Lohse2010small Scale

ANRV400-FL42-15
ARI
ANNUAL
REVIEWS
13 November 2009
14:9
Further
Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Access provided by Indian Institute of Science - Bangalore on 08/05/16. For personal use only.
Click here for quick links to

Annual Reviews content online,
including:
Other articles in this volume
Top cited articles
Top downloaded articles
Our comprehensive search
Small-Scale Properties of
Turbulent Rayleigh-Benard
Convection
Detlef Lohse1 and Ke-Qing Xia2
1
Physics of Fluids Group, Department of Science and Technology, J.M. Burgers Center for
Fluid Dynamics, and Impact-Institute, University of Twente, 7500 AE Enschede,
The Netherlands; email: d.lohse@utwente.nl
Department of Physics, The Chinese University of Hong Kong, Shatin, Hong Kong, China;
email: kxia@phy.cuhk.edu.hk
Annu. Rev. Fluid Mech. 2010. 42:33564
Key Words
First published online as a Review in Advance on

September 2, 2009
thermal convection, turbulence, structure functions, Bolgiano scaling
The Annual Review of Fluid Mechanics is online at

uid.annualreviews.org
Abstract
This articles doi:

10.1146/annurev.uid.010908.165152
c 2010 by Annual Reviews.
Copyright
All rights reserved
0066-4189/10/0115-0335$20.00
The properties of the structure functions and other small-scale quantities in

turbulent Rayleigh-Benard convection are reviewed, from an experimental,
theoretical, and numerical point of view. In particular, we address the question of whether, and if so where in the ow, the so-called Bolgiano-Obukhov
scaling exists, i.e., S (r) r 2/5 for the second-order temperature structure
function and Su (r) r 6/5 for the second-order velocity structure function.
Apart from the anisotropy and inhomogeneity of the ow, insufciently high
Rayleigh numbers, and intermittency corrections (which all hinder the identication of such a potential regime), there are also reasons, as a matter of
principle, why such a scaling regime may be limited to at most a decade,
namely the lack of clear scale separation between the Bolgiano length scale
LB and the height of the cell.
335
ANRV400-FL42-15
ARI
13 November 2009
14:9
1. INTRODUCTION

Thermally driven turbulence is of tremendous importance to various areas of science, technology,

and the environment and in the geophysical and astrophysical context. For ow in the atmosphere,
thermal convection (e.g., see Hartmann et al. 2001) is relevant both on smaller length scales and
timescales for weather predictions and on larger scales for climate calculations. In the ocean
(e.g., see Marshall & Schott 1999), thermohaline convection (Rahmstorf 2000) is a main driving
mechanism of deep-ocean circulation. Thermally driven convection takes place both in Earths
outer core (see, e.g., Cardin & Olson 1994) and in its mantle (see, e.g., McKenzie et al. 1974).
The idealized system of thermally driven turbulence is turbulent Rayleigh-Benard (RB)
convectiona uid in a box of height L strongly heated from below and cooled from above.
This system may even be considered as a raw model for turbulent ow in general. It is not
only a mathematically well-dened problem, in principle given by the Boussinesq equations (see
Section 2) and the appropriate boundary conditions for the velocity eld u(x, t) and the temperature eld (x, t), but it can also be straightforwardly realized experimentally. Figure 1 shows a
shadowgraph image of turbulent thermal convection in an aspect-ratio-one cylindrical cell, together with two instantaneous two-dimensional (2D) velocity eld snapshots, taken with particle
image velocimetry (PIV).
In contrast, the theoretical concept of homogeneous and isotropic turbulence (e.g., see
Batchelor 1953, Frisch 1995, Kolmogorov 1941, Monin & Yaglom 1975, Pope 2000) is much
harder to realize in a laboratory. Nonetheless, most theoretical (Frisch 1995; e.g., for a more
recent review, see Falkovich et al. 2001) and numerical (recently reviewed in Ishihara et al. 2009)
work on the (small scale) scaling of structure functions (SFs) and energy spectra focused on
0.0
2.5
0.0
Velocity (cm s1)
1.1
Velocity (cm s1)
Figure 1
(Middle panel ) Shadowgraph image of turbulent thermal convection in an aspect-ratio-one cylindrical cell (Ra = 6.8 108 , Pr = 596).
The thermal plumes in the image play important roles in the small-scale properties of the systems velocity and temperature elds. (Left
and right panels) Instantaneous high-resolution 2D velocity eld measured in the sidewall (left panel ) and central regions (right panel )
marked by the red rectangle and blue square in the shadowgraph, respectively. Because of the large disparity in the number of plumes,
these two regions exhibit different scaling properties of velocity and temperature elds, as found by Sun et al. (2006). Shadowgraph
taken from Xi et al. 2004, PIV taken from Zhou et al. 2008.
336
Lohse
Xia
ANRV400-FL42-15
ARI
13 November 2009
14:9
homogeneous, isotropic turbulence, comparing these results with laboratory and eld measurements for which the conditions of homogeneity and isotropy were only approximately fullled.
In their Annual Review of Fluid Mechanics article, Sreenivasan & Antonia (1997) describe the phenomenology of this small-scale turbulence. In the inertial range, the second-order velocity SF
Su (r) = (u(x + r) u(x))2 x,t is basically found to scale with the Kolmogorov (1941) scaling
r2/3 , in short called K41, apart from small intermittency corrections, which are found to be remarkably universal (Arneodo et al. 1996). K41 follows from pure dimensional analysis, assuming
that in the inertial range, apart from the scale r itself, the only relevant parameter is the mean
energy dissipation rate u , giving

Su (r) (u r)2/3 .
(1)
Obukhov (1949) and Corrsin (1951) generalized Kolmogorovs argument to the uctuations of a
passive scalar (x, t), giving
S (r) u1/3 r 2/3
(2)
for the second-order SF S (r) = ( (x + r) (x)) x,t of (x, t). Here is its dissipation rate.
Again, apart from small intermittency corrections, Equation 2 is in reasonable agreement with the
experimental data (see Warhaft 2000).
In analogy to K41, pure dimensional analysis can also be done for RB turbulence. Assuming that
next to the scale r, the only relevant parameters are the mean thermal dissipation rate and the
product of the thermal expansion coefcient and gravity g, one obtains the so-called BolgianoObukhov scaling (BO59), which was rst suggested for stably stratied convection (Bolgiano 1959,
Obukhov 1959). It reads
2
2/5
Su (r) (g)4/5 r 6/5 ,
(3)
S (r) (g)2/5 r 2/5 .
(4)
4/5
According to the same dimensional argument, the mixed SF between temperature and vertical
velocity Su3 (r) = ((x + r) (x))(u 3 (x + r) u 3 (x)x,t should scale as
3/5
Su3 (r) (g)1/5 r 4/5 .
(5)
When comparing Equations 1 and 3, on one hand, or Equations 2 and 4, on the other hand, one
can read off the so-called Bolgiano length LB as crossover scale, with
3/4
L B = u5/4
(g)3/2 .
(6)
For L B r L, one expects BO59 scaling, whereas for r L B one still expects K41
scaling (see Figure 2). (Here, the height L of the cell is identied with the outer length scale
and = 3/4 / 1/4 is the Kolmogorov length, i.e., the inner length scale.) However, the scaling as
sketched in Figure 2 has hitherto not been identied in experimental or in numerical data, neither
for the SFs nor in the respective Fourier transforms, for which BO59 would mean Eu (k) k11/5
for the energy spectrum and E (k) k7/5 for the thermal spectrum.
In this article we review the experimental and numerical attempts to at least partly identify
BO59 scaling. Owing to space limitations, we do not strive to review all aspects of the problem
that may be called small-scale. For example, we do not discuss the extraction and characterization
of individual thermal plumes (Ching et al. 2004b, Funfschilling et al. 2008, Julien et al. 1999,
Puthenveettil & Arakeri 2005, Shishkina & Wagner 2008, Zhou & Xia 2002, Zhou et al. 2007)
and studies of the energy and thermal dissipation rates (Benzi et al. 1998, Ching & Kwok 2000,
He & Tong 2009, He et al. 2007, Kaczorowski & Wagner 2009, Kerr 2001, Shishkina & Wagner
2006, Verzicco & Camussi 2003), nor do we review the small-scale properties in 2D turbulent
www.annualreviews.org Turbulent Rayleigh-Benard Convection
337
ANRV400-FL42-15
ARI
13 November 2009
14:9
2/5
log(Su(r)), log(S(r))
6/5
10
2/3
LB
15

20
10
log(r/)
10
Figure 2
Sketch of the second-order velocity (blue curve) and temperature ( purple curve) structure functions, as they
follow from BO59 dimensional analysis: L B = 106 and L = 1011 are assumed, and there is a crossover at
10 (e.g., see Efnger & Grossmann 1987) between the viscous subrange and the inertial subrange. For
10 < r < LB , the BO59-type dimensional analysis gives K41 scaling (Equations 1 and 2), and for LB < r < L
it gives BO59 scaling (Equations 3 and 4). Such pronounced scaling as sketched here has hitherto never been
observed.
thermal convection (Celani et al. 2001, 2002; Zhang & Wu 2005; Zhang et al. 2005) and those
in the Lagrangian frame (Gasteuil et al. 2007, Schumacher 2008). Rather, we focus on the issue
regarding the existence of BO59 and how the properties of an active scalar as a small-scale quantity
compare with those of a passive scalar in convective ows. Section 2 presents numerical data on
the Bolgiano scale LB (which turns out to be the crucial quantity) and demonstrates that the BO59
dimensional argument is too simplistic. We then point out further difculties, both practical and
as a matter of principle, in identifying BO59 scaling: the inhomogeneity and anisotropy of the
ow and intermittency. Section 3 presents experimental and numerical data on the SFs, whereas
Section 4 compares the ow characteristics of an active scalar, such as temperature in RB convection, with that of a passive scalar.
The review aims to give an update on the statistics of small-scale quantities in RB convection
to the status as described in Siggia (1994). An update on large-scale quantities and in particular
on how the Nusselt number (Nu) and the Reynolds number (Re) depend on the Rayleigh number
(Ra) and the Prandtl number (Pr) has already been given by Ahlers et al. (2009). Along with Ra,
Pr, and the shape of the cell, the other control parameter of the system is the aspect ratio ; here
we focus on 1.
2. BOLGIANO LENGTH SCALE AND DIFFICULTIES

IN REALIZING BO59 SCALING
2.1. Bolgiano Length Scale
The rst step in judging whether BO59 scaling is possible is to determine the Bolgiano length
LB from Equation 6. Assuming spatial homogeneity, one can do this based on rigorous relations,
within the approximations of the Boussinesq equation (see, e.g., Shraiman & Siggia 1990),
u (i u j (x, t))2 V,t =
338
Lohse
Xia
3
(Nu 1)RaPr2 ,
L4
(7)
ANRV400-FL42-15
ARI
13 November 2009
14:9
2
Nu,
(8)
L2
for the volume and time-averaged kinetic and thermal dissipation rates, respectively. Here Ra =
gL3 /() and Pr = / are the Rayleigh and Prandtl numbers, with the applied temperature
difference, and and the kinematic viscosity and the thermal diffusivity, respectively. For Nu
1, these relations imply
(i (x, t))2 V,t =
L B /L Nu1/2 /(PrRa)1/4 .
(9)
1/3
L B /L Ra1/12 Pr1/4 ,
(10)
giving hope that at least for very large Ra and large Pr a separation of scales between LB and L
may arise.
However, Relation 9 and the above argumentation are extremely misleading, as the SFs and
the energy and thermal dissipation rates should be locally dened (i.e., as function of x), whereas
Equation 9 is a global relation. Therefore, one must consider local Bolgiano lengths
3/4
L B (x) = u5/4 (x)
(x)(g)3/2 .
(11)
This quantity became only recently accessible, thanks to local multiple probe measurements (e.g.,
see He & Tong 2009, He et al. 2007, Wittmer et al. 1998) and in particular to the advancement
of numerical simulations (reviewed in Ahlers et al. 2009). Figure 3 reveals the very strong inhomogeneity of the kinetic and thermal dissipations rates, which of course is not surprising, given
the boundary conditions of the ow. The strong peaks of the time-averaged energy dissipation
rate u (x) at all boundaries reect the no-slip boundary condition. In contrast, the time-averaged
log10(u)
2.2
1.0
log10()
1.5
2.4
0.4
3.2
0.6
3.5
0.4
0.2
r/L
0.6
0.6
0.4
0.4
0.2
0.3
0.4
4.0
0.0
0.0
0.2
0.2
0.4
z/L
0.6
0.8
1.0
0.2
4.5
3.6
0.0
0.1 0.2 0.3 0.4 0.5
0.8
0.5
3.0
3.4
0.2
0.6
z/L
3.0
1.0
2.5
z/L
2.8
0.7
0.8
2.6
0.6
2.0
0.8
0.8
LB
1.0
LB(z)/L
1.0
z/L

Assuming the large-Ra power law Nu Ra , which follows from Grossmann & Lohses (2000)
theory and is consistent with the experimental data of Niemela et al. (2000), Nikolaenko et al.
(2005), and Niemela & Sreenivasan (2006), one obtains
0.0
0.1 0.2 0.3 0.4 0.5 5.0
r/L
0.1
0.0
0.1 0.2 0.3 0.4 0.5
r/L
Figure 3
(a) Energy dissipation rate u (x), (b) thermal dissipation rate (x), and (c) Bolgiano length LB (x), all obtained from the numerical
simulations of Kunnen et al. (2008) in a cylindrical cell with = 1, Ra = 109 , and Pr = 6.4. (d) Vertical prole of LB (z), as obtained
from the numerical simulations of Calzavarini et al. (2002) with periodic boundary conditions in the horizontal direction with
Ra = 3.5 107 and Pr = 1. Here, LB (z) has been obtained in two independent ways: by means of Equation 11 (red circles) and by
evaluating u r and gr Su3 (r) (the two terms on the right-hand side of Equation 18) and determining the (height-dependent) scale
when they cross. These scalesmultiplied by an appropriate overall prefactor of order unityare shown as blue crosses.
339
ARI
13 November 2009
14:9
thermal dissipation rate (x) is only peaked at the top and at the bottom wall, where there are
strong temperature gradientsat the sidewall there is no peak because of the adiabatic sidewall
boundary conditions. The resulting local Bolgiano length (Equation 11) is shown in Figure 3c,
again reecting this strong inhomogeneity. Except close to the plates, where L B /L 0.1, the
Bolgiano length scale is not really separated from the outer length scale L. This result was found
earlier in Benzi et al.s (1998) and, with better precision, Calzavarini et al.s (2002) simulations,
which assume periodic boundary conditions in the horizontal direction: In the bulk of the ow,
LB is the same order of magnitude as the height of the cell (see Figure 3d ) and no BO59 can be
expected. According to these numerical ndings, the best chance to observe BO59 is close to the
upper and lower plates. Consistent with this nding, in homogeneous RB convectionconvection
with an imposed vertical temperature gradient and periodic boundary conditions also in the vertical direction (see also Calzavarini et al. 2005, 2006; Lohse & Toschi 2003) so that no boundary
layers can developBiferale et al. (2003) nd LB L and correspondingly no BO59 scaling at all.
The corresponding SFs, including their subleading corrections, are discussed in Section 3.5.
From the global relation (Equation 10), one may expect that the separation of scales will
improve for larger Ra, but again the global argument is misleading. When tting the numerical
values for LB (x) between 0.1 < r/L < 0.4 at half-height in the regime 108 Ra 1010 , Kunnen
et al. (2008) obtained the relation L B (x in bulk)/L = 0.024 Ra0.1070.016 with a positive power-law
exponent, suggesting that for larger Ra, what would be a potentially small regime of BO59 scaling
will vanish altogether. This indeed is found in the experiments of Sun et al. (2006) and Kunnen
et al. (2008) (and numerical simulations as well for the latter; see Section 3).
We try to rationalize the positive power-law exponent in L B (x in bulk)/L = 0.024 Ra0.1070.016 .
When we take the bulk estimates of the Grossmann-Lohse theory, u U 3 /L and U2 /L,
we obtain L B (x in bulk)/L Re3 Pr3/2 /Ra3/2 . For regime Iu of Grossmann & Lohses (2000, 2001,
2002, 2004) theory, where one has Re Ra1/2 Pr5/6 , this would imply L B (x in bulk)/L Pr1 ;
for regime IVu with Re Ra4/9 Pr2/3 , it would imply L B (x in bulk)/L Ra1/6 Pr1/2 . Neither
gives a positive power-law exponent with Ra, for which we therefore do not have any explanation.

ANRV400-FL42-15
2.2. Cascade Picture

Based on Richardsons (1926) cascade picture of turbulence (e.g., see Frisch 1995 for a more
modern presentation of this picture), on which Kolmogorovs 1941 theory is based, we now further
explain why a realization of BO59 scaling is so difcult, if not impossible. In Kolmogorovs view,
the vortices are hierarchically ordered. The relevant energy transport from larger scales to small
scales is local in scale, i.e., from scale r to typically scale r/2 and from there to scale r/4, and so
on, down to a scale where viscosity dominates (at the Kolmogorov scale ) (see Figure 4a). The
external forcing of homogeneous and isotropic turbulence only takes place at large scales of the
order of the size L of the system. The energy input rate ein on the largest scales then equals the
transport rate Tru ur3 /r from any scale r in the inertial subrange to scale r/2. Here and below, ur is
the typical velocity difference of uid particles separated over a distance r. In the viscous subrange
(i.e., at scales and smaller), the energy is dissipated at a rate u , which equals ein because of
statistical stationarity. The scale independence of Tru = e in = u immediately implies ur r 1/3 or
Relation 1. The structure of the energy transfer rate Tru ur3 /r is read off from the Navier-Stokes
equation, if multiplied by the velocity, by dimensional analysis. Correspondingly, for a passive
scalar , in the Obukhov-Corrsin picture, the thermal transfer rate Tr r2 ur /robtained from
multiplying the temperature advection equation with is independent of r and is equal to both
the thermal input rate and the thermal dissipation rate . Together with ur r 1/3 , this gives
r r 1/3 or Relation 2.
340
Lohse
Xia
ANRV400-FL42-15
ARI
13 November 2009
14:9
b
Thermal plume
input
euin
grur

T ur
u3r
r
T ur
grur
K41
balance
u3r
r
K41
balance
B059
balance
u
u
Figure 4
(a) Sketch of the Richardson cascade for homogeneous isotropic turbulence. The K41 balance results in the
K41 scaling ur r 1/3 . (b) Attempted sketch of the situation in Rayleigh-Benard ow: On one hand, thermal
plumes drive the large-scale convection roll. On the other hand, the buoyancy term gr ur , whose r
dependency is not a priori known, supplies kinetic energy on scale r. Both the K41 and the BO59 balances
are shown. If the cascade is downscale, the energy transfer term Tru transports the sum of all energies that
have been put into the various scales from L to r down the cascade.
What changes for an active scalar , i.e., in RB convection, where the underlying equations are
now the Oberbeck-Boussinesq equations (Landau & Lifshitz 1987)
t ui + u j j ui = i p + 2j ui + gi3 ,
t + u j j =
2j ,
(12)
(13)
together with the incompressibility conditions i ui = 0? [Here p(x, t) is kinematic pressure;

scaling-wise the corresponding term i p has the same structure as the advection term u j j ui .]
The thermal balance does not change; i.e., Tr r2 ur /r remains scale-independent. The
situation is rather different for the kinetic balance. On one hand, a large-scale convection roll
develops, which was rst described by Krishnamurti & Howard (1981). It is fed and driven by the
small-scale thermal plumes detaching from the boundary layers, as nicely sketched by Kadanoff
(2001). This implies an interaction that is scale-wise nonlocal. On the other hand, the large-scale
convection roll decays to smaller vortices, so energy is also transported from the large scales toward
the small scales. We try to sketch the situation in Figure 4b.
If one wanted to obtain BO59 scaling within the cascade picture, the balance gr ur Tru
3
ur /r would be required for all scales in the inertial subrange for the energy equation. Indeed, from
this balance, together with above thermal balance Tr , one immediately obtains BO59 scaling.
However, this would imply that T2ru Tru , as on scale r the energy gain from larger scales would
be required to be neglible as compared to the energy input through buoyancy gr ur on scale r. It
therefore would imply that gr ur g2r u 2r for all scale r in the inertial range. Employing the
341
ANRV400-FL42-15
ARI
13 November 2009
14:9
framework of Efnger & Grossmanns (1987) mean eld theory, Procaccia & Zeitak (1989, 1990)
made these assumptions, obtaining BO59 scaling. Similarly, in the framework of shell models (e.g.,
see Biferale 2003), such balances were also implemented by Brandenburg (1992), Ching & Cheng
(2008), Ching & Ko (2008), and Ching et al. (2008a,b), again giving BO59 scaling. Grossmann &
Lvov (1993) discuss these balances in Fourier space.
However, within the cascade picture, it is impossible to fulll the conditions T2ru Tru and
gr ur g2r u 2r over many scales r in the inertial range. Instead, the energy input through
buoyancy from the outer length scale L down to the length scale r will accumulate, and the relevant
balance on scale r is
=L
r
Tru ur3 /r
gr
ur
.
(14)
r
=r
When assuming some power-law scaling for r ur with some positive exponente.g.,
with 4/5 as suggested by the BO59 scaling relation (Equation 5) itselfthe sum
r
=L
4/5
+ (L/2)4/5 + (L/4)4/5 + + r 4/5 in this balance is then a quickly convergr
=r gr
u r
L
ing geometric series because r L. In particular, it quickly becomes independent of r, resulting in
K41 scaling. This illustrates the paradoxical nature of the BO59 scaling. Within the framework of
the reduced wave vector set approximation (see Eggers & Grossmann 1991, Grossmann & Lohse
1992a), Grossmann & Lohse (1991) indeed found the balance (Equation 14) to dynamically establish, giving basically K41 scaling and LB L. Only when locally injecting, with decreasing
scale, an increasing amount of kinetic energy into the cascadesomehow mimicking the kinetic
driving of the plumes detaching from the boundary layerswere Grossmann & Lohse (1992b)
able to achieve a scaling less steep than K41 for r but for ur , no scaling steeper than K41 could
be achieved.
2.3. Further Obstacles in Identifying BO59 Scaling

Apart from the inhomogeneity of the ow and the unavoidable limited separation of the length
scales between LB and L, there are further obstacles in identifying BO59 scaling. As pointed out by
Lohse (1994), a clear identication of the, at most, short BO59 regime (as shown above) is further
complicated by the proximity of the exponents for the BO59 scaling and for the scaling in shear
ows. Moreover, because of the large-scale convection roll, there is considerable shear in the RB
ow eld, making it strongly nonhomogeneous and anisotropic. Either from dimensional analysis
(Grossmann et al. 1994, Kuznetsov & Lvov 1981, Lohse 1994, Lohse & Muller-Groeling
1996,
Lumley 1967, Tennekes & Lumley 1972) or from a more systematic SO(3) decomposition of the
velocity eld (e.g., see Arad et al. 1998, 1999a,b; von der Heydt et al. 2001; and Biferale & Procaccia
2005 for a recent review), one obtains for shear ow either dominantly or subdominantly
u1/2 /s 3/2
Su (r) u2/3 L2/3

r 4/3 ,
s
(15)
1/3
S (r) u1/3 L1/3
,
s r
(16)
where Ls
is the shear length scale and s the typical shear rate. In Fourier space, the corresponding spectra for the kinetic energy and the thermal uctuations are Eu (k) u2/3 L2/3
k7/3
s
1/3 1/3 4/3
and E (k) u Ls k
, respectively. We note that this type of scaling has been observed
in various shear ows, although often only as an anisotropic correction to the (isotropic) K41
scaling. Experimental and numerical examples for the kinetic energy spectrum include Arad et al.
1998, Biferale & Toschi 2001, Biferale et al. 2002b, Kurien & Sreenivasan 2000, and Saddoughi
& Veeravalli 1994for a review, we again refer the reader to Biferale & Procaccia (2005)and
Sreenivasan (1991) gives an example for the temperature uctuation spectrum. Indeed, close to
342
Lohse
Xia
ANRV400-FL42-15
ARI
13 November 2009
14:9
the top and bottom plates, Kunnen et al. (2008, their gure 18) nd a shear spectrum Su (r) r 4/3
for the second-order SF of the vertical velocity.
A proper analysis of the ow properties clearly requires one to disentangle the shear effects
and thermal effects, with the help of the SO(3) decompositions, as has been attempted for RB
turbulence (Biferale et al. 2003, Rincon 2006). For the longitudinal p-th order velocity SF Su( p) (r) =
(u(x + r) u(x)) r/r) p x,t , the SO(3) decomposition reads
Su( p) (r) =
m= j

( p)
S j m (r)Y j m (r/r),
(17)
j =0 m= j
( p)

where Y j m (r/r) is the spherical harmonics (see Biferale & Procaccia 2005). The projections S j m
( p)
on the different anisotropic sectors are expected to scale with an m-independent exponent, S j m
j
r u ( p) . Yakhot (1992) has generalized the von KarmanHowarth equation (see Monin & Yaglom
1975) to the case of thermally driven turbulence, obtaining
Su(3) (r) r + gre z S u3 (r).
(18)
Here ez is the unit vector in the vertical direction, and for simplicity we have suppressed the
tensorial structure. The rst term on the right-hand side is isotropic ( j = 0), whereas in the
second term, the j = 1 factor gre z couples to S u3 (r) with its j = 1, 3, . . . contributions, giving
projections on all sectors j = 0, 1, 2, . . . . Biferale et al. (2003) point out that the dimensional
j =1
isotropic balance in the j = 0 sector reads Su(3), j =0 (r) r + gre z S u3 (r), with the second term
being subdominant, implying K41-type scaling, whereas in the anisotropic sectors j > 0, one
obtains
j 1
Su(3), j (r) gre z S u3 (r).
(19)
Relation 19 reveals that it is the anisotropic contribution of the third-order velocity SF that,
according to the BO59 dimensional argument, should balance the buoyancy term. In Section 3.5,
we report Biferale et al.s (2003) numerical results on homogeneous RB convection, showing that
Relation 19 does not hold. The clean SO(3) analysis of Biferale et al. (2003) was possible only
because the authors restricted themselves to the kind of articial case of homogeneous RB ow,
i.e., the case with an imposed vertical temperature gradient, but without any walls. In real RB ow,
walls are present, and close to the walls, the SO(3) analysis must be replaced by an SO(2) analysis,
as done by Biferale et al. (2002b) for shear ow. We do not further discuss this complication here.
Apart from the SO(3) decomposition [or close to walls, SO(2)], another helpful tool for data
analysis may be the property of extended self-similarity (ESS), as rst found by Benzi et al.
(1993). The method, plotting (logarithms of) SFs of different orders p against each other and
not against the scale r, is usually effective in revealing mutual scaling relations and in particular intermittency. In the context of RB convection, it has rst been applied by Benzi et al.
(1994) and later by Cioni et al. (1995), Ching (2000), Zhou & Xia (2001), Skrbek et al. (2002),
Ching et al. (2003b), and Kunnen et al. (2008). Independent of the scaling of the second-order
velocity SF (K41, BO59, shear ow scaling), dimensional scaling implies Su( p) (r) (Su(2) (r)) p/2 .
However, even for homogeneous and isotropic ows, there are major deviations from this dimensional scalingthe so-called intermittency corrections (see Frisch 1995, Ishihara et al. 2009,
Sreenivasan & Antonia 1997 for reviews). Such intermittency corrections are also present in RB
ow for both the velocity and the temperature SFs (see Berschadskii et al. 2004; Ching 2000;
Ching & Cheng 2008; Ching et al. 2003b, 2008a; Sun et al. 2006; Zhang & Wu 2005). Disentangling the anisotropic effects and intermittency effects (Biferale & Toschi 2001, Biferale et al.
2002a, Toschi et al. 1999) and effects from the buoyancy-driven mechanism remains a major
challenge.
343
ANRV400-FL42-15
ARI
13 November 2009
14:9
We would like to stress that even if the above-mentioned complications (e.g., shear, intermittency) were overcome, the intrinsic difculties of the BO59 scaling as discussed in Section 2.2 still
remain.
3. TEMPERATURE AND VELOCITY SPECTRA

AND STRUCTURE FUNCTIONS
3.1. Overview of Main Experimental Methods

Above we discuss the theoretical difculties in detecting BO59 scaling, namely anisotropy, inhomogeneity, intermittency, and the lack of a wide inertial range that can provide a sufcient
separation of the relevant length scales (, LB , Ls , and L). However, in addition there is also a
lack of suitable experimental techniques for direct and high-resolution simultaneous multipoint
measurements of the velocity and temperature elds.
By far most experimental studies on temperature and velocity uctuations employ only point
measurements, resulting in temperature and velocity time series acquired at a single point in the
convection cell. At best, such measurements are performed at several locations in the ow. The
basic methods at hand are thermometry or bolometry for the temperature time series and later
laser Doppler velocimetry (LDV) for the velocity time series. To connect the time-domain results
to the theoretical predictions made for the spatial domain, investigators invoke Taylors (1938)
frozen-ow hypothesis, either explicitly or implicitly. The validity of this hypothesis requires that
turbulent velocity uctuations are much smaller than the mean ow velocity. However, the mean
velocity is approximately zero in the central region of the convection cell and is comparable to the
root-mean-square (rms) velocity near the sidewall and plate regions (Qiu et al. 2000, Sun et al.
2005, Xia et al. 2003). Therefore, the condition for the Taylor hypothesis is often not met in
turbulent RB convection, and its applicability to the system is at best doubtful. Nonetheless, we
discuss results from such point measurements in Sections 3.2 and 3.3.
Only recently, with the maturing of PIV (see Adrian 1991) and other similar techniques for the
full velocity eld measurement, have the experimental limitations of single-point measurements
become less of a problem. We discuss the results of recent PIV measurements of the velocity eld
in RB convection in Section 3.4. Finally, the numerical results on temperature and velocity spectra
and SFs are reviewed in Section 3.5. Here, the full spatial information is available, but Ra is limited.
3.2. Time-Domain Measurements

The rst systematic measurements in turbulent RB convection by Libchaber and coworkers
(Castaing et al. 1989, Heslot et al. 1987, Procaccia et al. 1991, Sano et al. 1989, Wu et al. 1990)
were restricted to temperature time series, from which temperature frequency power spectra were
obtained through Fourier transformation. These temperature power spectra, and also those based
on later temperature time-series measurements at Pr 0.7 and larger, display a scaling exponent
around 1.3 0.1 (e.g., Procaccia et al. 1991) to 1.4 (e.g., Wu et al. 1990), similar to the BO59
value 7/5 (but also similar to the shear spectrum value 4/3). In contrast, low-Pr liquid metals
(such as mercury, Pr 0.025) show more complicated behavior, depending on the value of Ra.
We rst discuss experiments made in uids with Pr 0.7.
Libchaber and coworkers early experiments (Castaing et al. 1989, Heslot et al. 1987, Procaccia
et al. 1991, Sano et al. 1989, Wu et al. 1990; reviewed in Siggia 1994) used low-temperature helium
gas as the working uid and measured temperature frequency power spectra with miniaturized
semiconductor bolometers placed inside the convection cell. Wu et al.s (1990) experiments were
344
Lohse
Xia

ANRV400-FL42-15
ARI
13 November 2009
14:9
made in a = 0.5 cylindrical cell with Ra spanning from 7 106 to 4 1014 . Their main
result is that up to1 at least Ra = 7 1010 , the temperature power spectra exhibit a universal
shape characterized by a power-law range in the low-frequency region and an exponential decay
in the high-frequency tails, which is characteristic of dissipation range behavior. All power spectra
can be collapsed onto a single curve by a simple rescaling of power density and frequency, i.e.,
shifts in log-log plot. A slope of 1.4 was observed in the low-frequency power-law region, and
the authors noted that the scaling range increases with increasing Ra, starting from Ra 108 ,
where the scaling range is almost zero. Interestingly, Wu et al. (1990) did not take the nding
of the exponent 1.4 in the temperature power spectra as indicative of the existence of BO59
scaling; rather they seem to regard this as a mere coincidence. Instead, their main message is that
a kind of universality may exist in the cascades of turbulent uctuations in the dissipative region
of thermal turbulence. Wu et al. (1990) suggested that this may represent a generalization of
Kolmogorovs picture of turbulence; i.e., a lack of universality in the original Kolmogorov spirit
may be restored by a proper multifractal transformation, which is referred to as a multifractal
universality (Frisch 1995). In their experiment, Wu et al. measured temperature at the cell center,
where the convective ow may be regarded as homogeneous but the mean ow is essentially zero,
and the authors themselves point out that Taylors frozen-ow hypothesis is not believed to be
applicable. We shall take the same view when discussing below the other experiments conducted
under similar circumstances.
Whereas Wu et al. (1990) focused primarily on whether the temperature power spectra has
a universal shape, Chilla et al. (1993) studied both the shape and the scaling properties of the
power spectra and SFs of the temperature eld. Their experiment was carried out in a rectangular
cell using water as the convecting uid, with Pr 4 and Ra between 106 and 4 108 . The 2D
temperature eld was measured using a laser sweeping technique in which the deection of the
laser beam is proportional to the local temperature gradient averaged along the beam path (Rubio
et al. 1989). The temperature was then reconstructed by integrating the measured temperature
gradient. Because the measured temperature gradient was integrated along the optical path of the
laser beam, this technique required that the ow eld be mainly 2D. These authors also used a
thermocouple placed inside the uid to measure the local temperature. Their results show that
wave-number power spectra of the temperature from the laser sweeping method and the frequency
power spectra measured by the thermocouple have similar shape. In fact these two types of spectra
largely overlap when normalized properly. The authors suggest that this coincidence may be
taken as an indirect verication of Taylors frozen-ow hypothesis (actually a modied version of
it because it involved averaged quantities), although they also conceded that it did not constitute a
rigorous proof of the hypothesis. Using the spatially averaged temperature eld, they also studied
low-order SFs. In a later study, Cioni et al. (1995), employing ESS, found consistency of Chilla
et al.s (1993) results with BO59.
Several later studies focused primarily on the scaling of the power spectra. For example, in a
helium gas experiment, Niemela et al. (2000) reported observing both BO59- and K41-type scaling
in the frequency power spectrum obtained from measured temperature time series. However, as
seen from Figure 5a, for both regimes the scaling range is less than a decade, and it is not known
whether the power spectrum remains the same shape at the high end of Ra that was reached in
The transition at Ra 7 1010 , which Wu et al. (1990) and later Procaccia et al. (1991) found in various quantities
including the temperature spectra themselveshas consistently been interpreted to be caused by temperature-uctuation
averaging effects around the bolometer (Grossmann & Lohse 1993). Although already Siggia (1994) favors Grossmann &
Lohse (1993)s interpretation, it has not been strictly proven. Due to space limitations, we do not further discuss this issue in
this review.
345
ANRV400-FL42-15
ARI

14:9
Power 7/5
102
Ra = 6
Slope = 1.4
1011
100
Power 5/3
5/3
0.01
102
102
103
104
104
105
7/5
0.01
101
2P()
Compensated power spectrum
P()/P(P)
Temperature power spectrum
13 November 2009
106
0.10
(Hz)
1.00
106
10.00
108
0.10
1.00
Frequency (Hz)
10.00
0.01
0.01
0.10
(Hz)
1.00
0.10
10.00
1.00
10.00
/P
Figure 5
(a) Frequency power spectrum of temperature measured in the center of an = 0.5 cylinder lled with helium (Pr = 0.7) at Ra =
6 1011 . (Inset) The K41- and BO59-compensated spectra. Figure taken from Niemela et al. 2000. (b) Frequency power spectra of
temperature measured in water (Pr = 4.3) with Ra from 2 109 to 1.3 1011 . Measurements were made at the center of cylindrical
cells of = 1 and 0.5. Spectra are presented after a simple scaling by the peak frequency fp [and P( fp )] of the temperature dissipation
spectra f 2 P( f ); an example of the latter is shown in the inset of the gure. Figure adapted from Zhou & Xia 2001.
their experiment (1017 ); it is also unclear why the spectra that Wu et al. (1990) measured under
similar conditions (Ra, Pr, ) showed only a single scaling exponent. Skrbek et al. (2002), using the
same apparatus as Niemela et al. (2000) but with = 1, measured local temperature in helium at
Ra = 1.5 1011 near the sidewall. Their SFs obtained directly from the temperature time series
did not show a reasonable scaling range, so the authors resorted to the ESS method, which yielded
good scaling behavior. As the ESS can only give relative scaling exponents, it is not possible to
distinguish between BO59 and K41 scaling behaviors (at least for lower orders).
Ashkenazi & Steinberg (1999) conducted a convection experiment in a rectangular box of
aspect ratio 0.7, with sulfur hexauoride (SF6 ) gas as the convecting uid. By working near the
gas-liquid critical point of SF6 , one can vary the Rayleigh and Prandtl numbers of this system
over a wide range, Ra from 109 to 1014 and Pr from 1 to 300, although at the price of having
pronounced non-Boussinesq effects at the edge of this domain. Using the scattering produced by
the critical density uctuations, Ashkenazi & Steinberg (1999) measured the local velocity with
a modied LDV technique (Lukaschuk et al. 2001). Their measured frequency power spectra
of the vertical velocity exhibit power-law scaling, and the tted exponent averaged over various
values of Ra and Pr is 2.4, which is close to the BO59 value of 11/5 predicted for the wavenumber spectra. They also measured the local temperature uctuations at both the cell center
and near the sidewall by placing small thermistors in the uid, observing an interesting feature
in the measured temperature frequency power spectra. Even for Ra up to 5 1012 and Pr less
than 50, the measured spectra did not show a noticeable scaling range. But for Pr from 90 to
300 and Ra from 8 1013 to 3 1014 , the measured spectra exhibited about one decade of
scaling with a tted exponent of 1.45. Another interesting feature of Ashkenazi & Steinbergs
(1999) experiment is their construction of the velocity and temperature cross-spectra using the
measured time series of the two quantities at the same location (although it is not clear how far
the temperature probe is separated from the LDV measuring volume). They reported a scaling
exponent of 1.85 for the obtained cross-spectrum, which is close to the theoretical value 9/5
predicted for the BO59 scaling, i.e., the Fourier transform of Equation 5.
346
Lohse
Xia

ANRV400-FL42-15
ARI
13 November 2009
14:9
In a convection experiment in water, Zhou & Xia (2001) measured local temperature uctuations in cylindrical cells of aspect ratio 1 and 0.5, with Ra varying from 4 108 to 1 1011
while maintaining Pr 4. Most of their measurements were made at the cell center, but for one
Ra (1.85 1010 ), they scanned the position from the bottom plate to the center. Both the power
spectra and SFs were investigated from the measured temperature time series. Extending an idea
rst introduced in the wave-number domain by She & Jackson (1993), that the peak wave-number
kp of the dissipation spectrum k2 E(k) can be used as a characteristic scale to collapse the energy
spectrum E(k), Zhou & Xia (2001) found that a simple scaling of the form P ( f )/P ( f p ) versus
f /fp could collapse all measured frequency temperature power spectra P( f ), with fp the peak frequency of the temperature dissipation spectrum f 2 P( f ). As seen from Figure 5b, all spectra with
Ra > 1 1010 exhibit a well-dened scaling range with an exponent of 1.4 (the inset of the
gure shows an example of the dissipation spectrum). In contrast to the method used by Wu
et al. (1990), this simple scaling does not need any adjustable phenomenological parameters. In
addition, Zhou & Xia did not observe a change of the spectra shape around Ra 7 1010 as found
by Wu et al. (1990). The simple scaling using the peak frequency of the dissipation spectra as the
characteristic scale to collapse frequency power spectra was later found to be also applicable to
velocity. Using LDV, Shang & Xia (2001) measured the vertical velocity in a cylindrical cell lled
with water. They found that the simple scaling method works well for velocity frequency power
spectra measured near the sidewall over the range of Ra from 108 to 1010 .
Whereas most of the temperature measurements in the time domain made at Pr 0.7 show
scaling behavior with an exponent close to the expected BO59 value, the behavior in low-Pr liquid
metal appears to be different. This is in coincidence with the theoretical expectation from Equation 10: Smaller Pr should result in larger LB and thus a more pronounced K41 scaling regime
and a less pronounced BO59 scaling regime. Cioni et al. (1995) made a comparative study of local
temperature uctuations in both water (Pr 7) and mercury (Pr = 0.025). They used a cylindrical
cell of aspect ratio one, and made the measurements above the center of the bottom plate at a
distance of one-quarter of the cell height. For water, the temperature frequency power spectrum
has a slope of 1.4 (Ra 1010 ). For mercury, a rather short and curved scaling range (which
actually straddles the Bolgiano frequency f B = U/L B ) of the spectrum (at Ra 109 ) may be tted
with an exponent 1.63, which is close to the Kolmogorov scaling. Using Taylors hypothesis,
Cioni et al. (1995) then studied the corresponding temperature spatial SFs, which showed no
scaling behavior. When using the ESS analysis, they found that relative scaling exponents are
consistent with those found for passive scalars in homogeneous and isotropic turbulence. Based
on the temperature time-series measurements from both water and mercury, Cioni et al. (1995)
concluded that temperature in bulk RB convection behaves as a passive scalar for the low-Pr case
with power-law exponents close to the K41 value, whereas for Pr 1, the exponents are close
to the BO59 value. They further attributed this to the difference in the Bolgiano scale for the
two cases, as, according to Equation 10, at least the global LB depends on Pr. This conclusion
appears to be supported by Takeshita et al. (1996), who studied mercury convection over the range
106 < Ra < 108 . Their measured temperature frequency power spectra are similar to those found
by Cioni et al. (1995), i.e., a limited scaling range with fB straddled in the middle. To shed light
on the force balances in the bulk, Takeshita et al. (1996) estimated, based on the measured temperature uctuations and velocity, that the buoyancy force is only approximately 1% of the inertial
force in the central region of the cell interior. This again suggests that the temperature behaves
like a passive scalar in this parameter range. In a later experiment by the same group (Glazier et al.
1999), a much higher Ra (5 1010 ) is reached. Because of the very low Pr, the corresponding
Reynolds number is approximately 5 105 , much higher than those reached at high-Pr uids.
Indeed, the measured temperature frequency power spectrum shows a wider scaling range than
347
ARI
13 November 2009
14:9
previous cases, and the estimated fB is well above the noise cutoff frequency. But even in this case,
the tted exponent (1.47) still differs from the BO59 prediction.
Although extremely important for the BO59 scaling, the force balance relation gr ur ur3 /r
has not been directly tested experimentally, and only a few indirect tests exist. Using a generalized
ESS scheme, Ching (2000) analyzed the Chicago data measured in the cell center in terms of the
normalized temperature temporal SFs and found the existence of two scaling regimes separated
by the Bolgiano timescale B = L B /U. Using velocity data from a numerical study (Benzi et al.
1996) and the Chicago temperature data, Ching (2000) also tested the force balance (in the time
domain) and found that it does not hold. In a later study of simultaneous velocity and temperature
data measured in water (Shang et al. 2003), both in the time domain, Ching et al. (2004a) found
the existence of cross-scaling between the normalized velocity and temperature SFs measured
in the cell center; i.e., when plotting one against the other on log-log scale, a linear relation
emerges, whereas none exhibited scaling behavior on its own. However, as shown in Figure 6, the
cross-scaling exponent is different from that implied by the BO59 force balance, thus, again, not
supporting BO59 scaling. We stress that the above cross-scaling exists only for timescales above
B (indicated in Figure 6), which Ching et al. (2004a) found to be a buoyancy-relevant scale from
the cross-correlation function between the increments of the vertical velocity and of temperature;
i.e., a strong correlation exists between the two increments above B but not below it. Similar
properties have recently been found to exist also between the temperature and concentration
elds in turbulent convection (Zhou & Xia 2008). Thus, although the BO59 scaling itself is not
supported, there does exist a range of scales above which buoyancy is important, and that may be
termed the Bolgiano regime.
0.36
Cell center
Sidewall
Bottom plate
p=6
b
103
p=5
0.32
p=4
102
Su(p)
log10 (Su(3)/[Su(2)]3/2)

ANRV400-FL42-15
0.28
p=2
101
p=1
100
0.24
Slope
change
101
0.20
0.08
0.12
0.16
0.20
0.24
101
100
101
Su(3)
log10 (S(3/2)/[S(1)]3/2)
Figure 6
(3)
(2)
Extended self-similarity plot of (a) time structure functions (SFs) and (b) space SFs. (a) log(Su ( )/[Su ] 3/2 ( )) versus
(3/2)
(1)
log(S
( )/[S ] 3/2 ( )) measured at the cell center (red squares). The solid line (implying cross-scaling between the two quantities) is a
least-squares t to the data points in the Bolgiano regime (indicated by the arrow, representing B ) but is plotted for the whole range of
. Results for measurements taken near the sidewall (blue triangles) and the bottom plate ( purple diamonds) are also shown, but for which
the cross-scaling is less certain. The dotted line has slope 1, which is implied by the Bolgiano scaling. Figure adapted from Ching et al.
( p)
(3)
2004a. (b) log Su3 (r) versus log Su3 (r) from the numerical simulations of Kunnen et al. (2008) at Ra = 1.11 108 with p = 1, 2, 4, 5,
and 6, from bottom to top. The dotted line indicates the local LB estimate. At the dashed line, a slope change is found in all curves.
Figure adapted from Kunnen et al. 2008.
348
Lohse
Xia
ANRV400-FL42-15
ARI
13 November 2009
14:9
As mentioned above, because of the absence of a sufciently large mean ow, the conditions for
Taylors frozen-ow hypothesis are not satised in most parts of the convection cell. This in turn
raises serious questions regarding the interpretation of results from time-domain measurements.
However, as the direct real-space measurement techniques such as PIV cannot be used conveniently in uids such as gas and liquid metal, time-domain measurement will continue to play an
important role. It is therefore highly desirable to be able to connect time or frequency domain data
with spatial domain theoretical predictions in a meaningful way. The standard Taylor hypothesis
assumes that velocity time series can be converted into spatial series according to v(x = Ut) v(t),
where the mean ow velocity U should be much larger than the rms velocity. When this condition
is not met, the so-called local Taylor hypothesis may be used (e.g., see Tennekes & Lumley 1972).
Sun et al. (2006) recently applied one such method, introduced by Pinton & Labbe (1994), to turbulent RB convection. By assuming that small-scale structures in the inertial range are advected by
large eddies at the integral scale, Pinton & Labbe (1994) propose that the velocity v(x) at location
t
)d , where v(
) is a local running average of
x should be related to the velocity v(t) by x = 0 v(

1 +Tint /2
) = Tint
v(t)d
t.
v(t) over the integral timescale Tint , v(
Tint /2
To test this method, Pinton & Labbe (1994) used hot-wire data from swirling ows and found
that for both the velocity power spectrum and the SF, the corrected spatial domain data exhibit a
longer scaling range and an exponent closer to the predicted value (K41) than those obtained with
time-domain data. Recently, Sun et al. (2006) applied this method to turbulent convection. They
used velocity time series measured with the LDV technique by Shang & Xia (2001) and compared
it with their own PIV data measured under similar conditions. They used the large-eddy turnover
time of the system as Tint to convert v(t) into v(x). As shown in Figure 7, compared with the
SF calculated with the time domain, the space-domain SF is similar to that from the PIV data in
terms of both the slope and the scaling range. After correcting with the local Taylor method, the
space-domain data clearly exhibit a longer scaling range that moreover sets in near the expected
1.5
100
Exponents
101
Su(3)(r/ or /)

3.3. Converting Time-Domain Data into Space-Domain

Data Using Local Taylor Hypothesis
PIV data
1.0
PIV data
LDV data
Local Taylor
hypothesis
0.5
LDV
data
Local Taylor hypothesis
101
101
r/ or /
102
0.0
1
Figure 7
(a) Third-order longitudinal velocity structure functions (SFs) (based on the modulus) calculated from time-domain laser Doppler
velocimetry (LDV) data ( purple), LDV data converted into space-domain data using the local Taylor hypothesis (blue), and particle
image velocimetry (PIV) data (red ). Their slopes in the scaling range are 1.08, 1.16, and 1.14, respectively. The vertical velocities used
were measured near the sidewall of aspect-ratio-one cylindrical cells lled with water (Ra = 7 109 and Pr = 4.3). The spatial
resolution of the PIV measurement is approximately 1.6, which is not enough to resolve dissipative range scales, which presumably
leads to the too large PIV-measured structure function at small scales. (b) Exponents of the three types of SFs as in panel a, for orders 1
to 4; the symbol colors correspond to those in panel a. Data taken from Shang & Xia 2001 and Sun et al. 2006.
349
ARI
13 November 2009
14:9
scale r 10. The difference between the time-domain and space-domain SFs becomes larger at
larger-order p, and thus with increasing order of the SF, the correction becomes more important.
It is clear from these results that corrections to the time-domain data can be quite signicant even
in the presence of a relatively strong mean ow (the LDV data were measured near the sidewall
of the convection cell, where the large-scale ow mean velocity v = 11.7 mm s1 and the rms
velocity vrms = 6.5 mm s1 ). Clearly the local Taylor hypothesis can be used to effectively convert
time-domain velocity series into spatial series in situations in which the usual condition for the
conventional Taylor hypothesis is not satised; i.e., the mean ow either is absent or is comparable
to the rms velocity. However, without velocity data, this method cannot be applied to convert the
temperature time series, which is more readily available and can be more conveniently measured
in thermal convection. Therefore, it is highly desirable to directly study SF, which indeed has been
done in recent years thanks to the developments in PIV.

ANRV400-FL42-15
3.4. Space-Domain Measurements

To our knowledge, the rst spatial-domain measurements of velocity differences in turbulent thermal convection were made by Tong & Shen (1992). Using the technique of homodyne photoncorrelation spectroscopy, they measured velocity differences of particle pairs in a scattering volume
of linear dimension r in an aspect-ratio-one cylindrical cell lled with water. Their results show
that ur r 0.6 over about a decade of r and for 5 107 Ra 1010 . Thus, their results appear to
agree with the BO59 prediction for the rst-order velocity SF. However, they should be interpreted
with caution. The measured photon-correlation function records differences in the Doppler shifts
of all particle pairs in the scattering volume of size , with separations from zero up to ( varied
in the experiment and was taken as r). Therefore, the measured ur is essentially an integration of
velocity differences of particles with separations from 0 to r. Another space-domain measurement
of the velocity eld was made in mercury by Mashiko et al. (2004). Using the same apparatus as
Glazier et al. (1999), the authors measured the vertical velocity prole along the cylindrical axis
of the cell by the technique of ultrasonic velocimetry. Their wave-number energy spectrum at
Ra = 5 1010 (at the cell center) showed about a decade of power-law behavior with an exponent consistent with the BO59 value of 11/5. They also computed the velocity SFs and
found that the second-order SF compensated by r6/5 showed a very small scaling range consistent with BO59. We note that Mashiko et al. (2004) used a mode decomposition method to
separate the slow dynamics related to the mean ow and the fast dynamics related to the turbulent cascades and that in the calculation of their energy spectra and SFs, mean ow and several low-order modes were subtracted. It is thus clear that, although both Tong & Shen (1992)
and Mashiko et al. (2004) found some evidence for BO59-like scaling, no direct spatial velocity SFs have been measured. Additionally, no direct spatial temperature SFs have been measured in the studies discussed above. Obviously, both velocity and temperature should be in the
same class of scaling behavior (either K41, BO59, or shear-ow scaling), over the same range of
scales, and under the same conditions if we are to agree on an acceptable and convincing cascade
dynamics.
In an attempt to determine the cascades of both temperature and velocity variances in turbulent
thermal convection, Sun et al. (2006) made high-resolution multipoint measurements of both the
velocity and the temperature elds in water, using a cylindrical cell of aspect ratio one (Ra
1.0 1010 and Pr = 4.3). Using PIV and the multi-thermistor-probe technique, these authors
measured the 2D velocity eld and the temperature difference along the vertical direction, from
which they obtained the real-space longitudinal and transverse SFs for both the horizontal and
vertical velocity components and spatial temperature SFs, respectively. The spatial resolutions
350
Lohse
Xia
104
ARI
13 November 2009
105
Scaling range
S(p)(r)/rp/3
Su(p)(r)/rp/3
101
101
100
101
r (mm)
101
K41
SFs
c
p=1
BO59
SFs
101
102
100

100
Scaling range
103
102
100
104
103
101
14:9
Su(p), compensated
ANRV400-FL42-15
102
100
101
r (mm)
100
LB
101
r (mm)
102
Figure 8
(a,b) Compensated real-space velocity and temperature structure functions (SFs) of order 1 to 8 ( from bottom to top) measured at the
central region of an aspect-ratio-unity cylindrical cell lled with water (Ra 1 1010 and Pr = 4.3). Data inside and outside the
scaling range are represented by different colors, and the lines represent power-law ts to data in the inertial range. Figure taken
from Sun et al. 2006. (c) Compensated transverse SFs of the vertical velocity measured near the cell center in a cylindrical cell
(Ra = 1.1 109 , Pr = 6.37, and = 1). Blue circles represent BO59-compensated SFs, and lines represent K41-compensated SFs.
The order p increases from 1 to 4 from top to bottom for BO59. The red dotted line is the p = 1 SF with K41 compensation. Similarly,
lines with triangles, squares, and upside-down triangles represent p = 2, 3, and 4, respectively. The estimated local LB is indicated by
the purple dashed line. Figure taken from Kunnen et al. 2008.
in both the velocity and temperature measurements are comparable to the Kolmogorov scale
and are approximately 10 times smaller than the global Bolgiano scale, so the inertial-range
features are well-resolved. An important aspect of this study is that for all orders p = 1 to p = 8,
both the velocity and temperature SFs exhibit their respective scaling behavior over the same
range of scales. An example of the real-space velocity and temperature SFs (compensated by K41
predictions) measured in the central region of the convection cell is shown in Figure 8a,b. These
SFs show a high level of convergence even at order 8, as evidenced by their integration kernels
(Sun et al. 2006). The power-law ts to data are in the inertial range 10 r L, where the
Kolmogorov scale is 0.4 mm, and L 30 mm is the integral length obtained by integrating the
longitudinal velocity autocorrelation function. The global quantity-based Bolgiano scale in this
case is LB 5 mm. Thus the entire inertial range should be above LB . Nonetheless, these results
clearly show that the low-order SFs are closer to K41 behavior. Deviations become progressively
larger with increasing order, presumably because of intermittency effects, which can be described
by the hierarchy models of She & Leveque (1994) for velocity and of Ruiz-Chavarria et al. (1996)
for passive scalars. Figure 9a shows exponents of both longitudinal and transverse velocity SFs
and of temperature SFs, and their comparisons with various model predictions. The results show
convincingly that for both the temperature and the velocity elds, in the inertial range and for
orders 1 to 8 consistently, the cascades of the temperature and velocity uctuations in the cell center
do not obey the BO59 scaling. Instead, K41-type scalings with intermittency properly taken into
account can well describe the observed scaling behavior. Thus in the cell center, both velocity
and temperature exhibit the same scaling behavior that one would nd for the velocity and for a
passive scalar in homogeneous and isotropic Navier-Stokes turbulence. However, in the sidewall
region, Sun et al. (2006) found that, whereas the SFs of horizontal velocity are consistent with
those in the central region, the exponents for the vertical velocity and temperature are different
from those in the central region and do not follow BO59, K41, or She & Leveques (1994) model
predictions. By considering the co-action of buoyancy and inertial forces and using dimensional
351
ANRV400-FL42-15
ARI
Structure function
exponents
14:9
3.0
BO59 (velocity)
BO59 (temperature)
K41
SL94
RCBC96
pL,w
pT,u
p
2.5
2.0
1.5
2.0
9/5
1.5
3u
1
1.0
1.0
3/5
0.5
0.5
0.0

13 November 2009
0.0
0.08
0.16
0.25
z/L
0.33
0.42
0.50
Figure 9
(a) Exponents of structure functions (SFs) measured at the center of an aspect-ratio-unity cylindrical cell lled with water
(Ra 1 1010 and Pr = 4.3). Here pL,w is the exponent of the p-th-order longitudinal SF of the vertical velocity, pT,u is that of the
transverse SF of the horizontal velocity, and p is the exponent of the temperature SF. The pink and purple lines represent BO59
predictions for velocity and temperature, respectively; the red line represents K41 and OC (Obukhov-Corrsin) predictions for velocity
and passive scalar; and the gold and blue lines represent, respectively, values resulting from the hierarchical model of She & Leveque
(1994) for the velocity and of Ruiz-Chavarria et al. (1996) for the passive scalar, both with appropriately chosen parameters. Figure
adapted from Sun et al. 2006. (b) Scaling exponents 3u and 3 as a function of the distance z/L from the plate, at Ra = 3.5 107 and
Pr = 1, based on numerical simulations with periodic boundary conditions in the horizontal direction. The scaling exponents are
measured in an r regime where extended self-similarity holds. Figure adapted from Calzavarini et al. 2002.
( p)
analysis, Sun et al. (2006) found Su( p) r 2 p/5 and S r 3 p/10 , which agree with the measured
low-order SFs (as intermittency is not considered, the dimensional analysis is not expected to be
correct for higher-order SFs). In a later investigation of the velocity eld in the central region of
the same cell, Zhou et al. (2008) made a systematic test of the local homogeneity and isotropy
using a number of criteria. Their results show that homogeneity and isotropy hold to an excellent
degree in the central region of the RB cell, which suggests that the effect of buoyancy is negligible
on small-scale turbulence in the central region. They attribute this to the fact that in aspect-ratiounity cells, the large-scale circulation carries most of the thermal plumes along the perimeter of
the system, so there are few plumes passing through the central region.
Recently, Kunnen et al. (2008) also measured spatial velocity SFs based on the stereoscopic PIV
technique. The experiment was done in water (Pr = 6.37) in the regime 1.1108 Ra 1.1109 .
From their numerical result (discussed in Section 3.5), the local Bolgiano scale (Equation 11),
shown in Figure 3c, was known. In the center, for which the SFs are reported, and for
Ra = 1.1 108 , it is approximately 1/5 of the external length scale L, but it grows to approximately
L/3 at Ra = 1.1 109 . Between LB and L, the BO59 compensated second-order velocity SFs (i.e.,
the SFs divided by r6/5 ) indeed give a straight line (see Figure 8c), but the range is rather limited
and shrinks with rising Ra. Clearly, there is no extended BO59 scaling. Close to the plates, Kunnen
et al. (2008) reveal the shear scaling (Equation 15) in the second-order velocity SF. As pointed out
by Kunnen et al. (2008), their results and those of Sun et al. (2006) are in fact complementary to
each other, despite the seeming contradiction, as shown in Figure 8. The key properties here are
the local Bolgiano scale and the size of the measurement area. According to their numerical result,
the local Bolgiano scale in Sun et al.s (2006) case (Ra = 1.0 1010 ) equals LB 56 mm, rather
than 5 mm as estimated from global quantities. Sun et al.s (2006) measurement area is 40
40 mm2 ; thus K41-like behavior is expected for their measurement in the cell center. To verify
this experimentally, one would need to measure local energy and thermal dissipation rates (and
hence local LB ).
352
Lohse
Xia

ANRV400-FL42-15
ARI
13 November 2009
14:9
Kunnen et al. (2008) also applied ESS, plotting the p-th order velocity SF against Su(3) ; the
relative scaling exponent is called p , dened by Su( p) (r) (Su(3) (r)) p . Two regimes can be identied
(see Figure 6b): For r L B /2, the relative exponents p resemble the ones one would expect for
intermittency-corrected K41 scaling and that are well described by She & Leveques (1994) model
(e.g., 6 = 1.78). For r L B /2, the p are smaller and Ra-dependent, however, with an error
bar within the range of what has been found by Toschi et al. (1999) for the p in channel-ow
turbulence (e.g., 6 = 1.44).
Zhou et al. (2008) measured the velocity circulation SFs in the central region of the RB system. Their result shows that the circulation SFs also exhibit K41-like behavior, but they are
more sensitive to local anisotropy than the velocity eld itself. These results give further support to Sun et al.s (2006) ndings that in the central region of the RB system, K41-like rather
than BO59-like dynamics prevails. Putting together results from these two studies, we see the
emergence of a consistent picture regarding the cascades of turbulent velocity and temperature
uctuations; i.e., for the directly measured SFs, no BO59-like behavior is observed in turbulent RB
convection.
3.5. Structure Functions and Spectra From Numerical Simulations

The tremendous advantage of numerical simulations of RB ow is that in principle all data are
available. However, due to computational costs, only limited Ra and Pr can be achieved, and only
in the past 15 years or so have turbulent 3D RB simulations become possible. For a more detailed
discussion, we refer the reader to the review by Ahlers et al. (2009).
Kerr (1996) provides one of the rst numerical simulations for really turbulent RB convection,
achieving Ra = 2 107 at Pr = 0.7 with periodic boundary conditions at the sidewalls. He nds
about one decade of K41-type scaling Eu (k) k5/3 and less steep scaling for the temperature
spectrum, consistent with E (k) k1 , both inconsistent with the BO59 picture.
Calzavarini et al. (2002) provide a numerical simulation in a similar Ra and Pr regime with a
Lattice-Boltzmann scheme. The SFs do not scale. However, some scaling regimes can be identied
in ESS plots of the SFs against each other. When measuring the local scaling exponents u( p) and
( p)
in the respective SFs for that ESS scaling regime as a function of the distance z from the
plates, they obtain values consistent with K41 scaling in the bulk, where L B (z) L, and values
consistent with BO59 scaling close to the plates, where L B (z) 0.2L (see Figure 9b for the local
scaling exponents and Figure 3c for the corresponding local Bolgiano scale).
Camussi & Verzicco (1998) provide temperature and velocity frequency spectra for RB ow
in mercury at = 1/2 and Pr = 0.022 and for 5 104 Ra 106 , based on their numerical
simulations. To allow for a one-to-one comparison, they calculated the spectra from velocity
and temperature time series taken at the cell center and close to the cell wall. No indication
of BO59 scaling was reported. Moreover, Camussi & Verzicco (2004) employed numerical time
series to calculate spectra and SFs Su ( ) and S ( ), but now for Pr = 0.7 and for Ra up to 2
1011 . The temperature frequency spectrum showed a power-law exponent close to 5/3 (K41)
in the center and close to 7/5 (BO59) at mid-height close to the sidewall, where S ( ) showed
the corresponding slope 2/5. Depending on the position and velocity component, the velocity
frequency spectrum showed either a slope close to 5/3, or even a less steep slope (smaller
modulus of slope)in fact similar to the value 7/5 of the temperature frequency spectrumbut
clearly no indication of BO59 scaling. Rincon (2006) performed numerical simulations at Ra =
106 , Pr = 1, and = 5, employing the SO(3) analysis to properly treat isotropic and anisotropic
projections of the SF tensor. However, the Rayleigh number in these simulations is too small to
reveal any scaling, and the focus of the paper is on energy balances resulting from the Boussinesq
353
ARI
13 November 2009
14:9
equations. The paper again stresses the importance of disentangling anisotropy, inhomogeneity,
and buoyancy effects.
The hitherto most complete numerical calculations of real-space velocity and temperature
SFs were performed by Kunnen et al. (2008). For the velocity SFs in the center of the cell, they
found similar results as those obtained from their stereoscopical PIV measurements, including the
relative ESS scaling exponents. In particular, with increasing Ra, the BO59 regime becomes even
less limited, which is consistent with the trend in the local Bolgiano length. In the small regime
beyond r > LB (less than half a decade even for the best case Ra = 108 ), the BO59-compensated
temperature spectra are more horizontal than the K41-compensated ones, although both show
an unexplained increase at the very large scales, r L. With increasing Ra, this regime becomes
even smaller.
All the numerical simulations reported above in this subsection are for the ow in real cells with
no-slip boundary conditions, at least at the top and bottom walls, leading to inhomogeneities in the
ow. As discussed in Section 2.3, the inhomogeneities can be avoided in the so-called homogeneous
RB ow. Biferale et al. (2003) applied the SO(3) decomposition for this type of ow. They not
only found K41 scaling (with the standard intermittency corrections) in the isotropic j = 0 sector,
but they also found no evidence for BO59 scaling in the subleading, anisotropic j = 1 sector:
The BO59 dimensional scaling relation (Equation 19) for j = 1 clearly is not fullled. Instead,
the anisotropic uctuations in the RB system are anomalous and universal, with similar relative
scaling exponents as in random Kolmogorov ow or in other types of shear ows (Toschi et al.
1999).

ANRV400-FL42-15
4. SMALL-SCALE MIXING OF ACTIVE AND PASSIVE SCALARS

In Grossmann & Lohses (2000, 2001, 2002, 2004) theory, the kinetic energy and thermal dissipation rates have been decomposed into boundary layer and bulk contributions, and a laminar
Prandtl-Blasius boundary layer, in a time-averaged sense, has been assumed. This theory can successfully describe and predict Nu and Res dependency on Ra and Pr (e.g., see the recent review
in Ahlers et al. 2009). In a recent high-resolution measurement of the properties of the velocity
boundary layer, Sun et al. (2008) have found that, despite the intermittent emission of plumes, the
Blasius-type laminar boundary condition is indeed a good approximation, in a time-averaged sense,
both in terms of its scaling and its various dynamic properties. In the Prandtl-Blasius boundarylayer theory, temperature is treated as a passive scalar, but because of buoyancy, in the inertial range
and above the Bolgiano scale, temperature behaves as an active scalar. One naturally wonders what
would be the properties of the temperature eld at these small scales. Specically, how does the
temperature as an active scalar compare with a passive scalar under the same conditions?
4.1. Small-Scale Anisotropy of the Temperature Field

The persistence of nonzero skewness of the scalar derivatives is a hallmark of passive scalar mixing,
as it invalidates the local isotropy assumption of the classical Kolmogorov-Obukhov-Corrsin
theory of passive scalars (Sreenivasan 1991). The nonzero skewness is believed to result from sharp
fronts or gradients of the scalar. By examining the skewness S S() ( )3 /( )2 3/2
of the temperature and the skewness St S(t ) of its time derivative in thermal convection,
Belmonte & Libchaber (1996) studied the small-scale mixing properties of the temperature eld,
in particular the question of active versus passive. Their experiment was conducted in a cubic
cell of aspect ratio one lled with pressurized gas (helium, nitrogen, or sulfur hexauoride) at
room temperature and spanned a range of Ra from 2 107 to 1 1011 with Pr = 0.7. Sharp
354
Lohse
Xia
ARI
13 November 2009
14:9
Active scalar
<>
Passive scalar
S > 0
St > 0
S > 0
St < 0
S(t+), S(t)
ANRV400-FL42-15
6
4
+
S < 0
St < 0
S < 0
St > 0
f
S(t+), S(t)
<>

Ra = 5.4 108
Ra = 1.1 1010
Time
Time
0.0
0.1
0.2
z/L
0.3
0.4
0.5
Figure 10
(ad ) A sketch of four scenarios of temperature asymmetries (relative to the mean), according to the signs of their skewness S and the
skewness St of their time derivative. Panels a and b correspond to an active scalar for which S St > 0, and panels c and d correspond
to a passive scalar for which S St < 0. Figure adapted from Belmonte & Libchaber 1996. (e, f ) Skewness of the plus (red circles) and
minus (blue triangles) temperature derivative as functions of the scaled distance z/L for two values of Ra. The width of the mixing zone is
characterized by the gap in the z direction within which S(t + ) and S(t ) differ from each other. Figure taken from Zhou & Xia
2002.
fronts in passive scalars manifest themselves as ramp-cliff (slow rise and fast decay) structures
in the time series, which would produce positive S and negative St when the passive scalar is
temperature measured near a mildly heated plate (Sreenivasan & Antonia 1977). Extending this
to active scalars, Belmonte & Libchaber (1996) argued that, when measured near the hot plate in
thermal convection, temperature time series would possess cliff-ramp structures that would give
rise to positive S and St . These cliff-ramp structures are believed to be associated with plumes.
Based on this, Belmonte & Libchaber (1996) classied four different scenarios according to the
signs of S and St . As shown in Figure 10, passive and active scalars produced at the hot and
cold plates (corresponding to hot and cold plumes for the active case) can be identied by the
sign of the product of S St . Belmonte & Libchabers (1996) measured prole St (z) from the
cold plate is mostly negative outside the boundary layer, so they concluded that the temperature
is active. Moreover, they argued that the region over which St remains negative corresponds to
the mixing zone but did not quantify this further. In a recent study of temperature uctuations
measured inside the conducting plates of the convection cell, Sun & Xia (2007) found that indeed
the product of S St is positive in both the cold and hot plates and can be described by the powerlaw Ra0.22 over the range 5 107 1 1011 , which may be taken as a quantitative measure of the
increased strength of plume eruptions.
Emran & Schumacher (2008) analyzed the skewness of the temperature and its spatial derivative
(in vertical direction) numerically, up to Ra = 109 , always nding non-Gaussian probability density
functions (PDFs) of the temperature statistics. In contrast, the statistics for the spatial derivative of
the temperature is similar to that of a passive scalar. In the bulk, Sz decreases with increasing Ra.
Moreover, proles S (z) are provided, but none for St (z), so that a classication of the regimes
according to Belmonte & Libchaber (1996) is not possible.
355
ARI
13 November 2009
14:9
In a somewhat different approach, Zhou & Xia (2002) studied the skewness of the so-called plus
and minus temperature increments = (| | )/2, where = (t + )(t) is the temperature
increment over a time interval . [Plus and minus increments were rst used in the study of ramp
structures in the velocity eld in Navier-Stokes turbulence (Vainshtein & Sreenivasan 1994) and
later in the context of the temperature eld by Aivalis et al. (2004).] When a (hot) plume passes
through a temperature probe, it would produce a spike with a steep rising edge (due to its cap) and
a relatively gentle falling edge (due to its tail) in the measured time series. The plus and minus
temperature increments thus capture, respectively, the slope of the rising and falling edges for
smaller than the size of a plume (in the time domain). Therefore, in regions where hot plumes
dominate over cold ones, + would be expected to be larger than , and vice versa in regions
where cold plumes dominate. Indeed, Zhou & Xia found that for < B = L B /U, the skewness
S(+ ) > S( ) in regions where hot plumes dominate. This also suggests that the cap thickness (in
time domain) of a typical plume should be smaller than B but larger than . For > B , features
smaller than the corresponding scale are averaged out, resulting in the same values for the plus
and minus skewness. Zhou & Xia (2002) also found that S(t + ) and S(t ) are the same in the
boundary layer and central regions of the cell but differ in regions where plumes dominate, which
is shown in the right panel of Figure 10 for two values of Ra. The width (in the z direction) of
the region in which the two quantities differ shrinks as Ra increases, which in fact has a scaling
exponent as that predicted for the width of the mixing zone (Castaing et al. 1989). The mixing
zone plays an important role in the Chicago model for hard turbulence (Castaing et al. 1989),
and the plus and minus skewness provides a useful denition for its quantitative characterization.
Recently, it has been found that quantities such as the prole of vorticity uctuations (Zhou et al.
2007) and skewness of spatial temperature gradient and local thermal dissipation rate (He & Tong
2009) can all be used to characterize the mixing zone.

ANRV400-FL42-15
4.2. Comparative Studies of Active and Passive Scalars

A fascinating property of passive scalars is the saturation of the scaling exponent of their high
order SFs; i.e., the exponent p approaches a constant
for large enough order p. The saturation
is believed to be related to the ramp-cliff structures (or fronts) in the scalar (Celani et al. 2000,
Moisy et al. 2001). With the temperature time series in thermal convection found to exhibit clifframp structures (Belmonte & Libchaber 1996, Zhou & Xia 2002), it is natural to ask whether the
temperature eld as an active scalar would possess the saturation property. In a 2D numerical study
of thermal convection, Celani et al. (2001) found that the temperature indeed exhibits saturation
phenomenon. Experimentally, Zhou & Xia (2002) measured long time records (consisting of 1.2
108 data points, corresponding to 37,000 large-eddy turnover times) of local temperature both near
the sidewall region and at the center in an aspect-ratio-one cylindrical cell lled with water (Ra =
1.8 1010 , Pr = 4). As the high-order moments are dominated by large temperature excursions,
the saturation of the SF exponent is equivalent to the scaled PDF r P (r ) becoming independent
of r for inertial range values of r rms . Because high-order SFs are very sensitive to the far tails
of the PDF, collapsing of the scaled PDF is a statistically more reliable test of the saturation (Celani
et al. 2000, 2001). By examining both the scaled PDF P ( ) of temperature increments over

varying timescales in the inertial range and the cumulated probability P ( )d , Zhou & Xia
found evidence for the saturation of the temperature SFs near the sidewalls, but not at the cell
center. Because plumes are abundant at the sidewall but scarce in the center, this nding suggests
that the saturation is caused by the presence of plumes. As mentioned in Section 3.2, Skrbek et al.
(2002) analyzed temperature time series measured in helium (Ra = 1.5 1011 ), also near the
sidewall, using the ESS method. They nd no saturation of the ESS (relative) scaling exponents
356
Lohse
Xia
ANRV400-FL42-15
ARI
13 November 2009
14:9
102
Temperature
Concentration
101
p=4
100
P(C/Crms)
Exponents
0.8
0.6
0.4
p=2
0.2
102
101
P(/rms)
1.0
101
102
103
0.1
0.3
0.5
0.7
101
102
103
104
104
0.0
100
105
10
C/Crms
10
10
/rms
10

Figure 11
(a) Local scaling exponents of temperature (circles) and concentration (triangles) structure functions for orders 2 and 4, from 2D
simulations made at Ra = 107 . Figure taken from Celani et al. 2002. (b,c) Probability density functions of concentration and
temperature increments normalized by their standard deviations for four time intervals ( from top to bottom): / B = 0.05, 0.15, 1, and
10. The curves have been shifted vertically for clarity, and the gray dashed curve indicates a Gaussian distribution of variance 1.
Measurements were made near the sidewall, at mid-height, of a rectangular cell lled with water (Ra = 1.2 1010 and Pr = 5.3).
Figure taken from Zhou & Xia 2008.
p with increasing order, up to p = 16. However, from that result it cannot be concluded that the
scaling exponents p of the temperature SFs do not saturate either.
A fruitful way to investigate the temperature eld with respect to the question of whether it
behaves more as an active or more as a passive scalar in thermal convection would be to make
a direct comparative study with a real passive scalar (such as the concentration of a dispersed
dye) under the same conditions. Such an approach has been taken mostly in numerical studies.
Celani et al. (2002) investigated the scaling properties of both the temperature and a passive scalar
governed by the Boussinesq equations in a 2D numerical study and found that the two scalar elds
have the same even-order SF exponents, as shown in Figure 11. They conjectured that this may
be attributed to the existence of so-called statistically preserved structures in the temperature eld,
as is the case for passive scalars. In a shell model study of thermal convection, Ching et al. (2002,
2003a) found numerical evidence that the equivalence of even-order statistics of active and passive
scalars can indeed be understood from the presence of statistically preserved structures.
On the experimental side, there exist only a few comparative studies of active and passive
scalars advected by the same ow. Using liquid crystal imaging and planar laser-induced uorescence techniques, Gluckman et al. (1993) studied the geometric properties of isothermal and
isoconcentration contours in turbulent convection in water (Ra 107 108 ). They found that
the isoconcentration contours show a limited fractal scaling range. In contrast, the isotherms do
not exhibit fractal scaling, which does not support Procaccia et al.s (1991) suggestion that the
temperature eld appears fractal above a certain scale. Conversely, the local curvature of both the
isoconcentration and isothermal contours possesses a stretched exponential PDF with asymmetric
tails. By injecting a uorescent dye into a rectangular cell lled with water, Zhou & Xia (2008)
made simultaneous and separate measurements of the local temperature, using a small thermistor,
and dye-concentration uctuations, using laser-induced uorescence, in close proximity (near the
sidewall at mid-height). They showed that although the two scalars are advected by the same
turbulent ow, the quantities themselves and their small-scale increments distribute differently.
Below B , the temperature behaves as a passive scalar, and above B it behaves as an active scalar,
where it also possesses a higher level of intermittency as compared with the passive scalar (dye concentration). Figure 11b,c plots the PDF of concentration and temperature increments over several
time intervals, respectively. The PDFs of temperature increments change from non-Gaussian in
small scales to near Gaussian in large scales, whereas those of concentration increments remain
357
ARI
13 November 2009
14:9
non-Gaussian for all scales investigated; i.e., as changes from small to large scales, the shapes
of the concentration-increment PDFs are more similar among themselves than those of the temperature, and hence the passive scalar is less intermittent than the active one. The nding that
an active scalar is more intermittent than a passive one is consistent with the saturation of the SF
exponent for the temperature. The saturation exponent

is found to be 0.8 for an active scalar
from experiment (Zhou & Xia 2002) and 2D simulations (Celani et al. 2001) and 1.45 for a passive
scalar measured in a swirling ow (Moisy et al. 2001). This suggests that

deviates more from
the dimensional predictions for active scalars than passive ones, implying that the active scalar
is more intermittent. As a possible future experiment, the saturation of both active and passive
scalars advected by the same ow should be measured.
As a closing remark, we point out that many of the studies reviewed in this section are made
in the time domain; therefore, they also face the question regarding the validity of the Taylor
hypothesis, and one should always bear this in mind when interpreting the results and connecting
them to theoretical predictions.

ANRV400-FL42-15
5. CONCLUSIONS AND OUTLOOK

The discussions in this review reect that the issue regarding the nature of cascades of the velocity
and temperature uctuations in turbulent thermal convection is not fully settled. On one hand,
both temperature and temperature gradient PDFs and their skewnesses and force balances on
large scales carry the feature of an active scalar, pronouncedly different from those of a passive
scalar like a concentration eld. On the other hand, BO59 scaling has not been found over a large
range of scales and many hints point toward the sketch of Figure 2 being unrealistic. According
to Kunnen et al. (2008), as the Ra dependency of the local Bolgiano scale LB (x) is an increasing
function of Raat least in the center of the cell, where the effect of anisotropy and shear is
smallestfuture attempts to nd BO59 scaling should be made with lower Ra and higher Pr to
increase the potential BO59 range, if it exists at all. This, of course, strongly limits the possible
inertial scaling regime, again revealing the conict of requirements for extended BO59 scaling:
This fundamental obstaclethe lack of separation of length scales between L and LB appears to
place conicting demands on the control parameters of the system, and an extended BO59 scaling
regime therefore may never be achieved.
Ultimately, a true test of the Bolgiano argument should consist of observing K41-like scaling,
for both velocity and temperature, below LB . In contrast, above LB , one should observe BO59like scaling, again for both velocity and temperature, just as sketched in Figure 2. Because the
Kolmogorov scale and the local Bolgiano scale have opposite dependency on Ra, this means
that simultaneous observation of K41 and BO59 scalings at the same location is unlikely, if not
impossible. In light of this, several different approaches may yield fruitful results. The rst one
is to select an appropriate range of Ra and Pr, and suitable system size and aspect ratio, and look
for different scalings in different regions of the convection cell; for example, BO59-like scaling
near the boundary layer and K41-like scaling in the cell center. The second approach is to look
for different scalings in the same location of the cell by tuning the local LB to be either larger
or smaller than the experimentally accessible scale, which could be achieved by using different
working uids, for example. Attempts along the line of the rst approach have been made by
Sun et al. (2006) and Kunnen et al. (2008), whereas the second type of approach has not been
attempted in a systematic way. A third approach may be to use cells of wider aspect ratio, 1,
to allow for a large length scale at least in the horizontal direction and for a possible separation
of scales between LB and L. For any of these efforts, it will in addition be helpful to employ
the SO(3) decomposition of the ow to better disentangle scaling corrections due to shear and
358
Lohse
Xia
ANRV400-FL42-15
ARI
13 November 2009
14:9
scaling corrections due to buoyancy. Also for future studies, a quantitative test of the force balance
relation gr ur ur3 /r in direct spatial measurements of the velocity and temperature elds will
need to be madeas done in Calzavarini et al. (2002) for numerical simulations (see Figure 3d).
As pointed out in Section 1, the RB system may be ideal to study turbulence as it is experimentally accessible and as the governing equations and boundary conditions are exactly known.
However, thanks to the richness of its dynamics, it clearly is not the easiest system to study clean
and extended scaling regimes of SFs over a large range of scales.

DISCLOSURE STATEMENT
The authors are not aware of any afliations, memberships, funding, or nancial holdings that
might be perceived as affecting the objectivity of this review.
ACKNOWLEDGMENTS
We thank all our colleagues and collaborators on this subject for various discussions over the past
20 years. We have learned tremendously from them and enjoy and appreciate the joint research
efforts to better understand turbulent thermal convection. D.L. acknowledges FOM for the continuous support and K.Q.X. the Hong Kong Research Grants Council (grant nos. CUHK 403806
and 403807).
LITERATURE CITED
Adrian RJ. 1991. Particle-imaging techniques for experimental uid mechanics. Annu. Rev. Fluid Mech. 23:261
304
Ahlers G, Grossmann S, Lohse D. 2009. Heat transfer and large scale dynamics in turbulent Rayleigh-Benard
convection. Rev. Mod. Phys. 81:50337
Aivalis KG, Schumacher J, Sreenivasan KR. 2004. Sign-symmetry of temperature structure functions. Phys.
Rev. E 69:066315
Arad I, Biferale L, Mazzitelli I, Procaccia I. 1999a. Disentangling scaling properties in anisotropic and inhomogeneous turbulence. Phys. Rev. Lett. 82:504043
Arad I, Dhruva B, Kurien S, Lvov VS, Procaccia I, Sreenivasan KR. 1998. Extraction of anisotropic contributions in turbulent ows. Phys. Rev. Lett. 81:533033
Arad I, Lvov V, Procaccia I. 1999b. Correlation functions in isotropic and anisotropic turbulence: the role of
the symmetry group. Phys. Rev. E 61:675365
Arneodo A, Baudet C, Belin F, Benzi R, Castaing B, et al. 1996. Structure functions in turbulence, in various
ow congurations, at Reynolds number between 30 and 5000, using extended self-similarity. Europhys.
Lett. 34:41116
Ashkenazi S, Steinberg V. 1999. Spectra and statistics of velocity and temperature uctuations in turbulent
convection. Phys. Rev. Lett. 83:476063
Batchelor GK. 1953. The Theory of Homogeneous Turbulence. Cambridge, UK: Cambridge Univ. Press
Belmonte A, Libchaber A. 1996. Thermal signature of plumes in turbulent convection: the skewness of the
derivative. Phys. Rev. E 53:489398
Benzi R, Biferale L, Ciliberto S, Struglia MV, Tripiccione R. 1996. Generalized scaling in fully developed
turbulence. Phys. D 96:16281
Benzi R, Ciliberto S, Tripiccione R, Baudet C, Massaioli F, Succi S. 1993. Extended self-similarity in turbulent
ows. Phys. Rev. E 48:R2932
Benzi R, Tripiccione R, Massaioli F, Succi S, Ciliberto S. 1994. On the scaling of the velocity and temperature
structure functions in Rayleigh-Benard convection. Europhys. Lett. 25:34146
Benzi R, Toschi F, Tripiccione R. 1998. On the heat transfer in the Rayleigh-Benard system. J. Stat. Phys.
93:90118
359
ARI
13 November 2009
14:9
Berschadskii A, Niemela JJ, Praskovsky A, Sreenivasan KR. 2004. Clusterization and intermittency of temperature uctuations in turbulent convection. Phys. Rev. E 69:056314
Biferale L. 2003. Shell models of energy cascade in turbulence. Annu. Rev. Fluid Mech. 35:44168
Biferale L, Calzavarini E, Toschi F, Tripiccione R. 2003. Universality of anisotropic uctuations from numerical simulations of turbulent ows. Europhys. Lett. 64:46167
Biferale L, Daumont I, Lanotte A, Toschi F. 2002a. Anomalous and dimensional scaling in anisotorpic turbulence. Phys. Rev. E 66:056306
Biferale L, Lohse D, Mazzitelli I, Toschi F. 2002b. Probing structures in channel ow through SO(3) and
SO(2) decomposition. J. Fluid Mech. 452:3959
Biferale L, Procaccia I. 2005. Anisotropy in turbulent ows and in turbulent transport. Phys. Rep. 414:43164
Biferale L, Toschi F. 2001. Anisotropies in homogeneous turbulence: hierarchy of scaling exponents and
intermittency of the anisotropic sectors. Phys. Rev. Lett. 86:483134
Bolgiano R. 1959. Turbulent spectra in a stably stratied atmosphere. J. Geophys. Res. 64:222629
Brandenburg A. 1992. Energy spectra in a model for convective turbulence. Phys. Rev. Lett. 69:6058
Calzavarini E, Doering CR, Gibbon JD, Lohse D, Tanabe A, Toschi F. 2006. Exponentially growing solutions
of homogeneous Rayleigh-Benard ow. Phys. Rev. E 73:R035301
Calzavarini E, Lohse D, Toschi F, Tripiccione R. 2005. Rayleigh and Prandtl number scaling in the bulk of
Rayleigh-Benard turbulence. Phys. Fluids 17:055107
Calzavarini E, Toschi F, Tripiccione R. 2002. Evidences of Bolgiano-Obhukhov scaling in three-dimensional
Rayleigh-Benard convection. Phys. Rev. E 66:016304
Camussi R, Verzicco R. 1998. Convective turbulence in mercury: scaling laws and spectra. Phys. Fluids 10:516
27
Camussi R, Verzicco R. 2004. Temporal statistics in high Rayleigh number convective turbulence. Eur. J.
Mech. B Fluids 23:42742
Cardin P, Olson P. 1994. Chaotic thermal convection in a rapidly rotating spherical shell: consequences for
ow in the outer core. Phys. Earth Planet. Inter. 82:23559
Castaing B, Gunaratne G, Heslot F, Kadanoff L, Libchaber A, et al. 1989. Scaling of hard thermal turbulence
in Rayleigh-Benard convection. J. Fluid Mech. 204:130
Celani A, Lanotte A, Mazzino A, Vergassola M. 2000. Universality and saturation of intermittency in passive
scalar turbulence. Phys. Rev. Lett. 84:238588
Celani A, Matsumoto T, Mazzino A, Vergassola M. 2002. Scaling and universality in turbulent convection.
Phys. Rev. Lett. 88:054503
Celani A, Mazzino A, Vergassola M. 2001. Thermal plume turbulence. Phys. Fluids 13:213335
Chilla F, Ciliberto S, Innocenti C, Pampaloni E. 1993. Boundary layer and scaling properties in turbulent
thermal convection. Il Nuovo Cim. D 15:122949
Ching ESC. 2000. Intermittency of temperature eld in turbulent convection. Phys. Rev. E 61:R3336
Ching ESC, Cheng WC. 2008. Anomalous scaling and rened similarity of an active scalar in a shell model
of homogeneous turbulent convection. Phys. Rev. E 77:015303
Ching ESC, Chui KW, Shang XD, Qiu XL, Tong P, Xia KQ. 2004a. Velocity and temperature cross-scaling
in turbulent thermal convection. J. Turbul. 5:27
Ching ESC, Cohen Y, Gilbert T, Procaccia I. 2002. Statistically preserved structures and anomalous scaling
in turbulent active scalar advection. Europhys. Lett. 60:36975
Ching ESC, Cohen Y, Gilbert T, Procaccia I. 2003a. Active and passive elds in turbulent transport: the role
of statistically preserved structures. Phys. Rev. E 67:016304
Ching ESC, Guo H, Cheng WC. 2008a. Understanding the different scaling behavior in various shell models
proposed for turbulent thermal convection. Phys. D 237:200914
Ching ESC, Guo H, Lo TS. 2008b. Rened similarity hypotheses in shell models of homogeneous turbulence
and turbulent convection. Phys. Rev. E 78:026303
Ching ESC, Guo H, Shang XD, Tong P, Xia KQ. 2004b. Extraction of plumes in turbulent thermal convection.
Ching ESC, Ko TC. 2008. Ultimate-state scaling in a shell model for homogeneous turbulent convection.
Phys. Rev. E 78:036309

ANRV400-FL42-15
360
Lohse
Xia

ANRV400-FL42-15
ARI
13 November 2009
14:9
Ching ESC, Kwok CY. 2000. Statistics of local temperature dissipation in high Rayleigh number convection.
Phys. Rev. E 62:R758790
Ching ESC, Leung CK, Qiu XL, Tong P. 2003b. Intermittency of velocity uctuations in turbulent thermal
convection. Phys. Rev. E 68:026307
Cioni S, Ciliberto S, Sommeria J. 1995. Temperature structure functions in turbulent convection at low Prandtl
number. Europhys. Lett. 32:41318
Corrsin S. 1951. On the spectrum of isotropic temperature uctuations in anisotropic turbulence. J. Appl.
Phys. 22:46973
Efnger H, Grossmann S. 1987. Static structure function of turbulent ow from the Navier-Stokes equation.
Z. Phys. B 66:289304
Eggers J, Grossmann S. 1991. Does deterministic chaos imply intermittency in fully-developed turbulence?
Phys. Fluids A 3:195868
Emran MS, Schumacher J. 2008. Fine-scale statistics of temperature and its derivatives in convective turbulence. J. Fluid Mech. 611:1334
Falkovich G, Gawe dzki K, Vergassola M. 2001. Particles and elds in uid turbulence. Rev. Mod. Phys. 73:913
75
Frisch U. 1995. Turbulence. Cambridge, UK: Cambridge Univ. Press
Funfschilling D, Brown E, Ahlers G. 2008. Torsional oscillations of the large-scale circulation in turbulent
Rayleigh-Benard convection. J. Fluid Mech. 607:11939
Gasteuil Y, Shew WL, Gibert M, Chilla F, Castaing B, Pinton JF. 2007. Lagrangian temperature, velocity,
and local heat ux measurement in Rayleigh-Benard convection. Phys. Rev. Lett. 99:234302
Glazier JA, Segawa T, Naert A, Sano M. 1999. Evidence against ultrahard thermal turbulence at very high
Rayleigh numbers. Nature 398:30710
Gluckman BJ, Willaime H, Gollub J. 1993. Geometry of isothermal and isoconcentration surfaces in thermal
turbulence. Phys. Fluids A 5:64761
Grossmann S, Lohse D. 1991. Fourier-Weierstrass mode analysis for thermally driven turbulence. Phys. Rev.
Lett. 67:44548
Grossmann S, Lohse D. 1992a. Intermittency in Navier-Stokes dynamics. Z. Phys. B 89:1119
Grossmann S, Lohse D. 1992b. Scaling in hard turbulent Rayleigh-Benard ow. Phys. Rev. A 46:90317
Grossmann S, Lohse D. 1993. Characteristic scales in Rayleigh-Benard convection. Phys. Lett. A 173:5862
Grossmann S, Lohse D. 2000. Scaling in thermal convection: a unifying view. J. Fluid. Mech. 407:2756
Grossmann S, Lohse D. 2001. Thermal convection for large Prandtl number. Phys. Rev. Lett. 86:331619
Grossmann S, Lohse D. 2002. Prandtl and Rayleigh number dependence of the Reynolds number in turbulent
thermal convection. Phys. Rev. E 66:016305
Grossmann S, Lohse D. 2004. Fluctuations in turbulent Rayleigh-Benard convection: the role of plumes. Phys.
Fluids 16:446272
Grossmann S, Lohse D, Lvov V, Procaccia I. 1994. Finite size corrections to scaling in high Reynolds number
turbulence. Phys. Rev. Lett. 73:43235
Grossmann S, Lvov VS. 1993. Crossover of spectral scaling in thermal turbulence. Phys. Rev. E 47:416168
Hartmann DL, Moy LA, Fu Q. 2001. Tropical convection and the energy balance at the top of the atmosphere.
J. Clim. 14:4495511
He X, Tong P. 2009. Measurements of the thermal dissipation eld in turbulent Rayleigh-Benard convection.
Phys. Rev. E 79:026306
He X, Tong P, Xia KQ. 2007. Measured thermal dissipation eld in turbulent Rayleigh-Benard convection.
Heslot F, Castaing B, Libchaber A. 1987. Transition to turbulence in helium gas. Phys. Rev. A 36:587073
Ishihara T, Gotoh T, Kaneda Y. 2009. Study of high-Reynolds number isotropic turbulence by direct numerical
simulation. Annu. Rev. Fluid Mech. 41:16580
Julien K, Legg S, McWilliams J, Werne J. 1999. Plumes in rotating convection. Part 1. Ensemble statistics
and dynamical balances. J. Fluid Mech. 391:15187
Kaczorowski M, Wagner C. 2009. Analysis of the thermal plumes in turbulent Rayleigh-Benard convection
based on well-resolved numerical simulations. J. Fluid Mech. 618:89112
361
ARI
13 November 2009
14:9
Kadanoff LP. 2001. Turbulent heat ow: structures and scaling. Phys. Today 54(8):3439
Kerr R. 1996. Rayleigh number scaling in numerical convection. J. Fluid Mech. 310:13979
Kerr RM. 2001. Energy budget in Rayleigh-Benard convection. Phys. Rev. Lett. 87:244502
Kolmogorov AN. 1941. The local structure of turbulence in imcompressible viscous uid for very large
Reynolds numbers. Dokl. Akad. Nauk. SSSR 30:299303
Krishnamurti R, Howard LN. 1981. Large scale ow generation in turbulent convection. Proc. Natl. Acad. Sci.
USA 78:198185
Kunnen RPJ, Clercx HJH, Geurts BJ, Bokhoven LJA, Akkermans RAD, Verzicco R. 2008. A numerical and
experimental investigation of structure function scaling in turbulent Rayleigh-Benard convection. Phys.
Rev. E 77:016302
Kurien S, Sreenivasan KR. 2000. Anisotropic scaling contributions to high-order structure functions in highReynolds-number turbulence. Phys. Rev. E 62:220612
Kuznetsov EA, Lvov VS. 1981. On developed hydrodynamic turbulence spectra. Phys. D 2:20317
Landau LD, Lifshitz EM. 1987. Fluid Mechanics. Oxford: Pergamon
Lohse D. 1994. Temperature spectra in shear ow and thermal convection. Phys. Lett. A 196:7075
Lohse D, Muller-Groeling
A. 1996. Anisotropy and scaling corrections in turbulence. Phys. Rev. E 54:395405
Lohse D, Toschi F. 2003. The ultimate state of thermal convection. Phys. Rev. Lett. 90:034502
Lukaschuk S, Ashkenazi S, Lebedev V, Steinberg V. 2001. New light scattering technique based on phase
time derivative correlation function. Europhys. Lett. 56(6):80814
Lumley JL. 1967. Similarity and turbulent energy spectrum. Phys. Fluids 10:85558
Marshall J, Schott F. 1999. Open-ocean convection: observations, theory, and models. Rev. Geophys. 37:164
Mashiko T, Tsuji Y, Mizuno T, Sano M. 2004. Instantaneous measurement of velocity elds in developed
thermal turbulence in mercury. Phys. Rev. E 69:036306
McKenzie DP, Roberts JM, Weiss NO. 1974. Convection in the Earths mantle: towards a numerical simulation. J. Fluid Mech. 62:465538
Moisy F, Willaime H, Andersen J, Tabeling P. 2001. Passive scalar intermittency in low temperature helium
ows. Phys. Rev. Lett. 86:482730
Monin AS, Yaglom AM. 1975. Statistical Fluid Mechanics. Cambridge, MA: MIT Press
Niemela J, Skrbek L, Sreenivasan KR, Donnelly R. 2000. Turbulent convection at very high Rayleigh numbers.
Nature 404:83740
Niemela J, Sreenivasan KR. 2006. The use of cryogenic helium for classical turbulence: promises and hurdles.
J. Low Temp. Phys. 143:163212
Nikolaenko A, Brown E, Funfschilling D, Ahlers G. 2005. Heat transport by turbulent Rayleigh-Benard
convection in cylindrical cells with aspect ratio one and less. J. Fluid Mech. 523:25160
Obukhov AM. 1949. Structure of the temperature eld in a turbulent ow. Izv. Akad. Nauk. SSSR Ser. Geog.
Geoz. 13:5869
Obukhov AM. 1959. On the inuence of Archimedean forces on the structure of the temperature eld in a
turbulent ow. Dokl. Akad. Nauk. SSR 125:124648
Pinton JF, Labbe R. 1994. Correction to the Taylor hypothesis in swirling ows. J. Phys. II France 4:146168
Pope SB. 2000. Turbulent Flow. Cambridge, UK: Cambridge Univ. Press
Procaccia I, Ching ESC, Constantin P, Kadanoff LP, Libchaber A, Wu XZ. 1991. Transition to convective
turbulence: the role of thermal plumes. Phys. Rev. A 44:8091102
Procaccia I, Zeitak R. 1989. Scaling nonisotropic convective turbulence. Phys. Rev. Lett. 62:212831
Procaccia I, Zeitak R. 1990. Scaling exponents in thermally driven turbulence. Phys. Rev. A 42:82130
Puthenveettil BA, Arakeri JH. 2005. Plume structure in high-Rayleigh-number convection. J. Fluid Mech.
542:21749
Qiu XL, Yao SH, Tong P. 2000. Large-scale coherent rotation and oscillation in turbulent thermal convection.
Phys. Rev. E 61:R607578
Rahmstorf S. 2000. The thermohaline ocean circulation: a system with dangerous thresholds? Clim. Change
46:24756
Richardson LF. 1926. Atmospheric diffusion shown on a distance-neighbor graph. Proc. R. Soc. Lond. Ser. A
110:70937

ANRV400-FL42-15
362
Lohse
Xia

ANRV400-FL42-15
ARI
13 November 2009
14:9
Rincon F. 2006. Anisotropy, inhomogeneity and inertial-range scalings in turbulent convection. J. Fluid Mech.
563:4369
Rubio MA, Bigazzi P, Albavetti L, Ciliberto S. 1989. Spatio-temporal regimes in Rayleigh-Benard convection
in a small rectangular cell. J. Fluid Mech. 209:30934
Ruiz-Chavarria G, Baudet C, Ciliberto S. 1996. Scaling laws and dissipation scale of a passive scalar in fully
developed turbulence. Phys. D 99:36980
Saddoughi SG, Veeravalli SV. 1994. Local isotropy in turbulent boundary layers at high Reynolds number.
J. Fluid Mech. 268:33372
Sano M, Wu XZ, Libchaber A. 1989. Turbulence in helium-gas free-convection. Phys. Rev. A 40:642130
Schumacher J. 2008. Lagrangian dispersion and heat transport in convective turbulence. Phys. Rev. Lett.
100:134502
Shang XD, Xia KQ. 2001. Scaling of the velocity power spectra in turbulent thermal convection. Phys. Rev. E
64:065301
Shang XD, Qiu XL, Tong P, Xia KQ. 2003. Measured local heat transport in turbulent Rayleigh-Benard
She ZS, Jackson E. 1993. On the universal form of energy spectra in fully developed turbulence. Phys. Fluids
A 5:152628
She ZS, Leveque E. 1994. Universal scaling laws in fully developed turbulence. Phys. Rev. Lett. 72:33639
Shishkina O, Wagner C. 2006. Analysis of thermal dissipation rates in turbulent Rayleigh-Benard convection.
Shishkina O, Wagner C. 2008. Analysis of sheetlike thermal plumes in turbulent Rayleigh-Benard convection.
Shraiman BI, Siggia ED. 1990. Heat transport in high-Rayleigh number convection. Phys. Rev. A 42:365053
Siggia ED. 1994. High Rayleigh number convection. Annu. Rev. Fluid Mech. 26:13768
Skrbek L, Niemela JJ, Sreenivasan KR, Donnelly RJ. 2002. Temperature structure functions in the Bolgiano
regime of thermal convection. Phys. Rev. E 66:036303
Sreenivasan KR. 1991. On local isotropy of passive scalars in turbulent shear ows. Proc. R. Soc. Lond. A
434:16582
Sreenivasan KR, Antonia RA. 1977. Skewness of temperature derivatives in turbulent shear ows. Phys. Fluids
20:198688
Sreenivasan KR, Antonia RA. 1997. The phenomenology of small-scale turbulence. Annu. Rev. Fluid Mech.
29:43572
Sun C, Cheung YH, Xia KQ. 2008. Experimental studies of the viscous boundary layer properties in turbulent
Rayleigh-Benard convection. J. Fluid Mech. 605:79113
Sun C, Xia KQ. 2007. Multi-point local temperature measurements inside the conducting plates in turbulent
thermal convection. J. Fluid Mech. 570:47989
Sun C, Xia KQ, Tong P. 2005. Three-dimensional ow structures and dynamics of turbulent thermal convection in a cylindrical cell. Phys. Rev. E 72:026302
Sun C, Zhou Q, Xia KQ. 2006. Cascades of velocity and temperature uctuations in buoyancy-driven thermal
Takeshita T, Segawa T, Glazier JA, Sano M. 1996. Thermal turbulence in mercury. Phys. Rev. Lett. 76:146568
Taylor GI. 1938. The spectrum of turbulence. Proc. R. Soc. Lond. A 164:47690
Tennekes H, Lumley JL. 1972. A First Course in Turbulence. Cambridge, MA: MIT Press
Tong P, Shen Y. 1992. Relative velocity uctuations in turbulent Rayleigh-Benard convection. Phys. Rev. Lett.
69:206669
Toschi F, Amati G, Succi S, Benzi R, Piva R. 1999. Intermittency and structure functions in channel ow
Vainshtein S, Sreenivasan K. 1994. Kolmogorovs 4/5 law and intermittency in turbulence. Phys. Rev. Lett.
73:308588
Verzicco R, Camussi R. 2003. Numerical experiments on strongly turbulent thermal convection in a slender
cylindrical cell. J. Fluid Mech. 477:1949
von der Heydt A, Grossman S, Lohse D. 2001. Scaling exponents in anisotropic turbulence from the NavierStokes equations. J. Fluid Mech. 440:38190
363
ARI
13 November 2009
14:9
Warhaft Z. 2000. Passive scalars in turbulent ows. Annu. Rev. Fluid Mech. 32:20340
Wittmer KS, Devenport WJ, Zsoldos JS. 1998. A four-sensor hot-wire probe system for three-component
velocity measurement. Exp. Fluids 24:41623
Wu XZ, Kadanoff L, Libchaber A, Sano M. 1990. Frequency power spectrum of temperature-uctuation in
free convection. Phys. Rev. Lett. 64:214043
Xi HD, Lam S, Xia KQ. 2004. From laminar plumes to organized ows: the onset of large-scale circulation
in turbulent thermal convection. J. Fluid Mech. 503:4756
Xia KQ, Sun C, Zhou SQ. 2003. Particle image velocimetry measurement of the velocity eld in turbulent
thermal convection. Phys. Rev. E 68:066303
Yakhot V. 1992. 4/5 Kolmogorov law for statistically staionary turbulence: application to high Rayleigh number
Benard convection. Phys. Rev. Lett. 69:76971
Zhang J, Wu XL. 2005. Velocity intermittency in a buoyancy subrange in a two-dimensional soap lm convection experiment. Phys. Rev. Lett. 94:234501
Zhang J, Wu XL, Xia KQ. 2005. Density uctuations in strongly stratied two-dimensional turbulence. Phys.
Rev. Lett. 94:174503
Zhou Q, Sun C, Xia KQ. 2007. Morphological evolution of thermal plumes in turbulent Rayleigh-Benard
Zhou Q, Sun C, Xia KQ. 2008. Experimental investigation of homogeneity, isotropy, and circulation of the
velocity eld in buoyancy-driven turbulence. J. Fluid Mech. 598:36172
Zhou Q, Xia KQ. 2008. Comparative experimental study of local mixing of active and passive scalars in
turbulent thermal convection. Phys. Rev. E 77:056312
Zhou SQ, Xia KQ. 2001. Scaling properties of the temperature eld in convective turbulence. Phys. Rev. Lett.
87:064501
Zhou SQ, Xia KQ. 2002. Plume statistics in thermal turbulence: mixing of an active scalar. Phys. Rev. Lett.
89:184502

ANRV400-FL42-15
364
Lohse
Xia
AR400-FM
ARI
13 November 2009
15:33

Contents
Annual Review of
Fluid Mechanics
Volume 42, 2010
Singular Perturbation Theory: A Viscous Flow out of Gottingen

Robert E. OMalley Jr. p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 1
Dynamics of Winds and Currents Coupled to Surface Waves
Peter P. Sullivan and James C. McWilliams p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p19
Fluvial Sedimentary Patterns
G. Seminara p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p43
Shear Bands in Matter with Granularity
Peter Schall and Martin van Hecke p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p67
Slip on Superhydrophobic Surfaces
Jonathan P. Rothstein p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p89
Turbulent Dispersed Multiphase Flow
S. Balachandar and John K. Eaton p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 111
Turbidity Currents and Their Deposits
Eckart Meiburg and Ben Kneller p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 135
Measurement of the Velocity Gradient Tensor in Turbulent Flows
James M. Wallace and Petar V. Vukoslavcevic p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 157
Friction Drag Reduction of External Flows with Bubble and
Gas Injection
Steven L. Ceccio p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 183
WaveVortex Interactions in Fluids and Superuids
Oliver Buhler
p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 205
Laminar, Transitional, and Turbulent Flows in Rotor-Stator Cavities
Brian Launder, Sebastien Poncet, and Eric Serre p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 229
Scale-Dependent Models for Atmospheric Flows
Rupert Klein p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 249
Spike-Type Compressor Stall Inception, Detection, and Control
C.S. Tan, I. Day, S. Morris, and A. Wadia p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 275
vii
AR400-FM
ARI
13 November 2009
15:33
Airow and Particle Transport in the Human Respiratory System

C. Kleinstreuer and Z. Zhang p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 301
Small-Scale Properties of Turbulent Rayleigh-Benard Convection
Detlef Lohse and Ke-Qing Xia p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 335
Fluid Dynamics of Urban Atmospheres in Complex Terrain
H.J.S. Fernando p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 365

Turbulent Plumes in Nature

Andrew W. Woods p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 391
Fluid Mechanics of Microrheology
Todd M. Squires and Thomas G. Mason p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 413
Lattice-Boltzmann Method for Complex Flows
Cyrus K. Aidun and Jonathan R. Clausen p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 439
Wavelet Methods in Computational Fluid Dynamics
Kai Schneider and Oleg V. Vasilyev p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 473
Dielectric Barrier Discharge Plasma Actuators for Flow Control
Thomas C. Corke, C. Lon Enloe, and Stephen P. Wilkinson p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 505
Applications of Holography in Fluid Mechanics and Particle Dynamics
Joseph Katz and Jian Sheng p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 531
Recent Advances in Micro-Particle Image Velocimetry
Steven T. Wereley and Carl D. Meinhart p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 557
Indexes
Cumulative Index of Contributing Authors, Volumes 142 p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 577
Cumulative Index of Chapter Titles, Volumes 142 p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 585
Errata
An online log of corrections to Annual Review of Fluid Mechanics articles may be found
at http://uid.annualreviews.org/errata.shtml
viii
Contents

Lohse2010small Scale

Uploaded by

Copyright:

Available Formats

Lohse2010small Scale

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lohse2010small Scale

Uploaded by

Copyright:

Available Formats

ANRV400-FL42-15

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Click here for quick links to

Annu. Rev. Fluid Mech. 2010. 42:33564

First published online as a Review in Advance on

thermal convection, turbulence, structure functions, Bolgiano scaling

The Annual Review of Fluid Mechanics is online at

This articles doi:

The properties of the structure functions and other small-scale quantities in

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Thermally driven turbulence is of tremendous importance to various areas of science, technology,

Velocity (cm s1)

Velocity (cm s1)

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Su (r) (u r)2/3 .

Su (r) (g)4/5 r 6/5 ,

S (r) (g)2/5 r 2/5 .

Su3 (r) (g)1/5 r 4/5 .

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

2. BOLGIANO LENGTH SCALE AND DIFFICULTIES

L B (x) = u5/4 (x)

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

2.2. Cascade Picture

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

together with the incompressibility conditions i ui = 0? [Here p(x, t) is kinematic pressure;

2.3. Further Obstacles in Identifying BO59 Scaling

Su (r) u2/3 L2/3

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Su(3), j (r) gre z S u3 (r).

3. TEMPERATURE AND VELOCITY SPECTRA

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

3.2. Time-Domain Measurements

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

www.annualreviews.org Turbulent Rayleigh-Benard Convection

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Compensated power spectrum

Temperature power spectrum

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

3.3. Converting Time-Domain Data into Space-Domain

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

3.4. Space-Domain Measurements

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

3.5. Structure Functions and Spectra From Numerical Simulations

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

4. SMALL-SCALE MIXING OF ACTIVE AND PASSIVE SCALARS

4.1. Small-Scale Anisotropy of the Temperature Field

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

4.2. Comparative Studies of Active and Passive Scalars

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

exponent for the temperature. The saturation exponent

scalar measured in a swirling ow (Moisy et al. 2001). This suggests that

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org

5. CONCLUSIONS AND OUTLOOK

Annu. Rev. Fluid Mech. 2010.42:335-364. Downloaded from www.annualreviews.org