PHYSICAL REVIEW E 76, 036711 共2007兲

Kinetic lattice Boltzmann method for microscale gas flows: Issues on boundary condition,
relaxation time, and regularization
Xiao-Dong Niu,* Shi-Aki Hyodo, and Toshihisa Munekata
Computational Physics Laboratory Toyota Central R&D Laboratories, Inc., Nagakute, Aichi, 480-1192, Japan

Kazuhiko Suga
Department of Mechanical Engineering, Osaka Prefecture University, Sakai 599-8531, Japan
共Received 7 May 2007; published 27 September 2007兲
It is well known that the Navier-Stokes equations cannot adequately describe gas flows in the transition and
free-molecular regimes. In these regimes, the Boltzmann equation 共BE兲 of kinetic theory is invoked to govern
the flows. However, this equation cannot be solved easily, either by analytical techniques or by numerical
methods. Hence, in order to efficiently maneuver around this equation for modeling microscale gas flows, a
kinetic lattice Boltzmann method 共LBM兲 has been introduced in recent years. This method is regarded as a
numerical approach for solving the BE in discrete velocity space with Gauss-Hermite quadrature. In this paper,
a systematic description of the kinetic LBM, including the lattice Boltzmann equation, the diffuse-scattering
boundary condition for gas-surface interactions, and definition of the relaxation time, is provided. To capture
the nonlinear effects due to the high-order moments and wall boundaries, an effective relaxation time and a
modified regularization procedure of the nonequilibrium part of the distribution function are further presented
based on previous work 关Guo et al., J. Appl. Phys. 99, 074903 共2006兲; Shan et al., J. Fluid Mech. 550, 413
共2006兲兴. The capability of the kinetic LBM of simulating microscale gas flows is illustrated based on the
numerical investigations of micro Couette and force-driven Poiseuille flows.

DOI: 10.1103/PhysRevE.76.036711 PACS number共s兲: 47.11.⫺j, 51.10.⫹y

I. INTRODUCTION In the Grad’s moment method 关9–11兴, the BE is replaced

Gas flows in microscale devices have received particular
attention over the past decade with the rapid development in
microelectromechanical systems 共MEMS兲 关1–3兴. In these mi-
croscale devices, the flows are usually distinguished by rela-
tively small Mach numbers 共Ma= U / cs ⱕ 0.3, where U is the
characteristic velocity of the flow and cs is the sound speed兲
and large Knudsen numbers 共Kn= ␭ / H ⱖ 0.01, where ␭ is the
molecular mean free path of fluid and H is the characteristic
length of the flow domain兲. It is well known that processes in
these kinds of flows are described by the Boltzmann equation
共BE兲 of the kinetic theory 关4,5兴. Since the numerical solution
of the BE, either directly 关6兴 or via the direct simulation
Monte Carlo 共DSMC兲 method 关7兴, is very time expensive,
there is a strong desire for accurate models which allow
simulations of processes in microscale gas flows at lower
computational cost.
Currently, there are three well-known approaches toward
this goal, the Chapman-Enskog 共CE兲 method 关4,5,8兴, the
Grad’s moment method 关9–11兴, and the lattice Boltzmann
method 共LBM兲 关12,13兴. In the CE method, the phase density
is expanded in powers of Kn, and to different successive
orders the expansion yields the Euler, Navier-Stokes 共NS兲,
Burnett equations, and so on. The NS equations cease to be
accurate for Kn larger than 0.01, and although the Burnett
equations are theoretically valid for larger Knudsen numbers,
their numerical solutions become linearly unstable for pro-
cesses involving small wavelength or high frequencies 关14兴.
*Author to whom correspondence should be addressed;
e1351@mosk.tytlabs.co.jp
by a set of moment equations, first-order partial differential
equations for the moments of the distribution function. For
the closure of the equations, the phase density is approxi-
mated by a expansion in the Hermite orthogonal polynomials
about local Maxwellian equilibrium distribution, and the co-
efficients of the polynomials are related to the moments.
Only a few moments have an intuitive physical meaning, i.e.
density, momentum density, energy density, heat flux, and
pressure tensor. The set of 13-moments forms the basis of
Grad’s well-known 13-moment equations 关9兴. However, the
13-moment set does not allow the computation of Knudsen
boundary layers 关15,16兴 and, with an increasing number of
moments 关17兴, the Knudsen boundary layers can be com-
puted, but solving these moment equations becomes formi-
dable due to its complexities.
The LBM 关12,13兴 is a simplified solver of the BE on a
discrete lattice. The LBM is originated from the lattice gas
cellular automaton models 关18–20兴, and quickly accepted as
an efficient computational fluid dynamics 共CFD兲 solver 关21兴
due to its distinctive computational features: easy to imple-
ment, intrinsically parallelizable, and straightforward for
handling complex geometries. Because of its intrinsic kinetic
nature, the LBM seems to allow the microscopic physics
responsible for many complex fluid phenomena to be mod-
eled more directly 关22–28兴. However, since the early LBM
models only satisfy the rotational invariance of hydrody-
namic properties up to the Navier-Stokes order, an argument
关29兴 usually arises, especially when they are applied in simu-
lations of high-Knudsen number flows 关30–39兴, in which the
higher-order moments are manifested.
Recently, inspired by the Grad’s moment method, a rigor-
ous and systematic theoretical procedure for the LBM, in-

1539-3755/2007/76共3兲/036711共8兲 036711-1 ©2007 The American Physical Society

NIU et al. PHYSICAL REVIEW E 76, 036711 共2007兲

cluding effects of the higher-order moments, was developed ⳵ f␣ f ␣ − f ␣共0兲

关40兴. By using the Hermite expansion approach, hydrody- + ␰␣ · ⵜf ␣ = − + F␣ , 共4兲
⳵t ␶
namic moments at various orders can be precisely and ex-
plicitly determined at a given order of truncations of the where F␣ = −a · ⵜ␰␣ f ␣ is the contribution of the force term,
Hermite polynomials. On this point, one can think that the
and f ␣共0兲is the nth-order truncated Hermite expansion of the
LBM is equivalent to the Grad moment method.
Maxwellian distribution 共2兲 at a set of discrete velocities ␰␣ 共
In this paper, we further develop the above kinetic theory
␣ = 0 , 1 , 2 , . . . , d, and d depending the truncation order N of
of the LBM by addressing the issues of the boundary condi-
the Hermite expansion兲 and is written as
tion, the relaxation time, and the nonequilibrium moments.
In Sec. II, we give a brief description of the lattice Boltz- N
mann equation 共LBE兲 including higher-order moment effects 1 共n兲 共n兲
f ␣共0兲 = ␻␣ 兺 a H 共␰␣兲, 共5兲
关40兴. The accuracy issue required by describing fluid proper- n=0 n! 0
ties is discussed. In Sec. III, three main topics are further
共0兲 共n兲
discussed; first, the boundary condition. Derivation of the where a共n兲
0 ⬇ 兺␣=1 f ␣ H 共␰␣兲 is the Hermite expansion coef-
d
diffuse scattering boundary condition 关30,41,42兴 on the Her- ficient, ␻␣ is the weight function, and H共n兲共␰␣兲 is the
mite basis is presented. Second, we discuss the relaxation nth-order Hermite polynomial, the first few terms of which
time. In the presence of the solid wall, an effective relaxation can be expressed as
time is introduced. The last topic is related to the nonequi-
librium distribution. To preserve the nonequilibrium hydro- H共0兲共␰␣兲 = 1,
dynamic moments, a regularization procedure 关38,43兴 of the
nonequilibrium distribution is discussed, and its effects on
the Knudsen layer are highlighted. Sec. IV is devoted to the H共1兲共␰␣兲 = ␰␣i ,
numerical illustrations of the LBM in simulating the high-
Knudsen number flows and concluding remarks are given in
Sec. V. H共2兲共␰␣兲 = ␰␣i␰␣ j − ␦i,j ,

II. LBE WITH HIGHER-ORDER MOMENTS

H共3兲共␰␣兲 = ␰␣i␰␣ j␰␣k − ␰␣i␦ jk − ␰␣ j␦ik − ␰␣k␦i,j , 共6兲
The BE 关4,5兴 has been well accepted as a mathematical
model simulating the microscale fluidic gas in the entire where ␦ij is the Kronecker delta function. In the discrete
Knudsen regime. The BE describes evolutions of a single velocity space, the fluid variables are calculated as
particle velocity distribution function f共x , ␰ , t兲 in the phase
space 共x , ␰兲, where x denotes the position and ␰ denotes the d d d

velocity of the particle, and can be written as the following ␳= 兺 f ␣,

␣=0
␳u = 兺 f ␣␰ ␣,
␣=0
D␳RT + ␳u =2
兺 f ␣␰␣2 .
␣=0
BGK form:
共7兲
⳵f f − f 共0兲
+ ␰ · ⵜf + a · ⵜ␰ f = − . 共1兲 Obviously from Eqs. 共4兲–共7兲, the truncation level of the
⳵t ␶
Hermite expansion determines the accuracy of Eq. 共4兲 to
Here, ␶ is the characteristic relaxation time of collisions to approximate Eq. 共1兲, and the level of accuracy is increased as
equilibrium, ⵜ␰ the gradient operator in velocity space, and a higher-order terms in the truncated expansion are retained.
the acceleration due to the external or self-generated body As revealed by the CE method, by retaining up to the fourth-
force. f 共0兲 represents a local equilibrium distribution and is order terms in the Hermite expansion, the Burnett level ac-
the Maxwellian curacy pertaining to the fluid momentum evolution for iso-

f 共0兲 =
␳
共2␲RT兲D/2
exp − 冉
共␰ − u兲2
2RT
, 冊 共2兲
thermal systems can be satisfied. For most microscale gas
flows in MEMS 关1–3兴, the Burnett description is said to be
accurate enough for modeling them. For isothermal MEMS
where ␳, T, and u are the fluid density, temperature, and fluids, a third-order Hermite expansion is enough to model
velocity, respectively, D is the dimension of the space, and R the momentum equation at the Burnett level due to the small
is the gas constant. Hydrodynamic variables, such as the den- Mach numbers and the equilibrium distribution function f ␣共0兲
sity, velocity, and temperature are defined, respectively, as in has the following form 关40兴:

再 冋 册
the following form:
␰␣ · u 1 共␰␣ · u兲2 u2
␳= 冕 fd␰, ␳u = 冕 f ␰d␰, D␳RT + ␳u2 = 冕 f ␰ 2d ␰ .
f ␣共0兲共x,t兲 = ␻␣␳ 1 +
RT
+
2 共RT兲2
−
RT

共3兲 + 冋
1 共␰␣ · u兲2
6 共RT兲2
− 3
u2 ␰␣ · u
RT RT
.册 冎 共8兲
As argued by Grad 关9兴, the Boltzmann BGK equation 共1兲
can be projected onto the Hermite orthogonal basis and we Similarly, the force distribution term F␣ in Eq. 共4兲 can be
have 关40兴 written as

036711-2
KINETIC LATTICE BOLTZMANN METHOD FOR… PHYSICAL REVIEW E 76, 036711 共2007兲

TABLE I. D2Q21 and D3Q39 discrete velocity models and the
corresponding weighting functions.
time ␶ in terms of Kn 关27–32,35,36,38–40,42,43兴. On the
other hand, when solid wall boundaries are presented,
the slip boundary condition should be imposed
关27–33,35,39,41–43兴 and its effects on the relaxation time
should be reflected 关2,29兴. Furthermore, as seen from the
above Gauss-Hermite procedure, the nonequilibrium distri-
bution f ␣⬘ generally does not lie in the space spanned by the
leading N Hermite orthogonal basis, and therefore the Gauss-
Hermite procedure introduces an error to the described fluid
systems. When Kn is small, such an error is small and ignor-
able. However, as Kn increases, the error becomes larger and
the correct hydrodynamics of the system will be blurred. To
avoid this and improve isotropy of the system, a “regulariza-
tion” procedure 关38,43兴 is introduced.

再冋 冉
A. Diffuse scattering boundary condition

F␣共x,t兲 = ␻␣␳
␰␣ · a
RT
1+
␰␣ · u
RT
−
a·u
RT
冊 册 From the kinetic theory, the slip boundary condition in
solving the BE 共1兲 is described by the gas-surface interaction

冋 冉 冊 册冎
law 关4,5兴. Taking into account the wall type via the non-
1 共␰ · a兲 ␰ a␰ negative scattering kernel R共␰⬘ − ␰ , x , t兲, representing a prob-
+ 关␴ij + ␳u2兴: ␣ H共2兲 ␣ − 2 ␣ ,
2␳RT 共RT兲 RT RT ability density that particles colliding the wall with velocity
共9兲 between ␰⬘ and ␰⬘ + d␰⬘ at location x and time instance t will
be reflected with velocity between ␰ and ␰ + d␰, the boundary
where ␴ij = 兺␣d =0 f ␣⬘ ␰␣i␰␣ j is the stress tensor and f ␣⬘ = f ␣ − f ␣共0兲 condition for the distribution f共x , ␰ , t兲 can be given as
is the nonequilibrium part of the distribution function f ␣. The
choice of ␰␣ depends on the quadrature in evaluating the
moments of Eq. 共7兲. In general, the quadrature must be ac-
f共x, ␰,t兲 = 冕cn⬘⬍0
兩cn⬘兩R共␰⬘ → ␰,x,t兲f共x, ␰⬘,t兲d␰⬘ , 共12兲
curate enough so that not only the conservation constraints
are preserved, but also necessary symmetry is retained. Ac- where cn⬘ = 共␰⬘ − uw兲 · n, with n being the unit wall vector nor-
cording to previous work 关40兴, the Gauss-Hermite quadra- mal to the boundary surface and the subscription w meaning
tures employed for solving the third-order truncated system the wall. If there is no absorption on the wall, the kernel
are based on the D2Q21 and D3Q39 discrete velocity mod- satisfies the normalization and reciprocity conditions, which
els for the 2D and 3D Cartesian lattices, respectively. In are expressed respectively as
these two quadratures, the sound speed cs = 冑RT is uniquely
equal to 冑2 / 3. The discrete velocities ␰␣ and weights ␻␣ are
listed in Table I. 冕 cn⬘⬎0
R共␰⬘ → ␰,x,t兲d␰⬘ = 1, 共13兲
With the above Cartesian quadratures, Eq. 共4兲 can be sim-
ply discretized in physical space and time, and yields the and
following LBE:
兩cn⬘兩R共␰⬘ → ␰,x,t兲f w共0兲共x, ␰⬘,t兲
␦t
f ␣共x + ␰␣␦t,t + ␦t兲 = f ␣共x,t兲 − „f 共x,t兲 − f ␣共0兲共x,t兲… = 兩cn兩R共− ␰ → − ␰⬘,x,t兲f w共0兲共x, ␰,t兲, 共14兲
␶ + 0.5␦t ␣
␶␦t where f w共0兲共x , ␰ , t兲 is the wall equilibrium distribution func-
+ F 共x,t兲 共10兲 tion. Different types of the scattering kernels express differ-
␶ + 0.5␦t ␣
ent gas-surface interactions and the most well-known Max-
with the density, velocity, and pressure calculated as well’s kernel has the following form:

␳= 兺
d

␣=0
f ␣, ␳u =
d

兺 f ␣␰␣ + 2 a,
␣=0
␦t
p = ␳cs2 . 共11兲 R共␰⬘ − ␰,x,t兲 =
1
␳
冑 2␲ 共0兲
f 共x, ␰,t兲.
RT w
共15兲

By using the Gauss-Hermite quadrature 关40兴, Eqs. 共12兲–共14兲

yield the following Maxwell’s diffuse-scattering boundary
III. BOUNDARY CONDITION, RELAXATION TIME, condition in the LBM frame 关30,41,42兴:
AND REGULARIZATION

In microscale gas flows, the degree of systems deviating 兺 兩共␰␣⬘ − uw兲 · n兩f ␣⬘共x,t兲
␣⬘
from the equilibrium is measured by the Knudsen number. f ␣共x,t兲 = f ␣共0兲,w共x,t兲,
Therefore, in order to use the LBE 共10兲 to model the micro- 兺 兩共␰␣⬘ − uw兲 · n兩f ␣⬘ 共x,t兲
共0兲

scale gas flow systems, it is better to define the relaxation ␣⬘

036711-3
NIU et al. PHYSICAL REVIEW E 76, 036711 共2007兲

Bounded wall
boundary
far away enough 共continuum flows兲, ␭* should approximate
Collision Collision to ␭ and thus satisfies limKn→0⌿共Kn兲 = 1. According to pre-
vious investigations 关2,29兴, the function ⌿ can be best ex-
λ∗ pressed as
λ
2
⌿共Kn兲 = arctan共aKn−b兲, 共22兲
␲
Collision Collision
where the coefficients a and b depend on the boundary types.
(a) (b) For microscale flows confined between two infinite parallel
walls, it is best to set a = 冑2 and b = 3 / 4 based on empirical
FIG. 1. Definition of molecular mean free path in unbounded investigation 关29兴. Accordingly, Eq. 共22兲 gives the effective
and bounded systems. 共a兲 Unbounded system. 共b兲 Bounded system. viscosity and relaxation time as

关共␰␣⬘ − uw兲 · n ⬍ 0; 共␰␣− uw兲 · n ⬎ 0兴, 共16兲 ␮* = ␮⌿共Kn兲, 共23兲

and
where f ␣共0兲,w共x , t兲 = f ␣共0兲共␳w , uw , x , t兲. Previous theoretical analy-
sis 关42兴 based on a 2D constant density flow along a straight
plate has shown that the above equation generates the fol- ␶* = 冑 2
␲
KncsH⌿共Kn兲, 共24兲
lowing higher-order slip boundary condition

冉 冊 冉 冊
respectively.
⳵u ␭2 ⳵ u 2
us − uw = ␭ + + ¯, 共17兲
⳵n w 2 ⳵n w
C. Regularization
provided that the Cartesian lattice spacing is in the order of The regularization procedure was introduced by Chen et
the molecular mean free path. al. 关38兴 to guarantee the nonequilibrium moments of the
LBM satisfied in the Hermite space. The regularization is
B. Relaxation time implemented before the collision of the particles. In the regu-
In kinetic theory 关4,5兴, the relaxation time ␶ can be de- larization procedure, the nonequilibrium distribution f ␣⬘ is
fined in terms of viscosity ␮ as also projected on the N Hermite-truncated basis by using the
Gauss-Hermite procedure, and we have
␮
␶= 共18兲
冉冑 冊
, ⬃ N
1 共n兲 共n兲 ␰␣
p f ⬘ = ␻␣ 兺 a H 共25兲
␣ n=0 n! RT
and the viscosity is proportional to a qualitatively defined
molecular mean free path ␭, and for hard sphere gases it is with the Hermite coefficients a共n兲 = 兺␣d =0 f ␣⬘ H共n兲共␰␣ / 冑RT兲. Cor-
expressed as responding to the first three Hermite-truncated expansions

␭=
␮
p
冑 ␲
2RT
. 共19兲
where the LBE 共10兲 is obtained, Eq. 共25兲 can be written as
the following generic form by using the mass and momen-
tum conservations:
Consequently, the relaxation time can be further written as

␶= 冑 2
冑 2
⬃
f ␣⬘ = ␻␣ 冋 1
2cs2
H 冉 冊兺
共2兲 ␰␣
cs
d

␣=0
f ␣⬘ ␰␣i␰␣ j

册
␭cs = Kn csH. 共20兲

␲ ␲
Here the definition of the sound speed cs = 冑RT is used.
It is important to note that the mean free path given by
Eq. 共19兲 is only valid for unbounded systems. In a micro-
scale gas flow system confined by the solid boundaries, some
molecules will hit walls and their flight paths may be shorter
than the molecular mean free path ␭ defined in the un-
bounded systems 共see Fig. 1兲. Therefore, the molecular mean
free path in a bounded system should be modified to reflect
the boundary wall effects 关2,29兴. Here we use an effective
mean free path ␭* to denote the property of gas flows in the
bounded system, and it can be formally expressed as
␭* = ␭⌿共Kn兲.
Here Kn is still the conventional Knudsen number without
boundary effects. Intuitively, when the boundary walls are
冉 冊兺
d
B ␰␣
+ 3 H共3兲
cs
f ␣⬘ ␰␣i␰␣ j␰␣k , 共26兲
6cs ␣=0
where the additional coefficient B is introduced in the second
term and is properly given by B = 1 − ⌿共Kn兲. Obviously,
when Kn goes to zero, the contribution of the third-order
nonequilibrium moment vanishes, and the Navier-Stokes or-
der hydrodynamics is recovered.
Mathematically, the regularization procedure serves as a
filter and ensures the nonequilibrium distribution remains in-
side the defined Hermite space by filtering out the higher-
order nonequilibrium moments not supported by the defined
共21兲 Hermite basis. Besides improvement of stability and isotropy
of the LBE 关38兴, the other significance of the regularization
procedure that needs to be addressed here is the Knudsen
layer, a region that the non-Newtonian stress/strain-rate rela-

036711-4
KINETIC LATTICE BOLTZMANN METHOD FOR… PHYSICAL REVIEW E 76, 036711 共2007兲

tionship exists, can be given correctly to some extent. This is 1

Kn = 0.1 (17 × 17)
because the nonlinear fluid behaviors in the Knudsen layer Kn = 0.1 (33 × 33)
are mostly influenced by the higher-order nonequilibrium 0.8
Kn = 0.1 (65 × 65)
Kn = 1 (17 × 17)
moments, and the regularization procedure ensures the con- Kn = 1 (33 × 33)
sistent description of the higher-order hydrodynamic mo- Kn = 1 (65 × 65)
Kn = 10 (17 × 17)
ments in the LBM frame. With the above regularization, the 0.6 Kn = 10 (33 × 33)
Kn = 10 (65 × 65)
discrete distribution function f ␣ can be written as

Y
⬃
0.4
f ␣ = f ␣共0兲 + f ␣⬘ , 共27兲
and the LBE 共10兲 becomes 0.2
␶ − 0.5␦t ⬃
f ␣共x + ␰␣␦t,t + ␦t兲 = f ␣共0兲共x,t兲 + f ⬘ 共x,t兲
␶ + 0.5␦t ␣ 0
0 0.2 0.4 0.6 0.8 1
␶␦t U
+ F 共x,t兲 共28兲
␶ + 0.5␦t ␣ FIG. 2. Normalized velocity profiles for planar Couette flows at
different Knudsen numbers and grid sizes.
when the flow system is unbounded.

which compares the normalized velocity profiles of Kn

IV. NUMERICAL ILLUSTRATIONS
To demonstrate advantages of the kinetic LBM introduced
above, the planar Couette and force-driven Poiseuille flows
for a range of Knudsen numbers are simulated in this section
and accuracy of the present method is verified against DSMC
approach 关44兴 and the linearized BE method 关45兴. For con-
venience, hereafter we shall refer to the method based on Eq.
共10兲 as the LBE method and the one based on Eq. 共28兲 as the
LBE-REG method. Both methods employ Eqs. 共8兲, 共9兲, 共16兲,
and 共24兲 with the D2Q21 discrete velocity models.
A. Planar Couette flows
We consider a planar Couette flow confined between two
plates parallel to the x axis at y = ± H. The upper plate moves
with a constant velocity U0 = 0.1 and the lower plate remains
stationary. Initially, a linear velocity distribution is set in the
flow field. The diffuse-scattering boundary conditions of Eq.
共16兲 are used to describe the gas-surface interactions on the
plates, while periodic boundary conditions are implemented
at the inlet and outlet. To demonstrate grid independence,
simulations are carried out on three uniform grid sizes of
17⫻ 17, 33⫻ 33, and 65⫻ 65, respectively. Figure 2 illus-
trates the normalized velocity 共U = u / U0兲 profiles of Kn
= 0.1, 1, and 10 as a function of distance 共Y = y / H兲 between
the plates at different grid sizes. As shown in this figure, the
velocity profiles demonstrate the Knudsen layers near the
plates and a linear property inside the domain. With Kn in-
= 0.1, 1, and 10 between the plates obtained by the LBE and
LBE-REG methods. As shown in Fig. 3, the LBE method
gives the linear velocity profiles between the plates for all
three Knudsen numbers as compared to the LBE-REG solu-
tions. Physically, the nonlinear fluid behaviors of the Knud-
sen layer are contributions of the high-order nonequilibrium
moments. Therefore, we can rationally figure out that the
regularization of the nonequilibrium moments makes the
LBE well consistent with the coupling physics 共Navier-
Stokes order hydrodynamics and beyond兲 at macro and mi-
cro scales. The accuracy of the present LBE-REG is shown
in Fig. 4, in which the normalized velocity profiles of the
planer Couette flows at Kn= 0.1 and 1 have been compared
to the DSMC data 关44兴. The DSMC method has been gener-
ally accepted as an accurate method to model the large
Knudsen number flows. Figure 4 shows that the LBE-REG
solutions are in good agreement with the DSMC data. The
differences between the two simulations imply that more
共higher than third-order兲 terms in the Hermite polynomials
should be included in the present LBE-REG model.
1
LEB-REG (Kn = 0.1)
LBE-REG (Kn = 1)
LBE-REG (Kn = 10)
0.8 LBE (Kn = 0.1)
LBE (Kn = 1)
LBE (Kn = 10)
0.6
Y

creasing, the Knudsen layers increase. Slip velocities on the

plates are also clearly observed for these three Knudsen 0.4
numbers and their magnitudes increase as Kn becomes
larger. Also see in this figure, the results are consistent with
0.2
each other as the grid size becomes larger than 33⫻ 33.
Therefore, in the following, the results based on the grid size
of 33⫻ 33 are shown only. 0
It is known that the Knudsen layer comes from the non- 0 0.2 0.4 0.6 0.8 1
U
linear fluid behaviors near the boundaries. However, from
our earlier experiences 关30,31兴, the LBE 共10兲 cannot capture FIG. 3. Regularization effects on the normalized velocity pro-
these phenomena and this can be clearly seen from Fig. 3, files for planar Couette flows at different Knudsen numbers.

036711-5
NIU et al. PHYSICAL REVIEW E 76, 036711 共2007兲

1 1.2
LEB-REG (Kn = 0.1)
LBE-REG (Kn = 1)
DSMC (Kn = 0.1) 1
0.8 DSMC (Kn = 1)

0.8
0.6
Y

0.6

U
0.4 DSMC (Kn=0.1)
0.4 DSMC (Kn=1.0)
DSMC (Kn=10)
0.2 LBE-REG (Kn = 0.1)
0.2 LBE-REG (Kn = 1)
LBE-REG (Kn = 10)
0
0 0.2 0.4 0.6 0.8 1 0
U 0 0.2 0.4 0.6 0.8 1
Y
FIG. 4. Comparison of the normalized velocity profiles for pla- FIG. 6. Comparison of the normalized velocity profiles for
nar Couette flows at different Knudsen numbers obtained by the force-driven Poiseuille flows at different Knudsen numbers ob-
LBE-REG and DSMC approaches. tained by the LBE-REG and DSMC approaches.

However, the present method is still shown to be capable of
providing a reasonable description for the microscale gas
flows.
B. Force-driven Poiseuille flows
The second numerical illustration is the force-driven Poi-
seuille flows in a 2D channel with height H. In numerical
simulations, the flow is assumed to be static initially, and a
constant force a = 0.01 is applied in the streamwise direction
while periodic boundary conditions are used at the inlet and
outlet. To ensure the grid independence of the solutions, two
uniform grid sizes of 33⫻ 33 and 65⫻ 65 are also used for
the flows to be presented in the following, and the consistent
velocity profile results of the flows at Kn= 1 on these two
grid sets are displayed in Fig. 5.
Figure 6 shows the normalized velocity profiles, U
= u / Umax 共Umax is the maximum value in the channel兲, across
the channel for three Knudsen numbers at 0.1, 1, and 10. To
show the accuracy of the present method, the DSMC results
1.2
1
0.8
0.6
U
关44兴 are also included. As observed from this figure, the re-
sults obtained by the LBE-REG are in good agreement with
the DSMC predictions. With the increase of the Knudsen
numbers, the slip velocities at the channel walls increase. As
a result, the velocity profiles become flatter and flatter.
One of the major successes in kinetic theory is the predic-
tion of a minimum of the mass flow rate as a function of the
Knudsen number at Kn⬃ 1. It was reported in 关43兴 that the
LBE model presented in that work can predict the Knudsen
minimum. However, the simulations indicate a large differ-
ence to the linearized BE and DSMC methods 关2,45兴 due to
no consideration of boundary effects on the relaxation time
and the third-order nonequilibrium moments. The nondimen-
sionalized mass flow rate, Q = 兺Hy=0␳u共y兲 / 共␳aH2 / cs兲, as a
function of Knudsen numbers obtained by the present model,
is plotted in Fig. 7. For comparison, this figure also includes
the DSMC results 关2兴 and the linearized BE solutions 关45兴 as
well as the analytical asymptotic expressions in the zero and
infinite Knudsen limits 关4兴,
Q0 = 冑␲/共12Kn兲 + s0 + 共2s20 − 1兲2Kn/冑␲ , 共29a兲
1000
Linearized BE
DSMC
100 Asympotic solution Eq.(29a)
Asympotic solution Eq. (29b)
LBE-REG

Q 10

0.4 33 × 33
65 × 65 1

0.2

0.1
0 0.01 0.1 1 10 100
0 0.2 0.4 0.6 0.8 1
Y Kn

FIG. 5. Normalized velocity profiles for force-driven Poiseuille FIG. 7. Nondimensionalized mass flow rate as a function of
flows of Kn= 1 at different grid sizes. Knudsen numbers predicted by different methods.

036711-6
KINETIC LATTICE BOLTZMANN METHOD FOR…
Q⬁ = s⬁␲−1/2ln共2Kn/冑␲兲, 共29b兲
with s0 = 1.01615 and s⬁ = 3. As expected, the LBE-REG pre-
dicts a minimum mass flow rate around Kn= 1. As Kn de-
creases to zero limit, all simulation results agree with each
other and converge to the asymptotic solutions of Eq. 共29a兲.
This implies that Navier-Stokes order hydrodynamics are re-
covered correctly by all schemes at vanishing Knudsen num-
bers. At higher Kn, the LBE-REG simulations are close to
the asymptotic solution of Eq. 共29b兲 and exhibit an excellent
agreement with the results of the linearized BE and DSMC
methods, suggesting that the present model with effective
relaxation time and modification of the third-order nonequi-
librium moments can indeed capture the microscale gas fluid
properties at high Knudsen numbers.
V. CONCLUSIONS
In conclusion, a systematic description of the issues of the
kinetic lattice Boltzmann method for simulating the micro-
scale gas flows is presented in this paper. By using the
Gauss-Hermite quadrature, the diffuse-scattering boundary
关1兴 C. M. Huo and Y. C. Tai, Annu. Rev. Fluid Mech. 30, 579
共1998兲.
关2兴 G. K. Karniadakis and A. Beskok, Microflows: Fundamentals
and Simulation 共Springer, New York, 2001兲.
关3兴 D. J. Beebe, G. A. Mensing, and G. M. Walker, Annu. Rev.
Biomed. Eng. 4, 261 共2002兲.
关4兴 C. Cercignani, Theory and Application of the Boltzmann Equa-
tion 共Scottish Academic Press, Edinburgh, 1975兲.
关5兴 S. Chapman and T. G. Cowling, The Mathematical Theory of
Non-Uniform Gases 共Cambridge University Press, Cambridge,
England, 1970兲.
关6兴 T. Ohwada, Phys. Fluids 8, 2153 共1996兲.
关7兴 G. Bird, Molecular Gas Dynamics and the Direct Simulation
of Gas Flows 共Clarendon Press, Oxford, 1994兲.
关8兴 J. H. Ferziger and HG. G. Kaper, Mathematical Theory of
Transport Processes in Gases, 共North-Holland, Amsterdam,
1972兲.
关9兴 H. Grad, Commun. Pure Appl. Math. 2, 331 共1949兲.
关10兴 I. Müller and T. Ruggeri, Rational Extended Thermodynamics
共Springer, New York, 1998兲.
关11兴 H. Struchtrup and M. Torrilhon, Phys. Fluids 15共9兲, 2668
共2003兲.
关12兴 D. A. Wolf-Gladrow, Lattice-Gas Cellular Automata and Lat-
tice Boltzmann Models 共Springer, Berlin, 2000兲.
关13兴 S. Succi, The Lattice Boltzmann Equation: For Fluid Dynam-
ics and Beyond 共Numerical Mathematics and Scientific Com-
putation Series兲 共Oxford University Press, New York, 2001兲.
关14兴 P. Rosenau, Phys. Rev. A 40, 7193 共1989兲.
关15兴 H. Struchtrup, ZAMP 51, 346 共2000兲.
关16兴 H. Struchtrup, Some Remarks on the Equations of Burnett and
Grad Transport in Transition Regimes, IMA Volume Series
135 共Springer, New York, 2003兲.
关17兴 D. Reitebuch and W. Weiss, Continuum Mech. Thermodyn.
036711-7
PHYSICAL REVIEW E 76, 036711 共2007兲
condition is directly obtained by projecting the Maxwell ki-
netic boundary condition on the Hermite space. The relax-
ation time is linked to the Knudsen number, and a concept of
the effective mean free path is introduced in determining the
relaxation time by taking into account the boundary effects.
Significances of the high-order nonequilibrium moments and
their regularization procedure are addressed. With a link to
the numerical simulations of the Couette and Poiseuille
flows, it is argued that by introducing the effective molecular
mean free path with boundary effects and regularization pro-
cedure to guarantee the nonequilibrium moments of the
LBM to be in the Hermite space, the microscale gas flows at
a range of the Knudsen numbers can be modeled.
ACKNOWLEDGMENTS
This work were supported by Core Research for Evolu-
tional Science and Technology 共CREST兲 of the Japan Sci-
ence and Technology 共JST兲 Agency 共Grant No: 228205R兲
and the Japan Society for the Promotion of Science through
a Grant-in-Aid for Scientific Research 共B兲 共Grant No.
18360050兲.
11, 217 共1999兲.
关18兴 U. Frisch, B. Hasslacher, and Y. Pomeau, Phys. Rev. Lett. 56,
1505 共1986兲.
关19兴 S. Wolfram, J. Stat. Phys. 45, 471 共1986兲.
关20兴 G. D. Doolen, Lattice Gas Methods for Partial Differential
Equations 共Addison-Wesley, Reading, PA, 1989兲.
关21兴 S. Chen and G. D. Doolen, Annu. Rev. Fluid Mech. 30, 329
共1998兲.
关22兴 X. Shan and H. Chen, Phys. Rev. E 47, 1815 共1993兲; X. Shan
and H. Chen, Phys. Rev. E 49, 2941 共1994兲.
关23兴 X. Shan and G. Doolen, J. Stat. Phys. 81, 379 共1995兲.
关24兴 S. Chen, H. Chen, D. Martinez, and W. Matthaues, Phys. Rev.
Lett. 67, 3776 共1991兲.
关25兴 A. J. C. Ladd, J. Fluid Mech. 271, 285 共1994兲; 271, 311
共1994兲.
关26兴 S. Succi, O. Filippova, H. Chen, and S. Orszag, J. Stat. Phys.
107, 261 共2002兲.
关27兴 X. Nie, G. Doolen, and S. Chen, J. Stat. Phys. 107, 279
共2002兲.
关28兴 C. Y. Lim, C. Shu, X. D. Niu, and Y. T. Chew, Phys. Fluids 14,
2299 共2002兲.
关29兴 Z. Guo, T. S. Zhao, and Y. Shi, J. Appl. Phys. 99, 074903
共2006兲.
关30兴 X. D. Niu, C. Shu, and Y. T. Chew, Europhys. Lett. 67, 600
共2004兲.
关31兴 C. Shu, X. D. Niu, and Y. T. Chew, J. Stat. Phys. 121, 239
共2005兲.
关32兴 Y. T. Chew, X. D. Niu, and C. Shu, Int. J. Numer. Methods
Fluids 50, 1321 共2006兲.
关33兴 B. Li and D. Kwok, Phys. Rev. Lett. 90, 124502 共2003兲.
关34兴 S. Ansumali and I. V. Karlin, Phys. Rev. Lett. 95, 260605
共2005兲.
关35兴 T. Lee and C. Lin, Phys. Rev. E 71, 046706 共2005兲.
NIU et al.
关36兴 Y. H. Zhang, X. J. Gu, R. W. Barber, and D. R. Emerson, Phys.
Rev. E 74, 046704 共2006兲.
关37兴 H. Chen, R. Zhang, I. Staroselsky, and M. Jhon, Physica A
362, 125 共2006兲.
关38兴 Y. Zhou, R. Zhang, I. Staroselsky, H. Chen, W. T. Kim, and M.
Jhon, Physica A 362, 68 共2006兲.
关39兴 F. Toshi and S. Succi, Europhys. Lett. 69, 549 共2005兲.
关40兴 X. Shan, X. F. Yuam, and H. Chen, J. Fluid Mech. 550, 413
共2006兲.
关41兴 S. Ansumali and I. V. Karlin, Phys. Rev. E 66, 026311 共2002兲.
036711-8
PHYSICAL REVIEW E 76, 036711 共2007兲
关42兴 X. D. Niu, C. Shu, and Y. T. Chew, Int. J. Mod. Phys. C 16,
1927 共2005兲.
关43兴 R. Zhang, X. Shan, and H. Chen, Phys. Rev. E 74, 046703
共2006兲.
关44兴 M. J. McNenly, M. A. Gallis, and I. D. Boyd, Proceedings of
the 36th AIAA Thermophysics Conference 共AIAA, Orlando,
Florida, 2003兲.
关45兴 F. Sharipov and V. Seleznev, J. Phys. Chem. Ref. Data 27, 657
共1998兲.