J. Fluid Mech. (2006), vol. 550, pp. 175–184.

c 2006 Cambridge University Press 175
doi:10.1017/S0022112005008438 Printed in the United Kingdom

Approach to the 4/5 law in homogeneous

isotropic turbulence
By R. A. A N T O N I A AND P. B U R A T T I N I
Discipline of Mechanical Engineering, University of Newcastle, NSW 2308, Australia

(Received 29 August 2005 and in revised form 6 December 2005)

Kolmogorov’s similarity hypotheses and his 4/5 law are valid at very large Reynolds
numbers. For flows encountered in the laboratory, the effect of a finite Reynolds
number and of the non-stationarity or inhomogeneity associated with the large scales
can affect the behaviour of the scales in the inertial range significantly. This paper
focuses on the source of inhomogeneity in two types of flows, those dominated mainly
by a decay of energy in the streamwise direction and those which are forced, through a
continuous injection of energy at large scales. Results based on a parameterization of
the second-order velocity structure function indicate that the normalized third-order
structure function approaches 4/5 much more rapidly for forced than for decaying
turbulence. This trend is supported by grid turbulence measurements and numerical
data in a periodic box.

1. Introduction
The similarity hypotheses proposed by Kolmogorov (1941b) (or K41) and their
subsequent revision (Kolmogorov 1962, or K62) to account for intermittency have had
a major impact on turbulence research. A fundamental element of these hypotheses
is the assumption that the small-scale motion, which includes dissipative and inertial-
range scales, is isotropic. Also, K41 and K62 require that the Reynolds number is very
large so that the small-scale motion is independent of the invariably anisotropic large-
scale motion. The major outcome of the first similarity hypothesis is the prediction
(δu∗ )n  = fun (r ∗ ), (1.1)
where the increment δu ≡ u(x + r) − u(x) (u is the velocity fluctuation along x; the
separation r is aligned with x, and the angular brackets denote averaging). For each
value of n, fun is a universal function, in the sense that it is expected to depend only
on r ∗ ≡ r/η (η ≡ (ν 3 /)1/4 is the Kolmogorov length scale, ν is the kinematic viscosity
and  is the mean energy dissipation rate). The asterisk denotes normalization by the
Kolmogorov velocity scale uK (≡ ν 1/4 1/4 ) and/or η. The second similarity hypothesis
yields the famous inertial-range (η  r  L; L is the integral length scale) result
(δu∗ )n  = Cun r ∗n/3 , (1.2)
when K41 is used (Cun is a universal constant).
An important exact relation between Buu ≡ (δu)2  and Buuu ≡ − (δu)3  was obtained
by Kolmogorov (1941a), starting with the Kármán–Howarth (von Kármán & Howarth
1938) equation for homogeneous isotropic turbulence,
4 ∂
Buuu = r − 6ν Buu . (1.3)
5 ∂r
176 R. A. Antonia and P. Burattini
In the inertial range, the viscous term can be neglected and (1.3) reduces to the 4/5
law,
Buuu = 45 r (1.4)
∗ ∗
or Buuu = 4r /5.
It is important to underline that (1.3) and (1.4), as well as the hypotheses in K41
and K62, apply only at very large Reynolds numbers. It is not surprising, therefore,
that for flows normally encountered in the laboratory, (1.1) appears to be satisfied
only in the dissipative range (typically r ∗ < 10) (see § 7 of Chassaing et al. 2002),
although the evidence is not altogether convincing especially when the isotropic form
of ,
iso = 15ν(∂u/∂x)2 , (1.5)
∗ ∗
is used in forming δu and r . With a few exceptions, the laboratory data also indicate
an asymptotic approach to (1.2) and (1.4) as the Reynolds number (usually represented
by Rλ and defined by u2 1/2 λ/ν, where λ is the Taylor microscale u2 1/2 /(∂u/∂x)2 )
increases. When Rλ is finite, deviations from (1.2)–(1.4) and indeed (1.5) can be quite
significant. For a fixed Rλ , the deviations may also depend on the nature of the
flow, thus casting doubt on any claim of universality, at least for scales extending
beyond the dissipative range. For the same flow and Rλ , departures from (1.2) and
(1.5) can still depend on the initial conditions that are used. It seems reasonable to
ascribe these deviations to a lack of homogeneity in laboratory flows, the source of
inhomogeneity depending on a number of parameters, such as the Reynolds number,
type of flow and initial conditions. For the relatively large Reynolds numbers which
occur in the atmospheric surface layer, the evidence in support of the ‘4/5’ law is
rather inconclusive. This is partly due to the uncertainty in estimating . The data
of Sreenivasan & Dhruva (1998) for Rλ  104 indicated, however, that there is no
discernible range over which ∂Buuu /∂r is constant over a convincing range. This is
cause for concern since the existence of the inertial range has been traditionally
linked to the linear increase of Buuu with r. These authors further noted that an
inertial range could not be identified unambiguously from the local slopes of the
even-order moments of δu. They also stressed that the scaling of (δu)n  cannot be
assessed effectively without first understanding the effects of finite shear and finite
Rλ . These observations fuel speculation about the validity of the scaling exponents
that have been inferred from laboratory data and also the corresponding inferences
regarding the departures of these exponents from the K41 or K62 predictions.
In deriving (1.3), Kolmogorov ignored the non-stationarity term in the Kármán–
Howarth equation. If ∂/∂t is of order /u2 , then it can be readily shown that
∂(δu)2 /∂t is negligible, provided (δu∗ )2 /Rλ  1 or (δu)2 /u2   1 (Saffman 1968).
The first of these requirements is satisfied when Rλ is very large. The second can be
satisfied even at relatively small Rλ , provided r is small compared to L. This latter
requirement is of interest since it provides some insight into why Kolmogorov’s first
similarity hypothesis continues to be relevant at least when r lies in the dissipative
range. It also reinforces the possibility that this range may have some claim to
universality. The non-stationarity has since been considered by a number of authors
(e.g. Danaila et al. 1999; Lindborg 1999; Lundgren 2003) in the context of decaying
grid turbulence (which, strictly, is only locally homogeneous because the spatial decay
precludes global homogeneity). There have also been attempts at identifying this non-
stationarity in more complicated flows, e.g. the centreline region of a fully developed
channel flow (Danaila et al. 2001), a homogeneous uniform shear flow (Casciola et al.
2003; Danaila, Antonia & Burattini 2004; Qian 1999, 2002) and the region near the
 Approach to the 4/5 law in turbulence 177
axis of a circular jet (Danaila et al. 2004). Stationary isotropic turbulence is often
studied numerically by adding a forcing term to the Navier–Stokes equation for low
wavenumbers (e.g. Gotoh, Fukayama & Nakano 2002). Forcing has also been used
in experiments (e.g. the flow in a cylindrical container between counter-rotating end
disks, Moisy, Tabeling & Willaime 1999) in order to achieve stationarity.
In this paper, the focus is mainly on the importance of the non-stationarity (or
inhomogeneity) which is neglected in (1.3). Specifically, we distinguish between flows
characterized by a decay of energy and those which are forced and examine in some
detail how the 4/5 law may be reached. In grid turbulence, the turbulent energy at each
scale decays in the streamwise direction. Temporally decaying as well as stationary
turbulence have been numerically simulated within a periodic box. Although Qian
(1999) and Lundgren (2003) have considered analytically the approach to the 4/5
law for both decaying and forced turbulence, they did not compare their predictions
with experimental or numerical results. A comparison is provided in this paper using
available data as well as new data in grid turbulence; for the latter data, the non-
stationarity is determined as accurately as possible, whilst measurements at relatively
small Rλ of Buuu /r are compared with analytical results based on a more realistic
model of Buu than considered by Lundgren (2003).

2. Experimental and numerical data – comparison with model

When decay or forcing is taken into account, the transport equation for Buu is
4 ∂Buu
Buuu = r − 6ν − Iu (2.1)
5 ∂r
where Iu reflects the influence of the non-stationarity. Like the Kármán–Howarth
equation, (2.1) expresses the budget of the energy for any particular scale. In spatially
decaying grid turbulence, Iu can be written as (Danaila et al. 1999)


U r 4 ∂Buu
Iu = −3 4 s ds, (2.2)
r 0 ∂x
where U is the mean velocity in the streamwise direction x and s is a dummy
integration variable. Since Iu is positive, its appearance in (2.1) maintains the
magnitude of Buuu below its asymptotic value of 4r/5. As a consequence, measured
∗
or calculated values of Buuu /r ∗ cannot exceed 4/5, unless  is evaluated incorrectly.
Iu can be estimated directly when the temporal or spatial decay of Buu is known.
Such information is readily available from DNS data and can be acquired for grid
turbulence when measurements of Buu are made at sufficiently fine steps in the x-
direction. When such measurements are not available, the assumption of similarity
allows ∂Buu /∂x to be approximated satisfactorily. As noted earlier, the non-stationarity
can be neglected either when r  L or when Rλ → ∞. In either case, it can be readily
shown (Batchelor 1947) that self-similarity of Buu and Buuu based on the Kolmogorov
variables uK and η satisfies (1.3). This result provides strong justification for the first
similarity hypothesis of Kolmogorov (1941b). When the non-stationarity is retained,
George (1992) postulated an equilibrium similarity, or similarity for all scales in
homogeneous isotropic turbulence, based on the Taylor microscale λ and the energy
u2 . When Rλ → ∞, this postulate is fully consistent with Kolmogorov similarity. It
is however less accurate than Kolmogorov similarity when Rλ is finite and r is small,
as was demonstrated by Antonia & Orlandi (2004), since the effect of Iu is negligible
under these conditions. As the effect of Iu increases, with increasing r, similarity based
178 R. A. Antonia and P. Burattini
1.0
(a) 4/5 (b) 4/5
0.8

0.6

0.4 Rλ

0.2

0
10–1 100 101 10–1 100 101
r/λ r/λ

Figure 1. (a) Terms in (2.1), divided by r, for grid turbulence (Rλ  40). O, viscous term;
䊊, third-order structure function; , Iu ; solid horizontal line: 4/5; dashed line: sum of the
viscous term, third-order structure function term and Iu . (b) Variation with Rλ of the terms
in (2.1), divided by r, for the grid turbulence data of Zhou & Antonia (2000) (Rλ increases
between 27 and 100 in the direction of the arrows). Solid lines: Buuu /r; dashed lines: Iu
(calculated by difference after assuming the validity of (2.1)); dash-dotted lines: viscous term;
solid horizontal line: 4/5.

on λ and u2  becomes more relevant. For this reason, we have chosen to estimate Iu
with the use of George’s similarity proposal. In grid turbulence, the expression for Iu
is

 4  
r/λ  5  
3U 5 du2  r/λ s Buu s 2 dλ 4 s ∂ Buu s
Iu = 4 λ d − u  λ d .
r dx 0 λ u  2 λ dx 0 λ ∂(r/λ) u 2 λ
(2.3)
Its detailed derivation can be found in Antonia et al. (2003). The terms in (2.1) have
been measured in the turbulence generated by a biplane grid composed of square
rods (solidity 0.35) using an X-wire located at x/M = 52 (Rλ  40). For this flow, 
was obtained reliably from the turbulent energy budget
1 dq 2 
 = − U , (2.4)
2 dx
where q 2  = u2  + v 2  + w2 . Here, q 2  was approximated by q 2  = u2  + 2v 2 
since v 2   w2 . The results are shown in figure 1(a) after normalizing by r.
The imbalance, or difference between 4/5 and the sum of the remaining normalized
terms in (2.1), is negligible at nearly all values of r/λ, reflecting partly the appro-
priateness of the similarity assumption used to estimate Iu and also the accuracy with
which  has been obtained. As noted by Antonia et al. (2000b), the presence of Iu
in (2.1) allows compliance with two important results. The energy budget, (2.4), is
retrieved at large r, whilst the decay of  or, equivalently, the decay of enstrophy
in homogeneous turbulence, is correctly reproduced in the limit of r → 0. At r ≈ λ,
(2.1) is almost perfectly satisfied by the data, implying that  can be estimated from
this equation only from a knowledge of Buu and Buuu for values of r close to λ. That
the influence of Iu on scales corresponding to the maximum value of the normalized
third-order structure function, i.e.
Au = max(Buuu /r), (2.5)
should diminish as Rλ increases can be readily inferred from figure 1(b); here the data
in Zhou & Antonia (2000) have been used. Although the range of Rλ is limited, the
trend is unmistakable. The implication is that the onset of a scaling range becomes
 Approach to the 4/5 law in turbulence 179
1.0
4/5
0.8

0.6

0.4

0.2

0
10–1 100 101
r/λ

Figure 2. Terms in (2.1), divided by r, using the DNS data of Fukayama et al. (2000)
at roughly the same Rλ ( 70). Solid symbols: forced turbulence; open symbols: temporally
decaying turbulence. 䊊, Third-order structure function; , Iu ; O, viscous term; dashed lines:
sum of the viscous term, third-order structure function term and Iu ; solid horizontal line: 4/5.

more realistic with increasing Rλ . Near r = λ, the magnitude of Iu , estimated here from
∗ ∗
the measured values of Buu and Buuu and by assuming the validity of (2.1), decreases
with Rλ at least as rapidly as that of the viscous term.
For a fixed Rλ , Iu /r is unlikely to be universal, even for a given type of flow. In
grid turbulence, Iu /r has been found to depend, but only slightly, on the shape and
solidity of the grid (Lavoie et al. 2005). Antonia & Burattini (2004) pointed out that
for the wake data of Antonia, Zhou & Romano (2002), obtained at the same distance
(x/D = 70) downstream of 5 different wake generators, the magnitude and shape
of Buuu /r can vary significantly, even though Rλ was nominally the same ( 200)
in each case. The implication is that Iu /r can vary between differently generated
wakes, reflecting the different degrees of organization in each wake.
For forced turbulence, Moisy et al. (1999) assumed (following Novikov 1993)
2 r3
Iu =  2 , (2.6)
7 Lf
where Lf is an integral scale which characterizes the forcing and was found to be
nearly constant. The value of  was obtained via (2.1) and (2.6) by fitting to the
measured third-order structure function. Increasing Lf corresponds to stirring the flow
at relatively larger scales. Lf is limited by the size of the test section in experiments
and the size of the DNS box. Gotoh et al. (2002) and Fukayama et al. (2000) used


2 3 ∞ 2
Iu = r k F (k) dk, (2.7)
35 0

in the context of DNS for forced box turbulence. Here, F (k) is the spectrum of
the random force, constant over a small, low-wavenumber range and zero elsewhere.
Fukayama et al. (2000) presented all the terms of (2.1) for Rλ  70. Their data are
plotted in figure 2 using a linear scale for the ordinate and r/λ along the abscissa.
The isotropic relation λ/η = 151/4 Rλ1/2 was used to convert the values of r/η given
by Fukayama et al. (2000). They also carried out simulations of temporally decaying
box turbulence; their results for Rλ  70 are included in figure 2. Although there is
practically no difference between the viscous terms, the shape of Iu differs between
180 R. A. Antonia and P. Burattini
2.5 1.0
(a) (b) 106 4/5
2.0
106
B*uu/r*2/3 1.5
102
Rλ = 104 0.5
1.0 Rλ = 104
102
0.5
0 0
10–4 100 104 100 105
r/λ r/λ

Figure 3. (a) Compensated second-order structure function. Solid lines: model (2.8); 䊊,
measured grid turbulence data of Zhou & Antonia (2000) at Rλ = 100; dashed line: asymptotic
value corresponding to Cu2 and model of Lindborg (1999) and Lundgren (2003). (b) Variation
with Rλ of the terms in (2.1), divided by r, in decaying isotropic turbulence assuming (2.8).
Dotted line: third-order structure function; dash-dotted line: Iu ; dashed line: viscous term.

the two cases. With forcing, Iu is smaller for small separations and rises steeply at
large separations. As a consequence, Au is bigger and located at slightly larger r/λ. In
decaying turbulence, Buu approaches a plateau slowly, at large r. In forced turbulence,
the shape of Buu varies according to the type of forcing that is applied. To examine
the dependence on Rλ of all terms in (2.1), we follow the approach of Antonia et al.
∗
(2003) who used a description of Buu (Kurien & Sreenivasan 2000) which extends
from the smallest (Kolmogorov) length scale to the integral length scale L,
∗ r ∗2 (1 + βr ∗ )
Buu = . (2.8)
15 (1 + (r ∗ /rcu
∗ )2 )c

∗
Here, rcu is identified with the crossover between the dissipative and inertial ranges,
c ≡ (1 − ζu /2) and β ≡ L∗−1 . Equation (2.8) is a modification, for finite Reynolds
∗
numbers, of the model for Buu first proposed by Batchelor (1951) with β = 0, figure 3.
For isotropic turbulence, λ = 151/4 Rλ1/2 , u∗2  = Rλ /151/2 and L∗ ≡ 15−3/4 C Rλ3/2 ,
∗

where L has been identified with C u2 3/2 /. The dimensionless energy dissipation
rate parameter C is expected to become constant at sufficiently large Rλ but its
magnitude should depend on the initial conditions. In general, one expects the shape
∗
of Buu to depend on the type and level of organization in a particular flow (e.g.
Lavoie et al. 2005), which most likely reflects the influence of the initial conditions.
A value of 1 for C was assumed, as in Antonia et al. (2003). We have also assumed
∗
that rcu = (15Cu2 )3/4 , with a value of 2 for the Kolmogorov constant Cu2 while the
∗
K41 value of 2/3 was assigned to ζu . Although Cu2 , ζu and rcu may vary slightly with
Rλ (e.g. Antonia, Pearson & Zhou 2000a) when Rλ is small, the present estimates of
Iu and Buuu should be sufficiently accurate to provide a reasonable indication of how
∗
Kolmogorov’s 4/5 law is approached. Figure 3(a) shows that Buu /r ∗2/3 becomes close
to 2 (the Kolmogorov constant) at Rλ  10 , although a reasonably sized plateau
4

for this quantity is achieved only at Rλ  106 . Lindborg (1999) and Lundgren (2003)
assumed, using Kolmogorov’s second similarity hypothesis, that
 r 2/3
Buu = Cu2 u2  = Cu2 (r)2/3 (2.9)
L
with Cu2 = 2. This approach neglects both the dissipative range and the Rλ -dependence
of the inertial range. The grid turbulence data (Rλ = 100) of Zhou & Antonia (2000)
have been included in figure 3(a) to show that (2.8) approximates the measured data
adequately.
 Approach to the 4/5 law in turbulence 181
0.45

0.40

0.35

Au

0.30

0.25

0.20
20 30 40 50 60 70
Rλ

Figure 4. Variation of Au with Rλ for Rλ . 60. Solid line: based on (2.8), see text for details;
symbols, grid turbulence behind biplane grids of different shapes and solidities (Lavoie et al.
2005): O, round bars, solidity 0.35; 䊊, round bars, solidity 0.44; : square bars, solidity 0.35.

Buuu was evaluated after introducing (2.8) into (2.1), and assuming a value of
the decay rate exponent m for the turbulent kinetic energy of −1.25 (close to that
measured by Antonia et al. 2003). This is shown in figure 3(b), where the viscous
term and Iu are also included.† As Rλ increases, the latter two shift to the left and
right respectively, and Buuu appears to exhibit a convincing plateau only when Rλ is
about 106 . Experimental values of Au in low-Rλ (. 60) grid turbulence are compared
with the model in figure 4. The data were obtained for different grid geometries and
solidities (see Lavoie et al. 2005). Notwithstanding the scatter in the data for each of
the grids, the small effect of the different initial conditions at the grids, and the lack
of information in the model on the effect of initial conditions, the agreement between
the model and the data is satisfactory, lending support to (2.8) and the hypotheses
assumed in applying (2.2) to grid turbulence. The agreement also reflects the fact that
 was estimated accurately via (2.4) for these data, and measurements were made at
small intervals (equal to M) in the x-direction. This approach eliminated the need to
use a similarity hypothesis since ∂Buu /∂x in (2.2) could be estimated directly. For all
data in figure 4, the departure from isotropy, as measured by the ratio u2 /v 2  say,
is of order 10 %; using a secondary contraction downstream of each grid resulted in
this ratio being closer to 1 but the magnitude of Au was essentially unchanged (Lavoie
2005). The solid curve in figure 4 has been extended to much higher values of Rλ in
figure 5. Included in figure 5 are box turbulence data as well as previously published
grid turbulence data (for clarity, the data in figure 4 are not shown). Clearly, 4/5 is
reached much more rapidly for the forced than for the decaying data, the values of
Rλ that are required being about 103 and 106 respectively. As already noted from the
distributions in figure 2 at the same Rλ , the difference in the trends can be attributed

† As Lindborg (1999) and Lundgren (2003) evaluated Buuu by using (2.9), Buuu /r does not go
to zero at small and large r (figures 1 and 2 of Lindborg (1999) and figure 1 of Lundgren (2003)).
182 R. A. Antonia and P. Burattini
1.0
Solid symbols: forced
0.9
Open symbols: decaying
4/5
0.8

0.7
100
0.6
Au 0.5
Based on 10–1 –2/3
0.4 (2.8) Fu

0.3 10–2
–6/5
0.2 Lindborg,
Linearly 10–3
Lundgren
0.1 forced 101 102 103 104
Rλ
0
101 102 103 104 105
Rλ

Figure 5. Variation with Rλ of Au . Solid line: based on (2.8); H, DNS data for forced box
turbulence (Gotoh et al. 2002); J, DNS data for forced box turbulence (Fukayama et al. 2000);

, DNS data for decaying box turbulence (Fukayama et al. 2000); ⊗, DNS data for decaying
box turbulence (Orlandi & Antonia 2002); , measurements of Moisy et al. (1999) for forced
turbulence; 䉫, grid turbulence (Gagne et al. 2004); 䊊, grid turbulence (Zhou & Antonia 2000);
M, grid turbulence (van de Water & Herweijer 1999); , active grid turbulence (Mydlarski &
Warhaft 1996); solid line: model based on (2.8); dash-dotted line: model of Lundgren (2003)
for linearly forced turbulence; dashed line: model of Lindborg (1999) and Lundgren (2003) for
decaying turbulence; dotted line: fit to the data of Moisy et al. (1999); dashed horizontal line:
4/5. Inset: Fu ≡ (4/5 − Au ) as a function of Rλ on a log-log scale. The slope of the dashed
and dash-dotted lines is −2/3 (Lindborg 1999; Lundgren 2003; Qian 1999), while the slope of
the dotted line is −6/5 (Moisy et al. 1999; Qian 1999).

to Iu (although the difference between the two values of Au in figure 2 is small, it is

nonetheless significant, as noted by the authors). The agreement with the DNS data
of Orlandi & Antonia (2002) and the grid turbulence data of Gagne et al. (2004),
Zhou & Antonia (2000) and van de Water & Herweijer (1999) is good (for the latter,
only the data at the centre of the working section were used). However, there is poor
agreement with some of the active grid turbulence data of Mydlarski & Warhaft
(1996), which exhibit a large scatter, reflecting the uncertainty in estimating  in
those experiments. For Rλ & 200, the model becomes almost indistinguishable from
the predictions of Lindborg (1999) and Lundgren (2003). For smaller values of Rλ ,
the present model (based on (2.8)) is superior because it accounts for the restricted
∗
scaling range, vis-à-vis (2.9). In (2.9), the assumed values of Buu are consistently higher
than those of (2.8) (see figure 3a). For Rλ & 40, results for linearly forced turbulence
(Lundgren 2003) lie between those for decaying and forced turbulence; this is not
surprising given that the energy level is sustained across all turbulent scales. The
dotted line was provided by Moisy et al. (1999) as a fit to their data. The inset
in figure 5 shows the variation of Fu ≡ (4/5 − Au ) on a log-log scale, as used by
Zhou et al. (2000). For Rλ & 100, the data follow two separate power-law regimes.
For decaying turbulence, the power-law exponent is −2/3 as predicted by Lindborg
(1999), Lundgren (2003) and Qian (1999). The latter author arrived at this result by
 Approach to the 4/5 law in turbulence 183
using the spectral equation and a model for the turbulent energy spectrum which has
a slope of −5/3 in the inertial range. The present model is also consistent with the
−2/3 power-law behaviour when Rλ exceeds about 200; below this value, it indicates
a much slower rate of decay. Qian (1999) emulated forcing by considering a Dirac
δ-function injection of energy at small wavenumbers and predicted a decay exponent
for Fu of −6/5, in accord with the data of Gotoh et al. (2002) and Moisy et al.
(1999), although it should be kept in mind that the latter authors did not measure
 directly in (2.1).

3. Concluding comments
The results in figure 5 indicate that the asymptotic value of 4/5 is approached more
rapidly for forced than for decaying turbulence. Correspondingly, the Kolmogorov
inertial range is likely to be established more rapidly when forcing is applied; this
vindicates the use of forcing for emulating Kolmogorovian turbulence. Figures 3 and 5
indicate that, for a substantial range to exist, Rλ should exceed about 103 when forcing
is applied and 106 when the turbulence is decaying; the latter estimate is consistent
with that proposed by Antonia et al. (2003). The difference in the magnitude of
Au , the maximum value of Buuu /r, between the two cases reflects a difference in
the magnitude of Iu , the non-stationarity or inhomogeneity in (2.1). This difference
suggests that turbulence statistics associated with scales for which Buuu /r is close
to its maximum are unlikely to be universal, at least for Reynolds numbers normally
encountered in the laboratory. Consequently, results for forced and decaying turbu-
lence should be interpreted in the context of this difference, even when they have been
obtained at the same Rλ . It is worth pointing out that the present difference between
forced and decaying results was anticipated by Lindborg (1999) who suggested that
experiments aimed at testing intermittency should be carried out in stationary flows.

The support of the Australian Research Council is acknowledged. We are especially

grateful to L. Danaila for her contribution to various aspects of this work.

REFERENCES
Antonia, R. A. & Burattini, P. 2004 Small-scale turbulence: How universal is it? In 15th
Australasian Fluid Mechanics Conference (ed. M. Behnia, W. Lin & G. D. McBain),
p. AFMC00184. The University of Sydney.
Antonia, R. A. & Orlandi, P. 2004 Similarity of decaying isotropic turbulence with a passive
scalar. J. Fluid Mech. 505, 123–151.
Antonia, R. A., Pearson, B. R. & Zhou, T. 2000a Reynolds number dependence of second-order
velocity structure functions. Phys. Fluids 12, 3000–3006.
Antonia, R. A., Smalley, R. J., Zhou, T., Anselmet, F. & Danaila, L. 2003 Similarity of energy
structure functions in decaying homogeneous isotropic turbulence. J. Fluid Mech. 487, 245–
269.
Antonia, R. A., Zhou, T., Danaila, L. & Anselmet, F. 2000b Streamwise inhomogeneity of
decaying grid turbulence. Phys. Fluids 12, 3086–3089.
Antonia, R. A., Zhou, T. & Romano, G. P. 2002 Small-scale turbulence characteristics of two-
dimensional bluff body wakes. J. Fluid Mech. 459, 67–92.
Batchelor, G. K. 1947 Kolmogoroff’s theory of locally isotropic turbulence. Proc. Camb. Phil. Soc.
43, 533–559.
Batchelor, G. K. 1951 Pressure fluctuations in isotropic turbulence. Proc. Camb. Phil. Soc. 47,
359–374.
Casciola, C. M., Gualtieri, P., Benzi, R. & Piva, R. 2003 Scale-by-scale budget and similarity
laws for shear turbulence. J. Fluid Mech. 476, 105–114.
184 R. A. Antonia and P. Burattini
Chassaing, P., Antonia, R. A., Anselmet, F., Joly, L. & Sarkar, S. 2002 Variable Density Fluid
Turbulence. Kluwer.
Danaila, L., Anselmet, F., Zhou, T. & Antonia, R. A. 1999 A generalization of Yaglom’s equation
which accounts for the large-scale forcing in heated decaying turbulence. J. Fluid Mech. 391,
359–372.
Danaila, L., Anselmet, F., Zhou, T. & Antonia, R. A. 2001 Turbulent energy scale budget
equations in a fully developed channel flow. J. Fluid Mech. 430, 87–109.
Danaila, L., Antonia, R. A. & Burattini, P. 2004 Progress in studying small-scale turbulence using
‘exact’ two-point equations. New J. Phys. 6 (128), 1–23.
Fukayama, D., Oyamada, T., Nakano, T., Gotoh, T. & Yamamoto, K. 2000 Longitudinal structure
functions in decaying and forced turbulence. J. Phys. Soc. Japan 69, 701–715.
Gagne, Y., Castaing, B., Baudet, C. & Malécot, Y. 2004 Reynolds number dependence of
third-order velocity structure functions. Phys. Fluids 16, 482–485.
George, W. K. 1992 The decay of homogeneous isotropic turbulence. Phys. Fluids A 4, 1492–1509.
Gotoh, T., Fukayama, D. & Nakano, T. 2002 Velocity field statistics in homogeneous steady
turbulence obtained using a high-resolution direct numerical simulation. Phys. Fluids 14,
1065–1081.
von Kármán, T. & Howarth, L. 1938 On the statistical theory of isotropic turbulence. Proc. R.
Soc. Lond. A 164, 192–215.
Kolmogorov, A. N. 1941a Dissipation of energy in the locally isotropic turbulence. Dokl. Akad.
Nauk SSSR 32 (1), English translation in Proc. R. Soc. Lond. A 434 (1991), 15–17.
Kolmogorov, A. N. 1941b Local structure of turbulence in an incompressible fluid for very large
Reynolds numbers. Dokl. Akad. Nauk SSSR 30 (4), 299–303.
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.
Kurien, S. & Sreenivasan, K. R. 2000 Anisotropic scaling contributions to high-order structure
functions in high-Reynolds-number turbulence. Phys. Rev. E 62, 2206–2212.
Lavoie, P. 2005 The effect of initial conditions in grid-generated turbulence. PhD thesis, University
of Newcastle (in preparation).
Lavoie, P., Burattini, P., Djenidi, L. & Antonia, R. A. 2005 Effect of initial conditions on decaying
grid turbulence at low Rλ . Exps. Fluids 39, 865–874.
Lindborg, E. 1999 Correction to the four-fifths law due to variations of the dissipation. Phys. Fluids
11, 510–512.
Lundgren, T. S. 2003 Linearly forced turbulence. CTR Annual Research Briefs.
Moisy, F., Tabeling, P. & Willaime, H. 1999 Kolmogorov equation in a fully developed turbulence
experiment. Phys. Rev. Lett. 82, 3994–3997.
Mydlarski, L. & Warhaft, Z. 1996 On the onset of high-Reynolds-number grid-generated wind
tunnel turbulence. J. Fluid Mech. 320, 331–368.
Novikov, E. A. 1993 Statistical balance of vorticity and a new scale for vortical structures in
turbulence. Phys. Rev. Lett. 71, 2718–2720.
Orlandi, P. & Antonia, R. A. 2002 Dependence of the non-stationary form of Yaglom’s equation
on the Schmidt number. J. Fluid Mech. 451, 99–108.
Qian, J. 1999 Slow decay of the finite Reynolds number effect of turbulence. Phys. Rev. E 60,
3409–3412.
Qian, J. 2002 Scaling of structure functions in homogeneous shear-flow turbulence. Phys. Rev. E
65, 036301.
Saffman, P. G. 1968 Lectures on homogeneous turbulence. In Topics in Nonlinear Physics (ed.
N. Zabusky), pp. 485–614. Berlin: Springer.
Sreenivasan, K. R. & Dhruva, B. 1998 Is there scaling in high-Reynolds-number turbulence? Prog.
Theor. Phys. Suppl. 130, 103–120.
van de Water, W. & Herweijer, J. A. 1999 High-order structure functions of turbulence. J. Fluid
Mech. 387, 3–37.
Zhou, T. & Antonia, R. A. 2000 Reynolds number dependence of the small-scale structure of grid
turbulence. J. Fluid Mech. 406, 81–107.
Zhou, T., Antonia, R. A., Danaila, L. & Anselmet, F. 2000 Approach to the four-fifths ‘law’ for
grid turbulence. J. Turb. 1 (5), 1–12.