Você está na página 1de 27

Journal of Scientific Computing, Vol. 7, No.

l, 1992

The Renormalization Group, the e-Expansion and


Derivation of Turbulence Models
Victor Yakhot 1 and Leslie M. Smith 2

Received January 7, 1992

We reformulate the renormalization group (RNG) and the e-expansion for


derivation of turbulence models. The procedure is developed for the Navier-
Stokes equations and the transport equations for the kinetic energy N and
energy dissipation rate & The derivation draws on the works of Yakhot and
Orszag (1986) and Smith and Reynolds (1992), and all results are found at low
order in the underlying perturbation expansion in powers of e. The sum of the
source terms in the g-equation is known to be O(1) due to the balance at
leading order of O(R~ 2) terms. Smith and Reynolds (1992) showed the cancella-
tion of some of the O(R~ '2) terms generated by the RNG procedure. Here we
show that including the random-force contribution to g-production results in
the cancellation of all the O(R~ 2) terms. We find that two of the O(1) terms in
the RNG equation for the mean dissipation rate ~ have the same form as those
in the widely used model ~-equation. The values of the coefficients of the
familiar terms are close to those used in practice. An extra production term is
predicted which is small for slow distortions, bat important for rapid distor-
tions. Hence, it may be a term that should be added to the g model equation.
We believe that the present derivation places the ~ model equation on a more
solid theoretical basis.

KEY WORDS: Renormalization group; turbulence models; dissipation rate


equation.

1. I N T R O D U C T I O N
Many turbulence models are based on ideas borrowed from the kinetic
theory of gases, where macroscopic variables depend on microscopic
fluctuations only through kinetic coefficients such as viscosity and heat

1Applied and Computational Mathematics Program, Princeton University, Princeton,


New Jersey 08544.
2 The Center for Turbulence Research, Stanford University, Stanford, California 94305.
35
0885-7474/92/0100-0035506.50/0 9 1992 Plenum Publishing Corporation
36 Yakhot and Smith

conductivity. The microscopic motions in a fluid close to thermal equi-


librium are characterized by the mean free path )~, and in typical fluids
such as air, 2 ~ 10-6 cm under normal conditions, which is much smaller
than the typical macroscopic scale L = O(1) cm. The ratio 2/L is a small
parameter which allows perturbative treatment of the fluid motions. Using
perturbative techniques, one can derive the basic relations for a fluid near
equilibrium relating various fluxes to the gradients of macroscopic
variables. For example, the momentum-flux tensor is given by

2J~ 6 ~
(1)il)J)= -Y~ ViVj)~ 3 (1.1)

where v is the velocity field of thermal fluctuations, V is the mean large-


scale velocity field, ~ is the mean kinetic energy of the fluctuations and
V~- ~/Oxi. The molecular viscosity v0 is Vo ~ (2TM)m2 ,.~ Vrms, where T
and M are the temperature and the mass of the molecules, respectively, and
Vrms is the root-mean-square velocity of the thermal fluctuations.
In a turbulent flow, this small parameter ,~/L does not exist since all
scales, from the smallest to the largest corresponding to the mean velocity
field, are strongly coupled. Nevertheless, a relation similar to (1.1) intro-
ducing the concept of the turbulent viscosity YT,

2.~
(v~vj) = - v ~(vj vi + vi vj) + - - - - 0 (1.2)
3

has been widely used in turbulence modeling. In (1.2) v is the velocity of


fluctuations in the turbulent fluid. In this formulation, it is tempting to
define VT'~VrmsL, where Vrms is the mean value of the turbulent velocity
fluctuations. However, the length scale L is not easily defined in general. To
avoid the problem of defining L, the turbulent viscosity can be found using
dimensional considerations. Assuming that the only parameters needed to
describe the flow are the local values of the mean kinetic energy 2~(7"and
dissipation rate of energy g, the turbulent viscosity is
o,~ 2
VT=C~,-- (1.3)
g

where c, is a dimensionless constant.


To use (1.3) one must have equations of motion for o,~ and g. The two
dominant source terms in the g-equation are O(R~2), where R r = ~2/gv o
is the turbulence Reynolds number. This is not surprising, as ~ depends
on the small-scale velocity fluctuations with wave number near the
RNG Turbulence Models 37

Kolmogorov cutoff. However, for large RT, these O(R~ '2) terms cancel
exactly at leading order to produce a net O(1) effect.
The sum of the source terms in the ~-equation is usually modeled by
an O(1) term responsible for the dissipation of ~, and an O(1) term
responsible for the production of J. For example, the dissipation of ~ is
usually taken to be

o~2
C~2 - - (1.4)

by dimensional analysis. The dimensionless coefficient C82 is independent of


R r for large RT.
The exact cancellation of O(R~ 2) terms in the g-equation as Rr--' oo
follows from intuitive scaling arguments (Tennekes and Lumley, 1972), but
a stronger theoretical basis for this cancellation, as well as for the model
terms, would be most welcome. Here we develop the dynamic renormaliza-
tion group and the e-expansion in an attempt to place the high Reynolds
number ~ - o~ equations on more solid ground.
The ~ - o~ equations are derived by averaging the basic "micro-
scopic" Navier-Stokes equations over the fluctuating velocity field. In the
RNG theory, the small-scale velocity fluctuations are driven by a universal
random force f. The force f represents the average effect of the large-scale
features of the flow, including the effect of initial and boundary conditions,
which cannot be treated in a homogeneous analysis. After the elimination
of small scales is complete, the force is dropped from the resulting equa-
tions of motion and initial and boundary conditions are restored.
The above procedure implies that the small-scale velocity fluctuations
are generated by hydrodynamic instability which occurs at the large scales
and strongly depends on large-scale features such as boundary conditions
and geometry. It assumes that the small scales, produced by the nonlinear
terms in the Navier-Stokes equations, are relatively weakly influenced by
the details of the large-scale structures.
An effort to derive the ~7~- ~ equations using RNG methods was
previously made by Yakhot and Orszag (1986). However, Smith and
Reynolds (1992) pointed out several problems with the derivation, the
most important being that the method did not yield a term responsible for
production in the ~-equation. These findings have led us to reexamine the
derivation, in which some important issues were overlooked. For example,
the inclusion of the random-force contribution to g-production leads to
cancellation of all O(R~ 2) terms generated by the RNG procedure.
We show that the renormalized 8-equation contains the term (1.4)
38 Yakhot and Smith

with a coeffient Ce2 in the range of the data for the decay of isotropic
turbulence. The introduction of a mean flow leads to O(1) contributions to
or-production. We found that a Reynolds decomposition of the source
terms into mean and fluctuating velocities is necessary to derive production
terms, implying that the artificial force does not reflect the direct influence
of mean shear on the small scales. RNG analysis then predicts an
or-production term of the form commonly used, and its coefficient Ce~
agrees well with the data for homogeneous shear flow. This method also
predicts another term which may contribute to ~-production, with
tensorial contractions different from the standard production term. Using
an energy spectrum tensor model for weakly anisotropic flow (Shih et al.,
1990), the coefficient of this term approaches zero when the model constant
is chosen to match the rapid pressure-strain closure model of Launder,
Reece, and Rodi (1975). However, the two production terms are equal
when the model constant is the value found in the RDT limit. Therefore,
the new production term may have interesting consequences for rapidly
strained turbulence, and perhaps also for other kinds of turbulence for
which the "standard" model does not work well, and its behavior should
be explored.
In Section 2 we introduce the basic equations on which the RNG
averaging procedure will be applied to derive the J7~- ~ model. Section 3
provides a brief review of the theory applied to the forced Navier-Stokes
equations, with emphasis on the meaning and properties of the e-expan-
sion. The RNG model equations for ~ and ~ are derived in Sections 4
and 5, respectively.

2. T H E M O D E L

We consider an infinite domain in which a Newtonian fluid is stirred


by a Gaussian random force. The force is defined by its correlation function
in wavevector and frequency space,

(L(k, co)fs.(k', ~ ' ) )


=2Do(2rc)a+lk-YPq(k)6(k +k')6(co+co'), AL < k < ~
=0, As<k<~AL (2.1)

where k is the wavevector, co is the frequency, k = Ik[ and d is the number


of space dimensions. The parameter y is chosen to give the Kolmogorov
form of the energy spectrum and in three dimensions y = d = 3. Statistical
homogeneity in space and time is guaranteed through 6(k + k')6(co + co').
The projection operator P,7(k)= ~ij-kikj/k 2 makes the force statistically
RNG Turbulence Models 39

isotropic and divergence-free. In the limit of an infinite Reynolds number


AL ~ O , but AL ~>A s ~ O .
The forcing function (2.1) reflects the fact that turbulence is usually
driven by hydrodynamic instability with the most unstable mode at k = A L.
The traditional view is that this primary instability generates velocity fluc-
tuations with k > A L such that there is a direct energy cascade to small
scales. The energy cascade may lead to generation of the universal random
force (2.1), but this has not yet been demonstrated rigorously (Migdal et
al., 1990).
The production mechanisms in turbulence most often depend on
initial and boundary conditions, for example, the lifting of vortices in near-
wall turbulence and the generation of plumes at the heated surface of a
convective system. Thus it is known that turbulence is not primarily
produced at large scales. On the other hand, we know that we can generate
turbulence by stirring the large scales. The model of large-scale driven
turbulence is only one of the many possibilities of turbulence generation.
It is not clear how this model is related to reality.
Another way to generate turbulence is with the force (2.1). This has
been demonstrated by Panda, Sonnad, and Clementi (1989), who numeri-
cally solved the forced Navier-Stokes equations subject to (2.1) with a
high-wavenumber cutoff to allow for the development of a dissipation
range. They showed that the statistics of the resultant flow field were in
good agreement with experimental data for turbulence.
Thus a reasonable model of turbulence is the forced Navier-Stokes
equations,

jvj ,= -V,p+ oVyj i+ f,

Vivi = 0 (2.2)

where the Gaussian force f is given by (2.1) and the density p has been
absorbed in the pressure p. A model similar to (2.1) and (2.2) has been
extensively used for theoretical investigation of hydrodynamic turbulence
(Yakhot and Orszag, 1986; Forster et al., 1977; DeDominicis and Martin,
1979; Fournier and Frisch, 1983). The only new feature of the model (2.1)
and (2.2) is the infrared cutoff of the random force, ( f J j . ) = 0 when
0 < k < Ac. This property is needed if we are interested in the derivation of
the equation for the mean rate of energy dissipation g.
The dynamical equations for the kinetic energy per unit mass
X = viv]2 and the homogenous part of the instantaneous rate of energy
dissipation per unit mass 8---vo(Vjvi) 2 are derived using (2.2),
40 Yakhot and Smith

8o~f
Ot + viV~X = - g + v0VgV~zf - V~(v~p) + v~fi (2.3)
Yflg T1

- ~ + vgVS = 2Vo(Vjvi)(VJ,-) - 2vo(Vjv,)(gjv,)(V,v,)


T2

- 2vg(VjV,v,) 2 - 2vo(Viv~)(V~Vjp ) + voV,V~g (2.4)

In the equation for g driven by large-scale features such as boundary and


initial conditions, TI balances T2 at leading order. As we shall see, for the
steady-state flow (2.1) and (2.2) sustained by the force f, T2 is balanced by
both T1 and the random-force contribution to g-production ~e.
We seek equations for the mean values V - - ( v } , ~ - (Y{'), and
g - ( g } , averaged over an ensemble of the fluctuating velocity field. To
find such equations, we shall use the dynamic renormalization group and
the e-expansion. Since the renormalized equations for ~ and J may not be
trivially related to the renormalized Navier-Stokes equations, the RNG
procedure must be applied to all of the equations (2.2)-(2.4).

3. FIXED POINT OF THE NAVIER-STOKES EQUATIONS


The details of the procedure have been described in various publi-
cations (Yakhot and Orszag, 1986; Smith and Reynolds, 1992; Forster
et al., 1977; DeDominicis and Martin, 1979; Fourier and Frisch, 1983;
Dannevik et al., 1987). Here we would like to present a discussion of the
meaning and properties of the e-expansion. First one introduces an ultra-
violet cutoff Ao where the viscosity is Vo and where the scale elimination
will begin. The results of the theory are universal (independent of A o and
Vo) in the limit k < Ao. Thus they apply to the high Reynolds number limit,
in which there is large scale separation, AL < Ao.
The Fourier transform of (2.2) leads to

t3,(f~) = G~ ~(~) - ~i2o G~ d~


Pem,,(k) f z3,,,(~) ~3,,(~:- ~) (2~) a+l
(3.1)

where ~ = (k, o~), G~ = ( - io~ + Vok2)-1, 2o = 1, 0 < k < Ao, Piton(k) =


kmPin(k) + knPim(k) and d~ denotes the (d+ 1)-dimensional integral over
the wavevector and frequency components of ~. One introduces a variable
cutoff A1 =Ao e-r and derives the equation of motion for the large-scale
Fourier components of the velocity field ~<(k, o~) with k ~ A l and o~ 4 0 .
The steps of the procedure include a perturbation expansion of ~ > (~) with
A1 < k < do in powers of the local Reynolds number at Ao. [See Smith and
R N G Turbulence M o d e l s 41

Reynolds (1992) for a detailed discussion of the perturbation expansion of


r used at each iteration of the scale removal procedure.] Following
substitution of the perturbation expansion for ~ > (1~), A~ < k < A0, into the
equaltion for ~ <([), k ~ A~, one takes an ensemble average over the force
~>(~) with A1 < k < A0. The average is performed assuming that all large-
scale quantities are the same for each realization of the ensemble. This
condition on the large-scale velocity ~ < ([) is approached for k ~ A~.
After averaging over the small-scale force field, the nondimensional
equations of motion are written in real space as

- - A V ftlV~Vju? = (fl)? --~IViP < ~-VjVjv?


Ot
(1 - e -~)
+ ~ Ba~(e)2~O((vjVj)"v~) (3.2)
/3 n=2

2
The nondimensionalization is with respect to the time tc=lc/Vl, length
lc = 1/A1 and velocity scale Vc= [DoSd/(2~) d] 1/2lc/V]/2 , where vl is the effec-
tive viscosity acting on modes k ~ A 1 and Sd is the area of the unit sphere
in d dimensions. The last terms on the right-hand side of (3.2) are the
higher-order nonlinearities generated by the procedure. The dimensionless
coupling constant 21 = vcIc/Vl is an effective Reynolds number of the flow
at A~, averaged over the small-scale fluctuations with A~ < k < A0,

;,~_DoSd 1 (3.3)
(2~Z)a V~A]

where /3= 4 + y - d and y = d = 3 for a Kolmogorov spectrum in three


dimensions. The coefficients Ba(/3) are functions of the space dimension d
and of the parameter/3. The effective viscosity Vl is

Vl=Vo+Aa(e)Vo[f~2+Cd~(/3)~4+O(~6)](e"r- 1) (3.4)
/3

where

d2-d-e
Ad(e) = (3.5)
2d(d+ 2)
DoSa 1
202= (2~) d v3oA~o (3.6)

and the coefficients C~(e) depend on the dimension d and e.


42 Yakhot and Smith

The effective force fl in (3.2) is

fl = f + F (3.7)

where the induced "eddy" force F has the two-point correlation function

(~,.(~)pj(~,)>=2Dok2pu(k)6(~+~,) _ _ 1 (e(*+Y+2)'-l) ~ ~b.(a)2oa


-2.
A~ +2 e+y+2 .=x
(3.8)
Again the ~b,a(e) are dimensionless coefficients. In the limit k--+0 the
induced force F, with (PiPj> oc k 2, is small in comparison with the bare
force f for A L < k < Ao. However, in the interval 0 < k < AL, the induced
force is not small and it accounts for the energy generated by the small-
scale velocity fluctuations. This effect is called backscattering, or an inverse
energy cascade. The equations of motion given by (3.2)-(3.8) are formally
exact.
Now, consider the case of small e. In this case, the problem (3.2)-(3.8)
can be solved exactly. The differential relation for the effective viscosity v(r)
is
dr(r) OoSa[A~ e--+O (3.9)
dr (2re)a v2(r) A~(r) '

where A(r)=-Ao e-r. Notice that the dimensional factor A-~(r) is common
to all orders in ~ and therefore need not (and must not by dimensional
analysis) be expanded in powers of g. From (3.5), AO=(d2-d)/
[2d(d+2)]. Neglecting the O(e) correction, the solution of (3.10) with
initial conditions A(0)= Ao and v(Ao)= Vo is

v(A)=vo[1 3A~176 A ~ ]1/3 (3.10)

with (3.10) strictly valid only for Ac ~ A0 where it is independent of Vo


and Ao.
Using (3.10), one finds for the coupling constant 2(A)

[DoSd 1 11/2
2(A) =- L(2rcVv3(A)--h--~J

=2oeZr[ l +3A~176g (-~)a(A-~ Aoe)1-1/2


- (3.11)

where (3.11) is strictly valid only for AL 4 Ao. We emphasize that ~, based
on the modified viscosity v, is an effective Reynolds number of the large-
RNG Turbulence Models 43

scale flow at A, averaged over the small-scale fluctuations A < k < Ao. In
the limit A L < A ~ A o and for e > 0 ,

f~2 DoScl 1 e
( 2 ~z) a v , A ~ ~ -~-~6d~ O (3.12)

Thus, all the higher-order nonlinearities proportional to higher powers of


)~ are negligibly small when e--+ 0.
Expressions (3.10) and (3.11) are the correct leading-order terms in
an expansion in powers of e (but the leading term may not be a good
approximation for sufficiently large e). Indeed, keeping higher-order terms
in the expansion of Aa(e), one finds instead of (3.12)

,~2 8
--+-f-~a+ O(e2) (3.13)

Then, to be consistent, one must account for the higher-order contributions


proportional to 2-2,o , n = (2, or) in the expression (3.4) for vl, and all
higher-order nonlinearities generated in the Navier-Stokes equations.
Thus, keeping terms to lowest order in the e-expansion, (3.10) with A ~ is
correct even when e is not small.
In the same order of the e-expansion, the induced force is Gaussian
(Forster et al., 1977) and given by ( k ~ A ) ,

d 2- 2 ~2
(F'(~)/'J(~') > = 2D~ ik2pu(k) ~(~ + ~') 20d(d+ 2) A y+2 (3.14)

This forcing function is not negligible in the range 0 < k < AL where the
bare force is zero.
When e ~ 0 , one may eliminate all modes r with q > k without
including the higher-order nonlinearities. In this case, the scale-dependent
viscosity is obtained from (3.10) as

(3A~ (3.15)
v(k)..~\ 2 e / 1_(2~) a j

for AL < k ~ A o . Using (3.15), the energy spectrum in this wave-number


range is found to be (Yakhot and Orszag, 1988)

1 [2DoSa] 2/3 (2A~ (3.16)


= [_(-577 VJ \57/
44 Yakhot and Smith

The energy spectrum for k < A L is determined (Forster et al., 1977) using
the correlation of the induced force F', the eddy viscosity v(AL) and the
coupling constant ~(AL) (see Forster et al., 1977, model C),

E_(k) =Ak 2 (3.17)

where A =A(AL) is found using (3.14).


The theory extends the above results for e ~ 0 to the case of e = O(1).
The Kolmogorov spectrum corresponds to (3.15) and (3.16) with e = 4,

[2D~ 1/3 k -4/3 (3.18)


v(k) = 0.42 [_ (2r~)a _]

F2DoSa] 2/3 k - s/3 (3.19)


E+(k) = 1.17 [_ (27t)a j

To this point, the theory is an extrapolation of conventional ideas into


the range of O(1) expansion parameter e. Dannevik, Yakhot, and Orszag
(1987) went further to find the unknown amplitude Do from overall energy
conservation by the nonlinear terms. Using the fact that the Reynolds
number based on the effective viscosity is O(e 1/2) at lowest order in the
~-expansion, they solved (3.2) to second order in the e-expansion. In this
case, the so-called E D Q N M equations introduced by Orszag (1970) are
derived. These equations express the mean dissipation rate g in terms of
the energy spectrum and the effective viscosity,

g = g(v(k), E+ (k)) (3.20)

Combining (3.18), (3.19), and (3.20), they derived the relation

1.59J = 2Do Sa
(27r)a (3.21)

Substituting (3.21)into (3.18) and (3.19), one finds for A L < k


E+ (k) = 1.6192/3k-5/3 (3.22)
v(k) = 0.4991/3k - 4/3 ( 3.23 )

The above results are based on the lowest order of expansion in powers of
e = 4, and agree well with the experimentally known parameters. This
suggests that the leading term in the expansion is adequate even with e = 4.
We do not have a mathematical proof of the validity of the theory at
low order in e. However, some recent works give hope that a firm mathe-
matical foundation for the theory is not impossible. Recently, Majda
RNG Turbulence Models 45

and Avillaneda (1990) considered the model problem of a passive scalar


diffusing in a random velocity field. They were able to solve this problem
exactly and compared the outcome with the RNG results obtained at
lowest order in e. In the range of relevant parameters, when the velocity
field is close to a Kolmogorov spectrum, the exact results are extremely
close to the outcome of RNG. A different problem involving a passive
scalar was considered by Yakhot and Orzsag (1990), for which they con-
structed an approximate infinite series for the effective diffusivity,

)~= Zo ~ ~O.u(~)~n (3.24)


n=0

An exact Borel summation of this series differs from the low-order RNG
conclusion only by 10 %. These facts, of course, cannot be used as a proof
or even as mathematical justification of the theory, although they give a
hint that the basis for success of the z-expansion in the theory of turbulence
might be of a similar nature as that in the theory of phase transitions.

4. THE ~-EQUAT1ON

4.1. The Fourier Representation


To derive the equation for ~ = (Vo(Vjvi)2), one must develop the
RNG theory for (2.4) together with the forced Navier-Stokes equations
(2.2). To begin, we use the result of Yakhot and Orszag (1986) that the
contribution from pressure-velocity correlation term is equal to zero to
second order in 2o. We also use their result that the advection of g by the
velocity field at small scales renormalizes as an effective diffusivity. Both
the Yakhot-Orszag and the present derivations neglect the high-wave-
number contribution to the nonlinear interaction of v with the source terms
T1, T2, ~e and the pressure-velocity correlation. These contributions are
assumed to be small compared to the terms with modified transport coef-
ficients. With this assumption, the analysis of T1, T2, and ~e decouples
from the analysis of the rest of the g-equation.
The Fourier transform of (2.4) minus the pressure-velocity correlation
is

[gO(i)]-~ ~ ( ~ ) = _i2ok, f f,(r ~(~ -r (2re)d+ 1


y,(}) - Y~(~)+ ~(})

(4.1)
46 Yakhot and Smith

where ~(~) is the transform of 8, gO(i) = (-i~o + vok2) -1 and

d~ at)
Ya= -aoi f qjQj(kz-qt-Qt) 6~(~)oz(t))o,(~-~l-t)) (2rc)2a+2 (4.2)

d~
Y2=bo f q21k-ql2 O~(~l) ee(~-~t)(27t)a+~ (4.3)

with ao = 2Vo and bo = 2vo2. The random-force contribution to the produc-


tion of ~ is
d~
~e= - c f q~(k~-q~) 6j(~l) Fj-(~-~l) (2x) (4.4)
a+l
with Co= 2Vo.

4.2. The Scale Elimination

Again one introduces the ultraviolet cutoff Ao where the viscosity is Vo.
We look for solutions to the RNG differential equations that are inde-
pendent of Ao and Vo. To eliminate the small-scale velocity fluctuations
from the problem (4.1)-(4.4), one introduces the modes 0>([) and ~>([)
with Aoe-r < k < Ao, and 0<([) and ~<([) with A L < k < Ao e-r. Then, for
example, we have

d~
Y2= Y~ + bo f q2[k_ql2[20<(~)~>([_~)+O~(~) O~([-~)] (270d+l
(4.5)

~ = ~ ~ - Cof qi(kt- qi)[0;< ( ~ ) f f (~ - tl) + i f (~1)f ~ (~ - ~t)


d~ (4.6)
+ U ( ~ ) f T ( ~ - ~)] (2~)~+1
where Y~ and ~ { are given by (4.3) and (4.4) with 0 everywhere replaced
by 0 <. Next one repeatedly substitutes the forced Navier-Stokes equations
for 0 > into (4.5) and (4.6). This formally eliminates the modes 0 > from the
problem.

4.3. Cancellation of O(R~/2) Terms


First consider the zeroth-order-in-2o contributions from Y2 and ~e.
We are interested in the limit ~ ~ 0 in which mean values are determined.
In the limit ~ ~ 0, Y2 and ~e give, respectively,
RNG Turbulence Models 47

-- h ( 2D~ yPii(q) ddq dO


6yo- ~,oj if22-k v2q4 (2~) a 2--~- (4.7)

6~eo= Cof 2D~ ddq dO (4.8)


S~+v-~ (2~r)a 2z~

The frequency integrations in (4.7) and (4.8) are carried out over the
interval - c ~ < (2 < oo. Upon performing the frequency integrations, one
sees that 6yo- 6~eo= 0 and thus that the zeroth-order-in-2o corrections to
(4.1) cancel. In the nondimensional equations they are zeroth order in 2,
where )~ oc el/2 to lowest order in the e-expansion.
The integrals (4.7) and (4.8) scale as k 2, and are thus dominated by
high wave numbers near Ao. Their Reynolds number dependence is deter-
mined from the relation between Ao and the Kolmogorov wavenumber,

Ao = 7 (4.9)

where 7 is a constant. Using (4.9) and Do oc ov by (3.21), one finds


O~2 ~2
yo= o 2 (4.10)
voAo -~

The removal of scales Ao e-r < k < Ao generates corrections to (4.1)


from both Y1 and Y2 of the form

~d(e) 2DoS d 1 [(Aoe-r)2-~-A~ -~] d~ (4.11)


(2~) d vo (g--2) f q2~/<(~) ~ ( _ ~ ) (2n) a+l

in the limit ~--+ 0. In the case of Y1, these appear at first order in the
expansion in powers of 2o, while in the case of I12, they appear at second
order. In addition, Y1 will produce

2DoS a l [(Aoe-r) 2 - - A 2-~]


i~Pa(~) (2=) a v~ ~---5

dO
x f QjO2(Q) v~5(~) ~ ( - ~ - Q) (2n)2a+ 2 (4.12)

at second order in 20. Scaling analysis of (4.11) and (4.12) shows that they
are O(R~2), reflecting the fact that g is dominated by fluctuations in the
dissipation range. However, Smith and Reynolds (1992) showed that all
of these corrections also exactly cancel Thus RNG provides theoretical
48 Yakhot and Smith

support for expressing the equation for g in terms of O(1) inertial-range


parameters.
One sees that when ~,e is included, all O(R~ 2) terms cancel at low
order in e. The contribution from ~'e is therefore necessary for a consistent
theory of the inertial range. The consequences of including Ne were
previously overlooked (Yakhot and Orszag, 1986; Smith Reynolds, 1992).

4.4. Self-Similarity at Low Order in c


To second order in the ~o-expansion, the g-equation is self-similar. By
self-similar we mean that no terms different from those in the original
8-equation are generated by the averaging procedure, and only the original
transport coefficients are modified. We believe that this self-similarity holds
to all orders in A0 (or e in the nondimensional equations). However, we
do not need this result for the theory based on the lowest order of the
~-expansion.
Since (4.1) is self-similar (at low order in 20), the equation for g<,
averaged over the velocity modes with Aoe-'< k < Ao, is

+ v<VJ < = - al(Vjv? )(V,v < )(Vjvt < ) - bl(VjVml.)? )2


~'t
VjO IVIVjO*'e< -~-Cl(Vjv<)(Vjf < (4.13)

where Vl is given by (3.4) in the limit e ~ 0, al = a0 + ha, bl = b0 + 6b and


c1=co+6C. The inverse Prandtl number ~1 was previously found by
Yakhot and Orszag (1986). Iteration of the averaging procedure leads to
the renormalized coefficients v = v(A) given by (3.18), a = a(A), b = b(A),
and c = c(A) for A <~Ao. For high Reynolds number flow corresponding to
Vo/V--*O, Yakhot and Orszag (1986) derived the value ~= 1.39. We note
that the coefficient of the pressure-velocity correlation is not renormalized
(Yakhot and Orszag, 1986) up to second order in 20. In the final equation
for the mean value g = ~ < ( [ ) for on--.0 and k ~ A L ~ O , we drop the
random-force contribution to the production of g and restore initial and
boundary conditions. Thus we now give only the derivation of a and b.

4.5. The Dissipation of


The first nonvanishing corrections to Y: appear at second order in the
perturbation series in powers of 20,

Y2 "= Y~ + 6Y2 (4.14a)


RNG TurbulenceModels 49

where for ~ ~ 0 we have

2DoS d 1 (e e r - 1) f d~
(~2 :/3bo (2zt)a voAo3~ ~ j q4t3/<(q) g/<(-q) (2n)d+-----------7(4.14b)

with the coefficient/3 > 0. The value of/3 computed at lowest order in e is
unimportant, and this shall be demonstrated in what follows. Notice that
the corrections to Y2 have the same form as Y2 itself.
Combining Y2 and 6Y2, we have in the limit ~ ~ 0

d~ ~
Y2 = hi f q4t3/<(I]) t3< (--I]) (27z)a+ (4.15)

bl=bo E1 + Ad(0)22 t416


The differential recursion relation for b(r) is then

db fl b(r) dr(r)
(4.17)
dr Ad(0 ) v(r) dr

where (3.9) has been used. The recursion relation (4.17) is to be solved
subject to the initial conditions b ( 0 ) = 2v 2 and v(0)= v0. The result is

b = 2v 2 '/Aa(~176 (4.18)

The solution independent of v0 is obtained only when /3 = 2Ad(0). Thus,


b ( r ) = 2v2(r). The equality /3= 2Aa(0), following from (4.17), is a conse-
quence of the symmetry of the problem. It will be shown in Sec. 4.8 that the
relation b(r)= 2v2(r) can be derived directly from the fixed-point Navier-
Stokes equations for the large scales.

4.6. The Production of


Now we consider the term Y1 (4.2), which is the Fourier transform of

T1 = 2vo(Vjvi)(Vjvz)(V, vi) (4.19)

Smith and Reynolds (1992) showed that Y1 does not generate corrections
similar to itself, and this it is small and may be neglected in comparison to
the terms with enhanced transport coefficients. However, we expect that Y1
is at least partly responsible for the O(1) contribution of g-production.
This points to a shortcoming in the representation f, as it does not reflect
the O(1) or-production due to the effect of mean shear on the small scales.

854/7/1-4
50 Yakhot and Smith

Here we investigate the consequences of incorporating the effect of


mean shear by performing a Reynolds decomposition of T1 given by (4.19)
into mean V and fluctuating v velocities. Notice that a Reynolds decom-
position of T2 does not lead to any new contributions to the equation for
the dissipation rate of the fluctuations v. From (4.19) we must consider the
additional contribution to Eq. (4.13) for g < of the three terms
T3 T4

- - 2 1 ) 0 ( V j Vi)(gjJ)l)(VlVi) -- 2vo(Vj V,)(V;vi)(V,vi) - 2vo(Vjvi)(Vjv,)(V, Vi)


(4.20)

The ensemble average of the first term vanishes in homogeneous flow, and
thus we argue that it should be relatively small in regions of high Reynolds
number flow. Indeed, (T3) and (T4) are the only other terms to appear
with (T1) and (T2) on the right-hand side of the exact transport equa-
tion for ff in homogeneous flow. We hereafter omit the brackets ( ) and
refer to the ensemble-averaged quantities as T,.
RNG predicts that the O(1) production of i is due to T3 and T4. As
we shall show, the magnitudes of T3 and T4 are determined by the small
scales through effective transport coefficients, while their symmetry is deter-
mined by the large scales. This is consistent with the analysis of Durbin
and Speziale (1991), which shows that the small scales respond to a mean
rate of strain. Some recent Y f - g models (Mansour and Shih, 1989;
Durbin, 1990) also attritude O(1) g-production to T3 and T4.
First we will consider T4, which has the same tensorial contractions as
the if-production term commonly used. Using the RNG scale elimination
procedure, we will average over the high wave numbers AL<k <Ao. In
this manner, we will find T4 in terms of the energy spectrum tensor E~ (k)
of the low wave numbers 0 < k < AL, which we allow to be anisotropic.
Assuming homogeneous flow in which VI V~ is constant, we use the
RNG method to evaluate

Y4 = ao( (Vjvi)(VjVl) ) (4.21)

where a0 = 2v0. One finds for Y4 = lim~ ~ 0);4(~)

Y4 = Y~ q- c~Y4 (4.22a)

where
(e ar- 1) a~
6y4=Tao i2 f q26~(~) 6~<(-~) - (4.22b)
(2re)a+l
RNG Turbulence Models 51

with ~:> 0. Thus it follows that

y4=al
d~ ~
fj q26~<(q) 6t<(-~) (27z)d+ (4.23)

with the coefficient a~ given by

7 Ad(0)i2 (4.24)
al=a~ I+A--~ e

The differential recursion relation for a(r) is then

da(r) 7 a(r) dr(r)


dr - Ad(0 ) v(r) dr (4.25)

with a(0)=2v 0 and v(0)=Vo. The Vo-independent solution of (4.25)


demands 7 =Ad(0) and a(r)= 2v(r).
We must also consider the contribution to g-production from T3 in
(4.20). We use the RNG procedure to evaluate

Y3 = a'o( (Vjv~)(V~v~) ) (4.26)

where a;=2vo. In exactly the same manner as above, one finds the
renormalized coefficient a'(r)= 2v(r).

43. The Equation for


After elimination of all fluctuating modes, the equation for the mean
dissipation rate is
T2
~3~
~--~-~ ViVi~=ViO~YVi ~ - 2Y 2 f d~
q46~<(~)t~S(-~) (2~z)d+,

T3
dfi
- 2v(Vj Vi) f qiqje? (~) e # ( - r (2=)d+ 1

T4
r
d~ t
- 2v(Vj V~) f q2gr (~]) e / ( - ~) (2n)d+ (4.27)

where g=o~(K) and V=r for o ~ 0 , k--*AL--*O.


The wave-vector integrations in (4.27) are carried out over the interval
O<k<~AL. The knowledge of the energy spectrum in this interval will
52 Yakhot and Smith

enable us to evaluate the integrals and obtain /'2 and T4 in familiar forms.
The importance of this interval for the derivation of the equation for the
mean dissipation rate was first addressed by W. C. Reynolds (1976). As we
shall discuss in Sec. 4.9, T3 is likely to be small for turbulent flows that are
well-modeled by the ; , ~ - g equations, but may be a useful addition for
others.

4.8. The o~-Equation from the Fixed-Point Navier-Stokes Equations


In this section, we show that the coefficients of T2, T3, and T4
in (4.27) can be obtained directly from the fixed-point Navier-Stokes
equations
Ov~ + vfVjv? = - V i p < + v(A)VjVjv? (4.28)
Ot

One can approximate ~ using the modified viscosity v(A)


e ~ Yo(Vjl)i) 2 ,~ v(A)(Vjv~ )2 (4.29)

where v < is the velocity field averaged over the high wave numbers (small
scales) A < k < Ao with AL < A <~Ao. The approximate equality in (4.29)
can be checked easily using the corresponding integral
A
g ~ 2 v ( A ) IA k2E+(k) dk (4.30)
L

where S E+(k) dk = (v~v~ )/2. Substituting E+(k)= 1.61~2/3k-5/3 into the


right-hand side of (4.30) gives 1.18g for A L ~ 0 . The overestimate by
the integral in (4.30) could be because the dissipation-rate spectrum of the
large-scale flow may begin to fall off from k 1/3 well before A, where the
energy spectrum falls off from k-5/3.
The equation of motion for g based on the large-scale velocity field
and modified viscosity is derived by taking the time derivative of v(Vjv~ )2,
followed by substitution from the fixed-point equations (4.28) above.
This procedure leads to the same coefficients of T2, T3, and 7"4 given in
(4.27). However, the equality form of (4.29) is implicitly assumed by this
procedure.

4.9. Closure of the o~ Model Equation


The g model-equation (4.27) is closed using the energy spectrum
tensor Eij (k) oc k 2 in the interval As<k<AL (see Sec. 3). As we will see,
the analysis depends only on the power of k in E~ (k).
R N G Turbulence Models 53

Consider first T2,

7"2= 2v2(Ar) JAo q4~<(fl) v<(-- ft) (2~)a+d~ = 4v2(AL) fAAfq4E_(q) dq


(4.31)

We also use a relation similar to (4.30) for the range As < k < At,

e ~ 2v(AL) "jALq2E_(q)dq (4.32)

Using E ( k ) = A k 2 and assuming equality in (4.32), one finds for A , ~ 0

T2 = ~ v(AL) A~g (4.33)

The previously derived (Yakhot and Orszag, t986) RNG relations for fully
developed turbulence

3F = IOv2(Az)A~ (4.34)

and
~(~2
v(AL) = 0.085 - - (4.35)

may then be used to eliminate v(AL) and AL. One finds


~2
T 2 = 1.68-- (4.36)

Thus the term responsible for the dissipation of ~ is recoverd in the


familiar form -Ce2~2/J~. The coefficient Ce= 1.68 is within 10% of the
value Ce2 = 11/6, which is found (Reynolds, 1976) for isotropic decay using
a low-wave-number spectrum that varies a s k 2. Notice that ~ is the kinetic
energy in the inertial range, and that the quantity SA~E ( k ) d k does not
have physical meaning in the model. We use the form of the energy spec-
trum E_(k) oc k 2 only to fix the flux of energy at k=Az.
Similarly, for T4 we write

~L det
T4=2v(AL)(VjVi) , q20<(~) 0j<(_~) (2~) a+t

AL
= 2V(AL)(Vj Vi) fAs q2E~ (q) dq (4.37)
54 Yakhot and Smith

where ~E~v(k)dk=(v~v~) in the range A~<k<AL. We take


E,j(k) oc k ~' as suggested by (3.8), but relax the constraint of isotropy
enforced by P0(k). We can write T4 in the form C<)~/a$" by using the
definition of ) ,

~--- - - ( V j Vi)(vivj) = --(Vj Vi) AsL Eij ( q ) d q (4.38)

Using (4.38), one can verify that


m

r4=-76 v(AL)A2) = _ 1.42) ~ (4.39)

where we have again used (4.34) and (4.35).


It is not apparent how one should close T3. However, a model energy
spectrum tensor for weakly anisotropic flow (Shih et al., 1990) may be used
to close both T3 and T4 and thereby provide estimates for their relative
sizes. Using the energy spectrum given in Shih et al. (1990), one finds

T3 2(27 - 1 )
(4.40)
T4 5 + 2 ( 1 - ~ )

where ~ is a model constant which affects the model's region of realizability.


The region of realizability on the invariant map of the anisotropy tensor
is the largest for 7~0.5, in which case T3~0. The value ~=0.527
corresponds to the rapid pressure-strain model of Launder et al. (1975).
The value ~ = 1.5 is found in the limit of RDT and has a small region of
realizability. The ratio for y = 1.5 is found from (4.40) to be T3/T4= 1,
indicating that they are equally important in regions of extremely rapid
strain. The ratio decreases from 1 to 0 as ~ decreases from 1.5 to 0.5, while
the region of realizability increases.
We next consider the opposite case of a strong homogeneous shear
flow in which eddies are stretched in the direction of the shear, and argue
that T 3 is small compared to T4. In homogeneous shear flow, the ratio of
T3 and T4 is

T3 (v,ViVjvl)
(4.41)
T4 (viVlVt~)
In a flow where the dominant structures are elongated in sheets or
"spaghetti," the mixed-derivative contribution from T3 with i # j is
relatively small since it involves derivatives in the direction parallel to the
elongated structures.
RNG Turbulence Models 55

Although T3 may be unimportant for slowly strained turbulence, it


will contribute significant production in regions of rapid distortion and
possibly other special cases. It can perhaps be used to improve the o~ model
equation in these cases. For example, as suggested by (4.40) for weakly
anisotropic turbulence, it may be possible to incorporate increased
production due to T3 by allowing the coefficient of T 4 to vary with the
mean strain rate. In ~ - ~ calculations of near-wall turbulence by Durbin
(1990), this kind of correction of Cel was necessary to account for
increased production due to local anisotropy near the wall. Thus the coef-
ficient of the standard ~-production term would increase from 1.42 in
regions of rapid strain. The value C~1 = 1.42 is in good agreement with the
typical value for slowly strained turbulence (Patel et aL, 1985) C~, = 1.44.
Finally, the high Reynolds number form of the RNG o~ model equa-
tion is written as

v ~ + VzVS = 1.42~
- 1:68 -:- + V/~vV;o~

- 2v(Vj V;) f q;qj~[ (~t) ~ (-~t) (2~)dt]d+' (4.42)

where v is given by (4.35) and ~ = 1.39 (Yakhot and Orszag, 1986).

5. THE ~ - E Q U A T I O N
The equation for the mean kinetic energy is derived by RNG applied
to (2.3) and the Navier-Stokes equations simultaneously. Again we use the
results of Yakhot and Orszag (1986) that the velocity-pressure correlation
does not contribute to second order in the 20-expansion, and that the
advection of Yl by the velocity field at small scales renormalizes as an
effective diffusivity. Contributions from the nonlinear interaction of v
with g and the random-force term are also neglected.
The Fourier transform of (2.3) minus the velocity-pressure correlation
is

Eg~ d~(~) = -i,~ok, f~;(ej)~(~-@) d~l ~(~) + ~ ( ~ ) (5.1)


(2rt)a+ 1

where Sr(l~)is the transform of ~ , again gO(~)= ( - i c o + v0k2) -1 and


a~
d{l~) = -e f qi(kj - qj) 6;(~) ~,(~ - ~]) (5.2)
(272) a+l
56 Yakhot and Smith

with eo = %. The random-force contribution to the production of o,@ is

. - ~ d~i (5.3)
~,(K) = h f v,(q) fi(q - K) (2~)d+ ~

with ho = 1. Following the same procedure as described for the g-equation,


we separate the modes ~<, ~<, ~>, and ~>, and repeatedly substitute the
forced Navier-Stokes equation for r >
At zeroth order in 2o, ok(~) and @(~) give, respectively,

C 2D~ daq dO
6~o=e (5.4)
J -- ig'2 + V o q 2 (27~) d 27~
f. 2Doq2- y p ii(q) ddq dO
a /o = h j 0 2 -t- v 2 q 4 (27r)a 27z
(5.5)

The corrections ~ o and ~ 0 grow logarithmically with wave number.


However, upon performing the frequency integrations over - oo < f2 < o%
one sees that 6 g 0 - 6 ~ 0 = 0 . Thus, the random-force contribution to
X-production balances the high-wave-number contributions from g.
Terms logarithmic in wave number generated by Y2 in the g-equation were
also found to exactly cancel (Yakhot and Orszag, 1986).
In is much easier in the case of oU than g to show that no new terms
are generated at low order in 20, and that only the bare coefficients are
modified. Thus we proceed directly to find the coefficient e in (5.2). At
second order in 20 and for ~ ~ 0, the correction to oe(~) is

64=aeo2DoSa 1 (e~r-1)f d~ (5.6)


(2~z)a voA
3
o
e q2~,.<(~) ~f ( - t ] ) (2g)a+ ~

where a > 0 . From experience with the g-equation, one sees that the
differential recursion relation for e(r) is given by

de(r) a e(r)dv(r)
m
(5.7)
dr Ad(O)v(r) dr
the solution of which is
e = v~ o/A~(O~v,/A~O) (5.8)

The vo-independent solution requires that o = Ad(0) and thus e(r)= v(r).
Next we recognize that the production of J f must be extracted by a
Reynolds decomposition of the velocity field, and can only arise from the
term vjVj)F = vivjVjvi. For the purpose of identifying the production term,
RNG TurbulenceModels 57

we think of the large-scale field as homogeneous, but anisotropic. This


leads to the inclusion of the J{-production term

=- -(vivj) V~ Vi (5.9)

RNG applied to (5.9) does not yield any new contributions to the s f <
equation, nor does it generate corrections to itself. However, N < cannot be
dropped from the equation for Y < because its coefficient of unity is not
small compared to the effective viscosity and diffusivity.
The equation for ~ is found for ~(~ = ~ < (~), g = g < (~), ~ = ~ < (~)
and Vi=f~<(~) in the limits co--, 0 and k - - , A L ~ O ,
a~
0---T+ v i V , ~ = - g + Vi~vV,s + ~ (5.10)

The complete RNG J ~ - g model for R r ~ oo is given by (4.35), (4.42),


and (5.10) with ~=1.39. Notice that RNG predicts ee=c~x-=l.39.
A "typical" value (Patel et al., 1985) for ~x- is e:c = 1.0.

6. CONCLUSIONS

We have shown that renormalization group theory provides a


theoretical basis for the exact cancellation of the O(R~ 2) terms in the
transport equation for the dissipation rate g. Furthermore, at low order in
the e-expansion, the equation for g averaged over small scales has the form
of the original equation for g with scale-dependent transport coefficients.
The calculation of the effective transport coefficients leads directly
to a model term responsible for g-dissipation of the form commonly
used. The additional introduction of a mean flow leads to model
terms responsible for ~-production. One of these has the standard form
proportional to the production of kinetic energy. The other is small
in slowly strained turbulence, but becomes comparable to the standard
production term in rapidly strained flow. It may be possible to improve
the performance of the g model equation in regions of rapid strain by
incorporating this new term.
The quality of the derived constants deserves discussion. The RNG
value Cel = 1.42 can be checked directly for the case of homogeneous shear
flow. Extensive investigation by Speziale et al. (1991) led to the conclusion
that Ce, = 1.4 gives the best results for homogeneous shear. The commonly
used value C~ = 1.44 is also close to the RNG value.
The value Ce2 = 1.92 has been widely used by modelers. The value of
58 Yakhot and Smith

this coefficient governs the rate of decay of isotropic turbulence, in which


case the model equations are

& = -g (6.1)

aO~ ~2
~ - = -Cg2 ~ (6.2)

From (6.1) and (6.2) it follows that


~'~ oCt -1/(c~2 1 ) = t - n (6.3)

Early experimental data (Millionshtchikov, 1939; Batchelor and Townsend,


1948; Tan and Ling, 1963) for the decay of turbulence behind a grid gave
n = 1.1, which led to the value Ce2~ 1.9. With improvement of experimen-
tal procedures to generate isotropic flow, the value of n was found (Comte-
Bellot and Corrsin, 1966) in the range n = 1.1-1.3. Recent numerical
simulations (Lee and Reynolds, 1985; Yakhot et al., 1988) on the decay of
isotropic tubulence gave values in the range n = 1.3-1.6. Thus the present
value Ce2 = 1.68 leading to n = 1.47 gives a resonable decay exponent.
Our value of ~ e = 1.39 differs from ~e=0.77 commonly used by
modelers to fit the value of the von Karman constant K. The yon Karman
constant can be estimated from the standard model equations assuming a
logarithmic mean profile and a constant value for the dimensionless mean
kinetic energy,

~ = (Ce2--~-Fgt~ 1/2 (6.4)

where

= c ; 1/2 (6.5)

Taking the RNG values c,=0.085, e = 1.39, Ce2= 1.68, and Cex = 1.42,
(6.5) gives the low value x=0.23. With the standard values ee=0.77,
c~ = 0.09, C~1 = 1.44, and Ce2 = 1.92, the desired value tr = 0.43 is obtained
and serves to justify the parameter values used in the literature. However,
the situation is not so clear. As we have shown, our model contains an
additional term T 3 that is small in weakly strained turbulence and large in
rapidly distorted flows. It is likely that the rapid distortion contribution
to the model cannot be neglected in the logarithmic layer, leading to a
substantial effect on the value of ~c.
R N G Turbulence Models 59

A recent paper by Yakhot, Thangam, Gatski, Orszag, and Speziale


(1922) proposes the following closure for T3:
T3 _ C j / 3 ( 1 -- q / q o ) if2 (6.6)
1 +/~q3 ~.

q= (6.7)

S = (2S~jSJ, S o = I(Vj Vi + V, Vj) (6.8)


The nondimensional parameter q is the ratio of the turbulence time scale
Jg'/g to the time scale of the mean S, and is large in regions of rapid distor-
tion. The constant qo = 4.38 is the fixed point for homogeneous shear flow
of the RNG g ( 7 - o~ model equations (4.35), (4.42) and (5.10) without T3,
and/? is an undetermined constant. The closure (6.6) is an approximation
to the sum of the divergent series expansion for T3 in powers of t/, which
satisfies a number of important consistency requirements (Yakhot et al.,
1992). For example, T3 --+ 0 in the limit of weak strain where q ~ 0, and T3
is singular in the rapid distortion limit q -+ oo.
Using (6.6), the new model in the logarithmic layer is

1.42 -~
g v(Vy Vx) 2 -- C ~ ~
if2 + Vyc~vVyg= 0 (6.9a)

C* = 1.68 -t c/13(1 - V/t/~


~f2 1 +/?q3 (6.9b)

- f f + v(Vy Vx) 2 + Vy~vVyff = 0 (6.10)


where y is the direction normal to the wall and Vx is in the downstream
direction x parallel to the wall. Choosing the value of/? such that (6.4) with
C*2 in place of Ce2 gives x = 0.4, leads to/~ = 0.012. Thus it is clear that the
value e = 0.77 is not the only way to obtain the correct value of the yon
Karman constant.
Finally, we would like to point out that RNG together with the
e-expansion is a general method that can be used to derive models for
turbulence at any level of complexity, for example, including rotation,
stratification, or compressibility. Such flows are described by several
dimensionless parameters, and are not easy to model using heuristic
methods. RNG provides a systematic procedure to derive sophisticated
models for complex flow systems.

ACKNOWLEDGMENTS
We are grateful to William Reynolds for many invaluable discussions.
He was the first to suggest the importance of the new model term for rapid
60 Yakhot and Smith

distortions. We would also like to thank Paul Durbin for very helpful com-
ments about the model and the manuscript. Victor Yakhot received sup-
port from the Office of Naval Research under contract No. N00014-82-C-
0451, the Air Force Office of Scientific Research under grant No. AFOST-
90-0124, and DARPA under contract No. N00014-86-K-0759. Leslie Smith
received support from NASA-Ames Research Center through the Center
for Turbulence Research.

REFERENCES
Batchelor, G. K., and Townsend, A. A. (1948). Decay of turbulence in the final period, Proc.
R. Soc. London A194, 527.
Comte-Bellot, G., and Corrsin, S. (1966). The use of contraction to improve the isotropy of
grid-generated turbulence, J. Fluid Mech. 25, 657.
Dannevik, W. P., Yakhot, V., and Orszag, S. A. (1987). Analytical theories of turbulence and
the e-expansion, Phys. Fluids 30, 2021.
De Dominicis, C., and Martin, P. C. (1979). Energy spectra of certain randomly stirred fluids,
Phys. Rev. A 19, 419.
Durbin, P. A. (1990). Turbulence closure modeling near rigid boundaries, Annual Research
Briefs--1990, Center for Turbulence Research, Vol. 3.
Durbin, P. A., and Speziale, C. G. (1991). Local anisotropy in strained turbulence at high
Reynolds numbers, A S M E J. Fluids Eng. 113, pp. 707-709.
Forster, D., Nelson, D. R., and Stephen, M. J. (1977). Large-distance and long-time properties
of a randomly stirred fluid, Phys. Rev. A 16, 732.
Fournier, D., and Frisch, U. (1983). Remarks on the renormalization group in statistical fluid
dynamics, Phys. Rev. A 28, 1000.
Launder, B. E., Reece, G. J., and Rodi, W. (1975). Progress in the development of a Reynolds-
stress turbulence closure, J. Fluid Mech. 68, 537.
Lee, M. J., and Reynolds, W. C. (1985). Report TF-24, Mechanical Engineering Department,
Thermosciences Division, Stanford University.
Majda, A., and Avillaneda, M. (1990). Mathematical models with exact renormalization for
turbulent transport, Commun. Math. Phys. 131, 381.
Mansour, N. N., and Shih, T.-H. (1989). Forum on Trubulent Shear Flows--1989, FED
Vol. 76, Am. Society of Mech. Eng., New York.
Migdal, A. A., Orszag, S. A., and Yakhot, V. (1990). Intrinsic stirring force in turbulence and
the g-expansion, Princeton University preprint.
Millionshtchikov, M. D. (1939). Decay of turbulence in wind tunnels, Dokl. Akad. Nauk
SSSR 22, 236.
Orszag, S. A. (1970). Analytical theories of turbulence, J. Fluid Mech. 41, 363.
Panda, R., Sonnad, V., and Clementi, E. (1989). Turbulence in a randomly stirred fluid, Phys.
Fluids A 1, 1045.
Patel, V. C., Rodi, W., and Scheurer, G. (1985). Turbulence models for near-wall and low
Reynolds number flows: A review, AIAA Z 23, 1308.
Reynolds, W. C. (1976). Computation of turbulent flows, Ann. Rev. Fluid. Mech. 8, 183.
Shih, T.-H., Reynolds, W. C., and Mansour, N. N. (1990). A spectrum model for weakly
anisotropic turbulence, Phys. Fluids A 2.
Smith, L. M., and Reynolds, W. C. (1992). On the Yakhot-Orszag renormalization group
method for deriving turbulence statistics and models, Phys. Fluids A 2, 364.
RNG Turbulence Models 61

Speziale, C., Gatski, T. B., and Fitzmaurice, N. (1991). An analysis of RNG-based turbulence
models for homogeneous shear flow, Phys. Fluids. A 3, 2278.
Tan, H. S., and Ling, S. C. (1963). Final stage decay of grid-generated turbulence, Phys. Fluids
6, 1693.
Tennekes, H., and Lumley, J. L. (1972). A First Course in Trubulence, MIT Press, Cambridge,
MA.
Yakhot, V., and Orszag, S. A. (1986). Renormalization group analysis of turbulence. I. Basic
theory, J. Sci. Comput. 1, 3.
Yakhot, V., and Orszag, S. A. (1990). Analysis of the e-expansion in turbulence theory:
Approximate renormalization group for diffusion of a passive scalar in a random velocity
field, Princeton University preprint.
Yakhot, V., Orszag, S., and Panda, R. (1988). Computational test of the renormalization
group theory of turbulence, J. Sci. Comput. 3, 139.
Yakhot, V., Thangam, S., Gatski, T. B., Orszag, S. A., and Speziale, C. G. (1992). Develop-
ment of turbulence models for shear flows by a double expansion technique, to appear
in Phys. Fluids A, 7.

Você também pode gostar