Computers & Fluids 35 (2006) 1154–1168

www.elsevier.com/locate/compfluid

Flow over a bluff body from moderate to high Reynolds

numbers using large eddy simulation
A. Sohankar
Department of Mechanical Engineering, Faculty of Engineering, Yazd University, Yazd, Iran

Received 6 September 2004; received in revised form 2 March 2005; accepted 17 May 2005
Available online 3 November 2005

Abstract

Large eddy simulation (LES) is performed to study the uniform approach flow over a square-section cylinder with different Rey-
nolds numbers, ranging from 103 to 5 · 106. Two different sub-grid scale models, the Smagorinsky and a dynamic one-equation
model, are employed. An incompressible finite-volume code, based on a non-staggered grid arrangement and an implicit fractional
step method with second-order accuracy in space and time, is used.
The structure of the flow is studied with the instantaneous and the mean quantities such as pressure, turbulent stresses, turbulent
kinetic energy, vorticity, the second invariant of velocity gradient and streamlines. The Strouhal number, the mean and RMS values
of the lift and drag are computed for various Reynolds numbers, which show a good agreement with the available experimental
results. It is found that the effect of Reynolds number on the global quantities, the mean and the large scale instantaneous flow-struc-
tures is not much at the higher Reynolds numbers, i.e. Re > 2 · 104. In this range of Reynolds numbers, the small scales of the
instantaneous structures are more complex and chaotic as they compare with the larger ones.
 2005 Elsevier Ltd. All rights reserved.

1. Introduction shear layers, dominant large scale structures and high

The flow over objects such as high-rise buildings,
marine risers, bridges, cooling towers, chimneys and
tube banks in heat exchangers are encountered exten-
sively in engineering applications. These objects, which
under normal circumstances usually create a massive
wake region downstream, are called bluff bodies.
Motion of objects such as submarines, ships and cars
through fluids are examples of a pivotal role of such
bluff objects in industrial applications. Thus, the knowl-
edge of the flow field around these structures is of major
importance in a multitude of applications such as civil,
wind engineering and technical problems associated
with energy conversion and structural design. This type
of flow often contains many complex phenomena such
as separation, wake flow, vortex shedding, curved free
E-mail address: asohankar@yazduni.ac.ir
turbulence.
In these bodies, the behaviour of flow changes as the
Reynolds number (Re) is increased. At this point, it may
be suitable to attend to special type of bluff bodies,
circular and square prisms. At the Reynolds numbers
below about unity, the flow is fully attached with no sep-
aration. When Re increases, the flow separates and a
pair of steady symmetric vortices forms behind the
body. At higher Reynolds numbers, the formation vor-
tex length of the re-circulation region behind the body
grows with increasing Re. At a critical onset Reynolds
number [1,2], the twin-vortex arrangement becomes
unstable, and a time-periodic oscillation wake and a
staggered vortex street form. The separated vortices
shed alternately from the upper and the lower side of
the body. The periodic phenomenon is referred to as
vortex shedding, whereas the anti-symmetric wake flow
pattern is referred to as the von Kármán vortex street.
This is the first 2D wake transition which is called

0045-7930/$ - see front matter  2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compfluid.2005.05.007
 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168 1155

Bénard–von Kármán instability [1,2]. This transition short review on LES mathematical formulation and
contains a 2D instability from a steady wake to a peri- the sub-grid scale models used in this work presents in
odic wake. By increasing the Reynolds number, a 3D Section 3. A description of the numerical solution proce-
transition develops at around Reynolds number of dure is given in Section 4. The analysis of the global,
about 200, and the three-dimensional flow effects appear time-averaged and instantaneous results is presented in
[3,4]. As Re still increases higher, the effects of flow three three subsections in Section 5. Finally, some conclusions
dimensionality and turbulence become more and more are drawn in Section 6.
pronounced.
The fundamental fluid dynamics problems of bluff
bodies such as circular and rectangular cylinders have 2. Problem description
been examined extensively in both numerical and exper-
imental, especially at low and moderate Reynolds num- The geometry of the domain of interest is shown in
bers [1–4]. In recent years, researchers attention has Fig. 1. As seen, a square cylinder with a side d is exposed
turned to the use of LES for studying the turbulent flow to a constant free stream velocity U. The flow is
around bluff bodies at higher Reynolds numbers. A LES described in a Cartesian coordinate system (x, y, z) in
Workshop was held in June 1995 in Germany, and its which the x-axis is aligned with the inlet flow direction,
results were published in Ref. [5], while they were com- the z-axis is parallel with the cylinder axis and the y-axis
pared with LDV-measurements of Lyn et al. [6]. The is perpendicular to both x and z. The origin of the coor-
same flow was considered as a test case LES2 at the First dinate system is located at the centre of the body, see
ERCOFTAC Workshop on Direct and Large Eddy Fig. 1. In order to obtain the governing equations and
Simulation in March 1994. Seven groups took part in
the LES2 exercise, and the results of this exercise were
reported by Voke [7]. Sohankar et al. [8,9] have investi-
gated flow around a rigid prism of a square cross-section H
with one side facing the oncoming flow at Re = 2.2 · 104
using LES with three different sub-grid scale models, the
Smagorinsky, the dynamic and a dynamic one-equation
model. The influence of a finer grid, shorter time step A
and the larger computational aspect ratio on their
results was also studied.
The reason for this focus on LES for the study of the
flow around bluff bodies has to do with poor results,
L2
when one uses statistical turbulence models. Most pro- U
bably, this has to do with complicating factors such d
as a strongly retarded stagnation flow, massive flow
separation, streamline curvature, transition from lami- L1
nar to turbulent flow, re-circulation, vortex shedding
and, perhaps most important, the existence of inher- 4
ent three-dimensional flow structures [3,5]. In recent
years, the most of the LES works for flow over a square
cylinder have been performed for a single Reynolds
number, i.e. Re = 2.2 · 104 and the effect of Reynolds 2
numbers on the results of this case is not studied
comprehensively.
The main purpose of the present study is to investi- y
0 x
gate the turbulent flow around a square cylinder for dif-
ferent Reynolds number ranging from 103 to 5 · 106.
The computation of the global results and the study of
the mean and instantaneous flow structures are also the -2
aims of this work. All computations were performed
using large eddy simulations (LES) with two different
sub-grid scale models, the Smagorinsky (SM) and a
-4
dynamic one-equation model (OEM). -2 0 2 4
The organisation of this paper is as follows. Section 2 Fig. 1. Flow configuration and the computational domain (top); and
is devoted to describe the flow configuration, computa- close-up of the grid near the body, the non-uniform grid 266 · 162 in
tional domain, grid and the boundary conditions. A the xy-plane.
1156 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168
different quantities in their non-dimensional form, the
solution variables are normalized as follows. All geo-
metrical lengths are scaled with d. The dimensions of
the computational domain are denoted with L1, L2, d,
H and A (see Fig. 1) and their normalized values are
selected respectively as 7.5, 16, 1, 16 and 4 for the pres-
ent study. Velocities are normalized with the inlet free
stream velocity, U, and the physical time with d/U. Scal-
ing with d/U also applies for the shedding frequency,
where St = f d/U is the Strouhal number. The Reynolds
number is defined as Re = U d/v. In the y-direction, the
vertical distance between the upper and the lower walls,
H, defines the solid blockage of the confined flow
(blockage parameter, b = d/H).
Based on the refinement study of the previous work
of the author [8], all simulations are performed for two
resolutions of 186 · 106 · 26 and 266 · 162 · 26
(denoted as fine grid, see Fig. 1) points in x, y and z,
respectively. The grid distribution is uniform with a
constant cell size, outside a region from the body,
which extended two units upstream, downstream and
sideways (in the x- and y-directions). The minimum
distance from the cylinder surface to the nearest grid
point approximately is 0.008. The hyperbolic tangent
function was used for stretching the cell sizes around
the body. A uniform grid was used in the spanwise
direction. A uniform flow (u = 1, v = w = 0) was
prescribed at the inlet as a boundary condition. At
the outlet, the convective boundary condition ðouoti þ
U ou
ox
i
¼ 0Þ [1] was applied for all velocity components.
No-slip conditions were prescribed at the body sur-
faces. Symmetry conditions simulating a frictionless
wall ðou
oy
¼ ow
oy
¼ v ¼ 0Þ were used at the upper and lower
boundaries. A periodic boundary condition was
employed in the spanwise direction.
3. Governing equations and sub-grid model
In LES, all scales larger than a selected size (e.g.
grid size) are computed exactly. Since the small scales
tend to be more dissipative, isotropic, short-lived,
homogeneous, random, universal, and less affected by
the boundary conditions than the larger ones, they
are not explicitly simulate. Thus, the effect of the small
scales of turbulence in LES is modelled. Three-dimen-
sional and the time dependent solution of the
Navier–Stokes equations are used in LES. The LES
governing equations are accomplished using a general
spatial filtering of the instantaneous equations, e.g. a
box filter, to limit the range of the scales in the flow
field. In this procedure, a general variable is decom-
posed into a large scale and a sub-grid scale compo-
nent, e.g. f ¼ f~ þ f sgs . For an incompressible flow of
a Newtonian fluid, the filtered Navier–Stokes equations
take the following form [9]:
~ui;i ¼ 0 ð1Þ
1
ot ~ui þ ð~ui ~uj Þ;j þ ~p;i þ sij;j ¼ Re ~ui;jj ð2Þ
The influence of the small scales on the large (resolved)
scales takes place through the sub-grid scale stresses
(SGS), sij ¼ r ~ij  ~ui ~ui ðrij ¼ ui uj Þ, which are unknown
and must be modelled with a sub-grid model. The sub-
grid stresses in LES are only a fraction of the total stres-
ses, thus, it is expected that the modelling error should
not affect the overall accuracy of the results as much
as in RANS approach. The details of the LES formula-
tion are found in [10].
3.1. Sub-grid models
The most of the sub-grid models are based on the
eddy-viscosity model. They use the Boussinesq hypothe-
sis, which expresses a direct proportionality between the
deviatoric part of the turbulent stress and the velocity
gradients, i.e. saij  sij  dij skk =3 ¼ vt ð~ui;j þ ~uj;i Þ.
The most widely used SGS model is Smagorinsky
model [11] (SM). In this model, the eddy viscosity,
which is the product of a length (l / D ~ ¼ ðdV Þ1=3 ) and
qffiffiffiffiffiffiffiffiffiffi
a velocity (qsgs / l S~ij S~ij ) scales, is calculated algebrai-
qffiffiffiffiffiffiffiffiffiffiffiffiffi
~ 2 2S~ij S~ij , where S~ij ¼ 0:5ð~
cally as: vt ¼ ðC s DÞ ui;j þ ~
uj;i Þ.
In the above relations, S~ij ; dij ; vt ; D;
~ C s and dV are the
strain tensor, the Kronecker delta, the sub-grid eddy vis-
cosity, the filter width (e.g. grid spacing), the propor-
tionality factor, and the local cell volume, respectively.
In this model, Cs is a constant value that must be
assigned a priori to simulation. In the present study,
Cs was set to 0.1. This model is incapable of taking into
account the reduction of the length scales near the solid
walls. A remedy is to use a damping function. In addi-
tion, the Smagorinsky model is absolutely dissipative
and cannot account for backscatter, but energy is always
transferred from large to small scales.
Dynamic models, which are capable of removing
some of the drawbacks of the Smagorinsky model, are
a suitable alternative. The first attempt to introduce a
dynamic SGS eddy viscosity model was developed by
Germano et al. [12]. In this model, the proportionality
factor is calculated during the computation process. A
new filter, test filter, is used to calculate this coefficient.
In this model, in addition to the grid filter (here denoted
by a tilde), a test filter (here denoted by an overbar) is
also employed, whose width is larger than the grid filter
width (e.g. D  ¼ 2D). ~ One of the drawbacks of the
dynamic model is the numerical instability associated
with the negative values and the large variation of the
proportionality coefficient (C) [8,9]. To avoid numerical
instability owing to an extensive variation of C in time
and space, spatial averaging in the homogeneous direc-
tion and additional local averaging are performed on C.
 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168 1157

In the present study, a one-equation dynamic sub-
grid model (OEM) is also employed. This model is able
to capture the backscattering phenomena and it is insen-
sitive to the existence of homogeneous directions and
also more stable than standard dynamic model
[8,9,13]. This model was developed by Davidson [13]
and has been successfully applied to different flow con-
figurations [8,9,13–16]. In this model, the eddy viscosity
viscosity field including a significant partition with neg-
ative values [9,13], which is destabilizing in numerical
simulations. The C Khom is determined at each time step
with requirement that the sub-grid dissipation of the
resolved kinetic energy Pksgs in the whole computational
domain remains the same as it is with the local coeffi-
cient, C k , i.e.
1 1

~ 0:5 ) is based on a length scale (l / D ~¼ h2C k k 2Ksgs S~ij S~ij Di

~ k ~ ~ ~
xyz ¼ 2C hom hk Ksgs S ij S ij Dixyz
2
ð7Þ
(vt ¼ C K Dk ksg pffiffiffiffiffiffiffi
1=3 sgs where h ixyz denotes the space averaging over the entire
ðdV Þ ) and a new velocity scale (q / k sgs ). ksgs is
the sub-grid turbulent kinetic energy and it is deter- domain.
mined by its modelled transport equation [13] as: The idea is to include all local dynamic information
through the source terms of the transport equation for
uj k sgs Þ;j ¼ ½ðv þ C khom k 0:5
ot k sgs þ ð~ ~
sgs DÞk sgs;j ;j ksgs. This is probably physically more sound since large
~
þ P Ksgs  C k k 1:5 =D; ð3Þ local variations in the dynamic coefficients appear only
 sgs
in the source term, and the effect of the large fluctuations
P Ksgs ¼ saij ~
ui;j ;
in the dynamic coefficients will be smoothed out. In this
~ k k 0:5 S~ij ;
saij ¼ sij  dij skk =3 ¼ 2vt S~ij ¼ 2DC sgs way, it turns out that the need to average or limit the
S~ij ¼ 0:5ð~ui;j þ ~uj;i Þ ð4Þ dynamic coefficients Ck in Eq. (5) and ðC k Þnþ1 in Eq.
(6) is eliminated altogether. The coefficients in the one-
In Pksgs, the dynamic coefficient Ck is computed in a way equation model affect the stresses in only an indirect
similar to that used in the standard dynamic model [12], way, while in the standard dynamics model the C coef-
i.e. ficient is linearly proportional to the stresses, which
‘ij M ij can make it numerically unstable.
Ck ¼ ; ‘ij ¼ ð~ ui ~ ui 
uj Þ  
~ ~
uj ; K ¼ k sgs þ 0:5‘ii ;
2M ij M ij
 0:5 S
M ij ¼ DK ~ 0:5 S~ij ;
~ij þ Dk 4. Numerical solution procedure
sgs

ð5Þ
where ‘ij denotes the dynamic Leonard stresses.
K  0.5Tii and Tij are the sub-grid kinetic energy and
stresses on the test level, respectively (test filter opera-
tion denoted with an overbar in the relations). The S~ij
is denoted as the strain rate tensor on the grid level.
The Smagorinsky model is based on the assumption
of local equilibrium of SGS turbulent energy, i.e.
Pksgs  eksgs = 0, where eksgs is the dissipation of ksgs.
A slightly better assumption for estimating the coeffi-
cient ck in the dissipation term would be to assume that
the filtered right-hand side of the ksgs equation is equal
to that of the K equation, i.e.,
k nþ1
P ksgs  C k ðk 1:5 ~ k 1:5 
sgs =DÞ ¼ P K  C  ðK =DÞ ) ðC  Þ
¼ ½P K  P ksgs þ ðC k Þn ðk 1:5 ~  1:5
sgs =DÞðD=K Þ
ð6Þ
 0:5 
where P K ¼ 2C DK S ij S~ij . The above relation gives the
k ~
coefficient C k in the dissipation term for time step n + 1.
Note that ðC k Þn in Eq. (6) has been kept inside the filter-
ing process.
The dissipation cannot be negative, which requires
that we limit C k to positive values, i.e. C k P 0. To
ensure numerical stability, a homogenous value of Ck
in space (C Khom ) is used in the momentum equations
and in the diffusion term in Eq. (3). The reason is that
the local coefficient Ck yields a highly oscillating eddy
An incompressible 3D finite volume code with a collo-
cated grid arrangement is used for solving the filtered
Navier–Stokes equations. Eqs. (1) and (2) are discretized
and they are solved with an implicit fractional two-step
method. Convective, viscous and sub-grid fluxes are
approximated by central differences of the second-order
accuracy. All terms in the momentum equations are
advanced in time using the Crank–Nicolson scheme. The
momentum equations are solved with the Gauss–Seidel,
whereas a multigrid V-cycle is used for the acceleration
of the convergence, when solving the pressure equations.
The time-marching calculations were started with the fluid
at rest, and a constant time step, Dt = 0.02 was selected,
which was smaller than the chosen value for the most of
the cases in the previous work of author [8].
The filtered Navier–Stokes equations are integrated
over a time interval and over the control volumes, thus
the discrete form of Eq. (2) can be written as:
~unþ1 ¼ ~uni þ DtF ð~uinþ1 ; ~uni Þ  aDtP~ ;i
nþ1 n
 ð1  aÞDtP~ ;i ð8Þ
i
where F ð~unþ1
i ; ~uni Þ includes convection and viscous terms
and also the sub-grid stresses [9].
The numerical procedure for obtaining ~ui at each time
step can be summarized as follows:
~inþ1 .
1. Solve the discretized equation (8) for obtaining u
At this stage, the most recent values of the pressure
and velocities are used.
1158 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168
2. Intermediate velocities are defined as:
n author [8] for this flow geometry. In Ref. [8], it was
~u#
i ¼ uni
~ þ DtF ð~uni ; ~
unþ1
i Þ  ð1  aÞDtP~ ;i
By using Eq. (8), u#
i can be calculated as:
nþ1
u#
~ uinþ1 þ aDtP~ ;i .
i ¼ ~
The intermediate velocity field (~ u#
i ) does not satisfy
the continuity equation. Velocities at control volume
faces (~u#
face ) are calculated by interpolating from
neighbouring intermediate velocities at the nodes.
These interpolated cell face velocities are used to cal-
culate the mass fluxes and convection terms.
3. To satisfy continuity for ~ nþ1
uiðfaceÞ , take the divergence of
Eq. (10) and insert ~ nþ1
ui;iðfaceÞ ¼ 0, thus
nþ1 1
P~ ;ii ¼ ðaDtÞ ~u#
iðfaceÞ;i
This is the Poisson equation for obtaining pressure on
the level (n + 1). This equation is solved with a multi-
grid Poisson solver. A Neumann boundary condition
was applied on the Poisson equation for the pressure.
4. Compute the cell face velocities (which satisfy conti-
nuity) from the pressure intermediate velocity as
nþ1
unþ1 u# ~
~ i;face ¼ ~ i;face  aDtðP ;i Þface
These velocities are used to calculate the mass fluxes
at the faces of the control volume.
Steps 1–4 are carried out for each iteration until a
global convergence criterion reaches at each time step.
Finally, the turbulent viscosity is computed and the next
time step starts.
5. Results and discussion
Large eddy simulations were performed to study the
uniform approach flow over a square-section cylinder
for various Reynolds numbers with two different sub-
grid scale models, the Smagorinsky (SM) and a dynamic
one-equation model (OEM). The simulations were car-
ried out for the selected Reynolds numbers of 103
(OEM), 3 · 103 (OEM), 8.7 · 103 (SM, OEM), 1.5 ·
104 (OEM), 2.2 · 104 (SM, OEM), 5 · 104 (SM), 105
(SM), 5 · 105 (OEM), 106 (OEM) and 5 · 106 (OEM).
A series of time- and spanwise-averaged resolved veloc-
ity, pressure, turbulent stresses, and streamlines were
calculated. The time-averaged global quantities such as
the Strouhal number, the base suction pressure, the
mean and the RMS values of the drag force, the lift
force and the pressure were also computed over about
20 shedding cycles and compared with available experi-
mental results. All simulations were carried out for a
solid blockage of b = 6%.
Influence of the spatial and the temporal resolutions,
the blockage effects and the computational spanwise
length was studied in the previous work of the present
ð9Þ
shown that using the finer resolutions in the space and
time and also employing a larger spanwise dimension
can improve the agreement between the predictions
ð10Þ and the experimental results, especially when also taking
into account the blockage effects. Based on the results of
that work, the suitable resolutions are chosen for the
present study, see also Section 2.
5.1. Global quantities
A comparison between the data on various quantities
of the present LES simulations (1000 6 Re 6 5 · 106)
and the published experimental results [6,18–24] are pro-
ð11Þ vided in Figs. 2 and 3. The 2D/3D numerical results of
the previous works of the present author (45 6 Re 6
500) [4,17] are also included in this figure. It is important
to mention that the RMS and mean drag (CD, CD 0 ) and
also the RMS lift (CL 0 ) coefficients were computed from
the time-dependent, spanwise-averaged forces on the
cylinder for all simulations. The Strouhal numbers were
calculated from the fluctuating lift signal.
The variations of the Strouhal number and drag coef-
ð12Þ
ficient with Reynolds number are shown in Fig. 2. As
seen, there is an increase in the Strouhal number with
increasing the Reynolds number from about 50 to 200,
whereas the drag coefficient continuously decreases in
this range of the Reynolds numbers, see Fig. 2. In the
lower limit of this range, the phenomenon of vortex
shedding results (Re  50) and a 2D instability from a
steady wake to a periodic wake forms, which gives rise
to the fluctuating drag and lift forces [1]. Effect of three
dimensionality appears close to upper limit of this range
(150 < Re < 200 [4]), which causes to increase the St and
decrease CD with increasing the Reynolds number. The
St reaches to a local minimum value at around
Re = 1000, which is about 0.12 and it shows a good
agreement with the experimental result of Okajima
[18]. With increasing the Reynolds number from 103 to
about 2 · 104, the St increases slightly, and then reaches
approximately to a constant level at about 0.13 for the
higher Reynolds number ranging about 2 · 104–5 ·
106. This Reynolds-independent of the Strouhal number
for Re > 2 · 104 is also observed for drag coefficient,
RMS lift and drag in Figs. 2 and 3. As seen form Figs.
2 and 3, there is some difference between the results of
the present study with the experimental ones, but this
Reynolds-independent of the results for Re > 2 · 104 is
also seen in the experimental results. As it was shown
in the previous work of the present author [8], we can
improve the agreement between the predictions and
the experiments results by using finer resolutions in
space and time and also taking into account the block-
age effects. Some disagreements between the present
results and experimental ones are due to use of the
 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168 1159

Fig. 2. Strouhal number (St) and mean drag coefficient (CD) versus Reynolds number, comparison with published experimental [6,18–24] and
numerical [4,17] results.

Fig. 3. RMS lift and drag versus Reynolds number, comparison with published experimental [19,20,22,23,25] and numerical [4,17] results.

different inlet turbulence intensity and different blockage
effects in the experiments [8,22,23]. For example, it is
shown that when the turbulence intensity increases or
the blockage decreases, the drag value reduces [25,26].
It is important to mention that the uniform flow at inlet
and a blockage of 6% are applied for the present
investigation.
From simulations of the present study (1000 6 Re 6
5 · 106), it is found that the ratio of the mean drag coef-
ficient and the base suction pressure coefficient (CD/
(CPb)) is approximately constant and equals to
1.47 ± 0.04. This finding is in agreement with the exper-
iments and also the previous work of the present author
[8]. If one computes this ratio from the published exper-
imental results, they are about 1.45 [18], 1.52 [19], 1.54
[20], 1.53 [21]. At low Reynolds numbers, this finding
is not valid, where it is about 3, 2.32, 1.93, 1.74 and
1.48 for Re = 45, 100, 150, 200, 500 [4,17], respectively.
The results of this study show that the stagnation pres-
sure coefficient approximately varies from 1.03 to 1.01
for Reynolds number ranging from Re = 103 to
Re = 5 · 106, respectively.
5.2. Mean-flow structure
Both time- and spanwise-averaged resolved and fluctu-
ation quantities such as pressure, turbulent stresses, tur-
bulent kinetic energy and streamlines are provided for
different Reynolds numbers (1000 6 Re 6 5 · 105) to
gain an understanding of the mean flow structure. It is
worth to mention that the time-averaged quantities were
calculated over about 20 shedding cycles. The computed
resolved quantities are decomposed into a time-mean
component and a fluctuation (denoted by 0 ) component.
For example, the resolved velocities are decomposed as,
~ui ¼ h~uit þ u0i . Hereinafter, the tilde and bracket symbols
are dropped form the resolved and the time- and span-
wise-averaged quantities for simplicity.
1160 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168
The time- and spanwise-averaged pressure coeffi-
cients (Cp), turbulent shear stresses (u 0 v 0 ) and turbulent
kinetic energy (k = 0.5(u 0 u 0 + v 0 v 0 + w 0 w 0 )) contours are
compared for two chosen Reynolds numbers, i.e.
Re = 3 · 103 and 106. One can easily observe that the
time-and spanwise-averaged quantities are symmetric
and a slight asymmetry in these plots is due to the insuf-
ficient averaging time, which is taken to obtain the
mean-values.
The time- and spanwise-averaged pressure coefficient
contours are shown in Fig. 4(top). As seen, the differ-
ence between these two plots is not much. For example,
the minimum pressure coefficient behind the body is
located on the centreline at around x = 1 with values
of 2.04 and 1.94 for Re = 3 · 103 and 106, respec-
tively. The pressure coefficient at the centre of recirculat-
ing regions on the sides of the body are about 2.01 and
2.15 for Re = 3 · 103 and 106, respectively. Similar
behaviour is found for the other selected Reynolds num-
ber of the present study.
The time-and spanwise-averaged turbulent shear
stress contours (u 0 v 0 ) are compared in Fig. 4(middle).
Two plots are very similar with a slight difference
between the maximum values, where their absolute val-
ues are about 0.20 (x = 1.37, y = 0.45) and 0.21
(x = 1.28, y = 0.45) for Re = 3 · 103 and 106, respec-
tively. Fig. 4(bottom) shows a comparison between tur-
bulent kinetic energy contours (k) for two Reynolds
numbers. Here, the similarity between these two cases
is also observed and the maximum values are located
on the centreline at around x = 1.5 with k = 0.50 and
k = 0.47 for Re = 3 · 103 and 106, respectively.
In general, similar behaviours are observed for the
other quantities such as streamlines, velocity compo-
nents, pressure fluctuation (p 0 p 0 ) and the components
of turbulent stresses (u 0 u 0 , v 0 v 0 , w 0 w 0 ) for different Rey-
nolds numbers in the range of the present study, for
instance, see turbulent shear stress contours (u 0 v 0 ) in
Fig. 5 for Re = 2.2 · 104, (this case is provided for a fine
grid, 266 · 162 · 26). The similarity between the time-
and spanwise-averaged quantities becomes closer at
the Reynolds numbers which are approximately greater
than 2 · 104, where the dependency of the results to the
Reynolds number is weak, see also Figs. 2 and 3.
Because of similarity of the mean quantities for various
Reynolds numbers, the results of only one chosen case
(Re = 2.2 · 104) is explained here and it is quantitatively
valid for the other Reynolds numbers of this
investigation.
The time- and spanwise-averaged streamlines are
shown in Fig. 5(bottom and right-side). In this plot, it
is seen that the flow separates from the upstream corners
and it is not reattached to the upper and lower faces.
Consequently, two large recirculating regions, bubbles,
on the sides of the body and a pair of counter-rotating
vortices behind the body form. The position and the
value of the pressure coefficient at the centre of these
bubbles on the sides of the body are (x = 0.19,
y = ±0.65) and 2.12, respectively. As seen from
Fig. 5, the maximum of the pressure fluctuation (p 0 p 0 )
is also located in the centre of these bubbles, which is
about 0.22. The minimum pressure coefficient behind
the body is 2.05, which is located at the end of the
reverse flow region on the centreline of the cylinder at
around x = 1. These minimum pressure coefficients
and their locations on the wake centreline and on the
side of the body are very close to the aforementioned
values for Re = 3 · 103 and 106 in Fig. 4. The length
of the time-mean recirculation zone behind the body,
lr, is about 1.05 for this case, see Fig. 5. The variation
of lr and the maximum mean absolute velocity value in
the backflow on the wake centreline, UBF, are also rela-
tively constant at high Reynolds numbers. For example,
the predicted values of lr and UBF are about (1.12, 0.1),
(1.05, 0.11), (1.02, 0.08), (0.97, 0.08) and (1.03, 0.07) for
Re = 3 · 103, Re = 22 · 103 (high resolution case),
Re = 5 · 105, Re = 106 and Re = 5 · 105, respectively.
The normalized time- and spanwise-averaged of the
different components of the resolved normal (u 0 u 0 , v 0 v 0 ,
w 0 w 0 ) and shear (u 0 v 0 ) stresses are shown also in Fig. 5.
It can be found from this figure that the maximum of
the streamwise (u 0 u 0 ) and cross-flow (v 0 v 0 ) normal stres-
ses are larger than the magnitude of the shear stresses
(u 0 v 0 ). The maximum values of u 0 u 0 are located some-
where in the shear layer along the sides of the body,
which is about 0.60, while the maximum value of v 0 v 0
is located on the wake centreline (x = 1.5), which is
about 0.80. In Fig. 5, the contours of time- and span-
wise-averaged shear stresses (u 0 v 0 ) are also plotted. The
magnitude of the maximum value of u 0 v 0 is located
somewhere in the wake (x = 1.35, y = 0.46), and its
value is about 0.22, which is in a good agreement with
the cases that are shown in Fig. 4, especially for the case
with Re = 106. In this figure, the time- and spanwise-
averaged normal stresses isolines (w 0 w 0 ) is also plotted.
As seen, the maximum value is located on the wake
centreline (x = l), and it is about 0.12.
In Figs. 6 and 7, the time- and spanwise-averaged
streamwise velocity (u), pressure coefficient (Cp), turbu-
lent normal stresses (u 0 u 0 , v 0 v 0 ), turbulent shear stresses
(u 0 v 0 ) and turbulent kinetic energy ðk ¼ 0:5u0i u0i Þ are com-
pared for different Reynolds numbers and available
experimental results [6,27]. The results along the wake
centreline, y = 0, (Fig. 6) and at the streamwise posi-
tions, x = 0, 0.5, 1.5, (Fig. 7) are provided for different
Reynolds numbers and sub-grid models, i.e. 3 · 103
(OEM), 8.7 · 103 (SM), 2.2 · 104 (SM and fine grid
OEM), 5 · 104 (SM) and 106 (OEM).
Turbulent normal stresses (u 0 u 0 , v 0 v 0 ) and turbulent
kinetic energy (k) are shown together with the experi-
mental results [6] at the wake centreline for different
Reynolds numbers in Fig. 6. As seen, the agreement is
 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168 1161

Cp -0
Cp -0 .
44
Re=3000 -0 . . 5 Re=1000000
6 -0 .
1 -0 . 1 68
8
-1
-1 .
04

-2
-1 . -1 .4
9
-1 .6 -1 .4
-1 .8 -1 .6 4
-2 -1 .8 8
0
1

y
y

-1 -1

0 1 2 0 1 2
x x

u' v' u' v'

-0 .0 1 Re=1000000 -0 .0 1
Re=3000 5
-0 . 0 -0 .0 5
1 -0 .0 9 1 -0 .0 9
- 0 .1 3 -0.13
-0
-0 . .1
1 7
-0 .1 9 7 -0 .2
1

0. 03 0 .0 3
0 0
y

-1 -1

0 1 2 0 1 2
x x

k 2
Re=1000000
2
0 .0 6 k 0 .0 0 .0 6
Re=3000 0 .0
0 .1 2 0 .1 4
1 1
0.2 0 .2
0 .3 0 .3
0 .3 0 .3
0. 6
8 42
0.
44
0 .4 6
0.26
0 .2 6

0 0
y

y
0 .4 8

-1 -1

0 1 2 0 1 2
x x

Fig. 4. Time- and spanwise-averaged pressure coefficients (Cp), turbulent shear stresses (u 0 v 0 ) and turbulent kinetic energy ðk ¼ 0:5u0i u0i Þ contours for
Re = 3 · 103 and Re = 106.

good between the results of this study and the experi- these values, especially k, for Re > 2 · 104 is not much.
mental one, especially for Re = 2.2 · 104. In general, This means that although the values of u 0 u 0 , v 0 v 0 and also
the results are qualitatively similar for all cases. For w 0 w 0 change slightly with Reynolds number, but the sum
example, the maximum values of u 0 u 0 , v 0 v 0 and k are of these values at the high Reynolds numbers is approx-
about (0.212, 0.719, 0.495), (0.14, 0.755, 0.494), (0.138, imately constant. Thus, the dependency of the time- and
0.736, 0.479), (0.147, 0.716, 0.478) and (0.151, 0.699, spanwise-averaged quantities with Re are weak at the
0.470) for Reynolds numbers of 3 · 103 (OEM), high Reynolds numbers as it was mentioned before.
8.7 · 103 (SM), 2.2 · 104 (SM), 5 · 104 (SM) and 106 Pressure coefficient (Cp) at the wake centreline
(OEM), respectively. As seen, the difference between together with experimental results [27] are shown in
1162 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168

u' u' v' v'

03 Re= 22000 0 .0
5
Re= 22000 0. 0 .0 9
1 0 .1 8 1
0 0 .1 5
0. 3 6 . 3 0 0 .3 0
0.40

0 .6
5

0.

0.
15

75
0 0
y

y
-1 -1

0 1 2 0 1 2
x x

w' w' u' v' 1

Re= 22000 1 Re= 22000 -0 .0
1 0 .0 1 -0 .0 7
0 .0 4 -0 .1 3
-0.17
0. 07 -0 .2 1
0 .0 9
0 .1 0 .0
1 3
0 0
y

y
3
-0 .0

1
0 .2
17
0.
0 .13
-1 -1 0 .0 7
0 .0 1

0 1 2 0 1 2
x x

p' p' 0.02 Re= 22000

Re= 22000 0.03
1 1
0.06

0.22
0 .0 9

1
0 .1
0 0
y
y

0 .0
6
-1 -1

0 1 2 0 1 2
x x
Fig. 5. Time- and spanwise-averaged streamlines, pressure fluctuation (p 0 p 0 ), the components of turbulent normal stresses (u 0 u 0 , v 0 v 0 , w 0 w 0 ) and
turbulent shear stresses (u 0 v 0 ) contours for Re = 2.2 · 104.

Fig. 6, where the minimum pressure coefficient is located mental results [6] at one streamwise position (x = 1.5)
at x = l with values of 2.04, 1.95, 1.86, 1.95 and the streamwise velocity (u) at three positions
1.84, 1.94 for Reynolds number of 3 · 103(OEM), (x = 0, 0.5, 1.5) are shown for different Reynolds num-
8.7 · 103 (SM), 2.2 · 104 (SM), 2.2 · 104 (OEM), bers. A relatively sufficient spatial resolution is used
5 · 104 (SM) and 106 (OEM), respectively. close the body. For example, the maximum negative
Time- and spanwise-averaged pressure coefficient velocity (u = 0.26) occurs at the 7th node from the
(Cp), turbulent shear stresses (u 0 v 0 ) and turbulent kinetic body at around y = 0.57 for Re = 106 for the coarser
energy ðk ¼ 0:5u0i u0i Þ versus y together with the experi- grid while this point is located at the 9th node for the
 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168 1163

Re
0. 8 3000, OEM
8700, SM Re
22000, SM 3000 , OEM
22000, OEM 8700, SM
50000, SM 22000, SM
1000000, OEM 22000, OEM
21400, Lyn et al. 50000, SM
v' v'

1000000, OEM
0. 5

k
0. 4

0 0
5 10 15 5 10 15
x x

0 0. 5

-0 .5
Re
3000, OEM
8700, SM
22000, SM
Re 22000, OEM
-1
50000, SM
3000, OEM
u'u'
Cp

8700, SM 0. 25 1000000, OEM

22000, SM 21400, Lyn et al.
-1 .5 22000, OEM
50000, SM
1000000, OEM
67000, Nakamura & Ohya
-2

-2 .5 0
5 10 15 5 10 15
x x
Fig. 6. Time- and spanwise-averaged pressure coefficient (Cp), turbulent normal stresses (u 0 u 0 , v 0 v 0 ) and turbulent kinetic energy ðk ¼ 0:5u0i u0i Þ versus x
at the centreline (y = 0) for different Reynolds numbers. Experiments: Lyn et al. [6], Nakamura and Ohya [27].

finer grid. As seen, the agreement between the present
results and the experiments is good and there are not
large differences between the results for various Rey-
nolds numbers, especially at higher Reynolds numbers.
It is important to mention that some of the differences
between the results for different Reynolds numbers are
due to the effect of sub-gird models, for example, see this
difference between the values of minimum pressure coef-
ficient for case Re = 2.2 · 104. The effect of sub-grid
model on the results was also reported in the previous
work of the present author [8]. In general, it is seen that
the effect of Reynolds number on the time- and span-
wise-averaged quantities is not much at high Reynolds
number, Re > 2 · 104, especially when one employs the
same grid and sub-grid model.
It is worth mentioning that the previous numerical
work by the author [17] and later analysis [1] show that
the separation for Re 6 100, at all times in the fully
saturated state, occurred from the rear corners, predom-
inantly from the rear corners at Re = 125 with occa-
sional upstream corner separation, predominantly
from upstream corners at Re = 150 and finally, at all
times, from upstream corners for Re P 175. The results
of this study also verify that the separation points
remains at the upstream corners at all times for the
higher Reynolds numbers. Thus the mean-structure of
the flow is not much sensitive to where the boundary
layer is laminar or turbulent. In general, the sharp-edged
bodies, which tend to cause flow separation regardless of
the character of the boundary layer, are insensitive to
the Reynolds number. This point was also mentioned
in Ref. [28], where a relatively constant value for the
mean drag coefficient (CD  2.1 for Re > 104) was
reported for flow over a square cylinder.
5.3. Instantaneous-flow structure
To gain further insight into the similarity and dissim-
ilarity of the flow structure for different Reynolds num-
bers, the instantaneous wake flow structures were
1164 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168

4 4 4
x= 0 x= 0.5 x= 1.5

3 3 3

2 2 2
y

y
1 1 1

0 0 0
-0.5 0 0.5 1 1.5 -0.5 0 0.5 1 1.5 0 0.5 1 1.5
u u u

4 4 4
Re Re
3000, OEM 3000, OEM
8700, SM 8700, SM
22000, SM 22000, SM
22000F, OEM 22000F, OEM
3 3 3
50000, SM 50000, SM
1000000, OEM 1000000, OEM
21400, Exp.

2 2 2
y

1 1 1

0 0 0
-0.2 0 0.2 -1.5 -1 -0.5 0 0 0.2 0.4 0.6
u'v' Cp k

Fig. 7. Time- and spanwise-averaged streamwise velocity (u at x = 0, 0.5, 1.5, see u 0 v 0 -plot for legend), pressure coefficient (Cp), turbulent shear
stresses (u 0 v 0 ) and turbulent kinetic energy ðk ¼ 0:5u0i u0i Þ versus y at x = 1.5 for different Reynolds numbers, Exp.: Lyn et al. [6].

provided for a number of selected Reynolds numbers in
the range of 103 6 Re 6 5 · 106. Due to the lack of
space, the results of only some cases with the same grid
and subgrid model (OEM) are compared here, i.e.
Re = 3 · 103 and Re = 5 · 106 in Figs. 8 and 9, and
Re = 103 and Re = 106 in Fig. 10. Please note that the
x-axis in Figs. 8–10 is originated from the inlet of the
computational domain (the body centre is located at
x = 7.85), while it is measured from the body centre in
Figs. 4–7, see also Fig. 1.
To study the instantaneous-flow structure, it is neces-
sary to recognize the significant vortical regions in the
flow field. It is important to mention that a vortex
should have a vortex core with a net vorticity and the
geometrical characteristics of the identified vortex must
be Galilean invariant [29]. Because an agreed definition
of a vortex does not yet exit, thus it is difficult to recog-
nize the general vortical structure for all flow cases with
only one method, where these requirements for the vor-
tex core may not yield with a single scheme. Therefore,
several methods such as using streamlines, pathlines,
vorticity, pressure, the rate of strain tensor, etc. are pro-
pounded for identification and visualization of vortical
structures [29]. In the present work, the instantaneous
flow structures at the mid-span and wake centre planes
are studied using vorticity, pressure and the second
 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168 1165

Fig. 8. Instantaneous pressure coefficient (top), spanwise vorticity (middle) and the second invariant of velocity gradient (bottom) contours in the
streamwise direction at the mid-span (z = 0). Left: Re = 3 · 103 (OEM), right: Re = 5 · 106 (OEM), here the x-coordinate is measured from the inlet.

Fig. 9. Instantaneous pressure coefficient (bottom), cross stream vorticity (middle) and the second invariant of velocity gradient (top) contours in the
streamwise direction at the wake-centre (y = 0). Left: Re = 3 · 103 (OEM), right: Re = 5 · 106 (OEM), here the x-coordinate is measured from the
inlet.
1166 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168
Fig. 10. Iso-surfaces of the second invariant of velocity gradient
(Q = 0.4, 1). Top: Re = 106 (OEM), bottom: Re = 103 (OEM), here the
x-coordinate is measured from inlet.
invariant of the velocity gradient tensor. It is worth
mentioning that a region with a pressure value lower
than the ambient one and a positive second invariant
of velocity gradient tensor (Q = 0.5(ui,iuj,j  ui,juj,i) > 0)
is also defined as a vortex core [30].
Instantaneous pressure coefficient, vorticity and the
second invariant of velocity gradient (Q) contours at
the wake-centre plane (y = 0, xz-plane) and at the mid-
span plane (z = 0, xy-plane) are provided for two differ-
ent Reynolds numbers, i.e. Re = 3 · 103 (CL = 2.39)
and Re = 5 · 106 (CL = 2.05) at one instant, where
the lift coefficient is close to its minimum value, see Figs.
8 and 9. Iso-surfaces of the second invariant of velocity
gradient (Q) are shown at one instant (t = 160) for
Re = 103 (CL = 1.61) and Re = 106 (CL = 1.31) in
Fig. 10. As it is seen from these figures, although there
are some differences in the size and the strength of the
small-scale vortices, but the large scale instantaneous
flow-structures are approximately similar for both
Reynolds numbers. These similarity and dissimilarity
are observed in the instantaneous structures for the
other selected Reynolds numbers of the present study
(these results are not shown here due to the lack of
space).
On the downstream part of the body the spanwise
(von Kármán) vortices of the alternating sign form,
see the corresponding black and white colours in
Fig. 8(middle). The strength of the vorticity decreases
rapidly in the streamwise direction, especially beyond
the vortex formation region (x > 11, approximately).
For example, the pressure coefficient and the magni-
tude of vorticity at the core of four vortices (Fig. 8 at
x > 11 and Re = 3000) are (2.72, 8.51), (1.9, 6.5),
(1.85, 4.53) and (1.13, 3), respectively. As seen in
Fig. 8, the three methods of vortex identification pre-
dict the same structure and vortex core at the mid-
span plane, while the similarity of the pressure and the
second invariant of velocity gradient patterns are much
closer.
It was reported in the previous work of the present
author [4] that a 2D/3D transition develops at
Re = Rec  150–200 for flow over a square cylinder
and both spanwise instability modes (A and B) are pres-
ent in the wake transitional process, similar to the case
of the flow around a circular cylinder [3]. At Re = 200
the spanwise wavelength of mode A instability associ-
ated with the initiation of transition from 2D to 3D flow
is approximately five diameters (diameter = side dimen-
sion of cylinder) [4]. The short-wavelength mode B
instability, appearing at around Re  250, has a wave-
length with the order of one diameter [4].
From the present results (e.g. Figs. 8–10), the stream-
wise-oriented tubular flow structures similar to the mode
B instability are approximately observed. The A mode
vortices are not seen in the present results, but close to
the cylinder they appear to have some similarities with
the mode B instability as found at lower Reynolds num-
ber [4] for Re P 250 (in Ref. [4] the highest Re is 500). It
is not possible to determine accurately the spanwise dis-
tance between the pairs of these tubular-like, streaky
structures from this instantaneous flow field. However
within the different planes (e.g. xz-plane in Fig. 9), the
number of pairs along lines, e.g. x = 12 at Re = 5 ·
106, appears to be in the same order as the spanwise
dimension. This implies as a rough estimate that the
spanwise distance of these structures are of the order
of one diameter. This compares with findings at high
Reynolds numbers for the flow around a circular
cylinder [31].
As it was shown in Sohankar et al. [4], the strength
and the spanwise coupling of the non-spanwise vortical
structures in the near wake are closely related to the fluc-
tuating drag. As can be seen in Fig. 8, there is a high
activity of the streamwise vorticity with high peak levels
in the near wake but with a relatively small degree of
the spanwise coupling. The low spanwise coupling is
consistent with the fact that for this case the ratio
between the spanwise-averaged and the spanwise-mean
sectional RMS drag (cD [4]) was as low as about 0.7.
Because the RMS drag and cD approximately constant
 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168
at Re > 2 · 104, it is expected that the non-spanwise
vortical structures become approximately similar, as it
is also seen in Figs. 3 and 9.
Fig. 10 shows 3D contour surfaces of (positive) Q.
Only two contour surfaces are plotted in this figure.
The spanwise rollers of the von Kármán vortices are
captured from this figure and also from the surfaces of
the constant pressure (only 2D plots for pressure are
shown in Fig. 9 at wake centreline plane) for different
Reynolds numbers. The fluctuating lift is closely con-
nected to the strength and near-wake development of
the spanwise (von Kármán) vortices [4]. The ratio
between the spanwise-averaged and the spanwise-mean
sectional RMS lift (cL [4]) for this study was about
0.96. The sectional RMS lift was about 6 times larger
than the sectional RMS drag. The near-unity value of
cL implies, in the mean sense, a strong spanwise coupling
of the sectional lift forces. As can be seen in Figs. 9 and
10, there is a relatively high degree of spanwise coupling
for the developing spanwise vortex still connected to the
cylinder. For other spanwise (von Kármán) vortices,
which can be traced in Figs. 8–10, some appear to be
strong and highly correlated across the span with only
little undulation, e.g. the vortex centred in Fig. 9 at
approximately x = 8.5 (Re = 3 · 103 and Re = 5 · 106)
and x = 10.5 (Re = 5 · 106), see also Fig. 10 at the posi-
tion of x  14 for Re = 103 and Re = 106. While some
vortices appear to be rather distorted across the span
and with a much higher degree of spanwise undulation,
see Figs. 9 and 10. Although the large-scale structure of
the spanwise vortices is similar for different Reynolds
numbers, but it seems that the spanwise undulation of
the von karman vortices slightly increases at the higher
Reynolds numbers, see Figs. 9 and 10.
In general, while the large scale spanwise vortical
structures become approximately similar for Re > 2 ·
104 due to approximately constant values of the RMS
lift and cL in this range of Reynolds numbers, see also
Figs. 3 and 8–10, but there is some dissimilarity in the
size and the strength of the small scale vortical struc-
tures at the higher Reynolds numbers. Some of these
dissimilarity may result from insufficient grid resolution
and incorrect behaviour of the SGS stresses near the
solid walls, especially for the higher Reynolds num-
bers. The presence of a solid wall can restrain the growth
of the small scales and can alter the exchange mecha-
nisms between the resolved and unresolved scales in
LES [32]. Thus, if the grid is very fine close to the solid
walls, the real flow-structure will appear exclusively
through the resolved scales of LES. When the grid is
coarsened, part of the flow-structure becomes due to
the unresolved scales and most of the energy production
may occur at the sub-grid scale level. Based on the
refinement study of the previous work of the author
[8], two resolutions are employed here for all simula-
tions. When the results of these resolutions are com-
1167
pared, no significant changes is observed in the global,
mean and instantaneous flow-structure. Thus, it is
expected that the accurate results can be obtained with
the fine resolution.
The correct prediction of the SGS stresses near the
solid walls is also important to capture the real flow-
structure. Based on the previous publications regarding
the LES with OEM model [8,9,13–17], it is believed that
the OEM model can predict the SGS stresses near the
body more accurate than the Smagorinsky and the stan-
dard dynamic models. This model is able to capture the
backscattering phenomena and it is insensitive to the
existence of homogeneous directions and also more sta-
ble than standard dynamic model [8,9,14]. Thus, it is
expected that the effect of the sub-grid model, especially
OEM model, on the results and the flow-structure is not
much.
6. Conclusions
Viscous uniform approach flow over a square-section
bluff body is studied using large eddy simulation with
two sub-grid scale models and different Reynolds
numbers, ranging from 103 to 5 · 106. The calculation
of global results and the study of mean and instanta-
neous flow structures for different Reynolds numbers
are the aims of this work. The Strouhal number, the
mean and RMS values of lift and drag are computed
for various Reynolds numbers and they are com-
pared with the available experimental results. The global
results show a good agreement with experiments and
they are approximately Reynolds-independent for
Reynolds number greater than 2 · 104. It is found that
the ratio of mean drag coefficient and base suction pres-
sure is approximately constant for various Reynolds
numbers.
It is also found that the effect of Reynolds number
on the mean and the large scale instantaneous flow-
structure is not much at the high Reynolds number,
i.e. Re > 2 · 104, especially when one employs the
same grid and sub-grid model. The results of this
study also show that the separation points remains at
the upstream corners at all times for the various
Reynolds numbers. Thus, it is expected that the flow-
structure, especially the mean one, is not much sensitive
to where the boundary layer is laminar or turbulent. In
general, the sharp-edged bodies, which tend to cause
flow separation regardless of the character of the
boundary layer, are insensitive to the Reynolds number,
especially at higher Reynolds numbers. Finally, while
the large scale instantaneous flow-structures become
approximately similar for various Reynolds numbers,
but there is some dissimilarity in the size and the
strength of the small-scale structures at higher Reynolds
numbers.
1168 A. Sohankar / Computers & Fluids 35 (2006) 1154–1168
References
[1] Sohankar A, Norberg C, Davidson L. Low-Reynolds number
flow around a square cylinder at incidence: study of blockage,
onset of vortex shedding and outlet boundary condition, I. J
Numer Meth Fluids 1998;26:39–56.
[2] Provansal M, Mathis C, Boyer L. Bénard–von Kármán instabil-
ity: transient and forced regimes. J Fluid Mech 1987;182:1–22.
[3] Williamson CHK. Three-dimensional wake transition. J Fluid
Mech 1996;328:345–407.
[4] Sohankar A, Norberg C, Davidson L. Simulation of unsteady 3D
flow around a square cylinder at moderate Reynolds number.
Phys Fluids 1999;11(2):288–306.
[5] Rodi W, Ferziger JH, Breuer M, Pourquie M. Status of large eddy
simulations: results of workshop. Trans ASME: J Fluids Eng
1997;119:248–62.
[6] Lyn DA, Einav S, Rodi W, Park JH. A laser-Doppler velocimetry
study of ensemble-averaged characteristics of the turbulent near
wake of a square cylinder. J Fluid Mech 1995;304:285–319.
[7] Voke PR. Flow past a square cylinder: test case Les2. In: Chollet
JP et al., editors. Direct and large eddy simulation II. Kluwer
Academic Publishers; 1997.
[8] Sohankar A, Davidson L, Norbeg C. Large eddy simulation of
flow past a square cylinder: comparison of different subgrid scale
models. ASME J Fluids Eng 2000;122:39–47. See also Erratum
J Fluids Eng, vol. 122, p. 643.
[9] Sohankar A. Numerical study of laminar, transitional and
turbulent flow past rectangular cylinders. PhD thesis, Department
of Thermo and Fluid Dynamics, Chalmers University of Tech-
nology, Gothenburg, 1998.
[10] Sagaut P. Large eddy simulation for incompressible flows, an
introduction. Springer Verlag; 2001.
[11] Smagorinsky J. General circulation experiments with the primitive
equations. Month Weather Rev 1993;91:99–165.
[12] Germano M, Piomelli U, Moin P, Cabot WH. A dynamic
subgrid-scale eddy viscosity model. Phys Fluids A 1991;3:1760–5.
[13] Davidson L. Large eddy simulation: a dynamic one-equation sub-
grid model for three-dimensional re-circulating flow. In: 11th int
symp on turbulent shear flow, Grenoble, vol. 3, 1997. p. 26.1–6.
[14] Sohankar A, Davidson L, Norbeg C. A dynamic one-equation
sub-grid model for simulation of flow around a square cylinder.
In: 4th international symposium on engineering turbulence
modeling and measurements, Corsica, France, May 24–26, 1999.
[15] Krajnovic S, Davidson L. Large eddy simulation of the flow
around a bluff body. AIAA J 2002;40(5):927–36.
[16] Schmidt S, Thiele F. Comparison of numerical methods applied to
the flow over wall-mounted cubes. Int J Heat Fluid Flow
2002;23:330–9.
[17] Sohankar A., Davidson L, Norberg L. Numerical simulation of
unsteady flow around a square 2D cylinder. In: Bilger RW, editor.
Proc 12th Australasian fluid mechanics conference, 1995. p. 517–
20.
[18] Okajima A. Strouhal numbers of rectangular cylinders. J Fluid
Mech 1982;123:379–98.
[19] Luo SC, Yazdani MG, Chew YT, Lee TS. Effects of incidence and
afterbody shape on flow past bluff cylinders. J Wind Engng Ind
Aero 1994;53:375–99.
[20] Vickery BJ. Fluctuating lift and drag on a long cylinder of square
cross-section in a turbulent stream. J Fluid Mech 1966;25:481–94.
[21] Norberg C. Flow around rectangular cylinders: pressure forces
and wake frequencies. J Wind Engng Ind Aero 1993;49:187–96.
[22] Lee BE. The effect of turbulence on the surface pressure field of a
square prism. J Fluid Mech 1975;69:263–82.
[23] Tamura T, Miyagi T. The effect of turbulence on aerodynamic
forces on a square cylinder with various corner shapes. J Wind
Engng Ind Aero 1999;83:135–45.
[24] Saha AK, Muralidhar K, Biswas G. Experimental study of flow
past a square cylinder at high Reynolds numbers. Exper Fluids
2000;29:553–63.
[25] McLean I, Gartshore C. Spanwise correlations of pressure on a
rigid square section cylinder. J Wind Engng Ind Aero 1992;41–
44:779–808.
[26] Saathoff P, Melbourne WH. Effects of free stream turbulence on
streamwise pressure measured on a square-section cylinder.
J Wind Engng Ind Aero 1999;79:61–78.
[27] Nakamura Y, Ohya Y. The effect of turbulence on the mean flow
past 2D rectangular cylinders. J Fluid Mech 1984;149:255–73.
[28] White FM. Fluid mechanics. 4th ed. McGraw-Hill; 1999.
[29] Jeong J, Hussain F. On the identification of a vortex. J Fluid
Mech 1995;285:69–94.
[30] Hunt JCR, Wray A, Moin P. Eddies, stream, and convergence
zones in turbulent flows. Center of Turbulence Research Report
CTR-S88, 1988. p. 193.
[31] Chyu CK, Rockwell D. Evolution of patterns of stream-wise
vorticity in the turbulent near wake of a circular cylinder. J Fluid
Mech 1996;320:117–37.
[32] Piomelli U, Chasnov JR. Large eddy simulation: theory and
applications. In: Hallback M, Henningson DS, Johansson AV,
Alfredsson PM, editors. Turbulence and transition modelling.
Ercoftac Series. Kluwer Academic Publishers; 1996.