Rayleigh–Bénard convection in liquid metal layers under the influence of a

vertical magnetic field

Ulrich Burr and Ulrich Müller

Citation: Phys. Fluids 13, 3247 (2001); doi: 10.1063/1.1404385

View online: http://dx.doi.org/10.1063/1.1404385
View Table of Contents: http://pof.aip.org/resource/1/PHFLE6/v13/i11
Published by the American Institute of Physics.

Related Articles
Bi-stability in turbulent, rotating spherical Couette flow
Phys. Fluids 23, 065104 (2011)
Turbulent mixing and passive scalar transport in shallow flows
Phys. Fluids 23, 016603 (2011)
Numerical simulation of reciprocating turbulent flow in a plane channel
Phys. Fluids 21, 095106 (2009)
Eigenmode analysis of scalar transport in distributive mixing
Phys. Fluids 21, 093601 (2009)
Decomposition driven interface evolution for layers of binary mixtures. II. Influence of convective transport on
linear stability
Phys. Fluids 21, 062104 (2009)

Additional information on Phys. Fluids

Journal Homepage: http://pof.aip.org/
Journal Information: http://pof.aip.org/about/about_the_journal
Top downloads: http://pof.aip.org/features/most_downloaded
Information for Authors: http://pof.aip.org/authors

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
PHYSICS OF FLUIDS VOLUME 13, NUMBER 11 NOVEMBER 2001

Rayleigh–Bénard convection in liquid metal layers under the influence

of a vertical magnetic field
Ulrich Burr
Eidgenössische Technische Hochschule Zürich, Institut für Hydromechanik und Wasserwirtschaft,
ETH Hönggerberg, CH-8093 Zürich, Switzerland
Ulrich Müller
Forschungszentrum Karlsruhe GmbH, Institut für Kern- und Energietechnik, Postfach 3640,
D-76021 Karlsruhe, Germany
共Received 31 May 2000; accepted 26 July 2001兲
The influence of a vertical magnetic field on the integral heat transfer and the temporal dynamics of
liquid metal Rayleigh–Bénard convection is studied in an experiment using a small Prandtl number
共Pr⬇0.02兲 sodium potassium alloy Na22Kr78 as a test fluid. The test section is a rectangular box of
large aspect ratio 20 : 10 : 1 that covers a parameter range of Rayleigh numbers, 103 ⬍Ra⬍105 , and
Chandrasekhar numbers, 0⬍Q⬍1.44⫻104 . The integral heat transfer across the layer is evaluated
from the measured temperatures at the upper and the lower boundary and the applied heat flux.
Local, time-dependent temperatures are obtained from a four-element temperature probe placed in
the middle of the liquid metal layer. The noncoplanar arrangement of the thermocouples enables the
evaluation of the time-dependent temperature gradient vector that allows us to estimate the local
isotropy properties of the time-dependent flow. From the damping effect of Joule dissipation, the
convective heat transport decreases monotonically with increasing Chandrasekhar numbers.
Fluctuations of the temperature field are damped significantly by the magnetic field. However, this
effect is selective with respect to frequency. Long period fluctuations are strongly damped whereas
short period fluctuations are less damped or may even be amplified. The observations show that
significant convective heat transport is practically always associated with time-dependent flow. The
fluctuating part of the local temperature gradient confirms the horizontal isotropy of the velocity
field; no predominant orientation of time-dependent flow structures is established either with or
without a magnetic field. © 2001 American Institute of Physics. 关DOI: 10.1063/1.1404385兴

I. INTRODUCTION where ␤ ⫽1/␳ •d ␳ /dT is the volumetric thermal expansion

coefficient defined as the relative change of the density ␳
Natural convective flows of electrically well conducting
with temperature T, g is the intensity of the acceleration of
fluids such as liquid metals are significantly influenced by
gravity, ⌬T⫽T b ⫺T t is the temperature difference between
electrodynamic forces when magnetic fields are imposed. A
the lower and the upper fluid boundary, h is the height of the
magnetohydrodynamic 共MHD兲 natural convective flow with
layer, ␯ is the kinematic viscosity, and ␬ ⫽␭/ ␳ c p is the ther-
significantly different flow structure and intensity of the ve-
mal diffusivity with the thermal conductivity ␭ and the spe-
locity compared to an ordinary hydrodynamic 共OHD兲 flow is
cific heat capacity c p . Due to thermal expansion, the fluid is
established. Thereby, the convective transport of heat and
lighter at the bottom whereas the fluid at the upper wall is
mass as well as the temporal behavior of the flow is changed.
heavier. If the temperature gradient exceeds a certain critical
This matter is of interest in several metallurgical processes
value, a convective motion sets in, in the form of stationary
such as casting or crystal pulling, where the control of the
buoyant convective rolls of a distinct critical spatial wave-
convective flow by a magnetic field may influence the qual-
ity of the produced material 共see Davidson1兲. length ␭ c . If both bounding surfaces are isothermal and non-
In this article we present experimental results obtained slip conditions hold, the onset of convection occurs at a criti-
for a Rayleigh–Bénard system with an imposed vertical cal Rayleigh number Rac ⫽1707.8 with a horizontal wave
magnetic field. Rayleigh–Bénard convection is a fundamen- number a c ⫽3.117/h 共see Pellew and Southwell2 and Reid
tal problem of natural convective heat transfer with very and Harris3兲. With the critical wavelength ␭ c ⫽2 ␲ /a c ⬇2h,
well-defined boundary conditions. It is characterized by a the lateral dimensions of the convective rolls are about equal
vertical temperature gradient aligned with the acceleration of to the height h of the layer at the point of marginal instabil-
gravity g being maintained over an infinite horizontal layer ity.
of fluid. The most significant dimensionless parameter that If the Rayleigh number is increased further, the flow
governs the flow is the Rayleigh number, undergoes a transition from steady two- to three-dimensional
共3-D兲 flow patterns and furthermore to time-dependent 3-D
␤ g⌬Th 3 and finally turbulent flow. The limits of these different flow
Ra⫽ , 共1兲
␯␬ regimes do not solely depend on the Rayleigh number. As

1070-6631/2001/13(11)/3247/11/$18.00 3247 © 2001 American Institute of Physics

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
3248 Phys. Fluids, Vol. 13, No. 11, November 2001 U. Burr and U. Müller

inertial forces become more important in the momentum bal-

ance, the Prandtl number,

␯
Pr⫽ , 共2兲
␬
enters the problem as an additional dimensionless parameter
that controls the coupling between the velocity and the tem-
perature field. The crucial influence of the Prandtl number on
the stability of the various flow regions was shown by
Krishnamurti.4 In case of very small Prandtl numbers, typical
FIG. 1. Magnetic damping of a convective roll by Lorentz forces FL gen-
for liquid metals, the transition from laminar to turbulent
erated from the interaction of a vertical magnetic field B and the current
convection occurs through a variety of intermediate states in density j.
a very small region of Rayleigh numbers just above critical
conditions. The structure of the intermediate convective
states at supercritical Rayleigh numbers has been addressed magnetic Reynolds number Rm⫽ ␮ ␴ v 0 h, where ␴ is the
extensively in the literature 共see, e.g., Clever and Busse,5 electrical conductivity of the fluid and v 0 the characteristic
Busse,6 and Chu and Goldstein7兲. The regions of steady and convective velocity that is unknown. Reasonable estimates
time-dependent 3-D flow can partly be characterized by well- of v 0 show that for liquid metal flows Rm is always small
organized flow patterns such as Eckhaus, zig–zag, cross-roll, and the induced magnetic field may therefore be neglected.
or knot instabilities. Turbulent convection is characterized by The coupling between fluid motion and the induced Lor-
thermal boundary layers at the upper and the lower boundary. entz forces in a Rayleigh–Bénard convective roll is sketched
They are, however, less distinct in liquid metals with low in Fig. 1 for the case of a purely vertical magnetic field. The
Prandtl numbers 共see Horanyi et al.8兲. From these layers horizontal motions induce potential differences parallel to
small-scale turbulent vortices are released, resulting in large the roll axis. As the potential gradients on the upper and the
convective heat transport from the bottom to the top. Within lower side oppose each other, electric currents immediately
the boundary layers, heat is transported mainly by heat con- close within the convective roll itself and due to the short
duction. current paths a high current density is produced. The induced
The effectiveness of the convective heat transport is Lorentz forces are opposing the horizontal motions and
characterized by the dimensionless Nusselt number, thereby cause strong electromagnetic damping by Joule dis-
sipation on the time scale ␶ JD⬃ ␳ / ␴ B 20 , where B 0 is the mag-
q
Nu⫽ , 共3兲 nitude of the applied magnetic field 共see Shercliff11兲. As hori-
q0 zontal motion cannot be avoided in recirculating natural
defined by the ratio of the total heat flux q to the conductive convective flows, the Lorentz force stabilizes the fluid and
one q 0 ⫽⫺␭⌬T/h. Pure conductive heat transfer is de- the onset of convection is shifted to higher critical Rayleigh
scribed by Nu⫽1, whereas the occurrence of convective heat numbers. This effect is characterized by the ratio of the time
transport results in Nusselt numbers Nu⬎1. In general, the scale of viscous dissipation ␶ v ⫽h 2 / ␯ to the time scale of
Nusselt number depends on both the Rayleigh and the Joule dissipation ␶ JD , which gives the Chandrasekhar num-
Prandtl number and has been investigated experimentally in ber,
the past by, e.g., Rossby9 for Pr⫽0.025, Pr⫽7.0 and Pr B 20 h 2 ␴
⫽200, and Threlfall10 for Pr⫽0.7 and Horanyi et al.8 for Q⫽M ⫽ 2
, 共4兲
␳␯
Pr⫽0.006. At constant, moderate Rayleigh numbers, the
Nusselt numbers are significantly smaller at smaller Prandtl as an additional dimensionless parameter. The Chan-
numbers. drasekhar number is the square of the Hartmann number M
External magnetic fields can change the characteristics commonly used in MHD duct flows. In MHD flows the elec-
of Rayleigh–Bénard convection in liquid metal layers sig- trical properties of walls have a significant influence. They
nificantly. A flow of velocity v within the magnetic field B are described by the wall conductance ratio,
induces an electric field E⫽v⫻B. Electric currents driven ␴ Ws
along the gradient of the electric field form current loops that c⫽ , 共5兲
␴b
either close within the liquid metal or through conducting
confining walls. There are two effects arising from the pres- where ␴ W is the electrical conductivity of the wall material,
ence of a current density j in the fluid. First, a magnetic field s is the thickness of the wall considered, and b is the half-
ⵜ⫻B⫽ ␮ "j is induced which is superimposed on the applied width of the flow region in the direction of the magnetic field
one, where ␮ is the magnetic permeability of the fluid. Sec- 共see, e.g., Walker12兲.
ond, the interaction of the current density with the magnetic The critical Rayleigh numbers for the onset of convec-
field causes Lorentz forces FL ⫽j⫻B that act perpendicular tion have been calculated by Chandrasekhar13 using a linear
to the current density and the magnetic field. stability analysis. Aside from the strong increase of the criti-
The ratio of the magnetic field induced by the fluid mo- cal Rayleigh number with an increasing Chandrasekhar num-
tion to the externally applied magnetic field is defined as ber, he finds an increase of the horizontal wave number a c

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
Phys. Fluids, Vol. 13, No. 11, November 2001 Rayleigh–Bénard convection at vertical magnetic field 3249

associated with smaller lateral dimensions of the convective

rolls. This is explained by the lack of Joule dissipation in
zones of vertical motion. The system seems to minimize
Joule dissipation by reducing horizontal motions up to the
extent where viscous dissipation compensates the benefit.
At high Chandrasekhar numbers viscous forces become
less important and the onset of instability is governed by the
balance of buoyant and electromagnetic forces only. Neglect-
ing viscous forces and assuming nonslip boundary conditions
at the upper and the lower interface, the layer is determined
to become asymptotically unstable at a critical Rayleigh
number,
Rac ⫽ ␲ 2 Q, 共6兲
FIG. 2. Cross section along the shorter side of the test section. The liquid
and for a critical wave number, metal layer is confined between two copper plates with embedded thermo-
couples. Heat is supplied to the lower boundary by electrically powered
␲ 4 1/6 heating rods and is removed at the upper boundary by an evaporation cooler
a c⫽ Q . 共7兲
2h with the condenser placed just above.

The critical Rayleigh numbers for the onset of convec-

tion were confirmed with reasonable accuracy by experi-
ments of Nakagawa 共Refs. 14 –16兲. The effect of a vertical
magnetic field on the flow patterns at moderate supercritical well-defined boiling process with constant superheating of
Rayleigh numbers was investigated theoretically by Busse the wall, the boiling pool is filled with glass spheres of 5 mm
and Clever 共Refs. 17 and 18兲. These authors showed that the diam.
stability limits of distinct flow regimes like stationary or time The magnetic field is provided by an electromagnet that
periodic flow patterns are shifted to higher Rayleigh numbers generates a homogeneous vertical magnetic field in an area
and characteristic time scales are reduced as the Chan- of 800⫻483 mm and within a magnetic gap of 168 mm. The
drasekhar number is increased. magnetic field of an intensity up to 2 T is constant to an
Although the problem has been discussed extensively in accuracy better that 1%. The eutectic sodium–potassium al-
the literature, experimental results for heat transfer and local loy Na22K78, with 22% weight sodium and 78% weight po-
temporal behavior are not available for highly supercritical tassium, was chosen as a test fluid. The temperature-
conditions. Since this region is most relevant for technical dependent thermophysical properties are calculated for the
applications, the design of the experimental test apparatus, mean temperature of the fluid T m based on fitting curves
described in the next section, was optimized to achieve high derived from data taken from O’Donnel et al.,19 Lyon,20 and
Rayleigh numbers rather than to investigate the flow struc- Foust.21 With this experimental setup the range of param-
ture at marginally supercritical conditions. Nevertheless, eters, 103 ⬍Ra⬍105 and 0⬍Q⬍6⫻106 , can be covered.
with the realized large aspect ratio, the influence of lateral Due to technical peculiarities of the experimental setup the
walls was reduced and the obtained results, presented in Sec. mean temperature of the fluid increases with the Rayleigh
III, can be compared with theoretical predictions derived for number and because of the temperature dependence of the
infinite fluid layers. physical properties the Prandtl number varies in the range
0.017⬍Pr⬍0.021.
The temperatures at the lower and the upper boundary
II. EXPERIMENTAL SETUP AND PERFORMANCE
are measured by five Cu–CuNi thermocouples situated in
The test section provides a rectangular liquid metal layer each copper plate. In Fig. 3 the horizontal positions of the
of height 20 mm, width 200 mm, and length 400 mm, result- thermocouples T i, j are shown, where i denotes the horizontal
ing in a large aspect ratio of 1 : 10 : 20. Figure 2 shows a position and j indicates the lower and upper wall by b and t,
cross section along the shorter side of the used test section. respectively. The temperature difference across the layer
The lower and upper walls are copper plates with 20 mm, the ⌬T⫽T b ⫺T t and the mean temperature of the fluid T m
side walls are stainless steel sheet metal with 1.5 mm thick- ⫽(T b ⫹T t )/2 are calculated from the average values of the
ness, respectively. From definition 共5兲 the wall conductance temperatures at the lower and the upper wall T b and T t ,
ratios c Cu ⬵4.5 and c ss ⬵0.08 for the horizontal and the ver- respectively. The heat flux q is determined from the mea-
tical walls are calculated. sured electric power. There are three main sources of heat
The lower copper plate is heated by 40 electrically pow- losses that have to be taken into account: First, Ohmic losses
ered heating rods in a parallel arrangement. At a maximum in the wiring and cold ends of the heater rods being propor-
power of P⬇11 kW a heat flux of q⬇1.35⫻105 W/m2 is tional to the electrical power; second, due to nonperfect ther-
obtained. The heat is removed from the upper wall by the mal insulation there are heat losses of the heated copper plate
evaporation of water in a boiling pool. The steam is con- to the environment that are estimated, taking into account the
densed on 12 longitudinal copper tubes just above the pool measured temperature difference between the copper plate
that are water cooled from an external loop. In order to get a and the environment; and third, heat conduction through the

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
3250 Phys. Fluids, Vol. 13, No. 11, November 2001 U. Burr and U. Müller

FIG. 3. A schematical sketch of the test facility and the positions of instru-
mentations. Five Cu–CuNi thermocouples T i, j are placed at the upper and
lower fluid–wall interface. In the middle of the fluid layer, between the
thermocouples T S,b and T S,t , a four-element temperature probe is placed.
All lengths are given in mm.
FIG. 4. The sensing tip of the four-element temperature probe. 共a兲 A com-
parison with a match. 共b兲 The geometry and coordinates. All lengths are
given in mm.

stainless steel sidewalls that are easily assessed by Fourier’s

law. spatial structure of the convective flow. As a measure for the
Notice that the integral quantities T b , T t , ⌬T, T m , and intensity of the temperature fluctuations, the variance of the
q as well as the characteristic numbers Ra, Nu, Q, and Pr, are temperature,
evaluated as spatial and temporal mean values only and the
averaging is therefore not indicated.
Local, time-dependent flow quantities are determined
T ⬘P2 ⫽
1
tm
冕 tm

0
T ⬘P2 dt, 共9兲
from a four-element temperature probe placed in the middle
is evaluated, where t m is a reasonably long time interval. The
of the layer, between the thermocouples T S,b and T S,t , as
integral time scale ␶ I of the flow is obtained from the expres-
shown in Fig. 3. Figure 4共a兲 shows the sensing tip of the
sion
probe in comparison with a match. The probe shaft is made
of an isolating ceramic tube with a diameter of 2 mm. Four
Ni–CrNi thermocouples of 0.25 mm diam are sticking out of ␶ I⫽
2␲
t0
冕 0
tI
A 共 t 兲 dt, 共10兲
the tube into the fluid and are measuring the local tempera-
ture T P . Due to the high thermal conductivity of the test where A(t) is the autocorrelation function normalized with
fluid, the thermal response time of the thermocouples is es- T ⬘P2 of the temperature signal, t I is the time shift where A(t)
timated to be of the order O(10⫺2 s), that is sufficiently becomes zero, and t 0 ⫽h 2 / ␬ ⬇15.5 s is the thermal diffusion
small for the flow problem considered here. The precise ar- time that renders ␶ I dimensionless. The integral time scale
rangement of the thermocouples and the distances between provides an estimate of the predominant time scale of turbu-
them is sketched in Fig. 4共b兲. From the noncoplanar arrange- lent convective flow.
ment of the thermocouples the probe is able to sense the
fluctuating part of the local temperature gradient ⵜT ⬘P III. EXPERIMENTAL RESULTS

冋 册
⫽( ⳵ x T ⬘P , ⳵ y T ⬘P , ⳵ z T ⬘P ) by finite differences, namely,
Experiments are conducted for six different Chan-
T 3 ⫹T 4 ⫺T 1 ⫺T 2 drasekhar numbers including OHD flow denoted by Q⫽0. In
x 3 ⫹x 4 ⫺x 1 ⫺x 2 Table I the intensities of the corresponding applied magnetic
fields B, the critical Rayleigh number for the onset of con-
T 2 ⫺T 1 vection Rac and the associated wave numbers a c are listed.
ⵜT ⬘P ⫽ , 共8兲
y 2 ⫺y 1 The latter two are obtained from an interpolation of the re-
T 3 ⫺T 4 sults of Chandrasekhar,13 valid for two rigid boundaries, to
z 3 ⫺z 4 the Chandrasekhar numbers used in our experiment.
Integral flow properties are evaluated from a first series
where x i , y i , and z i are the coordinates of the thermocouple of experiments, where the electrical power is chosen as a
tips. The vector ⵜT ⬘P provides additional information on the parameter. In this case the temperature difference across the

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
Phys. Fluids, Vol. 13, No. 11, November 2001 Rayleigh–Bénard convection at vertical magnetic field 3251

TABLE I. Hartmann numbers M, intensities of the applied magnetic field B, critical Rayleigh numbers for the
onset of convection Rac , and critical wave numbers a c for the Chandrasekhar numbers Q realized in the
experiments. The data for Rac and a c refer to predictions of Chandrasekhar 共Ref. 13兲.

Q 0 100 400 784 1600 14 400

M 0 10 20 28 40 120

B关T兴 0 7.2⫻10⫺3 1.45⫻10⫺2 2.02⫻10⫺2 2.85⫻10⫺2 8.74⫻10⫺2

Rac 1707.8 3768.7 8654.9 14 134 24 942 174 042
a c 关 h ⫺1 兴 3.117 4.00 4.88 5.44 6.14 8.89

layer ⌬T adjusts according to the heat transport, and the A. Integral heat transfer
Rayleigh number Ra is therefore not kept constant. In a sec-
In Fig. 6 the integral heat transfer across the layer is
ond series of experiments the more time consuming experi-
characterized by a graph of Nusselt versus Rayleigh num-
mental procedure of adjusting the Rayleigh number to the
bers. The different symbols represent measurements at dif-
constant value Ra⫽5⫻104 is chosen.
ferent Chandrasekhar numbers. The critical Rayleigh num-
If not mentioned otherwise, the experimental results are bers for the onset of convection are marked by vertical solid
presented in nondimensional form by using the height of the lines crossing the abscissa. Although the applied power is
layer h, the temperature difference across the layer ⌬T, the known with high accuracy, the Nusselt numbers do not indi-
thermal diffusion time t 0 ⫽h 2 / ␬ , and its inverse f 0 ⫽1/t 0 as cate the state of pure heat conduction by Nu⫽1. This effect
scales for length, temperature, time and frequency.
is most pronounced for Q⫽14 400, where all displayed mea-
As the temperature at the upper boundary of the liquid
surements are in the theoretically subcritical region. As the
metal layer is fixed to the boiling temperature of the water in
results of linear stability analysis are commonly confirmed
the pool 共see Fig. 2兲, the mean temperature of the liquid
by experiments, it is conjectured that this observation results
metal increases with Rayleigh number and due to the tem-
from uncertainties in the physical properties of the test fluid
perature dependence of the physical properties the Prandtl
used for the evaluation.
number is changed. In Fig. 5 the Prandtl number, as obtained
At large enough supercritical Rayleigh numbers, the
in the experiments at different Chandrasekhar numbers, is
Nusselt numbers are monotonically increasing with increas-
plotted versus the Rayleigh number. Larger Rayleigh num-
ing Rayleigh number for all Chandrasekhar numbers, includ-
bers are associated with lower Prandtl numbers, whereas the
ing OHD convection. If we select a constant Rayleigh num-
influence of different Chandrasekhar numbers is not signifi-
ber, an increase of the Chandrasekhar number always causes
cant. Although the exact value of the Prandtl number is im-
a decrease of the convective heat transport. Thus, the mea-
portant for a comparison with results, e.g., from numerical
simulations, the variations are, from a practical point of view,
not significant and we neglect the influence of Prandtl num-
ber changes in the further discussion.

FIG. 6. The increase of Nusselt numbers Nu with Rayleigh number Ra for

FIG. 5. Prandtl numbers Pr evaluated from the fluid properties at the mean different Chandrasekhar numbers Q. The critical Rayleigh numbers obtained
temperature of the fluid T m that varies for different values of the Rayleigh from linear theory are given for each Chandrasekhar number by vertical
number Ra and the Chandrasekhar number Q realized in the experiment. lines.

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
3252 Phys. Fluids, Vol. 13, No. 11, November 2001 U. Burr and U. Müller

FIG. 8. Intensity of temperature fluctuations for different Chandrasekhar

numbers characterized by the variance of temperature fluctuations T ⬘P2 mea-
FIG. 7. Nusselt numbers of MHD convection plotted versus the reduced sured by the probe.
Rayleigh number Rar ⫽Ra/Rac ⫺1.

Nu⫺1⫽0.125•Rar2/3 . 共12兲
sured Nusselt numbers show that a vertical magnetic field
has only a damping effect on the convective heat transport. Notice that the Chandrasekhar number enters the above rela-
Regions of enhanced heat transfer, as they might be expected tions 共11兲 and 共12兲 only by the critical Rayleigh number Rac
from an ordering influence of the magnetic field, are not for the onset of convection. In the asymptotic limit of high
observed. magnetic fields, Rac is given by relation 共6兲, and we derive
In order to derive scaling laws for the Nusselt number,
Nu⬀ 共 Ra/Q 兲 2/3 共13兲
OHD convection is treated separately from the MHD flows.
Rossby9 derived from his measurements in heated layers of as a scaling law valid for large supercritical Rayleigh num-
mercury with Pr⬇0.025 and Ra⬎2⫻104 , the relationship bers (RaⰇ1) and high Chandrasekhar numbers (QⰇ1).
Nu⫽0.147⫻Ra0.257⫾0.004 共see the dotted line in Fig. 6兲. Al- Scaling of the convective heat transport with Ra/Q was also
though the Prandtl number of mercury is similar to that of observed recently by Davoust et al.22 for a horizontal cylin-
Na 22K78, this correlation overestimates the convective heat drical cell with axial heat flux and vertical magnetic field. It
transport observed in our facility. Fitting the measurements seems therefore to be reasonable whenever the main vorticity
for Ra⬎104 with the same type of power function, we get of convective vortices is perpendicular to the applied mag-
Nu⫽0.1403⫻Ra0.250⫾0.004 with a remarkably similar power netic field.
law compared to the one of Rossby’s correlation.
In MHD convection, Joule dissipation additionally en- B. Temporal structure
ters the problem and the Chandrasekhar number should enter The linear stability analysis of Chandrasekhar13 shows
the scaling laws. In Fig. 7 the modified Nusselt numbers that in the limiting case of small magnetic Reynolds numbers
Nu⫺1 of MHD convection are plotted versus the reduced RmⰆ1, convection starts as stationary flow. Oscillatory con-
Rayleigh number Rar ⫽(Ra/Rac ⫺1), where Rac is obtained vection sets in only after a second bifurcation at higher criti-
from linear theory. The measured Nusselt numbers of all cal Rayleigh number Rat . The marginal conditions for this
Chandrasekhar numbers level off to the value Nu⫽1 at Rar transition to occur are investigated here experimentally by
⬇0.5. Thus significant convective heat transport occurs for
using the temperature variance T ⬘P2 , calculated from Eq. 共9兲
Raⲏ1.5•Rac . In the range 0.5⬍Rar ⬍2, the increase of the
using temperature fluctuations T ⬘P recorded by the probe. In
Nusselt number is well described for all Chandrasekhar num-
Fig. 8 the variances of all experiments at different Chan-
bers by one single equation in the form
drasekhar numbers are plotted versus the Rayleigh number.
Nu⫺1⫽0.066•Rar3/2 . 共11兲 The different symbols represent the measured values. The
level of noise is given by the straight solid line that scales as
In a transition region 2⬍Rar ⬍5 the Nusselt numbers are Ra⫺2 . The noise level has been determined separately using
increasing more strongly with Rar at higher Chandrasekhar thermally neutral instrumentation tests without heating the
numbers. However, for Rar ⲏ5, the Nusselt numbers for Q facility. The vertical dotted lines are indicating the onset of
⫽100 and Q⫽400 increase uniformly. For Q⫽100 a fit of convection as a stationary motion, predicted from linear
the available data gives the relation theory. Provided that the level of noise is reasonably small,

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
Phys. Fluids, Vol. 13, No. 11, November 2001 Rayleigh–Bénard convection at vertical magnetic field 3253

FIG. 10. The time series of the fluctuating part of the temperature T ⬘P re-
FIG. 9. Integral time scales ␶ I for OHD flow and MHD flows at different
corded by the probe for MHD flow at Q⫽400. The Rayleigh number is
Chandrasekhar numbers Q, plotted versus the supercritical Rayleigh number
increased from the subcritical value Ra⫽6062 to supercritical values up to
Ras ⫽Ra⫺Rac .
Ra⫽86 872.

the flow may be considered as stationary if the measured ent Chandrasekhar numbers is plotted versus the supercritical
variances are smaller than the level of noise. The onset of Rayleigh number Ras ⫽Ra⫺Rac . For Rayleigh numbers
time-dependent flow is then given by the points where the larger than for the onset of time-dependent flow, the integral
values of the variance plots T ⬘P2 emerge from the noise level. time scale of both MHD and OHD flow decrease. This shows
This can be estimated from the intersection of the solid fit- that instabilities set in as slow fluctuations that are becoming
ting curves of the measured variances with the noise level faster with increasing Rayleigh numbers. At Ras ⬇4000, the
curve. For OHD flow as well as for MHD flow at Q⫽100 integral time scale of OHD flow achieves a plateau value of
and 400, the onset of time-dependent flow is observed at ␶ I ⬇6, which remains constant up to Ras ⬇4⫻104 . A further
slightly higher Rayleigh numbers compared to those for sta- increase of Ras causes a significant decrease of ␶ I , which is
tionary convection. This is consistent with the linear theory related to faster dynamics of the flow.
of the onset of stationary convection. For the higher Chan- Contrary to this observation in OHD flow, the integral
drasekhar numbers Q⫽784 and 1600. The onset of time- time scale of MHD flow is decreasing monotonically for all
dependent convection is observed to occur at Rayleigh num- Rayleigh numbers, and above Ras ⬎104 much faster dynam-
bers below those for the onset of stationary convection. It is ics are observed as compared to OHD convection. By plot-
conjectured that this observation originates from technical ting the integral time scale ␶ I vs Ras , as it is done here, all
problems in maintaining homogeneous conditions at very evaluated values collapse into one line. A correlation ␶ I
high heat fluxes. However, if we take into account that in the ⬃Ras⫺1 may be given as a rough estimate for the governing
present case the onset of time-dependent flow is always ob- time scale of this type of MHD convection.
served in a region where no significant stationary convective According to the calculations of Clever and Busse23 and
heat transport is detected, the following important conclusion Busse and Clever,18 the flow at marginal and weakly super-
may be drawn: Significant convective heat transport in liquid critical conditions is characterized by regular flow patterns
metal layers, i.e., Nu⬎1, is always associated with a time- whose temporal characteristics are given by a few governing
dependent flow. frequencies only. However, such behavior is not observed in
Above conditions for the onset of oscillatory convection, the experiments conducted in this work, not even for high
the intensities of the temperature signals are steeply increas- intensities of the magnetic field, where a high sensitivity of
ing with Rayleigh number toward individual saturation lev- the probe is assured. Time series of temperature fluctuations
els. In the entire range of investigated parameters, an in- T ⬘ recorded by one thermocouple of the probe 共see Fig. 4兲
crease of the Chandrasekhar number causes a decrease of the are plotted in Figs. 10共a兲–10共f兲. At a constant Chandrasekhar
fluctuating intensity of the temperature field originating from number of Q⫽400 the Rayleigh number is increased from
Joule dissipation corresponding to the monotonic decrease of subcritical up to highly supercritical values. The small fluc-
the Nusselt numbers. tuations, visible in Figs. 10共a兲 and 10共b兲 for the subcritical
For some applications in material processing, the inter- Rayleigh number Ra⫽6062 and for the weakly supercritical
esting question arises as to whether damping by Joule dissi- value Ra⫽9407 can be considered to be related to random
pation can be made selective with respect to frequency. In noise, as the flow at both parameters is determined from Fig.
Fig. 9, the integral time scales defined by Eq. 共10兲 for differ- 8 to be stationary. A real time dependent signal is identified

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
3254 Phys. Fluids, Vol. 13, No. 11, November 2001 U. Burr and U. Müller

FIG. 11. The time series of the fluctuating part of the temperature T ⬘P re-
corded by the probe at constant Rayleigh number Ra⫽5⫻104 for OHD for
low and MHD flow at different Chandrasekhar numbers.

in Fig. 10共c兲 for Ra⫽15 089. The temporal dynamics at this

moderately supercritical Rayleigh number are characterized
by significant amplitudes, slow chaotic fluctuations with FIG. 12. Power spectra S of OHD flow and MHD flow at different Chan-
some characteristic time scale. A further increase of the Ray- drasekhar numbers and for a constant Rayleigh number Ra⫽5.0⫻104 ,
leigh number results in more intensive temporal dynamics evaluated from the temperature time recordings T ⬘P of the probe. At Chan-
drasekhar numbers Q⫽0, 400, 1600, and 3600 Nusselt numbers Nu⫽2.06,
showing large-amplitude fluctuations with a decreasing char- 1.41, 1.07, and 0.98 are obtained.
acteristic time scale, as it is already reflected in the integral
scales 关see Figs. 10共d兲–10共f兲兴.
In Figs. 11共a兲–11共e兲 the time recording of OHD flow is
compared with MHD flows at Q⫽400, Q⫽1600, and Q the upper abscissa in Fig. 12. Compared with the OHD flow
⫽3600 for a Rayleigh number of Ra⫽5⫻104 . The OHD at Q⫽0, the power spectra of MHD flow at Q⫽400 show
flow 关Fig. 11共a兲兴 is characterized by large-amplitude fluctua- significantly lower values in the low-frequency range ( f
tions of long duration in the range between 10 and 50 s. ⱗ1.8). Above this frequency the power spectra of MHD
However, there are also fluctuations of significant smaller flow, exceed those of OHD flow, indicating that high-
time scales. We may interpret the long time variations to frequency fluctuations with higher energies are generated in
result from a movement of convective cells in the horizontal the presence of the external magnetic field. If the magnetic
plane, as they are known from visual observations, e.g., in field is further increased to Q⫽1600 and beyond, a system-
the experiments of Rossby9 and Krishnamurti,24 and the atic damping of the fluctuations is observed in the whole
short events to be related to small-scale vortices that are frequency range.
created by the release of buoyant plumes out of the thermal The dynamics of the convective flow are to some extent
boundary layers. Even when a weak magnetic field corre- characterized by the slopes of the energy spectra in the high-
sponding to Q⫽400 关Fig. 11共b兲兴 is imposed to the convec- frequency range. In Fig. 13 the spectra of OHD flow and the
tive flow the long term variations are damped out and the MHD flows at Q⫽400 and Q⫽1600, divided by factors 103
signal is characterized by fast fluctuations with large ampli- and 104 already shown in Fig. 12, are replotted and charac-
tudes. Increasing the Chandrasekhar number further 关see teristic slopes are emphasized by the straight lines. At OHD
Figs. 11共c兲 and 11共d兲兴 the fluctuating amplitudes are system- flow two different regions with distinct slopes become appar-
atically reduced. At Q⫽1600 the small-scale fluctuations are ent. In the range 1⬍ f ⬍10 the energy decays as S⬃ f ⫺2 ,
more and more suppressed and fluctuations with a longer above f ⬇10 the spectrum matches a slope of approximately
duration become more pronounced. This is consistent with ⫺4. These results are not consistent with the predictions of
the findings in Fig. 9 as for a fixed Rayleigh number, the the spectral model of Batchelor,25 which predicts a behavior
critical Rayleigh number for the onset of convection is re- of S⬃ f ⫺17/3 for low Prandtl number fluids in the inertial-
duced by an increase of the Chandrasekhar number. diffusive subrange of the thermal energy spectrum. However,
Another characterization of the temporal dynamics is ob- if we calculate the Reynolds number from Re⫽(Ra/Pr) 1/2,
tained from the phase-averaged power spectra S of the tem- we obtain Re⬇1600. At this low value, a significant energy
perature recordings of the probe. In Fig. 12 the power spectra transfer from the large vortices to eddies of smaller scales
of OHD flow and MHD flows are shown at different Chan- cannot be expected. There exists no inertial subrange and the
drasekhar numbers and for a constant Rayleigh number of temperature fluctuations decay by molecular diffusion only.
Ra⫽5⫻104 . Here S is plotted versus the nondimensional Hinze26 outlines that in the case of smaller convective than
frequency f obtained from scaling with the thermal diffusion diffusive transport, the power spectrum S T of a passive scalar
time t 0 . Dimensional values of the frequency are given from quantity such as the temperature scales as

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
Phys. Fluids, Vol. 13, No. 11, November 2001 Rayleigh–Bénard convection at vertical magnetic field 3255

FIG. 13. The characteristic slopes of power spectra S of OHD flow Q⫽0 FIG. 14. Power spectra S x , S y , and S z of the temperature gradient ⵜT ⬘P ,
and MHD flows Q⫽400 and Q⫽1600 共divided by factors 103 and 104 ) for recorded by the probe. The spectra are assorted according to the three spatial
a constant Rayleigh number Ra⫽5.0⫻104 . directions x, y, and z where x and z are the horizontal directions, perpen-
dicular to the magnetic field and y is the vertical direction aligned with
gravity and the magnetic field. At a fixed Rayleigh number of Ra⫽5
⫻104 , the Chandrasekhar number is increased from Q⫽0 to Q⫽400 and
S T 共 f ,t 兲 ⫽H• ␬ ⫺3 •S v 共 f ,t 兲 • f ⫺4 , 共14兲 Q⫽1600.

where H is constant and S v is the energy spectrum of the

velocity field. In a nondissipative, purely convective range of
the velocity field, S v can be assumed to be constant and the 12. The spectra are assorted according to the three spatial
power spectra of the passive scalar quantity decays purely on directions x, y, and z, where x and z are the horizontal direc-
a diffusion process as S T ( f ,t)⬃ f ⫺4 . This power law is in tions perpendicular to the magnetic field and y is the vertical
good agreement with our measurements in the frequency direction aligned with gravity and the magnetic field. Similar
range f ⬎10. Similar findings were reported by Horanyi to the observations concerning the power spectra of the tem-
et al.8 from measurements in heated sodium layers. If the perature signals in Fig. 12, the spectral energy of all three
above argument holds, the less steep decrease of the tem- components of the temperature gradient are strongly reduced
perature spectrum in the range 1⬍ f ⬍10 is associated with by the magnetic field. However, nonisotropic electromag-
an increase of the velocity power spectrum as a power f 2 . To netic damping is observed for Q⫽400 共gray curves兲. The
confirm this conjecture, additional velocity measurements enhanced spectral density of temperature fluctuations ob-
are necessary. served for this particular intensity of the magnetic field in the
The power spectra of MHD flows at Q⫽400 and Q region f ⲏ1.8 in Fig. 12 is reflected here only in the horizon-
⫽1600 are essentially flat in the frequency range below f tal components of the power spectra of the temperature gra-
⬇1. In an intermediate frequency range of 1ⱗ f ⱗ10, the dient. The vertical direction is not affected by this phenom-
slopes are decreasing continuously, but finally constant ena.
slopes are obtained. For Q⫽400 we obtain above a fre- The similarity of the power spectra of the two horizontal
quency f ⬇20 a decay law as f ⫺5 and for Q⫽1600 the decay directions already suggests that there is no predominant ori-
is even steeper; above f ⬇10 the spectra falls off as f ⫺6 . As entation of time-dependent flow structures in the horizontal
convective heat transport is even more reduced compared to plane and the flow may therefore be called horizontally iso-
OHD flow, Eq. 共14兲 still holds. It is conjectured that the tropic. This property may be proven more thoroughly by
steeper decay of the spectrum is caused by a reduction of scatter plots of the fluctuating parts of the horizontal tem-
power density of the velocity field S v related to Joule dissi- perature gradient vector ⵜxz T ⬘P ⫽( ⳵ x T ⬘P , ⳵ z T ⬘P ). Figures
pation. 15共a兲–15共d兲 show such scatter plots extracted from the same
Because of the noncoplanar arrangement of the four measurements as the power spectra in Fig. 12 are evaluated
thermocouples of the probe 共see Fig. 4兲, the fluctuating part from. For all magnetic fields, including OHD flow, almost
of the temperature gradient vector ⵜT ⬘P can be measured and perfectly circular distributions of the values of the horizontal
the power spectra of the temperature gradients S x , S y , and temperature gradient are found and all flows behave horizon-
S z in all three spatial directions can be evaluated. In Fig. 14 tally isotropic. In contrast to that, the scatter plots of the two
such spectra are plotted for the same parameters as in Fig. vertical temperature gradient vectors ⵜxy T ⬘P ⫽( ⳵ x T ⬘P , ⳵ y T ⬘P )

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
3256 Phys. Fluids, Vol. 13, No. 11, November 2001 U. Burr and U. Müller

FIG. 15. Scatter plots of the fluctuating parts of the horizontal temperature
gradient ⵜxz T ⬘P ⫽( ⳵ x T ⬘P , ⳵ z T ⬘P ). At a Rayleigh number of Ra⫽5.0⫻104 . The
Chandrasekhar number is increased from 共a兲 OHD flow at Q⫽0 to values
共b兲 Q⫽400, 共c兲 Q⫽1600, and 共d兲 Q⫽3600.

and ⵜz,y T ⬘P ⫽( ⳵ z T ⬘P , ⳵ y T ⬘P ), shown in Figs. 16共a兲–16共f兲, be-

come less circular from the action of the magnetic field and
thereby demonstrate that fluctuations in the direction of the
magnetic are stronger reduced. These findings are consistent
FIG. 16. Scatter plots of the fluctuating parts of the vertical temperature
with the well-known tendency of the magnetic field to re-
gradient vectors ⵜxy T ⬘p ⫽( ⳵ x T ⬘P , ⳵ y T ⬘P ) 共left column兲 and ⵜzy T ⬘p
move velocity gradients along its direction as it is known ⫽( ⳵ z T ⬘P , ⳵ y T ⬘P ) 共right column兲. At a Rayleigh number of Ra⫽5.0⫻104 . The
from MHD turbulence and other vortical flows 共see, e.g., Chandrasekhar number is increased from 共a兲, 共b兲 OHD flow at Q⫽0 to
Someria and Moreau27 and Davidson28兲. values 共c兲, 共d兲 Q⫽400 and 共e兲, 共f兲 Q⫽1600.

IV. CONCLUDING REMARKS rived from the measurements. However, it is suggested from
general considerations that for large Rar the Nusselt numbers
The basic physical mechanisms of the interaction of scale as Nu⬀(Ra/Q) 2/3.
natural convection in a horizontal liquid metal layer with an From the variances of the temperature fluctuations, it is
imposed vertical magnetic field and its impact on the heat concluded that in liquid–metal MHD convection, significant
transfer are evaluated in a well-defined experimental appara- convective heat transport is associated with time-dependent
tus. It is well known from a linear stability analysis, that for flow. Stationary convective flow is practically not observed.
Rayleigh–Bénard convection in liquid metals with an im- At large supercritical Rayleigh numbers the intensity of the
posed vertical magnetic field, convective motions are temperature fluctuations is always decreased by an increas-
strongly inhibited and that the horizontal length scales of the ing of the intensity of the magnetic field. Thus, the magnetic
convective structures are considerably reduced. The mea- field always damps the intensity of fluctuations.
sured Nusselt numbers confirm these findings and show fur- The integral time scale of MHD flow is significantly
ther that the magnetic field exerts a systematic damping ef- shorter than for OHD flow and is continuously decreasing
fect on the heat transport up to large supercritical Rayleigh with increasing Rayleigh number. It scales with the super-
numbers, i.e., for a constant Rayleigh number Ra the Nusselt critical Rayleigh number Ras ⫽Ra⫺Rac as ␶ I ⬃Ra⫺1 c . The
number Nu decreases when the Chandrasekhar number Q energy spectra of temperature fluctuations suggest that the
increases. For magnetohydrodynamic 共MHD兲 convection at damping of fluctuations by the magnetic field can be made
moderate supercritical Rayleigh numbers, 0.5⬍Rar ⬍2 the selective with respect to frequency. Long term fluctuations
Nusselt and Rayleigh numbers are correlated by Nu⫺1 are damped more strongly by the magnetic field than short
⫽0.066 Rar2/3 , where Rar ⫽Ra/Rac ⫺1 is the reduced Ray- term fluctuations, which, compared to OHD flow, are even
leigh number. This implies that below Rar ⬇0.5, no signifi- enhanced.
cant convective heat transport is observed. In an intermediate From an evaluation of the energy spectra and the statis-
range of 2⬍Rar ⬍10, no unique scaling law could be de- tical properties of the time-dependent temperature gradient it

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
Phys. Fluids, Vol. 13, No. 11, November 2001 Rayleigh–Bénard convection at vertical magnetic field 3257

is concluded that the convective flow remains horizontally 11

J. A. Shercliff, A Textbook of Magnetohydrodynamics 共Pergamon, Oxford,
isotropic for all magnetic field intensities investigated up to 1965兲.
12
J. S. Walker, ‘‘Magnetohydrodynamic duct flows in rectangular ducts with
date.
thin conducting walls I,’’ J. Mec. 20, 79 共1981兲.
13
S. Chandrasekhar, Hydrodynamic and Hydromagnetic Stability 共Dover,
ACKNOWLEDGMENTS New York, 1961兲.
14
Y. Nakagawa, ‘‘An experiment on the inhibition of thermal convection by
The authors are grateful to Dr. L. Barleon for his con- a magnetic field,’’ Nature 共London兲 175, 417 共1955兲.
tinuous advice. Without his technical support and experience 15
Y. Nakagawa, ‘‘Experiments on the inhibition of thermal convection by a
the experiments could not have been conducted successfully. magnetic field,’’ J. Sci. Earth 4, 108 共1956兲.
16
Special thanks are given to K.-J. Mack for his unique tech- Y. Nakagawa, ‘‘Experiments on the stability of a layer of mercury heated
from below and subject to the simultaneous action of a magnetic field and
nical and cooperative assistance in overcoming severe prob-
rotation,’’ Proc. R. Soc. London, Ser. A 249, 138 共1959兲.
lems and for his patience during long measuring campaigns. 17
F. H. Busse and R. M. Clever, ‘‘Stability of convection rolls in the pres-
The experiments have been performed in the MEKKA facil- ence of a vertical magnetic field,’’ Phys. Fluids 25, 931 共1982兲.
ity of the Institut für Kern- und Energietechnik 共IKET兲 of the 18
F. H. Busse and R. M. Clever, ‘‘Three-dimensional convections in the
Forschungszentrum Karlsruhe. presence of strong vertical magnetic field,’’ Eur. J. Mech. B/Fluids 15, 1
共1996兲.
19
1
J. O. O’Donnell, P. G. Papanicolaou, and C. B. Reed, ‘‘The thermophysi-
P. A. Davidson, ‘‘Magnetohydrodynamics in materials processing,’’ Annu. cal and transport properties of eutectic NaK near room temperature,’’
Rev. Fluid Mech. 31, 273 共1999兲. Technical report, Argonne National Laboratory ANL/FPP/TM-237, 1989.
2
A. Pellew and R. V. Southwell, ‘‘On maintained convective motion in a 20
R. N. Lyon, Liquid Metals Handbook, 2nd ed., Washington, Atomic En-
fluid heated from below,’’ Proc. R. Soc. London, Ser. A 176, 312 共1940兲.
3 ergy Commission 共Department of the Navy, Washington, DC, 1952兲.
W. H. Reid and D. L. Harris, ‘‘Some further results on the Bénard prob- 21
O. J. Foust, Sodium–NaK Engineering Handbook 共Gordon and Breach,
lem,’’ Phys. Fluids 1, 102 共1958兲.
4 New York, 1972兲, Vol. 1.
R. Krishnamurti, ‘‘Some further studies on the transition to turbulent con- 22
L. Davoust, M. D. Cowley, R. Moreau, and R. Bolcato, ‘‘Buoyancy-driven
vection,’’ J. Fluid Mech. 60, 285 共1973兲.
5
R. M. Clever and F. H. Busse, ‘‘Transition to time dependent convection,’’ convection with a uniform magnetic field. Part 2. Experimental investiga-
J. Fluid Mech. 65, 625 共1974兲. tion,’’ J. Fluid Mech. 400, 59 共1999兲.
23
6
F. H. Busse, ‘‘Non-linear properties of thermal convection,’’ Rep. Prog. R. M. Clever and F. H. Busse, ‘‘Nonlinear oscillatory convection in the
Phys. 41, 1929 共1978兲. presence of a vertical magnetic field,’’ J. Fluid Mech. 201, 507 共1989兲.
24
7
T. Y. Chu and R. J. Goldstein, ‘‘Turbulent convection in a horizontal layer R. Krishnamurti, ‘‘On the transition to turbulent convection. Part 2. The
of water,’’ J. Fluid Mech. 60, 141 共1973兲. transition to time-dependent flow,’’ J. Fluid Mech. 42, 309 共1970兲.
25
8
S. Horanyi, L. Krebs, and U. Müller, ‘‘Turbulent Rayleigh–Bénard con- G. K. Batchelor, ‘‘Small-scale variation of convected quantities like tem-
vection in low Prandtl-number fluids,’’ Int. J. Heat Mass Transf. 42, 3983 perature in turbulent fluid,’’ J. Fluid Mech. 5, 113 共1959兲.
共1999兲.
26
J. O. Hinze, Turbulence 共McGraw-Hill, New York, 1975兲.
9 27
H. T. Rossby, ‘‘A study of Bénard convection with and without rotation,’’ J. Sommeria and R. Moreau, ‘‘Why, how and when, MHD turbulence
J. Fluid Mech. 36, 309 共1969兲. becomes two-dimensional,’’ J. Fluid Mech. 118, 507 共1982兲.
10 28
D. C. Threlfall, ‘‘Free convection in low-temperature gaseous helium,’’ J. P. A. Davidson, ‘‘Magnetic damping of jets and vortices,’’ J. Fluid Mech.
Fluid Mech. 67, 17 共1975兲. 299, 153 共1995兲.

Downloaded 07 May 2012 to 132.236.27.111. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions