HAL Id: hal-02086501
https://hal.archives-ouvertes.fr/hal-02086501
Submitted on 1 Apr 2019
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Lie-symmetry group and modelling in non-isothermal fluid mechanics
Aziz Hamdouni, Dina Razafindralandy, Nazir Al Sayed
To cite this version:
Aziz Hamdouni, Dina Razafindralandy, Nazir Al Sayed. Lie-symmetry group and modelling in non- isothermal fluid mechanics. 6th AIAA Theoretical Fluid Mechanics Conference, 2011, Honolulu, United States. �hal-02086501�
Fluid Mehanis
AzizHamdouni
⋆
,DinaRazandralandy
⋆⋆
,NazirAlSayed
⋆⋆⋆
LEPTIAB,UniversitédeLaRohelle,AvenueMihelCrépeau,17042LaRohelleCedex1,Frane
E-Mail:
⋆
ahamdoununiv-lr.fr,
⋆⋆
drazanuniv-lr.fr,
⋆⋆⋆
nalsay01univ-lr.fr
1 Introdution
Thephysialpropertiesofaphenomenon whihhas beenmodelizedintoaset ofequations
are ontained in the symmetry group of these equations. This fat is well illustrated by
N÷ther's theorem [5, 7℄ whih states that for a Lagrangian system, eah symmetry of
the equations orresponds to a onservation law. For example, a system having a time-
translation symmetry isenergy onserving.
In uid mehanis, the symmetry group theory has been used tostudy, among others,
the dynamis of vorties in a uid ow [3, 4℄ and to investigate spetral properties of the
ow [9℄. Oberlak [6℄ also used the symmetry theory to dedue saling laws, suh as the
algebrai law whih has been onrmed in the enter and in the near-wall region of a
hannel ow, the usual logarithmi wall law, a linear law whih is observed in the enter
region of a ouette ow. He also found a new exponential law whih has been onrmed
later inthe mid-wakeregion of the boundary layer.
TherstpartofthispaperextendstheworkofOberlaktonon-isothermalandparallel
ow, suh as a pipe ow. Saling laws for both veloity and temperature will be given.
The seond part is devoted to the onstrution of a lass of turbulene models of a non-
isothermal ow, whih respet the physisof the ow.
Indeed,inspiteofmanydeadesofstudies,turbulenemodellingstillonstitutesanim-
portanthallengeinengineering siene. No turbulenemodel, eitherwith RANS method
or with large-eddy simulation (LES) approah, is really satisfying regarding their math-
ematial and physial properties. Moreover, the subgrid heat ux model is, most of the
time, simply dedued from the subgrid stress tensor model by a Reynolds analogy. This
analogy limitsthe sope of the models for strongly oupledproblems (natural and mixed
onvetion). Inthis paper, wepropose agenerallassof subgridstress andheat uxmod-
els based on the preservation of the symmetry group (and then the underlying physial
properties)of the non-isothermal NavierStokesequations (1).
2 Saling laws in a parallel ow
Consider the NavierStokes equations for an inompressible, non-isothermal Newtonian
uid, with kinemativisosity ν and thermaldiusivityκ:
∂u
∂t + div(u⊗u) + 1
ρ∇p− div τr−βgθe2= 0
∂θ
∂t + div(θu)− div hr= 0 div u= 0
(1)
In these expressions, τr = 2νSand hr=κ∇θ are respetively the visous stress tensor and the heat ux, and S = ∇u+2T∇u is the strain rate tensor. β is the thermal expansion
and g the aeleration due togravity.
UsingtheRANSapproah,wedeomposetheveloity,thepressureandthetemperature
into amean and a utuating quantities:
f =F +f′. (2)
As we are seeking for stationnary bidimensionalsaling laws in aparallelow, we assume
that the mean pressure is of the form P(x2)−x1K, that U2 = U3 = 0 and that U1 and
Θ depend only on the wall normal oordinate x2. The utuating quantities verify the
followingset ofequations:
∂u′i
∂t +U1
∂u′i
∂x1
+u′2
dU1
dx2
δi1− K+ν∂2U1
∂x22
!
δi1+ ∂P
∂x2
−βgΘ−βgθ′
! δi2
+∂u′iu′k
∂xk + ∂p′
∂xi−ν∂2u′i
∂x2k = 0
∂θ′
∂t +∂θ′u′j
∂xj −κ∂2θ′
∂x2j −κ∂2Θ
∂x22 +u′2∂Θ
∂x2
+U1
∂θ′
∂x1
= 0
∂u′k
∂xk = 0
(3)
Forsimpliity,we alsodesignate system (3)by
E(t,x,U,u′, P, p′,Θ, θ′, ν, κ) = 0 (4)
Saling laws are self-similar solutions of (3). In what follows, we use Lie's theory to nd
the symmetries of the equationsand to dedue the self-similarsolutions.
A (Lie point-)symmetryof (3) isa transformation
(t,x,U,u′, P, p′,Θ, θ′, ν, κ)7→(bt,x,b Ub,ub′,P ,b pb′,Θ,b θb′,bν,bκ) (5)
dependingontinuously ona parametera,whihmaps asolution of (3)intoanothersolu-
tion. Knowingtransformation (5)is equivalentto knowing its generator
X =ξt∂
∂t+ξxi
∂
∂xi
+ξU1
∂
∂U1
+ξP ∂
∂P +ξΘ
∂
∂Θ+ξu′i ∂
∂u′i+ξp′ ∂
∂p′ +ξθ′ ∂
∂θ′ (6)
where
ξq= ∂qb
∂a a=0
. (7)
Aording toLie's theory, transformation (5)is asymmetry of (4) if and onlyif
E = 0 ⇒ X.E = 0. (8)
This last ondition permits toompute the omponents of the symmetry generators X of
(3), whihare
ξν = nν, ξκ = mκ,
ξt = [n−2m]t+c1,
ξx1 = [n−m]x1+α1(t) +c2, ξx2 = [n−m]x2+c3,
ξx3 = [n−m]x3+α2(t) +g,
ξU1 = mU1−α3(t) +α4(x2) + ˙α1(t),
ξP = 2mP +βgx2α5(t, x1, x3) +α6(t, x1, x3) +α9(t, x2), ξΘ = [3m−n] Θ−α7(x2) +α5(t, x1, x3),
ξu′1 = mu′1+α3(t)−α4(x2), ξu′2 = mu′2,
ξu′3 = mu′3+ ˙α2(t),
ξp′ = 2mp′ + [K(n−3m)−α˙3(t)]x1 −α¨2(t)x3+α8(t)−α9(t, x2), ξθ′ = [3m−n]θ′ +α7(x2).
(9)
where n, m, the c′is and the α′is, whih are arbitrary onstants and funtions, are the
parameters of the symmetries.
(U1,Θ)represents a self-similarsolution of (3)if
X.U1 = 0 and X.Θ = 0,
that is,if
dU1
ξU1
= dΘ ξΘ
= dx2
ξx2
(10)
Inthe partiularasewherethe αi'sare onstant,theresolution of(10)leads topartiular
saling laws.
• Ifm 6= 0, n6=m, n 6= 3m, we get the following algebraisalinglaw:
U1 =C1
x2+ b n−m
nm
−m
+C3
m, Θ =C2
x2+ b n−m
3mn −n
−m
+ C4
3m−n. (11)
• A logarithmi veloity prolearises if m = 0 and n6= 0: U1 = C1
n ln
x2+ b n
+C3, Θ =C2
x2+ b
n −1
+ C4
n . (12)
• A linear lawappears when m =n = 0:
U1 =C1x2+C3, Θ =C2x2 +C4. (13)
• We obtain anexponentiallawwhen m=n6= 0: U1 =C1expmx2
b
+ C3
m, Θ = C2exp
2mx2
b
+ C4
2m. (14)
As notied in the introdution, the orresponding saling laws in the isothermal ase
haveall been onrmed either by experimentsor by DNS data.
As weould see, the symmetry groupare ofparamountimportaneinmodellingphys-
ial phenomena. It ontains, among other, the saling laws of uid ows. In partiular,
turbulene models should alsopreserve the symmetry group of the equations. In the next
setion,webuildalass ofLESturbulenemodels,whihareompatible withthesymme-
try group of the non-isothermalNavierStokes equations. These models should naturally
reproduethe physialproperties of the ow, suh assaling laws.
Letusonsider equations(1). Usingthe sameapproahasabove,the symmetrygenerator
of these equations an be omputed. Its omponents are
ξt = c1+ 2tc3,
ξx1 = c2x2+ [c3 +c5]x1+α1(t), ξx2 = −c2x1+ [c3 +c5]x2+α2(t), ξx3 = [c3 +c5]x3 +α3(t), ξu1 = c2u2+ [c5−c3]u1+ ˙α1(t), ξu2 = −c2u1+ [c5−c3]u2+ ˙α2(t), ξu3 = [c5 −c3]u1+ ˙α3(t),
ξp = ζ(t) + 2[c5−c3]p−ρ(xα(t)) +¨ c4βgx3, ξθ = [c5−3c3]θ+c4/ρ,
ξν = 2c5ν, ξκ = 2c5κ.
(15)
With the generator, one get the following list of symmetries of the non-isothermal
NavierStokes equations(1):
• the group of time translations: (t,x,u, p, θ) 7−→ (t+a,x,u, p, θ)
• the group of pressure translations: (t,x,u, p, θ) 7−→ (t,x,u, p+ζ(t), θ)
• the group of pressure-temperature translations:
(t,x,u, p, θ) 7−→ (t,x,u, p+a βg x3, θ+a/ρ)
• the group of horizontal rotations: (t,x,u, p, θ) 7−→ (t,Rx,Ru, p, θ)
where R isa 2D (onstant) rotationmatrix,
• the group of generalized Galilean transformations:
(t,x,u, p, θ) 7−→ (t,x+α(t),u+α(t), p˙ +ρxα(t), θ)¨
• the group of the rst sale transformations:
(t,x,u, p, θ) 7−→ (e2at,eax,e−au,e−2ap,e−3aθ)
• and the group of the seond sale transformations:
(t,x,u, p, θ, ν, κ) 7−→ (t,eax,eau,e2ap,eaθ,e2aν,e2aκ).
Equations(1)ownotherknownsymmetries. Thesesymmetriesarenotloalontinuous
transformationslikethepreviousones andannotbededuedfromLie'stheory. Theyare:
• the reetions whih are disretesymmetries:: (t,x,u, p, θ)7→(t,Λx,Λu, p, ι3θ)
where Λis areetion matrix, Λ=diag(ι1, ι2, ι3) with ιi =±1,
• and the material indierene in the limit of a 2D horizontal ow [2℄ whih is a
time-dependent rotation: (t,x,u, p)7→(t,x,b u,b p)b
with
b
x=R(t) x, ub=R(t) u+ ˙R(t) x, pb=p−3ωφ+ω2kxk2/2
where R(t) is an horizontal 2D rotation matrix with angle ωt, ω a real parameter, φ the usual 2D stream funtion suh that (u = curl(φe3)) and k•k indiates the
Eulidian norm. The material indierene onstitutes a non-loal symmetry of the
equations.
The ombinationof all these symmetriesonstitutes the symmetry group of (1).
In LES approah, one does not solve diretly equations (1). Instead, one omputes an
approximate solution (u, p, θ) whih ontains only the large sales of the atual veloity u, pressure p and temperature θ of the uid. The separation of large and small sales is
dened by a ltering. The equationsof (u, p, θ)are:
∂u
∂t + div(u⊗u) + 1
ρ∇p− div(τr−τ)−βgθe2= 0
∂θ
∂t + div(θu)− div(hr−h) = 0 divu= 0.
(16)
In these equations, τ =u⊗u−u⊗u is the subgrid stress tensor and h=θu−θu the
subgridheat ux. Inorder tolose the equations, thesetwoterms must bemodelled,i.e.,
replaed by funtions of (u, p, θ), alled turbulene model. A good turbulene model is
one with whih (u, p, θ) has the same properties as (u, p, θ) from ertain point of view.
In our approah, we require that (u, p, θ) has the same symmetry properties as (u, p, θ).
Thisrequirementisimportantbeause, asunderlinedearlier,thesymmetrygroupontains
importantphysialinformationon the ow.
More preisely, themodelshouldbesuh thateahsymmetryof(1)appliedto(u, p, θ)
is alsoasymmetry of (16) appliedto (u, p, θ). When itis the ase, the modelwillbesaid
invariant. Unfortunately, as analyzed in ([8℄), the major part of existing LES turbulene
models do not omply with this requirement. In what follows, we propose new models
based onthe preservation of the symmetry groupof the equations.
Time, pressure, and pressure-temperature translations, and the galilean transforma-
tions remain symmetriesof (16) if τ and h depend onlyon Sand T=∇θ:
−τ =−τ(S,T), −h=−h(S,T) (17)
Next,rotations,reetionsandthematerialindierenearesymmetriesof(16)ifthemodel
has the following form
−τd=E1S +E2( AdjS)d +E3(T⊗T)d +E4
S(T⊗T)d
+E5
S(T⊗T)Sd
−h=E6T +E7S T +E8S2T
(18)
where the oeients Ei are salar funtionsof:
X = trS2, ξ = detS, ϑ=T2, ω1 =TS T, ω2 =S TS T (19)
and Adj is the adjoint operator. Next, equations (16) are invariant under the rst sale
transformations if
τb= e−2aτ and Tb = e−4aT. (20)
Ei(X, ξ, ϑ, ω1, ω2) =XsiEi′(v1, v2, v3, v4) (21)
where the vi'sare the invariants:
v1 = ξ
X3/2, v2 = ϑ
X2, v3 = ω1
X5/2, v4 = ω2
X3. (22)
and
s1 = 0, s2 =−12, s3 =−32, s4 =−2, s5 =−52, s6 = 0, s7 =−12, s8 =−1
Finally,the seondsale transformation are symmetries of equations(16) if
Ei′ =νFi, i= 1, ...,5 and Ei′ =κFi, i= 6, ...,8.
To sum up, we get the following lass of subgrid models whih are onsistent with the
symmetry group of (1):
−τd=νF1S+νX−1/2F2AdjdS+νX−3/2F3(T⊗T)d +νX−2F4[S(T⊗T)]d] +νX−5/2F5S[(T⊗T)S]d]
−h=κ
F6+X−1/2F7S+X−1F8S2 T
(23)
In order toredue the degree offreedom of the model,wepropose torestrainthe lass
(23) to models whih derive from a potential. This restrition is legitimated by the fat
that, like τr andhr, τ and h represent respetivelya (subgrid)stress and a(subgrid) heat ux. Moreover,τr andhr arederived fromthepotentialsν trS2 andκ||T||2/2,inthesense
that
τr= ∂ν trS2
∂S
and hr = ∂
∂T 1 2κ||T||2
!
(24)
Hene, τ and h should also be derived from potentials. This ondition leads to the followinglass of models (see [8℄):
−τd=ν
"
2gm−3v1
∂gm
∂v1
−4v2
∂gm
∂v2
−5v3
∂gm
∂v3
−6v4
∂gm
∂v4
# S
+ν
"
X−1/2∂gm
∂v1
AdjdS+X−3/2∂gm
∂v3
(T⊗T)d+ 2X−2∂gm
∂v4
[S(T⊗T)]d
#
−h=κ ∂gt
∂v2
I3+X−1/2∂gt
∂v3
S+X−1∂gt
∂v4
S2
! T
(25)
where gm and gt are arbitrary funtions of the vi's. Note that the assumption that the
modelderives froma onvex potential ensures the stability of the model[8℄.
Further simpliationsan be done on lass (25) aording to the type of modelthat
we wish toobtain. Forexample,
• if gm and gt are only funtionsof v1 and v2,we get a strongly oupledmodel:
−τd=ν 2gm−3v1
∂gm
∂v1
−4v2
∂gm
∂v2
!
S+ν 1
||S||
∂gm
∂v1
AdjdS, −h=κhtT (26)
whereht= ∂gt
∂v2
.
• Adeoupled modelan be obtainedif gm isfuntiononlyof v1 andgtfuntion ofv2:
−τd=ν(2gm−3vg˙m)S+ν 1
||S|| g˙m AdjdS, −h=κht T. (27)
• Finally,we get a linear modelwhen gm and ht are linearfuntions of v:
−τd=νCm −detS 1 S
3S+ AdjdS 1 S
!
, −h=κCtdetS S
3 T (28)
where Cm are Ct are the onstants of the model, whih may depend on the grid
size. These onstants may also be evaluated dynamially, using the Germano-Lilly
proedure.
The simpliest model (28) (alled invariant model) and its dynamial version (alled
dynami invariant model) have been implemented for the simulation of an air ow in a
ventilated 3D room (see Figure 1). The oor is heated at 35◦C and the other walls are
maintainedat15◦C. The ode used forthe simulationis basedona nitevolumesheme,
with Crank-Niholson sheme in time [1℄. The grid size is (86×86×12).
Figure 1: Ventilated andheated room.
Themeanveloityandtemperatureprolesalongthevertialline(x= 0.502, z = 0.35)
are shown on Figure2. A goodagreement with experimentaldata [10℄, and better results
ompared to Smagorinsky and its dynamial version an be observed. These results are
partiularlygoodnear the walls(no wallmodelwere used). This may beexplainedby the
preservation of saling laws in the onstrution of the invariantmodels.
4 Conlusions
Weshowed that the symmetrytheory providesapowerfullmodellingtoolinuid mehan-
is, as well for theoretial as for numerial developpements. In partiular, we saw that
it ould be used to establish saling laws, whih are essential for the understanding of
turbulene. The symmetry approahhas alsobeenused forthe onstrutionof turbulene
models whih naturallyhave goodphysialproperties, sine they preserve the symmetries
of the equations. The numerialtest onrms this goodbehaviour.
Figure 2: Mean streamwise veloity(left)and meantemperature (right)at x =0.502m.
Referenes
[1℄ F.Arhambeau,N.Mehitoua,andM.Sakiz.Codesaturne: anitevolumeodeforthe
omputationof turbulent inompressibleows -industrialappliations. International
Journal On Finite Volumes, (1), 2004.
[2℄ B.J. Cantwell. Similaritytransformations for the two-dimensional, unsteady, stream-
funtion equation. Journal of FluidMehanis, 85:257271,1978.
[3℄ V. Grassi, R.A. Leo, G. Soliani, and P. Tempesta. Vortiies and invariant surfaes
generated by symmetries for the 3D Navier-Stokes equation. Physia A, 286:79108,
2000.
[4℄ V. Grassi,R.A. Leo, G.Soliani,and P. Tempesta. Temperaturebehaviourof vorties
of a 3D thermoonduting visous uid. Physia A: Statistial Mehanis and its
Appliations, 305(3-4):371 380, 2002.
[5℄ E. N÷ther. Invariante Variationsprobleme. In Königlihe Gesellshaft der Wis-
senshaften,pages 235257,1918.
[6℄ M. Oberlak. A unied approah for symmetries in plane parallel turbulent shear
ows. Journal of FluidMehanis, 427:299328,2001.
[7℄ P. Olver. Appliations of Lie groups to dierential equations. Graduate texts in
mathematis.Springer-Verlag, 1986.
[8℄ D. Razandralandy, A. Hamdouni, and M. Oberlak. Analysis and development of
subgrid turbulene models preserving the symmetry properties of the NavierStokes
equations. European Journal of Mehanis/B,26, 2007.
[9℄ G.Ünal.Appliationofequivalenetransformationstoinertialsubrangeofturbulene.
Lie Group and Their Appliations, 1(1):232240,1994.
[10℄ W. Zhang and Q.Chen. Large eddy simulationof indoor air ow with a ltered dy-
namisubgridsalemodel.InternationalJournal ofHeatandMassTransfer,43:3219
3231, 2000.