Documente Academic
Documente Profesional
Documente Cultură
A Lattice Boltzmann model for simulating micro flows has been proposed by us recently
(Europhysics Letters, 67(4), 600–606 (2004)). In this paper, we will present a further
theoretical and numerical validation of the model. In this regards, a theoretical analysis
of the diffuse-scattering boundary condition for a simple flow is carried out and the
result is consistent with the conventional slip velocity boundary condition. Numerical
validation is highlighted by simulating the two-dimensional isothermal pressure-driven
micro-channel flows and the thin-film gas bearing lubrication problems, and comparing
the simulation results with available experimental data and analytical predictions.
1. Introduction
Micro-electromechanical systems (MEMS) have become a subject of active research
in a growing discipline.1 – 3 As all the micro devices have to be operated in a fluid
media, the understanding of flow at the micro level is fundamental to the devel-
opment of MEMS. To effectively design and optimize the micro devices, one has
to understand the fundamental features of the micro flow such as film damping of
resonant structures, viscous forces, heat transfer in mass flow sensors and unsteady
pressure fields around micro-valves, etc.
In spite of its importance, the research on micro flows is still at a preliminary
stage although the mechanical properties of some micro devices are reasonably
well studied. The main reason behind this is that at micro-level, the continuum
assumption is no longer valid since the mean free path of gas molecules is the same
∗ Corresponding author.
1927
February 7, 2006 14:45 WSPC/141-IJMPC 00844
order as the typical geometric dimension of the device.4 As a result, the conventional
governing equation of motion (the Navier–Stokes equation) and numerical tools that
seek to solve this equation are not applicable. The usual ways to study the micro
flows are molecular dynamics (MD),5 the direct simulation Monte Carlo (DSMC)
approach6 and solutions of full Boltzmann equation (BE).7,8 However, the required
computational effort of the MD and the DSMC is usually very large, even with the
use of most powerful supercomputer. The schemes used for solving the full BE are
more complicated than those usually used for the Navier–Stokes equations.
In the past decades, as an alternative computational fluid dynamics (CFD)
tool, the lattice Boltzmann method (LBM) has received considerable attention by
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
bution functions of fictitious particles on the links of the lattice. The particles then
evolve in a discrete time according to certain rules that guarantee the satisfactions
of some desirable macroscopic behaviors, e.g., compressible thermal or isothermal
fluids, emerging at scales larger than the lattice spacing. Unlike the MD and DSMC
methods, the number of particles distributed in the computational field in the LBM
is not related to the number of molecules. Therefore, the LBM is intuitively more
computationally efficient than the MD and DSMC methods. Furthermore, since the
LBM solver is based on a simplified kinetic model — lattice Boltzmann equation
(LBE), one avoids using complicated schemes to solve the full BE.
In recent years, a few endeavors of applying the LBM to simulate micro
flows14 – 16 have been carried out. The critical issues of the LBM modeling of micro
flows are how to simulate the Knudsen effects and the slip boundary condition. The
Knudsen effects can be described by the Knudsen number Kn, where Kn is the
ratio of the molecular mean free path λ to the characteristic length H. In our recent
work,17 an entropy lattice Boltzmann model for simulating micro flows is proposed
based on the entropy lattice Boltzmann equation (ELBE).18 In this model, the
Knudsen effects are modeled by relating the relaxation time τ with Kn according
to the kinetic theory7,20 and the slip boundary condition is implemented by the
diffuse- scattering boundary condition (DSBC), which is derived from the classic
gas-surface interaction theory.7,19,20 Although very interesting results are obtained
for the shear driven flow between two parallel plates, the detailed implementation of
the model is not given in Ref. 17. One may also be interested to know whether the
DSBC can generate a consistent slip velocity condition at macroscopic level, which is
extensively used in the slip velocity model. This paper seeks to answer to these ques-
tions. To further validate the model developed in the work of,17 the present study
focuses on two objectives: first, presenting a theoretical analysis of the DSBC; sec-
ond, applying this model to simulate more challenging micro flows. In this regards,
a theoretical analysis of the DSBC for a simple flow is presented and numerical
simulations of the two-dimensional (2D) pressure-driven isothermal micro-channel
flows and the thin-film gas bearing lubrication problems are carried out. For con-
venience, the standard lattice Boltzmann equation (SLBE)11 is used since there
February 7, 2006 14:45 WSPC/141-IJMPC 00844
is no intrinsic difference between the SLBE and the ELBE, except stability. The
theoretical analysis and the numerical results obtained are compared with available
experimental data,21 – 24 numerical simulations24 and theoretical analysis.22 – 24
The rest of the paper is organized as follows. Section 2 briefly reviews the lat-
tice Boltzmann BGK model from Ref. 17 by replacing the ELBE with the SLBE.
In Sec. 2.1, kinetic definition of the relaxation times in terms of the kinetic the-
ory is presented. A diffuse-scattering boundary condition (DSBC) is introduced in
Sec. 2.2. Section 3 presents a theoretical analysis of the DSBC based on a sim-
ple flow. Numerical simulations of the 2D pressure-driven isothermal micro-channel
flows and the thin-film gas bearing lubrication problems are presented in Secs. 4
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
with
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.
AN
Rf (eα0 → eα ) = ((eα − uw ) • n)fαeq |u=uw , (7)
ρw
where α0 and α are directions of the incident and reflected particles, respectively,
and AN is a normalization coefficient and can be obtained by satisfying zero normal
flux conditions on the walls. This coefficient is also dependent on the velocity model
used in the LBM. The essence of the DSBC is that the particles in the LBM arriving
at the wall lose their information and are redistributed in a way consistent with the
mass-balance and normal-flux conditions.
To illustrate the above DSBC clearly, we take a straight solid wall with velocity
uw and choose the D2Q9 discrete velocity model9 – 13 as an example (see Fig. 1).
As shown in Fig. 1, the solid arrow lines (α = 1, 5, 6, 7, 8) represent the particles
streaming from the flow domain with the known distribution functions while the
dashed arrow lines (α = 2, 3, 4) are the particles reflected from the wall boundary
with the unknown distribution functions, which are to be determined by the DSBC
4 3 2
n Flow field
5 1
6 7 8 Wall boundary
Fig. 1. Sketch of the diffuse scattering boundary condition based on the D2Q9 model (n is the
inward normal vector of the wall boundary; solid arrow lines represent functions of the particles
streaming from the flow field while dashed arrow lines represent functions of the particles diffused
from the boundary).
February 7, 2006 14:45 WSPC/141-IJMPC 00844
AN eq
f2 = f (ρw , uw )(f6 + f7 + f8 ) ,
ρw 2
AN eq
f3 = f (ρw , uw )(f6 + f7 + f8 ) , (8)
ρw 3
AN eq
f4 = f (ρw , uw )(f6 + f7 + f8 ) ,
ρw 4
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
where AN = 6 obtained by
P
|(eα − uw ) • n|fα
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.
α
AN = ρ w eq (9)
n|fα |u=uw α |(e0α −
P
|(eα − uw ) • uw ) • n|fα0
u2w
τ + 0.5δt uw 9 uw 2 3
f˜2j=1 = 1+3 + − (f6j=1 + f7j=1 + f8j=1 )
6τ c 2 c 2 c2
!!
δt uj=1 9 uj=1 2 3 u2j=1
− ρ 1+3 + − , (10c)
72τ c 2 c 2 c2
u2w
2(τ + 0.5δt) 3
f˜3j=1 = 1− (f6j=1 + f7j=1 + f8j=1 )
3τ 2 c2
!!
δt 3 u2j=1
− ρ 1− , (10d)
18τ 2 c2
February 7, 2006 14:45 WSPC/141-IJMPC 00844
u2w
τ + 0.5δt uw 9 uw 2 3
f˜4j=1 = 1−3 + − (f6j=1 + f7j=1 + f8j=1 )
6τ c 2 c 2 c2
!!
δt uj=1 9 uj=1 2 3 u2j=1
− ρ 1−3 + − , (10e)
72τ c 2 c 2 c2
!!
1 uj=1 9 uj=1 2 3 u2j=1
f˜5j=1 = ρ 1−3 + − , (10f)
9 c 2 c 2 c2
!!
δt uj=2 9 uj=2 2 3 u2j=2
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
f˜6j=1 = ρ 1−3 + −
36(τ + 0.5δt) c 2 c 2 c2
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.
τ − 0.5δt ˜j=2
+ f , (10g)
τ + 0.5δt 6
!!
δt 3 u2j=2 τ − 0.5δt ˜j=1
f˜7j=1 = ρ 1− + f , (10h)
9(τ + 0.5δt) 2 c2 τ + 0.5δt 7
!!
δt uj=2 9 uj=2 2 3 u2j=2
f˜8j=1 = ρ 1+3 + −
36(τ + 0.5δt) c 2 c 2 c2
τ − 0.5δt j=2
+ f , (10i)
τ + 0.5δt 8
where j = 1 represents the wall boundary and j = 2 is its adjacent layer. The
weighting coefficients wα in Eq. (2) are chosen as 4/9 for α = 0, 1/9 for α = 1, 3,
5, 7 and 1/36 for α = 2, 4, 6, 8, respectively. The equilibrium distribution function
under these coefficients corresponds to the third-order Hermite formula.25
From Eqs. (10a)–(10i), one can easily prove the normal moment flux ρvj=1 = 0.
The tangential moment flux ρuj=1 can be written as
(
8τ − δt uj=1 τ + 0.5δt ρuw
ρuj=1 = c ρ +
12τ c 6τ c
)
τ ρuj=2 τ − 0.5δt ˜j=2 ˜j=2
+ − (f − f4 ) , (12)
3(τ + 0.5δt) c τ + 0.5δt 2
which further yields
(
2 uj=1 δt ρuw
ρuj=1 = c ρ +
3 c 3(τ + 0.5δt) c
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
)
τ ρuj=2 δt uj=1
+ − ρ . (13)
3(τ + 0.5δt) c 6(τ + 0.5δt) c
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.
Through simple algebraic operation for Eq. (13), the slip velocity on the wall can
then be obtained
τ Kn
Us = uj=1 − uw = (uj=2 − uj=1 ) = (uj=2 − uj=1 ) , (14)
δt δ
where δ is the normalized minimum mesh spacing, which is usually the normalized
interval between the boundary point and its neighboring point. By applying the
Taylor series expansion to approximate uj=2 , we can get
∂u Knδ ∂ 2 u
Us = Kn + +··· . (15)
∂n 2 ∂n2
Clearly, the DSBC generally yields a wall slip velocity, which is dependent on
the Knudsen number and the flow gradients near the wall. Interestingly, if the
neighboring layer is set at one mean free path away from the boundary (δ = Kn)
as used in the analysis of the Beskok,26 we get
Kn2 ∂ 2 u
∂u
Us = Kn + + ··· , (16)
∂n w 2 ∂n2 w
which is exactly the same as the high-order slip velocity model derived by Beskok. 26
This finding shows that the lattice Boltzmann model with the DSBC correctly
captures the Knudsen effects on the solid boundaries and provides a firm theoretical
foundation of the present model.
4. Numerical Simulations
To further validate the LBE model presented in previous sections numerically, two
typical 2D cases, the pressure-driven isothermal micro-channel flows and the thin-
film gas bearing lubrication problems, are simulated in this part. All simulation
results have been verified by grid-independence studies. The first case has been
partially simulated in the early work.17 Here we review it using more detailed
results. The second case chosen in this paper represents a more general flow because
of the complicated boundary geometry simulations.
February 7, 2006 14:45 WSPC/141-IJMPC 00844
the second-order extrapolation. The initial inter-flow field is set with zero velocity
and constant pressure gradient dp/dx = (pi − p0 )/L. Usually, if the uniform grid
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.
is used, it will need at least 100 points in the cross section of the channel for
the standard LBE to get grid-independence solutions. To save computational re-
sources, non-uniform grids and the Taylor series expansion and least square-based
LBM (TLLBM)27 are used in this work. The TLLBM is a method developed for
irregular grid systems and has been proven to be an efficient and accurate solver
for simulating the continuum flows.27 – 29
To test the accuracy of the present LBM model, the pressure-driven slip flow
with pressure ratio of θ = pi /p0 = 2.02 and the Knudsen number of 0.053 is first
studied on three different non-uniform grids of 1001 × 29, 1001 × 41 and 1001 × 61,
respectively. The grid points are uniformly distributed along the channel and clus-
tered near the channel centerline and walls so that a better resolution for the Knud-
sen layer and pressure distribution along the channel centerline can be achieved.
The minimal grid spacings in three grids are between 0.01H and 0.04H, which are
the order of the Knudsen numbers simulated. Table 1 gives a detailed comparison of
the peak of the normalized pressure deviation from the linear pressure distribution
P̄ = (p − p1 )/p0 , normalized slip velocities Ūslip = Uslip /Umax-inlet|theoretical at three
locations of the channel and the ratio of slip mass flow rate to the corresponding
∗
no-slip mass flow rate of the analytical prediction M /MN S . The linear pressure dis-
tribution along the channel is defined as pl = p0 [1+(θ −1)(1−X)], where X = x/L.
Table 1. Grid-dependence study of present LBM in simulating the pressure- driven mi-
cro-channel flows (Kn = 0.053, θ = 2.02).
P̄peak (X/L)
∗
Grid size (0.071143 Ūslip(X/L=0) Ūslip(X/L=0.5) Ūslip(X/L=1) M /MN S
(0.57)) (0.094982) (0.149617) (0.340347) (1.210596)
The values in the first row are the analytical results based on the formula of Arkilic
et al.22 – 24 As shown in Table 1, grid dependence LBM is obtained for all three grid
sizes. The present results based on the three grids are consistent with each other
and compare well with the theoretical analysis of Arkilic et al.23 The maximum
relative error of the present results to the theoretical analysis is within 6%. Hence
in the following of this part, all simulation results presented are based on the grid
of 1001 × 29.
Figures 2 and 3 show the normalized pressure deviation from the linear pressure
distribution along the micro-channel for the flows at Kn = 0.053 and θ = 2.02, and
Kn = 0.155 and θ = 2.05, respectively. The experimental data of Pong et al. 21
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
are also included in these figures for comparison. As shown in the figures, the
simulation results show qualitatively the same nonlinear trend of the analytical
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.
prediction and experimental data. The differences between the simulation results
and the experimental data are probably due to uncertainty of the experiments.
The discrepancies between our simulation results and analytical predictions can
be attributed to the fact that effect of the Knudsen number is only accounted for
in the analytical predictions by incorporating the first-order slip on the boundary
with the Navier-Stokes solver in the inter-flow field. Figures 2 and 3 also show that,
the rarefaction effect serves to decrease the curvature of the pressure caused by the
compressibility effect, and this finding is consistent with the previous studies.21 – 23
A comparison of the normalized slip velocity along the channel wall between
the simulation results and the analytical prediction for flows at Kn = 0.053 and
0.06
0.04
0.02
0
0 0.2 0.4 0.6 0.8 1
X/L
Fig. 2. Normalized pressure deviation from linear pressure drop for Kn = 0.053 and θ = 2.02.
February 7, 2006 14:45 WSPC/141-IJMPC 00844
0.04
0.02
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
0
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.
-0.02
0 0.2 0.4 0.6 0.8 1
X/L
Fig. 3. Normalized pressure deviation from linear pressure drop for Kn = 0.155 and θ = 2.05.
1
Normalized Slip Velocity
Analytical [22]
0.8 LBM
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
X/L
Fig. 4. Normalized slip velocity variations along the channel wall for Kn = 0.053 and θ = 2.02,
and Kn = 0.155 and θ = 2.05, respectively.
θ = 2.02, and Kn = 0.155 and θ = 2.05 are presented in Fig. 4. We can see that the
slip velocities are increased as flow goes forward due to the pressure drop and hence
the increment of the local Knudsen number along the channel. The comparisons
of the mass flow rate as a function of the pressure ratio at Kn = 0.053 and 0.165
February 7, 2006 14:45 WSPC/141-IJMPC 00844
are demonstrated in Figs. 5 and 6, respectively. Included in these figures are the
present results, experimental data, and Arklic’s analytical results.23 The No-Slip
Navier–Stokes analytical results23 are also shown in the figures for comparison. It is
found that our simulations are in favorable agreement with the experimental data
and analytical predictions.
1.2
Exe. data, Argon [21]
1 No-Slip
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
LBM
Mass Flow Rate
0.6
0.4
0.2
0
1 1.5 2 2.5 3 3.5 4 4.5 5
Pressure Ratio
Fig. 5. Mass flow variation with the pressure ratio for Kn = 0.053.
5
Exe. data, Helium [22]
4.5 No-Slip
4 Analytical [22]
Mass Flow Rate
LBM
3.5
3
2.5
2
1.5
1
0.5
0
1 1.5 2 2.5 3
Pressure Ratio
Fig. 6. Mass flow variation with the pressure ratio for Kn = 0.165.
February 7, 2006 14:45 WSPC/141-IJMPC 00844
simulations. At the inlet and outlet of the channel, a pressure P0 at the ambient
condition is given and the velocities are extrapolated from the inner flow field.
For the gas lubrication problems, the bearing number Λ(= 6µUw L/P0 H02 ) is
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.
an important parameter used to describe the fluid behavior in the slider bearings.
For convenient comparison of the present LBM simulation with the work available
in the literature, the slider bearings with three bearing numbers of 10, 61.6, and
123.2 are chosen for study and all the following simulations using the present LBM
have been tested to be grid independent and only the results based on 1001 × 41
non-uniform grid with minimal grid spacing of 0.02H0 are shown.
Figures 8 and 9 show the slider bearing pressure profiles of Kn = 1.24 along
the channel for Λ = 61.6 and 123.2, respectively. The results obtained by using
the Reynolds equation with the second-order model30 and the stress-density ratio
model31 are also included in these two figures for comparison. As shown in the
figures, the results obtained by the present LBM simulations are in good agreement
with those of the second-order model but significantly different from those of the
stress-density ratio model. RL
The load carrying capacity W (= 1/L 0 (P (x) − P0 )/P0 dx) of the upper plate
√
as a function of the inverse Knudsen number D(= π/(2Kn)) is shown in Fig. 10.
To show the accuracy of the present LBM, the results obtained by using the lin-
ear Boltzmann equation,32 the Reynolds equation with continuum assumption, the
second-order model30 and the stress-density ratio model31 are also plotted in this
0.6
2nd- order model
( P P0 )
0.5 stress-density ratio model
P0
0.4 LBM
0.3
0.2
0.1
0
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
0.8
2nd-order model
( P P0 ) 0.7 stress-density ratio model
P0 0.6 LBM
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
X/L
Knudsen Number, Kn
0
101 100 10-1
10
/ =10
Load Carrying Capacity, W
10-1
Boltzmann
-2
10
Continuum
stress-density ratio model
second-order model
-3
LBM
10
-1 0
10 10
Inverse Knudsen Number, D
Fig. 10. Comparison of non-dimensional load capacity versus inverse Knudsen number.
February 7, 2006 14:45 WSPC/141-IJMPC 00844
figure. It was found from this figure that the present LBM produces a very close
result to the linear Boltzmann equation for the Knudsen number ranging from 1 to
10 and a slightly lower prediction than those of other approaches for 0.1 ≤ Kn < 1.
5. Conclusions
The lattice Boltzmann model with the DSBC developed in the work of Ref. 16 has
been further validated in this paper numerically and theoretically. A theoretical
analysis of the DSBC based on the D2Q9 discrete velocity model for a simple
2D plane flow shows that the DSBC yields a wall slip velocity depending on the
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
Knudsen number and flow gradients near the wall. If the neighboring layer is set
at one mean free path away from the boundary, the wall slip velocity is exactly the
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.
References
1. C. M. Ho and Y. C. Tai, Annu. Rev. Fluid Mech. 30, 579 (1998).
2. K. Gabriel, J. Javris and W. Trimmer, Small machine, large opportunities, Technical
Report (NSF, 1998).
3. L. O’Connor, MEMS: Microelectromechanical systems, Mech. Eng. 114(2), 40 (1992).
4. S. A. Schaaf and P. L. Chambre, Flow of Rarefied Gases (Princeton University Press,
Princeton, New Jersey, 1961).
5. J. Koplik and J. R. Banavar, Ann. Rev. Fluid Mech. 27, 257 (1995).
6. G. A. Bird, Molecular Gas Dynamics and the Direct Simulation of Gas Flows (Oxford
Science Publications, 1994).
7. F. Sharipov and V. Seleznev, J. Phys. Chem. Ref. Data 27(3), 657 (1998).
8. T. Ohwada, Y. Sone and K. Aoki, Phys. Fluids A 1(12), 2042 (1989).
9. Y. H. Qian, D. d’Humiéres and P. Lallemand, Europhys. Lett. 17, 479 (1992).
10. S. Chen, H. Chen, D. Martinez and W. H. Mathaeus, Phys. Rev. Lett. 67, 3766 (1991).
11. X. He, S. Chen and G. D. Doolen, J. Comp. Phys. 146, 282 (1998).
12. S. Chen and G. D. Doolen, Annu. Rev. Fluid Mech. 30, 329 (1998).
13. X. He, Q. Zou, L.-S. Luo and M. Dembo, J. Stat. Phys. 87, 115 (1997).
14. C. Y. Lim, C. Shu, X. D. Niu and Y. T. Chew, Phys. Fluids 14, 2299 (2002).
15. X. Nie, G. D. Doolen and S. Chen, J. Stat. Phys. 107, 279 (2002).
16. R. Benzi, L. Biferale, M. Sbragaglia, S. Succi and F. Toschi, Meso-
scopic Modeling of Heterogeneous Boundary Conditions for Microchannel Flows
http://arxiv.org/PS cache/nlin/pdf/0502/0502060.pdf
17. X. D. Niu, C. Shu and Y. T. Chew, Europhys. Lett. 67, 600 (2004).
18. S. Ansumali and I. V. Karlin, J. Stat. Phys. 107, 291 (2002).
19. J. C. Maxwell, The Scientific Papers of James Clerk Maxwell, Vol. 2, ed. W. D. Niven
(Dover Publications, New York, 1952).
February 7, 2006 14:45 WSPC/141-IJMPC 00844
20. C. Cercignani, R. Illner and M. Pulvirenti, The Mathematical Theory of Dilute Gases
(Springer Verlag, Berlin, 1994).
21. K. C. Pong, C. M. Huo, J. Liu and Y. C. Tai, Application of Microfabrication to Fluid
Mechanics, ASME Winter Annual Meeting, Vol. 197 (1994), p. 51.
22. E. B. Arkilic, K. S. Breuer and M. A. Schmidt, J. Fluid Mech. 437, 29 (2001).
23. E. B. Arkilic, M. A. Schmidt and K. S. Breuer, Application of Microfabrication to
Fluid Mechanics, ASME Winter Annu. Meet. Chicago, IL, Vol. 197 (1994), p. 51.
24. E. B. Arkilic, M. A. Schmidt and K. S. Breuer, J. MEMS 6, 167 (1997).
25. X. He and L.-S. Luo, Phys. Rev. E 55, 6333 (1997).
26. A. Beskok, Simulations and Models for Gas Flows in Micrigeometries, Ph.D. Thesis,
UMI Dissertation Services (1996).
27. C. Shu, Y. T. Chew and X. D. Niu, Phys. Rev. E 64, 045701-1 (2001).
Int. J. Mod. Phys. C 2005.16:1927-1941. Downloaded from www.worldscientific.com
28. C. Shu, X. D. Niu and Y. T. Chew, Phys. Rev. E 65, 036708-1 (2002).
29. Y. T. Chew, C. Shu and X. D. Niu, Int. J. Mod. Phys. C 13, 719 (2002).
by UNIVERSITY OF WARSAW on 12/22/14. For personal use only.