School of Mechanical Aerospace and Civil Engineering

TPFE MSc Advanced Turbulence Modelling

Introduction

Linear Eddy-Viscosity Models

T. J. Craft

The first part of this lecture aims to recall the linear eddy-viscosity
schemes studied in previous courses:

To consider some alternative forms of models.

To consider implementation in numerical flow solvers.

In the second part of the lecture, attention will be turned to the near-wall
region.

The aim is to explore how viscous and other near-wall effects are built
into linear EVMs to allow them to be applied across the thin near-wall
viscous sub-layer.

We will examine a number of approaches, with some examples of

applications.

Some of the modelling strategies mentioned or referred to will be met

again, in more detail, later.

George Begg Building, C41

Reading:
S. Pope, Turbulent Flows
D. Wilcox, Turbulence Modelling for CFD
Closure Strategies for Turbulent and Transitional
Flows, (Eds. B.E. Launder, N.D. Sandham)
Notes:
Blackboard and CFD/TM web server:
http://cfd.mace.manchester.ac.uk/tmcfd
- People - T. Craft - Online Teaching Material
Linear Eddy-Viscosity Models

2011/12

1 / 38

Linear Eddy-Viscosity Models

Linear Eddy-Viscosity Models

2011/12

Mixing-Length Models

The problem is to obtain approximations for the Reynolds stresses that

appear in the mean momentum equations:


Ui
DUi
1 P

(1)

+
ui uj
=
Dt
xi xj
xj
Linear eddy-viscosity models employ a stress-strain relation of the form


Ui Uj
(2)
+
ui uj = 2/3 k ij t
xj
xi

In simple mixing-length models the mixing-length lm is prescribed

algebraically, and t obtained from




2 U 
t = lm
 y 
or, in a more general formulation:
s
2
t = lm

The Reynolds stresses are linearly related to mean strains via the
turbulent, or eddy, viscosity.
lm

3 / 38

2
(4)

A typical formulation with the Van-Driest

near-wall damping term is
lm = min( yD, )

2011/12

Ui Uj
+
xj
xi

(3)

The lengthscale lm is taken to vary linearly with distance from the wall,
but with a damping term to account for near-wall viscous effects.

The turbulent viscosity t is obtained in a different manner, depending on

the level of modelling employed.

Linear Eddy-Viscosity Models

2 / 38

Linear Eddy-Viscosity Models

D = 1 exp(y + /A)
2011/12

4 / 38

One-Equation Models

Two-Equation Models

In a typical 1-equation model, a transport equation is solved for the

turbulent kinetic energy k :



k
Dk
( + t /k )
= Pk +
Dt
xj
xj

The generation rate is given by Pk = ui uj Ui / xj .

The dissipation rate is modelled as = k 3/2 /l .

The turbulent viscosity is then taken as

(5)

2011/12

5 / 38

is not the only variable that can be solved for to determine the
lengthscale.

Wilcox (1988), for example, proposed solving an equation for ( /k )



D
Pk

( + t / )
(9)
= c 1
c 2 2 +
Dt
k
xj
xj

(8)

The model constants c 1 , c 2 , c , and are tuned for simple equilibrium

shear flows.

In the near-wall region, where viscous effects are important, a damping

function is introduced into c , and additional source terms are sometimes
included in the equation.

In a two-equation model there is no need to prescribe a lengthscale for

the turbulence; it now emerges from the solution (l = k 3/2 / ).

Linear Eddy-Viscosity Models

2011/12

6 / 38

Implementation of the Momentum Equations

Representing the Reynolds stresses using a linear EVM, the momentum

equations become


 

DUi

Ui Uj

Ui
1 P

t
+

+
+
(2/3 k ) (11)
=
Dt
xi xj
xj
xj
xj
xi
xi

The final term, containing the gradient of k , can be absorbed into the
pressure term.

Adopting this treatment, the variable in which the pressure is stored is

actually representing P + (2/3) k .

The diffusive terms can be written as





Uj

( + t ) i +
t
xj
xj
xj
xi

(10)

which is claimed to give better results in near-wall adverse pressure

gradient flows.
Some alternative lengthscale equations will be considered in more detail
in a later lecture.

Linear Eddy-Viscosity Models

The turbulent viscosity is then modelled as

Numerical Implementation

t = c k /

(7)

(6)

Both lengthscales l and l are typically prescribed as increasing linearly

with distance from the wall although they may have different viscous
damping terms associated with them to account for very near-wall
viscous effects.

Linear Eddy-Viscosity Models

In a 2-equation model a second variable (often ) is solved for:



D

( + t / )
= c 1 Pk c 2 +
Dt
k
k
xj
xj

t = c k 2 /

t = c k 1/2 l

2011/12

7 / 38

Linear Eddy-Viscosity Models

(12)

2011/12

8 / 38

The first of these diffusion terms appears similar to the laminar form,
except with a modified effective viscosity, eff = t + .

Zero-Equation Models

The second term (which is negligible in thin shear flows), must simply be
added into the momentum equation as a source term.
From a computational point of view, the above equation can thus be
treated almost identically to the case of a laminar flow except that the
viscosity is now varying in space, and cannot be taken as a constant.

In this case, all that need be done, having updated the velocity field, is to
re-calculate t using the new velocity gradients.

One- and Two-Equation Models

In these models, separate transport equations are solved for k (and ).

These equations are still of the general form



D

+ S
=
Dt
xj
xj

The linear eddy-viscosity formulation is thus relatively easy to implement,

and is generally fairly stable, since the turbulent viscosity related diffusion
terms can mostly be treated in an implicit manner.

(13)

so convection and diffusion terms can be discretized in the same way as

was done when considering the momentum equations.

In a finite volume approach, integrating the k equation over a cell leads to

a discretized equation of the form
ap kp = ai ki + Su

(14)

where the major contribution to Su is typically from the integration of

Pk over the control volume.
Linear Eddy-Viscosity Models

2011/12

9 / 38

Linear Eddy-Viscosity Models

(15)

In the case of the k and equations, there are often quite significant
source terms on the right hand side of the difference equations.

This is generally unhelpful from a numerical point of view, as it leads to

slow convergence, and can give rise to instabilities.

In an iterative scheme, some of these terms can be linearized and treated

in a manner that improves stability.

The source term Su in the k equation can be split into two parts:

where Pkp and p are evaluated at the cell centre.

10 / 38

Linearizing Source Terms

Such a source term is normally approximated as

(Pkp p ) Vol

2011/12

The generation rate Pkp is simply evaluated from the turbulent viscosity
and velocity gradients.

p is either obtained as a result of a corresponding transport equation,

or is algebraically related to kp , depending on which model is employed.

Su = Su + Sp kp

(16)

where Su contains positive contributions to Su and Sp is negative.

Sp can then be transferred onto the left hand side of the equation:
(ap Sp )kp = ai ki + Su

Linear Eddy-Viscosity Models

2011/12

11 / 38

Linear Eddy-Viscosity Models

(17)

2011/12

12 / 38

Low-Reynolds-Number Eddy-Viscosity Models

This increases the diagonal dominance of the coefficient matrix, thus

improving stability and convergence rates.

For example, in a 1-equation model, the dissipation term in the k

3/2
equation leads to a source term (kp /l )Vol. The term is negative, so
can be transferred to the left hand side by setting
1/2

Sp = (kp /l )Vol

(18)

In a two-equation model, since k must be positive, its dissipation term

can still be transferred to the left hand side by writing
Sp = (p /kp )Vol

(19)

An accurate representation of near-wall regions may be necessary in

highly non-equilibrium flows, or in computing heat-transfer problems.

Later lectures will consider wall function approaches which aim to avoid
or simplify the resolution of this region.

In mixing length models, viscous damping terms were introduced to

reduce the lengthscale in the near-wall (viscosity-affected) region.

Here, we mainly consider low-Reynolds-number two-equation EVMs,

which can be applied across the viscous sublayer, down to the wall.

When solving transport equations for turbulence quantities, a number of

issues need to be considered:

The source/sink terms in the equation can be treated in a similar

manner.

Linear Eddy-Viscosity Models

2011/12

13 / 38

Inclusion of Viscous Effects

Viscous diffusion should be included

Model coefficients may need changing in low-Re regions

Additional, near-wall, model terms may need to be included

Wall boundary conditions

Linear Eddy-Viscosity Models

2011/12

Exact terms, such as viscous diffusion.

In a simple shear flow, the linear stress-strain relation leads to



uv
k U
= c S
= c
k
y

Modelled source terms.

where S is the non-dimensional strain rate.

Terms in transport equations explicitly involving molecular viscosity

Damping functions, which act to modify model coefficient values,

and which may depend on quantities such as

Turbulent Reynolds number, Rt = k 2 /( ).

Wall distance,

In the fully turbulent near-wall region of a local equilibrium boundary

layer, S 3.3 and |uv /k | 0.3, giving c 0.09.

Plotting uv /(kS) across channel and boundary layer flows:

= yk 1/2 / .

Terms involving wall-distances are less preferable as they become

difficult to define in complex geometries.

In general, low-Reynolds-number models should asymptote to the

high-Re forms seen earlier in regions far from walls.
Plane channel (Kim et al, 1987)

Linear Eddy-Viscosity Models

14 / 38

The Turbulent Viscosity

Viscous, or near-wall, effects can be included in a number of ways:

2011/12

15 / 38

Linear Eddy-Viscosity Models

Boundary layer (Spalart, 1988)

2011/12

16 / 38

The turbulent viscosity (effectively c ) needs to be reduced across the

viscous region.

A damping function f is usually introduced:

t = f c

k2

giving

Wall-Limiting Behaviour

uv
= f c S
k

The wall-limiting behaviour of flow quantities can be explored using Taylor

series expansions.

Close to a no-slip wall one can write

u = a1 y + a2 y 2 + a3 y 3 +
v = b1 y + b2 y 2 + b3 y 3 +

f is typically a function of turbulent Reynolds number.

w = c1 y + c2 y 2 + c3 y 3 +

where y is distance from the wall, and the as, bs and cs are functions of
x , z and time, but not y .

For example, the Launder-Sharma

model takes


3.4
f = exp
(1 + Rt /50)2

Continuity ( u/ x + v / y + w/ z = 0) gives b1 = 0.

Hence u y , w y , but v y 2 at the wall.

The near-wall behaviour of the stresses is thus

u2 y 2

Linear Eddy-Viscosity Models

Since we now have

2011/12

17 / 38

w2 y2

v2 y4

Linear Eddy-Viscosity Models

uv y 3
2011/12

The Equation

k2 U
uv = c f
y

The dissipation rate reaches a maximum at the wall

a suitable form of f can be chosen to give the correct uv behaviour at

the wall.

This non-zero value should balance

with the viscous diffusion of k .

As k y 2 and is O(1) near the wall (see later), the above implies that
f 1/y at the wall.

This explains the upturn in the effective c seen earlier in the immediate
near-wall vicinity DNS data.

One can show that at the wall

!2
k 1/2
w = 2
y

Note, however, that in t above is sometimes replaced by a slightly

different variable, which will imply a different near-wall behaviour of f .

Although this could be used as a

boundary condition, it is not a very
convenient condition to apply.

With a linear EVM, one cannot obtain the correct near-wall (or often, in
fact, the outer) behaviour of all the normal stress components.

Instead of solving for itself, many -based models use the isotropic
dissipation rate, , defined by
!2
k 1/2
= + 2
xj

Linear Eddy-Viscosity Models

18 / 38

2011/12

19 / 38

Linear Eddy-Viscosity Models

2011/12

20 / 38

Note that this results in = 0 at the wall, which is a convenient boundary

condition.

Selected Low-Reynolds Number k - Schemes

Written in a general form:


k
( + t /k )
xj


Pk

D
( + t / )
= c 1
c 2 f2 + E +
Dt
Tt
Tt
xj
xj

Typically, beyond around y 5, and are essentially identical.

Since y 2 at the wall, if the turbulent viscosity is now defined as

Dk
= Pk ( + D) +
Dt
xj

t = c f k 2 /
then the damping function f should be proportional to y at the wall.

Additional near-wall source terms are often also included in the modelled
(or ) equation.

Model
LS

c 1
1.44

c 2
1.92

f2
1 0.3 exp(Rt2 )

Tt

CH

1.35

1.8

1 0.3 exp[( R6t )2 ]

YS

1.44

1.92

MK

1.4

1.8

1.0
h
ih
i2
+
1 29 exp{( R6t )2 } 1 exp( y5 )

SZS

1.5

1.83

1.0

 1/2

LS: Launder & Sharma (1974); CH: Chien (1982); YS: Yang & Shih (1993);
MK: Myong & Kasagi (1990); SZS: So et al (1991)
Linear Eddy-Viscosity Models

2011/12

Model
LS

0.09

exp(3.4/(1 + Rt /50)2 )

1.0

1.3

1.0

1.3

1.0

1.3

1.4

1.3

0.75

1.45

CH

0.09

YS

0.09

MK

0.09

SZS

0.096
Model
LS

1 exp(0.0115y )

1/2
1 exp(a1 Rey a3 Rey3 a5 Rey5 )
i
h
1/2
[1 exp(y + /70)]
1 + 3.45/Rt
i
h
1/2
[1 exp(y + /115)]
1 + 3.45/Rt
D

2
k 1/2 / xj

CH

2 k /y 2

YS

MK

SZS

Linear Eddy-Viscosity Models

E
2t

2U

i / xj xk

21 / 38

Linear Eddy-Viscosity Models

2011/12

Alternative Lengthscale Equations

Models which solve for quantities other than also require

low-Reynolds-number modifications in near-wall regions.

Wilcoxs (1991) k - model ( /k ) takes t = c f k / :



D
Pk

( + t / )
= c 1 f1
c 2 2 +
Dt
k
xj
xj
f1 =

2

o + Rt /R
c f (1 + Rt /R )

0
h
i
2
exp((Rt /64)2 ) 2 k + 1.5 (k)
23 / 38

f =

o + Rt /Rk
1 + Rt /Rk

The wall boundary condition is applied at the first interior node:

1 =

2 exp(y + /2)/y 2

2
t 2 Ui / xj xk

2011/12

22 / 38

6
c 2 y12

The formulation does have some advantages, particularly with regard

to near-wall lengthscales (see later lecture).

However, at the wall may not be particularly attractive.

Linear Eddy-Viscosity Models

2011/12

24 / 38

Elliptic Relaxation Approaches (v 2 -f Models)

The k - model of Speziale et al (1990) ( = k / ) takes:

P
k
2
D
= (1 c 1 ) k + (c 2 f1 f2 1) + ( + t / 1 )
Dt
k
k
xj xj


2

( + t / 2 )
( + t / 2 )
+

xj xj xj
xj

f1 = 1

2
exp{(Rt /6)2 }
9

t = c f k


2
f2 = 1 exp(y + /4.9)
"

f = 1 +

3.45
1/2

Rt

In the forms considered so far we have taken t = c f k 2 / , with f

accounting for the reduction in t needed near a wall.

Durbin (1991) noted that DNS

data showed c k v 2 / to give a
good fit to the implied turbulent
viscosity.

He proposed solving a transport

equation for v 2 the wall-normal
stress component.

In simple shear flow a modelled v 2 transport equation can be written as

!
Dv 2

v2

2
( + t )
= 22 v
+
Dt
k xj
xj

#
tanh(y + /70)

Like the -based schemes above, the details differ from model to model,
but the principles are generally similar to those already outlined.

Some of these alternatives will be examined in more detail in a later

lecture.

Linear Eddy-Viscosity Models

2011/12

represents a combination of redistribution and dissipation

where 22
processes (see later lecture on stress transport models).
25 / 38

Durbin proposed closing the system by solving a further differential

.
equation for 22

= kf , and solved an elliptic relaxation equation for f , of

He wrote 22
22
22
the form
2 f22
(20)
L2
f22 = 22 /k (v 2 /k 2/3)/T
xj2

Linear Eddy-Viscosity Models

2011/12

Durbin showed good results in plane channel flow.

A number of modified forms of this v 2 -f model approach have been

proposed.

26 / 38

where the length and time scales L and T are

L = CL max(k 3/2 / , c ( 3 / )1/4 )

T = max(k / , CT ( / )1/2 )

22 is a redistribution model taken from the stress transport model of

Launder, Reece & Rodi (1975):
22 = c1 (v 2 2k /3)/T + c2Pk

Far away from a wall f22 should revert to the form of the LRR model, but
the differential operator in equation (20) modifies this behaviour near a
wall.

Linear Eddy-Viscosity Models

2011/12

27 / 38

Linear Eddy-Viscosity Models

2011/12

28 / 38

Simple Near-Wall Flows

Accelerating Boundary Layers

Channel Flow

Boundary Layer

Most models
perform reasonably.

Note some models

are tuned to give
correct near-wall
asymptotes in these
flows.

Linear Eddy-Viscosity Models

2011/12

29 / 38

DNS data of Spalart (1986) at K = 2.5 106

Launder-Sharma

Acceleration parameter K defined as

K=

Near-wall k profiles
for several low-Re
EVMs (from Sarkar
& So, 1997).

dU
U2 dx

Jones & Launder (1972)

Linear Eddy-Viscosity Models

2011/12

30 / 38

Skewed Channel Flow

Wilcox k -

Starts with fully developed channel flow.

Walls are then impulsively

moved in direction
perpendicular to flow direction.

W
x
z

Symbols: DNS at K = 0 and 2.5 106

Lines: Calculations at k = 0 (solid) and 2.5 106 (broken).

Linear Eddy-Viscosity Models

2011/12

31 / 38

DNS data of Howard & Sandham (2000) shows an initial reduction in

turbulence energy and shear stress, followed by a recovery as the flow
re-aligns to the new direction.

Linear Eddy-Viscosity Models

2011/12

32 / 38

By-Pass Transition

Skewed Channel: Shear Stress at Selected Times

Launder-Sharma k -
1.0

1.0

uv

0.75

uv+ (T=0.3)
0.75

0.5

0.5

0.25

0.25

0.25

0.25

0.5

0.75

1.0

y/h

0.0
0.0

0.25

0.5

0.75

1.0

y/h

0.0
0.0

Skin friction

0.25

0.5

0.0
0.0

1.0

0.75

Reasonable results returned by the Launder-Sharma scheme.

0.75

0.5

0.25

Transition position is dependent on free-stream turbulence levels.

uv+ (T=1.0)

uv+ (T=0.4)

0.75

0.5

0.0
0.0

1.0

1.0

0.25

0.5

0.75

Velocity profiles

1.0

y/h

y/h

k - schemes
1.0

1.0

0.0

1.0

0.3

1.0

0.4

0.75

0.75

0.5

0.5

0.5

0.5

0.25

0.25

0.25

0.25

0.0
0.0

0.25

0.5

0.75

1.0

0.0
0.0

0.75

0.25

0.5

0.75

1.0
1.0

0.0
0.0

1.0

0.75

0.25

0.5

0.75

1.0
1.0

0.0
0.0

0.25

0.5

0.75

1.0

Symbols: DNS data

The E source term in the Launder-Sharma model is crucial in obtaining

the reduction in peak turbulence levels during the initial skewing.

Linear Eddy-Viscosity Models

2011/12

33 / 38

Impinging Flow

Linear Eddy-Viscosity Models

2011/12

34 / 38

Summary

Previous flows have mainly involved obtaining the correct shear stress.

We have revisited the general forms adopted for linear EVMs.

In more complex flows, EVMs can fail badly.

Numerical implementation of linear EVMs is fairly straightforward: the

inclusion of turbulent viscosity can help stabilize the system of equations.

Impinging jet heat transfer with Launder-Sharma scheme.

Low-Re-number EVMs generally contain damping terms and near-wall

source terms, dependent on viscosity and/or wall distance.

Examples have been presented of k - schemes and some alternative

approaches.

Most perform acceptably in channel or simple boundary layer flow.

Additional source terms included in some models aid in more complex

flows such as accelerating or transitional boundary layers.

In flows with more complex strains, linear EVMs (high or low Re) may fail.

All these low-Re schemes require a very fine near-wall grid, which can be
expensive in 3-D. Alternative strategies will be examined in a later lecture.

Solid line: Original k- form;

Broken line: k- with lengthscale
correction.

We will revisit this problem in a later lecture.

Linear Eddy-Viscosity Models

2011/12

35 / 38

Linear Eddy-Viscosity Models

2011/12

36 / 38

References

Durbin, P.A., (1991), Near-wall turbulence closure modeling without

damping functions, Theoret. Comput. Fluid Dynamics, vol. 3, pp. 1-13.
Howard, R.J.A., Sandham, N.D., (2000), Simulation and modelling of a
skewed turbulent channel flow, Flow, Turbulence & Combustion, vol. 65,
pp. 83-109.
Kim, J., Moin, P., Moser, R., (1987), Turbulence statistics in fully
developed channel flow at low Reynolds number, J. Fluid Mech., vol. 177,
pp. 133-166.
Jones, W.P., Launder. B.E., (1972), The prediction of laminarization with
a two-equation model of turbulence, Int. J. Heat Mass Transfer, vol. 15,
pp. 301-314.

Launder, B.E., Reece, G.J., Rodi, W., (1975), Progress in the

development of a Reynolds stress turbulence closure, J. Fluid Mech., vol.
68, p. 537.

Launder, B.E., Sharma, B.I., (1974), Application of the energy-dissipation

model of turbulence to the calculation of flow near a spinning disc, Lett. in
Heat Mass Transfer, vol. 1., pp. 131-138.

Linear Eddy-Viscosity Models

2011/12

37 / 38

Myong, H.K., Kasagi, N., (1990), Prediction of anisotropy of the near wall
turbulence with an anisotropic low-Reynolds number k - turbulence
model, ASME J. Fluids Eng., voli. 112, pp. 521-524.

Sarkar, A., So, R.M.C., (1997), A critical evaluation of near-wall

two-equation models against direct numerical simulation data, Int. J. Heat
and Fluid Flow, vol. 18, pp. 197-208.
Spalart, P.R., (1986), Numerical study of sink-flow boundary layers, J.
Fluid Mech., vol. 172, pp. 307-328.

Spalart, P.R., (1988), Direct simulation of a turbulent boundary layer up to

R = 1410, J. Fluid Mech., vol. 187, pp. 61-98.

Wilcox, D.C., (1988), Reassessment of the scale determining equation for

advanced turbulence models, AIAA J., vol. 26, pp. 1299-1310.

Wilcox, D.C., (1991), Progress in hypersonic turbulence modelling, Proc.

AIAA 22nd Fluid Dynamics, Plasmadynamics & Laser Conference,
Honolulu.

Yang, Z., Shih, T-H., (1993), A Galilean and tensorial invariant k - model
for near-wall turbulence, NASA Tech. memo 106263, NASA Langley
Research Center.

Linear Eddy-Viscosity Models

2011/12

38 / 38