15th Australasian Fluid Mechanics Conference The University of Sydney, Sydney, Australia 13-17 December 2004

Turbulent Stress Invariant Analysis: Clarication of Existing Terminology

address: SINTEF Energy Research, Energy Processes. Kolbjrn Hejes v. 1b, N-7065 Trondheim, Norway of Energy and Process Engineering, The Norwegian University of Science and Technology, N-7491 Trondheim, Norway

2 Department

Abstract

As a means of studying the structure of the turbulence, the time mean invariants, dened according to the theory of Lumley [3] has proven to be a useful and popular tool. According to the theory there is a domain, known as the Lumley triangle, within which all realizable Reynoldss stress invariants must lie. The borders of this domain describes different states of the turbulent stress tensor. It has been found that there is some confusion in the terminology used when describing these states. The confusion is related to whether the notation is used to describe the shape of the stress tensor or the eddies of the turbulence. Choi and Lumley [1] noted the same controversy in terminology, but since the confusion is still found to exist the current work describes the fundamental relationships which exist between the shape of the stress tensor and its invariants. Also an analysis is given to clarify the existing terminology.
Introduction

Assuming incompressible axisymmetric turbulence with y as the symmetry axis, and using cylindrical coordinates, the invariants may be expressed as I1 I2 I3 = = = 0 1 2 2 + b2 [b2 yy + b + 2bry ] 2 rr [brr byy b2 ry ]b (8)

In solving the characteristic equation, equation (5), for a given set of invariants the eigenvalues and the related eigenvectors of the tensor are revealed. This gives the main stress directions of the stress tensor and their magnitude. Several solutions are possible depending on the stress eld, but there are also denitive bounds with respect to the realizability of the ow which have led to the anisotropy invariant map (initially proposed by Lumley [3]) and therefore often referred to as the Lumley triangle) given in gure 1.

Any symmetric tensor, like the stress tensor i j , may be divided into a traceless deviator, i j and an isotropic part, i j i j = i j + i j where 1 = kk i j 3

1 Component 0.3 0.25

(1)

(2)
I2

2 Component 0.2 2 Component Axisymmetric

and

1 i j = i j kk i j 3

(3)

0.15 0.1

In the case of true isotropy all the elements in i j would become zero. The non-dimensional form of the anisotropy tensor is given by i j i j 1 bi j = = i j (4) kk kk 3 By using the Cayley-Hamilton theorem, the characteristic equation for any second order symmetric tensor may be written as I1 + I2 I3 = 0
3 2

Axisymmetric I3>0

0.05 Axisymmetric I3<0 0 0.04 0.02 0 I3

Isotropic 0.02 0.04 0.06 0.08

Figure 1: The anisotropy invariant map, AIM. At the origin (I2 = I3 ) the isotropic state is found. From this point two limiting lines are found where the ow is assumed to be axisymmetric (i.e. bry = 0 and two of the diagonal elements are equal e.g. brr = b ). This leads to the relation I3 = 2(I2 /3)3/2 . Each of these lines have a limiting value. In the rst case the two identical diagonal terms are smaller than the third element and tend to zero. This leads to a 1 component state of turbulence (1C in gure 1). The other possibility is that the two identical elements dominate so that the third component is ignorable. This leads to the two component axisymmetric state indicated in gure 1. The line joining 1- and 2-component axisymmetric turbulence represents the all other possible states where only two diagonal components exist. As a tool in describing the state of the turbulence Lee and

(5)

where the I s are the rst, second, and third invariants of the tensor and is the principal stress. The invariants are related to the tensor terms according to the relations I1 I2 I3 = = = bkk 1 bi j b ji 2 det (bi j ) (6) (7)

where the ow has been assumed to be incompressible.

Reynolds[2] designated different shapes of the stress eld to the value of the invariants in equation (5) such as rod and disk like turbulence. However, in their description Lee and Reynolds were focusing on the turbulent eddies rather than the stress eld itself. This has over the years lead to a confusion when describing the stress eld since the characteristic shapes related to the stresses and the eddies are not generally the same. If not aware of this the axisymmetric limits of the Lumley triangle will be described by shapes that do not relate to the stress tensor the invariants represent. In the following the relations between the stress eld and the invariants are shown and then related to the shape of the stress tensor. This will be compared to the shape of the turbulent eddies for the same type of turbulence. It is must be emphasized that this analysis does not have an inuence on turbulence modelling, but have implications in understanding and visualization of the stress eld. It must also to be noted that the aim of the current paper is to clarify the terminology used.
The characteristic shape related to the stress tensor

By using the equation (10) and the matrix relation (AB)T = BT AT the following is obtained xT x = yT 2 y = 1 or y1 1
2

(15)

(16)
2

y2 2

y3 3

=1

(17)

where relation 13 is used. This describes the shape of a spheroid in the coordinate system (y1 , y2 , y3 ). In order to establish a relationship between the shape of the stress tensor and the invariants of the anisotropy tensor we need 1 T 3 I) and T. From matrix to relate the eigenvalues of b = ( T kk theory we know that if for two matrices A and B AB = BA (18)

In terms of the stress tensor the visualization of the stress eld is determined by the relative sizes of the three main stresses and the designations of the axisymmetric limits are based upon the following analysis. The Reynolds stress tensor, i j , is a second order symmetric tensor. As stated by the spectral theorem every real symmetric matrix can be diagonalized by an orthonormal matrix. In other words the Reynolds stress matrix, 11 12 13 T = i j = 21 22 23 31 32 33 may be written as where T = Q Q1 1 = 0 0 0 2 0 0 0 3 (9)

then the matrices A and B share the same eigenvectors. Since the tensor 1/3i j is equal to the identity matrix I/3, relation 18 always holds between the matrices T and I. They therefore share the same eigenvector n and the following relationship will hold. T 1 bn = ( I)n (19) Tkk 3 The relationship between the eigenvalues of b and T then becomes i 1 i = . (20) Tkk 3 For the non-dimensional anisotropy tensor, b, related shapes of the ellipsoid formed by the Reynolds stresses are illustrated in gure 2 and characteristics of the ow are given in table 1.

is the eigenvalue matrix of T and i are the eigenvalues. Q is an orthonormal matrix with the properties QQ1 = QQT = I. (10)

Now, recalling that the stress, f , in the direction of a vector n is represented by the linear operator T on R3 (T: R3 R3 where both n, f R3 ): f = T n (11) By the use of relation 9 and left-multiplying with Q1 this may be rewritten as y = Q1 f = Q1 n = x, (12)

which means that by changing the old coordinate system into a new coordinate system, coinciding with the principal axis of the stress tensor, T becomes a diagonal matrix with the principal stresses of T on its diagonal. If x is considered as the normalized stress vector of an isotropic eld describing the unit sphere we have x xT x = 1 = x1 , x2 , x3
2 2 2 x1 + x2 + x3 = 1.

(13) Figure 2: Illustration of the ellipsoid shapes formed by the Reynolds stress tensor in different regions of the ow.

Equation (12) is now representing a distortion of this stress vector due to the mapping T. Left-multiplying equation (12) with 1 gives x = 1 y (14)

State of turbulence Isotropic Axi.sym. (One large ) Axi.sym. (One small )

Invariants I2 = I3 = 0
3 2/3 2 = ( I2 ) I3

Eigenvalues of bi j i = 0
1 0 < 1 < 1 3 , 6 < 2 = 3 < 0

Shape of stress tensor (see eq. 17) Sphere Prolate spheroid Oblate spheroid

Symbol in gures 1 and 2 a, isotropic b, axisymmetric (I3 > 0) c, axisymmetric (I3 < 0) d, 1 Component e, 2 Component, (axisymmetric) f, 2 Component

I3 2/3 2 I3 = ( 2 )

1 3 < 2 < 0, 0 < 1 = 3 <

1 6

One-comp. Two comp., axisym.

2 I3 = 27 , I2 = 1 3 1 I3 = 108 , 1 I2 = 12 1 I2 = 3( 27 + I3 )

Line Disk

Two component

1 1 + 3 = 1 3 , 2 = 3

Ellipsoid

Table 1: Characteristics of the turbulence stress tensor and anisotropic tensor.

Clarication of Terminology

Now it is time to address the misconception encountered in the designation of the limits of Lumleys invariant map. Axisymmetric turbulence means that two of the principal stresses i , or i , are equal. Writing b in terms of the principal direction we have 1 0 0 (21) b = 0 2 0 0 0 3 and for axisymmetric turbulence we have e.g. 2 = 3 . Together with relationship 8, 1 + 2 + 3 = 0, this gives for the invariants I2 and I3 3 I2 = 2 4 1 , 1 I3 = 3 4 1 (22)

The error in the designation encountered in articles and textbooks is that the two axisymmetric limits with rod like and disk like turbulence are switched because some researchers talk about the shapes of the turbulent eddies rather than those of the stress tensor. In terms of vorticity the visualization of the turbulence would be opposite to that of the stresses. This may be exemplied by the ow through an axisymmetric contraction as shown in gure 4. Figure 4A) illustrates a streamwise vortex tube before and after a contraction. For an incompressible uid the transport equation for vorticity is given by the Helmholtz equation D = ( )U + 2 , (26) Dt where U is the uid velocity and the vorticity is dened by the curl of the velocity = U. (27)

From the previous analysis it is clear that if we have one large principal value 1 and two equal but smaller principal values 2 = 3 , the turbulence eld will be stretched in the n1 direction creating the prolate spheroid of gure 2 b). On the other hand, if 1 is smaller than the other two, the shape of the stress tensor would be the oblate spheroid of gure 2 c). This gives the following relationship between the principal values, invariants, and shape. 1 > 0 1 < 0 I3 > 0 Rod-like turbulence I3 < 0 Disk-like turbulence (23) (24)

The rst term on the right of (26) is called the vortex stretching term whereas the second term is a viscous diffusion term. Through the contraction the streamwise component of the vorticity 1 = is increased by vortex stretching term since it includes the term 1 U1 /x1 . U1 /x1 is positive due to the decrease in ow area and therefore a streamwise vortex tube is stretched and obtains a rod-like shape after the contraction. At the same time, due to the continuity requirement U = 0, the vorticity in the directions normal to the streamwise direction, 2 and 3 , are attenuated. For the turbulent stress tensor, on the other hand, the production of the streamwise turbulence component is suppressed by the secondary production term 11 U1 /x1 (see [4]). 11 is always positive and since the streamwise gradient is positive the production term will become negative. The result is that the stress ellipsoid, according to the analysis in the previous section, becomes attened in the streamwise direction and obtains a disk-like shape as illustrated in gure 4B).

Eliminating 1 from the relations in 22 we get I3 = A 2 I2 3

(25)

The term in the parenthesis will always be greater than zero since I2 0 (see equation (7)), so the sign of I3 will be determined by A. Comparing this with the statements 23 and 24, A will have the following values A= 1 1 , , Rod-like turbulence Disk-like turbulence

Summary

With respect to the invariant map of Lumley and in terms of the shape of the stress tensor this gives the designation of axes shown in gure 3.

The aim of the current paper is to clarify the terminology used when the axisymmetric limits of the Lumley triangle are related

Figure 3: Anisotropy invariant map with correct designation of the axisymmetric states.

X2

X2

X3

X1

X3

X1

Figure 4: Illustration of streamwise vortex tube stretching(A) and stress ellipsoid deformation(B).

to shapes of the turbulence. As demonstrated by the above discussion the shape designated to the axisymmetric limits in the invariant map may be interpreted in two ways. Although this has no effect on turbulence modelling it does lead to some confusion when visualizing the turbulence. Since the invariant map itself is based on the turbulent stress tensor, it would be preferable that the designation of its axis is related to this quantity only.

[4] Torbergsen, L., Experiments in turbulent pipe ow, Ph.D. thesis, Norwegian University of Science and Technology, 1998.

References

[1] Choi, K. and Lumley, J., The return to isotropy of homeneous turbulence, J. Fluid Mech., 436, 2001, 5984. [2] Lee, M. and Reynolds, W., Numerical experiments on the structure of homogeneous turbulence, Technical Report Rep. TF-24, Thermoscience Division, Stanford University, 1985. [3] Lumley, J., Computational modelling of turbulent ows, Adv.Appl.Mech., 18, 1978, 123176.