DNS/LES Simulations of Separated Flows at High Reynolds

Numbers
P. Balakumar
Flow Physics and Control Branch, NASA LaRC, MS 170, Hampton, VA 23681-2199
Email: ponnampalam.balakumar-1@nasa.gov, (757) 864-8453

Abstract
Direct numerical simulations (DNS) and large-eddy simulations (LES) simulations of flow through a
periodic channel with a constriction are performed using the dynamic Smagorinsky model at two
Reynolds numbers of 2800 and 10595. The LES equations are solved using higher order compact
schemes. DNS are performed for the lower Reynolds number case using a fine grid and the data are
used to validate the LES results obtained with a coarse and a medium size grid. LES simulations are
also performed for the higher Reynolds number case using a coarse and a medium size grid. The
results are compared with an existing reference data set. The DNS and LES results agreed well with
the reference data. Reynolds stresses, sub-grid eddy viscosity, and the budgets for the turbulent kinetic
energy are also presented. It is found that the turbulent fluctuations in the normal and spanwise
directions have the same magnitude. The turbulent kinetic energy budget shows that the production
peaks near the separation point region and the production to dissipation ratio is very high on the order
of five in this region. It is also observed that the production is balanced by the advection, diffusion, and
dissipation in the shear layer region. The dominant term is the turbulent diffusion that is about two
times the molecular dissipation.

I. Introduction

T
he major objectives of any CFD code are to predict global aerodynamic quantities such as lift and drag as well
as to predict local flow features such as skin friction, heat transfer, and pressure oscillations, accurately and
efficiently for a wide range of problems. The flow physics are fundamentally governed by the unsteady three-
dimensional Navier-Stokes (N-S) equations that state the three conservation laws for mass, momentum, and energy.
Turbulent flow is characterized by the existence of a vast range of length and time scales ranging from the smallest,
the Kolmogorov scale, to the largest, determined by the geometry.1,2 The required computer resources and time
constraints hinder the solution of these equations numerically for turbulent flows at high Reynolds numbers.
Analysis2 has shown that the number of grid points required increases as Re9/4 and the number of time steps required
increases as Re3/4. These challenging requirements restrict the solution of the full N-S equations to simple
geometries such as channels and flat plates at low Reynolds numbers. The first successful direct numerical
simulation3 (DNS) was performed for a channel flow at a Reynolds number of 3000. With increasing computational
capability, the application of DNS to more complex problems at high Reynolds numbers is currently being pursued
by many researchers. The recent status of DNS in turbulent flow is reviewed by Moin and Mahesh.4 However, DNS
usage is limited to understanding the turbulent physics in canonical problems and to extending the findings for
applications to turbulence model development. Hence, the current state of the art is to solve some approximate
versions of the N-S equations that require modeling.
The simplest and the most popular approximate method is the set of Reynolds-Averaged Navier-Stokes (RANS)
equations, where the equations are derived for time-averaged quantities. The unclosed Reynolds stresses are
modeled to close the equations. Wilcox5 gives a good account of different models and their applications. In general,
existing models provide good results compared to experiments in attached flows where turbulence is in equilibrium.
However, prediction of separated flows with existing RANS turbulence models is still not very satisfactory. It is
difficult to accurately capture both the separation and reattachment points (i.e., the size of the separation bubble). In
particular, it is believed that the flow field dynamics near the separation point are not predicted correctly. The RANS
computations performed for the flow over a two-dimensional hump, NASA Langley CFDVAL2004 Case 36,7,
consistently over-predicted the reattachment point compared to the experiment. The computed Reynolds stresses
were 2 to 3 times smaller than the measured values. Computations performed8 for two other massively-separated
flow cases (the 2-D periodic hill9 and the 3-D Ahmed body10) using different turbulence models also predicted too-
long separation bubbles. These computations further exemplify the deficiencies in predicting massively-separated
flows using existing RANS models.

  1  
 The second approximate method uses the Large-Eddy-Simulation (LES) equations, which are obtained for the
large-scale motions by filtering small scales. The unknown stresses introduced by the small scales (subgrid-scale
stresses, or SGS) are modeled to close the equations. Sagaut11 and Garnier et al.12 give account of different SGS
models and their applications in incompressible and compressible flows, respectively. This approach is
computationally more expensive than the RANS approach, but yields more accurate results without many ad-hoc
modifications to the SGS models. This methodology is being applied to a wide range of problems. Recently, several
DNS and LES simulations have been performed for separated flows over two- and three-dimensional humps in
channel flows.13-20
Flow through a periodic channel with a constriction (separated flow over a 2-D hill, Fig. 1) is widely used as a
test case to investigate the flow physics of turbulent separated flows over smooth surfaces and to validate different
RANS and LES solutions obtained on coarser grids with highly resolved DNS and LES data sets. Breuer et al.19
gave a brief history behind the selection of this geometry as a benchmark test case. They also have presented highly
resolved LES simulation results for a range of Reynolds numbers 700 to 10595. Frohlich et al.15 also performed a
highly resolved LES at a Reynolds number of 10595. The results include mean flow profiles, Reynolds stress
profiles, energy balances, and the instantaneous flow fields. Ziefle et al.18 performed LES simulations with a very
coarse grid using the approximate-deconvolution SGS model for compressible flow at a Reynolds number of 2800.
It was found that the grid resolution near the separation point is critical in capturing the reattachment point correctly
in the simulations. Even though this geometry is attractive for numerical simulations because of the simple geometry
and periodic boundary conditions in the streamwise direction, the flow becomes very unsteady and complex due to
the varying inflow conditions at the crest that is continuously influenced by the previous hill. In the previous
simulations, it took about 50-60 flow through times to obtain reasonable statistical averages. Another observation
from the LES and DNS was that the ratio of production over dissipation takes very high values in the separated
shear region compared to homogeneous flows. Whether this is one of the causes for the deficiency in RANS is not
clear. The turbulent dynamics between the separation point and the reattachment point is very complex.21 The
adverse pressure gradient near the reattachment point and spreading of the shear layer due to turbulence are
interconnected and the balance between these two factors determines the separation and the reattachment points.
The objective of the present research is to first validate our DNS/LES results for the periodic channel flow with a
constriction with the reference data of Breuer et al.17,19 Another objective is to investigate the turbulent dynamics of
this flow by analyzing the Reynolds stresses and the budget of different terms in the turbulent kinetic energy
equation. We also want to investigate the effects of the SGS model on the turbulent statistical quantities by
performing the LES with and without the model on the same grid. We solve the three-dimensional unsteady
compressible Navier-Stokes equations using higher-order compact schemes for space discretization and 3rd order
Runge-Kutta scheme for time integration. We employ the dynamic Smagorinsky SGS model in the LES
calculations. We first present the DNS results obtained on a fine grid at a low Reynolds number based on the bulk
velocity (Ub), density (ρb), the hill height (h), and molecular viscosity at the wall (µw) of Reb = ρb Ub h/µw = 2800,
and then present the wall-resolved LES (WRLES) results at low to high Reynolds numbers of Reb = 2800 and 10595.

I. Models and Flow Conditions

Computations were performed for a flow through a channel with a constriction (2D-hill). The geometry and the
coordinate systems are shown in Figs. 1(a) and 1(b). The flow was periodic in the streamwise and spanwise
directions. The lengths are non-dimensionalized by the hill height, h. The computational dimension in the
streamwise direction is Lx = 9, in the spanwise direction is Lz= 4.5, and in the normal direction above the crest of the
hill is Ly= 3.035. These are the same dimensions used in the DNS simulations15,19 and in the LES simulations.18
Simulations are performed for Reynolds numbers based on the bulk velocity, bulk density, and the hill height of Reb
= ρb Ub h/µw = 2800 and 10595. The wall temperature is fixed at a constant temperature of 300K. The Mach number
based on the bulk velocity is M0 = 0.2. The non-dimensional bulk density and the bulk velocity are defined as

1 "1 %
∫ ρ dv = 1, $ ∫ u dA' = 1 (1)
Vol vol # A area &x=0

  2  
 (a)

y, v
Ly
x, u
Lz
z, w
h=1
Lx

Figure 1. Periodic channel with a constriction and the coordinate systems used.  

II. Governing Equations

The partial differential equations solved are the filtered three-dimensional unsteady compressible Navier-
Stokes equations in conservation form22
 
Continuity:
∂ρ ∂ ( ρ u j )
+ =0
∂t ∂x j
(2)
Momentum:
∂ ( ρ ui ) ∂ ( ρ ui u j ) ∂p ∂ (σˆ ij + τ ij )
+ + + fiδi1 = + Σij
∂t ∂x j ∂xi ∂x j
(3 )

µ (T ) $" ∂ ui ∂ u j 2 ∂ uk $&

σˆ ij = # + − δij '. (4 )
Re $%∂ x j ∂ xi 3 ∂ xk $(

Energy:

( ) + ∂ ( ρ Ê + p ) u
∂ ρ Ê j
+ f1u1 =
∂ (σˆ ij ui ) ∂q̂ j
− − B1 − B2 − B3 + B4 + B5 + B6 + B7
∂t ∂x j ∂x j ∂x j
(5 )

µ (T ) ∂T
q̂ j = −
(γ −1) Re Pr M 20 ∂x j (6)

p = ρT (7)

Here the over-bar “-” denotes filtered variables and the tilde “~” denotes the Favre-averaged variables. The subgrid-
scale (SGS) terms appearing in the momentum equations are

  3  
 ∂(σ ij − σˆ ij )
Σij = ,
∂x j
(8)

(
τ ij = −ρ ui u j − ui u j . ) (9)

Here τij is the subgrid-scale stress tensor. The subgrid-scale terms appearing in the energy equations are
1 ∂
B1 =
(γ −1) ∂x j
(
pu j − pu j , )
(10)
∂uk ∂u
B2 = p −p k,
∂xk ∂xk (11)
∂
B3 =
∂x j
(τ kj uk ),
(12)
∂
B4 = τ kj uk ,
∂x j
(13)
∂ ∂
B5 = σ kj uk − σ kj uk ,
∂x j ∂x j
(14)
∂
B6 =
∂x j
(σ ij ui − σˆ ij ui ),
(15)
∂
B7 =
∂x j
(q j − q̂ j ).
(16)

The subgrid-scale viscous term Σij is orders of magnitude smaller than the subgrid-scale stress term and is neglected.
The deviatoric part of the SGS stress tensor is modeled using the Smagorinsky eddy viscosity model23

1

τ ij − τ kkδij = τ ijD = ρν t Sij (u)
3 (17)
ν t = (Cs Δ)2 S(u)  ,
2 1
 = Sij (u)S
S(u)  ij (u).

2 (18)

The value of the parameter Cs varies from 0.18 for isotropic turbulence to 0.10 for plane channel flow.24, 25 The
characteristic length scale Δ is determined by the local grid size.

1/3
Δ = ( Δx Δy Δz )
(19)

The model is very dissipative in transitional flows and does not vanish near the wall and in laminar shear flows.
One has to use Germano’s26 dynamic procedure or a simple Van Driest damping to remedy these issues. We
followed Vreman’s27 procedure in the derivation of the dynamic Smagorinsky model (DSM). In the dynamic eddy-
viscosity model, the deviotoric stress is written as a linear function of a variable coefficient Cd.

  4  
  Sij (u).
τ ijD = Cd Δ 2 S(u) 
(20)

The coefficient Cd is determined dynamically based on Germano’s identity, which relates the flow fields at the
filtered level and at a new test-filter level. Applying the Germano identity to the subgrid-scale stress term yields the
following system of equations for the coefficient Cd.

Cd M ij = Lij ,
(21)

where

Lij = ( ρui ρu j / ρ )ˆ − ( ρui )ˆ ( ρu j )ˆ / ρ

(22)

∧
 Sij (u)
M ij = −ρ̂ (κΔ 2 ) S(v) Sij (v) + ( ρΔ 2 S(u) ) ,
(23)

∧
( )
vi = ρui / ρ̂.
(24)

The equation (20) yields six equations for one unknown coefficient Cd. The model coefficient is determined using a
least square method

M ij Lij
Cd = .
M ij M ij
(25)

In the present simulations, we used a simple top-hat filter with the filter-width two-times the grid size. This is
equivalent of using κ=√5.

A. Solution Algorithm
The governing equations were solved using higher order compact schemes.28,29 Flow simulations were performed
using an 8th order compact scheme in the streamwise and spanwise directions and a 6th order compact scheme in the
wall normal direction. The orders are reduced to 4th and 3rd orders near the walls. We used higher order implicit
filtering30 of 10th and 8th orders in the homogeneous directions and in the normal directions, respectively to remove
the high wavenumber contents in the solution. We employed a 3rd-order total-variation-diminishing (TVD) Runge-
Kutta scheme for time integration. We used a body-fitted curvilinear grid system in all of the simulations. The
equations are transformed from the physical coordinate system (x, y, z) to the computational curvilinear coordinate
system (ξ, η, ζ). The grids were uniform in the streamwise and spanwise directions and were stretched in the normal
directions close to the walls. The flow is maintained by applying a body force f1(t) in the streamwise direction. This
is a function of time only and is determined at every step of the time marching by requiring that the average mass-
flux remains constant, Eq. (26). We adapted the procedure described in Ziefle et al.18 to determine the body force
f1(t) at every time step.
∂
∫ ρu dv = 0
∂t vol
(26)

III. Results
A. DNS
We first present the DNS results at a low Reynolds number of Reb = 2800. We used (513, 257, 289) points in the
streamwise, wall normal, and spanwise directions, respectively. The grid spacing in viscous units change along the
streamwise direction due to the change in friction velocity along the wall. Figure 2(a) shows the variations of the
grid spacing in wall units in the x, y, and z directions along the bottom wall and the grid spacing in the y direction in

  5  
the midway points between the bottom and top walls of the channel. The maximum grid spacing in the x- and z-
directions are below 3 in the region up to x ~ 7.5 and increase to Δx+ = 7.5 and Δz+ = 6.7 above the downstream hill
due to a large increase in the friction velocity in this region. The minimum grid spacing at the wall varies between
Δy+ = 0.1 to 0.5 and the maximum spacing in the middle of the channel is Δy+ = 3.3. Figure 2(b) shows the variation
of the ratio of grid spacing in each direction to the Kolmogorov length scale, η, along the normal direction at x =
1.0. The maximum grid size is about 2.7 times the Kolmogorov scale. Hence these grid distributions are sufficient to
resolve the flow field up to the dissipation scales.
(a) (b)
8 3.0
Δx+ Δx/η
+ Δy/η
Δy 2.5
Grid spacing

+ Δz/η
6 Δz
Δy+m 2.0
Δ/η
y
4 1.5

1.0
2
0.5

0 0.0
0 2 4 X 6 8 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Δx/η, Δy/η, Δz/η, Δ/η
Figure 2. (a) Grid spacing in wall units along the bottom wall, and (b) ratio of grid spacing to Kolmogorov scale along the
normal direction at x = 1.0. Reb = 2800.

Figure 3 displays the instantaneous u-velocity field in the (x-y) plane at the spanwise location z = 0. The blue
color denotes the reverse flow. The flow field can be divided into two regions. The first is the region below the line
connecting the two crests of the hill and the second is the region above this line and below the top wall. The flow
field in the second region appears as a plug flow through a two-dimensional channel. The first region is dominated
by the reverse flow extending from one crest to the other. It is also seen that patches of attached flow exist inside
and between the reverse flow regions.

Figure 3. Instantaneous u-velocity field in the (x-y) plane at z = 0.

Figure 4(a) displays the instantaneous u-velocity field in the cross sectional (y-z) plane across the separation
bubble at the streamwise location x = 2.0. It is seen that the distorted shear layer is confined between the upper plug
flow and the lower recirculation flow. Figure 4(b) shows the instantaneous u-velocity in the plan view along a fixed
normal grid (j=5, y+ ~ 2 ) plane ((x-z) plane) which is located a short distance above the lower wall. The white
region displays the region with the negative velocity. The figure shows that except near the windward side of the
downstream hill the streamwise velocity is negative in most of the domain. This phenomenon is common in
separated flows14. The instantaneous separation regions extend to larger domains compared to those derived from
the time averaged flow field. Figure 5(a) shows the variation of the instantaneous streamwise, u, and spanwise, w,
velocities in the x-direction along a fixed normal grid line (j=5) at a station z = 0. Similarly, Fig. 5(b) shows the
velocities in the z-direction at the station x = 2.0 and y = 1.0. This station is located in the middle of the shear layer.

  6  
 Figure 4. u-velocity contours (a) in the cross sectional (y-z) plane at x = 2.0, and (b) in the plan view (x-z) plane at a fixed
normal grid j=5.
(a)
0.4

U
0.2 W

0.0
U, W

-0.2

-0.4

0 2 4 6 8
X

(b)

1.0

0.5
U, W

0.0

-0.5 U
W

-1.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
Z

Figure 5. Variations of the instantaneous u, w velocity fields (a) in the x direction along a fixed grid line (j=5) at the station
z=0, and (b) in the z direction at x = 2.0 and y=1.0.

The statistical quantities are obtained by averaging the solution in the spanwise direction and in time. The
averaging in time was performed for about 40 flows through periods. Figure 6 shows the mean streamwise velocity
contours and the streamlines from the current DNS. We observe backflow and massive separation behind the first
hill, as expected. Figures 7(a) and 7(b) show the skin friction cf and the pressure coefficient cp along the lower and
the upper surfaces. We also include the DNS results17,19 obtained for the incompressible flow. In the current
compressible computations at M0 = 0.2, the flow separates at xsep = 0.23 and reattaches at xreatt = 5.45. The separation
point location is close to the DNS19 and the LES simulations18 predicted value of 0.21. The current predicted
reattachment point location is close to that predicted by the DNS19 and the LES18 simulations, where the
reattachment occurred at xreatt=5.4 and 5.3, respectively. The agreement of the skin friction coefficient and the
pressure coefficient from the present simulations with the DNS19 are quite good. The skin friction coefficient along
the upper surface is higher compared to the incompressible DNS19 results. The lower surface skin friction coefficient
becomes negative after the separation and reaches a minimum value near the middle of the separation bubble.
Beyond this minimum point, it slowly increases and takes a dip at the foot of the hill. The skin friction increases to
larger values over the windward part of the hill and decreases at the crest. The pressure first increases and remains
flat up to x ~ 2, and increases strongly up to the foot of the windward part of the hill due to the strong deceleration of
the flow, x ~ 7.5, before it decreases to the inflow value.

  7  
 Figure 6. Hill flow contours of the mean streamwise velocity and the streamlines, Reb=2800.
(a) Skin friction (b) Pressure
0.06 0.6
Lower Present DNS
Lower Presnt DNS Upper " "
0.05 Upper " " 0.4 Lower DNS Breuer et al. (2009)
Lower DNS Breuer et al. (2009) Upper " "
Upper " "
0.04 0.2
cf

0.03 0

0.02 cp -0.2

0.01 -0.4

0 -0.6

-0.01 -0.8
0 1 2 3 4 5 6 7 8 9 0 2 4 X 6 8
X

Figure 7. Hill flow variation of mean (a) skin friction coefficient, and (b) pressure coefficient along the lower and upper
walls. Reb = 2800.

In Figs. 8-10, the computed mean velocity <U> and the Favre-averaged Reynolds stresses <ρuʺ″uʺ″> and <ρuʺ″vʺ″>
profiles are compared with the results of Breuer et al.19 at five stations x = 0, 1, 2, 4, and 8. Mean velocity profiles
and Reynolds stress profiles show excellent agreement with the DNS results of Breuer et al.19. This comparison
validates the accuracy of the present computations. Figures 11(a-f) show all the Reynolds stress components profiles
at six stations x = 0, 1, 2, 4, 5 and 8. We also plotted the mean velocity profiles in these figures. The first station x =
0 is located at the hill crest and the flow separates slightly downstream of this point. The stations x = 1, 2, 4 and 5
are situated inside the separation zone and the last station x = 8 is located on the windward side of the downstream
hill. The first observation is that the Reynolds stresses in the separated region are confined to the separated shear
layer region and they become comparatively small near the wall. The maximum values for the normal Reynolds
stresses in the x-, y-, and z-directions at the station x = 1 are <ρuʺ″uʺ″> =0.09, <ρvʺ″vʺ″> =0.04, <ρwʺ″wʺ″> =0.05 and the
maximum Reynolds shear stress is <ρuʺ″vʺ″> =0.03. These maximum values occur near the middle of the shear layer
close to y ~ 1.0. The ratios of the three normal stresses are 1: 1/2.25 :1/1.8. This shows that the normal stresses in the
spanwise and the normal directions have almost the same magnitude. These ratios in a plane channel flow31 are 1:
1/9 :1/6.25. This implies that the turbulence in separated shear layers is more isotropic than in wall bounded flows. It
is also seen that when the flow approaches the reattachment point, the normal stresses in the streamwise and
spanwise directions become almost the same near the wall. This suggests that the flow near the reattachment region
behaves similarly to a stagnation flow impinging against a wall at an angle. The profiles near the downstream hill
reveal that the maximum fluctuations occur in the spanwise direction compared to x- and y-directions. This was also
observed in the simulations of Frohlich et al.15 at a high Reynolds number of Reb = 10595. This is due to the
impinging of downstream moving eddies against the windward side of the downstream hill. This also generates
substantial Reynolds stresses in the outer region above the crest as evident in Figs. 11 (a) and (f).

  8  
 (a) x=0 (b) x=1 (c) x=2 (d) x=4 (e) x=8
3 3 3 3 3

2 2 2 2 2

y y y y y

1 1 1 1 1

Breuer 2009
Present DNS
0 0 0 0 0
0 0.5 1 0 0.5 1 0 Um 0.5 1 0 Um 0.5 1 0 Um 0.5 1
Um Um

Figure 8. Hill flow mean streamwise velocity profiles at different stations. Reb = 2800.
(a) x=0 (b) x=1 (c) x=2 (d) x=4 (e) x=8
3 3 3 3 3

2 2 2 2 2

y y y y y

1 1 1 1 1

0 0 0 0 0
0 0.05 0.1 0 0.05 0.1 0 0.05 0.1 0 0.05 0.1 0 0.05 0.1
(ρu" u" ) (ρu" u" ) (ρu" u" ) (ρu" u" ) (ρu" u" )
Figure 9. Hill flow streamwise normal Reynolds stress profiles at different stations. Reb = 2800.

(a) x=0 (b) x=1 (c) x=2 (d) x=4 (e) x=8
3 3 3 3 3

2 2 2 2 2

y y y y y

1 1 1 1 1

Breuer 2009
Present DNS
0 0 0 0 0
-0.02 0 -0.02 0 -0.02 0 -0.02 0 -0.02 0
(ρu" v" ) (ρu" v" ) (ρu" v" ) (ρu" v" ) (ρu" v" )
Figure 10. Hill flow Reynolds shear stress velocity profiles at different stations. Reb = 2800.

  9  
 (a) x=0 (b) x=1 (c) x=2
3.0 3.0 3.0

2.5 2.5 2.5

2.0 2.0 2.0

y y y
1.5 1.5 1.5

1.0 1.0 1.0

<u'u'>
<v'v'>
0.5 <w'w'> 0.5 0.5
<u'v'>
<U>/10
0.0 0.0 0.0
0.00 0.05 0.10 0.00 0.05 0.10 0.00 0.05 0.10

(d) x=4 (e) x=5 (f) x=8

3.0 3.0 3.0

2.5 2.5 2.5

2.0 2.0 2.0

y y y
1.5 1.5 1.5

1.0 1.0 1.0

0.5 0.5 0.5

0.0 0.0 0.0

0.00 0.05 0.10 0.00 0.05 0.10 0.00 0.05 0.10
Figure 11. Normal and shear Reynolds stress components profiles at different stations. Reb = 2800.

The transport equations for the Reynolds stresses and the turbulent kinetic energy are given in Ref. 32. The
equation for the turbulent kinetic energy can be written as

1
K = ρ ui!!ui!!
2
∂K
= Σ − D −ε − C
∂t (27)

The left hand side of Eq. (27) is the local rate of change of turbulent kinetic energy and the terms on the right
hand side are: (1) production, (2) diffusion, (3) dissipation, and (4) advection. In the statistical steady state
conditions, the local rate of change should be zero and these four terms should balance each other to yield zero sum.
A non-zero left hand side points to lack of grid resolutions in the simulations. This non-zero value is termed as
numerical dissipation and in turbulent simulations this balance is added to the physical dissipation and considered as
the total physical dissipation in the analysis. Figures 12(a-e) depict the magnitudes of the four terms on the right
hand side of this equation and the balance or the total of these four terms at five stations x = 0, 1, 2, 4 and 8. It is
seen that the balance is below 3% of the dissipation in the whole domain. This confirms that the grid used in this
DNS resolved all the scales including the dissipation scales at this low Reynolds number. The first observation is
that turbulent production is the maximum at all the stations. The production peaks near the separation point x = 0
and 1 to a value of 0.10 and decreases to 0.05, 0.03 and 0.02 at the stations x = 2, 4 and 8, respectively. It is also
observed that these peaks occur in the middle of the shear layer. Let us consider the balance of the four terms at the
station x = 1, which is downstream of the separation point x = 0.23. The production peaks at a value of 0.10 near y ~
1. This is balanced in the negative side by the advection, diffusion and dissipation terms. The magnitudes of these
terms are 0.01, 0.06 and 0.03, respectively. Hence diffusion is about two times larger than the dissipation term.
When we move towards the lower wall, all the terms first decrease to zero and at the wall dissipation again
increases. The dissipation at the wall is balanced by the positive diffusion term. This description remains the same
up to the separation point except that the diffusion term decreases faster than the dissipation term.

  10  
 (a) x = 0 (b) x = 1 (c) x = 2
3.00 3.00 3.00

Advec Advec Advec

Prod Prod Prod
2.50 2.50 2.50
Diffu Diffu Diffu
Dissip Dissip Dissip
Total Total Total
2.00 2.00 2.00

Y Y Y
1.50 1.50 1.50

1.00 1.00 1.00

0.50 0.50 0.50

0.00 0.00 0.00

-0.05 0.00 0.05 0.10 -0.05 0.00 0.05 0.10 -0.05 0.00 0.05 0.10

(d) x = 4 (e) x = 8
3.00 3.00

Advec Advec
Prod Prod
2.50 2.50
Diffu Diffu
Dissip Dissip
Total Total
2.00 2.00

Y Y
1.50 1.50

1.00 1.00

0.50 0.50

0.00 0.00
-0.05 0.00 0.05 0.10 -0.05 0.00 0.05 0.10

Figure 12. Balance of different terms in the turbulent kinetic equation. Reb = 2800, DNS.

We also plotted the contours of the average statistical quantities: (a) production, (b) dissipation, (c) ratio of
production to dissipation, (d) kinetic energy, and (e) shear stress in Figs. 13(a-e). These quantities play important
roles in RANS modeling and development. The production is maximum near the start of the separated shear layer as
observed earlier and is confined to a region along the separation line. The maximum production occurs near x = 0.6
and y = 1.06, which is downstream of the separation point and slightly above the hill crest. Beyond the reattachment
point, the production is concentrated in the middle of the channel at a height of y ~ 1. There is negative production
near the reattachment region and along the shoulder of the hill. It is also noted that there exists a high production
region above the downstream hill. The maximum dissipation occurs along the wall on the windward side of the hill.
Outside the wall region, dissipation is confined to the shear layer and to its proximity. The ratio of production to
dissipation is shown in Fig. 13(c). This ratio takes a peak value of 4.5 near the start of the separated shear layer at x
= 0.6 and y = 1.06. It decreases to a value around 2 near x ~ 2 and y ~ 1. Recall that this ratio is about 1.8 in the
buffer region of a turbulent channel flow. It is also observed that this ratio remains at a constant value of about 1.5
up to the second hill along the line y ~ 1. Figure 13(d) shows the turbulent kinetic energy distribution in the
separated turbulent boundary layer. The kinetic energy is concentrated along the shear layer and it peaks near x ~ 1.4
and y ~ 1.06. After the reattachment point the energy is confined to region near y ~ 1. This agrees with the earlier
observations from the turbulent production contours. After the reattachment point, turbulence is confined to a region
away from and parallel to the wall at a height of y ~ 1. The turbulent shear stress contours in Fig. 13(e) also convey
the same picture that turbulent shear stress is concentrated near the start of the shear layer and near the y ~ 1 region
at x ~ 1.4.

  11  
 Figure 13. Hill flow contours of the turbulent statistical quantities (a) Production, (b) Dissipation, (c)
Production/Dissipation, (d) Kinetic energy, and (e) Shear stress. Reb = 2800.

  12  
B. LES
We have implemented the dynamic Smagorinsky model (DSM) into the existing code and performed the LES
simulations for the two Reynolds numbers Reb = 2800 and 10595 using the higher order compact scheme. The grids
that were used in the DNS and the LES are given in the Table 1.

Table 1. Grid sizes used in the DNS and LES simulations

Re DNS LES
2800 513*257*289 Grid 1 129*129*73
Grid 2 257*257*145
10595 Grid 1 193*193*145
Grid 2 257*257*217

(1) Case 1: Reb = 2800

Figure 14 shows the variations of the grid spacing in wall units for the two grids, Grid 1 and Grid 2, in the x, y,
and z directions along the bottom wall and the grid spacing in the y direction in the midway points between the
bottom and top walls of the channel. The maximum grid spacing in the x- and z-directions for Grid 1 are below 12 in
the region up to x ~ 7.5 and increase to Δx+ = 30 and Δz+ = 25 above the downstream hill due to large increases in the
friction velocity in this region. The minimum grid spacing at the wall varies between Δy+ = 0.70 to 2.0 and the
maximum spacing in the middle of the channel is Δy+ = 6. Similarly, the maximum spacing in the x- and z-directions
for Grid 2 are below 6 in the region up to x ~ 7.5. The minimum grid spacing at the wall varies between Δy+ = 0.1 to
0.9. These values are finer than the recommended values32 of Δy+ < 2, Δx+ ~ 50-150 and Δz+~15-40 for wall
resolved LES. Especially, the values for Grid 2 are much finer than the recommended values.

30

Δx+ Grid 1
25 Δy
+

+
Δz
Grid spacing

20 Δy+m
Grid 2
15

10

0
0 2 4 X 6 8

Figure 14. Grid spacing in wall units along the bottom wall.

Figures 15(a-d) show the contours and the streamlines obtained from (a) DNS, (b) LES without the SGS model
(ILES(1)) using Grid 1, (c) LES with the dynamic Smagorinsky model (LES(1)) using Grid 1, and (d) LES with the
dynamic Smagorinsky model (LES(2)) using Grid 2. We did not include the results obtained without the SGS model
using Grid 2 because the results are almost the same as LES(2). Qualitatively, the large scale features such as
separation point, reattachment point, bubble size and shape, and the velocity contours look almost the same in all
four figures. The bubble looks slightly smaller and the reattachment occurs earlier in the results obtained with
ILES(1). Figure 16 depicts variation of the skin friction at the lower wall. The reattachment point is located at xreatt =
5.4, 5.0, 5.2 and 5.4 from the four simulations DNS, ILES(1), LES(1) and LES(2), respectively. The corresponding
separation points are located close to x ~ 0.23, 0.23, 0.24 and 0.25, respectively. The implicit LES simulation using
the coarser grid (ILES(1)) predicts an early reattachment point compared to other cases. This is due to the fact that at
this grid level the dissipation scales are not resolved and the energy dissipated by the numerical scheme is not
sufficient to keep the turbulent kinetic energy at the correct level. Hence, there exist large turbulent fluctuations in
the shear layer and this strong mixing causes the shear layer to reattach earlier. In these simulations, we employed
uniform grid distributions in the x- and z-directions. In earlier simulations,17, 18, 19 a stretched grid was used in the x-

  13  
direction. This makes the grid finer near the separation point and coarser in the middle part of the channel. Whether
the stretched grid distribution causes changes in separation and reattachment points will be evaluated in the future.
Figure 17 shows the variation of the sub-grid eddy viscosity at different streamwise stations x = 0.0, 1.0, 2.0, 4.0,
8.0. As expected, the eddy viscosity is large in the separated shear layer region. However, the computed maximum
eddy viscosity is only 50% of the molecular viscosity for the Grid 1 and is about 30% for the Grid 2. This may be
the reason that at this low Reynolds number the SGS model did not affect the dynamics of the turbulence.

Figure 15. Hill flow contours of the mean streamwise velocity and the streamlines, Reb=2800. (a) DNS, (b) ILES (Grid 1),
(c) LES with DSM (Grid 1), and (d) LES with DSM (Grid 2).

  14  
 0.06
ILES
0.05 LES Dynamic Grid1
LES Dynamic Grid 2
Present DNS
0.04

cf
0.03

0.02

0.01

-0.01
0 1 2 3 4 5 6 7 8 9
X

Figure 16. Hill flow variation of mean skin friction coefficient along the lower wall. Reb = 2800.

3.0
Re = 2800
X

2.5 0.0 Grid 1

1.0
2.0
4.0
2.0 8.0
0.0 Grid 2
Y 1.0
1.5 2.0
4.0
8.0

1.0

0.5

0.0
0.0 0.2 0.4 vt / v 0.6 0.8 1.0

Figure 17. Variation of the sub-grid eddy viscosity at different stations.

Figures (18-21) depict the comparison of the mean and turbulent quantities obtained from the four simulations at
different streamwise locations x=0, 1, 2, 4. The mean streamwise and normal velocities obtained from the ILES and
LES simulations are in excellent agreement with the DNS results at x = 0, 1 and 2. The velocities at x = 4 are slightly
smaller than the DNS results near the wall region. The Reynolds stress components also show good agreement with
the DNS results. The maximum normal Reynolds stress <ρuʺ″uʺ″> at x = 4 obtained with LES(1) is under predicted
by about 5% compared to the DNS results.

  15  
 (a) (b)

3.0 3.0
Re=2800 Re=2800

0.0 ILES (129 * 129 * 73) 2.0 ILES

2.5 1.0 2.5 4.0
0.0 LES Dynamic (129 * 129 * 73) 2.0 LES Dynamic Grid 1
1.0 4.0
0.0 LES Dynamic (257 * 257 * 145) 2.0 LES Dynamic Grid 2
2.0 1.0 2.0 4.0
0.0 DNS (513 * 257 * 289) 2.0 DNS
Y 1.0 Y 4.0

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2
<U> <U>

   
Figure 18. Comparison of mean streamwise velocity <U> profiles obtained from the DNS, ILES and LES simulations at
the stations (a) x = 0 and 1, and (b) x = 2 and 4.

(a) (b)

3.0 3.0
Re=2800 Re=2800
0.0 ILES (129 * 129 * 73)
1.0 2.0 ILES
4.0
2.5 0.0 LES Dynamic (129 * 129 * 73) 2.5 2.0 LES Dynamic Grid 1
1.0
0.0 LES Dynamic (257 * 257 * 145) 4.0
1.0 2.0 LES Dynamic Grid 2
0.0 DNS (513 * 257 * 289) 4.0
1.0 2.0 DNS
2.0 2.0 4.0

Y Y

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
-0.10 -0.05 0.00 0.05 0.10 0.15 0.20 0.25 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20 0.25
<V> <V>

   
Figure 19. Comparison of mean vertical velocity <V> profiles obtained from the DNS, ILES and LES simulations at the
stations (a) x = 0 and 1, and (b) x = 2 and 4.

  16  
 (a) (b)

3.0 3.0
Re=2800 Re=2800

0.0 ILES (129 * 129 * 73) 2.0 ILES

2.5 1.0 2.5 4.0
0.0 LES Dynamic (129 * 129 * 73) 2.0 LES Dynamic Grid 1
1.0 4.0
0.0 LES Dynamic (257 * 257 * 145) 2.0 LES Dynamic Grid 2
2.0 1.0 2.0 4.0
0.0 DNS (513 * 257 * 289) 2.0 DNS
Y 1.0 Y 4.0

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10
<uu> <uu>

   
Figure 20. Comparison of turbulent intensity < ρuʺ″uʺ″ > profiles obtained from the DNS, ILES and LES simulations at the
stations (a) x = 0, and (b) x = 4.

(a) (b)

3.0 3.0
Re=2800 Re=2800
0.0 ILES (129 * 129 * 73)
1.0 2.0 ILES
2.5 0.0 LES Dynamic (129 * 129 * 73) 2.5 4.0
1.0
0.0 LES Dynamic (257 * 257 * 145) 2.0 LES Dynamic Grid 1
1.0 4.0
0.0 DNS (513 * 257 * 289) 2.0 LES Dynamic Grid 2
2.0 1.0 2.0 4.0
2.0 DNS
Y Y 4.0

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
-0.04 -0.03 -0.02 -0.01 0.00 0.01 0.02 -0.04 -0.03 -0.02 -0.01 <uv> 0.00 0.01 0.02
<uv>

   
Figure 21. Comparison of turbulent shear stress < ρuʺ″vʺ″ > profiles obtained from the DNS, ILES and LES simulations at
the stations (a) x = 0, and (b) x = 4.

(2) Case 2: Reb = 10595

Figures (22-33) similarly show the LES results at the higher Reynolds number of 10595. Figure 22 shows the
variations of the grid spacing in wall units for the two grids, Grid 1 and Grid 2. The maximum grid spacing in the x-
and z-directions for Grid 1 are below 20 in the region up to x ~ 7.5 and increase to Δx+ = 55 and Δz+ = 36 above the
downstream hill. The minimum grid spacing at the wall varies between Δy+ = 0.20 to 4.0 and the maximum spacing
in the middle of the channel is Δy+ = 10. Similarly, the maximum spacing in the x- and z-directions for Grid 2 are
below 15 in the region up to x ~ 7.5. The minimum grid spacing at the wall varies between Δy+ = 0.20 to 1.0. These
values for Grid 2 are finer than the recommended values31 in the whole domain. The grid spacing near the wall for
Grid 1 is slightly higher than the recommended values.

  17  
 50
Δx+ Grid 1
+
Δy
+
40 Δz
Δy+m

Grid spacing
Grid 2
30

20

10

0
0 2 4 X 6 8

Figure 22. Grid spacing in wall units along the bottom wall, Re = 10595.

Figures 23(a-d) show the contours and the streamlines obtained from (a) LES without the SGS model (ILES(1))
using Grid 1, (b) LES(1) with the dynamic Smagorinsky model using Grid 1, (c) LES without the SGS model
(ILES(2)) using Grid 2 and (d) LES(2) with the dynamic Smagorinsky model using Grid 2. The separation bubble
appears slightly smaller and the reattachment point occurs earlier in the results obtained with ILES(1) compared to
LES(1). Figure 24(a) depicts the variation of the skin friction at the lower wall. Figure 24(b) shows the pressure
coefficient along the lower and upper walls. The pressure coefficients agree very well with the Breuer et al.19 data
except the implicit LES simulations with Grid 1. The pressure coefficient up to x ~ 2 is lower than that was
predicted by other simulations. The reattachment point is located at xreatt = 3.75, 4.2, 4.4 and 4.4 from the four
simulations ILES(1), LES(1), ILES(2) and LES(2), respectively. The separation points are located close to x ~ 0.22,
0.23, 0.22 and 0.20 from the four simulations. The separation point location is close to the DNS19 and the LES17,18
predicted value of 0.21. The predicted reattachment point locations by the DNS15,19 and the LES18 vary between 4.6-
4.7. The implicit LES simulation using the coarser grid predicts an early reattachment point compared to other cases.
This was also observed in the simulations at the lower Reynolds number Reb = 2800. As we discussed earlier, this is
due to the low molecular dissipation of the turbulent kinetic energy by the resolved scales. Figure 25 shows the
variation of the sub-grid eddy viscosity at different streamwise stations x = 0.0, 1.0, 2.0, 4.0, and 8.0. As we
observed earlier, the eddy viscosity is large in the separated shear layer region. The computed maximum eddy
viscosity is about 1.8 times the molecular viscosity for the coarser Grid 1 and it is about 1.2 for Grid 2. Hence, with
the increasing Reynolds number, the effect of the SGS increases and this is reflected in the separation lengths
obtained with and without the SGS model.

  18  
 Figure 23. Hill flow contours of the mean streamwise velocity and the streamlines, Re=10595. (a) ILES(1), (b) LES(1), (c)
ILES(2) and (c) LES(2).

  19  
 (a) Skin friction (b) Pressure
0.03 0.6
Lower LES No Model Grid 1
Lower LES Dynamic Grid 1 0.4
Lower LES Dynamic Grid 2
0.02 Lower Breuer et al. 2009
0.2
cf

cp
0.01 -0.2 Lower LES No Model Grid 1
Upper
Lower LES Dynamic Grid 1
-0.4 Upper
Lower LES Dynamic Grid 2
0 -0.6 Upper
Lower Breuer et al. 2009
Upper
-0.8
0 1 2 3 4 5 6 7 8 9 0 2 4 X 6 8
X

Figure 24. Hill flow variation of mean (a) skin friction coefficient, and (b) pressure coefficient along the lower and upper
walls, Re = 10595.

3.0
Re = 10595
X
0.0 Grid 1
2.5 1.0
2.0
4.0
8.0
2.0 0.0 Grid 2
1.0
Y
2.0
1.5 4.0
8.0

1.0

0.5

0.0
0.0 0.5 1.0 vt / v 1.5 2.0

Figure 25. Variation of the sub-grid eddy viscosity at different stations, Re = 10595.

Figures (26-29) depict the comparison of the mean and turbulent quantities obtained from the four simulations,
ILES(1), LES(1), ILES(2), and LES(2) at different streamwise locations x = 0 and 4 with the data of Breuer et al.19
The mean velocity profiles show good agreement with the LES results of Ref. 19. The results from the dynamic LES
are better than that obtained with the ILES(1). The normal mean velocity at x = 4 is slightly under predicted
compared to the data from Ref. 19. The Reynolds stresses obtained with the dynamic LES also depict good
agreement at these stations with the data from Ref. 19.

  20  
 (a) X = 0 (b) X = 4

3.0 3.0
Re=10595 Re=10595
X X
0.0 ILES(1) 4.0 ILES(1)
2.5 0.0 LES (1) 2.5 4.0 LES (1)
0.0 ILES(2) 4.0 ILES(2)
0.0 LES(2) 4.0 LES(2)
0.0 Breuer et al. (2009) 4.0 Breuer et al. (2009)
2.0 2.0
Y Y

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2
<U> <U>

   
Figure 26. Comparison of mean streamwise velocity <U> profiles obtained from the ILES(1), LES(1), ILES(2) and LES(2)
simulations at the stations (a) x = 0, and (b) x = 4.

(a) X = 0 (b) X = 4

3.0 Re=10595 3.0 Re=10595

X X
0.0 ILES(1) 4.0 ILES(1)
2.5 0.0 LES (1) 2.5 4.0 LES (1)
0.0 ILES(2) 4.0 ILES(2)
0.0 LES(2) 4.0 LES(2)
0.0 Breuer et al. (2009) 4.0 Breuer et al. (2009)
2.0 2.0
Y Y

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
-0.10 0.00 0.10 0.20 -0.10 0.00 0.10 0.20
<V> <V>

   
Figure 27. Comparison of mean vertical velocity <V> profiles obtained from the ILES(1), LES(1), ILES(2) and LES(2)
simulations at the stations (a) x = 0 , and (b) x = 4.

  21  
 (a) X = 0 (b) X = 4

3.0 Re=10595 3.0 Re=10595

X X
0.0 ILES(1) 4.0 ILES(1)
2.5 0.0 LES (1) 2.5 4.0 LES (1)
0.0 ILES(2) 4.0 ILES(2)
0.0 LES(2) 4.0 LES(2)
0.0 Breuer et al. (2009) 4.0 Breuer et al. (2009)
2.0 2.0
Y Y

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10
<ρu"u"> <ρu"u">

   
Figure 28. Comparison of turbulent intensity <ρuʺ″uʺ″> profiles obtained from the ILES(1), LES(1), ILES(2) and LES(2)
simulations at the stations (a) x = 0, and (b) x = 4.

(a) X = 0 (b) X = 4

3.0 Re=10595 3.0 Re=10595

X X
0.0 ILES(1) 4.0 ILES(1)
0.0 LES (1) 4.0 LES (1)
2.5 2.5
0.0 ILES(2) 4.0 ILES(2)
0.0 LES(2) 4.0 LES(2)
0.0 Breuer et al. (2009) 4.0 Breuer et al. (2009)
2.0 2.0
Y Y

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
-0.04 -0.02 0.00 0.02 -0.04 -0.02 0.00 0.02
<ρu"v"> <ρu"v">

   
Figure 29. Comparison of turbulent shear stress <ρuʺ″vʺ″> profiles obtained from the ILES(1), LES(1), ILES(2) and
LES(2) simulations at the stations (a) x = 0, and (b) x = 4.

Figures 30(a-f) show all the Reynolds stress components profiles at six stations x = 0, 1, 2, 4, 5 and 8 obtained
from LES(2). We also plotted the mean velocity profiles in these figures. The first station x = 0 is located at the hill
crest and the flow separates slightly downstream of this point. The stations x = 1, 2, and 4 are situated inside the
separation zone, the station x = 5 is located downstream of the reattachment point, and the last station x = 8 is
located on the windward side of the downstream hill. The conclusions are similar to what we observed in the low
Reynolds number case (Figs. 11(a-f)). The maximum Reynolds stresses in the separated region are confined to the
separated shear layer region and they become small near the wall. The maximum values for the normal Reynolds
stresses in the x-, y-, and z-directions at the station x = 1 are < ρuʺ″uʺ″ > = 0.08, < ρvʺ″vʺ″ > = 0.05, < ρwʺ″wʺ″ > = 0.06
and the maximum Reynolds shear stress is < ρuʺ″vʺ″ > = 0.04. These maximum values occur near the middle of the
shear layer close to y ~ 1.0. The ratios of the three normal stresses are 1: 1/1.6 :1/1.3. As we observed earlier, the
turbulence in shear layers is more isotropic than in wall bounded flows. It is also noted by comparing these ratios
with that for the low Reynolds number Reb = 2800 case that the flow in the shear layer region becomes more
isotropic with the increasing Reynolds number. It is also seen as we observed for the low Reynolds number case that

  22  
when the flow approaches the reattachment point, Figs. 30(d) and (e), the normal stresses in the streamwise and
spanwise directions become almost the same near the wall. As we described in the low Reynolds number case, this
is due to the impinging of the shear layer against the bottom wall. The profiles near the downstream hill reveal that
the maximum fluctuations occur in the spanwise direction compared to x- and y-directions. These observations agree
with the simulations of Frohlich et al.15 ʺ″

(a) x=0 (b) x=1 (c) x=2

3.0 3.0 3.0

2.5 2.5 2.5

2.0 2.0 2.0

y y y
1.5 1.5 1.5

1.0 1.0 1.0

<u'u'>
<v'v'>
0.5 <w'w'> 0.5 0.5
<u'v'>
<U>/10
0.0 0.0 0.0
0.00 0.05 0.10 0.00 0.05 0.10 0.00 0.05 0.10

(d) x=4 (e) x=5 (f) x=8

3.0 3.0 3.0

2.5 2.5 2.5

2.0 2.0 2.0

y y y
1.5 1.5 1.5

1.0 1.0 1.0

0.5 0.5 0.5

0.0 0.0 0.0

0.00 0.05 0.10 0.00 0.05 0.10 0.00 0.05 0.10

Figure 30. Normal and shear Reynolds stress components profiles at different stations. Reb = 10595.

Figures 31(a-e) depict the magnitudes of the four terms in the energy equation Eq. (27), and the balance or the
total of these four terms at five stations x = 0, 1, 2, 4 and 8 obtained from LES(2). The observations are again similar
to what we discussed in the low Reynolds number case, Figs. 12(a-e). However, in this high Reynolds number case,
it is observed that the balance term is not zero for the grid we used. This points to the lack of grid resolution in the
simulations. The maximum deficit occurs in the separated shear layer regions at x = 1 and 2. The percentages of the
maximum deficit compared to the resolved dissipation are 25, 60, 50, 45, 40 at the stations x = 0, 1, 2, 4, and 8,
respectively. These percentage differences were also observed in other simulations.14, 15, 19 The maximum production
at the hill crest location x = 0 is about 0.2, which is two times higher than that in the low Reynolds number case.
However, downstream of the separation point, the magnitudes and the profiles for the four terms are almost the same
as that in the low Reynolds number case. The maximum production is about 0.10 at the station x = 1 and it decreases
to 0.04, 0.02, and 0.01 at the stations x = 2, 4, and 8, respectively. It is also observed that these peaks occur in the
middle of the shear layer. Comparing the figures for the low and high Reynolds numbers, Fig. 12 and 31, it is seen
that the role of different terms in the turbulent kinetic equation budget is almost the same in both cases. Near the
separation point region, x= 0 and 1, the production is balanced by the advection, diffusion, and dissipation terms. It
is also seen that diffusion is about two times larger than the dissipation term in this region. When we move towards
the lower wall, all the terms first decrease to zero and at the wall dissipation again increases and this is balanced by
the positive diffusion term.
We also again plotted the contours of the average statistical quantities: (a) turbulent production, (b) turbulent
dissipation, (c) ratio of turbulent production to dissipation, (d) turbulent kinetic energy, and (e) turbulent shear stress
in Figs. 32(a-e). The observations are similar to the low Reynolds number case. The production is maximum near
the start of the separated shear layer and is confined to a region along the separation line. Beyond the reattachment
point x ~ 4.4, the production is concentrated in the middle of the channel at a height of y ~ 1. There is negative

  23  
production near the reattachment region and along the shoulder of the hill. The maximum dissipation occurs along
the wall on the windward side of the hill. Outside the wall region, dissipation is confined to the shear layer and to its
proximity. The ratio of production to dissipation is shown in Fig. 32(c). This ratio takes a peak value of 6.5 near the
start of the separated shear layer x ~ 0.4 and y ~ 1.0. It decreases to a value around 2 near x ~ 6 and y ~ 1. Figure
32(d) shows the turbulent kinetic energy distribution in the separated turbulent boundary layer. The kinetic energy is
concentrated along the shear layer and it peaks near x ~ 1 and y ~ 1.0. After the reattachment point, the energy is
confined to a region near y ~1. After the reattachment point, turbulence is confined to a region away from and
parallel to the wall at a height of y ~ 1. The turbulent shear stress contours in Fig. 32(e) also convey the same picture
that turbulent shear stress is concentrated near the start of the shear layer and near the y ~ 1 region at the foot of the
hill.
To investigate the effects of the SGS in the LES simulations, we compared the Reynolds stresses and the
different terms in the turbulent kinetic energy equation obtained with LES with no model ILES(1) and LES with the
dynamic Smagorinsky model LES(1) using Grid 1. Figures 33(a) and (b) show the comparison at the station x = 1.
The solid and dashed lines denote the results for ILES(1) and LES(1), respectively. We also plotted the production
term obtained from LES(2) as the thick black line in Fig. 33(b). It is seen that turbulent production obtained with
ILES(1) is 30% higher compared to LES(1) and LES(2) at this station. The second observation is that the resolved
dissipation with ILES(1) is two times smaller than that obtained with LES(1). This will manifest into higher
Reynolds stresses in the ILES(1) computations compared to LES(1) computations. Figure 33(a) shows that all the
Reynolds stresses are higher in the ILES(1) compared to LES(1). This confirms the role and the importance of the
SGS in the coarse grid LES simulations. The sub-grid stresses extract the energy from the large scales and dissipate
it.

(a) x = 0 (b) x = 1 (c) x = 2

3.00 3.00 3.00

Advec Advec Advec

Prod Prod Prod
2.50 2.50 2.50
Diffu Diffu Diffu
Dissip Dissip Dissip
Total Total Total
2.00 2.00 2.00

Y Y Y
1.50 1.50 1.50

1.00 1.00 1.00

0.50 0.50 0.50

0.00 0.00 0.00

-0.05 0.00 0.05 0.10 -0.05 0.00 0.05 0.10 -0.05 0.00 0.05 0.10

(d) x = 4 (e) x = 8
3.00 3.00

Advec Advec
Prod Prod
2.50 2.50
Diffu Diffu
Dissip Dissip
Total Total
2.00 2.00

Y Y
1.50 1.50

1.00 1.00

0.50 0.50

0.00 0.00
-0.05 0.00 0.05 0.10 -0.05 0.00 0.05 0.10

Figure 31. Balance of different terms in the turbulent kinetic equation. Reb = 10595.

  24  
 Figure 32. Hill flow contours of the turbulent statistical quantities (a) Production, (b) Dissipation, (c)
Production/Dissipation, (d) Kinetic energy, and (e) Shear stress. Reb = 10595.

  25  
 (a) X=1 Reynolds stresses (b) X = 1 K.E Budget

1.2 1.20

1 1.00
Y

0.8 0.80
<ρu"u">
<ρv"v"> Advec.
<ρw"w"> Prod
0.6
<ρu"v"> 0.60 Diffu
Dissip
Total
0.4 0.40
-0.05 0 0.05 0.1 -0.10 0.00 0.10
Reynolds stresses

Figure 33. Comparison of (a) Reynolds stresses and (b) balance of different terms in the turbulent kinetic equation
obtained from ILES(1) (solid lines), LES(1) (dashed lines). Black thick solid line in Fig. (b) production only from LES(2).

IV. Conclusions
We investigated the turbulent flow through a channel flow with a constriction (2-D hill) using DNS and LES
calculations at two Reynolds numbers Reb = 2800 and 10595. The Navier-Stokes equations are solved using a higher
order compact scheme for space discretization and a 3rd order Runge-Kutta scheme for time discretization. We
employed the dynamic Smagorinsky sub-grid scale model in the LES calculations. DNS simulations were performed
using a fine grid at the low Reynolds number of 2800 and the results are compared with the reference data of Breuer
et al.19 The separation and reattachment points, skin friction, pressure coefficient, mean flow, and the turbulent
quantities agree well with the reference data. The instantaneous flow fields identified two distinct regions. One is the
region between the line joining the crests and the bottom wall and the other is above this line and the top wall. The
upper region resembles a plug flow through a channel. The bottom region is dominated by the recirculation flow and
large eddies. These two regions are separated by a strong shear layer. The results showed that the Reynolds stresses
peak in the separated shear layer region near the separation point. The simulations also showed that the normal
Reynolds stresses in the wall normal direction and in the spanwise directions have almost the same magnitude in the
shear layer region. Close to the reattachment point the magnitudes of the normal stresses in the streamwise and the
spanwise directions become almost the same. The budget of the terms in the turbulent kinetic energy equation
showed that the production is balanced by the diffusion, dissipation and advection terms in the shear layer region
near the separation point. It is also observed that the diffusion is two times larger than the dissipation in the
proximity of the separation point. The ratio of production to dissipation takes a maximum value of 4.5 near the
separation point. Another general observation is that all the quantities, shear stress, turbulent kinetic energy,
production, remain at a reasonable level in a region away from and parallel to the wall at the hill height.
LES simulations were performed at two Reynolds numbers Reb = 2800 and 10595 using two grids in each case.
We also performed simulations with and without the SGS model using the same grid. The mean flow and the
turbulent quantities agree well with the reference data. The predicted reattachment point is about 4% smaller
compared to the reference data19. The computed eddy viscosity showed that for the grids and the algorithm used in
these simulations, the effect of SGS is small at the lowest Reynolds number of 2800. The effect of the SGS increases
with increasing Reynolds number. The Reynolds stresses and the budget terms have similar behavior at both low
and high Reynolds numbers. The ratio of production to dissipation increases to 6.0 in the high Reynolds number
case. The simulations with and without the SGS model showed that the production is larger without the model
compared to with the model. The dissipation decreases without the model compared to with model. These two
effects cause the Reynolds stresses to increase and force the flow to reattach earlier compared to that with the model.

  26  
 References
1 nd
Hinze, J. O., “Turbulence,” 2 ed, New York, McGraw-Hill, 1975.
2
Pope, S. B., “Turbulent Flows,” Cambridge University Press, 2000.
3
Kim, J., Moin, P., and Moser, R., “Turbulent Statistics in a Fully Developed Channel Flow at Low Reynolds Number,” J. Fluid
Mech., 177, 1987, pp. 133-166.
4
Moin, P., and Mahesh, K, “Direct Numerical Simulation: a Tool for Turbulence Research,” Annu. Rev. Fluid Mech., 30, 1998,
pp. 539-578.
5
Wilcox, D. C., “Turbulence Modeling for CFD,” La Canada, CA, DCW Industries, 2006.
6
Greenblatt, D., Paschal, K. B., Yao, C. S., Harris, J., Schaeffler, N. W., and Washburn, A. E., “Experimental Investigation of
Separation Control Part 1: Baseline and Steady Suction,” AIAA Journal, Vol. 44, No. 12, 2006, pp. 2820-2830.
7
Rumsey, C. L., Gatski, T. B., Sellers, W. L., Vatsa, V. N., and Viken S. A., “Summary of the 2004 Computational Fluid
Dynamics Validation Workshop on Synthetic Jets,” AIAA Journal, Vol. 44, No. 2, 2006, pp. 194-207.
8
Rumsey, C. L., “Effect of Turbulence Models on Two Massively-Separated Benchmark Flow Cases,” NASA/TM-2003-212412,
2003.
9
Jang, Y. J., Temmerman, L. and Leschziner, M. A., “Investigation of Anisotropy-Resolving Turbulence Models by Reference to
Highly-Resolved LES Data for Separated Flows,” ECCOMAS Computational Fluid Dynamics Conference 2001, Swansea,
Wales.
10
Lienhart, H., Stoots, C., and Becker, S., “Flow and Turbulence Structures in the Wake of a Simplified Car Model (Ahmed
Model),” DGLR Fach Symp. Der STAB, Stuttgart Univ., Nov. 15-17, 2000.
11
Sagaut, P., “Large Eddy Simulation for Incompressible Flows: An Introduction,” 3rd ed., Springer-Verlag, New York, 2005.
12
Garnier, E. Adams, N, and Sagaut, P., “Large Eddy Simulation for Compressible Flows,” Springer-Verlag, New York, 2009.
13
Marquille, M., Laval, J. P., and Dolganov, R., “Direct Numerical Simulation of a Separated Channel Flow with a Smooth
Profile,” Journal of Turbulence, Vol. 9, No. 1, 2008, pp. 1-23.
14
Bentaleb, Y., Lardeau, S., and Leschziner, M. A., “Large-Eddy Simulation of Turbulent Boundary Layer Separation from a
Rounded Step,” Journal of Turbulence, Vol. 13, 2012, pp. 1-28.
15
Frohlich, J., Mellen, C. P., Rodi, W., Temmerman, L., and Leschziner, M. A., “Highly Resolved Large-Eddy Simulation of
Separated Flow in a Channel with Streamwise Periodic Constraint,” J. Fluid Mech., Vol. 526, 2005, pp. 19-66.
16
Temmerman, L., and Leschziner, M. A., “Large Eddy Simulation of Separated Flow in a Streamwise Periodic Channel
Constriction,” Turbulence and Shear Flow Phenomena 2, Taylor & Francis, London, Vol. 3, 2001, pp. 399-404.
17
Breuer, M., Jaffrezic, B., Peller, N., Manhart, M., Frohlich, J., Hinterberger, C., Rodi, W., Deng, G., Chikhaoui, O., Saric, S.,
and Jakirlic, S., “A Comparative Study of the Turbulent Flow Over a Periodic Arrangement of Smoothly Contoured Hills,”
Direct and Large-Eddy Simulation VI, Vol. 10, ERCOFTAC Series, Springer Science, Dordrecht, The Netherlands, Sixth
International ERGOFTAC Workshop on DNS and LES: DLES-6, Poitiers, France, 12-14 Sept. 2005, 2006, pp. 635-642.
18
Ziefle, J., Stolz, S., and Kleiser, L., “Large-Eddy Simulation of Separated Flow in a Channel with Streamwise-Periodic
Constrictions,” AIAA Journal, Vol. 46, No. 7, July 2008, pp. 1705-1718.
19
Breuer, M., Peller, N., Rapp, Ch., and Manhart, M., “Flow over periodic hills – Numerical and experimental study in a wide
range of Reynolds numbers,” Computer & Fluids 38 (2009) 433-457.
20
Temmerman, L., Wang, C., and Leschziner, M. A., “A Comparative Study of Separation From a Three-Dimensional Hill Using
Large-Eddy Simulation and Second-Moment_closure RANS Modeling”, European Congress on Computational Methods in
Applied Sciences and Engineering, ECCOMAS 2004, Jyvaskyla, 24-28 July 2004.
21
Simpson, R. L., “Turbulent Boundary-Layer Separation,” Ann. Rev. Fluid Mech., Vol. 21, 1989, 205-34.
22
Lenormand, E., Sagaut, P., and Ta Phuoc, L., “Large Eddy Simulation of Subsonic and Supersonic Channel Flow at Moderate
Reynolds Number,” Int. J. Numer. Meth. Fluids 2000; 32: 369-406.
23
Smagorinsky, J. “General cicrculation experiments with the primitive equations”, Mon. Weather Rev., 91, 99-164.
24
Deardorff, J. W., “A numerical study of three-dimensional turbulent channel flow at large Reynolds numbers,” J. Fluid Mech.,
Vol. 41, 1970, pp. 453-480.
25
Deardorff, J. W., “On the magnitude of the subgrid scale eddy viscosity coefficient,” J. Comp. Phys., Vol. 7, 1971, pp. 120-133.
26
Germano, M., Piomelli, U., Moin, P., and Cabot, W., “A Dynamic Subgrid-Scale Eddy Viscosity Model,” Physics of Fluids,
Vol. 3, No. 12, 1991.
27
Vreman, A. W., Direct and Large-Eddy Simulation of the Compressible Turbulent Mixing Layer, Ph.D thesis, Department of
Applied Mathematics, University of Twente, November 1995.
28
Lele, S. K., “Compact finite difference schemes with spectral-like resolution,” J. Comut. Phys., Vol. 103, 16 (1992).
29
Liu, X., Zhang, S., Zhang, H., and Shu, C. W., “A new class of central compact schemes with spectral-like resolution I: Linear
schemes,” J. Comput. Phys., Vol. 248, 2013, pp. 235-256.
30
Visbal, M. R., and Gaitonde, D. V., “On the Use of Higher-Order Finite-Difference Schemes on Curvilinear and Deforming
Meshes”, J. Compu. Phys., Vol. 181, 2002, pp. 155-185.
31
Balakumar, P., Rubinstein, R., and Rumsey, C. L., “DNS, Enstrophy Balance, and the Dissipation Equation in a Separated
Turbulent Channel Flow,” AIAA Paper 2013-3723, 2013.
32
Piomelli, U., and Chasnov, J. R., “Large-eddy simulations: theory and applications”. In Turbulence and Transition Modeling
(ed. M. Hallback, D. S. Henningson, A. V. Johansson and P. H. Alfredson), pp. 269-331. Kluwer.

  27