Reynold stress model (RSM)

Rixin Yu, 2017

1
 Reynold stress model (RSM)
• Part 1.1
• Characterize the Reynold stress “anisotropy”.
• Lumley triangle

• Part 1.2
– Reynold stress equations.
• Derivation
• Brief examination

2
 What we plan to do now!
• Reynold stresses contain information more than k
– k is only the isotropic part of Reynold stress!
• To characterize the Reynold stress “anisotropy”
– Using Lumley triangle to represent “anisotropy”.
• Tensor “invariant” concept
– It is related to Eigenvalue (Linear algebra) concept.

3
Characterization of Reynold-stress anisotropy

2
− 2
= 3 = − /2
2 3

= ( + + )

1: 2nd order tensor
2: symmetrical
2: real valued
3: zero trace
4: realizable, e.g. ≤( + ) + ( ) ≤ ⋅
….

Tensor invariant for ?

 Recap: Tensor element values change at a different coordinate

= 1, 0 , 0
1st order tensor(vector)
#
= 0, 1 , 0

0 0 0
= 0 0 0
0 0 #

2nd order tensor

0 0
#
= 0 0 0
Note: such a is NOT realizable, 0 0 0 # 5
but we just use here for simple demonstration #
 Characterize of Reynold-stress anisotropy
Recap: Tensor elements change values at different coordinate:

Example: a 2nd order tensor ( ) formed as the product of two 1st order tensor (vector) : and .

#
Original coordinate A rotated coordinate
#

#
#

= 1 0 0 #
= 0 1 0

#
= 1 0 0
= 0 0 1

0 0 0 0 0
# # #
≡ = 0 0 0 ≡ = 0 0 0
0 0 0 0 0
# 6
In coordinate In coordinate
•
Tensor invariants
Tensor elements change values as coordinate changes (kept to be orthogonal)

= 1 0 0 #
= 0 1 0

0 0 0 0 0 #
# # #
≡ = 0 0 0 ≡ = 0 0 0 #
0 0 0 0 0
• Can we find some expressions formed using the tensor element values, those expressions will
keep constant under coordinate changes ?
– For 1st order tensor, i.e. = ( , , )
# # #
• Length of the vector: = + + = + +

– For 2nd order tensor ,

# #
• ( ,… , )= ( ,… , )
– for example , ( ,… , )≡ +2 +⋯
# #
• ,… , = ( ,… , )
• ???
7
 Tensor invariants for 2nd order symmetrical tensor
• Yes, we can!

= − /2 = ×

– Treat as matrix, find its 3 eigenvalues

= − =0 | − |=0 + ⋅ + ⋅ + =0

Characteristic polynomial

3 principle = + +
( − )( − )( − )=0
invariants to = + +
coordinate =
change!
8
 Characterization of Reynold-stress anisotropy
-The Lumley triangle
(Real, symmetrical, zero trace tensor)

= + + = =0

= + + = −

1 1 1
= = − +
6 2 3

Tensor*Tensor ÆTensor
collapse indexÆ low order tensor
If reach to first order tensor, scalar (invariant)

6η 2 = −2 II b = bii2
= ≠
6ξ 3 = 3III b = bii3 Compact notation
NOT realizable if
outside the triangle

10
 /
= / Îb=
/
/ /
= / Î b= /
− /

/
= ⇒ = − /
− /

Lumley triangle

= + + =0
= + + = −3
= =2
 Two-equations models vs. RSM
Mean continuity: =0

Mean momentum: 1 ̅ −
=− +

1 transport equations+5 relations

RSM : 6 transport equations
= ( ) + − = +Π + −

Eddy-viscosity-assumption.
2 2
−
3
=− ( + ) +. .
3
1
=
2

∗
= ( )+ − = ( )+ −
12
 Derive Reynold stress equations
Denote = + + − =0
Non-conservative form

( + ) ( + ) 1
Expand first two terms: + + − =0

( + ) ( + ) 1
⋅ = + + − =0

( + ) ( + ) 1
⋅ = + + − =0

Not expand yet

+ = + + + + +
1
+ + − ( + )=0
13
 Expand then take average: + =0

1
+ + + + + + +

+ + =0
( + )

=0
0

+ + + +

1
+ + − 2 + =0

+ = −2 + 14
 Reynold stress equations

+ = − +Π − +

Production tensor =− + Viscous

Turbulent diffusion
transport
Dissipation tensor =2

Velocity-pressure-gradient tensor Which terms

need modelling ?
Π =− +

= + − +

= −

Pressure-rate-
Pressure
of-strain tensor 15
transport tensor
 Reynold stress equations contain the (= ) equation

+ = − +Π − +

Take half trace

(collapse j to i )

′ 1
Standard k-eq + = − − − +
2

1 2
= = =
2 2

Π = + − +
1 1
= =− + ′
2 2 = 0 − 2
= −
16
How those RS terms looks like in a turbulent
boundary layer?

+ = − +Π − +

Mean Convection Prod. Dissp. Pressure Turb. conv Vis. diff.

0
0
0 0

+ = − +Π − +

+ = − +Π − +

+ = − +Π − +

+ = − +Π − +

′ 1
+ = − − − +
2
17
Distribution of Reynold stres
0
< > < > 0
< > < > 0 0
0 0 < > 0 0

18
 Budget of k equation

′ 1
+ = − − − +
2
Mean Convection Prod. Dissp. Pressure Turb. conv Vis. diff.

1
2

19
∑ =1 Normalized by
 Put together 0
0
0 0

Π Π

Π
Π

Π
20
 Dissipation rate

+ = − +Π − +

Accurate dissipation tensor =2 =

3 + +
Simplification: use = =
Kolmogorov’s local 2 3
isotropy hypothesis:

2 2 0 0
= = 0 0
3 3
0 0
 The dissipation equation
1 1
= =− +
2 2
= −
2
=
3

= + −
=⋯

equation
= + −
In k −

=− 2 ( + )

22
 A note for dissipation
• Near wall ( y+<20)( = is not valid anymore )

Π Π

Π Π

Π
Π

Π Π
23
Reynold stress models
-part 2

24
 Reynold stress equations
• Model the pressure-rate-of-strain
term:
• Contribution by the slow pressure
• Contribution by the rapid pressure
– Rapid Distortion Theory (RDT)
• Contribution by the harmonic pressure

25
 Reynold stress equations

+ = − +Π − +

Production tensor =− + Viscous

Turbulent diffusion
transport
Dissipation tensor =2

Velocity-pressure-gradient tensor Which terms need to

be modeled?
Π =− +

= + − +

= −

Pressure-rate-
Pressure
of-strain tensor 26
transport tensor
 The redistribution role of 0
0
+. . = − + Π …. 0 0

Reynolds stress at a
certain direction ( , ) is
Π
produced and consumed
by a joint contriubtions
from three terms: the
production term , the
Π Π
disspation term and the
Π term (or ).
The contribution by the last
term is signficant, it
servers to redistribing the Π
reynolds stresses among Π
different directions.
Indeed summation of
over all direction gives Π
zero, i.e = 0 .
 Pressure-rate-of-strain tensor

= +

Responsible for redistribition of fluctuations beween different directions

=⋯ ∉ , =0
For pressure fluctuations:

1 ′
= −2 − −

=0
Decompose pressure 1
+ =− +⋯
( ) ( ) ( )

Rapid: Slow: Harmonic: 28

 Pressure-rate-of-strain tensor
1
′ + ′ + ′ = −2 − −

1
Æ∞ rapid = −2

→0 1
slow = − −

harmonic =0 only important close to walls

The same can be done for the pressure-rate-of-strain tensor

( ) ( )
( ) ( )
= + = + 29
 First: turn to the ”slow” pressure
1
′ + ′ + ′ = −2 − −

Consider homgeneous decaying turbulence ( = 0, but anistropy is still present):

+ = − +Π − +

Slow velocity Π = ( ( ) ( ) ( )
+ + ) −
press-gradient
( )
Slow pressure ( )
= +
rate-of-strain

A greatly-simplified
Reynold stress equation for ( )
homogenous turbulence
= −
30
 Understand the slow pressure-rate-of-strain
- in decaying homgeneous anistropic turbulence

( )
= −

1
= −
2 3

Homogenous flow,
1 1
= ⋅ − ⋅ no production
2 2
=−
reformulated − 1 −
in terms of = − +
2 3
anisotropy
2
tensor − + 2 +
3
=
2 Kolmogorov local
isotropy hypothesis

= + ε ij = 23 εδ ij
2
31
 Model the slow pressure-rate-of-strain

=+ + =+ ⋅ ⇒ =
2

If term is zero, all 6 elements of anisotropy tensor will increase exponently in time

A modelling Ideal: assume that turbulence has a natural tendency to

become less anisotropic as it decays.

Linear, return-to-isotropy
Rotta(1951) :
=− −1

2
= −2 =− −
3
We need this
term in
Reynolds stress
eq. 6η 2 = bii2
32
6ξ = b
3 3
ii
 (Nonlinear) Model of slow pressure-rate-of-strain

=− −1 Linear: c⋅

Nonlinear: C ⋅

1 ( )
1 1
= 1+ + − ?
2 2 3
Rotta(1951) :
Although is zero trace, is not nessary zero trace.

… can be (linearly) expressed by

and ; due to Cayley-Hamilton theorem
for 2nd order tensor .
(Sakar & Speziale, 1990)
Characteristically polynomial:
( )
= −3.4; = 4.2 + ⋅ + ⋅ + =0
Plug into the characteristic polynomail
33
+ ⋅ + ⋅ + =0
 Now turn to rapid pressure : Rapid distortion theroy(RDT)
What is Rapid distortion?
Turbulence-to-Mean strean time scale ratio is large! = ≫1
1/ /

Rapid mean-flow
distortion:
→ −∞ ; → +∞
satisfy continuity,
+ + =0

turbulent velocity fluctuations ( )

 Rapid distortion theroy(RDT) for homogenous turbulence
The governing equations for fluctuating velocity and pressure reduce to a set of linear P.D.E.s.

Eq. for pressure

1
′ + ′ = −2 − −

Eq.s for the fluctuating velocity

1 + ′ ′
+ + + + − =0

There is no nonlinear term anymore (powerful assumption),

analytical results can be found.
How? Firstly lets start from a single Fourier wave.
 A 1D example
“Rapid distortion” of a almostly frozen “scalar fluctuation”

A kind of elastic process

Initially
+ +
- -

⋅ The wave
number
| ( )|
increases
with time
After a while + +
for this
- - particular
“distortion”
⋅
A glimpse of rapid distortion theroy in 2D
“rapid” deformation of 2D “frozen” fluctuating scalar field

-
+ + -
- -
- +
+ + ( )
- -
+ ⋅
- “Initially” a single flutuating wave
( )⋅
evolve as
- -
+ +
- -
+ +
- -
+ +
- -

Mean strain( ⇐ / =0 Î Same area after distortion) 37

Through , ( ) rotates, changes magnitude | |
Rapid distortion theroy: single-wave result

Rotate ̂
= =− ̂ ( − ̂ ̂)
direction

′
=− ′ ( −2 ̂ ̂ )

Following this “single” time-evolving wave (t), the fluctuation amplitude ( )

evolves independent of length of wave-vector | |, only on its orientation .

⋅
′

Actual turbulence fluctuations should be summed for all possible wave 38

vectors. Since RDT equations is linear, allowing each wave to evolve
independently(then to be summed)
 Rapid distortion theroy: Linearly-summed-all-waves
Rapid-straining turbulence in a periodic domain : ⋅
′

( )
= +

( )
=2 ( + )

∗
= ̂ ̂

is a fourth order tensor, depends on direction of Fourier mode

(partially contained in the velocity-spectrum) , which is not contained in
the Reynolds-stress ( ).
Two turbulence states may start with the
same but different , their temporal
evolutions are not uniquely determined by .
( ) 39
This can be a limitation for modelling , Two routes return-
( ) ( )
same applies to . to-isotropy for
 An example: shear rapid distortion

direction 1

t=0 t>0

direction 2

2
−
= 3 Later, anisotropy ++
2

( )
= +
Initial, isotropic turbulence Evolve in time due to combined 40
( )
effect of
 ( )
RDT model for rapid pressure
Initial, isotropic turbulence under rapid, mean distortion :

RDT =− −

( ) 4 3
= =− 2
5 5 =− +Ω −⋯
3

( ) 3 2 =−
Only valid for =− ( − )
5 3
= , zero trace
( )
= +

Anisotropy Counteracts 60%

production of production
 Part 3
• Reynolds stress models
– Combined models for the pressure-rate-of-strain terms
• Slow pressure model + rapid pressure model
– Model the turbulence and pressure transport terms
• Gradient diffusion
– Model dissipation
• Isotropic model
• Non-isotropic model (Near wall consideration)

• A hierarchy of other turbulence models

– Algebraic RS models
• Simple but effective approaches inspired from RS-models
– Nonlinear eddy-viscosity models
– Elliptical relaxation model
42
 Put things together
A basic combined model for pressure-rate-of-strain term

Linear Rotta model 2

= −2 =− −
for slow pressure : 3

Isotropization of ( ) 3 2
Production for rapid =− ( − )
5 3
pressure

( ) ( ) 2 2
LRR-IP = + =− − − ( − )
3 3

(Launder, Reece, Rodi , 1975)

Simply, widely used, resonable-performence

 Other “advanced” pressure-rate-of-strain models
All these ”advanced” models can Key modelling ingredients:
be described in a general form: 1: Input tensor: ,
Tensor requirement
= 1.1 Symmetry (antisymmetric)
Æ Ω
=
1.2 deviatric (zero trace)
1 2
= − −
3 3
= 1.3 Normalize (non-dimentional)
2 = ⋅
= + −
3 2: Linear / nonlinear
=Ω − Ω Linear:
2 slow: c ⋅
= + − rapid: c ⋅ or c ⋅ Ω
3
=Ω − Ω Nonlinear:
slow: C ⋅
1
= − rapid: none ( Mijkl is linear)
3 …….
Other “advanced” pressure-rate-of-strain models
some examples with given coeffcients

LRR-IP LRR-QI SSG

Launder et al 1975 Launder et al 1975 Speziale et al 1991 =

−2 −2 ∗
− −

0 0
4 4 − 6 ∗

3 5
2 6
2+3
11
2 2
10 − 7
11
0 0 0
0 0 0
0 0 0

Note that only SSG has non-linear return to isotropy.

 Part 3
• Reynolds stress models
– Combined models for the pressure-rate-of-strain terms
• Slow pressure model + rapid pressure model
– Model the turbulence and pressure transport terms
• Gradient diffusion
– Model dissipation
• Isotropic model
• Non-isotropic model (Near wall consideration)

• A hierarchy of other turbulence models

– Algebraic RS models
• Simple but effective approaches inspired from RS-models
– Nonlinear eddy-viscosity models
– Elliptical relaxation model
46
 Modelling Reynold stress equations

+ = − +Π − +

Production tensor =− + Viscous

Turbulent diffusion
transport
Dissipation tensor =2

Velocity-pressure-gradient tensor

Π =− +

= + − +

= −

Pressure-rate-
Pressure
of-strain tensor 47
transport tensor
 Inhomogenous tubulence(Model turbulent-transport term)

ui′ p′ u ′j p′
Tijk(t ) = ui′u ′j uk′ + δ jk + δ ik
Turbulent ρ ρ
transport
Normally the pressure transport is neglected (or Pressure
implicitly included in the modelling of the turbulent transport
transport) on the basis of the energy budget.
2 ∂ u ′u ′
Gradient diffusion models: k
Tijk(t ) = −Cs Shir (1971)
i j
ε ∂xk
k ∂ui′u ′j
Tijk = −Cs uk′ ul′
(t ) Daly & Harlow (1970)
ε ∂xl
k 2 ⎛⎜ ∂u′j uk′ ∂ui′uk′ ∂ui′u ′j ⎞⎟ Mellor & Herring
ui′u ′j uk′ = −Cs ⎜ + +
ε ⎝ ∂xi ∂x j ∂xk ⎟⎠ (1973)

k⎛ ∂u ′j uk′ ∂ui′uk′ ∂ui′u ′j ⎞

ui′u ′j uk′ = −Cs ⎜⎜ ui′ul′ + u′j ul′ + uk′ ul′ ⎟ Hanjalic & Launder
ε⎝ ∂xl ∂xl ∂xl ⎟⎠ (1972) 48
 Part 3
• Reynolds stress models
– Combined models for the pressure-rate-of-strain terms
• Slow pressure model + rapid pressure model
– Model the turbulence and pressure transport terms
• Gradient diffusion
– Model dissipation
• Isotropic model
• Non-isotropic model (Near wall consideration)

• A hierarchy of other turbulence models

– Algebraic RS models
• Simple but effective approaches inspired from RS-models
– Nonlinear eddy-viscosity models
– Elliptical relaxation model
49
 Dissipation rate

+ = − +Π − +

Accurate dissipation tensor =2 =

3 + +
Simplification: use = =
Kolmogorov’s local 2 3
isotropy hypothesis:

2 2 0 0
= = 0 0
3 3
0 0
 Dissipation rate

Near wall effects:

• Low Reynolds number
• High shear rate
• Two-component turbulence ( fluctuaiton in direction is grealty inhibited)
• Wall blocking (pressure reflections)

Hence, dissipation can be anisotropic close to walls, but also in

moderate Reynolds number flow and for large mean rate of strain.
 A demonstration of anisotropic dissipation near wall using DNS

• Near wall ( y+<20), = does not hold.

(To be isotropic, it should be ≈ ≈ and ≈ 0)

Π Π

Π Π

Π
Π

Π Π
52
 Dissipation rate Note: All 6 Reynold
stresses(containing ) will
Note: the isotropic part of always be zero at wall.
disspation is not zero
at the wall = 0!
′ ′
= ≠ 2 ≠2

Two component behavior ′ ′

when approching wall → 0 =2 ≠ 2
(Exact analysis: series
expansion) ′ ′
=4

∗
′ ′ a simple model by Rotta (1951)
Near walls region(anisotopic) =
(Not very accurate!, but not very far off)
2
Far from wall (isotropic) =
3

Using a blending function = 1 at y = 0 and → 0 →∞

For all regions: ∗ 2

= + (1 − ) 53
3
 Dissipation rate

A more accurate model accounting for the two-component behaviour close

to the wall:

∗
′ ′ + ′ ′ + ′ ′ + ′ ′ ⁄
=
5
1+ ′ ′
2

: Wall normal direction vector

Simiarly blend this near-wall model with the isotropic model

∗ 2
For all regions: = + 1−
3
 The harmonic pressure
At y=0 the pressure fluctuations are:

′ ′
=

Now consider the Poisson equation for pressure fluctuations:

1 ∂ 2 p′ ∂ui ∂u ′j ∂2
= −2 − (ui′u′j − ui′u′j )
ρ ∂xk ∂xk ∂x j ∂xi ∂xi ∂x j
This is independent of viscous effects. Hence for the rapid
and slow parts one uses the inviscid boundary condition and
for the harmonic part the viscous boundary condition, i.e.:

′
=0 =0 =

Note that p(h) does not affect beyond y+=15 55

 Part 3
• Reynolds stress models
– Combined models for the pressure-rate-of-strain terms
• Slow pressure model + rapid pressure model
– Model the turbulence and pressure transport terms
• Gradient diffusion
– Model dissipation
• Isotropic model
• Non-isotropic model (Near wall consideration)

• A hierarchy of other turbulence models

– Algebraic RS models
• Simple but effective approaches inspired from RS-models
– Nonlinear eddy-viscosity models
– Elliptical relaxation model
56
Can Reynold-stress equations be reduced?

In previous slides, ”modeling” of all nessary terms in 6-

Reynold Stress equations and 1 disspation equaiton are
presented.

However solving 7 RS model equations are more

expensive than standard 2-equation models , is there other
simplficaton to avoid solving 5 extra equations while
retaining some benefits of RS models in terms of better
describing RS anistropy. (It should be better than those
models based on Boussineq eddy-viscosity assmption).

Yes, that is algerbric stress model.

57
 Compare Reynold stress and k-equations

+ = − +Π − +

Regroup R-S equation

Six eqs. 2
+ + = + −
3

contains time/spatial-derivative
Let’s call it , then for the unknown ,
this must be called can we simplify 6-elemement- by
represented it using the single-
element- without derivative ?

j=i , take half

Æ one k-eq. + + = −
 Algebraic stress model
Whatever pressure-models:
LRR-IP ; LRR-QI ;SSG , etc…

2 Isotropic
Six eqs. + + = + − dissipation
3

Rodi’s Ideal
1972

Six Reynolds stress

multiply one k-eq × + + = − ×

Price to pay: (Weak equilibrium assumption) Gained Benefits:

(Simple Algebraic relations to
Approximation ≈ × Î = , compute , no need to solve
may not be accurate. P.D.E.s)
 Algebraic stress models
Weak-equilibrium assumption (Rodi, 1972)

∂ u′u′ ∂ u′u′ u′u′ ⎡ ⎤ ⎡ ⎛ u′u′ ⎞ ⎛ u′u′ ⎞⎤

i j i j i j ∂k ∂k ⎢ ∂ ⎜ i j ⎟ ∂ ⎜ i j ⎟⎥
+u = ⎢ +u ⎥ +k +u
∂t k ∂x k ⎢ ∂t k ∂x ⎥ ⎢ ∂t ⎜ k ⎟ k ∂x ⎜⎜ k ⎟⎟⎥
⎜ ⎟
k ⎣ k⎦ ⎣⎢ ⎝ ⎠ k⎝ ⎠⎥⎦

1
≈ × Î =
2

2
= 2 +
3
 Algebraic stress model
Whatever pressure-models:
LRR-IP ; LRR-QI ;SSG , etc…

2 Isotropic
+ + = + − dissipation
The Algebraic stress model replaces boussineq 3
assumption.
− =− + = −2
3: Solve five light weight algebraic
to be used in the relations to compute the remained
(A) Three averaged momentum eq.
(B) Production terms ( ) 2
− = + −
3

1: Solve one expensive

transport equation for k
+ + = − ×

2: Solve another expensive transport −

equation for
Algebraic stress models vs. k-e model

62
 Non-linear eddy viscosity model
and Ω are the symmetrical
and antisymmertical part of
mean velocity strain tensor

Linear: c⋅ , or c ⋅ Ω
2 Nonlinear: C ⋅ Ω , or C ⋅
− = −2
3

+ + + Ω + Ω + Ω Ω + Ω Ω

Nonlinear model with possible combinations of

Match the unit tensor product of 2nd order,
dimension

63
 Elliptical relaxation model
( )
Motivation: “Non-local” model of the anisotropic redistribution term ( ≡Π − Π ) for
inhomogeneous turbulence (Wall bounded flow)
Previous "local" model (note: neglect the difference − = − )
∗ ∗ ∗ ∗ |
, = , , , , , , ∗

Desired model
( ∗, ) =
(.., ∗
,..) + remote effects

# ( )
+ + = + −

∗ ( )
+ + = + +

Use this particular form of eq. is to (i) avoid extra ( ) ( )

( ) ≡ −( − )
model for , (ii) = 0 at wall, easy to handle 64
 Elliptical relaxation model
( )
Now is able to receive non-local, remote contributions, it
∗
= model if L = 0 and ?
also recovers the original

∗ ( )
+ + = + +

6 simple relations: , , , = , , , x, y, z, t

6 Elliptical relaxation Poisson equations to solve . − =

( , , , )
(x, y, z, t) − L ( , , , ) + + =
( , , , )

If this 1 is removed, it is standard LRR-IP

= max ( , ) model for the term (different than )

−1 2 2
, , , =− − − ( − )
Time scale: 3 3

= max ( , 6 )
65
 ( )
This slide summarizes the relations between and ; and

compare eq. (11.138) in page 429 with eq. (7.187) in page 315, pope’s book.
Decomposition 1:
Eq. 7.193, p. 319
1
Π =− + ( ) 1
= ( + )
Eq. 11.3,
p. 388 1 1
+ − +

= − Eq. 11.3, page 388

Eq. 7.186, p.317
Eq. 7.192, p.319
Eq. 7.187, p.317 1
Π =− +
Decomposition 2:
1 1
Π = (Π − Π ) +( Π ) ( Π )=− +
3 3
( )
Eq. 11.5, p. 389 = − ( ) 1
− 2 −
=
( )
−
( ) 3

2
−
3
Eq. 11.6, p.389 ( ) 2 2
= =−
3 3
Eq. 11.140, page 429 66
Additional slides

67
 Reynold stress models
A) ( , )

B) , ≡ , , , , ,

, ∗ ∗
Select ( , ) from this set
C) , ≡ , , ∗
, , ,

D) , ≡ = ( , )

Transport equation of in general P.D.E. form : (Be aware of boundary conditions)

∗ ∗ ∗
, + , + ( , )= , − ,
68
 When Reynolds stress model equations are solved, we will get all 6
elements of Reynold stress at our disposal, our interests goes to
characterize the Reynold stress anisotropy.

Perfect 2
∗ ∗ 0 0
isotropy: 3
, ≡ ∗ 2
, = 0 0
3
2
0 0
3
some tentative examples of deviation from
perfect isotropy base on “tensor structure”

a) unequal diagonal terms” b): nonzero off-diagonal term” c): combined a & b
2
2 2 − 0
− 0 0 0 3
3 3 2
2 2 − 0
0 − 0 0 3
3 3 2
2 2 0 0 +( + )
0 0 +( + ) 0 0 3 69
3 3
To characterize the Reynold stress anisotropy, it naturally goes with
(i) first remove from the all diagonal terms and
(ii) then normalize by 2 ,
that gives us the anisotropic tensor :
2
− 2
= 3 = − /2
2 3

a) unequal diagonal terms” b): nonzero off-diagonal term” c): combined

2 2 2
− 0 0 0 − 0
3 3 3
2 2 2
0 − 0 0 − 0
3 3 3
2 2 2
0 0 +( + ) 0 0 0 0 +( + )
3 3 3

− ∗ 0 0 0 ∗
0 − ∗ ∗
0
0 − ∗ 0 ∗
0 0 ∗
− ∗ 0
∗ ∗
0 0 + 0 0 0 0 0 ∗
+ ∗

Superscript * represented division by 2 70

Using tensor invariants (i.e. expression formed by eigenvalues of )
is a better way to characterize the anisotropic tensor than a simple
tensor-structure based classification of anisotropy

a) unequal diagonal terms” b): nonzero off-diagonal term” c): combined

− ∗ 0 0 0 ∗
0 − ∗ ∗
0
0 − ∗ 0 ∗
0 ∗
− ∗
0
∗
0 0 + ∗ 0 0 0 0 ∗
+ ∗

∗
− 0
∗ ∗ ∗, ∗,
− = − = − =0 = =− =0
0 −

Choosing another coordinate, Tensor invariants gives a

∗ coordinate-independent
the (b) “non-zero off-diagonal” + 0 0
problem is equivalent to (a) ∗
description of anisotropy
0 − 0
“unequal diagonal terms” 0 0 0
71
Plot the resolved 5 anisotropic Reynolds stresses into the Lumley triangles of 2-dimentions

2
= − /2
3

= + + =0
= + + = −3
72
= =2