Computers & Fluids 66 (2012) 1–9

Contents lists available at SciVerse ScienceDirect

Computers & Fluids

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m p fl u i d

Numerical simulation for unsteady compressible Micropolar fluid flow

James Chen a,⇑, Chunlei Liang b, James D. Lee b
a
Division of Business and Engineering, Pennsylvania State University, the Altoona college, Altoona, PA 16601, USA
b
Department of Mechanical and Aerospace Engineering, The George Washington University, Washington, DC 20052, USA

a r t i c l e i n f o a b s t r a c t

Article history: This paper reports the advancement of computational Micropolar fluid dynamics (CMFD) using the spec-
Received 8 November 2011 tral difference (SD) method. The fundamentals, linear constitutive equations and generalized Stokes’
Received in revised form 14 May 2012 hypothesis for Micropolar fluid dynamics are briefly introduced. The additional degrees of freedom in
Accepted 23 May 2012
MFD, gyration, can help to understand the coherent structures of vortices from the two-level energy
Available online 31 May 2012
transfer in the balance law of energy. The spectral difference (SD) method is proposed for solving
unsteady compressible viscous Micropolar flow problems. The analytical and exact solution of compress-
Keywords:
ible Micropolar Couette flow is solved and used to demonstrate the order of numerical accuracy of the SD
Micropolar fluid dynamics
Computational Micropolar fluid dynamics
method. As a numerical example, a 2D spectral difference solver is developed to simulate flow past a cyl-
Spectral difference method inder. The instantaneous gyration contours are plotted for the observation of vortex shedding behind the
Unsteady compressible flow cylinder. A new physical phenomenon, the coupling effect, is discussed.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction flow physics taking the interaction of macromotion and micromo-

tion of fluid particles into account. Such interaction between
The compressible Navier–Stokes equations have been widely micromotion and macromotion gives a possibility to identify the
used for computational aerodynamics. Classical continuum theory, coherent structure of vortices.
from which the Navier–Stokes equations are, does not include a In order to simulate unsteady fluid problem, a proper numerical
mechanism of characterizing the self-spinning motion, e.g. vorti- scheme should be developed. Recent numerical analysis mainly fo-
ces. A common practice is to use vorticity (a mathematical defini- cuses on the numerical solution of Hagen–Poiseuille flow and
tion which is not consistent with vortex physically) for nano/microfluidic system [6–8]. These studies only solve steady
characterizing vortex dynamics. It should be noticed that the vor- problems and are not suitable for unsteady problem. In 2011, Chen
ticity is computed as the curl of velocity field. In other words, it is et al. [9,10] incorporated Fu and Hodges’ Time-Centered Split
not an independent variable. Also, considerable confusion sur- method (TCSM) [11], Chorin’s Projection Method [12] and finite
rounds the longstanding question of what constitutes a vortex difference (FD) method for unsteady incompressible Micropolar
and a consistent and universal vorticity-based method to identify fluid problems. This solver is limited with second order accuracy
and analyze vortex structure is still out of reach [1]. These are in space and time, and for structured grids. The current develop-
the fundamental reasons for the incapability of vorticity on the ments of numerical schemes with high order accuracy mainly fo-
vortex characterization. cus on Navier–Stokes equations. A numerical scheme for higher
Microcontinuum Theory, consisting of Micropolar, Microstretch order of numerical accuracy, e.g. Spectral Difference (SD) Method,
and Micromorphic (3 M) theories, developed by Eringen [2–5], pro- Discontinuous Galerkin (GK) Method [13] and Spectral Volume
vide a mathematical foundation to capture such self-spinning mo- (SV) Method [14], for incompressible and/or compressible Micro-
tion. In 3 M theories, each particle has a finite size and contains a polar fluids is still out of reach.
microstructure, which can rotate and deform independently In this article, a numerical method with high order accuracy,
regardless of the motion of the centroid of the particle. In Micropo- Spectral Difference Method, is developed for compressible Micropo-
lar Theory, there are three additional degrees of freedom, gyration, lar fluid. By considering the micromotion, gyration, and its interac-
to determine the rotation of the microstructure. Micropolar fluid tion with macromotion, Micropolar theory provides an opportunity
dynamics (MFD) is the application of Micropolar continuum theory of revealing the multiscale energy transfer in fluid flows, especially
on fluent medium. MFD equations can be used to characterize the vortex-dominated flows. The exact and analytical solutions of plane
Couette flow of compressible Micropolar fluid is provided and used
to evaluate the order of the accuracy of the numerical solutions from
⇑ Corresponding author. the Spectral Difference Method. A benchmark problem, flow past a
E-mail address: [email protected] (J. Chen). cylinder, is demonstrated. In the benchmark example, the coupling

0045-7930/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compfluid.2012.05.015
2 J. Chen et al. / Computers & Fluids 66 (2012) 1–9

effect between micromotion and macromotion is studied and dis- may be deduced by calculating the material time derivative of
cussed. This coupling effect opens the door to study the multiscale the spatial deformation tensors. For Micropolar fluid, two objective
transport phenomenon in fluid mechanics. deformation-rate tensors are [2–5,9,10]
This article is organized as follows: Section 2 briefly introduces
akl ¼ v l;k þ elkm xm
Micropolar fluid theory. It also discusses about the linear constitu- ð6Þ
tive equations and the generalized Stokes Hypothesis for Micropo-
bkl ¼ xk;l
lar fluid theory. Section 3 describes the Spectral difference (SD)
v l is the velocity of the centroid of the particle; elkm is the permuta-
method as numerical approach and solution methods. In order to
tion symbol; v l;k is velocity gradient; xm is gyration vector, the
validate the spatial accuracy of the code, Section 4 presents the
additional rotating degrees of freedom of the particle. The mean
analytical and exact solutions of Micropolar compressible Couette
free path of a fluid is larger than that of a solid, i.e., a fluid molecule
flow. Section 4 also shows and discusses the simulation results ob-
has more space to move around before colliding into another fluid
tained for laminar viscous flow past a cylinder. A literature is com-
molecule. Hence when a fluid particle in Micropolar continuum,
pared with the current numerical results. Finally, Section 5
which can be either a fluid molecule or a group of fluid molecules,
summarizes this current study.
rotates, the effect of gyration appears and such phenomena cannot
be observed in classical continuum theory. On the contrary, the
2. Theory of Micropolar fluid dynamics gyration vector in Micropolar fluid dynamics (MFD) is a natural
choice to reveal such effect.
2.1. Kinematics
2.2. Balance laws
A Micropolar continuum is a collection of continuously distrib-
uted finite size particles that can rotate. A material point, P, in the
The thermodynamic balance laws of Micropolar continuum can
reference configuration is identified by a position vector, X K , K = 1,
be expressed as [2–5,9,10]
2, 3, and three directors vkK ðX; tÞ attached to the point P. These
three directors in Micropolar continuum are rigid.
Conservation of Mass
The motion, at time t, carries the finite-size particle to a spatial
point and rotates the three directors to a new orientation. Thus, @q
þ ðqv i Þ;i ¼ 0 ð7Þ
such motion is similar to the motion of the Earth. It cannot only re- @t
volve around the Sun, which is macromotion, but also can spin on Conservation of Linear Momentum
its own axis, which is micromotion. These motions and their in-
verse motion for Micropolar continuum can be (cf. Fig. 1) ex- t lk;l þ qðfk  v_ k Þ ¼ 0 ð8Þ
pressed by [2–5,9,10] Conservation of Angular Momentum
xk ¼ xk ðX; tÞ; XK ¼ X K ðx; tÞ
ð1Þ mlk;l þ ekij t ij þ qðlk  jx
_ kÞ ¼ 0 ð9Þ
nk ¼ vkK ðX; tÞNK ; NK ¼ v  Kk nk
Conservation of Energy
and
vkK vlK ¼ dkl ; v Kk v Lk ¼ dKL ð2Þ qe_  tkl akl  mkl blk þ qk;k  qh ¼ 0 ð10Þ

It is straightforward to prove that Clausius–Duhem Inequality

vkK ¼ v Kk ð3Þ
_ þ t kl akl þ mkl blk  qk h;k P 0
qðw_ þ ghÞ ð11Þ
Consequently, the later part of Eq. (2) becomes h
vkK vkL ¼ dKL ð4Þ where
Here and throughout, an index followed by a common denotes a q ¼ mass density; v k ¼ center velocity;
partial derivative, e.g. lk ¼ body moment density; t kl ¼ Cauchy Stress tensor;
@xk @X K f k ¼ body force density; mkl ¼ moment stress tensor;
xk;K ¼ ; X K;k ¼ ð5Þ
@X K @xk e ¼ internal energy density; g ¼ entropy density;
qk ¼ heat flux factor; h ¼ energy source density;
For fluid flow, deformation-rate tensors are crucial to the char- h ¼ absolute temperature; xk ¼ gyration vector;
acterization of the viscous resistance. Deformation-rate tensors
j ¼ microinertia; eklm ¼ permutation symbol;
X3 w ¼ Helmholtz free energy density
B :t = 0
(
P X ,Ξ )
defined as

Ξ
(
b : t = t p x, ξ ,t ) w  e  hg ð12Þ

ξ It is noticed that Clausius–Duhem inequality, Eq. (11), is in fact

X the thermodynamic second law. All the constitutive equations
should satisfy this inequality. This second law is not a restriction
x on the kind of processes that occur in nature, but a restriction on
o the kind of material properties that physical systems occurring in
X2
nature can be [10,15]. Richard Feynman also mentioned in his
book, ‘‘So we see that a substance’s properties must be limited in
X1 a certain way; one cannot make up anything he wants; . . .This
principle, this limitation, is the only real rule that comes out of
Fig. 1. Deformation of microelement. thermodynamics [16].’’
 J. Chen et al. / Computers & Fluids 66 (2012) 1–9 3

One can understand microinertia through the concept of mo- rotational motion in MFD involves the relative motion between
ment of inertia. Microinertia is a measure of the resistance of the fluid particles and the self-spinning motion. Therefore, the tDkl akl
finite-size particle to changes to its rotation and can be defined as term in Eq. (19) indicates that the work done by Cauchy stress cov-
R ers two different scales of vortex structure: one is the relative mo-
q0 ðx; nÞnk nl dv 0
ikl  RDv  hnk nl i tion between two fluid particles and another one is the self-
Dv
q0 ðx; n; tÞdv 0 spinning motion for a single fluid particle. Another objective defor-
jkl  imm dkl  ikl ð13Þ mation rate tensors, bkl (cf. Eq. (6)), expresses the gradient of gyra-
1 tion between two fluid particles. In other words, at the smallest
j  jmm scale MFD can determine, the mkl blk term tells how the energy flows
3
from one fluid particle to another. This two-level energy transfer
The integration, in Eq. (13), is to be carried out over the volume gives a feasible explanation for the coherent structure of vortices.
of the finite-size particle. If the finite-size particle is assumed to be
a solid sphere with a radius, d, and a constant density, q, the micr- 2.4. Conservation form of MFD equations
oinertia can be computed as
2 2 The thermodynamic balance laws can be recalled as Eqs. (7)–
j¼ d ð14Þ (11). Multiplying by velocity in Eq. (8) and gyration in Eq. (9), they
5
can be written as
It shows that the microinertia for a spherical particle is the mo-
ment of inertia of a solid sphere divided by its mass. To determine  
D 1
the radius of the spherical particle, one can adopt the experimental tkl;k v l þ qfl v l ¼ q v lv l ð20Þ
Dt 2
data of the Lagrangian velocities of tracer particles [8].  
D 1
mkl;k xl þ elmn t mn xl þ qll xl ¼ qj xl xl ð21Þ
2.3. Linear constitutive equations Dt 2
The summation of Eqs. (19)–(21) leads to
The linear constitutive equations for Cauchy stress, moment
stress, and heat flux are derived to be [2–5,9,10]
qE_ ¼ ðtkl v l Þ;k þ ðmkl xl Þ;k  qk;k þ qh þ qfl v l þ qll xl
t kl ¼ pdkl þ ktrðamn Þdkl þ ðl þ jÞakl þ lalk  pdkl þ D tkl 1
E ¼ e þ ðv l v l þ jxl xl Þ ð22Þ
2
a
mkl ¼ eklm h;m þ atrðbmn Þdkl þ bbkl þ cblk ð15Þ where E is defined as total energy density. Also the pressure can be
h
related to the total energy by
K
qk ¼ h;k þ aeklm xm;l p
h e ¼ cv h ¼ cv
qðcp  cv Þ
p is the pressure; l is the viscosity; j is the coupling coefficient; c is ð23Þ
p 1
the gyration viscosity; k is the second coefficient of viscosity; K is qE ¼ þ qðv l v l þ jxl xl Þ
the thermal conductivity; ða; bÞ is the newly introduced material cp =cv  1 2
constants, whose physical meanings are still unknown. For the present study, the ratio of two specific heats, cp =cv , is set
Substitute the constitutive equations into all the balance laws to be 1.4 for air.
and the governing field equations of MFD can be re-written as For simplicity, the conservative form of the governing equations
[2–5,9,19]. is written in 2D as

Conservation of Mass
@Q @F @G
þ þ ¼S ð24Þ
@t @x @y
q_ þ qv l;l ¼ 0 ð16Þ
where
Conservation of Linear Momentum 8 9
>
> q > >
>
> >
>
rp þ ðk þ lÞrr  v þ ðl þ jÞr2 v þ jr  x þ qf > >
< qv x >
> =
¼ qv_ ð17Þ Q¼ yqv ð25Þ
>
> >
Conservation of Angular Momentum
>
>
> qjx >
>
>
>
>
: >
;
qE
ða þ bÞrr  x þ cr2 x þ jðr  v  2xÞ þ ql ¼ qjx
_ ð18Þ
Conservation of Energy F ¼ F inv  F v is ;
8 9
>
> qv x >
>
ðqe_  D t T : a  m : b þ r  q  qh ¼ 0 ð19Þ >
> >
>
> >
< qv x þ p
2
> >
=
F inv ¼ qv x v y ;
Eq. (19) could be used to interpret how the energy transfers in >
> >
>
>
> qjxv x >
>
the coherent structure of vortices. One of the objective deforma- >
> >
>
: ;
tion rate tensors, akl (cf. Eq. (6)), contains two parts. The first part ðqE þ pÞv x
8 9
is the gradient of fluid particle velocity, expressing the relative mo- >
>  0 >
>
tion between two fluid particles. The second part is the gyration, >
> @v x @v y vx >
>
>
> k þ þ ð2l þ jÞ @@x >
>
which illustrates the self-spinning motion of the fluid particle. It >
< 
@x @y
   >
=
is noticed that the symmetric part of, akl , is the strain-rate tensor F v is ¼
>
l @@yv x þ @@xv y þ j @@xv y  x >
ð26Þ
>
> >
>
in classical fluid dynamics; however the antisymmetric part is >
> c @x >
>
>
> @x >
>
the angular velocity (half of the curl of the velocity) minus the : ;
F v is½2 v x þ F v is½3 v y þ F v is½4 x  qx
gyration. This new antisymmetric part points out that the
4 J. Chen et al. / Computers & Fluids 66 (2012) 1–9

G ¼ Ginv  Gv is ;
8 9
>
> qv y > >
>
> >
>
> >
< qv x v y >
> =
Ginv ¼ qv 2y þ p ;
>
> >
>
>
> qjxv y > >
>
> >
>
: ;
ðqE þ pÞv y
8 9
>
>
>  0   >
>
>
>
> l @v x
þ
@v y
þ j @v x
þ x >
>
>
> @y @x @y >
>
>
<   >
=
@v x @ v @ v
Gv is ¼ k @x þ @y þ ð2l þ jÞ @y
y y
ð27Þ
>
> >
>
>
> @x >
>
>
> c >
>
>
> @y >
>
: ;
Gv is½2 v x þ Gv is½3 v y þ Gv is½4 x  qy

8 9
>
> 0 >
>
>
> >
>
>
> qfx >
>
>
< >
=
S¼ qfy ð28Þ
>   >
>
> @v y v x  2x >
>
>
> j  @@y >
> Fig. 2. Distribution of flux and solution points for the fourth order SD scheme.
>
>
@x >
>
: ;
qðfx v x þ fy v y þ lx þ hÞ
3. Introduction to spectral difference method

2.5. Generalized Stokes’ hypothesis The Spectral Difference (SD) Method is a newly developed effi-
cient high-order approach based on the differential form of govern-
From Eqs. (6) and (15), we can write the linear constitutive ing equations. The SD Method combines elements from finite
equation as volume and finite difference techniques. Similar to the Discontinu-
ous Galerkin (DG) [13] and Spectral Volume (SV) methods [14], the
t ij ¼ ðp þ kamm Þdij þ ðl þ jÞaij þ laji SD scheme achieves high-order by locally approximating the solu-
ð29Þ
aij ¼ v j;i þ ejik xk tions as a high degree polynomial inside each cell. However, being
based on differential form of the equations, it is formulating is sim-
Similar to the Stokes’ hypothesis in Newtonian fluid, the hydro-
ilar to than that of the DG and SV methods as no test function or
static pressure is identified with the mean stress, i.e. p ¼  13 t ii . We
surface integral is involved. Conservation properties are still main-
can also further derive Eq. (29) as
tained by a judicious placement of the nodes at quadrature points
t ii ¼ 3ðp þ kHÞ þ ð2l þ jÞH of the chosen simplex.
ð30Þ To achieve an efficient implementation for Eq. (24), all elements
) 3p ¼ t ii þ ½3k þ ð2l þ jÞH
in the physical domain (x, y) are transformed into a standard ele-
ments ð0 6 g 6 1; 0 6 n 6 1Þ as shown in Fig. 2. The transformation
where H ¼ r  v .
can be written as
Therefore, we have the generalized Stokes’ hypothesis
2 3
x1
3k þ ð2l þ jÞ ¼ 0 ð31Þ 6y 7
6 17
6 7
Hence the Cauchy stress is now re-written as
6 x2 7
   6 7
x N1 0 N2 0 N3 0 N4 0 6 7
6 y2 7
1 ¼ 6 7 ð33Þ
t ij ¼ pdij  ð2l þ jÞHdij þ ðl þ jÞaij þ laji ð32Þ y 0 N1 0 N2 0 N3 0 N4 6 7
6 x3 7
3 6y 7
6 37
The generalized Stokes’ hypothesis implies that there is no dif- 6 7
4 x4 5
ference between mechanical pressure and thermodynamic pres-
y4
sure. It should be mentioned that this assumption is supported
by kinetic theory when the fluid is a monatomic gas. This hypoth- where shape function N for 4-node elements is chosen and is a func-
esis is adopted throughout the simulations in this chapter. tion of ðn; gÞ; ðxi ; yi Þ are the Cartesian coordinates of those points.

Table 1
Velocity error analysis of compressible plane Couette flow.

21 42 84

Third order L1 Error 7:46  104 1:21  104 2:38  105
Order 2.6 2.34
L2 error 8:20  104 1:46  104 2:88  105
Order 2.49 2.34
Fourth order L1 Error 1:93  104 2:24  105 2:15  106
Order 3.1 3.38
L2 error 2:54  104 2:59  105 2:32  106
Order 3.3 3.48
 J. Chen et al. / Computers & Fluids 66 (2012) 1–9 5

Other types of shape function can be also chosen, e.g. 8-node qua- In summary, the algorithm to compute the inviscid flux deriva-
dratic elements and 12-element cubic elements. tives consists of the following steps [19]:
The metrics and the Jacobian, J, can be expressed as
 2 3 (1) Given the conservative variables at the solution point, they are
 x1 y1 
 @x  6 
 @x   N N2;n N3;n N4;n 6 x2 y2 7  computed at the flux points.
 @n @g   1;n 7
J ¼  @y  ¼  6 7 ð34Þ (2) The inviscid fluxes at the interior flux points are computed
 @y   N1;g N2;g N3;g N4;g 4 x3 y3 5
@n @g  directly using the solutions computed at Step (1).
 x4 y4  (3) The inviscid fluxes at the element interfaces are computed
where ðN i;j ; i ¼ 1  4; j ¼ n or gÞ represent the partial derivatives using the Rusanov/Roe [15–17] solver. Given the normal direc-
of the shape function with respect to the coordinates of the stan- tion of the interfaces, the averaged normal velocity compo-
dard element. nents and the sound speed, the inviscid fluxes on the
The governing equations in the physical domain are then trans- interfaces can be determined.
formed into the computational domain. They take the following (4) The derivatives of the fluxes are computed at the solution
form. points using the derivatives of the Lagrange operator l as
shown in Eq. (39).
e @ Fe @ G
@Q e
þ þ ¼e
S ð35Þ
@t @x @y On the other hand, the solution procedures to get viscous fluxes
can be specified in the following steps.
where
   
e ¼ J F @g þ G @g ; e
e ¼ JQ ; Fe ¼ J F @n þ G @n ; G
Q S ¼ JS:
(1) Reconstruct the conservative variables at the flux points from
@x @y @x @y those at the solution points using Eq. (38)
(2) Average the field of conservative
  variables on the element
In the standard element, two sets of points are defined, namely interfaces as Q f ¼ 12 Q Lf þ Q Rf . Meanwhile, appropriate
the solution points and the flux points, as illustrated in Fig. 2. The boundary conditions shall be applied for specific edge flux
solution and flux points are represented using circles and squares, points.
respectively. (3) Evaluate the derivatives of the conservative variables on the
In order to construct a degree (N  1) polynomial in each coor- solution points.
dinate direction, N solution points are required. The solution points (4) Reconstruct the derivatives of the conservative variables
in 1D are chosen to be the Gauss points defined by and apply appropriate boundary conditions for specific flux
  
1 2s  1 points. Average them on the element interfaces similar to
Xs ¼ 1  cos p ; s ¼ 1; 2; . . . ; N ð36Þ Step (2).
2 2N
(5) Use the conservative variables and the derivatives to compute
The flux points are selected to be the Legendre–Gauss quadra- the viscous flux vectors described in Eqs. (26) and (27) at the
ture points. element interfaces.
Using N solution points, a degree ðN  1Þ polynomial can be
built using the following Lagrange basis: It should be mentioned that a simple averaging operator is uti-
Y
N   lized to obtain common viscous fluxes at face interfaces. It is stable
X  Xs
hi ðXÞ ¼ ð37Þ for our simulations of all viscous flows. In practice, other operator
Xi  Xs
s¼1;s–i such as BR2 [20] and the LDG2 [21] can also be chosen. Neverthe-
Similarly, using ðN þ 1Þ flux points, a degree N polynomial can less, the current formulation is simpler than that of the DG and SV
be built for the fluxes using a similar Lagrange basis: methods as no test function or surface integral is involved.
! It is worthwhile to remind that, in the recent publications
Y
N X  X sþ1 [22,23], it is known that the Spectral Difference Method has several
liþ1 ¼ 2
ð38Þ
2 X iþ1  X sþ1 instability issues for the triangular grids. In this study, the quadri-
s¼0;s–i 2 2
lateral cells are utilized throughout and the above mentioned
The reconstructed solution for the conserved variable in the instability are overcome and avoided completely.
standard element is the tensor product of the two one-dimensional For all the computations in this study, the unsteady term is
polynomials, treated with a fourth-order strong-stability-preserving five-stage
Runge–Kutta scheme.
X
N X
N
e ðn; gÞ ¼
Q e ij hi ðnÞhj ðgÞ
Q ð39Þ
j¼1 i¼1
4. Computational example
Similarly, the reconstructed flux polynomials take the following
form:
4.1. Plane Couette flow
X
N X
N
Fe ðn; gÞ ¼ e
F ij liþ1 ðnÞhj ðgÞ In order to demonstrate the order of spatial accuracy of the SD
2
j¼1 i¼0
ð40Þ methods, the numerical results of compressible plane Couette flow
X
N X
N
are compared with the analytical and exact solution of such flow.
e gÞ ¼
Gðn; e ij l 1 ðgÞhi ðnÞ
G jþ
2 Consider a compressible Micropolar fluid with a top plate, at tem-
j¼0 i¼1
perature h1 , in height of h, moving at a velocity U 0 and a fixed bot-
The reconstructed fluxes are element-wise continuous, but dis- tom plate, at temperature h0 (cf. Fig. 3) and the following
continuous across cell interfaces. For the inviscid flux, an approxi- conditions are assumed: (1) zero velocity in y and z directions;
mate Riemann solver [17,18] is employed to compute a common (2) zero gyration in x and y directions; (3) the final solution is stea-
flux at interfaces to ensure conservation and stability. In this study, dy and the flow is fully developed, i.e. both x-direction velocity and
we have adopted the Rusanov flux [17,18] for inviscid fluxes at cell z-direction gyration are functions of y only; and (4) no body force is
interfaces. considered in this case.
6 J. Chen et al. / Computers & Fluids 66 (2012) 1–9

4.1.1. Analytical and exact solutions

For completeness, the analytical and exact solutions of velocity,
gyration and temperature are derived for the steady compressible
Micropolar plane Couette fluid flow. These solutions are
j  
vx ¼  C 2 eMy  C 3 eMy þ 2ðC 2 þ C 3 Þy þ C 4
Mðl þ jÞ
lþj
xz ¼ C 2 eMy þ C 3 eMy  C1
2l þ j
jB2 1  2 2My 
Fig. 3. Illustration of plane Micropolar Couette flow. h¼ C e þ C 23 e2My
aðl þ jÞ 4M2 2
2j B1 B2   jC 2 C 3 B2 2 cM 2 C 2 C 3 2
þ C 2 eMy þ C 3 eMy  y þ y
aðl þ jÞ M 2 aðl þ jÞ a
c  2 2My  B2 B2
 C2e þ C 23 e2My þ A1 y þ A2  1 y2
4a a
ð43aÞ
where

M2 ¼ jcð2ðllþþjjÞÞ ; C 1 ¼ 2llþþjj ðC 2 þ C 3 Þ;

C2 ¼  U0  ;
jð1eMh Þ Mh Mh
2 þ e Mhe h
MðlþjÞ e 1 ð43bÞ
1eMh j ðC
C3 ¼ C ;
eMh 1 2
C4 ¼ MðlþjÞ 2  C 3 Þ;
B1 ¼ C 2 þ C 3 ; B 2 ¼ 2l þ j
Fig. 4. Illustration of uniform flow past a cylinder.
!
A2 h 1 1 jB2  
A1 ¼  þ þ cþ C 22 e2Mh þ C 23 e2Mh
h h 4ah ðl þ jÞM 2

!
2jB1 B2   cM2 C 2 C 3 jC 2 C 3 B21 B2
 C 2 eMh þ C 3 eMh    h
haðl þ jÞM 2 a aðl þ jÞ a
  !
2 2
C2 þ C3 jB2 2jB21 B2
A2 ¼ h0 þ cþ 
4a ðl þ jÞM 2
haðl þ jÞM 2
ð43cÞ

4.1.2. Order demonstration of numerical accuracy

The numerical solutions of the compressible Micropolar Couette
flow problem are computed for the 3rd and 4th order SD methods
on the same grids. All the computations are initialized using con-
stant density and pressure. The temperature boundary conditions
are consistent. The L1 and L2 errors of the velocity are evaluated.
The three meshes consist of 2  1; 4  2 and 8  4, respectively.
Fig. 5. Time history of lift coefficient. The details of numerical order calculation and verification for
velocity are shown in Table 1. The temperature error is also com-
With the above conditions, the governing equations in Micropo- puted and the studies of numerical accuracy of the third order case
lar compressible fluid reduce to are shown in Table 2. The results obtained in the table clearly indi-
2
d Vx dxz cate that the SD methods exhibit desired optimal order of accuracy.
ðl þ jÞ 2
þj ¼0 ð42aÞ The average orders of accuracies for L1 and L2 errors of velocity in
dy dy
  the third order case are both 2.47 and 2.415, respectively. The aver-
2
d x dV x age orders of accuracy for the L1 and L2 errors of velocity in the
c 2z  j þ 2xz ¼ 0 ð42bÞ
dy dy fourth order case are 3.24 and 3.39, respectively. Moreover, the
  
2 average orders of accuracy for L1 and L2 errors of temperature
d dV x dV x d h
Vx l þj þ xz ¼a 2 ð42cÞ are also found to be 3.01 and 3.05, respectively, in the third order
dx dy dy dy case.

Table 2
Temperature error analysis of compressible plane Couette flow.

21 42 84

Third order L1 Error 7  104
8:53  10 5
1:06  105
Order 3.03 3.00
L2 Error 7:75  104 9:11  105 1:12  105
Order 3.08 3.02
 J. Chen et al. / Computers & Fluids 66 (2012) 1–9 7

The Cauchy stress is utilized to calculate the drag and lift coef-
ficients as
FD
Cd  1
2
qU 20 AD
FL
Cl  1
2
qU 20 AL ð46-aÞ
Z Z Z
FD ¼ p cos hdA þ sw sin h dA
dF x ¼
Z Z Z
F L ¼ dF y ¼  p sin hdA þ sw cos h dA

where sw is the Cauchy stress on the cylinder wall.

It should be noticed that in MFD, once the gyration (micromo-
tion) is concerned, the Cauchy stress is no longer symmetric. In a
2D case, the four components of Cauchy stress are, respectively,
Fig. 6. Time history of drag coefficient. @v x
t xx ¼ p þ kH þ ð2l þ jÞ
@x
@v y
t yy ¼ p þ kH þ ð2l þ jÞ
Table 3 @y
    ð46-bÞ
Strouhal number and vortex shedding period of flow past a cylinder with Re ¼ 150 @v x @v y @v y
and Ma ¼ 0:2. t xy ¼l þ þj x
@y @x @x
   
Case I Case II Case III @v x @v y @v x
t yx ¼l þ þj þx
Stouhal number 0.2 0.21 0.192 @y @x @y
Vortex shedding period 25.0325 24.112 26.047
where again H ¼ r  v .
In Case I, the drag coefficient (cf. Fig. 6) oscillates between
1.1949 and 1.2923. The average drag coefficient is 1.2436. In Case
4.2. Flow past a cylinder II, it oscillates between 1.1282 and 1.2513. The average one for
Case II is 1.19. In Case III, the oscillation exists between 1.2564
In this section, the SD methods are used to solve a uniform flow and 1.333 while the average is 1.2949. It is also found that as the
past a cylinder as shown Fig. 4, whose radius is one at a Reynolds coupling coefficient increases, the drag coefficient decreases. For
number of 150 and a freestream Mach number of 0.2. the reader’s reference, the experimentally determined value of
drag coefficient is 1.5 [24].
4.2.1. Coupling effect Figs. 7–9 show the instantaneous gyration contours in three
For incompressible Micropolar fluid, after the dimension analy- cases. It is noticed that the gyration here is an independent variable
sis, Eq. (17) can be re-written as and has nothing to do with vorticity, which is the curl of velocity
field.
b
@V 1 ^2b 1 ^ 1
b r
þV ^Vb ¼ r
^p^þ r Vþ r ^ þ 2 ^f
x ð44Þ
@^t Re Er Fr 4.2.2. Mesh refinement and higher order scheme demonstration
where the definition of dimensionless parameter are defined as Fig. 10 shows the gyration contour of Case II solved by the
b ¼ V ; ^t ¼ tD ; r
V ^ ¼ r; p ^ ¼ xDU0 .
^ ¼ p2 ; x fourth order SD method. Comparing Figs. 8 and 10, it is observed
U0 U0 D qU 0
The symbol with hat, ^, is dimensionless physical quantity or that the fourth order SD method retains the shape of the gyration
operator. Fr is the Froude number. Er is a new dimensionless num- better. In the 3rd order case, the gyration field is affected by the
ber, Eringen number, defined as Er ¼ qUj0 D. The physical meaning of numerical dissipation. By adopting a higher order scheme for the
Eringen number indicates how strong the coupling effect between same mesh, it improves the retainability of gyration and reveals
microscale and mascroscale is. Re is the definition of Reynolds a better resolution of the flow physics. The only disadvantage is
number in MFD, and it is slightly different from that in a Newto- that 4th order scheme takes longer time to converge.
nian fluid. In the previous cases, only 32 elements (regular mesh) are ar-
ranged around the periphery of the cylinder. In order to improve
qU 0 D the simulation result, a mesh with 40 elements (fine mesh) ar-
Re ¼ ð45Þ
ðl þ jÞ

where l is viscosity and j is coupling coefficient.

Three different cases are modeled: (I) l=j ¼ 1; (II) l=j ¼ 0:3;
 Fig. 5 shows the time history of lift coefficients
(III) l=j ¼ 3  3.
in three cases while Fig. 6 plots the time history of drag
coefficients.
The Strouhal numbers, St ¼ UfD0 (f is the frequency of gyration
shedding), are estimated from the lift coefficient profile. They are
0.2, 0.21 and 0.19 for Case I, II and III, respectively. For reference,
the experimentally determined value of Stouhal number is 0.183
[24]. Other quantities are listed in Table 3. It is observed that the
frequency of vortex shedding increases with increasing the cou-
pling coefficients. One can also find that from Fig. 5, the peak value
of the lift coefficients drops as the coupling coefficients decrease. Fig. 7. Instantaneous gyration contour of Case I. The freestream velocity is 0.2.
8 J. Chen et al. / Computers & Fluids 66 (2012) 1–9

Fig. 8. Instantaneous gyration contour of Case II. The freestream velocity is 0.2. Fig. 11. Instantaneous gyration contour of Case II solved by a finer mesh (40
elements around the cylinder). The freestream velocity is 0.2.

Fig. 9. Instantaneous gyration contour of Case III. The freestream velocity is 0.2.

Fig. 12. Comparisons of drag coefficients in the cases of different order schemes
and fine mesh.

fluid theory and concluded that when the coupling coefficient in-
creases, the shedding frequency decreases. This is the opposite to
our simulation. The reason is that Mafhouz did not define the Rey-
nolds number in a Micropolar fluid correctly, but borrowed the
definition for Newtonian fluids directly. Reynolds number is a
dimensionless parameter determining the ratio of viscous force
Fig. 10. Instantaneous gyration contour of Case II solved by 4th order SD method.
and inertia force. The classical definition in Newtonian fluid can
The freestream velocity is 0.2.
be derived by rendering Navier–Stokes equations to be dimension-
less. Through similar analysis, the Reynolds number in Micropolar
fluid can be defined as Eq. (45).
ranged along the periphery of the cylinder is used. Fig. 11 shows In Mafhouz’s case, when the coupling coefficient increases, the
the gyration contour. Comparing Figs. 8, 10 and 11, one can con- Reynolds number is altered from 50 to 16 23 according to Eq. (45)
clude that the 4th order SD scheme and the fine mesh obtain con- because Micropolar theory is utilized. That is the reason his study
verged and similar numerical results for the gyration field. concluded that the coupling coefficient decreases the shedding
Fig. 12 shows improvement of drag coefficient from 4th order frequency. Instead, after fixing the Reynolds number and changing
SD scheme and fine mesh (40 elements around the cylinder). In the ratio between viscosity and coupling coefficient, the numerical
the case using 3rd order SD scheme, the drag coefficient oscillates results show clearly an opposite trend: For a fixed Reynolds num-
between 1.13 and 1.25 while the average is 1.19. In the case using ber, the larger the coupling coefficient is, the faster the gyration
4th) order scheme, the drag coefficient oscillates between 1.29 and sheds.
1.43 with the average of 1.36. In the case with fine mesh (40 ele-
ments around the cylinder), the drag coefficient oscillates between
1.27 and 1.41 owing the average of 1.34. In Fig. 12, a more consis- 5. Concluding remarks
tent result on drag coefficient is shown by comparing regular mesh
with 3rd order scheme, regular mesh with 4th order scheme, fine Micropolar fluid dynamics provides an alternative way for mod-
mesh with 3rd order scheme. eling vortex dynamics. Micropolar continuum contains continuous
distribution of finite size particles. Each finite-size fluid particle is
4.2.3. Comparisons with a literature equipped with microsturcture. The existence of microstructure al-
In 2007, Mahfouz [25] analytically used stream functions to lows additional degrees of freedom, gyration. The extra degrees of
study the problem of flow past a circular cylinder using Micropolar freedoms, gyration, can be independently solved from the balance
 J. Chen et al. / Computers & Fluids 66 (2012) 1–9 9

laws of angular momentum. In addition, the physical meanings of [5] Eringen AC. Microcontinuum field theories II: Fluent media. Springer; 2001.
[6] Papautsky I, Brazzle J, Ameel T, Frazier AB. Laminar fluid behavior in
gyration, an independent variable, can be read as a self-spinning
microchannel using Micropolar fluid theory. Sensor Actuator 1999;73:101–8.
motion, e.g. the smallest vortex. From the balance law of energy, [7] Ye S, Zhu K, Wang W. Laminar flow of micropolar fluid in rectangular
one can understand how the energy flows inside the coherent microchannels. Acta Mech Sinica 2006;22:403–8.
structures of vortices. [8] Hansen JS, Davis PJ, Todd BD. Molecular spin in nano-confined fluidic flows.
Microfluid Nanofluid 2009;6:785–95.
To practically tackle flow problems with MFD, an efficient and [9] Chen J, Liang C, Lee JD. Theory and simulation of Micropolar fluid dynamics. J
accurate numerical solver is needed. A high-order unstructured Nanoeng Nanosyst 2011;224:31–40.
solver with the spectral difference method for Micropolar com- [10] Chen J, Lee JD, Liang C. Constitutive equations of Micropolar electromagnetic
fluids. J Non-Newtonian Fluid 2011;166:867–74.
pressible viscous flows is developed and verified by comparing [11] Fu Shipeng, Hodges BR. Time-center split method for implicit discretizatino of
the analytical solution of Micropolar plane Couette flow. The SD unsteady advection problems. J Eng Mech 2009;135:256–64.
method is then applied to simulate flow past a cylinder problem. [12] Chorin AJ. Numerical solution of the Navier–Stokes equations. Math Comput
1968;22:745–62.
The instantaneous contour plot of gyration is used to visualize [13] Cockburn B, Shu C-W. The Runge–Kutta discontinuous Galerkin method for
the vortex shedding behind the cylinder. A new physical phenom- conservation laws V: multidimensional systems. J Compt Phys
enon in MFD, the coupling effect, is discussed. As the coupling coef- 1998;141:199–224.
[14] Wang ZJ, Liu Y. Extension of the spectral volume method to high-order
ficient is increased, the lift force increases but the drag decreases. boundary representation. J Compt Phys 2006;211:154–78.
This numerical results are also compared with Mahfouz [25] and [15] Coleman BD, Noll W. The thermodynamics f elastic materials with heat
clarify the correct definition of Reynolds number in MFD. conduction and viscosity. Arch Rational Mech Anal 1963;13:167–78.
[16] Feynman R, Lighton R, Sands M. The Feynman lectures on physics. Addison
The additional degrees of freedom in Micropolar fluid dynamics
Wesley Longman; 1970.
give a possibility to explain the multi-length scale dynamics and [17] Roe PL. Approximate Riemann solver, parameter vectors and difference
motions in fluid flows. Especially, this is essential for rotational scheme. J Comput Phys 1981;43:357–72.
turbulent flows, such as the Taylor–Couette flow or turbulent con- [18] Rusanov VV. Calculation of interaction of non-steady shock wave with
obstacles. Comput Math Math Phys 1961;1:261–79.
vection in the sun. Conventionally, the energy-carrying large-scale [19] Liang C, Jameson A, Wang ZJ. Spectral difference method for two-dimensional
structures are directly resolved by the grids in the LES computa- compressible flow on unstructured greids with mixed elements. J Comput
tion, while the small scales are modeling using one of sub-grid Phys 2009;228:2847–58.
[20] Bassi F, Rebay S. A high-order accurate discontinuous finite element method
scale (SGS) models. These SGS models are largely based on eddy- for the numerical solution of the compressible Navier-Stokes equations. J
viscosity approaches and similar to each other. Moreover, they car- Comput Phys 1997;131:267–79.
ry a priori coefficients. The newly introduced degrees of freedom in [21] Kannan R, Wang ZJ. LDG2: a variant of the LDG flux formulation for the
spectral volume method. J Sci Comput 2011;26:314–28.
the MFD could provide promising possibilities to construct SGS [22] den Abeele KV, Lacor C, Wang ZJ. On the stability and accuracy of the spectral
models. difference method. J Comput Phys 2008;37:162–8.
[23] Liang C, Kanana R, Wang ZJ. A p-mulgrid spectral difference method with
explicit and implicit smoothers on unstructured triangular grids. Comput Fluid
References 2009;38:254–65. .pffiffiffiffiffiffi
[24] Williamson CHK, Brown GL. A Series in 1 Re to represent the Strouhal–
[1] Jeong J, Hussain F. On the identification of a vortex. J Fluid Mech Reynolds number relationship of the cylinder wake. J Fluid Struct
1995;285:69–94. 1998;12:1073–85.
[2] Eringen AC. Simple microfluids. Int J Eng Sci 1964;2:205–17. [25] Mahfouz FM. Laminar forced convection from a circular cylinder placed in a
[3] Eringen AC. Theory of micropolar fluids. J Appl Math Mech 1964;16:1–8. Micropolar fluid. J Heat Transf 2007;129:256–64.
[4] Eringen AC. Microcontinuum field theories I: Foundations and solid. Springer;
1999.