Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
University of Aix-Marseille & C.N.R.S. U.M.R. 6595, IUSTI, 5 rue E. Fermi, 13453 Marseille Cedex 13, France
University of Aix-Marseille & C.N.R.S. U.M.R. 7340, Case 322, Av. Escadrille Normandie-Niemen, 13397 Marseille Cedex 20, France
a r t i c l e
i n f o
Article history:
Received 14 March 2011
Received in revised form 21 September 2011
Accepted 21 November 2011
Available online 11 April 2012
Dedicated to Professor Victor Berdichevsky
on the occasion of his 65th birthday
Keywords:
Reynolds stress tensor
Capillary uids
Variational principle
a b s t r a c t
The dynamics of the Reynolds stress tensor for turbulent ows is described with an evolution equation coupling both geometric effects and turbulent source terms. The effects of
the mean ow geometry are shown up when the source terms are neglected: the Reynolds
stress tensor is then expressed as the sum of three tensor products of vector elds which
are governed by a distorted gyroscopic equation. Along the mean ow trajectories, the uctuations of velocity are described by differential equations whose coefcients depend only
on the mean ow deformation. If the mean ow vorticity is small enough, an approximate
turbulence model is derived, and its application to shear shallow water ows is proposed.
Moreover, the approximate turbulence model admits a variational formulation which is
similar to the one of capillary uids.
2012 Elsevier Ltd. All rights reserved.
1. Introduction
The system describing turbulent compressible barotropic ows is composed of equation of mass balance, equation of
average momentum and evolution equation for the Reynolds stress tensor. In the following, we see that the Reynolds stress
tensor equation is mainly driven by the velocity gradient tensor of the mean motion and this equation is the main object of
our study when the source term is negligible.
The reason for considering the simplied turbulence model without the source term is twofold. First, in numerical studies
of compressible turbulent ows, this homogeneous equation is a natural step in applying the splitting-up technique (see for
example (Berthon, Coquel, Hrard, & Uhlmann, 2002)). Secondly, such a homogeneous system appears as an exact asymptotic model of weakly shearing ows of long waves over a at bottom (see (Teshukov, 2007)). The only difference is the space
dimension: two dimensions are considered for shallow water ows instead of three dimensions for the general case.
We use the spectral decomposition of the Reynolds stress tensor and in the homogeneous case, we obtain a simpler
dynamical system for the eigenvalues and the eigenvectors of the Reynolds stress tensor. The system admits a simple physical interpretation: the motion of each point of the turbulent ow is analogous to the motion of a free rigid body moving
along the mean ow and rotating with an angular velocity which is different from the mean ow vorticity. The angular velocity is completely determined by the mean ow velocity. The moments of inertia of the free rigid body are not constant, they
are also determined by the mean ow.
When the mean ow vorticity is small enough, an approximate turbulence model is obtained, which admits a variational
formulation.
Corresponding author.
E-mail addresses: sergey.gavrilyuk@polytech.univ-mrs.fr (S. Gavrilyuk), henri.gouin@univ-amu.fr (H. Gouin).
0020-7225/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijengsci.2012.03.008
66
8
>
< hqit hqiU i ;i 0;
hqiU i t hqiU i U j hpidij hqui uj i;j 0;
>
: hqu u i hqu u iU hqu u iU hqu u iU S
i j t
i j
i k
ij
k ;k
k j
i;k
j;k
where brackets mean the averaging, coma means the derivation with respect to the Eulerian coordinates x = {xi},
i 2 {1, 2, 3} and index t means the partial derivative with respect to time, q is the uid density, U = {Ui}, i 2 {1, 2, 3} is the
mass average velocity, p is the pressure, u = {ui}, i 2 {1, 2, 3} is the velocity uctuation verifying hqui = 0. Repeated indices
mean summation. Here S = {Sij} is a source term, and its explicit expression can be written as:
R hqu ui;
Rij hqui uj i:
8
@hqi
>
>
divhqiU 0;
>
>
>
@t
>
>
<
dU
hqi
rhpi div RT 0;
dt
>
>
>
T
> dR
>
@U
@U
>
>
S;
R divU
RR
:
dt
@x
@x
where d/dt means the material derivative with respect to the mean motion
d
@
UT r:
dt @t
The superscript T means the transposition. Using the mass conservation law, the equation for the volume Reynolds stress
tensor R can be rewritten as the equation for the specic (or per unit mass) Reynolds stress tensor
T
dP @U
@U
S
PP
;
dt @x
@x
hqi
where
R
:
hqi
The structure of source term S has generated much debate within physical and mathematical communities. Our goal is not to
add new closure hypotheses, but to study the structure of the master Eq. (3) when S = 0. The reason is twofold:
rstly, in the numerical study of compressible turbulent ows, this is a natural step in applying the splitting-up technique
(see for example (Berthon et al., 2002)),
secondly, system (2) also appears as an exact asymptotic model of weakly shearing ows of long waves (turbulent shallow water ows) over a at bottom (see (Teshukov, 2007)):
8
@h
>
>
divhU 0;
>
>
@t
>
>
!
>
>
2
< dU
gh
div RT 0;
r
h
dt
2
>
>
>
>
T
>
>
>
dR
@U
@U
>
:
R divU
RR
0:
dt
@x
@x
In system (4), h is the uid depth playing the role of the average density; the average pressure is given by hpi = gh2/2, g is
the gravity acceleration, and
Z
0
h
i
e U U
e U dz;
U
hU
Z
0
e
Udz;
67
e is the instantaneous velocity. Equations are written for three-dimensional long waves and the production term is
where U
zero in the limit of weakly shearing ows. Eq. (4) are hyperbolic (see (Teshukov, 2007) for proof). Finally, we will focus on the
equation of the Reynolds stress tensor per unit mass
T
dP @U
@U
PP
0:
dt @x
@x
The case S = 0 corresponds to conservative motions of turbulent compressible ows; these motions verify the equations
8
@hqi
>
>
divhqiU 0;
>
>
>
@t
>
>
>
<
dU
rhpi div RT 0;
hqi
dt
>
>
>
T
>
>
>
dR
@U
@U
>
>
R
divU
R
0:
:
dt
@x
@x
The particular case rotU = 0 was investigated in Debive, Gouin, and Gaviglio (1982). In such a case (@U/@x)T = @U/@x and Eq.
(5) corresponds to a two-covariant tensor convected by the mean ow. This means that P has a zero Lie derivative dL with
respect to the velocity eld U and the tensor P0, image of P in Lagrange coordinates (t, X), only depends on X = {Xi}, i 2 {1, 2, 3}
dL P
dP @U
@U
PP
0;
dt @x
@x
P F T 1 P0 XF 1 ;
3
P
k2a ea ea
a1
3
P
a1
k2a ea eTa :
The eigenvalues k2a ; a 2 f1; 2; 3g are non-negative; the case k2a > 0 is a generic one. For the two-dimensional case, k23 0. Let
us denote
aa ka ea ;
Then,
3
P
3
P
aa aa
a1
aa aTa :
a1
"
T #
3
dP P
daa T
daa
:
aa aa
dt a1 dt
dt
T
3
P
daa @U
daa @U
aa aTa
aa aTa 0:
@x
@x
dt
a1 dt
10
The vector daa/dt + (@U/@x)aa can be developed in the local basis {ab}, b 2 {1, 2, 3} of eigenvectors; one obtains
3
P
daa @U
aa Aba ab ;
@x
dt
b1
a 2 f1; 2; 3g
11
where Aba, (a, b 2 {1, 2, 3}) are the scalar components to be determined. By using Eq. (11), Eq. (10) leads to
3
P
Aaa gaa
a1
3
P
ab1
where gaa 2aa aTa and gab gba aa aTb ab aTa ; a; b 2 f1; 2; 3g, are six independent symmetric tensors. Consequently,
a; b 2 f1; 2; 3g:
68
daa @U
aa K ipea
@x
dt
a 2 f1; 2; 3g;
12
where a diagonal matrix K and an antisymmetric matrix i(p) are determined in the basis eb, b 2 {1, 2, 3} as
k1
KB
@0
0
k2
0
C
0 A;
k3
B
ip @ A21
A13
A13
A21
C
A32 A:
0
A32
aTa
dab
daa
aTb
0:
dt
dt
13
8a b 2 f1; 2; 3g;
@U
@U
ab aTb K ipea
aa 0;
aTa K ipeb
@x
@x
or
2ka kb eTa Deb aTa K ipeb aTb K ipea k2a eTa ipeb k2b eTb ipea ;
14
where
T !
1 @U
@U
D
2 @x
@x
is the rate of deformation tensor corresponding to the mean ow. We denote the mixed product of three vectors {a, b, c} as
(a, b, c) aT(b ^ c). Hence, Eq. (14) yields
2ka kb eTa Deb k2a k2b ea ; p; eb k2b k2a p; ea ; eb k2b k2a pT ec ;
where {a, b, c} is a cyclic permutation of the triplet {1, 2, 3}. Finally, we get
pT ec
15
ka
dea dka
@U
ea K ipea 0;
ea ka
@x
dt
dt
eTa
dea
0;
dt
a 2 f1; 2; 3g:
16
Since
dka T
ea Dea ka 0:
dt
By multiplying the left side of Eq. (16) with the projector I ea eTa , we get
ka
@U
dea
ea K ipea 0:
ka I ea eTa
@x
dt
dea
P ^ ea :
dt
Such a vector P should verify the condition
@U
ea K ipea 0:
ka iPea ka I ea eTa
@x
17
69
ka PT ec kb pT ec ka eTb
@U
ea :
@x
18
k2b eTa
PT e c
T !
@U
@U
eb
@x
@x
k2b
ka
eTb
@U
ea
@x
eTb k2b
!
T
@U
@U
ea
k2a
@x
@x
k2b k2a
R hqi
3
P
a1
k2a ea ea :
8
dea
>
>
P ^ ea ;
<
dt
2
>
>
: d Lnka 2l ;
a
dt
a 2 f1; 2; 3g;
19
where
eTb
la eTa Dea ; PT ec
k2b
!
T
@U
2 @U
ea
ka
@x
@x
k2b k2a
The triplet {a, b, c} corresponds to a cyclic permutation of the triplet {1, 2, 3}.
Eq. (191) is similar to the equations of a rigid body (see (Marsden & Ratiu, 1994)). The vectors ea form a natural moving
frame fea g3a1 whose evolution is determined by the mean rate of deformation tensor. The eigenvalues k2a of the Reynolds
stress tensor are determined by the evolution Eq. (192). Let us note that, if ka are initially positive, they will be positive
for any time. Hence, it means that the tensor P will always be positive denite. Due to the mass conservation law (21)
and Eq. (192), we obtain the following quantity conserved along trajectories of mean ow:
3
Q
d
hqi2 k2a 0:
dt
a1
Consequently, system (19) admits an invariant scalar along the trajectories of mean ow. This invariant was earlier obtained
in Debive, Gouin, and Gaviglio (1982); Debive et al. (1982) in a different form. Let us introduce the turbulent specic
energy
eT
3
1
1P
tr P
k2 :
2
2 a1 a
In the incompressible (isochoric) case, we have dhq i/dt = 0; the turbulent energy is minimal in the isotropic case when the
three eigenvalues k2a are equal k21 k22 k23 k2 . In this case, the orthonormal eigenvectors ea, a 2 {1, 2, 3} of the Reynolds
stress tensor P are also the orthonormal eigenvectors of the mean rate of deformation tensor D and la are the corresponding
eigenvalues.
In the compressible isotropic case eT = hqi2/3j, j = 3k2/(2h qi2/3), and j is a classical invariant of isotropic compressible
turbulence.
Q
In presence of shock waves the quantity hqi2 3a1 k2a is not conserved through shocks; it increases like the classical entropy in compressible uid dynamics. The estimation of the jump of turbulence entropy in isotropic case was given in Gavrilyuk and Saurel (2006).
As a consequence, the governing Eq. (6) admit the energy conservation law
@
1
1
div hqiU jUj2 ei eT hpiI RU 0
hqi jUj2 ei eT
@t
2
2
where the internal specic energy ei is dened by
1
dei hpid qi
h
70
and the mean pressure hpi is supposed to be a given function of hqi. Indeed, using (192) we immediately obtain
@
1
1
deT
div hqiU jUj2 ei eT hpiI RU hqi
hqi jUj2 ei eT
trRD
@t
2
2
dt
3
3
P 2
hqi d P 2
k hqitr
ka la 0:
2 dt a1 a
a1
System (6) is a conservative and Galilean invariant system of equations which is the counterpart of the Euler equations for
the turbulent compressible ows. The equations for the Reynolds stress tensor (63) are rewritten in a simpler form admitting
a clear physical interpretation.
4. An approximate model
In this Section, we derive a useful approximation of model (6) describing compressible turbulent ows for motions characterized by a small average vorticity.
4.1. Preliminaries
Eq. (5) can be rewritten as:
T
dP
@U
@U
PP
irotU; P
dt
@x
@x
20
with
irotU
T
@U
@U
:
@x
@x
Let s be a characteristic time scale and x a characteristic value of the mean vorticity norm krot Uk; we assume that
sx 1:
21
Relation (21) is veried, in particular, for motions which are close to one-dimensional ones. Eq. (20) gets the form
T
dP
@U
@U
0:
PP
dt
@x
@x
22
Using the solution (7) when P0(X) = I, we consider the Reynolds stress tensor in the form
P
a
rua rua ;
where index a ranges over a nite number of integers and ua are generalized Lagrangian coordinates:
dua
0:
dt
Covectors
T
ba
@ ua
@x
dba
T @U
0;
ba
@x
dt
T
corresponding to a zero Lie derivative of ba with respect to the mass average velocity. In such a case, Eq. (22) is identically
veried. Symmetric tensor P is determined by six scalar elds ua, a 2 {1, . . . , 6}. If, initially, vectors rua, a 2 {1, 2, 3} constitute an orthogonal system corresponding to the eigenvectors of P, we can choose ua = 0, a 2 {4, 5, 6} and consequently,
3
P
a1
rua rua :
23
It is worth to note that since the Reynolds stress tensor is positive denite, it can be considered as a metric tensor of a Riemannian space associated with the metric
Pij
3 @u @u
P
a
a
:
a1 @xi @xj
71
a1
i;j
It is interesting to note that this special structure of the Reynolds stress tensor implies a variational structure of Eq. (6).
4.2. The Hamilton principle
The aim of this Section is to prove that in special form (23), System (6) admits a variational formulation which is similar to
the one of capillary uids (Cahn & Hilliard, 1958; Casal, 1963, 1972; DellIsola, Gouin, & Rotoli, 1996; Eglit, 1965; Truskinovsky, 1982; van der Waals, 1979). However, in our case, the expression of the capillary energy is determined by the gradients of
three scalar order parameters transported along the trajectories of the mean ow.
We consider the specic internal energy in the form
ei ei hqi:
The mean density is submitted to the constraint
@hqi
divhqiU 0:
@t
Let us dene the specic turbulent energy as
eT
3 jru j2
P
a
;
2
a1
dua
0;
dt
a 2 1; 2; 3:
For a material volume Dt of the mean motion, the Hamilton action calculated between times t1, t2 is
t2
Z Z Z
t1
hqiLdv dt;
Dt
1 T
U U ei eT :
2
The uid motion is a C2-diffeomorphisme / from a three-dimensional space D0 into the physical space Dt:
x /X; t or xi /i X 1 ; X 2 ; X 3 ; t;
i 2 f1; 2; 3g:
Let a one-parameter family of virtual motions denoted by {/e}, possessing continuous derivatives up to the second order and
expressed in the form:
x UX; t; e;
with e 2 O; where O is an open interval containing 0 and such that U(X, t; 0) = /(X, t) (the real motion of the continuous medium is obtained when e = 0). The derivation with respect to e when e = 0 is denoted by d. Derivation d is named variation and
the virtual displacement d/ is the variation of the motion of the medium. At time t, the virtual displacement of the particle x
is dx obtained when dX = 0 and de = 1 at e = 0; the virtual displacement corresponds to the eld of tangent vectors to Dt
x 2 Dt ! f wx
@U
2 T x Dt ;
@ e
e0
da 0:
We have the following general results (see (Berdichevsky, 2009; Casal, 1963; Gavrilyuk & Gouin, 1999; Serrin, 1959)):
R
8 R t RR R
2
>
d
h
q
iLd
v
dt X hqidLdv dt;
>
t
D
t
1
>
>
>
>
>
df
>
<
dU ;
dt
>
>
dhqi hqidivf;
>
>
>
>
@ ua @f
> @ ua
>
:d
;
@x
@x @x
72
hqidL hqiUT
R
X
T
3
P
df
@ei
@ ua @f @ ua
:
hqi2
divf hqi
dt
@hqi
@x @x @x
a1
Let us denote hpi = hqi2@ei/@hqi, the mean pressure scalar eld of the uid; due to the identities,
8
>
@hqiUT f
dUT
T df
T
>
>
divh
h
q
iU
q
iU
f
h
q
i
f;
>
>
dt
@t
dt
>
>
>
<
@hpi
f;
hpidivf divhpif
@x
>
!
!
!
>
T
T
T
>
T
>P
3
3
3
3
>
P
P
P
@ ua @f @ ua
@ ua @ ua @f
@ ua @ ua
@ ua @ ua
>
>
h
q
i
tr
h
q
i
div
h
q
i
div
h
q
i
f
f:
>
:
@x @x @x
@x
@x @x
@x
@x
@x
@x
a1
a1
a1
a1
!
T
@hqiUT f
@ ua @ ua
; divhqiUT f; divhpif and div hqi
f can be integrated on @ X
Stokes formula implies that terms
@t
@x
@x
where f vanishes. We get for each eld of virtual motion,
!)
T
Z (
3
dUT @hpi P
@ ua @ ua
da
hqi
div hqi
f dv dt;
@x
dt
@x
@x
X
a1
and the fundamental lemma of variational calculus yields the equation of motion
hqi
3
P
dU
rhpi div hqi rua rua 0;
dt
a1
73
Ganapathisubramani, B., Longmire, E. K., & Marusic, I. (2006). Experimental investigation of vortex properties in a turbulent boundary layer. Physics of Fluids,
18, 055105.
Gavrilyuk, S., & Gouin, H. (1999). A new form of governing equations of uids arising from Hamiltons principle. International Journal of Engineering Science,
37, 14851520.
Gavrilyuk, S., & Saurel, R. (2006). Estimation of the turbulence energy production across a shock wave. The Journal of Fluid Mechanics, 549, 131139.
Marsden, J. E., & Ratiu, T. S. (1994). Introduction to mechanics and symmetry. Series in applied mathematics (vol. 17). Berlin: Springer.
Mohammadi, B., & Pironneau, O. (1994). Analysis of the K-epsilon turbulence model. Research in applied mathematics. New York: John Wiley & Sons.
Pope, S. B. (2005). Turbulent ows. Cambridge University Press.
Serrin, J. (1959). Mathematical principles of classical uid mechanics. In Encyclopedia of Physics VIII/1 (pp. 125263). Berlin: Springer.
Teshukov, V. M. (2007). Gas dynamic analogy for vortex free-boundary ows. Journal of Applied Mechanics and Technical Physics, 48, 303309.
Truskinovsky, L. (1982). Equilibrium phase boundaries. Soviet Physics Doklady, 27, 551553.
van der Waals, J. D. (1979). The thermodynamic theory of capillarity under the hypothesis of a continuous density variation. J. Stat. Phys., 20, 197244.
Translated by J. S. Rowlinson (1893).
Wallace, J. M. (2009). Twenty years of experimental and direct numerical simulation access to the velocity gradient tensor: What have we learned about
turbulence? Physics of Fluids, 21, 021301.
Wilcox, D. C. (1998). Turbulence modeling for CFD. La Caada, California: DCW Industries, Inc.