however, be
cautious about this conclusion, since the accuracy of the
calculated results can be very large, near or over 100 K. force field for different systems may vary.
However, if the coordinate number is reduced to a value The philosophy of this approach is relatively simple:
of 4, the calculated results become quite consistent with use a simple theoretical model which contains parameters
the experimental values. having a definite physical meaning or molecular origin (z
It is very difficult to accurately predict critical tem- and Awl2 in this case) and then calculate these parameters
peratures of a polymer mixture by use of the Flory-Hug- with computer simulation. The theory need not be as
gins theory. The heat of formation for an unlike pair, elementary as the Flory-Huggins theory. However, the
Aw12, is usually small. The Awl2 of the polyisoprene and
molecular origin of the parameters is required as well as
polystyrene system treated here is only about 0.01 (kcal/ the capability for calculation in order to avoid arbitrary
mol of polymer segment) near room temperature. The fitting. In principle, specific knowledge about the system
degree of polymerization has to be less than 30 for the is unnecessary as information which is unavailableby either
system to exhibit an upper critical solution temperature purely theoretical or experimental means may be gener-
near room temperature. By increasing the x to about 100, ated.
the system becomes immiscible even at reasonably high
temperatures. The small value of Awl2 and large x , the Conclusions
degree of polymerization, sensitize the critical tempera-
ture to both Awl2 and z. Thus both Awl2 and z need be A systematical approach to calculate the temperature
accurately calculated, a condition that cannot be easily dependence of the x parameter in the Flory-Huggins
satisfied in real calculations. The calculations carried out theory of a binary mixture is developed in this study. The
in this study provide the means to describe the miscibility method consists of a Monte Carlo sampling and a
behavior of a binary system, with small scaling of the z consideration of excluded-volume contraint in the calcu-
taken into account. lation of pairwise interactions (Aw12) and coordination
The methodology outlined in this study can be used for number z. Calculated results for a solvent-solvent and a
any system as long as a proper force field exists to describe solvent-polymer system compare favorably with the
all the interatomic interactions. The major obstacle for experimental data, even with a generic force field. In the
 3676 Fan et al. Macromolecules, Vol. 25, No. 14, 1992
calculation of a polymer-polymer binary mixture a small Pesci, A. I.; Freed, K. F. Macromolecules 1990,23, 1181.
scaling on the coordination number z is necessary in order (7) Burtkert, U.; Allinger, N. L. Molecular Mechanics; American
to match the experimental data. The accuracy of this Chemical Society: Washington, DC, 1982.
(8) Mayo, S. L.; Olafson, B. D.; Goddard, W. A. J. Phys. Chem.
type of calculation depends on the theoretical model, the 1990,94,8897.
force field, and the limitation of the method itself. Further (9) Fan, C. F.; Hsu, S. L. Macromolecules 1991,24,6244.
improvements in all three aspects are required for a more (10) Yang, X.; Hsu, S. L. Macromolecles 1991,24,6680.
definite prediction of the behavior of binary mixtures. (11) Fan, C. F.; Hsu, S. L. Macromolecules 1992, 25, 266.
Nevertheless, the calculations in this study clearly dem- (12) Rappe, A. K.; Goddard, W. A., I11 J. Phys. Chem. 1991, 95,
onstrate the applicability of molecular simulations in this 3358.
exciting area. (13) Blanco, M. J. Comput. Chem. 1991,12, 237.
(14) Allen, M. P.; Tildesley, D. J. Computer Simulation ofliquids;
References and Notes Oxford University Press: New York, 1987.
(15) Arfken, G. Mathematical Methods for Physicists; Academic
(1) Flory, P. J. Principles ofPolymerChemistry;Comell University Press: New York, 1985.
Press: Ithaca, NY, 1953. (16) Sobol,I. M. The Monte Carlo Method;The Universityof Chicago
(2) Paul, D. R., Ed. Polymer Blends; Academic Press: New York, Press: Chicago, IL, 1974.
1978. (17) Chandler, D. Introduction to Modern Statistical Mechanics;
(3) Schweizer, K. S.; Curro, J. G. J. Chem. Phys. 1989, 91, 5059. Oxford University Press: New York, 1987.
(4) Freed, K. F. J . Phys. A 1985,18,871. Bawendi, M. G.; Freed, (18) Atkins, P. W. Physical Chemistry, 2nd ed.; Oxford University
K. F.; Mohanty, U. J. Chem. Phys. 1986,84, 7036; 1987,87, Press: New York, 1982. Moore, W. J. Physical Chemistry, 5th
5534. Nemirovsky, A. M.; Bawendi, M. G.; Freed, K. F. J.Chem. ed.; Prentice-Hall: Englewood Cliffs, NJ, 1972.
Phys. 1987,87,7272. Bawendi, M. G.; Freed, K. F. J. Chem. (19) McIntyre, D.; Rounds, N.; Campos-Lopez, E. Polym. Prepr.
Phys. 1988,88,2741. Pesci, A. I.; Freed, K. F. J. Chem. Phys. (Am. Chem. SOC.,Diu. Polym. Chem.) 1969,10, 531.
1989,90,2003.
( 5 ) Dudowicz, J.; Freed, K. F.; Madden, W. G. Macromolecules (20) Koningsveld, R.; Kleintjens, L. A. Macromol. Chem. 1973, 8,
1990,23,4803. Dudowin, J.;Freed, K. F. Macromolecules 1991, 197.
24, 5076 and 5112. Dudowicz, J.; Freed, M. S.; Freed, K. F. (21) Koningsveld, R.; Kleintjens, L. A.; Schoffeleers, H. M. Pure
Macromolecules 1991,24,5096. Appl. Chem. 1974,39, 1.
(6) Sariban, A,; Binder, K. J. Chem. Phys. 1987,86,5859. Sariban, (22) Binder, K. Colloid Polym. Sci. 1988, 266, 871 and reference
A.; Binder, K. Macromolecules 1988,21,711. Madden, W . G.; therein.

View publication stats