betaFeedback

WorksPhysics and Astronomy

Influence of Rotation on Fingering Convection in Planetary Cores

Abstract

Stably stratified layers are thought to develop at the top of the liquid metallic cores of many terrestrial planets. We consider the case where the thermal gradient is stable but the compositional gradient is unstable, a situation particularly relevant to Mercury. The strong contrast between molecular diffusivities of temperature and composition leads to fingering convection. We investigate this process using hydrodynamical simulations in a rotating spherical shell, systematically varying the stratification strength N relative to the rotation rate ΩΩ. In all regimes, the primary fingering mode forms narrow, elongated structures that shift orientation from the rotation axis to the direction of gravity as N2/Ω2N^2/Ω^2 exceeds 10. The fingers remain laminar, with transverse scales proportional to thermal stratification but independent of rotation. Fingering convection also drives secondary large-scale flows across most of the explored parameter space, producing diverse dynamics including zonal flows, hemispherical convection, axisymmetric poloidal bands, finger clusters, and toroidal gyres. In the rapidly-rotating regime, laterally inhomogeneous mixing generates zonal flows in thermo-compositional wind balance; zonal flow direction and amplitude depend on N2/Ω2N^2/Ω^2, with amplitude weakening for strong stratification N2/Ω2>10N^2/Ω^2>10. In the intermediate regime (N2/Ω2∼1N^2/Ω^2\sim 1), axisymmetric or spiraling poloidal bands emerge within the tangent cylinder, gradually overtaking the primary fingers. For stronger stratification, finger clusters and weak, large-scale density anomalies surrounded by toroidal gyres form in the upper domain. These diverse large-scale flows may interact with the dynamo-generated magnetic field in the deeper core, potentially influencing surface magnetic fields.

Introduction

Many terrestrial planets are expected to host a stably stratified layer at the top of their liquid metallic cores as a consequence of thermal and compositional evolution. Thermal stratification can arise when the heat flux across the core-mantle boundary becomes locally sub-adiabatic (e.g. Labrosse [36]; Greenwood et al. [21]) while compositional stratification may result from the accumulation of light elements near the top of the core (e.g. Buffett and Seagle [9]; Bouffard et al. [5]). For the Earth, seismological and geomagnetic observations have been interpreted as evidence for such a stable layer in the uppermost outer core (Helffrich and Kaneshima [29]; Buffett [8]), although its existence, properties and extent remain debated (Lesur et al. [37]; Huguet et al. [31]; Irving et al. [32]). Models of other planetary bodies, including Mercury and Ganymede, similarly suggest thermally stratified regions within their cores (Hauck et al. [28]; Dumberry and Rivoldini [14]; Davies et al. [12]; Knibbe et al. [34]). These stratified layers are often assumed to inhibit radial convective motions while permitting large-scale differential rotation, which may have important implications for the morphology of planetary magnetic fields. For example, Mercury's magnetic field is weak, dominantly dipolar, and unusually axisymmetric (Anderson et al. [1]), features that have been linked to the filtering effect of a stable layer on the dynamo-generated field (Christensen [10]; Christensen and Wicht [11]).

However, the common assumption that stable density gradients largely suppress vertical motions ignores an important aspect of stably stratified fluid dynamics. If one source of buoyancy (thermal or compositional) is destabilising within the stratified layer, the system can become susceptible to double-diffusive instabilities [47]. In particular, a configuration in which the temperature gradient is stabilising while the compositional gradient is destabilising (as is plausible for the upper cores of Mercury and Ganymede) gives rise to fingering convection, a type of double-diffusive convection in which the primary motions take the form of vertical plumes. Consequently, the assumption that radial convective flows are largely suppressed in the stable layer may not hold, raising questions about the influence of stratification on the observed planetary magnetic fields. The presence of fingering convection could have significant effects on the morphology of the magnetic fields. Yet, the interaction between fingering convection and magnetic fields remains largely unexplored [26, 15].

Fingering convection occurs when a stabilising thermal gradient is opposed by an unstable compositional gradient, and where thermal diffusivity greatly exceeds compositional diffusivity [60]. The potential energy linked to the destabilising composition is released through the rapid molecular diffusion of temperature. This process drives vertical motions that can greatly enhance the transport of heat and chemical composition in otherwise stable layers [65, 59]. Fingering convection has been widely explored in fields such as oceanography [54, 35], magma dynamics [58, 25], and stellar interior dynamics [17]. The potential for fingering convection to operate in the metallic cores of planets has recently received increasing attention [38, 43, 6, 41, 57, 39, 23, 63]. A key distinction between these different natural systems lies in the value of the Prandtl number (PrPr), which quantifies the ratio of kinematic viscosity to thermal diffusivity. While oceans and magmas typically have Pr>1Pr > 1, liquid metals such as those found in planetary cores exhibit lower values, on the order of Pr∼0.1Pr \sim0.1. Studies of fingering convection in stellar interiors, where Pr≪1Pr \ll1, have demonstrated that the Prandtl number plays a critical role in controlling the dynamics [16], thereby limiting the direct applicability of results across different natural systems. One notable example is the near-absence of thermo-compositional staircases at low Prandtl numbers [66, 7]. These staircases, which consist of well-mixed layers separated by sharp stratified interfaces, are known to substantially enhance mixing efficiency when present at high Prandtl numbers (e.g. [59]).

The motivation of this work is to characterise the dynamics of fingering convection under conditions relevant to metallic cores. As a first step in this direction, we use a simplified spherical shell model, where the dynamics of the stable layer is studied in isolation from the deeper convective layer. The study focuses on thick stable layers, which are expected to be particularly suitable for Mercury. Thermally stratified regions in liquid outer cores are also relevant for the Earth (e.g. [21]) and Ganymede [27]. This work builds on recent studies by Guervilly [23] and Tassin et al. [63], both of which explore fingering convection in spherical shell geometries across wide regions of parameter space. Tassin et al. [63] investigates the non-rotating regime, while Guervilly [23] considers the regime where rotational effects dominate over stratification (a regime that we will quantify below). In the latter case, at low forcing (i.e. small Rayleigh numbers), the flow organises into tall, columnar structures aligned with the rotation axis, with azimuthal size comparable to the layer thickness [41, 39]. As the forcing increases, these structures become increasingly sheet-like and thin in azimuth while remaining axially elongated. Radial motions remain largely laminar, and strong equatorially-symmetric zonal (i.e. axisymmetric and azimuthal) flows develop, eventually dominating the non-axisymmetric flow. In contrast, in the absence of rotation, Tassin et al. [63] finds that the fingers align radially. Tassin et al. further examines how thermal and compositional fluxes, velocities, and finger sizes scale with key dimensionless parameters, deriving new scaling laws. Additionally, they identify a secondary instability characterised by the formation of large-scale toroidal jets that distort the fingering structures and induce relaxation oscillations. Given the uncertainty surrounding the stratification strength in Mercury’s stable layer, the present study aims to bridge the gap between the strongly-stratified, non-rotating regime of Tassin et al. [63] and the weakly-stratified, rapidly-rotating regime of Guervilly [23]. We focus specifically on identifying the conditions under which the transition between these regimes occurs. As a first step in addressing this complex problem, magnetic fields are neglected in the present model.

The paper is structured as follows. Section 2 presents the mathematical and computational formulation of the model. Section 3 details the results, including the dynamics across weak, intermediate, and rapid rotation regimes, the emergence of differential rotation (i.e. zonal flows), and the efficiency of convective transport. Concluding remarks are provided in Section 4.

Model

Governing equations

We model a stably stratified layer of Boussinesq fluid confined within a spherical shell. The inner boundary, at radius RiR_i, represents the interface with the deeper convective core, while the outer boundary, at radius RoR_o, corresponds to the core-mantle boundary. The thickness of the stable layer at the top of Mercury’s core is not well constrained by thermal evolution models. Here we model a thick layer of depth 800 km, which is at the upper end of the proposed range [68, 12]. This corresponds to an aspect ratio of χ=Ri/Ro=0.6\chi= R_i/R_o = 0.6.

The fluid layer rotates about the zz-axis with rotation rate Ω\Omega. Gravity is directed radially inward and varies linearly with spherical radius rr, given by g=−g0r/Roer\boldsymbol{g}=-g_0r/R_o\boldsymbol{e}_r where g0g_0 is a constant. The fluid has constant viscosity ν\nu, thermal diffusivity κt\kappa_t and compositional diffusivity with κt>ν>κc\kappa_t > \nu> \kappa_c.

The density ρ\rho depends linearly on the temperature TT and the composition CC (i.e. the concentration in light elements) of the Boussinesq fluid:

ρρm=1−αt(T−Tm)−αc(C−Cm),(1)\frac{\rho}{\rho_m}=1-\alpha_t(T-T_m)-\alpha_c(C-C_m), \tag*{(1)}

where αt\alpha_t is the coefficient of thermal expansion, and αc\alpha_c the compositional analogue to αt\alpha_t, with constant αt,αc>0\alpha_t,\alpha_c>0. The subscript mm denotes a constant mean value. The temperature and composition are decomposed into a constant mean, a static profile (denoted by the subscript ss) and a perturbation:

T(r,θ,ϕ,t)=Tm+Ts(r)+Θ(r,θ,ϕ,t),(2)T(r,\theta,\phi,t)=T_m+T_s(r)+\Theta(r,\theta,\phi,t), \tag*{(2)}
C(r,θ,ϕ,t)=Cm+Cs(r)+ξ(r,θ,ϕ,t),(3)C(r,\theta,\phi,t)=C_m+C_s(r)+\xi(r,\theta,\phi,t), \tag*{(3)}

in spherical polar coordinates (r,θ,ϕ)(r,\theta,\phi) and where tt is time.

The equations are nondimensionalised using the layer thickness D=Ro−RiD=R_o-R_i as a unit for lengths, D2/νD^2/\nu for times, ν2/αtg0D3\nu^2/\alpha_tg_0D^3 for temperature, and ν2/αcg0D3\nu^2/\alpha_cg_0D^3 for composition. The dimensionless inner and outer radii are thus ri=Ri/D=1.5r_i=R_i/D=1.5 and ro=Ro/D=2.5r_o=R_o/D=2.5. The dimensionless governing equations are

∂u∂t+(u⋅∇)u+2Ekez×u=−∇p+∇2u+(Θ+ξ)rroer,(4)\frac{\partial\boldsymbol{u}}{\partial t}+(\boldsymbol{u}\cdot\nabla)\boldsymbol{u}+\frac{2}{Ek}\boldsymbol{e}_z\times\boldsymbol{u}=-\nabla p+\nabla^2\boldsymbol{u}+(\Theta+\xi)\frac{r}{r_o}\boldsymbol{e}_r, \tag*{(4)}
∂Θ∂t+u⋅∇Θ+urdTsdr=1Pr∇2Θ,(5)\frac{\partial\Theta}{\partial t}+\boldsymbol{u}\cdot\nabla\Theta+u_r\frac{\mathrm{d}T_s}{\mathrm{d}r}=\frac{1}{Pr}\nabla^2\Theta, \tag*{(5)}
∂ξ∂t+u⋅∇ξ+urdCsdr=1Sc∇2ξ,(6)\frac{\partial\xi}{\partial t}+\boldsymbol{u}\cdot\nabla\xi+u_r\frac{\mathrm{d}C_s}{\mathrm{d}r}=\frac{1}{Sc}\nabla^2\xi, \tag*{(6)}

where u=(ur,uθ,uϕ)\boldsymbol{u}=(u_r,u_\theta,u_\phi) is the solenoidal velocity and pp the pressure. The dimensionless numbers are the Ekman, Prandtl, and Schmidt numbers defined respectively as

Ek=νΩD2,Pr=νκt,Sc=νκc.(7)Ek=\frac{\nu}{\Omega D^2},\qquad Pr=\frac{\nu}{\kappa_t},\qquad Sc=\frac{\nu}{\kappa_c}. \tag*{(7)}

The radial gradients of the static fields are obtained by solving the diffusion equations in the absence of motions. At the inner boundary, the static temperature and composition gradients are imposed:

Ts′(ri)=dTsdr∣ri=−RatPr1ri,Cs′(ri)=dCsdr∣ri=−RacSc1ri,(8)T_s'(r_i)=\left.\frac{\mathrm{d}T_s}{\mathrm{d}r}\right|_{r_i}=-\frac{Ra_t}{Pr}\frac{1}{r_i},\qquad C_s'(r_i)=\left.\frac{\mathrm{d}C_s}{\mathrm{d}r}\right|_{r_i}=-\frac{Ra_c}{Sc}\frac{1}{r_i}, \tag*{(8)}

where the Rayleigh numbers are dimensionless numbers defined as

Rat=αtg0qiRiD3νκt,Rac=αcg0fiRiD3νκc,(9)Ra_t=\frac{\alpha_tg_0q_iR_iD^3}{\nu\kappa_t},\qquad Ra_c=\frac{\alpha_cg_0f_iR_iD^3}{\nu\kappa_c}, \tag*{(9)}

where qi∝−Ts′(ri)q_i\propto-T_s'(r_i) and fi∝−Cs′(ri)f_i\propto-C_s'(r_i) are the dimensional static temperature and composition fluxes at the inner boundary. In fingering convection, the static temperature gradient through the layer is stable while the static compositional gradient is unstable, so qi<0q_i<0 and fi>0f_i>0, and so Rat<0Ra_t<0 and Rac>0Ra_c>0. At the core-mantle boundary, we assume that there is no flux of light elements to/from the mantle, hence Cs′(ro)=0C_s'(r_o)=0. We therefore include a sink term in the diffusion equation for CsC_s to compensate the inflow of light elements at the inner boundary and achieve a steady state (see further detail in Guervilly [23]). For the static temperature, the outward heat flux integrated over the spherical surface at ror_o equals the incoming heat flux integrated over the spherical surface at rir_i to ensure a steady state, so no source term are included in the diffusion equation for TsT_s. The gradient of the static fields in the domain are thus

Ts′=dTsdr=−RatPrrir2.(10)T_s' = \frac{\mathrm{d}T_s}{\mathrm{d}r} = -\frac{Ra_t}{Pr}\frac{r_i}{r^2}. \tag*{(10)}
Cs′=dCsdr=RacSc(1−χ3)(riro3r−rir2).(11)C_s' = \frac{\mathrm{d}C_s}{\mathrm{d}r} = \frac{Ra_c}{Sc(1-\chi^3)}\left(\frac{r_i}{r_o^3}r-\frac{r_i}{r^2}\right). \tag*{(11)}

The boundary conditions for the perturbations are zero radial fluxes at r=rir=r_i and r=ror=r_o,

∂Θ∂r∣ri=∂ξ∂r∣ri=∂Θ∂r∣ro=∂ξ∂r∣ro=0.(12)\left.\frac{\partial\Theta}{\partial r}\right|_{r_i}=\left.\frac{\partial\xi}{\partial r}\right|_{r_i}=\left.\frac{\partial\Theta}{\partial r}\right|_{r_o}=\left.\frac{\partial\xi}{\partial r}\right|_{r_o}=0. \tag*{(12)}

We neglect any dynamical interaction with the deep convective core at r=rir=r_i for simplicity. We therefore use impenetrable boundary conditions at r=rir=r_i and r=ror=r_o. We use stress-free boundary conditions at r=rir=r_i to model the interface with the convective liquid core and no-slip boundary conditions at r=ror=r_o to model the rigid core-mantle boundary. The choice of impenetrable and stress-free boundary condition at r=rir=r_i does not capture the interaction between a stably stratified layer and an underlying convective region. This choice is made primarily for numerical convenience, enabling a systematic exploration of parameter space. This simplification also allows us to study the intrinsic dynamics of the stably stratified layer in isolation.

The diffusivity coefficients are held constant throughout this study. Following Monville et al. [41]; Guervilly [23], we adopt a Prandtl number of Pr=0.3Pr=0.3 and a Schmidt number of Sc=3Sc=3, which yields a Lewis number Le=κt/κc=Sc/Pr=10Le=\kappa_t/\kappa_c=Sc/Pr=10 for all simulations. Although this value is lower than that expected in planetary cores (Le=O(103)Le=O(10^3)), it remains computationally tractable and allows for a meaningful exploration of double-diffusive fingering instabilities within a relevant parameter regime as described below.

Stratification and domain of fingering convection

The stratification of the layer measures the stability of the fluid to vertical displacements. It is quantified by the buoyancy frequency NN, also called the Brunt-Väisälä frequency, which corresponds to the frequency of oscillations of a vertically-displaced fluid parcel under the restoring force of gravity. N2N^2 is proportional to minus the radial density gradient, so in a stably stratified layer, where the density decreases with radius, N2>0N^2>0. The system is unstably stratified when N2<0N^2<0. We measure the buoyancy frequency from the static profiles (10)-(11), which gives in dimensionless units,

N~2=N2ν2/D4=rro(Ts′+Cs′).(13)\widetilde{N}^2=\frac{N^2}{\nu^2/D^4}=\frac{r}{r_o}\left(T_s'+C_s'\right). \tag*{(13)}

For our choice of static gradients, N2N^2 varies slightly with radius, with the stratification being strongest at the bottom of the layer and N2(ro)=0.85N2(ri)N^2(r_o)=0.85N^2(r_i). In the rest of the paper, values of N2N^2 will always be quoted at r=rir=r_i.

In a rotating domain, it is convenient to measure NN relative to the rotation rate Ω\Omega:

N2Ω2=Ek2ro(RatPr+RacSc)=Ek2roRacSc(Rρi−1),(14)\frac{N^2}{\Omega^2}=\frac{Ek^2}{r_o}\left(\frac{Ra_t}{Pr}+\frac{Ra_c}{Sc}\right)=\frac{Ek^2}{r_o}\frac{Ra_c}{Sc}\left(R_\rho^i-1\right), \tag*{(14)}

where the density ratio RρR_\rho is

Rρ=∣Ts′∣∣Cs′∣,Rρi=Rρ(ri)=∣Rat∣LeRac.(15)R_\rho=\frac{|T_s'|}{|C_s'|},\qquad R_\rho^i=R_\rho(r_i)=\frac{|Ra_t|Le}{Ra_c}. \tag*{(15)}

Double-diffusive instabilities are commonly characterised by the density ratio, which delineates the conditions under which they may arise. In non-rotating, planar layers with uniform static gradients, the system becomes linearly unstable to the double-diffusive fingering instabilities when

1<Rρ<Le,(16)1<R_\rho<Le, \tag*{(16)}

assuming Le≫1Le \gg1 and sufficiently high Rayleigh numbers ([60, 4]). Values of Rρ<1R_{\rho} < 1 correspond to unstably stratified systems with N2<0N^2 < 0. Rotational effects do not significantly alter this critical range for large RacRa_c ([56, 41]). However, in spherical geometry with rotation, large-scale double-diffusive modes with low azimuthal wavenumber can be preferred to small-scale fingers near the stability boundaries, especially for small RacRa_c ([57, 41, 23]). Fingering convection can also be preferred for Rρ<1R_{\rho} < 1 ([24, 55, 33, 70]). Since Le≫1Le \gg1 in planetary cores, the parameter regime prone to fingering instability is therefore broad, suggesting that fingering convection may play a role across a wide range of thermal and compositional stratifications.

In our study, RρR_{\rho} increases with radius (see Figure 2 in Guervilly [23]). At the bottom of the domain, the unstable range (16) corresponds to

Isosurfaces of the radial velocity for two representative cases

Figure 2. Isosurfaces of the radial velocity for two representative cases of (a) N2/Ω2≫1N^2/\Omega^2 \gg1 (Ek=10−1Ek = 10^{-1} in Series 1) and (b) N2/Ω2≪1N^2/\Omega^2 \ll1 (Ek=10−4Ek = 10^{-4} in Series 1). Red (blue) indicates positive (negative) values of uru_r and the isosurfaces corresponds to 10% of the maximum values of uru_r in a given snapshot.

RacLe<∣Rat∣<Rac.(17)\frac{Ra_c}{Le} < \lvert Ra_t \rvert< Ra_c. \tag*{(17)}

The density ratio goes to infinity at the outer boundary as C′C' is zero there. RρR_{\rho} crosses the theoretical onset of fingering convection for Rρ(r=rs)=LeR_{\rho}(r = r_s) = Le. This implies that, while the bottom of the domain might be prone to fingering instabilities, the top of the domain might not be. As the radial profile of RρR_{\rho} depends on ∣Rat∣/Rac\lvert Ra_t \rvert/Ra_c (and not on RatRa_t or RacRa_c individually), fixing this ratio leaves radius rsr_s unchanged. In this study, we fix Rat=−Rac/3Ra_t = -Ra_c/3 i.e. Rρ=Le/3R_{\rho} = Le/3. This implies that rs=2.26r_s = 2.26, so we would expect fingering convection to occur in the inner 76% of the radial domain at large Rayleigh numbers.

Numerical method

All the simulations are performed with the open-source code XSHELLS (Schaeffer [51]; Schaeffer et al. [52]). XSHELLS is a C++ pseudo-spectral code that solves the governing equations (4)-(6) in a 3D spherical geometry. To ensure that the velocity field is solenoidal, u\mathbf{u} is decomposed into poloidal and toroidal scalars, UpU_p and UtU_t,

u=∇×∇×(Upr)+∇×(Utr),\mathbf{u} = \nabla\times\nabla\times(U_p\mathbf{r}) + \nabla\times(U_t\mathbf{r}),

where r\mathbf{r} is the position vector. All the scalars (UpU_p, UtU_t, Θ\Theta and ξ\xi) are expanded in spherical harmonics Ylm(θ,ϕ)Y_l^m(\theta, \phi) of degree ll and order mm, which are truncated at Lmax⁡L_{\max} and Mmax⁡M_{\max} respectively. In the radial direction, the code uses a second-order finite difference scheme with NrN_r points. The code uses a second-order time-stepping scheme with an implicit treatment of the diffusive terms and explicit treatment of the non-linear terms. Details of the numerical resolution used for each simulation are provided in Appendix § 9.

Results

Overview

Parameter survey

We first give an overview of the results before describing the flow in the different dynamical regimes in later sections. All the simulations were run at a constant density ratio Rρi=Le/3R_{\rho}^{i} = Le/3. As shown in Figure 1, the ratio N2/Ω2N^2/\Omega^2 is varied in two ways: (1) varying Ekman number at fixed Rayleigh number (Rac=2×107Ra_c = 2 \times10^7) (Series 1); (2) varying Rayleigh numbers (i.e. varying RacRa_c and RatRa_t by the same factor to maintain constant RρiR_{\rho}^{i}) at fixed Ekman number (Ek=10−4Ek = 10^{-4}) (Series 2). Tables A.1 and A.2 in Appendix § 9 list the input parameters and numerical resolutions of all the simulations. The range of the dynamical lengthscales increases with the Rayleigh numbers, implying that the exploration of the regime N2/Ω2≫1N^2/\Omega^2 \gg1 is more computationally demanding in Series 2 and, consequently, restricted to N2/Ω2≲25N^2/\Omega^2 \lesssim25. Simulations with increasing N2/Ω2N^2/\Omega^2 in Series 1 correspond to Ek→1Ek \to1, so the weakening of rotation relative to stratification is accompanied by enhanced viscous effects (e.g. cases with N2/Ω2>10N^2/\Omega^2 > 10 correspond to Ek>10−3Ek > 10^{-3}). While these simulations offer useful insights to complement the results from Series 2, their interpretation should be approached carefully in the context of planetary core applications where Ek≪1Ek \ll1.

Simulations in the parameter space $Ek$ – $Ra_c$

Figure 1. Simulations in the parameter space EkEk – RacRa_c. All the simulations were performed with Le=10Le = 10, Pr=0.3Pr = 0.3 and Rρi=Le/3R_{\rho}^{i} = Le/3. The lower (upper) dashed line represents N2/Ω2=1N^2/\Omega^2 = 1 (N2/Ω2=10N^2/\Omega^2 = 10 respectively). Series 1 simulations are represented in red and Series 2 simulations in blue. Columnar/radial/bands refer to the structure of the radial flows in the rapidly-rotating/weakly-rotating/intermediate regimes. The location of the finger clusters (§ 4.2) is indicated by large open circles.

For convenience, we define the weakly-rotating regime as N2/Ω2>O(1)N^2/\Omega^2 > \mathcal{O}(1), understood to be relative to the strength of the stratification. Conversely, we refer to N2/Ω2<O(1)N^2/\Omega^2 < \mathcal{O}(1) as the rapidly-rotating regime. In single-component rotating convection, the convective Rossby number is often used as a control parameter that determine regime transitions (e.g. Aurnou et al. [3]). Here we base its definition on the control parameters of compositional buoyancy,

Roc=RacEk2/ScRo_c = \sqrt{Ra_c Ek^2/Sc}. From equation (14), N2/Ω2N^2/\Omega^2 is directly related to the convective Rossby number for fixed density ratio. Here Rρi=3.33R_\rho^i = 3.33, so N2/Ω2≈Roc2N^2/\Omega^2 \approx Ro_c^2; Values of RocRo_c are given in Tables A.1 and A.2.

Two cases representative of N2/Ω2≫1N^2/\Omega^2 \gg1 and N2/Ω2≪1N^2/\Omega^2 \ll1 are illustrated in Figure 2, which show snapshots of the isosurfaces of the radial velocity. Comparison of the two cases show one immediate difference: whilst the structure of the radial velocity in the weakly-rotating case is essentially aligned with gravity (“radial” fingers), it is aligned with the rotation axis in the rapidly-rotating case (“columnar” fingers). In both cases, the fingers do not occupy the whole volume of the sphere: the density ratio increases with radius in our system so the fingers develop preferentially close to the inner boundary. Additionally, fingering convection within the tangent cylinder is inhibited for rapid rotation, and only sets in at higher values of N2/Ω2N^2/\Omega^2. This is due to the detrimental effect of rotation on the buoyancy-driven instability, which is particularly pronounced when the rotation axis is aligned with gravity, similar to the behaviour observed in single-component overturning convection (e.g. Gastine and Aurnou, [19]). Fingering convection can occur inside the tangent cylinder for N2/Ω2<1N^2/\Omega^2 < 1 as shown in Section § 5.

Finger scale

In the weakly- and rapidly-rotating regimes, the fingers have a specific lengthscale in the transverse direction (i.e. the direction tangential to the rotation axis for rapid rotation and to the radial direction for weak rotation). In non-rotating plane layer, the fingering instability develops as elongated fingers with a transverse lengthscale that follows the dimensionless “Stern scale” Ltheo=(∣Rat∣/ro)−1/4L_{\mathrm{theo}} = (|Ra_t|/r_o)^{-1/4} for large Rayleigh numbers and when 1≪Rρ≪Le1 \ll R_\rho\ll Le (Stern, [60]; Schmitt [53]). In spherical shell simulations without rotation, Tassin et al. [63] found that the observed transverse finger scale closely follows the Stern scale’s dependence on RatRa_t, although it also exhibits sensitivity to the density ratio. In the rapidly-rotating regime, Guervilly [23] demonstrated that, at a fixed density ratio and sufficiently large Rayleigh numbers, the transverse finger scale remains well described by the Stern scale across varying Ekman numbers. This suggests that, for N2/Ω2<1N^2/\Omega^2 < 1, rotation does not significantly influence the transverse finger scale.

To measure a typical transverse finger scale, we estimate the most energetic scale of the poloidal flow from the power spectra of the poloidal kinetic energy in Series 1 and 2 cases (Figure 3). The power spectra density of the kinetic energy is used to identify the dominant energy-carrying horizontal lengthscales of the flow and the location of the spectral peak might reflect transitions between different dynamical regimes. In Series 1, where RatRa_t is held constant, the spectra exhibit a well-defined peak for both N2/Ω2<1N^2/\Omega^2 < 1 and N2/Ω2>1N^2/\Omega^2 > 1. In both cases, the peak occurs at approximately the same wavenumber, consistent with the Stern scaling, indicating that it is largely insensitive to variations in EkEk. The most noticeable difference is that length scales at or above the Stern scale are more energetic for N2/Ω2>1N^2/\Omega^2 > 1. In Series 2, where RatRa_t increases, a well-defined spectral peak is again observed for both N2/Ω2<1N^2/\Omega^2 < 1 and N2/Ω2>1N^2/\Omega^2 > 1. As N2/Ω2N^2/\Omega^2 increases, the flow becomes more energetic across all scales. When the spherical harmonics degree ll is normalised by LtheoL_{\mathrm{theo}} on the xx-axis (Figure 3c), the spectral peaks align, indicating consistency with the predicted scaling. The intermediate regime (N2/Ω2≈1N^2/\Omega^2 \approx1) is visibly different: in Series 1, the spectrum displays a flat slope at low wavenumbers up to the Stern scale; in Series 2, two separate spectral peaks emerge: one at the Stern scale and another at approximately half that wavenumber.

Figure 3. Top row: Power Spectral Density (PSD) of the poloidal kinetic energy as a function of the spherical harmonics degree ll for selected cases in Series 1 and 2 in (a)-(b). The spectra are calculated from a snapshot and averaged in radius excluding the boundary layers. In (c), the degree ll is normalised by Ltheo=(∣RaT∣/ro)−1/4\mathcal{L}_{\mathrm{theo}} = (|\mathrm{Ra}_{T}|/r_{o})^{-1/4}. N2/Ω2=0.06N^{2}/\Omega^{2} = 0.06 has the same parameters in both Series. Bottom row: Variation of the lengthscale measured from the spectral peak L\mathcal{L} normalised by Ltheo\mathcal{L}_{\mathrm{theo}} as a function of N2/Ω2N^{2}/\Omega^{2}. Data from Monville et al. [41] (M19) and Guervilly [23] (G22) are included for smaller EkEk.

To quantify the departure from the Stern scale, we measure the characteristic lengthscale of the most energetic poloidal flow by identifying the spectral peak (spherical harmonic degree lpeakl_{\mathrm{peak}}) in each simulation. This peak is computed from the ll spectra averaged radially, excluding the boundary layers. The corresponding length scale, L=πri/lpeak\mathcal{L} = \pi r_i/l_{\mathrm{peak}}, is shown in Figure 3(d), where it is normalised by LtheoL_{\mathrm{theo}}. We also include data from Monville et al. [41] and Guervilly [23] at lower Ekman numbers (and same PrPr and LeLe) for comparison. Across all studies, at small Rayleigh numbers (RacEk≲103Ra_c Ek \lesssim10^{3}–10410^{4}), the columnar flows exhibit a large azimuthal extent, comparable to the layer thickness (as also observed by Mather and Simitev, [39]). As the Rayleigh number increases, these structures evolve into narrow, sheet-like fingers (see Figure 2(b)), whose typical transverse scale progressively approaches the Stern scale and is well approximated by L=5Ltheo\mathcal{L} = 5L_{\mathrm{theo}}, independent of EkEk. This behaviour confirms that, at sufficiently large Rayleigh numbers and for a fixed density ratio, the transverse scale of both columnar and radial fingers is governed by the Stern scale. In contrast, the intermediate regime is characterised by the emergence of larger scales, which is particularly evident in the data points from Series 2. Section § 5.3 will explore this novel feature.

Scaling of the radial velocity

To evaluate how the radial variation of RρR_\rho influences the vigour of fingering convection throughout the layer, Figure 4 shows the radial profile of the root mean square (r.m.s.) radial velocity for increasing values of N2/Ω2N^2/\Omega^2.

Figure 4. Radial profile of the r.m.s. radial velocity for selected simulations in Series 2 (0.25≤N2/Ω2≤24.90.25 \le N^2/\Omega^2 \le24.9) and Series 1 (N2/Ω2=62240N^2/\Omega^2 = 62240). The vertical dashed line corresponds to the radius rsr_s where Rρ(r=rs)=LeR_\rho(r = r_s) = Le.

radial profiles are used to assess the radial localisation of the kinetic energy, which indicates how the active regions of the dynamics shift with control parameters. In all cases, radial motions are strongest near the bottom of the layer, just outside the lower boundary layer, and decrease monotonically with radius. Only weak radial flows are observed to cross the upper region beyond r>rsr > r_s (where Rρ(r=rs)=LeR_\rho(r = r_s) = Le).

The saturation of the flow amplitude is commonly assessed via the dependence of the Reynolds number on the control parameters. Here we measure the Reynolds number ReRe as a dimensionless measure of the r.m.s. poloidal velocity VpV_p,

Re=VpDν,(18)Re = \frac{V_pD}{\nu}, \tag*{(18)}

where VpV_p is calculated from the volume and time average of the poloidal kinetic energy. The variation of ReRe with the parameters can be compared with the theoretical scalings that have been proposed in non-rotating (e.g. Brown et al. [7]; Yang et al. [71]; Hage and Tilgner [24]; Tassin et al. [63]) and rotating fingering convection (Sengupta and Garaud [56]) for regimes of weakly and strongly driven fingering convection, defined by the reduced density ratio (Traxler et al. [65]):

rρ=Rρi−1Le−1.(19)r_\rho= \frac{R_\rho^i - 1}{Le - 1}. \tag*{(19)}

The regime of weakly driven convection corresponds to rρ≈1r_\rho\approx1 (i.e. close to the linear onset for large Rayleigh numbers, Rρi≈LeR_\rho^i \approx Le), whilst the regime of strongly-driven fingering convection, close to neutral stability (Rρi=1R_\rho^i = 1), corresponds to 0<rρ≪10 < r_\rho\ll1. Note that here rρ=0.26r_\rho= 0.26, so our simulations have moderate driving by that definition. Nevertheless, we can attempt to compare our numerical data to theoretical scalings proposed for rρ≪1r_\rho\ll1.

A simple theoretical scaling can be derived by assuming that the timescale for thermal diffusion and advection are similar at the finger scale L\mathcal{L}, meaning that the typical radial velocity of the fingers is of the order of κt/L\kappa_t/\mathcal{L}, i.e. the local Péclet number is order unity (Radko [47]):

ReL=ReL∼1Pr.(20)Re_{\mathcal{L}} = Re\mathcal{L} \sim\frac{1}{Pr}. \tag*{(20)}

However, for small Prandtl numbers, Brown et al. [7] and Sengupta and Garaud [56] argue that the saturation of the fingering instability is caused by a secondary shear instability of the fingers (Radko and Smith [48]) and propose an alternative scaling depending on rρr_\rho. In particular, for the range of parameters that is closest to our study, rρ≪Pr≪1r_\rho\ll Pr \ll1 and Le≫1Le \gg1, Brown et al. [7] found that

ReL∼1Pr1/2.(21)Re_{\mathcal{L}} \sim\frac{1}{Pr^{1/2}}. \tag*{(21)}

An alternative scaling was proposed by Tassin et al. [63] for non-rotating fingering convection by considering that the buoyancy sources must balance the viscous dissipation in a steady state. The scaling assumes that the viscous dissipation occurs at the finger size and that the convective transport follows the 1/3-law between the compositional Nusselt number and RacRa_c (see discussion in § 5.6). At rρ≪1r_\rho\ll1, this theoretical scaling is

Re∼Rac2/3∣Rat∣−1/4Sc−1.(22)Re \sim Ra_c^{2/3} |Ra_t|^{-1/4} Sc^{-1}. \tag*{(22)}

Tassin et al. found a good agreement between their data, those of Yang et al. [71] and this theoretical scaling. The scaling can be re-written for the local Reynolds number, assuming that L∝∣Rat∣−1/4\mathcal{L} \propto|Ra_t|^{-1/4},

ReL=ReL∼Rac1/6Rρ−1/2Pr−1/2Sc−1/2.(23)Re_{\mathcal{L}} = Re\mathcal{L} \sim Ra_c^{1/6} R_\rho^{-1/2} Pr^{-1/2} Sc^{-1/2}. \tag*{(23)}

Figure 5 shows the variation of the local Reynolds number, ReL=ReLRe_{\mathcal{L}} = Re\mathcal{L}, where L\mathcal{L} is measured from the spectral peak, as a function of N2/Ω2N^2/\Omega^2 in our simulations. The data from Monville et al. [41] and Guervilly [23] are also included. First we note that the local Reynolds numbers are close to or below one, meaning that, at the finger scale, the poloidal flow remains laminar. The local Reynolds number remains fairly constant throughout. The data cannot distinguish between the two scalings (20) and (21) since the Prandtl number is fixed and not particularly small (Pr=0.3Pr = 0.3). Similarly, the density ratio is fixed, so we can only test the dependence of ReRe on one of the Rayleigh numbers. The numerical data exhibit a weak increasing trend that appears consistent with the scaling (23) derived by Tassin et al. [63], although the shallow dependence on RacRa_c makes numerical confirmation challenging. This trend is also visible in Series 1 for N2/Ω2<100N^2/\Omega^2 < 100, despite RacRa_c being held fixed. For N2/Ω2>100N^2/\Omega^2 > 100, ReLRe_{\mathcal{L}} remains approximately constant, suggesting that rotational effects become dynamically negligible as expected at high N2/Ω2N^2/\Omega^2 in Series 1, i.e. as Ek→1Ek \to1. This behaviour would not necessarily persist in Series 2 if RacRa_c could be increased further. The critical Rayleigh number for the onset of fingering convection, RaccritRa_c^{\mathrm{crit}}, scales as Raccrit∼RacEk−1Ra_c^{\mathrm{crit}} \sim Ra_c Ek^{-1} for Ek≪1Ek \ll1 (Monville et al. [41]). It is therefore worth noting that the increase of the supercriticality Rac/Raccrit∼RacEkRa_c/Ra_c^{\mathrm{crit}} \sim Ra_c Ek in Series 1 may partially account for the weak increase in the local Reynolds number for N2/Ω2<100N^2/\Omega^2 < 100 in this case.

Figure 5. Variation of the local Reynolds number ReLRe_{\mathcal{L}} as a function of N2/Ω2N^2/\Omega^2. The Reynolds numbers is based on the mean poloidal velocity. The theoretical scaling of Tassin et al. [63], ReL∼Rac1/6Re_{\mathcal{L}} \sim Ra_c^{1/6}, is indicated by a dashed line. Data from Monville et al. [41] (M19) and Guervilly [23] (G22) are included for smaller EkEk.

In summary, we find that the local Reynolds numbers in our simulations remain nearly constant and close to unity. The scaling (23) proposed by Tassin et al. [63] for non-rotating convection, which predicts a weak dependence of ReLRe_{\mathcal{L}} on RacRa_c, is consistent with our simulations, but a more definitive confirmation would require a wider exploration of the parameter space. Remarkably, comparison with the results of Monville et al. [41] reveals good agreement between full-sphere and spherical-shell configurations, despite differences in geometry, boundary conditions, and stratification profiles.

In the following, we will briefly describe the flows in the weakly- and rapidly-rotating regimes (§ 4 and 3.3) before presenting the intermediate regime in more detail (§ 5.3). Section 3.5 presents the formation and saturation of the axisymmetric and azimuthal (i.e. zonal) flows. In § 5.6, we discuss the transport properties of the fingering convection in the different regimes.

Weak rotation

Radial fingers

Figure 6 shows the radial velocity projected on a spherical surface at r=1.8r = 1.8 for simulations in Series 1 and 2 representative of the weakly-rotating regime with N2/Ω2>1N^{2}/\Omega^{2} > 1. Since the thermal stratification is significantly stronger for our chosen case from Series 2 (∣RaT∣|Ra_{T}| is 400 times larger), the finger size is visibly smaller. In either cases, no noticeable difference in the radial flows of the equatorial and polar regions is visible, suggesting that the dynamics of the fingers are weakly influenced by rotation. The meridional slices of uru_{r} in Figure 6 shows that, as expected from the radial profiles in Figure 4, the radial fingers are preferentially located in the inner region where the density ratio is smallest. Similarly to the radial fingers described in Tassin et al. [63] in the absence of rotation, the fat fingers at low RacRa_{c} (Series 1) are tubular structures with a radial lengthscale that extend over more than half of the layer depth, whilst at larger RacRa_{c} (Series 2), the thin fingers have undulating structures with shorter radial lengthscale.

Snapshots of the radial velocity for simulations in the weakly-rotating regime

Figure 6. Snapshots of the radial velocity for simulations in the weakly-rotating regime: (a)-(b) Aitoff projection at r=1.8r = 1.8 (left) and meridional slice (right) for (a) N2/Ω2=62240N^2/\Omega^2 = 62240 in Series 1 (Ek=10−1Ek = 10^{-1}, Rac=2×107Ra_c = 2 \times10^7) and (b) N2/Ω2=25N^2/\Omega^2 = 25 in Series 2 (Ek=10−4Ek = 10^{-4}, Rac=8×109Ra_c = 8 \times10^9). Plots (c) and (d) are close-ups of the spherical projections. In the Aitoff projections, the gray lines indicates the location of the tangent cylinder.

The spherical projections of uru_{r} in Figure 6 (especially in Series 1) show an asymmetry in the distribution of the upward and downward flows, with concentrated upflows amongst more dilute downflows. The net radial velocity must be zero to respect mass conservation, but the distribution can be skewed. The Probability Distribution Function (PDF) of the radial velocity at r=1.8r = 1.8 is shown in Figure 7. The distribution is skewed for both cases from Series 1 and 2, with a long tail in the positive values. The skewness is very evident for Series 1, with a prominent shoulder corresponding an accumulation of weak negative values. We calculate the skewness of the distribution as

PDF of the radial velocity for Series 1 and its mirrored distribution

(a) N2/Ω2=62240N^2/\Omega^2 = 62240 (Series 1)

PDF of the radial velocity for Series 2 and its mirrored distribution

(b) N2/Ω2=25N^2/\Omega^2 = 25 (Series 2)

Radial profile of the skewness

(c)

Figure 7. (a) and (b) PDF of the radial velocity at r=1.8r = 1.8 for N2/Ω2>1N^2/\Omega^2 > 1. The distribution is mirrored with respect to x=0x = 0 to highlight the asymmetry between positive and negative values. (c) Radial profile of the skewness γ(r)\gamma(r) of the radial velocity.

γ(r)=⟨ur3⟩S(r)⟨ur2⟩S(r)3/2,\gamma(r) = \frac{\langle u_{r}^{3}\rangle_{S(r)}}{\langle u_{r}^{2}\rangle_{S(r)}^{3/2}},

where the angular brackets correspond to an average over the spherical surface S(r)S(r). Figure 7c shows the radial profile of γ(r)\gamma(r) for these two cases. Note that a limited number of data snapshots were available for the case of Series 2, so the profile shows significant fluctuations at large radii. In either case, the radial velocity is positively skewed for most of the domain and the skewness increases with radius. At radius greater than r>2.3r > 2.3, where the flow is weak, the skewness drops and becomes negative in the outer boundary layer.

The pattern of skewness for r<2.3r < 2.3 can be understood from the observations above, that the fingers are tubular (i.e. of constant width), and more pronounced and consistent for Series 1. The finger width is broadly constant as the buoyant fluid rises because this is controlled by the Stern scale. In the spherical geometry, the filling factor of the regions of upwelling fluid therefore decreases with radius, and from conversation of mass, the outward radial velocity in the upwellings increasingly dominates over the inward velocity elsewhere, leading to the positive skewness. The stronger skewness observed for Series 1 in Figure 7c is linked to the heavier positive tail noted in Figure 7a, and the more consistent extension of the fingers throughout the shell.

Finger clusters and toroidal gyres

In some cases with N2/Ω2>1N^{2}/\Omega^{2} > 1, the dynamics change in the upper half of the domain (r≳2r \gtrsim2) with the localisation of radial fingers into large-scale patches (or clusters) at high latitudes. This is illustrated in Figure 8 showing the spherical projection of uru_{r} and uθu_{\theta} at r=2.1r = 2.1 for N2/Ω2=12.4N^{2}/\Omega^{2} = 12.4 in Series 2. Fingering convection in the lower half of the domain is more vigorous and there is no evidence of clustering, similarly to the dynamics observed in Figure 6(b). The patches in the upper domain are much larger scale than the transverse finger size and are accompanied by toroidal gyres. Figure 9 shows the poloidal and toroidal kinetic energy spectra as a function of the azimuthal order mm at different radii for the case of Figure 8. In the lower domain, the poloidal and toroidal flows involve a wide range of azimuthal wavenumbers from m=1m = 1 to m≈50m \approx50, whilst in the upper domain, the flow is dominated by smaller azimuthal wavenumbers (m≲20m \lesssim20) and is prominently toroidal. The zonal flow (toroidal m=0m = 0 mode) dominates at all radii and will be discussed separately in section 3.5.

Aitoff projection of the radial velocity

(a) uru_r

Aitoff projection of the latitudinal velocity

(b) uθu_\theta

Figure 8. Aitoff projection at r=2.1r = 2.1 of a snapshot of the radial and latitudinal velocities for N2/Ω2=12.4N^2/\Omega^2 = 12.4 in Series 2.

PSD kinetic energy spectra of the poloidal component

(a) poloidal

PSD kinetic energy spectra of the toroidal component

(b) toroidal

Radial profiles of the filtered r.m.s. latitudinal velocity and composition perturbation

(c)

Figure 9. (a)–(b) Spectra of the poloidal and toroidal kinetic energy as a function of the azimuthal order mm at different radii for N2/Ω2=12.4N^{2}/\Omega^{2}=12.4 in Series 2. (c) Radial profiles of the filtered r.m.s. latitudinal velocity and composition perturbation. The composition perturbation is normalised by CsC_{s} at r=rir=r_{i}.

The clusters are associated with the presence of large-scale density anomalies that emerge in the upper domain. The fingers tend to cluster in the heavier anomalies, and are nearly absent in the lighter anomalies. Large-scale upwelling and downwelling flows are associated with these density anomalies, but they are very weak compared with the

fingers in the lower domain as indicated by the spectra of Figure 9; they are also not visible in the radial profiles of the r.m.s. velocity in Figure 4 for N2/Ω2>1N^2/\Omega^2 > 1. The large-scale anomalies are more evident in the composition perturbation, so we use the composition to understand the spatial relation between anomalies and toroidal gyres. Large-scale temperature perturbations also exist, which counteract the composition perturbations; however, their amplitude is smaller, resulting in the composition perturbations dominating the density anomalies. Figure 9c shows the radial profiles of the r.m.s. uθu_\theta and ξ\xi separated into large azimuthal scales (1≤m≤201 \le m \le20) and smaller scales (m>20m > 20). The large-scale latitudinal velocity dominates in the upper domain and the r.m.s. profile has two local peaks (located at r≈2.1r \approx2.1 and r≈2.4r \approx2.4) that surround a peak of the large-scale composition perturbation.

Since the flow is dominated by the toroidal component at large radii, we plot in Figure 10 isocontours of the non-axisymmetric toroidal scalar UtU_t and composition perturbation at radius r=2.1r = 2.1 and r=2.4r = 2.4. The location of the toroidal gyres and composition patches is clearly correlated. At r=2.1r = 2.1, anti-clockwise gyres (Ut>0U_t > 0) are associated with lighter patches (ξ>0\xi> 0) in the northern hemisphere, and vice-versa in the southern hemisphere. This hemispherical dependence indicates the importance of the Coriolis force in the formation of these toroidal flows. The composition patches extend through the upper domain and keep their coherence. The toroidal gyres change sign of rotation at r=2.4r = 2.4, i.e. clockwise gyres (Ut<0U_t < 0) are now associated with lighter patches. These observations suggest that the slow, large-scale upflows of lighter fluid draw convergent horizontal flows at the lower roots around r≈2.1r \approx2.1 and divergent horizontal flows at their top (r≈2.4r \approx2.4) (poloidal uθu_\theta and uϕu_\phi not shown here); and vice-versa for downflows of heavier fluid. These convergent/divergent flows are deflected by the Coriolis force into anti-clockwise/clockwise toroidal gyres in the northern hemisphere (and vice-versa in the southern hemisphere).

Isocontours of non-axisymmetric $\xi$ and $U_t$ in an Aitoff projection at $r = 2.1$ and $r = 2.4$

Figure 10. Isocontours of non-axisymmetric ξ\xi (colour) and UtU_t (black lines) in an Aitoff projection at r=2.1r = 2.1 and r=2.4r = 2.4 for N2/Ω2=12.4N^2/\Omega^2 = 12.4 in Series 2. The composition is normalised by CsC_s at r=rir = r_i. For UtU_t, the solid (dashed) lines indicates positive (negative) contours. Anti-clockwise gyres have positive UtU_t.

The finger clusters occur in a limited range of the parameter space indicated by open circles in Figure 1. They are not visible in Series 1, which suggests that sufficiently low Ekman number might be required. To test their sensitivity to changes in N2/Ω2N^2/\Omega^2, we ran a number of simulations at intermediate values of RacRa_c indicated in Figure 1. The clusters are present for N2/Ω2≈100N^2/\Omega^2 \approx100, but disappear for N2/Ω2≈1000N^2/\Omega^2 \approx1000, suggesting that they occur in the limited range N2/Ω2≈1N^2/\Omega^2 \approx1–100100 for Ek≲6×10−4Ek \lesssim6 \times10^{-4}.

Finger clustering was previously observed in Cartesian box simulations at high PrPr and low density ratio [44], but this is likely through a different mechanism than observed here. Indeed, in these box simulations, the clustering is defined as the formation of large-scale buoyant structures and is thought to occur via entrainment as the local Reynolds number might exceed unity. Unlike in our simulations, this occurs in the absence of rotation. Here the clustering corresponds to a spatial localisation of fingering convection. It is linked to the development of large-scale density anomalies: within the heavier anomalies, the local compositional Rayleigh number, estimated from the local radial compositional gradient, is up to four greater than in the lighter anomalies. Consequently, fingering convection is preferentially promoted in the heavier anomalies. This feedback is suggestive of the presence of large-scale instabilities (e.g. [66, 47]). The occurrence of large-scale buoyant patches in the presence of latitudinal gradients of temperature and composition (arising from rotationally induced latitudinal inhomogeneities in the mixing, see § 5.6) is reminiscent of lateral intrusions, which manifest as laterally interleaving layers ([30, 50, 40]). Moreover, lateral intrusions are known to develop at large density ratios (Rρ>LeR_{\rho} > Le), outside the parameter regime unstable to the primary fingering instability, consistent with the observation that the large-scale patches occurs in the upper part of the layer in our simulations. Although intrusions typically form quasi two-dimensional, tilted horizontal layers, it is plausible that spherical geometry and Coriolis effects disrupt these layers into patches of finite azimuthal extent. The possible formation of intrusions in our simulations warrants further investigation, which we plan to pursue in future work, particularly in simplified configurations (e.g. in the strongly-stratified regime Rρ>LeR_{\rho} > Le, as in Medrano et al. [40]).

Rapid rotation

Columnar fingers

Flows in the rapid-rotation regime were described in Guervilly [23] in the same configuration and using a larger dataset including lower Ekman numbers (Ek∈[10−6,10−4]Ek \in[10^{-6}, 10^{-4}]). The main features are summarised in this short subsection for completeness.

Figure 11 shows spherical projections and equatorial slices of the radial velocity for two cases of Series 2 with N2/Ω2<1N^2/\Omega^2 < 1. For the smallest stratification, columnar fingers occur mainly outside of the tangent cylinder, impinging slightly inside the tangent cylinder. Larger N2/Ω2N^2/\Omega^2 leads to the development of columnar fingers in the polar regions. The radial flows gradually become stronger in the polar regions than in the equatorial regions as N2/Ω2N^2/\Omega^2 increases, whilst fingering convection become localised in the equatorial region. Zonal flows develop in the equatorial region (see § 5.5), leading to a visible shear and disruption of fingering convection in that region, which might explain why radial flows are relatively more vigorous in the polar regions. The localised fingering structures in the equatorial region drift azimuthally as they are advected by the zonal flow.

Snapshot of the radial velocity for simulations in the rapidly-rotating regime in Series 2

Figure 11. Snapshot of the radial velocity for simulations in the rapidly-rotating regime in Series 2 (Ek=10−4Ek = 10^{-4}): Aitoff projection at r=1.8r = 1.8 (left) and equatorial slice (right).

Equatorially antisymmetric mode

We observe an interesting behaviour (not reported in Guervilly [23]) from an extended running time for the simulation with N2/Ω2=0.06N^2/\Omega^2 = 0.06 (which is shared by Series 1 and 2) as shown in the time series of Figure 12. The flow is columnar and located outside the tangent cylinder. In the initial steady state (Figure 12(b)), the flow is symmetric with respect to the equatorial plane (equatorially symmetric – ES) with a dominant non-axisymmetric component. Then, after over 60 viscous timescales, the flow gradually transitions to a different steady state. The transition corresponds to a change in the equatorial symmetry of the flow. After the transition (Figure 12(c)), the zonal velocity dominates and is now largely equatorially antisymmetric (EAS). Whilst the non-axisymmetric flow remains mainly ES after the transition, its amplitude is divided by a factor two (so is then only a factor of two larger than the non-axisymmetric EAS component). Note that in all the plots showing global averages (ReRe, Re0Re_0 and NucNu_c below), we use the values averaged before the transition for simplicity. The output quantities calculated after the transition to the EAS mode are given in Table A.2.

Figure 12. (a) Time series of the volumetric kinetic energy for N2/Ω2=0.06N^2/\Omega^2 = 0.06 and Ek=10−4E_k = 10^{-4}. Sym/anti-sym refers to the equatorial symmetry. (b)-(c) snapshots of the radial velocity at r=1.8r = 1.8 and meridional slice of the zonal velocity in the initial steady state, shortly after the beginning of the simulation and in the final state.

The transition to the EAS mode was previously observed in the simulations of fingering convection in a full sphere geometry of Monville et al. [41]. Monville et al. reported that this transition occurs for Rayleigh numbers above the linear onset for the EAS axisymmetric mode, which suggests that this mode grows linearly. This mode corresponds to hemispherical convection with EAS axisymmetric composition and temperature perturbations, where the northern hemisphere is enriched in light elements and colder than the southern hemisphere. This is the only instance where the EAS axisymmetric mode becomes dominant in our simulations, even after running simulations for multiple tens of viscous timescales. For smaller stratification, the mode decays; for larger stratification, the mode saturates at a smaller amplitude than the ES axisymmetric mode. This simulation shows that the EAS axisymmetric mode can occur in spherical shells, even for relatively large aspect ratio (here ri/ro=0.6r_i/r_o = 0.6). As argued in Monville et al. [41], simulations spanning many viscous times are required to observe this transition.

Intermediate regime

Banded structures

We now explore the intermediate regime where N2/Ω2≈1N^2/\Omega^2 \approx1. Over timescales of the order of a viscous timescale, we observe the formation and saturation of bands of poloidal flows as shown in the spherical projection of uru_r in Figure 13. These bands are persistent and only form inside the tangent cylinder. While in some cases the bands are spiraling (corresponding predominantly to an azimuthal order m=1m = 1, Figure 13(a)), in other cases the bands are predominantly axisymmetric (corresponding to m=0m = 0, Figure 13(b)). The bands have a narrow latitudinal extent compared with their azimuthal extent. The meridional slices of Figure 13 show the axial extent of the bands. The bands are essentially concentric cylindrical structures aligned with the rotation axis. The radial flows in the equatorial regions are comparatively weak, which might be due to the presence of strong zonal flows that can disrupt convection in this region. Indeed, the amplitude of the zonal flow is maximal in the intermediate regime, exceeding the amplitude of the radial flows by one order of magnitude (see section § 5.5).

Snapshots of the radial velocity for simulations in the intermediate regime

Figure 13. Snapshots of the radial velocity for simulations in the intermediate regime: (left) Aitoff projections at r=1.8r = 1.8; (right) meridional slice at a given longitude. For (b), the meridional slice shows the axisymmetric component.

Figure 14 shows the times series of the kinetic energy and the power spectra of the poloidal kinetic energy at different times of the simulation at N2/Ω2=1.24N^2/\Omega^2 = 1.24 of Series 2. The fingers initially develop at a size close to the Stern

Figure 14. (a) Times series of the volumetric kinetic energy for N2/Ω2=1.24N^{2}/\Omega^{2} = 1.24 of Series 2 decomposed into poloidal/toroidal components and axisymmetric/non-axisymmetric components. (b)-(c) Spectra of the poloidal kinetic energy as a function of the spherical harmonics degree ll and order mm at different times of the simulation. The spectra are averaged in radius excluding the boundary layers.

scale (around l≈m≈80l \approx m \approx80), but the energy then builds up at small mm, with a dominant m=0m = 0, and around about l≈40l \approx40. Meanwhile the energy at the Stern scale gradually decreases before reaching saturation at a lower energy level than the bands. The fingers might be losing energy to the benefit of the bands, either due to a decrease of the mean composition gradient in the bulk or due to non-linear energy transfers, or because of the progressive increase of the zonal flow, which finally reaches saturation after t>5t > 5. At saturation (t=4.6t = 4.6), the bands carry a similar amount of kinetic energy to the initial fingers at the Stern scale (at t=0.3t = 0.3). Their associated local Reynolds number is thus greater (as their latitudinal scale are larger), but remains below unity.

Figure 15 shows a space-time diagram of the axisymmetric radial velocity at a fixed radius close to the inner boundary (r=1.6r = 1.6), where the horizontal axis corresponds to co-latitude. This representation allows us to track the temporal evolution of the axisymmetric bands in the saturated regime and reveals their latitudinal drift. The time period corresponds to approximately one viscous timescale (one unit of dimensionless time). For latitudes smaller than ±70∘\pm70^\circ, the bands slowly drift latitudinally over time towards the equatorial region, eventually disappearing on the tangent cylinder. New drifting bands nucleate over time around latitudes of ±70∘\pm70^\circ. At higher latitudes, weak meandering bands can be seen and appear to be drifting towards the poles. The latitudinal drift is not attributable to meridional circulation, as the meridional velocity is very weak within the tangent cylinder and would not produce the observed drift direction, making this mechanism unlikely. While advection by the zonal flow could account for the drift of the spiraling structures, it does not explain the latitudinal drift observed in the axisymmetric bands in Figure 15.

Space-time diagram of the axisymmetric radial and azimuthal velocities

Figure 15. Space-time diagram of the axisymmetric radial (top) and azimuthal (bottom) velocities at r=1.6r = 1.6 for N2/Ω2=1.24N^2/\Omega^2 = 1.24 of Series 2. The spherical harmonics degree l=1l = 1 has been removed from the azimuthal component to filter out the dominant equatorial prograde zonal flow. The dashed line corresponds to the intersection with the tangent cylinder for this radius.

Azimuthal velocities are associated with the presence of the bands of poloidal flows. Similarly, they show a spiraling or axisymmetric pattern. They can be seen inside the tangent cylinder in Figure 17(c) in § 5.5 with a smaller amplitude than the equatorial prograde zonal. To observe their temporal evolution, we filter out the spherical harmonics degree l=1l = 1 in the post-processed data, which is the main component of the equatorial zonal flow. The space-time diagram of the filtered axisymmetric azimuthal velocity is shown in Figure 15 and reveals that the latitudinal drift is also seen in the banded azimuthal flow.

The bands have a narrow range of existence, only appearing for parameters around N2/Ω2≈1N^2/\Omega^2 \approx1 (see Figure 1). Varying the buoyancy frequency by more than one order of magnitude above or below this threshold value leads to a transition to radial or columnar fingers respectively. The bands are robust features in the sense that they occur for N2/Ω2≈1N^{2}/\Omega^{2}\approx1 irrespective of EkEk or RacRa_c (i.e. they appear in both Series 1 and 2 and in additional simulations shown in Figure 1), at least for the density ratio used here.

Origin of the bands

We conclude this section by studying the origin of the bands. In particular, we want to understand the role of the non-linear interactions in the formation of the bands. First, we compare the relative importance of the forces in the Navier-Stokes equation at different scales. This is most easily done by comparing the sources of the vorticity ω=∇×u\boldsymbol{\omega}=\nabla\times\mathbf{u}. Since the bands correspond essentially to axisymmetric or m=1m=1 poloidal flows, we analyse the relative strengths of the terms in the equation for the ϕ\phi-component of the vorticity:

∂ωϕ∂t+[(u⋅∇)ω]ϕ−[(ω⋅∇)u]ϕ−2Ek∂uϕ∂z=[∇2ω]ϕ−1ro(∂Θ∂θ+∂ξ∂θ).(24)\frac{\partial\omega_{\phi}}{\partial t}+\left[(\mathbf{u}\cdot\nabla)\boldsymbol{\omega}\right]_{\phi}-\left[(\boldsymbol{\omega}\cdot\nabla)\mathbf{u}\right]_{\phi}-\frac{2}{Ek}\frac{\partial u_{\phi}}{\partial z}=\left[\nabla^{2}\boldsymbol{\omega}\right]_{\phi}-\frac{1}{r_o}\left(\frac{\partial\Theta}{\partial\theta}+\frac{\partial\xi}{\partial\theta}\right). \tag*{(24)}

For a given snapshot, we calculate the power spectra of each term as a function of the spherical harmonics degree ll or order mm. The goal here is to determine whether the force balance differs for the bands and the fingers. Figure 16 shows the power spectra for the case of Figure 14. The spectra are averaged in radius between 1.6 and 2, where the bands have the strongest amplitude, and calculated at a time where the system is in a statistically steady-state. For spherical harmonics degrees and orders smaller than l≲100l\lesssim100 and m≲30m\lesssim30, the dominant vorticity sources are the thermal and compositional buoyancy terms (last term on the right-hand side (RHS) of equation (24)). They partially cancel each other as shown by the lower value of the total buoyancy for these scales. The thermal diffusion wipes out smaller scales and the thermal buoyancy drops out more rapidly than the compositional buoyancy in the spectral tail. At the scale of the bands (m=0,1m=0,1 and l≈40l\approx40), the main balance for the generation of the azimuthal vorticity is between the buoyancy term and the Coriolis term (last term on the left-hand side (LHS) of (24)). At the finger scale (l≈80l\approx80), the viscous term (first term on the RHS) also enters this balance. The non-linear inertial terms (2nd and 3rd terms in the LHS) are significantly smaller than buoyancy, viscous and Coriolis terms at all scales, except for the advection by the zonal velocity which is dominant at the finger scale. Using snapshots at earlier times during the growth of the bands gives the same picture, suggesting again that inertia is unimportant for the formation of the bands.

Figure 16. Spectral representation of each terms in the equation for ϕ\phi-component of the vorticity (equation (24)) for the case N2/Ω2=1.24N^{2}/\Omega^{2}=1.24 of Series 2. The spectra are averaged in radius between 1.6 and 2. IcI_c corresponds to the Coriolis term; IbtI_{bt} and IbcI_{bc} to the thermal and compositional buoyancy terms; IvI_v to the viscous term; Iadv0I_{\mathrm{adv}}^{0} and Iadv−Iadv0I_{\mathrm{adv}}-I_{\mathrm{adv}}^{0} to the advection by the zonal and non-zonal velocity; IstrI_{\mathrm{str}} to vortex stretching. The total buoyancy Ib=Ibt+IbcI_b=I_{bt}+I_{bc} and the residual Ic+Ib+IvI_c+I_b+I_v are calculated in the physical space before calculating the spectral decomposition.

To confirm this, we performed a linear simulation for N2/Ω2=1.24N^2/\Omega^2 = 1.24 of Series 2, where the non-linear terms (u⋅∇)u(\mathbf{u} \cdot\nabla)\mathbf{u}, u⋅∇Θ\mathbf{u} \cdot\nabla\Theta and u⋅∇ξ\mathbf{u} \cdot\nabla\xi are switched off in equations (4)-(6) in XSHELLS (i.e. calculating the growth or decay of perturbations from the static state): the mode m=0m = 0 grows and forms banded structures inside the tangent cylinder similar to those described in the non-linear calculations. We therefore conclude that the bands corresponds to a linear mode of fingering convection, which is preferred when N2/Ω2≈1N^2/\Omega^2 \approx1 for this density ratio. In the linear calculation, the bands do not drift in the latitudinal direction, suggesting that the drift observed in the full non-linear simulations arises from a non-linear mechanism.

Zonal Flow

Formation of the zonal flow

Fingering convection produces persistent zonal flows in the presence of rotation. They are equatorially symmetric (except for the case N2/Ω2=0.06N^2/\Omega^2 = 0.06 reported in §3.3.2) and their speed can exceed the amplitude of the radial velocities. Guervilly [23] showed that the zonal flow is produced by a linear process and is in a thermo-compositional wind balance, where the latitudinal variations of the axisymmetric density perturbations are balanced by axial variations of the zonal velocity. The thermo-compositional wind balance is derived from the equation for ωϕ\omega_\phi (equation (24)) in an axisymmetric, steady state and neglecting the non-linear and viscous terms,

2Ek∂uϕ‾∂z≈1ro(∂Θ‾∂θ+∂ξ‾∂θ),(25)\frac{2}{Ek}\frac{\partial\overline{u_\phi}}{\partial z} \approx\frac{1}{r_o}\left(\frac{\partial\overline{\Theta}}{\partial\theta}+\frac{\partial\overline{\xi}}{\partial\theta}\right), \tag*{(25)}

where the overline denotes an azimuthal average. For rapidly-rotating fingering convection, the latitudinal difference in the mixing of composition and temperature is a source of baroclinicity, driving a meridional circulation that is deflected into zz-dependent zonal flows by the Coriolis force. Composition and temperature gradients oppose each other, but the slower diffusion of the composition leads to sharper composition gradients, which dictate the direction of the meridional circulation and zonal flows.

Figure 17 shows meridional slices of the zonal velocity and the axisymmetric composition perturbation for increasing N2/Ω2N^2/\Omega^2. For the smallest N2/Ω2N^2/\Omega^2, the zonal flow is retrograde in the equatorial region and is associated with a compositional mixing that solely takes place outside the tangent cylinder, as the polar regions are not convecting (Figure 11(a)). As fingering convection onsets in the polar regions at larger N2/Ω2N^2/\Omega^2 (Figure 11(b)), the mixing becomes more efficient at high than low latitudes and the latitudinal compositional gradient is reversed. This leads to a reversal in the direction of the zonal flow, which is strongest outside the tangent cylinder. Further increase of N2/Ω2N^2/\Omega^2 up to approximately 10 leads to an amplification of the latitudinal dependence in the mixing and an intensification of the prograde zonal flow. As N2/Ω2N^2/\Omega^2 is increased further in the weak-rotation regime, the mixing becomes increasingly homogeneous with latitude, leading a gradual weakening of the zonal flow. At our largest stratification (N2/Ω2=62240N^2/\Omega^2 = 62240), the zonal flow is significantly weaker than the radial flows; we expect that it would eventually completely disappear as N2/Ω2→∞N^2/\Omega^2 \to\infty.

Figure 17. Meridional slices of the zonal velocity (top row) and the axisymmetric composition perturbation (bottom row) for simulations in Series 2 (N2/Ω2∈[0.12,25]N^{2}/\Omega^{2} \in[0.12, 25]) and Series 1 (N2/Ω2=62240N^{2}/\Omega^{2} = 62240). The composition is normalised by the basic state composition CsC_{\mathrm{s}} at r=rir = r_{\mathrm{i}}.

Scaling of the zonal velocity

Figure 18(a) shows the variation of the Reynolds number based on the zonal flow amplitude, Re0=V0D/νRe_0 = V_0D/\nu, where V0V_0 is the mean zonal velocity calculated from the volume and time average of the axisymmetric toroidal kinetic energy. Re0Re_0 increases until N2/Ω2≈10N^2/\Omega^2 \approx10 for both Series, and then decreases for larger values. At its maximum, Re0Re_0 reaches a value approximately 10 -- 30 times larger than ReRe. This suggests that large values of the zonal flow can be reached at low Ekman numbers.

Figure 18. Variation of (a) the zonal Reynolds number and (b) the zonal Rossby number as a function of N2/Ω2N^{2}/\Omega^{2}. The two solutions of the bistable case at N2/Ω2=0.2N^{2}/\Omega^{2} = 0.2 are linked by a dashed line in (a). The dashed line in (b) corresponds to Ro0=Re0Ek=10−2N2/Ω2\mathrm{Ro}_{0} = \mathrm{Re}_{0} \mathrm{Ek} = 10^{-2}N^{2}/\Omega^{2}, while the dotted line corresponds to Ro0=0.1(N2/Ω2)−0.45\mathrm{Ro}_{0} = 0.1(N^{2}/\Omega^{2})^{-0.45}, which is discussed in the conclusion section. Data from Monville et al. [41] (M19) and Guervilly [23] (G22) are included for smaller EkEk in (b). In Monville et al. [41], the Rossby number is calculated from the total kinetic energy, while in Guervilly [23] and in the present study, the Rossby number is based on the mean zonal velocity.

Around N2/Ω2=0.2N^2/\Omega^2 = 0.2, a discontinuity is visible in the variation of Re0Re_0. This corresponds to the onset of fingering convection inside the tangent cylinder, which is accompanied by a reversal of the direction of the zonal flow, from retrograde (as in Figure 17(a)) to prograde (Figure 17(b)). The case N2/Ω2=0.2N^2/\Omega^2 = 0.2 of Series 1 exhibits a transition from the initial (weaker) retrograde state to the final (stronger) prograde state after a long integration time of approximately 35 viscous timescales.

In the rapidly-rotating regime, Guervilly [23] obtain an approximate scaling for the zonal flow amplitude by assuming a compositional wind balance, where only the latitudinal gradients of the composition balance with the axial gradient of the zonal velocity on large scales. Since the diffusion of the composition perturbations is slow compared with its advection, the scaling then assumes that the composition perturbations are of the same order of magnitude as

the variation of the static composition over a mixing length. Assuming that the mixing length is of the order of the layer depth, the scaling becomes

Ro0=Re0Ek∼1ri(Riρ−1)N2Ω2.(26)Ro_{0}=Re_{0}Ek\sim\frac{1}{r_{i}(R_{i}^{\rho}-1)}\frac{N^{2}}{\Omega^{2}}. \tag*{(26)}

Figure 18(b) shows that the data (including the data from Monville et al. [41] and Guervilly [23]) follow reasonably well the linear dependence on N2/Ω2N^{2}/\Omega^{2} predicted by the theoretical scaling for N2/Ω2≲10N^{2}/\Omega^{2}\lesssim10. We determine empirically that Ro0=10−2N2/Ω2Ro_{0}=10^{-2}N^{2}/\Omega^{2}, meaning that the maximum value for Ro0Ro_{0} is expected to be 0.1 (for N2/Ω2=10N^{2}/\Omega^{2}=10) irrespective of the Ekman number. This scaling also provides a good fit to the data of Monville et al. [41], suggesting that mean flows driven by fingering convection in a rotating sphere or spherical shell saturate at similar amplitudes despite differences in model configuration, including boundary conditions and the radial dependence of the stratification. For N2/Ω2≥100N^{2}/\Omega^{2}\geq100, the Rossby number becomes a less meaningful measure of zonal flow amplitude as Ek→1Ek\to1 in Series 1. Simulations at higher Rayleigh numbers and constant Ekman numbers, analogous to Series 2 but currently beyond computational reach, would likely reveal a decreasing trend of Ro0Ro_{0} with increasing N2N^{2}.

Compositional transport

In this last result section, we study the mixing of the composition throughout the layer. We do not consider the convective heat transport since it is small compared with the conductive heat flux in fingering convection at low Prandtl numbers (Garaud, [16]; Guervilly, [23]; Tassin et al., [63]). We first look at the radial distribution of the convective flux, then its latitudinal dependence in the different regimes, and, finally, the global efficiency measured by a Nusselt number.

Radial and latitudinal dependence

The averaged convective and conductive fluxes of composition over a spherical surface S(r)S(r) are defined respectively as

F‾cv=Sc∫S(r)urξ dS,F‾cd=−∫S(r)∂ξ∂r dS+Fs(r)(27)\overline{F}_{cv}=Sc\int_{S(r)}u_{r}\xi\,\mathrm{d}S,\qquad\overline{F}_{cd}=-\int_{S(r)}\frac{\partial\xi}{\partial r}\,\mathrm{d}S+F_{s}(r) \tag*{(27)}

where the double overbar denotes a spherical average and the static flux is Fs(r)=−Cs′F_{s}(r)=-C_{s}'.

Figure 19(a)-(Figure 19c) shows the radial profiles of F‾cv\overline{F}_{cv} and F‾cd\overline{F}_{cd} for the three different dynamical regimes for selected cases in Series 1. The sum of the convective and conductive fluxes equal the static flux at all radii, as expected from a steady state. For N2/Ω2≫1N^{2}/\Omega^{2}\gg1, the amplitude of the convective flux is comparable to the conductive flux in the lower domain of the layer where fingering convection is most vigorous. The conductive fluxes in the polar regions (F‾cd(θ=0)\overline{F}_{cd}(\theta=0), where the single overbar denotes an azimuthal average) and in the equatorial region (F‾cd(θ=π/2)\overline{F}_{cd}(\theta=\pi/2)) are similar, thereby confirming that there is no notable latitudinal difference when the rotational effects are weak.

Six-panel radial profiles of composition fluxes and buoyancy frequencies

Figure 19. (a)-(c) Radial profiles of the convective (FcvF_{cv}) and conductive (FcdF_{cd}) fluxes of composition for (a) the weakly-rotating, (b) the intermediate, and (c) the rapidly-rotating regimes from Series 1. The static flux in the absence of motion, FsF_s, is shown for comparison. Azimuthal averages are denoted by a single overbar and spherical averages by a double overbar. (d)-(f) Radial profiles of the buoyancy frequency (normalised by Ω2\Omega^2) based on the static gradients (solid lines) for the density (N2N^2), temperature (Nt2N_t^2) and composition (Nc2N_c^2) and based on the total gradients (dashed lines) for the density (Ntot2N_{\mathrm{tot}}^2), temperature (Ntot,t2N_{\mathrm{tot},t}^2) and composition (Ntot,c2N_{\mathrm{tot},c}^2). The profiles have been calculated from snapshots, but are representative of the statistically steady state.

For N2/Ω2≈1N^{2}/\Omega^{2}\approx1, a reduction in convective flux is observed. In the polar regions, where the bands are present, the conductive flux slightly diminishes at mid-radius compared with other latitudes, indicating that the bands enhance compositional mixing somewhat more effectively than the sheared fingers in the equatorial region.

For N2/Ω2≪1N^{2}/\Omega^{2}\ll1, the convective flux is further reduced and significantly smaller than the conductive flux throughout the layer. A large disparity between high and low latitudes is observed: only the region outside the tangent cylinder is convecting, so the conducting flux in the polar regions is superposed onto the static flux.

To assess how the stratification is modified by fingering convection in the statistically steady state, Figure 19(d)-(Figure 19f) shows radial profiles of the buoyancy frequency computed from the total gradients (i.e. including both the perturbations and the static background) for density, temperature, and composition. These profiles are compared with the corresponding static profiles for the same three cases as Figure 19(a)-(Figure 19c). The stabilising temperature stratification remains largely unchanged relative to the static state in all cases. However, the reduction of the destabilising compositional gradient leads to an increase in the overall stratification. This enhancement is most pronounced near the bottom of the layer, where it reaches up to 20% in the weakly rotating case.

To study systematically the latitudinal dependence of the transport efficiency as a function of N2/Ω2N^{2}/\Omega^{2}, we measure the normalised composition difference across the layer at a given latitude

ΔC(θ)=C‾(ro,θ)−C‾(ri,θ)Cs(ro)−Cs(ri),(28)\Delta C(\theta)=\frac{\overline{C}(r_{o},\theta)-\overline{C}(r_{i},\theta)}{C_{s}(r_{o})-C_{s}(r_{i})}, \tag*{(28)}

where, as previously, the overbar denotes an azimuthal average. Because fixed compositional fluxes are imposed at both the inner and outer boundaries, we expect ΔC\Delta C to decrease as the mixing efficiency increases. Figure 20(a) shows the values of ΔC\Delta C averaged over a ±5∘\pm5^\circ angle around three latitudes for Series 1: the north pole (θ=0\theta= 0), the equator (θ=π/2\theta= \pi/2), and the latitude where the tangent cylinder intersects the outer boundary (θ=0.93 rad\theta= 0.93\ \mathrm{rad}).

Variation of composition difference and compositional Nusselt number

Figure 20. (a) Variation of ΔC\Delta C at different latitudes as a function of N2/Ω2N^2/\Omega^2 for Series 1. (b) Variation of Nuc−1Nu_c - 1 as a function of N2/Ω2N^2/\Omega^2. The two solutions of the bistable case at N2/Ω2=0.2N^2/\Omega^2 = 0.2 are linked by a dotted line.

For N2/Ω2≪1N^2/\Omega^2 \ll1, as expected, ΔC=1\Delta C = 1 at the poles since there is initially no convection inside the tangent cylinder and transport occurs only in the equatorial region. The behaviour becomes notably non-monotonic in the rapidly-rotating regime at low latitudes, with a distinct discontinuity at N2/Ω2=0.2N^2/\Omega^2 = 0.2, coinciding with the onset of convection in the polar regions. This discontinuity is associated with the reversal of the zonal flow and a pronounced increase in its amplitude (Figure 18(b)), indicating an intensification of radial shear. Given the slight reduction in the poloidal Reynolds number following this transition (see Table A.1), the observed increase of ΔC\Delta C at low latitudes is likely a consequence of the enhanced shear, which may act to suppress convective transport.

For weak rotation (N2/Ω2>100N^2/\Omega^2 > 100), ΔC\Delta C attains similar values across latitudes, although the poles tend to exhibit slightly smaller values. This homogenisation of mixing is consistent with the gradual disappearance of zonal flows in this regime.

In the intermediate regime (1≤N2/Ω2≤101 \le N^2/\Omega^2 \le10), ΔC\Delta C at high latitudes is slightly smaller than at the equator, confirming moderately enhanced mixing associated with the bands inside the tangent cylinder.

Nusselt number

The efficiency of convective transport is commonly quantified by a global measure, the Nusselt number, which compares the total heat flux to the conductive heat flux at rest. In systems with fixed-flux boundary conditions, the Nusselt number can be determined from the spherically-averaged temperature difference across the layer [42, 41]. Thus, by analogy, we define the “compositional” Nusselt number (sometimes called the Sherwood number) as

Nuc=Cs(ro)−Cs(ri)C‾(ro)−C‾(ri).(29)Nu_c = \frac{C_s(r_o)-C_s(r_i)}{\overline{C}(r_o)-\overline{C}(r_i)}. \tag*{(29)}

It is inversely related to the previously defined ΔC\Delta C, but provides a global measure across the entire layer, allowing its variation with control parameters to be compared with scaling predictions from the literature.

A power-law scaling of the form Nuc−1=ARac1/3Nu_c - 1 = A Ra_c^{1/3} (where the constant AA depends on PrPr, LeLe, and RρR_\rho) has long been proposed based on dimensional arguments [67]. This scaling has been observed in non-rotating, plane-layer experiments and numerical simulations for Pr>1Pr > 1 [e.g. 49, 24]. Extending the Rayleigh–Bénard convection theory of Grossmann and Lohse [22], Yang et al. [70] demonstrated that the 1/31/3 law is expected in the asymptotic regime of large RacRa_c for fingering convection. In contrast, for small PrPr and in a triply periodic (unbounded) domain, Brown et al. [7] proposed an alternative scaling in which NucNu_c remains constant for fixed values of PrPr, LeLe, and RρR_\rho within the regime rρ≪(1/Le,Pr)≪1r_\rho\ll(1/Le,Pr) \ll1 (noting that, although this regime is the closest for our simulations where Pr=0.3Pr = 0.3, Le=10Le = 10 and Rρ=10R_\rho= 10, it is not strictly satisfied). In the absence of rotation, Tassin et al. [63] reported a best-fit scaling of Nuc−1∼Rac0.27Nu_c - 1 \sim Ra_c^{0.27} from numerical simulations spanning a broad range of Prandtl numbers (Pr∈[0.03,7]Pr \in[0.03,7]). They attributed the deviation from the 1/31/3 law to the non-negligible contribution of dissipation within the boundary layers, while the 1/31/3 law in Grossmann and Lohse [22] assumes bulk-dominated dissipation. In the rapidly-rotating regime with N2/Ω2<1N^2/\Omega^2 < 1, Guervilly [23] found that the 1/31/3 scaling remains a reasonable fit for constant density ratio, albeit with an additional dependence on the Ekman number.

We now examine how our simulations, which span a broad range of N2/Ω2N^2/\Omega^2 at fixed density ratio, compare with the scaling behaviours discussed above. Figure 20(b) shows the compositional Nusselt number for both Series as a function of N2/Ω2N^2/\Omega^2. Across all cases, the convective enhancement remains modest, Nuc−1<1Nu_c - 1 < 1, indicating that convective fluxes are relatively weak. In the rapidly-rotating regime, the behaviour of NucNu_c is non-monotonic, mirroring the behaviour of ΔC\Delta C with a discontinuity at N2/Ω2=0.2N^2/\Omega^2 = 0.2. The regional behaviours inside and outside the tangent cylinder act in opposition: the onset of fingering convection inside the tangent cylinder enhances convective transport, whereas the intensified shear in the equatorial region following the zonal flow reversal suppresses it. Overall, this results in a slight decrease in NucNu_c at the transition.

For the range 0.2<N2/Ω2<10.2 < N^2/\Omega^2 < 1, NucNu_c remains approximately constant in Series 1, in line with the plateau observed in the poloidal Reynolds number over the same interval. In the intermediate regime, 1≤N2/Ω2≤101 \le N^2/\Omega^2 \le10, NucNu_c begins to increase, mirroring the trend observed in Series 2. As anticipated for N2/Ω2>100N^2/\Omega^2 > 100, NucNu_c asymptotes to a plateau, indicating the transition to the Ek→1Ek \to1 dynamics in Series 1. To facilitate comparison with previous studies, we fit a power-law scaling of the form Nuc−1∼(N2/Ω2)αNu_c - 1 \sim(N^2/\Omega^2)^\alpha over the range 1<N2/Ω2<301 < N^2/\Omega^2 < 30, despite noticeable scatter in the data. The best fit across both Series yields an exponent α≈0.27\alpha\approx0.27, closely matching the non-rotating scaling reported by Tassin et al. [63]. In the case of rotating fingering convection, the radial shear of the fingers due to the presence of the zonal flows indirectly modifies the transport efficiency and the intensification of shear with increasing RacRa_c (at least up to the peak in zonal flow amplitude at N2/Ω2≈7.5N^2/\Omega^2 \approx7.5) may account for the observed shallower scaling relative to the classical 1/3 law.

Conclusions

Summary of the results

We have conducted numerical simulations of fingering convection in a thick spherical shell under the influence of rotation. Our study examined how the dynamics change as a function of the stratification strength NN measured relative to the rotation rate Ω\Omega, by varying the Ekman number at fixed Rayleigh numbers (Series 1) or by varying the Rayleigh numbers at fixed Ekman number (Series 2). Throughout, the density ratio was fixed at Rρ=3.33R_\rho= 3.33 at the inner radius, as were the diffusivity ratios (Pr=0.3Pr = 0.3 and Le=10Le = 10). We found that the dynamics depend strongly on the parameter N2/Ω2N^2/\Omega^2, with three distinct regimes emerging: a weakly-rotating regime for N2/Ω2>10N^2/\Omega^2 > 10, a rapidly-rotating regime for N2/Ω2<1N^2/\Omega^2 < 1, and an intermediate regime for 1<N2/Ω2<101 < N^2/\Omega^2 < 10.

In all regimes, the primary fingering mode forms narrow, elongated fingers that align with the rotation axis when N2/Ω2<10N^2/\Omega^2 < 10, and with gravity when N2/Ω2>10N^2/\Omega^2 > 10. The fingers remain laminar, with local Reynolds numbers below unity. Their transverse scale L\mathcal{L} approximately follows the Stern scale Ltheo=(∣Rat∣/ro)−1/4\mathcal{L}_{\mathrm{theo}} = (|Ra_t|/r_o)^{-1/4} and, for our fixed density ratio, is well described by L=5Ltheo\mathcal{L} = 5\mathcal{L}_{\mathrm{theo}}, independent of the Ekman number. For Mercury’s core with a stable layer of thickness 800 km, the predicted transverse finger scales is approximately 50 – 0.5 m for N2/Ω2=10−4−104N^2/\Omega^2 = 10^{-4} - 10^4 for the moderate density ratio considered here. In non-rotating fingering convection simulations, Tassin et al. [63] found that the finger size depends only weakly on the density ratio. Consequently, primary fingers in planetary cores are expected to develop only on small spatial scales, on the order of 1 m. Note that an exception may occur close to the stability boundaries of fingering convection, especially when the stratification is extremely weak at low Rayleigh numbers. In this regime, columnar flows can develop with an azimuthal extent comparable to the layer thickness, as described by Monville et al. [41]; Mather and Smithev [39]; Guervilly [23]. In the limit Ek≪1Ek \ll1, the lower stability boundary occurs at Rac≈100Ek−1Ra_c \approx100Ek^{-1} (Monville et al. [41]), corresponding to N2/Ω2≈10−11N^2/\Omega^2 \approx10^{-11} for Mercury.

Beyond the primary fingering mode, additional large-scale flows emerge depending on the regime. In the range 1<N2/Ω2<1001 < N^2/\Omega^2 < 100 and at sufficiently small Ekman numbers, we observe clustering of fingers in the upper domain, where the local density ratio is larger and fingering convection less vigorous. These clusters are associated with weak, large-scale density anomalies and surrounded by toroidal gyres driven by the Coriolis force. These toroidal gyres dominate the local dynamics. They differ from the toroidal jets reported by Tassin et al. [63], which flow along concentric shells about an arbitrary rotation axis. The size of the density anomalies and associated gyres does not appear to vary significantly with the control parameters, although this cannot be established with certainty from our simulations given that they occur over a limited range of RacRa_c and EkEk. Their typical azimuthal order is m≈5−10m \approx5 - 10. If their size is indeed insensitive to these parameters, we may infer that such structures would have characteristic horizontal scales of approximately 500 – 1000 km under Mercury’s core conditions.

In the intermediate regime (1<N2/Ω2<101 < N^2/\Omega^2 < 10), fingering convection exhibits a novel feature: axisymmetric or spiraling bands of poloidal flow confined within the tangent cylinder. These bands form concentric cylindrical structures that are nearly invariant along the rotation axis and slowly drift equatorward. They are highly anisotropic, with latitudinal widths greater than the Stern scale (roughly twice the Stern scale, though their domain of existence is too narrow to confirm this reliably) but azimuthal extents far larger. They grow and saturate over viscous timescales, gradually superseding primary fingering, and contribute to compositional mixing slightly more efficiently than the surviving columnar fingers in the equatorial region. The bands are robustly found within 1≲N2/Ω2≲101 \lesssim N^2/\Omega^2 \lesssim10 but do not persist outside this range.

In the rapidly-rotating regime (N2/Ω2<1N^2/\Omega^2 < 1), the mixing is laterally inhomogeneous, initially more vigorous near the equator. This generates latitudinal compositional gradients that are in a thermo-compositional wind balance, analogous to the thermal wind in single-component thermal convection (e.g. Aubert, [2]). The resulting radial shear is disruptive to fingering convection outside the tangent cylinder, weakening the mixing of composition in that region. As N2/Ω2N^{2}/\Omega^{2} increases, enhanced mixing at higher latitudes leads to a reversal of the latitudinal compositional gradient and, in turn, of the direction of the zonal flow, from retrograde to prograde. The zonal flow amplitude, quantified by the zonal Rossby number Ro0\mathrm{Ro}_{0}, scales linearly with N2/Ω2N^{2}/\Omega^{2} and reaches a maximum value of Ro0≈0.1\mathrm{Ro}_{0} \approx0.1 at N2/Ω2=10N^{2}/\Omega^{2}=10. For stronger stratification, the zonal flow gradually diminishes as rotation becomes less influential and the lateral inhomogeneity of the mixing fades.

For rapid rotation at N2/Ω2=0.06N^{2}/\Omega^{2}=0.06, we obtained a case in which an equatorially-antisymmetric and axisymmetric mode is favoured. This mode corresponds to hemispherical convection, where the fluid in the northern hemisphere is colder and richer in light elements than in the southern hemisphere. The associated equatorially-antisymmetric zonal flow (slower in the northern hemisphere) becomes dominant over the usual equatorially-symmetric zonal flow. This behaviour occurs only for this specific parameter set: the mode decays at smaller stratification and becomes subdominant at larger stratification. It closely resembles the linear mode identified by Monville et al. [41] in a full-sphere geometry, indicating that it can also arise in a thick spherical shell.

Interaction with magnetic fields

In most of the explored parameter space, we therefore find that fingering convection generates large-scale flows in addition to the primary small-scale fingers. The resulting large-scale dynamics is remarkably diverse, exhibiting axisymmetric bands of poloidal flows, finger clusters, toroidal gyres, hemispherical convection, and zonal flows. Although these flows generally have small local Reynolds numbers (except for the zonal flows), they may possess interesting dynamo properties or interact with ambient magnetic fields in ways that remain to be elucidated.

Few studies have investigated the interaction between fingering convection and ambient magnetic fields. In a plane-layer geometry and in the absence of rotation, Harrington and Garaud [26] showed that the range of density ratios that are linearly unstable to fingering convection (equation (16)) is unchanged by the presence of a vertical magnetic field. They found that the finger flows become increasingly rigid along the direction of the ambient field and that magnetised fingering convection saturates at higher amplitudes than its hydrodynamic counterpart, resulting in a substantial enhancement of convective transport. For simulations with Prandtl and Lewis numbers comparable to those considered here (Pr=0.1Pr=0.1 and Le=10Le=10), the magnetically dominated regime is reached when the Alfvén velocity, vA=B0/(ρμ0)1/2v_{\mathrm{A}}=B_{0}/(\rho\mu_{0})^{1/2} (where B0B_{0} is the strength of the ambient magnetic field and μ0\mu_{0} the magnetic permeability) becomes comparable to the characteristic finger velocity, vf=κT/Ltheov_{\mathrm{f}}=\kappa_{T}/\mathcal{L}_{\mathrm{theo}}. This corresponds to magnetic field strengths satisfying B0>(κTρμ0)/LtheoB_{0}>(\kappa_{T}\rho\mu_{0})/\mathcal{L}_{\mathrm{theo}}. Applied to metallic planetary cores, this estimate suggests that magnetic fields exceeding ∼10\sim10 nT may be sufficient to significantly enhance fingering convection, a condition met even by Mercury’s relatively weak magnetic field [1]. It should be noted, however, that the simulations of Harrington and Garaud [26] were conducted at magnetic Prandtl numbers of order unity. As a result, this threshold may be overestimated for fluids with low magnetic Prandtl numbers [15].

One property of the small-scale fingers that may be relevant for magnetic interactions is their pronounced radial asymmetry in the regime weakly influenced by rotation (N2/Ω2>10N^{2}/\Omega^{2}>10), characterized by intense, concentrated upflows surrounded by more diffuse downflows. This asymmetry could enable topological magnetic pumping, as first proposed by Drobyshevski and Yuferev [13] in the context of the solar interior. In the solar convection zone, compressible convection manifests as narrow downwelling lanes separating broader, weaker upflows; in this configuration, horizontal magnetic fields are expected to be carried downward by the convection [e.g. 64]. In contrast, the radial asymmetry observed here is reversed, suggesting by analogy that magnetic flux might instead be pumped outward from the stable layer. However, it is unclear whether the degree of radial asymmetry associated with Boussinesq fingering convection would be sufficient for such magnetic flux pumping to operate effectively.

Of perhaps more direct relevance to the role of stable layers in shaping planetary magnetic fields is the spontaneous generation of strong zonal flows, even in the absence of imposed boundary heterogeneity. These zonal flows, and the associated shear, are often invoked as a mechanism to attenuate non-axisymmetric magnetic fields passing through a stable layer, thereby promoting axisymmetrisation of the surface magnetic field [62]. According to Stevenson [62]’s linear analysis, achieving a dipole tilt below one degree at Mercury’s surface requires Ro0≳10−4\mathrm{Ro}_{0}\gtrsim10^{-4} for an 800 km-thick stable layer (see discussion in Guervilly [23]). Our results suggest zonal flows sufficient to induce effective axisymmetrisation over N2/Ω2∈[10−2,25]N^{2}/\Omega^{2}\in[10^{-2},25]. For stronger stratification, the flows may still generate adequate shear, but further simulations (particularly in Series 2 at fixed Ek=10−4Ek=10^{-4}) are needed to confirm this. (Note that in Series 1, where Ek→1Ek\to1, the Rossby number becomes less meaningful for N2/Ω2>100N^{2}/\Omega^{2}>100.) To estimate a possible upper threshold at stronger stratification, we consider the last three data points of Series 2 in Figure 18(b), which exhibit a decreasing trend of Ro0Ro_0 with increasing N2/Ω2N^2/\Omega^2. A least-squares fit over the range N2/Ω2≥7.5N^2/\Omega^2 \ge7.5 yields Ro0=0.1(N2/Ω2)−0.45Ro_0 = 0.1(N^2/\Omega^2)^{-0.45}. Extrapolating this scaling suggests that Ro0Ro_0 would decrease to ∼10−4\sim10^{-4} at N2/Ω2≈5×106N^2/\Omega^2 \approx5 \times10^6. While highly speculative, this estimate provides a tentative indication that zonal flows driven by fingering convection could provide adequate shear over a wide range of stratification strength.

Ultimately, nonlinear magnetohydrodynamic simulations that include the feedback of the Lorentz force are required to properly test the interaction of rotating fingering convection with magnetic fields.

Limitations of the model

Finally, our results demonstrate that simulations extending over long integration times are essential to capture the rich dynamics of fingering convection: the equatorially-antisymmetric mode observed at N2/Ω2=0.06N^2/\Omega^2 = 0.06 grows and saturates only very slowly, becoming dominant after roughly 60 viscous timescales; similarly, the reversal of zonal flow direction at N2/Ω2=0.2N^2/\Omega^2 = 0.2 occurs only after several tens of viscous timescales; the bands of poloidal flow in the intermediate regime also require several viscous timescales to saturate. As a result, global simulations of fingering convection are computationally demanding, owing to the wide range of spatial and temporal scales involved. This limits our ability to approach realistic core conditions, particularly at low Ekman numbers. For example, in our exploration of the regime N2/Ω2>1N^2/\Omega^2 > 1, we were restricted to Ekman numbers Ek≥10−4Ek \ge10^{-4}, whereas in Mercury’s core Ek≈10−12Ek \approx10^{-12}. It is therefore difficult to determine whether the slow dynamics observed over viscous timescales in our simulations would remain relevant under planetary core conditions, where the characteristic viscous timescales at large scales exceed several billion years. These processes may plausibly evolve on shorter timescales in realistic core regimes, but confirming this would require computationally intensive simulations at lower Ekman numbers.

In this study the Lewis number was fixed at Le=10Le = 10 due to computational constraints, whereas values in metallic planetary cores are expected to be closer to Le∼103Le \sim10^3. On small scales, the use of a moderately large Lewis number is not expected to alter fundamentally the physics of fingering convection provided that Le≫1Le \gg1 ([61, 46]). However, the larger disparity between thermal and compositional diffusivities leads to enhanced vertical transport of composition, as reflected in the scaling laws for NucNu_c derived by Traxler et al. [65]; Brown et al. [7] at Pr<1Pr < 1 and Yang [69] at Pr>1Pr > 1. On larger scales, the Lewis number also influences the emergence of thermohaline staircases: increasing LeLe promotes the γ\gamma-instability by shifting the minimum of the flux ratio γ\gamma toward larger density ratios ([53, 45]). Although this instability is expected to be inactive at low Prandtl number ([65]), staircase formation may still arise through alternative mechanisms ([18]), which could themselves be sensitive to LeLe. As a first step toward exploring more realistic parameter regimes, we examined how increasing LeLe affects the poloidal flow bands identified in the intermediate regime, which constitute a novel feature of this work. We performed an additional simulation with Le=100Le = 100 (with Pr=0.1Pr = 0.1 and Sc=10Sc = 10) at Ek=10−4Ek = 10^{-4} and N2/Ω2=0.62N^2/\Omega^2 = 0.62, adjusting the Rayleigh numbers to maintain the density ratio at Riρ=3.33Ri_\rho= 3.33. We find that the poloidal bands persist under these conditions, indicating that this feature is robust to at least moderately larger LeLe. Simulations at high Lewis numbers will be required to determine whether these results remain valid under planetary-relevant parameter regimes.

Lastly, our study neglects any dynamical interaction with the deeper convective core. We plan to extend this work by modelling the coupling between convective and stable layers in the future.

Acknowledgements

This work was supported by the UK Science and Technology Facilities Council [ST/W001039/1; UKRI[1201]. This research used the Rocket High Performance Computing service at Newcastle University and the DiRAC Data Intensive service (CSD3) at the University of Cambridge, managed by the University of Cambridge University Information Services on behalf of the STFC DiRAC HPC Facility (www.dirac.ac.uk). The DiRAC component of CSD3 at Cambridge was funded by BEIS, UKRI and STFC capital funding and STFC operations grants. DiRAC is part of the UKRI Digital Research Infrastructure. For the purpose of open access, the authors have applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising from this submission. The authors thank the two referees for suggestions that have improved the manuscript.

Data statement

Data sets for this research are available on the Figshare powered Newcastle University research data repository (https://data.ncl.ac.uk) [20]. The numerical code (XSHELLS) used for this research is openly available at https://nschaeff.bitbucket.io/xshells/.

List of simulations

EkEkN2/Ω2N^2/\Omega^2RocRo_c(Nr,Lmax⁡,Mmax⁡)(N_r,L_{\max},M_{\max})ReReRe0Re_0NucNu_cΔtavg\Delta t_{\mathrm{avg}}
5×10−55 \times10^{-5}0.0160.13(192, 120, 110)2.01.81.0985
1×10−41 \times10^{-4}†0.0620.26(192, 120, 110)2.53.71.1085
1.2×10−41.2 \times10^{-4}0.0900.31(192, 120, 110)2.73.81.12810
1.4×10−41.4 \times10^{-4}0.120.36(192, 120, 110)2.83.61.15915
1.5×10−41.5 \times10^{-4}0.140.39(192, 120, 110)3.03.21.17412
1.6×10−41.6 \times10^{-4}0.160.41(192, 120, 110)3.32.61.19012
1.8×10−41.8 \times10^{-4}*0.200.47(192, 120, 110)3.52.11.20710
1.8×10−41.8 \times10^{-4}**0.200.47(192, 120, 110)3.36.81.15912
2×10−42 \times10^{-4}0.250.52(192, 120, 110)3.08.91.16415
2.5×10−42.5 \times10^{-4}0.390.65(192, 120, 110)3.413.51.17010
3×10−43 \times10^{-4}0.560.78(192, 120, 110)3.318.31.16710
3.33×10−43.33 \times10^{-4}0.690.86(192, 120, 110)2.921.51.1625
4×10−44 \times10^{-4}1.01.0(192, 120, 110)3.127.21.1625
5×10−45 \times10^{-4}1.61.3(192, 120, 110)3.432.51.21515
6.66×10−46.66 \times10^{-4}2.81.7(192, 120, 110)3.644.11.2705
8×10−48 \times10^{-4}4.02.1(192, 120, 110)3.843.71.3225
1×10−31 \times10^{-3}6.22.6(192, 120, 110)4.042.31.3785
2×10−32 \times10^{-3}24.95.2(192, 120, 110)4.924.71.5305
4×10−34 \times10^{-3}99.610(192, 120, 110)5.513.11.6495
6×10−36 \times10^{-3}22416(192, 120, 110)5.68.61.7025
8×10−38 \times10^{-3}39821(192, 120, 110)5.76.31.7315
1×10−21 \times10^{-2}62326(192, 120, 110)5.75.21.7545
3.33×10−23.33 \times10^{-2}690286(192, 120, 110)5.62.01.8275
6.66×10−26.66 \times10^{-2}27619172(192, 120, 110)5.51.31.7545
1×10−11 \times10^{-1}62240258(192, 120, 110)5.41.01.8737

Table A.1. Simulations from Series 1 with constant Rac=2×107Ra_c = 2 \times10^7, Rat=−Rac/3Ra_t = -Ra_c/3 and Le=10Le = 10. The numerical resolution is given by the number of radial points, NrN_r, and the truncation degree and order of the spherical harmonics decomposition, Lmax⁡L_{\max} and Mmax⁡M_{\max}. RocRo_c is the convective Rossby number calculated from the compositional Rayleigh number, Roc=RacEk2/ScRo_c = \sqrt{Ra_c Ek^2/Sc}. ReRe is the Reynolds number for the poloidal flow, Re0Re_0 the Reynolds number for the zonal flow. Δtavg\Delta t_{\mathrm{avg}} is the run time used to compute the averaged diagnostics (in units of a viscous timescale). (†) For Ek=10−4Ek = 10^{-4}, the output quantities are calculated before the transition in the equatorial symmetry (see Table A.2). (*) – () The output quantities for the simulation at Ek=1.8×10−4Ek = 1.8 \times10^{-4} averaged before (after) the reversal of the direction of the zonal velocity are denoted by * ( respectively).

RacRa_cN2/Ω2N^2/\Omega^2RocRo_c(Nr,Lmax⁡,Mmax⁡)(N_r,L_{\max},M_{\max})ReReRe0Re_0NucNu_cΔtavg\Delta t_{\mathrm{avg}}
2×1072 \times10^7†0.0620.26(192, 163, 163)2.53.81.10540
2×1072 \times10^7††0.0620.26(192, 163, 163)2.34.61.10755
4×1074 \times10^70.120.36(220, 163, 163)4.03.51.19620
6×1076 \times10^70.190.45(250, 180, 160)4.315.31.17510
8×1078 \times10^70.250.52(250, 180, 160)4.724.21.17910
1.2×1081.2 \times10^80.370.63(250, 180, 160)5.141.61.17210
2×1082 \times10^80.620.82(300, 200, 180)6.183.41.19430
4×1084 \times10^81.21.2(300, 220, 220)6.9182.91.2507
8×1088 \times10^82.51.6(300, 220, 220)8.5303.81.3466
1.2×1091.2 \times10^93.72.0(300, 240, 220)10.2360.81.3693
1.6×1091.6 \times10^95.02.3(300, 240, 220)11.5391.01.3976
2.4×1092.4 \times10^97.52.8(300, 240, 220)13.7402.61.4316
4×1094 \times10^912.43.7(504, 380, 255)17.9309.21.5113
8×1098 \times10^924.95.2(608, 485, 340)23.4238.41.5722

Table A.2. Simulations from Series 2 with constant Ek=10−4Ek = 10^{-4}, RaI=−Rac/3Ra_I = -Ra_c/3 and Le=10Le = 10. (†) – (††) The output quantities for the simulation at Rac=2×107Ra_c = 2 \times10^7 averaged before (after) the transition in the flow equatorial symmetry are denoted by † (†† respectively).

References

Anderson, B.J., Johnson, C.L., Korth, H., Winslow, R.M., Borovsky, J.E., Purucker, M.E., Slavin, J.A., Solomon, S.C., Zuber, M.T., McNutt, R.L., 2012. Low-degree structure in Mercury’s planetary magnetic field. J. Geophys. Res.: Planets 117.

Aubert, J., 2005. Steady zonal flows in spherical shell dynamos. J. Fluid Mech. 542, 53–67.

Aurnou, J.M., Horn, S., Julien, K., 2020. Connections between nonrotating, slowly rotating, and rapidly rotating turbulent convection transport scalings. Phys. Rev. Research 2, 043115.

Baines, P., Gill, A., 1969. On thermohaline convection with linear gradients. J. Fluid Mech. 37, 289–306.

Bouffard, M., Choblet, G., Labrosse, S., Wicht, J., 2019. Chemical convection and stratification in the Earth’s outer core. Frontiers Earth Sci. 7, 99.

Bouffard, M., Labrosse, S., Choblet, G., Fournier, A., Aubert, J., Tackley, P.J., 2017. A particle-in-cell method for studying double-diffusive convection in the liquid layers of planetary interiors. J. Comp. Physics 346, 552–571.

Brown, J.M., Garaud, P., Stellmach, S., 2013. Chemical transport and spontaneous layer formation in fingering convection in astrophysics. Astrophys. J. 768, 34.

Buffett, B., 2014. Geomagnetic fluctuations reveal stable stratification at the top of the earth’s core. Nature 507, 484–487.

Buffett, B.A., Seagle, C.T., 2010. Stratification of the top of the core due to chemical interactions with the mantle. J. Geophys. Res.: Solid Earth 115.

Christensen, U.R., 2006. A deep dynamo generating Mercury’s magnetic field. Nature 444, 1056–1058.

Christensen, U.R., Wicht, J., 2008. Models of magnetic field generation in partly stable planetary cores: Applications to Mercury and Saturn. Icarus 196, 16–34.

Davies, C.J., Pommier, A., Greenwood, S., Wilson, A., 2024. Thermal and magnetic evolution of Mercury with a layered Fe-Si (-S) core. Earth Planet. Sci. Lett. 641, 118812.

References

  1. [1]Anderson, B.J., Johnson, C.L., Korth, H., Winslow, R.M., Borovsky, J.E., Purucker, M.E., Slavin, J.A., Solomon, S.C., Zuber, M.T., McNutt, R.L., 2012. Low-degree structure in Mercury’s planetary magnetic field. J. Geophys. Res.: Planets 117.
  2. [2]Aubert, J., 2005. Steady zonal flows in spherical shell dynamos. J. Fluid Mech. 542, 53–67.DOI
  3. [3]Aurnou, J.M., Horn, S., Julien, K., 2020. Connections between nonrotating, slowly rotating, and rapidly rotating turbulent convection transport scalings. Phys. Rev. Research 2, 043115.arxiv.org/abs/2009.03447
  4. [4]Baines, P., Gill, A., 1969. On thermohaline convection with linear gradients. J. Fluid Mech. 37, 289–306.DOI
  5. [5]Bouffard, M., Choblet, G., Labrosse, S., Wicht, J., 2019. Chemical convection and stratification in the Earth’s outer core. Frontiers Earth Sci. 7, 99.DOI
  6. [6]Bouffard, M., Labrosse, S., Choblet, G., Fournier, A., Aubert, J., Tackley, P.J., 2017. A particle-in-cell method for studying double-diffusive convection in the liquid layers of planetary interiors. J. Comp. Physics 346, 552–571.DOI
  7. [7]Brown, J.M., Garaud, P., Stellmach, S., 2013. Chemical transport and spontaneous layer formation in fingering convection in astrophysics. Astrophys. J. 768, 34.DOI
  8. [8]Buffett, B., 2014. Geomagnetic fluctuations reveal stable stratification at the top of the earth’s core. Nature 507, 484–487.DOI
  9. [9]Buffett, B.A., Seagle, C.T., 2010. Stratification of the top of the core due to chemical interactions with the mantle. J. Geophys. Res.: Solid Earth 115.DOI
  10. [10]Christensen, U.R., 2006. A deep dynamo generating Mercury’s magnetic field. Nature 444, 1056–1058.pubmed.ncbi.nlm.nih.gov/17183319
  11. [11]Christensen, U.R., Wicht, J., 2008. Models of magnetic field generation in partly stable planetary cores: Applications to Mercury and Saturn. Icarus 196, 16–34.DOI
  12. [12]Davies, C.J., Pommier, A., Greenwood, S., Wilson, A., 2024. Thermal and magnetic evolution of Mercury with a layered Fe-Si (-S) core. Earth Planet. Sci. Lett. 641, 118812.DOI
  13. [13]Drobyshevski, E., Yuferev, V., 1974. Topological pumping of magnetic flux by three-dimensional convection. J. Fluid Mech. 65, 33–44.DOI
  14. [14]Dumberry, M., Rivoldini, A., 2015. Mercury’s inner core size and core-crystallization regime. Icarus 248, 254–268.
  15. [15]Fraser, A.E., Reifenstein, S.A., Garaud, P., 2024. Magnetized fingering convection in stars. Astrophys. J. 964, 184.arxiv.org/abs/2302.11610
  16. [16]Garaud, P., 2018. Double-diffusive convection at low Prandtl number. Ann. Rev. Fluid Mech. 50, 275–298.DOI
  17. [17]Garaud, P., 2021. Journey to the center of stars: the realm of low Prandtl number fluid dynamics. Phys. Rev. Fluids 6, 030501.arxiv.org/abs/2103.14214
  18. [18]Garaud, P., Medrano, M., Brown, J., Mankovich, C., Moore, K., 2015. Excitation of gravity waves by fingering convection, and the formation of compositional staircases in stellar interiors. Astrophys. J. 808, 89.
  19. [19]Gastine, T., Aurnou, J.M., 2023. Latitudinal regionalization of rotating spherical shell convection. J. Fluid Mech. 954, R1.DOI
  20. [20]Gray, M., Guervilly, C., Sarson, G., 2026. Data from Influence of Rotation on Fingering Convection in a Spherical Stably Stratified Layer. doi:10.25405/data.ncl.31334053.
  21. [21]Greenwood, S., Davies, C.J., Mound, J.E., 2021. On the evolution of thermally stratified layers at the top of Earth’s core. Phys. Earth Planet. Int. 318, 106763.
  22. [22]Grossmann, S., Lohse, D., 2000. Scaling in thermal convection: a unifying theory. J. Fluid Mech. 407, 27–56.
  23. [23]Guervilly, C., 2022. Fingering convection in the stably stratified layers of planetary cores. J. Geophys. Res.: Planets 127, e2022JE007350.
  24. [24]Hage, E., Tilgner, A., 2010. High Rayleigh number convection with double diffusive fingers. Phys. Fluids 22.
  25. [25]Hansen, U., Yuen, D.A., 1990. Nonlinear physics of double-diffusive convection in geological systems. Earth-Science Reviews 29, 385–399.
  26. [26]Harrington, P., Garaud, P., 2019. Enhanced mixing in magnetized fingering convection, and implications for red giant branch stars. Astrophys. J. Lett. 870, L5.
  27. [27]Hauck, S.A., Aurnou, J.M., Dombard, A.J., 2006. Sulfur’s impact on core evolution and magnetic field generation on Ganymede. J. Geophys. Res. Planets 111.
  28. [28]Hauck, S.A., Dombard, A.J., Phillips, R.J., Solomon, S.C., 2004. Internal and tectonic evolution of Mercury. Earth Planet. Sci. Lett. 222, 713–728.
  29. [29]Helfrich, G., Kaneshima, S., 2010. Outer-core compositional stratification from observed core wave speed profiles. Nature 468, 807–810.
  30. [30]Holyer, J.Y., 1983. Double-diffusive interleaving due to horizontal gradients. J. Fluid Mech. 137, 347–362.
  31. [31]Huguet, L., Amit, H., Alboussière, T., 2018. Geomagnetic dipole changes and upwelling/downwelling at the top of the Earth’s core. Frontiers in Earth Science 6, 170.
  32. [32]Irving, J.C., Cottaar, S., Lekić, V., 2018. Seismically determined elastic parameters for Earth’s outer core. Science advances 4, eaar2538.
  33. [33]Kellner, M., Tilgner, A., 2014. Transition to finger convection in double-diffusive convection. Phys. Fluids 26.
  34. [34]Knibbe, J.S., Rivoldini, A., Zhao, Y., Van Hoolst, T., 2025. On the thermal evolution and magnetic field generation of planet Mercury. Phys. Earth Planet. Int. 363, 107348.
  35. [35]Kunze, E., 2003. A review of oceanic salt-fingering theory. Progress in Oceanography 56, 399–417.DOI
  36. [36]Labrosse, S., 2015. Thermal evolution of the core with a high thermal conductivity. Phys. Earth Planet. Int. 247, 36–55.
  37. [37]Lesur, V., Whaler, K., Wardinski, I., 2015. Are geomagnetic data consistent with stably stratified flow at the core–mantle boundary? Geophys. J. Int. 201, 929–946.DOI
  38. [38]Manglik, A., Wicht, J., Christensen, U., 2010. A dynamo model with double diffusive convection for Mercury’s core. Earth Plant. Sci. Lett. 289, 619–628.DOI
  39. [39]Mather, J.F., Simitev, R.D., 2021. Regimes of thermo-compositional convection and related dynamos in rotating spherical shells. Geophys. Astrophys. Fluid Dyn. 115, 61–84.DOI
  40. [40]Medrano, M., Garaud, P., Stellmach, S., 2014. Double-diffusive mixing in stellar interiors in the presence of horizontal gradients. Astrophys. J. Lett. 792, L30.DOI
  41. [41]Monville, R., Vidal, J., Cébron, D., Schaeffer, N., 2019. Rotating double-diffusive convection in stably stratified planetary cores. Geophys. J. Int. 219, 195–218.DOI
  42. [42]Mound, J.E., Davies, C.J., 2017. Heat transfer in rapidly rotating convection with heterogeneous thermal boundary conditions. J. Fluid Mech. 828, 601–629.DOI
  43. [43]Net, M., Garcia, F., Sánchez, J., 2012. Numerical study of the onset of thermosolutal convection in rotating spherical shells. Phys. Fluids 24, 064101.DOI
  44. [44]Paparella, F., Von Hardenberg, J., 2012. Clustering of salt fingers in double-diffusive convection leads to staircaselike stratification. Phys. Rev. Lett. 109, 014502.pubmed.ncbi.nlm.nih.gov/23031108
  45. [45]Radko, T., 2003. A mechanism for layer formation in a double-diffusive fluid. J. Fluid Mech. 497, 365–380.DOI
  46. [46]Radko, T., 2008. The double-diffusive modon. J. Fluid Mech. 609, 59–85.DOI
  47. [47]Radko, T., 2013. Double-diffusive convection. Cambridge University Press.
  48. [48]Radko, T., Smith, D.P., 2012. Equilibrium transport in double-diffusive convection. J. Fluid Mech. 692, 5–27.DOI
  49. [49]Radko, T., Stern, M.E., 2000. Finite-amplitude salt fingers in a vertically bounded layer. J. Fluid Mech. 425, 133–160.DOI
  50. [50]Ruddick, B., Kerr, O., 2003. Oceanic thermohaline intrusions: theory. Progress in oceanography 56, 483–497.DOI
  51. [51]Schaeffer, N., 2013. Efficient spherical harmonic transforms aimed at pseudospectral numerical simulations. Geochem, Geophys, Geosyst. 14, 751–758.DOI
  52. [52]Schaeffer, N., Jault, D., Nataf, H.C., Fournier, A., 2017. Turbulent geodynamo simulations: a leap towards Earth’s core. Geophys. J. Int. 211, 1–29.DOI
  53. [53]Schmitt, R.W., 1979. The growth rate of super-critical salt fingers. Deep Sea Res. 26, 23–40.DOI
  54. [54]Schmitt, R.W., 1994. Double diffusion in oceanography. Ann. Rev. Fluid Mech. 26, 255–285.
  55. [55]Schmitt, R.W., 2011. Thermohaline convection at density ratios below one: A new regime for salt fingers. J. Marine Res. 69, 779–795.DOI
  56. [56]Sengupta, S., Garaud, P., 2018. The effect of rotation on fingering convection in stellar interiors. Astrophys. J. 862, 136.DOI
  57. [57]Silva, L., Mather, J.F., Simitev, R.D., 2019. The onset of thermo-compositional convection in rotating spherical shells. Geophys. Astrophys. Fluid Dyn. 113, 377–404.DOI
  58. [58]Sparks, R.S.J., Huppert, H.E., Turner, J.S., 1984. The fluid dynamics of evolving magma chambers. Phil. Trans. Royal Soc. London. Series A 310, 511–534.DOI
  59. [59]Stellmach, S., Traxler, A., Garaud, P., Brummell, N., Radko, T., 2011. Dynamics of fingering convection. Part 2 The formation of thermohaline staircases. J. Fluid Mech. 677, 554–571.DOI
  60. [60]Stern, M.E., 1960. The “salt-fountain” and thermohaline convection. Tellus 12, 172–175.DOI
  61. [61]Stern, M.E., Radko, T., Simeonov, J., 2001. Salt fingers in an unbounded thermocline. Journal of Marine Research 59, 355–390.DOI
  62. [62]Stevenson, D.J., 1982. Reducing the non-axisymmetry of a planetary dynamo and an application to Saturn. Geophys. Astrophys. Fluid Dyn. 21, 113–127.DOI
  63. [63]Tassin, T., Gastine, T., Fournier, A., 2024. Fingering convection in a spherical shell. J. Fluid Mech. 988, A18.arxiv.org/abs/2309.01602
  64. [64]Tobias, S.M., Brummell, N.H., Clune, T.L., Toomre, J., 2001. Transport and storage of magnetic field by overshooting turbulent compressible convection. Astrophys. J. 549, 1183.
  65. [65]Traxler, A., Garaud, P., Stellmach, S., 2011a. Numerically determined transport laws for fingering (“thermohaline”) convection in astrophysics. Astrophys. J. Lett. 728, L29.
  66. [66]Traxler, A., Stellmach, S., Garaud, P., Radko, T., Brummell, N., 2011b. Dynamics of fingering convection. Part 1 Small-scale fluxes and large-scale instabilities. J. Fluid Mech. 677, 530–553.
  67. [67]Turner, J., 1967. Salt fingers across a density interface. Deep-Sea Res. 14, 599–611.DOI
  68. [68]Wardinski, I., Amit, H., Langlais, B., Thé bault, E., 2021. The internal structure of Mercury’s core inferred from magnetic observations. J. Geophys. Res. Planets 126, e2020JE006792.DOI
  69. [69]Yang, Y., 2020. Double diffusive convection in the finger regime for different Prandtl and Schmidt numbers. Acta Mechanica Sinica 36, 797–804.DOI
  70. [70]Yang, Y., Van Der Poel, E.P., Ostilla-Mónico, R., Sun, C., Verzicco, R., Grossmann, S., Lohse, D., 2015. Salinity transfer in bounded double diffusive convection. J. Fluid Mech. 768, 476–491.DOI
  71. [71]Yang, Y., Verzicco, R., Lohse, D., 2016. Scaling laws and flow structures of double diffusive convection in the finger regime. J. Fluid Mech. 802, 667–689.DOI

Paper details

vlnc:W_be3kq52t
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

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