betaFeedback

WorksPhysics and Astronomy

Covering convection with a thermal blanket: numerical simulation and stochastic modeling

Abstract

Adding moving boundaries to convective fluids is known to result in nontrivial and surprising dynamics, leading to spectacular geoformations ranging from the kilometer-scale karst terrains to the planetary-scale plate tectonics. On one hand, the moving solid alters the surrounding flow field, but on the other hand, the flow modifies the motion and shape of the solid. This leads to a two-way coupling that is significant in the study of fluid-structure interactions and in the understanding of geomorphologies. In this work, we investigate the coupling between a floating plate and the convective fluid below it. Through numerical experiments, we show the motion of this plate is driven by the flow beneath. However the flow structure is also modified by the presence of this plate, leading to the "thermal blanket" effect where the trapped heat beneath the plate results in buoyant and upwelling flows that in turn push the plate away. By analyzing this two-way coupling between moving boundary and fluid, we are able to capture the dynamical behaviors of this plate through a low-dimensional stochastic model. Geophysically, the thermal blanket effect is believed to drive the continental drift, therefore understanding this mechanism has significance beyond fluid dynamics.

Introduction

The interior of Earth has fascinated generations of scientists [19]. Among them, Leonardo da Vinci (1452-1519) was one of the pioneers who noticed the incessant geological movements of our planet, as he observed the presence of marine fossils in the mountains. We now know the continents of Earth do not stay in place and instead undergo tectonic motions, and thermal convection in Earth’s mantle is believed to be the driving force of these motions [8].

Thermal convection occurs when uneven temperatures of fluid lead to uneven density and buoyancy, so warm fluid rises while cold fluid sinks. The definition of fluids here can be very broad, as modern geologists confirm that even the mantle flows like fluids at a large time scale [22]. The Prandtl number (Pr\mathrm{Pr}) there, defined as the ratio between the mantle’s kinematic viscosity and thermal diffusivity, is estimated to be around 102310^{23} [16].

The core of Earth is much warmer than its surface, and the destabilizing buoyancy is strong enough to drive mantle convection. As a measure of relative strength between buoyancy and viscous effects, the Rayleigh number (Ra\mathrm{Ra}) is around 10610^{6} in the mantle [21]. In the well-studied case of Rayleigh-Bénard convection, such a high Ra is known to lead to turbulent fluid motions [1]. With the mantle convecting like a fluid, its surface flow transports the continental plates resulting in their tectonic motions.

Due to the large spatial scale of Earth and the long time scale of mantle convection, the geophysical study of plate tectonics focuses on the current state of continents as well as predicting its important consequences like earthquakes [19]. On the other hand, numerical simulations [5, 24, 26, 15, 14, 25] and lab-scale experiments (Elder 1967; Zhang & Libchaber 2000; Zhong & Zhang 2005, 2007a,b; Whitehead et al. 2011) have proven to be an effective means of understanding the dynamics of plate tectonics.

Early experiments of Elder [2] showcase how one can recover the tectonic motion in the lab, where a paraffin fluid layer was used to model the mantle while a thin sheet of plastic floating on top served as a model continental plate. When the paraffin is heated from below, convection occurs and the plate moves due to shearing of the convective flow beneath. Such a simple experimental setup also displays nontrivial dynamics, as Zhang & Libchaber [27] observed the plate moves periodically between the two bounding walls of the fluid surface. Through more detailed investigations [28, 29 b], the size of floating plate is shown to strongly affect the plate motion, where small plates display periodic motion while large plates stay trapped in the middle of the fluid surface. Placing a moving heat source on top of a thermally convecting fluid also yields plate motions, and this is experimentally investigated by Howard et al. [5]; Whitehead [24]. Although the geometry, physical parameters, and time-scales presented in these works are very different from the mantle convection, they reveal surprising dynamics and most importantly, provide invaluable insight into the fluid-structure interaction mechanism behind continental drift.

The numerical exploration of plate tectonics has developed rapidly in the past several decades. Gurnis [3] provided the first time-dependent numerical simulations of continental drift, where multiple continents were allowed to merge and diverge. As in this work, many other numerical and theoretical endeavors (Zhong & Gurnis 1993; Lowman & Jarvis 1993, 1995; Lowman & Gable 1999; Lowman & Jarvis 1999; Lowman et al. 2001; Zhong et al. 2000) employ geophysical parameters of the mantle and enable rapid advancement of our understanding of the interior of Earth. It has also become clear that the two-way coupling between the continental plates and the mantle convection results in diverse dynamics [3, 31, 18, 26]. Most notably, large plates are observed to have more consistent motions, while small plates tend to move sporadically [3, 26]. These observations were recently examined by Mao et al. [15]; Mao [14] through resolved numerical simulations.

The aforementioned works confirm that continental plates are not only passive to the mantle flow advection beneath, but are also affecting the flow structure through the thermal blanket effect: The continental crust is known to have a much lower heat flux compared to the oceanic crust due to its large crust depth [14], so the continental plates essentially serve as a blanket that prevents heat from escaping and warms up the mantle beneath. The warm and light mantle tends to rise, forming an upward convective flow. As this flow moves towards the surface of Earth it diverges and creates a fluid forcing beneath the continental plate, transporting the plate away. This is the current understanding of continental drift, and the thermal blanket effect has been verified both numerically and experimentally [3, 28, 29 b].

This manuscript aims to provide a new angle for modeling plate tectonics through a low-dimensional model. After conducting direct numerical simulations (DNS) of the plate-flow interaction in a 2-dimensional periodic domain, we propose a stochastic model of the plate motion and show how the moving plate mechanically and thermally couples to a convecting fluid flow beneath it. Only retaining the most basic physics of thermal convection, this model recovers the dynamics observed in the full DNS and captures the transition of plate dynamics seen in Gurnis [3]; Whitehead & Behn [26]; Mao et al. [15]; Mao [14].

In what follows, we summarize the equations and numerical methods in § 2, and present the numerical results in § 3. The stochastic model will be systematically derived in § 4, and its application to the convection domain with various aspect ratios will be discussed in § 5. Finally, we summarize and discuss our results in § 6.

Numerical model

Flow equations

The configuration of our numerical simulation is shown in Figure 1, where a solid plate centered at location x=xpx = x_p floats on top of a convecting fluid that is bounded in the yy direction and periodic in the xx direction. Throughout this study, all lengths are rescaled by the fluid depth HH, time is rescaled by the diffusion time H2/κH^2/\kappa (κ\kappa is thermal diffusivity), and temperature is rescaled by the temperature difference ΔT\Delta T between the bottom and top free surface. The xx direction of the fluid domain is periodic with period Γ=D/H\Gamma= D/H (DD is the domain width), so the overall computational domain is x∈(0,Γ)x \in(0,\Gamma) and y∈(0,1)y \in(0,1) as shown in Figure 1. With the Boussinesq approximation, the resulting PDEs for flow speed u=(u,v)\mathbf{u} = (u,v), pressure pp, and temperature θ∈[0,1]\theta\in[0,1] are

Figure 1. Schematics of the moving plate and convecting fluid. The fluid domain is bounded between y∈(0,1)y \in(0,1) and periodic in xx, and the floating plate of width dd has center location xpx_p.

DuDt=−∇p+Pr ∇2u+Ra Pr θ ey(1)\frac{D\mathbf{u}}{Dt} = -\nabla p + \mathrm{Pr}\,\nabla^2\mathbf{u} + \mathrm{Ra}\,\mathrm{Pr}\,\theta\,\mathbf{e}_y \tag*{(1)}
∇⋅u=0(2)\nabla\cdot\mathbf{u} = 0 \tag*{(2)}
DθDt=∇2θ(3)\frac{D\theta}{Dt} = \nabla^2\theta \tag*{(3)}

Here, the Rayleigh number is Ra⁡=αgΔTH3/νκ\operatorname{Ra} = \alpha g \Delta T H^{3}/\nu\kappa and the Prandtl number is Pr⁡=ν/κ\operatorname{Pr} = \nu/\kappa, where ν\nu, α\alpha and gg are the kinematic viscosity, the thermal expansion coefficient of the fluid, and the acceleration due to gravity. Simple modifications to the flow solver can be adapted for the geophysical mantle convection, but as we wish to consider a more general case of fluid-structure interactions and to apply our theory to future laboratory experiments, we preserve the inertia of both the fluid and the solid plate in this study.

Boundary conditions

Without the presence of a plate, the boundary conditions are straightforward: the flow velocity u=(u,v)\mathbf{u} = (u,v) is no-slip at the fluid/solid boundary and shear-free at the air/fluid interface; The temperature θ\theta is 1 at the bottom and 0 at the air/fluid boundary. This yields the boundary conditions for the bottom surface y=0y = 0,

θ=1,u=v=0at y=0.(4)\theta= 1,\quad u = v = 0\quad\text{at }y = 0. \tag*{(4)}

At the top surface, the plate is effectively shielding the heat from escaping, we thus take θn=0\theta_n = 0 there while set the flow to be no-slip with respect to the moving plate. The resulting boundary conditions are

θ=0,uy=v=0for y=1 and x∉P,(5)\theta= 0,\quad u_y = v = 0\quad\text{for }y = 1\text{ and }x \notin P, \tag*{(5)}
θy=0,u=up, v=0for y=1 and x∈P.(6)\theta_y = 0,\quad u = u_p,\ v = 0\quad\text{for }y = 1\text{ and }x \in P. \tag*{(6)}

Alternatively, these conditions can be enforced as

{(1−1P)θ+1Pθy=0,1Pu+(1−1P)uy=upat y=1,v=0(7)\begin{cases} (1-\mathbb{1}_P)\theta+\mathbb{1}_P\theta_y=0,\\ \mathbb{1}_P u+(1-\mathbb{1}_P)u_y=u_p & \text{at }y=1,\\ v=0 \tag*{(7)} \end{cases}

where 1P\mathbb{1}_P is an indicator function that takes the value of 1 under the plate and 0 otherwise.

Plate dynamics

The fluid shear force directly drives the plate motion, so

mu˙p=−Pr⁡∫P∂u∂y(x,1,t) dx.(8)m\dot{u}_p=-\operatorname{Pr}\int_P\frac{\partial u}{\partial y}(x,1,t)\,dx. \tag*{(8)}

Here up=x˙pu_p=\dot{x}_p is the plate velocity, m=ρdm=\rho d is the dimensionless mass of the plate with linear density ρ\rho and width dd, and the integration area P={x∣x∈(xp−d/2,xp+d/2)}P=\{x\mid x\in(x_p-d/2,x_p+d/2)\} is the region under the plate.

Parameters and numerical method

The numerical method solving Eqs. (1) to (3) with Eqs. (4), (7) and (8) is detailed in Appendix § 7, where we use a Fourier-Chebyshev spectral method to obtain resolved and accurate numerical solutions. In all simulations, we choose Ra⁡=106\operatorname{Ra}=10^{6}, Pr⁡=7.9\operatorname{Pr}=7.9, Γ=1−16\Gamma=1-16, and m=4dm=4d (dd is the plate width), matching the parameters of water convection in experiments [27, 28].

Numerical results

Several trajectories of plates with various sizes are shown in Figure 2(a), and we can immediately see how the plate’s size affects its motion. We define the covering ratio Cr=d/ΓCr=d/\Gamma to measure how much of the free surface is covered by the plate of width dd, where the fluid aspect ratio is Γ=4\Gamma=4 for all the results in this section. For small plates, their net displacement is small, which can be better seen from the total displacement dp(t)=∫0t∣up(t′)∣ dt′d_p(t)=\int_{0}^{t}|\boldsymbol{u}_p(t')|\,dt' shown in Figure 2(b). Increasing the plate size, linear motion appears as CrCr becomes greater than 0.33, as seen in the green trajectories in Figure 2(a). These trajectories are subject to reversals, as there is an effective noise from the turbulent fluid forcing. As CrCr further increases, the linear motion becomes more persistent, as the reversals of plate motion become rare when Cr→1Cr\to1 in Figure 2(a). We note that similar dynamical behaviors have been seen in geophysical Stokes flow simulations [14], therefore the coupling mechanism between the moving plate and flow beneath must be similar for different flow regimes.

Figure 2. Motion of a plate floating on top of a convective fluid. (a) Trajectories of plates of various sizes. The covering ratio is Cr=d/ΓCr=d/\Gamma and the time t0t_0 marks the beginning of plate motion. (b) The total displacement of the plates, where dpd_p is defined through d˙p=∣x˙p∣\dot{d}_p=|\dot{x}_p|. (c) Average plate speed vp=⟨∣x˙p∣⟩v_p=\langle|\dot{x}_p|\rangle becomes high when Cr>0.33Cr>0.33 and reaches a maximum at Cr=0.56Cr=0.56. In these simulations, Γ=4\Gamma=4, ρ=4\rho=4, Ra=106\mathrm{Ra}=10^6, and Pr=7.9\mathrm{Pr}=7.9.

From the total displacement dpd_p, one can see that a maximum plate speed is achieved at around Cr≈0.5Cr\approx0.5, and this can be confirmed by plotting the time-averaged plate speed vp=⟨∣x˙p∣⟩v_p=\langle|\dot{x}_p|\rangle in Figure 2(c). The average velocity vpv_p remains low for small plates, but increases significantly for Cr>0.33Cr>0.33 and reaches a maximum around Cr=0.5Cr=0.5.

To investigate the transition between dynamical states, the typical flow and temperature distributions in the fluid are shown in Figure 3. In Figure 3(a), a small plate with Cr=0.125Cr=0.125 is placed on the convecting fluid and it is attracted by the center of downwelling fluid at xmx_m [Figure 1], where the surface flow forms a sink. This sink is a stable equilibrium for the plate, as any deviations from this sink will result in a restoring fluid force acting on the plate. The structure of this flow sink can be further seen in Figure 3(b), where both the yy-averaged temperature θˉ=∫01θ dy\bar{\theta}=\int_{0}^{1}\theta\,dy and the yy-averaged vertical flow velocity vˉ=∫01v dy\bar{v}=\int_{0}^{1}v\,dy reach their minima.

Dynamics of floating plate with $Cr=0.125$ and $Cr=0.5$

Figure 3. Dynamics of floating plate with Cr=0.125Cr=0.125 [(a)–(d)] and Cr=0.5Cr=0.5 [(e)–(h)]. (a) A typical snapshot of the flow and temperature distributions under the plate with Cr=0.125Cr=0.125. (b) The yy-averaged temperature and vertical flow velocity corresponding to (a), the shaded area indicates the position of the plate. (c) The plate displacement is a random process. (d) Plate velocity is a random variable with mean 0, whose PDF is a Gaussian distribution as indicated by the histogram. (e) At Cr=0.5Cr=0.5, typical flow and temperature distributions showing the plate is transported by the surface flow. (f) The yy-averaged temperature and vertical flow velocity corresponding to (e). (g) The plate displacement is linear in time, indicating a unidirectional translation. (h) Plate velocity has nonzero mean. Supplementary movies of these simulations are included.

Following this surface flow pattern, the plate displacement xpx_p is stochastic as shown in Figure 3(c). Due to the random forcing from turbulent flows, the plate location is subject to noise that can be seen affecting the plate velocity upu_p in Figure 3(d), whose histogram shows a Gaussian distribution. It is rare but not impossible for the plate to experience a strong “wind” from the flow, which can push the plate away from the flow sink, across the flow source, and back to the sink again, resulting in the jumps in xpx_p seen in Figure 3(c).

Figure 3(e) shows the dynamics of a plate with Cr=0.5Cr=0.5. In this case, the plate motion is unidirectional as shown in Figure 3(g)–(h), with velocity upu_p that has a nonzero mean. Shown in Figure 3(e)–(f), the moving plate tends to situate between the flow sink and source. As the surface flow pushes the plate towards its sink, the distribution of flow temperature also shifts, leading to a moving plate chasing a moving surface flow sink.

This is a direct consequence of the thermal blanket effect: When the plate is large enough, the temperature increases beneath it as heat cannot escape there. This local warming modifies the flow temperature and effectively pushes the cold, downwelling fluid away, resulting in a shift of the flow sink location. Overall, the plate moves towards the cold flow sink while simultaneously pushing the sink away. Thus a simple dynamics exists for this seemingly complicated fluid-structure interaction problem, and we will derive a model from these observations.

Stochastic model

As seen in Figure 3, the variation of the yy-averaged temperature θˉ\bar{\theta} strongly affects the flow pattern. To capture the variational and periodic nature of θˉ\bar{\theta}, we approximate it with its lowest nontrivial Fourier mode,

θˉ(x,t)=α−αcos⁡r[x−xm(t)],(9)\bar{\theta}(x,t)=\alpha-\alpha\cos r[x-x_m(t)], \tag*{(9)}

where xmx_m is the location of surface flow sink in Figure 1, r=2πΓ−1r=2\pi\Gamma^{-1} is the wavenumber, and the constant α\alpha measures the strength of temperature variation.

Induced by this temperature distribution, the surface flow velocity U(x,t)=u(x,1,t)U(x,t)=u(x,1,t) can be approximated as

U(x,t)=−βsin⁡r[x−xm(t)],(10)U(x,t)=-\beta\sin r[x-x_m(t)], \tag*{(10)}

where β>0\beta>0 is the surface flow strength. Indeed, this surface flow profile has a sink at x=xmx=x_m: Small deviations from xmx_m results in U>0U>0 for x<xmx<x_m and U<0U<0 for x>xmx>x_m, so the flow locally points towards x=xmx=x_m.

We note that this surface flow profile does not match the plate velocity at the solid/fluid boundary, and the mismatch between UU and upu_p allows us to estimate ∂u/∂y(x,1,t)\partial u/\partial y(x,1,t) and the resulting shear stress, leading to the plate acceleration

u˙p(t)=−Prm∫Pup(t)−U(x,t)δ dx+σW˙(t).(11)\dot{u}_p(t)=-\frac{\mathrm{Pr}}{m}\int_P\frac{u_p(t)-U(x,t)}{\delta}\,dx+\sigma\dot{W}(t). \tag*{(11)}

Here, δ\delta is the momentum boundary layer thickness [20] that is determined by Ra\mathrm{Ra} and Pr\mathrm{Pr}, so we assume it to be constant in our study. We also include a white noise with standard deviation σ\sigma, representing the turbulent fluid forcing. Overall, Eqs. (10) and (11) represent a flow profile that has both the large-scale circulations and the small-scale turbulent flows, consistent with observations of high-Ra\mathrm{Ra} thermal convection [1, 6].

To model the moving plate as a thermal blanket, we look at the yy-averaged heat equation,

∂θˉ∂t=∂2θˉ∂x2+q(x,t).(12)\frac{\partial\bar{\theta}}{\partial t}=\frac{\partial^2\bar{\theta}}{\partial x^2}+q(x,t). \tag*{(12)}

Here we have ignored the flow advection, and q(x,t)=∂θ∂y(x,1,t)−∂θ∂y(x,0,t)q(x,t)=\frac{\partial\theta}{\partial y}(x,1,t)-\frac{\partial\theta}{\partial y}(x,0,t) is the vertical heat flux passing through location xx. Assuming the heat leaving the fluid-air interface obeys Newton’s law of cooling and no heat penetrates the plate, we can rewrite the heat equation as

∂θˉ∂t=∂2θˉ∂x2−γθˉ(1−1P).(13)\frac{\partial\bar{\theta}}{\partial t}=\frac{\partial^2\bar{\theta}}{\partial x^2}-\gamma\bar{\theta}(1-\mathbb{1}_P). \tag*{(13)}

The indicator function 1P(x)\mathbb{1}_P(x) is 1 when x∈Px\in P and 0 otherwise, and the constant γ\gamma models the rate of cooling. We now plug in the value of θˉ\bar{\theta} from Eq. (9) and integrate over xx, which leads to an ODE for xmx_m,

x˙m(t)=γπ∫P[1−cos⁡r(xm−x)]sin⁡r(xm−x) dx.(14)\dot{x}_m(t)=\frac{\gamma}{\pi}\int_P[1-\cos r(x_m-x)]\sin r(x_m-x)\,dx. \tag*{(14)}

The integrals in Eqs. (11) and (14) can be evaluated exactly. Defining a phase angle ϕ=r(xp−xm)\phi=r(x_p-x_m), we arrive at a closed dynamical system for (up,ϕ)(u_p,\phi),

u˙p=−βλπCrsin⁡(πCr)sin⁡ϕ−λup+σW˙,(15)\dot{u}_p=-\frac{\beta\lambda}{\pi Cr}\sin(\pi Cr)\sin\phi-\lambda u_p+\sigma\dot{W}, \tag*{(15)}
ϕ˙=rup+2γπsin⁡(πCr)sin⁡ϕ−γ2πsin⁡(2πCr)sin⁡(2ϕ),(16)\dot{\phi}=ru_p+\frac{2\gamma}{\pi}\sin(\pi Cr)\sin\phi-\frac{\gamma}{2\pi}\sin(2\pi Cr)\sin(2\phi), \tag*{(16)}

where λ=Pr/(ρδ)\lambda=\mathrm{Pr}/(\rho\delta). Once the dynamics of (up,ϕ)(u_p,\phi) is known, the dynamics of (xp,xm)(x_p,x_m) can be calculated through x˙p=up\dot{x}_p=u_p and xm=xp−r−1ϕx_m=x_p-r^{-1}\phi.

There are four parameters in this model: β\beta as the strength of surface flow, λ=Pr/(ρδ)\lambda=\mathrm{Pr}/(\rho\delta) as a damping coefficient, γ\gamma as the rate of surface cooling, and σ\sigma as the random fluid forcing. Physically, the surface cooling rate γ\gamma is affected by the surface flow strength β\beta, so we take γ=rβ\gamma=r\beta in this model which results in the correct dynamics. The remaining parameters can be estimated from the numerical simulations, and their values and estimation procedures are included in Appendix B.

Starting with ϕ(0)=0\phi(0)=0 and a random value of up(0)u_p(0), Figure 4(a)-(Figure 4c) show some typical trajectories of xp(t)x_p(t) at different CrCr. In Figure 4(a), the trajectory of the small plate (Cr=0.2Cr=0.2) is noise-driven. For the medium plate of Cr=0.5Cr=0.5, Figure 4(b) shows its trajectory is composed of linear translations with reversals. For the large plate of Cr=0.8Cr=0.8, Figure 4(c) indicates that the translation is unidirectional. The plate speed in Figure 4(c) is comparable to the full DNS result in Figure 2, and this speed decreases as CrCr further increases. The typical displacement xpx_p for plates with various sizes is shown in Figure 4(d), which resembles Figure 2(a) and has a transition between the noise-driven and linear motions at Cr≈0.3Cr \approx0.3. Thus, this simple model captures all the key features of the full numerical simulation.

Figure 4. Simulated trajectories from the stochastic model. (a) The dynamics of the small plate is a random walk. (b) The medium-sized plate has a nonzero translational velocity whose direction is subject to reversals. (c) The translational motion of the large plate is more persistent. (d) Trajectories of all simulated paths, whose dynamics recover Figure 2(a).

Without noise, the critical behavior of the dynamical system Eqs. (15) and (16) can be further analyzed. For small CrCr, it is easy to see that Eqs. (15) and (16) have up=0u_p = 0, ϕ=2πN\phi= 2\pi N (where NN is an integer) as equilibria, which are stable and reflect the passive state of plate motion. Increasing CrCr, new equilibria appear at Cr∗=1/3Cr^* = 1/3. For Cr>Cr∗Cr > Cr^*, it becomes possible to have a nonzero plate velocity up∗=(β/π)u^p∗u_p^* = (\beta/\pi)\hat{u}_p^*, where

u^p∗=−sin⁡πCrCrsin⁡ϕ∗(17)\hat{u}_p^* = -\frac{\sin\pi Cr}{Cr}\sin\phi^* \tag*{(17)}

and the equilibrium ϕ∗\phi^* can be determined from

cos⁡ϕ∗=1−(2Cr)−1cos⁡πCr.(18)\cos\phi^* = \frac{1-(2Cr)^{-1}}{\cos\pi Cr}. \tag*{(18)}

These states represent the translation of the plate. We note that Eqs. (17) and (18) are only functions of CrCr, and thereby independent of all other parameters assumed in this model. To recover the dimensional plate velocity up∗u_p^*, one only needs to know the flow speed factor β/π\beta/\pi.

The possible values of u^p∗\hat{u}_p^* and ϕ∗\phi^* are shown in Figure 5. For Cr<Cr∗=1/3Cr < Cr^* = 1/3, up∗=0u_p^* = 0 and ϕ∗=2πN\phi^* = 2\pi N are the only possible equilibria which reflect the passive nature of the small plate that is always attracted by the surface flow sink. For Cr>Cr∗Cr > Cr^*, new phases appear as ϕ∗=(2N+1)π±arccos⁡([(2Cr)−1−1](cos⁡πCr)−1)\phi^* = (2N+1)\pi\pm\arccos\left([(2Cr)^{-1}-1](\cos\pi Cr)^{-1}\right), which are solutions of Eq. (18) and become stable for large CrCr. They indicate that the larger plate tends to sit between the surface flow sink (ϕ=2Nπ\phi= 2N\pi) and source [ϕ=(2N+1)π\phi= (2N + 1)\pi], confirming our observations in Figure 3. As the surface flow points from its source to its sink [see arrows in Figure 5(b)], these new phases indicate two possible plate velocities that are given by Eq. (17) and shown in Figure 5(a). Furthermore, Figure 5(a) resembles Figure 2(c) as the plate velocity vanishes for Cr→Cr∗Cr \to Cr^{*} and Cr→1Cr \to1, and obtains its maximum around Cr=0.5Cr = 0.5.

Equilibrium values of the plate velocity and phase angle

Figure 5. Equilibrium values of the plate velocity and phase angle. (a) Dimensionless plate velocity u^p\hat{u}_p is 0 for small CrCr, but becomes translational as CrCr increases. (b) The phase angle ϕ\phi. Blue lines represent the passive state where the plate is attracted by the flow sink, and red/orange curves show the equilibrium phase of the translational state. The surface flow direction is labelled with arrows.

Through this simple model, we see clear physics of how the solid plate interacts with the fluid beneath, and the covering ratio CrCr serves as a measure of the strength of the thermal blanket effect. For small CrCr, the thermal blanket effect is weak thus the plate motion is passive to the fluid. Increasing CrCr beyond Cr∗Cr^{*}, the thermal blanket effect is strong enough to alter the flow and temperature distributions, generating an upwelling that can lead to plate motion. Both Figure 2(c) and Figure 5(a) suggest that the plate speed peaks at Cr≈0.5Cr \approx0.5 and vanishes as Cr→1Cr \to1, which also reflects the competition between the thermal blanket effect and the flow convection. As CrCr increases, more area of the free surface is covered by the plate and the fluid force beneath is averaged in a larger domain. As this domain may cover both the upwelling and downwelling flows, the total fluid force is affected by CrCr. This can be seen in Eq. (11), where the surface flow velocity UU provides plate acceleration. In Eq. (10), UU is modelled as sinusoidal and this profile is integrated in Eq. (11), therefore increasing the integration area in Eq. (11) to half of the open surface (Cr=1/2Cr = 1/2) will cover the highest contribution of UU. Further increasing the covering area will thereby decrease the contribution of UU, as the integration domain is more than half a period of the sine function. In an extreme case, Cr=1Cr = 1 indicates an integration of UU for a full period, leading to zero fluid force as seen in Figure 5(a).

Aspect ratio effect

All the results discussed so far focus on the convective fluid domain of aspect ratio Γ=4\Gamma= 4, which matches the geometry of many experimental investigations. Varying the domain aspect ratio will certainly affect the dynamics of the convecting fluid and moving plate, as a more complicated, multi-roll flow structure emerges [1]. In this section, we investigate the plate dynamics at Γ=2\Gamma= 2, 44, 88, and 1616 while keeping other dynamical parameters the same as described in § 2 to § 4.

Figure 6 shows some typical temperature distributions of DNS results at Γ=2−16\Gamma= 2 - 16. Clearly, a more complicated flow structure emerges as more convection cells appear with increasing Γ\Gamma. Figure 7 shows trajectories of the plate at different Γ\Gamma; a common feature reemerges as the small plate moves little while the large plate translates.

Figure 6. Snapshots of the convective fluid at various aspect ratio Γ\Gamma. (a) Γ=2\Gamma= 2. (b) Γ=8\Gamma= 8. (c) Γ=16\Gamma= 16. In these simulations, all other parameters are fixed at Cr=0.2Cr = 0.2, ρ=4\rho= 4, Ra=106Ra = 10^{6}, and Pr=7.9Pr = 7.9.

Figure 7. Simulated trajectories from the DNS at the aspect ratio of (a) Γ=2\Gamma= 2, (b) Γ=8\Gamma= 8, and (c) Γ=16\Gamma= 16. In these simulations, ρ=4\rho= 4, Ra=106\mathrm{Ra} = 10^{6}, and Pr=7.9\mathrm{Pr} = 7.9, as in Figure 2 and Figure 3. (d) The critical Cr∗Cr^{*} separating the passive and translating states has a −2/3-2/3 power-law scaling with Γ\Gamma.

To extend our simple model to cases of high aspect ratio, we consider a bulk temperature and surface flow profile with a more complicated spatial dependence,

θˉ(x,t)=α−αcos⁡kr[x−xm(t)],(19)\bar{\theta}(x,t) = \alpha- \alpha\cos kr[x-x_m(t)], \tag*{(19)}
U(x,t)=−βsin⁡kr[x−xm(t)],(20)U(x,t) = -\beta\sin kr[x-x_m(t)], \tag*{(20)}

where the integer kk is the most dominant wave number in the Fourier spectrum. The inclusion of wave number kk is inspired by the fact that the convection might have a multi-roll structure, so the temperature and flow profiles above indicate that there are 2k2k convection rolls in the fluid with kk surface sinks and kk surface sources. We track one of the surface sinks xm(t)x_m(t), which is also the location of lowest fluid temperature.

Through similar derivations as in § 4, we can again obtain a stochastic dynamical system for the phase ϕ=r(xp−xm)\phi= r(x_p-x_m) and the plate velocity upu_p,

u˙p=−βλkπCrsin⁡(kπCr)sin⁡(kϕ)−λup+σW˙,(21)\dot{u}_p = -\frac{\beta\lambda}{k\pi Cr}\sin(k\pi Cr)\sin(k\phi)-\lambda u_p+\sigma\dot{W}, \tag*{(21)}
ϕ˙=rup+2γk2πsin⁡(kπCr)sin⁡(kϕ)−γ2k2πsin⁡(2kπCr)sin⁡(2kϕ).(22)\dot{\phi} = ru_p+\frac{2\gamma}{k^2\pi}\sin(k\pi Cr)\sin(k\phi)-\frac{\gamma}{2k^2\pi}\sin(2k\pi Cr)\sin(2k\phi). \tag*{(22)}

Without noise, it is easy to verify that Eqs. (21) and (22) have passive equilibria up∗=0u_p^{*}=0 and ϕ∗=2mπ/k\phi^{*}=2m\pi/k, where mm is any integer. The Jacobian of such passive states is

J=[−λ−βλ(kπCr)−1sin⁡(kπCr)r2γ(kπ)−1sin⁡(kπCr)(1−cos⁡(kπCr))].(23)J = \begin{bmatrix} -\lambda& -\beta\lambda(k\pi Cr)^{-1}\sin(k\pi Cr) \\ r & 2\gamma(k\pi)^{-1}\sin(k\pi Cr)(1-\cos(k\pi Cr)) \end{bmatrix}. \tag*{(23)}

Evaluating the trace and determinant of JJ in the limit of Cr→0Cr \to0 yields

tr⁡(J)→−λ<0,(24)\operatorname{tr}(J) \to-\lambda< 0, \tag*{(24)}
det⁡(J)→krβλ>0.(25)\det(J) \to kr\beta\lambda> 0. \tag*{(25)}

Therefore both eigenvalues of JJ are negative, indicating stable passive states for small CrCr and confirming our numerical observations.

The stability of passive states is lost when one or both eigenvalues of JJ become positive, therefore det⁡(J)=0\det(J)=0 provides an equation to determine the critical Cr∗Cr^{*}. After simplification, we have

Cr∗k(1−cos⁡kπCr∗)=βr2γ.(26)\frac{Cr^{*}}{k}(1-\cos k\pi Cr^{*})=\frac{\beta r}{2\gamma}. \tag*{(26)}

The root of Eq. (26) can be determined numerically, given that the parameters β\beta and γ\gamma are known. In laboratory experiments and geophysical plate tectonics, the surface flow speed scale β\beta and ventilation coefficient γ\gamma can be properly estimated and used to determine Cr∗Cr^{*}, therefore Eq. (26) offers the possibility of determining plate mobility through physical parameters.

We note that Cr∗Cr^{*} is usually small for large Γ\Gamma as shown in Figure 7, and the root of Eq. (26) in this limit can be shown as

Cr∗≈(βrπ2γk)1/3∼Γ−2/3.(27)Cr^{*} \approx\left(\frac{\beta r}{\pi^{2}\gamma k}\right)^{1/3} \sim\Gamma^{-2/3}. \tag*{(27)}

Here we have used the relations r∼Γ−1r \sim\Gamma^{-1} and k∼Γk \sim\Gamma, with the latter indicating the number of convection rolls is proportional to the aspect ratio Γ\Gamma. Equation (27) can indeed be verified– as shown in Figure 7(d), the critical Cr∗Cr^{*} measured from DNS data does follow the -2/3 power law with Γ\Gamma.

Much more can be investigated through the dynamics of Eqs. (21) and (22), and future experimental studies can certainly be used to further address the interaction between the plate and convective fluid below.

Discussion

In this work, we have numerically and theoretically explored the mechanical and thermal coupling between a moving plate and a convecting fluid beneath it. Inspired by present and past works [3, 26, 15, 14], we propose a stochastic model showing that the plate size (covering ratio) is a deciding factor for the strength of the thermal blanket effect. For small plates, the shielding effect is weak and the plate motion is passive to the flow structure; For large plates, the flow beneath becomes warm enough and an upwelling center is formed, pushing the plate away and resulting in its translation. The proposed simple model consists of minimal assumptions about the flow and its mechanical and thermal coupling to the plate, however it is capable of predicting the dynamical transition of the plate motion. Although the DNS is conducted at a parameter regime similar to laboratory experiments [28, 29 b], this stochastic model is Reynolds number independent, therefore allowing it to be applied to both laboratory and geophysical scenarios.

Laboratory-scale experiments are usually conducted in a bounded convection system, therefore the plate is limited to move between two walls. The flow structure and its coupling to the plate is very different in bounded convection, as previous works show the flow has two counter-rotating large-scale circulations, whose strengths are modulated by the location of the plate [28, 29 b]. The plate is also bounded by the solid boundaries, so it stops moving once touching the wall. Modeling the plate dynamics in this case is not trivial, and advanced tools such as stochastic variational inequalities [7] can serve as a mathematical means of analyzing such plate-boundary interactions. A periodic domain of convecting fluid is necessary to experimentally verify our stochastic model, and such an experiment would also resemble the mantle convection more closely. Future experiments consisting of an annular convection domain could provide more details to validate the stochastic model.

Although here we only investigate the dynamics of a single plate, our numerical method is capable of handling multiple plates providing that their interactions can be properly modeled. We are currently investigating such interactions, which has led to even more diverse and unpredictable dynamics. For example, if two small plates each have Cr<1/3Cr < 1/3 but their combined size reaches Cr>1/3Cr > 1/3, we have seen that each individual plate moves randomly but their combined “super continent” can translate. In the geophysical case of plate tectonics, plate interactions are the reason for many volcanic activities and mountain formations, therefore understanding the converging and diverging motion of nearby plates might offer new insights into the fluid mechanics behind these geophysical events.

For simplicity, the simulation and model in this work are both two dimensional, and extending our results to three dimensions is a current priority. Using the Chebyshev-Fourier-Fourier method, we have implemented the numerical solver for evolving a plate sitting on top of a three-dimensional domain that is periodic in two horizontal directions. Moreover, the simulation of plate tectonics on a spherical shell is also possible through the Chebyshev-Chebyshev-Fourier method, which is a configuration closer to the geophysical plate tectonics. Through analyzing the direct numerical simulation results there, we wish to further develop our stochastic model and use it to address the fluid-structure interactions happening inside Earth.

Finally, we note that the geophysical plate tectonics is much more complicated than any experiments or numerical simulations conducted so far, as the interior of Earth is such a complex environment and is still being explored by modern science. But although current simple models cannot fully capture the dynamics of continental drifts, they can hopefully still offer some fluid mechanical insights into the geophysics of Earth.

Numerical method

The Navier-Stokes Boussinesq Eqs. (1) to (3) in 2D can be written in the vorticity & stream function format,

DωDt=Pr⁡∇2ω+Pr⁡Ra⁡∂θ∂x,(28)\frac{D\omega}{Dt} = \operatorname{Pr}\nabla^2\omega+ \operatorname{Pr}\operatorname{Ra}\frac{\partial\theta}{\partial x}, \tag*{(28)}
−∇2ψ=ω,u=∇⊥ψ,(29)-\nabla^2\psi= \omega,\quad\mathbf{u} = \nabla_\perp\psi, \tag*{(29)}
DθDt=∇2θ,(30)\frac{D\theta}{Dt} = \nabla^2\theta, \tag*{(30)}

where the zz-component of vorticity ω=z^⋅∇×u\omega= \hat{\mathbf{z}}\cdot\nabla\times\mathbf{u} and the stream function defined by u=∇⊥ψ=(ψy,−ψx)\mathbf{u} = \nabla_\perp\psi= (\psi_y,-\psi_x) are solved for, alleviating the difficulty of solving for pressure pp.

In the vorticity & stream function format, the boundary conditions Eqs. (4) and (7) can be enforced as

θ=1, ψ=ψy=0at y=0.(31)\theta= 1,\ \psi= \psi_y = 0 \quad\text{at } y = 0. \tag*{(31)}

and

{(1−1P)θ+1Pθy=01Pψy+(1−1P)ψyy=uPat y=1.ψ=0(32)\begin{cases} (1-1_P)\theta+ 1_P\theta_y = 0\\ 1_P\psi_y + (1-1_P)\psi_{yy} = u_P \quad\text{at } y = 1.\\ \psi= 0 \tag*{(32)} \end{cases}

In the numerical simulations we soften the edge of indicator function 1P1_P, making it smoother in order to reduce numerical error.

The time derivatives in Eqs. (28) and (30) are approximated with the 2nd order Adam-Bashforth Backward Differentiation method (ABBD2). At time step tn=nΔTt_n = n\Delta T, we denote ωn(x,y)=ω(x,y,nΔT)\omega_n(x,y) = \omega(x,y,n\Delta T), ψn(x,y)=ψ(x,y,nΔT)\psi_n(x,y) = \psi(x,y,n\Delta T), and θn(x,y)=θ(x,y,nΔT)\theta_n(x,y) = \theta(x,y,n\Delta T), and Eqs. (28) to (30) become

∇2ωn−σ1ωn=fn,(33)\nabla^2\omega_n-\sigma_1\omega_n = f_n, \tag*{(33)}
∇2θn−σ2θn=hn,(34)\nabla^2\theta_n-\sigma_2\theta_n = h_n, \tag*{(34)}
−∇2ψn=ωn,(35)-\nabla^2\psi_n = \omega_n, \tag*{(35)}

where

σ1=32Pr⁡Δt,σ2=32Δt(36)\sigma_1 = \frac{3}{2\operatorname{Pr}\Delta t}, \qquad\sigma_2 = \frac{3}{2\Delta t} \tag*{(36)}
fn=Pr⁡−1[2(u⋅∇ω)n−1−(u⋅∇ω)n−2]−(2Pr⁡Δt)−1(4ωn−1−ωn−2)−Ra⁡(∂θ∂x)n,(37)f_n = \operatorname{Pr}^{-1}\left[2(\boldsymbol{u}\cdot\nabla\omega)_{n-1}-(\boldsymbol{u}\cdot\nabla\omega)_{n-2}\right] -\left(2\operatorname{Pr}\Delta t\right)^{-1}(4\omega_{n-1}-\omega_{n-2})-\operatorname{Ra}\left(\frac{\partial\theta}{\partial x}\right)_n, \tag*{(37)}
hn=[2(u⋅∇θ)n−1−(u⋅∇θ)n−2]−(2Δt)−1(4θn−1−θn−2).(38)h_n = \left[2(\boldsymbol{u}\cdot\nabla\theta)_{n-1}-(\boldsymbol{u}\cdot\nabla\theta)_{n-2}\right]-\left(2\Delta t\right)^{-1}(4\theta_{n-1}-\theta_{n-2}). \tag*{(38)}

Equations (33) to (35), together with the inhomogeneous Robin boundary conditions Eqs. (31) and (32), are Helmholtz equations that can be solved by standard spectral methods ([17]). More details of this numerical solver will be included in future publications.

Nonlinear terms like u⋅∇θ\boldsymbol{u}\cdot\nabla\theta and u⋅∇ω\boldsymbol{u}\cdot\nabla\omega in Eqs. (37) and (38) are computed pseudospectrally with a simple and efficient anti-aliasing filter ([4]). With given initial and boundary data, (34) can be solved first to obtain θn\theta_n, which is inserted in fnf_n so (33) can be solved next. Finally, (35) is solved with the known ωn\omega_n.

After solving for the flow and temperature fields, the plate acceleration can be determined as

ap,n=−Pr⁡m∫P∂2ψn∂y2(x,1) dx.(39)a_{p,n}=-\frac{\operatorname{Pr}}{m}\int_P\frac{\partial^2\psi_n}{\partial y^2}(x,1)\,\mathrm{d}x. \tag*{(39)}

The plate velocity up,nu_{p,n} and plate location xp,nx_{p,n} can then be computed through a 2nd order Adam-Bashforth method,

xp,n=xp,n−1+Δt2(3up,n−1−up,n−2),(40)x_{p,n}=x_{p,n-1}+\frac{\Delta t}{2}(3u_{p,n-1}-u_{p,n-2}), \tag*{(40)}
up,n=up,n−1+Δt2(3ap,n−1−ap,n−2).(41)u_{p,n}=u_{p,n-1}+\frac{\Delta t}{2}(3a_{p,n-1}-a_{p,n-2}). \tag*{(41)}

At Γ=4\Gamma=4, there are typically 256 Fourier modes in the xx direction and 64 Chebyshev nodes in the yy direction, with a time step size of Δt=10−6\Delta t=10^{-6}. These parameters are tested to yield resolved and accurate numerical solutions.

Model parameters

The four parameters involved in the stochastic model can be estimated from the DNS data and auxiliary numerical tests. The detailed procedures are listed below.

  1. β≈400\beta\approx400 is directly measured from the numerical simulation in Figure 3.

  1. λ≈200\lambda\approx200 is estimated from Pr⁡=7.9\operatorname{Pr}=7.9, ρ=4\rho=4, and δ=0.01\delta=0.01. The boundary layer thickness δ\delta is estimated from the relation δ∼(2Nu⁡)−1\delta\sim(2\operatorname{Nu})^{-1}, where the Nusselt number Nu⁡\operatorname{Nu} is at the order of 10110^1 as measured from the simulation.

  1. σ≈200\sigma\approx200 is estimated from the variance of the plate center xpx_p for small CrCr. From Eq. (15), we have σ2≈2λVar⁡(xp)\sigma^2\approx2\lambda\operatorname{Var}(x_p), where Var⁡(xp)≈100\operatorname{Var}(x_p)\approx100 is measured from the numerical simulation shown in Figure 3.

We note that β\beta and σ\sigma model the surface flow, therefore can be estimated from numerical simulations without floating plate or calculated through scaling relations ([1]).

Supplementary Materials

Supplementary movies are available at https://math.nyu.edu/~jinzi/research/convectivePlate-1p/Movie.

References

  1. [1]AHLERS, G., GROSSMANN, S. & LOHSE, D. 2009 Heat transfer and large scale dynamics in turbulent Rayleigh-Bénard convection. Rev. Mod. Phys. 81 (2), 503.DOI
  2. [2]ELDER, J. 1967 Convective self-propulsion of continents. Nature 214 (5089), 657–660.DOI
  3. [3]GURNIS, M. 1988 Large-scale mantle convection and the aggregation and dispersal of supercontinents. Nature 332 (6166), 695–699.DOI
  4. [4]HOU, T. Y. & LI, R. 2007 Computing nearly singular solutions using pseudo-spectral methods. Journal of Computational Physics 226 (1), 379–397.DOI
  5. [5]HOWARD, L., MALKUS, W. & WHITEHEAD, J. 1970 Self-convection of floating heat sources: A model for continental drift. Geophysical and Astrophysical Fluid Dynamics 1 (1-2), 123–142.DOI
  6. [6]HUANG, J. M. & ZHANG, J. 2022 Rayleigh–Bénard thermal convection perturbed by a horizontal heat flux. J. Fluid Mech. 954.arxiv.org/abs/2112.06768
  7. [7]HUANG, J. M., ZHONG, J.-Q., ZHANG, J. & MERTZ, L. 2018 Stochastic dynamics of fluid–structure interaction in turbulent thermal convection. J. Fluid Mech. 854.DOI
  8. [8]KIOUS, W. J. & TILLING, R. I. 1996 This dynamic Earth: the story of plate tectonics. DIANE Publishing.
  9. [14]MAO, Y. 2021 An insulating plate drifting over a thermally convecting fluid: the effect of plate size on plate motion, coupling modes and flow structure. Journal of Fluid Mechanics 916, A18.DOI
  10. [15]MAO, Y., ZHONG, J.-Q. & ZHANG, J. 2019 The dynamics of an insulating plate over a thermally convecting fluid and its implication for continent movement over convective mantle. Journal of Fluid Mechanics 868, 286–315.DOI
  11. [16]MEYERS, R. A. et al. 1987 Encyclopedia of physical science and technology. Academic Press.
  12. [17]PEYRET, R. 2002 Spectral methods for incompressible viscous flow, , vol. 148. Springer Science & Business Media.DOI
  13. [18]PHILLIPS, B. R. & BUNGE, H.-P. 2005 Heterogeneity and time dependence in 3d spherical mantle convection models with continental drift. Earth and Planetary Science Letters 233 (1-2), 121–135.DOI
  14. [19]PLUMMER, C. C., MCGEARY, D., CARLSON, D. H. et al. 2001 Physical geology. McGraw-Hill Boston.
  15. [20]SCHLICHTING, H. & GERSTEN, K. 2016 Boundary-layer theory. springer.
  16. [21]SELLEY, R. C., COCKS, L. R. M. & PLIMER, I. R. 2005 Encyclopedia of geology. Elsevier Academic.
  17. [22]TURCOTTE, D. L. & SCHUBERT, G. 2002 Geodynamics. Cambridge University Press.
  18. [24]Whitehead, J. A. 1972 Moving heaters as a model of continental drift. Phys. Earth Planet. In. 5, 199–212.DOI
  19. [25]Whitehead, J. A. 2023 Convection cells with accumulating crust: Models of continent and mantle evolution. Journal of Geophysical Research: Solid Earth 128 (4), e2022JB025643.DOI
  20. [26]Whitehead, J. A. & Behn, M. D. 2015 The continental drift convection cell. Geophys. Res. Lett. 42 (11), 4301–4308.
  21. [27]Zhang, J. & Libchaber, A. 2000 Periodic boundary motion in thermal turbulence. Physical Review Letters 84 (19), 4361.pubmed.ncbi.nlm.nih.gov/10990686
  22. [28]Zhong, J.-Q. & Zhang, J. 2005 Thermal convection with a freely moving top boundary. Phys. Fluids 17 (11), 115105.DOI
  23. [29]Zhong, J.-Q. & Zhang, J. 2007a Dynamical states of a mobile heat blanket on a thermally convecting fluid. Physical Review E 75 (5), 055301.pubmed.ncbi.nlm.nih.gov/17677125
  24. [31]Zhong, S. & Gurnis, M. 1993 Dynamic feedback between a continentlike raft and thermal convection. Journal of Geophysical Research: Solid Earth 98 (B7), 12219–12232.DOI

Paper details

vlnc:W_v3xcn277
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 7

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