WorksPhysics and Astronomy
Ultimate regime of Rayleigh-Benard turbulence: Sub-regimes and their scaling relations for Nu vs. Ra and Pr
Abstract
We offer a new model for the heat transfer and the turbulence intensity in strongly driven Rayleigh-Benard turbulence (the so-called ultimate regime), which in contrast to hitherto models is consistent with the new mathematically exact heat transfer upper bound of Choffrut et al. [J. Differential Equations 260, 3860 (2016)] and thus enables extrapolations of the heat transfer to geo- and astrophysical flows. The model distinguishes between four subregimes of the ultimate regime and well describes the measured heat transfer in various large-Ra experiments. In this new representation, which properly accounts for the Prandtl number dependence, the onset to the ultimate regime is seen in all available large-Ra data sets, though at different Rayleigh numbers, as to be expected for a non-normal-nonlinear instability.
Knowing the heat and/or mass transfer in large-scale turbulent flows is of utmost importance for many questions in climate research, in geophysical or astrophysical systems, or in industrial flow systems. Examples are thermally driven flows in the ocean, in the atmosphere [1], or in the outer core of Earth, other planets, or stars [2]. In all these cases, very strong turbulence is achieved, due to the strong thermal driving. For such systems, however, direct measurements under controlled conditions are not feasible, and neither are direct numerical simulations, given the many degrees of freedom of such systems, though the underlying dynamical equations (the advection-diffusion equations for the temperature and/or the mass transport, coupled to the Navier–Stokes equations) are known. Given this, in order to get an estimate for the heat and/or mass transfer in such systems, one has to rely on more controlled model systems on much smaller scale and then extrapolate towards larger systems with stronger thermal driving.
The most popular controlled model system for heat transfer is the Rayleigh–Bénard (RB) system, consisting of a container of height filled with fluid, heated from below and cooled from above [3, 4, 5, 6, 7, 8, 9, 10, 11]. The control parameters of this thermally driven convective flow are the Rayleigh number (the dimensionless temperature difference between top and bottom plates, as measure of the driving strength), the Prandtl number (the ratio between kinematic viscosity and thermal diffusivity ), and the aspect ratio (the width of the system divided by its height). The main global response parameters are the Nusselt number (the dimensionless heat transfer from bottom to top) and the so-called wind Reynolds number , which quantifies the velocity of the large scale convective flow. The key question is: How do the Nusselt and the Reynolds number depend on the control parameters, and ? For not too strong thermal driving (the so-called classical regime) this question can meanwhile be answered and there is good agreement between various experiments and numerical simulations and a good understanding of the flow physics, namely in terms of the “Grossmann–Lohse-theory” or in short “GL-theory”, cf. [12, 13, 14, 15], see also the reviews [6, 11].
This is not so for very strong thermal driving, i.e., for the regime of very large Rayleigh numbers, which is called the “ultimate regime” and to which, for large enough , the RB system is believed to undergo a transition of non-normal–nonlinear type [11, 16], as typical in strongly driven sheared wall-bounded flows [17] and as here is the case in the boundary layers. It is this very ultimate regime which is relevant for climate research and the above mentioned geophysical and astrophysical convective flows, due to the very strong thermal driving in these systems. Therefore, extrapolations from the classical regime to the ultimate regime are required. Typically, these extrapolations are sought for as scaling laws, but this only makes sense once there is no transition towards a different state of turbulence. If there is such a transition, the extrapolation with a simple scaling law becomes meaningless. But then, how to upscale the RB system and how to understand and predict the heat (and mass) flux for very large , as it occurs in geo- and astrophysical applications?
To answer these questions, various theoretical heuristic models of different degrees of rigor have been developed, based on some assumptions and speculations on the flow physics in this ultimate regime [18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29]. The most famous and influential one of all these models may be the one by Kraichnan [20], who for very large and small suggested . For very large and moderate , he suggested a slightly different -dependence, namely .
All these models obviously should obey the mathematically strict upper bounds for , which can be derived from the underlying dynamical equations (heat transfer equation and Navier–Stokes equations in their Boussinesq approximation). The well-known upper bound was found already in the second half of the last century [30, 31, 32, 33] and the best-known (smallest) prefactor was calculated in [34]. Although this upper bound leads to values much higher than the experimentally measured or numerically calculated , it excludes the universality of the scaling relation (with any logarithmic corrections). This scaling relation was proposed in several models for the ultimate regime [24, 27, 28], but due to the upper bound it can only hold for . Moreover, in 2016 Choffrut et al. [35] succeeded to sharpen the upper bound in a large- subrange of the ultimate regime, namely to for (all subject to logarithmic corrections). Thus in this subregime of the ultimate regime, in which grows faster than but slower than (that is, as , ), Kraichnan’s model predicts with , in direct contradiction to Choffrut et al.’s strict upper bound. Similarly, also other models [24, 27, 28], which propose the growth of faster than for the ultimate regime, cannot hold in this subrange of the – parameter space.
This discrepancy calls for revisiting the suggested scaling laws in the ultimate regime, in view of the improved and sharpened strict upper bound [35]. To do so is the objective of this paper. We will first suggest a new model for the heat transfer in the ultimate regime, which is based on the flow physics in the turbulent boundary layers and which respects the new mathematically strict upper bounds in the various subregimes of the ultimate regime in the parameter space. We will then show that the available experimental and numerical data for strongly driven RB convection can be well described with our new model.
We start with the boundary layer equations for the horizontal velocity and temperature in a turbulent flow next to a rigid horizontal wall:
(where denotes time, and , respectively, the horizontal and vertical coordinates, the hydrodynamic pressure, and the density) and conduct the Reynolds decomposition of the flow components into their time-averages and fluctuations: , .
We assume that the flow is highly turbulent, so that the convective contributions from the mean, time-averaged flow are negligible compared to the contributions from the Reynolds stresses, i.e. and , so that the following relations hold: , . Using these relations, we average Eqs. (1), (2) in time and over a horizontal cross-section (i.e., apply ) under further natural assumptions that the long averages in time of the temporal derivatives vanish, , , as well as the averages in the horizontal direction of the horizontal derivatives, , , , , . With this we obtain the following reduced equations:
Integrating equations (3)–(4) from 0 to , introducing the eddy viscosity and the eddy thermal diffusivity ,
and taking into account the vanishing fluctuations at the plate and that the Nusselt number is defined by , we obtain
for (the square of) the friction velocity and
To close the system (7) and (8), we need to know the functional dependences of the eddy viscosity and the eddy thermal diffusivity . Near the plate, within the viscous sublayer of the thickness , both the eddy viscosity and the eddy thermal diffusivity , behave as cubic functions of the distance from the plate, [36, 37, 38, 39], and therefore the contribution of the eddy viscosity and eddy thermal diffusivity within the viscous sublayer is negligible.
To estimate the mean vertical profiles of , and outside the viscous sublayer, we follow Landau [40] and assume that (i) the turbulent Prandtl number is independent of (or only weakly dependent on) the molecular Prandtl number and (ii) that the mean vertical profiles of , , and , are determined exclusively by the momentum transferred by the fluid to a solid wall, i.e. the friction velocity , and by the distance to the plate, subject to a certain Prandtl-number dependence, i.e. .
These assumptions, by dimensional analysis, imply that, outside the viscous sublayer, , , and should scale as
with some positive constants , , and .
We further propose that the turbulent diffusivities and are controlled by the smallest of the two fluid characteristics of diffusion, i.e. either by or by . In other words, both, and , should be proportional to for small , and to for large , which implies for and for . This proposition is related to the fact that if any of the two quantities, or , equals zero, convection is fully suppressed, , independently from which value takes the other characteristics of diffusion ( or , respectively).
Now, from the exact relation for the time- and volume-averaged kinetic energy dissipation rate [see, e.g., 26] and Eqs. (7)–(11) we can derive the scaling relations, which we propose for the ultimate regime, for both small and large : Dividing both sides of Eq. (7) by , substituting (9), and integrating the resulting equation in from the edge of the viscous sublayer, , to the location of the maximal wind velocity we obtain
Analogously, dividing both sides of Eq. (8) by , substituting (10), and integrating the resulting equation in from to , we obtain
In Eqs. (12) and (13) behind the second tilde-sign we have neglected the -dependences and keep only the leading terms in the log-corrections. Next we consider the profile of the kinetic energy dissipation rate. As , the contribution to the mean kinetic energy dissipation rate from the viscous sublayer is smaller than . In contrast, the contribution from the core part of the domain is scalingwise larger, as one can see from integrating (11),
Here we used the fact that for all Prandtl numbers. Since the main contribution to the total kinetic energy dissipation rate comes from the bulk, with the exact relation for we obtain
From relations (14) and (15) it follows
Combining (12), (13) and (16) we obtain
and
with for and for . Note that in relations (17) and (18) we again neglected the -dependences in the logarithmic corrections. Thus, finally we obtain the following scaling relations for the heat transport
The derived scaling relation (20) is the same as in Grossmann and Lohse’s model for the ultimate regime [28]; therefore as in that paper we call it “regime ”. Relation (19) gives an extension of that model towards large and we call that regime “regime ”. The derived regimes are sketched in Figure 1.
Figure 1. A sketch of the proposed scaling relations in the ultimate regime of Rayleigh–Bénard convection in the – parameter space, where the ultimate regime is split into the subregimes , , and . The numbers in color boxes show the scaling exponents in the relations , (subject to logarithmic corrections). The straight lines indicate the slopes of the transitions between the neighbouring regimes, , where the values of are written next to the lines. The dotted line indicates where the laminar kinetic boundary layer is expected to become turbulent (i.e., where the shear Reynolds number achieves a critical value, ).
The transition between the scaling regimes and takes place at a constant , where grows slightly slower than as (see the horizontal line for in the – plane in Figure 1, which indicates the transition between the neighbouring regimes and ). Another boundary for the regime is for (marked with a blue line in Figure 1). While moving along this line for an increasing and , the Nusselt number remains constant, and any steeper transition slope from regime would imply an unphysical limit along that line. The blue line in Figure 1 indicates the slope of the transition to the regime , which has the very same scaling exponents for and as the classical regime of the GL-theory [12].
Analogously, one can conclude that the slope of the upper boundary of the regime should not be steeper than so that along this line the Nusselt number remains constant for increasing . However, as we will explain below, the transition from regime has a significantly more gentle slope, namely with (marked with a pink line in the sketch of Figure 1).
Indeed, for the no-slip boundary conditions, each component of the velocity field vanishes at the (Lipschitz) boundary of the domain, and therefore the Friedrichs inequality holds, where is the lowest (positive) eigenvalue of the Laplace operator in the considered domain with the corresponding boundary conditions, that depends only on the geometrical characteristics and has the dimension of inverse squared length [10].
Therefore, for any RB flow
where is based on the kinetic energy . In regime (for small ), this relation is always fulfilled within the discussed boundaries, since (21) then means . However, in regime , the requirement (21) means , as it follows from the combination of the relations (17), (19) with (21). Therefore, regime can exist only for . This is consistent with the upper bound of Choffrut et al. [35], who, as mentioned above, derived that the upper bounds for the heat transport for large Prandtl numbers cannot exceed (all up to logarithmic corrections).
While moving along the line with increasing (the red line in Figure 1), the Nusselt number effectively scales as . We assume that this transition line connects regime with regime (as we call it), in which the scaling exponents are exactly the same as in the classical regime of the GL-theory [12], namely . This result is again consistent with the new strict upper bound of Choffrut et al. [35].
The sketch in Figure 1 summarizes the four subregimes of the ultimate regime, namely , , and , which all can be interpreted as ultimate in the sense that one can approach infinite within these regimes. All these subregimes lie to the right of the pink dotted line that indicates a constant associated with the onset of a turbulent boundary layer. In the regimes and the scaling exponent in the relation is, however, smaller than . Regimes and can be considered as the “true” ultimate regimes in the sense that only there .
The proposed model thus indeed suggests that the scaling exponent in the scaling relation can be asymptotically achieved within the regimes and , but only for almost constant Prandtl numbers. As soon as changes as a power law of (with some small , and here it does not matter whether is positive or negative), one should expect an asymptotic reduction of the effective scaling exponent as .
We now want to compare the available experimental data for large (close to or in the ultimate regime) with the model Eqs. (19)-(20), cf. Figure 1. According to the model, in the ultimate regime (for not extremely small or extremely large ), the following scaling should hold: , where a negative exponent () for should be taken for large and a positive one () for small . Thus, the Nusselt number is a function of with for and for . In Figure 2a, the considered experimental data for are plotted vs. . One clearly sees that all data sets, including the Oregon data, follow a scaling close to with for smaller values of and between 0.4 and 0.5 at the highest values of . The exact onset value of the steeper scaling varies from experiment to experiment, which is consistent with the view that the transition to the ultimate regime is of non-normal–nonlinear nature [11, 16].

Figure 2. (a) Nusselt number vs. (with for and for ) and (b) compensated Nusselt number vs. (where the function smoothly connects the two regimes for and for ), as obtained in the various different RB experiments of refs. [16, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50] under (nearly) Oberbeck–Boussinesq conditions in cylindrical containers, distinguished by the aspect ratio and where it was done. The blue curve shows the predictions of the GL-theory for the classical regime, and . All data sets at the highest achieved Rayleigh numbers show the transition to the ultimate regime, with slopes about (brown, cyan, green, pink and magenta thin lines in (b)).
For better visibility of the onset, Figure 2b provides a compensated plot of vs. . Here, is a function of that substitutes the discontinuous change of from to at by a smooth function that matches the small- and large- regimes. (Here, of course, different options are possible to match the small- and large- regimes, in particular, by optimizing the constant , which in Figure 2b
equals .) Again, all data show a transition for very large . The inclined lines in Figure 2b highlight the scaling exponent . We interpret the results in Figure 2b as support for our new model for the ultimate regime and its subregimes.
In summary, we have suggested a new model for the heat transfer in the ultimate regime of RB turbulence, which distinguishes between four subregimes of the ultimate regime and which for each of these subregimes gives the scaling relations as shown in Figure 1. In contrast to prior models, it obeys the mathematically strict upper bounds of Choffrut et al. [35]. It moreover is consistent with the experimental data on of the various large- RB experiments of refs. [16, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50]. We emphasize again that in this new representation, which take the -dependence into account, the onset of the ultimate regime is seen in all data sets, though at different numbers, as to be expected for a non-normal–nonlinear instability. Our model thus offers a reliable basis to estimate the heat transfer for systems with even larger , which cannot be achieved in today’s experiments, and for geophysical and astrophysical systems.
Acknowledgements: We thankfully acknowledge all colleagues who have made their experimental data available to us and to the whole community.
References
References
- [1]G. K. Vallis, Atmospheric and oceanic fluid dynamics (Cambridge University Press, Cam- bridge, 2017).
- [2]C. Clarke and B. Carswell, Principles of astrophysical fluid dynamics (Cambridge University Press, Cambridge, 2007).
- [3]S. Chandrasekhar, Hydrodynamic and Hydromagnetic Stability (Clarendon Press, ADDRESS, 1961).
- [4]E. Bodenschatz, W. Pesch, and G. Ahlers, Recent developments in Rayleigh–Bénard convec- tion, Annu. Rev. Fluid Mech. 32, 709 (2000).DOI
- [5]L. P. Kadanoff, Turbulent heat flow: Structures and scaling, Phys. Today 54, 34 (2001).DOI
- [6]G. Ahlers, S. Grossmann, and D. Lohse, Heat transfer and large scale dynamics in turbulent Rayleigh–Bénard convection, Rev. Mod. Phys. 81, 503 (2009).DOI
- [7]D. Lohse and K.-Q. Xia, Small-scale properties of turbulent Rayleigh–Bénard convection, Annu. Rev. Fluid Mech. 42, 335 (2010).DOI
- [8]F. Chillà and J. Schumacher, New perspectives in turbulent Rayleigh–Bénard convection, Eur. Phys. J. E 35, 58 (2012).DOI
- [9]K.-Q. Xia, Current trends and future directions in turbulent thermal convection, Theor. Appl. Mech. Lett. 3, 052001 (2013).
- [10]O. Shishkhina, Rayleigh-Bénard convection: The container shape matters, Phys. Rev. Fluids 6, 090502 (2021).DOI
- [11]D. Lohse and O. Shishkina, Ultimate turbulent thermal convection, Phys. Today 76, 26 (2023).DOI
- [12]S. Grossmann and D. Lohse, Scaling in thermal convection: A unifying theory, J. Fluid Mech. 407, 27 (2000).DOI
- [13]S. Grossmann and D. Lohse, Thermal convection for large Prandtl numbers, Phys. Rev. Lett. 86, 3316 (2001).DOI
- [14]S. Grossmann and D. Lohse, Prandtl and Rayleigh number dependence of the Reynolds number in turbulent thermal convection, Phys. Rev. E 66, 016305 (2002).pubmed.ncbi.nlm.nih.gov/12241479
- [15]R. J. A. M. Stevens, E. P. van der Poel, S. Grossmann, and D. Lohse, The unifying theory of scaling in thermal convection: The updated prefactors, J. Fluid Mech. 730, 295 (2013).DOI
- [16]P.-E. Roche, The ultimate state of convection: a unifying picture of very high Rayleigh numbers experiments, New J. Phys. 22, 073056 (2020).DOI
- [17]M. Avila, D. Barkley, and B. Hof, Transition to turbulence in pipe flow, Annu. Rev. Fluid Mech. 55, 575 (2023).
- [18]C. H. B. Priestley, Convection from a large horizontal surface, Australian J. Phys. 7, 176 (1954).DOI
- [19]M. V. R. Malkus, The heat transport and spectrum of thermal turbulence, Proc. R. Soc. London A 225, 196 (1954).DOI
- [20]R. Kraichnan, Turbulent thermal convection at arbitrary Prandtl number, Phys. Fluids 5, 1374 (1962).DOI
- [21]J. R. Herring, Investigation of problems in thermal convection, J. Atmos. Sci. 20, 325 (1963).doi.org/10.1175/1520-0469(1963)020<0325:iopitc>2.0.co;2
- [22]K. Stewartson, On almost rigid rotation. Part 2, J. Fluid Mech. 26, 131 (1966).DOI
- [23]P. H. Roberts, in Non-Equilibrium Thermodynamics, Variational Techniques and Stability, edited by R. D. et al. (University of Chicago Press, ADDRESS, 1966), pp. 125–162.
- [24]E. A. Spiegel, Convection in stars, I. Basic Boussinesq convection, Annu. Rev. Astron. Astrophys. 9, 323 (1971).DOI
- [25]B. Castaing, G. Gunaratne, F. Heslot, L. Kadanoff, A. Libchaber, S. Thomae, X.-Z. Wu, S. Zaleski, and G. Zanetti, Scaling of hard thermal turbulence in Rayleigh–Bénard convection, J. Fluid Mech. 204, 1 (1989).DOI
- [26]B. I. Shraiman and E. D. Siggia, Heat transport in high-Rayleigh-number convection, Phys. Rev. A 42, 3650 (1990).pubmed.ncbi.nlm.nih.gov/9904456
- [27]X. Chavanne, F. Chillà, B. Castaing, B. Hebral, B. Chabaud, and J. Chaussy, Observation of the ultimate regime in Rayleigh–Bénard convection, Phys. Rev. Lett. 79, 3648 (1997).DOI
- [28]S. Grossmann and D. Lohse, Multiple scaling in the ultimate regime of thermal convection, Phys. Fluids 23, 045108 (2011).DOI
- [29]S. Grossmann and D. Lohse, Logarithmic temperature profiles in the ultimate regime of thermal convection, Phys. Fluids 24, 125103 (2012).DOI
- [30]L. N. Howard, Heat transport by turbulent convection, J. Fluid Mech. 17, 405 (1963).DOI
- [31]F. H. Busse, On Howard’s upper bound for heat transport by turbulent convection, J. Fluid Mech. 37, 457 (1969).DOI
- [32]L. N. Howard, Bounds on flow quantities, Annu. Rev. Fluid Mech. 4, 473 (1972).DOI
- [33]C. R. Doering and P. Constantin, Variational bounds on energy dissipation in incompressible flows. III. Convection, Phys. Rev. E 53, 5957 (1996).pubmed.ncbi.nlm.nih.gov/9964954
- [34]S. C. Plasting and R. R. Kerswell, Improved upper bound on the energy dissipation rate in plane Couette flow: the full solution to Busse’s problem and the Constantin–Doering–Hopf problem with one-dimensional background field, J. Fluid Mech. 477, 363 (2003).DOI
- [35]A. Choffrut, C. Nobili, and F. Otto, Upper bounds on Nusselt number at finite Prandtl number, J. Differential Equations 260, 3860 (2016).arxiv.org/abs/1412.4812
- [36]R. A. Antonia, Behaviour of the turbulent Prandtl number near the wall, Int. J. Heat Mass Transfer 23, 906 (1980).DOI
- [37]R. A. Antonia and J. Kim, Turbulent Prandtl number in the near-wall region of a turbulent channel flow, Int. J. Heat Mass Transfer 34, 1905 (1991).DOI
- [38]O. Shishkina, S. Horn, S. Wagner, and E. S. C. Ching, Thermal boundary layer equation for turbulent Rayleigh–Bénard convection, Phys. Rev. Lett. 114, 114302 (2015).DOI
- [39]O. Shishkina, S. Horn, M. S. Emran, and E. S. C. Ching, Mean temperature profiles in turbulent thermal convection, Phys. Rev. Fluids 2, 113502 (2017).DOI
- [40]L. D. Landau and E. M. Lifshitz, Fluid Mechanics, Vol. 6 of Course of Theoretical Physics, 2 ed. (Butterworth Heinemann, Exeter, 1987).
- [41]X. He, D. Funfschilling, H. Nobach, E. Bodenschatz, and G. Ahlers, Transition to the ultimate state of turbulent Rayleigh–Bénard convection, Phys. Rev. Lett. 108, 024502 (2012).pubmed.ncbi.nlm.nih.gov/22324688
- [42]X. He, D. Funfschilling, E. Bodenschatz, and G. Ahlers, Heat transport by turbulent Rayleigh–Bénard convection for Pr ∼ 0.8 and 4 × 10¹¹ ≲ Ra ≲ 2 × 10¹⁴: Ultimate-state transition for aspect ratio Γ = 1.00, New J. Phys. 14, 063030 (2012).
- [43]J. J. Niemela, L. Skrbek, and R. J. Donnelly, Ultra-high Rayleigh number convection in cryogenic helium gas, Physica B 284–288, 61 (2000).DOI
- [44]J. J. Niemela, L. Skrbek, K. R. Sreenivasan, and R. J. Donnelly, Turbulent convection at very high Rayleigh numbers, Nature 404, 837 (2000).pubmed.ncbi.nlm.nih.gov/10786783
- [45]J. J. Niemela, L. Skrbek, C. Swanson, S. Hall, K. R. Sreenivasan, and R. J. Donnelly, New results in cryogenic helium flows at ultra-high Reynolds and Rayleigh numbers, J. Low Temp. Phys. 121, 417 (2000).
- [46]P.-E. Roche, F. Gauthier, R. Kaiser, and J. Salort, On the triggering of the Ultimate Regime of convection, New J. Phys. 12, 085014 (2010).DOI
- [47]J. J. Niemela and K. R. Sreenivasan, Confined turbulent convection, J. Fluid Mech. 481, 355 (2003).
- [48]J. J. Niemela and K. R. Sreenivasan, Rayleigh-number evolution of large-scale coherent motion in turbulent convection, Europhys. Lett. 62, 829 (2003).DOI
- [49]P. Urban, P. Hanzelka, V. Musilová, T. Králík, M. L. Mantia, A. Srnka, and L. Skrbek, Heat transfer in cryogenic helium gas by turbulent Rayleigh–Bénard convection in a cylindrical cell of aspect ratio 1, New J. Phys. 16, 053042 (2014).DOI
- [50]P. Urban, P. Hanzelka, T. Králík, M. Macek, V. Musilová, and L. Skrbek, Elusive transition to the ultimate regime of turbulent Rayleigh–Bénard convection, Phys. Rev. E 99, 011101 (2019).DOI