betaFeedback

WorksPhysics and Astronomy

Inertia-gravity waves in geophysical vortices

Abstract

Pancake-like vortices are often generated by turbulence in geophysical flows. Here, we study the inertia-gravity oscillations that can exist within such geophysical vortices, due to the combined action of rotation and gravity. We consider a fluid enclosed within a triaxial ellipsoid, which is stratified in density with a constant Brunt-Väisälä frequency (using the Boussinesq approximation) and uniformly rotating along a (possibly) tilted axis with respect to gravity. The wave problem is then governed by a mixed hyperbolic-elliptic equation for the velocity. As in the rotating non-stratified case considered by Vantieghem (2014, Proc. R. Soc. A, 470, 20140093, doi:10.1098/rspa.2014.0093), we find that the spectrum is pure point in ellipsoids (i.e. only consists of eigenvalues) with smooth polynomial eigenvectors. Then, we characterise the spectrum using numerical computations (obtained with a bespoke Galerkin method) and asymptotic spectral theory. Finally, the results are discussed in light of natural applications (e.g. for Mediterranean eddies or Jupiter's vortices).

Introduction

Geophysical flows are often influenced by the action of density stratification and rotation. In particular, buoyancy supports the propagation of internal (gravity) waves in stratified fluids [1]. Similarly, the Coriolis force sustains inertial waves in homogeneous rotating fluids [2]. Combining density stratification and global rotation then gives birth to a new wave family [3], usually called inertia-gravity waves (IGWs). Such various waves, which are ubiquitous in rotating and stratified environments, are often believed to be key for understanding the (turbulent) dynamics of geophysical flows [4, 5].

Another characteristic of rotating stratified fluids is that long-lived coherent vortices are often observed in geophysical conditions. This is a direct manifestation of the high-Reynolds-number turbulence of geophysical flows. Rotating turbulence can sustain vortices with a large vertical extent without stratification (due to a nearly geostrophic balance [7]). However, such vortices become strongly flattened when density stratification is strong enough, leading to a pancake-like ellipsoidal shape [8]. Notably, the elliptical shape strongly depends on the interplay between rotational effects, background shear, and the difference between the vortex’s stratification and that of the ambient medium [9, 10, 11]. For instance, Mediterranean eddies are long-lived anticyclones of lenticular shape, with radii 10–100 km and thicknesses of less than 1 km (Figure 1a), formed by the accumulation of warm salty water flowing from the Mediterranean Sea in the Atlantic Ocean [12]. Flattened vortices are also observed in planetary atmospheres. The most striking example is Jupiter’s Great Red Spot (GRS, Figure 1b), which has persisted for more than a century [13].

Figure 1. (a) Cross-section of the temperature anomaly (acoustic tomography) in a Mediterranean eddy offshore in the Eastern Atlantic. Adapted from figure 3 in [6]. (b) Picture of Jupiter’s Great Red Spot (GRS) taken on 21 April 2014 with Hubble. Credits: NASA, ESA and A. Simon (Goddard Space Flight Center). Jupiter’s axis of rotation is Ω\Omega.

The origin of the long-term stability of such (nearly) isolated vortices has thus received much attention in geophysical fluid dynamics. For mathematical tractability, the quasi-geostrophic approximation is often employed to model a geophysical vortex by a fluid ellipsoid with spatially uniform potential vorticity [14]. It has also been reported that such quasi-geostrophic vortices are often quite stable over time [15, 16]. However, various hydrodynamic instabilities could exist in geophysical vortices [17, 18]. Ultimately, these instabilities may sustain (small-scale) bulk turbulence, which would likely affect the long-term vortex stability (because of additional dissipation). In particular, the strong elliptical shape of geophysical vortices may trigger the so-called elliptical instability [19]. This is a parametric instability, due to nonlinear couplings between an elliptical flow and two waves (e.g. inertial or internal gravity waves). Yet, it remains unclear whether the elliptical instability could be triggered for realistic parameters in rotating stratified fluids [20, 21, 22, 23]. Before assessing the relevance of this mechanism for geophysical vortices, a preliminary step is to understand the wave properties in rotating stratified fluids.

Although IGWs have been extensively studied in unbounded fluids [3], the properties of the global oscillations in bounded domains (e.g. the inertia-gravity modes, IGMs) are far from being fully understood. First insight into the mathematical problem can be obtained by considering the non-stratified regime, which has been extensively examined after Greenspan [2]. For homogeneous fluids viewed in a frame rotating at angular velocity Ω\Omega (with respect to an inertial frame), inertial modes in a bounded domain VV are governed by the Poincaré problem (called after Cartan [24] who revisited Poincaré’s pioneering paper [25]). The latter is given by

ω2∇2Φ=(2Ω⋅∇)2Φ,∇Φ⋅n∣∂V=n⋅(u∧2Ω)∣∂V,(1)\omega^2\nabla^2\Phi=(2\boldsymbol{\Omega}\cdot\nabla)^2\Phi,\qquad\left.\nabla\Phi\cdot\boldsymbol{n}\right|_{\partial V}=\left.\boldsymbol{n}\cdot(\boldsymbol{u}\wedge2\boldsymbol{\Omega})\right|_{\partial V}, \tag*{(1)}

where uu is the incompressible fluid velocity, Φ\Phi is the (reduced) hydrodynamic pressure and nn is the unit vector normal to the fluid boundary ∂V\partial V. Time dependence was assumed to be exp⁡(iωt)\exp(i\omega t), where ∣ω∣<2∣Ω∣|\omega|<2|\Omega| is the real-valued angular frequency [2]. Equation (1.1a) is hyperbolic, but boundary condition (1.1b) is neither of Dirichlet-type or Neumann-type (it is a mixed-typed condition, sometimes called oblique condition). Hence, problem (1.1) is an ill-posed Cauchy problem [26]. Nonetheless, it admits smooth polynomial eigenvectors when the geometry is a triaxial ellipsoid [27, 28]. Moreover, these eigenvectors form a complete set in the Hilbert subspace of complex-valued divergenceless fields tangent to the boundary with the L2L^2 norm in ellipsoids [29]. Given the formal analogy between rotation and stratification [30], internal (gravity) modes also obey an ill-posed problem in non-rotating stratified fluids [31]. In the presence of stratification and rotation, the mathematical problem becomes even more complicated because IGMs obey a mixed hyperbolic-elliptic operator [32]. As for pure inertial [33, 34] and internal (gravity) waves [35], IGWs can also converge after multiple reflections on solid boundaries to wave attractors [36]. Attractors are interesting geometrical structures [37], which are capable of focusing the wave energy at small length scales.

This work builds upon and extends previous studies of IGMs in confined geometries. Allen [38] and Friedlander & Siegmann [39] paid attention to IGMs in spherical (and cylindrical) containers when gravity is constant and parallel to the rotation axis. Misaligned cases were later considered in [32] with arbitrary gravity fields. Here, we aim to study IGMs using a simple model retaining the key ingredients to describe pancake-like geophysical vortices. Briefly, we consider a fluid ellipsoid subject to a constant (ambient) gravity field and uniformly rotating along an axis that is tilted with respect to gravity (to account for the full planetary rotation). This setup allows us to extend prior results about pure inertial modes in ellipsoids [27, 28, 29], while preserving polynomial eigenvectors with stratification. Finally, we characterise the wave spectrum using the mathematical theory recently presented in Colin de Verdière & Vidal [40] for inertial modes in ellipsoids. The manuscript is divided as follows. We first formulate the wave problem for general fluid volumes in § 2. Then, we assume that the domain is an ellipsoid in § 3 to model geophysical vortices. The wave spectrum for an ellipsoid is mathematically described and compared to numerical solutions in § 4. The results are discussed in light of geophysical applications in § 5, and we end the paper in § 6. Additional (technical) details are provided in the Electronic Supplementary Material (ESM).

Formulation of the general model

(a) Primitive fluid dynamics equations

As illustrated in Figure 2(a), we model a vortex by a fluid domain of volume VV and (fictive) boundary ∂V\partial V, embedded within an ambient fluid (e.g. a planetary atmosphere or ocean) that is rotating at the angular velocity Ω=Ωs1Ω\boldsymbol{\Omega} = \Omega_s\mathbf{1}_{\Omega} with respect to an inertial frame (where 1Ω\mathbf{1}_{\Omega} is a unit vector). We work in the frame rotating at an angular velocity Ω\Omega, and denote the position vector by r=(x,y,z)T\mathbf{r} = (x,y,z)^{\mathsf{T}} using the Cartesian coordinates. The vortex is internally stratified in density under the action of gravity g\boldsymbol{g}. The vortex’s stratification is quantified by the Brunt-Väisälä (BV) frequency NN. Homogeneous fluids are such that N2=0N^2=0, whereas stably stratified fluids are characterised by N2>0N^2>0. As a starting point, we employ the Boussinesq approximation to account for density variations [41]. The vortex is differentially rotating with respect to the ambient fluid outside VV, which can be modelled by a background flow U0\boldsymbol{U}_0. Laboratory experiments [10, 11] and numerical calculations [9] show that, for incompressible fluids, U0\boldsymbol{U}_0 is often close to a uniform-vorticity flow. The latter can be sought as U0=ω×r+∇Ψ\boldsymbol{U}_0=\boldsymbol{\omega}\times\boldsymbol{r}+\nabla\Psi, where ω\boldsymbol{\omega} is the vortex’s rotation vector (either cyclonic or anticyclonic) and Ψ\Psi is a scalar ensuring that the flow does not cross ∂V\partial V (e.g. [42] in an ellipsoid). The amplitude of U0\boldsymbol{U}_0 is characterised by the Rossby number Ro=∣ω∣/Ωs\mathrm{Ro}=|\boldsymbol{\omega}|/\Omega_s, and we are interested in modelling stratified vortices in the regime ∣Ro∣≪(N/Ωs)2|\mathrm{Ro}|\ll(N/\Omega_s)^2. Thus, we neglect the effects of U0\boldsymbol{U}_0 onto the background density and pressure. We further simplify the model by discarding baroclinic (and centrifugal effects), the very weak vortex’s self-gravity, and the (weak) lateral variations of the planet’s gravity at the size of a vortex. Hence, the fluid is only subject to the ambient gravity g=−g(z)1z\boldsymbol{g}=-g(z)\boldsymbol{1}_z within VV, where 1z\boldsymbol{1}_z is the unit vector along the vertical zz-axis. We denote by ρ∗+ρ0(z)\rho_*+\rho_0(z) the reference density within VV (with the mean density ρ∗\rho_*), which is given by the hydrostatic equilibrium ∇P0=(ρ∗+ρ0)g\nabla P_0=(\rho_*+\rho_0)\boldsymbol{g} where P0P_0 is the hydrostatic pressure.

Figure 2. Sketch of the mathematical model for geophysical vortices (not to scale). (a) Meridional cross-section of a vortex centred at colatitude θ\theta, which is embedded within an ocean or an atmosphere (blue region). (b) Front view of a flattened ellipsoidal vortex of semi-axes a≫b≫ca \gg b \gg c, which is stratified under the uniform planet’s gravity g\mathbf{g} (i.e. the background density ρ0\rho_0 decreases with increasing zz).

Now, we seek small-amplitude perturbations around the reference state [U0,ρ0,P0][\boldsymbol{U}_0,\rho_0,P_0]. Without diffusion, the velocity v(r,t)\boldsymbol{v}(\boldsymbol{r},t) obeys the linearised Euler equation given in the rotating frame by

∂tv+(U0⋅∇)v+(v⋅∇)U0+2Ω×v=−∇π+ρρ∗g(2)\partial_t\boldsymbol{v}+(\boldsymbol{U}_0\cdot\nabla)\boldsymbol{v}+(\boldsymbol{v}\cdot\nabla)\boldsymbol{U}_0+2\boldsymbol{\Omega}\times\boldsymbol{v}=-\nabla\pi+\frac{\rho}{\rho_*}\boldsymbol{g} \tag*{(2)}

together with the incompressible condition ∇⋅v=0\nabla\cdot\boldsymbol{v}=0 for an incompressible fluid, and where π=p/ρ∗\pi=p/\rho_* is the reduced pressure. The density ρ(r,t)\rho(\boldsymbol{r},t) is then governed by

∂tρρ∗+U0⋅∇ρρ∗=(v⋅1z)N2(z)g(z),(3)\frac{\partial_t\rho}{\rho_*}+\boldsymbol{U}_0\cdot\frac{\nabla\rho}{\rho_*}=(\boldsymbol{v}\cdot\boldsymbol{1}_z)\frac{N^2(z)}{g(z)}, \tag*{(3)}

where N2(z)=g⋅∇(ρ0/ρ∗)N^2(z)=\boldsymbol{g}\cdot\nabla(\rho_0/\rho_*) is the BV frequency within the Boussinesq approximation. The background flow has two main effects in the linear theory. First, it is responsible for a shift in frequency of the wave motions (e.g. see in [43] for a discussion in unbounded fluids). Second, it can be responsible for the onset of hydrodynamic instabilities. In particular, when U0\boldsymbol{U}_0 is an elliptical flow [10, 11], the so-called elliptical instability could develop [19] and be responsible for the transition towards a wave turbulence regime [44]. For mathematical simplicity, we assume U0=0\boldsymbol{U}_0=0 in equations (2)–(3) to focus on the free wave motions given by

∂tv+2Ω×v=−∇π−N(z)ξ1z,∇⋅v=0,∂tξ=(v⋅1z)N(z),(4)\begin{aligned} \partial_t\boldsymbol{v}+2\boldsymbol{\Omega}\times\boldsymbol{v}&=-\nabla\pi-N(z)\xi\boldsymbol{1}_z, & \nabla\cdot\boldsymbol{v}&=0,\\ \partial_t\xi&=(\boldsymbol{v}\cdot\boldsymbol{1}_z)N(z), \tag*{(4)} \end{aligned}

where we have introduced ξ=(ρ/ρ∗)g(z)/N(z)\xi=(\rho/\rho_*)g(z)/N(z). The background flow will be reintroduced in future work (to investigate the outcome of the elliptical instability).

Finally, equations (2.3a–c) are supplemented with boundary conditions (BCs). We impose that the fields are regular at r=0\boldsymbol{r}=0. Then, realistic conditions would be to couple the dynamics inside the vortex with the exterior flow on ∂V\partial V. This would amount to considering continuity conditions across ∂V\partial V, and decaying solutions at infinity (e.g. for Rankine [45] or Lamb-Oseen [46] vortices). For mathematical simplicity, we here neglect the coupling with the exterior fluid and assume that ∂V\partial V is a rigid boundary. Consequently, the velocity obeys the no-penetration BC v⋅n∣∂V=0\boldsymbol{v}\cdot\boldsymbol{n}|_{\partial V}=0 on ∂V\partial V, where n\boldsymbol{n} is the unit (outward) vector normal to the boundary. With no diffusion, no other physical BCs have to be explicitly enforced in the model. Indeed, the density and pressure BCs automatically follow from the velocity BC. In practice, the pressure satisfies a mixed BC that is given by the continuity of the normal component of equation (2.3a) on ∂V\partial V (see the ESM).

(b) Mixed hyperbolic-elliptic problem

We seek modal solutions of equations (2.3a–c) given by

[v,π,ξ](r,t)=[u,Φ,ζ](r)exp⁡(iωt),(5)[\boldsymbol{v},\pi,\xi](\boldsymbol{r},t)=[\boldsymbol{u},\Phi,\zeta](\boldsymbol{r})\exp(\mathrm{i}\omega t), \tag*{(5)}

where [u,Φ,ζ][\boldsymbol{u},\Phi,\zeta] are complex-valued amplitudes depending on space, and with ω∈C\omega\in\mathbb{C}. The wave properties are deeply tied to the nature of the mathematical problem. Instead of solving the

Spectral problem for the velocity

primitive equations, it is worth considering a master equation that governs the evolution of either u\boldsymbol{u} or Φ\Phi. If we multiply Euler equation (2.3a) by iωi\omega and replace iωζi\omega\zeta by buoyancy equation (2.3c), we obtain for ω≠0\omega\ne0

−ω2u+2Ω×(iωu)+N2(z)(u⋅1z)1z=−iω∇Φ,∇⋅u=0,(6)-\omega^{2}\boldsymbol{u}+2\boldsymbol{\Omega}\times(i\omega\boldsymbol{u})+N^{2}(z)(\boldsymbol{u}\cdot\boldsymbol{1}_{z})\boldsymbol{1}_{z}=-i\omega\nabla\Phi,\qquad\nabla\cdot\boldsymbol{u}=0, \tag*{(6)}

Equations (2.5a,b) and the no-penetration BC u⋅n∣∂V=0\boldsymbol{u}\cdot\boldsymbol{n}|_{\partial V}=0 show that the velocity is an appropriate variable to explore the wave properties. Note that a similar equation can be obtained for the fluid particle displacement vector [47]. On the contrary, non-oscillatory modes with ω=0\omega=0 require a specific consideration. For unstratified fluids with N=0N=0, they are called geostrophic modes because they obey the geostrophic balance 2Ω×u=−∇Φ2\boldsymbol{\Omega}\times\boldsymbol{u}=-\nabla\Phi. When N(z)>0N(z)>0, the steady modes are two-dimensional (2D) such that u⋅1z=0\boldsymbol{u}\cdot\boldsymbol{1}_{z}=0, and the density perturbation obeys the diagnostic equation ρ∗(2Ω⋅∇u)=∇ζ×1z\rho_{*}(2\boldsymbol{\Omega}\cdot\nabla\boldsymbol{u})=\nabla\zeta\times\boldsymbol{1}_{z} (resulting from the thermal wind balance).

Because of historical reasons (dating back to Poincaré [25]), the problem is generally recast into a scalar equation for the pressure Φ\Phi [32, 39]. As shown in the ESM, the pressure is governed in VV by a second-order equation whose higher-order terms are given by

[ω2−N2(z)]∇2Φ+N2(z)(1z⋅∇)2Φ−(2Ω⋅∇)2Φ+⋯=0(7)[\omega^{2}-N^{2}(z)]\nabla^{2}\Phi+N^{2}(z)(\boldsymbol{1}_{z}\cdot\nabla)^{2}\Phi-(2\boldsymbol{\Omega}\cdot\nabla)^{2}\Phi+\cdots=0 \tag*{(7)}

when ∣ω∣≠ω±(z)|\omega|\ne\omega_{\pm}(z) with

2ω±2(z)=[N2(z)+4∣Ω∣2]±[N2(z)+4∣Ω∣2]2−16N2(z)(Ω⋅1z)2.(8)2\omega_{\pm}^{2}(z)=[N^{2}(z)+4|\boldsymbol{\Omega}|^{2}]\pm\sqrt{[N^{2}(z)+4|\boldsymbol{\Omega}|^{2}]^{2}-16N^{2}(z)(\boldsymbol{\Omega}\cdot\boldsymbol{1}_{z})^{2}}. \tag*{(8)}

The lower-order terms in equation (7), which vanish when NN is constant, are here omitted for concision but are given in the ESM. When ∣ω∣=ω±(z)|\omega|=\omega_{\pm}(z), the pressure instead obeys a first-order equation (see the ESM). Finally, the pressure obeys a mixed BC on ∂V\partial V (i.e. neither a pure Dirichlet nor Neumann BC), which is difficult to take into account for numerical computations. However, the nature of the mathematical problem can be simply determined from the pressure equation by using microlocal analysis [48]. This is a branch of mathematics, which studies the solutions of PDE using (notably) asymptotic geometric techniques. Microlocal tools have already proven useful in physics to study wave problems [49, 50]. In particular, the nature of equation (7) is governed by a scalar quantity that is called the principal symbol. The latter is obtained from the equation by freezing the variable coefficients and keeping only the highest-degree terms. In practice, it amounts to transforming the derivatives as ∇↔ik\nabla\leftrightarrow i\boldsymbol{k}, where k=(k1,k2,k3)T\boldsymbol{k}=(k_{1},k_{2},k_{3})^{\mathsf{T}} is a real-valued wave vector at the position (x,y,z)(x,y,z). The principal symbol is then a homogeneous polynomial in k\boldsymbol{k}, which encapsulates many properties of the spectral problem [37, 51]. The problem is elliptic when the principal symbol is invertible, and hyperbolic when the principal symbol vanishes. The vanishing condition on the principal symbol gives

∣k∣2ω2−[N2(z)∣1z×k∣2+(2Ω⋅k)2]=0,(9)|\boldsymbol{k}|^{2}\omega^{2}-\left[N^{2}(z)|\boldsymbol{1}_{z}\times\boldsymbol{k}|^{2}+(2\boldsymbol{\Omega}\cdot\boldsymbol{k})^{2}\right]=0, \tag*{(9)}

which is the dispersion relation of IGWs in unbounded fluids when NN is constant [3]. The wave-like equation is thus hyperbolic when equation (9) is satisfied for some non-zero wave vectors k∈R3\boldsymbol{k}\in\mathbb{R}^{3}. Otherwise, it is elliptic (except when ω=0\omega=0). Next, the bounds for the hyperbolic domain can be obtained from equation (9). The latter defines a quadric cone given by (g∗)ijkikj=0(\mathsf{g}^{*})^{ij}k_{i}k_{j}=0 (using Einstein notation) when ω≠0\omega\ne0, where g∗\mathsf{g}^{*} is the associated metric tensor. The nature of this quadric cone depends on the eigenvalues of g∗\mathsf{g}^{*}, which are given by μ0=ω2−N2(z)\mu_{0}=\omega^{2}-N^{2}(z) and μ±=ω2−ω±2(z)\mu_{\pm}=\omega^{2}-\omega_{\pm}^{2}(z). When the eigenvalues are non-zero and of the same sign, the surface is elliptic. On the contrary, the surface is hyperbolic when the eigenvalues are non-zero but of different signs. The above microlocal description agrees with the analysis presented in Section 3 of Friedlander & Siegmann [32]. The wave-like equation is thus hyperbolic in VV when ω−(z)<∣ω∣<ω+(z)\omega_{-}(z)<|\omega|<\omega_{+}(z), whereas it is elliptic in VV when 0<∣ω∣<ω−(z)0<|\omega|<\omega_{-}(z) or ∣ω∣>ω+(z)|\omega|>\omega_{+}(z).

Spectral problem for the velocity

Further wave properties can be obtained by considering the velocity equation. We introduce the Hilbert space V\mathcal{V} spanned by complex-valued vector fields that are square-integrable in VV. The complex-valued inner product in this Hilbert space is given by

⟨u,v⟩:=∫Vu†⋅v dV,(10)\langle u,v\rangle:= \int_V u^\dagger\cdot v\,\mathrm{d}V, \tag*{(10)}

where †\dagger is the complex conjugate. The associated norm is ∥u∥=⟨u,u⟩1/2\lVert u\rVert= \langle u,u\rangle^{1/2}. Then, we denote by V0⊂V\mathcal{V}^0 \subset\mathcal{V} the closed subspace that is orthogonal, with respect to inner product (10), to the space spanned by vector fields made of gradients of smooth functions. A smooth element in V0\mathcal{V}^0 is divergenceless and tangent to the boundary [29], that is

V0:={u∈V∣∇⋅u=0 in V, u⋅n=0 on ∂V}.(11)\mathcal{V}^0 := \{u \in\mathcal{V} \mid\nabla\cdot u = 0\ \text{in } V,\ u \cdot n = 0\ \text{on } \partial V\}. \tag*{(11)}

Finally, we introduce the orthogonal projector L\mathbb{L} from V\mathcal{V} onto V0\mathcal{V}^0 (called Leray projector after [52]). For any vector e∈Ve \in\mathcal{V} written as e=∇Ψ+∇×Ae = \nabla\Psi+ \nabla\times A (using a Helmholtz decomposition) for some potentials [Ψ,A][\Psi,A], the projected vector is defined by L(e):=e−∇Ψ\mathbb{L}(e) := e - \nabla\Psi. This projector is used in the mathematical analysis of incompressible flows [53], but also in some numerical studies [54].

Equipped with the above definitions, we can seek solutions of equation (6) with u∈V0u \in\mathcal{V}^0 in ellipsoids. We apply the orthogonal projector to the equation to remove the pressure term, and find that (ω,u)(\omega,u) obey a quadratic eigenvalue problem (QEP). The latter is given by

Qω(u)=0,Qω:=−ω2I+ωiC+K,(12)\mathcal{Q}_{\omega}(u)=0,\qquad\mathcal{Q}_{\omega}:=-\omega^2\mathcal{I}+\omega i\mathcal{C}+\mathcal{K}, \tag*{(12)}

where I\mathcal{I} is the identity operator, and with the two operators

iC(u):=iL[2Ω×u],K(u):=L[N2(z)(u⋅1z)1z].(13)i\mathcal{C}(u):=i\mathbb{L}[2\boldsymbol{\Omega}\times u],\qquad\mathcal{K}(u):=\mathbb{L}[N^2(z)(u\cdot\mathbf{1}_z)\mathbf{1}_z]. \tag*{(13)}

The operator iCi\mathcal{C}, called the Poincaré operator [40], is a bounded self-adjoint operator in V0\mathcal{V}^0 [29]. The buoyancy operator K\mathcal{K} has properties given by Proposition 2.1.

Proposition 2.1. The operator K\mathcal{K} is bounded, self-adjoint and positive in V0\mathcal{V}^0.

Proof. We must show that K\mathcal{K} is symmetric, bounded, and positive in V0\mathcal{V}^0. We verify that K\mathcal{K} is symmetric since ⟨a,K(b)⟩=⟨K(a),b⟩\langle a,\mathcal{K}(b)\rangle= \langle\mathcal{K}(a),b\rangle for any (a,b)∈V0×V0(a,b)\in\mathcal{V}^0\times\mathcal{V}^0. Moreover, we have

∥K(u)∥2=⟨K(u),N2(u⋅1z)1z⟩−⟨K(u),∇Φ⟩=⟨K(u),N2(u⋅1z)1z⟩\lVert\mathcal{K}(u)\rVert^2 = \langle\mathcal{K}(u),N^2(u\cdot\mathbf{1}_z)\mathbf{1}_z\rangle-\langle\mathcal{K}(u),\nabla\Phi\rangle= \langle\mathcal{K}(u),N^2(u\cdot\mathbf{1}_z)\mathbf{1}_z\rangle

since ⟨K(u),∇Φ⟩=0\langle\mathcal{K}(u),\nabla\Phi\rangle=0 using the definition of L\mathbb{L}. Using Cauchy-Schwarz inequality, we then obtain ∥K(u)∥2≤∥K(u)∥∥N2(u⋅1z)1z∥≤max⁡(N2)∥K(u)∥∥u∥\lVert\mathcal{K}(u)\rVert^2\leq\lVert\mathcal{K}(u)\rVert\lVert N^2(u\cdot\mathbf{1}_z)\mathbf{1}_z\rVert\leq\max(N^2)\lVert\mathcal{K}(u)\rVert\lVert u\rVert, showing that K\mathcal{K} is bounded. Finally, K\mathcal{K} is positive because ⟨u,K(u)⟩≥0\langle u,\mathcal{K}(u)\rangle\geq0 if N2≥0N^2\geq0. ■

We want to characterise the spectrum σ\sigma of the quadratic spectral form Qω\mathcal{Q}_{\omega} in ellipsoids, which is the set of complex numbers ω\omega for which Qω\mathcal{Q}_{\omega} is not continuously invertible. Actually, the spectral problem Qω(u)=0\mathcal{Q}_{\omega}(u)=0 can be converted into a standard eigenvalue problem (SEP) for a bounded self-adjoint operator in V0×V0\mathcal{V}^0\times\mathcal{V}^0.

Proof. This is a general result for spectral families of the form Qω:=−ω2+ωA+B\mathcal{Q}_{\omega}:=-\omega^2+\omega\mathcal{A}+\mathcal{B} acting on a Hilbert space H\mathcal{H}, where [A,B][\mathcal{A},\mathcal{B}] are bounded self-adjoint operators and with B≥0\mathcal{B}\geq0 [55, 56]. When ω≠0\omega\neq0, the QEP Qω(u)=0\mathcal{Q}_{\omega}(u)=0 can be recast as a SEP for the operators L\mathcal{L} or T\mathcal{T} given by

L:=(0IBA),T:=(AB1/2B1/20),(14)\mathcal{L}:=\begin{pmatrix}0&\mathcal{I}\\\mathcal{B}&\mathcal{A}\end{pmatrix},\qquad \mathcal{T}:=\begin{pmatrix}\mathcal{A}&\mathcal{B}^{1/2}\\\mathcal{B}^{1/2}&0\end{pmatrix}, \tag*{(14)}

which are both acting on the Hilbert space H×H\mathcal{H}\times\mathcal{H} equipped with the inner product ⟨z,ζ⟩=⟨z1,ζ1⟩+⟨z2,ζ2⟩\langle z,\zeta\rangle=\langle z_1,\zeta_1\rangle+\langle z_2,\zeta_2\rangle with z=(z1,z2)Tz=(z_1,z_2)^{\mathsf{T}} and ζ=(ζ1,ζ2)T\zeta=(\zeta_1,\zeta_2)^{\mathsf{T}}. Here, T\mathcal{T} is a bounded self-adjoint operator (where B1/2\mathcal{B}^{1/2} is the square root of B\mathcal{B}, which is also self-adjoint [57]), whereas L\mathcal{L} has no particular symmetries. The spectra of Qω\mathcal{Q}_{\omega}, L\mathcal{L} and T\mathcal{T} are identical outside ω=0\omega=0. ■

The spectrum σ(Qω)\sigma(\mathcal{Q}_{\omega}) is thus the disjoint union of the point spectrum given by

σP(Qω):={ω∈C∣∃ u≠0 Qω(u)=0},(15)\sigma_{P}(\mathcal{Q}_{\omega}) := \{\omega\in\mathbb{C} \mid\exists\, u \ne0\ \mathcal{Q}_{\omega}(u)=0\}, \tag*{(15)}

which is the subset spanned by the eigenvalues of QEP (2.11), and the remaining continuous spectrum (i.e. the set of ω\omega for which T−ωI\mathcal{T}-\omega I is injective and has a dense range, but is not surjective). This mathematical distinction is of direct physical interest to IGMs. Indeed, the point spectrum is associated with (regular) square-integrable eigenvectors in diffusionless fluids. On the contrary, a non-empty continuous spectrum is often characterised by (almost) periodic attracting trajectories obtained from ray tracing techniques [37]. These geometrical structures, called attractors [37], are associated with nearly singular (i.e. non-square-integrable) velocity fields when diffusion is vanishingly small [58]. Given the symmetries of the different operators, some general properties of the spectrum are given in Proposition 2.2.

Proposition 2.2. The spectrum of QEP (2.11) is real-valued and even. Moreover, the spectrum is bounded by ∣ω∣≤ωmax⁡|\omega| \le\omega_{\max} with ωmax⁡≤Ωs+Ωs2+max⁡(N2)\omega_{\max} \le\Omega_s + \sqrt{\Omega_s^2+\max(N^2)}. Finally, two eigen-pairs (ω1,u1)(\omega_1,u_1) and (ω2,u2)(\omega_2,u_2) with ω1≠ω2\omega_1 \ne\omega_2 are not orthogonal with respect to inner product (10), but satisfy the two integral properties

ω1ω2⟨u2,u1⟩=−⟨u2,K(u1)⟩,(ω2+ω1)⟨u2,u1⟩=⟨u2,iC(u1)⟩.(16)\omega_1\omega_2\langle u_2,u_1\rangle=-\langle u_2,\mathcal{K}(u_1)\rangle,\qquad(\omega_2+\omega_1)\langle u_2,u_1\rangle=\langle u_2,i\mathcal{C}(u_1)\rangle. \tag*{(16)}

Proof. ω∈R\omega\in\mathbb{R} result from the fact that the operator T\mathcal{T} defined in equation (2.13b) is self-adjoint when K≥0\mathcal{K} \ge0. Then, the complex-conjugate of QEP (2.11) gives

−ω2u†−ωiC(u†)+K(u†)=0,-\omega^2u^\dagger-\omega i\mathcal{C}(u^\dagger)+\mathcal{K}(u^\dagger)=0,

which shows that (−ω,u†)(-\omega,u^\dagger) is an eigen-pair of the QEP. The upper bound of ωmax⁡\omega_{\max} can be obtained using a min-max principle [55]. Finally, the orthogonality conditions can be found using vector manipulations of the QEP [59].

Ellipsoidal model for a constant BV frequency

We now simplify the general model to consider an idealised configuration of geophysical vortices. To get physical insight into the problem, it is useful to consider a shape that is amenable to mathematical analysis. For simplicity, we assume that the boundary ∂V\partial V is a triaxial ellipsoid of semi-axes [a,b,c][a,b,c], where the cc-axis is aligned with gravity (Figure 2b). The use of ellipsoids has a long-standing history in the study of geophysical vortices [14], and is supported by recent fluid dynamic experiments [10, 11]. In the rotating frame, the boundary is expressed using the Cartesian coordinates as F(x,y,z)=1F(x,y,z)=1 with F(x,y,z):=(x/a)2+(y/b)2+(z/c)2F(x,y,z):=(x/a)^2+(y/b)^2+(z/c)^2, such that the unit normal is n:=∇F/∥∇F∥\boldsymbol{n}:=\nabla F/\lVert\nabla F\rVert. Moreover, geophysical vortices with N≫2ΩsN\gg2\Omega_s are usually strongly flattened with a,b≫ca,b\gg c (e.g. Jupiter’s GRS is expected to have a depth of only a few hundreds of km [11]). In such vortices, the ambient gravity is expected to weakly varies with depth. Thus, we also assume that g=−g1z\boldsymbol{g}=-g\boldsymbol{1}_z is uniform and consider a background density ρ∗+ρ0(z)\rho_*+\rho_0(z) that varies linearly with zz as

ρ0(z)/ρ∗=−(N2/g)z,(17)\rho_0(z)/\rho_* = -(N^2/g)z, \tag*{(17)}

where NN is a constant BV frequency. With these assumptions, the eigensolutions can be characterised using finite-dimensional arguments in ellipsoids.

(a) Polynomial description

Since an ellipsoid is a quadratic surface and expression (17) is a polynomial, we can seek eigenvectors using polynomial functions in the Cartesian coordinates. We denote by Pn:R3→C\mathcal{P}_n:\mathbb{R}^3\to\mathbb{C} the space spanned by the scalar Cartesian monomials xiyjzkx^iy^jz^k of degree i+j+ki+j+k smaller or equal to nn with n≥1n\ge1. Its dimension is N(Pn)=(n+1)(n+2)(n+3)/6N(\mathcal{P}_n)=(n+1)(n+2)(n+3)/6. Then, we denote by Vn\mathcal{V}_n the space of vectors in ellipsoids with coefficients in Pn\mathcal{P}_n and define the subspace Vn0:=Vn∩V0\mathcal{V}_n^0:=\mathcal{V}_n\cap\mathcal{V}^0 of dimension N(Vn0)=n(n+1)(2n+7)/6N(\mathcal{V}_n^0)=n(n+1)(2n+7)/6, whose elements are divergenceless vector fields that are tangent to the boundary ∂V\partial V. Finally, we introduce the subspaces Wn0:=(Vn−10)⊥∩Vn0\mathcal{W}_n^0 := (\mathcal{V}_{n-1}^0)^\perp\cap\mathcal{V}_n^0 of dimension N(Wn0)=n(n+2)N(\mathcal{W}_n^0)=n(n+2), which are spanned by vectors in Vn0\mathcal{V}_n^0 of degree nn that are orthogonal to polynomial vectors of degree ≤n−1\le n-1. These polynomial spaces are key to characterise the spectrum because of Proposition 3.1.

Proposition 3.1. The spaces Vn0\mathcal{V}_n^0 with n≥1n\ge1 are invariant by the action of C\mathcal{C} and K\mathcal{K} in triaxial ellipsoids, that is C∣Vn0⊆Vn0\mathcal{C}|_{\mathcal{V}_n^0}\subseteq\mathcal{V}_n^0 and K∣Vn0⊆Vn0\mathcal{K}|_{\mathcal{V}_n^0}\subseteq\mathcal{V}_n^0.

Proof. The invariance of Vn0\mathcal{V}_n^0 under the action of C\mathcal{C} has been demonstrated elsewhere [29, 40]. Here, we can use similar arguments to show that it is invariant by the action of K\mathcal{K}. Briefly, we must prove that the Leray projector L\mathsf{L} is well-defined on Vn0\mathcal{V}_n^0. This is ensured by Theorem 2.4 in [60], which shows that there is a unique scalar Ψ∈Pn+1\Psi\in\mathcal{P}_{n+1} in the definition of L\mathsf{L}. ■

We deduce from Proposition 3.1 that Vn0\mathcal{V}_n^0 is invariant by the action of Qω\mathcal{Q}_\omega. Moreover, since ⨁n∈N∗Vn0\bigoplus_{n\in\mathbb{N}^*}\mathcal{V}_n^0 is dense in V0\mathcal{V}^0 [40], it shows that σ(Qω)\sigma(\mathcal{Q}_\omega) is pure point in ellipsoids (i.e. σ=σP\sigma=\sigma_P). The continuous spectrum is thus empty, and equation (6) admits square-integrable eigenvectors in Vn0\mathcal{V}_n^0. From a physical viewpoint, an empty continuous spectrum means that there are no singular velocity fields associated with wave attractors. This is a difference with other geometries, in which wave attractors can exist even when NN is constant (e.g. in a trapezoid [36]). However, this does not preclude the existence of (almost) periodic wave trajectories using ray tracing techniques. The classical example is the pure inertial wave problem within a sphere. Periodic trajectories have been found in this geometry using ray theory [61] but, since the continuous spectrum is also empty in a full sphere [29, 62], these structures are not associated with singular global modes for vanishingly small diffusion. Here, because the spectrum is pure point, we can solve the eigenvalue problem by restricting uu to Vn0\mathcal{V}_n^0 in ellipsoids.

(b) Galerkin algorithm

An accurate numerical description can be developed by seeking u∈Vn0u\in\mathcal{V}_n^0 as u=∑j≥1αjeju=\sum_{j\ge1}\alpha_j e_j, where {αj}j≥1\{\alpha_j\}_{j\ge1} are complex-valued coefficients, and {ej}j≥1\{e_j\}_{j\ge1} are real-valued basis vector elements of Vn0\mathcal{V}_n^0 [63] normalised such that ∥ej∥=1\lVert e_j\rVert=1. Then, we substitute the above expansion into the QEP Qω(u)=0\mathcal{Q}_\omega(u)=0, and project the entire equation on every eie_i using inner-product (10) to cancel out the spatial dependence (Galerkin method [63]). This gives the finite-dimensional QEP

(−ω2Mn+ωiCn+Kn)α=0(18)\left(-\omega^2 M_n+\omega\mathrm{i}C_n+K_n\right)\alpha=0 \tag*{(18)}

where α=(α1,α2,…)T\alpha=(\alpha_1,\alpha_2,\ldots)^\mathsf{T} is the eigenvector, and where [Mn,Cn,Kn][M_n,C_n,K_n] are three real-valued matrices of size N(Vn0)×N(Vn0)N(\mathcal{V}_n^0)\times N(\mathcal{V}_n^0) and of elements Mn,ij=⟨ei,ej⟩M_{n,ij}=\langle e_i,e_j\rangle, Cn,ij=⟨ei,2Ω×ej⟩C_{n,ij}=\langle e_i,2\Omega\times e_j\rangle and Kn,ij=⟨ei,N2(ej⋅1z)1z⟩K_{n,ij}=\langle e_i,N^2(e_j\cdot\mathbf{1}_z)\mathbf{1}_z\rangle. The matrix MnM_n is positive definite but non-diagonal (because the chosen basis elements are not mutually orthogonal by construction in triaxial ellipsoids [63]). The matrix CnC_n is antisymmetric, and KnK_n is a real-valued positive matrix. Further properties of [Cn,Kn][C_n,K_n] are given below in Proposition 3.2.

Proposition 3.2. dim⁡(ker⁡Cn)=⌊n/2⌋\dim(\ker C_n)=\lfloor n/2\rfloor and dim⁡(ker⁡Kn)=n(n+1)(n+2)/6\dim(\ker K_n)=n(n+1)(n+2)/6.

Proof. When N=0N=0, ker⁡(Cn)\ker(C_n) is associated with geostrophic modes of degree nn. Its dimension was proved in [40]. When N≠0N\ne0, ker⁡(Kn)\ker(K_n) is associated with 2D motions such that u⋅1z=0u\cdot\mathbf{1}_z=0. The latter can be sought as u=∇×(Ψ1z)u=\nabla\times(\Psi\mathbf{1}_z) where Ψ=p[1−F(x,y,z)]\Psi=p[1-F(x,y,z)] is a stream function and with p∈Pn−1p\in\mathcal{P}_{n-1}. The dimension of ker⁡(Kn)\ker(K_n) is thus dim⁡(Pn−1)=n(n+1)(n+2)/6\dim(\mathcal{P}_{n-1})=n(n+1)(n+2)/6. ■

For numerical convenience, we convert the finite-dimensional QEP into an equivalent finite-dimensional SEP when ω≠0\omega\ne0. The finite-dimensional counterpart of the self-adjoint operator defined in equation (2.13b) is not well suited for numerical computations (because of the square root). Instead, we consider the SEP Lnz=ωzL_n z=\omega z with z=(α,β)Tz=(\alpha,\beta)^\mathsf{T} and the matrix LnL_n given by

Ln:=(0IBniAn),An:=Mn−1Cn,Bn:=Mn−1Kn,(19)L_n:=\begin{pmatrix} 0 & I \\ B_n & \mathrm{i}A_n \end{pmatrix},\qquad A_n:=M_n^{-1}C_n,\qquad B_n:=M_n^{-1}K_n, \tag*{(19)}

where II is the identity matrix. As illustrated in Figure 3, the matrices AnA_n and BnB_n are sparse and have an upper triangular block structure (which directly results from Proposition 3.1). It has a practical consequence in the unstratified case N=0N=0. The eigenvalues of the pure inertial modes, given by Anα=ωαA_n\alpha=\omega\alpha, can indeed be computed separately for every space Wn0W_n^0 by seeking the eigenvalues of every diagonal block of AnA_n. Yet, since the matrix LnL_n has not an upper triangular block structure, we cannot compute separately the eigenvalues for every degree nn when N≠0N\ne0.

Figure 3. Sparse upper triangular block matrices AnA_n (left) and BnB_n (right) for n=5n=5 and 1Ω=1z\mathbf{1}_{\Omega}=\mathbf{1}_z.

We have modified the bespoke numerical code SHINE, initially developed for compressible fluids [64, 65], to implement the above algorithm for Boussinesq fluids. The matrices [Mn,Cn,Kn][M_n,C_n,K_n] are first built analytically. Since the basis vector elements are not mutually orthogonal in an ellipsoid [63], the matrices become numerically ill-conditioned when the polynomial degree is large. Hence, we perform the computations using extended-precision algorithms. We can illustrate the Galerkin algorithm by considering the six polynomial solutions u∈V10u\in V_1^0. The latter are sought in the form of uniform-vorticity flows as [42]

u=α×r+∇ψ,ψ=αxc2−b2b2+c2yz+αya2−c2a2+c2xz+αzb2−a2a2+b2xy,(20)u=\alpha\times r+\nabla\psi,\qquad\psi=\alpha_x\frac{c^2-b^2}{b^2+c^2}yz+\alpha_y\frac{a^2-c^2}{a^2+c^2}xz+\alpha_z\frac{b^2-a^2}{a^2+b^2}xy, \tag*{(20)}

and with α=(αx,αy,αz)T\alpha=(\alpha_x,\alpha_y,\alpha_z)^\mathsf{T}. Next, the block matrices of L1L_1 are given by

B1=(N2c2b2+c2000N2c2a2+c20000),A1=(0−2Ωza2a2+c22Ωya2a2+b22Ωzb2b2+c20−2Ωxb2a2+b2−2Ωyc2b2+c22Ωxc2a2+c20),(21)B_1=\begin{pmatrix} \frac{N^2c^2}{b^2+c^2} & 0 & 0\\ 0 & \frac{N^2c^2}{a^2+c^2} & 0\\ 0 & 0 & 0 \end{pmatrix},\qquad A_1=\begin{pmatrix} 0 & -\frac{2\Omega_z a^2}{a^2+c^2} & \frac{2\Omega_y a^2}{a^2+b^2}\\ \frac{2\Omega_z b^2}{b^2+c^2} & 0 & -\frac{2\Omega_x b^2}{a^2+b^2}\\ -\frac{2\Omega_y c^2}{b^2+c^2} & \frac{2\Omega_x c^2}{a^2+c^2} & 0 \end{pmatrix}, \tag*{(21)}

and with Ω=(Ωx,Ωy,Ωz)T\Omega=(\Omega_x,\Omega_y,\Omega_z)^\mathsf{T}. Finally, the eigenvalue-eigenvector solutions of L1L_1 can be solved analytically to extend the linear solutions obtained by Vantieghem [27] when N=0N=0 and 1Ω=1z\mathbf{1}_{\Omega}=\mathbf{1}_z. However, we must rely on numerical computations to obtain LnL_n for higher degrees nn.

Enumeration of the eigenvalues

We can explicitly count the number of non-zero and zero eigenvalues of the QEP when u∈Vn0u \in\mathcal{V}_n^0. Since the spectra of the primitive equations, the one of QωQ_\omega and the one of L\mathcal{L} are identical when ω≠0\omega\ne0, their finite-dimensional counterparts have the same number of non-zero eigenvalues.

Proposition 3.3. The number of non-zero eigenvalues of QEP (18) is n(n+1)(n+5)/3n(n+1)(n+5)/3.

Proof. The number of non-zero eigenvalues can be estimated from the primitive equations

iωu+2Ω×u=−∇π−ζN1z,∇⋅u=0,iωζ=(u⋅1z)N.\mathrm{i}\omega u + 2\boldsymbol{\Omega} \times u = -\nabla\pi- \zeta N\mathbf{1}_z,\qquad\nabla\cdot u = 0,\qquad\mathrm{i}\omega\zeta= (u \cdot\mathbf{1}_z)N.

Solutions of the latter equations belong to the space spanned by u∈Vn0u \in\mathcal{V}_n^0 and ζ∈Pn\zeta\in\mathcal{P}_n, whose dimension is N(Vn0)+N(Pn)=(n+1)(n2+4n+2)/2N(\mathcal{V}_n^0) + N(\mathcal{P}_n) = (n+1)(n^2 + 4n + 2)/2. Next, we count the number of zero eigenvalues. From the density equation, we deduce that ω=0\omega= 0 is associated with 2D flows such that u⋅1z=0u \cdot\mathbf{1}_z = 0. Such 2D flows u∈Vn0u \in\mathcal{V}_n^0 can be sought as u=∇×(Ψ1z)u = \nabla\times(\Psi\mathbf{1}_z) where Ψ=p[1−F(x,y,z)]\Psi= p[1 - F(x,y,z)] is a stream function and with p∈Pn−1p \in\mathcal{P}_{n-1}. Assuming that Ω=(Ωx,Ωy,Ωz)⊤\boldsymbol{\Omega} = (\Omega_x,\Omega_y,\Omega_z)^\top, we obtain 2Ωz∂xΨ=−∂xπ2\Omega_z\partial_x\Psi= -\partial_x\pi and 2Ωz∂yΨ=−∂yπ2\Omega_z\partial_y\Psi= -\partial_y\pi from the horizontal components of the momentum equation. Then, we deduce that π=−2ΩzΨ+h(z)\pi= -2\Omega_z\Psi+ h(z), where h(z)h(z) is an arbitrary polynomial function of zz and of degree n+1n+1 (with n≥1n \ge1). Finally, the density perturbation is obtained from the vertical component of the momentum equation, which gives ζN=2(Ωy∂yΨ+Ωx∂xΨ)−∂zπ\zeta N = 2(\Omega_y\partial_y\Psi+ \Omega_x\partial_x\Psi) - \partial_z\pi. Thus, ω=0\omega= 0 is associated with an eigenspace of dimension dim⁡(Pn−1)+(n+1)=(n+1)(n2+2n+6)/6\dim(\mathcal{P}_{n-1}) + (n+1) = (n+1)(n^2 + 2n + 6)/6. Therefore, the number of non-zero eigenvalues is given by n(n+1)(n+5)/3n(n+1)(n+5)/3.

Equipped with Proposition 3.3, we can enumerate the zero eigenvalues of the matrix LnL_n, which has 2N(Vn0)2N(\mathcal{V}_n^0) eigenvalues. Hence, the algebraic multiplicity of the zero eigenvalue ω=0\omega= 0 is given by n(n+1)(n+2)/3n(n+1)(n+2)/3. However, ω=0\omega= 0 is not only associated with standard eigenvectors. The geometric multiplicity mLm_L of ω=0\omega= 0 is the dimension of ker⁡Ln\ker L_n. The null space of the equation Ln(α,β)⊤=0L_n(\alpha,\beta)^\top= 0 is spanned by the vector elements such that β=0\beta= 0 and Knα=0K_n\alpha= 0. Therefore, we have ker⁡Ln=ker⁡Kn⊕{0}\ker L_n = \ker K_n \oplus\{0\} and ω=0\omega= 0 has the geometric multiplicity mL=n(n+1)(n+2)/6m_L = n(n+1)(n+2)/6 according to Proposition 3.2. However, since LnL_n is not Hermitian, the eigenvalue ω=0\omega= 0 has an algebraic multiplicity rL>mLr_L > m_L. The missing rL−mL=n(n+1)(n+2)/6r_L - m_L = n(n+1)(n+2)/6 polynomial solutions are associated with generalized eigenvectors, which belong to ker⁡(Ln2)\ker(L_n^2).

Proposition 3.4. The eigenvalue ω=0\omega= 0 of the matrix LnL_n has the geometric multiplicity mL=n(n+1)(n+2)/6m_L = n(n+1)(n+2)/6, and the algebraic multiplicity rL=2mLr_L = 2m_L (because of generalized eigenvectors).

The vector space associated with ω=0\omega= 0 is thus partly spanned by generalized eigenvectors, which here satisfy Lnz≠0L_n z \ne0 and Ln2z=0L_n^2 z = 0. Back in the temporal domain, they correspond to solutions that are not steady but linear in tt. However, these admissible solutions are mathematical artefacts due to the formulation of the quadratic problem in terms of the matrix LnL_n. Indeed, QEP (18) can be converted into an equivalent self-joint SEP [59], which is the finite-dimensional counterpart of the self-adjoint operator defined in equation (2.13b). This shows that, from a physical viewpoint, ω=0\omega= 0 is only associated with standard eigenvectors. Therefore, we will not further elaborate on the (artificial) generalized eigenvectors of LnL_n below. Following our prior investigation of the non-stratified problem [40], we can combine polynomial computations for fixed values of nn and spectral theory for n→∞n \to\infty to study the eigensolutions in ellipsoids.

Point spectrum in ellipsoids

We have shown that σ(Qω)\sigma(Q_\omega) is pure point in ellipsoids when NN is constant. For this configuration, we can further characterise the eigensolutions. In particular, the spectrum can be divided into two subsets. If an eigenvalue is isolated in the spectrum and of finite multiplicity (when n→∞n \to\infty), it belongs to the discrete spectrum σD\sigma_D. Otherwise, it belongs to the essential spectrum σE:=σ∖σD\sigma_E := \sigma\setminus\sigma_D. (i.e. the complement to the discrete spectrum). Since ⨁n∈N∗Vn0\bigoplus_{n\in\mathbb{N}^{*}} V_{n}^{0} is dense in V0V^{0}, we conclude from Proposition 3.4 that the eigenvalue ω=0\omega= 0 has an infinite multiplicity when n→∞n \to\infty (i.e. ω=0\omega= 0 belongs to σE\sigma_{E}). Now, it remains to characterise the other eigensolutions.

Figure 4. Sketch of the bounded pure point spectrum in triaxial ellipsoids (not to scale) when NN is constant. The spectrum is symmetric with respect to 00 and bounded by ∣ω∣≤ω+\lvert\omega\rvert\leq\omega_{+} (Proposition 4.1). The velocity equation is either hyperbolic or elliptic in volume. The spectrum is spanned by IGMs and low-frequency (surface) modes.

When NN is constant, the frequencies ω±\omega_{\pm} defined in equation (8) are constant in the volume. Moreover, it is possible to find a better estimate of ωmax⁡\omega_{\max} (compared to the one given in Proposition 2.2). In the aligned case, it was shown in [39] that ωmax⁡=ω+\omega_{\max} = \omega_{+} with ω+=max⁡(N,2Ωs)\omega_{+} = \max(N,2\Omega_{s}). Thus, σ3\sigma_{3} is empty in ellipsoids when Ω\boldsymbol{\Omega} and 1z\mathbf{1}_{z} are collinear. We also show below in Proposition 4.1 that ωmax⁡=ω+\omega_{\max} = \omega_{+}, even when Ω\boldsymbol{\Omega} and 1z\mathbf{1}_{z} are not collinear.

Proposition 4.1. When NN is constant, the pure point spectrum is bounded by ω+\omega_{+} in an ellipsoid.

Proof. We introduce the projector operator P\mathcal{P} acting on a pair z=(u,ζ)⊤z = (u,\zeta)^{\top} by P(z):=(Lu,ζ)⊤\mathcal{P}(z) := (\mathcal{L}u,\zeta)^{\top}, where L\mathcal{L} is the Leray projector introduced in § 2. Then, we can reformulate the spectral problem as the spectral properties of the linear operator P\mathcal{P} defined by P(z):=P(EP(z))\mathcal{P}(z) := \mathcal{P}(\mathcal{E}\mathcal{P}(z)) with

E=i(0−2Ω32Ω202Ω30−2Ω10−2Ω22Ω10N00−N0).\mathcal{E} = \mathrm{i} \begin{pmatrix} 0 & -2\Omega_{3} & 2\Omega_{2} & 0 \\ 2\Omega_{3} & 0 & -2\Omega_{1} & 0 \\ -2\Omega_{2} & 2\Omega_{1} & 0 & N \\ 0 & 0 & -N & 0 \end{pmatrix}.

P\mathcal{P} is a self-adjoint operator for square-integrable fields. We have ∥P∥=1\lVert\mathcal{P}\rVert= 1, and the eigenvalues of the Hermitian matrix E\mathcal{E} are ±ω±\pm\omega_{\pm} such that ∥E∥≤ω+\lVert\mathcal{E}\rVert\leq\omega_{+}. Thus, the spectrum is bounded by ω+\omega_{+}.

Therefore, we can divide the allowable frequency range into disjoint intervals with I1=[−ω+,−ω−]∪[ω−,ω+]I_{1} = [-\omega_{+},-\omega_{-}] \cup[\omega_{-},\omega_{+}] and I2=]−ω−,ω−[∖{0}I_{2} = ]-\omega_{-},\omega_{-}[ \setminus\{0\}. The velocity equation is hyperbolic in VV when ω∈I1\omega\in I_{1}, and elliptic in VV when ω∈I2\omega\in I_{2}. We also define the associated spectra σi=σP∩Ii\sigma_{i} = \sigma_{P} \cap I_{i} for i={1,2}i = \{1,2\} and, when uu is restricted to Vn0V_{n}^{0}, the subsets are denoted by [σ1,n,σ2,n][\sigma_{1,n},\sigma_{2,n}] such that σi,n−1⊆σi,n\sigma_{i,n-1} \subseteq\sigma_{i,n}. In unbounded fluids, a continuum of IGWs exists with angular frequencies ω∈I1\omega\in I_{1} given by dispersion relation (9). We will show below that the spectrum σ1⊂σE\sigma_{1} \subset\sigma_{E} is dense in I1I_{1} within ellipsoids. On the contrary, σ2\sigma_{2} is known to be empty in unbounded fluids (which agrees with prior experimental observations [66]). This is consistent with the fact that the velocity satisfies an elliptic-type equation when ω∈I2\omega\in I_{2} (i.e. plane-wave solutions are evanescent in these intervals). Yet, a countable number of eigenfrequencies can exist in some bounded geometries when ∣ω∣<ω−\lvert\omega\rvert< \omega_{-} [38, 39]. The eigensolutions with ω∈σ1\omega\in\sigma_{1} are referred as class-I solutions, and those with ω∈σ2\omega\in\sigma_{2} are called class-II solutions in [38, 39]. We describe the two spectra below.

Low-frequency (surface) modes

Illustrative examples

Contrary to the free-space case, it is known that σ2\sigma_{2} is not empty when Ω∝1z\boldsymbol{\Omega} \propto\mathbf{1}_{z} in some bounded geometries (e.g. in cylinders or spheres [21, 39]). In triaxial ellipsoids, we can also show that σ2≠∅\sigma_{2} \neq\varnothing when Ω⋅1z≠0\boldsymbol{\Omega}\cdot\mathbf{1}_z\ne0 (otherwise, ω−=0\omega_-=0 and σ1\sigma_1 fills the entire interval 0<∣ω∣<ω+0<|\omega|<\omega_+). To this end, we first consider the eigenvalues for u∈V10u\in\mathcal{V}_1^0, which are the roots of the characteristic equation det⁡(L1−ωI)=0\det(L_1-\omega I)=0. The six roots ωj\omega_j of the characteristic polynomial are the two degenerate solutions ω1,2=0\omega_{1,2}=0 and the four solutions

ω3,4=±β42Δ+(β4/Δ)2−4β2/Δ2,ω5,6=±β42Δ−(β4/Δ)2−4β2/Δ2,(22)\omega_{3,4}=\pm\sqrt{\frac{\beta_4}{2\Delta}+\frac{\sqrt{(\beta_4/\Delta)^2-4\beta_2/\Delta}}{2}},\qquad\omega_{5,6}=\pm\sqrt{\frac{\beta_4}{2\Delta}-\frac{\sqrt{(\beta_4/\Delta)^2-4\beta_2/\Delta}}{2}}, \tag*{(22)}

where we have introduced the positive coefficients

β4=N2c2(a2+b2)(a2+b2+2c2)+4Ωx2b2c2(b2+c2)+4Ωy2a2c2(a2+c2)+4Ωz2a2b2(a2+b2),\beta_4=N^2c^2(a^2+b^2)(a^2+b^2+2c^2)+4\Omega_x^2b^2c^2(b^2+c^2)+4\Omega_y^2a^2c^2(a^2+c^2)+4\Omega_z^2a^2b^2(a^2+b^2),
β2=c4N2[4(Ωx2b2+Ωy2a2)+N2(a2+b2)],\beta_2=c^4N^2[4(\Omega_x^2b^2+\Omega_y^2a^2)+N^2(a^2+b^2)],
Δ=a4b2+a4c2+a2b4+2a2b2c2+a2c4+b4c2+b2c4.\Delta=a^4b^2+a^4c^2+a^2b^4+2a^2b^2c^2+a^2c^4+b^4c^2+b^2c^4.

The only non-zero solutions in the non-stratified case N=0N=0 are ω3,4=±β4/Δ\omega_{3,4}=\pm\sqrt{\beta_4/\Delta}, which reduce to the eigenvalues of the pure inertial modes in V10\mathcal{V}_1^0 obtained by Vantieghem [27] when $\Omega_x=\Omega_y=0. Solutions (4.1) are illustrated in Figure 5 for a sphere and a flattened (triaxial) ellipsoid. Several points are worth commenting on in the figure. First, ω3\omega_3 and ω5\omega_5 smoothly vary as a function of NN. We observe that ω3∈σ1,1\omega_3\in\sigma_{1,1} for any value of NN but, on the contrary, ω5\omega_5 can either belong to σ1,1\sigma_{1,1} or σ2,1\sigma_{2,1} when NN is varied. For instance, the transition occurs here at N/Ωs>4N/\Omega_s>4 in a sphere and at a much larger value N/Ωs≳25N/\Omega_s\gtrsim25 in the pancake-like ellipsoid.

Figure 5. Spectrum σ2\sigma_2 in the aligned case 1Ω=1z\mathbf{1}_{\Omega}=\mathbf{1}_z. (a) Eigenvalues ω3,5≥0\omega_{3,5} \ge0 in σ1,1\sigma_{1,1} for a sphere with b/a=c/a=1b/a=c/a=1 (solid purple curves) and an ellipsoid with b/a=0.7b/a=0.7 and c/a=0.1c/a=0.1 (dashed olive curves). Red region shows σ2\sigma_2 in which the problem is hyperbolic. (b) Number of eigenvalues in σ1,n\sigma_{1,n} for a sphere (b/a=c/a=1b/a=c/a=1) and a flattened ellipsoid (b/a=0.7b/a=0.7 and c/a=0.1c/a=0.1). Colour bar shows the value of N/ΩsN/\Omega_s.

Next, we show in Figure 5(b) the evolution of dim⁡(σ2,n)\dim(\sigma_{2,n}) as a function of nn for some illustrative cases. We see that we have always σ2,n≠∅\sigma_{2,n}\ne\varnothing for a sufficiently large degree nn, and that dim⁡(σ2,n)\dim(\sigma_{2,n}) increases when nn is increased. This suggests that σ2\sigma_2 has an infinite (but countable) number of eigenvalues when n→∞n\to\infty. However, dim⁡(σ2,n)\dim(\sigma_{2,n}) is found to strongly depend on the geometry and the value of N/ΩsN/\Omega_s. More specifically, dim⁡(σ2,n)\dim(\sigma_{2,n}) is reduced when NN is increased, but the variation is more pronounced in the sphere than in pancake-like ellipsoids with c/a≪1c/a\ll1. Finally, the number of eigenvalues seems to be bounded by n(n+1)n(n+1), which corresponds to the number of spherical harmonics YlmY_l^m of degree 2≤l≤n+12\le l\le n+1 and order 0<∣m∣<l0<|m|<l. This shows that σ2\sigma_2 is related to solid (spherical) harmonics.

Further insight into the link between σ2\sigma_2 and the theory of spherical harmonics can be gained by considering the aligned case 1Ω=1z\mathbf{1}_{\Omega}=\mathbf{1}_z with N=2ΩsN=2\Omega_s. For this particular configuration, we have ω−=ω+\omega_-=\omega_+ from equation (8). Then, pressure equation (7) and its associated BC for eigenvalues ω∈σ2\omega\in\sigma_{2} reduce to

∇2Φ=0,n⋅∇Φ=−1iω(2Ω×n)⋅∇Φ  on ∂V.(23)\nabla^{2}\Phi=0,\qquad\boldsymbol{n}\cdot\nabla\Phi=-\frac{1}{\mathrm{i}\omega}(2\boldsymbol{\Omega}\times\boldsymbol{n})\cdot\nabla\Phi\ \text{ on }\partial V. \tag*{(23)}

Laplace equation (23) can then be solved using harmonic functions in spherical or ellipsoidal geometries, and the eigenvalues ω\omega are the roots of a transcendental equation obtained from BC (4.2b). In a sphere, the pressure associated with u∈Vn0u \in\mathcal{V}_{n}^{0} is given in spherical coordinates (r,θ,ϕ)(r,\theta,\phi) by the solid harmonics Φ(r)=rlYlm\Phi(r)=r^{l}Y_{l}^{m} with ω=±2Ωsm/l\omega=\pm2\Omega_{s}m/l, where YlmY_{l}^{m} is the spherical harmonic of degree l≤n+1l\le n+1 and azimuthal order m<lm<l. The modes with m>0m>0 have a negative frequency ω<0\omega<0 when Ωs>0\Omega_{s}>0 (respectively positive ω>0\omega>0 if m<0m<0). Given our phase convention ∝exp⁡[i(ωt+mϕ)]\propto\exp[\mathrm{i}(\omega t+m\phi)], these modes have a positive azimuthal phase velocity cϕ=−ω/m>0c_{\phi}=-\omega/m>0 when Ωs>0\Omega_{s}>0 (i.e. prograde direction). The explicit solutions show that the spectrum σ2,n\sigma_{2,n} is here isomorph to the set of rational numbers Q\mathbb{Q}, such that dim⁡(σ2,n)=n(n+1)\dim(\sigma_{2,n})=n(n+1) in this case. Moreover, it can explicitly be shown that there is a one-to-one correspondence between the solutions in the sphere and those in the ellipsoid for this specific configuration [39]. Since Q\mathbb{Q} is dense in R\mathbb{R}, σ2\sigma_{2} belongs to the essential spectrum σE\sigma_{E} when 1Ω=1z\mathbf{1}_{\Omega}=\mathbf{1}_{z} and N=2ΩsN=2\Omega_{s}.

General properties

The above examples strongly suggest that σ2⊂σE\sigma_{2}\subset\sigma_{E}. This can be shown by further looking at the pressure problem. Assuming that ∣ω∣≠ω±|\omega|\ne\omega_{\pm}, the pressure is governed by a boundary-value problem given by equation (7) in VV and a mixed-type BC on ∂V\partial V. This boundary-value scalar problem is elliptic if the equation is elliptic in VV (which occurs here when 0<∣ω∣<ω−0<|\omega|<\omega_{-}), but also if some invertibility conditions are satisfied by the BC on ∂V\partial V [48] (see the ESM). If both requirements were met in VV and everywhere on ∂V\partial V in a given interval [ω1,ω2][\omega_{1},\omega_{2}], then the spectrum would be discrete in this interval (see the ESM). Otherwise, the spectrum would be essential in such an interval. Here, we can prove that σ2\sigma_{2} belongs to the essential spectrum in an ellipsoid when Ω\boldsymbol{\Omega} and 1z\mathbf{1}_{z} are aligned (see the ESM). We also conjecture that σ2∈σE\sigma_{2}\in\sigma_{E} in the misaligned case. Because the spectrum is pure point in ellipsoids, an infinite number of modes would thus exist with 0<∣ω∣<ω−0<|\omega|<\omega_{-} when n→∞n\to\infty.

Finally, we explore the properties of the corresponding eigenvectors. We remind the reader that these modes only exist when ω−≠0\omega_{-}\ne0. Otherwise, pure inertial modes (when N=0N=0), pure internal (gravity) modes (when Ω=0\Omega=0) or pure IGMs (when Ω≠0,1z=0\Omega\ne0,\mathbf{1}_{z}=0) fill the entire interval 0<∣ω∣<ω+0<|\omega|<\omega_{+}. Two modal families are shown in Figure 6. When N/Ωs≤2N/\Omega_{s}\le2, the energy of the lowest-frequency modes with ∣ω∣→0|\omega|\to0 is nearly aligned with the rotation axis. This structure is reminiscent of geostrophic modes without stratification, and fully agrees with asymptotic analysis when N/Ωs≪1N/\Omega_{s}\ll1. As investigated by Allen [38] when Ω∝1z\Omega\propto\mathbf{1}_{z}, these modes appear because a weak stratification imposes some spatial restrictions on the steady geostrophic modes without stratification. The modes are nearly zz-independent throughout the fluid (at the leading order), and have a phase travelling in the prograde direction. Note that this origin seems similar to the appearance of low-frequency Rossby modes in a volume VV without closed geostrophic contours [2]. On the contrary, the higher-frequency modes with ∣ω∣≲ω−|\omega|\lesssim\omega_{-} have an energy that is maximum near the boundary and decays more or less significantly away from the boundary. These low-frequency (surface) modes seem similar to the (trapped) Kelvin waves in shallow-water models [67]. Indeed, as for the Kelvin waves, the wave energy of these low-frequency (surface) modes is mainly trapped near the boundary and decays away from ∂V\partial V (see Figure 6). Therefore, the presence of a boundary introduces a new subset of low-frequency (surface) modes in a rotating stratified fluid (which differ from the higher-frequency IGMs).

Spatial structure of the low-frequency surface modes

Figure 6. Spatial structure of the low-frequency (surface) modes for u∈V100u \in\mathcal{V}_{10}^{0} in the aligned case 1Ω=1z\mathbf{1}_{\Omega}=\mathbf{1}_{z} as a function of N/ΩsN/\Omega_{s}, in a flattened ellipsoid with b/a=1b/a=1 and c/a=0.1c/a=0.1. Meridional view of the absolute value of the normalised density perturbation ∣ζ∣|\zeta| (in logarithmic scale) in the (Oxz)(Oxz) plane (with rescaled axes to have a 1:1 ratio).

(b) High-frequency IGMs

We investigate IGMs with angular frequencies belonging to σ1\sigma_{1}. A few illustrative modes are shown in Figure 7. IGMs are mainly inertial modes modified by gravity when N≪2ΩsN\ll2\Omega_{s}, whereas they are gravity modes more or less affected by rotation when N≫2ΩsN\gg2\Omega_{s}. In between these two limits, rotation and stratification can have competing effects. Moreover, some of their characteristics are directly controlled by the dispersion relation of IGWs. For example, the modes exhibit a pancake-like structure when ω/Ωs→2\omega/\Omega_{s}\to2 and 1Ω=1z\mathbf{1}_{\Omega}=\mathbf{1}_{z}. From dispersion relation (9), the angle ϑ\vartheta between 1z\mathbf{1}_{z} and k\boldsymbol{k} is such that ϑ→0\vartheta\to0 when ω/Ωs→2\omega/\Omega_{s}\to2. Since the group velocity is orthogonal to k\boldsymbol{k} for IGWs [3], the energy propagates in the horizontal direction (in agreement with the observed pancake-like structure). On the contrary, we have θ→π/2\theta\to\pi/2 for high-frequency IGWs with ω→N\omega\to N, such that the energy varies in the vertical direction. IGMs with ω∈σ1\omega\in\sigma_{1} are thus the direct counterparts in ellipsoids of IGWs in unbounded fluids.

Spatial structure of IGMs for $u \in \mathcal{V}_{10}^{0}$ in the aligned case $\mathbf{1}_{\Omega}=\mathbf{1}_{z}$ as a function of $N/\Omega_{s}$

Figure 7. Spatial structure of IGMs for u∈V100u \in\mathcal{V}_{10}^{0} in the aligned case 1Ω=1z\mathbf{1}_{\Omega}=\mathbf{1}_{z} as a function of N/ΩsN/\Omega_{s}, in a flattened ellipsoid with b/a=1b/a=1 and c/a=0.1c/a=0.1. Meridional view of the absolute value of the normalised density perturbation ∣ζ∣|\zeta| (in logarithmic scale) in the (Oxz)(Oxz) plane (with rescaled axes to have a 1:1 ratio).

Such as IGWs without boundaries, IGMs are dense in this frequency interval when n→∞n\to\infty. Yet, the eigenfrequencies are likely not equiprobable within this interval in ellipsoids. The frequency density is worth obtaining for physical applications, since it will strongly constrain the likelihood of obtaining (resonant) IGMs at a given frequency in geophysical vortices. The asymptotic density of the eigenvalues in σ1\sigma_{1} can be obtained using mathematical analysis when n→∞n\to\infty, and then compared to the density obtained from the polynomial computations at fixed values of nn. We define the probability measure πn\pi_{n} of the eigenvalues {ωj}\{\omega_{j}\} of Qω∣Wn0\mathcal{Q}_{\omega}\vert_{\mathcal{W}_{n}^{0}} given by

πn:=(1/dn)∑j=1dnδ(ωj)\pi_{n}:=(1/d_{n})\sum_{j=1}^{d_{n}}\delta(\omega_{j}) with dn=2N(Wn0)d_{n}=2N(\mathcal{W}_{n}^{0}), where δ\delta is the Dirac function. As explained above, the eigenvalues of SEP (3.3) cannot be easily separated in different subsets for every degree nn. Therefore, we rather consider the joint repartition of the eigenvalues of Qω∣Vn0\mathcal{Q}_{\omega}|_{\mathcal{V}_{n}^{0}}. Naturally, the two repartitions have the same asymptotic distribution when n→∞n\to\infty. Asymptotic properties of the essential spectrum of PDEs can generally be investigated using microlocal analysis. Such an approach usually neglects the boundary effects but, here, it is possible to account for them while keeping a microlocal description. The asymptotic behaviour is given in Proposition 4.2.

Proposition 4.2. The measure πn\pi_{n} converges weakly to the asymptotic measure π∞\pi_{\infty} when n→∞n\to\infty. The asymptotic measure is characterised by a cumulative distribution function given by

∫−∞λdπ∞=18πArea⁡(Sλ∩S2)\int_{-\infty}^{\lambda} \mathrm{d}\pi_{\infty}=\frac{1}{8\pi}\operatorname{Area}(\mathcal{S}_{\lambda}\cap\mathcal{S}^{2})

where S2\mathcal{S}^{2} is the surface of the unit sphere, Sλ\mathcal{S}_{\lambda} is the surface in R3\mathbb{R}^{3} defined by

Sλ:={k~∈R3∣ω~(k~)≤λ},k~=(A1kx,A2ky,A3kz)T,\mathcal{S}_{\lambda}:=\left\{\widetilde{\mathbf{k}}\in\mathbb{R}^{3}\mid\widetilde{\omega}(\widetilde{\mathbf{k}})\leq\lambda\right\},\qquad\widetilde{\mathbf{k}}=\left(\sqrt{A_{1}}k_{x},\sqrt{A_{2}}k_{y},\sqrt{A_{3}}k_{z}\right)^{\mathsf{T}},

with A1=1/a2A_{1}=1/a^{2}, A2=1/b2A_{2}=1/b^{2} and A3=1/c2A_{3}=1/c^{2}, and where ω~(k~)\widetilde{\omega}(\widetilde{\mathbf{k}}) is the plane-wave dispersion relation given by equation (9) for the rescaled wave vector k~\widetilde{\mathbf{k}}. Hence, the asymptotic measure has a probability density function f∞:=dπ∞/dωf_{\infty}:=\mathrm{d}\pi_{\infty}/\mathrm{d}\omega that is even and non-vanishing when ω−2≤ω2≤ω+2\omega_{-}^{2}\leq\omega^{2}\leq\omega_{+}^{2}.

Proof. This is the most technical part of the paper. The proof closely follows the strategy presented in Colin de Verdière & Vidal [40] for the non-stratified problem, so we only outline the key steps below. The proof can be extended to matrix operators that only involve pseudo-differential operators of degree 0. This condition is not satisfied by the operator T\mathcal{T} defined in equation (2.13b). Thus, the starting point is to consider the matrix operator L\mathcal{L} defined in equation (14). The fact that this operator is not self-adjoint is not a problem in the analysis (i.e. in the trace formula). Then, we apply the tools of microlocal analysis in the presence of boundary effects. They allow us to get the asymptotics of ∑j=1dnf(ωj)\sum_{j=1}^{d_n}f(\omega_j) when n→∞n\to\infty for any function f∈C0(R∖{0})f\in C_0(\mathbb{R}\setminus\{0\}) (i.e. the set of real-valued continuous functions vanishing near zero). The analysis heavily relies on the principal symbol of the operator L\mathcal{L} when ω≠0\omega\ne0. This gives the asymptotic measure in the interval I1I_1 (where the velocity equation is hyperbolic). Note that we avoid the value ω=0\omega=0 because L\mathcal{L} is not clearly related to the steady solutions of the original physical problem (due to the existence of generalized eigenvectors). ▪

Proposition 4.2 shows that σ1\sigma_{1} is dense in I1I_{1} when n→∞n \to\infty. Next, we compare in Figure 8 the asymptotic density with the density obtained from polynomial computations at a fixed degree n=35n = 35. The asymptotic density is normalised such that ∫−∞∞dπ∞=1\int_{-\infty}^{\infty} \mathrm{d}\pi_{\infty} = 1 but, since Proposition 4.2 heavily relies on the dispersion relation of IGWs, it is assumed that dπ∞/dω=0\mathrm{d}\pi_{\infty}/\mathrm{d}\omega= 0 when ω∈I2\omega\in I_{2}. To have a correct comparison, we have thus removed the spectra σ2,n\sigma_{2,n} before normalising the density. An excellent agreement is then obtained with the asymptotic density (even for the moderate polynomial degree n=35n = 35 considered here). We have also represented a particular configuration for which 1Ω⋅1z=0\mathbf{1}_{\Omega} \cdot\mathbf{1}_{z} = 0, such that ω−=0\omega_{-} = 0 from expression (8). In this case, IGMs are dense in the entire interval ]0,ω+[]0,\omega_{+}[. Note that a similar agreement is found for other configurations (not shown). Hence, Proposition 4.2 can be used to obtain the density of σ1\sigma_{1}.

Density of IGMs in $\sigma_1$

Figure 8. Density of IGMs in σ1\sigma_1. Comparison between asymptotic density dπ∞/dωd\pi_{\infty}/d\omega given in Proposition 4.2 and σ2,35\sigma_{2,35} obtained from polynomial computations. Left: Flattened ellipsoid b/a=1b/a=1 and c/a=0.1c/a=0.1 in the aligned case 1Ω=1z\mathbf{1}_{\Omega}=\mathbf{1}_{z} with N/Ωs=5N/\Omega_{s}=5. Right: Sphere b/a=c/a=1b/a=c/a=1 in the perpendicular case 1Ω⋅1z=0\mathbf{1}_{\Omega}\cdot\mathbf{1}_{z}=0 with N/Ωs=10N/\Omega_{s}=10.

The asymptotic behaviour is further illustrated in Figure 9 for different parameters and geometries. The density of pure inertial modes is known to be uniform within the frequency interval ]−2Ωs,2Ωs[] - 2\Omega_{\mathrm{s}}, 2\Omega_{\mathrm{s}}[ for a sphere [40]. Yet, the density of pure internal gravity modes without rotation is not uniform in the sphere (it is larger for the high-frequency modes than for the low-frequency modes). Between these two limits, the density is very close to the non-rotating case when N/Ωs≫1N/\Omega_{\mathrm{s}} \gg1. On the contrary, the density of the high-frequency IGMs with ∣ω∣/ωmax⁡→1|\omega|/\omega_{\max} \to1 approaches the density of the non-stratified modes when N/Ωs≤2N/\Omega_{\mathrm{s}} \le2. We also observe that the geometry does modify the spectral density. When N/Ωs<2N/\Omega_{\mathrm{s}} < 2, the density is maximal near ∣ω∣→ω−|\omega| \to\omega_{-} in elongated (i.e. prolate) ellipsoids with c/a≫1c/a \gg1, whereas it is maximal near ∣ω∣→ω+|\omega| \to\omega_{+} in flattened (i.e. oblate) ellipsoids with c/a≪1c/a \ll1. Instead, low-frequency IGMs with ∣ω∣→ω−|\omega| \to\omega_{-} are favoured in flattened ellipsoids when N/Ωs>2N/\Omega_{\mathrm{s}} > 2.

Figure 9. Asymptotic density dπ∞/dω~\mathrm{d}\pi_{\infty}/\mathrm{d}\widetilde{\omega} of IGMs with normalised eigenvalues ω~:=ω/ω+\widetilde{\omega} := \omega/\omega_{+}, given by Proposition 4.2. Left: Sphere b/a=c/a=1b/a = c/a = 1 in the aligned case 1Ω=1z\mathbf{1}_{\Omega} = \mathbf{1}_{z}. Right: Axisymmetric ellipsoids with b/a=1b/a = 1 and varying c/ac/a ratios. Top panel computed for N/Ωs=0.5N/\Omega_{\mathrm{s}} = 0.5, and bottom one for N/Ωs=5N/\Omega_{\mathrm{s}} = 5.

Finally, the situation ω+=ω−\omega_{+} = \omega_{-} occurring when 1Ω=1z\mathbf{1}_{\Omega} = \mathbf{1}_{z} and N/Ωs=2N/\Omega_{\mathrm{s}} = 2 deserves a specific consideration. The spectrum σ1\sigma_{1} is then reduced to the degenerate eigenvalue ω/Ωs=±2\omega/\Omega_{\mathrm{s}} = \pm2, whose multiplicity can be determined. Indeed, we know that dim⁡(σ2,n)=n(n+1)\dim(\sigma_{2,n}) = n(n+1) from the explicit solutions of equations (4.2a,b). Because the total number of non-zero eigenvalues is fixed for a degree nn (Proposition 3.3), we have dim⁡(σ1,n)=n(n+1)(n+2)/3\dim(\sigma_{1,n}) = n(n+1)(n+2)/3 when ω+=ω−\omega_{+} = \omega_{-}.

Physical discussion

The idealised model with a constant NN has allowed us to identify the key properties of the bounded oscillations in rotating stratified ellipsoids. Next, physical insight into geophysical vortices could be obtained by plugging realistic parameters into the model. Typical geophysical estimates for some large-scale vortices are given in Table 1. These vortices are strongly stratified with N/Ωs≫1N/\Omega_s \gg1 and, consequently, they usually have a very small aspect ratio c/a≤10−2c/a \le10^{-2}. However, the value of NN is still subject to strong uncertainties. Indeed, the remote determination of the ellipsoidal geometry is only sensitive to the difference between the BV frequency within the vortex and that of the ambient fluid [9, 10, 11]. The detection of IGWs (or IGMs) might be another way to estimate the internal stratification of such vortices. Various mechanisms could indeed excite such wave motions within a vortex (e.g. as interactions with the surrounding fluid, or bulk turbulence [68]). The bounds given by formula (8), which depend on the orientation of Ω\boldsymbol{\Omega} and on the ratio N/ΩsN/\Omega_s, can be simplified in the relevant regime N/Ωs≫1N/\Omega_s \gg1. It gives

ff [rad.s−1^{-1}]N/fN/faa [km]bb [km]cc [km]
Jupiter’s GRS∼1.4×10−4\sim1.4 \times10^{-4}O(102)O(10^{2})∼9000\sim9000∼5000\sim5000∼100\sim100
Jupiter’s Oval BA∼1.9×10−4\sim1.9 \times10^{-4}O(102)O(10^{2})∼3000\sim3000∼3000\sim3000∼50\sim50
Mediterranean eddies∼8.5×10−5\sim8.5 \times10^{-5}10−5010-5010−10210-10^{2}10−10210-10^{2}0.5−10.5-1

Table 1. Typical estimates for some Jovian vortices and Mediterranean eddies (obtained from [10, 11]). Rotation is usually described using the ff-plane approximation with the Coriolis parameter f=2Ω⋅1z=2Ωzcos⁡(θ)f = 2\boldsymbol{\Omega} \cdot\mathbf{1}_z = 2\Omega_z \cos(\theta), where θ\theta the mean colatitude of the vortex (Figure 2).

ω+2≃N2+4[∣Ω∣2−(Ω⋅1z)2],ω−2≃4∣Ω⋅1z∣2.(24)\omega_{+}^{2} \simeq N^{2} + 4\left[\lvert\boldsymbol{\Omega}\rvert^{2} - \left(\boldsymbol{\Omega} \cdot\mathbf{1}_z\right)^{2}\right], \qquad\omega_{-}^{2} \simeq4\lvert\boldsymbol{\Omega} \cdot\mathbf{1}_z\rvert^{2}. \tag*{(24)}

We see that the upper bound approaches ω+2→N2\omega_{+}^{2} \to N^{2} for realistic values N/Ωs≫1N/\Omega_s \gg1. Detecting IGMs within a vortex may thus yield an estimate of the vortex’s stratification.

A striking property of the global modes is that they admit, under our assumptions, polynomial solutions. This behaviour directly results from Proposition 3.1, and IGMs in other configurations can be very different. To illustrate this point, we consider a sphere of radius RR that is stratified under the radial gravity g=−(g/R)r\boldsymbol{g} = -(g/R)\boldsymbol{r}. The background hydrostatic density is then ∇ρ0=−(β/R)r\nabla\rho_0 = -(\beta/R)\boldsymbol{r} (where β>0\beta> 0 is a constant). The BV frequency is thus given by N2=N02(r/R)2N^{2} = N_{0}^{2}(r/R)^{2} using the Boussinesq approximation, where N0=βg/ρ∗N_0 = \sqrt{\beta g/\rho_*} is the constant value of NN at r=Rr = R. Then, the IGMs obey a mixed hyperbolic-elliptic equation that changes of type when [32, 58]

s2(ω2−4Ωs2)+z2ω2=1N02,(25)\frac{s^{2}}{\left(\omega^{2} - 4\Omega_s^{2}\right)} + \frac{z^{2}}{\omega^{2}} = \frac{1}{N_0^{2}}, \tag*{(25)}

where s=x2+y2s = \sqrt{x^{2} + y^{2}} is the cylindrical radius measured from the rotation axis Ω=Ωs1z\boldsymbol{\Omega} = \Omega_s\mathbf{1}_z. For a given angular frequency ω\omega, equation (25) defines a turning surface that separates regions where the equation is hyperbolic and elliptic in VV. Such turning surfaces strongly impact the spatial structure of IGMs, which are mainly trapped in the hyperbolic region [58]. Moreover, IGMs with ∣ω∣→2Ωs<N\lvert\omega\rvert\to2\Omega_s < N become trapped in the equatorial waveguide when N/Ωs≫1N/\Omega_s \gg1 [69]. Finally, such turning surfaces are also likely responsible for a non-empty continuous spectrum without diffusion [58]. On the contrary, the eigenmodes in a stratified ellipsoid with a constant NN are always smooth and can penetrate deep inside the volume.

We have also shown the existence of low-frequency (surface) modes when 0<∣ω∣<ω−0 < \lvert\omega\rvert< \omega_-. Their properties are very similar to those of the (trapped) Kelvin waves in rotating shallow-water models [67], which owe their existence to topological features [70, 71]. Our study brings here a complementary viewpoint. Indeed, we show that such low-frequency modes can exist in a bounded geometry with rigid walls if the pressure problem fails to be elliptic (due to the BC, see the ESM). Since this situation is expected to be rather generic, such low-frequency (surface) modes could also exist in more complicated models of stratified vortices.

Finally, we turn our attention to the lifetime of geophysical vortices. Without bulk motions, the slow temporal evolution of a stratified vortex is governed by diffusion processes that operate on the slow time scale [16, 72] τν=(a2/ν)(2Ωs/N)2\tau_\nu= (a^{2}/\nu)(2\Omega_s/N)^{2}, where ν\nu is the kinematic viscosity of the fluid within the vortex. Yet, it is known that bulk dissipation can be significantly increased in the presence of IGMs shaped by wave attractors [31, 58]. Hence, we could wonder how much energy could be dissipated if some IGMs were forced inside an elliptical vortex. Without stratification, a pure inertial mode (ω,u)(\omega, \mathbf{u}) satisfies in an ellipsoid the intriguing integral property [27, 28]

∫Vu†⋅∇u2u dV=∫Vu†⋅L(∇u2u) dV=0.(26)\int_V \mathbf{u}^{\dagger}\cdot\nabla_{\mathbf{u}}^2\mathbf{u}\,\mathrm{d}V=\int_V \mathbf{u}^{\dagger}\cdot\mathcal{L}(\nabla_{\mathbf{u}}^2\mathbf{u})\,\mathrm{d}V=0. \tag*{(26)}

This could be interpreted as a vanishing viscous dissipation of the inertial modes in an ellipsoid if there were no boundary layers. However, when viscous BCs are properly taken into account on the boundary (either with no-slip or stress-free BCs), the inertial modes have a non-zero viscous dissipation [2, 73]. From a mathematical viewpoint, formula (26) is valid in an ellipsoid because (i) L(∇u2u)∈Vn−2\mathcal{L}(\nabla_{\mathbf{u}}^2\mathbf{u})\in\mathcal{V}^{n-2} when u∈Vn\mathbf{u}\in\mathcal{V}^{n} and (ii) two inertial eigenvectors (ui,uj)(\mathbf{u}_i,\mathbf{u}_j) are orthogonal with respect to inner product (10) such that ⟨ui,uj⟩=δij\langle\mathbf{u}_i,\mathbf{u}_j\rangle=\delta_{ij} (when properly normalised). Here, the IGMs do not satisfy property (26) because they are not mutually orthogonal with respect to inner product (10) as explained in Proposition 2.2. This shows that the IGMs are (at least) subject to viscous dissipation in the bulk on the slow time scale 1/E1/E in a stratified vortex. In addition, they will also be affected by thermal diffusion (if the stratification has a thermal origin, otherwise by molecular diffusion). Since the IGMs are here smooth polynomial solutions, it is possible to account for the leading-order dissipative effects using boundary-layer theory (see the ESM). This also shows that the IGMs would be dissipated on a typical time scale that is of the same order as τν\tau_\nu in the absence of motions. The presence of IGMs may thus affect the slow temporal evolution of stratified vortices. Moreover, small-scale turbulence is often present inside stratified vortices, and IGMs could be involved in nonlinear interactions to generate (weakly) turbulent motions [19, 68, 74]. Therefore, IGMs may play a role in the transition towards turbulence and the long-term evolution of stratified vortices.

Concluding remarks

We have investigated the normal modes of a diffusionless fluid enclosed in a uniformly rotating ellipsoidal vortex, which is stratified in density under a background gravity. Indeed, they could be important for the dynamics of pancake-like stratified vortices (which are ubiquitous in geophysical flows). We have employed the Boussinesq approximation for simplicity, and shown that the problem can be recast as a mixed hyperbolic-elliptic equation for the fluid velocity. Next, we have simplified the model by considering a constant BV frequency. This idealised model retains the key ingredients to account for the dynamics of geophysical vortices. Under these assumptions, we have proved that the problem admits smooth polynomial eigenvectors in triaxial ellipsoids. Given the polynomial form of the eigenvectors, we have characterised the spectrum by using a bespoke numerical algorithm and asymptotic spectral theory following our prior study without stratification [40]. The normal modes consist of high-frequency IGMs and low-frequency (surface) modes. Using microlocal analysis, we have shown that the boundary plays a key role in sustaining such low-frequency (surface) modes. Finally, we have discussed our results in the light of geophysical applications, arguing that accounting for IGMs in the models could be useful to (better) understand the dynamics of stratified vortices.

Several questions have remained unanswered by our work (even in the linear theory). A more in-depth investigation of the properties of the low-frequency (surface) modes will be presented in a mathematical work [75]. From a physical viewpoint, the model could also be extended to account for additional ingredients (while keeping polynomial solutions). For instance, a uniform-vorticity background flow U0U_0 will be taken into account to study the effects of the vortex’s differential rotation (as measured by the Rossby number RoRo). Large-scale vortices (e.g. Jupiter’s vortices or Mediterranean eddies) are expected to be in the low-Rossby regime ∣Ro∣≲O(10−1)\lvert Ro\rvert\lesssim\mathcal{O}(10^{-1}), whereas smaller-scale vortices are in the opposite regime ∣Ro∣≳O(1)\lvert Ro\rvert\gtrsim\mathcal{O}(1) (e.g. vortices at sub-mesoscales ≲1−10\lesssim1-10 km in the Earth’s oceans [6]). We have also entirely discarded magnetic effects. This is a reasonable assumption since, for realistic parameters, magnetic fields are expected to only weakly modify the high-frequency IGMs at the size of a vortex. However, magnetic fields are known to affect the low-frequency spectrum in rotating, stratified and electrically conducting fluids [21, 76]. Moreover, there is a class of magnetic fields leaving invariant the three-dimensional polynomial vector spaces in an ellipsoid [77, 78]. Therefore, our ellipsoidal model could be used as a toy model to investigate the properties of these Magneto-Archimedean-Coriolis (MAC) modes, by combining microlocal analysis and polynomial computations. Other physical ingredients (e.g. a variable BV frequency) would break the exact polynomial nature of the eigenvectors. Hence, the spectrum should be investigated using another numerical method. Finally, future studies could build upon our work to understand the onset of turbulence within geophysical vortices. Turbulent mixing and dissipation may indeed (partly) explain why some stratified vortices are remarkably stable over long time scales, while some others have much shorter lifetimes. Several routes exist for the excitation of turbulence in stratified fluids, but direct applications of our work include the study of elliptical instabilities. So far, the flows driven by elliptical instabilities have only received scant attention in stratified environments [79, 80, 81, 82]. The polynomial Galerkin algorithm presented in this work can be extended in the linear theory to determine the onset of the instabilities upon an elliptical flow. Therefore, elliptical instabilities will be considered in future work to explore, as a function of the Rossby number, the transition towards turbulence in stratified vortices.

Data Accessibility. The paper has an Electronic Supplementary Material (ESM). The source code is available at https://bitbucket.org/vidalje/.

Authors’ Contributions. This work is an idea of JV to pursue the interdisciplinary collaboration with YCdV after their joint work [40]. JV designed the study and performed the Galerkin computations, whereas YCdV conducted the microlocal analysis. Both authors discussed the results presented in the paper and drafted the manuscript before giving final approval for submission.

Competing Interests. The authors declare that they have no competing interests.

Declaration of AI use. The authors did not use AI-assisted technologies in creating this article.

Funding. JV received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 847433, THEIA project). The numerical computations were performed using both laboratory resources and using the GRICAD infrastructure (https://gricad.univ-grenoble-alpes.fr), which is supported by Grenoble research communities. A CC-BY public copyright license has been applied by the authors to the present document and will be applied to all subsequent versions up to the Author Accepted Manuscript arising from this submission, in accordance with the grant’s open access conditions.

Acknowledgements. JV would like to dedicate this work to the memory of Norman R. Lebovitz (1935-2022) and Stijn Vanthieghem (1983-2023), who both revivified the topic and made pioneering contributions. JV acknowledges Bernard Valette and David Cébron for fruitful discussions over the years. YCdV acknowledges Gerd Grubb for her clarification of the definition of a boundary-value elliptic problem. The authors acknowledge the three anonymous referees, whose valuable comments helped them to improve the clarity and quality of the manuscript.

References

  1. [1]Mowbray DE, Rarity BSH. 1967 A theoretical and experimental investigation of the phase configuration of internal waves of small amplitude in a density stratified liquid. J. Fluid Mech. 28, 1–16. (doi:10.1017/S0022112067001867).DOI
  2. [2]Greenspan HP. 1968 The theory of rotating fluids. Cambridge, UK: Cambridge University Press.
  3. [3]LeBlond PH, Mysak LA. 1981 Waves in the ocean. Amsterdam: Elsevier.
  4. [4]Dauvois T, Joubaud S, Odier P, Venaille A. 2018 Instabilities of internal gravity wave beams. Annu. Rev. Fluid Mech. 50, 131–156. (doi:10.1146/annurev-fluid-122316-044539).
  5. [5]Le Bars M. 2023 Wave turbulence in geophysical flows. J. Fluid Mech. 962, F1. (doi:10.1017/jfm.2023.219).DOI
  6. [6]McWilliams JC. 2016 Submesoscale currents in the ocean. Proc. R. Soc. A 472, 20160117. (doi:10.1098/rspa.2016.0117).DOI
  7. [7]Hopfinger EJ, Van Heijst GJF. 1993 Vortices in rotating fluids. Annu. Rev. Fluid Mech. 25, 241–289. (doi:10.1146/annurev.fl.25.010193.001325).DOI
  8. [8]Billant P, Chomaz JM. 2001 Self-similarity of strongly stratified inviscid flows. Phys. Fluids 13, 1645–1651. (doi:10.1063/1.1369125).DOI
  9. [9]Hassanzadeh P, Marcus PS, Le Gal P. 2012 The universal aspect ratio of vortices in rotating stratified flows: theory and simulation. J. Fluid Mech. 706, 46–57. (doi:10.1017/jfm.2012.180).DOI
  10. [10]Aubert O, Le Bars M, Le Gal P, Marcus PS. 2012 The universal aspect ratio of vortices in rotating stratified flows: experiments and observations. J. Fluid Mech. 706, 34–45. (doi:10.1017/jfm.2012.176).DOI
  11. [11]Lemasquerier D, Facchini G, Favier B, Le Bars M. 2020 Remote determination of the shape of Jupiter’s vortices from laboratory experiments. Nat. Phys. 16, 695–700. (doi:10.1038/s41567-020-0833-9).DOI
  12. [12]Richardson PL, Bower AS, Zenk W. 2000 A census of Meddies tracked by floats. Prog. Oceanogr. 45, 209–250. (doi:10.1016/S0079-6611(99)00053-1).DOI
  13. [13]Marcus PS. 1993 Jupiter’s Great Red Spot and other vortices. Annu. Rev. Astron. Astrophys. 31, 523–569. (doi:10.1146/annurev.aa.31.090193.002515).DOI
  14. [14]McKiver WJ. 2015 The ellipsoidal vortex: A novel approach to geophysical turbulence. Adv. Math. Phys. 2015. (doi:10.1155/2015/613683).DOI
  15. [15]Tsang YK, Dritschel DG. 2015 Ellipsoidal vortices in rotating stratified fluids: beyond the quasi-geostrophic approximation. J. Fluid Mech. 762, 196–231. (doi:10.1017/jfm.2014.630).DOI
  16. [16]Facchini G, Le Bars M. 2016 On the lifetime of a pancake anticyclone in a rotating stratified flow. J. Fluid Mech. 804, 688–711. (doi:10.1017/jfm.2016.549).DOI
  17. [17]Sipp D, Lauga E, Jacquin L. 1999 Couette, Poiseuille flows in rotating systems: centrifugal, elliptic and hyperbolic type instabilities. Phys. Fluids 11, 3716–3728. (doi:10.1063/1.870180).DOI
  18. [18]Yim E, Billant P, Meunésguen C. 2016 Stability of an isolated pancake vortex in continuously stratified-rotating fluids. J. Fluid Mech. 801, 508–553. (doi:10.1017/jfm.2016.402).DOI
  19. [19]Kerswell RR. 2002 Elliptical instability. Annu. Rev. Fluid Mech. 34, 83–113. (doi:10.1146/annurev.fluid.34.081701.171829).
  20. [20]Miyazaki T, Fukumoto Y. 1992 Three-dimensional instability of strained vortices in a stably stratified fluid. Phys. Fluids 4, 2515–2522. (doi:10.1063/1.858438).DOI
  21. [21]Kerswell RR. 1993 Elliptical instabilities of stratified, hydromagnetic waves. Geophys. Astrophys. Fluid Dyn. 71, 105–143. (doi:10.1080/03091929308203599).DOI
  22. [22]Le Bars M, Le Dizès S. 2006 The thermo-elliptical instability in a rotating cylindrical shell. J. Fluid Mech. 563, 189–198. (doi:10.1017/S0022112006001674).DOI
  23. [23]Vidal J, Cébron D, Ud-Doula A, Alecian E. 2019 Fossil field decay due to nonlinear tides in massive binaries. Astron. Astrophys. 629, A142. (doi:10.1051/0004-6361/201935658).DOI
  24. [24]Cartan ME. 1922 Sur les petites oscillations d’une masse fluide. Bull. Sci. Math. 46, 317–369.
  25. [25]Poincaré H. 1885 Sur l’équilibre d’une masse fluide animée d’un mouvement de rotation. Acta Math. 7, 259–380.
  26. [26]Rieutord M, Georgeot B, Valdettaro L. 2000 Wave attractors in rotating fluids: a paradigm for ill-posed Cauchy problems. Phys. Rev. lett. 85, 4277. (doi:10.1103/PhysRevLett.85.4277).DOI
  27. [27]Vandermeighen S. 2014 Inertial modes in a rotating triaxial ellipsoid. Proc. R. Soc. A 470, 20140093. (doi:10.1098/rspa.2014.0093).
  28. [28]Ivers DJ. 2017 Enumeration, orthogonality and completeness of the incompressible Coriolis modes in a tri-axial ellipsoid. Geophys. Astrophys. Fluid Dyn. 111, 333–354. (doi:10.1080/03091929.2017.1330412).DOI
  29. [29]Backus G, Rieutord M. 2017 Completeness of inertial modes of an incompressible inviscid fluid in a corotating ellipsoid. Phys. Rev. E 95, 053116. (doi:10.1103/PhysRevE.95.053116).DOI
  30. [30]Veronis G. 1970 The analogy between rotating and stratified fluids. Annu. Rev. Fluid Mech. 2, 37–66. (doi:10.1146/annurev.fl.02.010170.000345).DOI
  31. [31]Rieutord M, Noui K. 1999 On the analogy between gravity modes and inertial modes in spherical geometry. Eur. Phys. J. B 9, 731–738. (doi:10.1007/s100510050818).DOI
  32. [32]Friedlander S, Siegmann WL. 1982 Internal waves in a rotating stratified fluid in an arbitrary gravitational field. Geophys. Astrophys. Fluid Dyn. 19, 267–291. (doi:10.1080/03091928208208959).DOI
  33. [33]Brunet M, Dauxois T, Jülicher F, Favier B. 2019 Linear and nonlinear regimes of an inertial wave attractor. Phys. Rev. Fluids 4, 034801. (doi:10.1103/PhysRevFluids.4.034801).DOI
  34. [34]Favier B, Le Dizès S. 2024 Inertial wave super-attractor in a truncated elliptic cone. J. Fluid Mech 980, A6. (doi:10.1017/jfm.2024.5).DOI
  35. [35]Maas LRM, Benielli D, Sommeria J, Lam FPA. 1997 Observation of an internal wave attractor in a confined, stably stratified fluid. Nature 388, 557–561. (doi:10.1038/41509).DOI
  36. [36]Pacary C, Dauxois T, Erm anyuk E, Metz P, Moulin M, Joubaud S. 2023 Observation of inertia-gravity wave attractors in an axisymmetric enclosed basin. Phys. Rev. Fluids 8, 104802. (doi:10.1103/PhysRevFluids.8.104802).DOI
  37. [37]Colin De Verdière Y, Saint-Raymond L. 2020 Attractors for two-dimensional waves with homogeneous Hamiltonians of degree 0. Commun. Pure Appl. Math. 73, 421–462. (doi:10.1002/cpa.21845).DOI
  38. [38]Allen JS. 1971 Some aspects of the initial-value problem for the inviscid motion of a contained, rotating, weakly-stratified fluid. J. Fluid Mech. 46, 1–23. (doi:10.1017/S0022112071000363).DOI
  39. [39]Friedlander S, Siegmann WL. 1982 Internal waves in a contained rotating stratified fluid. J. Fluid Mech. 114, 123–156. (doi:10.1017/S002211208200007X).DOI
  40. [40]Colin de Verdière Y, Vidal J. 2023 The spectrum of the Poincaré operator in an ellipsoid. arxiv:2305.01369. (doi:10.48550/arXiv.2305.01369).DOI
  41. [41]Spiegel EA, Veronis G. 1960 On the Boussinesq approximation for a compressible fluid. Astrophys. J. 131, 442–447. (doi:10.1086/146849).DOI
  42. [42]Noir J, Cébron D. 2013 Penetration-driven flows in non-axisymmetric ellipsoids. J. Fluid Mech. 737, 412–439. (doi:10.1017/jfm.2013.524).
  43. [43]Lam H, Delache A, Godeferd FS. 2023 Supply mechanisms of the geostrophic mode in rotating turbulence: interactions with self, waves and eddies. J. Fluid Mech. 971, A10. (doi:10.1017/jfm.2023.644).DOI
  44. [44]Le Reun T, Favier B, Barker AJ, Le Bars M. 2017 Inertial wave turbulence driven by elliptical instability. Phys. Rev. Lett. 119, 034502. (doi:10.1103/PhysRevLett.119.034502).DOI
  45. [45]Park J, Billant P. 2013 Instabilities and waves on a columnar vortex in a strongly stratified and rotating fluid. Phys. Fluids 25. (doi:10.1063/1.4816512).
  46. [46]Le Dizès S. 2008 Inviscid waves on a Lamb–Oseen vortex in a rotating stratified fluid: consequences for the elliptic instability. J. Fluid Mech. 597, 283–303. (doi:10.1017/S0022112007009780).DOI
  47. [47]Friedlander S. 1985 Internal oscillations in the Earth’s fluid core. Geophys. J. Int. 80, 345–361. (doi:10.1111/j.1365-246X.1985.tb05999.x).DOI
  48. [48]Grubb G. 2008 Distributions and operators. Graduate Texts in Mathematics. Springer, New York (USA): Springer.
  49. [49]Faure F. 2023 Manifestation of the topological index formula in quantum waves and geophysical waves. Ann. Henri Lebesgue 6, 449–492. (doi:10.5802/ahl.169).DOI
  50. [50]Venaille A, Onuki Y, Perez N, Leclerc A. 2023 From ray tracing to waves of topological origin in continuous media. SciPost Phys. 14, 062. (doi:SciPostPhys.14.4.062).DOI
  51. [51]Colin de Verdière Y. 2020 Spectral theory of pseudodifferential operators of degree 0 and an application to forced linear waves. Anal. PDE 13, 1521–1537. (doi:10.2140/apde.2020.13.1521).DOI
  52. [52]Leray J. 1934 Sur le mouvement d’un liquide visqueux emplissant l’espace. Acta Math. 63, 193–248.DOI
  53. [53]Foias C, Manley O, Rosa R, Temam R. 2001 Navier-Stokes equations and turbulence. Cambridge, UK: Cambridge University Press.DOI
  54. [54]Teed RJ, Dormy E. 2023 Solenoidal force balances in numerical dynamos. J. Fluid Mech. 964, A26. (doi:10.1017/jfm.2023.332).DOI
  55. [55]Barston EM. 1967 Eigenvalue Problem for Lagrangian Systems. II. J. Math. Phys. 8, 1886–1892. (doi:10.1063/1.1705433).DOI
  56. [56]Valette B. 1989 Étude d’une classe de problèmes spectraux. C. R. Acad. Sci. Paris 309, 785–788.
  57. [57]Wouk A. 1966 A note on square roots of positive operators. SIAM Rev. 8, 100–102. (doi:10.1137/1008008).DOI
  58. [58]Dintrans B, Rieutord M, Valdettaro L. 1999 Gravito-inertial waves in a rotating stratified sphere or spherical shell. J. Fluid Mech. 398, 271–297. (doi:10.1017/S0022112099006308).DOI
  59. [59]Barston EM. 1967 Eigenvalue problem for Lagrangian systems. J. Math. Phys. 8, 523–532. (doi:10.1063/1.1705227).DOI
  60. [60]Axler S, Shin PJ. 2018 The Neumann problem on ellipsoids. J. Appl. Math. Comput. 57, 261–278. (doi:10.1007/s12190-017-1105-4).DOI
  61. [61]Rabitti A, Maas LRM. 2014 Inertial wave rays in rotating spherical fluid domains. J. Fluid Mech. 758, 621–654. (doi:10.1017/jfm.2014.551).DOI
  62. [62]Ivers DJ, Jackson A, Winch D. 2015 Enumeration, orthogonality and completeness of the incompressible Coriolis modes in a sphere. J. Fluid Mech. 766, 468–498. (doi:10.1017/jfm.2015.27).DOI
  63. [63]Lebovitz NR. 1989 The stability equations for rotating, inviscid fluids: Galerkin methods and orthogonal bases. Geophys. Astrophys. Fluid Dyn. 46, 221–243. (doi:10.1080/03091928908208913).DOI
  64. [64]Vidal J, Su S, Cébron D. 2020 Compressible fluid modes in rigid ellipsoids: towards modal acoustic velocimetry. J. Fluid Mech. 885, A39. (doi:10.1017/jfm.2019.1004).DOI
  65. [65]Vidal J, Cébron D. 2020 Acoustic and inertial modes in planetary-like rotating ellipsoids. Proc. R. Soc. A 476, 20200131. (doi:10.1098/rspa.2020.0131).DOI
  66. [66]Peacock T, Weidman P. 2005 The effect of rotation on conical wave beams in a stratified fluid. Exp. Fluids 39, 32–37. (doi:10.1007/s00348-005-0955-y).DOI
  67. [67]Thomson W. 1880 On gravitational oscillations of rotating water. Proc. R. Soc. Edinburgh 10, 92–100. (doi:10.1017/S0370164600043467).
  68. [68]Lanchon N, Mora DO, Monsalve E, Cortet PP. 2023 Internal wave turbulence in a stratified fluid with and without eigenmodes of the experimental domain. Phys. Rev. Fluids 8, 054802. (doi:10.1103/PhysRevFluids.8.054802).
  69. [69]Stewartson K, Walton IC. 1976 On waves in a thin shell of stratified rotating fluid. Proc. R. Soc. Lond. A 349, 141–156. (doi:10.1098/rspa.1976.0064).DOI
  70. [70]Tauber C, Delplace P, Venaillé A. 2019 A bulk-interface correspondence for equatorial waves. J. Fluid Mech. 868, R2. (doi:10.1017/jfm.2019.233).DOI
  71. [71]Venaillle A, Delplace P. 2021 Wave topology brought to the coast. Phys. Rev. Res. 3, 043002. (doi:10.1103/PhysRevResearch.3.043002).DOI
  72. [72]Le Bars M. 2021 Numerical study of the McIntyre instability around Gaussian floating vortices in thermal wind balance. Phys. Rev. Fluids 6, 093801. (doi:10.1103/PhysRevFluids.6.093801).DOI
  73. [73]Vidal J, Cébron D. 2023 Precession-driven flows in stress-free ellipsoids. J. Fluid Mech. 954, A5. (doi:10.1017/jfm.2022.976).DOI
  74. [74]Boury S, Maurer P, Joubaud S, Peacock T, Odier P. 2023 Triadic resonant instability in confined and unconfined axisymmetric geometries. J. Fluid Mech. 957, A20. (doi:10.1017/jfm.2023.58).DOI
  75. [75]Colin de Verdière Y, Vidal J. 2024 On gravito-inertial surface waves. Preprint. (doi:hal-04461197).arxiv.org/abs/2402.12992
  76. [76]Friedlander S. 1987 Hydromagnetic waves in the Earth's fluid core. Geophys. Astrophys. Fluid Dyn. 39, 315–333. (doi:10.1080/03091928708208816).DOI
  77. [77]Malkus WVR. 1967 Hydromagnetic planetary waves. J. Fluid Mech. 28, 793–802. (doi:10.1017/S0022112067002447).DOI
  78. [78]Gerick F, Jault D, Noir J, Vidal J. 2020 Pressure torque of torsional Alfvén modes acting on an ellipsoidal mantle. Geophys. J. Int. 222, 338–351. (doi:10.1093/gji/ggaa166).DOI
  79. [79]Cébron D, Maubert P, Le Bars M. 2010 Tidal instability in a rotating and differentially heated ellipsoidal shell. Geophys. J. Int. 182, 1311–1318. (doi:10.1111/j.1365-246X.2010.04712.x).DOI
  80. [80]Le Reun T, Favier B, Le Bars M. 2018 Parametric instability and wave turbulence driven by tidal excitation of internal waves. J. Fluid Mech. 840, 498–529. (doi:10.1017/jfm.2018.18).DOI
  81. [81]Vidal J, Cébron D, Schaeffer N, Hollerbach R. 2018 Magnetic fields driven by tidal mixing in radiative stars. Mon. Not. R. Astron. Soc. 475, 4579–4594. (doi:10.1093/mnras/sty080).DOI
  82. [82]Onuki Y, Joubaud S, Dauxois T. 2023 Breaking of internal waves parametrically excited by ageostrophic anticyclonic instability. J. Phys. Oceanogr. 53, 1591–1613. (doi:10.1175/JPO-D-22-0152.1).DOI

Paper details

vlnc:W_pvq9dadu
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents