betaFeedback

WorksPhysics and Astronomy

Convective heat transfer in the Burgers-Rayleigh-Bénard system

Abstract

The dynamics of heat transfer in a model system of Rayleigh-Bénard (RB) convection reduced to its essential, here dubbed Burgers-Rayleigh-Bénard (BRB), is studied. The system is spatially one-dimensional, the flow field is compressible and its evolution is described by the Burgers equation forced by an active temperature field. The BRB dynamics shares some remarkable similarities with realistic RB thermal convection in higher spatial dimensions: i) it has a supercritical pitchfork instability for the onset of convection which solely depends on the Rayleigh number (Ra)(Ra) and not on Prandlt (Pr)(Pr), occurring at the critical value Rac=(2π)4Ra_c = (2π)^4 ii) the convective regime is spatially organized in distinct boundary-layers and bulk regions, iii) the asymptotic high RaRa limit displays the Nusselt and Reynolds numbers scaling regime Nu=RaPr/4Nu = \sqrt{RaPr}/4 for Pr≪1Pr\ll 1, Nu=Ra/(4π)Nu=\sqrt{Ra}/(4\sqrtπ) for Pr≫1Pr\gg1 and Re=Ra/Pr/12Re = \sqrt{Ra/Pr}/\sqrt{12}, thus making BRB the simplest wall-bounded convective system exhibiting the so called ultimate regime of convection. These scaling laws, derived analytically through a matched asymptotic analysis are fully supported by the results of the accompanying numerical simulations. A major difference with realistic natural convection is the absence of turbulence. The BRB dynamics is stationary at any RaRa number above the onset of convection. This feature results from a nonlinear saturation mechanism whose existence is grasped by means of a two-mode truncated equation system and via a stability analysis of the convective regime.

Introduction

Thought experiments, toy-models, low-dimensional representations are keys to the scientific thinking, and allow to get insight into the complex physics of many real systems. In fluid-dynamics research, reduced models obtained, e.g., via expansion and truncations of the original dynamical equations have been used to conceptualize and to understand, for instance, the chaotic dynamics of flows (Lorenz system [1]), or the physics of energy cascade in developed turbulence (Shell models [2, 3]). In this study we focus on the problem of thermal convection and, in the spirit of the one-dimensional (1D) toy model for granular media introduced by Du, Li and Kadanoff [4, 5], we here introduce a stripped-down mock-up of the classical Rayleigh-Bénard (RB) system [6].

The RB has been extensively studied either in its spatial three-dimensional or in its two-dimensional version [7, 8]. We are not aware of any study of the system in one-dimension. This is after all quite understandable, since in 1D the incompressibility condition for the flow does not hold and one expects a rather different physical behaviour. This is indeed already the case for the 1D version of the Navier-Stokes equation, i.e., the Burgers equation [9]. It is well known that the Burgers equation does not display a turbulent behaviour, because it can be recast in term of a diffusion equation via the Hopf-Cole transformation [11] [10]. However, stochastically forced Burgers equation does produce a special kind of turbulence, dubbed Burgulence [12], that has drawn the attention of recent research [13].

We show in this study that a 1D deterministically-forced version of the RB system that we dub Burgers-Rayleigh-Bénard (BRB) system can be defined. Interestingly, this system possesses a certain number of similarities with thermal convection in higher spatial dimensions: it has a supercritical linear instability for the onset of convection, the convective regime is spatially organized in distinct boundary-layers and bulk regions, and the asymptotic high RaRa limit displays the so-called ultimate Nusselt and Reynolds number scalings [8], although it lacks of any turbulent behaviour. It also admits shock-like solutions that are peculiar of the Burgers dynamics.

The article is organized as follows. We first define the BRB system, next we examine its most relevant symmetries and its global properties. In particular, we introduce the definition of the Nusselt and Reynolds numbers, which are the two global response parameters of the system. Secondly, we perform a theoretical analysis on the system dynamics, focusing on the calculation of the linear instability threshold for convection, on the subsequent non-linear saturation mechanism and on a derivation of a steady matched asymptotic solution for the very intense convection state. Third, we push forward the analysis by means of a numerical approach. In particular we show that the system is stationary at all Rayleigh numbers, this is first revealed empirically then verified by means of numerically-based stability analysis. We then show that the Nusselt and Reynolds number asymptotically approaches the ultimate state of thermal convection, this both in their Rayleigh and Prandtl number dependencies. Finally, we discuss the implications of our findings and possible perspectives.

The Burgers-Rayleigh-Bénard model system

Equations of motion

We study the spatio-temporal evolution of a single-component velocity W(Z,τ)W(Z,\tau) and temperature T(Z,τ)T(Z,\tau) fields in a one-dimensional domain Z∈[0,H]Z \in[0,H], described by the coupled system of differential equations:

Wτ+WWZ=νWZZ+βg(T−Tc)(1)W_{\tau} + W W_{Z} = \nu W_{ZZ} + \beta g(T-T_{c}) \tag*{(1)}
Tτ+WTZ=κTZZ,(2)T_{\tau} + W T_{Z} = \kappa T_{ZZ}, \tag*{(2)}

with Dirichlet boundary conditions

W=0,T=Δ2in Z=0 (bottom),(3)W=0,\quad T=\frac{\Delta}{2}\quad\text{in } Z=0\text{ (bottom),} \tag*{(3)}
W=0,T=−Δ2in Z=H (top),(4)W=0,\quad T=-\frac{\Delta}{2}\quad\text{in } Z=H\text{ (top),} \tag*{(4)}

where ν\nu and κ\kappa denotes respectively the viscosity and thermal diffusivity, β\beta the thermal expansion coefficient, gg the gravitational acceleration intensity, and TcT_{c} the linear profile given by Tc(Z)=−(Δ/H)Z+Δ/2T_{c}(Z)=-(\Delta/H)Z+\Delta/2, which is also said conductive because it represents a solution for the temperature field when W=0W=0 in all the domain. Furthermore, to keep the similarity with realistic RB convection we adopt the additional constraint that the global value of velocity and temperature fields are null,

∫0HW dZ=∫0HT dZ=0(no-zero mode condition).(5)\int_{0}^{H} W\,\mathrm{d}Z=\int_{0}^{H} T\,\mathrm{d}Z=0\quad\text{(no-zero mode condition)}. \tag*{(5)}

This prevents the possibility for the system to acquire a vertical mean flow and to heat up/cool off. We will comment later on the consequence of this constraint.

As we have already mentioned, the above model constitutes an oversimplified representation of the Rayleigh-Bénard system. It can be loosely obtained from the Navier-Stokes-Boussinesq set of equations for the three-dimensional velocity U=(U,V,W)\mathbf{U}=(U,V,W) and temperature TT, by assuming that (i) the vertical component of the velocity, WW, and the temperature depend only on the vertical direction, ZZ, (ii) by removing the hydrodynamic pressure field and (iii) by expressing the buoyancy force as proportional to the temperature deviation from the local conductive temperature profile. With the above assumptions, the equations for TT and WW decouple from the ones for horizontal components U,VU,V and can be treated separately. As a consequence the vertical velocity gradient WZW_{Z} becomes unconstrained and the corresponding unidimensional velocity field is compressible. We stress that the BRB model can not be regarded as a low-dimensional mean-field form of the Boussinesq system, neither a model of convection in compressible gases (where the continuity equation would have a different form). However, we believe that despite its incompleteness this model is useful to get an insight in what does/does not occur in the realistic system.

The equations (1-2) can be made dimensionless by means of the linear size of the domain (or height HH), the free-fall velocity Uf=βgHΔU_{f}=\sqrt{\beta g H\Delta} and the global temperature gap Δ\Delta (i.e. the difference between the top temperature and the bottom one). This leads to the two control parameters in the system: the Rayleigh number Ra=(UfH)2/(νκ)Ra=(U_{f}H)^{2}/(\nu\kappa) and the Prandtl number Pr=ν/κPr=\nu/\kappa. With these choices the equations can be conveniently rewritten in term of the velocity, w=W/Ufw=W/U_{f}, and the temperature deviation from the conductive profile, θ(z,t)=(T(Z,τ)−Tc(Z))/Δ\theta(z,t)=(T(Z,\tau)-T_{c}(Z))/\Delta, as:

wt+wwz=PrRa wzz+θ(6)w_{t}+w w_{z}=\sqrt{\frac{Pr}{Ra}}\,w_{zz}+\theta \tag*{(6)}
θt+wθz=1PrRa θzz+w,(7)\theta_{t}+w\theta_{z}=\frac{1}{\sqrt{PrRa}}\,\theta_{zz}+w, \tag*{(7)}

with w=θ=0w=\theta=0 in z=0z=0 and z=1z=1 and ⟨w⟩=⟨θ⟩=0\langle w\rangle=\langle\theta\rangle=0, where ⟨⋯ ⟩=∫01⋯ dz\langle\cdots\rangle=\int_{0}^{1}\cdots\,\mathrm{d}z is the spatial average (all lowercase letter denote dimensionless variables). Equation (6) is the 1D forced Burgers equation, which is coupled to the advection-diffusion equation for a scalar field (7) which is in turn forced by ww.

Symmetries

The system (6-7) enjoys a series of symmetries which greatly affect its dynamics. We describe them in detail in this section. To begin with we note that when Pr=1Pr=1, θ=w\theta=w is a permitted solution of the BRB model system.

Second, the set of equations (6)-(7) is invariant with respect to the transformation (z,θ)→(1−z,−θ)(z,\theta) \to(1-z,-\theta) which also implies w→−ww \to-w because by definition w=dz/dtw = dz/dt. This means that if the couple θ(t,z)\theta(t,z), w(t,z)w(t,z) indicates a solution of the equations, than also −θ(t,1−z)-\theta(t,1-z), −w(t,1−z)-w(t,1-z) is a solution. Combining this with the condition ⟨w⟩=⟨θ⟩=0\langle w\rangle= \langle\theta\rangle= 0, it entails that both θ\theta and ww are odd functions with respect to z=1/2z = 1/2, and so θ(z=1/2)=w(z=1/2)=0\theta(z = 1/2) = w(z = 1/2) = 0.

A third symmetry is the following:

z→{z+1/2if z≤1/2z−1/2if z>1/2orz→z+sign⁡(12−z)12.(8)z \to\begin{cases} z + 1/2 & \text{if } z \le1/2 \\ z - 1/2 & \text{if } z > 1/2 \end{cases} \quad\text{or} \quad z \to z + \operatorname{sign}\left(\frac{1}{2}-z\right)\frac{1}{2}. \tag*{(8)}

It corresponds to swapping the spatial interval [0,1/2][0,1/2] with the one [1/2,1][1/2,1]. As shown in the sketch in Figure 1(a)-(b) this symmetry transforms what we call a “boundary-layer” type solution to a “shock” type solution (more on this later). In other words, due to the zero boundary conditions and to the second symmetry, the functions w(z)w(z) and θ(z)\theta(z) can be seen as periodic odd functions. This means that adding a phase of half the period is still a solution of the system. Equivalently one can say that, z→z+1/2z \to z + 1/2 is a symmetry of the system.

Image not extracted

Figure 1. Symmetries of the system of equations: (a) odd symmetry of ww and θ\theta with respect to the position z=1/2z = 1/2; (b) swap symmetry with resect to the system mid-point (8); (c) rescaling transformation (9)-(10) with n=1n = 1.

A fourth remarkable symmetry of the system is the following: Let’s call w(t,z;Ra,Pr)w(t,z;Ra,Pr), θ(t,z;Ra,Pr)\theta(t,z;Ra,Pr) the solution of the system for a given value of the parameters RaRa, PrPr. The system is then invariant with respect to the transformation:

w(t,z;Ra,Pr)→w(t,2nz;Ra2n,Pr)/(2n)(9)w(t,z;Ra,Pr) \to w\left(t,2nz;\frac{Ra}{2n},Pr\right)/(2n) \tag*{(9)}
θ(t,z;Ra,Pr)→θ(t,2nz;Ra2n,Pr)/(2n)(10)\theta(t,z;Ra,Pr) \to\theta\left(t,2nz;\frac{Ra}{2n},Pr\right)/(2n) \tag*{(10)}

where nn is a positive integer number. This “rescaling transformation” symmetry is illustrated in Figure 1(c) for the case n=1n = 1.

Global response parameters: Nusselt and Reynolds

To derive the expression for the global heat flux it is convenient to resort to the dimensional notation. The temperature equation (2) in conservative form reads: Tτ+(JT)Z=0T_\tau+ (J_T)_Z = 0, where

JT(Z,τ)=WT−κTZ−∫0ZTWZ′ dZ′,(11)J_T(Z,\tau) = WT - \kappa T_Z - \int_0^Z T W_{Z'}\,dZ', \tag*{(11)}

is the local and instantaneous heat flux at position ZZ and time τ\tau. Averaging the conservative form equation over time (here denoted as overline) and assuming a steady state gives the expression of the mean global heat-flux:

JT‾(Z)=WT‾−κT‾Z−∫0ZTWZ′‾ dZ′=const.(12)\overline{J_T}(Z) = \overline{WT} - \kappa\overline{T}_Z - \int_0^Z \overline{TW_{Z'}}\,dZ' = \mathop{\rm const.} \tag*{(12)}

The integral term in the above expression, which is absent in the mean heat flux of the RB system, is a consequence of the compressibility of the velocity field. The mean Nusselt number is defined by adimensionalizing the mean global heat flux with respect to the conductive heat flux (i.e. the state where W=0W=0 and T=TcT=T_c):

Nu≡JT‾JTc=const.(13)Nu \equiv\frac{\overline{J_T}}{J_{T_c}} = \mathop{\rm const.} \tag*{(13)}

We observe that by plugging into the above expression the dimensionless temperature fluctuation θ\theta, and evaluating the expression either in z=0z=0 or z=1z=1, gives the following equivalent expressions for NuNu:

Nu=1−θ‾z(0)=1−θ‾z(1).(14)Nu = 1-\overline{\theta}_z(0)=1-\overline{\theta}_z(1). \tag*{(14)}

One can remark that this same expression for NuNu is obtained in the RB flow ruled by the Boussinesq system of equations. On the contrary, if one considers the spatial average of NuNu (spatial average of eq. (13)) one gets:

Nu=1+PrRa(⟨w‾θ⟩−⟨∫0z(θw‾z′+w‾)dz′⟩)(15)Nu=1+\sqrt{PrRa}\left(\langle\overline{w}\theta\rangle-\left\langle\int_0^z\left(\theta\overline{w}_{z'}+\overline{w}\right)dz'\right\rangle\right) \tag*{(15)}

which is different from the RB expression by the appearance of the integral term on the r.h.s., which originates, as already mentioned, by the flow compressibility. The volume averaged expression of the Nusselt number is convenient for numerical calculations, as it is less affected by discretization and numerical errors (we will use this expression in the numerical calculations presented in this article). Finally, we note that the Reynolds number defined as a system response parameter is here:

Re≡⟨W2‾⟩1/2Hν=RaPr⟨w2‾⟩1/2.(16)Re\equiv\frac{\langle\overline{W^2}\rangle^{1/2}H}{\nu}=\sqrt{\frac{Ra}{Pr}}\langle\overline{w^2}\rangle^{1/2}. \tag*{(16)}

The BRB dynamics: theoretical analysis

This section presents some notable analytical results on the dynamics of the BRB model system. First, we perform the linear stability analysis to determine the transition from the conductive to the convective state. Second, we address the non-linear saturation mechanism that is responsible for the stabilization of the flow after the inception of convection. Third, by means of a standard matched asymptotic (mama) analysis, we solve the BRB system of equations in steady condition in the limit of large RaRa numbers. Finally, based on the mama solution we derive the asymptotic-in-Rayleigh scaling laws for the Nusselt and Reynolds numbers.

Onset of convection

The linearization of the system (6–7) with respect to ww and θ\theta satisfies solutions of the form

w=∑n=1∞wneσntsin⁡(nkz),θ=∑n=1∞θneσntsin⁡(nkz),(17)w=\sum_{n=1}^{\infty}w_n e^{\sigma_n t}\sin(nkz),\qquad\theta=\sum_{n=1}^{\infty}\theta_n e^{\sigma_n t}\sin(nkz), \tag*{(17)}

where k=2πk=2\pi and nn is an integer value, and with the growth rate

σn=(Pr+1)2(nk)4+4Pr(Ra−(nk)4)−(Pr+1)(nk)22RaPr.(18)\sigma_n=\frac{\sqrt{(Pr+1)^2(nk)^4+4Pr\left(Ra-(nk)^4\right)-(Pr+1)(nk)^2}}{2\sqrt{RaPr}}. \tag*{(18)}

Therefore, for Ra>Rac=k4=(2π)4≃1558Ra>Ra_c=k^4=(2\pi)^4\simeq1558 and at any PrPr value the system becomes linearly unstable (σ1>0\sigma_1>0). The critical Rayleigh number happens to be the same as in the three-dimensional three-periodic homogeneous Rayleigh-Bénard system [14, 15] although in that case the perturbation form is different as it depends only on the horizontal coordinates. We observe that the relative amplitude of the velocity and temperature field is w1/θ1=(σ1+k2/PrRa)w_1/\theta_1=(\sigma_1+k^2/\sqrt{PrRa}), this implies that w1=θ1w_1=\theta_1 for Pr=1Pr=1. This prediction will be verified in Sec. IV by means of a numerical simulation starting from a tiny white noise perturbation on ww and θ\theta fields (see also Figure 6). The described exponentially growing solution is eventually saturated by the presence of the nonlinear terms, as we discuss in the next section.

Figure 6. Nusselt-Rayleigh scaling at Pr=1Pr=1. (a): NuNu vs. RaRa. We display the numerical simulation results (red circle symbols) and comparisons with different analytical predictions: (vertical dotted line) the critical Rayleigh number for the onset of convection Rac≃1558Ra_c \simeq1558; (solid line) the Nusselt number calculated from the two-mode steady solution (33); (dashed line) the NuNu expression computed via the matched asymptotic solution Nu=Ra/8Nu=\sqrt{Ra}/8. (b): The same data points for NuNu and theoretical predictions, here compensated with respect to Ra1/2Ra^{1/2}.

Non-linear saturation mechanism

Similarly to what occurs in a three-dimensional RB system in slightly supercritical conditions (Ra≳RacRa \gtrsim Ra_c) the exponential growth rate of the perturbation rapidly saturates into a convective steady state. This phenomenology can be promptly explained for the BRB at Pr=1Pr = 1 by means of a two modes Galerkin expansion, which we detail in the following. We assume

w(z,t)=∑n=1,2An(t)sin⁡(nkz),θ(z,t)=∑n=1,2Bn(t)sin⁡(nkz).(19)w(z,t) = \sum_{n=1,2} A_n(t) \sin(nkz), \qquad\theta(z,t) = \sum_{n=1,2} B_n(t) \sin(nkz). \tag*{(19)}

Upon its substitution into the equations of motion, retaining only terms in sin⁡(kz)\sin(kz) and sin⁡(2kz)\sin(2kz), we obtain the first-order differential system for the evolution of the amplitudes of the four considered modes:

A˙1=Rac1/42A1A2−RacRa/PrA1+B1(20)\dot{A}_1 = \frac{Ra_c^{1/4}}{2} A_1 A_2 - \sqrt{\frac{Ra_c}{Ra/Pr}} A_1 + B_1 \tag*{(20)}
A˙2=−Rac1/42A12−4RacRa/PrA2+B2(21)\dot{A}_2 = -\frac{Ra_c^{1/4}}{2} A_1^2 - 4\sqrt{\frac{Ra_c}{Ra/Pr}} A_2 + B_2 \tag*{(21)}
B˙1=Rac1/42(2A1B2−A2B1)−RacRaPrB1+A1(22)\dot{B}_1 = \frac{Ra_c^{1/4}}{2}(2A_1B_2 - A_2B_1) - \sqrt{\frac{Ra_c}{RaPr}} B_1 + A_1 \tag*{(22)}
B˙2=−Rac1/42A1B1−4RacRaPrB2+A2,(23)\dot{B}_2 = -\frac{Ra_c^{1/4}}{2} A_1B_1 - 4\sqrt{\frac{Ra_c}{RaPr}} B_2 + A_2, \tag*{(23)}

where we have taken into account that Rac=k4Ra_c = k^4. Such system can not be solved analytically, however that is possible for the special case Pr=1Pr = 1. In this condition, the equations for AiA_i and BiB_i become identical, hence the differential system reduces to a two-dimensional one:

A˙1=σ1A1+Rac1/42A1A2(24)\dot{A}_1 = \sigma_1 A_1 + \frac{Ra_c^{1/4}}{2} A_1 A_2 \tag*{(24)}
A˙2=σ2A2−Rac1/42A12,(25)\dot{A}_2 = \sigma_2 A_2 - \frac{Ra_c^{1/4}}{2} A_1^2, \tag*{(25)}

with σ1=1−Rac/Ra>0\sigma_1 = 1 - \sqrt{Ra_c/Ra} > 0 and σ2=1−4Rac/Ra<0\sigma_2 = 1 - 4\sqrt{Ra_c/Ra} < 0 for Rac<Ra<16RacRa_c < Ra < 16Ra_c. As the characteristic time-scale of the second mode τ2=∣σ2∣−1\tau_2 = |\sigma_2|^{-1} is smaller than the characteristic time of the first mode τ1=∣σ1∣−1\tau_1 = |\sigma_1|^{-1}, we can perform a so-called adiabatic elimination and take A˙2≈0\dot{A}_2 \approx0 in the vicinity of the bifurcation threshold. By doing so, we obtain the following Landau equation [16] for the evolution of the amplitude A1A_1:

A˙1=σ1A1−γA13(26)\dot{A}_1 = \sigma_1 A_1 - \gamma A_1^3 \tag*{(26)}

with γ=Rac4(4Rac/Ra−1)\gamma= \frac{\sqrt{Ra_c}}{4(4\sqrt{Ra_c/Ra}-1)}. Eq. (26) admits three steady solutions:

A1ss=0(conductive state)(27)A_1^{ss} = 0 \quad\text{(conductive state)} \tag*{(27)}
A1ss=±σ1/γ(two convective states).(28)A_1^{ss} = \pm\sqrt{\sigma_1/\gamma} \quad\text{(two convective states)}. \tag*{(28)}

Assuming the perturbation expansion A1(t)=A1ss+εA1p(t)+O(ε2)A_1(t) = A_1^{ss} + \varepsilon A_{1p}(t) + O(\varepsilon^2), the amplitude equation at order O(ε)O(\varepsilon) writes

A˙1p=A1p(t)(σ1−3γ(A1ss)2)(29)\dot{A}_{1p} = A_{1p}(t)(\sigma_1 - 3\gamma(A_1^{ss})^2) \tag*{(29)}

A solution of the linear and homogeneous equation above can be written as A1p(t)=A1p(0)exp⁡(−iωt)A_{1p}(t) = A_{1p}(0) \exp(-i\omega t), which leads to

σ1−3γ(A1ss)2+iω=0(30)\sigma_1 - 3\gamma(A_1^{ss})^2 + i\omega= 0 \tag*{(30)}

Since all the coefficients are real and ω=ωR+iωI\omega= \omega_R + i\omega_I, we get ωR=0\omega_R = 0 (meaning that solutions are stationary) and ωI=σ1−3γ(A1ss)2\omega_I = \sigma_1 - 3\gamma(A_1^{ss})^2. Hence, the growth rates of the trivial (27) and non-trivial (28) steady-states are ωI=σ1\omega_I = \sigma_1 and ωI=−2σ1\omega_I = -2\sigma_1 respectively. Therefore, for Ra<RacRa < Ra_c the trivial conductive solution is stable, whereas for Ra>RacRa > Ra_c the non-trivial convective solutions are the ones to be stable. This corresponds to a supercritical pitchfork bifurcation, as illustrated in the graph of Figure 2.

Figure 2. Steady solution of the amplitude A1A_1 of the two-mode truncated system as a function of the control parameter RaRa, as given in eq. (27),(28). The graph has a typical structure of a supercritical pitchfork bifurcation, solid lines denote stable branches while the dotted one denotes the unstable branch. We indicate as “conduction” the solution corresponding to A1=0A_1 = 0 (and A2=0A_2 = 0 too), with “shock” type solution the one presenting a sharp variation on the bulk of the system, which corresponds to the case A1>0A_1 > 0, and finally with “boundary layer” type the one where the variation occurs close to the boundaries (A1<0A_1 < 0 in this case). The insets show the complete steady two-mode solutions for θ(z)\theta(z) or w(z)w(z) of the shock and boundary layer types for the case Ra=2000Ra = 2000, marked with colored circles on the main panels.

The multiplicity of the above convective solutions is a consequence of the swap symmetry mentioned in the previous section, which in the case of a sinusoidal profile takes the form sin⁡(nkz)→(−1)nsin⁡(nkz)\sin(nkz) \to(-1)^n \sin(nkz) for integers nn. In the two mode expansion it therefore affects only the mode A1A_1. This produces a change from a solution with sharp variations at the boundaries, which we can denote as “boundary layer” type solution, to one with a sharp transition in the bulk of the domain, that we call “shock” type solution. While the boundary layer type solution is analogous to the vertical flow profile observed in the RB system, the shock type solution has no corresponding in 2D or 3D convective systems. For this reason, in our analysis we will mainly focus on the former kind of solution. The temperature/velocity spatial profiles corresponding to the complete convective steady solutions of (24)–(25), i.e., the two-mode truncated series (19) with

A1=±σ1/γ=±2Rac1/4(1−RacRa)1/2(4RacRa−1)1/2(31)A_1 = \pm\sqrt{\sigma_1/\gamma} = \frac{\pm2}{Ra_c^{1/4}}\left(1-\sqrt{\frac{Ra_c}{Ra}}\right)^{1/2}\left(4\sqrt{\frac{Ra_c}{Ra}}-1\right)^{1/2} \tag*{(31)}
A2=−2σ1/Rac1/4=−2Rac1/4(1−RacRa)(32)A_2 = -2\sigma_1/Ra_c^{1/4} = \frac{-2}{Ra_c^{1/4}}\left(1-\sqrt{\frac{Ra_c}{Ra}}\right) \tag*{(32)}

are traced in Figure 2 (insets) [17].

The above convective solutions allows to estimate the Nusselt number near to the onset, as

Nu=1−2π(A1+2A2).(33)Nu = 1 - 2\pi(A_1 + 2A_2). \tag*{(33)}

They also imply that

⟨w2⟩=⟨θ2⟩=⟨wθ⟩=12(A12+A22)=6Ra(1−RacRa),(34)\langle w^2\rangle= \langle\theta^2\rangle= \langle w\theta\rangle= \frac{1}{2}(A_1^2 + A_2^2) = \frac{6}{\sqrt{Ra}}\left(1-\sqrt{\frac{Ra_c}{Ra}}\right), \tag*{(34)}

so the Reynolds number becomes:

Re=6(Ra−Rac).(35)Re = \sqrt{6\left(\sqrt{Ra}-\sqrt{Ra_c}\right)} . \tag*{(35)}

These predictions will be tested in Sec. IV by means of direct numerical simulations.

Finally, we wish to comment on the role of the no-zero mode condition, ⟨w⟩=⟨θ⟩=0\langle w\rangle=\langle\theta\rangle=0, introduced in our model system. The removal of this condition allows for an anticipated onset of convection at Rac=π4Ra_c=\pi^4 (16 times smaller than in the present case). It also permits non-odd solutions characterized by a single boundary layer on one of the sides of the domain. By virtue of the system symmetries, these solutions are trivially linked to the ones described above: their period and their amplitude are doubled but all scaling properties remains identical.

Large-Ra asymptotic dynamics

What happens to the system dynamics at large RaRa? Does it keep stationary or rather does it become time-dependent, and possibly chaotic and turbulent? Replying to these questions on the basis of a theoretical analysis is challenging. We will try to reply to this question with the aid of numerical simulations, in Sec. IV. However, it is possible to derive a steady solution of (6)-(7) in the limit of large RaRa and at any PrPr, by means of the standard technique of matched asymptotic expansion. In the following we detail the derivation of such remarkable solution and we will then use it to give a prediction for the Nusselt and Reynolds scaling in the asymptotically large RaRa regime. The numerical simulations, described in Sec. IV, will prove that the steady mama solution describe strikingly well the real behaviour of the BRB system.

Approximate stationary solution with matched asymptotic expansion

The use of matched asymptotics to describe the structure of shocks in the Burgers equation at very small viscosities is a classical approach [18]. This technique has been applied in several studies to estimate the contribution of shocks to the anomalous dissipation of kinetic energy [19], to the energy spectrum of solutions [20], to propose closures for statistical theories of Burgers’ turbulence [21] and more recently to explain the spontaneous stochasticity of Lagrangian trajectories in Burgers equation [22].

Here we look for a solution of Eq. (6)-(7) under the assumption that both PrRa\frac{\sqrt{Pr}}{\sqrt{Ra}} and 1RaPr\frac{1}{\sqrt{RaPr}} are small parameters (or equivalently Ra−1≪Pr≪RaRa^{-1}\ll Pr\ll Ra). First, we consider the solution of the system far from boundaries, denoted as outer solution. In this region the dissipative terms are negligible, because they multiply the above mentioned small parameters, hence the system reduce to:

wwz=θ(36)w w_z = \theta \tag*{(36)}
wθz=w,(37)w\theta_z = w, \tag*{(37)}

From the second equation we get θout=z+c1\theta_{out}=z+c_1, which is then plugged into the first equation leading to wwz=z+c1w w_z=z+c_1 which admits the solution wout=z2+2c1z+c2w_{out}=\sqrt{z^2+2c_1z+c_2}. By using the condition that the solution should be an odd function in the domain (i.e. w(1/2)=θ(1/2)=0w(1/2)=\theta(1/2)=0) we determine the constants c1=−1/2c_1=-1/2, c2=1/4c_2=1/4 and so wout=θout=z−1/2w_{out}=\theta_{out}=z-1/2. Second, we consider the solution near to a boundary, denoted as inner solution (we choose here the boundary close to z=0z=0). In this region, the application of standard least degeneracy principle [23] leads to the system:

wwz=PrRawzz(38)w w_z = \frac{\sqrt{Pr}}{\sqrt{Ra}}w_{zz} \tag*{(38)}
wθz=1PrRaθzz,(39)w\theta_z = \frac{1}{\sqrt{PrRa}}\theta_{zz}, \tag*{(39)}

Here one first solve the equation for ww. This leads to win=−atanh⁡(aRaPrz2)w_{in}=-a\tanh\left(a\sqrt{\frac{Ra}{Pr}}\frac{z}{2}\right) where we have adopted the boundary condition in w(0)=0w(0)=0. The constant aa can be determined by matching the inner and outer solutions for ww, lim⁡z→∞win=lim⁡z→0wout\lim_{z\to\infty}w_{in}=\lim_{z\to0}w_{out}: −a=−1/2-a=-1/2, so a=1/2a=1/2.

win=−12tanh⁡(RaPrz4)(40)w_{in}=-\frac{1}{2}\tanh\left(\sqrt{\frac{Ra}{Pr}}\frac{z}{4}\right) \tag*{(40)}

By substituting now the inner solution winw_{in} in the equation for θ\theta we obtain

−12tanh⁡(RaPrz4)θz=1PrRaθzz(41)-\frac{1}{2}\tanh\left(\sqrt{\frac{Ra}{Pr}}\frac{z}{4}\right)\theta_z=\frac{1}{\sqrt{PrRa}}\theta_{zz} \tag*{(41)}

which can be rewritten as

−12PrRatanh⁡(RaPrz4)=ddzlog⁡θz(42)-\frac{1}{2}\sqrt{PrRa}\tanh\left(\sqrt{\frac{Ra}{Pr}}\frac{z}{4}\right)=\frac{d}{dz}\log\theta_z \tag*{(42)}

and integrated to

log⁡(cosh⁡(RaPrz4))−2Pr=log⁡θz+log⁡K(43)\log\left(\cosh\left(\sqrt{\frac{Ra}{Pr}}\frac{z}{4}\right)\right)^{-2Pr}=\log\theta_z+\log K \tag*{(43)}

where log⁡K\log K is a constant, hence removing the log:

(cosh⁡(RaPrz4))−2Pr=Kθz.(44)\left(\cosh\left(\sqrt{\frac{Ra}{Pr}}\frac{z}{4}\right)\right)^{-2Pr}=K\theta_z. \tag*{(44)}

This can be integrated in the interval [0,z][0,z]

θ(z)−θ(0)=1K∫0z(cosh⁡(RaPrz′4))−2Prdz′(45)\theta(z)-\theta(0)=\frac{1}{K}\int_0^z\left(\cosh\left(\sqrt{\frac{Ra}{Pr}}\frac{z'}{4}\right)\right)^{-2Pr} dz' \tag*{(45)}

We now apply the boundary condition θ(0)=0\theta(0)=0, while the value of KK is obtained by matching the inner and outer solutions for θ\theta, lim⁡z→∞θin=lim⁡z→0θout\lim_{z\to\infty}\theta_{in}=\lim_{z\to0}\theta_{out}: −12=1Klim⁡z→∞∫0z(cosh⁡(RaPrz′4))−2Prdz′-\frac{1}{2}=\frac{1}{K}\lim_{z\to\infty}\int_0^z\left(\cosh\left(\sqrt{\frac{Ra}{Pr}}\frac{z'}{4}\right)\right)^{-2Pr} dz'. It follows:

θin(z)=−12∫0z(cosh⁡(RaPrz′4))−2Prdz′∫0∞(cosh⁡(RaPrz′4))−2Prdz′.(46)\theta_{in}(z)=-\frac{1}{2}\frac{\int_0^z\left(\cosh\left(\sqrt{\frac{Ra}{Pr}}\frac{z'}{4}\right)\right)^{-2Pr} dz'}{\int_0^\infty\left(\cosh\left(\sqrt{\frac{Ra}{Pr}}\frac{z'}{4}\right)\right)^{-2Pr} dz'}. \tag*{(46)}

The complete perturbative solution is obtained by summing up the inner and the outer solution and by subtracting their overlap, wma(z)=win(z)+wouter(z)−woverlapw_{ma}(z)=w_{in}(z)+w_{outer}(z)-w_{overlap}, where woverlap=lim⁡z→∞win=lim⁡z→0wout=−12w_{overlap}=\lim_{z\to\infty}w_{in}=\lim_{z\to0}w_{out}=-\frac{1}{2}. This leads to the final expression for the matched asymptotic solution:

wma(z)=z−12tanh⁡(RaPrz4)(47)w_{ma}(z)=z-\frac{1}{2}\tanh\left(\sqrt{\frac{Ra}{Pr}}\frac{z}{4}\right) \tag*{(47)}
θma(z)=z−12∫0z(cosh⁡(RaPrz′4))−2Prdz′∫0∞(cosh⁡(RaPrz′4))−2Prdz′(48)\theta_{ma}(z)=z-\frac{1}{2}\frac{\int_0^z\left(\cosh\left(\sqrt{\frac{Ra}{Pr}}\frac{z'}{4}\right)\right)^{-2Pr} dz'}{\int_0^\infty\left(\cosh\left(\sqrt{\frac{Ra}{Pr}}\frac{z'}{4}\right)\right)^{-2Pr} dz'} \tag*{(48)}

We note that if Pr=1Pr=1 the solutions takes the simpler form

wma(z)=θma(z)=z−12tanh⁡(Raz4).(49)w_{ma}(z)=\theta_{ma}(z)=z-\frac{1}{2}\tanh\left(\sqrt{Ra}\frac{z}{4}\right). \tag*{(49)}

The solution (49) for Pr=1Pr=1 coincides with the shock solution derived by Saffman [24] for the randomly forced Burgers equation in the limit of t→+∞t\to+\infty and large ReRe (when Ra\sqrt{Ra} is replaced by ReRe), see also [25] for a recent discussion. Remark that the matched asymptotic solutions (47),{xref}(48)` are valid only in the interval z∈[0,1/2]z\in[0,1/2], but they can be applied to the interval z∈[1/2,1]z\in[1/2,1] with the transformation z→z−1z\to z-1. This at the price of accepting a discontinuity in the origin, because e.g. for the velocity wma(z=1/2)=ε≠0w_{ma}(z=1/2)=\varepsilon\ne0. However, such a discontinuity goes as 2ε=1−tanh⁡(RaPr18)2\varepsilon=1-\tanh\left(\sqrt{\frac{Ra}{Pr}}\frac{1}{8}\right), therefore it vanishes asymptotically with RaRa. We also note that the perturbative solution is not an exact solution of the original system. However, it is asymptotically correct. This can be seen by plugging it into the stationary equations and considering the Ra→∞Ra \to\infty limit of the residuals

lim⁡Ra→∞wmawma,z−PrRawma,zz−θma=0\lim_{Ra\to\infty} w_{ma}w_{ma,z}-\sqrt{\frac{Pr}{Ra}}w_{ma,zz}-\theta_{ma}=0
lim⁡Ra→∞wmaθma,z−1PrRaθma,zz−wma=0.\lim_{Ra\to\infty} w_{ma}\theta_{ma,z}-\frac{1}{\sqrt{PrRa}}\theta_{ma,zz}-w_{ma}=0.

Another observation is in order about the the shape of the solution. We remark here that by choosing that the internal solution occurs at z=0z=0, we have implicitly selected the boundary-layer type of the solution. A different, equally admissible choice, is to place the inner solution close to z=1/2z=1/2, this leads to a shock type solution. One can trivially go from the former type of solution to the latter, by applying the third symmetry transformation {(xref)}(8) discussed in Sec.II B. We observe that these two type of solutions are characterized by the same Reynolds number, as Re∼⟨w2⟩1/2Re \sim\langle w^2\rangle^{1/2}, while they differ for the Nusselt number, because Nu=1−θz(0)Nu=1-\theta_z(0). The latter observation implies that while boundary layer (BL) type solutions are characterized by an increasing NuNu as Ra→∞Ra\to\infty, in the shock type solution the Nusselt number vanishes (leading to a perfectly insulating system for Ra→∞Ra\to\infty). As we mentioned before, although these two convective states are equally probable, we will limit our considerations to the case of BL type solution, as it offers a better analogy with the dynamics of the realistic RB system which motivates this study.

Upper bounds and asymptotic scalings for the Nusselt and Reynolds numbers

The matched asymptotic solution allows to promptly compute asymptotic expressions for all global quantities in the system, we focus here on the two main output observables of the BRB system, the Nusselt and the Reynolds number.

Nusselt number

We evaluate Nu=1−θma,z(0)Nu=1-\theta_{ma,z}(0). In the general PrPr case, by using {(xref)}(48) we get

Nu=12∫0∞(cosh⁡(RaPrz′4))−2Pr dz′(50)Nu=\frac{1}{2\int_{0}^{\infty}\left(\cosh\left(\sqrt{\frac{Ra}{Pr}}\frac{z'}{4}\right)\right)^{-2Pr}\,dz'} \tag*{(50)}

First of all let us note that the RaRa dependence can be factorized by introducing the auxiliary variable z~=Ra z′\tilde{z}=\sqrt{Ra}\,z' in the integral, so:

Nu=Ra2∫0∞(cosh⁡(z~4Pr))−2Pr dz~.(51)Nu=\frac{\sqrt{Ra}}{2\int_{0}^{\infty}\left(\cosh\left(\frac{\tilde{z}}{4\sqrt{Pr}}\right)\right)^{-2Pr}\,d\tilde{z}}. \tag*{(51)}

Because the denominator only depends on PrPr it is therefore clear that the scaling Nu∼RaNu\sim\sqrt{Ra} is to be expected asymptotically in RaRa for any PrPr.

We now focus on the PrPr dependence. Using the property ex/2≤cosh⁡x≤exe^x/2\le\cosh x\le e^x for x>0x>0, one can write,

22Pr+1Pr≥∫0∞(cosh⁡(z~4Pr))−2Pr dz~≥2Pr,\frac{2^{2Pr+1}}{\sqrt{Pr}}\ge\int_{0}^{\infty}\left(\cosh\left(\frac{\tilde{z}}{4\sqrt{Pr}}\right)\right)^{-2Pr}\,d\tilde{z}\ge\frac{2}{\sqrt{Pr}},

and finally by using {(xref)}(51):

RaPr4Pr+1≤Nu≤RaPr4.\frac{\sqrt{RaPr}}{4^{Pr+1}}\le Nu\le\frac{\sqrt{RaPr}}{4}.

This bounding relation is relevant in the limit of small PrPr, because 4Pr→14^{Pr}\to1, leading to the scaling law

Nu≃RaPr4(Pr small).(52)Nu\simeq\frac{\sqrt{RaPr}}{4}\qquad(Pr\ \text{small}). \tag*{(52)}

In the limit of large PrPr as (cosh⁡(z~4Pr))−2Pr→e−(z~/4)2\left(\cosh\left(\frac{\tilde{z}}{4\sqrt{Pr}}\right)\right)^{-2Pr}\to e^{-(\tilde{z}/4)^2} and using {(xref)}(51) one obtains:

Nu=Ra4π(Pr→∞).(53)Nu=\frac{\sqrt{Ra}}{4\sqrt{\pi}}\qquad(Pr\to\infty). \tag*{(53)}

Which is an expression that does remarkably not depend on the PrPr value. It is worth observing that the saturation of the NuNu number for large PrPr values (53) is a feature also observed in the realistic RBRB system (see e.g. [26] Figure 5). Finally, for the intermediate case, Pr=1Pr = 1, taking advantage of the simpler form of the matched asymptotic solution (49), one can exactly derive:

Nu=Ra8(Pr=1).(54)Nu = \frac{\sqrt{Ra}}{8}\quad(Pr = 1). \tag*{(54)}
Reynolds number

We now turn the attention to the Reynolds number. Asymptotically in RaRa we observe that the mama velocity solution approaches the behaviour wma≃wout=z−1/2w_{ma} \simeq w_{out} = z - 1/2, while the boundary layers (BL) become thinner and thinner. This points to the existence of the upper bound for the velocity variance: ⟨w2⟩≤∫01(z−1/2)2 dz=1/12\langle w^2 \rangle\leq\int_0^1 (z - 1/2)^2\,\mathrm{d}z = 1/12, which implies:

Re=RaPr⟨w2⟩1/2≤112RaPr(55)Re = \sqrt{\frac{Ra}{Pr}}\langle w^2\rangle^{1/2} \leq\frac{1}{\sqrt{12}}\sqrt{\frac{Ra}{Pr}} \tag*{(55)}

A lower bound for ReRe can be obtained by just considering that the asymptotic outer solution is the one most contributing to the global velocity variance, so:

⟨w2⟩≳∫δ1−δwout2 dz≃∫δ1−δ(z−12)2 dz=(1−2δ)312\langle w^2\rangle\gtrsim\int_{\delta}^{1-\delta} w_{out}^2\,\mathrm{d}z \simeq\int_{\delta}^{1-\delta}\left(z-\frac{1}{2}\right)^2\,\mathrm{d}z = \frac{(1-2\delta)^3}{12}

where δ\delta is an estimation for the thickness of the kinetic boundary layer, which we define as the height zz where the argument of the hyperbolic function in winw_{in} is one, i.e., δ=4/Ra/Pr\delta= 4/\sqrt{Ra/Pr}. This implies

Re≳112RaPr(1−8PrRa)3/2(56)Re \gtrsim\frac{1}{\sqrt{12}}\sqrt{\frac{Ra}{Pr}}\left(1-8\sqrt{\frac{Pr}{Ra}}\right)^{3/2} \tag*{(56)}

We will show that the above predictions for the Nusselt and the Reynolds numbers approach quite well the result that we obtain from the numerical simulations of the BRB system in the asymptotic large-RaRa limit (see below Sec. IV C).

Scaling laws of the form Nu∼Ra PrNu \sim\sqrt{Ra\,Pr} and Re∼Ra/PrRe \sim\sqrt{Ra/Pr} identifies the so called ultimate regime of thermal convection. Physically it can be interpreted as a regime where the microscopic diffusion material properties, i.e. the viscosity and the thermal diffusivities, have a negligible role in the determination of the intensity of the heat transport and of the kinetic energy in the system. The ultimate regime has been predicted to occur in the RB systems in the asymptotic high-RaRa limit [27] (see also [28]). However, its verification in RB experiments and simulations is still debated (see [29] for a recent concise account). On the opposite, it has been clearly observed in the so called homogeneous-Rayleigh-Benard (HRB) model system, which is a three-dimensional vertically unbounded system, either triperiodic [14] or laterally confined [30], which can be realized only in numerical simulations. It is important to note that the HRB model has no horizontal wall boundaries, as such it misses the corresponding kinetic and thermal boundary layers, i.e. well identified regions where dissipation has the dominant role with respect to inertial transport terms. Experimental realizations of the Nu∼RaNu \sim\sqrt{Ra} regime have been achieved only in bulk dominated convective systems, such as in the vertical channel setup [31, 32] or in systems where the wall thermal heating has been replaced by volumetric radiative heating [33, 34]. More recently [35] have numerically demonstrated the occurrence of the ultimate regime of convection also in a RB system with permeable walls. This system possesses thermal and kinetic BL but does not enforce the cancellation of the vertical velocity on the top-bottom walls. In the light of this, the verification of eqs. (54) and (55) for the BRB system in high-RaRa regime would make it the first bounded system, with kinetic and thermal boundary layers, to display the ultimate regime. As we will numerically demonstrate in the next section, the BRB clearly possesses this feature.

The BRB dynamics: numerical simulation analysis

In order to test the predictions presented in the previous section and to get deeper insight into the BRB dynamics, we performed direct numerical simulations (DNS) of the system of equations. This is conveniently done by means of a Fourier pseudo-spectral method. For this study we use spatial resolutions ranging from N=213N=2^{13} to 2162^{16} grid points and explore the two-dimensional parameter space Ra−PrRa-Pr. The simulations evolve in time from an initial state where ww and θ\theta are null except for a tiny random uniform spatially-uncorrelated perturbation. The adopted numerical methods and protocols are described in detail in the Appendix A.

Temperature and velocity profiles

We numerically find that the system displays a steady solution at any RaRa, up to 101010^{10} simulated in this study, and at any PrPr in the range [10−2,102][10^{-2},10^2]. Furthermore, when Pr=1Pr=1 the ww and θ\theta profiles are always coincident. As illustrated in Figure 3 and 4 when the Rayleigh number is slightly beyond the critical threshold Ra=O(103)Ra=O(10^3), the two-modes solution (31)(32) perfectly approximate the numerically computed temperature and velocity profiles. On the other hand for Ra=105Ra = 10^{5} the agreement with the matched asymptotic solution (49) is already excellent. It is also evident that as Ra→+∞Ra \to+\infty the profiles approaches the z−1/2z - 1/2 linear shape, with vanishingly thin boundary layers. Note that the temperature TT is the sum of the conductive profile and the fluctuation θ\theta, this implies that TT is essentially constant in the well mixed bulk of the system and changes sharply only in the boundary layer, in close analogy with the mean vertical temperature profile in the turbulent high-RaRa RB system. Figure 4 reports the corresponding numerical results for non unit Prandtl numbers (Pr=5Pr = 5 and 1/51/5) at Ra=105Ra = 10^{5}. One can appreciate that for Pr>1Pr > 1 the thermal boundary layer is thinner than the kinetic one and vice-versa for Pr<1Pr < 1. Note also that the two cases are distinct, because Pr→Pr−1Pr \to Pr^{-1} is not a symmetry of the system. Again we observe a good agreement with the matched asymptotic solutions (47),(48). In the asymptotic high-PrPr limit one shall expect that θ\theta will become almost everywhere equal to z−1/2z - 1/2, while in the low-PrPr limit the same will happen for ww.

Figure 3. Graph of the numerical solution for the temperature fluctuation θ(z)\theta(z) at Pr=1Pr=1 (identical to the velocity w(z)w(z)) for several RaRa values from Ra=1700=1.09RacRa=1700=1.09Ra_c, to Ra=105=64.16RacRa=10^5=64.16Ra_c. We show the two-modes truncated solution for Ra=1700Ra=1700 as well as the matched asymptotic one for Ra=105Ra=10^5. The solution approach the bulk behaviour, z−1/2z-1/2, at increasing the RaRa number.

Figure 4. Graph of the numerical solution for the temperature fluctuation θ(z)\theta(z) and velocity w(z)w(z) at Ra=105Ra=10^5 and Pr=5Pr=5 (a) and Pr=1/5Pr=1/5 (b), and comparison with the matched asymptotic solutions for both fields. Note that we show here for better visibility the profile in the bottom half of the domain.

Is the convective stationary regime stable?

As mentioned above, numerically we find that the convective state of the system displays a stationary solution at any RaRa and PrPr values. This aspect might seem surprising as it is different from what happens in the RB system, where successive bifurcations occur as RaRa is increased, leading first to temporally periodic solutions, then chaotic ones and finally to progressively more turbulent states. In the BRB case, although we can not rule out the existence of sub-critical bifurcations that might lead to the existence of such unsteady states, we can prove that the convective state displayed by the system is linearly stable. This point is addressed in the current section.

In section III A we studied the stability of the conductive state. It was shown that, beyond Rac=k4=(2π)4Ra_c = k^4 = (2\pi)^4, the system bifurcates to a convective stationary state, hereafter denoted as ws,θsw_s,\theta_s. Depending on the value of RaRa this state can be approximated in three ways:

(i) by a two-mode expansion eq. (19) and (31),(32) valid near the critical threshold RacRa_c;

(ii) by a matched asymptotic solution eqs. (47),(48), valid in the limit of large Rayleigh numbers;

(iii) by an interpolation of the discretized numerical solution obtained from the DNS, which is valid in the whole RaRa range.

We can now study the stability of these states by applying the linear stability analysis. We adopt the Galerkin method of weighted residuals [36]. In short, the idea is to choose trial functions that satisfy the boundary conditions exactly and solve the differential equations in an averaged sense, by imposing the condition that the residuals are orthogonal to the trial functions. The result is a set of homogeneous equations whose non-trivial solution leads to an eigenvalue problem. Here we denote with σ~(n)\tilde{\sigma}(n) the series of the eigenvalues, which are determined in terms of the RaRa and PrPr parameters. Similarly to the previous case, we write

w=∑n=1Nw~nsin⁡(nkz)eσ~(n)t+ws,θ=∑n=1Nθ~nsin⁡(nkz)eσ~(n)t+θs(57)w = \sum_{n=1}^{N} \tilde{w}_n \sin(nkz)e^{\tilde{\sigma}(n)t} + w_s,\qquad\theta= \sum_{n=1}^{N} \tilde{\theta}_n \sin(nkz)e^{\tilde{\sigma}(n)t} + \theta_s \tag*{(57)}

with n∈Nn \in\mathbb{N} and k=2πk = 2\pi. The number of modes NN is chosen so that the convergence is assured for the different parameter values. Figure 5a shows the evolution of the perturbation growth rate σ~=σ~(1)\tilde{\sigma} = \tilde{\sigma}(1) with Ra(>Rac)Ra(> Ra_c), for the three different approximated base solutions and Pr=1Pr = 1. As expected, the two-mode solution (case i) agrees well with the numerical convective state (case iii) for moderate values of the Rayleigh number, while the matched asymptotic does it for large values. The evolution of the larger growth rate σ~\tilde{\sigma} for different values of Prandtl obtained with the numerical base state is depicted on Figure 5b. It can be seen that the growth rate is always negative, indicating that the first convective stationary state never loses its stability, even for large values of RaRa, in agreement with the observations from the DNS.

Figure 5. Results of linear stability analysis of the convective steady state ws,θsw_s,\theta_s. Perturbation growth rate σ~\tilde{\sigma} as a function of RaRa: (a) for the case Pr=1Pr=1 computed from three different representation of the base state: the numerical (valid in the full range of RaRa), the two-mode Galerkin truncation base state (valid at small RaRa) and the matched asymptotic one (valid for Ra→+∞Ra \to+\infty); (b) for different Prandtl numbers Pr∈[0.1,102]Pr \in[0.1,10^2] using the numerical base state. The dotted vertical line it is traced at Ra=RacRa=Ra_c.

Measure of Nusselt and Reynolds number asymptotic scalings

Form the numerically computed θ(z)\theta(z) and w(z)w(z) profiles one can estimate the corresponding Nusselt and Reynolds numbers and analyze their functional dependencies with RaRa and PrPr. The behaviour of NuNu as a function of RaRa, for Pr=1Pr = 1, is shown in Figure 6(a). We observe that for Ra<RacRa < Ra_c Nu≃1Nu \simeq1 (vertical dotted line), hence only conductive heat-transfer occurs, as expected from the theoretical analysis. Beyond RacRa_c convection starts and NuNu progressively increases. Close to the onset, for Ra≲2Rac∼3×103Ra \lesssim2Ra_c \sim3 \times10^3, the two-modes expression (33), (solid line), approaches well the numerical results. On the other hand the matched asymptotic prediction agrees with the data from Ra∼104Ra \sim10^4 (dashed line) up to the highest explored RaRa number (101010^{10}). Indeed, as it is better appreciated in the compensated plot Figure 6(b) at the largest RaRa the normalized data approaches the value 1/8 (dashed-dotted line) meaning that Nu∼Ra/8Nu \sim\sqrt{Ra}/8 in excellent agreement with the mama prediction (54).

We now look at the heat-flux dependence with respect to PrPr. This is illustrated in Figure 7(a), where Nu(Pr)Nu(Pr) is traced for various Rayleigh numbers Ra=106,107,108Ra=10^6,10^7,10^8. Also in this case we see an excellent agreement with the matched asymptotic solution (50) and with the small/large-PrPr asymptotic behaviours (52),(53) derived in the previous section.

Figure 7. Nusselt-Prandtl scaling behaviour at various Rayleigh numbers Ra=106Ra=10^6, 10710^7, 10810^8. (a): Nusselt as a function of Prandtl. We display the numerical simulation results (symbols) and the corresponding matched asymptotic predictions (colored dotted-dashed lines). The behaviours RaPr/4\sqrt{Ra}Pr/4 for small PrPr (grey dashed line) and Ra/π/4\sqrt{Ra/\pi}/4 for large PrPr (black dotted line). (b): Compensated graph Nu/RaNu/\sqrt{Ra} vs. PrPr. The value 1/81/8 corresponding to Pr=1Pr=1 is also traced. One can appreciate the collapse of the measurements on a single curve.

We clearly observe a saturation of the Nusselt number for Pr≫1Pr \gg1. Furthermore, Figure 7(b) shows how all the data points can be collapsed on a single curve, by means of the rescaling Nu/RaNu/\sqrt{Ra}. This means that, in agreement with the matched asymptotic solution, the RaRa and PrPr dependence can be factorized (see eq. (51)).

Similar observations can be made for the dependence of the Reynolds number versus RaRa for Pr=1Pr=1 (Figure 8(a,b)) and Re(Pr)Re(Pr) figure Figure 9 (a)). At high-RaRa agreement with the mama predictions is satisfactory in all cases. The Re(Ra,Pr)Re(Ra,Pr) functional relation is also well described by the expression (56), which is simpler in form than the mama expression (see again Figure 8(b) and Figure 9 (a)). In Figure 9(a), we observe that the compensated Reynolds number expression Re/Ra/PrRe/\sqrt{Ra/Pr}, which is equivalent to the mean root-mean-squared velocity ⟨w2⟩\sqrt{\langle w^2\rangle}, decreases for large PrPr. This is consistent with the fact that the kinematic boundary layer becomes thicker hence the velocity reduce in intensity. The opposite is true for the fluctuations of the temperature field ⟨θ2⟩\sqrt{\langle\theta^2\rangle}, which is reported in Figure 9(b), and for which again the mama solution offers an excellent approximation.

Figure 8. Reynolds-Rayleigh scaling at Pr=1Pr = 1. (a): ReRe vs. RaRa. We display the numerical simulation results (red circle symbols) and comparisons with different analytical predictions: (vertical dotted line) the critical Rayleigh number for the onset of convection Rac≃1558Ra_c \simeq1558; (violet solid line) the Reynolds number calculated from the two-mode steady solution (35); (black solid line) the ReRe expression computed via the matched asymptotic solution; (green dotted-dashed line) the upper bound value Re=Ra/12Re = \sqrt{Ra/12} (55). (b): The same data points for ReRe and theoretical predictions, here compensated with respect to Ra1/2Ra^{1/2}. We include also the approximated expression (56) (blue dashed line) which has a much simpler form as the mama prediction and fits equally well the measurements at large RaRa.

Figure 9. Reynolds number scaling versus PrPr. (a) Compensated Reynolds number Re/Ra/PrRe/\sqrt{Ra/Pr}, equivalent to the root-mean-square value of the velocity ⟨w2⟩\sqrt{\langle w^2\rangle} as a function of Pr∈[10−2,102]Pr \in[10^{-2},10^2] for Ra=106,107,108Ra = 10^6, 10^7, 10^8; (b) root-mean-square value of the temperature fluctuation ⟨θ2⟩\sqrt{\langle\theta^2\rangle}. In both cases we display the numerical simulation results (symbols) and comparisons with the dependencies computed upon integration of the matched asymptotic solutions (47),(48) and the upper bound 1/121/\sqrt{12} (green dashed-dotted) that can be computed from the bulk solution. In (a) the algebraic approximation (56) (dotted) is also reported.

Overall the DNS confirms the realization of the ultimate regime for the Reynolds and Nusselt numbers in respect to both the Rayleigh and Prandtl dependence. To our knowledge the occurrence of this regime was previously assessed only for the 3D homogeneous-Rayleigh-Bénard system in [14] and more recently in a wider RaRa and PrPr range in [37]. Despite its great degree of abstraction (1D, compressible flow) the BRB system represents a second convective model system where this flow regime takes place. Moreover the saturation of NuNu for large-PrPr is a feature of the RB model [8] which is here present, while it was instead missing in the HRB system.

Conclusions

The BRB dynamics shares remarkable similarities with realistic thermal convection in higher spatial dimensions, i.e., the Rayleigh-Bénard system under Oberbeck-Boussinesq conditions. In this work we have shown that: i) BRB has a supercritical linear instability for the onset of convection which solely depends on the Rayleigh number and not on Prandtl (same as in RB), occurring at the critical value Rac≈1558Ra_c \approx1558 which is of the same order as in the RB system; ii) the convective regime is spatially organized in distinct boundary-layers and bulk regions, although shock-like solutions are equally admitted; iii) the asymptotic high RaRa limit displays the ultimate Nusselt and Reynolds numbers scaling regime Nu=RaPr/4Nu=\sqrt{Ra}Pr/4 for Pr≪1Pr\ll1, Nu=Ra/(4π)Nu=\sqrt{Ra}/(4\sqrt{\pi}) for Pr≫1Pr\gg1 and Re=Ra/Pr/12Re=\sqrt{Ra/Pr}/\sqrt{12} thus making BRB the simplest convective system with boundaries exhibiting the ultimate regime of convection. A major difference with realistic higher dimensional natural convection is the absence of turbulence. The BRB dynamics is stationary at all RaRa numbers above the onset of convection for all PrPr values, a feature that results from a nonlinear saturation mechanism.

One may object that the odd symmetry in ww and θ\theta makes the BRB de facto a periodic system, as the fields can be expressed in term of sine series. For this reason in the future it would be interesting to explore the dynamics of this system in a higher (two- or three-) dimensional space. In this case the convective state might be unstable due to the increased number of degrees of freedom and to the increased system symmetries available in higher dimensions. Whether the missing physics in the present model, namely the violation of the incompressibility, the absence of the pressure field or of the spatial lateral dimensions, might be related to the realization of the ultimate regime of heat transfer remains a question for further investigations. In this sense it would be interesting, and perhaps useful, to think up of a similar minimalistic model capable to reproduce the classical scaling of thermal convection (Nu∼Ra1/3Nu \sim Ra^{1/3}), in order to see which key mathematical terms and corresponding physical features are needed for the realization of this different convection regime.

Acknowledgments

The authors are grateful to Prof. M. N. Ouarzazi for useful comments.

Numerical simulation method

Equations (6)–(7) in Fourier space, denoted with ( ~\widetilde{\ }), read:

w~t+w~w~z=−k2PrRaw~+θ~(58)\widetilde{w}_t+\widetilde{w}\widetilde{w}_z=-k^2\sqrt{\frac{Pr}{Ra}}\widetilde{w}+\widetilde{\theta} \tag*{(58)}
θ~t+w~θ~z=−k21PrRaθ~+w~.(59)\widetilde{\theta}_t+\widetilde{w}\widetilde{\theta}_z=-k^2\frac{1}{\sqrt{PrRa}}\widetilde{\theta}+\widetilde{w}. \tag*{(59)}

The dissipative terms can be analytically integrated, while the nonlinear terms can be evaluated via pseudo-spectral algorithm. In our code we adopt the standard 2/3 dealiasing procedure for the computation of the nonlinear terms. Furthermore, we enforce the boundary conditions by imposing that both ww and θ\theta are in real space zero-mean odd-functions (i.e. we use sinus transform instead of Fourier). The temporal discretisation with time step δt\delta t, performed by means of a second order Adams-Bashfort algorithm, leads to:

w~n+1=(w~n+δt2(3[−w~w~z+θ~]n−[−w~w~z+θ~]n−1e−k2PrRaδt))e−k2PrRaδt(60)\widetilde{w}_{n+1}=\left(\widetilde{w}_n+\frac{\delta t}{2}\left(3\left[-\widetilde{w}\widetilde{w}_z+\widetilde{\theta}\right]_n-\left[-\widetilde{w}\widetilde{w}_z+\widetilde{\theta}\right]_{n-1}e^{-k^2\sqrt{\frac{Pr}{Ra}}\delta t}\right)\right)e^{-k^2\sqrt{\frac{Pr}{Ra}}\delta t} \tag*{(60)}
θ~n+1=(θ~n+δt2(3[−w~θ~z+w~]n−[−w~θ~z+w~]n−1e−k21PrRaδt))e−k21PrRaδt,(61)\widetilde{\theta}_{n+1}=\left(\widetilde{\theta}_n+\frac{\delta t}{2}\left(3\left[-\widetilde{w}\widetilde{\theta}_z+\widetilde{w}\right]_n-\left[-\widetilde{w}\widetilde{\theta}_z+\widetilde{w}\right]_{n-1}e^{-k^2\frac{1}{\sqrt{PrRa}}\delta t}\right)\right)e^{-k^2\frac{1}{\sqrt{PrRa}}\delta t}, \tag*{(61)}

where the subscript indexes indicate the discretized value of time. The time step width is chosen as δt=10−2/σ\delta t=10^{-2}/\sigma where σ\sigma is the growth rate of the most unstable mode, eq. (18), while the spatial resolution is increased till when the resulting velocity and temperature profiles become independent on the number of discretization points NN. However, we note that the existence of sharp variations in the solution is at odds with our discretization method based on sine-Fourier transform which is known being affected by from the Gibbs phenomenon. Indeed, at high-RaRa some Gibb’s like spurious fluctuations are seen in correspondence of the bulk to boundary layer transition. This fluctuations do not affect the scaling laws presented in this work. The simulations of the BRB system have been also validated against a second code based on finite-difference discretization.

The temperature and velocity fields in the simulations are initialized with a spatially uncorrelated pseudo-random noise. These perturbations lead in 50% of cases to the BL-type solution and in the remaining cases to the shock solution. Although these two states correspond to different global heat-transfer modes, the resulting velocity and temperature profiles can be transformed one into another by the swap transformation, eq. (8). In the present analysis, focused on the scaling of Nusselt in the BL type state, we take advantage of the swap transformation to maximize the the number of realizations of BL solutions. However, we note that it is possible to direct the instability towards one of the two possible convective states, e.g. by adding a sinusoidal modulation to the initial white noise. A modulation of the form, −sin⁡(2πz)-\sin(2\pi z), leads to BL state while its opposite, sin⁡(2πz)\sin(2\pi z), favours the transition towards the shock type solution. This aspect has an analogous in the RB system where one can control the large-scale circulation direction of the convective cells by initializing the flow with an horizontally asymmetric perturbation of the temperature field. The codes used in this study are available at https://github.com/ecalzavarini/BurgersRB.

References

References

  1. [11]J. D. Cole, On a quasi-linear parabolic equation occurring in aerodynamics, Quarterly of Applied Mathematics 9, 225 (1951).DOI
  2. [12]U. Frisch and J. Bec, Burgulence, in New trends in turbulence Turbulence: nouveaux aspects: 31 July – 1 September 2000, edited by M. Lesieur, A. Yaglom, and F. David (Springer Berlin Heidelberg, Berlin, Heidelberg, 2001) pp. 341–383.
  3. [13]J. Bec and K. Khanin, Burgers turbulence .
  4. [14]E. Calzavarini, D. Lohse, F. Toschi, and R. Tripiccione, Rayleigh and Prandtl number scaling in the bulk of Rayleigh–Bénard turbulence, Physics of Fluids 17, 055107 (2005).DOI
  5. [15]E. Calzavarini, C. R. Doering, J. D. Gibbon, D. Lohse, A. Tanabe, and F. Toschi, Exponentially growing solutions in homogeneous Rayleigh-Bénard convection, Phys. Rev. E 73, 035301 (2006).pubmed.ncbi.nlm.nih.gov/16605590
  6. [16]L. D. Landau and E. M. Lifshitz, Fluid Mechanics, Second Edition: Volume 6 (Course of Theoretical Physics), 2nd ed., Course of theoretical physics / by L. D. Landau and E. M. Lifshitz, Vol. 6 (Butterworth-Heinemann, 1987).
  7. [17]The stability of the three fixed points can be also assessed by means of the analysis of the eigenvalues of the Jacobian of the system of equations (24),(25), and in particular their dependence with respect to the control parameter Ra. This leads to the fact that the equilibrium point A₁ = A₂ = 0 changes from stable to unstable when Ra > Rac, while the two fixed points (31),(32) become stable as long as Rac < Ra < 16Rac. This identifies a supercritical pitchfork bifurcation, see Fig.2.
  8. [18]S. A. Bender, C. M.; Orszag, Advanced Mathematical Methods for Scientists and Engineers (Springer, 199).
  9. [19]J. Goodman and Z. Xin, Viscous limits for piecewise smooth solutions to systems of conservation laws, Arch. Rational Mech. Anal. 121, 235 (1992).DOI
  10. [20]J. P. Boyd, The energy spectrum of fronts: time evolution of shocks in Burgers’ equation, J. Atmo. Sci. 49 (1991).doi.org/10.1175/1520-0469(1992)049<0128:tesoft>2.0.co;2
  11. [21]E. Weinan and E. Vanden Eijnden, Asymptotic theory for the probability density functions in Burgers turbulence, Phys. Rev. Lett. 83 (1999).
  12. [22]G. L. Eyink and T. D. Drivas, Spontaneous stochasticity and anomalous dissipation for Burgers equation, J. Stat. Phys. 158, 386 (2015).
  13. [23]M. Van Dyke, Perturbation methods in fluid mechanics (The Parabolic Press, Stanford,California,USA, 1975).
  14. [24]P. G. Saffman, in Topics in Nonlinear Physics, edited by N. J. Zabusky, (Springer, Berlin, 1968), 485–614 (1968).
  15. [25]S. Alam, P. K. Sahu, and M. K. Verma, Universal functions for burgers turbulence, Phys. Rev. Fluids 7, 074605 (2022).DOI
  16. [26]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, 295–308 (2013).DOI
  17. [27]R. H. Kraichnan, Turbulent thermal convection at arbitrary prandtl number, Phys. Fluids 5, 1374 (1962).DOI
  18. [28]S. Grossmann and D. Lohse, Scaling in thermal convection: a unifying theory, J. Fluid Mech. 407, 27–56 (2000).
  19. [29]C. Doering, Turning up the heat in turbulent thermal convection, PNAS 117, 9671 (2020).DOI
  20. [30]L. E. Schmidt, E. Calzavarini, D. Lohse, F. Toschi, and R. Verzicco, Axially homogeneous Rayleigh–Bénard convection in a cylindrical cell, J. Fluid Mech. 691, 52–68 (2012).DOI
  21. [31]M. Gibert, H. Pabiou, F. Chillà, and B. Castaing, High-Rayleigh-number convection in a vertical channel, Phys. Rev. Lett. 96, 084501 (2006).pubmed.ncbi.nlm.nih.gov/16606185
  22. [32]M. R. Cholemariy and J. H. Arakeri, Axially homogeneous, zero mean flow buoyancy-driven turbulence in a vertical pipe, Journal of Fluid Mechanics 621, 69–102 (2009).
  23. [33]S. Lepot, S. Aumaître, and B. Gallet, Radiative heating achieves the ultimate regime of thermal convection, PNAS 115(36), 8937 (2018).DOI
  24. [34]V. Bouillaut, S. Lepot, S. Aumaître, and B. Gallet, Transition to the ultimate regime in a radiatively driven convection experiment, J. Fluid Mech. 861, R5 (2019).DOI
  25. [35]K. Kawano, S. Motoki, M. Shimizu, and G. Kawahara, Ultimate heat transfer in ‘wall-bounded’ convective turbulence, J. Fluid Mech. 914, A13 (2021).arxiv.org/abs/2004.08831
  26. [36]B. A. Finlayson, The Method of Weighted Residuals and Variational Principles (Academic Press, 1972).
  27. [37]A. Barral and B. Dubrulle, Asymptotic ultimate regime of homogeneous Rayleigh–Bénard convection on logarithmic lattices, J. Fluid Mech. 962, A2 (2023).DOI

Paper details

vlnc:W_n4a2nxnp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents