Molecular Modeling Approach to Determine Flory-Huggins

Interaction Parameter for Polymer Miscibility Analysis

Connor P. Callaway1, Kayla Hendrickson1, Nicholas Bond1, Seung Min Lee1, Parveen Sood1, and
Seung Soon Jang1,2,3,*

1
Computational NanoBio Technology Laboratory, School of Materials Science and Engineering,
Georgia Institute of Technology, 771 Ferst Drive NW, Atlanta, GA 30332-0245, USA
2
Institute for Electronics and Nanotechnology, Georgia Institute of Technology, Atlanta, GA,
USA
3
Parker H. Petit Institute for Bioengineering and Bioscience, Georgia Institute of Technology,
Atlanta, GA, USA

*
Corresponding author. Email address: seungsoon.jang@mse.gatech.edu

1
Abstract
In this work, we present a thorough procedure for estimating the Flory-Huggins χ-parameter for
use in atomistic and mesoscale molecular simulations in computational materials science. In
particular, we propose improvements upon traditional Flory-Huggins theory by implementing a
Connolly volume normalization (CVN). We apply this technique to several test systems, including
a blend of poly(epichlorohydrin) and poly(methyl acrylate), a blend of polyethylene glycol and
poly(methyl methacrylate), a blend of polystyrene and deuterated polystyrene, and three
molecular-weight variants (monomer, dimer, and trimer) of a triblock copolymer for use in
multicompartment micelle applications. Our results demonstrate that the newly developed
procedure offers high accuracy and efficiency in predicting the Flory-Huggins χ-parameter for
miscibility analysis than traditional experimental and computational methods. There yet remain
several factors which cause the magnitude of the χ-parameter to vary between simulations
performed on molecular species with the same identity but different degrees of polymerization;
although we discuss possible explanations for these factors, this is nonetheless a primary focus for
further exploration into this new methodology.

Keywords: molecular modeling, Flory-Huggins theory, χ parameter, molecular interaction,

multiscale simulation

2
Introduction
Modern applications of polymeric materials increasingly utilize polymer blends instead of
single-component polymer systems. The burgeoning manufacturing field, for example, often
makes use of blends such as acrylonitrile-butadiene-styrene (ABS) and polycarbonate (PC) [1].
Whereas designing new polymers from the ground up to obtain any target properties is typically
expensive, slow, and subject to experimental error, polymer blend approaches are popular because
they offer a convenient way to easily tailor the properties of a system by mixing given polymeric
materials [2, 3].
However, many polymer systems will not readily blend, which can pose problems for
applications calling for polymers with specified characteristics. For this reason, it can be extremely
desirable to be able to probe the miscibility behavior of a given pair of polymers. Indeed, the design
of a chemical process relies heavily on the miscibility of constituent polymers. Likewise, the utility
of polymer phase diagrams – whether based on already existent polymers or composed of novel
molecules – is highly appreciated, as they offer valuable information about the regions in which a
useful polymer blend may be produced.
Even now, new applications for polymer miscibility analysis are arising, among which the
concept of the micelle nanoreactor has become a topic of intense and widespread interest in the
chemical community in recent years [4-7]. In particular, the multicompartment micelle
nanoreactor, a prominent development in immobilized catalysis [8, 9], heavily relies on miscibility
analysis to probe the effectiveness of a given micelle design with respect to reactant and product
transporting through the micelle layers. Moreover, in order to achieve the multicompartment
nature of the nanoreactor, different compartments in micelle must be mutually immiscible. With
all of this in mind, the ability to determine the miscibility of a given pair of polymers is clearly an
attractive prospect.
To this end, the model proposed by Flory and Huggins offers an interaction parameter χ
which quantifies the favorability of mixing between two polymers [10-12]. This interaction
parameter depends on many process conditions, not least of which are the temperature and
composition of a solution, in order to properly estimate the extent of the interaction between a
given pair of polymer chains. Unfortunately, experimental techniques to determine this parameter
are fraught with challenges and shortcomings. For example, producing chemical solutions for the

3
purpose of miscibility testing can be costly and time-consuming, especially given that it is often
necessary to perform many trials to satisfactorily conclude the χ-parameter for a substance.
Previous work related to the χ-parameter has highlighted the large spread of values
generated by experimental measurements (e.g., by inverse gas chromatography [13-15],
differential scanning calorimetry [16-19], or small-angle neutron scattering [20-23]), necessitating
very precise control of experimental conditions. To compound all of these difficulties, even when
such precision is possible, there is no guarantee that the requisite data (Hildebrand or Hansen
solubility parameters, cohesive energy densities, etc.) are known accurately, if at all.
Due to these significant limitations of experimental techniques for estimating the χ-
parameter, computational miscibility analysis providing accurate χ-parameter may be
advantageous to consider. For one, calculating the χ-parameter through computer simulations
instead of experimentation would greatly expedite the process of determining the miscibility of a
given pair of molecules. In addition, the development of a temperature- and composition-sensitive
model for estimating the χ-parameter would pave the way for full constructions of phase behavior.
In this work, we present a molecular modeling approach to calculate the Flory-Huggins χ-
parameter for a given pair of molecules. Combined with the existing theories and molecular
modeling techniques, such as Flory-Huggins theory, density functional theory, molecular
mechanics/dynamics simulation, and configurational statistics, we establish a new procedure of
normalizing the interaction energies using the molecular volume enclosed by the Connolly surface,
thereby creating an interaction energy density which is analogous to, but distinct from, the
cohesive energy density. From this study, it is apparent that this newly developed procedure
produces accurate estimations for polymer miscibility.

Results and Discussion

Miscibility of PECH and PMA
We performed miscibility simulations for a blend consisting of PECH (Figure 1a) and PMA
(Figure 1b). The PECH-PMA blend has previously been studied via experimental approaches
through inverse gas chromatography (IGC), as discussed previously [13-15]. Table 1 displays the
minimum, average, maximum, and standard deviation of the experimental 𝜒-values obtained via
IGC in comparison with the results from our simulations using both our updated model from
equation (13) and the traditional Flory-Huggins formulation from equation (4).

4
 There are several striking results from this simulation. First, the method we present here
quantitatively matches the data obtained via experimentation to a very satisfactory degree, with
our averaged 𝜒-value falling well within the range of values obtained from IGC for this polymer
blend system at both 349 K and 398 K, the temperatures considered in Al-Saigh and Munk’s work.
[13, 14]. Moreover, the spread of the data obtained in our simulations is far less than that found in
the experimental 𝜒-values. Specifically, the computational methodology shown here reduces the
standard deviation of the obtained data to only 10% of the data obtained via IGC experiments.
Finally, the results here were refined essentially as little as possible: only the monomers
were considered, and the number of energy samples used did not exceed 1×106. By improving
some of these simulation conditions, the accuracy and consistency of our miscibility analysis could
be improved far beyond what can be extracted from physical experiments.
It is noteworthy that Knopp et al., have suggested that modeling the entire polymer is
unnecessary to obtain reliable miscibility data and instead, there is a threshold number of repeat
units beyond which the 𝜒-parameter effectively converges and provides consistent miscibility
behavior, even with an extension of the molecular chain [17]. This suggests that we can obtain
appreciably accurate results with relatively small molecules in well-established simulation
procedure.

Miscibility of PEG and PMMA

We also performed miscibility simulations for a blend of PEG (Figure 1c) and PMMA
(Figure 1d). The PEG-PMMA blend has been studied experimentally using small angle neutron
scattering (SANS) [21]. Knopp et al. reported that using the monomer species to estimate polymer
miscibility may be flawed in cases where the chemistry is changed upon polymerization [17].
Thus, in order to more accurately capture the chemistry of PEG, the trimer form was used instead
of the monomer, whereas only the monomer was considered for PMMA. The difference in
polymerization is normalized in our process when calculating 𝜒.
Figure 2 shows the experimentally determined 𝜒 parameter in comparison to that obtained
from our simulations. Notably, Ito, Russell, and Wignall found that the value of 𝜒 parameter is
very small – under the 𝜃-solvent condition of 𝜒 = 0.5, indicating favorable miscibility[21]. Our
simulation method resulted in smaller 𝜒 parameter in comparison with the experimental values,
which are all smaller than of 𝜒 = 0.5, meaning that it is favorable for mixing.

5
 It is noted that our method resulted in a temperature-dependency of 𝜒 parameter in this
temperature range, which is not observed in SANS approaches. Therefore, we will keep
investigating such discrepancy in the temperature-dependency of 𝜒 parameter.

Miscibility of Triblock Copolymer for Multicompartment Micelle

Having verified that this new methodology is robust and accurate, we applied the procedure
to a triblock copolymer system consisting of poly(acrylic acid) (Figure 1e), poly(6-(4-
formylphenoxy) hexyl acrylate) (Figure 1f), and poly(pentafluorostyrene) (Figure 1g) for blocks
A, B, and C, respectively. This system has been studied for multicompartment micelle applications
in collaboration with Weck’s group in New York University (please see SI). The triblock
copolymer under investigation would be synthesized following an A-B-C structure. In this case,
we employed monomer, dimer, and trimer models in order to investigate the effect of the number
of monomeric units on the calculation of χ-parameter.
We conducted miscibility simulations on these molecules under a variety of conditions.
First, simulations were performed on the monomer models, both with consideration given to
dielectric screening effects (i.e., in the implicit presence of a water-like solvent with dielectric
constant κ = 78.4) and without any dielectric screening. All simulations performed with dielectric
screening used a dielectric constant of κ = 78.4, while those performed with no dielectric screening
used κ = 1. Following this, the simulations were repeated for the dimer and trimer models with
dielectric screening in order to analyze the effect of number of monomeric units. All of the
simulations mentioned here were performed multiple times, to examine the reproducibility of the
present methodology.
Miscibility Analysis for Monomer Models. χ-parameter values for each pair using monomer
models as a function of temperature with and without dielectric screening are shown in Figure 3.
According to Flory-Huggins theory, 𝜒 = 0.5 corresponds to the 𝜃 solvent condition; moreover, 𝜒-
values of less than 0.5 are known to be indicative of miscibility[24]. Based on this, our results
from the monomer model simulations in Figure 3a suggest full immiscibility in each pairing when
dielectric screening is present, with pronounced immiscibility in the A-B and C-A blends.
Here, an important consideration arises with regards to the interactions between blocks. In
solution, while the hydrophilic species A will preferentially interact with water, the hydrophobic
species B and C will be segregated from water, ultimately forming the core of the micelle. Because

6
species A will be in contact with water, water-like dielectric screening (κ = 78.4) should be
included for both the A-B and the C-A pairs.
However, there is small probability for species B and C to be in contact with water in the
equilibrium condition of the system (i.e., after micelle formation). Thus, it is expected that
modeling the B-C pair interaction with dielectric screening by a water-like solvent is not necessary.
Therefore, water-like dielectric screening ought to be omitted for the B-C interaction.
When these calculations are performed without dielectric screening, significantly different
results are observed, as shown in Figure 3b. Clearly, when the dielectric screening is removed, the
B-C pair interaction becomes significantly more favorable. Indeed, the value of 𝜒-parameter
become negative in this region. Based on the difference between Figure 3a and 3b, the importance
of carefully selecting the dielectric constant should be quite clear. In the DPD simulations
performed here, the χ values for A-B and C-A pairs were obtained from simulations with dielectric
screening and the χ values for B-C pair from simulations without dielectric screening.
These results suggest full phase separation between A and B, but only partial or no
separation between B and C. As such, it would be expected that in a micellization experiment
comprised of these blocks, there would only be two-compartment micelles formed. This was
indeed confirmed via cryo-TEM by collaborators at New York University (See Figure SI),
presenting that the hydrophilic corona made of block A is not visible due to its association with
water while the observed part of the micelle does not have internal phase segregation due to the
miscibility of blocks B and C. This indicates remarkable success of the miscibility simulation and
corresponding analysis presented here, especially since this was based on the monomers alone.
Moreover, it can be seen from Figure 3 that the 𝜒-parameters calculated from our molecular
simulations are highly precise for the monomer model case. To illustrate this, Tables 2 and 3
display simulation results which are used to calculate 𝜒 at 295 K both with and without dielectric
screening, respectively. It is noteworthy that these are largely consistent between independent runs.
Miscibility Analysis for Dimer Models. In order to analyze the miscibility of the dimer model,
note that we have performed additional geometry optimization prior to Blends simulation mainly
due to the markedly larger conformational diversity of the dimer models in comparison to the
monomer models. This additional geometry optimization was carried out in Forcite with a smart
convergence algorithm consisting of steepest descent, ABNR, and quasi-Newton techniques.

7
 From the data displayed in Figure 4, it should be evident that the dimers behave quite
similarly to the monomers (as seen in Figure 3). Comparing Figures 5a and 6a, both the monomers
and the dimers display χ-values of close magnitudes. It is clear that the B-C pair is miscible in the
dimer case regardless of dielectric screening, in contrast to same pair in the monomer case.
Comparing the cases without dielectric screening, as observed in Figures 5b and 6b, it is clear that
the dimer and monomer models show similar miscibility behavior, although the dimer model
results demonstrate slightly more immiscibility for the A-B pair than the monomer model results.
Overall, the relative miscibility behavior is in excellent agreement with the monomer
results in each case – with dielectric screening present, as in Figures 5a and 6a, the C-A pair
interaction is the least favorable, while the B-C pair interaction appears to be most favorable. On
the contrary, the cases without dielectric screening (Figures 5b and 6b) point to the A-B pair as the
least miscible, while the B-C pair is still the most miscible of the three. Tables SI-1 and SI-2 reveal
the numerical data which can be used to calculate the χ-values of these simulations.
Miscibility Analysis for Trimer Models. For the trimer model, we again performed an additional
geometry optimization prior to Blends simulations, in order to combat the significantly higher
conformational diversity of the trimer molecules. The settings for this additional geometry
optimization are identical to those outlined in the previous section. These runs also produced
surprisingly precise results, especially considering the sizable conformation space for each trimer.
The miscibility data for the trimer blends are shown in Figure 5.
Figure 5a demonstrates that the B-C pair is highly miscible while the A-B and C-A pairs
are immiscible. Again, these results are in good agreement with cryo-TEM results, confirming that
the phase segregation between blocks B and C does not occur in the micelle, while the phase
segregation between blocks A and B does indeed occur. Specific values of the relevant simulation
results used in calculating the trimer χ-parameter at 295 K are shown in Tables SI-3 and SI-4.
It should be noted that the results from the trimer models are in agreement with those from
the monomer and dimer models, although the relative ordering of the χ-parameters for the A-B
and C-A pairs is reversed in the trimer model. We believe that this result can be attributed to the
effect of conformational diversity, since we already normalized the interactions by the number of
repeating units as well as Connolly volume, as discussed earlier. This appears probable due to the
reversal of the relative ordering of the A-B and C-A pairs, as well as the fact that the B-C pair
interaction is unfavorable in the monomer model, but becomes significantly more favorable in the

8
dimer and trimer models. Further investigation is presently underway and will be the focus of a
later work corresponding to thorough refinement of this methodology.
Finally, several points are evident in all of the studies performed on the miscibility of
species A, B, and C. Principally, it is clear that B-C pair is always the least immiscible pair, and
in fact it is predicted to be completely miscible in every study except that of monomers with
dielectric screening, where the attractive interactions between these species are shielded by the
dielectric medium (Figure 3a). Additionally, we note that in all cases, the magnitude of the 𝜒-
parameter is significantly higher when dielectric screening is not present, which is of course to be
expected. From this, it could be carefully suggested that while it is not necessary to perform time-
consuming explicit solvation modeling in order to probe miscibility, it is essential to give careful
consideration of the implicit solvent effects in order to correctly replicate the behavior of the
polymer system. Overall, these results demonstrate that molecular modeling and simulation can
provide an accurate and predictive method for phase separation of polymer-polymer systems,
which should be a useful tool to design polymer architectures for multicompartment micelle-based
nanoreactors.
Dissipative Particle Dynamics for Triblock Copolymer System. To examine the application of
our miscibility studies to micelle system, dissipative particle dynamics (DPD) simulations were
performed on the ABC triblock copolymer system. As recalled from equation (17), the repulsion
parameter of the DPD method can be connected directly to the 𝜒-parameter[25]:
𝑎AB = 25 + 3.5𝜒AB (17)
Results from these DPD simulations can be seen in Figure 6. As is clear in Figure 6a, the
outermost A-block displays complete microphase separation from the other two blocks, due to the
hydrophilic behavior it displays with the water in the system (water beads omitted from Figure 9a
for visibility). The cross-sectional view in Figure 6b shows that there is no microphase separation
within the core of the micelle – that is, between blocks B and C. This is, of course, to be expected
from the miscibility results obtained from our 𝜒-parameter analysis.
It is worth noting that cryo-TEM work performed on micelles of the same system (with
higher block lengths) show the same behavior; the Supplementary Information for this work
contains cryo-TEM results (Figure SI) in which there is no visible microphase separation between
the B and C blocks. The outer shell of the micelle (i.e., the A block) is not easily visible in TEM
image due to its inherent lack of contrast with the amorphous ice background.

9
CONCLUSIONS
In this work, we propose a computational miscibility analysis scheme that simplifies
calculating the Flory-Huggins interaction parameter (χ) for rapid and reliable determination of the
miscibility behavior of two polymers.
Many advantages of our approach over traditional experimental methods are apparent from
the results presented herein. One of the foremost strengths lies in the reproducibility of the results.
In the majority of the cases considered here, the 𝜒-parameter can be determined with remarkable
precision, delivering consistent values across several independent simulations. Finally, miscibility
analysis via computational methods offers a straightforward way to enhance the precision of a
particular result: by increasing the number of samples and extending the number of repeat units of
a molecule in a calculation, the spread in 𝜒-values can be drastically reduced while still yielding
enhanced accuracy. Performing additional geometry optimizations prior to miscibility calculations
can also offer much higher consistency; a striking example of this was shown in Figure 6, where
the trimer simulation results were highly consistent even despite the exponentially larger number
of conformations.
It should be noted that there is still room for refinement even within the advancements
offered by this method. For example, it is currently unknown exactly how the number of samples
needed in a particular calculation depends on the number of bonds which can undergo dihedral
rotation (which directly influences the conformational diversity). In addition, the particular number
of low-energy configurations used for Boltzmann statistics appears to have a marked effect on the
variance in the 𝜒 -values obtained from the calculation. Advanced sampling techniques are
currently under investigation by the authors in order to improve this methodology. Nonetheless,
the methodology presented here is already quite robust, and enables accurate and precise
estimation of the Flory-Huggins 𝜒 -parameter from even just the monomers, with improved
accuracy and consistency being possible from calculations with higher molecular weights.

MATERIALS AND SIMULATION METHODS

In the present work, we have principally performed miscibility analyses on four systems.
The first system is a blend consisting of poly(epichlorohydrin) (PECH) (Figure 1a) and

10
poly(methyl acrylate) (PMA) (Figure 1b). The second consists of polyethylene glycol (PEG)
(Figure 1c) and poly(methyl methacrylate) (PMMA) (Figure 1d).
The last system is a triblock copolymer consisting of poly(acrylic acid) (Figure 1e), poly(6-
(4-formylphenoxy)hexyl acrylate) (Figure 1f), and poly(pentafluorostyrene) (Figure 1g) as blocks
A, B, and C, respectively. All of the χ-parameters reported in this study were calculated for blends
with equally divided volume ratios.
Flory-Huggins Theory
As discussed before, the Flory-Huggins solution model defines an interaction parameter χ
which is useful for investigating miscibility [10-12]. This model begins with the traditional
definition of Gibbs free energy of mixing, given as
Δ𝐺𝑚 = Δ𝐻𝑚 − 𝑇Δ𝑆𝑚 (1)
where 𝑇 represents the absolute temperature. Flory-Huggins theory reduces the solution to a lattice
model. Thus, the entropy of mixing Δ𝑆𝑚 can be transformed through configurational statistics on
the polymer chains:
Δ𝑆𝑚 = 𝑅(−𝑛1 ln 𝜙1 − 𝑛2 ln 𝜙2 ) (2)
where 𝑅 is the gas constant and 𝑛𝑖 and 𝜙𝑖 represent the number of moles and the volume fraction
of species 𝑖, respectively. Likewise, the enthalpy change upon mixing (Δ𝐻𝑚 ) may be expressed
through the change in interaction energies between the unmixed and mixed states:
Δ𝐻𝑚 = (𝑛1 𝑁𝐴 )𝜙2 Δ𝑤12 (3)
where the mixing energy Δ𝑤12 is given as
𝑍12 +𝑍21 1
Δ𝑤12 = ( ) 𝜀12 − (𝑍11 𝜀11 + 𝑍22 𝜀22 ) (4)
2 2

Here 𝜀ij represents the interaction energy between individual molecules of species 𝑖 and 𝑗, while
𝑍ij represents the coordination number of the species 𝑗 around the species 𝑖. Then the χ-parameter
is defined as
Δ𝑤12
𝜒12 = (5)
𝑅𝑇

and its relationship with the enthalpy of mixing is given as

Δ𝐻𝑚 = 𝑛1 𝜙2 𝜒12 𝑅𝑇 (6)
Thus, the change in free energy upon mixing is expressed in terms of 𝜙1 , 𝜙2 , 𝜒12 , and 𝑅𝑇 via the
combination of equations (1), (2), and (6):
Δ𝐺𝑚 = 𝑅𝑇(𝑛1 ln 𝜙1 + 𝑛2 ln 𝜙2 + 𝑛1 𝜙2 𝜒12 ) (7)

11
 All of these parameters are easily obtainable in experiment except for the value of 𝜒12 .
From Flory-Huggins theory, then, it should be evident that obtaining 𝜒-parameter is the only
barrier for calculating the free energy of mixing of a given pair of molecules. Therefore, it is
tempting to seek experimental techniques that allow the value of 𝜒 to be calculated or measured.
Experimental Approaches for Determination of 𝝌-Parameters
Unfortunately, most of the traditional experimental methods have been unreliable to
varying extents. For example, it is possible to estimate the 𝜒-parameter between two polymers via
inverse gas chromatography (IGC) by first determining the miscibility between a probe species
(designated 1) and each of two polymer species (designated 2 and 3). Each of these interactions
can be determined according to the expression
273.15𝑅𝑣2 𝑉 𝐵11 −𝑉1
𝜒12 = ln ( ) − 1 + 𝑀 1𝑣 − ( ) 𝑃1 ° (8)
𝑉𝑔 °𝑉1 𝑃1 ° 2 2 𝑅𝑇

where 𝑉1 , 𝑃1 °, and 𝐵11 represent the molar volume, saturated vapor pressure, and second virial
coefficient (in the gaseous state), respectively, of the probe species. 𝑣2 and 𝑀2 represent the
specific volume and molecular weight of a polymer, respectively. The parameter 𝑉𝑔 ° represents
the reduced specific retention volume. An analogous expression exists for 𝜒13 . Although most of
these can be determined via experiment to satisfactory accuracy, Al-Saigh and Munk have noted
that the parameters 𝑉1 , 𝑃1 °, and 𝐵11 are all subject to a relatively large amount of uncertainty,
leading to uncertainty in 𝜒12 and 𝜒13 [13]. An expression that avoids the use of these parameters
can be derived as well:
1 𝑉𝑔,blend 𝑉𝑔,2 𝑉𝑔,3
𝜒23 = [ln ( ) − 𝜙2 ln ( ) − 𝜙3 ln ( )] (9)
𝜇𝜙2 𝜙3 𝑤2 𝑣2 +𝑤3 𝑣3 𝑣2 𝑣3

where 𝜇 is a simple volumetric ratio and 𝑤𝑖 and 𝜙𝑖 are the weight and volume fractions of species
𝑖, respectively. However, even if this expression is employed, the very low values of the retention
volume 𝑉𝑔 lead to very large error in 𝜒23 [13]. For these reasons, although IGC offers an avenue
for estimating the 𝜒-parameter, it presents prohibitively large uncertainty to be a practical method.
It is also possible to estimate 𝜒 using small angle neutron scattering (SANS) [21-23, 26].
This procedure measures the angle of refraction of neutrons through an experimental medium, in
this case a solution of two polymer species, while varying angle of incidence in a manner
analogous to small angle x-ray scattering. The interaction parameter between the two species can
be calculated using the following equation [21]:

12
 1 −1 −1
𝜒AB = [[𝜙𝑍A (𝑃(0) + 𝑁𝑄(0))] − (𝜙𝑍A )−1 − ((1 − 𝜙)𝑍B ) ] (10)
2

Here 𝜙 represents volume fraction of B in A, 𝑍𝐴 and 𝑍𝐵 are the degree of polymerization of the
two species, 𝑁 represents the total number of B polymer chains per unit volume, and 𝑃(𝑞) and
𝑄(𝑞) represent the intramolecular and intermolecular components of scattering, respectively. 𝑞
can be expressed using the equation
4𝜋 𝜃
𝑞= sin ( ) (11)
𝜆 2

where 𝜆 represents the wavelength of the neutrons and 𝜃 represents scattering angle. Ito, Russell,
and Wignall argue that 𝜙, 𝑍𝐴 , and 𝑍𝐵 are known from the composition of the solution and from
the synthesis of each species and assume that 𝑃(0) = 1. Therefore, extrapolating 𝑄(𝑞) at 𝑞 = 0
yields the 𝜒-parameter [21].
On the other hand, it is well established that 𝜒 varies with temperature. In particular, 𝜒 may
be expressed as the sum of a temperature-independent entropic term and a temperature-dependent
enthalpic term, as shown below [24, 27]:
𝐵
𝜒(𝑇) ≈ 𝐴 + (12)
𝑇

Thus, determining an empirical equation for 𝜒 of the form 𝜒 ≈ 𝐴 + 𝐵𝑇 −1 requires that

SANS be performed at each step in a range of temperatures and the resulting 𝜒-values be fitted as
a linear function of 𝑇 −1 . For this reason, although SANS produces useful miscibility data, the time
needed to prepare and test solutions at each desired temperature makes it a slow and quite
cumbersome method of determining 𝜒.
Revised Theory Through Molecular Simulation
In this study, we aimed at revising the theory by utilizing molecular modeling methods.
For this, the following procedure was implemented. To prepare molecules for miscibility analysis,
we built the molecular structures in Cerius2 (Molecular Simulations Inc., San Diego, CA, USA).
Following this, basic partial charge estimates and coarse geometry optimization were carried out
via the charge equilibration (QEq) method [28] and molecular mechanics [29], respectively.
Coarse geometry optimization via molecular mechanics in this stage improves convergence in the
next geometry optimization using the density functional theory (DFT) method.
After the coarse geometry optimization, the DFT geometry optimization was performed
using Jaguar [30] in Maestro (Schrödinger, New York, USA) with GGA PBE functional and the
6-31G** basis set to refine the molecular structures. The atomic partial charges were obtained

13
using Mulliken population analysis [31]. The self-consistent field procedure was performed with
a convergence threshold of 5×10-5 Hartree.
Next, molecular dynamics (MD) simulations were performed on these molecules via the
Forcite module of Materials Studio (Accelrys, San Diego, USA) to sample low-energy
conformations, which were employed for the miscibility analysis. These simulations were carried
out at 298 K with a time step of 1 fs for a total simulation time of 50 ps. It should be noted that
while 50 ps is nominally too short for molecular dynamics simulation, in this case the quite limited
conformational space of our small molecules renders the use of a longer timescale unnecessary. A
trajectory file was generated by saving the structure every 500 steps. The energy was calculated
using Dreiding force field [32] and an NVT ensemble with random initial velocities. These same
procedures were repeated for a second molecule with the same conditions. The process flow of the
initial preparation of one molecule is shown in Figure 7.
The Dreiding force field was selected because it has been extensively demonstrated in
previous work by the authors to be accurate and reliable for organic molecular systems [33-38].
Further, the simple implementation of force expressions makes the molecular dynamics
simulations proceed much more efficiently.
With molecular structures for molecules 1 and 2 stored in trajectory files, miscibility
simulations were performed using the Blends module of Materials Studio at 298 K. When
undertaking this calculation, 1×106~1×107 samples were collected with an energy bin width of
0.02 kcal/mol. In order to determine the coordination number of each pairing of molecules, 1×104
cluster samples were taken with 20 iterations per cluster. The 1,000 lowest-energy sample frames
were retained for analysis. Similar to the Forcite calculation, here the Dreiding force field was
employed to determine intermolecular interaction energies.
It should be noted that Blends module of Materials Studio has been observed to give results
with a large variation [39-41]. To combat this, we utilized post-processing and data analysis steps
developed by the authors which improved the accuracy of the resultant χ-values significantly. A
molecular mechanics task was executed in Perl within Materials Studio, as shown in Figure 8.
During this process, a secondary geometry optimization was performed on each individual
molecule and pair of molecules in the lowest-energy frames output by the Blends simulation,
which allowed us to determine the energies and Connolly volumes of both the individual molecules
and the pair of these two molecules.

14
 Once this was completed for each of the 1,000 lowest-energy frames, additional data
refinement was carried out, in which the interaction energies were normalized by the Connolly
volume and then averaged via Boltzmann statistics. These two steps capture a more realistic
snapshot of the interaction energy at any given point in time and fairly consider the effect that
molecular size has on the apparent interaction strength.
We propose this technique of Connolly volume normalization (CVN) since a direct
calculation of the interaction energies between molecules, which the Blends module does [42],
does not fairly assess intermolecular interaction. For example, two large molecules with weak
interactions can appear to have the same affinity for one another that two small molecules with
strong interactions have, which would lead to faulty conclusions about the miscibility of these two
species. Thus, in the expression for the Flory-Huggins χ-parameter between species 1 and 2 as
shown in equation (5), we suggest an alternative expression for the exchange energy between the
unmixed and mixed states, Δ𝑤12 :
∗
𝑉ref 𝐸12 ∗
𝐸11 ∗
𝐸22
Δ𝑤12 = [ (𝑍12 + 𝑍21 ) − ( 𝑍11 + 𝑍22 )] (13)
2 𝑉12 𝑉11 𝑉22

where 𝑉𝑖𝑗 refers to the enclosed Connolly volume of the combined pair of molecules 𝑖 and 𝑗. The
reference volume 𝑉ref and Boltzmann-averaged energy 𝐸 ∗ are given as
𝜙1 𝑉1 𝜙2 𝑉2
𝑉ref = + (14)
𝑛1 𝑛2

and
𝐸𝑘
1𝑁f
𝐸 ∗ = ∑𝑘=1 𝐸𝑖 𝑒 −𝑅𝑇 (15)
𝑄

respectively, where 𝑁f is the number of low-energy frames returned from the blends calculation
and 𝑄 is the partition function of the corresponding system of 𝑁f frames:
𝐸𝑖
𝑁
f
𝑄 = ∑𝑘=1 𝑒 −𝑅𝑇 (16)
Some discussion and justification of the new terms in equation (13) is warranted. The
interaction energy 𝐸𝑖𝑗 is weighted by the Boltzmann factor to become 𝐸𝑖𝑗∗ in order to capture the
statistical mechanical probability that would favor low-energy frames in a physical system.
Further, the interaction energy density 𝐸𝑖𝑗∗ ⁄𝑉𝑖𝑗 is fundamentally a size-mitigation term. The
∗
contribution of an interaction 𝐸11 to the overall exchange energy Δ𝑤12 may be overestimated if
the species 1 is particularly large. It would be unclear how much of that contribution arises from
the species’ size and how much arises from its intrinsic interaction strength. Thus, the energy is

15
normalized by the size of the molecule such that molecules only contribute to Δ𝑤12 based on their
intrinsic interaction strength.
The introduction of the volume normalization term renders the expression dimensionally
inconsistent with Δ𝑤12 , so some volume prefactor is necessary. The 𝜒-parameter is nominally a
segment interaction parameter[43], so to preserve this quality the volume prefactor should
represent the “average” segment volume. 𝑉1 𝑛1−1 gives the Connolly volume of an individual
segment of species 1 (and likewise for 𝑉2 𝑛2−1 ). Then 𝑉ref is computed as the volume average of
the individual segments, to obtain a representative “average” segment volume.
Finally, to further test the accuracy of this method with respect to experimental results on
the triblock copolymer system described in Figure 1, dissipative particle dynamics (DPD)
simulations were carried out using the 𝜒 -values obtained as detailed above. Using periodic
boundary conditions in a simulated box size of 30×30×30 at a reduced temperature of 1.0, the
simulation required a duration of 2.5×104 reduced DPD units with a time step of 0.05 to reach
equilibrium. Temperature and pressure were monitored to ensure that the system reached
equilibrium. The system was composed of 10% triblock copolymer and 90% water with a bead
density of 3.0. The reduced-ratio polymer topology used in this simulation was A5–B11–C13.
Repulsion interaction parameters 𝑎AB were calculated directly from the 𝜒-parameter results via the
expression
𝑎AB = 25 + 3.5𝜒AB (17)
which is valid for a bead density of 3, as outlined by Groot and Warren [25]. All dissipation
parameters were equal to 4.5.

Acknowledgements
We thank the U.S. Department of Energy, Office of Basic Energy Sciences, for funding this work
through Catalysis Contract DE-FG02-03ER15459. The authors are also grateful to Aaron Cohen
and Prof. Marcus Weck at New York University for their cryo-TEM work which validated our
computational procedures and for their continued advice and discussions on promising systems to
model.

References

16
[1] D. Roberson, C. M. Shemelya, E. MacDonald, R. B. Wicker Rapid Prototyping J. 2015, 21,
137-143.
[2] D. Drummer, K. Wudy, F. Kühnlein, M. Drexler Phys. Proc. 2012, 39, 509-517.
[3] K. E. Fagelman, J. T. Guthrie Surf. Coat. Int. Pt. B-Coat. Trans. 2006, 89, 1-14.
[4] A. H. Gröschel, F. H. Schacher, H. Schmalz, O. V. Borisov, E. B. Zhulina, A. Walther, A.
H. E. Müller Nat. Commun. 2012, 3.
[5] F. H. Schacher, A. Walther, M. Ruppel, M. Drechsler, A. H. E. Müller Macromolecules.
2009, 42, 3540–3548.
[6] L. Wang, J. Lin Soft Matt. 2011, 7, 3383-3391.
[7] S. Kubowicz, J.-F. Baussard, J.-F. Lutz, A. F. Thünemann, H. v. Berlepsch, A. Laschewsky
Angew. Chem. 2005, 44, 5262–5265.
[8] B. J. Chun, C. C. Fisher, S. S. Jang Phys. Chem. Chem. Phys. 2016, 18, 6284-6290.
[9] Y. Liu, Y. Wang, Y. Wang, J. Lu, V. Piñón, M. Weck J. Am. Chem. Soc. 2011, 133, 14260–
14263.
[10] P. J. Flory J. Chem. Phys. 1942, 10, 51-61.
[11] M. L. Huggins J. Phys. Chem. 1942, 64, 1712-1719.
[12] T. M. Aminabhavi, R. H. Balundgi J. Chem. Educ. 1986, 63, 581-581.
[13] Z. Y. Al-Saigh, P. Munk Macromolecules. 1984, 17, 803-809.
[14] C. Panayiotou Polymer. 2013, 54, 1621-1638.
[15] G. DiPaola-Baranyi, J. E. Guillet Macromolecules. 1978, 11, 228–235.
[16] M. H. Chiu, E. J. Prenner J. Pharm. Bioallied Sci. 2011, 3, 39-59.
[17] M. M. Knopp, N. E. Olesen, P. Holm, P. Langguth, R. Holm, T. Rades J. Pharm. Sci. 2015,
104, 2905-2912.
[18] M. E. S. R. Silva, V. Mano, R. R. J. Pacheco, R. F. S. Freitas J. Mod. Phys. 2013, 4, 45-51.
[19] H. Yang, S. Ricci, M. Collins Macromolecules. 1991, 24, 5218-5220.
[20] K. Yurekli, R. Krishnamoorti J. Polymer Sci. B. 2004, 42, 3204–3217.
[21] H. Ito, T. P. Russell, G. D. Wignall Macromolecules. 1987, 20, 2213-2220.
[22] T. P. Russell Macromolecules. 1993, 26, 5819-5819.
[23] F. S. Bates, G. D. Wignall Phys. Rev. Lett. 1986, 57, 1429-1432.
[24] M. Rubinstein, R. H. Colby, Polymer Physics, Oxford University Press, New York, 2003.
[25] R. D. Groot, P. B. Warren Journal of Chemical Physics. 1997, 107, 4423-4435.
[26] R. K. Koray Yurekli J. Polym. Sci. Part B: Polym. Phys. 2004, 42, 3204–3217.
[27] H. B. Eitouni, N. P. Balsara in Thermodynamics of Polymer Blends, Vol., Springer, New
York, 2007.
[28] A. K. Rappe, W. A. Goddard J. Phys. Chem. 1991, 95, 3358-3363.
[29] U. Burkert, N. L. Allinger, Molecular Mechanics, American Chemical Society, Washington,
DC, 1982.
[30] A. D. Bochevarov, E. Harder, T. F. Hughes, J. R. Greenwood, D. A. Braden, D. M. Philipp,
D. Rinaldo, M. D. Halls, J. Zhang, R. A. Friesner Int. J. Quantum Chem. 2013, 113, 2110-
2142.
[31] R. S. Mulliken Journal of Chemical Physics. 1955, 23, 1833-1840.
[32] S. L. Mayo, B. D. Olafson, W. A. Goddard J. Phys. Chem. 1990, 94, 8897-8909.
[33] G. Brunello, S. G. Lee, S. S. Jang, Y. Qi Journal of Renewable and Sustainable Energy.
2009, 1.
[34] G. F. Brunello, W. R. Mateker, S. G. Lee, J. I. Choi, S. S. Jang Journal of Renewable and
Sustainable Energy. 2011, 3.

17
[35] S. G. Lee, J. I. Choi, W. Koh, S. S. Jang, J. Kim, G. Kim IEEE Transactions on Components,
Packaging and Manufacturing Technology. 2011, 1, 1533.
[36] S. G. Lee, S. S. Jang, J. Kim, G. Kim IEEE Transactions on Advanced Packaging. 2010, 33,
333-339.
[37] S. G. Lee, W. Koh, G. F. Brunello, J. I. Choi, D. G. Bucknall, S. S. Jang Theoretical
Chemistry Accounts. 2012, 131, 1206.
[38] S. G. Lee, T. A. Pascal, W. Koh, G. F. Brunello, W. A. Goddard, S. S. Jang J. Phys. Chem.
C. 2012, 116, 15974–15985.
[39] A. Ahmadi, J. J. Freire Polymer. 2009, 50, 3871–3876.
[40] H. Rezaei, S. Amjad-Iranagh, H. Modarress Energy Fuels. 2016, 30, 6626–6639.
[41] K. Pajula, M. Taskinen, V.-P. Lehto, J. Ketolainen, O. Korhonen Mol. Pharmaceutics. 2010,
7, 795-804.
[42] R. L. C. Akkermans, N. A. Spenley, S. H. Robertson Mol. Simul. 2013, 39, 1153-1164.
[43] A. Chremos, A. Nikoubashman, A. Z. Panagiotopoulos The Journal of Chemical Physics.
2014, 140.

18
Table 1. Comparison of 𝜒-parameter for PECH-PMA pair between simulation and experiment
𝝌 (experimental, via IGC [13, 14])
𝜒min 𝜒avg 𝜒max 𝜒𝜎
349 K -0.09 0.252 1.470 0.437
398 K -0.07 0.146 1.010 0.259
𝝌 (simulated, via eq. 13)1
𝜒min 𝜒avg 𝜒max 𝜒𝜎
349 K 0.431 0.496 0.553 0.039
398 K 0.379 0.434 0.481 0.033
𝝌 (simulated, via traditional Flory-Huggins model)1
𝜒min 𝜒avg 𝜒max 𝜒𝜎
349 K 1.583 1.734 1.854 0.089
398 K 1.389 1.518 1.619 0.074
1The simulation data are averaged across 10 independent calculations, using dielectric screening.

19
Table 2. Parameters used to calculate 𝜒 at 295 K for the monomer case with dielectric screening. These parameters are defined in
greater detail in the Methodology section.
∗
𝚫𝒘𝟏𝟐 𝑬𝟏𝟏 𝑬∗𝟏𝟐 𝑬∗𝟐𝟐
𝝌 𝚫𝒘𝟏𝟐 𝑽𝟏𝟏 𝑽𝟏𝟐 𝑽𝟐𝟐 𝒁𝟏𝟏 𝒁𝟏𝟐 𝒁𝟐𝟏 𝒁𝟐𝟐
𝑽𝐫𝐞𝐟 𝑽𝟏𝟏 𝑽𝟏𝟐 𝑽𝟐𝟐

kcal kcal kcal kcal kcal

- Å3 Å3 Å3 - - - -
mol mol ∙ Å3 mol ∙ Å3 mol ∙ Å3 mol ∙ Å3

5.24 3.07 0.0168 -0.0229 -0.0181 -0.0219 160.0 383.4 615.7 5.55 3.69 8.24 5.55
AB 5.03 2.95 0.0162 -0.0215 -0.0177 -0.0222 160.2 383.1 616.0 5.56 3.70 8.21 5.57
5.41 3.17 0.0174 -0.0228 -0.0181 -0.0223 160.1 383.3 616.1 5.56 3.69 8.21 5.57
3.10 1.82 0.0084 -0.0230 -0.0219 -0.0250 615.3 462.0 314.7 5.56 7.07 4.34 5.55
BC 3.24 1.90 0.0087 -0.0223 -0.0216 -0.0250 615.9 462.0 314.8 5.59 7.05 4.32 5.55
3.29 1.93 0.0089 -0.0222 -0.0214 -0.0250 615.8 462.0 314.8 5.55 7.08 4.34 5.56
6.01 3.52 0.0318 -0.0250 -0.0179 -0.0227 314.8 235.2 160.2 5.55 6.54 4.71 5.55
CA 6.00 3.52 0.0318 -0.0250 -0.0179 -0.0227 314.8 235.1 160.2 5.56 6.55 4.72 5.56
6.03 3.53 0.0319 -0.0250 -0.0179 -0.0228 314.8 235.1 160.1 5.55 6.54 4.71 5.54

20
Table 3. Parameters used to calculate 𝜒 at 295 K for the monomer case without dielectric screening.
∗
𝚫𝒘𝟏𝟐 𝑬𝟏𝟏 𝑬∗𝟏𝟐 𝑬∗𝟐𝟐
𝝌 𝚫𝒘𝟏𝟐 𝑽𝟏𝟏 𝑽𝟏𝟐 𝑽𝟐𝟐 𝒁𝟏𝟏 𝒁𝟏𝟐 𝒁𝟐𝟏 𝒁𝟐𝟐
𝑽𝐫𝐞𝐟 𝑽𝟏𝟏 𝑽𝟏𝟐 𝑽𝟐𝟐

kcal kcal kcal kcal kcal

- Å3 Å3 Å3 - - - -
mol mol ∙ Å3 mol ∙ Å3 mol ∙ Å3 mol ∙ Å3
34.38 20.14 0.1108 -0.0811 -0.0322 -0.0283 157.2 378.5 613.1 5.54 3.68 8.27 5.55
AB 34.50 20.21 0.1112 -0.0817 -0.0318 -0.0270 157.3 378.4 612.9 5.53 3.67 8.26 5.57
33.72 19.76 0.1087 -0.0810 -0.0317 -0.0264 157.3 378.6 613.2 5.54 3.68 8.26 5.59
-2.44 -1.43 -0.0066 -0.0268 -0.0246 -0.0214 612.9 457.2 310.6 5.56 7.10 4.31 5.55
BC -3.24 -1.90 -0.0087 -0.0261 -0.0247 -0.0215 612.8 457.2 310.6 5.57 7.12 4.30 5.56
-2.16 -1.27 -0.0058 -0.0272 -0.0248 -0.0214 613.0 457.5 310.6 5.59 7.12 4.29 5.56
20.05 11.74 0.1062 -0.0215 -0.0316 -0.0811 310.5 232.2 157.5 5.57 6.56 4.72 5.55
CA 19.88 11.65 0.1053 -0.0214 -0.0317 -0.0811 310.6 232.2 157.2 5.56 6.56 4.72 5.54
19.97 11.70 0.1058 -0.0215 -0.0316 -0.0811 310.5 232.3 157.3 5.55 6.56 4.71 5.54

21
 Cl O O
CH3
O
n n

(a) (b)
O O CH
3

H OH
O n n

CH3
(c) (d)

n n

n
O F F
O O
O OH O
F F
H F
(e) (f) (g)

Figure 1. Chemical structures of species studied in this work: (a) poly(epichlorohydrin), (b) poly(methyl
acrylate), (c) polyethylene glycol, (d) poly(methyl methacrylate), (e) poly(acrylic acid) (block A), (f)
poly(6-(4-formylphenoxy) hexylacrylate) (block B), and (g) poly(pentafluorostyrene) (block C).

22
 1
0.8
0.6
χ (dimensionless)
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
350 360 370 380 390 400

Temperature (K)

Figure 2. Change in χ-parameter as a function of temperature for the PEG-PMMA blend from both small-
angle neutron scattering (red dashed line) and miscibility simulations (blue lines).

23
 7
6

χ (dimensionless)
5
4
3
AB
2
BC
1 CA
0
294 302 310 318 326
Temperature (K)

(a)

40

30
χ (dimensionless)

20
AB
10 BC
CA
0

-10
294 302 310 318 326
Temperature (K)

(b)

Figure 3. Change in χ-parameter as a function of temperature for each monomer model of triblock
copolymer (Figures 1f-1g) (a) with dielectric screening and (b) without dielectric screening.

24
 6

χ (dimensionless)
2
AB
0 BC
CA
-2

-4

-6
294 302 310 318 326
Temperature (K)

(a)

48

38
χ (dimensionless)

AB
28
BC
18 CA

-2

-12
294 302 310 318 326
Temperature (K)

(b)

Figure 4. Change in χ-parameter as a function of temperature for each dimer model of triblock copolymer
(Figures 1f-1g) (a) with dielectric screening and (b) without dielectric screening.

25
 8
6

χ (dimensionless)
4
2
AB
0 BC
CA
-2
-4
-6
294 302 310 318 326
Temperature (K)

(a)

25

20
χ (dimensionless)

15

10
AB
5 BC
CA
0

-5
294 302 310 318 326
Temperature (K)

(b)

Figure 5. Change in χ-parameter as a function of temperature for each trimer model of triblock
copolymer (Figures 1f-1g) (a) with dielectric screening and (b) without dielectric screening.

26
 (a)

(b)

Figure 6. DPD simulations results for micellization of triblock copolymer system: (a) Full micelle. (b)
Cross-sectional view. Please note that the block A is phase-separated as the shell (blue color) from the core
while Blocks B (green color) and C (red color) have no phase-separation in the core. The light blue color
indicates water.

27
 Build molecular Estimate coarse partial Perform coarse geometry
structure charges using QEq optimization using molecular
mechanics simulation

Perform fine geometry optimization and estimate

MD simulations to generate
partial charges using DFT method and Mulliken
most probable configurations
population analysis

Trajectory file for molecules 1 and 2

Figure 7. Preparation of molecules for miscibility analysis. Steps shown with a dashed border are
optional, depending on the conformational diversity of the particular species.

28
 Trajectory file for molecule 1 Miscibility simulation to obtain low-
Trajectory file for molecule 2 energy pairing configurations

Molecular mechanics simulation

Data analysis
for geometry optimization

χ-parameter for the molecules 1 and 2

Figure 8. Miscibility simulations are performed using the trajectories as inputs. The simulation results are
analyzed to obtain a value for 𝜒12 .

29