You are on page 1of 16

Journal of Colloid and Interface Science 283 (2005) 251266

www.elsevier.com/locate/jcis
Comparison between theoretical values and simulation results of viscosity
for the dissipative particle dynamics method
Akira Satoh
a,
, Tamotsu Majima
b
a
Department of Machine Intelligence and System Engineering, Faculty of System Science and Technology, Akita Prefectural University, 84-4, Ebinokuchi,
Tsuchiya-aza, Honjo 015-0055, Japan
b
Graduate School of Science and Technology, Chiba University, 1-33, Yayoicho, Inage-ku, Chiba 263-8522, Japan
Received 10 October 2003; accepted 13 September 2004
Abstract
In order to investigate the validity of the dissipative particle dynamics method, which is a mesoscopic simulation technique, we have derived
an expression for viscosity from the equation of motion of dissipative particles. In the concrete, we have shown the FokkerPlanck equation
in phase space, and macroscopic conservation equations such as the equation of continuity and the equation of momentum conservation.
The basic equations of the single-particle and pair distribution functions have been derived using the FokkerPlanck equation. The solutions
of these distribution functions have approximately been solved by the perturbation method under the assumption of molecular chaos. The
expressions of the viscosity due to momentum and dissipative forces have been obtained using the approximate solutions of the distribution
functions. Also, we have conducted nonequilibrium dynamics simulations to investigate the inuence of the parameters, which have appeared
in dening the equation of motion in the dissipative particle dynamics method. The theoretical values of the viscosity due to dissipative forces
in the HoogerbruggeKoelman theory are in good agreement with the simulation results obtained by the nonequilibrium dynamics method,
except in the range of small number densities. There are restriction conditions for taking appropriate values of the number density, number
of particles, time interval, shear rate, etc., to obtain physically reasonable results by means of dissipative particle dynamics simulations.
2004 Published by Elsevier Inc.
Keywords: Dissipative particle dynamics method; Mesoscopic approach; Nonequilibrium dynamics method; Transport coefcients; Viscosity
1. Introduction
The hydrodynamic solution for a three-particle system
has to be combined into a simulation method in order to take
into account multibody hydrodynamic interactions among
colloidal particles more precisely. However, it is highly dif-
cult even to solve analytically the ow eld for a three-
particle system [1,2], and, therefore, it seems to be almost
hopeless to obtain an analytical solution for a nonspherical
particle system such as a system composed of rodlike par-
ticles. Thus, we have the choice to take another approach
to develop a more precise simulation method for colloidal
dispersions. If both colloidal particles and solvent mole-
*
Corresponding author. Fax: +81-184-27-2188.
E-mail address: asatoh@akita-pu.ac.jp (A. Satoh).
cules are simulated, we can obtain the particle motion and
the solution of the ow eld simultaneously. However, if
molecules themselves are treated in simulations, we cannot
develop an effective simulation method, since the character-
istic time for the motion of colloidal particles is signicantly
different from that of solvent molecules. In other words, this
kind of molecular-dynamics-like method is unrealistic as a
technique for simulation of a colloidal dispersion from a
simulation time point of view. To circumvent this difculty,
the concept of uid particles seems to be promising. Hooger-
brugge and Koelman [3,4] have developed the dissipative
particle dynamics method in terms of uid particles. In their
theory, a uid is regarded as being composed of such virtual
particles, and the ow eld is solved by simulating these par-
ticles. The uid particles interact with each other, exchange
momentum, and should make random motion like Brownian
0021-9797/$ see front matter 2004 Published by Elsevier Inc.
doi:10.1016/j.jcis.2004.09.050
252 A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266
particles. From now on, such uid particles are called dissi-
pative particles.
For the simulation of the ow eld for a colloidal dis-
persion, the motion of colloidal particles is dependent on
the interaction between colloidal particles themselves, the
interaction between colloidal particles and dissipative par-
ticles, and the interaction between colloidal particles and
an applied eld such as a magnetic eld. Hence, in this
simulation technique, the solution of pair or three-body hy-
drodynamic interactions need not be combined with it, in
order to simulate colloidal particles. This is in high con-
trast with the ordinary simulation methods such as Stokesian
dynamics methods [511]. Multibody hydrodynamic inter-
actions among colloidal particles are automatically taken
into account from the interactions of colloidal particles with
dissipative ones. This is a signicant feature in the dissi-
pative particle dynamics method compared with ordinary
simulation methods. It is clear from these discussions that
the concept of the dissipative particle dynamics method is
straightforwardly applicable to a colloidal dispersion com-
posed of nonspherical particles, and, therefore, seems to be a
highly promising technique from a simulation point of view.
As will be shown later, the equation of motion for the dis-
sipative particle dynamics, which has been presented to date
[3,4,1214], is not unique, but has considerable freedom;
for example, the equation of motion has unknown constants
which are parameters representing the strength of conserv-
ative and dissipative forces. However, whether or not the
equation of motion can yield physically reasonable solu-
tions has not been sufciently claried [3,4,15], although
there is uncertainty in the equation of motion for dissipa-
tive particles. Hence, in order to apply the dissipative par-
ticle dynamics method widely to the usual ow problems
and physicsengineering problems of colloidal dispersions,
it is very important to investigate the inuence of such un-
certainty in the equation of motion on the solution of the
ow properties such as the ow eld and transport coef-
cients. This may be accomplished by comparing simulation
results with theoretical solutions concerning transport coef-
cients, which can be derived analytically from the equation
of motion.
The present study, therefore, attempts rst to derive an
analytical expression for viscosity using the equation of mo-
tion, secondly to obtain the numerical data in terms of dis-
sipative particle dynamics simulations, and nally to com-
pare the simulation results with the analytical solutions. In
the concrete, we show the FokkerPlanck equation in phase
space, and macroscopic conservation equations such as the
equation of continuity and the equation of momentum con-
servation. The basic equations of the single-particle and pair
distribution functions are derived using the FokkerPlanck
equation. These equations of the distribution functions are
approximately solved to obtain an expression for the viscos-
ity due to momentumand dissipative forces. Nonequilibrium
dynamics simulations have been conducted under circum-
stances of simple shear ow. Finally, these simulation results
are compared with the theoretical solutions to investigate the
inuence of the model parameters, which have appeared in
denitions of the equation of motion, number density, num-
ber of particles, time interval, etc., on this transport coef-
cient and also the internal structure of a dissipative particle
system. The present study may stimulate the development of
a new equation of motion for the dissipative particle dynam-
ics, which gives rise to more physically reasonable simula-
tion results.
2. Dynamics of dissipative particles
2.1. Kinetic equation of dissipative particles
The equation of motion for dissipative particles has to
obey some physical constraints in order for the solution
obtained by the dissipative particle dynamics method to
agree with that obtained by the NavierStokes equation. The
conservation of the total momentum of a system and the
Galilean invariance have to be satised by the equation of
motion as a minimum request [3,4,12,13].
We concentrate our attention on particle i and consider
the forces acting on this particle. The following three kinds
of forces may be physically reasonable as forces acting on
particle i: a repulsive conservative force F
C
ij
exerted by the
other particles, a dissipative force F
D
ij
providing a viscous
drag to the system, and a random or stochastic force F
R
ij
in-
ducing the thermal motion of particles. With these forces, the
equation of motion of particle i can be written as [1214]
(1) m
dv
i
dt
=

j (=i)
F
C
ij
+

j (=i)
F
D
ij
+

j (=i)
F
R
ij
,
in which m is the mass of particle i, v
i
is the velocity, and,
concerning the subscripts, for example, F
C
ij
is the force act-
ing on particle i from particle j.
Now we have to embody specic forms of the above-
mentioned forces. It may be reasonable to assume that the
conservative force F
C
ij
depends only on the relative position
r
ij
(= r
i
r
j
), and not on the particle velocities. An ex-
plicit expression for this force will be shown later. Since the
Galilean invariance has to be satised, the dissipative force
F
D
ij
and the random force F
R
ij
should not be dependent on
the position r
i
and velocity v
i
themselves, but should be
functions of the relative position r
ij
and relative velocity v
ij
(= v
i
v
j
), if necessarily. Additionally, it may be reason-
able to assume that the randomforce F
R
ij
does not depend on
the relative velocity but the relative position r
ij
alone. Fur-
thermore, we have to take into account the isotropy of the
particle motion and the decrease in the magnitude of forces
with particleparticle separation. The following expressions
for F
D
ij
and F
R
ij
satisfy these physical requirements [1214],
(2) F
D
ij
= w
D
(r
ij
)(e
ij
v
ij
)e
ij
,
(3) F
R
ij
=w
R
(r
ij
)e
ij

ij
,
A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266 253
in which r
ij
= |r
ij
|, and e
ij
is the unit vector denoting the
direction of particle i from particle j, expressed as e
ij
=
r
ij
/r
ij
. Also
ij
is a random variable inducing the random
motion of particles and has to satisfy the following stochastic
properties:
_

ij
(t )
_
= 0,
(4)
_

ij
(t )
i

j
(t

)
_
= (
ii

jj
+
ij

ji
)(t t

).
This variable satises the characteristic of the symmetry

ij
=
ji
, which ensures that the total momentum of the
system is conserved. The w
D
(r
ij
) and w
R
(r
ij
) are weight
functions to reproduce the decrease in forces with particle
particle separation, and and are constants specifying the
magnitude of forces. These constants can be related to the
system temperature and friction coefcient, which will be
shown later. It is seen from the expression for F
D
ij
that this
force acts in such a way as to relax the relative motion of
particles i and j. On the other hand, F
R
ij
acts as if inducing
the thermal motion of particles, but this force satises the
actionreaction formula, so that the conservation of the total
momentum of the system is ensured.
The substitution of Eqs. (2) and (3) into Eq. (1) leads to
the following equation:
m
dv
i
dt
=

j (=i)
F
C
ij
(r
ij
)

j (=i)
w
D
(r
ij
)(e
ij
v
ij
)e
ij
(5) +

j (=i)
w
R
(r
ij
)e
ij

ij
.
If this equation is integrated with respect to time over a small
time interval fromt to t +t , then the nite difference equa-
tions governing the particle motion in simulations can be
obtained as
(6) r
i
= v
i
t ,
(7)
v
i
=
1
m
_

j (=i)
F
C
ij
(r
ij
)

j (=i)
w
D
(r
ij
)(e
ij
v
ij
)e
ij
_
t
+
1
m

j (=i)
w
R
(r
ij
)e
ij
W
ij
,
in which
(8) W
ij
=
t +t
_
t

ij
d.
This W
ij
has to satisfy the following stochastic properties,
which arise from Eq. (4):
W
ij
= 0,
(9) W
ij
W
i

j
=(
ii

jj
+
ij

ji
)t .
If a new stochastic variable
ij
is introduced from the deni-
tion W
ij
=
ij

t , the third term in Eq. (7) can be written


as
(10)
1
m

j (=i)
w
R
(r
ij
)e
ij

ij

t ,
in which
ij
has to satisfy the following stochastic proper-
ties:
(11)
ij
= 0,
ij

j
=(
ii

jj
+
ij

ji
).
It is seen fromEq. (11) that the stochastic variable
ij
is sam-
pled froma uniformor normal distribution with zero average
value and unit variance [3,4,1214].
The equation of motion in Eq. (5) satises the conserva-
tion law of the total momentum of the system, but not the
energy conservation law. The equation of motion has to be
modied to satisfy the conservation of the system energy
[16,17].
The coefcients and and the weight functions
w
D
(r
ij
) and w
R
(r
ij
) cannot be determined independently,
because there are some relationships among these quanti-
ties. The derivation of these relationships will be shown in
the next section.
2.2. FokkerPlanck equation
We use the notation r
i
for the position vector of particle i
and v
i
for the velocity vector. Also, for simplicity of expres-
sion, the vector v
N
is used for describing the velocity vectors
of all the particles (that is, v
N
represents v
1
, v
2
, v
3
, . . .), and
similarly r
N
is for the position vectors of all the particles
(that is, r
N
represents r
1
, r
2
, r
3
, . . .). If the probability that
a particle position and velocity are found within the range
from (r
N
, v
N
) to (r
N
+ r
N
, v
N
+ v
N
) is denoted by
W(r
N
, v
N
, t ) dr
N
dv
N
, then the probability density function
W(r
N
, v
N
, t ) satises the following stochastic formula, i.e.,
the ChapmanKolmogorov equation,
W(r
N
, v
N
, t ) =
_ _
W(r
N
r
N
, v
N
v
N
, t t )
(12)
(r
N
r
N
, v
N
v
N
; r
N
, v
N
) d(r
N
) d(v
N
),
in which (r
N
r
N
, v
N
v
N
; r
N
, v
N
) is the transi-
tion probability for the state (r
N
r
N
, v
N
v
N
) trans-
ferring to another state (r
N
, v
N
) during a time interval t
and this does not depend on the time point t . If the veloc-
ity v
N
does not change appreciably during the small time
interval t , then can be written as
(r
N
r
N
, v
N
v
N
; r
N
, v
N
)
(13)
=(r
N
r
N
, v
N
v
N
; v
N
)(r
N
v
N
t ),
in which is the transition probability and is independent
of r
N
. The substitution of Eq. (13) into Eq. (12) leads to
the following different form of the ChapmanKolmogorov
equation:
W(r
N
+v
N
t , v
N
, t +t )
=
_ _
W(r
N
, v
N
v
N
, t )
(14) (r
N
, v
N
v
N
; v
N
) d(v
N
).
254 A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266
The FokkerPlanck equation in phase space (or the Chan-
drasekhar equation) can be derived by reforming Eq. (14) in
terms of Taylor series expansions with the nite difference
equations (6) and (7). Since the detailed derivation proce-
dure is shown in Ref. [18], we here show the nal result:
W
t
+

i
v
i

W
r
i
+

j
(i=j)
F
C
ij
m

W
v
i
=

j
(i=j)
e
ij


v
i
_
1
m
w
D
(r
ij
)(e
ij
v
ij
)W
_
+
1
2

j
(i=j)
1
m
2

2
w
2
R
(r
ij
)
(15) e
ij


v
i
_
e
ij


v
i
e
ij


v
j
_
W.
2.3. Final expression of equation of motion and its
nondimensional form
If a system composed of dissipative particles is in equi-
librium, then the equilibrium distribution W
eq
becomes the
canonical distribution for an ensemble which is specied by
a given particle number N, volume V , and temperature T .
With the notation U for the potential energy, W
eq
is written
as
(16) W
eq
=
1
Z
exp
_

1
kT
_
N

i=1
m
v
2
i
2
+U
__
,
in which Z is the partition function, and U is related to the
conservative force F
C
ij
by
(17) F
C
ij
=
U
r
ij
.
The equilibrium distribution W
eq
has to satisfy the Fokker
Planck equation in phase space in Eq. (15). Since the left-
hand side in Eq. (15) vanishes for the substitution of W
eq
,
the right-hand side also has to become zero. This is accom-
plished by the requirements
(18) w
D
(r
ij
) =w
2
R
(r
ij
),
2
= 2 kT,
in which k is Boltzmanns constant. The second equation is
the uctuationdissipation theorem for the dissipative parti-
cle dynamics.
We nowhave to determine the explicit expressions for the
conservative force F
C
ij
(r
ij
) and the weight function w
R
(r
ij
).
F
C
ij
(r
ij
) is a repulsive force preventing unphysical exces-
sive overlaps between particles, and w
R
(r
ij
) has to be set
so that interparticle forces decrease with increasing particle
particle separations. These requirements are satised by the
expressions
(19) F
C
ij
=w
R
(r
ij
)e
ij
,
(20) w
R
(r
ij
) =
_
1
r
ij
r
c
for r
ij
r
c
,
0 for r
ij
>r
c
,
in which is a constant representing the magnitude of the re-
pulsive forces. By substituting these equations into Eqs. (6)
and (7) with considering Eq. (10), the nal expression for
the equation of motion of the dissipative particle dynamics
can be written as
(21) r
i
= v
i
t ,
(22)
v
i
=

m

j (=i)
w
R
(r
ij
)e
ij
t

j (=i)
w
2
R
(r
ij
)(e
ij
v
ij
)e
ij
t
+
(2 kT )
1/2
m

j (=i)
w
R
(r
ij
)e
ij

ij

t ,
in which the expression for w
R
(r
ij
) has already been shown
in Eq. (20). The stochastic variable
ij
has to obey the sto-
chastic properties shown in Eq. (11) and is sampled from a
uniform or normal distribution with zero average and unit
variance, as already pointed out. The dissipative particle dy-
namics method uses Eqs. (21) and (22) to simulate the mo-
tion of dissipative particles.
Lastly, we showan example of the nondimensionalization
method used for actual simulations. To nondimensionalize
each quantity, the following representative values are used:
(kT/m)
1/2
for velocities, r
c
for distances, r
c
(m/kT )
1/2
for
time, (1/r
3
c
) for number densities, etc. With these represen-
tative values, Eqs. (21) and (22) can be nondimensionalized
as
(23) r

i
= v

i
t

,
(24)
v

i
=


j (=i)
w
R
(r

ij
)e
ij
t


j (=i)
w
2
R
(r

ij
)(e
ij
v
ij
)e
ij
t

+(2

)
1/2

j (=i)
w
R
(r

ij
)e
ij

ij

,
in which
(25) w
R
(r

ij
) =
_
1 r

ij
for r

ij
1,
0 for r

ij
>1,
(26)

=
r
c
kT
,

=
r
c
(mkT )
1/2
.
In these equations, the quantities with superscript
*
are di-
mensionless. Equations (23) and (24) clearly show that one
can start a dissipative particle dynamics simulation if appro-
priate values of the nondimensional parameters

and

are adopted, and also the time interval t

, the number den-


sity n

0
(=r
3
c
N/V , V is the system volume), and the particle
number N are properly specied. Results obtained by the
A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266 255
simulations should not depend on the nondimensional para-
meters which have been introduced in embodying the equa-
tion of motion; one has to be, therefore, careful in setting
appropriate values for these values in conducting dissipa-
tive particle dynamics simulations. For example, since time
is nondimensionalized by the representative value based on
the mean velocity v ( (kT/m)
1/2
) and the radius r
c
, the
magnitude of the dimensionless time interval t

may have
to be taken sufciently smaller than unity [19].
3. Transport equations
In the present section, the dissipative particle dynamics
method, the equation of continuity, and the momentumequa-
tion of the uid are derived using the equation of motion of
the dissipative particle dynamics.
If an arbitrary physical quantity A(r
N
, v
N
) is not de-
pendent on time explicitly, the time average A can be
expressed, using the probability density function W which
satises Eq. (15), as
(27) A =
_ _
AW(r
N
, v
N
, t ) dr
N
dv
N
,
in which
(28)
_ _
W(r
N
, v
N
, t ) dr
N
dv
N
= 1.
Hence, the time variation of A can be expressed, from
Eq. (15), as

t
A =
_ _
A
W
t
dr
N
dv
N
=
_ _
A
_

i
v
i

W
r
i

j
(i=j)
F
C
ij
m

W
v
i
+

j
(i=j)
e
ij


v
i
_
1
m
w
D
(r
ij
)(e
ij
v
ij
)W
_
+
1
2

j
(i=j)
1
m
2

2
w
2
R
(r
ij
)
(29)
e
ij


v
i
_
e
ij


v
i
e
ij


v
j
_
W
_
dr
N
dv
N
.
By applying integration by parts, this equation is written as

t
A =
_

i
v
i

A
r
i

j
(i=j)
F
C
ij
m

A
v
i

j
(i=j)

m
w
D
(r
ij
)(e
ij
v
ij
)e
ij

A
v
i
+
1
2

j
(i=j)
1
m
2

2
w
2
R
(r
ij
)
_
e
ij


v
i
_
(30)
_
e
ij


v
i
e
ij


v
j
_
A
_
.
Next we derive the equation of continuity and the momen-
tum equation of the uid using Eq. (30). If A is dened by
the equation
(31) A=
N

i=1
m(r r
i
),
then the average A, which is evaluated from Eq. (27), is
equal to a local density (r). That is,
(32) A =
_
N

i=1
m(r r
i
)
_
=(r).
If A is taken as
(33) A =
N

i=1
mv
i
(r r
i
),
then the average A is now
(34) A =(r)u(r),
in which u(r) is a macroscopic uid velocity at position r.
The substitution of Eq. (31) into Eq. (30) leads to the fol-
lowing expression:
(35)

t
+

r
(u) = 0.
This is no other than the equation of continuity.
Next the substitution of Eq. (33) into Eq. (30) leads to the
following equation:

t
(u) =
_

i
mv
i


r
i
_
v
i
(r r
i
)
_
+

j
(i=j)
F
C
ij


v
i
_
v
i
(r r
i
)
_

j
(i=j)
w
D
(r
ij
)(e
ij
v
ij
)
_
e
ij


v
i
_
v
i
(r r
i
)
_
_
+
1
2

j
(i=j)
1
m

2
w
2
R
(r
ij
)
_
e
ij


v
i
_
(36) e
ij

_

v
i
_
v
i
(r r
i
)
_


v
j
_
v
j
(r r
j
)
_
_
_
.
Thus, by taking into account Eq. (2), and conducting a simi-
lar reformation, Eq. (36) can be simplied as
256 A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266

t
(u) =
_


r

_

i
mv
i
v
i
(r r
i
)
_
(37) +

j
(i=j)
_
F
C
ij
+F
D
ij
_
(r r
i
)
_
.
By taking account of the relation F
D
ij
=F
D
ij
e
ij
, the following
equation can be obtained:

j
(i=j)
F
D
ij
(r r
i
)
=
1
2

j
(i=j)
_
F
D
ij
(r r
i
) +F
D
ji
(r r
j
)
_
(38) =
1
2

j
(i=j)
F
D
ij
e
ij
_
(r r
i
) (r r
j
)
_
.
If the following relationship concerning the Dirac delta func-
tion is taken into account,
(r r
j
) =(r r
i
+r
ij
)
(39)
=(r r
i
) +

r

_
e
ij
r
ij
_
0
(r r
i
+

e
ij
) d

_
,
then Eq. (38) can be reformed further as

j
(i=j)
F
D
ij
(r r
i
)
(40)
=
1
2

j
(i=j)

r

_
F
D
ij
e
ij
e
ij
r
ij
_
0
(r r
i
+

e
ij
) d

_
.
By conducting a similar reformation concerning F
C
ij
, Eq. (37)
can nally be written as
(41)

t
(u) =

r
(uu) +

r
(
K
+
U
),
in which

K
=
_

i
m(v
i
u)(v
i
u)(r r
i
)
_
,
(42)

U
=
1
2
_

j
(i=j)
_
F
C
ij
+F
D
ij
_
e
ij
e
ij
r
ij
_
0
(r r
i
+e
ij
) d
_
.
If the equation of continuity in Eq. (35) is taken into consid-
eration, Eq. (41) reduces to the momentum equation of the
uid,
(43)
_
u
t
+u
u
r
_
=

r
(
K
+
U
),
in which
K
and
U
are stress tensors due to the particle
momentum and the forces acting between particles, respec-
tively. It has now been shown that the equation of motion of
the dissipative particle dynamics gives rise to the momen-
tum equation of the uid. The virial equation of state can be
derived by evaluating directly
K
and
U
in Eq. (42).
4. Expressions for transport coefcients
4.1. Distribution functions
The local number density n(r, t ) at position r at time t
can be expressed using W(r
N
, v
N
, t ) as [18]
n(r, t ) =
_
N

i=1
(r r
i
)
_
=
_

_

i
(r r
i
)W(r
N
, v
N
, t ) dr
N
dv
N
=N
_

_
W(r, r
2
, . . . , r
N
, v
N
, t )
(44) dr
2
. . . dr
N
dv
N
.
Similarly, the pair correlation function g(r, r

, t ) can be writ-
ten as [18]
g(r, r

, t )
=
1
n
2
0
_

_ N

i
N

j
(i=j)
(r r
i
)(r

r
j
)
W(r
N
, v
N
, t ) dr
N
dv
N
=
N(N 1)
n
2
0
(45)

_

_
W(r, r

, r
3
, . . . , r
N
, v
N
, t ) dr
3
. . . dr
N
dv
N
,
in which n
0
is the mean number density, given by n
0
=
N/V , as dened before.
If the distribution function f (r, v, t ) is dened as [14]
(46) f (r, v, t ) =
_
N

i=1
(r r
i
)(v v
i
)
_
,
f (r, v, t ) can be written, using W(r
N
, v
N
, t ), as
(47)
f (r, v, t ) =N
_

_
W(r, r
2
, . . . , r
N
, v, v
2
, . . . , v
N
, t )
dr
2
. . . dr
N
dv
2
. . . dv
N
.
By comparing Eq. (47) with Eq. (44), it is seen that there is
a relationship between f (r, v, t ) and n(r, t ):
(48) n(r, t ) =
_
f (r, v, t ) dv.
A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266 257
Similarly,
(49) n(r, t )u(r, t ) =
_
N

i
v
i
(r r
i
)
_
=
_
vf (r, v, t ) dv.
Next, if we dene the pair distribution f
(2)
(r, r

, v, v

, t )
by the equation [14]
f
(2)
(r, r

, v, v

, t )
=
_
N

i=1
N

j=1
(i=j)
(r r
i
)(r

r
j
)(v v
i
)(v

v
j
)
_
=N(N 1)
(50)

_

_
W(r, r

, r
3
, . . . , r
N
, v, v

, v
3
, . . . , v
N
, t )
dr
3
. . . dr
N
dv
3
. . . dv
N
,
then the pair distribution can be related to the pair correlation
function g(r, r

, t ) as
(51) n
2
0
g(r, r

, t ) =
_ _
f
(2)
(r, r

, v, v

, t ) dv dv

.
It is clear from Eq. (30) that the average of an arbi-
trary quantity A(r
N
, v
N
) can be evaluated using f (r, v, t )
and f
(2)
(r, r

, v, v

, t ) without obtaining the expression for


W(r
N
, v
N
, t ). Hence, we here rst show the basic equations
for f and f
(2)
, and then solve the equations analytically to
obtain the approximate solutions.
If we set A =

i
(r r
i
)(v v
i
) in Eq. (30), and re-
formthe equation with the formulae of vector operations, the
following equation can be obtained:

t
f (r, v, t ) +v

r
f (r, v, t )
=

m
_ _
w
D
(R)

R
:

v
_
(v v

)f
(2)
(r, r +R, v, v

, t )
_
dRdv

+

2
2m
2
_ _
w
2
R
(R)

R

R
(52) :

2
vv
f
(2)
(r, r +R, v, v

, t ) dRdv

.
It has been assumed in deriving this equation that there are
no conservative forces. In Eq. (25),

R is the unit vector,
denoted by

R = R/|R|. It is seen from Eq. (52) that the
pair distribution function f
(2)
(r, r

, v, v

, t ) is necessary for
obtaining the solution of f (r, v, t ), and, similarly, the multi-
body distribution functions more than the pair distribution is
required to get the solution of f
(2)
. This clearly shows that
Eq. (52) is not closed, but a hierarchy equation. To close the
equation for f , the following assumption of molecular chaos
is introduced [14]:
(53) f
(2)
(r, r

, v, v

, t ) f (r, v, t )f (r

, v

, t ).
This assumption theoretically enables us to solve the equa-
tion of the distribution function. The substitution of Eq. (53)
into Eq. (52) leads to the following equation:

t
f (r, v, t ) +v

r
f (r, v, t )
=

m
_ _
w
D
(R)f (r +R, v

, t )

R

R
:

v
_
(v v

)f (r, v, t )
_
dRdv

+

2
2m
2
_ _
w
2
R
(R)f (r +R, v

, t )

R
(54) :

2
vv
f (r, v, t ) dRdv

.
The validity of solutions which are obtained from Eq. (54)
is strongly dependent on the physical reasonability of the
assumption of molecular chaos.
The nondimensional formof Eq. (54) may be more under-
standable to be solved analytically. According to the nondi-
mensionalization method, which has been shown in Sec-
tion 2.3, Eq. (54) is nondimensionalized as

(r

, v

, t

) +v

(r

, v

, t

)
=

0
_ _
w
D
(R

)f

(r

+R

, v

, t

R
:

v

_
(v

v

)f

(r

, v

, t

)
_
dR

dv

+

0
_ _
w
D
(R

)f

(r

+R

, v

, t

R
(55) :

2
v

(r

, v

, t

) dR

dv

,
in which the distribution function f has been nondimension-
alized as f

= (1/n
0
)(kT/m)
3/2
f . If we set = 1/

0
and assume to be much smaller than unity, the perturba-
tion method is applicable. Hence, f

is expanded, with the


perturbation parameter , as
f

(r

, v

, t

) =f

0
(r

, v

, t

) +f

1
(r

, v

, t

)
(56) +
2
f

2
(r

, v

, t

) + .
By substituting this equation into Eq. (55) and neglecting the
higher-order terms, we can obtain the following expression:

_

t

0
+v

0
_
+
2
_

t

1
+v

1
_
=
_ _
w
D
(R

)
_
f

0
(r

+R

, v

, t

)
+f

1
(r

+R

, v

, t

)
_

R
:

v

_
(v

v

)
_
f

0
(r

, v

, t

)
+f

1
(r

, v

, t

)
__
dR

dv

+
_ _
w
D
(R

)
_
f

0
(r

+R

, v

, t

)
+f

1
(r

+R

, v

, t

)
_

R
258 A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266
(57)
:

2
v

_
f

0
(r

, v

, t

) +f

1
(r

, v

, t

)
_
dR

dv

.
The equation for the zeroth order of can be written, from
this equation, as
_ _
w
D
(R

)f

0
(r

+R

, v

, t

R
:

v

_
(v

v

)f

0
(r

, v

, t

)
_
dR

dv

+
_ _
w
D
(R

)f

0
(r

+R

, v

, t

R
(58) :

2
v

0
(r

, v

, t

) dR

dv

= 0.
Similarly, the equation for the rst order of can be ex-
pressed as

0
+v

0
=
_ _
w
D
(R

)
_
f

0
(r

+R

, v

, t

R
:
_

v

(v

v

)f

1
(r

, v

, t

)
_
+f

1
(r

+R

, v

, t

R
:

v

_
(v

v

)f

0
(r

, v

, t

)
_
_
dR

dv

+
_ _
w
D
(R

)
_
f

0
(r

+R

, v

, t

R
:

2
v

1
(r

, v

, t

)
+f

1
(r

+R

, v

, t

R
(59) :

2
v

0
(r

, v

, t

)
_
dR

dv

.
Equation (58) can straightforwardly be solved as
(60) f

0
(r

, v

, t

) =n

_
1
2
_
3/2
exp
_

1
2
(v

)
2
_
.
By reforming Eq. (59) with the formulae of vector opera-
tions, the following equation is obtained:

0
+v

0
=
_
n

(r

+R

, t

)w
D
(R

R
(61) :
_

v

_
(v

)f

1
(r

, v

, t

)
_
+

2
v

1
(r

, v

, t

)
_
dR

.
If it is assumed that n

(r

+R

, t

) (=n/n
0
) is independent
of the position and is constant, and the integration is carried
out using the polar axis coordinates of

R, then Eq. (61) can
be simplied as

0
+v

0
=
[w]
3
_

v


_
(v

)f

1
(r

, v

, t

)
_
(62) +

v

1
(r

, v

, t

)
_
,
in which [w] is dened as
(63) [w] =
1
_
0
w
D
dR

=
1
_
0
w
D
4R
2
dR

.
The left-hand side of Eq. (62) can nally be reformed, by
taking account of Eq. (60), as

0
+v

0
(64) =
1
2
f

0
U

:
_

r

+
_

r

_
t
_
,
in which the notation U

, dened by U

= (v

), has
been used. Also, the superscript t means a transposed ten-
sor. Now we have prepared for solving Eq. (62) to get the
solution of f

1
.
As the rst trial to nd the solution of f

1
, we substitute
f

0
U

into f

1
in Eq. (62) to get the following expression
for the right-hand side of Eq. (62):
(65)
1
3
[w]
_
2U

0
+2If

0
_
.
As the second trail, we substitute f

0
U
2
I

into f

1
in
Eq. (62) to get the following equation for the right-hand side
of Eq. (62):
(66)
1
3
[w]
_
2U
2
f

0
I +6f

0
I
_
.
As the third trial, we substitute f

0
I

into f

1
in Eq. (62) to
obtain zero for the right-hand side of Eq. (62). Finally, if the
expression for f

1
is assumed to be
(67) U

0

1
3
U
2
f

0
I,
then the right-hand side of Eq. (62) becomes the following
expression:
(68)
2
3
[w]
_
U

0

1
3
U
2
f

0
I
_
.
After all, from these considerations, the solution of Eq. (62)
can straightforwardly be seen to be the following expression:
f

1
=
3
2
1
n

0
[w]
_
U

0

1
3
U
2
f

0
I
_
(69) :
1
2
_

r

+
_

r

_
t
_
.
If the equation of continuity, /r

= 0, is taken into
account, it is seen that the following relationship is satised:
(70) I :
_

r

+
_

r

_
t
_
= 0.
A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266 259
We now have obtained the zeroth and rst order solutions
of f

,
(71) f

=f

0
+f

1
,
in which f

0
and f

1
are given in Eqs. (60) and (69), respec-
tively.
4.2. Viscosity due to momentum
The viscosity due to momentum,
K
, can be obtained
from Eqs. (42) and (71) as

K
= n

0
_ _
U

_
f

0
+f

1
_
dv

= n

0
I +
3
2
_
1
2
_
3/2
n

0
[w]

_
U

_
U


1
3
U
2
I
_
exp
_

1
2
U
2
_
dU

:
1
2
_

r

+
_

r

_
t
_
(72) = n

0
I +
3
2
1

[w]
_

r

+
_

r

_
t
_
,
in which the following relationship has been used:
_
U

_
U


1
3
U
2
I
_
: D

exp
_

1
2
U
2
_
dU

(73)
=(2)
3/2
_
2D

11
D

12
+D

21
D

13
+D

31
D

12
+D

21
2D

22
D

23
+D

32
D

13
+D

31
D

23
+D

32
2D

33
_
.
In this equation, D

is dened by
(74) D

=
1
2
_

r

+
_

r

_
t
_
.
Hence, by comparing Eq. (72) with the general expression
of a stress tensor for Newtonian uids,
(75)

= P

I +

_

r

+
_

r

_
t
_
,
the viscosity due to momentum,
K
, can be expressed as
(76)
K
=
3
2
1

[w]
=
45
4
1

.
The dimensional expression for this expression is written as
(77)
K
=
3
2
mkT
r
3
c
[w]
.
4.3. Viscosity due to dissipative forces
We derive an expression for the stress tensor
D
. With
the following Taylor series expansion of the Dirac delta
function,
(r r
i
+e
ij
)
=(r r
i
) +e
ij


r
(r r
i
)
(78) +
1
2!

2
e
ij
e
ij
:

r

r
(r r
i
) + ,
the stress tensor due to dissipative forces can be written as

D
=
1
2
_

j
(i=j)
w
D
(r
ij
)(e
ij
v
ij
)e
ij
r
ij
(r r
i
)
_
+
1
4

r

_

j
(i=j)
w
D
(r
ij
)
(79) (e
ij
v
ij
)e
ij
r
ij
r
ij
(r r
i
)
_
+ .
The second term on the right-hand side in this equation is
negligible unless there is a signicant nonuniformity of the
system. Equation (79) is, therefore, rewritten as

D
=
1
2
n
2
0

_

_
R

w
D
(R

)
_

R (v

v

)
_

R
(80) f
(2)
(r

, r

+R

, v

, v

, t

) dR

dv

dv

.
By the assumption of molecular chaos, written in Eq. (53),
with Eq. (71) for f

, the expression for f


(2)
can be written
as
f
(2)
(r

, r

, v

, v

, t

) =f

0
(r

, v

, t

)f

0
(r

, v

, t

)
+f

0
(r

, v

, t

)f

1
(r

, v

, t

)
+f

0
(r

, v

, t

)f

1
(r

, v

, t

)
(81) +
2
f

1
(r

, v

, t

)f

1
(r

, v

, t

).
By evaluating Eq. (80) with Eq. (81), the expression for
D
can be obtained, which is now shown in the following.
We rst derive the expression for the stress
D
0
due to the
rst term on the right-hand side in Eq. (81). From Eq. (80),
we get the following expression:

D
0
=
1
2
n
2
0

_

_
R

w
D
(R

)
_

R (v

v

)
_

R
(82) f

0
(r

, v

, t

)f

0
(r

, v

, t

) dR

dv

dv

.
If we take into account the relationship
(83)
v

v

=(v

) (v

u

) R

(r

, t

),
then Eq. (82) can be rewritten as
(84)
D
0
=
1
2
n
2
0

_
w
D
(R

)R
2

R

R : D

dR

.
The integral on the right-hand side in this equation can be
carried out using the polar coordinate systemto give the nal
260 A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266
expression for
D
0
,
(85)
D
0
=

n
2
0
15
[R
2
w]
1
2
_

r

+
_

r

_
t
_
,
in which
[R
2
w] =
_
R
2
w
D
(R

) dR

(86) =
1
_
0
R
2
w
D
(R

)4R
2
dR

.
Next, we have to evaluate the stresses due to the second,
third, and fourth terms on the right-hand side in Eq. (81).
From a similar procedure, it can straightforwardly be shown
that all these stresses become zero.
By taking into account all these results, the expression for

D
can easily be derived as
(87)
D
=
D
0
.
Hence, the comparison of Eq. (85) with Eq. (75) leads to
the nal expression for the viscosity
D
due to dissipative
forces:
(88)
D
=

n
2
0
30
[R
2
w] =
2
1575
n
2
0

.
The dimensional form is expressed as
(89)
D
=
n
2
0
30
_
r
3
c
R
2
w
_
.
The viscosity is nally obtained as the sum of
K
and
D
as =
K
+
D
in the present case.
It is clearly seen from the above discussion that the per-
turbation method with perturbation parameter using the
nondimensional basic equation written in Eq. (55) can be
signicantly understandable in notifying the validity range
of the analytical expression of the viscosity, Eq. (88) or
Eq. (89), although the result essentially agrees with that
which was derived by Marsh et al. [14].
4.4. Relationship with HK theory
In order to compare the present results with the HK the-
ory [3,4] more straightforwardly, we rewrite Eq. (89) as
(90)
D
=
n
2
0
30
_
r
2
_
1
r
r
c
_
2
dr =
n
2
0
30
_
r
2
w
D
(r) dr.
The viscosity due to dissipative forces,
D
HK
, in the HK
theory is written as
(91)
D
HK
=
mn
2
0

30t
_
r
2
w
HK
(r) dr,
in which w
HK
(r) is a weight function in the HK theory
and this weight function corresponds to w
D
(r) in the present
theory. However, it should be noted that w
HK
(r) does not
satisfy the uctuationdissipation theory [12,13], which is,
in constant, satised by the present theory. If we take into
account that the notation in the HKtheory is related to the
notations here as = t /m, it is seen that
D
essentially
agrees with
D
HK
. This clearly shows that the expression in
the HK theory,
D
HK
, has a similar validity range to the
present expression
D
, which has derived for the cases of
small values of .
5. Evaluation of viscosity by means of the
nonequilibrium dynamics method
5.1. Evaluation by GreenKubo expression
There are two methods for evaluating transport coef-
cients by means of simulations: the rst method is for
thermodynamic equilibrium and evaluates them using the
GreenKubo expression; the second one is the nonequilib-
rium dynamics method, shown in the following, in which
transport coefcients are evaluated under circumstances of a
simple shear ow.
The theory of the nonequilibrium molecular dynamics
method [20] for a molecular system is applicable to the
present case, if a simple shear ow is considered as a ow
eld. In this case, the viscosity
yx
is evaluated fromthe fol-
lowing equation in simulations [18,20],
(92)
yx
=
1
V
J
yx

ne
,
in which
J
yx
(t ) =
N

i=1
_
mv
iy
(t )v
ix
(t ) +y
i
(t )F
ix
(t )
_
(93) =
N

i=1
mv
iy
(t )v
ix
(t ) +
N

i=1
N

j=1
(i<j)
y
ij
(t )F
ijx
(t ),
and F
ijx
is the x-component of the force vector F
ij
. In the
dissipative particle dynamics method, the force acting be-
tween particles, F
ij
, is taken as F
ij
= F
C
ij
+ F
D
ij
. Also, is
the shear rate, and J
yx

ne
is the time average of J
yx
under
circumstances of a simple shear ow, which is assumed to
act in the x-axis direction.
By nondimensionalizing the shear rate by (kT/m)
1/2
/r
c
,
the dimensionless form of Eq. (92) can be written as
(94)

yx
=

yx
(mkT )
1/2
/r
2
c
=
1

_
J

yx
_
ne
,
in which
(95) J

yx
(t

) =
N

i=1
v

iy
(t

)v

ix
(t

) +
N

i=1
N

j=1
(i<j)
y

ij
(t

)F

ijx
(t

).
The LeesEdwards boundary condition [18,20] has to be
used in order to generate a simple shear ow, and, further-
more, the equation of motion is slightly modied in many
A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266 261
cases for a molecular system. According to the nonequi-
librium molecular dynamics method, the nite difference
equations of the equation of motion for the present system
are modied as [20]
x

i
=
_
v

ix
+

i
_
t

, y

i
=v

iy
t

,
(96) z

i
=v

iz
t

,
v

ix
=
_
F

ix

iy
_
t

, v

iy
=F

iy
t

,
(97) v

iz
=F

iz
t

,
in which F
i
denotes the force acting on particle i.
5.2. Parameters for simulations
Simulations have been conducted under various condi-
tions in order to clarify the characteristics of the equation
of motion for dissipative particles dynamics by comparing
simulation results with theoretical solutions. Typical cases
such as n

0
= 0.5, 3.16, and 10.0 were taken for the number
density, and the particle number N was 500, unless specif-
ically notied. From sufcient numbers of simulations for
a small system, the time interval t

was taken as t

=
0.0001, and the total time steps were adopted as 5,000,000,
which ensures sufcient accuracy of average values. The er-
ror analysis [20] concerning almost all simulation results
was conducted to show the error ranges of simulation data;
the error ranges of
K
and
D
are denoted by
K
error
and

D
error
, respectively, and these values will be shown with
results in Section 6. Also, the parameters

and

, which
have appeared in the equation of motion for dissipative parti-
cles, were taken as

= 10, 25 and

/10, unless
specically noted.
6. Results
6.1. Inuence of

on viscosity
Figs. 1 and 2 show the inuence of the values of

on
the viscosity, which were obtained by the nonequilibrium
dynamics method. Fig. 1 is for the viscosity
D
due to dis-
sipative forces and Fig. 2 is for the viscosity
K
due to
momentum.
It is seen from Fig. 1 that the results obtained by the
nonequilibriumdynamics simulations are in good agreement
with the theoretical solutions shown in Eq. (88) for both
cases of

/10 and

. However, for the cases


of small number densities such as n

0
= 0.5, the simula-
tion results have signicant errors and, therefore, qualitative
properties are relatively difcult to identify. That the num-
ber density n

0
is smaller than unity means that a sufciently
large number of particles do not exist around an arbitrary
particle within the range of the radius r
c
from its center. In
this situation, the accuracy of the viscosity data is strongly
dependent on whether or not there are other particles which
interact with the particle of interest. Hence such a situation
(a)
(b)
Fig. 1. Inuence of

on viscosity due to dissipative forces obtained


by nonequilibrium dynamics method (a) for

/10 and (b) for

(
D
error
= (5.2, 1.5) for n
2
0

= (10,000, 1000), respec-


tively, for n

0
= 10, (1.5, 0.32) for n
2
0

= (1000, 100), respectively,


for n

0
= 3.16, and (0.049, 0.022) for n
2
0

= (25, 2.5), respectively, for


n

0
=0.5, in (a); similar errors are included in (b)).
causes signicant errors of the simulation results of viscos-
ity, which has been seen for n

0
= 0.5 in Fig. 1. We may
conclude from this fact that it is desirable to take the number
density as n

0
1.0 in simulating a ow problem by means
of the dissipative particle dynamics method.
Fig. 2 clearly shows that the simulation data are larger
than the theoretical solutions, and this tendency becomes
more signicant for larger values of

. When the inuence


of dissipative forces is of the same order of the momentum,
that is, when

1, the simulation results of


K
agree
well with the theoretical solutions. On the other hand, at

1, where dissipative forces are more dominant than


the momentum, the discrepancy between the simulation re-
sults and the analytical solutions has a tendency to increase
with values of

. As will be shown later in Fig. 8, the ra-


dial distribution function does not signicantly change for
values of

, so that this discrepancy is presumed to be due


to the mechanism other than the assumption of molecular
chaos, adopted here. Some researchers attempted to explain
the discrepancy between simulation results and theoretical
solutions concerning the viscosity due to the momentum, in
terms of the pair collision theory [21].
262 A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266
(a)
(b)
Fig. 2. Inuence of

on viscosity due to momentum obtained by non-


equilibrium dynamics method (a) for

/10 and (b) for

(
K
error
= (0.57, 1.47) for 1/

= (0.01, 0.1), respectively, for n

0
= 10,
(0.40, 0.62) for 1/

= (0.01, 0.1), respectively, for n

0
= 3.16, and
(0.44, 0.12) for 1/

= (0.01, 0.1), respectively, for n

0
= 0.5, in (a);
similar errors are included in (b)).
6.2. Inuence of

and number density on viscosity


Fig. 3 shows the inuence of

on the viscosity due to


dissipative forces. It is seen from Fig. 3 that the values of

do not have signicant inuence on the viscosity for both


cases of n

0
= 3.16 and 10.0. As will be shown later in Fig. 9,
the radial distribution function changes signicantly for dif-
ferent values of

. Hence, this means that the viscosity due


to dissipative forces,
D
, is not strongly dependent on the
shape of the radial distribution function.
Fig. 4 shows the dependence of the viscosity on the num-
ber density of particles: Fig. 4a is for the viscosity
D
due
to dissipative forces, and Fig. 4b is for the viscosity
K
due to momentum. Since the theoretical solutions are valid
for n

1.0, the simulation results are in good agree-


ment with the analytical solutions for the values of n

0
which
satisfy the condition n

1.0 for a given value of

.
This agreement becomes worse with decreasing values of n

0
smaller than unity; especially, the results for

= 10 and

= 1 signicantly deviate from the theoretical values with


large error ranges. As already pointed out, this is presumed
to be due to the situation in which the ambient particles of an
Fig. 3. Inuence of

on viscosity due to dissipative forces (


D
error
=
(1.2, 3.1) for

= (100, 1), respectively, in the case of n

0
= 10 and

= 25, and (0.17, 0.32) for

= (40, 1), respectively, in the case of


n

0
= 3.16 and

= 10).
(a)
(b)
Fig. 4. Inuence of number density on viscosity: (a) for viscosity due
to dissipative forces and (b) for viscosity due to momentum (
D
error
=
(2.7, 0.12) for n

0
= (10, 1), respectively, in the case of

= 25 and

= 2.5, and (0.78, 0.022) for n

0
= (10, 0.5), respectively, in the case
of

= 10 and

= 1, in (a);
K
error
= (1.1, 0.24) for n

0
= (10, 1),
respectively, in the case of

= 25 and

= 2.5, and (1.0, 0.12) for


n

0
= (10, 0.5), respectively, in the case of

= 10, and

= 1, in (b)).
arbitrary particle come to be less for decreasing the number
density smaller than unity, which leads to the large uctu-
ation in the contribution of dissipative particles to physical
A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266 263
quantities such as transport coefcients with time. Although
the viscosity
K
due to momentum shown in Fig. 4b has a
large error range, it is clearly seen that there is a signicant
discrepancy between the simulation and theoretical values,
and this tendency does not strongly depend on the values
of n

0
. As already pointed out in Fig. 2, since the radial distri-
bution function does not signicantly deviate from g(r) = 1
(shown later in Fig. 10a), it is presumed that this discrepancy
for

= 10 and

= 1 is induced by the mechanism other


than the invalidity of the molecular chaos assumption.
6.3. Inuence of time interval on mean velocity of particles
and viscosity
Fig. 5 shows the inuence of the time interval h

(=
t

), which is used in conducting simulations based on the


nite difference equations, on the mean velocity of particles
and the viscosity. It is seen from Fig. 5a that the viscos-
ity
D
does not strongly depend on the time interval for
h

0.1. On the other hand, as shown in Fig. 5b, the vis-


cosity due to momentum,
K
, is signicantly dependent
on the values of h

. Especially, this dependence comes to


appear from h

0.001, and values of


K
signicantly
increase with increasing values of the time interval. A sim-
ilar tendency can be observed in Fig. 5c, in which the mean
velocity of particles signicantly increases from h

0.01
with the values of h

. Since the mean velocity v is nondi-


mensionalized in such a way of being v

= 1, this condition
has to be satised in actual simulations. We may, therefore,
conclude that the simulations have to be conducted using a
sufciently small time interval such as h

= 0.0001, adopted
here.
6.4. Inuence of shear rate and number of particles on
viscosity
Fig. 6 shows the inuence of the shear rate of a simple
shear ow and the number of particles on the viscosity. It
is seen that the simulation results come to have signicant
errors with decreasing shear rate, since just a weak simple
shear ow is generated for such situations. This character-
istic is observed for both cases of
D
in Fig. 6a and
K
in Fig. 6b, especially in the results of
K
. It is seen from
Fig. 6c that the mean velocity v

signicantly increases with


the shear rate after

1. Hence, there is an appropriate


range for the shear rate

, which is sufciently small for


satisfying the condition of v

= 1, and sufciently large for


generating a simple shear ow with an appropriate shear
strength. By taking account of these constraints, we here
adopted

= 0.01 in conducting simulations.


Fig. 7 shows the inuence of the number of particles on
the viscosity
D
due to dissipative forces. It is seen from
Fig. 7 that a larger system is required for a smaller num-
ber density to remove the dependence of the results on the
number of particles. This means that the interactions of the
(a)
(b)
(c)
Fig. 5. Inuence of time interval on (a) viscosity due to dissipative forces,
(b) viscosity due to momentum, and (c) average velocity of particles
(
D
error
= (0.43, 2.7) for h

= (0.01, 0.0001), respectively, in the case


of n

0
= 10,

= 25, and

= 2.5 (Case A), and (0.051, 0.31) for


h

= (0.1, 0.0001), respectively, in the case of n

0
= 3.16,

= 10, and

= 1 (Case B), in (a);


K
error
= (0.34, 1.1) for h

= (0.01, 0.0001),
respectively, in Case A, and (1.1, 0.62) for h

= (0.1, 0.0001), respec-


tively, in Case B, in (b)).
particle of interest with the ambient particles come to de-
pend on the dimensions of the system, unless a sufciently
large system is used for the cases of small number densities.
As already pointed out, the number density has to be taken
as sufciently large, and a large system such as N = 500,
adopted here, at least has to be adopted to remove the de-
pendence of results on the system size.
264 A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266
(a)
(b)
(c)
Fig. 6. Inuence of shear rate on (a) viscosity due to dissipative forces,
(b) viscosity due to momentum, and (c) average velocity of particles
(
D
error
= (0.055, 2.7) for

= (0.5, 0.01), respectively, in Case A,


and (0.0064, 0.32) for

= (0.5, 0.01), respectively, in Case B, in (a);

K
error
= (0.028, 1.1) for

= (0.5, 0.01), respectively, in Case A, and


(0.017, 0.62) for

= (0.5, 0.01), respectively, in Case B, in (b)).


6.5. Inuence of

, and number density on radial


distribution function
Finally, we show the results of the radial distribution
function [18] in Figs. 810 in order to clarify the quantitative
dependence of the internal structure on

, and the num-


ber density of particles. Figs. 8, 9, and 10 show the inuence
of

, and the number density, respectively.


It is seen from Fig. 8 that the radial distribution function
is almost independent of the values of

, and the curve for

= 10 agrees well with that for

= 100 for both cases


Fig. 7. Inuence of number of particles on viscosity due to dissipative forces
(
D
error
= (1.7, 6.1) for N = (1372, 108), respectively, in Case A, and
(0.19, 0.68) for N =(1372, 108), respectively, in Case B).
(a)
(b)
Fig. 8. Dependence of radial distribution function on

for

= 10 (a) for
n

0
= 3.16 and (b) for n

0
= 10.
of n

0
= 3.16 and 10.0. Although the theoretical solutions of
the viscosity have been derived under circumstances of ne-
glecting the internal structure of the system, that is, under
the molecular chaos assumption, it is very interesting that
the theoretical solutions agree well with the simulation re-
sults (Fig. 1), irrespective of the system being in a gas-like
or liquid-like internal structure, as shown in Fig. 8.
Fig. 9 clearly shows that the radial distribution function
strongly depends on the values of

. A more signicant
repulsive force acts between particles with increasing the
A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266 265
(a)
(b)
Fig. 9. Dependence of radial distribution function on

(a) for n

0
= 3.16
and

= 10, and (b) for n

0
= 10 and

= 25.
values of

, so that the radial distribution function shows a


liquid-like structure for

= 25 and

= 40. That the value


of

is much larger than that of

means that the inuence


of dissipative and random forces is much smaller than that
of repulsions due to conservative forces. This means that the
merits of the dissipative particle dynamics method cannot
function effectively under such circumstances. It is, there-
fore, presumed that

has to be taken as

<

to make
the dissipative particle dynamics method effective. It is seen
from Fig. 9a that the molecular chaos assumption is satised
to a certain degree for

= 0.4. To the contrary, the results


for

= 4 and 25 cannot sufciently satisfy this assumption.


Nevertheless, as already shown in Fig. 3, the viscosity
D
due to dissipative forces is almost independent of the values
of

. We may conclude form these results that dissipative


forces do not strongly depend on the internal structure of the
system.
It is seen from Fig. 10 that the radial distribution function
is not strongly dependent on the number density. However,
the curve for the case of n

0
= 6.0, in which the particle of
interest can interact with a sufcient number of ambient par-
ticles, slightly deviates from the results for n

0
= 0.2 and 1.0,
in which there are not sufcient particles around the particle
of interest.
(a)
(b)
Fig. 10. Dependence of radial distribution function on number density of
particles (a) for

= 10 and

= 1, and (b) for

=25 and

=2.5.
Fig. 11. Results of radial distribution function obtained by Espanol et al.
[22].
Finally, we compare the present results of the radial distri-
bution function with other simulation results obtained by Es-
panol et al. [22]. They investigated the relationship between
the dissipative particle dynamics and the smoothed particle
dynamics methods. In the concrete, the behavior of the clus-
ters of particles, which correspond to dissipative particles
in the present study, was investigated by means of conduct-
ing equilibriummolecular dynamics simulations of a system
composed of particles with repulsive interactions. Fig. 11
266 A. Satoh, T. Majima / Journal of Colloid and Interface Science 283 (2005) 251266
shows one of the results concerning the radial distribution
function, which were obtained by Espanol et al. [22]. From
the potential curves of the mean force in their paper [22], it
is seen that R in their notation roughly corresponds to 3r

in the present one, and that lower curves correspond to re-


sults for a larger value of

in the present study. We clearly


see fromcomparing Fig. 9 (also Fig. 10) with Fig. 11 that the
shape of the curves are quite similar between the present and
their results, in which a small increase at very short distances
can be observed more signicantly with decreasing values
of

. A system with larger values of

approaches a usual
system such as a molecular one, so that dissipative parti-
cles cannot penetrate each other due to a high-energy barrier.
This leads to the fact that the curves of the radial distribution
function comes to have an ordinary shape for a liquid or gas,
which has no small increase within a short distance range.
7. Conclusions
In order to investigate the validity of the dissipative par-
ticle dynamics method, which is a mesoscopic simulation
technique, we have derived expressions for transport coef-
cients such as viscosity, from the equation of motion of
the dissipative particles. In the concrete, we have shown the
FokkerPlanck equation in phase space, and macroscopic
conservation equations such as the equation of continuity
and the equation of momentum conservation. The basic
equations of the single-particle and pair distribution func-
tions have been derived using the FokkerPlanck equation.
The solutions of these distribution functions have approx-
imately been solved by the perturbation method under the
assumption of molecular chaos. The expressions of the vis-
cosity due to momentum and dissipative forces have been
obtained using the approximate solutions of the distribution
functions. Also, we have conducted nonequilibrium dynam-
ics simulations to investigate the inuence of the parame-
ters, which have appeared in dening the equation of motion
in the dissipative particle dynamics method. The theoreti-
cal values of the viscosity due to dissipative forces in the
HK theory are in good agreement with the simulation re-
sults obtained by the nonequilibrium dynamics method, ex-
cept in the range of small number densities. There are restric-
tion conditions for taking appropriate values of the number
density, number of particles, time interval, shear rate, etc., to
obtain physically reasonable results by means of dissipative
particle dynamics simulations.
References
[1] P. Mazur, W. van Saarloos, Physica A 115 (1982) 21.
[2] W. van Saarloos, P. Mazur, Physica A 120 (1983) 77.
[3] P.J. Hoogerbrugge, J.M.V.A. Koelman, Europhys. Lett. 19 (1992) 155.
[4] J.M.V.A. Koelman, P.J. Hoogerbrugge, Europhys. Lett. 21 (1993) 263.
[5] G. Bossis, J.F. Brady, J. Chem. Phys. 80 (1984) 5141.
[6] J.F. Brady, G. Bossis, J. Fluid Mech. 155 (1985) 105.
[7] L. Durlofsky, J.F. Brady, G. Bossis, J. Fluid Mech. 180 (1987) 21.
[8] A. Satoh, R.W. Chantrell, G.N. Coverdale, S. Kamiyama, J. Colloid
Interface Sci. 203 (1998) 233.
[9] A. Satoh, G.N. Coverdale, R.W. Chantrell, J. Colloid Interface Sci. 231
(2000) 238.
[10] A. Satoh, J. Colloid Interface Sci. 243 (2001) 342.
[11] A. Satoh, J. Colloid Interface Sci. 255 (2002) 98.
[12] P. Espanol, P. Warren, Europhys. Lett. 30 (1995) 191.
[13] P. Espanol, Phys. Rev. E 52 (1995) 1734.
[14] C.A. Marsh, G. Backx, M.H. Ernst, Phys. Rev. E 56 (1997) 1676.
[15] E.S. Boek, P.V. Coveney, H.N.W. Lekkerkerker, P. van Schoot, Phys.
Rev. E 55 (1997) 3124.
[16] P. Espanol, Europhys. Lett. 40 (1997) 631.
[17] J.B. Avalos, A.D. Mackie, Europhys. Lett. 40 (1997) 141.
[18] A. Satoh, Studies in Interface Science: Introduction to Molecular-
Microsimulation of Colloidal Dispersions, Elsevier Science, Amster-
dam, 2003.
[19] K.E. Novik, P.V. Coveney, J. Chem. Phys. 109 (1998) 7667.
[20] M.P. Allen, D.J. Tildesley, Computer Simulation of Liquids, Claren-
don, Oxford, 1987.
[21] A.J. Masters, P.B. Waeewn, Europhys. Lett. 48 (1999) 1.
[22] P. Espanol, M. Serrano, I. Zuniga, Int. J. Mod. Phys. C 8 (1997) 899.