Sunteți pe pagina 1din 8

On the Evolution of the Subgrid-Scale Energy

and Scalar Variance: Effect of the Reynolds


and Schmidt numbers

C. B. da Silva and J.C.F. Pereira

Instituto Superior Técnico, Pav. Máquinas I, 1o andar/LASEF, Av. Rovisco Pais,


1049-001 Lisboa, Portugal csilva,jose@navier.ist.utl.pt

A promising trend in subgrid-scale modeling consists in using transport equa-


tions for the subgrid-scale kinetic energy as e.g. in Ghosal et al. [1]. In the
present work Direct Numerical Simulations (DNS) of forced homogeneous
isotropic turbulence are used to analyze some of the assumptions often used
in these models, and their dependence on the Reynolds number and implicit
filter size. In particular, three key issues are analyzed: (a) the relative im-
portance between production and diffusion of SGS kinetic energy; (b) the
modeling of the viscous SGS dissipation and; (c) the acceleration statistics
of the local and convective terms. Finally, the same analysis is applied to an
equation for the evolution of the SGS scalar variance.

1 Introduction
In the context of Large-Eddy Simulations (LES), the hypothesis of (a) equi-
librium and (b) self-similarity, were often invoked in the past, and are still
often used today, in order to develop and improve new subgrid-scale models.
Indeed, most subgrid-scale models today use one of these assumptions, either
to derive mathematical expressions or to compute model constants.
However, it has been recognized that, particularly the equilibrium hypoth-
esis, does not work very well. As shown by da Silva and Métais [2], and da
Silva and Pereira [3], even for statistically stationary turbulence the large and
small scales of the velocity field are only weakly correlated. The same oc-
curs for any passive scalar field. This problem is even more important when
considering flows that are statistically unsteady at the large scales e.g. in
periodically forced jets or in unsteady boundary layers. For these cases, the
”past history” of the flow field has to be taken into consideration if one is to
compute accurately the local turbulence characteristics.
One interesting trend in Large-Eddy Simulations that overcomes these
difficulties by dropping out the equilibrium and self similarity assumptions, is
2 C. B. da Silva and J.C.F. Pereira

to consider simplified transport equations for the subgrid-scale stresses or the


subgrid-scale kinetic energy. Examples of this approach include the works of
Ghosal et al. [1], Wong [4], Debliquy et al. [5], Schiestel and Dejoan [6] and
Chaouat and Schiestel [7].
The estimation of the subgrid-scale kinetic energy is interesting in other
situations as for instance the evaluation of the Reynolds stresses in LES [8, 9].
Also, in numerical simulations of turbulent reacting flows, knowledge of the
SGS scalar variance might be useful. For these reasons it is important to
analyze the dynamics of the SGS kinetic energy and SGS scalar variance,
and hopefully to try to develop simple or approximated expressions for their
evolution.
The present work uses Direct Numerical Simulations (DNS) of statisti-
cally stationary (forced) homogeneous isotropic turbulence to analyze three
key issues related to the modeled or simplified expressions for the evolution
of the subgrid-scale kinetic energy: (a) the relative importance between pro-
duction and diffusion of SGS kinetic energy; (b) the modeling of the viscous
SGS dissipation and; (c) the acceleration statistics of the local and convective
acceleration terms. Finally, a similar analysis is applied to an equation for the
evolution of the SGS scalar variance.

2 Transport equations: exact and simplified forms


2.1 Evolution of the SGS kinetic energy

The classical filtering operation used to obtain the Filtered Navier-Stokes


equations, is defined by,

φ< (x) = φ(x )G∆ (x − x )dx , (1)

where φ is a given flow variable, and G∆ (x) is the filter kernel. When applied
to the Navier-Stokes equations the filtering operation yields the subgrid-stress
tensor, τij = (ui uj )< − u< <
i uj . The exact equation for the subgrid-scale kinetic
energy, τii , reads,
   <  <
∂t (τii ) + ∂j τii u< = ∂j (ui ui )< u<
j − (ui ui uj ) j − (puj )
+ 2∂j p< u<
       j
    
I II IIIa IIIb
  
+ ν∂j2 (τii ) − 2ν <
(∂j ui ∂j ui ) − ∂j u< <
i ∂j ui j − 2τii ∂j (ui ) .
+ 2∂j τii u< <
           
IV V VI V II
(2)

where ui and p are the velocity and pressure fields, respectively, and ν is the
molecular viscosity. In equation (2) the terms I and II account for the total
(local and convective) variation of resolved SGS kinetic energy, terms IIIa ,
On the evolution of subgrid-scales energy and scalar variance 3

IIIb , IV and V I represent diffusion terms, V is the viscous dissipation of SGS


kinetic energy and term V II is the SGS production (also called subgrid-scale
dissipation). The physical mechanisms, relative importance and topology of
each term in equation (2) was analyzed in great detail by da Silva and Métais
[2]. A simplified equation describing the evolution of the subgrid-scale kinetic
energy, τii , usually takes the following form [1, 4, 5],
 
∂t (τii ) + ∂j τii u<
j = −∂j Qj + P ∆ − ε∆ (3)

where the first term on the right hand side has to model the mechanisms de-
scribed by terms IIIa , IIIb , IV and V I in equation (2), P ∆ = −τij ∂u<
j /∂xj ,

and ε models term V .

2.2 Evolution of the SGS scalar variance

The exact equation for the evolution of the SGS scalar variance, qθ = (θ2 )< −
θ<2 , is given by,
 
 <  2 <
∂t (qθ ) + ∂j qθ u< = ∂j θ2 u<j − θ uj + γ∂j2 (qθ )
       j
    
Iθ IIθ IVθ
IIIθ
 <  <
− 2γ (Gj Gj ) − G< <
j Gj + 2∂j qj θ − 2qj G<
j (4)
        
Vθ V Iθ V IIθ

where qj = (θuj )< − θ< u< < <


j represents the SGS scalar flux, Gj = ∂θ /∂xj
is the filtered scalar gradient,and γ is the scalar diffusivity. Apart from the
absence of the pressure, each term in equation (4) has a similar role to the
term with the same number in equation (2). A model equation describing the
evolution of the SGS scalar variance can be defined as,
 
∂t (qθ ) + ∂j qθ u<
j = −∂j Rj + Pθ∆ − ε∆
θ (5)

where each term has a similar role to the corresponding term in equation (3).

3 Direct numerical simulations of isotropic turbulence

The numerical code used in the present simulations is a standard pseudo-


spectral code in which the temporal advancement is made with an explicit
3rd order Runge-Kutta scheme. The physical domain consists in a periodic
box of sides 2π and the simulations were fully dealised using the 3/2 rule.
Three DNS of statistically steady (forced) homogeneous isotropic turbu-
lence using N = 192 collocation points in each direction were carried out.
Table 1 lists the details of the simulations. Both the velocity and scalar large
4 C. B. da Silva and J.C.F. Pereira

Table 1. Details of the direct numerical simulations. Reλ = u λ/ν: Reynolds num-

ber based on the Taylor micro-scale and r.m.s. of the velocity fluctuations u ; ν:
molecular viscosity; kmax : maximum resolved wave number; Sc = ν/γ: Schmidt
number (γ is the scalar diffusivity); L11 : velocity integral scale; η: Kolmogorov micro-
scale; ηB : Batchelor micro-scale; S(φ) and F(φ) are the skewness and flatness of φ.

Reλ ν Sc kmax η kmax ηB L11 η(×10−2 ) ηB (×10−2 ) S( ∂u ) F( ∂u ) S[( ∂u )( ∂θ )2 ]


∂x ∂x ∂x ∂x

39.4 0.02 3.0 4.3 2.5 1.11 6.8 3.9 -0.46 +3.75 -0.55
95.6 0.006 0.7 1.8 2.1 1.24 2.8 3.3 -0.49 +4.63 -0.46
95.6 0.006 0.2 1.8 4.1 1.24 2.8 6.2 -0.49 +4.63 -0.50

scales were forced in order to sustain the turbulence using the method de-
scribed by Alvelius [10]. The forcing was imposed on 3 wave numbers concen-
trated on kp = 3.
After an initial transient that lasts about 10Tref where Tref = (Vc kp )−1 ,
Vc = (P/kp )1/3 , and P is the forcing intensity[10], the flow reaches a state
where all the turbulence quantities are statistically stationary. The analysis
was made using 10 instantaneous fields taken from this region, separated by
about 0.5Tref .
Notice that L > 4L11 in all simulations, where L is the box size and L11
is the integral scale, so that the size of the computational domain does not
affect the larger flow structures[11]. Also, to insure a good resolution of the
dissipative range we have kmax η > 1.5 and kmax ηB > 1.5 in all simulations,
where η = (ν 3 /ε)1/4 and ηB = η/Sc1/2 are the Kolmogorov and Batchelor
micro-scales, respectively.

Reλ=39.4
8
10 Reλ=95.6
(-5/3)
106 ∆/∆x=2
∆/∆x=4
4
∆/∆x=8
10
∆/∆x=16
E(k)/(εν5)1/4

2
10

100

-2
10
(filters:)
10-4
Reλ=95.6
Reλ=39.4
-6
10
0.5 1 1.5 2

Fig. 1. Kinetic energy spectra for all the simulations with the location of the several
filters used in this work.

Figure 1 shows the kinetic energy spectra for all the simulations, with the
location of the filters used in the subsequent analysis. That the dissipative
On the evolution of subgrid-scales energy and scalar variance 5

scales are indeed being well resolved is attested by the small upturns at the end
of the wave number range. For the case Reλ = 95.6 the velocity spectra has a
−5/3 range which shows also that, at least for that simulation, the existence
of an inertial range region. For each simulation the values of the skewness
and flatness of the velocity derivative oscillate around about −0.5 and 4.0
respectively and increase slowly with the Reynolds number. The present values
are quite close to the ones of Jimenez et al. [11].
In this study the separation between grid and subgrid-scales was made
using a box filter with filter widths equal to ∆m = m∆x, with m = 2, 4, 8, 16.
Notice that the implicit cut-off wave number for the filter with ∆/∆x = 16 is
within that region.

4 Results and discussion


4.1 Relative importance between GS/SGS diffusion and SGS
production

The relative importance between the GS/SGS diffusion (term V I from equa-
tion 2) and SGS production (term V II from equation 2) was studied by da
Silva and Métais [2]. It was observed that the local values of V I are one order
of magnitude greater than V II and this poses a challenge for subgrid-scale
models based on transport equations for τii . The same result is valid for the
analogous terms of the SGS scalar variance as shown in figures 2 (a) and (b).
The first figure shows PDFs of V Iθ and V IIθ where one sees that V Iθ is
a much more intermittent variable than V IIθ and also attains much higher
values (particularly for the backscatter part). Also, in figure 2 (b) we see that
RM S(V Iθ ) > RM S(V IIθ ), which means that there is much more local ”ac-
tivity” due to V Iθ than due to V IIθ , and the ”difference” increases with the
filter size. These facts have to be taken into account when modeling equation
(5).

4.2 Modeling of ε∆ and ε∆


θ

Arguably, the biggest challenge for modeling in equations (3) and (5) comes
from the viscous dissipation of SGS kinetic energy and SGS scalar variance,
represented by ε∆ and ε∆θ , respectively. Usually ε

is modeled using the fol-
lowing approximation, [1, 4, 5],
3/2
τii
ε∆ = C ∆ . (6)

For the SGS scalar variance a similar expression would be,
1/2
τii qθ2
ε∆ ∆
θ = Cθ . (7)

6 C. B. da Silva and J.C.F. Pereira

Rms(VIθ) Sc=0.2
0
10 2.5 Sc=0.7
VIθ Sc=3.0
VIIθ Rms(VIIθ) Sc=0.2
10
-1
2 Sc=0.7
Sc=3.0

Rms(VIθ,VIIθ)
PDF(VIθ,VIIθ)
10-2 1.5

10-3 1

10-4 0.5

-5
10 0
-15 -10 -5 0 5 10 15 2 4 6 8 10 12 14 16
VIθ,VIIθ
(a) ∆/∆x (b)

Fig. 2. (a) PDF of the GS/SGS diffusion (V Iθ ) and SGS production (V IIθ ); (b)
Rms of GS/SGS diffusion and SGS production for the SGS scalar variance;

Figure 3 (a) shows the constants C ∆ and Cθ∆ for several Reynolds and Schmidt
numbers, and 4 filter sizes obtained by putting the exact values for ε∆ and ε∆ θ
i.e. terms V and Vθ , into equations (6) and (7). The constants tend to decrease
with an increase in the Reynolds and Schmidt numbers, and with the filter
size. The fact that Cθ∆ seems to change more than C ∆ indicates that it is
more difficult to model ε∆ ∆
θ than ε . On the other hand it is encouraging to
see (figure 3 b) that the local values of both ε∆ and ε∆θ are very well correlated
with terms V and Vθ , respectively, particularly for large i.e. inertial range filter
sizes. For the SGS kinetic energy this result agrees with Meneveau and O’Neil
[12] and da Silva and Métais [2] who noticed that the correlation between
term V and ε∆ is quite good (above 0.6) for inertial range filters.

4 Reλ=39.4 Reλ=39.4
10 1
Reλ=95.6 Reλ=95.6
Sc=0.2 0.9 Sc=0.2
Corr (V,ε∆) , Corr(Vθ,ε∆θ)

103
Sc=0.7 Sc=0.7
0.8
Sc=3.0 Sc=3.0
102 ∆ 0.7
C (Schumann)
ε
C∆ε , C∆εθ


C (Schumann)
εθ 0.6
1
10
0.5

0
10 0.4

0.3
-1
10
0.2

10-2 0.1
2 4 6 8 10 12 14 16 2 4 6 8 10 12 14 16

∆/∆x (a) ∆/∆x (b)

Fig. 3. (a) Constants in equations (6) and (7); (b) Correlations between terms V
and Vθ , and the SGS kinetic energy τii and SGS scalar variance qθ .
On the evolution of subgrid-scales energy and scalar variance 7

4.3 Acceleration statistics for τii and qθ

Tsinober et al. [13], and Yeung and Sawford [14] have reported the existence
of partial cancellation of the local and advective variations for the velocity
and scalar fields, respectively. It is interesting to see if the same occurs for
the SGS kinetic energy and SGS scalar variance, as this is related to the
nonexistence of local equilibrium in equations (2) and (4), and thus to the
need for employing transport equations in order to get correct estimates for
both τii and qθ . Concerning τii we computed correlation coefficients for terms
I and II and observed that there is indeed a strong anti-correlation (−0.9) for
filter sizes at the dissipative region (∆/∆x = 2). However, for inertial range
filters (∆/∆x = 16) the (anti) correlation decreases to about −0.5, which
confirms the lack of equilibrium in equation (2). The figures 4 (a) and (b)
show JPDFs from terms I and II that illustrate that the degree of (anti)
correlation decreases with an increase in the filter size.

2 2

1 1
II’

II’
0 0

-1 -1

-2 -2
-2 -1 0 1 2 -2 -1 0 1 2
I’ I’
(a) (b)

Fig. 4. Correlation between local and advective variation of τii (terms I and II).
(a) ∆/∆x = 2; (b) ∆/∆x = 16.

5 Conclusions
Direct numerical simulations of (forced) isotropic turbulence were used to
analyze three problems related to the modeling of the transport equations
for the SGS kinetic energy and SGS scalar variance. The results showed that
both the GS/SGS diffusion for the velocity and scalar fields is more important
than the SGS production, respectively, and this ”difference” increases with the
filter size, Reynolds and Schmidt numbers.
Concerning the classical equations used to model the SGS viscous dissi-
pation i.e. equations (2) and (4), the results indicate that the constant C ∆ is
more sensitive than Cθ∆ to the filter size, the Reynolds and Schmidt numbers.
8 C. B. da Silva and J.C.F. Pereira

However, locally the approximations work equally well for the velocity and
scalar fields, as long as the filter size is not at the dissipative range region.
Finally, in a similar way as for the velocity and scalar field, we observed
the existence of a partial cancellation of the local and convective variation
terms for τii and qθ when the filter is at the dissipative scales, but the level
of (anti) correlation decreases with an increase in the filter size.

References
1. S. Ghosal, T. Lund, P. Moin, and K. Akselvol. A dynamic localisation model
for large-eddy simulation of turbulent flows. J. Fluid Mech., 286:229–255, 1995.
2. C. B. da Silva and O. Métais. On the influence of coherent structures upon
interscale interactions in turbulent plane jets. J. Fluid Mech., 473:103–145,
2002.
3. C. B. da Silva and J. C. F. Pereira. On the local equilibrium of the subgrid-
scales: The velocity and scalar fields. Phys. Fluids, 17:108103, 2005.
4. V. C. Wong. A proposed statistical-dynamic closure method for the linear or
nonlinear subgrid-scale stresses. Phys. Fluids, 4(5):1080–1082, 1992.
5. O. Debliquy, B. Knapen, and D. Carati. A dynamic subgrid-scale model based
on the turbulent kinetic energy. In B. J. Geurts, R. Friedrich, and O. Métais,
editors, Direct and large-eddy simulations IV, pages 89–96. Kluwer Academic
Publishers, 2001.
6. R. Schiestel and A. Dejoan. Towards a new partially integrated transport model
for coarse grid and unsteady turbulent flow simulations. Theor. Comput. Fluid.
Dyn., 18:443, 2005.
7. B. Chaouat and R. Schiestel. A new partially integrated transport model for
subgrid stresses and dissipation rate for turbulent developing flows. Phys. Fluids,
17:065106, 2005.
8. G. Winckelmans and H. Jeanmart. On the comparison of turbulence intensi-
ties from large-eddy simulation with those from experiment or direct numerical
simulation. Phys. Fluids, 14(5):1809–1811, 2002.
9. B. Knaepen, O. Debiliquy, and D. Carati. Subgrid-scale energy and pseudo-
pressure in large-eddy simulation. Phys. Fluids, 14(12):4235–4241, 2002.
10. K. Alvelius. Random forcing of three-dimensional homogeneous turbulence.
Phys. Fluids, 11(7):1880–1889, 1999.
11. J. Jiménez and A. Wray. On the characteristics of vortex filaments in isotropic
turbulence. J. Fluid Mech., 373:255–285, 1998.
12. C. Meneveau and J. O’Neil. Scaling laws of the dissipation rate of turbulent
subgrid-scale kinetic energy. Phys. Review E, 49(4):2866–2874, 1994.
13. A. Tsinober, P. Vedula, and P. Yeung. Random taylor hypotesis and the be-
haviour of local and convective accelerations in isotropic turbulence. Phys.
Fluids, 13(7):1974–1984, 2002.
14. P. Yeung and B. Sawford. Random-sweeping motion hypotesis for passive scalars
in isotropic turbulence. J. Fluid Mech., 459:129–138, 2002.

S-ar putea să vă placă și