Sie sind auf Seite 1von 33

Computer Simulations of Nano-Scale

Phenomena Based on the Dynamic Density


Functional Theories
Applications of SUSHI in the OCTA System

Takashi Honda1 and Toshihiro Kawakatsu2


1
Japan Chemical Innovation Institute, and Department of Organic and Polymeric
Materials, Tokyo Institute of Technology, Ookayama, Meguro-ku, Tokyo
152-8552, Japan
2
Department of Physics, Tohoku University, Aoba, Aramaki, Aoba-ku, Sendai
980-8578, Japan

1 Introduction
Multicomponent polymeric materials such as polymer blends, polymer melts,
block copolymers, and polymer solutions, often show macro and micro phase
separations that generate domains of the length scales of 1–100 nm. These
polymeric materials with phase-separated domains are promising candidates
for functional materials in nano-technologies [1–3]. The characteristic length
scales of these domain structures are much larger than atomic length scales
but are still smaller than hydrodynamic length scales. For phenomena on the
micro and macroscopic length scales, there are well-established simulation
techniques. For example, microscopic phenomena on atomic length scales can
be dealt with using particle simulation techniques such as molecular dynamics
(MD) simulations. On the other hand, macroscopic hydrodynamic phenomena
are simulated with the finite element method (FEM). Compared to these
extreme length scales, there have been very few simulation techniques for the
intermediate length scales (the so-called mesoscopic scales) where the phase-
separated domains locate.
To study the phase separated domains on mesoscopic scales, very use-
ful tools are the density functional theories (DFTs) [4–7], where the phase-
separated domains are described in terms of the density distributions of
monomers and solvents. One of the important features of DFT is that it
can take into account the conformational entropy of polymer chains with any
molecular architectures, i.e. the monomer sequence and the branching struc-
tures. Using this DFT, one can predict the equilibrium state of polymeric
systems with mesoscopic structures, which is not easily accessible by the par-
ticle simulations or the fluid dynamics simulations. Therefore the DFT plays
462 Takashi Honda and Toshihiro Kawakatsu

an important role in bridging between microscopic particle simulations and


macroscopic fluid dynamics simulations.
The DFTs for polymers are categorized into two major classes. One is
based on the self-consistent field (SCF) theory where the free energy of the
system is evaluated using the path integrals of the chain conformations, and
the other is based on the phenomenological power series expansion (the so-
called Ginzburg–Landau (GL) expansion) of the free energy where the random
phase approximation (RPA) is used in evaluating the expansion coefficients.
These DFTs have successfully reproduced the experimental results of poly-
meric systems, especially for block copolymers. The phase diagrams of diblock
copolymers are predicted using the RPA [8] and the SCF theory [9,10], respec-
tively. For the dynamical aspects, the physical responses of the micro domains
in diblock copolymers to an external field, such as a shear flow or an electric
field, are simulated using a dynamical extension of the SCF theory [11, 12].
The order–order and disorder–order transitions in diblock copolymers are also
investigated with the DFTs [13–17].
Despite the success of the DFTs on mesoscopic phenomena in polymeric
systems, there have been very few general multi-purpose simulation tools for
DFTs. We have been developing the “Open Computational Tool for Advanced
material technology (OCTA)” system which is tailored for mesoscopic scale
simulations. All source codes of the OCTA system are open to the public,
and anyone can download these codes from the internet site and can use
the OCTA system for nano-scale studies. In the OCTA system, there are
several simulation engines. One of these engines is the “Simulation Utilities
for Soft and Hard Interfaces (SUSHI)” which is designed to implement DFTs
in general [18].
SUSHI can carry out static and dynamic calculations by using the SCF
theory for any melts and blends of polymers with any types of architectures.
With the static calculation, one can get the equilibrium state of the sys-
tem. On the other hand, with the dynamic calculation, one can reproduce
the time-dependent phase transition behavior such as the phase separation,
the adsorption, and the order–order transition between microphase separated
structures, and so on. The SCF theory presents a quantitatively accurate cal-
culation method to deal with these phenomena because the conformational
changes of polymers are taken into account in the calculations using the path
integral technique. Moreover, with the SCF theory, one can obtain the physical
quantities associated with the interfaces and domains such as the interfacial
tension, critical micelle concentration, and so on.
In return for the quantitative accuracy of the calculations, the SCF theory
demands much computer power. To reduce the computational cost, it is worth
introducing a DFT based on the series expansion of the free energy, such as GL
theory combined with the RPA. We will abbreviate this method as GRPA [19].
Although the range of validity of GRPA is limited only to the weak segregation
regime because of the truncation of the free energy expansion used in the GL
theory, the computational efficiency is much higher than SCF theory.
Computer Simulations of Nano-Scale Phenomena 463

The basic assumption commonly used among the DFTs is the Gaussian
chain approximation. In the next section, this Gaussian chain approximation
is briefly summarized. Then, we give an overview of the general features of
DFTs (SCF theory and GRPA) in Sect. 3. The technical details of the SCF
theory and the GRPA are given in Sects. 4 and 5, respectively. We also give
several examples of the simulation results at the end of these sections. Finally,
we discuss the perspective of the DFTs in Sect. 6.

2 Gaussian Chain Model


The DFTs can treat mixtures of polymers with any molecular architectures
and solvents. A polymer is a chain molecule composed of a sequence of repeat-
ing units. Here, the repeating unit is in general different from the monomer.
For example, an ethylene monomer becomes two methylene units in polyethy-
lene upon polymerization. In the DFTs, the minimal structural unit of the
polymer chain is called a segment, which consists of several repeating units.
The number of repeating units in a segment is chosen so that the consecutive
bond vectors connecting the segments are statistically independent. Such a
chain is referred to as a Gaussian chain because, as will be shown below, the
probability distribution of the chain conformation is described by a Gaussian
distribution. The average length of the bond vector connecting the segments
is called an effective bond length.
A Gaussian chain is also called an ideal chain, which means that the
segment-segment interactions are negligible. Modeling a polymer chain on a
coarse-grained scale with a Gaussian chain is called the Gaussian approxima-
tion. Due to the screening effect, this Gaussian approximation becomes valid
for highly concentrated polymer systems, i.e., polymer melts and blends, and
dense polymer solutions in which the concentration is semi-dilute or more [7].
To explain the statistical nature of the Gaussian chain, we focus on a
subchain which is a part of the whole ideal chain. As is shown in Fig. 1, it is
convenient to introduce a lattice model of the polymer chain where the lattice
constant corresponds to the effective bond length b of the ideal chain.
In Fig. 2, we show the possible positions of a nearest neighbor segment
that is bonded to the segment located at the center of the cubic cell. The
assumption of the ideal chain model is that the nearest neighbor segment can
take any nearest neighbor sites freely irrespective of the occupation of the
lattice sites by other segments.
Let us calculate the statistical weight of a subchain whose end segments
are specified by the indices i and i (i < i) and are fixed at the sites r  and r,
respectively. We denote the statistical weight of this subchain by Q(i , r  ; i, r).
Due to the chain connectivity, the statistical weight Q(i , r  ; i + 1, r) is related
to Q(i , r  ; i, r) through the following recurrence formula
1
Q(i , r  ; i + 1, r) = Q(i , r  ; i, r  ), (1)
z 
r
464 Takashi Honda and Toshihiro Kawakatsu

Fig. 1. Schematic figures of lattice models of ideal chains

Fig. 2. Possible positions of the nearest neighbor segment in a lattice model of a


polymer chain

where z is the number of nearest neighbor sites (z = 6 in the case shown in


Fig. 2) and r  is one of the possible positions of the nearest neighbor sites to
r. Equation (1) means that Q(i , r  ; i + 1, r) is the sum of the contributions
from the statistical weight of the i-th segment placed on the nearest neighbor
sites r  . Since each nearest-neighbor position is statistically equivalent in the
ideal chain model, each contribution is proportion to the probability 1/z.
From the discretized expression of the recurrence formula (1), let us derive
a continum formula. Subtracting Q(i , r  ; i, r) from both sides of (1) leads to
1
Q(i , r  ; i + 1, r) − Q(i , r  ; i, r) = Q(i , r  ; i, r  ) − Q(i , r  ; i, r). (2)
z 
r
Computer Simulations of Nano-Scale Phenomena 465

In the continuum limit, the left hand side becomes the derivative of Q(i , r  ; i, r)
with respect to the index i and the right hand side becomes the Laplacian
of Q(i , r  ; i, r). Thus (2) reduces to the following three dimensional (z = 6)
continuum equation:
∂ b2
Q(s , r  ; s, r) = ∇2 Q(s , r  ; s, r), (3)
∂s 6
where s is used as a continuous parameter instead of the discrete index i and
the b2 factor appears when the finite difference is replaced by the Laplacian.
Equation (3) can be regarded as a three dimensional diffusion equation
of Q(s , r  ; s, r) where the parameter s corresponds to time and the diffusion
constant is given by D = b2 /6. Equation (3) can be solved when the initial
condition at the end segment of the subchain is given. For example, if the end
segment is fixed at the position r  , the initial condition is given by

Q(s , r  ; s , r) = δ(r − r  ). (4)

The solution of the diffusion equation (3) under this initial condition is given
by

3 3/2 3|r − r  |2
Q(s , r  ; s, r) = 
exp − . (5)
2π|s − s |b 2 2|s − s |b2
As this statistical weight distribution is a Gaussian distribution, the ideal
chain obeys Gaussian statistics. Several physical properties of the Gaussian
chain are obtained as follows.
The probability distribution of the end-to-end vector R of a Gaussian
chain composed of N segments is obtained from (5) as [7]
3 3/2 3R2
P (R) = Q(0, 0; N, R) = exp − . (6)
2πN b2 2N b2
The average squared end-to-end vector < R2 > is given by
 ∞
< R >=
2
dR 4πR4 P (R) = N b2 , (7)
0

where we assume that the chain is isotropic, i.e. the probability distribution
function P (R) is a function only of R = |R|. The average squared end-to-end
vector of a subchain whose end segments are s and s segments at r and r  is
also given by

< (r − r  )2 >= |s − s |b2 . (8)

Using this relation, we can calculate the square of the radius of gyration
RG that is defined as the root mean square of the distance between each
segment and the center of mass of the chain. The result leads to [20]
466 Takashi Honda and Toshihiro Kawakatsu
 N  N
1
2
RG = ds ds < (r − r  )2 >
2N 2 0 0
 N  N
1 1 2
= ds ds|s − s |b2 = Nb . (9)
2N 2 0 0 6

The Fourier transformation of the distribution of the end-to-end vector of


a subchain is given by
 −∞ 3/2
3 3r 2
dr exp ( iq · r) exp −
−∞ 2π|s − s |b2 2|s − s |b2
b2
= exp ( − |s − s ||q|2 ). (10)
6
This expression gives the scattering intensity from a pair of segments s and
s . Thus the total scattering function from the whole chain, i.e., the structure
factor for a Gaussian chain, is given by [8].
 
1 N  N b2
0
S (q) = ds ds exp(− |s − s ||q|2 )
N 0 0 6
2N
= 2 (e−x − 1 + x), (11)
x
where x ≡ (1/6)N b2 |q|2 = RG 2
|q|2 and the function 2(e−x − 1 + x)/x2 in (11)
is called the Debye function.

3 An Overview of the SCF Theory and GRPA


A polymer chain in a polymer blend is not an ideal Gaussian chain if there
are strong segment–segment interactions between different types of segments
or strong segment–object interactions exist, where “object” means impurity
particles, such as the filler particles, walls and surfaces. For example, in a
system with a wall, the ideal Gaussian distribution of segments is distorted
in the vicinity of the wall.
The SCF theory is applicable to such systems under strong interactions.
The distorted chain conformations are evaluated with the use of the path in-
tegral technique, where the segment-segment and segment-object interactions
are treated as external fields acting on the ideal Gaussian chain. To obtain
such external fields, an iterative refinement procedure is necessary, which in-
creases the computational cost considerably.
On the other hand, there is no iteration procedure in the GRPA, and its
computation is much faster than the SCF theory. However, in the GRPA, the
free energy is expanded in power series around a uniform reference state, which
means that the way of introducing external objects into GRPA is not trivial.
Although the quantitative accuracy of the GRPA theory is poorer than that
Computer Simulations of Nano-Scale Phenomena 467

of the SCF theory, the GRPA is very useful in evaluating the phase behavior
and the mesoscopic structures of polymer melts and blends qualitatively.
In the subsequent sections, we explain the SCF theory and the GRPA in
more detail.

3.1 Definition of the Molecular Architecture of a Polymer and the


Characters of a Solvent

A polymer is composed of subchain(s). For example, a homopolymer consists


of a single subchain and a block copolymer consists of several subchains of
different kinds. Each segment has three important characteristics that are used
in SCF calculations. These characteristics are the chemical species (segment
type), effective bond length, and the specific volume. The molecular structure
of a polymer is defined by the topology of the connectivity of the subchains,
the type of the segments comprising each subchain, and the length of each
subchain.
We regard a solvent molecule as a special kind of segment without a bond.
Thus, a solvent has two characteristics, i.e. the chemical type and the specific
volume.
We use superscript indices p and s (p, s = 1, 2, 3, · · ·) to identify the type
of the polymer and the solvent, respectively, and we use a subscript index K
to identify the chemical type of the segment (including the solvent as a special
case).
Subscript indices i and j (i, j = 1, 2, 3, · · ·) are used to identify all subchains
in the system where we denote the number of subchains in a p-type ' polymer
by n(p) and the total number of subchains in the system by n = p n(p) . We
also define that the number of segments of the i-th subchain of the p-type
(p)
polymer is Ni and the total number of segments of the p-type polymer is
' (p)
N (p) = i Ni .
Figure 3 shows a typical topology of the subchain connectivity in a polymer
chain (p-type polymer). Each subchain is numbered and the chemical type
of the subchain is distinguished by the line style. A subchain is joined to
other subchains by chemical bonds. We will define an end connected to other
subchains as a junction point.

4 SCF Theory

We will use the SCF theory to derive the expressions of density distributions
of the segments and the solvents and the expression of the free energy.
The SCF theory relies on the mean-field approximation which assumes
that the same types of segments feel an identical external mean field. In other
words, the SCF theory adopts a single body picture where a single chain is
placed in an external mean field. Under this approximation, the fluctuations
468 Takashi Honda and Toshihiro Kawakatsu

Fig. 3. Typical p-type polymer

among the chains are neglected and we only assume that the same types of
polymers show the same statistical nature.

4.1 Solvents

The mean-field approximation can also be applied to solvents, which are


molecules consisting of a single segment. The density of a K-type solvent
at position r is in proportion to the Bolzmann factor as
(s) (s) (s)
φK (r) = CK exp[−βρK VK (r)], (12)
(s)
where β = 1/(kB T ), CK is a normalization constant, ρK is the specific vol-
(s)
ume of the K-type solvent, and VK (r) is an external potential acting on
(s)
the K-type solvent. The external potential VK (r) is determined so that the
constraints imposed on the system, such as the incompressibility condition,
(s)
are satisfied. The actual procedure to determine VK (r) is given in the next
(s)
section. The normalization constant CK depends on the statistical ensemble
we use. There are two typical ensembles. One is the canonical ensemble which
keeps the total volume fractions of each type of component in the system
constant. The other is the grand canonical ensemble in which the system is
assumed to be in equilibrium with the bulk reservoir, and therefore the total
volume fractions of each type of component fluctuate.

Canonical Ensemble

One can use the canonical ensemble for both finite systems surrounded by
impenetrable walls and infinite systems (surrounded by periodic boundaries
or reflective boundaries in the actual simulations). This canonical ensemble is
useful in calculating for example interfaces in an A/B polymer blend.
(s)
In this case, the normalization constant CK is given by
Computer Simulations of Nano-Scale Phenomena 469

(s) M (s) M (s)


CK =  (s)
= (s)
, (13)
dr exp[−βρK VK (r)] ZK

where M (s) is the total number of K-type solvent molecules in the system
(s)  (s)
and ZK = dr exp[−βρK VK (r)] is the partition function of the K-type
solvent.

Grand Canonical Ensemble

When one considers a system with a wall to which polymers are adsorbed
from a bulk reservoir, it is convenient to use the grand canonical ensemble
rather than the canonical ensemble.
(s)
In this case, CK is given by

(s) m(sb) m(sb)


CK = (b)
= (sb)
, (14)
exp[−βρK WK ] ZK

where the superscript b means that the parameter is evaluated in the bulk
uniform reservoir, i.e. m(sb) is the total number of K-type solvent molecules
(b)
per volume, WK is the segment-segment interaction potential acting on a K-
(sb) (b)
type solvent molecule, and ZK = exp[−βρK WK ] is the partition function
of the K-type solvent per volume.
(b)
The explicit expression of WK is given by
(b)
 (b)
WK = KK  φ̄K  , (15)
K

where φ̄K  is the total volume fraction of the K  -type segment in the bulk
(b)

reservoir, and KK  is the interaction energy between K-type and K  -type


segments.

4.2 Polymers

The statistical nature of the distribution of the segments of polymers is dif-


ferent from that of solvents because there are extra contributions from the
conformational degrees of freedom of the polymer chains.

Path Integral

We consider the i-th subchain composed of the K-type segments in the p-


type polymer. As we have already defined, the total number of segments in
(p)
this subchain is Ni . The K-type segment has two physical characteristics:
the effective bond length bK (the same as b in (3) ) and the specific volume
ρK . As there is an arbitrariness in the definition of the effective bond length
470 Takashi Honda and Toshihiro Kawakatsu

bK (or equivalently the definition of the size of a segment), we should fix this


(p)
parameter bK and the total number of segments in a subchain Ni simul-
taneously so that the radius of gyration of the i-th subchain RGi obeys the
2 (p)
relation (9), i.e., RGi = (1/6)Ni b2K . The specific volume ρK is needed to
relate the SCF calculation to the mass density of the repeating unit in the
real system.
We define the statistical weight of the i-th subchain as Qi (s , r  ; s, r) and
the external potential which acts on the segments of the i-th subchain as
Vi (r). When the values of bK , ρK and Vi (r) are given, (3) is modified as

∂  b2 
Qi (s , r  ; s, r) = K ∇2 − βρK Vi (r) Qi (s , r  ; s, r). (16)
∂s 6
This equation is no longer a simple diffusion equation because the total
amount of the diffusing material is not conserved due to the term con-
taining Vi (r). Equation (16) is called the Edwards equation. The solution
Qi (s , r  ; s, r) deviates from the ideal Gaussian statistics when Vi (r) becomes
large. The solution of (16) is the sum of Boltzmann weights for all possible
chain conformations, which is identified as the path of a diffusing particle in
(16). Therefore, Qi (s , r  ; s, r) is called the path integral.
Vi (r) in (16) can be divided into two contributions: a segment-segment
interaction potential and an external potential due to the constraints on the
segment density distributions, such as the incompressibility condition. These
potentials are derived as follows.
We denote the local segment density of the i-th subchain as φi (r) and
assume that the segment-segment interactions are short range interactions
that are proportional to the product of the pair of the local segment densities
with the interaction energy KK  introduced in (15). The contribution from
the segment-segment interaction potential is then given by

WK (r) = KK  φK  (r), (17)
K

where φK  (r) is the sum of the segment densities of the subchains composed
of K  -type segment and K  -type solvent densities as
 (s)
φK  (r) = φj (r) + φK  (r), (18)
j∈K 

where we assume that the size of a segment and the size of a solvent molecule
are the same.
The interaction parameter KK  can be related to the Flory–Huggins in-
teraction parameter χ. For an A/B binary mixture, we obtain
1
χ ≡ zβ{ AB − ( AA + BB )}. (19)
2
Computer Simulations of Nano-Scale Phenomena 471

The external potential for the constraint on the segment density distribu-
tion is a potential that keeps the profile of the segment density distribution
at a given profile. Therefore, this constraining potential should balance with
the chemical potential of the segment μi (r) so that the change in the pro-
files of the segment density do not occur. This leads to the result that the
constraining potential is given by −μi (r).
As a result, the self-consistent potential Vi (r) is given by

Vi (r) = WK (r) − μi (r). (20)

The same relation as (20) is also applicable to the K-type solvent as


(s) (s)
VK (r) = WK (r) − μK (r), (21)
(s)
where μK (r) is the chemical potential of the K-type solvent.
The potential Vi (r) and φi (r) are related to each other under the con-
straining condition and the total incompressibility condition [4, 7, 9, 21].
The total incompressibility condition is evaluated using the sum of the
segment densities φi (r) over the index i. The segment density at the position
(s, r) of the i-th subchain is calculated using two path integrals starting from
both ends of the subchain. As was explained in (4) for the ideal homopolymer
case, if the starting segment for the path integral calculation is a free end of
the chain, the initial condition for (16) is given by Qi (0, r ; 0, r) = δ(r−r ). On
the other hand, if the starting segment of the subchain is a junction point, the
initial condition for (16) is a product of the path integrals of the subchains
that meet at this junction point except for the subchain for which we are
calculating the path integral. When we denote these initial statistical weights
at both ends of the sumchain s = 0, Ni as qi0 (0, r ) and qi0 (Ni , r ), we
(p) (p)

can introduce the following integrated path integrals from both ends of the
subchain as

qi (s, r) ≡ dr qi0 (0, r )Qi (0, r ; s, r) (22)

q;i (s, r) ≡ dr qi0 (Ni , r )Qi (0, r ; s, r).
(p)
(23)

It is easy to confirm that qi (s, r) and q;i (s, r) also satisfy (16). We will call
these qi (s, r) and q;i (s, r) the normal direction integrated path integral and
the reverse direction integrated path integral, respectively.
When the end of a subchain is a free end, the initial value of the statistical
weight of this free end is uniform everywhere, and we have qi0 (0, r ) = 1 or
qi0 (Ni , r ) = 1. Otherwise the end is a junction point as shown in Fig. 3,
(p)

where the initial statistical weight of the end must be a product of all the
statistical weights of the other subchains connected to this junction point.
To calculate the statistical weight of a branching chain, the path integrals
of all subchains should be evaluated sequentially starting from the free ends.
472 Takashi Honda and Toshihiro Kawakatsu

We show, for example, a typical order of evaluation by the arrows with circled
numbers on the subchains in Fig. 4. An arrow with a circled number along a
subchain indicates the direction of the normal direction path integral and the
opposing arrow on the same subchain indicates the direction of the reverse
direction path integral. Having such an order of evaluation, one can avoid a
loss of evaluation of the statistical weights of all subchains.

Fig. 4. Typical p-type polymer with order of evaluation of path integrals

Segment Density Calculation

When the path integrals are evaluated, the segment density of the i-th sub-
chain at position r is given by

(p)
φi (r) = C (p) qi (Ni − s, r),
ds qi (s, r); (24)
s∈i−th subchain

(p)
where C is a normalization constant and the integral over s is taken over the
whole i-th subchain. The integrand on the right-hand side is the statistical
weight of the whole polymer chain under the condition that the segment s
is located at position r because it is a product of the path integrals of the
two parts of the whole chain meeting at this segment. To accumulate the
contributions from all the segments of the i-th subchain, the integrand is
integrated over s in the subchain. As in the case of the solvent, the expression
for C (p) depends on the ensembles we use.

Canonical Ensemble

The normalization constant C (p) is given by


M (p) M (p)
C (p) =  = , (25)
dr qi (s, r);
qi (Ni
(p)
− s, r) Z (p)
Computer Simulations of Nano-Scale Phenomena 473

where M (p) is the total number of p-type polymer chains to which the i-th
subchain belongs and Z (p) is the partition function of a p-type chain. Note
that this partition function is independent of the subchain index i in the
p-type chain.

Grand Canonical Ensemble

The normalization constant C (p) is given by

m(pb) m(pb)
C (p) =  = (pb) , (26)
exp[−β
(p)
ρK Ni WK (b)] Z
i∈p

where m(pb) is the total number of p-type polymers per volume in the bulk
reservoir [4], and Z (pb) is the partition function of the p-type polymer per
volume in the bulk reservoir.

Note that the C (p) for the grand canonical ensemble is a constant de-
termined only by the condition of the bulk reservoir while the C (p) for the
canonical ensemble changes depending on the internal domain structure in
order to keep the total volume fractions of polymers in the system.

4.3 Free Energy

The Helmholtz free energy of the system can be given as a function of the
(s)
segment densities of subchains {φi } and the solvent densities {φK } as follows
[4, 7, 9].
/ (s) 0 1  (p) 1  (p)
F {φi }{φK } = − M ln Z (p) + M ln M (p)
β p β p
1  (s) 1  (s)
− M ln Z (s) + M ln M (s)
β s β s

1 
+ dr KK  φK (r)φK  (r)
2 
K K
  (s) (s)
− drVi (r)φi (r) − drVK (r)φK (r). (27)
i K

In this expression, the first and second terms correspond to the confor-
mational entropy and the ideal mixing entropy of the polymers, respectively,
and the third and fourth terms correspond to those of the solvents. The fifth,
sixth and seventh terms are the interaction energy among the segments and
solvents, the interaction energy between the external fields and the segments,
and the interaction energy between the external fields and the solvents, re-
spectively. The most important factor is the first term that comes from the
474 Takashi Honda and Toshihiro Kawakatsu

conformational entropy of polymer chains. The advantage of SCF theory is


that this conformational entropy can be quantitatively evaluated.
A functional derivative of the free energy (27) with respect to φi (r) or
(s) (s)
φK (r) gives μi (r) or μK (r), which reduces to the relations (20) and (21).

4.4 Static Density Functional Theory

In the equilibrium state of the system, the chemical potential differences be-
tween any pair of segment species and solvents are spatially uniform so that no
mutual diffusion takes place. In this case, the total incompressibility condition
is imposed on the segment densities φi (r) instead of constraining each seg-
(s)
ment density profile. Therefore, all the constraint potentials μi (r) and μK (r)
are identical to the constraining potential due to the total incompressibility
condition. (We assume that all the segments and the solvent molecules have
the same volume.) Therefore we obtain
(s)
μi (r) = μK (r) = μe (r), (28)

where μe (r) is the chemical potential due to the total incompressible condi-
tion.

4.5 Dynamic Density Functional Theory

For dynamical simulations, we introduce a simple diffusion process for each


segment density. Here we assume Fick’s law of linear diffusion. Then, the
time evolution equation for the segment density of the i-th subchain is given
by [22, 23]


φi (r, t) = ∇ · [Li (r, t)∇{μi (r, t) + λ(r, t)}] + ξi (r, t), (29)
∂t
where Li (r, t) is the local mobility, λ(r, t) is a Lagrange multiplier for the
local incompressibility condition, and ξi (r, t) is the random noise due to ther-
mal fluctuation. The Lagrange multiplier λ(r, t) is determined using the local
incompressibility condition, and is expressed as
'
Li (r, t)∇μi (r, t)
∇λ(r, t) = − i ' . (30)
i Li (r, t)

The thermal fluctuation ξi (r, t) satisfies the following fluctuation–dissipation


relation

ξi (r, t)ξi (r  , t ) = 2kB T Lij (r, t)∇2 δ(r − r  )δ(t − t ). (31)

Here, Lij (r, t) on the right-hand side is given by the local mobility Li (r, t) as
Computer Simulations of Nano-Scale Phenomena 475

Li (r, t)Lj (r, t)


Lij (r, t) = Li (r, t)δij +  , (32)
Li (r, t)
i

where the second term corresponds to a Lagrange multiplier for the local
incompressibility condition.
The same type of equation as (29) can be used for the time evolution of
(s)
the solvent density φK (r, t).
The local mobility Li (r, t) in general depends on the environment. In the
simplest model, Li (r, t) is assumed to be a constant. This assumption is valid
when the segment density fluctuations are small compared to the average
densities. Another simple model is to assume that Li (r, t) depends on the
segment density as
Li (r, t) = L0 φi (r, t), (33)
where L0 is a constant. This model can be used for the Rouse dynamics or
the reptation dynamics [20, 24].
In the dynamic simulations, each subchain has a different chemical po-
tential that depends on its instantaneous density profile. For example, ho-
mopolymers with different chain lengths have different chemical potentials
even if these are composed of the same type of segments. Thus we have to
treat all μi (r, t) as independent variables. The total incompressibility condi-
tion imposes a constraint on these μi (r, t)’s. Therefore, the μi (r, t)’s are not
independent of each other.

4.6 SCF Method

In the SCF simulations, the value of Vi (r) in (16) is obtained by recursive


calculations. As is shown in Fig. 5, the quantities φi (r), Vi (r) (including the
(p)
solvents in general), qi (s, r) and q̃i (Ni − s, r) are mutually related and form
a self-consistent loop. To solve this self-consistent condition, one has to rely
on a recursive calculation, where Vi (r) is adjusted so that the total incom-
pressibility condition and the constraint on each segment density profile φi (r)
is fulfilled. When this recursive calculation converges, Vi (r) becomes the self-
consistent field.
This SCF scheme is basically a multi-dimensional optimization problem.
The total number of parameters that should be adjusted is in proportion to the
number of external fields at all mesh points in the system. (We use a spatial
mesh to solve the Edwards equation.) Therefore, when the system size is large,
a numerical method that uses second derivatives of the judge function with
respect to the adjusting parameters, such as the conjugate gradient method,
can not be used because the number of second derivatives is too large to
be treated numerically. To avoid this difficulty, we propose a straightforward
SCF method that is similar to the steepest descent method, where the second
derivatives are not required. Such a straightforward SCF method falls into
two categories.
476 Takashi Honda and Toshihiro Kawakatsu

Fig. 5. A recursive calculation scheme in the SCF simulations

One is the static method explained in Sect. 4.4, by which one can generate
the equilibrium state starting from an arbitrarily chosen initial state. For
example, a convenient choice of the initial condition is to give a small random
distribution of external potential and then introduce the segment-segment
interactions gradually into the SCF calculations. In this case, the evolution of
the system does not correspond to the actual time evolution.
The other is the dynamic method explained in Sect. 4.5 where the equilib-
rium state is achieved by tracing the time evolution of the segment densities
according to the diffusion process.

Static SCF Method

In the static calculation, all segments of the same type are subjected to the
same self-consistent field irrespective of the subchains they belong to. As was
(s)
discussed (28), we can assume that all μi (r)’s and μK (r)’s are equal to the
same field μe (r). In order to get the equilibrium state, a simultaneous updating
of WK (r) in (17) and μK (r) should be performed starting from an appropriate
initial distribution. This updating is realized in the following manner.
 
WK (r) −→ WK (r) + αW × 
KK K φ (r) − W K (r) (34)
K
  
μe (r) −→ μe (r) − αV × 1 − φK (r) , (35)
K

where αW and αV are appropriately chosen constants.


Equation(34) introduces the segment-segment interactions gradually. Equation(35)
guarantees that the total incompressibility condition is achieved when the it-
eration converges.

Dynamic SCF Method

As an initial condition for a dynamic calculation, segment density distributions


(s)
of subchains {φi (r)} and those for solvents {φK (r)} are given appropriately
under the total incompressibility condition:
Computer Simulations of Nano-Scale Phenomena 477
  (s)
φi (r) + φK (r) = 1, (36)
i K

where, on the left-hand side, the first term is the sum of the segment densities
of subchains and the second term is the sum of the solvent densities.
As in the case of the static SCF calculation (see (35)), the chemical po-
tential μi (r) for the given segment density φi (r) is evaluated by recursive
calculations as
$ %
μi (r) −→ μi (r) − αV × φi (r) − φi (r) , (37)
where φi (r) is the segment density calculated by the SCF calculation, which
should be equal to the given density distribution φi (r). As described in
Sect. 4.5, individual chemical potentials are independent in the dynamic cal-
culations, even if the subchains are composed of the same type of segments.
On the other hand, recursive calculations for solvents are not needed to
(s)
obtain the chemical potentials. Solving (12) with respect to μK (r) for the
(s)
given solvent density φK (r) gives

1  φ (r) 
(s)
(s)
μK (r) = log K (s) + WK (r)
βρK C
(s)
= −VK (r) + WK (r). (38)

4.7 Features of SUSHI for the SCF Calculations

In this section, we show the typical futures of SUSHI as a simulation tool for
the SCF theory. Many functions are implemented in SUSHI [18].

Polymer Structure

SUSHI can treat any types of polymers such as homopolymers, block poly-
mers, comb polymers, star (multi-miktoarm) polymers and ring polymers. Any
topological structures can be treated by SUSHI except for those that contain
many loops such as cross-linking polymers, because an SCF calculation on
many loop polymers requires extra recursive calculations.
SUSHI can also simulate a tapered polymer that consists of several types
of segments that are distributed along the chain with a gradient. By assigning
a path integral to each type of segment, SUSHI evaluates the path integral of
a tapered polymer [4].

Various Spatial Meshes

SUSHI can treat various types of system such as an infinite periodic system,
a system that is in equilibrium with a bulk reservoir, an isolated micelle or a
vesicle in a solvent, etc. For simulation of these various situations, the following
spatial meshes are prepared in SUSHI.
478 Takashi Honda and Toshihiro Kawakatsu

1. Regular and rectangular meshes are available to treat Cartesian coordi-


nate systems with a rectangular shape. Depending on the spatial struc-
tures of the target system, the user can select one, two or three dimensional
meshes.
2. A cylindrical mesh is available for simulating micelles, vesicles and so on.
This cylindrical mesh is a two dimensional mesh system with radial and
horizontal axes.
3. A spherical mesh is available for simulating micelles. This spherical mesh
is a one dimensional mesh system with only radial coordinate.

Boundary Conditions

In SUSHI, a periodic boundary condition, a Dirichlet boundary condition and


a Neumann boundary condition (reflective boundary condition) can be used.
The periodic boundary condition is used when simulating an infinite system
where the effects of the finite size of the simulation box should be eliminated
(or reduced). The periodic boundary condition is also useful for simulating
periodic domain structures in micro phase separations. The Dirichlet bound-
ary condition is used for the calculation of the path integrals at the surface
of a wall. On the other hand, the Neumann boundary condition is used in
the dynamic SCF simulation for the segment density fields at the surface of
a wall. This Neumann boundary condition is also used for the boundaries of
the cylindrical and the spherical meshes where the boundary of the simula-
tion box is not a realistic wall but a hypothetical boundary beyond which the
system continues.
By combining these Dirichlet and Neumann boundary conditions, one can
simulate such a system with a wall and a system in equilibrium with a bulk
reservoir.

Other Features

In SUSHI, one can perform simulations on the following situations.


• a solid surface onto which polymers are grafted
• a polydisperse homopolymer system (static calculations only)
• chemical reactions
• polyelectrolyte and electrostatic interactions [4]
• a system onto which a shear flow is imposed [11, 12]
• a system with hydrodynamic interactions.

4.8 Examples of the SCF Simulations on Block Copolymers

As demonstrations of the usefulness of the SCF simulations, we show several


simulation results of microphase separations of block copolymer systems.
Computer Simulations of Nano-Scale Phenomena 479

A block copolymer is composed of several blocks (subchains) whose ends


are connected together by covalent bonds. When the blocks are mutually
immiscible, a phase separation takes place. However, the connectivity of the
different blocks prevents a macro phase separation. As a result of these two
effects, a block copolymer shows a micro phase separation that has a periodic
domain structure with a characteristic length scale of the order of 10–100 nm.
AB type diblock copolymers are known to show typical microphase-
separated structures, and phase diagrams were obtained experimentally for
various types of diblock copolymers. These phase diagrams have a similarity
that does not depend on the chemical detail of the block copolymers. Such
a universal nature is a target of the DFTs. Theoretical phase diagrams were
actually obtained using RPA and SCF theories [8, 10]. Thin films of diblock
copolymers are expected to be used as nano-scale functional materials for data
storage, photonic materials, etc [3].
The microphase-separated domains become much more complex when the
numbers of blocks are increased. Making use of such complex nature, multi-
block copolymers are used as industrial functional materials. For example,
ABA type triblock copolymers form a network of micelles. One of these tri-
block copolymers is the SIS (S=styrene, I=isoprene) thermoplastic elastomer.
Since the glass transition temperature of PS is higher than room temperature
and that of PI is lower than room temperature, the PS micro domains perform
as physical cross-linking points and the SIS shows elasticity [25, 26].
Star triblock polymers (three-miktoarm polymers) have also received con-
siderable attention because of their fascinating microphase-separated struc-
tures [27–29].

Microphases of Diblock Copolymers

The stable region of each microphase of a diblock copolymer is specified by two


parameters, χN and f . Here, χ is the Flory–Huggins interaction parameter
between the blocks, N is the total number of segments in the chain, and f is
the block ratio defined as the number of segments in one of the two blocks
divided by N .
Figure 6 shows snapshot pictures of typical microphase- separated domains
of a diblock copolymer obtained by SCF simulations, where the isosurfaces of
the volume fraction of the segment of the minor phase are shown. Figs. 6a-e
correspond to body-centered cubic sphere (BCC), hexagonally packed cylin-
der (HEX), bicontinuous double gyroid (GYR), hexagonally perforated lamel-
lar (HPL), and lamellar structures (LAM), respectively. Except for the HPL
phase, all phases have their stable regions in the parameter space composed
of χN and f . The HPL phase is a metastable phase and two types of stacking
layers of hexagonally perforated lamellae are observed. The stacking layers are
characterized by the relative positions of the hexagonally packed holes. One is
an AB type and the other is an ABC type which is shown in Fig. 6d [30–32].
480 Takashi Honda and Toshihiro Kawakatsu

Fig. 6. Typical microphase separated structures of diblock copolymers

Fig.7 shows the details of the GYR structures, where a super cell (2×2×2
unit cells) is shown in Fig. 7a. Figure 7b shows a cross section of the GYR
structure, which shows a continuous “U” shaped structure that was actually
observed in TEM observations [33].

Fig. 7. Super cell and a cross section of a bicontinuous double gyroid structure. (a)
reprinted from [12]. Copyright (2006) American Chemical Society.

Epitaxial Transition of a Diblock Copolymer

Upon a structural phase transition between different types of micro phases, if


the lattice constant and the positions of the periodic domains are preserved,
Computer Simulations of Nano-Scale Phenomena 481

the transition is called ”epitaxial transition”. For example, the GYR domain
transforms into HEX domains aligned to the [111] direction of the GYR unit
cell. This transition is a first order transition where the nucleation and growth
processes of HEX domains are expected. Figure 8 shows an epitaxial transition
from GYR phase to HEX phase in a diblock copolymer (f = 0.35 and χN =
20.0) melt under a shear flow [12].
Figure 8a is a cross section of a unit cell of a GYR phase, observed from
the [111] direction. Under a shear flow and a sudden temperature change to
χN = 15.0, the GYR phase transforms into the HEX phase as shown in
Fig. 8b. The black arrows indicate the direction of the shear flow and the gray
arrows show the time flow. In this simulation, the expected nucleation of the
HEX phase is observed.

Fig. 8. Epitaxial transition from a GYR phase to a HEX phase in a diblock copoly-
mer melt. (b) reprinted from [12]. Copyright (2006) American Chemical Society.

Thin Films of Diblock Copolymers

Figure 9 shows two dimensional cross sections of thin films of an AB type


diblock copolymer. The block ratio of the A block of this diblock copolymer
is f = 0.34 and the interaction parameter is set to χN = 15.0. The upper
black regions are the air phase which is simulated using a poor solvent, and
the boundaries at the bottom are surfaces of solid substrates with which the
minor A-type segments have more affinity than the B-type segments. The
482 Takashi Honda and Toshihiro Kawakatsu

value of the interaction parameters between the segments and the surface are
assumed to be χsA = −0.5 and χsB = 0.0. The gray scales show the densities
of the A-type segments of the diblock copolymer. We can clearly observe a
HEX phase in these cross sections.
The interaction parameters between the air phase and the polymers are
changed in Fig. 9a-c. We call these interaction parameters χAV and χBV
where V denotes the air (Void). At the free surfaces between the film and the
air, both A-type and B-type segments exist when χAV = χBV as shown in
Fig. 9a. When χAV > χBV , the A-type segments are depressed at the interface
as shown in Fig. 9b. Contrarily, when χAV < χBV , the A-type segments cover
the surface as shown in Fig. 9c.

Fig. 9. Microphase separations in thin films of an AB diblock copolymer with a


block ratio f = 0.34. Figures a-c are for different interaction parameters between
the air and the segments of the diblock copolymer, i.e., (a) χAV = χBV = 5.0, (b)
χAV = 5.0 and χBV = 2.5, and (c) χAV = 2.5 and χBV = 5.0, where A and B refer
the two types of segments of the block copolymer and V refers to the air.

Local Segment Density Distributions in an ABA-type Triblock


Copolymer in HEX Phase

The degree of the physical cross-linking of SIS is related to the segment distri-
butions at the nano-scale. The static SCF method is an appropriate method
to investigate such segment distributions.
We calculated a two dimensional HEX structure of an ABA triblock
copolymer as shown in Fig. 10. Figure 10a is a HEX structure of an ABA
triblock copolymer. To this structure, we add the same ABA triblock copoly-
mer with a volume fraction of 0.0001 and confine one end of it to one of the
cylinder domains as shown in Fig. 10b by defining the statistical weight of
this end segment only in the specified cylinder domain. Figure 10b shows the
spatial distribution of this end segment.
Under this condition, we can obtain the distributions of the other segments
of the added ABA triblock copolymer. The results are shown in Fig. 10. The
Computer Simulations of Nano-Scale Phenomena 483

schematic picture of the polymer at the center of Fig. 10 shows the speci-
fied segments and Figs. 10b-f show the distributions of these segments. As is
shown in Fig. 10f, we can obtain the distribution of the other end segment.
The fractions of the distribution of this end segment in the central cylinder
domain and in the surrounding cylinder domains correspond to the fraction of
the loop conformations and the bridge conformations, respectively. From this
simulation data, one can predict the viscoelastic properties of the cylindrical
domains due to physical cross-linking [26].

Fig. 10. Segment distributions of an A10 B40 A10 triblock copolymer in a HEX struc-
ture with χN = 30.0. (a) The HEX structure of the ABA triblock copolymer, and
(b)-(f) segment distributions of the segment specified in the central schematic picture
under the condition that one end segment is constrained in the central cylindrical
domain. The details are described in the text.

Microphase Structures of Star Triblock Copolymers


We calculated two-dimensional microphase structures of star triblock copoly-
mers. The typical structures of these microphase structures are shown in
Fig. 11. In these simulations, a matching between the natural periodicity of
the domains and the side lengths of the simulation box has a crucial effect on
the simulation results. Thus, we optimized the side lengths of the simulation
box so that the free energy density of the systems is minimized.
The regions with different gray colors in these figures correspond to the
domains of individual types of segments. For this calculation, all the interac-
tion parameters χij are fixed to 0.6 and the numbers of the segments in the
individual blocks are changed. Figs. 11a-b show LAM and HEX phases and
Figs. 11c-d show complex structures.
484 Takashi Honda and Toshihiro Kawakatsu

Fig. 11. Microphase structures of star triblock polymers. (a) A20 /B20 /C8 , (b)
A20 /B20 /C20 , (c) A20 /B36 /C12 , (d) A20 /B40 /C40 with χij = 0.6.

5 A DFT for General Polymer Structures Using RPA


In this section, we explain the GRPA applied to the dynamic density func-
tional theory [19].

5.1 Free Energy Model of GRPA

The range of validity of the free energy obtained with the Ginzburg–Landau
theory is in general restricted to the weak segregation regime because the
expression of the Ginzburg–Landau free energy relies on a Taylor series ex-
pansion of the free energy with respect to the density fluctuations of segments
around the uniform mixed state [7] as
 
1  −1
F[{φi (r)}] = F0 [{φ̄i }] + dr dr Sij (r − r )δφi (r)δφj (r ) + · · ·
2β ij
(39)

1  −1
= F0 [{φ̄i }] + dqSij (q)δφi (q)δφj (−q) + · · · , (40)
2β ij

where φ̄i is the averaged segment density of the i-th subchain in the system
and the first term F0 [{φ̄i }] denotes the free energy of the reference uniformly
mixed state. The local segment density fluctuation of the i-th subchain at
position r is defined as
Computer Simulations of Nano-Scale Phenomena 485

δφi (r) ≡ φi (r) − φ̄i . (41)


−1
The quantity Sij (r − r ) in (39) is the inverse of the density-density auto-
correlation function between the segment density fluctuations of the i-th and
j-th subchains at positions r and r , respectively. It can be converted to the
−1
inverse of the scattering function Sij (q) in Fourier space as in (40) where
q means a scattering wave vector. The dots at the ends of (39) and (40)
indicate the higher order terms in the Taylor series expansion in δφi (r) or
δφi (q), respectively.
Bohbot-Raviv and Wang have combined the second term of (40) with the
Flory–Huggins free energy as follows [34].
  φi (r)   1
1 1
F[{φi }] = dr ln φi (r) − dr δφi (r)δφi (r)
β i
Ni 2β i
Ni φ̄i

1  −1
+ dqSij (q)δφi (q)δφj (−q), (42)
2β ij

where the first term on the right-hand side is the Flory–Huggins entropy
term of the subchains and the second term is the second order term in the
expansion of the entropy term in δφi (r). This second term is necessary to
cancel the double counting of the second order terms in the expansion of the
entropy term which is explicitly given by the third term.
Equation (42) is regarded as a modification of the right-hand side of (40).
The first term F0 [{φ̄i }] is eliminated because it is a constant and does not
affect the phase separation in the system. The higher order terms (dots) are
replaced by the entropy term of the Flory–Huggins free energy. The third
term on the right-hand side of (42) can be calculated using the auto-correlation
function of the segment density fluctuations. Thus it can account for the short
range interactions between segments, such as the interaction energy of the
Flory–Huggins free energy and the interfacial energy of a polymer blends [35],
and can also account for the long-range effects such as the interaction between
domains composed of different kinds of segments in a block copolymer [36].
Therefore the free energy (42) is a generalized free energy model for poly-
mers with arbitrary topological structures. For this model to be used, one has
−1
to find the expression of the Sij (q) for a given polymer.

5.2 Subchain Scattering Function


−1
We can derive an explicit expression for Sij (q) according to Leibler [8].
Vectors of the segment density fluctuations of each subchain and the ex-
ternal potentials acting on each segment density with the wave vector q in
the Fourier space are defined, respectively, as

x(q) ≡ {δφi (q)} (43)


u(q) ≡ {ui (q)}. (44)
486 Takashi Honda and Toshihiro Kawakatsu

The same vectors but in the real space are defined as:
x(r) ≡ {δφi (r)} (45)
u(r) ≡ {ui (r)}. (46)
The relation between x and u are given by
1
u(q) = − S−1 (q)x(q), (47)
β
where S−1 is the inverse of the scattering function matrix derived by the
RPA under the incompressibility condition. The detail of the derivation of
S−1 is described in Appendix A. The incompressibility condition sets the
n-th elements of u and x as dependent variables on the other elements, and
these elements are not used in the dynamic calculations.

5.3 Free Energy and Chemical Potential


By using the inverse of the scattering function matrix, S−1 , the free energy
of the system is given by
n  n 
1 φi (r) 1  1
F[{φi }] = dr ln φi (r) − dr δφi (r)δφi (r)
β i Ni 2β i Ni φ̄i

+ dr xT (r)u(r), (48)

where xT (r) is the transposed vector of x(r). The vector u(r) in the third
term of (48) is calculated by using the matrix S−1 (q) and the Fourier trans-
formation of x.
The last component, i.e. the n-th component of the segment densities is
calculated as follows:

n−1
φn (r) = 1 − φi (r) (49)
i

n−1
δφn (r) = − δφi (r). (50)
i

We can obtain the chemical potential of the i-th subchain under the in-
compressibility condition (49) and (50) as a functional derivative of (48) with
respect to φi (r) as
δF[{φi }]
μi (r) = (51)
δφi (r)
1  ln φi (r) 1 ln φn (r) 1 1 1 
= + − − − δφi (r) + ¯ δφn (r)
β Ni Ni Nn Nn φ̄i Ni φn Nn
+ui (r).
(52)
Computer Simulations of Nano-Scale Phenomena 487

5.4 Dynamic DFT for GRPA

Similar to the dynamic SCF simulations, the chemical potentials obtained in


(52) can be used in the dynamic DFT that was defined in Sect. 4.5. Let us
call this simulation a dynamic GRPA simulation.
It should be noted that in the dynamic GRPA simulation, one should
use the segment density dependent mobility given in (33). This is because
the Taylor series expansion in terms of the density fluctuation δφi (r, t) does
not guarantee the positiveness of the value of φi (r, t). Moreover, in order to
eliminate such a difficulty, the variable time mesh technique which varies the
simulation time step width to keep the positiveness of the value of φi (r, t) is
recommended for the dynamic GRPA.

5.5 Examples of the Dynamic GRPA

The dynamic GRPA method is appropriate to study the dynamical transition


of a large scale system. We will show the dynamical transition of the ABC
layers of the HPL phase as shown in Fig. 6d under thermal noise for example.

Dynamical Structure Change of HPL with ABC-Type Stacking

The HPL phase is observed as an intermediate metastable phase from the


HEX phase to the GYR phase under a deep quench [37]. To study the tran-
sition from HPL to GYR, we performed a dynamic GRPA simulation. The
initial phase of the system is the ABCABC six layers as shown in Figs. 12a
and d whose period is optimized with the static SCF method. The system is
composed of diblock copolymers with block ratio f = 0.35 and χN = 20.0.
For these parameters, the equilibrium phase is GYR phase [10]. We generate
the thermal noise in (29) in the following way:

ξi (r, t) = ∇v G (r, t), (53)

where v G (r, t) is a vector whose elements are white Gaussian random numbers
with a mean 0 and a standard deviation 0.001. We change the value of χN
from 20.0 to 15.0 for the dynamic GRPA calculation because the stable phase
for χN = 15.0 is the GYR phase.
The initial ABC layers are shown in Fig. 12a. Upon the change in χN , this
structure changes into the GYR as shown in Figs. 12b and c as time goes on.
The lower figures Figs. 12d to f correspond to the same structures as those in
the Figs. 12a to c, respectively, but observed from a different view point. In
the initial step, the HPL layers are undulating (Figs. 12b and e), and then one
of the ABC three layers deforms largely (Figs. 12c and f) to form a three-fold
junction (the circle in Fig. 12f). Such a three-fold junction is a characteristic
partial structure of the GYR phase. Therefore, we expect that this simulation
result shows a typical phase change from HPL to GYR.
488 Takashi Honda and Toshihiro Kawakatsu

Fig. 12. A transition from ABC type HPL phase to GYR phase under a thermal
noise, (a) and (d) are the initial structure at t = 0, (b) and (e) are the structure at
t = 2537, and (c) and (f) are the one at t = 7965

6 Perspectives

As shown in this chapter, the DFTs including the SCF theory and the GRPA
are useful theories for the study of phase behaviors of polymeric materials
because they cover the range from the small scale of polymer interfaces to the
large scale of macro phase separations of polymer blends. One of the important
factors in these DFTs is the quantitative evaluation of the conformational
entropy of the polymer chains using the statistical mechanics of Gaussian
chains.
It is difficult to reproduce the same simulation results as DFTs by us-
ing particle dynamics simulations such as the molecular dynamics simulations
and the dissipative particle dynamics simulations [38] because these parti-
cle dynamics simulations demand a great deal of CPU power to obtain the
full equilibrium structures of the system and to treat phenomena with large
systems such as macro phase separations. However, multi-scale modeling com-
bining coarse-grained molecular dynamics and the dynamic SCF theory has
potential for the study of nano-scale materials as done for the ABA triblock
copolymer [26].
On the other hand, the dynamic DFTs introduced in this chapter do not
include the rheological effects which are caused by the entanglements of poly-
mers. Only recently, several trials have started to introduce rheological effects
to the SCF simulation [39–41].
Computer Simulations of Nano-Scale Phenomena 489

As the SCF theory demands much CPU power when dealing with large
systems, a development of the technique of parallel computations is impor-
tant, and such a trial has already been started in the MesoDyn project [11].
To overcome this difficulty in a different manner, we proposed a new DFT
by combining the SCF and GRPA theories to accelerate the dynamic SCF
calculation without the loss of the quantitative accuracy of SCF theory [19].
The dynamic DFTs will be further extended and will be more useful by
introducing new theoretical features such as rheological effects and fluid dy-
namics. At the same time, the graphic user interface (GUI) combined with
the computer-aided design (CAD) systems for modeling meso-scale structures
will be enhanced, which promotes the ease of meso-scale simulations.
For the development of this field, an accumulation of knowledge and expe-
rience is important. For such a purpose, open source code projects which have
been developed in the field of quantum chemistry will be required in polymer
science. The OCTA project is one such attempt.

Acknowledgement. The authors thank Dr. A. Zvelindovsky for giving us the op-
portunity to contribute this chapter. This study is executed under the national
project on nanostructured polymeric materials, which has been entrusted to the
Japan Chemical Innovation Institute by the New Energy and Industrial Technology
Development Organization (NEDO) under METI’s Program for the Scientific Tech-
nology Development for Industries that Creates New Industries. This study is also
supported by a grant-in-aid for science from the Ministry of Education, Culture,
Sports, Science and Technology, Japan.

Appendices

A. Subchain Scattering Function by the RPA


Hereafter, the expressions x and u without arguments are only used for those
in the Fourier space.
We define a matrix of the segment interaction energies and a matrix of the
scattering functions between segments of subchains of ideal Gaussian chains
as

C ≡ {z ij } (54)
S0 ≡ {Sij
0
(q)}, (55)

where ij denotes the interaction energy between the segments of the i-th and
0
the j-th subchains and Sij (q) = 0 if the i-th and j-th subchains belong to
different polymers. The detail of S0 is described in Appendix 6.
We introduce a static pressure originating from the incompressibility con-
dition in the Fourier space, and denote it as u∗ (q). This static pressure u∗ is
the same for all subchains and can be derived as a function of u, S0 and C as
490 Takashi Honda and Toshihiro Kawakatsu

u∗ = f (u, S0 , C). (56)


Using the vectors and matrices mentioned above, the self-consistent equa-
tion is given by the RPA as [8]
x = −βS0 (u + Cx + u∗ e), (57)
where e is a vector whose elements are all unity. This equation is the funda-
mental equation which retains terms up to the second order in the expansion
of the free energy. Solving (57) with respect to x gives
x = −β(E + βS0 C)−1 S0 (u + u∗ e)
= −β(Bu + u∗ Be), (58)
where E is the identity matrix and B = (E + βS0 C)−1 S0 . Here the incom-
pressibility condition must be satisfied for x as

n
xi = 0. (59)
i=1

From (58) and (59), u∗ is obtained as



n 
n
Bij uj
∗ i=1 j=1
u = − n n . (60)

Bij
i=1 j=1

Substituting (60) into (58) leads to


x = −βSu, (61)
 
where S = B + B and the element of the matrix B is defined as
 n  n 
 
Bij  Bi j
 j  =1 i =1
Bij =− . (62)

n 
n
Bi j 
i =1 j  =1

If the inverse of S can be obtained in (61), we can get u = (−1/β)S−1 x


and this u can be used for the coefficient of the free energy. However, due
to the incompressibility condition (59), the matrix S is a singular matrix, i.e,
the rank of S is n − 1. This means that the segment densities of n subchains
are not independent. Then, we take the segment densities of the subchains
i ≤ n − 1 as the independent variables and the segment density of the n-th
subchain as a dependent one. In this case, we should calculate the inverse
matrix of the (n − 1) × (n − 1) block in the upper left part of S. Then, we
define an n × n matrix S−1 , where the (n − 1) × (n − 1) block in the upper
left part is filled with the inverse matrix obtained above and the n-th row and
column are filled with 0 elements.
Computer Simulations of Nano-Scale Phenomena 491

B. Scattering Functions from an Ideal Gaussian Chain


In a similar manner to which the Debye function in (11) is derived, the element
0
Sij (q) in (55) can be obtained. We denote the indices of the subchains in the
p-type polymer by i and j  . Then the scattering function from the p-type
polymer which is assumed as an ideal Gaussian chain is given by
(p)
1 2N 
Si0p
 i (q) =
(p)
S 0 (q)|N =N (p) = (p)i 2 (e−x − 1 + x) (63)
N i  N x
 N (p)  N (p)
1 i j  q2
Si0p
 j  (q) =
(p)
ds ds exp(− |s − s |b2K )
N 0 0 6
Ni Nj  e−z
(p) (p)
= (e−x − 1)(e−y − 1). (64)
N (p) xy
(p) (p) (p)
Here, S 0 (q)|N =N (p) is given by (11) with N = Ni , Ni and Nj  being the
i
number of segments of the i -th and j  -th subchains, respectively, and x, y,
and z are given by

x ≡ RGi
2
 |q| , y ≡ RGj  |q| , andz ≡ RGi j  |q| ,
2 2 2 2 2
(65)

where RGi and RGj  are the radii of gyration defined by (9) of the i -th
and j  -th subchains, respectively, and RGi j  is the radius of gyration of a
sequence of subchains that connects the i -th and j  -th subchains. Let us
explain the meaning of this RGi j  using the polymer shown in Fig. 13 as an
0p
example. When we calculate S15 (q) (scattering function between segments
of the 1st and 5th subchains) in Fig. 13a, it is enough to consider only the
subchains shown in Fig. 13b, i.e. the 1st, 5th, and a sequence of the 3rd
and 4th subchains. (The value of the effective bond length is also averaged
so that the average end to end distance is unchanged after the averaging.)
This simplification is due to the Gaussian chain approximation where each
sunbchain is statistically independent.

Fig. 13. Ideal Gaussian chain model for the calculation of the scattering functions
492 Takashi Honda and Toshihiro Kawakatsu

0
As a result, the explicit expression of the Sij (q) is obtained as a product
of the average volume fraction of the p-type polymer in the system φ¯p and
the Si0p
 j  (q) as

0
Sij (q) = φ¯p Si0p
 j  (q) (66)

p−1

i= n(p ) + i , (67)
p

where S0 is a symmetric matrix that satisfies Sij 0 0


(q) = Sji (q) and the diagonal
0
element Sii (q) is the self scattering function of the i-th subchain.

References
1. F. S. Bates, G. H. Fredrickson: Physics Today 52, 32 (1999).
2. I. W. Hamley: Block Copolymers; Oxford University Press: Oxford, 1999.
3. C. Park, J. Yoon, E. L. Thomas: Polymer 44, 6725 (2003).
4. G. J. Fleer, M. A. Cohen Stuart, J. M. H. M. Scheutjens, T. Cosgrove, B.
Vincent: Polymers at Interfaces; Chapman & Hall: London, 1993.
5. M. W. Matsen, F. S. Bates: Macromolecules 29, 1091 (1996).
6. M. W. Matsen: J. Phys. Cond. Matt. 14, R21 (2002).
7. T. Kawakatsu: Statistical Physics of Polymers - An Introduction; Springer-
Verlag, Berlin, 2004.
8. L. Leibler: Macromolecules 13 1602 (1980).
9. E. Helfand, Z. R. Wasserman: Macromolecules 9, 879 (1976). E. Helfand, Z.
R. Wasserman: Macromolecules 11, 960 (1978). E. Helfand, Z. R. Wasserman:
Macromolecules 11, 994 (1980).
10. M. W. Matsen, M. Schick: Phys. Rev. Lett. 72, 2660 (1994).
11. A. V. Zvelindovsky, G. J. A. Sevink, B. A. C. van Vlimmeren, N. M. Maurits,
J. G. E. M. Fraaije: Phys. Rev. E 57, R4879 (1998). A. V. Zvelindovsky, B.
A. C. van Vlimmeren, G. J. A. Sevink, N. M. Maurits, J. G. E. M. Fraaije: J.
Chem. Phys. 109, 8751 (1998). A. V. Zvelindovsky, G. J. A. Sevink, J. G. E.
M. Fraaije: Phys. Rev. E 62, R3063 (2000). A.V. Zvelindovsky, G. J. A. Sevink:
Europhys. Lett. 62, 370 (2003).
12. T. Honda, T. Kawakatsu: Macromolecules 39, 2340 (2006).
13. M. Laradji, A.-C. Shi, J. Noolandi, C. R. Desai: Macromolecules 30, 3242
(1997).
14. M. W. Matsen: Phys. Rev. Lett. 80, 4470 (1998). M. W. Matsen: J. Chem.
Phys. 114, 8165 (2001).
15. S. Qi, Z. G. Wang: Phys. Rev. Lett. 76, 1679 (1996). S. Qi, Z. G. Wang: Pys.
Rev. E 55, 1682 (1997). S. Qi, Z. G. Wang: Polymer 39, 4639 (1998).
16. M. Nonomura, T. Ohta: J. Phys. Soc. Jpn. 70, 927 (2001). M. Nonomura, T.
Ohta: Physica A 304, 77 (2002). M. Nonomura, T. Ohta: J. Phys.: Condens.
Matt. 15, L423 (2003).
17. K. Yamada, M. Nonomura, T. Ohta: Macromolecules 37, 5762 (2004).
Computer Simulations of Nano-Scale Phenomena 493

18. (http://octa.jp) T. Honda,H. Kodama, J.-R. Roan, H. Morita, S. Urashita, R.


Hasegawa, K. Yokomizo, T. Kawakatsu, M. Doi: SUSHI Users Manual ; OCTA:
Nagoya, Japan, 2004.
19. T. Honda and T. Kawahatsu Macromolecules 40, 1227 (2007).
20. M. Doi, S. F. Edwards: The Theory of Polymer Dynamics; Oxford Science:
Oxford, 1986.
21. K. M. Hong, J. Noolandi: Macromolecules 14, 727 (1981).
22. J. G. E. M. Fraaije: J. Chem. Phys. 99, 9202 (1993).
23. R. Hasegawa, M. Doi: Macromolecules 30, 5490 (1997).
24. P. G. de Gennes: Scaling Concepts in Polymer Physics Cornell University Press,
Ithaca, 1979.
25. A. Hotta, S. M. Clarke, E. M. Terentjev: Macromolecules 35, 271 (2002).
26. T. Aoyagi, T. Honda, M. Doi: J. Chem. Phys. 117, 8153 (2002).
27. R. Stadler, C. Auschra, J. Beckmann, U. Krappe, I. Voigt-Martin, L. Leibler:
Macromolecules 28, 3080 (1995).
28. W. Zheng, Z.-G. Wang: Macromolecules 28, 7215 (1995).
29. T. Gemma, A. Hatano, T. Dotera: Macromolecules 35, 3225 (2002).
30. I. W. Hamley, K. A. Koppi, J. H. Rosedale, F. S. Bates, K. Almdal, K.
Mortensen: Macromolecules 26 5959 (1993).
31. S. Foerster, A. K. Khandpur, J. Zhao, F. S. Bates, I. W. Hamley, A. J. Ryan,
W. Bras: Macromolecules 27, 6922 (1994).
32. M. E. Vigild, K. Almdal, K., Mortensen,I. W. Hamley, J. P. A Fairclough, A.
J. Ryan: Macromolecules 31, 5702 (1998).
33. D. A. Hajduk, P. E. Harper, S. M. Gruner, C. C. Honeker, G. Kim, E. L.
Thomas, L. J. Fetters: Macromolecules 27, 4063 (1994).
34. Y. Bohbot-Raviv, Z.-G. Wang: Phys. Rev. Lett. 85, 3428 (2000).
35. P. G. de Gennes: J. Phys. (Paris) 31, 235 (1970).
36. T. Ohta, K. Kawasaki: Macromolecules 19, 2621 (1986).
37. C-Y. Wang, T. P. Lodge: Macromol. Rapid. Commun. 23, 49 (2002).
38. R. D. Groot, T. J. Madden: J. Chem. Phys. 108, 8713 (1998).
39. T. Shima, H. Kuni, Y. Okabe, M. Doi, X.-F. Yuan, T. Kawakatsu: Macro-
molecules 36, 9199 (2003).
40. M. Mihajlovic, T. S. Lo, Y. Shnidman: Phys. Rev. E 72, 041801-1-26 (2005).
41. B. Narayanan, V. A. Pryamitsyn, V. Ganesan: Macromolecules 37, 10180
(2004).

Das könnte Ihnen auch gefallen