betaFeedback

WorksPhysics and Astronomy

The Lagrangian kinetic energy cascade in Rayleigh-Bénard convection

Abstract

Rayleigh-Bénard convection at high Rayleigh number exhibits turbulence superimposed on large-scale circulation. While buoyancy forces drive the flow at certain scales, how kinetic energy is transfers across the scales is not understood. Here, utilizing a Lagrangian description of the kinetic energy flux, we present experimental evidence of a split cascade where energy flows downwscale at small scales and upscale at large scales. The flow topology of these energy transfer events differ profoundly, and the transition between them occurs gradually, over a broad range of scales.

Introduction— Fluid density differences can drive fluid motion through conversion between potential and kinetic energies. This mechanism, so-called natural convection, is at the heart of numerous flows widely abundant in nature and industry. One of the canonical examples is the Rayleigh-Bénard convection, in which heat supplied to the bottom of a fluid-filled vessel and extracted from its top lead to fluid flow. For sufficiently high Rayleigh numbers (Ra≳106\mathrm{Ra} \gtrsim10^{6}) turbulent flow is superimposed upon large-scale circulation (LSC) [2, 3] [1]. Due to the relative simplicity of its configuration and the complex physical phenomena it displays, the Rayleigh-Bénard convection has been widely used in the past to study the fundamental aspects of stratified turbulent flows (see [3, 4] and references therein).

A hallmark of turbulent flows is the energy cascade: kinetic energy is transferred from one scale to another through inertial flow structures [5]. In the inertial range of three-dimensional turbulence the average energy flux is in the downscale direction—from larger to smaller scales. Thus, kinetic energy produced at large scales by an external forcing is transferred toward small scales to be dissipated there by viscous friction [6]. In certain situations, such as in two-dimensional, rotational, and certain convective turbulent flows [7, 8, 9, 10, 11, 12, 13], the kinetic energy cascade can be reversed; there, the mean kinetic energy flux is in the upscale direction—from smaller to larger scales—and dissipation occurs due to friction of the large-scale flow structures against the domain boundaries. However, in turbulent Rayleigh-Bénard convection, the picture of the cascade is more complex [14, 15]. Forcing is applied to the fluid through buoyancy forces of thermal plumes that detach from the hot and cold plates [16, 17], which can occur on a wide range of scales [5]. Then, kinetic energy transfer across scales is split: it flows both downscale, and upscale [12, 18], dissipating due to viscous friction in the bulk and the boundary layers [19]. Interestingly, a similar co-existence of LSC and turbulent fluctuations occur in buoyancy-driven bubbly flows [20]. Overall, this split in energy pathways leads to the following questions: What are the characteristics of the transition (in scale) between downward and upward energy transfers, and how is this picture manifested in the dynamics of flow observables?

Energy fluxes in turbulence are often examined through the Eulerian approach, in which flow observables are analyzed in a reference frame fixed in space, e.g. [21, 22, 23, 24, 25]. Here, however, we utilize the Lagrangian framework, analyzing flow observables along flow-tracing particles’ trajectories. Advances in the Lagrangian description of turbulence have enhanced our understanding of transport and dispersion processes [26, 27], and energy transfer in turbulent flows [28, 29, 30, 31]. In addition, conservation laws are often more naturally understood in the Lagrangian framework. Therefore, here we leverage the Lagrangian description to better understand turbulence dynamics in Rayleigh-Bénard convection. As we show below, there is a crossover in the direction of kinetic energy transfer across the scales from downscale at small scales to upscale at large scales. Detailed analysis shows that the transition is gradual, through an intermediate mixed regime, and that the flow structures associated with the downscale and upscale fluxes have different topology.

A Lagrangian view energy flux— Temporal changes of tracer particles’ kinetic energy are related to energy transfer in the flow [13, 28, 31, 32, 33, 34]. Under the Boussinesq approximation the acceleration of fluid particles is governed by the Navier-Stokes equation

v˙=∇⋅σ+fb=−∇p+2ν∇⋅S+fb(1)\dot{\boldsymbol{v}}=\nabla\cdot\boldsymbol{\sigma}+\boldsymbol{f}_b=-\nabla p+2\nu\nabla\cdot\boldsymbol{S}+\boldsymbol{f}_b \tag*{(1)}

where σ≡−pI+2νS\boldsymbol{\sigma}\equiv-p\boldsymbol{I}+2\nu\boldsymbol{S} is the stress tensor, fb\boldsymbol{f}_b is the buoyancy force density, S=12(∇v+∇vT)\boldsymbol{S}=\frac{1}{2}(\nabla\boldsymbol{v}+\nabla\boldsymbol{v}^{\mathsf{T}}) is the rate of strain tensor, pp is the pressure divided by density and ν\nu is the viscosity. Multiplying eq. (1) by v\boldsymbol{v} and averaging in space and time gives

⟨12v˙2⟩+∇⟨T⟩=ϵb−ϵk(2)\left\langle\frac{1}{2}\dot{v}^{2}\right\rangle+\nabla\langle\boldsymbol{T}\rangle=\epsilon_b-\epsilon_k \tag*{(2)}

Here, T=pv−2νv⋅S\boldsymbol{T}=p\boldsymbol{v}-2\nu\boldsymbol{v}\cdot\boldsymbol{S}, and the average of its divergence is ⟨∇T⟩=0\langle\nabla\boldsymbol{T}\rangle=0 due to Stokes theorem. The mean density of kinetic energy dissipation rate in eq. (2) is ϵk=2ν⟨T(SS)⟩\epsilon_k=2\nu\langle\boldsymbol{T}(\boldsymbol{S}\boldsymbol{S})\rangle. In addition, mechanical work is applied to the fluid at a mean rate ϵb≡⟨vfb⟩\epsilon_b\equiv\langle\boldsymbol{v}\boldsymbol{f}_b\rangle. In statistically steady states ⟨v˙2⟩=0\langle\dot{v}^{2}\rangle=0, so the work and dissipation are equal ϵb=ϵk\epsilon_b=\epsilon_k.

As our goal is to explore energy transfer across scales we consider the relative motion between fluid tracers. The instantaneous relative velocity between two fluid tracers is δrv≡v(x+r,t)−v(x,t)\delta_r\mathbf{v}\equiv\mathbf{v}(\mathbf{x}+\mathbf{r},t)-\mathbf{v}(\mathbf{x},t), and the kinetic energy of their relative motion is k=12δrv2k=\frac{1}{2}\delta_r\mathbf{v}^2. Following Mann et al. [32], we identify the mean rate of change of the relative motion kinetic energy, ⟨k˙⟩r\langle\dot{k}\rangle_r, as the flux of kinetic energy from scales larger than rr to those smaller than rr (where ⟨⋅⟩r\langle\cdot\rangle_r is a space and time average over all trajectory pairs with distance ∣r∣=r|\mathbf{r}|=r). Similarly, δrv\delta_r\mathbf{v} holds the minimum information needed to define the coarse-grained velocity and stress resolved at a scale rr which also supports identifying ⟨k˙⟩r\langle\dot{k}\rangle_r as the kinetic energy flux [28, 35].

To obtain an equation for ⟨k˙⟩r\langle\dot{k}\rangle_r we subtract eq. (1) at point x1\mathbf{x}_1 from itself at x2\mathbf{x}_2 with ∣x1−x2∣=r|\mathbf{x}_1-\mathbf{x}_2|=r, and then multiply by δrv\delta_r\mathbf{v}

k˙=(v1−v2)(∇σ1−∇σ2+fb1−fb2).(3)\dot{k}=(\mathbf{v}_1-\mathbf{v}_2)\left(\nabla\sigma_1-\nabla\sigma_2+\mathbf{f}_{b1}-\mathbf{f}_{b2}\right). \tag*{(3)}

Averaging over time and space, and recalling that ∇⟨T⟩=0\nabla\langle T\rangle=0, we obtain

⟨k˙⟩r=2ϵb−2ϵk−2⟨v∇σ∗⟩r−2⟨vfb∗⟩r(4)\langle\dot{k}\rangle_r=2\epsilon_b-2\epsilon_k-2\langle\mathbf{v}\nabla\sigma^*\rangle_r-2\langle\mathbf{v}\mathbf{f}_b^*\rangle_r \tag*{(4)}

where terms of the form ⟨vX∗⟩r\langle\mathbf{v}X^*\rangle_r represent a covariance between the velocity at one point and a property XX at a point separated by distance rr. As shown in eq. (2), the ϵk\epsilon_k and ϵb\epsilon_b terms cancel each other out. Furthermore, as the strain rate is a small scale property in turbulence [36] we expect that ⟨v∇S∗⟩r≈0\langle\mathbf{v}\nabla S^*\rangle_r\approx0 for rr above the dissipation range (r≫ηr\gg\eta where η=(ν3/ϵk)1/4\eta=(\nu^3/\epsilon_k)^{1/4} is the Kolmogorov scale). Overall, the flux results from spatial correlations between the fluid velocity and buoyancy force, and between the fluid velocity and pressure across distance rr

⟨k˙⟩r=2⟨v∇p∗⟩r−2⟨vfb∗⟩r(for η≪r).(5)\langle\dot{k}\rangle_r=2\langle\mathbf{v}\nabla p^*\rangle_r-2\langle\mathbf{v}\mathbf{f}_b^*\rangle_r\qquad(\text{for }\eta\ll r). \tag*{(5)}

In contrast, in the inertial range of homogeneous isotropic turbulence (HIT), ⟨k˙⟩r=−2ϵk\langle\dot{k}\rangle_r=-2\epsilon_k for η≪r≪L\eta\ll r\ll L as the flux is constant [32, 33].

Methods— In what follows, we explore the statistics of the rate of change of the kinetic energy of the relative velocity (k˙\dot{k}) in a Rayleigh-Bénard convection experimentally. The flow in the experiment was in a cubic cell with side lengths L=300mmL=300\mathrm{mm}, containing water as the working fluid. The flow was heated from below and cooled from above with two aluminum plates kept at a constant temperature difference of δT=6.5∘±0.1∘C\delta T=6.5^\circ\pm0.1^\circ\mathrm{C}. Tracer particles with 20 μm20\,\mu\mathrm{m} mean diameter and density 1.03 g cm−31.03\,\mathrm{g\,cm^{-3}} were seeded and tracked in the flow using proPTV—an open source 3D particle tracking velocimetry (3D-PTV) software [37, 38]. In this work, we analyzed a dataset taken from a 30 hour experiment carried out to study the LSC [39]. The dataset we used for the analysis contained ∼2.5×105\sim2.5\times10^5 trajectories taken from time windows of approximately 1.5 hours. Figure 1a visualizes the flow with a 3D rendering of trajectories from the experiment. The Rayleigh number, characterizing the thermal forcing in the experiment was Ra=2.5×109\mathrm{Ra}=2.5 \times10^{9}, and the mean free-fall time, characterizing the average period it takes a Lagrangian tracer to complete one half of the circulation cycle, is approximately 40s (Figure 1b, Figure 1c). Full experimental details are given in Ref. [39].

Figure 1. Characterization of the experimental system. (a) 3D rendering of approximately 2000 trajectories measured in the experiment. Colors represent the instantaneous vertical velocity where blue is directed with gravity and red is directed against it. (b) Examples of the zz coordinate of four tracer particles; the vertical line shows the center of the tank, zc=12Lz_c=\frac{1}{2}L and τzc\tau_{z_c} is the time duration between two consecutive z=zcz=z_c crossings. (c) Histogram of τzc\tau_{z_c} with a dashed line showing its mean value, ⟨τzc⟩=39\langle\tau_{z_c}\rangle=39 s. (d) Picture of the experimental system including the water tank and, illumination and photography equipment used in the measurements. (e) Longitudinal second order structure function shown in log-log scales where its logarithmic slope is shown in (f). Two scaling law are shown corresponding to the K41 and BO59 theories.

To characterize the scales relevant for the turbulent flow in the experiment we show in Figure 1e the longitudinal velocity second order structure function,

S2,L(r)=⟨(δrv)2⟩,(6)S_{2,L}(r)=\left\langle\left(\delta_r v\right)^2\right\rangle, \tag*{(6)}

and in Figure 1f we show its local logarithmic slope,

ζ(r)=dlog⁡S2,Ldlog⁡r.(7)\zeta(r)=\frac{d\log S_{2,L}}{d\log r}. \tag*{(7)}

We obtained these results by binning samples of δrv\delta_r v according to rr, each bin holding several millions of samples (between ∼3×106\sim3 \times10^{6} for the smallest rr bin and up to ∼50×106\sim50 \times10^{6}), and calculated statistics over these ensembles. During bin assembly we took samples from the whole region of measurement which consists the whole fluid tank excluding region very near to the walls. This technique was used to calculate the results discussed throughout this paper. Mean field theory for Rayleigh-Bénard convection predicts S2,L∝rζ(r)S_{2,L}\propto r^{\zeta(r)}, with three characteristic regimes [15]. For r≪ηr\ll\eta the scaling ζ(r)=2\zeta(r)=2 corresponds to the dissipation range, which is not resolved in our experiment. For η≪r≪LB\eta\ll r\ll L_{B} an inertial range is expected with the K41 scaling of ζ(r)=23\zeta(r)=\frac{2}{3} (LBL_{B} is the Bolgiano scale). A narrow inertial range scaling is seen in our data for 0.15L≲r≲0.25L0.15L\lesssim r\lesssim0.25L in which ζ=0.67±0.06\zeta=0.67\pm0.06. The lower end of this range suggests that the Kolmogorov scale in our flow is on the order of a millimeter η∼O(1) mm\eta\sim O(1)\,\mathrm{mm}, and the velocity dissipation rate is on the order of ϵk∼O(1) mm2s−3\epsilon_k\sim O(1)\,\mathrm{mm}^{2}\mathrm{s}^{-3}. Above LBL_{B} buoyancy forces become dominant and the Bolgiano-Obukhov theory (BO59) [40] leads to ζ=65\zeta=\frac{6}{5}. In Figure 1f, a narrow range of ζ=1.3±0.07\zeta=1.3\pm0.07 is seen for 0.3L≲r≲0.4L0.3L\lesssim r\lesssim0.4L, slightly higher than the BO59 scaling. The OB59 scaling is known to be elusive in empirical studies [41], and the higher value of ζ\zeta observed here can be attributed to shear effects associated with the LSC [15]. All in all, we estimate the Bolgiano scale in our experiment to be approximately LB≈0.3LL_{B}\approx0.3L.

Flux statistics

The statistical behavior of k˙\dot{k} as a function of rr in the Rayleigh-Bénard convection is characterized in Figure 2. The distribution of k˙\dot{k} is non-Gaussian and has exponential tails (for example, Figure 2a, inset). Consequently, extreme events of k˙\dot{k} being many standard deviations away from the mean occur at probabilities much higher relative to a normal distribution at all scales we tested. For example, the skewness factor of k˙\dot{k} (F(r)=⟨(k˙′)4⟩r/⟨(k˙′)2⟩r2F(r)=\left\langle(\dot{k}')^{4}\right\rangle_r/\left\langle(\dot{k}')^{2}\right\rangle_r^2 where k˙′=k˙−⟨k˙⟩r\dot{k}'=\dot{k}-\langle\dot{k}\rangle_r) decreased with rr from approximately 200 at r=0.05Lr=0.05L to approximately 30 at r=0.5Lr=0.5L (not shown); on the other hand, assuming the velocity and the acceleration were normally distributed and uncorrelated we would expect the flatness factor of k˙\dot{k} to be 9 (as k˙=∑iviv˙i\dot{k}=\sum_i v_i\dot{v}_i). It is indeed a well-known feature of three dimensional turbulence that small scale properties, such as the acceleration, have a high flatness factor—a phenomenon called intermittency [42, 43, 44]. Nevertheless, while in HIT the flatness of velocity increments decreases with rr throughout the inertial range and reaches close to Gaussian values (F(50η)≈4F(50\eta)\approx4 in HIT [45]), the flatness of k˙\dot{k} observed here remained very high all the way up to r=12Lr=\frac{1}{2}L, which is above the inertial range and above LBL_{B} (Figure 1). We believe that the high flatness at such large scales is due to the signature of the driving mechanism in the Rayleigh-Bénard cell, namely the thermal plumes.

Figure 2. Statistics of k˙\dot{k} as a function of separation. Panel (a) shows the rr conditional mean of k˙\dot{k}, and a PDF of k˙\dot{k} for r=0.32Lr=0.32L is shown in the inset and compared with a Gaussian distribution. Panels (b), (c) and (d) show the second, third and fourth central moments of k˙\dot{k} as a function of rr respectively. The dotted and dashed lines show the transitional values, r∗r^{*} and r†r^{\dagger} respectively (see text).

We observe two transitions in the behavior of k˙\dot{k} as a function of rr. The first transition, shown as dotted line in Figure 2, is identified by the observation that for r<r∗r<r^{*} where r∗=(0.06±0.013)Lr^{*}=(0.06\pm0.013)L the mean ⟨k˙⟩r\langle\dot{k}\rangle_r is negative and it decreases with rr, while for r>r∗r>r^{*}, ⟨k˙⟩r\langle\dot{k}\rangle_r increases with rr; the same is observed for the third moment, ⟨(k˙′)3⟩r\langle(\dot{k}')^{3}\rangle_r. Furthermore, for r<r∗r<r^{*} the second and fourth moment of k˙\dot{k} increase sharply with rr while they grow more gradually with rr for r>r∗r>r^{*}. In the inertial range of HIT we would expect the mean ⟨k˙⟩r\langle\dot{k}\rangle_r and the third moment ⟨(k˙′)3⟩r\langle(\dot{k}')^3\rangle_r to plateau with rr (as ⟨k˙⟩r=−2ϵk\langle\dot{k}\rangle_r=-2\epsilon_k, and see [34]), and such plateaus are not observed in our Rayleigh-Bénard flow (recall, the inertial range coincides with 0.15L≲r≲0.25L0.15L \lesssim r \lesssim0.25L, slightly above r⋆r^{\star}). The growth of ⟨k˙⟩r\langle\dot{k}\rangle_r in the inertial range suggests that the downscale kinetic energy flux is not constant in the inertial range as it decreases with rr.

The second transition, shown with dashed lines in Figure 2, is associated with a crossover of ⟨k˙⟩r\langle\dot{k}\rangle_r from negative to positive values, at r†=0.32±0.01Lr^{\dagger}=0.32\pm0.01L. At the same scale, there is also an increase in the growth rate of the second and fourth moments with rr signaling on higher ∣k˙∣|\dot{k}| values for r>r†r>r^{\dagger}, in addition to a slight decrease in the growth rate of the third moment of k˙\dot{k} with respect to rr. Having a negative mean, ⟨k˙⟩r<0\langle\dot{k}\rangle_r<0, is expected at small scales due to the downscale directed cascade [32, 33]. This transition roughly coincides with the Bolgiano scale, LBL_B. Thus, the transition to ⟨k˙⟩r>0\langle\dot{k}\rangle_r>0 denotes a transition to upscale cascade for r>LBr>L_B in which thermal plumes drive the LSC (eq. (4)). This is in agreement with the split cascade observed in previous Eulerian studies [18, 46].

Flow structure and energy flux– The downward kinetic energy flux in three dimensional turbulence can occur through two processes: strain self-amplification and vortex stretching [23, 36]; these processes support downscale energy transfer if the strain skewness is negative and vortex stretching is positive respectively. Previous studies in three dimensional turbulence have shown that the downscale cascade is mostly correlated with straining regions [23, 28] which suggests that strain self-amplification is the more dominant mechanism for downscale energy transfer. Furthermore, the flow structure around intense energy transfer events has a saddle-node topology where the coarse-grained rate-of-strain tensor has two positive and one negative eigenvalues, resulting in bi-axial strain [25]. On the other hand, the flow topology responsible for the upscale transfer is much less known. Two mechanisms have been suggested: equal-sign vortex merger and vortex thinning by large scale strain, where the latter has received stronger support from numerical simulations [47, 48]. Notably, these mechanisms were suggested for purely 2D turbulent flows which can be different from the split cascade in 3D flows [49].

To better understand the flow structure associated with the two transitions observed above we examine the joint distribution and conditional mean of k˙\dot{k} with the separation velocity between particles

δvr≡δv⋅rr.(8)\delta v_r \equiv\delta\mathbf{v}\cdot\frac{\mathbf{r}}{r}. \tag*{(8)}

Cases in which δvr<0\delta v_r<0 correspond to converging trajectories while δvr>0\delta v_r>0 means separating trajectories. Notably, in bi-axial strain there are two divergence axes versus only one convergence axis so in such a flow topology converging trajectories will tend to have higher ∣δvr∣|\delta v_r| than diverging trajectories; in axial strain the opposite

will occur and separating trajectories will tend to have higher ∣δvr∣|\delta v_r|. Thus, observing values of δvr\delta v_r can hint of topological aspects of the flow.

In Figure 3 we show joint statistics of k˙\dot{k} and δvr\delta v_r as a function of rr. In the small scale regime, r≲r⋆r\lesssim r^{\star}, the joint PDF is biased towards negative k˙\dot{k} and δvr\delta v_r values. In particular, the conditional mean ⟨k˙∣δvr⟩r\langle\dot{k}\mid\delta v_r\rangle_r for δvr<0\delta v_r<0 (namely for converging trajectories) is negative and it increases monotonically with δvr\delta v_r, while it is approximately ⟨k˙∣δvr⟩r=0\langle\dot{k}\mid\delta v_r\rangle_r=0 for δvr>0\delta v_r>0 (namely separating trajectories). Because ⟨k˙⟩r<0\langle\dot{k}\rangle_r < 0 characterizes the downward cascade, this behavior supports the association of downscale kinetic energy flux with bi-axial flow topology for r<r∗r < r^{*}. Namely, our results support the view that the Rayleigh-Bénard convection at small scales displays cascade characteristics similar to the three dimensional Kolmogorov turbulence.

Joint probability distributions and conditional means

Figure 3. (a) Joint probability distribution functions of the relative kinetic energy rate of change and the particles’ separation velocity normalized using their standard deviations. Data shown for four rr bins. Dashed lines show the conditional mean ⟨k˙∣δvr⟩r\langle\dot{k}\mid\delta v_r\rangle_r as a function of the separation velocity where red dots show its zero crossings. (b) Conditional mean ⟨k˙∣δvr⟩r\langle\dot{k}\mid\delta v_r\rangle_r for ten rr values uniformly spread starting from r=0.078Lr=0.078L and up to r=0.47Lr=0.47L.

For scales r≳r∗r \gtrsim r^{*} and up to slightly above r†r^{\dagger}, there is a different behavior in which an inflection point appears for the conditional mean ⟨k˙∣δvr⟩r\langle\dot{k}\mid\delta v_r\rangle_r at δvr≈0\delta v_r \approx0 (Figure 3b). Strongly converging trajectories (δvr≪0\delta v_r \ll0 events) are still associated with ⟨k˙∣δvr⟩r\langle\dot{k}\mid\delta v_r\rangle_r being negative, yet for weaker δvr<0\delta v_r < 0 trajectories ⟨k˙∣δvr⟩r\langle\dot{k}\mid\delta v_r\rangle_r is positive. For separating trajectories the opposite occurs: strongly separating trajectories (δvr≫0\delta v_r \gg0 events) are associated with ⟨k˙∣δvr⟩r\langle\dot{k}\mid\delta v_r\rangle_r being positive, while weaker separating trajectories are associated with ⟨k˙∣δvr⟩r\langle\dot{k}\mid\delta v_r\rangle_r being negative. These behaviors, characterized by the negative slope at δvr=0\delta v_r = 0, intensify as rr increases in this range.

Lastly, for r≫r†r \gg r^{\dagger} the inflection point disappears and ⟨k˙∣δvr⟩r\langle\dot{k}\mid\delta v_r\rangle_r decreases monotonically with δvr\delta v_r crossing zero from above at δvr≈0\delta v_r \approx0. Therefore, for scales r≫r†r \gg r^{\dagger}, strongly converging trajectories are associated with ⟨k˙∣δvr⟩r>0\langle\dot{k}\mid\delta v_r\rangle_r > 0 while separating trajectories are associated with ⟨k˙∣δvr⟩r<0\langle\dot{k}\mid\delta v_r\rangle_r < 0. As ⟨k˙⟩r\langle\dot{k}\rangle_r characterizes the direction of kinetic energy flux, this suggests that the flow structures associated with energy transfer at the large scales regime differ from those at the small-scale regime. Specifically, the mean energy transfer in this regime is in the upscale direction (⟨k˙⟩r>0\langle\dot{k}\rangle_r > 0), and the events that contribute to this upscale transfer are more associated with converging trajectories. Furthermore, the downscale transfer events in this regime (k˙<0\dot{k} < 0) occur mostly in association with separating trajectories; this is in stark contrast with the observation above that downscale flux events at small scales are correlated with converging trajectories.

Discussion– The transitions in the behavior of k˙\dot{k} with respect to scale thus correspond to three regimes of energy transfer in the Rayleigh-Bénard flow. At small scales near the dissipation range the mean energy transfer is downscale, and, similar to HIT flows, the flux is associated with converging trajectories flow topology. At large scales, r>LBr > L_B, the energy transfer is mostly in the upscale direction and associated with converging trajectories, while the downscale energy transfer events in this regime are associated with separating trajectories. Therefore, in the small and large scale regimes both the mean direction of energy transfer and the correlation between flow topology and energy flux are reversed. This suggests that the flow structures associated with downscale and upscale energy transfer are not simply time-reversed images of one-another, so they occur due to different physical mechanisms. The third regime, r∗<r<LBr^{*} < r < L_B, is a mixed inertial regime laying between the former two; there, intense energy transfer events (namely farther from the mean) are of the type observed in the HIT-like downscale cascade, while the less intense events are of the type observed in upscale cascade regime. The intermediate regime manifests a smooth transition between the upscale-downscale cascades. A possible explanation for the behavior observed, presented conceptually in Figure 4, is that downscale and upscale transfer events occur simultaneously over a wide range of inertial scales above r∗r^{*} with different proportions, and these balance each other out around LBL_B.

Figure 4. A conceptual sketch for the energy transfer regimes in Rayleigh-Bénard turbulence.

Conclusions– This work explores the transfer of kinetic energy across scales in Rayleigh-Bénard convection through the Lagrangian framework. We observe three energy transfer regimes based on scale. At small scales the mean energy flux is in the downward direction while at large scale the mean flux is in the upward direction. The third regime is an intermediate one that mixes the behaviors of the former two where more extreme events were correlated with downscale transfer while less intense events were associated with upscale transfer. Furthermore, the flow topology observed for the different downscale and upscale cascade regimes was different in the small and large scales regimes. This shows that split cascade turbulent flows present a mixture of inertial energy transfer mechanisms that occur simultaneously but are separated in scale-space. Furthermore, it shows that the physical mechanisms responsible for the down- and upscale energy transfer turbulent flows are not simply time-reversed images, but are profoundly different from one another, thus highlighting the irreversibly associated with turbulent flows.

Acknowledgments

The authors thank Prof. Alain Pumir for insightful comments on the initial draft. RS acknowledges support from ISF grants 1244/24 and 2586/24 and from the Alon Scholarship.

References

References

  1. [2]A. Pandey, J. D. Scheel, and J. Schumacher, “Turbulent superstructures in Rayleigh-Bénard convection,” Nature Communications 9 (2018), 10.1038/s41467-018-04478-0.
  2. [3]D. Lohse and O. Shishkina, “Ultimate Rayleigh-Bénard turbulence,” Rev. Mod. Phys. 96, 035001 (2024).DOI
  3. [4]D. Lohse, “Asking the right questions on Rayleigh–Bénard turbulence,” Journal of Fluid Mechanics 1000 (2024), 10.1017/jfm.2024.871.DOI
  4. [5]A. Alexakis and L. Biferale, “Cascades and transitions in turbulent flows,” Physics Reports 767, 1–101 (2018).DOI
  5. [6]U. Frisch, Turbulence: The legacy of A.N. Kolmogorov (Cambridge University Press, 1995).
  6. [7]R. H. Kraichnan, “Inertial ranges in two-dimensional turbulence,” Physics of fluids 10, 1417 (1967).DOI
  7. [8]J. Paret and P. Tabeling, “Experimental observation of the two-dimensional inverse energy cascade,” Phys. Rev. Lett. 79, 4162–4165 (1997).DOI
  8. [9]H. Xia, M. Shats, and G. Falkovich, “Spectrally condensed turbulence in thin layers,” Physics of Fluids 21 (2009), 10.1063/1.3275861.DOI
  9. [10]H. Xia, D. Byrne, G. Falkovich, and M. Shats, “Upscale energy transfer in thick turbulent fluid layers,” Nature Physics 7, 321–324 (2011).DOI
  10. [11]B. Favier, J. L. Silvers, and M. R. E. Proctor, “Inverse cascade and symmetry breaking in rapidly rotating Boussinesq convection,” Physics of Fluids 26 (2014), 10.1063/1.4859131.DOI
  11. [12]P. P. Vieweg, J. D. Scheel, R. Stepanov, and J. Schumacher, “Inverse cascades of kinetic energy and thermal variance in three-dimensional horizontally extended turbulent convection,” Physical Review Research 4, 043098 (2022).DOI
  12. [13]S. Gallon, A. Sozza, F. Feraco, R. Marino, and A. Pumir, “Lagrangian irreversibility and energy exchanges in rotating-stratified turbulent flows,” Phys. Rev. Lett. 133, 024101 (2024).DOI
  13. [14]V. S. L’vov, “Spectra of velocity and temperature fluctuations with constant entropy flux of fully developed free-convective turbulence,” Phys. Rev. Lett. 67, 687–690 (1991).pubmed.ncbi.nlm.nih.gov/10044963
  14. [15]D. Lohse and K.-Q. Xia, “Small-scale properties of turbulent Rayleigh-Bénard convection,” Annual Review of Fluid Mechanics 42, 335–364 (2010).DOI
  15. [16]E. M. Sparrow, R. B. Husar, and R. J. Goldstein, “Observations and other characteristics of thermals,” Journal of Fluid Mechanics 41, 793–800 (1970).DOI
  16. [17]L. P. Kadanoff, “Turbulent heat flow: structures and scaling,” Physics today 54, 34–39 (2001).DOI
  17. [18]R. Togni, A. Cimarelli, and E. De Angelis, “Physical and scale-by-scale analysis of Rayleigh-Bénard convection,” Journal of Fluid Mechanics 782, 380–404 (2015).DOI
  18. [19]B. Gayen, G. O. Hughes, and R. W. Griffiths, “Completing the mechanical energy pathways in turbulent Rayleigh-Bénard convection,” Physical Review Letters 111, 124301 (2013).pubmed.ncbi.nlm.nih.gov/24093264
  19. [20]R. Shnapp and M. Holzner, “Bubble-induced convection and flow instability in liquid vessels,” Journal of Fluid Mechanics 996, A2 (2024).DOI
  20. [21]C. Härtel, L. Kleiser, F. Unger, and R. Friedrich, “Subgrid-scale energy transfer in the near-wall region of turbulent flows,” Physics of Fluids 6, 3130–3143 (1994).DOI
  21. [22]G. L. Evink, “Local energy flux and the refined similarity hypothesis,” Journal of Statistical Physics 78, 335–351 (1995).
  22. [23]V. Borue and S. A. Orszag, “Local energy flux and subgrid-scale statistics in three-dimensional turbulence,” Journal of Fluid Mechanics 366, 1–31 (1998).DOI
  23. [24]J. A. Domaradzki, “Large eddy simulations of high Reynolds number turbulence based on interscale energy transfer among resolved scales,” Phys. Rev. Fluids 6, 044609 (2021).DOI
  24. [25]D. Park and A. Lozano-Durán, “The coherent structure of the energy cascade in isotropic turbulence,” Scientific Reports 15, 14 (2025).DOI
  25. [26]R. Shnapp, S. Brizzolara, M. M. Neamtu-Halic, A. Gambino, and M. Holzner, “Universal alignment in turbulent pair dispersion,” Nature Communications 14, 4195 (2023).DOI
  26. [27]P. P. Vieweg, A. Klünker, J. Schumacher, and K. Padberg-Gehle, “Lagrangian studies of coherent sets and heat transport in constant heat flux-driven turbulent Rayleigh-Bénard convection,” European Journal of Mechanics-B/Fluids 103, 69–85 (2024).DOI
  27. [28]A. Pumir, B. I. Shraiman, and M. Chertkov, “The Lagrangian view of energy transfer in turbulent flow,” Europhysics Letters 56, 379 (2001).DOI
  28. [29]M. Wan, Z. Xiao, C. Meneveau, G. L. Eyink, and S. Chen, “Dissipation-energy flux correlations as evidence for the Lagrangian energy cascade in turbulence,” Physics of Fluids 22 (2010), 10.1063/1.3447887.DOI
  29. [30]C. Meneveau, “Lagrangian dynamics and models of the velocity gradient tensor in turbulent flows,” Annual Review of Fluid Mechanics 43, 219–245 (2011).DOI
  30. [31]A. Pumir, H. Xu, E. Bodenschatz, and R. Grauer, “Single-particle motion and vortex stretching in three-dimensional turbulent flows,” Phys. Rev. Lett. 116, 124502 (2016).DOI
  31. [32]J. Mann, S. Ott, and J. S. Andersen, Experimental study of relative, turbulent diffusion, Denmark. Forskningscenter Risoe. Risoe-R No. 1036(EN) (1999).
  32. [33]G. Falkovich, K. Gawedzki, and M. Vergassola, “Particles and fields in fluid turbulence,” Reviews of modern Physics 73, 913 (2001).DOI
  33. [34]H. Xu, A. Pumir, G. Falkovich, E. Bodenschatz, M. Shats, H. Xia, N. Francois, and G. Boffetta, “Flight–crash events in turbulence,” Proceedings of the National Academy of Sciences 111, 7558–7563 (2014).DOI
  34. [35]M. Germano, “Turbulence: the filtering approach,” Journal of Fluid Mechanics 238, 325–336 (1992).DOI
  35. [36]A. Tsinober, An informal conceptual introduction to turbulence (Springer, 2009).
  36. [37]R. Barta, C. Bauer, S. Herzog, D. Schiepel, and C. Wagner, “proptv: A probability-based particle tracking velocimetry framework,” Journal of computational physics 514, 113212 (2024).DOI
  37. [38]R. Barta, A. Liberzon, and R. Shnapp, “Comparison of camera calibration methods for particle tracking velocimetry,” Measurement Science and Technology (2025), 10.1088/1361-6501/adc6a2.DOI
  38. [39]R. Barta and C. Wagner, “Large-scale reorientation in cubic Rayleigh–Bénard convection measured with particle tracking velocimetry,” Journal of Turbulence , 1–14 (2024).DOI
  39. [40]R. Jr. Bolgiano, “Turbulent spectra in a stably stratified atmosphere,” Journal of Geophysical Research 64, 2226–2229 (1959).DOI
  40. [41]M. Kaczorowski and K.-Q. Xia, “Turbulent flow in the bulk of Rayleigh-Bénard convection: small-scale proper- ties in a cubic cell,” Journal of Fluid Mechanics 722, 596–617 (2013).
  41. [42]A. Arnéodo, R. Benzi, J. Berg, L. Biferale, E. Bō-denschatz, A. Busse, E. Calzavarini, B. Castain, M. Cencini, L. Chevillard, et al., “Universal intermittent properties of particle trajectories in highly turbulent flows,” Physical Review Letters 100, 254504 (2008).
  42. [43]J. Schumacher, “Lagrangian studies in convective turbulence,” Physical Review E–Statistical, Nonlinear, and Soft Matter Physics 79, 056301 (2009).DOI
  43. [44]R. Shnapp, “On small-scale and large-scale intermittency of Lagrangian statistics in canopy flow,” Journal of Fluid Mechanics 913, R2 (2021).DOI
  44. [45]P. Tabeling, G. Zocchi, F. Belin, J. Maurer, and H Will aime, “Probability density functions, skewness, and flatness in large Reynolds number turbulence,” Phys-ical Review E 53, 1613 (1996).
  45. [46]G. Boffetta, F. De Lillo, A. Mazzino, and S. Musacchio, “Boligiano scale in confined Rayleigh–Taylor turbulence,” Journal of fluid mechanics 690, 426–440 (2012).DOI
  46. [47]S. Chen, R. E. Ecke, G. L. Eyink, M. Rivera, M. Wan, and Z. Xiao, “Physical mechanism of the two-dimensional inverse energy cascade,” Phys. Rev. Lett. 96, 084502 (2006).pubmed.ncbi.nlm.nih.gov/16606186
  47. [48]Z. Xiao, M. Wan, S. Chen, and G. L. Eyink, “Physical mechanism of the inverse energy cascade of two-dimensional turbulence: a numerical investigation,” Journal of Fluid Mechanics 619, 1–44 (2009).DOI
  48. [49]S. Musacchio and G. Boffetta, “Split energy cascade in turbulent thin fluid layers,” Physics of Fluids 29 (2017), 10.1063/1.4986001.DOI

Paper details

vlnc:W_32xqxaaq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

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