Beruflich Dokumente
Kultur Dokumente
H54
2000
c.2
v.3
Fourth Edition
KLAUS A. HOFFMANN
STEVE T. CHIANG
ODTO KUTUi"HANLSl
M. E. T. U. LIBRARY
www.EESbooks.com
" .. ~ \
1·
336037
QA9U H 54 2000
v.~
METULWRARY
0.2
ComputaUonal Fluid dynamics
Copyright @2000, 1998, 1993, 1989 by Engineering Education System. All rights
reserved. No part of this publication may be reproduced or distributed in any
form or by any means, mechanical or electronic, including photocopying, recording,
storage or retrieval system, without prior written permission from the publisher.
The data and information published in this book are for information purposes only.
The authors and publisher have used their best effort in preparing this book. The
authors and publisher are not liable for any injury or damage due to use, reliance,
or performance of materials appearing in this book.
ISBN 0-9623731-6-8
First Print: August 2000
This book is typeset by Jeanie Duvall dba SciTech Computer Typesetting of Austin,
Texas.
Or visit:
www.EESbooks.com
CONTENTS
Preface
Chapter Twenty:
Turbulence and Turbulent Flows 1
Chapter Twenty-One:
Turbulent Flow and Turbulence Models 20
21. 7 Applications 92
21. 7.1 Shock/Boundary Layer Interaction 92
21. 7.2 Two-Dimensional Base Flow 95
21. 7.3 Axisymmetric, Supersonic, Turbulent Exhaust Flow 100
21.8 Favre-Averaged Navier-Stokes Equations 105
21.8.1 Turbulent Viscosity 113
21.9 Concluding Remarks 115
Chapter Twenty-Two:
Compact Finite Difference Formulations 117
Chapter Twenty-Three:
Large Eddy Simulation and Direct
Numerical Simulation 139
IAppendices
Appendix J: Transformation of Turbulence Models from Physical Space
to Computational Space 155
Appendix K: The Transport Equation for the Turbulence Kinetic Energy 159
References 166
Index 170
PREFACE
~ )
--(J
S)
'\
K
C. Wake flow
The simple turbulent boundary layer over a flat plate and a free shear layer
will be used to introduce conceptual models. Several processes associated with
the production of turbulence will be described in this chapter. Before attempting
to introduce a conceptional model, a brief description of transition is warranted,
which is given next. Subsequently, some fundamental concepts and definitions are
reviewed leading to introduction of a conceptual model.
.;
4 Chapter 20
part of the fully turbulent zone. The remaining part of the turbulent boundary
layer is considered to be in the outer region.
The division of a turbulent boundary layer to several regions defined above is
commonly identified by definition of a non-dimensional velocity u+ and normal to
the surface spatial coordinate y+. These quantities are defined as
U
u+=-
Ur
and
Ur
y + =y-
v
where UT is know as the friction velocity given by
and T", is the wall shear stress. Now, the various regions are identified as
and
Observe that the division between each region in terms of y+ is approximate, and
different values have been used by investigators to identify each region. However, the
range given above is the most common. The various regions of turbulent boundary
layer are summarized in Figure 20-3.
In this section the stability theory is reviewed. The mathematical details are
deleted; instead the procedures to obtain the governing equations and the conclu-
sions drawn from the solution of the equations are emphasized. The mathematical
details may be found in texts by White [20.2], Panton [20.3]' and Landahl and
Mello-Christensen [20.4].
The governing equation of stability is primarily derived for an incompressible
flow with constant properties. Assume a vector Q to represent the flow variables and
Q' to represent perturbations (disturbances) imposed on the flow. The flow variable
Q + Q' is substituted into the Navier-Stokes equations, and the original equations
written in terms of Q are subtracted from it. The equations thus obtained involve
the perturbation variable Q' as the unknown. The resulting equations will be non-
linear in perturbation properties Q', thus, typically the equations are linearized by
neglecting the higher order terms in Q'. This is in fact a reasonable assumption,
because the disturbances are assumed to be small. Now, if for a given problem
the disturbance Q' will decay, the problem is stable. On the other hand, if the
disturbance Q' grows with time, the problem is unstable. The governing equation
of stability may be solved numerically for a wide range of applications. Thadition-
ally, analytical solutions of stability equation have been achieved for special cases
where additional assumptions had to be imposed to reduce the equation. One such
solution, which is relevant to boundary layer flows, is reviewed at this point.
The problem of interest is the nearly parallel viscous flow. With this assumption,
the stability equation is reduced to a single ordinary differential equation. Typically
a disturbance in the form of an exponential function is introduced and the resulting
equation is known as the Orr-Sommerfeld equation.
At this point, the fundamentals and procedures for obtaining the governing
equations of stability have been reviewed. Again deleting the mathematical details
or the solution of Orr-Sommerfeld equation for parallel flow, several observations
regarding the solution of the equations are presented.
Typical solution of the Orr-Sommerfeld equation is given graphically in Figure
20-4 for a Blasius boundary layer flow. The horizontal axis represents ReB' where
ReB' = (puo')//-L, and the vertical axis is 0.0', where 0. is the wave number in the
x-direction and 0' is the displacement thickness. A curve defining the boundaries of
the unstable waves which separates the stable and unstable regions is called marginal
(or neutral) stability curve. A point on the marginal stability curve corresponding
to lowest Re6' is called the critical point. Within the curve of marginal stability is
the range of unstable wave numbers. These unstable waves are known as Tollmein-
Schlichting waves, which represent the first indication of instability.
8 Chapter 20
Stable
0.30
afl' 0.20
0.10
Stable
o
2 5 2 5
20.3.2 Transition
Based on the discussion of stability just completed, it is well known that at some
critical Reynolds number instabilities in the form of Tollmein-Schlichting waves
which are two-dimensional are developed. Shortly thereafter, the unstable waves
become three dimensional and nonlinear effects come into play. As mentioned pre-
viously, these instabilities by themselves do not define transition. In fact several
mechanisms and processes subsequent to initial instabilities must take place to
complete the process of transition from laminar to turbulent flow. Once the three-
dimensional instabilities are set in, the nonlinearity effects and interaction with the
mean flow, result in changes in the mean flow velocity and spanwise stretching of
vorticity. In addition internal shear layers are formed which are highly unstable.
Subsequently a breakdown process sets in where the stretched vortices breakdown
to smaller vortices with random frequencies and amplitudes, producing localized re-
gions of high shear and intense fluctuations in the shear layer. As a result, random
and disorderly motion termed turbulence is generated which travels downstream
and grows in coverage. This process occurs in localized regions surrounded by lam-
inar flow and is called turbulent spot. Eventually the number of spots is increased
in the streamwise direction and the flow becomes fully turbulent. This in turn com-
pletes a process, which is termed transition. This brief description of transition is
based on a flow over a smooth flat plate. The transition process can be enhanced or
delayed by several factors identified earlier. A graphical presentation of transition
process as proposed by White [20.2] is shown in Figure 20-5.
)
,,
,
Laminar flow Transition !, Fully turbulent
,,
,,
,,
''"'~~
Spanwise
Vortex
U
Streamwise
Low speed Vortices
Streak Ci / / / Peak
(1~ ~ ~ Valley
()2 2 2
(b) Near wall region has low and high speed streaks and the counter-rotating stream-wise
vortices are associated with these streaks. [20.7)
Breakdown
~
~ ~1») Lift lip '-'.::/
777777775777777777777 7777777/
(c) low speed streak lift up off the wall leading to breakdown, known as bursting. [20.7)
Intermittently, the low speed fluid is ejected outward accompanied by inrush of high-
speed fluid. The ejection of the low speed flow, which is accompanied with oscillation
and breakup, is termed bursting. It is observed that majority of turbulence energy
is produced in the sublayer/buffer zone regions [20.8J.
It is well known that the edge of a turbulent boundary layer is irregular with
peaks and valleys on the scale of boundary layer thickness. Freestream fluid is
entrained into the turbulent boundary layer at the irregular interface. Furthermore,
it is observed that the flow in the outer region of the boundary layer under the bulges
develops rotational spanwise eddies [20.9J. This outer region large-scale eddy is
illustrated in Figure 20-6.
There is ample evidence that the turbulence in the inner region is self-sustaining
and that the outer layer has some effect on the near wall production process [20.10J.
Obviously, there is an interaction between the inner and ou.ter regions in that there
is a transfer of mass, momentum, and energy from the outer layer to the inner layer
and vice versa.
Vortical structures have also been used extensively in description of conceptual
models of turbulent boundary layers. A brief and limited discussion of voritcal
structures is described in this section. The purpose again is to familiarize the reader
with some of the proposed models. For a comprehensive review Reference [20.1J
is highly recommended. The proposed conceptual model populates the turbulent
boundary layer with various vortices, which are the elements in the turbulence
production and dissipation as well as transport of mass and momentum between
the inner and outer layers. One of the most common vortical structures used is
the hairpin vortex (in high Reynolds number region) and the horseshoe vortex (in
low Reynolds number region) within the inner region. These vortices are inclined
in the downstream direction at relatively small angles from the wall, in the range
of 10° '" 50°. The formation of these vortices can be visualized as lifting of the
streamwise vortices, which are stretched and deformed to a horseshoe vortex. As
discussed previously, spanwise vortices or large eddies are developed in the outer
region.
smaller eddies is smaller due to their smaller velocity fluetuations and their life
span is considerably shorter. Furthermore, recall that kinetic energy is transferred
from larger eddies to smaller and smaller eddies the so called cascading process.
Eventually, the kinetic energy is dissipated into heat due to the action of molecular
viscosity at the scale of smallest eddies.
Now consider a turbulent boundary layer with a freest ream velocity of Uo and
a boundary layer thickness 0. Define a turnover time for the large eddies by 6/Uo.
The energy of the large eddy is proportional to UJ. Hence, the rate of dissipation
f would be proportional to Ugj6.
The scales of turbulence can be easily established for smallest eddies if one uses
Kolmogorov universal equilibrium theorem which states that the rate of transfer of
energy from larger eddies to smaller eddies is approximately equal to the dissipation
of .energy to heat by the smallest eddies. Therefore, the primary parameters for
smallest eddies are the mean dissipation and kinematics viscosity. Hence, one can
use dimensional argument to establish the following length (1J), velocity (v), and
time (7) scales,
1J -
(:3)1/4
v - (Vf) 1/4
7 - U;) 1/2
which are known as the Kolmogorov scales of length, velocity, and time. The
Kolmogorov length scale 1/ is then a measure of smallest eddies which can withstand
the damping effect of molecular viscosity. It is important to note that the smallest
length scale of turbulence that we have defined as the Kolmogorov length scale is
still much larger than the mean free path of molecules, the primary requirement of
continuum model. Therefore, the continuum model is commonly used for turbulent
flows.
It is appropriate at this time to review the concept of isotropy, which is used
extensively in turbulent flows, and it has significant implications for turbulence
models and large eddy simulation. Isotropy generically implies invariance with
respect to orientation. The assumption of local isotropy is used for small turbulence
scales. Experimental and theoretical considerations indicate that this assumption
is reasonable under certain circumstances. Nonetheless, the assumption of local
isotropy is used extensively. The implication of this assumption on the accuracy of
solutions and alternate procedures are currently under intense investigations.
Turbulence and Turbulent Flows 15
': ..
'.,' ;
.....,; ..
',:" .....
breakdown into turbulence takes place. Thus, imposed on the quasi two-dimensional
coherent eddies are the three-dimensional small scale eddies. As in the case of
wall-bounded turbulent flows, a significant fraction of turbulent kinetic energy and
Reynolds stresses are due to large scale eddies.
i
modated. This is where limitations to DNS are imposed by present day computers.
To identify the problem, recall that the smallest scale of turbulence to consider is
the Kolmogorov length scale TJ, and one can consider the boundary layer thickness
! 6 as the largest scale. To adequately resolve small-scale eddies whose size is of the
order of Kolmogorov length scale, a minimum of 4 to 6 grid points are required in
each direction. It is rather simple to establish an estimate of the number of grid
points required for a DNS of turbulent flows. For example, consider a turbulent
boundary layer where the largest length scale is the boundary layer thickness 6.
For a three-dimensional computation, the number of grid points with a domain of
6 x 6 x 6 would be approximately
where 6 grid points in each direction are used to resolve small scale eddies. Substi-
e e U5/6
tution of TJ = {1/3/ F/4 and = yields
(-6)3 = [(
{j -
e)1/4]3 = [(U~63)1/4l3
6 - -- 9/4 = (Re6) 94
= (lfo6) /
TJ 1/3 64 v3 1/
where
Uo6
Re6=-
1/
Thus N is proportional to (Re6)9/4 for the three-dimensional DNS of turbulent
boundary layer. Simple estimate of the number of grid points for Re6 = 5000
Turbulence and Turbulent Flows 17
would be around 4.5 x 1010 grid points. Observe that this is the number of grid
points required for a cubical domain of 6 x 6 x 6. However, even for a simple
problem of flat plate or a channel flow, the domain of solution must be in the
order of several 6 in each direction, most likely several hundred 6 in the streamwise
direction. And of course for complex geometries it would be in the order of several
million. It is thus seen that the number of required grid points for relatively complex
geometries at realistic Reynolds numbers are currently beyond the capability of
available computers. Current applications of DNS have been limited to simple
geometries at relatively low Reynolds numbers. A review of these applications will
be presented in Chapter 23. We can then conclude that, due to the above-mentioned
limitations, application of DNS to a majority of problems of interest is ruled out
at present. However, this conclusion does not make DNS obsolete, on the contrary,
the available results from DNS provide valuable data on the process of transition
and development of turbulent flow. These data can be used for development and
improvements of turbulence models. Furthermore, DNS provides wide range of
data, some of which is very difficult or impossible to measure experimentally. Thus,
DNS plays an important role in our understanding and prediction of turbulence and
its importance will increase as more powerful computers are developed.
In addition to computer requirements of DNS, several numerical issues must
also be addressed. First, the numerical scheme used must be higher order accurate.
Recall that typical schemes used in CFD applications are second-order accurate
in space. In order to accurately resolve various scales of turbulence an order of
accuracy of four or preferably six is required. For this purpose, compact finite
difference formulations appear to be very attractive. The compact finite difference
formulations will be discussed in Chapter 22.
The second numerical issue is development of accurate boundary conditions.
Treatment of boundary conditions will be addressed in Chapter 15.
At this point in time application of DNS for practical problems and its use in
design process is ruled out. However, over the last several decades approximate
procedures have been developed which allow us to solve turbulent flow fields. The
scheme is based on averaging of the fluid properties whereby the high frequency
(small-scale) fluctuations are removed. These fluctuating terms are then related
to the mean flow properties by relations, which are known as turbulence models.
Turbulence models are primarily developed based on experimental data obtained
from relatively simple flows under controlled environment. That in turn limits the
range of applicability of turbulence models. That is, no single turbulence model
is capable of providing accurate solution over a wide range of flow conditions and
geometries. Nonetheless, turbulence models are a practical approach for solution of
turbulent flows and are used extensively. Discussion of several turbulence models
18 Chapter 20
• Large-scale eddies in shear layers are more or less regular and possess co-
herent structure even though it is possible for them to become irregular.
Imposed upon the large scale coherent quasi two-dimensional structures are
three-dimensional small scale eddies.
• The growth of large-scale eddies are primarily governed by inertia and pressure
effects, whereas viscosity is the primary factor for small scale eddies. In fact,
viscosity is the dominant factor limiting the size of Ilmall scale eddies.
• Kinetic energy contained in large eddies are transferred to smaller eddies and
ultimately dissipated into heat at the level of smallest eddies due to molecular
viscosity.
• The transfer of energy from larger eddies to smaller and smaller eddies is
defined as cascading.
• As the flow Reynolds number is increased, the range of eddy sizes becomes
wider with the smallest eddies getting smaller.
• The number of grid points for DNS of turbulent flows scales as Re9 / 4 for
three-dimensional problems and as Re 2 for two-dimensional problems. When
the smallest scales of the flow are filtered out and the remaining scales are
directly computed by the filtered Navier-Stokes equation, the procedure is
called large eddy simulation (LES). The filtered small scales are determined
with the subgrid scale models and thus introduce some degree of uncertainties
into the computations.
Chapter 21
Turbulent Flow and Turbulence Models
surface temperature, surface injection, and pressure gradient, the mixing of the
fluid increases and takes place at the macroscopic level; streamlines are no longer
well-behaved. This type of flow is known as turbulent flow. There is a transitional
region between laminar and turbulent boundary layers known as the transition
region, which was described in Section 20.3. The various boundary layer regions
are shown in Figure 21-1.
/~
/../Y A J
~~ .. /'-~
U
~???11??117~?7?77777777777777
I. .1. .1.
laminar Transition
.1Turbulent
Turbulent
!--~==-
o
__-:-__ u/u.
As a result of heavy mixing of the fluid in a turbulent boundary layer and the
associated large momentum flux, the velocity profile becomes fuller in a turbulent
boundary layer as compared to a laminar boundary layer; i.e., the velocity gradient
near the wall for a turbulent boundary layer is larger than the velocity gradient
of a laminar boundary layer. Typical velocity profiles for laminar and turbulent
boundary layers are shown in Figure 21-2.
In addition, boundary layer flows can be classified into velocity boundary layers
and thermal boundary layers. For most problems, velocity boundary layer thick-
ness and thermal boundary layer thickness are not identical. Typical velocity and
thermal boundary layers are shown in Figure 21-3.
f-- T, --l
for Pr < 1
Pr= 1
Pr <1 (b(6,)
Figure 21-4. The velocity and thermal boundary layer thicknesses for
various Prandtl numbers.
Note that the influence of the outer layer on the near wall velocity profile enters
through
(21-2)
where
(21-3)
24 Chapter 21
(21-4)
for a smooth wall. This relation is known as the law of the wall. In terms of the
different regions of turbulent boundary layer defined previously in Section 20.2, the
law of the wall is valid for the viscous layer, the buffer zone, and the fully turbulent
portions of the boundary layer. It should be noted that the total shear stress
composed of viscous stress and Reynolds stress is approximately constant within
the regime governed by the law of the wall.
The outer region velocity profile can be expressed as
This relation is known as the defect law. Observe that viscosity does not appear
in the functional relation (21-5). Recall that as previously discussed in Section 20-
20, the effect of viscosity is to dissipate the small eddies due to exist~nce of large
velocity gradient. This effect is dominant near the wall where the Reynolds number
is relatively low. However, the contribution of small eddies to the Reynolds stress
in the high Reynolds number outer region is in fact negligible. As a consequence it
is argued that the outer region of the velocity profile is essentially independent of
viscosity.
At this point, a relation for the near wall region known as the law of the wall,
given by the functional form (21-1) or equivalently by (21--2), and a relation for the
outer region known as the defect law, given by (21-5) or (21-6), has been established
based on dimensional analysis and experimental observations. It is then obvious
that these relations must merge together smoothly over a finite region between
the near wall region and the outer region. This region is known as the log layer
or the overlap layer. The length of the log layer is problem dependent, and, in
particular, a function of Reynolds number (uTo)/v. Typical range of log layer is
,...., 50::; y+ ::;,...., 400, or, in terms of y/o, ,. . , 0.01 < y/o ::;,...., 0.2.
Now consider a smooth flat plate with no imposed pressure gradient. Then,
from relation (21-4), the law of the wall is
~ = f (yu.,.)
UT V
= f (~ u.,.O)
0 V
(21-7)
Turbulent Flow and Turbulence Models 25
~ = u e + 9 (~) (21-8)
Ur Ur 6
Since, in the log layer relations, (21-7) and (21-8) must merge together, we set
f (-uur6) _
Ue
- --+9 - (Y) (21-9)
6 v Ur 6
Functional analysis indicates the validity of relation (21-9) if both f and 9 have
logarithmic form. Thus an equation in the following form is established
U 1 +
- = -iny +B (21-10)
Ur K.
U - Ue = ]: in ~ _ A (21-12)
Ur K. 6
where
A '" 2.35
It is important to emphasize that both relations given by (21-11) and (21-12)
are valid only for incompressible flows over smooth flat plates with zero pressure
gradient.
Now consider the velocity profile very near the wall within the viscous sublayer,
that is, y+ < 2 '" 8. This is a region where viscous shear dominates and the velocity
profile can be approximated to be linear. Therefore,
8u", U
Tw = /-L - = /-L-
8y y
or
Tw /-L U U
-=--=v-
p p y y
or
U
u r2 = v-y
from which
U UrY
-=-
26 Chapter 21
or
(21-13)
The velocity profile in the sublayer region given by (21-1~1) must smoothly merge
with the velocity profile for the log layer given by (21-11). A relation satisfying this
requirement is established using experimental data as
which is valid for the buffer zone, that is, 2 8 < y+ < 2 '" 50.
f'V
Now that several relations which essentially form the L1W of the wall have been
established, Figure 20-3 is repeated as Figure 21-5, where the equation for each
region is identified. The compressibility law of the wall for the log layer can be
derived for most turbulence models. The main assumption used in the derivation
is based on the fact that the convection effects are small within the log layer, and
therefore they are ignored in the equations of fluid motion and the turbulence models
are considered. The mathematical details are omitted here, however the details can
be found in References (21.1, 21.2). The resulting expressions from the k - f model
and the k - w model are, respectively,
(21-15)
where
1
B. = B + -en -::
)5/4 (p (21-16)
Itt p",
and
1
u+ = -eny+ +B (21-17)
/'i. w
where
(_)1 /4 1
Bw =B+ -en ~ (21-18)
/'i. w pw
and Pw is the density at the wall. The constants /'i.. and /'i.w are known as the effective
von Karman constants. In fact, /'i.. and /'i.w are not constants, but are variables for
which expressions can be developed, as in Reference (21.1). However, the variation
in the values of /'i.. and /'i.w is typically small, in the range of about 0.2 percent for
/'i. w and 5 percent for /'i.. when the Mach number range is up to about 5. Therefore,
+
L
I
Defect law region ...
u I
Inner region ~ Outer region
~I
25 I
Viscous
sublayer
~ Buffer zone Fully
turbulent
I
20 I
I I
I I
Eq. (21-13) I
I
15 ~/ I
I
~
~
I
I
-t~
10 I
Eq. (21-14) I
I I
5 I I
I I
0 y+
1 10 100 1000
The time interval, tl.t, used in the definitions above must be larger than the period
of fluctuating quantities but smaller than the time interval associated with the
unsteady flow. Thus the time interval is problem dependent, i.e., depends on the
geometry and physics of the flow field being investigated. Note that for a steady
flow the time-averaged mean value is constant while for an unsteady flow it is a
function of time.
In order to carry out the mathematical details, the following averaging rules are
applied:
7- 1 (21-22a)
Turbulent Flow and Turbulence Models 29
fg - ]9 (21-22c)
8J -
8]
(21-22d)
8x 8x
f'2 f 0 (21-22e)
f'g' f 0 (21-22f)
The instantaneous values are replaced by time-averaged mean and fluctuating values
to provide
a a
+ -a [(p+ p') (u+ u') (v+ v')] + - [(p+ p') (u+ u') (v+ v')] =
y y
~
ax {J.L [~~ ~~ (v + V')]} + ~
3ax (u + u') - 3ay ay {J.L [~(u
ay + u') + ~
ax (11 + V')]}
+ ay {J.L [~
ay (u + u') + ~
ax (11 + V')] - ~y~ [~ (v + V')] }
3 ax y
Since the fluctuating component of viscosity is usually small, it has not been in-
cluded in this equation. Now the entire equation is time averaged and subsequently
rearranged as
:y (puv + pu'v' + up'v' + vp'u' + p'u'v') + : (puv + pu'v + up'v' + vp'u' + p'u'v')
i
=
~ [J.L (~au _~ 8V)] + ~ [J.L (au + au)] + a [J.L (auay -+- 8V) _~y~ (J.L1!..)]
ax 3 ax 3 ay ay ay ax y ax 3 ax y
a(Txrp - pul2
ax --- p'u-
12 - )
up'u'a (Txy - pu'v'
+ ay --- p'u'v'- )+
- vp'u'
+ -ay [T
xy
2 ya
- - (v) pu'v'
- 11--
3 ax y
- - v{lii! - {lu'v']
- (21-24)
+ a(
y pv2 + vp'v'
-)
= a (Txy - pu'v'
ax - - up'v'
-- - - )
p'u'v'
Turbulent Flow and Turbulence Models 31
The energy equation given by the fourth component of (11-192) in terms of the
total energy per unit mass is expressed as
a a a a
at (pet) + ax [(pet + p) u] + ay [(pet + p) v] + y [(pet + p) v] =
a
+ ax [ur",xp + vr",y -
a
q",] + ay [ur",y + vry!i7' - qy]
2
+ ay [ur", Y + VT.!I"".r _ q!l _ ~3 J.L vy _ y~ (~J.L V2) _ y~
ay 3 y
(~J.L uv)]
ax 3 y
Note that the total energy is assumed to be composed of internal energy and ki-
netic energy. The mathematical procedure used to include turbulence is similar to
previous manipulations. The final form of the energy equation becomes
ata[pet +"2?
1 (-
u12 + -)
v12 + -p'e' + '1(-
2 p'u12 + -
p'v) - + vp'v'
12 + up'u' -]
1 (
+ '2v 12 + -
p'u- ) + P (e'v'
p'V'2 - + '2u12v'
1 - + '2v13
1-) + p ( - + vv-12)
uu'v'
32 Chapter 21
2~ ~
( l 4~]
+ Ii- -3 ,V1ax + v 7fii) + 31/)
+ -y a[UT,. !I
- + Ii- (2~'
+ VT!1!11' - kaT
oy -- u'- + -~'
3 ox u'-oy + --u'v'
3y
41_)
+Ii- (~V,av' _ ~v'au') _ ~!!:. (v2) _ ~!!:. (v'2)
3 oy 3 ox 3y 3y
where
o
o
Q= (21-28a) 0'= o (21-28b)
!p (u12 + vl2)
Turbulent Flow and Turbulence Models 33
p'fI pv
p'fJ? + p p'fIv
E= (21-28c) F= (21-28d)
puv p'ii+p
(pet + p)u (pet + p)v
pv o
puv Txxp
H=! (21-28e) E" = (21-28f)
Y pfi TX!J
(pet +p)v UTxxp + VTxl/ -7jx
o
TX!l
F,,= (21-28g)
T!l1IP
UT"'!I + VT!l1IP - q!l
1
H,,=- (21-28h)
Y
o
-pill
E' = _fJuIV' (21-29a)
-¥i5U (U 12 + vl2) - p (e1ul) - pu (u12 ) - pv (UIVI) - p'u'
+/1 (~Ul{}u' _ ~UIOv') + [_~ (vl{}u' + VIOv') + ~ ""']
/I
""3 ax 3 &y t'" 3 ax &y 3!1
(21-29b)
J
34 Chapter 21
H'= ~ (21-29c)
y
au
-
Ov
ax+ay
- - 0 (21-30)
au +v-
au 1 &pa2'[i a (17)
'[i-
ax ay - -p ax + ay2 - ay Uv
II
(21-31)
Turbulent Flow and Turbulence Models 35
T/
au
= IL- = pv-
au (21-34)
ay ay
In order to express the turbulent shear in a similar form, consider the concept of
eddy viscosity, where the turbulent shear stress is related to the gradient of the
mean flow velocity. This analogy was introduced by Boussinesq and is referred to
as the Boussinesq assumption. Thus, one writes
au
Tt = ILt ay = pVt ay
au (21-35)
where ILt (or Vt) is known as the turbulent viscosity or eddy viscosity. Hence,
Equation (21-31) may be expressed as
au au1 8p 11
u - + v - = - - - + (v + Vt)- = -- -
a2
1 8p 1
+ -(IL
a2u
+ ILt) -8 2y (21-36)
ax ay pax 2
a y pay p
In a similar fashion, turbulent conductivity is defined to combine laminar and
turbulent heat fluxes. For a laminar flow, the Fourier heat conduction law is ex-
pressed as
(!), = -k:
For turbulent heat flux, the following is written,
= -pCpaa;:; (21-37)
where k t is the turbulent conductivity and at is the turbulent diffusivity. Now, the
energy equation may be rearranged as
aT aT = 1 (0u)2 1 a2T
u-
ax + 'iJ-
ay -(J-t+ J-tt)
PCp ay + -(k
-
PCp
+ kt )ay2
- (21-40)
Note that, up to this point, the problem of closure has not been addressed,
Le., two additional unknowns still remain. All that has been accomplished is the
rewriting of the equations by introducing the Boussinesq assumption.
It is well known that in a laminar flow the mixing of fluid is at the molecular
level, and the viscous stresses and heat fluxes are due to momentum and energy
transfer by molecules traveling a distance of mean-free path before collision. A
similar concept is extended to turbulent flows, where it is assumed that lumps of
fluid travel a finite distance before collision and before losing their identity. The
resulting momentum and energy exchange give rise to what was defined as turbulent
shear stress and turbulent heat flux. This finite distance is defined as the mixing
length, lm. This concept was introduced by Prandtl and is known as the Prandtl
mixing length. The Prandtl hypothesis is expressed as
Similarly,
(21-42)
kt = pCpl.2 (aT)
ay (21-44)
The concept of turbulent viscosity and turbulent conductivity can be easily ex-
tended to general flow fields governed by the Navier-Stokes equations. For example,
for a 2-D problem,
(21-45)
What has been accomplished up to this point is the introduction of new param-
eters, and the central issue of closure is not complete as yet since one still has two
additional unknowns, now in the form of J-tt and kt , or, equivalently, lm and le.
Turbulent Flow and Turbulence Models 37
effect of turbulence, Equation (21-46) is used along with the following definitions
for the viscous stress and heat conduction terms.
1 [4 2
Txx = Re (J-L + !J.t) 3(~XUt + 1Jx u f/ + Crud - 3(~yVt + 1Jl,Vf/ + (yvd
oo
1 [4 2
Tzz - Re (J-L + !J.t) 3(~zw~ + 1Jz wf/ + (zwd - 3(~xu~ + 77x'Ur, + (xud
oo
qx - (21-5:~)
qy - (21-54)
where a is usually assigned a value of 0.0168 for flows where Reo (momentum
thickness Reynolds number) is greater than 5000 and o' is the kinetic displacement
thickness defined as
The algebraic model described above requires information regarding the bound-
ary layer thickness and flow properties at the boundary layer edge. When the
Navier-Stokes equations are being solved, it may be a difficult task to determine
the boundary layer thickness and the required properties at the edge. That is es-
pecially the case when flow separation exists within the domain. However, when it
is necessary to determine the extent of the viscous region within the domain, the
total enthalpy is usually used.
A turbulence model which is not written in terms of the boundary layer quanti-
ties was introduced by Baldwin and Lomax [21-9]. The inner region is approximated
by
(21-64)
where w is the vorticity defined as
W=---
av au (21-65)
ax ay
and 1 is given by (21-56). The nondimensional space coordinate y+ can be written
as
y+ = UTJL = J'Twl Pw y
Vw Pw fLw
or
Turbulent Flow and Turbulence Models 41
y+ = Jpwlrwl~
I-tw
where the subscript w indicates the wall quantities. The outer region is approxi-
mated by
I-tt = apCCpFwakeFKleb (21-66)
where a is assigned a value of 0.0168 (as in the Cebeci/Smith model) and
A typical value for Cwake is 1.0. In Equation (21-67), the following definitions are
employed,
(21-68)
where", is the von Karman constant, and I, the mixing length, is determined by the
van Driest function given by (21-56). The difference between the absolute values
of the maximum and minimum velocities within the viscous region is denoted by
tl. V. For wall bounded flows, the minimum velocity occurs at the surface where the
velocity is zero, then
tl.V = (u 2 + V2)~
For shear layer flows, tl. V is defined as the difference between the maximum velocity
and the velocity at the Ymax location, that is,
and U r is the friction velocity. The velocity gradient in (3* is calculated outside the
viscous region. Once the Klebanoff constant is evaluated, Ccp is determined from
3 - 4CKJeb
C CP = (21-71)
2CKJeb (2 - 3CKJeb + C~Jeb)
Finally, the turbulent viscosity distribution across the boundary layer is deter-
mined from
(21-72)
that is, Ilt is set to Ill.; from the wall to a location for which Ill.; exceeds the value of
Ill., at which point Ilt is set to Ilt•.
To model the eddy diffusivity, at, or equivalently the turbulent conductivity,
the Reynolds analogy may be used. Recall that the Reynolds analogy assumes a
similarity between the momentum transfer and heat transfer. Therefore, a turbulent
Prandtl number is defined as
Prt = Vt = IltS, (21-73)
at kt
For most flows, it is assumed that the turbulent Prandtl number remains con-
stant across the boundary layer. For air, Prt = 0.9 for wall bounded flows and 0.5
for shear layers. Thus, the turbulent conductivity is determined as
kt = IltS, (21-74)
Prt
where Ilt is provided by the turbulence models just described.
To initiate the computations, a transition location Xtr is specified, and the initial
value of the turbulent viscosity is set equal to zero everywhere within the domain.
Subsequently, in regions where x > Xtr, the turbulent viscosity is computed from
(21-72) and updated after each time step (or iteration).
of turbulence given by (21-152), and kinematic viscosity. To account for the near
wall regions, the turbulence Reynolds number is split into two parts as Her -
ReT!(ReT), where! is a damping function such that ReT""" ReT for large ReT.
The transport equation for turbulence Reynolds number ReT is written as
(21-77)
(21-79)
where Dl and D2 are damping functions which extend the validity of the model to
near wall regions and are given by
-
44 Chapter 21
a (vReT)+
at
- - -
V ·'\1(vReT) -
(21-83)
where
82 = 2 [(au)2
ax + (av)2
ay + auay av]
ax + (au)2 ax _~3 (auax + av)2
ay + (av)2 ay (21-84)
a - - -
at (vReT) + v ·'\1(vReT) Vt) 1 -
2 v+- '\1 2 (vReT)--'\1,vt'\1(vReT)
-
- ( (1. (1.
(21-85)
Note that the unknown in either Equation (21-83) or (21-85) is ReT' Once ReT is
determined, relation (21-79) is used to determine Vt.
21.4.2.1.1 Nondimensional form: Since the governing equations to be solved are
typically in nondimensional form, the equations for turbulence models are also
Thrbulent Flow and Thrbulence Models 45
v't -
Vt
k'=~
2
f " = fL-
3
Voo u00 u 00
k,2
Re~ -
V"f"
Now, the nondimensional form of the Baldwin-Barth one-equation turbulence
model corresponding to (21-83) or (21-85) is given by the following equations where
the asterisk has been dropped.
8 - ~ - 1 ( 1 1 ) -
-8 (vReT)+ V·V'(vReT) - -R v + -vt) V' 2(vReT)
-
- - R -(V'Vt· V'(vReT)
t eoo u, eoo u,
(21-86)
or
-2- ( v + -u,
vt ) V' 2 (vReT)
-
- -1- -V'
1 ·vtV'(vReT
- )
Re oo Reoo u,
(21-87)
82 82 82 82
8x 2
- f,~ 8f,2 + 2f,:r:'f/:r: 8f,8'f/ + 'f/~ 8'f/2
(21-90)
46 Chapter 21
(21-91)
IIReT = R
Then, the transformed Baldwin-Barth turbulence model given by Equation
(21-86) is written as follows. The details of transformation and mathematical ma-
nipulation is provided in Appendix 1.
(21-92)
U - €"'u +€yV
V - ."",U + 1]yv
and
a4 - €;+€; (21-93)
2
S = al 8€ (8u )2+ b ()2
81] + a2 ()2
8u
1 8€ + 1>2 ()2
8v 8v
81]
Turbulent Flow and Turbulence Models 47
(21-98)
where the coefficients al, a2, a3, b), b2, ha, Cll C2, C3 and C4 are given by (1l-21Oa), (11-
21Ob) , (11-21Oc), (1l-211a), (1l-211b), (1l-211c), (1l-212a), (11-212b), (1l-212c),
and (1l-212d), respectively. The transformation of Equation (21-87) is carried out
in a slightly different fashion, the details of which are given in Appendix I. The
transformed equation is given as
1 1 [aa a~ aa a~]
- Reoo u. t;,,,, at;, + t;,y at;, + 'fJ", a'fJ + 'fJy a'fJ
where
aR aR aR
A - ax = t;,,,, at;, + T/", 8'fJ (21-100)
aR aR aR
B - ay = t;,11 at;, + 'fJ1I 8T/ (21-101)
a - lit
(aR aR)
t;,,,, at;, + T/", aT/ (21-102)
~ = lit
(aR aR)
t;,11 at;, + T/y aT/ (21-103)
(21-106)
48 Chapter 21
M2{ - 1 1
- Re oo a,
(ao.
ex ae + ell a.B)
ae (21-107)
M2 - -1- -(ao.
1 TI-+TI- a.B) (21-108)
'1 xaTl
Re oo a, lIaTl
VI) (aA
M3{ - -2- ( V+-
Re oo ex ae + e aB)
a, a{ y (21-109)
M3 -
'1
VI)
-2- ( V+- (aA
Tlx aTl + TIll aB)
a1]- (21-110)
Re oo 0',
and
p = (C,d2 - c,dvcl'DlD2 SR
Ivl - (21-113)
XI + ~l
Tl
x - V
(21-114)
Turbulent Flow and Turbulence Models 49
where d is the distance to the wall, K is the von Karman constant, S is the magnitude
of the vorticity,
S= 18V aul
ax _ ay (21-116)
and
X
Iv2 = 1- 1 + XlvI (21-117)
It should be noted that expression (21-66) has also been used for S. The function
Iw is
(1g6+c!3
I
where
(21-119)
and
11
r = SKfJi (21-120)
Large values of r should be truncated to a value of about 10. The function It2 is
given by
(21-121)
and the trip function In is
The first two terms on the right-hand side of Equation (21-123) can be expanded
and recombined to provide the following equation.
+ ~(v + V)\l2V
1
: _ ( : Cb2) [\lev + V) . \lvj
(21-124)
(21-126)
2 2
8 11 8
+ 2cs 8~8rt + b 4
11
8rt 2 + 91 8Tl 8Tl]
8~ + 92 8rt + Cb1 (I - ft2)SI1-
( 1 )
Reoo (c:wdw)
(11)2
d
where
8u 8u 8v
S = ~!I 8~ + rty 8rt -~:z: 8~ - rt:r 8rt
8vl (21-128)
I
and, as defined previously,
a4 - ~;+~; (21-129)
8Tl8Tl
+ U 8~ + V 8rt =
8Tl (1)
Re (1 +a Cb2) {80 80 8f3 8f3 }
~:z: 8~ + 71:r 8rt + ~y 8~ + 71y 871
8t oo
ODTO KOTOPHANESI
- M. E. T. U. LIBRARY
52 Chapter 21
(21-134)
where
Ov Ov Ov
A -
ax = ~x a~ + T/x aT/ (21-135)
B -
Ov Ov
ay = ~y a~ + T/y
avaT/ (21-136)
a - (v + V) (av
~x a~ + T/x Ov)
aT/ (21-137)
(3 - (v + V) (av
~y a~ + T/y av)
aT/ (21-138)
(21-139)
where
(a) The term M represents the convection and diffusion of turbulence and, due
to numerical consideration, is decomposed as follows:
(21-140)
(21-141)
(21-142)
M2 -
e (_1 ) C+Cb2)
Re oo a
(~8a
8e + e11 8(3)
% 8e (21-144)
M3 = - Cb2 _ (8A
( - 1 ) -(v+v) 17%-+17- 8B) (21-147)
'1
Re oo a 8TJ 11 8TJ
(b) The second term in Equation (21-139) represents the production of turbulence
and is
(c) The third term D represents the destruction of turbulence and is given by
(21-149)
(d) And, finally, the last term in Equation (21-139) represents the trip term which
causes/ triggers transition from laminar flow to turbulent flow.
(21-150)
discussed in the previous section. When two transport equations are used, it is
known as a two-equation model.
Complex flowfields which include massively separated flows, unsteadiness, and
flows involving multiple-length scales occur frequently in fluid mechanics applica-
tions. In these types of complex flows, the lower order turbulence models, that is,
zero-, half-, or one-equation models, become very complicated and often ambiguous.
Two-equation models are developed to better represent the physics of turbulence
in these types of complex flowfields. In this section, k-E and k-w two-equation
turbulence models are reviewed.
Since it is important to identify the terms involved in these equations, and, for
the benefit of those who are not familiar with the origin and derivation of these
equations, the details ofthe kinetic energy of turbulence equation and interpretation
of terms are provided in Appendix K. Now, the standard k-f two-equation model is
expressed by the turbulent kinetic equation
(21-153)
P(-ak ak ak)
at -I- u -ax + v -ay = -a [( ILt) ak] a [( ILt) ak]
ax IL + -Uk -ax + -ay IL -I- -Uk -ay + Pk - pE
(21-155)
Turbulent Flow and Turbulence Models 55
and
8f 8f 8f)
P
( -+u-+v-
at 8x 8y
(21-156)
k2
Mt = pCp.- (21-157)
f
and
(21-159)
(21-161)
56 Chapter 21
where the length scale may be evaluated by the simple linear relation for the viscous
sublayer given by
l = "'Y (21-162)
Once the local values of the kinetic energy of turbulence and dissipation of
turbulence have been computed, turbulent viscosity given by (21-157) may be de-
termined. It is noted that correlations such as u t2 , vt2 , and w 12 can be determined
by semi-empirical relations. For example,
(21-163)
v t2 = 2a3k (21-164)
(21-165)
where a2 and a3 are structural scales usually assigned values of 0.556 and 0.15,
respectively.
21.4.3.1.1 Low Reynolds number k- E model: The difficulty with the standard
k-f model introduced in the previous section is that the equations become numeri-
cally unstable when integrated to the wall. To overcome this problem, a two-layer
approach is implemented, as discussed previously. However, a better approach is
perhaps to directly integrate the k-f equations through the viscous sublayer all the
way to the wall. In order to enable the integration to the wall and to improve the
capability of the standard k-f model, several modifications are introduced. The
resulting formulation is known as the low-Reynolds number k-f equations. The
first low-Reynolds number k-f. model was developed by Jones and Launder [21-13,
21-14], and subsequently it has been modified by several investigators. The pri-
mary modifications introduced by Jones and Launder were to include turbulence
Reynolds number dependency functions II and h in Equation (21-154) and I,.. in
relation (21-157). Furthermore, additional terms L", and L, were added to the equa-
tions to account for the dissipation processes which may not be isotropic. Thus the
low-Reynolds number k-f equation is written as
p dk = ~ [ (/1- + t
/1- ) ak ] + P", - pf + L", (21-166)
dt ax; a", ax;
and
P ~ = a~; [ (/1- + ~:) ::; ] + Cd II P", i- C,2 hP i+ L, (21-167)
11 - 1.0
h - 1- 0.3exp (-Re~)
I,.. - [ -2.5 ]
exp 1 + 0.02ReT
Lk -2JL (8Vie)
8Xj
2
and
L. - 2 JL JLt
p
(88x~2
U
i
r
Since the early 1970's several modifications to the low Reynolds number terms
and the model constants have been introduced. A summary of selected k-e models
are provided in Table 21.1.
Nagano-Hishida
I
-2v I 8 k)
Vie
ay
~
v Vt(1- I,..) (8{)y~ 2
r 0.09 1.45 1.90 1.0 1.30
Table 21.1 Functions and constants for the k-e turbulence models.
58 Chapter 21
Now, the k-f equations including the compressibility correction terms are written
as
a (pk) + -a
-a a (pujk) = -a
a [ ( J-L + -J-Lt) -a
ak] + Pk - p(f + fc) + -P"d" + Lk (21-169)
t Xj Xj Uk Xj
and
The term fc represents the contribution due to compressible dissipation, and the
term P" d" represents the pressure dilatation term. They are given respectively by
and
P" d" = -/'2 Pk Mt2 + /'3 pf M t2 (21-172)
The following values based on DNS have been suggested.
21.4.3.1.3 Initial and boundary conditions: The initial and boundary conditions
must be specified for both k and f. The initial condition for the turbulent kinetic
energy k is specified in terms of the freest ream turbulence intensity Tt, as follows
Turbulent Flow and Turbulence Models 59
and, therefore,
(1) At the inflow, the freestream values are specified, that is, f = foo and k = k oo •
(2) At the solid surface, k = 0, J.lt = 0, and f = fw = O. For some k-f models, a
value for fw is prescribed.
(21-174)
pdJ..J
dt
= ~
8xj
[(J.L + O'J.Lt) 8W] + 0 w P" _ /3pw2
8xj k
(21-175)
f. - /3'wk (21-177)
k~
i = (21-178)
w
The constants used in Equations (21-174) and (21-175) are
5 3 , 9 1 , 1
0=-
9
/3= -40 /3 = 100
0'=-
2
0' =-
2
To include the best features of each model, Menter [21-18] has combined different
elements of the k-€ and k-w models to form a new two-equation turbulence model.
This model incorporates the k-w model for the inner region of the boundary layer,
and it switches to the k-€ model for the outer and wake regions of the boundary
layer. Two versions of the model introduced by Menter are referred to as the baseline
(BSL) model and, a modified version of the BSL model, the shear-stress transport
(SST) model. It has been shown that the SST model performs well in the prediction
of flows with adverse pressure gradient.
The baseline model is identical to the k-w model given by Equations (21-174)
and (21-175) in the sub layer and the log layer ofthe boundary layer, and it gradually
switches to the k-€ model given by Equation (21-153) and (21-154) in the outer wake
region. These equations are repeated here for convenience. The k-w equations are
8 _
-8 (pk)
8 __
+ -8 (pUj k) =
8 [
-8 (1-£ + akll-£t) -8
8k]·._
+ Pk - f3 pkw (21-179)
t Xj Xj Xj
and
(21-180)
where
k
I-£t = p- (21-181)
W
and
€ = f3'w k (21-182)
Furthermore, the following are used
8 _
-(pk) 8 __
+ -(pujk) = -8 [(1-£ + -I-£t) 8k] + Pk -
-- _
p€ (21-183)
8t 8xj 8Xj Uk 8Xj
and
(21-184)
where
(21-185)
62 Chapter 21
and
€
W =- (21-186)
c!,k
Now, in order to combine the two sets of equations, the k-€ set is transformed into
a k-w formulation using relation (21-182). In the process of this transformation,
two additional terms appear in the new w-equation. One of these terms is a cross-
diffusion term and the other occurs if ak and a. are not equal. It has been shown
that the second term is relatively small and does not affect significantly the solution.
Therefore it is neglected. Subsequently, the transformed k-€ equations written in a
k-w formulation are written as
a (pk)
at
+ aaXj (pUj k) = aa
Xj
[(/1- + ak2J-Lt) aak]
Xj
+ Pk - fJ*pkw (21-187)
and
1 ak aw
+ ll2 W-P
k
k - fJ2PW
2
+ 2paw2- --
w aXj aXj
(21-188)
The relations between the coefficients in the original k-€ equations and the trans-
formed set are established as follows
1 1
a w2 = - ,
a.
fJ* = c!' , ll2 = (Cd - 1)
Now the two sets of equations are combined by the introduction of a blending
function F. The k-w set given by Equations (21-179) and (21-180) is multiplied
by F, and the transofmred k-€ set given by Equations (21-187) and (21-188) is
multiplied by (1- F). They are subsequently added together. The blending function
F is designed such that it will be equal to 1 in the vicinity of the wall region, thereby
activating the k-w model, and it will be zero away from the wall, thus activating
the transformed k-f. model.
The resUlting combined k-w / k-f. two-equation model is given by
:t (p k) + a~. (pUj k)
J
= a~. [(J-L + akJ-Lt) ::.] + Pk -
J J
fJ* {XJJk (21-189)
and
(21-190)
Turbulent Flow and Turbulence Models 63
21.4.3.3.1 Baseline model: The turbulent viscosity for the baseline model is de-
fined by
(21-193)
and the constants for 4; are set as follows.
The values of the constants for set 4;1 are specified according to
01 = (31
(3* - O'W1/'i,2/ Ii-:: ~9
y~'
02 = ~~ - O'W2/'i,2/#
(21-194)
(21-195)
64 Chapter 21
where y is the distance to the nearest surface, and CD""" is the positive portion of
the cross-diffusion term
18k 8w -20"
CD""" = max [2p<Jw 2- -8 -8 ,10 (21-196)
W Xi Xi .
Note that far away from solid surface as y --+ large, term argl approaches zero
due to 1/y and 1/y2. Furthermore, the three arguments within argl represent the
following:
2) The term (500V/wy 2) enforces Fl to be one in the sublayer region and l/y in
the log layer. Recall that w is proportional to l/y2 near the surface, and it is
proportional to l/y in the log region. Therefore, l/wy2 will be constant near
the surface and will approach zero in the log layer.
and L is a characteristic length of the problem, for example, the length of compu-
tational domain. At the solid surface, the boundary condition for w is set according
to
(21-197)
where t:.Yl is the distance of the first point away from the wall. This boundary
condition is specified for a smooth wall, and t:.yt must be less than about 3. At
the outflow boundary, extrapolation is used.
Turbulent Flow and Turbulence Models 65
where
O"kl = 0.85
21.4.3.3.4 Compressibility correction: Similar to that of the k-f model, the com-
pressibility correction term is included by introduction of turbulent Mach number.
Thus, Equations (17-189) and (17-190) are formulated as
-aat (pk) + -a
a (pui k)
Xj
= a [(J.L + O"kJ.Lt) -a
-a
Xj
ak] + Pk
Xj
and
(21-201)
(21-202)
and
dw
p- = p (aw
- + uaw
- + vaw)
- = -1- -a [ (IL + O'w ILt)-
aw'l
dt at ax ay Reoo ax ax.
+- 1 -a [
(ILa
+ O'ww ]
ILt)- + 2p(1 - F 1 )O'w2 -1 [{)k
-- -aw ak -
+- aw]
Re oo ay ay w ax ax ay ay
W 2
+ a-Pk - fJpw (21-203)
k
where the production term Pk is determined from
au,
Pk=T'j-= [ ILt (au.
-+- aUj
--6 aUk)
2 'j 2
- --pM'j ] -aU. (21-204)
aXj aXj ax, 3 aXk 3 aXj
21.4.3.3.6 k-f/ k-w Turbulence model in computational space: Using the expres-
sions for the Cartesian derivatives given by (21-88) and (21-89), the k-equation in
the computational space becomes
ak ak ak 1 { a a
p at = -pU a~ - pV a", + Reoo ~xa~ [(MUS)(KX)] + "'x a", [(MUS) (KX)] +
a
~y a~ [(MUS)(KY)]
a
+ "'y a", [(MUS) (KY)] }+ Pt. - fJ'pwk (21-205)
where
ak ak
(21-207)
KX -
~ a~ + "'x a",
ak ak
KY - (21-208)
~y a~ + "'y a",
Turbulent Flow and Turbulence Models 67
8w 8w 8w 1{8
Pm = - pU 8(, - pV 8TJ + Re oo (,x 8(, [(MUR) (OX)]
+ TJx :TJ [(MUR)(OX)] +(,y :(, [(MUR) (OY)] + TJy :TJ [(MUR) (OY)] }
where
MU R - IL + O"wILt (21-210)
OX (21-211)
(21-212)
The production term Pic given by (21-204) is first expanded and subsequently trans-
formed to the computational space as follows.
where
1'xx _ (_1_)
Reoo
ILt (~ 8u _ ~ 8V) -
3 8x 3 8y
~pk
3
(21-214)
(21-215)
1'yy = (_1_)
Reoo
ILt (~ 8v _ ~ 8U) -
3 8y 3 8x
~pk
3
(21-216)
Utilizing the expressions (21-88) and (21-89), the shear stresses defined by (21-214)
through (21-216) become
(21-218)
au (au av) au
PI: = Tx:< ax + Try ay + aX + Tyy ay
au au av av
- (PI) a~ + (P2) aTJ + (P3) a~ + (P4) aTJ (21-220)
where
Recall that the term M can be decomposed into terms involving ~ or 1J derivatives
as defined by (21-107) through (21-110). Thus, M is expressed as M = M~ + M~.
Now, Equation (21-229) is written as
t:..R - - -
t:..t - (M~ + M~ + P) t:..R = (M; + M; + pn) (21-230)
or
[1- (M~ + M~ + P)t:..t] t:..R = (M; + M; + pn)t:..t (21-231)
and decomposed as
(21-233)
Turbulent Flow and Turbulence Models 71
and
[1 - tlt(M'I + P)] tlR = tlR* (21-234)
The finite difference expressions for each one of the terms appearing in Equations
(21-233) and (21-234) are developed separately.
The convective term Ml given by (21-106) is approximated by a first-order
upwind formulation as follows.
Ml = _U 8R _ V 8R
8e 8T1
(21-235)
If a second-order approximation for the convective term is desired, then the formu-
lation is as follows.
Ml = _U 8R _ V 8R
8e 8T1
+ ~(U
2 'J
. . -IU"I) (-R;+2 J + 4R;+lJ - 3R;J)
'
J
2tle
Since all the terms within M2 given by (21-107) or (21-108) are similar, the de-
velopment of the finite difference expression will be illustrated in detail for one
72 Chapter 21
term only. Subsequently, the conclusion will be extended to the remaining terms.
Now, consider the term f.x(80:/8~) which is approximated by a second-order central
difference expression as follows.
8R 8R)
( f.x 8~ + 1}x 8T]
0: = lit (21-238)
(21-239)
and
(21-240)
(21-242)
where
Turbulent Flow and Turbulence Models 73
(21-243)
Similarly for M;,
(21-244)
(21-245)
(21-246)
and
8R 8R
A - f.x 8f. + TJx 8TJ (21-247)
8R 8R
B = ~Y 8f. + TJy 8TJ (21-248)
Now the finite difference approximation for Ml following (21-243) can be ex-
pressed as
(21-249)
At this point, the finite difference approximation of the terms in Equations (21-
233) and (21-234) have been identified. However, the bar quantities such as M
require some deliberation.
The procedure for obtaining these quantities is illustrated for M~ in detail, and
the remaining terms can be obtained in a similar fashion.
Recall that Ml is approximated by (21-235) when first-order approximation is
used. Thus,
(21-251)
Note that, as before, when second-order approximation is used, that is, Equation
(21-236), the terms at grid points i - 2 and i + 2 are taken to the right-hand side
of the equation and treated explicitly. Now from Equation (21-251) expressions for
-1
M{ are developed as follow.
-11,- " -
Me l' 8::!IHJ = {~(UiJ + IUiJI) (~€)}
- 8::!liJ = - {~(UiJ + IUiJI) (~€)
Turbulent Flow and Turbulence Models 75
Now, with all the terms in Equations (21-233) and (21-234) identified, the two
tridiagonal systems of equations are solved sequentially by any tridiagonal solver,
for example, by the scheme introduced in Appendix B.
(21-252)
Turbulent Flow and Turbulence Models 77
where
2 a .. (UHIJ - Ui_IJ)2 b. (Ui,i+1 - UiJ_l)2
SiJ = I.) 2~~ + I.,; 2~1]
(21-253)
_ _(U Ov + .v Ov
8~81]
)
(21-254)
MI = _ {!(U IU.I)
2 ',J + ',J
(317,j - 417,-1,;
26.~
+ 17,-2 j )
The terms M2 and M3 given by (21-144) through (21-147) are approximated fol-
lowing the procedure illustrated in the previous section.
where
(21-256)
and
M2
'I
= (_1
R eoo
) (1 + Cb2) (8a
U
8 T}x
T}
+T}y"
!!(3)
(IT}
where
Turbulent Flow and Turbulence Models 79
(21-257)
and
M; = - (
1 )
Reoo
Cb2 _
-;;:-(v + v)
(aA
ex ae + eyaE)
ae
where
(21-258)
and
where
- (v + /7) (Tlx aA
aTi + T}y aE)
aT} ='"
80 Chapter 21
(21-259)
K, = 0.41
11
Iv2 - 1- X X=-
1+ XlvI II
~
Ivi - CuI = 7.1
X3+~1
(21-260)
(21-261)
s·
&,J
_ lav aul
ax ay .. &,J
(21-262)
Turbulent Flow and Turbulence Models 81
ViJ
XiJ (21-263)
ViJ
3
Xi,j
fVliJ - (21-266)
xlJ + ~I
The destruction term is given by
where
9 - r + ewa(r6 - r) (21-269)
1 V
r - (21-270)
Reoo "S",2d2
(21-271 )
Thus,
D·· -
'" - Re
1
oo
(
c",JUJ;J - Cbt)
",2 ft2;J (V'Jr
d;J (21-272)
fWiJ -
gi,j
(1+d.'
gf,j + ;;t
)1 (21-273)
(21-274)
r;,j = (21-275)
where
(21-276)
and
(21-277)
flq: Is the difference between the velocities at the field point and at the trip point
at the wall. Since the velocity at the wall is zero (no suction), flq is simply
the velocity at the field point.
At this point, all the finite difference expressions are in place to develop a nu-
merical scheme. The ADI formulation, which was introduced in Chapter 3, will be
used for this purpose. Recall that the governing equation given by (21-139) is
&TJ
-=M+P-D+T (21-278)
at
Furthermore, Equation (21-278) is nonlinear due to the terms in M, P, and D.
Thus, before a numerical scheme is considered, Equation (21-278) must be lin-
earized. The linearization procedure developed in Chapter 6, and similarly devf)l-
oped in Chapter 11, and given by Relation (11-81), is used to linearize Equation
(21-278). The general expression is
where
A=aE
au
Now, to develop an implicit scheme, Equation (21-278) is applied at time level n+ l.
Therefore,
(21-280)
Turbulent Flow and Turbulence Models 83
where the trip term is treated as a source term evaluated at time level n. Now the
linearization (21-279) provides
(21-281)
pn+! _ ap
pn + av6.V = pn + P6.v
- (21-282)
= D
n aD
+ DV6.V = D
n -
+ D6.v (21-283)
As seen previously, each one of these equations, once applied to all of the grid
points, will result in a tridiagonal system. For example, Equation (21-286) will yield
(21-288)
The finite difference expressions for the terms in RHSi j of Equation (21-288) have
already been identified. The "bar" quantities are determined in a similar fashion
84 Chapter 21
-I
as that of Section 21.6.1.1. For example, the results for Me are as follows
-II
Me i+IJ
-II
Me 'J
-n+1
v'J
_ -~
v.,j
_ llt {!(u. .. + IU... I)
2'J'J
(V.J -ll€V'-IJ) + !(U.
2 'J
...-IU.'..J I) (Vi+lJ -
ll€
V'J)
Turbulent Flow and Turbulence Models 85
(21-289)
86 Chapter 21
ak ak ak 1 { a
at - -U a~ - v a", + pReoo ~:r a~ [(MUS) (KX)]
a a
+ "':r a", [(MUS) (KX)] + ~I! a~ [(MUS) (KY)]
n+l
kiJ = kn
iJ -
A
ut
[{
'12 (u.iJ + IU.)
iJ
kiJ - kr-lJ
!:l.~ +
Turbulent Flow and Turbulence Models 87
+ 'fly;,; p,,}.. ~e
oo
(2:)u. 7J
{[(MUS)(KY) ]'J+I - [(MUS)(KY) ]'J-I}
(21-291)
To illustrate the midpoint approximation, consider the first four terms on the
right-hand side of Equation (21-205). Keep in mind that the convective terms
are approximated by upwind scheme, as before. The four terms of (21-205) under
consideration are
1
Reoo er
0 [(MU S) (Ok
oe er oe + 7Jr Ok)]
07J
0 [(MU S) (Ok
+ Re1oo ~y o~ ~y oe + 'T/y Ok)
O'T/ ]
=
1
Re oo
{OA
er oe + ell oB
oe }+ Re1 {8A 8B}
'T/% O'T/ + 'T/ll O'T/ (21-292)
oo
l
88 Chapter 21
where
ak
A = (MUS) ( ~x af. + rtx ak)
art
and
ak
B = (MUS) ( f.y a~ + rty ak)
art
Now consider the finite difference expression for (~xaA/8~) where central differ-
ence approximation is used.
(21-293)
where
A.+I
, ~"'
and
(21-295)
Similarly,
B,+, '
, ~"
(21-297)
For example,
1
ki+iJ+i = "4(k'J + ki+1J + ki+1J+l + k"j+l)
Similarly, the second term of (21-292) which is given by
1 {BA BB}
Tlx BTl + Tly BTl
Re oo
is approximated by
(21-300)
The w-equation has similar terms as the k-equation and, therefore, the FDE just
investigated can be used for the w-equation by replacing k with w. However, recall
that the w-equation has an extra term which will be investigated at this point.
The seventh term of the w-equation given by (21-209) is
/
92 Chapter 21
(21-301)
21. 7 Applications
Several turbulence models have been introduced in this chapter. The appli-
cations of these models to three different flowfields are illustrated in this section.
Computations are performed with several turbulence models for each domain such
that the solutions can be compared to each other and the experimental data.
The flowfield is solved either by the first-order and/or second-order flux-vector
splitting schemes described in Section 14.4.1.2 and given by Equations (14-21) and
(14-22) or by the modified fourth-order Runge-Kutta scheme described in Section
14.4.1.3.
Oblique shock
Incoming Turbulent
Note that at this stagnation point higher values of pressure and heat transfer are
produced. (3) The shock/turbulent boundary layer interaction is highly unsteady.
The shock typically oscillates back and forth, creating large pressure fluctuations
at the surface. This physical phenomenon has been shown by several researchers,
for example, Dolling and his co-workers [21-19 - 21-21].
A detailed flowfield survey for several compression corners have been conducted
by Settles [21-23]. The flowfield corresponding to a compression corner of 24° is
used in the application presented in this section. The flow conditions are specified
as follow, Moo = 2.85(u oo = 1875 ft/sec), Too = 180 OR, and a constant wall tem-
perature of Ttl) = 498 oR. The Reynolds number base on 00, where 00 = 0.83 inches,
was 1.33 x 106 •
The grid system is composed of 101 x 81 grid points as shown in Figure 21-
8. The grid points clustering near the surface and at the compression corner is
enforced. The y+ for the first grid point off the wall, (i = 2), varied from about 1
at the inlet to about 4 at the downstream of reattachment point. It is important
to reiterate that for turbulent flow computations, a y+ of less than 5, preferably a
y+ of 1 '" 2 are necessary. Satisfying this requirement will ensure that at least one
grid point is located within the viscous sublayer.
l
94 Chapter 21
8
7
6
5
y/oo 4
3
2
1
0
0 2 4 6 8 10 12 14 16
x/o o
The initial conditions at the inlet can be obtained by several methods. For ex-
ample, a computation can be conducted over a flat plate and at a location where
the computed boundary layer thickness matches the experimental value, the data,
including the turbulent viscosity distribution, is stored to be used as the inlet con-
ditions.
The no slip boundary condition and the constant wall temperature are imposed
at the surface, that is, the lower surface. At the upper boundary, the free stream
conditions are specified. Finally, at the downstream boundary, a first-order extrap-
olation is used. The solution is initiated by imposing the specified flow condition
at the inlet over the entire domain.
Comparisons of the computed surface pressure distributions with the experi-
mental data are shown in Figures (21-9) and (21-10). The solutions shown in these
figures are obtained using the Spalart-Allmaras one-equation model and the Jones-
Launder two-equation k-€ model, respectively.
Recall that one of the physical phenomena of the shock/turbulent boundary layer
interaction identified previously is the unsteadiness of this interaction. However,
the solutions obtained in this example are developed for steady state. There are a
couple of reasons for doing so. First, the experimental data used in comparisons
are provided as their mean values. Second, and perhaps more important, is the
formulations used in the computations. Since an averaging process is used in the
development of the Reynolds Average Navier-Stokes equation, the occurrence of
any high frequency processes within the domain cannot be predicted. In order to
Turbulent Flow and Turbulence Models 95
4.5 4.5
4 4
3.5
3
!
0
.e-<>
2.5
1.5
1
·4 ·2 0 2 4
xl6 0 xl6 0
Figure 21-9. Surface pressure distri- Figure 21-10. Surface pressure distri-
butions, Spalart-Allmaras butions, Jones-Launder
Turbulence model. k-e Turbulence model.
accurately predict the unsteady nature of this problem, large eddy simulation or
direct numerical simulation should be used. Nonetheless, the RANS equations com-
plemented with a turbulence model can predict the overall behavior of the flowfield.
Comparisons of the velocity profiles at the compression corner and downstream of
the corner are shown in Figures (21-11) and (21-12). The computed values compare
reasonably well with the experimental data in light of the complexity of the flow-
field. Observe that the reverse flow at the corner is predicted fairly well, as shown
in Figure (21-11).
It should be noted that, due to the uncertainties in experimentally measured
turbulence quantities, the approximations used in the RANS equations, and the
uncertainties in turbulence models, a prediction with a range of ±20 percent can
be considered reasonable for complex flows involving turbulence.
2.5 r;=====;1;;===;----,--~--,----:-
~~ .. ~ 35 BB .
SA
2 ~t Jl
I BSl I BSl - ...
i
1 5 Ir' ,--,EoO'I5P,e""nm!.!!eSllin~",,~_,,---,
I
/' I i'
25 SST
Ex erimental i:J
,~j
15
05
OL-_~~~C4~~_~L-_-L_~
Figure 21-11. Comparison of the ve- Figure 21-12. Comparison of the ve-
locity profiles at station loeity profiles at station
22, x = 0.0. 47, x = 4.0 inches.
Shear Layer
Recirculation
Region
The objectives of this exercise are to numerically simulate this complex flow-
field and, in the process, to evaluate the performance of several turbulence models
discussed previously.
The flow properties at the upper and lower streams are designated by subscripts
1 and 2, respectively, and they are specified as
Ml = 2.56, Ul = 584.13 m/sec, PI = 27,566 N/m2
and
M2 = 2.05, U2 = 523.83 m/sec, P2 = 61, 110 N/m2
The corresponding total pressure and total temperature are 517KP and 299 K,
respectively. The properties of stream 1 are used to nondimensionalize all the flow
properties.
The domain of solution is shown in Figure 21-14, where the origin of the coor-
dinate system is located at the upper corner, as illustrated.
(Plj)Block 2 - (PIMI-lj)Block 1
(ulj)Block 2 - (UIMI-lj)Block 1
(vlj)Block 2 - (VIMI-lj)Block 1
-
98 Chapter 21
f
2H
BLOCK I
y
BLOCK 2
Figure 21-14. Domain of solution for Figure 21-15. Illustration of the multi-
the base flow. block domain of solu-
tion.
Il D F Il D F
I I
A
i=1
I I
C E
i=IMI
I II ~ I
G
i=IM
A
i=1
C E
i=IMI·1 HMI
D F
W
C E G
i=1 i=2 i=IM
Once the solution for block 2 is completed, the computed values of the variables
along the line EF of block 2 are used as the boundary values along line EF(i =
1M1) of block 1, and solution of block 1 is initiated for the next time level. The
specification of boundary condition from block 2 to block 1 i.s set according to
(PIMIJ)Block 1 - (P2j)Block 2
(UIMl';)Block 1 - (u2.;)Block 2
(VIMl.;)Block 1 - (v2';)Block 2
(etIMIJ)Block 1 - (et2,j)Block 2
.,
Turbulent Flow and Turbulence Models 99
Figure 21-18. Initial rectangular grid Figure 21-19. Adapted grid system,
block 1: 44x32, block block 1: 44x70, block
2: 100x131, block 3: 2: 100x245, block 3:
44x18. 44x50.
Computations for the base flow are performed on the rectangular multi-block
grid system shown in Figure 21-18. This initial computation is performed to obtain
an estimate of the shear layer in order to generate a more appropriate grid where
there is a clustering of grid in the shear layer region, and, furthermore, the grid
lines are more or less aligned along the shear layer. Thus, once a solution on the
grid system shown in Figure 21-18 is obtained, a second grid based on the flow
data is generated which is shown in Figure 21-19. Subsequently, the solution from
the initial rectangular grid is mapped onto the new grid system and used as the
initial condition. Turbulent flow computations are initiated on this grid. Transition
to turbulence is specified on the upper and lower surfaces in blocks 1 and 3 at a
distance of half base thickness downstream of the inflow. Observe that there is
clustering of grid points along the upper and lower walls and at the inlet location
in addition to grid clustering along the shear layer.
As the solution develops, the grid system may be updated in order to better align
the grid line along the shear layer and to impose more efficiently the grid clustering.
The inlet flow conditions are generated in a similar approach, as described in the
previous example.
Computed streamwise mean velocity profiles obtained by the Baldwin-Barth
one-equation model, the Spalart-Allmaras one-equation model, and the Baseline k-
f./k-w two-equation model with compressibility correction are shown in Figures 21-
20 through 21-22. The axial locations in Figures 21-20 through 21-22 are at
x = 5mm, 25mm, and 55 mm, respectively. Observation of the experimental data
indicates that the axial locations of x = 5mm and 25mm approximately correspond
to the separation/recirculation region and the shear layer mixing region, where an
axial location of x = 55mm corresponds to the recompression/reattachment region.
The streamwise velocity profiles obtained by the baseline two-equation turbu-
100 Chapter 21
lence model with compressibility correction provide the best prediction, as com-
pared to the experimental data. Comparison of the baseline pressure is shown in
Figure 21-23. The pressure values obtained from the baseline two-equation model
with compressibility correction and the Baldwin-Barth model compare well with
the experimental data, whereas the pressure values predicted by the baseline and
Spalart-Allmaras models are much lower.
40'-1r-------------,-------------_,r--------,
• Experiment
40"-.~E,-~-'~;m-.-"tr---~-------------,---------,
I~
JOlm B-B 30] ___ a-a lti-
_.- 5-A I::
,-.
20 _.- 5-A 20 ____ _
,~
;:s
10~ _
... BSl -•.. BSl
....
'
j .....
~
BSL-cc 10 ' _ BSL-cc
E
E :~
.5.. 0 .di~" ...,;?."",.......",............_.;;;..- .5. 0 o
-10 :-pi~~-- -00
-30
-40
j I
_ AD
[
- .20
I
.00
I
20 .40
I
U/Ul
i
.60
J .80
I
1.00
I
1.20 1.40
-30
- 40 L .:;,~;---;----;;:~:--r----,1~--
-.40 - .20 .00 .20 .40
U/UI
.60 80 '.00 1 20 , 40
Figure 21-20. Comparison of the streamwise mean Figure 21-21. Comparison of the streamwise mean
velocity profiles at x = 5mm. velocity profiles at x = 25mm.
40,,------------,------------,r-------,
• Experiment
70,000 1 I • Experiment (Ave.) ~
___ 8-8 I
<
E ~40.000
.5. 0 < _______ ~, 4-· • ,
-10 - ---i----- .~'7--\, :1 30.000
.t
-20 20,DOO
-30 tO,DDO
o:o·:c:":::":~··~:,·,::=:::"j
-40 ! ~I oI ' 1
-.40 -.20 .00 .20 ..04-0 .60 .80 1.00 1.20 1.40
-50 -40 -30 -20 -10 10 20
U/Ul
y (mm)
......
Figure 21-22. Comparison of the streamwise mean Figure21-23. Comparison of the pressure distribu- o
......
velocity profiles at x = 55mm. tion at the base.
102 Chapter 21
Frce Stream
===> Mach Disk Reflected Shock
Figure 21-25. The grid system for the jet exhaust flow, block 1:
56x52, block 2: 200x78
Turbulent Flow and Turbulence Models 103
cca) with 01 = 1.0, 02 = 004, 03 = 0.2, and 02 = 0.6. The best solution is obtained
when both compressibility and axisymmetric correction terms are included .
. Expefimenl§
······Ofi9inoI8-8
2.00 --- Modified 8-8
1.50
~ 1.00
ii:
0.50
.
O.OO1--,--.--,--.--"r-'--'r-'--'--'--'--'---"--~'--'i
o 5 10 15
x/d
• Experimentj
--- Originol S·-A .
2.00 , ... _.- Modified S-A
/r
~.: ~\"
'1\'"
~ ~ .~:
. '( ;'
\~ 'd\t l \'{"'
1.50
Il :/\
~ 1.00
ii:
t .J
. . 1-.. . ,:..
1
\.,~\ 'It "\\ I~\ U[_·_}
d" , . ~. ,.:~ ,! ~.r\ '._ '.\,_.
N \ \ ! i . ,~. ~~/ i..:..\,-.,... ... .,._...•, :_. .•
\ 1
\ I /.
r:
Ii
';\', i f . .
\\ I!
j
":\, r,.'
•\
-,
Ii
.\ .~
': \.J
. \
... :'
0.50 \+\J '~~.,:... ·.\l ..
\1 '.
•
O.OO+-~~-,.--...,....-,--
,~--,--.--..,......-r-~-,.---~~--,
o 5
x/d
• Experiment
...... BSL
- BSL-cca
~ 1.00
..-:
:
O.OO-+--r--'---"-'--;--"'T"'"-r--'--"'T---ir--r---'--~~-;
o 5 10 15
x/d
1=]+1" (21-304)
where] and f" are the Favre mean and Favre fluctuating part of the turbulent
motion. Relation (21-304) can be multiplied by p and time·.averaged to yield
pl=p]+pf" (21-305)
pl=p] (21-306)
1"=1-] (21-308)
(21-310)
(21-311)
Turbulent Flow and Turbulence Models 107
j - j (21-312)
pf" - 0 (21-313)
p'f'
f" - --- (21-314)
P
pig - p j 9 + p I" gil (21-315)
With the definitions and rules of Favre-averaging established, consider the time-
or Reynolds-averaged continuity equation (for a two-dimensional flow) as given by
Equation (21-23), that is,
ap a ( -) a ( -)
at + ax pfl+p'u' + ay pfJ + p'v' =0 (21-316)
Using relation (21-303), the Favre-averaged continuity equation now can be written
as
ap a ( __ ) a (--) 0
-+-pu+-pv= (21-317)
at ax ay
Observe that the Favre-averaged continuity equation retains the same form as the
Reynolds-averaged continuity equation, and, in fact, the same form as the original
continuity equation. This is precisely the reason for the definition of mass-averaged
quantities, as given by (21-303). That is, the mass-averaged quantities are defined
in such a way as to write the resulting equation in a similar form as the original
equations, while retaining the effect of density fluctuations. Furthermore, this is
the reason why it was stated previously that the concept of Favre-averaging is
mathematical.
Now consider the x-component of the momentum equation for a two-dimensional
flow given by
-a (pu) +-
a (pu2) +-
a (puv ) =--+--+--
ap aTxz aTxy (21-318)
at ax ay ax ax ay
For simplicity, each term will be considered separately. The procedure, as illus-
trated previously, begins with time-averaging, and subsequently the Favre-averaging
concept is applied. Now the first term is written as
a a [ a a
at (pu) = at (p+P') (fl+U')] = at (pfl+p'u') = at (pu)
Similarly, the remaining terms are modified and given by the following
a a
ay (puv) ay (,0 uv + P u" v")
and
_
Tx'J =
_(au
J.L
av)
ay + ax (21-319)
Note that the density in pU,,2 and pu" v" is the instantaneous density, and that
these terms represent Reynolds stresses.
Similarly, the y-component of the momentum equation is written as
~(pv)+~
at ax (puv)+~
2
ay (pv ) = - ap +~ (rx -pu"v")+~ (of. _pV/~)
ay ax 'J ay 'J'J (21-321)
The energy equation in terms of the total energy per unit mass expressed in Carte-
sian coordinates for two-dimensional problems is given by
a
at (pet)
a a
+ ax (pu et + pu) + ay (pv et + pv) =,
a {)
ax (UTxx + VTxy - qx) + ay (UTx'J + VTW - qy) (21-322)
Again, considering the modification of each term separately, the first term is
(21-323)
Observe that, in the equation above, the potential energy has been dropped.
Turbulent Flow and Turbulence Models 109
Now the instantaneous terms in Equation (21-323) are replaced by their mean
and fluctuating values, thus,
-- 1 1-- 1-
pc"T +"2 p(u 2 + v2 ) = c" (p+ e') (T + T') +"2 pu2 +"2 pv2
- - 1- 1-
= c,,(pT + efT') + - pu2 + - pv 2
2 2
or
(21-324)
The last term in relation (21-324) is defined as the turbulence kinetic energy (per
unit mass) as follows
or
(21-325)
It is important to note that the turbulence kinetic energy defined in the Favre-
averaged formulation is different from the turbulence kinetic energy defined in the
Reynolds-averaged formulation given by (21-75) as
1
k = - (U,2 + V/2) (21-326)
2
which is essentially based on incompressible flow considerations. Now relation (21-
292) can be written as
(21-327)
The second term of the energy equation (21-322) is considered next, which is rewrit-
ten as
(21-328)
I pc"T+p=pc"T+pRT=pT(c,,+R) =pepT
l
110 Chapter 21
puT=puT+puIT" (21-330)
+~ (pU
1l3
+ PUIlV lP )
Finally, the second term is written as
and
-
CppT=CvpT+p
-
where
The terms on the right-hand side of the energy equation include shear stress terms
such as UTxx ' These terms, for example, are written as
and rearranged as
+~ [-upu"v"-iipv,,2+UTXy+iiTyy]
The terms involving triple products such as p U't", P U"VI/2, etc. are typically smaller
than other terms and are usually dropped. Therefore, the energy equation is reduced
to
+~
ax [-ep p ul/T" - Q",] + ~
ay [-ep Ii v"T" - Q]
y
(21-332)
where
_ = --at
q
ep _at
k- = --J1,-
'" ax Pr ax
and
_ = --at
q
ep _at
k- = --J1,-
Y ay Pr ay
aQ aE aF aEv aF. u
-at+ - + - = - +
ax ay ax ay - - (21-333)
where
'pu
pu2+p
Q= (21-334) E= (21-335)
puv
. (pet + p)u
pv
piiu
F= (21-336)
pfj2 +P
(p et + p}ii
Turbulent Flow and Turbulence Models 113
o
fxx-P?
(21-337)
fxy - p U"V"
U (fxx - pUlP) + il (fxy - pUIlV Il ) - ijx - eppulT"
o
fxy - pUIV"
(21-338)
fw - pV'P
au
- P u" V" = I-Lt ( - +ail)
-
ay ax (21-339)
(21-340)
-,p =
-pv I-Lt (2 ___
ail 2"<"7
yo v-::) 2 -k
--p (21-341)
ay 3 3
The stress terms defined above can be expressed in a tensor notation given by
II II
-pu·u·
, J
=I-Lt (aUi
aXj
- - -2U
- + -aUj
aXi
aUk· ·
3 aXk 'J
£) --p
3
£
2 - k U··
'J
(21-342)
(s- aUk £)
I
2 -k
- puiuj
II II
= 2I-Lt ij - 31 aXk Uij - 3 P Uij
£
(21-343)
l
114 Chapter 21
where
(21-344)
Thrbulent heat flux terms are also written in a form such as to resemble the laminar
heat flux term. For this purpose, a turbulent thermal conductivity is introduced
such that
aT
Cppu"T" = -kt - (21-345)
ax
and
aT
Cop p v"T" = -kt - (21-346)
ay
It is a common practice to write the turbulent thermal conductivity in terms of
turbulent viscosity by the introduction of a turbulent Prandtl number defined as
Prt = Cp ILt
kt
This is the same approach taken in the Reynolds-averaged formulation. Typically
the value of the turbulent Prandtl number is selected as a constant, and for air is
specified as either 0.9 for wall bounded flows or 0.5 for free shear layers.
Now the stress and heat flux terms in (21-337) and (21-338) are redefined as
follows.
(21-348)
(21-349)
-ijx -
-
-ijx - Cppu"T" = (k + k t) ax = Cp
aT ( Pr
p, + ILt) aT
Prt ax (21-350)
aQ aE aF aE" aF"
-at+ -
ax
+-
ay
= -ax+ -ay- (21-352)
Turbulent Flow and Turbulence Models 115
where
pu
PU2 +p
Q= (21-353) E= (21-354)
pUV
(pet+p)u
pv
pvu
F - (21-355)
pv 2 + p
(pet + p) v
0
Txx
Ev - (21-356)
TXll
0
T IIX
Fv - (21-357)
TlIlI
model which can be used in diverse flow regimes does not exist. Some models
may perform reasonably well in certain flow regimes but result in inaccurate pre-
dictions under different situations. Therefore, turbulence models must be carefully
investigated for their range of applicability before being utilized.
An attempt has been made in this chapter to introduce some fundamental con-
cepts of turbulent flow, turbulence models, and limited munber of solution schemes.
It is hoped that this attempt will be beneficial as a first step toward the inclusion
of turbulence into the governing equations of fluid motion. An in-depth review of
turbulence may be found in various texts such as [21-25] through [21-27].
Chapter 22
Compact Finite Difference Formulations
all = i.
= 1,-2 - 8/i-l + 8/H1 - IH2 + O(D.X)4 (22-2)
ax,' 12D.x
The grid points involved in the computations are shown in Figure 22.1.
Essentially what is being accomplished in the estimation of the derivative I at
point i is a curve fit through several points of data and subsequently determination of
the derivative. In fact, the finite difference approximations can be directly obtained
by a curve fit of polynomials as shown in Chapter 2.
Next logical approach to increase the accuracy of the estimates of the deriva-
tives, in particular for problems involving shorter length scales or equivalently high
frequencies is to include the influence of the derivative's of neighboring points in
the calculations. However, keep in mind that these "neighboring" derivatives are
unknowns, and therefore the developed expression will involve more than one un-
known. Nonetheless, when this equation is applied to all the points within the
domain sufficient equations are produced which can be solved simultaneously for
the unknowns. This approach is analogous to the solution of a pde by an implicit
.';
L
118 Chapter 22
scheme (more than one unknown per equation), compared to an explicit scheme
(one unknown per equation). The resulting approximation is referred to as Com-
pact Finite Difference (CTFD) formulation. They have been also referred to as
Hermitian formulations. The derivation of the CTFD formulation also involves
Taylor series expansions. A simple example is used here to illustrate the procedure,
and subsequently a general formulation is provided.
f(x)
f i_, f; (.,
fi _2
~+2
,, ,,,
,, ,, ,,,
,, ~
,,
,,
,,, ,
,,,
,
,
, ,,
)< x
I~xl~xl~xl~xl
'" "'''' "'''' "'''' '"
Figure 22-1. Grid points appearing in formulations (22-1) and (22-2).
1 2' 1 ( )3 fll 1 )4 iv 1 (A )5 ~
al [ Ii + t!.Xli + i(t!.x) Ii + 6 t!.x Ji + 24 (t!.x Ii + 126 uX Ji +
I
Compact Finite Difference Formulations 119
7~O (~X)6 ft' + O(~X)7] + £lof, + a_I [f, - ~xl. + ~(~X)2f. - ~(~x)3f:" +
(22-5)
or
Now if the relation given by Equation (22-6) is exact, then the coefficients in (22-
6) should be zero. In reality, however, only a few of the coefficients are set to zero,
and the remaining higher order terms are truncated and will form the truncation
error (TE). To obtain a third-order scheme, set the coefficients of J;, f;, /;', and /;"
equal to zero.
al + £lo + a-I = 0 (22-7)
1
6(~x
)3( al - a_I) + 2(~x)(bl
1 + b_ l ) = 0 (22-10)
The coefficients are ah £LO, a_h and bo are solved in terms of bl> and b_ l to yield
(22-11)
120 Chapter 22
2
ao - t!.x (b l - b_ l ) (22-12)
1
a-I - 2t!.x (b l + 5b_d (22-13)
bo - 2(b l + b_ l ) (22-14)
+ ... (22-15)
(22-16)
where the equation is divided by bl • Now, the expressions given by (22-11) through
(22-14) are redefined as
al 1
- - ( - 5 -a) (22-17)
bl 2t!.x
ao - 2
-(1- a) (22-18)
bl t!.x
a_I 1
bl
- T ( l + 5a)
2 x
(22-19)
bo
- 2(1 + a) (22-20)
bl
where a - b_I/b l .
Substitutions of (22-17) through (22-20) into (22-16) yield an equation for the
derivatives iLl, 1;', and 1;'+1' as follows
Compact Finite Difference Formulations 121
where
13
TE = 12(~X) (1- a) n".+160(~x) (1 + a)N + 180(~x)
4 15 .
(1- a)lt + .. (22-22)
Note that for a = 1 the leading error coefficient vanishes and the scheme becomes
fourth-order, whereas, for other selections of a's the scheme is third-order. For
a = 1 the formulation is
(22-23)
Now a general five-point formulation for the approximation of the first derivative
can be written as
1.'-2 -
l - 2~ f." 4(~X)2 l" _ 8(~x? f.'" 16(~x)4 f." (22-25)
, x ,+ 2! ' 3! ,+ 4! '
, f.' - ~ f." (~X)2 f.1II _ (~X)3 ff" (AX)4 f."
1,-1 -
, X ,+ 2 ! ' 3!' + 4! ' (22-26)
I'~I -
l ,+ ~ x,f." + (~x)2l' (~X)3 ff" (~X)4 f."
2! ,+ 3! ,+ 4! ' (22-27)
{3 IL2 + alLl + !;' + a I'~l + {3 1'~2 = {3UL2 + 1'~2) + aU ;-1 + I'~l) + I,' (22-29)
Substitution of expansions (22-25) through (22-28) into (22-29) yields
l
122 Chapter 22
, 2 III 2(~X)4
(20: + 2(:1 + 1) Ii + (~x) (0: + 4(:1) Ii + 4! (0: + 16(:1) N (22-30)
The required Taylor series expansions for the terms on the right-hand side of
(22-24) are
1,. = 1,. - 3~ f..' 9(~X)2 f." _ 27(~X)3f..111 81(~x)4 r!v _ 243(~X)5f.~ (22- 1)
1-3 I X I + 2! I 3! I + 4! .1, 5! I 3
.
1,1-1 = 1,. - ~ f..' (~X)2f..11 _ (~X)3 ~.III (~X)4f.iV _ It.lX)5f..v (22-33)
I x I + 2! I 3! h + 4! I 5! I
.
1,1+2 = 1,. 2~ f..' 4(~X)2f..11 8(~x?f.." 16(~x)4f..j.v 32(~X)5f..v (22-35)
I + X I + 2! I + 3! ,+ 4! I + 5! I
Expansions (22-31) through (22-36) are substituted into the right-hand side of
(22-24) to provide
, 2 If' 2(Dox)4
(2a + 2{3 + 1) Ii + (Dox) (a + 4(3)Ii + 4! (a + 16(3) N =
Setting the coefficients of the differentials equal, the following constraints be-
tween the coefficients are established
7'
Eighth-order: 2 ;(a+26 {3) = a+26 b+36 c (22-42)
6
9'
Tenth-order: 2-i(a + 2'1(3) = a + 28 b + 38 c (22-43)
8.
To obtain the error term of a particular scheme, the truncated terms are con-
sidered. For example, for the fourth-order scheme, the truncated terms are
1 2
TE = [5! (a + zib + 34 c) - 4! (a + 24(3)](Llx)4N (22-44)
L
124 Chapter 22
The following observations can be made with regard to the schemes which are
given in Table 22.1:
1. Scheme I
(a) For a = 0 the values of a and b are ~ and -~, respectively, and relation
(22-24) provides
.' = fi-2 - 8f.-1 + 8fl+1 -- fm
1,I 126.x (22-50)
which is the fourth-order central difference formulation.
(b) For a = ~ and the resulting values of a = ~~ and b = ~, the scheme
becomes formally sixth-order due to vanishing of the leading term in the
truncation error.
2. Scheme II
(a) For a = l, then a = ~4, b = ~, and c = 0, and the result is identical to
formulation of Ib. Therefore, the formulation of Ib can be considered as
a member of II.
(b) For a = ~, then a = ~, b = k, and c = -io,
and the scheme is eighth-
order.
3. Scheme III.
(a) For a = ~, (3 = 0, a = !~, b = k, c = -io,
the scheme is identical to lIb.
(b) For a = !,
(3 = 2~' a = g, b = ~~, and c =, 160' the scheme becomes
tenth-order accurate.
Compact Finite Difference Formulations 125
(22-51)
and
I I 1
11M + ailMI = ll.x (ahM + bhMMI + ChMM2 + dhMM3) (22-52)
The relation between the coefficients are established in the same manner as pre-
sented previously and are given by the following.
1
Second-order: a = - 2(3 + a + 2d)
b - 2+3d
1
C - - 2(1 - a + 6d)
1
Third-order: a - - -(11 + 2a)
6
1
b - -(6 - a)
2
1
C - -(2a - 3)
2
1
d - -(2 - a)
6
with the truncation error of
(22-54)
Fourth-order: a - 3
17
a -
6
126 Chapter 22
3
b - -
2
3
c - -
2
1
d -
6
and
TE = :!(LlX)4N (22-55)
Several formulations for the second derivative are provided in Table 22.2.
Compact Finite Difference Formulations 127
Table 22.2. Selected compact finite difference schemes for the second order
derivative.
Observe that for the scheme given by Equation (22-62), when a = 0, the values
of a and b are determined to be ~ and -l respectively. Subsequently Equation
(22-56) provides
which is the fourth-order central difference approximation of ft. For the case of
a = 121' a = i~, and b = (1' the scheme is formally sixth-order accurate.
the spatial grid size, s = !.z and K = 21'" = ~~ are defined as the scaled wave
number. Now,
k=IM/2
f(s) = L Fk elKs (22-67)
k=-IM/2
or
k=IM/2
/(s) = L F: elKs (22-69)
k=-IM/2
where
(22-70)
Now, consider a finite difference approximation of the first derivative of f as
IM/2
f;i s ) = L (22-71)
k=-IM/2
or
IM/2
where
F:/d = I K' Fk (22-73)
and K' = K'(k) is the modified wave number. One obtains the exact value if
or
K'=K (22-75)
which would be a straight line in a wave number, modified wave number coordinate.
Now, define the error as
ER= K-K' (22-76)
K
from which one may write
K' = K(1-ER) (22-77)
From the definition given by (22-76), the following may be written
Now, consider an example where a sine wave is used. In this case, the Fourier
series is reduced to one term only, therefore,
and
, K x K
I (x) = -cos(K-) = -cos(Ks) (22-80)
tlx tlx tlx
:.:...
2.00
li
e
."=
~
"
~
".
0;:
:;; 1.00
Q
0.00 "''--~-r---r---r-'--r---r--r-'--r---r--,
Wave number K
Figure 22-2. Comparison of several schemes for the approximation of the first
derivative of (22-79).
A comparison of the wave number K versus the modified wave number K' is
shown in Figure 22-2 for several schemes. It is evident that the compact finite
difference follows the exact differentiation over a wider range of wave numbers com-
pared to the fourth-order central difference approximation. The approximate value
of maximum wave number for well resolved waves (1<,.) is provided in Table 22.3.
A measure of resolving efficiency of a scheme may be defined by e = ~. For ex-
ample, as seen from Table 22.3, the resolving efficiency of the fourth-order scheme
with Q = 5/14 is higher than the resolving efficiency of the sixth-order scheme with
"'=!3'
....
130 Chapter 22
Scheme kr e
Central finite difference (fourth-order) 0.6 0.191
Compact finite difference (fourth-order, a = ~) 1.0 0.318
Compact finite difference (sixth-order, a = ~) 1.25 0.398
Compact finite difference (fourth-order, a = :4) 1.50 0.477
The explicit first-order upwind scheme applied to Equation (22-81) yields the
following FDE.
U~+I
, = u~, - c(u~' -1u~
- I) (22-83)
The modified Runge-Kutta scheme is given by
U(I) u~
- (22-84)
• •
(2)
u; - u~ _ a ~t (au)p) (22-85)
, 4 ax'
(3)
u; - u~ _ a ~t (aU)~2) (22-86)
• 3 ax'
(4)
u; - u~ _ a ~t (au )(3) (22-87)
• 2 ax'
u~+! - u~ _ a~t(aU)~4) (22-88)
• • ax •
Two options are considered in the approximation of the first-order derivative
on the right hand side of Equations (22-85) through (22-88). In the first option, a
fourth-order central difference approximation given by
1. Recall that the error term for the explicit first-order upwind scheme is given
by Equation (4-77), repeated here for convenience.
2
ER = a~x (1 _ c) a u2 _ a(~x? (2C _ 3c + 1) {flu3 +
2 ax 6 ax
O[(~x?, (~X)(~t)2, (~X)2 (~t), (~x)3l (22-90)
Note that the dominant term involves a second derivative a 2u / ax 2 , and there-
fore the error produced in the solution is dissipative. That is, as the Courant
number is decreased from its maximum value of one, the amplitude of the
wave is decreased and the wave is dissipated to the neighboring points. The
solution at Courant number of 1 is exact as seen from Equation (22-90), where
the error becomes zero.
132 Chapter 22
The solution by the compact formulation is well behaved. for all the three spec-
ified Courant numbers. Of course, this increase in accuracy is accompanied by an
increase in computation time. However, it should be stated that in some applica-
tions it will be necessary to implement the compact formulation due to its resolution
efficiency of small scales, which may be present in these applications. Example of
such applications includes problems in computational acoustics and direct numerical
simulation.
Compact Finite Difference Formulations 133
2.0
1.5 -<>-1=0.0
-0-1=0.02
1.0
-fr- 1=0.04
u
0.5 _ _ 1=0.06
-JI(- 1=0.08
0.0
~t=0.10
0 5 10 15 20 25 30
-0.5
(a)
x
2.0
1.5 -<>-1=0.0
-0-1=0.02
1.0
-fr-1=0.04
u
0.5 _ _ t=0.06
-JI(- t=0.08
0.0
~t=0.10
0 5 10 15 20 25 30
-0.5
(b)
x
2.0
1.5 -<>-t=O.O
-0-1=0.02
1.0
-fr- 1=0.04
u
0.5 _ _ 1=0.06
-JI(- 1=0.08
0.0
~1=0.10
0 5 10 15 20 25 30
-0.5
(c)
x
Figure 22-3. Solutions of the wave equation by the first-order upwind scheme.
134 Chapter 22
2.0
15 . -<>-- t=O.O
--0-- t=002
10
---fr- t=004
u
05 _ _ t=0.06
-lIE- t=0.08
0.0
-0-1=0.10
0 5 10 15 20 25 30
-0.5
(a)
x
2.0
-lIE- 1=0.08
0.0
-0-1=0.10
0 5 10 15 20 25 30
-0.5
(b)
x
2.0
2.0
1.5 -<>-t=O.O
-o-t=0.02
1.0
---fr- t=0.04
u
0.5 ---*- t=0.06
-lI(- t=0.08
0.0
-o-t=0.10
0 5 10 15 20 25 30
-05
(a)
x
2.0
1.5 -<>-t-O.O
-o-t=0.02
1.0
---fr- t=0.04
u
0.5 ---*- t=0.06
-lI(- t=0.08
0.0
-0-1=0.10
0 5 10 15 20 25 30
-05
(b)
x
2.0
1
136 Chapter 22
2.0
1.9 U
1.8
1.7
1.6
1.5
1.'
1.3
12
1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.'
0.3
0.2
0.1 X
0.0
·0.1
(a) 0 5 10 15 20 25 30
2.0
1.9 U
1.8
1.7
1.6
1.5
1.'
1.3
1.2
1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1 X
0.0
-0.1
(b) 0 5 10 15 20 25 30
2.0
1.9 U
1.8
1.7
1.6
1.6
1.4
1.3
1.2
1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.'
0.3
0.2 X
0.1
0.0
·0.1
(e) 0 5 10 15 20 25 30
-- 0- ANALYTICAL --)(- FIRST UPWIND - []- STANDARD RK4 -tr- COMPACT RK4
22.6 Problems
22.1 Consider a wave which is propagating within a domain with the following
initial distribution and shown in Figure P22.1.
u(x,O) =0
1.00 f
V
sx 0.00 +-___J
i
=>
-1.00 L-.---,
0.0 5.0
- - - - - 1 ..---... --.
10.0 15.0 20.0 25.0
- ----i----.
30.0
X
(III) Modified fourth-order Runge-Kutta scheme with central compact finite differ-
ence approximations
Plot the solutions at intervals of 0.02 sec up to 0.1 sec for all three schemes and for
the three Courant numbers.
22.2 Repeat Problem 22.1 for the following initial velocity distribution given by
2.00
6'
~ 1.00
:::J
0.00 1----i-""'"
0.0 5.0 10.0 15.0 20.0 25.0 30.0
X
which may be interpreted as a filtering operator for which any scale smaller than
ll.x is filtered out. Now, a filtered quantity is defined as
- 1 1:1:+60/2 1:1:+60/2
J;(x) = A f(t;)~ = f(t;)G(x,t;)~ (23-3)
L.l. :1:-60/2 :1:-60/2
where, in this case, G(x, t;) = 1/ll., G is called a filter function, and ll. is the filter
width. Typically, a filter over the entire domain is defined such that
otherwise
(23-6)
1=/+1' (23-7)
With the definitions of the resolved and unresolved quantities completed, the
Navier-Stokes equations are now modified to yield the Filtered Navier-Stokes (FNS)
equations. The FNS equations govern the evolution of large-scale eddies. It will be
given for an incompressible flow initially, and, subsequently it will be extended to
compressible flows.
After the application of a filtering process, the incompressible Navier-Stokes
equations become
(23-8)
(23-9)
where Tij is the subgridscale stress term which represents the effect of small scales.
The subgrid scale stress is given by
i
142 Chapter 23
can be written as
(23-12)
Observe that the Leonard stress involves only the resolved quantities, and therefore
it can be explicitly computed. Furthermore, note that this term represents the
interaction of resolved scales which contribute and affect subgrid scales. The cross
term stress and the SGS Reynolds stress involve unresolved quantities and must be
modeled. The cross stress term represents the interaction of resolved and unresolved
scales, whereas the SGS Reynolds stress represents the interaction of unresolved
scales.
It is important to note that a filtered quantity represented by an overbar, that is,
/, is different from the averaging process used in the RANS equation, in particular
f-lf.
It is common to rearrange Equation (23-9) and rewrite it as
OU'
-' + 0
-(u,u·)
at OXj 1 - - ~pOX,
~ (p - ~prkk)
3
6,·
J
(23-13)
(23-14)
and
(23-15)
Large Eddy Simulation and Direct Numerical Simulation 143
8p+ 8- (--)
- fYUi =
0 (23-17)
8t 8Xi
(23-19)
or
8 ( __ ) 8 ___ ) 8 (_ 1 ) 8Ti; 8 ( 1 )
at PUi + 8x; (pu,Uj + 8Xi P - 37/:/: 0,; = 8x; + 8x; 7i; - 37/:/:Oi; (23-20)
(23-23)
where
(23-24)
(23-25)
l
144 Chapter 23
Furthermore,
(23-26)
is the subgrid scale stress.
Note that at this point thesubgridscale stress T'j appearing in Equation (23-20)
is an additional unknown which must be modeled. FUrthermore, the introduction of
Tkk in relations (23-21) and (23-22) and the question of how it may be computed need
to be deliberated. These issues will be addressed shortly. For now, the goal is to
establish the required set of filtered Navier-Stokes equations for LES of compressible
flows. To complete the system of equations, consider the filtered energy equation
expressed as
(23-27)
where
(23-28)
With the assumption of perfect gas, the total energy given by relation (23-28)
can be written as
1 -( -2 -2 -2) 1
pel = c"p.L + 2P u + v + w - 2Tkk
-- "r;;
(23-29)
which can be rearranged as
--
pet = c"P-(T- - Tkk )
2c"p + '2P
1 -(u-2 + v-2 + w-2) (23-30)
Subsequently
(23-32)
Consider also the equation of state for a perfect gas which, in terms of the filtered
quantities, is written as
p= pRT (23-33)
In order to write the equation of state given by (23-3:J) in terms of modified
pressure and modified temperature, consider the following
_ 1 _ - 1 Tkk _ Tkl=_
P - -Tkk = pRT - -Tkk
3 3
- - _ pR
~p
+ --=
2c"p
pR =
pR ( T- - --_
Tkk)
2c"p
+ Tk.\; (R2c" 31)
- - -
Large Eddy Simulation and Direct Numerical Simulation 145
or
P+ = pRT
_ ++ Tkk
(R- - -1) (23-34)
2c" 3
Now the energy equation given by (23-27) is written in terms of the modified
pressure. Consider the second term given by (pet + P)Ui and rewrite it as
(23-35)
where
Qi = -(pet + P)Ui + (pet + P+)Ui (23-36)
is the subgrid heat flux. Thus, the energy equation is now written as
a (pet
__ ) + -a
a [( pet
__ + P+) u,_) = -a a (k -a
aQ; + -a aT) + -a S)
a (JLUj'j (23-37)
at X; X; X; X; X;
and
JL Uj S'j = JL Uj S'j (23-39)
Thus, the system of equations composed of filtered continuity, momentum, and
energy equations is written as
ap + ~ (piL,) = 0 (23-40)
at ax;
a (pu;)
__ + -a
a (__ _ + ) aT.) a ( -)
pu;Uj + P 6' j = -a + -a JL 8'j (23-41)
at Xj XJ Xj
-a( pet
at
__ )+ -a
ax,
[(-- Ui = -aQ, + -a
pet + P+)-)
ax; ax;
(kID a (JLUj-
-ax, + -ax, S-)
;j (23-42)
These equations are now expanded in Cartesian coordinate and are written in a
flux vector formulation by defining
p piL
pil piL2+ p+
Q= pv (23-43) E= pilv (23-44)
pw piLw
pet (pet + p+)u
l
146 Chapter 23
pv pw
pvu pwu
F= pv 2 + p+ (23-45) G= pwv (23-46)
pvill - -2 -I- p+
pw
(pet + p+)v (pet -I- p+)ill
o
+ + J.l S-xx
Txx
E" = Til, + J.l SXll (23-47)
+ + J.l S-xz
Txz
8T+ - - ..
Qx + k(JX + J.l (uSxx + V SXll + ill Sxz)
o
T:X + J.l Syx
F" = T; + J.lSyy (23-48)
T:Z + J.lSI/Z
8T+
QI/ + kay + J.l (uSI/X
- -
+ V Syy + ill 8..11z)
o
T:" + J.lSzx
G" = Tzt + J.lSZy (23-49)
T~ + J.lSzz
8T+
Qz + k7JZ + J.l (uS- zx + V Szy
-
+ ill 8.-zz )
Now the flux vector formulation, similar to that of the Navier-Stokes equation
given by (14-1), is written as
8Q 8E 8F 8G 8E" 81"" aG"
-
at+ -
8x
+-
8y
+8z
-=-8x
+-
ay
+ .8z- (23-50)
The subgridscale terms Tij and Q. are typically expressed in terms of the eddy
viscosity and eddy diffusivity similar to that in RANS equa.tions and is presented
next.
been established, continuous reference to RANS and turbulence models will be made
as options for LES and subgridscale models are explored.
As in the case of RANS where turbulence quantities are written in terms of
eddy viscosity and eddy diffusivity, the subgridscale terms are also expressed in
terms of eddy viscosity and eddy diffusivity. For most applications in either RANS
or LES, a model for eddy viscosity is developed, and, subsequently, eddy diffusivity
is determined by the introduction of a turbulent Prandtl number.
Subgridscale models similar to turbulence models vary in sophistication from
simple algebraic models to one-equation models to multi-equation models.
Several concepts with regard to small scales of turbulence were identified in
Chapter One which are relevant to the development of SGS models and are re-
viewed at this point. The small scales are more uniform and isotropic than large
scales. Therefore, modeling of small scales by simple algebraic expression tends to
represent the physics fairly well and is applicable to a wider range of flow condi-
tions. Furthermore, subgridscale stress contributes a fraction of total stress, and,
therefore, small errors in modeling do not substantially effect the overall accuracy.
In addition to assumptions of isotropic and uniformity of small scales, the as-
sumption of equilibrium is also imposed for most applications. The assumption of
equilibrium is referred to as a condition where the energy contained in large ed-
dies is transferred to smaller eddies, and, ultimately, all are dissipated into heat at
the level of smallest eddies (subgridscale) due to molecular viscosity. Although, on
the average, the transfer of energy takes place from resolved scales to unresolved
scales, the reverse can occur (known as backscatter), which is not accounted for
when equilibrium assumption is used.
Based on these discussions, algebraic models appear to be good candidates for
subgridscale terms. Besides their simplicity, they require a minimum amount of
computational effort. In the following sections, algebraic SGS models are intro-
duced.
(23-51)
where
(23-52)
148 Chapter 23
Observe that, just as in the RANS equations, the eddy conductivity kt is replaced
by terms of eddy viscosity and turbulent Prandtl number. Again, just as in the
RANS equations, several models are available to determine Vt.
Sij = ~
2
(au, + au;)
aXj ax,
(23-57)
The length scale is taken as the filter width ~, and the vel.ocity scale qSGS "" llSI,
where
lSI = J2Si; 5'j (23-59)
Finally,
Vt "" ~2151 (23-60)
Large Eddy Simulation and Direct Numerical Simulation 149
(23-61)
where, if eli: = 1.4, then e. = 0.18. This value of Smagorinsky constant is not
universal. In fact, similar to the constants used in the algebraic turbulence models,
it's value would vary. However, for most applications, a constant in the range of
0.1 < e. < 0.24 have been used.
The Smagorinsky model and its application are simple and provide reasonable
results for some flows. However, there are several drawbacks to the model.
• Third, it may be necessary to reduce the model constant in the near wall
region.
• Fourth, since, for most applications, grid clustering will be used, specification
of filter width tl. may be difficult. Among several options available, one may
consider tl. = (tl.x 2 + tl.y2 + tl.z 2)1/2 or tl. = (tl.xtl.ytl.Z)1/3.
An improvement to the model can be made where the value of constant e. would
vary locally, thereby reducing or increasing eddy viscosity in locations as required.
This concept is the basis of the dynamic model which is provided next.
1
150 Chapter 23
-
fij = U.Uj - -
UiUj (23-62)
(23-64)
(23-65)
(23-66)
The resolved stress i i j can be computed explicitly from the resolved velocity
field, whereas the SGS stress and STS stress require modeling.
In the following, the dynamic model is used in conjunction with the Smagorinsky
model. In fact, the procedure can be applied to any eddy viscosity model. Now,
the Smagorinsky model applied to SGS and STS stresses is written as
(23-67)
and
(23-68)
Large Eddy Simulation and Direct Numerical Simulation 151
where
(23-69)
(23-70)
or
A 1 A
(23-75)
from which
(23-76)
Another approach is proposed by Lilly [23.4], in which the sum of the squares
of the residual is minimized. The result is
1 L'jM,j
c= 2 M~. (23-77)
'3
regions. This may also become a source of instability in the solution. Though a
certain value of negative c is desireable in the solution, large values could create
serious problems. The negative values of c are interpreted as allowing backscatter
in the solution. Introduction of the averaging procedure discussed above tends to
eliminate the problem of excessive negative values of c. However, note that, with
the introduction of the averaging process over the domain, the model can no longer
be considered strictly local.
Now the formulations of c are written in terms of the averaging scheme as follows.
In closing this section, it is concluded that the dynamic model overcomes some
of the problems associated with eddy viscosity models with eonstant coefficient. In
the process, however, the dynamic model also creates some difficulties as identified
above. However, some simple procedures were introduced by which these difficulties
can be removed.
in performance of DNS for turbulent flows over large domain at high Reynolds
numbers.
The second challenge is the development of a solution algorithm which is free
of significant numerical error. This aspect of the problem includes the grid system,
numerical scheme, and boundary conditions.
The grid system must be developed by a higher-order grid generation scheme
such as to provide a high quality grid, that is, a grid for which there is continuity in
the curvature and the metrics obtained by high-order approximations, for example,
fourth-order or higher. An example of such grid generation is the fourth-order
elliptic grid generation scheme.
The development of numerical schemes must include higher-order approximation
of the derivatives. The fourth-order Runge-Kutta scheme can be used for time
integration, whereas typically sixth-order compact finite difference approximation
is used for the convection and diffusion terms. The third element of the solution
algorithm which is crucial is the treatment of boundary conditions. The sponge layer
type boundary treatment is usually implemented for the inflow/outflow boundaries,
and the standard no-slip boundary condition is applied at the solid surface.
Some examples of DNS of transitional/turbulent flows which have been reported
in the literature are provided below. As the technology matures, the envelope and
the range of applications will increase to include higher Reynolds numbers and more
complex geometries. The following list is a very limited example of DNS and by no
means is complete. .
• Compressible free shear flow. Several computations are performed at convec-
tive Mach number (Me) of 0.38 (Ml = 1.5, M2 = 1.5), Me of 0.4 (Ml = 2.0,
M2 = 1.2), and Me of 0.8 (Ml = 4.0, M2 = 2.4). The Reynolds number range
was between 100 and 500, where Re = Pl(Ul-U2)6w /fJ,l. The subscripts 1 and
2 refer to high and low speed streams, and 6w is the initial vorticity thickness
[23.5].
• 'fransitional flow at Mach 0.5 and ReL of 160,000 over a Joukowsky airfoil.
The effect of suction/blowing is considered as a means for controlled transition
[23.6).
• Flow transition at Mach 1.6 over a Joukowsky airfoil. The Reynolds number
based on the half-thickness of the airfoil h was Reh = 15000 [23.7].
• Compressible flat plate boundary layer flow at Mach 2.25. The Reynolds
number based on the inlet conditions was 635000/in [23.8].
• Fully turbulent channel flow. Effect of wall injection and suction was investi-
gated. The Reynolds number based on the channel half width and averaged
154 Chapter 23
The details of the coordinate transformation for turbulence models are provided
in this appendix. Since the terms for most models in each category, e.g., one-
equation models, are similar, mathematical steps will be carried out for a typical
model only. Extension to similar turbulence models is straightforward.
L
156 AppendixJ:
(; (8& 8R
+ <"x 8~ 8~ + 8~
8T/x 8R)
8T/ + T/x
(8~x 8R 8T/x 8R)
8T/ 8~ + 8T/ 8T/
8 2R 8 2R &R
+ ~~ 8e + 2~y'T"/y 8~8T/ + T/~ 8T/2
(; (8~y 8R 8T/y 8R) (8~v 8R 8T/y 8R)
+ <"y 8~ 8~ + 8~ 8T/ + T/ll 8T/ 8t + 8T/ 8T/
Rearrange terms,
where
a4 - ~;+~~
b4 - 17; + 17~
Cs - &T/x + ~y17!1
8& 8~x 8~!I 8~y
gl -
& 8e + T/z 8T/ + ~11 8e + 'T"/yaTi
8T/x 8T/x e 817" 817y
g2 - & 8e + T/x BT/ + "Be + 'T"/yi.~T/
Transformation of Turbulence Models: Physical Space to Computational Space 157
Now, consider the Baldwin-Barth turbulence model given by (21-87). The trans-
formation of the convective term is exactly the same as the previous case, that is
~ aR aR aR aR
V ·V R = u - + v - = u- + V -
ax ay ae a'f/
The Laplacian in the diffusion term is first expanded in Cartesian coordinates
and expressed as
where
A= aR and B= aR
ax ay
Now relations (21-88) and (21-89) are used to provide
2 aA aA aB aB
V R = (. ae + 1J., a1J + ell ae + 1J1I a1J
aA aB aA aB
- (. ae + ell ae + 1J., a'f/ + 1J1I a1J
1
158 AppendixJ:
\1 ,vI\1R aR ~~ +VI-
aR~)
= \1. (VI-
ax ay J
- ~:r: :~ [VI (~:r: ~; + 11:r: ~~)] + 11:r: :11 [VI ((:r: ~; + 11:r: ~~) ]
+ ~1I :~ [VI (~y ~; + 1111 ~~) ] + 11y :11 [VI (~y ~; + 11y ~~) ]
ao ap
ao ap
- ~:r: a~ + ~y a~ + 11:r: a11 + 11!1 a11
where
aR aR)
o = VI ( ~:r: a~ + 11:r: a11
aR aR)
p = VI ( ~!I a~ + 11y a11
Finally, the following definitions are introduced which will be used in the formulation
of FDE.
8 ( _ - ,)
' 8 (_ -pI,) 8 (-p'-'--')
+ -8xj PU,Uj + ax.J
Uj Uj + 8x.
J
U,Uj =
8p 8f'j
-8Xi
- +8xj
- (K-5)
-8
8t
(u1)
2
+-
P- 8 ( PUjU
8xj
i
2) = - U8p
8Xi
Tij
i - + U8i -
8xj
(K-9)
Now substitute the sum of average and fluctuating quantities for the instanta-
neous quantities, so that Equation (K-9) becomes
8 8
- (u, + uD -8
Xi
(jj+ pi) + (iii + uD -8 (fij + Tfj )
Xj
Taking the time average following the rules set by (17-2), one obtains
"218( - - )
- -2 - (l (l - -,
at PUi + PUs + P'Ui + 2Ui P'Ui
+ "21 8x.
8 [_ (_
P UjU,
-2 + UjU,
- :;i2\J + Uj
- ----;-;rp'
U,
, + 2-UjUj
- -pI,
U,
J
The Transport Equation for the Turbulence Kinetic Energy 161
(K-lO)
_ 8p + _ 8T,j
-U,- U;- (K-ll)
8x, 8xj
Subtract Equation (K-ll) from Equation (K-lO) and collect terms to obtain
- 8 (2-,)
+ -128x. - 8u,
u·p'u. - ii;p'u'·-
• J J 8x.
J J
+ -8xj "
8(PU;U,u·
-- -) - U;- - ,,
- 8(-)
pu,U·
J 8xj J
(K-12)
162 AppendixK
The remaining terms can be combined in a similar fashion, and Equation (K-12) is
written as
1 a [_ _Ii - p' ft
+-- - ---,--;'i
2 ax· pu;·u·' +u;' u·, +pu·u·
-,-p:]
; , + p'1L·U·
;,
;
--nT
_I up I a-'
T'j
-u·-+u·- (K-13)
'ax, 'aXj
Observe that the sum of the first and fifth terms is zero, due to continuity, and one
can write
Iap' = - -
-u,- a (-I) ,.J au:
rlu, +1'- (K-14)
ax, ax, ax;
Now define the turbulent kinetic energy k as
(K-15)
a (-)
- -a al (-)
puj k - ! : -,rOu;
Uj p'k - PU,Uj ~
Xj uXj uX;
-
-p'u~uj !au, - au, au: - OT:
: l - p'u~ Uj ! : l + rI ! l + lL,
I
!l
j
(K-16)
UXj UXj uX, UXj
(K-17)
- -axa. (-)
pu'.k
J
a (-)
- -ax. U'p'k
J
-au,
- pu(u'.-
• J ax.
J J J
u, + fFD"'
- (U;- -P" a~
u,u;) ax.
J
I. The rate of change of TKE which is composed of a local change and a con-
vective term is
(K-20)
-~
ax, (p'u:) - ~
ax; (pu"J k) - ~
ax; (u; p'k) (K-21)
- p
_-,
u,u-, -a
ail; - (_Uj piu,, + -pi") Ou,
u,u -;,- (K-22)
j j
Xj OXj
and it represents the rate of transfer of kinetic energy of the mean flow to
turbulence.
(K-24)
and finally
(K-26)
=ji (ak
at
+u' ak) =jidk
J aXjdt
(K-27)
-8 (I-£t
--- 8k) (K-28)
8x; U'" 8x;
where Tij now represents the turbulent shear stress and is given by
or
Tij = {I-£t (8Ui + 8uj _ ~ 8u", 0;;) _O,j ~3 Pk} 8x;
8u; (K-31)
8x; 8x, 3 8x",
N. The work done by the turbulent viscous stress is
(K-32)
8 _
- (p k) + -
8 __
(p Uj k) = -8 [(
1-£ +
8k]
-I-£t) -- + P'" - _
pf (K-34)
8t 8xj 8x; u'" 8x;
REFERENCES
[20-1] Robinson, S.K. "Coherent Motions in the Turbulent Boundary Layer," An-
nual Review Fluid Mechanics, 23, pp. 601--639, 1991.
[20-2] White, F.M., "Viscous Fluid Flow," McGraw-Hill Book Company, 1974.
[20-3] Panton, RL., "Incompressible Flow," John Wiley and Sons, 1984.
[20-4] Landahl, M.T., and Mollo-Christensen, E., "Turbulence and Random Pro-
cesses in Fluid Mechanics," Second Edition, Cambridge University Press,
1992.
[20-8] Kim, H.T., Kline, S.J., and Reynolds, W.C., "The Production of Turbulence
Near a Smooth Wall in a Turbulent Boundary Layer," J. Fluid Mechanics,
Vol. 50, 1971, pp. 133-160.
[20-9] Blackwelder, RF., and Kovasznzy, L.S., "Time Scales and Correlations in
a Turbulent Boundary Layer," Physics of Fluids" Voli5, No.9, 1972, pp.
1545-1554.
[20-10] Myose, R.Y., and Blackwelder, RF., "On the Role of the Outer Region in the
Turbulent Boundary Layer Bursting Process," J. Fluid Mechanics, Vol. 259,
1994, pp. 345-373.
References 167
[21-1J Wilcox, D. C., "Turbulence Modeling for CFD," DCW Industries, Inc., 1993.
[21-3J Chen, C-J., and Jaw, S-Y., "Fundamentals of Turbulence Modeling," Taylor
and Francis, 1998.
[21-6] Kral, L. D., Mani, M., and Ladd, J. A., "On the Application of Turbulence
Models for Aerodynamic and Propulsion Flowfields," AIAA-96-0564, January
1996.
[21-7J Rumsey, C. L., and Natsa, V. N., "Comparison of the Predictive Capabilities
of Several Turbulence Models," Journal of Aircraft, Vol. 32, No.3, May-June
1995.
[21-8] Kern, S., "Evaluation of Turbulence Models for High-Lift Military Airfoil
Flowfields," AIAA-96-0057, January 1996.
[21-11] Baldwin, B. S., and Barth, T. J., "A One-Equation Turbulence 'fransport
Model for High Reynolds Number Wall-Bounded Flows," NASA-TM-102847,
August 1990.
[21-12J Spalart, P. R., and Allmaras, S. R., "A One-Equation Turbulence Model for
Aerodynamics Flows," AIAA-92-0439, January 1992.
[21-13J Jones, W. P., and Launder, B. E., "The Prediction of Laminarization with a
Two-Equation Model of Turbulence," International Journal of Heat and Mass
Transfer, Vol. 15, 1972, pp. 301-314.
[21-15J Hoffman, G. H., "Improved Form of the Low Reynolds Number k-t Turbulence
Model," Physics of Fluids, Vol. 18, No.3, March 1975, pp. 309-312.
[21-16J Nagano, Y., and Hishida, M., "Improved Form of the k-t Model for Wall
Turbulent Shear Flows," Journal of Fluids Engineering, Vol. 109, June 1987,
pp. 156-160.
[21-18J Menter, F. R., "Assessment of Higher Order Thrbulence Models for Com-
plex Two- and Three-Dimensional Flowfields," NASA-TM-103944, September
1992.
[21-19] Unalmis, O. H., and Dolling, D. S., "Experimental Study of Causes of Un-
steadiness of Shock-Induced Turbulent Separation," AIAA Journal, Vol. 36,
No.3, March 1998, pp. 371-378.
[21-20J Dolling, D. S., and Murphy, M. T., "Unsteadiness of the Separation Shock
Wave Structure in a Supersonic Compression Ramp Flowfield," AIAA Jour-
nal, Vol. 21, No. 12, December 1983, pp. 1628-1634.
[21-23J Hoffmann, K. A., Suzen, Y. B., and Papadakis, M., "Numerical Computation
of High Speed Exhaust Flows," AIAA-95-0758, January 1995.
[21-25J Cebeci, T. and Bradshaw, P., "Physical and Computational Aspects of Con-
vective Heat Transfer," Springer-Verlag, 1984.
[21-26J Kays, W. M. and Crawford, M. E., "Convective Heat and Mass Transfer,"
Second edition, McGraw-Hill, 1980.
References 169
[21-27] Landahl, M. T., and Mollo-Christensen, E., "Turbulence and Random Pro-
cesses in Fluid Mechanics," Second Edition, Cambridge University Press,
1992.
[23-3) Germano, M., Piomelli, U., Moin, P., and Cabot, W. H., "A Dynamic Subgrid-
Scale Eddy Viscosity Model," Phys. Fluids A., Vol. 3, pp. 1760-1765, 199?
[23-5] Lele, S. K., "Direct Numerical Simulation of Compressible Free Shear Flows,"
AIAA-89-0376, January 1989.
[23-6] Liu, Z., Zhao, W., and Liu, C., "Direct Numerical Simulation of Transition in
a Subsonic Airfoil Boundary Layer," AIAA-97-0752, January 1997.
[23-7] Liu, Z., Zhao, W., Xiong, G., and Liu, C., "Direct Numerical Simulation of
Flow Transition in High-Speed Boundary Layers Around Airfoils," AIAA-97-
0753, January 1997.
[23-8] Rai, M. M., Gatski, T. B., and Erlebacher, G., "Direct Simulation of Spatially
Evolving Compressible Turbulent Boundary Layers," AIAA-95-0583, January
1995.
[23-9] Sumitari, Y., and Kasagi, N., "Direct Numerical Simulation of Turbulent
Transport with Uniform Wall Injection and Suction," AIAA Journal," Vol.
33, No.7, pp. 1220-1228, July 1995.
[23-10] Hatay, F. F., and Biringen, S., "Direct Numerical Simulation of Low-Reynolds
Number Supersonic Thrbulent Boundary Layers," AIAA-95-0581, January
1995.
INDEX
The numbers 1, 2 or 3 preceding the page numbers refer to Volumes 1, 2 and 3,
respectively.
Linearization, 1(90, 317), 2(31, 100, 163, Numerical flux functions, 1(241}
225} Orr-Sommerfeld equation, 3(7}
Lagging, 1 (90)
Orthogonality at the surface, 1 (407)
Iterative, 1(91}
Newton's iterative, 1(91} Outflow boundary condition, 1 (326,
348), 2(120, 190, 192, 325}
Local derivative, 1 (447)
Parabolic equation, 1 (6, 60)
Local time step, 2(146}
Parabolic grid generation, 1(418}
Low Reynolds number k-€ model, 3(56}
Parbolized Navier-Stokes equations,
LU Decomposition, 2(291}
2(60, 218}
MacCormack method, 1(190, 211, 278, Two-dimensional planar or axisym-
286, 287), 2(270} metric, 2(81}
Marker and cell (MAC) formulation, Partial pressure, 2(338}
1 (330) Poisson equation for pressure, 1 (310,
Metrics of transformation, 1 (363), 311)
2(23} Prandtl mixing length, 3(36}
Midpoint leapfrog method, 1(133, 186} Prandtl number, 1(463}, 2(25}, 3(22}
Minmod, 1(246}, 2(114} Pressure dilatation, 3(58}
Mixed boundary condition, 1 (20) Production of turbulence, 3(12, 54}
Mixed partial derivatives, 1(51} Reaction rates
Modified equation, 1(143} Backward reaction, 2(341}
Modified wave number, 3{128} Forward reaction, 2(341}
Momentum thickness, 3(40} Real gas, 2(337}
Monotone schemes, 1(236} Recombination,2(341}
Multi-step methods, 1 (189) Reynolds Averaged Navier-Stokes
Navier-Stokes equations, 1(274, 455}, Equation, 3(27, 28}
2(28, 267} Reynolds number, 1 (306), 2(25}
Two-dimensional planar or axisym- Richardson method, 1 (64)
metric, 2(69} Richtmyer method, 1(189}
Newtonian fluid, 1(452} Riemann invariants, 1(435}, 2(191}
Neumann boundary condition, 1(20} Robin boundary condition, 1(20}
Neumann boundary condition for pres- Rotational energy, 2(339}
sure, 1(323}
Round off error, 1 (113)
Nodal point scheme, 2(389}
Runge-Kutta method, 1 (219)
Nonequilibrium flow, 2(339} Modified Runge-Kutta, 1 (225, 290),
Nonlinear PDE, 1(3} 2(112, 180, 275}
174 Index