Está en la página 1de 14

ME 543 Handout 6 March 2018

1 Elements of Kolmogorov’s Universal Equilibrium Theory


In the terminology of Kolmogorov, ‘universal’ implies that the theory applies to all flows for which
certain requirements are met. ‘Equilibrium’ implies that certain aspects of the flows considered are
in statistical equilibrium, i.e., they are statistically stationary.

1.1 Possibility of statistical equilibrium of the smaller-scale motions


Consider a turbulent flow with root-mean-square turbulent velocity u0 and integral length scale
`; the latter corresponds approximately to the wave number kp at the peak in the kinetic energy
spectrum, i.e., ` ∼ 2π/kp . The relevant time scale for the ‘large-eddies’ is then T` = `/u0 . Take
Tk = T (k) to be the time scale associated with wave number k. It is observed that the energy
spectrum E(k) has a maximum, kp , and then falls off, usually monotonically, as k → ∞. It
is intuitively plausible that Tk decreases as k increases. (This fact would have to be checked a
posteriori for any theory.) For example, given a wave number k, a length scale `0 associated with
it can be defined by
`0 ≡ |k|−1 = k −1 .
A convenient velocity scale at wave number k is, from dimensional arguments,

u0 = [kE(k)]1/2 .

Therefore a time scale associated with k is


1
Tk = `0 /u0 = .
[k 3 E(k)]1/2

Thus if E(k) decreases less rapidly than k −3 , then the local time scale would decrease as k increases.
Note that experimental observations, and the theory discussed below, indicate that (i) E(k) does
decay less rapidly than k −3 over a considerable range of k, and (ii) the time scale does decrease as
k increases. This implies that
Tk /T` → 0 as k → ∞ .
Note that, earlier in the course, it was found that the ratio of ` to the scale at which energy
dissipation occurs, η, the Kolmogorov scale (see below), behaves as

` 3/4 u0 `
= O(R` ) , where R` = .
η ν
Therefore, if R` is large enough, a dynamically relevant equilibrium range would be expected to
exist.
Hypothesis: Provided that the Reynolds number R` is large enough, there exists a region of wave
number space such that the associated characteristic time Tk is small compared to the time scale
of the large-scale flows T` , and therefore this region is in approximate statistical equilibrium.
Note that this hypothesis is assumed to hold for any turbulent flow, provided R` is large enough.
In addition, it does not say specifically how large R` should be.

1
Finally, in this equilibrium range the dynamic equation for E(k, t) reduces to, for statistically
∂E
isotropic flow, since = 0,
∂t
T (k) = 2νk 2 E(k) , or, (1)
Z ∞ Z ∞
T (k)dk = W (k) = 2ν k 2 E(k)dk , (2)
k k
where W (k) is the flux of energy from the wave number range [0, k) into the equilibrium wave
number range [k, ∞). This equation states that the flux of energy into the spectral region [k, ∞)
is just equal to the amount of energy dissipated there.
Note that this latter fact is related to the approximate balance of mean square vorticity (en-
strophy), as discussed in the handout ‘Vorticity Equations’, i.e.,
∂ωi ∂ωi
 
0 ' hωi ωj sif i − ν ,
∂xj ∂xj
the balance between the vortex stretching rate and the vortex dissipation rate. There it is pointed
out that (i) vortex stretching is a principal mechanism for the transfer of energy from large to small
scales, and (ii) the time scale for vortex stretching and vortex dissipation is much smaller than that
−1/2
for the mean flow, the ratio of the two behaving as R` for large R` .

1.2 Mutual independence of distant wave numbers


The main excitation of the turbulent field occurs at low wave numbers (near kp ), at length scales
characteristic of the generation mechanisms, e.g., the jet diameter. If the Reynolds number R` is
high enough, there will be a range of wave numbers which receive energy only from spectral energy
transfer. It is plausible that the direct influence of external conditions (excitations) is strongest for
small wave numbers, near kp , less strong for adjacent wave numbers, and disappears for very high
wave numbers. This should be accompanied by a loss of information as energy cascades to higher
wave numbers.
Hypothesis: Small-scale motions are statistically independent of the large-scale structure, provided
R` is large enough.
This hypothesis implies in particular that the small scale features do not directly depend on u0 and
`, but only indirectly.

1.3 Local isotropy

Pressure forces are known to have the effect of redistributing kinetic energy among the three velocity
components while conserving energy. As energy cascades from lower wave numbers, where it was
generated, to higher wave numbers, it is plausible then that the pressure forces tend to eliminate
the directional information that was in the larger-scale motions.
Hypothesis: Provided that the Reynolds number R` is large enough, the small-scale motions are
statistically isotropic.
This hypothesis has been verified, to some extent, experimentally. For example, for isotropic
turbulence, the following relationship should be satisfied:
2  2 
∂u1 1 ∂u2
 
= .
∂x1 2 ∂x1

2
These quantities depend on the small-scale motions, due to the fact that spatial derivatives are
taken, and the relationship (and other isotropic identities) has been approximately verified experi-
mentally for flows that are nonhomogeneous and nonisotropic at the large scales.
For the three hypotheses above, the value of R` to be exceeded for such a theory to hold must
be determined from experiments. Clearly what is required though is that kp  kue , where kue is
the lowest wave number in this equilibrium range.

1.4 Universal equilibrium range

From the previous hypotheses we have that, for R` sufficiently large, there is a range of equilibrium
wave numbers that are:

• in statistical equilibrium,

• locally isotropic,

• and statistically independent of the energy-containing range.

Figure 1: Cartoon depicting the equilibrium range.

The question is then, on what does this equilibrium range depend? It depends first of all on
the rate of insertion and removal of energy, which are the same, since
Z ∞ Z ∞ Z ∞
T (k)dk = W (k) = 2ν k 2 E(k)dk ' k 2 E(k)dt = ,
k k 0

the kinetic energy dissipation rate. (Note that this implies that the spectral peak in energy dissi-
pation occurs in this range, which is established below.) Secondly, it must depend on the viscosity
ν which determines the peak in the energy dissipation range.
Hypothesis: The statistical properties of motions associated with the equilibrium range of wave
numbers are uniquely determined by the parameters  and ν.
Note that this hypothesis is expected to be true for any flow as long as R` is large enough.
Furthermore, it is expected that there is no direct dependence of any statistical quantity in this
equilibrium range on u0 or `, but only dependence on  and ν.

3
This hypothesis, for example, implies that the spectral energy density E is given by E =
E(, ν, k). This is expected to be true for any flow, for large enough Reynolds number R` . A
length, time and velocity scale can be defined from  and ν, i.e.,

η = (ν 3 /)1/4 , the Kolmogorov length scale

τ = (ν/)1/2 , the Kolmogorov time scale


υ = (ν)1/4 , the Kolmogorov velocity scale
Note that the Reynolds number in this equilibrium range is then

Rη = υη/ν = (ν)1/4 (ν 3 /)1/4 /ν = 1 ,

implying that the inertial and viscous effects are in balance in this range. In wave number re-
gions with smaller k (corresponding to larger length scales), inertial effects become more dominant
compared to viscous effects. Therefore, viscous dissipation is expected to occur in this equilibrium
range.
From dimensional analysis, when all length scales are referred to η, and velocity scales to υ, the
motions associated with the equilibrium range are expected to have a universal form. A prediction
of this theory is then that
E(k) = υ 2 ηEe (kη) = 1/4 ν 5/4 Ee (kη)
where Ee is a dimensionless function of universal form. If the theory is right, then Ee (k) can be
determined once and for all from theory or experiment, as will be shown below.
Note that if all the dissipation occurs in the universal equilibrium range, then
2νυ 2 η
Z ∞ Z ∞
2
 = 2ν k E(k)dk = ζ 2 Ee (ζ)dζ
0 η3 0
Z ∞
1/2 3 −1/2 2
= 2ν(ν) (ν /) ζ Ee (ζ)dζ
0
Z ∞
= 2 ζ 2 Ee (ζ)dζ , so
0
Z ∞
1
ζ 2 Ee (ζ)dζ =
, a constraint on Ee (k).
0 2
The necessary condition for the existence of the equilibrium range is now

kη = η −1  kp ∝ 1/` , or

`(/ν 3 )1/4  1 .
With the rough approximation that  ≈ u03 /`, then
3/4
`(/ν 3 )1/4 = `(u03 /(`ν 3 ))1/4 = (u0 `/ν)3/4 = R` ,

so the condition becomes


3/4
R`  1 , as discussed above.
Note that, from dimensional reasoning, any dimensionless ratio which is independent of k and
depends only on the equilibrium range should be an absolute constant. An example of this is the
skewness of the velocity derivative, i.e.,
3  2 3/2
∂u1 ∂u1

S= .
∂x1 ∂x1

4
This has been found to be approximately constant, at a value of about -0.4, for all flows above a
certain Reynolds number.
Also note that there is a duality between the equilibrium wave-number range and velocity
differences, e.g., u(x + r) − u(x), where
1
 kp , i.e., `  |r| .
|r|

Analogous to the longitudinal velocity autocorrelation function f (r), defining a second-order lon-
gitudinal structure function D11 as

D11 (r) = h|u1 (x + re1 ) − u1 (x)|2 i ,

where e1 is the unit vector in the x1 direction, then in this case the theory predicts that

D11 (r) = υ 2 F (r/η) ,

where F is a universal function, related to Ee above.

1.5 The inertial subrange


When the Reynolds number R` is large enough, there may exist a significant portion of the equi-
librium range where viscous effects are unimportant; this range is called the inertial subrange. The
existence of this range would require a wave number k such that

kη = η −1  k  kp .

The lack of dissipation in this range implies that the flux of energy into this range equals the flux of
energy out, which is equal to . To see this, consider the energy flux W (k), defined in Equation (2),
into the range [kue , ∞), where kue is the lowest wave number in the universal equilibrium range.
From the definition of W (k),
∂W
= −T ,
∂k
where T (k, t) is the energy transfer spectrum. The equation for E(k, t) in the inertial subrange
reduces to
T (k, t) = 0 ,
so that W (k, t) satisfies, in the inertial subrange,

∂W
= 0,
∂k
that is, W (k, t) is constant in this range. To determine its value, integrate the equation for E(k, t),
Equation (1), across the entire equilibrium range, to obtain
Z ∞ Z ∞
T (k)dk = 2ν k 2 E(k)dk ,
kue kue
| {z } | {z }
W (kue ) 

so that W (kue ) = W |inertial range = . Therefore the statistical properties in the inertial subrange
should depend on the energy flux in this range, , but not on the viscosity ν.

5
Figure 2: Cartoon depicting the inertial subrange.

Hypothesis: The statistical properties of the motion in the inertial subrange depend only on the
parameter .
A prediction for the inertial subrange is that E = E(, k). The dimensions of E are (length)3 /(time)2 ,
and those of  are (length)2 /(time)3 , while those of k are (length)−1 . Therefore, dimensional anal-
ysis for this case gives
E(k) = α2/3 k −5/3 , (3)
where α is a universal constant, found from experiments to be α ≈ 1.4.
Although this prediction was made by Kolmogorov in 1941, it was not until 1962 (Grant,
Stewart, and Moilliet, “Turbulence spectra in a tidal channel”, J. Fluid Mech., 12:241, 1962)
that the theory was approximately verified. A plot of the spectra from the experiments of Grant,
Stewart, and Moilliet is given in Figure 3. Note that the plot is given in nondimensional form as
E(k)/1/4 ν 5/4 versus kη.
For the inertial subrange, the prediction for the second order structure function, which should
depend only on r and , is, from dimensional analysis,

D11 (r) = β2/3 r2/3 ,

where β is a universal constant, found to be approximately given by β ≈ 1.3α. Note that the
conditions for the existence of the inertial subrange in this case are

`  r  η,

where r = |r|. In fact, the prediction for the nth -order structure function is

h[u1 (x + re1 ) − u1 (x)]n i = Cn (r)n/3 .

Figure 4 is a plot of the prediction for the exponent of the nth -order structure function, ζn , as a
function of n. The straight line is the curve ζn = n/3, the prediction of the theory. The theory
is good up to a value of n of about 5, at which point it begins to significantly differ from the
experimental results. A modification of the theory to improve this by taking into account the
intermittency of the dissipation rate will be discussion in Section 2 below.
Kolmogorov considered his theory to be universal, i.e., to apply to any flow satisfying the
basic assumptions. Figure 5 is a plot of many sets of data for the energy spectrum, scaled with

6
Figure 3: Comparisons of the tidal channel data of Grant, Stewart, and Moilliet (and others) with
the original predictions of Kolmogorov.

Kolmogorov’s variables. (See Figure 6.14, page 235 of the text for a similar plot.) The scaling of
the data with Kolmogorov’s variables emphasizes the high wave-number range of the spectra. It
is seen that the data collapse fairly well for high wave numbers. Following the data from right to
left, it is seen that the data sets break off from the universal curve at different values of k1 η. The
lower Reynolds number data break off sooner (i.e., at larger values of k1 η), while the high Reynolds
number data exhibit a large inertial subrange and follow the universal curve to much lower wave
numbers.

1.6 The inertial-convective subrange


Consider the passive scalar θ(x, t) which satisfies the convection-diffusion equation, i.e.,

∂θ ∂θ ∂2θ
+ uj =D 2,
∂t ∂xj ∂xj

as discussed earlier in class. Arguments similar to the spectral transfer of kinetic energy can be
made about the spectral transfer of θ. For example, for the case of statistical homogeneity, assuming
the mean value of θ is zero for convenience, then the mean square of θ satisfies
2 
d1 2 ∂θ

hθ i = −D = −χ ,
dt 2 ∂xi
where χ is the dissipation rate of θ. An equilibrium range for θ is hypothesized for large enough
Reynolds number R` and Peclet number Pe = u0 `/D (or Schmidt number). In this case the various
dynamical quantities only depend on , ν, χ, and D, e.g., the energy spectrum of θ has the following

7
Figure 4: Measurements of the structure function exponential ζn compared to predictions of the
original Kolmogorov theory.

functional dependence:
Eθ = Eθ (, ν, χ, D, k) .
Furthermore, there is the possibility of a ‘convective’ subrange, where D is unimportant. If
the inertial and convective subranges overlap, then there is the possibility of an inertial-convective
subrange. In this case  and χ are the determining parameters. Dimensional analysis leads to

Eθ (k) = αθ χ−1/3 k −5/3 ,

which is called the Oboukov-Corrsin spectrum. The parameter αθ is assumed to be a universal


constant, similar to α in Equation (3) above. It has been determined to have a value of about
0.4. Several other spectral regions can exist, depending on the Peclet (or Schmidt) number. For
example, for very large Peclet number, ν may become important but D still is not. A theory has
been developed for this subrange, due to Batchelor.

1.6.1 The Pao Spectrum


An idea of a fairly simple model for the kinetic energy flux, W (k), while giving fairly good agreement
with laboratory data as shown in Figure 5, is given by the work of Pao, a student of Corrsin
(“Structure of turbulent velocity and scalar fields at large wave numbers”, Phys. Fluids, 8:1065-
1075). Following Kolmogorov’s arguments, the kinetic energy flux should only depend on E(k), 
and k, i.e.,
W (k) = f [E(k), , k] . (4)
Applying dimensional analysis gives:

W (k) h E(k) i
=g . (5)
 α2/3 k −5/3

Note that, in the inertial range, E(k) → α2/3 k −5/3 , W (k) → , so

g(1) = 1 . (6)

8
Figure 5: Energy spectra, scaled with Kolmogorov’s variables, from a variety of data.

The question is, what is the functional form for g. Pao assumed that g is a linear function of its
argument, i.e., that W (k) and E(k) are linearly related, or, with Equation 6,
W (k) E(k)
= 2/3 −5/3 , or
 α k
W (k) = α−1 E(k)1/3 k 5/3 . (7)
This can then be plugged into the equation for E(k) for isotropic turbulence, assuming stationarity,
to give,
d d n −1 1/3 5/3 o
T (k) = − W (k) = − α  k = 2νk 2 E(k) . (8)
dk dk
Integrating gives, after several steps,
 3 
E(k) = βk −5/3 exp − α(kη)4/3 , (9)
3
where β is a constant of integration. Note that, for kη  1, E(k) → βk −5/3 , so β = α2/3 , and
finally
 3 
E(k) = α2/3 k −5/3 exp − α(kη)4/3 . (10)
3
This spectrum form gives fairly good agreement with the laboratory data, as seen from Figure 5.
The model is obviously fairly simple. The accuracy probably indicates a weak sensitivity of the
results to the modeling assumptions (i.e., more ‘physical’ assumptions probably would not improve
the results much more). Also note that a similar approach gives a fairly accurate model for the
scalar spectrum.

9
2 Refinements to Kolmogorov’s Theory (K62)
Kolmogorov’s theory works reasonably well for both energy spectra and second-order structure
functions, assuming that the Reynolds number is large enough. However, there are some difficulties
with higher order quantities, and these difficulties give some insight into the limitations of the
theory.
Consider the n-th order structure function,

Dn (r) = h(∆r u)n i , with

∆r u = u1 (x + e1 r, t) − u1 (x, t) .
Assuming r is in the inertial range, i.e., `  r  η, where ` is the length scale of the energy-
containing range, and η is the Kolmogorov scale, then Dn only depends on  and r (specifically not
on ν), and dimensional analysis leads to

Dn (r) = Cn (r)n/3 ,

where Cn for n = 2, 3, . . . are ‘universal’ constants (n 6= 1). The prediction is good for n = 2 to 4,
but starts to break down strongly for n > 5.

Figure 6: Exponent ζp of structure functions of order p versus p. Inverted white triangles: data from
Van Atta and Park (1972); black circles, white squares and black triangles: data from Anselmet,
Gagne, Hopfinger and Antonia (1984) with Rλ = 515, 536, 852, respectively; + signs: S1 data
processed by ‘ESS’. Straight chain line: ζp = p/3 (K41); dashed line; β-model; solid line: lognormal
model with µ = 0.2; dotted line: log-Poisson model. From Frisch (1995).

Consider Dn (r) ∼ rζn , then ζn versus n is plotted in Figure 6 (see also Figure 6.31, page 257,
in the text). If the theory is right, then ζn = n/3. Clearly from the figure the theory breaks down
as n increases. The reason for this breakdown is thought to be the internal intermittency of the
small-scale turbulence, something not taken into account in the original theory (K41). Even in

10
flows that are continuously turbulent, the small-scale motions are found to be very intermittent.∗
For example, Figure 7 shows the velocity signal taken from a high Reynolds number jet; note the
continuous presence of the velocity signal. On the other hand Figure 8 is the high-pass filtered
signal for the same case, which emphasizes the higher frequency (higher wave number based upon
Taylor’s hypothesis (see pages 223 and 224 of the text)), showing intermittent bursts. Energy at
the small scales is very intermittent.

Figure 7: Velocity time signal from a jet with Rλ = 700 (Gagne, 1980). From Frisch (1995)

Figure 8: Same signal as in Figure 7, subject to high-pass filtering, showing intermittent bursts.
From Frisch (1995)

To deal with this intermittency, it is useful to distinguish between


• the instantaneous dissipation rate 0 (x, t) defined by

0 = 2νsij sij (usual definition; highly intermittent)

• the locally spherically-averaged dissipation rate, i.e., for a given radius r and volume V(r),
3
ZZZ
r (x, t) = 0 (x + r)dr , r = |r| (this is not intermittent if r is large enough)
4πr3 Vr

This is often called internal intermittency to distinguish it from the intermittency in localized turbulent flows,
such as jets or boundary layers, where the outer flow can vary between turbulent and non-turbulent.

11
Using r , which is a random function, Kolmogorov introduced a refinement to his original hypothesis
(K62).
Hypothesis: Assuming that the Reynolds number R` is large enough, and r  `, the statistics
of ∆r u conditioned on r are universal, and determined by r and ν. Therefore, for an inertial
subrange the following holds.
Hypothesis: Assuming that the Reynolds number R` is large enough, and `  r  η, the statistics
of ∆r u, conditioned on r , are universal, and determined by r . That is,

h(∆r u)n |r = ˆi = Cn (ˆ


r)n/3 ,

where ˆ is a value from the sample space of r . Note that in the original theory (K41), the dissipation
rate used was , the ensemble-averaged dissipation rate. Here the local dissipation rate r = ˆ is
used, so that there is still a need to average over r .

• Note from probability theory: Consider two correlated random variables (A, B) with the sam-
ple space (a, b). Then Z
hA|B = bi ≡ a pA (a|B = b)da .
a
But since pAB (a, b) = pA (a|B = b) pB (b), then
Z Z Z Z 
hAi = a pAB (a, b)dadb = a pA (a|B = b)da pB (b)db .
b a b
| a {z }
hA|B=bi

In the present case, the unconditional mean is obtained as


  Z
h(∆r u)n i = h(∆r u)n |r = ˆi = h(∆r u)n |r = ˆipr (ˆ
)dˆ


Z
= r)n/3 pr (ˆ
(ˆ  = Cn hn/3
)dˆ r ir
n/3
.

Generally to proceed further we need more information about the probability density function of
n/3
), or the mean hr i.
r , pr (ˆ
Note first that, for the special case n = 3,

h(∆r u)3 i = h(u1 (x + re1 ) − u1 (x, t))3 i = DLLL (r) = C3 hr ir , and
1 1
 ZZZ  ZZZ
hr i = 4 3
0 dV = 4 3 h0 i dV = ,
3 πr V(r) 3 πr V(r) |{z}


so DLLL (r) = C3 r. Therefore the modified theory (K62) gives the same result as the original
theory (K41) for n = 3.
Note that for the equilibrium range (see Equation 6.88 and Exercise 6.9 in the text), DLLL (r) =
−(4/5)r. So C3 = −4/5 for this case.
To go further, Kolmogorov (K62) conjectured that
 µ
h2r i `
' `rη (inertial subrange),
2 r

12
which has been verified to some extent experimentally with µ = 1/4. Therefore, for example, for
n = 6,
h(∆r u)6 i = C6 h2r ir2 ' C6 2 `µ r2−µ ,
with the exponent 2 − µ ' 1.75, consistent with data.
In addition, Kolmogorov assumed that r is log-normally distributed, i.e., that ln(r /) is nor-
mally distributed (see page 46 and the following in the text for a discussion of the log-normal
distribution). He based this assumption on a cascade model roughly like the cartoon in Fig-
ure 9. In this case as energy is cascaded to smaller and smaller scales, it becomes less and less
space filling, a property of a log-normal function. This assumption leads to Dn (r) ∼ rζn with
ζn = (n/3)[1 − (µ/6)(n − 3)]. This is the solid line shown in Figure 1 (the dashed line in Figure 6.31
of the text). Note that, for n = 2, ζ2 = 2/3 + µ/9 ' 2/3 + 1/36, not much different than 2/3
predicted by the original theory; it is very difficult to measure the difference. Similarly, for energy
spectra in the inertial subrange, E(k) ∝ k −p , with p = 5/3 + µ/9 ' 5/3 + 1/36, again not much
different than the original prediction of 5/3. For n = 3, then ζ3 = 1, giving the same result as the
original theory, as mentioned above. But significant differences occur for n > 3.

Figure 9: A cartoon of the energy cascade. Note that with each step the eddies become less and
less space-filling. From Frisch (1995)

References

Anselmet, F., Gagne, Yl, Hopfinger, E.J., & Antonia, R.A. 1984. High-order velocity structure
functions in turbulent shear flow, J. Fluid Mech. 140, 63-89.
Frisch, U. 1995. Turbulence, Cambridge University of Press.
Gagne, Y. 1980. Contribution à l’étude expérimentale de l’intermittence de la turbulence à petite
échelle, Thèse de Docteur ès-Sciences Physiques, Université de Grenoble.
(K41) Kolmogorov, A.N. 1941. The local structure of turbulence in incompressible viscous fluid for
very large Reynolds number, Dok.. Akad. Nauk SSR 30, 9-13 (reprinted in Proc. R. Soc. Lond. A
434, 15-17 (1991)).
(K62) Kolmogorov, A.N. 1962. A refinement of previous hypotheses concerning the local structure
of turbulence in a viscous incompressible fluid at high Reynolds number, J. Fluid Mech. 13, 82-85.

13
Van Atta, C.W., and Park, J. 1972. Statistical self-similarity and inertial subrange turbulence, in
Statistical Models and Turbulence, Lect. Notes in Physics. 12, 402-426, eds. M. Rosenblatt & C.W.
Van Atta, Springer, Berlin.

14

También podría gustarte