Journal o f Statistical Physics, Vol. 87, Nos.

1/2, 1997

Analytic Solutions of Simple Flows and

Analysis of Nonslip Boundary Conditions for the
Lattice Boltzmann BGK Model

Xiaoyi He, i.~_ Qisu Zou, t.3.5 Li-Shi Luo, 3'4'6 and Micah Dembo 2

Received July 14. 1995

In this paper we analytically solve the velocity of the lattice Boltzmann BGK
equation [LBGK) Ibr several simple flows. The analysis provides a framework
to theoretically analyze various boundary conditions. In particular, the analysis
is used to derive the slip velocities generated by various schemes for the nonslip
boundary condition. We find that the slip velocity is zero as long as Z=J~e~ = 0
at boundaries, no matter what combination of distributions is chosen. The
schemes proposed by Noble et al. and by Inamuro et aL yield the correct zero-
slip velocity, while some other schemes, such as the bounce-back scheme and
the equilibrium distribution scheme, would inevitably generate a nonzero slip
velocity. The bounce-back scheme with the wall located halfway between a flow
node and a bounce-back node is also studied Ibr the simple flows considered
and is shown to produce results of second-order accuracy. The momentum
exchange at boundaries seems to be highly related to the slip velocity at bound-
aries. To be specific, the slip velocity is zero only when the momentum
dissipated by boundaries is equal to the stress provided by fluids.

KEY W O R D S : Lattice Boltzmann BGK equations; nonslip boundary condi-

tions; analytic solutions of simple flows.

*Center for Nonlinear Studies (CNLS), MS-B258, Los Alamos National Laboratory,
Los Alamos, New Mexico 87545.
2Theoretical Biology and Biophysics Group (T-10), MS-K710, Theoretical Division,
Los Alamos l~lational Laboratory, Los Alamos, New Mexico 87545.
3Complex Systems Group (T-13), MS-B213, Theoretical Division, Los Alamos National
Laboratory, Los Alamos, New Mexico 87545.
4Computational Science Methods Group (XCM), MS-F645, X Division, Los Alamos
National Laboratory, Los Alamos, New Mexico 87545.
s Department of Mathematics, Kansas State University, Manhattan, Kansas 66506.
~' ICASE, Mail Stop 403, NASA Langley Research Center, Hampton, Virginia 23681.

115

0022-4715/97/0400-0115512.50/0 c. 1997PlenumPublishingCorporation
116 He e t a l .

1. I N T R O D U C T I O N

In recent years, the lattice Boltzmann equation (LBE) c4-8~has been success-
fully applied to various hydrodynamic problems which are difficult for con-
ventional numerical methods (e.g., finite-difference or fmite-element scheme).
Among these problems are multiphase, multicomponent flows, tg' i0~ magneto-
hydrodynamics, ~111and reactive flows. 1121From a computational viewpoint,
the notable advantages of the LBE method are parallelism of algorithm, sim-
plicity of programming, and ease of incorporating microscopic interactions.
Although it has been proved ~6-7-13~that the LBE recovers the Navier-
Stokes equation with a second order of accuracy in space in the interior of
flow domain, the real hydrodynamic boundary conditions have not been
fully understood. For a node near a boundary, some of its neighboring
nodes may locate outside the flow domain. The distribution functions from
these nonfluid nodes are therefore unknown after each streaming process.
The boundary condition is responsible for determining these unknown
distributions.
In general, there are two ways to define a boundary: placing the
boundary on grid nodes 11-3.141 or placing the boundary on links. I ~0.~s. 16) No
matter which method is used, to find out whether a boundary condition is
appropriate, one would like to know the slip velocity introduced by the
boundary condition. Many of the previous studies on the velocity bound-
ary condition are based on numerical experiments. In ref. 14 the distribu-
tion at a boundary node is set to the equilibrium distribution plus a
modification term based the Chapman-Enskog expansion, and the velocity
gradient in the modification is approximated by a finite difference. In ref. 1
the unknown distributions and density at a boundary node are obtained
from the known distributions and the known velocities at the boundary
for the triangular lattice. In refs. 2 and 3 the unknown distributions and
density at a boundary node are obtained from the known distributions and
the known velocities at the boundary with some additional assumptions of
the unknown distributions. With numerical experiments, one can only
observe the order of the slip velocity. How this slip velocity is generated
and what factors it depends on are not well known. Hence, an analytical
study is necessary for a better understanding of the model and the non-slip
boundary condition on stationary or moving walls.
Some theoretical studies have been carried out on the LGA or the
LBE model and the non-slip boundary condition on a stationary wall was
studied. ~15-181 The discussion in refs. 15 and 17 is based on a linearized
lattice Boltzmann equation with a global equilibrium with constant density
and isotropic velocity (zero velocity). In ref. 16 the boundary condition
of the 3D F C H C lattice Boltzmann model 119"2~ with a linear collision
Lattice Boltzmann BGK Model 117

operator represented by a collision matrix A was studied for the plane

Poiseuille flow and the plane stagnation flow. In that study, the first-order
and second-order deviations of the distribution function from equilibrium
were assumed to take an a priori form in terms of flow quantities. In ref. 18
analytical solutions of the distribution functions for Poiseuille flow and
Couette flows were found for the F H P or the square L B G K models
without any approximation. The study was restricted to situations that the
velocity profile extends beyond the boundary in a smooth way and to the
Poiseuille flow with square forcing given in ref. 21 for the F H P model.
Nevertheless, all the previous theoretical studies only treated bounceback
boundary conditions. The present study directly solves the L B G K equation
without any approximations and hence it can handle various boundary
conditions.
The paper is organized as follows. Section 2 analytically solves the nine-
bit lattice Boltzmann BGK equation for the two-dimensional Poiseuille
flow and the Couette flow with possible vertical injection at the boundaries.
The results are further used to analyze different schemes for the non-slip
boundary condition. Section 3 discusses the results and concludes the
paper. The Appendix briefly discusses some results of the seven-bit (FHP)
L B G K model.

2. A N A L Y T I C A L SOLUTION A N D NONSLIP
BOUNDARY CONDITION
In this section, we use the square nine-bit lattice L B G K model. The
procedure can be easily applied to the F H P model and some results are
given in the Appendix. The nine-bit lattice BGK model is on a square
lattice space with three speeds: 0, c, and v/'2c, where c=6,./6, and 6,.
and 6, are the lattice constant and the step size in time, respectively.
Figure 1 shows all possible velocities of the model. With the presence of a
body force F, the evolution equation of the system is

f~(x + e~6,, t + fi,)--f~(x, t ) = --~1 [f~(x, t)--f~q)(p, u)] + 6'2 <5.,.g~(1)

where r is the dimensionless relaxation time, and

((o, o), 0c=0

e~= ~ ( c o s [ ( ~ - 1) n/2], s i n [ ( ~ - 1) n/2]) c ~= 1,2,3,4
{ (cos [ ( ~ - 5 ) re/2 + 7r/4], sin [ (~ - 5 ) 1r/2 + rr/4 ] ) x/'2c 0~=5,6, 7, 8
(2)
118 He e t a l .

6 2 5

\/ ,
-x 3

7
/\ 4
1

8
-

Fig. 1. All possible velocities for the nine-bit lattice B G K model on a square lattice. This
figure also shows the arrangement for the channel flow and the Couette flow in simulations.

are the velocity vectors. The equilibrium distribution function f~cq~(p, u) is

given by

f'~qito u ) = w ~ p [ l + 3 / ' e ~ ' u ' x 9{/e~'u'~ 2 3 ( u ) 2] (3)

" '" -7

where Wo=4/9, wi.2.3.4= 1/9, and w5.6.7.8= 1/36. The local density of mass
p and the local density of momentum pu in Eq. (3) are given by

p =~f~ (4a)
ct

pu = ~ e~f~ (4b)

Finally, the forcing term g: is given by

I
0~c or
e~.F, cr 1,2, 3,4
g: = (5)

t 1-~c e," F, ct = 5, 6, 7, 8

The evolution of the lattice BGK system consists of two steps: colli-
sion and advection. First the distribution functions f: at the site x undergo
Lattice Boltzmann BGK Model 119

the collision process, prescribed by the right of Eq. (I) or otherwise. After
the collision, j~ moves to the next site, x + % 5 , , according to the
velocity %.
The Navier-Stoke equation can be derived from Eq. (1):

911
p ~ + pu" Vu = -V(c~p) + vpV2u + F (6)

where the kinetic viscosity is

(2~- l)6~
6 5,

and the speed of sound is

1
Cs--~T C

In the following analysis, we will only consider steady (time-independ-

ent) flows satisfying

Ou Ov
~x=O, ox-O, p =const (7)

The body force is assumed to be along the x direction, i.e., F =pGix. In

this type of flow, both the velocity and the distribution functions are only
functions of the y coordinate. Although being rather simple, they serve as
examples for us to carry out some theoretical analyses which can be further
used to study the nonslip boundary condition.
By substituting Eqs. (3) and (7) into Eq. (1), we obtain the following
equations:

fJ~ - - L 1 -- 5 ~,722+ ~ ) ] (8a)

= - - + 3 -~ vj . (8b)
c c- 2 75-" 3c-'
-)

f{,- = 9z[
P-- [ l + 3 v j - t + 3 c v-7~
c- '
23"-]c (8c)

5,.rpG
f{-P-[1-3uJ+3 -s (8d)
3--9 L C c- 2 3c 2
120 He etal.

pF ~;§ v~+t 3u2+,]~_r-l.+,

f~=GLI-3 c +3 c-~- 2 c 5 J - r "4
(8e)

p[ uj_, vj_, u~ v}_, uj_,vj_,]

f~=3-6-~r 1 + 3 c +3 c +3 ~'+3c_ ~ + 9 c2

r- 1 . OxpG
+ r f~-'+ ~ (8f)

- 3" ' c c '+3u c, +3 V,c

r- 1 3.,.pG
+ f~-'--- (8g)
r 12c 2

P l Uj+I Vj+I /'/7+1 /)7+ Uj+II')j+I]

f~=3--~r 1-3 c -3 c +3 ,c_ + 3 ,c_ + 9 c2

+ t - 1 fi7+, - -O.,.pG
- (8h)
r 12c 2

p uj+ I 1.)j+ I 2 I
1..Ij+ . 1
Oj+ tg+lq/+t ]
f{=3-~r 1+3 -3 +3 , +3 -9
c c c- c2 c -~ l

+ r - 1 f{+ ~+ -c~,.pG
- (8i)
r 12c 2

where f ~ is the distribution function and (uj, uj) is the velocity vector at
y=jr~,.. Equations (8) are valid for 2 < ~ j < ~ n - - 2 . The node j = 0 o r j = n
corresponds to lower or upper boundaries, respectively, where the evolu-
tion rule depends on the implementation of boundary conditions.
In the interior of the flow domain (2 < ~ j < ~ n - 2 ) , the x component of
the momentum density PUJ can be rewritten as

puj = cr (f'; - G ) + (G - f O + (G-f4)]

r+l 2r--1
= 3r pUj+'"~--~ p(t'tj-I+uj+l)

+2-~r (u:- t v : _ , - u j + iv~+ ,)

9
+6'pG
T
(9)
Lattice Boltzmann BGK Model 121

which further gives us

Uj+ll)j+l--Uj--IUj--I = v uj+l +uJ-I-2Ui + G (10)

25,. 52,
The above equation is exactly the second-order finite-difference form of the
simplified incompressible Navier-Stokes equation under the assumption
(7) and constant pressure:
a(uv) a2u
Oy =v--~y,_+G (11)

In the y direction, from pvj= c [ f { - f { +f~ + f J 6 - f ~ - f { ] , it is easy to

prove

v~+, - v.~_, = (2r - 1)(v/+, + vj_, - 2vj) c (12)

8 i
On the other hand, from p =Y'.i=ofi and the definition of PVi above, we
can prove

v]+ i +v.7-1-2v~ = ( 2 r - 1)(vj+ i - vj_ i) c (13)

Combination of Eqs. (12) and (13) yields

vj = v,.-- const (14)

It should be noted that this result is only valid in the interior of the flow
domain. Whether v,. is equal to the vertical velocity at boundaries v~
depends on the implementation of the boundary condition.

2.1. Poiseuille Flow

For the Poiseuille flow, the vertical velocity is zero and Eq. (10) has
a simple solution:

4 Uc
uj=--Tj(n --j) + U.,., j = 1..... n -- 1 (15)
n-

where Uc = L2G/8v is the centerline velocity without slips at boundaries,

with L = n5.,. being the width of the channel and U, is the slip velocity
depending on the implementation of the boundary condition for walls in a
particular scheme. The above velocity profile is a parabola with a shift
on the boundaries. The slip velocity U.,. is the only term which makes the
122 H e e t al.

solution of the L B G K possibly different from the exact solution of the

Poiseuille flow.
In order to find the slip velocity, we need to apply Eq. (9) at the grid
line next to the bottom wall ( j = 1) (analysis for the top boundary is the
same):

pu, = c[ (fl - f ~ ) + (J~ -f~6) + (f~ - f { ) ]

= 3 - - - ~ P U t + - - ~ r p(u~ r pG+ 1- p(Uo-Uo) (16)

which further yields

6-' 3(r-- 1)
uo + uz + - .,. G -t [ fro - uo] (17)
u, = 2 2v 2r-- 1

In the derivation, it is assumed that f ( ~ , f o , f o , f o follow the rules in Eqs.

(8), the equilibrium distribution at j = 0 is calculated using velocity (Uo, 0),
and PUo= c[(f~ - f 3 o) + ( f o - f 6 o) + ( f so - f v o) ] , w h e r e f ~ ~ depend on the
boundary condition implemented. Comparing Eqs. (15) and (17), we
obtain the explicit expression for the slip velocity:

U , . - 6 ( r - l![~o-Uo] (18)
9 2r- 1

With this result, we can easily analyze different schemes for the nonslip
boundary condition (Uo is set to be zero in what follows for the sake of sim-
plicity).
1. Bounce-back scheme. For the bounce-back scheme, the particle
colliding with the wall simply reverses the direction of its velocity, that is,
f'/--
- f 4 ,0 f5I =f7,0 f6I -=-iS
"o
( 19)
Notice that the collision process does not occur at the boundary in this
scheme, hence Eq. (18) does not apply. Using the bounce-back rule, we can
derive

2U,. [ ( 2 r - 1 ) ( 4 r - 3) - 3 n ] (20)
U,~= 3n-'---T

Clearly, the bounce-back rule generally yields a nonzero slip velocity.

Furthermore, the slip velocity is of the first order in space because it has
a term of O(l/n).
Lattice Boltzmann BGK Model 123

2. Modified bounce-back scheme. By the modified bounce-back rule,

we mean that collision and forcing still occur at boundary nodes. The
precollision unknown distribution is set equal to the value of the distribu-
tion along the opposite direction:

f , 0_ = f 04, f~ =f~, f 6o= f 8 o (21)

With the modified bounce-back rule, we can prove

1 6 r ( r - 1)
U,- 3n2 U,. (22)

Clearly, the modified bounce-back boundary condition also generate a

nonzero slip velocity as long as r :~ 1. However, compared to the bounce-
back rule, the error is reduced and the slip velocity is of the second order
in space.
3. Nonslip schemes. In the nonslip scheme proposed by Noble
et aU 2~ the unknown distributions and the boundary density are calculated
from the constraints given in Eq. (4) and an additional constraint given by

2pt=~,(e~-u)2f~ (23)

where e is the internal energy, which equals the square of the sound speed
in the limit of low Mach number and low Knudsen number. In the nonslip
scheme proposed by Inamuro et al. ~3~ it is assumed that the distributions at
a horizontal boundary can be written in the following equilibrium form:

f o= -_ _~f ( e q ) ( / },,.,
l u.,) (24)
--
wnere j*"(eq
~ ) is defined in Eq. (3) and u,.l = (Uo + u', %). The two parameters
p', u' and the density at the boundary are determined from the constraints
(4).
From their definitions, it is easy to see both schemes lead to

Uo = c [ ( f , - f 0 (f0.15 __fO)j6 + ( f ~ 1 7 6 =0 (25)

Therefore, the slip velocity is automatically zero and the nonslip boundary
condition is implemented correctly. Notice that the distribution functions
are different in these two schemes, and they are different from the analytical
solution in ref. 18, but all three distribution functions give the correct
velocity profile. In fact, any scheme which uses the k n o w n f ~ ~ 1 7 6 o o o
124 He e t al.

to generate fo,fo,fo according to E ~ f ~ e ~ = 0 and ensures the correct p

will give the correct velocity profile for the Poiseuille flow. For example, we
have obtained the boundary density from a consistency condition in Eq. (3)
as in refs. 2 and 3 and have set fo=p/fl with fl being an adjustable
parameter, and the numerical simulations yielded the correct density and
velocity profile of the Poiseuille flow for fl varying from 0.01 to 108.

To verify our theoretical analysis, we have further carried out numeri-

cal simulations for the Poiseuille flow. Figure 2 shows the velocity profiles
uj normalized by Uc for the Poiseuille flow. The system size is N,. x N,, =
16 x4. The amplitude of the uniform forcing along the x axis is 0.1. The
density of the system is set to be 1.0, The initial state of the system is
u = 0 and v = 0. The uj are measured at a cross section of the channel after
5000 times iteration. Shown in Fig. 2 are the velocity profiles with different
values of r and with different implementations of the boundary condition
for a stationary wall, i.e., the bounce-back rule, the modified bounce-back
rule, and the nonslip boundary conditions in refs. 2 and 3. The result with
the nonslip boundary condition is a perfect parabola without any slip
velocity at the wails, while the results with the bounce-back or the modified
bounce-back boundary condition have a nonzero slip velocity at the
boundaries. Also, the numerical results are in excellent agreement with our

1.5

1.0

0.5

0.0

-0.5

0 1 2 3 4
j
Fig. 2. The normalized velocity profile of the Poiseuille flow. Solid lines, our analytic results
with various types of boundary conditions and values of r. ( + ) The numerical results of lat-
tice BGK simulation with the scheme in refs. 2 and 3; ( II, [] ) the numerical results of lattice
BGK simulation with the bounce-back boundary conditions; ( , t , A ) the results with the
modified bounce-back boundary condition. The solid and open symbols represent the numeri-
cal results with r = 2.0 and 0.75, respectively.
Lattice Boltzmann BGK Model 125

~4
\
2

0 I 2 3 4 5

Fig. 3. The normalized slip velocity of the Poiseuille flow as a function of r and the number
of nodes across the channel, i1, with two types of boundary conditions. Solid lines, our
analytic results. I I-I + ) The numerical results by the lattice BGK simulation with the bounce-
back boundary condition and with n = 4 and 8, respectively; (,~, • the numerical results with
the modified bounce-back boundary condition and with n 4 and 8, respectively.
=

theoretical analysis. The error in the numerical results of uJU,. is of the

order o f 1 0 - to, which is of the order of the r o u n d o f f error.
Figure 3 shows the slip velocity Us, normalized by U,., as a function
of r and n for the Poiseuille flow with the b o u n c e - b a c k b o u n d a r y condi-
tion. The analytic results are given by Eq. (20). We chose n = 4 and 8 in the
simulations. The results of the simulations coincide perfectly with the
analytic results.

2.2. P l a n e C o u e t t e F l o w w i t h Possible I n j e c t i o n at B o u n d a r i e s

F o r the Couette flow with possible vertical injection at the boundaries,

the b o d y force G is zero. The lower and upper walls m o v e along the
horizontal direction at different velocities Uo and u,, respectively. In the
mean time, a vertical velocity is injected at a speed of vb at both walls.
E q u a t i o n (10) can be solved in the interior of the flow domain. T h r o u g h
some simple algebraic derivations we find

M- 1 2" - 2 ./
uj - 2 " - 1 (u,, + U,~) + ~ (Uo + U ~ (26)

We have 2 = (2 + R ) / ( 2 - - R ) , where R = vc3.,./v, with vc being the vertical

velocity in the interior of the flow channel. The U~ and U ~ stand for the
slip velocities at the top and b o t t o m walls, respectively. Once the slip
126 He e t a l .

velocities are set to zero, it is easy to prove that this solution is a second-
order approximation of the analytical solution:
e R e ,y/L __ 1 eRe _ ere .y/L
u eR'- 1 u,, + e Re- 1 u0 (27)

where the Reynolds number Re = vbL/v, with L being the width of the flow
region. Note that with Re in Eq. (27) replaced by Re* = n In(2), Eq. (26)
differs from the analytical solution only by the slip velocities at the
boundaries. The difference between Re and Re* reflects the finite-difference
errors and it vanishes when 6,./L tends to zero.
Again, the slip velocities depend on how the boundary condition is
implemented. Following the same procedure in deriving Eq. (10), we have
the following recurrence relationships for the velocity near the bottom wall:

U2[2--1,101.) 0 u,~+uo--2u I v - - 1
26 v - fi2 +----~ [~o--Uo] (28)

v{--vo=(2r-1)(vo-vt)c+2(r--1)(Oo--Vo)C (29)

where

~o = c[ ( f ~ _ f o ) + ( f ~ _ f o ) + ( f o _fo)]
~o=c[(f~176 + fs+f
0
~ -
0
(f7 +/~)]
'0

Recall that in the derivation, it is assumed that f r 1 7 6 .1 7 6 co r s follow the

rules (8), the equilibrium distribution at j - - 0 is calculated using velocity
(uo, %), and f o , f o f o depend on the boundary condition implemented.
Obviously, the vertical velocity in the interior of the flow domain may
not equal the injection velocity if a boundary condition does not satisfy
~o = Vo. The interior vertical velocity can be obtained by solving the quad-
ratic equation (29).
As to the slip velocity, by comparing Eq. (26) with Eq. (28), we have

UO = 1 2 ( r - 1) Uo- Uo (30)
2 r - 1 2 + Rh

where RI, = vhfi,./v. The same result can be obtained at the top boundary:

1 2 ( r - 1) ~ , , - u , ,
U.',!= 2 r - I 2--Rh (31)

With the above results, we can explicitly analyze the different schemes
for hydrodynamic boundary condition.
Lattice Boltzmann BGK Model 127

1. Equilibrhtm scheme. For a boundary condition with nonzero

velocity, the bounce-back scheme is obviously not valid. A common practice
in the LBE simulation is to assign the corresponding equilibrium distribu-
tion to the distribution functions at a boundary node. In our case, applying
this scheme at the bottom boundary gives

fo=~
PI1 + 3 v ~ c-;
_3.oi
2c2J

f~ l + 3 U~+ 3 v~
c + 3 ~-7+3--s+9c-v8

fo= 1 _ 3 u o + 3 Vo+3__~+
C C C-
3 -9
C-

Notice that assigning the equilibrium distribution to rJ l ~~J 3 c~ is consistent

with Eqs. (8), and the values off4,fT,J8
o 0 .o are irrelevant to the flow inside.
0 0 '0
Here we usef4,fv,J8 as given in Eqs. (8), so that Eqs. (28) and (29) are
still valid for a simpler presentation. With this information, it is easy to
derive

~o-Vo =
1( Vo-vl+V~
2)+ ( v l - vo) (32)

Substitution of this into Eq. (29) leads to vt = %, and thereafter, vj = vb, Vj.
In the horizontal direction, it is easy to prove

U~ ( I- ~(u,-uo)

and similarly at the top boundary

u',!= 1 - ~Z-ktu.,_,-.,,)

Substituting Eq. (26) with j = 1 and j = n - 1 into above two equations, we

can obtain, the final explicit expression for the slip velocities:

UO = (~- 1)(2- 1)(r2- r-2) (u,,-Uo) (33)

" 2"(r2-- r + 1) + 2 ( r 2 - r - 2)
( 3 - 1)(2-- 1) 2"(T2-- r + 1)
Utl
s 2,,(r2_r+l)+2(r2_r_2)(u,-uo) (34)

822/87/I-2-9
128 He e t a / .

Obviously, the equilibrium scheme generally yields a nonzero slip velocity.

Let ~,.--, 0; it is easy to prove

u? ~ ! u: ~ -1
// n

This states that the slip velocity is of first order in space.

2. Nonslip schemes. With the nonslip schemes in refs. 2 and 3 we

have

uo = uo, vo = Vo, /~,, = u,,, ~,, = v,, (35)

Obviously, this boundary condition yields a zero slip velocity and hence
the boundary condition can be implemented correctly.

We also carried out numerical simulations for the Couette flow with
injection at boundaries. The system is exactly the same as the one used in
simulations for the Poiseuille flow, i.e., N.,. • N,.= 16 • 4. The bottom wall
is fixed and the top wall moves at a speed of 0.1c. For consistency, we
adjusted the value of the vertical injection velocity to yield the same
Reynolds number Re = v~,L/v.
Figure 4 shows the velocity profile for Re = 2.0 with different values of r.
The theoretical solutions of the Navier-Stokes equation are also included
for comparison. The results with the equilibrium scheme yield a nonzero
slip velocity as long as r :~ 1, while the schemes in refs. 2 and 3 yield a zero
slip velocity regardless of the value of r. All the numerical results coincide
with our analytical solution. Figure 5 shows the slip velocity for Re = 1.0
with the equilibrium scheme as a function of r. Again, the numerical results
agree excellently with our analytic results.

2.3. Slip Velocity and Momentum Exchange

Another interesting phenomenon we found is that the slip velocity is
highly related to the m o m e n t u m exchange at boundaries. Let us introduce
the quantity

cV
AM=~[ " " .,, v (36)
o,A

where V is the volume of the unit cell and A is the surface area between
two adjacent cells through which the m o m e n t u m is transferred. In the 2D
nine-bit model, V = ~. and A = ~5.. The physical significance of A M is very
Lattice Boltzmann BGK Model 129

1.5, i i i i i

1.0

0.5

0.0

0 1 2 3 4
J
Fig. 4. The normalized velocity profile of the Couette flow with R e = 2 . 0 . Dashed line,
theoretical solution of the Navier-Stokes equation; solid lines, our analytic results with two
types of boundary conditions and different values of r. I D) The numerical results of lattice
BGK simulation with the schernes in refs. 2 and 3; ( + ) the numerical results with the eqtfi-
librium scheme and with r = 0 . 7 5 ; ( A ) the results with the equilibrium scheme and with
t- = 2.0.

0.5

0.0

-0.5

-l.O r i

0 1 2 3
T

Fig. 5. The normalized slip velocity IRe = 1.0) o f the Couette flow as a function o f t and the
nunlber of nodes across the channel, n , with the equilibrium scheme. Solid lines, our analytic
results, i D, + ) The slip velocity in nunaerical simulations at the top wall and with n = 4 and
8, respectively: ( O, x I the slip velocity in numerical simulations at the bottom wall and with
n = 4 and 8, respectively.
130 He e t al.

clear: 3M is the net tangential momentum exchange per time step and per
unit area across the middle surface between the wall layer and the first flow
layer. It can be shown that the slip velocity along the wall U., is a direct
consequence of inadequate momentum dissipation. For the Poiseuille flow,
we have

AM = c2[(f~ --f~) - ( f s o - f 7 ) ]o

= - c [f5 -f,,] - c-'[f~ -f~] + g puo +-6 6,pa

- - c [fs-f7]-~pu,--~6,pG

Addition of the last two equations gives

2zfM = (2--~) zfM--6 (2 - ~ ) p(u, --Uo) + (1--~) pc( ~o-- Uo) (37)

or

LI I ~ /,l 0
AM= - p v ~ +6 (3- 1) pc(~o-Uo) (38)
. v

Recall that the slip velocity

U~,.=6 ( v - 1) (Uo--Uo)
23- 1

Thus, keeping the slip velocity to be zero for an arbitrary r is equivalent

to having

2JM~- -Pll-5f~ j= l/2

provided the derivative is second-order accurate in 6.,.. In other words, the

momentum dissipation by the wall must be equal to the stress applied by
the fluid at the middle of the wall layer and the first flow layer.
A similar momentum exchange can be found for the Couette flow with
vertical injection:

Ul--UO . P.
dM-- -pv %---7---+ ~ tuL + Uo) v + (r - 1 ) pC(ao - Uo) (39)
Lattice Boltzmann BGK M o d e l 131

Again, a zero slip velocity implies that the momentum dissipation by the
wall is equal to the momentum transfer carried by the vertical velocity plus
the stress applied by the fluid (evaluated at the middle of the wall layer and
first flow layer).

2.4. Bounce Back with Halfway Wall

In the previous subsections, the bounce-back boundary condition has

been shown to yield a nonzero, first-order slip velocity at the bounce-back
row for the Poiseuille and Couette flows. On the other hand, the bounce-
back boundary condition has its own merit: easy implementation, which
has been considered as one of the advantages of LGA and LBE. Although
being attractive, the new boundary conditions proposed in refs. 1-3 have
difficulties in implementation for a complex geometry like that in porous
media, in which boundaries usually do not orient in either x or y direction.
Besides, some additional considerations are needed for corner nodes. If a
"complete" bounce-back rule, with every distribution simply reversing its
direction, can work with a halfway wall, it will be very useful in practice.
By a "halfway wall" we mean a wall placed at halfway between the bounce-
back row and the first flow row.
For a stationary wall, it has been proved that the bounce-back bound-
ary condition with the halfway wall has a second-order accuracy for some
simple flows in the cases of F H P LGA ~~51and L C H C LBE 1161for a suitable
choice of the second eigenvalue of the collision matrix. This linear analysis
was of the first order in gradient in ref. 15 and second order in ref. 16. In
ref. 18 the theoretical result of Poiseuille flow with the square forcing ~2~)
(equivalent to bounce-back boundary condition with a halfway wall) for
F H P L B G K indicates a second order of accuracy. Using the results presented
in previous subsections, we can prove that the bounce-back boundary con-
dition with halfway wall is of second-order accuracy for the present L B G K
models for the stationary walls of the simple flows considered. We will also
give the explicit expression for the slip velocity.
First let us consider the bounce-back with halfway wall for the
Poiseuille flow. The setup of the system is a little different from previous
subsections. Nodes with j = 1..... n - 1 are inside the flow and nodes j = 0,
n are the "bounce-back rows. After a complete bounce back, we have
f21 = f 4o, f ~ = f 7 o, f 6i= f s ,o. the exchange o f f t ~ with f o3 has no effect on the
flow. The walls are located at j = 1/2, j = n - 1/2. The evolution equation is
the same as given in Eq. (1), but the width of the channel now becomes
L=(n-1) fix instead of L=n~,.. It has been previously (in Section2.1)
shown that with the bounce-back boundary condition at both bottom and

822/87/I-2-10
132 He e t al.

top walls, the velocity profile of the L B G K model for the Poiseuille flow
[Eqs. (15) and (20)] is

3G6, j(n-j)-t G6, [(2z- 1)(4r-3)-3n] (40)

ui = 9r - 1 2(2r - 1 ) 172

For the setup of halfway wall, the forcing G is related to the exact central
velocity 0,. by U,. = L2G/8v, where L = (n - 1 ) 6.,. is the width of the chan-
nel in this setup. Using 0,., we find the velocity of the L B G K model for
Poiseuille flow:

40,. + 2 U,.[ ( 2 r - 1 ) ( 4 r - 3 ) - - 31,)]

uj = (n - 1) 2 j(n-jJ 3 ( n - 1 )2 (41)

The exact solution of Poiseuille flow is

U,.(2j- 1 ) ( 2 n - 2 j - 1)
u: - )2 (42)
' (17- 1

Thus, the error of the velocity from L B G K simulation is given by

U,.[4r(4r - 5) + 3]
Ui - u} = 3(n - 1 )2 (43)

While the error of velocity at the bounce-back row j = 0 is O ( 1 / ( n - 1 ) ) ,

indicating a first-order accuracy, the velocity inside the flow has a uniform
error O(1/(17- 1 )2), indicating a second-order accuracy for a fixed r. There
is no need to adjust parameters in the model. Notice that for a fixed lattice
size, the error goes to infinity as r ~ oo. The conclusion is the same for the
F H P L B G K modelJ ~t For practical purposes, there is no need to take a
large value of r in a simulation. In fact, the factor 4 r ( 4 r - 5 ) + 3 in the
error expression has a maximal absolute value of only 3.25 for
0.5 < r ~< 1.25. Hence the bounceback with halfway wall is an acceptable
boundary condition for the Poiseuille flow. It is also noted that in this
bounce back, f ~ and f ~ can be interchanged (a complete bounce back)
without having any effect on the flow. Thus, the boundary condition is
implemented in a very simple way without distinguishing flow directions
and other directions.
Next let us consider the bounce back with halfway wall at the bottom
for the Couette flow with zero injection velocity. We will not change the
flow setup at the top wall and use a correct boundary condition like that
Lattice Boltzmann BGK Model 133

in refs. 2 and 3 there. After some algebraic manipulations we obtain the

following x velocity:

uj=U 2(n-j) 2j-I U (44)

2n~ U = 2n---~

where U is the velocity at the top. While the velocity at the bounce-back
row ( j = 0 ) is O(1/(n- 1 )), indicating a first-order accuracy, the velocity at
the halfway wall ( j = 1/2) is exactly zero, indicating that the result is the
exact solution! These results also hold for the F H P L B G K model as given
in the Appendix. Hence the bounce back with halfway wall deserves more
attention in further simulations.

3. C O N C L U S I O N AND DISCUSSION

In this paper, we have solved the lattice Boltzmann BGK equation for
two simple flows under different boundary conditions. The solutions are
verified by independent numerical simulations. Using these solutions, we
further analyzed different schemes for implementation of the velocity
boundary condition. We found that the schemes proposed in refs 1-3 give
the zero slip velocity for the cases considered, and therefore are of second-
order accuracy in space. Other schemes, such as bounce-back, equilibrium,
etc., are only of first-order accuracy in general. The bounce-back scheme
with halfway wall, however, is of second-order accuracy for the flows con-
sidered. Another interesting phenomenon we found is that the slip velocity
is highly related to the momentum exchange at boundaries. A boundary
condition is correct only when the momentum dissipation by the boundary
is equal to the stress of the fluid. We hope this work can provide some help
for future studies of boundary conditions. Needless to say, our analysis is
only carried out for the 2D Poiseuille flow and the 2D Couette flow with
injection at the boundaries. The analytical solutions for these flows are still
too simple to use as general guidance for the boundary condition analysis.
Further studies on more complicated flows are necessary.

APPENDIX. R E S U L T S FOR F H P L B G K M O D E L

For the steady Poiseuille flow, the seven-bit BGK model on a tri-
angular lattice ( F H P L B G K model) with eo = 0 and e~ = [ c o s ( ( o ( - 1) 7r/3),
sin((c~-1) ~z/3)] c, ~ = 1,..., 6, can be reduced as the following recursion
forms:
134 He e t al.

f~ = -~rlp [6(1--r)+UJc']+(1-!)f~-'+6" 4cz pG

b,.
A = ~ p1 [ 6 ( I - , ' ) U -J c ' ] + ( 1 - - ~ ) f ~ - ' 4c,_PG (A1,

fJ4=~ p (1--r)-2Ug+3c ---~c'- rpG

f'~=gp 6{1-r)-"Jc' + 1- -4c--~pa

1 [ 6 ( 1 - , 9 + U'c']+(l-~)fJ6+t+~pG
fJ6=-~rp

where r is the fraction of rest particle mass density out of the total mass
density; it can be adjusted between 0 and 1. The body force F, with
amplitude [[F[I =pG, is along the direction of e~. The system satisfies the
Navier-Stokes equation (6) with the kinetic viscosity

(2r - 1 ) 6.~
v= - - (A2)
8 6,

and the speed of sound

1--r
c.,.=J-~-c (A3)

Inside the flow domain, it is easy to derive the same solution for u as
in the square-lattice case:

4U,.j(n -j)
us= n 2 + Us
W here Uc=L-G/8v is the central velocity as before. L=n6,., with
") " "

~,, = ~/3di,,/2 being the vertical distance between two adjacent lattice lines.
Lattice Boltzmann BGK Model 135

The slip velocity U, depends on the boundary condition. The boundary

conditions in refs. 1 and 3 generate a zero slip velocity, hence give the exact
solution of the Poiseuille flow.
With the bounce-back boundary condition, for example, at the bot-
tom, f ~ = f o , f 3 1_
- - f 6 , o the slip velocity can be derived as

2U,.(n+ I) 3(2zz-33+2)
Us - n2 + 2(2r - 1 ) 6,G (A4)

which is of first order in general.

Similarly, for the case of bounce back with halfway wall, the error of
the velocity of the L B G K is derived as

0c(432 - 63 + 1 )
uj-u~- (n- 1) 2 (A5)

the same result as in ref. 18, which is of second-order accuracy. The factor
4 3 2 - 6 3 + 1 takes a maximal absolute value of 1.25 for 0.5 <r~.< 1.25.
Notice also that a complete bounce back can be used for the bounce back
with a halfway wall.
For the Couette flow with v = 0 , the model at the beginning of this
appendix can be used with G = 0. Suppose that we use a correct boundary
condition as in refs. 2 and 3 at the top wall and use bounce back at the
bottom; then the x velocity is

uj= U 2(n-j) U 2j--__~l U (A6)

2n - 1 2n - 1

where U is the velocity at the top. This is the same result as in the square-
lattice case, and bounce back with halfway wall still gives the exact
solution.
Other boundary conditions can be studied as well.

REFERENCES
1. D. R. Noble, S. Chen, J. G. Georgiadis, and R. O. Buckius, A consistent hydrodynamics
boundary,condition for the lattice Boltzmann method, Phys. Fluids 7:203 (1995).
2. D. R. Noble, J. G. Georgiadis, and R. O. Buckius, Direct assessment of lattice Boltzmann
hydrodynamics and boundary conditions for recirculating flows, J. Stat. Phys. 81:17
(1995).
3. T. Inamuro, M. Yoshino, and F. Ogino, A non-slip boundary condition for lattice
Boltzmann simulations, Phys. Fluids 7:2928 (1995); Erratum, 8:1124 (1996).
4. G. McNamara and G. Zanetti, Use of the Boltzmann equation to simulate lattice-gas
automata, Phys. Rev. Lett. 61:2332 (1988).
136 He e t al.

5. F. Higuera and J. Jimenez, Boltzmann approach to lattice gas simnhttions, Europhys. Lett.
9:663 (1989).
6. H. Chen, S. Chen, and W. H. Matthaeus, Recovery of tile Navier Stokes equations using
a lattice Boltzmann method, Phys. Rev. A 45:R5339 (1991).
7. Y. H. Qian, D. d'Humi6res, and P. Lallemand, Lattice BGK models for the Navier-Stokes
equation, Europh.vs, Lett. 17:479 (1992).
8. R. Benzi, S. Succi, and M. Vergassola, The lattice Boltzmann equation: Theory and
applications, Phys. Rep. 222:145 (1992).
9. X. Shan and H. Chen, Lattice Boltzmann model for simulating flows with multiple phases
and components, Plo,s. Rev. E 47:1815 (1993); Simulation of non-ideal gases and liquid
gas phase transitions by the lattice Boltzmann equation, Phys. Rei,. E 49:2941 (1994).
10. A. J. C. Ladd, Numerical simulations of particulate suspensions via ti discretized
Boltzmann equation. Part I. Theoretical foundation, J. Fhdd Mech. 271:285 (1994):
Numerical simulations of particulate suspensions via a discretized Boltzmann equation.
Part II. Numerical restllts, J. Fhdd Mech. 271:311 (1994).
11. S. Chen, H. Chert, and W. H. Matthaeus. Lattice Boltzmann magnetohydrodynamics,
Phys. Rev. Lett. 67:3776 11991 ).
12, S. P. Dawson, S. Chen, and G. Doolen, Lattice Boltzmann computations Ibr reaction-
diffusion equations, J. Chem. Phys. 98:1514 ( 1993 ).
13. S. Hou. Q. Zou, S. Chen. G. D. Doolen, and A. Cogley, Simulation of cavity flow by the
lattice Boltzmann method, J. Comp. Phys. 118:329 (1995).
14, P. A. Skordos, Initial and boundary conditions for the hittice Boltzmann method, Phys.
Rev. E 48:4823 (1993).
15, R. Cornubert, D. d'Humi6res, and D. Levermore, A Knudsen hlyer theory tbr hittice
gases, Physica D 47(6):241 (1991).
16. I. Ginzbourg and P. M. Adler, Boundary condition analysis for the three-dimensional
lattice Boltzmann model, J. Phys. II France 4:191 (19941.
17. L-S. Luo, H. Chen, S. Chen, G. D. Doolen, and Y.-C. Lee. Generalized hydrodynamic
transport in lattice-gas automata, Phys. Rev. A 43:R7097 (1991).
18. Q. Zou, S. Hou, and G. D. Doolen. Analytical solutions of the lattice Boltzmann BGK
model, J. Star. Phys. 81:319 (1995).
19. U. Frisch. D. d'Humi6res. B. Hasslacher, P. Lallemand, Y. Pomeau, and J.-P. Rivet,
Lattice gas hydrodynamics in two or three dimensions, Complex Systems 1:649 (1987).
20. A. K. Gunstensen and D. H. Rothman, Microscopic modeling of immiscible fluids in three
dimensions by a lattice-Boltzmann method, MIT Porous Flow Project, Report No. 4, 20
1991 ).
21. L. P. Kadanoff, G, R. McNamara, and G. Zanetti, From automata to fluid flow: Com-
parisons of simulation and theory, Phys. Rer. A 40:4527 [1989).