ME431A/599H k- Equations 8 November 2010

In turbulence modeling, the Reynolds stresses are usually written in terms of a tur-
bulent (or “eddy”) viscosity, i.e.,
∂u
−u0 v 0 = νT ,
∂y
where the turbulent viscosity is defined by

νT = cT υT LT .

Here cT is a constant to be determined, and υT and LT are characteristic velocity and

length scales of the turbulence.
To obtain a predictive equation for υT , consider first the Reynolds decomposition of
the u-component of the velocity:
u = u + u0 .
Here u is the average, or mean, velocity, and u0 represents the fluctuations about the
average, the latter which is what contributes to the Reynolds stresses. Therefore in order
to get an estimate for υT defining the turbulent viscosity, an estimate for u0 is needed.
√
Note that u0 = 0, so that its average cannot be used. The average that is useful is u02 ,
the root-mean-square of u0 . In the modeling, all three components are considered, in terms
of the turbulent kinetic energy per unit mass, defined as
 
1 02 02 02
k= u +v +w .
2

A dynamic equation for k is obtained in the following way. Here only the component
1 02
u will be considered, as the derivation for the other two components is accomplished
2
in the same manner. First consider the x-component of the momentum equation for an
incompressible flow (neglecting the acceleration of gravity):
 2
∂ u ∂2u ∂2u

∂u ∂u ∂u ∂u 1 ∂p
(1) +u +v +w =− +ν + 2 + 2 .
∂t ∂x ∂y ∂z ρ ∂x ∂x2 ∂y ∂z
As shown in class, the time average of this equation is:
 2
∂ u ∂2u ∂2u

∂u ∂u ∂u 1 ∂p ∂ 0 0 ∂ 0 0 ∂ 0 0
(2) u +v +w =− +ν 2
+ 2+ 2 − uu − uv − uw .
∂x ∂y ∂z ρ ∂x ∂x ∂x ∂x ∂x ∂y ∂y
An equation for u0 can be obtained by first writing Equation (1) using the Reynolds
decomposition for all the variables, and then subtracting Equation (2) from (1). Examples
of terms to be dealt with are the following.
∂ ∂u0
(u + u0 ) − 0 =
∂t ∂t

1
 ∂ ∂u ∂u ∂u ∂u0 ∂u0 ∂u ∂u ∂u0 ∂u0
(u + u0 ) (u + u0 ) − u =u + u0 +u + u0 −u = u0 +u + u0
∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x

1 ∂ 1 ∂p 1 ∂p0
(p + p0 ) − =
ρ ∂x ρ ∂x ρ ∂x

∂2 0 ∂2u ∂2 0
ν 2 (u + u ) − ν 2 = ν 2 u
∂x ∂x ∂x
The result of this process is:

∂u0 ∂u0 ∂u0 ∂u0

+u +v +w
∂t ∂x ∂y ∂z
∂u ∂u ∂u
+ u0 + v0 + w0
∂x ∂y ∂z
0 0 0
0 ∂u 0 ∂u 0 ∂u
+u +v +w
∂x ∂y ∂z
0
 2 0
∂ 2 u0 ∂ 2 u0

1 ∂p ∂ u
=− +ν + +
ρ ∂x ∂x2 ∂y 2 ∂z 2
∂ 0 0 ∂ 0 0 ∂ 0 0
+ uu + uv + uw .
∂x ∂y ∂z

Now multiply this last equation by u0 and average. Terms such as the following are
obtained.
∂u0 ∂ 1 02
u0 = u = 0 (stationary)
∂t ∂t 2

∂u0 ∂ 1 02
u0 u =u u
∂x ∂x 2

∂u ∂u
u0 u0 = u02
∂x ∂x

∂u0 ∂ 1 02
u0 v 0 = v0 u
∂y ∂y 2

∂ 2 u0 ∂u0 ∂u0 ∂u0

 
∂
νu0 = ν u 0 − ν
∂x2 ∂x ∂x ∂x ∂x
∂ 2 1 02 ∂u0 ∂u0
=ν u − ν
∂x2 2 ∂x ∂x

2
The final results is:
∂ 1 02 ∂ 1 02 ∂ 1 02
u u +v u +w u
∂x 2 ∂y 2 ∂z 2
∂u ∂u ∂u
+ u02 + u0 v 0 + u0 w 0
∂x ∂y ∂z
∂ 0 1 02 ∂ 0 1 02 ∂ 0 1 02
+ u u + v u + w u
∂x 2 ∂y 2 ∂z 2
 2
1 0 ∂p0 ∂2 ∂ 2 1 02

∂
=− u +ν + 2+ 2 u
ρ ∂x ∂x2 ∂y ∂z 2
 0 2  0 2  0 2 
∂u ∂u ∂u
−ν + + .
∂x ∂y ∂z
Here the fact that the fluctuating velocity field (u0 , v 0 , w0 ) satisfies the incompressible form
of the conservation of mass was used.
When the same operations are carried out for the equations for v 0 and w0 , and the
three equations are added together, we obtain the equation for k:
∂k ∂k ∂k
(3) u +v +w = P + D − .
∂x ∂y ∂z
Here P, called the production term and usually resulting in an increase in k, is given by:
∂u ∂u ∂u
P = u02 + u0 v 0 + u0 w 0
∂x ∂y ∂z
∂v ∂v ∂v
+ u0 v 0 + v 02 + v 0 w0
∂x ∂y ∂z
∂w ∂w ∂w
+ u0 w 0 + v 0 w0 + w02 .
∂x ∂y ∂z
The term D is called the turbulent diffusion term, and results in the turbulent kinetic
energy k being moved around in space (from regions of high k to regions of low k) by the
turbulence without any overall change in k, and is:
     
∂ ∂ ∂
D=− p0 u 0 + u 0 k − p0 v 0 + v 0 k − p0 w 0 + w 0 k .
∂x ∂y ∂z
Finally , the kinetic energy dissipation rate, which is always positive, represents the
conversion of mechanical energy into internal energy (heat) by the turbulence, and is given
by:
 0 2  0 2  0 2
∂u ∂u ∂u
 =ν + +
∂x ∂y ∂z
 0 2  0  2  0  2
∂v ∂v ∂v
+ + +
∂x ∂y ∂z
 0 2  0 2  0 2 
∂w ∂w ∂w
+ + + .
∂x ∂y ∂z

3
Note that in the equation for k new modeling has to be introduced for the terms p0 u0 and
u0 k, and similarly for terms with v 0 and w0 , in the expression for the turbulence diffusion
D.
In addition to this equation for k, an equation for  can also be obtained (by another
even more tedious procedure). It is of the form:

∂ ∂ ∂
(4) u +v +w = P + D −  .
∂x ∂y ∂z

The terms on the right hand side have similar interpretations as for the corresponding terms
in the equation for k, i.e., P represents the production of , D its turbulent diffusion, and
 its dissipation rate. Additional modeling is needed for each of these terms.
If k is known (i.e., has been computed), then the velocity scale can be identified as
p
υT = k.

Also, given k and , a length scale can be defined as

3/2
LT = k / ,

the only dimensional combination of k and  possible. So the turbulent viscosity is

1/2 3/2 2
νT = cµ υT LT = cµ k k / = cµ k / ,

where the constant cµ , needs to be determined.

To close Equations (2), (3), and (4), some additional assumptions are needed. First
of all, in the production term P, there is the need to model velocity fluctuation products,
i.e., the Reynolds stresses, e.g., u0 v 0 . In analogy with the viscous stresses, e.g., τxy , (see
pages 306 and 307 of the text by Munson et al., or page 242 in White, for the Newtonian
viscous fluid expressions for the viscous stresses), the following assumptions are made.

2 ∂ ū 2 ∂v̄ 2 ∂ w̄
u02 = k̄ − νT , v 02 = k̄ − νT , w02 = k̄ − νT .
3 ∂x 3 ∂y 3 ∂z

Note that, using the conservation of mass,

 
∂ ū ∂v̄ ∂ w̄
u02 + v 02 + w02 = 2k̄ − νT + + = 2k̄ ,
∂x ∂y ∂z

as expected. Assumptions for the remaining Reynolds stress terms are the following.
     
∂ ū ∂v̄ ∂ ū ∂ w̄ ∂v̄ ∂ w̄
u0 v 0 = −νT + u0 w 0 = −νT + v 0 w0 = −νT + .
∂y ∂x ∂z ∂x ∂z ∂y

4
 The fluxes in the diffusion term D are also modeled in analogy with molecular diffusion.
The turbulent fluxes in the x-, y-, and z-directions are modeled as, respectively,

νT ∂ k̄ νT ∂ k̄ νT ∂ k̄
(u0 p0 + u0 k) = − , (v 0 p0 + v 0 k) = − , (w0 p0 + w0 k) = − .
σk ∂x σk ∂y σk ∂z

Here, in analogy with the Schmidt number, the turbulent Schmidt number is defined as
νT
σk = , where νk is the turbulent diffusivity for k̄. The constant σk thus needs to be
νk
determined.
In Equation (4) for , additional modeling is needed. To model the production of ,
i.e., P , it is assumed that P = c1 (/k̄)P. Here c1 needs to be determined. The fluxes of
 in the x-, y-, and z-directions are modeled as, again in analogy with molecular diffusion:

νT ∂ νT ∂ νT ∂
− , − , −
σ ∂x σ ∂y σ ∂z

where the constant σ needs to be determined.

Finally, the dissipation rate of , i.e.,  , is modeled as

 = c2 2 /k̄ ,

where the constant c2 needs to be determined.

Numerous experiments have been carried out to determine the constants, and the
generally accepted values are the following.
. . . . .
cµ = 0.09 , c1 = 1.44 , c2 = 1.92 , σk = 1.0 , σ = 1.3 .

The k̄- model tends to work fairly well for simple flows, like two-dimensional jets and
non-swirling jets. For swirling jets, however, it does not work as well, since the swirling
flow changes the dynamics, which is not incorporated into the k̄- model. The same is
true for flows in more complex geometries, or flows with additional effects, such as variable
densities and/or compressibility. So one has to generally use the k̄- model with caution.
If it is used with a more complex flow, then the results may be, at most, qualitatively or
semi-quantitatively correct.