You are on page 1of 10

PHYSICAL REVIEW E 84, 036704 (2011)

Axisymmetric lattice Boltzmann method revised


Jian Guo Zhou
School of Engineering, University of Liverpool, Liverpool L69 3GQ, United Kingdom
(Received 19 January 2011; revised manuscript received 19 July 2011; published 14 September 2011)
A reformulated lattice Boltzmann model is described for incompressible axisymmetric flows with or without
swirling. It is a further development and improvement on the author’s axisymmetric lattice Boltzmann method.
The main features of the revised scheme are (a) all the macroscopic variables such as velocities are determined
in the same formulas as those in the conventional lattice Boltzmann approach to the Navier-Stokes equations and
(b) the added sink or source and force terms are simple with no calculation for a derivative. Such features
distinguish the present method from the other existing simplified schemes, leading to a simple and efficient
model. The scheme is naturally suitable for generic incompressible axisymmetric rotational flows involving
more physical phenomena. The numerical solutions to Womersley and cylindrical cavity flows are presented to
demonstrate its accuracy and capability.

DOI: 10.1103/PhysRevE.84.036704 PACS number(s): 47.11.−j

I. INTRODUCTION tests [18]. The major drawbacks of these methods are that
the second source term contains several complicated terms
The development of the lattice Boltzmann method contin-
involving velocity gradients, which may introduce additional
ues rapidly, as its numerous potential capabilities continue
errors and cause numerical instability [19].
to be realized and demonstrated in different areas [1,2].
Recently, Chen et al. developed an axisymmetric lattice
It has become a very successful computational method in
Boltzmann method based on the vorticity-stream-function
interdisciplinary subjects, which is far beyond its original
equations, making the formulation automatically satisfy the
intention of simulating fluid flows described by the Navier-
continuity equation and easy to solve [20,21]. The downside
Stokes equations. For instance, Chang and Yang applied the
of this method is the difficulty in determining vorticity at
method to image denoising [3], Aminfar and Mohammad
used the method for the simulation of electrowetting [4], boundaries, thereby losing the feature of the easy treatment of
and Mendoza et al. proposed a lattice Boltzmann model for boundary conditions [19]. Guo et al. proposed an axisymmetric
relativistic fluids [5]. lattice Boltzmann model from the continuous Boltzmann equa-
The lattice Boltzmann method has also been extended tion in cylindrical coordinates [22], which is a complete lattice
to the simulation of incompressible axisymmetric flows, as Boltzmann model for axisymmetric flows with or without
they represent numerous important flow problems in practice swirling in the framework of the lattice Boltzmann approach.
[6–10]. The three-dimensional (3D) lattice Boltzmann method Furthermore, Li et al. presented an improved axisymmetric
has been applied to modeling 3D axisymmetric flows [11–13] lattice Boltzmann scheme including the rotational effect [19].
in which the cubic lattices and a treatment of the curved Both models of Guo et al. and Li et al. are suitable for general
boundary are used. This implies that one more dimensional axisymmetric rotational flows and the added source or force
lattices are required in the method used for the solution to the terms to the schemes contain no velocity gradient, which are
flow problems, and hence the efficiency is reduced. Mathe- simpler and easier to use compared to other existing methods.
matically, 3D axisymmetric flows are effectively 2D problems However, these two methods share the same weaknesses:
in a cylindrical coordinate system. To take full advantage of (a) the source or force terms contain more terms than those
this feature, Halliday et al. formulated a lattice Boltzmann in the original governing equations, (b) the expressions for
method for axisymmetric flows through the introduction of calculating the macroscopic variables, such as velocities, take
two source terms into the lattice Boltzmann equation [14]. complex forms instead of the conventional simple sum of the
This method has successfully been applied to a number of distribution functions due to the elimination of the implicitness
axisymmetric flow problems [6,7,15]. Later, Lee et al. [16] in the schemes, thereby complicating the algorithm, and
noticed that one term in the momentum equation that is (c) each gives its own expression for the viscosity that is
related to radial velocity is missing from the formulation of different from the standard definition in the lattice Boltzmann
Halliday et al., which causes large errors for axisymmetric dynamics. In addition, Zhou developed an axisymmetric lattice
flows with significant radial velocities in nonstraight pipes. Boltzmann method without swirling [23]. This method has
They corrected the mistake and obtained accurate solutions to many features close to the standard lattice Boltzmann approach
flows where radial velocities cannot be ignored. Furthermore, to the Navier-Stokes equations, e.g., a simple procedure, good
the method of Halliday et al. has been improved to simulate numerical stability, and standard calculations for macroscopic
multiphase flows by Premnath and Abraham [8] and two parameters, such as velocity, which have been confirmed in
phase flows with a large density ratio by Shiladitya and recent research by Huang and Lu [24], Li et al. [25], and Tang
Abraham [10]. Following a similar idea to that of Halliday et al. [26]. Its only disadvantage is that the introduced force
et al., Reis and Phillips [9,17] described a scheme for the term contains velocity gradients.
flows by modifying the source terms in a different manner. In order to overcome the above-mentioned weakness, the
This method was further successfully verified by numerical original axisymmetric lattice Boltzmann method by Zhou [23]

1539-3755/2011/84(3)/036704(10) 036704-1 ©2011 American Physical Society


JIAN GUO ZHOU PHYSICAL REVIEW E 84, 036704 (2011)

has been reformulated. The main purpose is to eliminate the Boltzmann scheme for axisymmetric flows [19]. In order to
requirement of the calculations for velocity gradients while simulate the axisymmetric flows in a simple way, we extend
all of the advantages of the original model are retained: the idea in a consistent manner to the original axisymmetric
(a) the source terms are just the same additional terms as those lattice Boltzmann model [23] to eliminate the calculations
in the governing equations except for the missing velocity of the velocity gradients. The reformulated lattice Boltzmann
gradients compared to the Navier-Stokes equations, which has equation with source or sink and force terms reads
the advantage of considering more physical phenomena in an  
easy way, and (b) the same standard calculations for density fα (x + eα t,t + t) − fα (x,t) = −τα fα − fαeq + wα θ t
and velocities as those in the conventional lattice Boltzmann t
+ eαi Fi , (5)
method for fluid flows are preserved, as distinguished from Ke2
the other existing simplified models. To enhance the revised eq
where fα is the distribution function of particles; fα is the
model with the similar features for generic axisymmetric
local equilibrium distribution function; t is the time step;
flows involving swirling, one more distribution function is
x is the space vector, i.e., x = (r,x); e = x/t; x is the
formulated for the solution of the azimuthal velocity. The
lattice size; wα is the weight given by Eq. (10); θ is the source
accuracy, capability, and applicability have been demonstrated
or sink term,
using two typical numerical tests, which are compared with
analytical solutions or available experimental and numerical ρur
θ =− ; (6)
results. r
Fi is the force term defined by
II. AXISYMMETRIC FLOW EQUATIONS ρui ur 2ρνui
Fi = − − δir ; (7)
The governing equations for the incompressible axisym- r r2
metric flows in a cylindrical coordinate system can be written eαi is the component of eα , which is the velocity vector of a
in tensor form as [27] particle in the α link; K is the constant and is determined from
∂uj ur 1  1 
=− , (1) K= 2 eαx eαx = 2 eαr eαr ; (8)
∂xj r e α e α
and τα is an effective relaxation time related to the single
∂ui ∂ui 1 ∂p ∂ 2 ui ν ∂ui νui
+ uj =− +ν 2 + − 2 δir , (2) relaxation time τ [29] as
∂t ∂xj ρ ∂xi ∂xj r ∂r r 
1
, r = 0,
where ρ is the density; p is the pressure; t is the time; ν τα = τ1 (2τ −1)eαr t

(9)
1+ , r = 0.
is the kinematic viscosity; i is the index standing for r or τ 2r
x; r and x are the coordinates in radial and axial directions, It can be seen from the following recovery section that the term
respectively; ui is the component of velocity in i direction; δij related to 1/r in the above equation recovers the second term,
is the Kronecker δ function defined by
ir /r, on the right-hand side of Eq. (30) that is zero according
 to l’Hôpital’s rule when r = 0, and hence it does not exist at
0, i = j,
δij = (3) r = 0 [8,18].
1, i = j, If the nine-velocity square lattice (D2Q9) shown in Fig. 1
and the repeated indexes are the Einstein summation conven- is used, wα is defined as
⎧4
tion, which means a summation over the space coordinates.
⎨ 9 , α = 0,

Such a convention is used throughout the paper without further
wα = 19 , α = 1, 3, 5, 7, (10)
indication. ⎪
⎩ 1
Application of the continuity equation (1) to the momentum 36
, α = 2, 4, 6, 8;
equation (2) results in [19] and eα is
  
∂ui ∂(ui uj ) 1 ∂p ∂ ∂ui ∂uj
+ =− +ν + (0,0), α = 0,
∂t ∂xj ρ ∂xi ∂xj ∂xj ∂xi eα =
(11)
 
(α−1)π (α−1)π
λα e cos 4 , sin 4 , α= 0;
ν ∂ui ∂ur ui ur 2νui
+ + − − 2 δir . with λα defined as
r ∂r ∂xi r r 
(4) 1, α = 1, 3, 5, 7,
λα = √ (12)
2, α = 2, 4, 6, 8.
III. LATTICE BOLTZMANN METHOD From Eq. (8), we have K = 6.
A. Lattice Boltzmann equation The fluid density ρ and velocity ui are determined from the
distribution function in the same manner as that in the lattice
In the lattice Boltzmann method, it is a common practice to Boltzmann method for the Navier-Stokes equations,
calculate velocity gradients using the nonequilibrium moments 
for increased efficiency [28]. Li et al. successfully applied 1
ρ= fα , ui = eαi fα . (13)
this to remove the velocity gradients in an improved lattice α
ρ α

036704-2
AXISYMMETRIC LATTICE BOLTZMANN METHOD REVISED PHYSICAL REVIEW E 84, 036704 (2011)

According to the Chapman-Enskog expansion, fα can be


written in a series of ε,

4 3 2 fα = fα(0) + εfα(1) + ε2 fα(2) + O(ε3 ). (19)


The centered scheme [30] is used for both source term θ and
force term Fi as
5 1  
θ = θ x + 12 eα ε,t + 12 ε (20)
and
6 7 8  
Fi = Fi x + 12 eα ε,t + 12 ε , (21)
which can also be written, via a Taylor expansion, as
 
1 1
FIG. 1. Nine-velocity square lattice (D2Q9). θ x + eα ε,t + ε
2 2
eq
 
The local equilibrium distribution function fα is 1 ∂ ∂
= θ (x,t) + ε + eαj θ (x,t) + O(ε2 ) (22)
  2 ∂t ∂xj
eαi ui 9 eαi eαj ui uj 3 ui ui
fαeq = wα ρ 1 + 3 2 + − , (14) and
e 2 e4 2 e2  
1 1
which can be shown to have the following properties: Fi x + eα ε,t + ε
2 2
 1  
ρ= fαeq , ui = eαi fαeq . (15) 1 ∂ ∂
ρ = Fi (x,t) + ε + eαj Fi (x,t) + O(ε2 ). (23)
α α 2 ∂t ∂xj
Compared to the original axisymmetric lattice Boltzmann After substitution of Eqs. (19), (22), and (23) into Eq. (18), the
model [23], there are two differences: (a) inclusion of wα equation to the order of ε0 is
and (b) introduction of τα associated with the term of [(2τ −
1)eαr t/2r]. The former retains the advantage of the same fα(0) = fαeq , (24)
additional term in the revised model as the original one, and to the order of ε is
the latter removes the calculation of the velocity gradients. The  
coefficient of (2τ − 1)/2 in the latter will lead to the correct ∂ ∂ f (1) 1
+ eαj fα(0) = − α + wα θ + 2 eαi Fi , (25)
viscosity as the standard definition in the conventional lattice ∂t ∂xj τ 6e
Boltzmann method. All these become clear in the following
and to the order of ε2 is
recovery of the hydrodynamic equations.    
∂ ∂ 1 ∂ ∂ 2 (0)
+ eαj fα(1) + + eαj fα
B. Recovery of the axisymmetric flow equations ∂t ∂xj 2 ∂t ∂xj
 
The Chapman-Enskog analysis is applied to show that the fα(2) (2τ − 1) 1 ∂ ∂
=− − eαr fα +
(1)
+ eαj (wα θ )
macroscopic equations (1) and (4) can be derived from the τ 2τ r 2 ∂t ∂xj
lattice Boltzmann equation (5). We assume t is small and  
1 ∂ ∂
equal to ε, + 2
+ eαj (eαi Fi ). (26)
12e ∂t ∂xj
t = ε. (16) By using Eq. (25), we can write the above equation as
Noting Eq. (9), we substitute the above into Eq. (5) and obtain  
(2τ − 1) ∂ ∂ f (2) (2τ − 1)
+ eαj fα(1) = − α − eαr fα(1) .
fα (x + eα ε,t + ε) − fα (x,t) 2τ ∂t ∂xj τ 2τ r
1  (2τ − 1)   (27)
= − fα − fαeq − eαr ε fα − fαeq
τ 2τ r From Eq. (25) + ε × Eq. (27), we have
ε    
+ wα θ ε + 2 eαi Fi . (17) ∂ ∂ (2τ − 1)ε ∂ ∂
6e + eαj fα +
(0)
+ eαj fα(1)
∂t ∂xj 2τ ∂t ∂xj
By taking a Taylor expansion to Eq. (17) in time and space at
point (x,t), we have 1  (2τ − 1)ε
= − fα(1) + εfα(2) − eαr fα(1)
    τ 2τ r
∂ ∂ 1 2 ∂ ∂ 2 1
ε + eαj fα + ε + eαj fα + O(ε3 ) + wα θ + 2 eαi Fi . (28)
∂t ∂xj 2 ∂t ∂xj 6e
1 (2τ − 1) The summation of the above equation over α provides
= − (fα − fαeq ) − eαr ε(fα − fαeq )
τ 2τ r ∂  (0) ∂ 
ε fα + eαj fα(0) = θ. (29)
+ εwα θ + 2 eαi Fi . (18) ∂t α ∂xj α
6e

036704-3
JIAN GUO ZHOU PHYSICAL REVIEW E 84, 036704 (2011)


The use of Eq. (24) and substitution of Eq. (15) into the above the term ∂/∂xk α eαi eαj eαk fα(0) is of the order of ρe2 Uc /Lc ,
equation result in the continuity equation (1), if the density based on which we obtain that the ratio of the former to the
variation is
small enough and can be neglected. latter terms has the order of
Taking eαi [Eq. (25) + ε × Eq. (27)] about α yields  
∂/∂t(0)ij
O 
∂  ∂(0)
ij ∂
ij
ir ∂/∂xk α eαi eαj eαk fα(0)
eαi fα +
(0)
= + + Fi , (30)    2  2
∂t α ∂xj ∂xj r ρUc2 /tc Uc Uc
=O 2
=O =O = O(M 2 ),
ρe Uc /Lc e Cs
where (0)
ij is the zeroth-order momentum flux tensor given by (40)
the following expressions:
 in which M = Uc /Cs is the Mach number. It follows that the
(0)
ij = eαi eαj fα(0) , (31) first term in Eq. (38) is very small compared with the second
α term and can be neglected if M  1, which is consistent with
the lattice Boltzmann dynamics; hence Eq. (38), after Eq. (39)
 is substituted, becomes
ε

ij = − (2τ − 1) eαi eαj fα(1) , (32) e2 ε ∂
2τ α (1)
ij = (2τ − 1) (ρui δj k + ρuj δki + ρuk δij ), (41)
6 ∂xk
and or
 
ε  ∂(ρui ) ∂(ρuj ) ∂(ρuk )

ir = − (2τ − 1) eαi eαr fα(1) . (33) (1)
ij =ν + + δij , (42)
2τ ∂xj ∂xi ∂xk
α
where ν is the kinematic viscosity and defined by
By evaluating the terms in Eq. (31) with Eq. (14), we have
e2 t
ν= (2τ − 1). (43)
(0)
ij = pδij + ρui uj , (34) 6
The insertion of Eq. (42) into Eq. (36) and evaluation of the
where p√= ρe2 /3 is the pressure, leading to a sound speed rest of the terms of the equation lead to
Cs = e/ 3. Substitution of the above equation into Eq. (30)  
∂(ρui ) ∂(ρuj ) ∂(ρuk )
produces
ij = ν + + δij − νθ δij . (44)
∂xj ∂xi ∂xk
∂(ρui ) ∂(ρui uj ) ∂p ∂
ij
ir
+ =− + + + Fi . (35) After applying the continuity equation (1) and Eq. (6) to the
∂t ∂xj ∂xi ∂xj r above, we obtain
 
By applying Eq. (25), we can rewrite Eq. (32) as ∂(ρui ) ∂(ρuj )

ij = ν + . (45)
 ∂xj ∂xi
ε

ij = (1)
ij − (2τ − 1) eαi eαj wα θ, (36) Similarly, we have
2 α  
∂(ρui ) ∂(ρur )

ir = ν + . (46)
with the first-order momentum flux tensor (1)
ij defined by ∂r ∂xi
   Combining Eqs. (7), (45), and (46) with Eq. (35) results in
ε ∂ ∂
(1)
ij = (2τ − 1) eαi eαj + eαk fα(0) , (37) ∂(ρui ) ∂(ρui uj )
2 α
∂t ∂xk +
∂t ∂xj
which can also be written using Eq. (31) as  
∂p ∂ ∂(ρui ) ∂(ρuj )
=− +ν +
ε ∂ ε ∂  ∂xi

∂xj ∂xj

∂xi
(1)
ij = (2τ − 1) (0)
ij + (2τ − 1) eαi eαj eαk fα(0) . ν ∂(ρui ) ∂(ρur ) ρui ur 2ρνui
2 ∂t 2 ∂xk α + + − − δir . (47)
r ∂r ∂xi r r2
(38)
Again, if the density variation is assumed to be small enough,
The second term in the above equation can be evaluated with the above is just the momentum equation (4).
Eqs. (14) and (24) as In the above recovery process, the centered scheme is
applied to evaluate the source term θ and force term Fi . They
∂  e2 ∂ may be calculated in one of the following three ways with ψ
eαi eαj eαk fα(0) = (ρui δj k + ρuj δki + ρuk δij ).
∂xk α 3 ∂xk for either θ or Fi :
(i) Implicit form,
(39)
  1
ψ x + 12 eα ε,t + 12 ε = [ψ(x,t) + ψ(x + eα t,t + t)];
If we assume characteristic velocity Uc , length Lc , and time tc , 2
we have that the term ∂/∂t(0) 2
ij is of the order of ρUc /tc , and (48)

036704-4
AXISYMMETRIC LATTICE BOLTZMANN METHOD REVISED PHYSICAL REVIEW E 84, 036704 (2011)

(ii) Semi-implicit form, ēα is the velocity vector of a particle on the D2Q4,
  1  
(α − 1)π (α − 1)π
ψ x + 12 eα ε,t + 12 ε = [ψ(x,t) + ψ(x + eα t,t)]; ēα = e cos , sin , α = 1, 2, 3, 4;
2 2 2
(49) (56)
(iii) Explicit form,
and τ̄α is also an effective relaxation time associated with the
  single relaxation time τ̄ ,
ψ x + 12 eα ε,t + 12 ε = ψ(x,t). (50)

As the implicit scheme needs iteration in simulations, which
1
, r = 0,
reduces efficiency, the semi-implicit and explicit schemes τ̄α = τ̄1 (2τ̄ −1)ēαr t

(57)
τ̄
1+ 2r
, r = 0.
become the obvious choices. Practical calculations have shown
eq
that both schemes can provide accurate solutions and hence the There are many expressions for f¯α and a simple one is
explicit scheme is preferred. In addition, the explicit scheme used here [19],
turns out to be second-order accurate due to the fact that it  
happens to be its average at previous and next time steps with 2ēαj uj ρuφ
f¯αeq = 1 + , α = 1, 2, 3, 4, (58)
the second-order truncation error, e2 4

ψ(x,t) = 12 [ψ(x − eα t,t − t) + ψ(x + eα t,t + t)]. in which ēαj is the component of ēα . It is easy to show that the
above equation has the following properties:
(51) 
This has been confirmed by Huang and Lu [24], Li et al. [25], f¯αeq = ρuφ , (59)
α
and Tang et al. [26] in their recent research.
It may be worthwhile to mention that the main features 
of the revised model are (a) it retains all the advantages of ēαi f¯αeq = ρui uφ , (60)
the original Zhou’s model [23], (b) the added source and α
force terms involve no velocity gradients and are simpler and
compared to those in the existing schemes [19,22], and 
(c) it preserves effectively the same simple procedure as that in ēαi ēαj f¯αeq = ρe2 uφ δij /2. (61)
the standard lattice Boltzmann method for the Navier-Stokes α
equations, which is distinguished from and simpler than the
The azimuthal velocity uφ is calculated as
recent available simplified schemes [19,22].
1 ¯
uφ = fα . (62)
C. Axisymmetric rotational flows ρ α
Axisymmetric rotational flows contain an azimuthal veloc- In order to prove that Eq. (52) can be recovered from
ity uφ , which is governed by the equation in a cylindrical the lattice Boltzmann equation (54), we apply the similar
coordinate system [27], Chapman-Enskog analysis to that given in Sec. III B and, after
∂uφ ∂(uj uφ ) ∂ 2 uφ ν ∂uφ 2ur uφ νuφ taking a Taylor expansion to Eq. (54) in time and space at point
+ =ν + − − 2 . (52) (x,t), we have
∂t ∂xj ∂xj2 r ∂r r r
   
Its further effect on the flow field is taken into account by ∂ ∂ 1 2 ∂ ∂ 2 ¯
ε + ēαj fα + ε
¯ + ēαj fα + O(ε3 )
adding an additional term to the force term Fi in Eq. (7) as ∂t ∂xj 2 ∂t ∂xj
ρu2φ 1  (2τ̄ − 1)   Sφ
ρui ur 2ρνui = − f¯α − f¯αeq − ēαr ε f¯α − f¯αeq + ε.
Fi = − − δ ir + δir . (53) τ̄ 2τ̄ r 4
r r2 r
(63)
It is realized that Eq. (52) is an advection-diffusion equation
and can be solved accurately and efficiently on either a D2Q4 The centered scheme [30] is again used for the term Sφ ,
or D2Q5 lattice Boltzmann model [20,31,32]. Consequently,  
the following lattice Boltzmann equation with a source or sink Sφ = Sφ x + 12 ēα ε,t + 12 ε , (64)
term using the D2Q4 is applied: which is further written, via a Taylor expansion, as
 
f¯α (x + ēα t,t + t) − f¯α (x,t) 1 1
Sφ t Sφ x + ēα ε,t + ε
= −τ̄α (f¯α − f¯αeq ) + , (54) 2 2
4  
1 ∂ ∂
eq = Sφ (x,t) + ε + ēαi Sφ (x,t) + O(ε2 ). (65)
where f¯α is the distribution function; f¯α is the local 2 ∂t ∂xi
equilibrium distribution function; Sφ is the source or sink term
defined by After substitution of Eqs. (19) and (65) into (63), Eq. (63) to
the order of ε0 is
2ρur uφ ρνuφ
Sφ = − − 2 ; (55) f¯α(0) = f¯αeq , (66)
r r

036704-5
JIAN GUO ZHOU PHYSICAL REVIEW E 84, 036704 (2011)

to the order of ε is By applying Eq. (61) into above, we have


 
∂ ∂ f¯(1) Sφ ∂(ρuφ )
+ ēαj f¯α(0) = − α + , (67) i = ν̄ , (77)
∂t ∂xj τ̄ 4 ∂xi
and to the order of ε2 is in which ν̄ is the kinematic viscosity defined by
   
∂ ∂ 1 ∂ ∂ 2 ¯(0) e2 t
+ ēαj fα +
¯(1)
+ ēαj fα ν̄ = (2τ̄ − 1). (78)
∂t ∂xj 2 ∂t ∂xj 4
 
f¯(2) (2τ̄ − 1) 1 ∂ ∂ Sφ Similarly, we have
=− α − ēαr f¯α(1) + + ēαi .
τ̄ 2τ̄ r 2 ∂t ∂xi 4 ∂(ρuφ )
r = ν̄ . (79)
(68) ∂r
The substitution of Eq. (67) into the above equation gives The substitution of Eqs. (59), (60), (77), and (79) into Eq. (70)
  results in the governing equation (52) if the density variation
(2τ̄ − 1) ∂ ∂ f¯(2) (2τ̄ − 1) is assumed to be small enough.
+ ēαi f¯α(1) = − α − ēαr f¯α(1) .
2τ̄ ∂t ∂xi τ̄ 2τ̄ r Consistency in the hydrodynamics requires the same
(69) kinematic viscosity, ν̄ = ν, which leads to

Taking [Eq. (67) + ε × Eq. (69)] yields τ̄ = 1
2
+ 13 (2τ − 1). (80)
∂  ¯(0) ∂  ∂i r As seen from the above formulas, there is no calculation for
fα + ēαi f¯α(0) = + + Sφ , (70)
∂t α ∂xi α ∂xi r a velocity gradient in the procedure and the added source or
sink term is the same as that in the governing equation without
where the velocity gradient.
(2τ̄ − 1)ε 
i = − ēαi f¯α(1) , (71)
2τ̄ α
IV. APPLICATIONS

and The method is applied to solve two flow problems: an


unsteady Womersley flow and a steady cylindrical cavity
(2τ̄ − 1)ε 
r = − ēαr f¯α(1) . (72) flow. The results are compared with either available analytical
2τ̄ α solutions or experimental data or other numerical results. The
terms involving singularities in Eqs. (6), (7), (53), and (55)
Inserting Eq. (67) into Eq. (71) leads to
related to 1/r at r = 0 are evaluated using l’Hôpital’s rule: they
  
ε ∂ ∂ are set to zeros at r = 0 [8,18]. All the dimensional physical
i = (2τ̄ − 1) ēαi + ēαj f¯α(0) (73) variables in the International System (SI) of units are used in
2 α
∂t ∂x j
the numerical computations.
or
 
ε ∂  ∂  A. 3D Womersley flow
i = (2τ̄ − 1) ēαi f¯α(0) + ēαi ēαj f¯α(0) .
2 ∂t α ∂xj α The 3D Womersley flow, or a pulsatile flow, is an unsteady
axisymmetric flow in a straight pipe. It is driven by a periodic
(74)

The order analysis indicates that ∂/∂t  ¯(0)
α ēαi fα has the order 0.1
of ρUc2 /tc from Eq. (60), and ∂/∂xj α ēαi ēαj f¯α(0) has the Numerical results
Analytical solutions
0.08
order of ρe2 Uc /Lc from Eq. (61), based on which the ratio of
n=3
the former to the latter has the order of 0.06
   n=2
∂/∂t α ēαi f¯α(0) 0.04
O  n=1
∂/∂xj α ēαi ēαj f¯α(0) 0.02
ux/Uc

   2  2 n=0
ρUc2 /tc Uc Uc 0
=O =O =O = O(M 2 ).
ρe2 Uc /Lc e Cs -0.02 n = 15

(75) -0.04 n = 14
n = 13
This suggests that the first term in Eq. (74) is much smaller -0.06
compared to the second and can be dropped if M  1, which n = 12
-0.08
again conforms to the lattice Boltzmann method; hence Eq. -1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1
(74) can be approximated by r/R

ε ∂ 
i = (2τ̄ − 1) ēαi ēαj f¯α(0) . (76) FIG. 2. Comparisons when ux is increasing at different times
2 ∂xj α t = nT /16, with n = 0,1,2,3,12,13,14,15.

036704-6
AXISYMMETRIC LATTICE BOLTZMANN METHOD REVISED PHYSICAL REVIEW E 84, 036704 (2011)

0.1 1
Numerical results
Analytical solutions 0.9
0.08 50x37
n=4 0.8
100x75
0.06
n=5 0.7 150x112
0.04 n=6 0.6 200x150

x/H
0.5 400x300
0.02 n=7
ux/Uc

0.4
0
n=8 0.3
-0.02 0.2
n=9
0.1
-0.04
n = 10 0
-0.06 -0.02 0 0.02 0.04 0.06 0.08 0.1 0.12
n = 11
ux/u0
-0.08
-1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1
r/R
FIG. 5. Lattice number effect on Case 2 with A = 1.5 and Re =
1290 (u0 = R).
FIG. 3. Comparisons when ux is decreasing at different times
t = nT /16, with n = 4,5,6,7,8,9,10,11. an equivalent periodic body force to the flow [33], i.e., an
additional body force is added to the existing force term Fi ,
pressure gradient at the inlet of the pipe and the pressure ρui ur 2ρνui
gradient is normally given as Fi = − − δir + p0 cos(ωt)δix . (84)
r r2
dp In the computation, ρ = 3, p0 = 0.001, D = 40, T =
= p0 cos(ωt), (81) 1200, α = 8, and UC = 1, which gives Re = 1200. Also,
dx τ = 0.6, and 80 × 40 lattices with x = 1 are used in the sim-
where p0 is the maximum amplitude of the pressure variation ulation. The periodic boundary conditions are applied to inflow
and ω = 2π/T is the angular frequency, in which T is the and outflow boundaries; the standard bounce-back scheme is
period. applied to no-slip boundary conditions along the pipe walls.
The Reynolds number is defined as Re = Uc D/ν, with the The numerical solutions at different times are obtained after
characteristic velocity Uc given by an initial running time of 10T . The corresponding results for
velocity ux are shown in Figs. 2 and 3, in which they are
p0 α 2 p0 R 2 further compared with the analytical solution (83), showing
Uc = = , (82) good agreements.
4ωρ 4ρν

in which α = R ω/ν is the Womersley number, where R is B. Cylindrical cavity flow
the pipe radius and D is the diameter. The analytical solution Steady cylindrical cavity flows have been investigated
for the velocity component in the axial direction is both experimentally and numerically [19,22,34,35]. The flow
    problem is sketched in Fig. 4 and is known to have different
p0 J0 (rφ/R) iωt complex structures depending on the combinations of the
ux (r,t) = Re 1− e , (83)
iωρ J0 (φ) aspect ratio A = H /R and the Reynolds number Re = R 2 /ν,
where  is the the constant angular velocity, R is the radius,
where J0 is the zeroth-order Bessel function
√ of the first type, and H is the height of the cylinder. Three cases with parameters
i is the imaginary unit, φ = (−α + iα)/ 2, and Re denotes
the real part of a complex number. The implementation of 0.05
the periodic pressure gradient can be achieved by applying

r
0.01
Er

Ω Numerical
2R 0 x 2.16
Er = 21.745 Kn

0.001
0.01 0.03 0.06
Kn = Δx/R
H
FIG. 6. Errors against lattice size for Case 2 with A = 1.5 and
FIG. 4. Definition sketch for cylindrical cavity flow. Re = 1290.

036704-7
JIAN GUO ZHOU PHYSICAL REVIEW E 84, 036704 (2011)

1.5 1.5

1 1
x/R

x/R
0.5 0.5

0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
r/R r/R

FIG. 7. Stream lines for Case 1 with A = 1.5 and Re = 990. FIG. 8. Stream lines for Case 2 with A = 1.5 and Re = 1290.

of (A, Re) = (1.5,990),(1.5,1290), and (2.5,1010) have been


used to validate the 3D lattice Boltzmann method by Bhaumik that at the equilibrium state, having little effect on the solution
and Lakshmisha [35] and the axisymmetric lattice Boltzmann (hydrodynamics). It should be noted that other schemes like
method by Guo et al. [22]. Li et al. also used the first two the nonequilibrium extrapolation approach by Guo et al. [36]
cases for the validation of their improved axisymmetric lattice may be applied for accurate solutions to flows with significant
Boltzmann scheme [19]. Consequently, these three cases are nonequilibrium effect.
applied to verify the proposed model. In order to establish solution independence from lattice size
In the simulations, R = 1, ρ = 1, and τ = 0.55. The steady and convergence order, different lattices in r and x directions as
flow state is reached after certain time steps when the following Nr × Nx , i.e., 50 × 37, 100 × 75, 150 × 112, 200 × 150, and
convergence criterion is satisfied [22]: 400 × 300, were used for Case 2 with (A, Re) = (1.5,1290),
as this is a complex flow involving vortex breakdown. The
V(t) − V(t − 1000t) results are shown in Fig. 5, indicating that solutions are
< 10−6 , (85)
V(t) independent of the choice of lattice number if more than
 150 × 112 lattices are used. If the maximum velocity for ux is
in which V(t) = [u2x (x,r) + u2r (x,r)]. The boundary chosen to calculate errors between the numerical solutions and
conditions are the experimental measurement (see Table I), then the errors
ux = ur = uφ = 0, x = 0, calculated using Er = (ux,max − ux,max ) are plotted against
ux = ur = uφ = 0, r = ±R, (86) the relative lattice size or Knudsen number kn = x/R in
ux = ur = 0, uφ = r, x = H, Fig. 6; here ux,max stands for the numerical result and ux,max
stands for the experimental data. It must be pointed out that
which are implemented in the proposed model by setting all of none of the results based on lattices larger than 150 × 112 are
the distribution functions to their respective local equilibrium shown in the figure because they are almost the same and are
eq eq
distribution functions, fα = fα and f¯α = f¯α , at the bound- no longer appropriate for the representation of convergence
aries. Although a real individual distribution function may be order. As seen from the figure, a trend line is best fitted
different from the local equilibrium distribution function, this through the points, suggesting that the model is second-order
can provide accurate solutions to the problem as the following accurate.
results show. The reasons are because (a) the macroscopic Based on the above investigation, the lattice size of x =
properties are the aggregate effect of all the microscopic 0.01 that corresponds to 200 × 150 for Case 2 is believed to
interactions, and the former hydrodynamics depends very little produce accurate solutions and is used in all the cases, and is
on the details of the latter dynamics, consistent with the theory also used in the previous studies for the same problems [19,22].
of lattice gas automata, and (b) the macroscopic property (mass After the steady solutions are reached, the stream lines are
and momentum) at the nonequilibrium state retains the same as depicted in Figs. 7–9, showing correct flow patterns, which are

TABLE I. Comparisons of maximum axial velocities.

Re = 990 Re = 1290 Re = 1010


Reference ux,max /u0 hmax /H ux,max /u0 hmax /H ux,max /u0 hmax /H

Present 0.0992 0.207 0.0706 0.147 0.105 0.448


Expt. [34] 0.097 0.21 0.068 0.14 0.103 0.46
3D LBM [35] 0.093 0.22 0.072 0.16 0.102 0.52
N-S [35] 0.099 0.19 0.0665 0.125 0.106 0.44
Li’s LBM [19] 0.0987 0.213 0.0716 0.147

036704-8
AXISYMMETRIC LATTICE BOLTZMANN METHOD REVISED PHYSICAL REVIEW E 84, 036704 (2011)

2.5 2.5

2 2

1.5 1.5
x/R

x/R
1 1

0.5 0.5

0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
r/R r/R

FIG. 9. Stream lines for Case 3 with A = 2.5 and Re = 1010. FIG. 10. Stream lines for Case 4 with A = 2.5 and Re = 2200.

consistent with those reported by the experimental and numer- The stream lines for the steady solution are plotted in Fig. 10
ical studies [19,22,34,35]. The maximum axial velocities and and clearly show that two breakdown vortices are well
their locations are shown in Table I and further compared with developed. The flow pattern is again in agreement with that
the previous experimental data and other numerical results observed in the experimental measurements [34].
from (i) the improved model by Li et al. [19], (ii) the 3D
lattice Boltzman method (LBM) and the Navier-Stokes (N-S)
V. CONCLUSIONS
solutions by Bhaumik and Lakshmisha [35], where the relative
errors calculated using (ux,max − ux,max )/ux,max for Re = 990 An axisymmetric lattice Boltzmann method revised
is 2.3%, which is among the improved model with 1.7%, the (AxLAB) R is presented for generic axisymmetric flows
3D LBM solution with 4.2% and the Navier-Stokes result with with or without swirling. Unlike existing simplified schemes,
2.1%, for Re = 1290 is 3.8%, which is among the improved it preserves effectively the same simple procedure as the
model with 5.3%, the 3D LBM solution with 5.9% and the conventional lattice Boltzmann approach to the Navier-Stokes
Navier-Stokes result with 2.2%, and for Re = 1010 is 1.9%, equations. There is no need for the determination of velocity
which is between the 3D LBM solution with 0.9% and the gradients in the model. The natural incorporation of the source
Navier-Stokes result with 2.9%. This indicates that the present or force terms into the lattice Boltzmann equation makes
model can produce accurate solutions, agreeing well with the the method easy and simple to simulate axisymmetric flows
previous investigations. involving more physical phenomena. It is also found that using
To demonstrate the potential of the described method in the local equilibrium distribution functions as the boundary
predicting complex vortex breakdowns, the fourth case for conditions is simple and can provide accurate solutions to com-
(A, Re) = (2.5,2200) is further simulated. This case has also plex cylindrical cavity flows. The numerical tests have shown
been investigated in the experiment [34], revealing that there that the proposed method is simple and accurate, which is
are two vortex breakdowns. The other calculating parameters suitable for both steady and unsteady axisymmetric rotational
remain the same as those in the three aforementioned cases. flows.

[1] S. Succi, The Lattice Boltzmann Equation for Fluid Dynamics [6] Y. Peng, C. Shu, Y. Chew, and J. Qiu, J. Comput. Phys. 186, 295
and Beyond (Oxford, London, 2001). (2003).
[2] M. C. Sukop and D. T. J. Thorne, Lattice Boltzmann Modeling: [7] X. D. Niu, C. Shu, and Y. T. Chew, Int. J. Mod. Phys. C 6, 785
An Introduction for Geoscientists and Engineers (Springer- (2003).
Verlag, Berlin, 2006). [8] K. N. Premnath and J. Abraham, Phys. Rev. E 71, 056706 (2005).
[3] Q. Chang and T. Yang, IEEE Trans. Image Proc. 18, 2797 (2009). [9] T. Reis and T. N. Phillips, Phys. Rev. E 75, 056703 (2007).
[4] H. Aminfar and M. Mohammad, Comput. Methods Appl. Mech. [10] S. Mukherjee and J. Abraham, Phys. Rev. E 75, 026701
Eng. 198, 3852 (2009). (2007).
[5] M. Mendoza, B. M. Boghosian, H. J. Herrmann, and S. Succi, [11] R. S. Maier, R. S. Bernard, and D. W. Grunau, Phys. Fluids 8,
Phys. Rev. Lett. 105, 014502 (2010). 1788 (1996).

036704-9
JIAN GUO ZHOU PHYSICAL REVIEW E 84, 036704 (2011)

[12] R. Mei, W. Shyy, D. Yu, and L. S. Luo, J. Comput. Phys. 161, [25] X. F. Li, G. H. Tang, T. Y. Gao, and W. Q. Tao, Int. J. Mod. Phys.
680 (2000). C 21, 1237 (2010).
[13] A. M. Artoli, A. G. Hoekstra, and P. M. A. Sloot, Int. J. Mod. [26] G. H. Tang, X. F. Li, and W. Q. Tao, J. Appl. Phys. 108, 114903
Phys. C 13, 1119 (2002). (2010).
[14] I. Halliday, L. A. Hammond, C. M. Care, K. Good, and [27] W. P. Graebel, Engineering Fluid Mechanics (Taylor & Francis,
A. Stevens, Phys. Rev. E 64, 011208 (2001). London, 2001).
[15] T. S. Lee, H. Huang, and C. Shu, Int. J. Numer. Methods Fluids [28] S. Hou, J. Sterling, S. Chen, and G. D. Doolen, Fields Inst.
49, 99 (2005). Commun. 6, 151 (1996).
[16] T. S. Lee, H. Huang, and C. Shu, Int. J. Mod. Phys. C 17, 645 [29] P. L. Bhatnagar, E. P. Gross, and M. Krook, Phys. Rev. 94, 511
(2006). (1954).
[17] T. Reis and T. N. Phillips, Phys. Rev. E 76, 059902(E) (2007). [30] J. G. Zhou, Lattice Boltzmann Methods for Shallow Water Flows
[18] T. Reis and T. N. Phillips, Phys. Rev. E 77, 026703 (2008). (Springer-Verlag, Berlin, 2004).
[19] Q. Li, Y. L. He, G. H. Tang, and W. Q. Tao, Phys. Rev. E 81, [31] S. P. Dawson, S. Chen, and G. D. Doolen, J. Chem. Phys. 98,
056707 (2010). 1514 (1993).
[20] S. Chen, J. Tölke, S. Geller, and M. Krafczyk, Phys. Rev. E 78, [32] J. G. Zhou, Int. J. Numer. Methods Fluids 61, 848 (2009).
046703 (2008). [33] J. A. Cosgrove, J. M. Buick, and S. J. Tonge, J. Phys. A 36, 2609
[21] S. Chen, J. Tölke, and M. Krafczyk, Phys. Rev. E 79, 016704 (2003).
(2009). [34] K. Fujimura, H. Yoshizawa, R. Iwatsu, and H. Koyama, J. Fluids
[22] Z. Guo, H. Han, B. Shi, and C. Zheng, Phys. Rev. E 79, 046708 Eng. 123, 604 (2001).
(2009). [35] S. Bhaumik and K. Lakshmisha, Comput. Fluids 36, 1163
[23] J. G. Zhou, Phys. Rev. E 78, 036701 (2008). (2007).
[24] H. B. Huang and X. Y. Lu, Phys. Rev. E 80, 016701 (2009). [36] Z. Guo, C. Zheng, and B. Shi, Phys. Fluids 14, 2007 (2002).

036704-10

You might also like