betaFeedback

WorksPhysics and Astronomy

Commutative Coarse-graining on the Sphere

Abstract

Filtering by spatial convolution underlies coarse-graining and Large Eddy Simulation, the analysis of multiscale dynamics, and data processing across the sciences. Yet, as commonly practiced on the sphere, it violates a basic consistency requirement: commutation with differential operators, which is necessary for deriving coarse-grained dynamics and preserving the physics across lengthscales. The root of the problem is nonlocality of filtering, which compares directions in different tangent planes. The same reason underlies the Foucault pendulum's precession, and it is why the commutation failure is intrinsically geometric rather than being due to the coordinate system or numerical grid. A recently proposed generalized spherical convolution is, for a given scalar kernel, the unique linear operator commuting with the radial projection, divergence, and curl. Any consistent linear operator, be it spectral truncation, wavelet transform, diffusion, Green's-function inversion, or a machine-learned convolution, necessarily takes this form. From a simulation of an unstable jet, we demonstrate commutation to within numerical round-off, whereas common filtering that ignores curvature produces significant spurious divergence where none exists and gradient errors comparable to the gradients themselves.

Introduction. –

A simple everyday observation already reveals the length scales at which we experience Earth’s curvature: standing on a beach at sea level, the visible horizon is only about 5 km away. This is a reminder that “flat space” intuition can already fail on O(1−10)O(1-10) km scales. While many general circulation models can be regarded as Large Eddy Simulations (LES) in spirit, there has, until recently, been no rational framework for deriving LES equations (in physical space) for flows on the globe. The difficulty stems from the fact that the standard convolution on the 2-dimensional (2D) spherical surface S2S^2 does not commute with spatial derivatives, a property essential for preserving the physical character of flows across scales (e.g., the divergence-free or vortical character). The main goal of this paper is to demonstrate this framework, developed in [1]. As we shall see, the problem is considerably more subtle than in Euclidean (or flat) spaces and, in places, runs counter to intuition from flat geometry. The generalized convolution of [1] is uniquely determined by commutation with radial projection, divergence, and curl: any linear operator compatible with them necessarily reduces to it. Below, we demonstrate the commutation numerically for the first time and quantify the large errors that arise from common filtering practice. Applications span geophysics, planetary and solar physics, astrophysics, cosmology, and inertial confinement fusion, as well as inverse problems and data assimilation on the sphere.

The applied mathematics literature on spherical harmonic analysis, convolutions, wavelets, and the Helmholtz decomposition of vectors and tensors on S2S^2 is vast (see [1]); while our work builds on or overlaps with some of those studies, a key difference is our interest in ensuring that filtering commutes with differential operators on S2S^2, an issue that, to our knowledge, had not been addressed prior to [1]. Indeed, many of these methods, including spectral truncation, wavelet transforms, diffusion-based smoothing, Green’s-function inversions such as the inverse Laplacian, and the rotation-equivariant layers of neural networks with learned kernels (e.g. [2]) are themselves spherical convolutions with particular kernels, so the question of how they extend consistently to vectors and higher rank tensors is answered, uniquely, by the construction of [1] as explained below.

As a matter of terminology, we distinguish between Cartesian coordinates and Euclidean space. One may choose non-Cartesian coordinates in Euclidean space, e.g., polar coordinates in R2\mathbb{R}^{2}. Alternatively, one may use Cartesian coordinates while being restricted to a curved surface, such as that of a sphere embedded in flat R3\mathbb{R}^{3}.

One might contend that, having modeled flows on spherical grids of varying resolution for decades, we already “know” how to write the governing equations at different scales. However, in traditional numerical modeling, resolution dependence enters the equations primarily through diffusive coefficients scaled with the grid spacing, whether empirically or via closure assumptions. What these equations lack is a parameter ℓ\ell associated with an operator that specifies which scales of the exact solution the simulated fields represent. Running two simulations at different resolutions does not, by itself, provide a one-to-one correspondence between the scale content of the two solutions. Some form of filtering or projection must be performed on both the high-resolution solution and the governing equations. One may suggest projecting onto spherical harmonics; this suffers from two problems: (i) spherical harmonics are not a basis for regional models or for the global ocean with land boundaries, and (ii) even for the global atmosphere, which variables of the solution should we project? The most obvious choice may be the meridional and zonal (northward and eastward) velocity components but, as we explain below, this is erroneous. A general method was provided in [1] that identifies the precise manner in which one must project the solution to ensure that it satisfies the coarse-grained equations on the sphere with spectral truncation being a special case.

While the problem of commutation with derivatives due to non-uniform numerical grid discretization or the presence of boundaries has received attention in LES (e.g. [3]), the problem due to curvature does not seem to have been appreciated. A partial exception is the work by Jordan [4], which makes uncontrolled approximations to reduce the problem to filtering in grid space (or token space, or logical space). In fact, this is how convolutions are performed in “token space” in recent machine learning (ML) applications (e.g. [5, 6, 7, 8]). The downside of such an approach is its dependence on a specific representation (e.g., a choice of coordinate system), which does not satisfy the underlying symmetries or guarantee generalizability. For ML applications, this is akin to approximating sin⁡(x)\sin(x) by polynomials: over a fixed interval x∈[a,b]x \in[a,b], such an approximation converges to arbitrary precision with a sufficiently high-order polynomial (by the Weierstrass approximation theorem), yet it fails outside the “fitting” interval [a,b][a,b]. Indeed, such failure motivated [2] to use global spherical harmonics for ML application. Even there, however, vector fields such as winds are treated as scalar channels and are filtered componentwise (eq. (2)).

Spherical Convolution of Vectors. –

As Figure 1 shows, convolutions, and thereby filtering, on the sphere are accomplished by rotating the kernel, not translating it. For an axisymmetric kernel GG (about the kernel’s center, also called a “zonal” kernel in harmonic analysis [1], e.g., the spherical red cap in Figure 1 or the Gaussian in Figure 5a), the spherical convolution of a scalar field ff can be written as

Three spherical filtering panels showing a kernel before and after rotation or translation

Figure 1. Left panel shows a filtering kernel in red, e.g., a top-hat function. Convolutions (and equivariant filtering) on the sphere are accomplished by rotating the kernel (middle panel): rotations play the role on the sphere that translations play in flat space as the underlying symmetry group. Although the sphere is embedded in 3D Euclidean space (shown here in units of radius aa), translating the kernel (right panel) is ill-defined: the translated kernel does not remain on the spherical surface.

Figure 5. Panel 5a: normalized Gaussian kernel GG (width ≈1000\approx1000 km), shown in dimensionless form aE2Ga_E^2 G. The generalized convolution G∗⟶vG\overset{\longrightarrow}{\ast}\mathbf{v} is evaluated two ways: panel 5c via the Edmonds basis (eq. (14)), and panel 5d via the Helmholtz scalars (eq. (16)). That both yield ∇3⋅(G∗⟶v)=0\nabla_3\cdot(G\overset{\longrightarrow}{\ast}\mathbf{v}) = 0 to within round-off is nontrivial and demonstrates the results in [1]. Panel 5d values are even smaller than those in Figure 4c because of filtering. In contrast, the ambient convolution (panel 5b) yields a spurious divergence roughly eighteen orders of magnitude larger.

G⊛f(x):=∫S2G(x⋅ya2)f(y) dA(y),(1)G \circledast f(\mathbf{x}) := \int_{\mathbb{S}^{2}} G\left(\frac{\mathbf{x}\cdot\mathbf{y}}{a^{2}}\right) f(\mathbf{y})\,dA(\mathbf{y}), \tag*{(1)}

where x,y∈S2\mathbf{x},\mathbf{y} \in\mathbb{S}^{2} are position vectors, x⋅y/a2\mathbf{x}\cdot\mathbf{y}/a^{2} is the cosine of the geodesic angle between them, and dAdA is the area element on S2\mathbb{S}^{2} of radius aa. Definition (1) and more general SO(3)SO(3) formulations are well known in the applied mathematics community and are routinely used in applications (see [1]). However, fluid dynamics studies prior to [1, 9], including those on subgrid modeling, do not seem to have appreciated that convolutions on the sphere are accomplished by rotations, utilizing instead the traditional flat convolution based on kernel translations or on averaging within grid cells. A reason may be that LES applications to geophysical flows, such as in the atmospheric boundary layer and subscale oceanic flows, have focused on systems of small spatial extent, O(1−10)O(1-10) km, over which Earth’s curvature is less important (e.g., [10, 11]).

Spherical Convolution of Vectors. –

In Euclidean space, the convolution of vectors and higher-rank tensors is a straightforward extension of the convolution of scalars; it is accomplished by convolving each of the tensor’s Cartesian components. Unfortunately, many studies in geophysics, and more recently in ML-based modeling, have relied on some form of local averaging or componentwise processing of the spherical vector components (i.e., the eastward, northward, and radial components) (e.g. [5, 6, 7, 8]). Treating the spherical basis vectors e^r,e^λ,e^ϕ\hat{\mathbf{e}}_{r},\hat{\mathbf{e}}_{\lambda},\hat{\mathbf{e}}_{\phi} (radial, zonal, and meridional, with λ\lambda denoting longitude and ϕ\phi latitude) as if they were spatially constant neglects both the metric terms arising from their spatial variation and, more fundamentally, the intrinsic curvature of S2\mathbb{S}^{2}. We shall call a convolution acting on the spherical vector components a componentwise convolution,

G⊛v:=(G⊛vr)e^r+(G⊛uλ)e^λ+(G⊛uϕ)e^ϕ,(2)G \circledast\mathbf{v} := (G \circledast v_{r})\hat{\mathbf{e}}_{r} + (G \circledast u_{\lambda})\hat{\mathbf{e}}_{\lambda} + (G \circledast u_{\phi})\hat{\mathbf{e}}_{\phi}, \tag*{(2)}

since it acts on each spherical component separately as if it were a scalar, without attempting to respect geometry. Here, vrv_{r}, uλu_{\lambda}, uϕu_{\phi} are the spherical vector components of v\mathbf{v}. The schematic in Figure 2 along with a numerical example in Figure 7 shows how the componentwise convolution fails to respect the very notion of lengthscale, let alone commuting with differential operators. It does not even respect the underlying rotational symmetry (it is not rotation-equivariant), since e^λ,e^ϕ\hat{\mathbf{e}}_{\lambda},\hat{\mathbf{e}}_{\phi} are tied to a fixed choice of pole.

Original zonal flow and componentwise convolution on a sphere

Figure 2. Componentwise convolution, i.e., locally averaging the spherical components of a vector field (eq. (2)), fails to decompose scales correctly on the sphere. In this example, the flow is purely zonal (eastward) with constant magnitude ∣u∣=c0|\mathbf{u}| = c_0 everywhere. On a flat manifold, such a flow would be constant, with no lengthscale except the largest possible (that of the domain). On the sphere, however, the flow has inherent gradients and, therefore, lengthscales: material elements are deformed, since the equatorial flow takes longer to complete a revolution than the flow at higher latitudes, and vorticity diverges toward the poles. Yet locally averaging the spherical components within the kernel’s footprint returns the field unaltered regardless of kernel size, incorrectly implying the absence of lengthscales smaller than the domain size.

Four spherical plots comparing Edmonds, ambient, Helmholtz, and componentwise convolutions

Figure 7. Panel (7a): the tangent gradient vector, filtered using the generalized convolution by an Edmonds transformation, remains tangent to within round-off. In contrast, (7b) shows that the ambient convolution yields a spurious radial component eleven orders of magnitude larger than (7a), reaching ∼20%\sim20\% of the gradient magnitude in (7c). Panel (7c): meridional component of the generalized convolution of the height gradient, evaluated via the Helmholtz scalars. Panel (7d): commutation error, {∇(G⊛h)}⋅e^ϕ−[G⊛∇h]⋅e^ϕ\left\{\nabla(G \circledast h)\right\}\cdot\hat{e}_{\phi} - [G \circledast\nabla h]\cdot\hat{e}_{\phi}, of the componentwise convolution in the meridional gradient, comparable to the magnitude of the gradient itself in (7c).

It is instructive to consider the purely zonal (or eastward) flow u=uλe^λ\mathbf{u}=u_{\lambda}\hat{\mathbf{e}}_{\lambda} in Figure 2, which has speed uλ=c0u_{\lambda}=c_{0} (in m/s) that is spatially uniform on S2\mathbb{S}^{2}. A componentwise

Spherical convolution.

convolution of this flow with a low-pass kernel is essentially a local averaging of uλ=c0u_\lambda= c_0, which yields the same flow unaltered. Therefore, a componentwise convolution sees the flow in Figure 2 as spatially uniform, the largest scale possible, whereas the flow is multiscale as evidenced by the deformation experienced by fluid particles and the intense vorticity near the poles (Figure 2).

For a laterally non-divergent (or solenoidal) flow as in Figure 2, [1] showed that the ambient convolution preserves the physical character of the flow, including its tangential and non-divergent properties. The ambient convolution treats the field as an object in the ambient flat space R3\mathbb{R}^{3}, performing a local average of the vector field on S2\mathbb{S}^{2} by spherical convolution of the Cartesian vector components:

G⊛u:=(G⊛u1)ı^+(G⊛u2)ȷ^+(G⊛u3)k^,(3)G \circledast\mathbf{u} := (G \circledast u_{1})\hat{\imath} + (G \circledast u_{2})\hat{\jmath} + (G \circledast u_{3})\hat{k}, \tag*{(3)}

where u1,u2,u3u_{1},u_{2},u_{3} are the Cartesian components of u\mathbf{u} and ı^,ȷ^,k^\hat{\imath},\hat{\jmath},\hat{k} are the fixed Cartesian unit vectors in R3\mathbb{R}^{3}. The ambient convolution exactly accounts for the spatial variation

of the spherical basis vectors, which the componentwise convolution (eq. (2)) ignores. Note, however, that the absence of explicit metric terms is because curvature information is contained in the Cartesian components themselves. The ambient convolution does not account for the intrinsic curvature. This is because by going to the ambient flat space R3\mathbb{R}^{3}, averaging of vectors tangent to S2\mathbb{S}^{2} at different points is not, in general, tangent at the target point. This is why, for general flows that include a divergent component (e.g. Figure 3), the ambient convolution does not commute with differential operators, e.g.,

Original tangential flow and ambient convolution near the pole

Figure 3. Consider a flow that is tangent to the spherical surface at three grid points along a great circle (the equator here), where the velocity vector has equal magnitude at two adjacent points and vanishes at the third. In this case, an ambient convolution (eq. (3)) that treats the vector field in the ambient space R3\mathbb{R}^{3}, equivalent to averaging the Cartesian components, yields a locally-averaged velocity field at the center-point that is no longer tangent to the surface but has a radial component. Therefore, the ambient convolution, despite exactly accounting for the basis vector variation (the source of metric terms), still changes the physical properties of the field.

G⊛(∇f)≠∇(G⊛f),G⊛(∇⋅u)≠∇⋅(G⊛u).(4)G \circledast(\nabla f) \ne\nabla(G \circledast f), \qquad G \circledast(\nabla\cdot\mathbf{u}) \ne\nabla\cdot(G \circledast\mathbf{u}). \tag*{(4)}

Numerical examples in Figure 5,Figure 7 show how the commutation failure of eq. (4) can lead to tremendous errors, exceeding O(1011)O(10^{11}), as we discuss below. A related issue concerns purely radial and purely tangent vector fields, fe^rf\hat{e}_{r} and u=uλe^λ+uϕe^ϕ\mathbf{u}=u_{\lambda}\hat{e}_{\lambda}+u_{\phi}\hat{e}_{\phi}, respectively, which do not remain radial or tangent when convolved according to (3),

G⊛(fe^r)≠(G⊛f)e^r,G⊛(u×e^r)≠(G⊛u)×e^r.(5)G \circledast(f\hat{e}_{r}) \ne(G \circledast f)\hat{e}_{r}, \qquad G \circledast(\mathbf{u}\times\hat{e}_{r}) \ne(G \circledast\mathbf{u})\times\hat{e}_{r}. \tag*{(5)}

Figure 3 illustrates the failure in eq. (5). The flow there is tangent to S2\mathrm{S}^{2} at three grid points along a great circle, with equal velocity magnitude at the middle point and one neighbor, and vanishing at the other. In this case, an ambient convolution that treats the vector fields in the ambient space R3\mathbb{R}^{3} yields a locally averaged velocity at the middle point that is no longer tangent to S2\mathrm{S}^{2} but acquires a radial component. Even the ambient convolution, which accounts for the metric terms and respects rotational symmetry, fails to preserve the field’s physical character and is ill-suited to derive tractable coarse-grained PDEs on S2\mathrm{S}^{2}.

Root of the Problem. {#sec:Root of the Problem}

The problem lies in the nonlocality of filtering, which compares vectors in different tangent planes. There is no path-independent way to compare directions on an intrinsically curved surface. In contrast, convolutions in Euclidean space commute with derivatives because the convolution is translation-equivariant and derivatives are the generators of translations (esddxf(x)=f(x+s)e^{s\frac{d}{dx}}f(x)=f(x+s)) (e.g. [12]). Commutation carries through to the gradient, divergence, and curl, since these involve only the generators and a parallel basis (the same at every point), such as ı^,ȷ^,k^\hat{\imath},\hat{\jmath},\hat{k}, both of which pass through the convolution. Such a basis exists because in flat space a vector can be transported from one point to another unambiguously, so directions at different points can be compared. On the sphere, no such basis exists. Suppose you hold a spear pointing north at the equator and walk straight north up the 0∘0^{\circ} meridian to the pole, where the spear points toward 180∘E180^{\circ}\mathrm{E}. Now walk sideways (without turning) down the 90∘E90^{\circ}\mathrm{E} meridian to the equator, then backward along the equator to where you began. Although all three legs are geodesics and you never turned the spear, it ends up pointing east. The same reason underlies the Foucault pendulum’s precession, and it is why the commutation failure is intrinsically geometric: it reflects the curvature of the underlying manifold, not the choice of coordinates or numerical grid. One way to see this is from an easily verifiable (perhaps not widely appreciated) fact: the Laplacian Δ\Delta on S2\mathrm{S}^{2}, which is a (high-pass) filter, does not generally commute with spatial derivatives, e.g. [Δ,∇]=Δ∇−∇Δ≠0[\Delta,\nabla]=\Delta\nabla-\nabla\Delta\ne0, a coordinate-invariant statement. The Laplacian on S2\mathrm{S}^{2}

Δf=1a2cos⁡ϕ∂∂ϕ(cos⁡ϕ∂f∂ϕ)+1a2cos⁡2ϕ∂2f∂λ2(6)\Delta f=\frac{1}{a^{2}\cos\phi}\frac{\partial}{\partial\phi}\left(\cos\phi\frac{\partial f}{\partial\phi}\right)+\frac{1}{a^{2}\cos^{2}\phi}\frac{\partial^{2}f}{\partial\lambda^{2}} \tag*{(6)}

for a scalar function f(x)f(\mathbf{x}). The tangent gradient is

∇f(ϕ,λ)=e^λ1acos⁡ϕ∂f∂λ+e^ϕ1a∂f∂ϕ.(7)\nabla f(\phi,\lambda)=\hat{\mathbf{e}}_{\lambda}\frac{1}{a\cos\phi}\frac{\partial f}{\partial\lambda}+\hat{\mathbf{e}}_{\phi}\frac{1}{a}\frac{\partial f}{\partial\phi}. \tag*{(7)}

Throughout, ∇\nabla, ∇⋅\nabla\cdot, and Δ\Delta without subscript ‘3’ denote intrinsic operators on S2\mathrm{S}^{2}. It is easy to check that Δ∇f=∇Δf−2a2e^rΔf\Delta\nabla f=\nabla\Delta f-\frac{2}{a^{2}}\hat{\mathbf{e}}_{r}\Delta f is not a tangent vector field in general, and has a radial component2. On the other hand, ∇Δf\nabla\Delta f is necessarily tangent to S2\mathrm{S}^{2}. This is why diffusion of a vector field on the sphere requires using the vector (Hodge–de Rham) Laplacian on S2\mathrm{S}^{2}, Δvec\Delta^{\mathrm{vec}}, which maintains the field’s geometric properties [13] and commutes with the gradient, [Δvec,∇]=0[\Delta^{\mathrm{vec}},\nabla]=0. On curved surfaces, the vector Laplacian is inherently distinct from the scalar Laplacian; it mixes vector components and involves additional metric-dependent terms arising from curvature. In Euclidean settings with Cartesian basis, the vector Laplacian reduces to the componentwise scalar Laplacian since there is no curvature. Note that the absolute (but not relative) commutation error [Δ,∇]f=−2a2e^rΔf[\Delta,\nabla]f=-\frac{2}{a^{2}}\hat{\mathbf{e}}_{r}\Delta f comes mostly from the smallest scales of f(x)f(\mathbf{x}). The generalization of convolutions to vectors and tensors in [1] parallels the need to define different Laplacians (filters) for tensors of different ranks.

From a symmetry perspective, the counterpart of translation on S2\mathrm{S}^{2} is rotation: an axisymmetric kernel yields a rotation-equivariant convolution, G⊛(f∘R)=(G⊛f)∘RG\circledast(f\circ\mathcal{R})=(G\circledast f)\circ\mathcal{R} for all rotations R∈SO(3)\mathcal{R}\in SO(3). Equivariance guarantees commutation of G⊛G\circledast (eq. (3)) with the generators of rotations (angular momentum operators x×∇3\mathbf{x}\times\nabla_{3}, where ∇3\nabla_{3} is the gradient in the ambient R3\mathbb{R}^{3}) [12], and hence with the scalar Laplacian, since (x×∇3)2(\mathbf{x}\times\nabla_{3})^{2} is a2Δa^{2}\Delta when restricted to S2\mathrm{S}^{2}. Commutation with the generators is also why the ambient convolution suffices for laterally unidirectional flows (eq. (13)) [1]. However, equivariance is not sufficient to guarantee commutation with the gradient or divergence on S2\mathrm{S}^{2}, which, unlike their Euclidean counterparts, involve the generators and the spatially varying basis vector e^r=x/a\hat{\mathbf{e}}_{r}=\mathbf{x}/a. The latter does not pass through the ambient convolution; instead, it shifts harmonic degree (eqs. (11),{xref}(12)`).

To further elucidate the failure of the ambient convolution (3) on S2\mathrm{S}^{2}, consider first a 2D flat plane embedded in R3\mathbb{R}^{3}, in which the Fourier scalar basis function eik⋅xe^{i\mathbf{k}\cdot\mathbf{x}} is an eigenfunction of the traditional flat convolution:

G∗eik⋅x=G^(k)eik⋅x,(8)G*e^{i\mathbf{k}\cdot\mathbf{x}}=\widehat{G}(\mathbf{k})e^{i\mathbf{k}\cdot\mathbf{x}}, \tag*{(8)}

where ∗* denotes the traditional flat (translational) convolution and G^(k)\widehat{G}(\mathbf{k}) is the Fourier transform of the kernel. The result carries over trivially to the Fourier vector basis functions eik⋅xe^me^{i\mathbf{k}\cdot\mathbf{x}}\hat{\mathbf{e}}_{m}, where e^m\hat{\mathbf{e}}_{m} represents any of ı^,ȷ^,k^\hat{\imath},\hat{\jmath},\hat{k}:

G∗eik⋅xe^m=G^(k)eik⋅xe^m,(9)G*e^{i\mathbf{k}\cdot\mathbf{x}}\hat{\mathbf{e}}_{m}=\widehat{G}(\mathbf{k})e^{i\mathbf{k}\cdot\mathbf{x}}\hat{\mathbf{e}}_{m}, \tag*{(9)}

because Cartesian unit vectors e^m\hat{\mathbf{e}}_{m} are spatially constant. On the sphere S2\mathrm{S}^{2}, the scalar spherical harmonic basis function Yn,j(λ,ϕ)Y_{n,j}(\lambda,\phi), of degree nn and order jj, is similarly an eigenfunction of the spherical convolution (eq. (1)),

G⊛Yn,j=G^(n)Yn,j,(10)G\circledast Y_{n,j}=\widehat{G}(n)Y_{n,j}, \tag*{(10)}

where G^(n)\widehat{G}(n) is the Legendre coefficient of degree nn of the axisymmetric kernel GG. Since the Laplacian is also diagonal on Yn,jY_{n,j}, with eigenvalue −n(n+1)/a2-n(n+1)/a^{2}, the spherical convolution commutes with it, [Δ,G⊛]=0[\Delta,G\circledast]=0, a property that the generalized convolution below also satisfies. However, the vector spherical harmonics basis Yn,j:=Yn,je^r\mathbf{Y}_{n,j}:=Y_{n,j}\hat{\mathbf{e}}_r, Φn,j:=a∇Yn,j/n(n+1)\mathbf{\Phi}_{n,j}:=a\nabla Y_{n,j}/\sqrt{n(n+1)}, Ψn,j:=ae^r×∇Yn,j/n(n+1)\mathbf{\Psi}_{n,j}:=a\hat{\mathbf{e}}_r\times\nabla Y_{n,j}/\sqrt{n(n+1)} are not all eigenfunctions of G⊛G\circledast (eq. (3)). To see this, define Yn,joutY_{n,j}^{\mathrm{out}} and Yn,jinY_{n,j}^{\mathrm{in}} as harmonic extensions of Yn,jY_{n,j} that decay over r>ar>a and are regular over r<ar<a, respectively. The Edmonds basis [14], Kn,j(i)\mathbf{K}_{n,j}^{(i)} for i=1,2,3i=1,2,3, decomposes any vector field on S2S^2 uniquely into potential fields that are source-free outside (Kn,j(1)∝∇3Yn,jout∣r=a)(\mathbf{K}_{n,j}^{(1)}\propto\nabla_3Y_{n,j}^{\mathrm{out}}|_{r=a}) or inside (Kn,j(2)∝∇3Yn,jin∣r=a)(\mathbf{K}_{n,j}^{(2)}\propto\nabla_3Y_{n,j}^{\mathrm{in}}|_{r=a}) the sphere, and non-divergent (solenoidal) fields on S2S^2 (Kn,j(3)∝x×∇3Yn,jin∣r=a)(\mathbf{K}_{n,j}^{(3)}\propto\mathbf{x}\times\nabla_3Y_{n,j}^{\mathrm{in}}|_{r=a}). Their key property is that the Cartesian components of Kn,j(i)\mathbf{K}_{n,j}^{(i)} are scalar harmonics of a single degree, deg⁡(i)=n+1,n−1,n\deg(i)=n+1,n-1,n for i=1,2,3i=1,2,3 [1]. By eq. (10), each is then an eigenfunction of G⊛G\circledast,

G⊛Kn,j(i)=G^(deg⁡(i))Kn,j(i),(11)G\circledast\mathbf{K}_{n,j}^{(i)}=\widehat{G}(\deg(i))\mathbf{K}_{n,j}^{(i)}, \tag*{(11)}

but with degree-shifted eigenvalues for i=1,2i=1,2. The Edmonds basis is related to the canonical basis by a unitary map, which we call the Edmonds transformation [1],

(Kn,j(1)Kn,j(2)Kn,j(3))=12n+1(n+1−n0nn+10002n+1)(Yn,jΦn,jΨn,j)(12)\begin{pmatrix} \mathbf{K}_{n,j}^{(1)}\\ \mathbf{K}_{n,j}^{(2)}\\ \mathbf{K}_{n,j}^{(3)} \end{pmatrix} = \frac{1}{\sqrt{2n+1}} \begin{pmatrix} \sqrt{n+1} & -\sqrt{n} & 0\\ \sqrt{n} & \sqrt{n+1} & 0\\ 0 & 0 & \sqrt{2n+1} \end{pmatrix} \begin{pmatrix} \mathbf{Y}_{n,j}\\ \mathbf{\Phi}_{n,j}\\ \mathbf{\Psi}_{n,j} \tag*{(12)} \end{pmatrix}

In the Edmonds basis, G⊛(⋅)G\circledast(\cdot) is diagonal (eq. (11)). Conjugating G⊛(⋅)G\circledast(\cdot) by the Edmonds transformation (eq. (12)) shows that G⊛Yn,jG\circledast\mathbf{Y}_{n,j} and G⊛Φn,jG\circledast\mathbf{\Phi}_{n,j} are degree-dependent mixtures of Yn,j\mathbf{Y}_{n,j} and Φn,j\mathbf{\Phi}_{n,j} with coefficients involving G^(n±1)\widehat{G}(n\pm1) (explicit expressions in [1]): ambient convolution of a purely radial field acquires a tangential component and vice versa, which explains Figure 3 spectrally. Only the solenoidal basis function is unaltered, Ψn,j=Kn,j(3)\mathbf{\Psi}_{n,j}=\mathbf{K}_{n,j}^{(3)}:

G⊛Ψn,j=G^(n)Ψn,j,(13)G\circledast\mathbf{\Psi}_{n,j}=\widehat{G}(n)\mathbf{\Psi}_{n,j}, \tag*{(13)}

which explains why the ambient convolution suffices for laterally non-divergent flows (Figure 2) but fails in general. The commutation requirement, which is uniquely satisfied by the generalized convolution [1], resolves the ambiguity in comparing directions on S2S^2.

Generalized Convolution. –

With the problem formulated as in the preceding discussion, the solution comes naturally: filter each Edmonds component with a spectrally shifted kernel that undoes the degree shift. Decomposing v=v(1)+v(2)+v(3)\mathbf{v}=\mathbf{v}^{(1)}+\mathbf{v}^{(2)}+\mathbf{v}^{(3)} by projection onto the Kn,j(i)\mathbf{K}_{n,j}^{(i)}, the generalized convolution is defined as [1]

G⊛→v:=∑i=13G(i)⊛v(i),(14)\overrightarrow{G\circledast}\mathbf{v}:=\sum_{i=1}^{3}G^{(i)}\circledast\mathbf{v}^{(i)}, \tag*{(14)}

where the shifted kernels are defined via their Legendre coefficients,

G^(1)(n):=G^(n−1),G^(2)(n):=G^(n+1),G(3):=G,(15)\widehat{G}^{(1)}(n):=\widehat{G}(n-1),\qquad\widehat{G}^{(2)}(n):=\widehat{G}(n+1),\qquad G^{(3)}:=G, \tag*{(15)}

so that every Kn,j(i)\mathbf{K}_{n,j}^{(i)} (and, by unitarity of (12), every Yn,j,Φn,j,Ψn,j\mathbf{Y}_{n,j},\mathbf{\Phi}_{n,j},\mathbf{\Psi}_{n,j}) is an eigenfunction of G⊛→\overrightarrow{G\circledast} with the same eigenvalue G^(n)\widehat{G}(n). Commutation with differential operators on S2S^2 follows [1]. In particular, since G⊛→\overrightarrow{G\circledast} preserves the radial, potential, and solenoidal characters of a vector field, a corollary in [1] shows that it can be evaluated without spherical harmonic transforms: for v=e^rvr+∇Φ+e^r×∇Ψ\mathbf{v}=\hat{\mathbf{e}}_r v_r+\nabla\Phi+\hat{\mathbf{e}}_r\times\nabla\Psi with radial (vr)(v_r), potential (Φ)(\Phi), and solenoidal (Ψ)(\Psi) Helmholtz scalars,

G⊛→v=(G⊛vr)e^r+∇(G⊛Φ)+e^r×∇(G⊛Ψ),(16)\overrightarrow{G\circledast}\mathbf{v}=(G\circledast v_r)\hat{\mathbf{e}}_r+\nabla(G\circledast\Phi)+\hat{\mathbf{e}}_r\times\nabla(G\circledast\Psi), \tag*{(16)}

which provides a practical method for computing generalized convolutions [15]. Below we evaluate both eq. (14) and eq. (16) numerically. The procedure in eq. (16) arguably obscures the nature of averaging vectors; {xref}eq. (14), formulated in the Edmonds basis, provides the insight that guided the counterexamples in {xref}Figs. 2,3. We emphasize that commutativity with differential operators is a property of the convolution operation itself and does not rely on any particular kernel $G$ beyond axisymmetry: admissible kernels include low-pass, high-pass, isotropic wavelets, Green’s functions (e.g., of the Laplacian, yielding its inverse), and kernels learned from data. Moreover, for a given scalar kernel, $\overrightarrow{G\circledast}$ is the unique linear operator commuting with the radial projection, divergence, and curl (see Supplementary Material, SM), and the construction extends to rank-2 tensors {citep}[1], as required for coarse-graining the subgrid stress in LES. This construction parallels the theory of Laplacians on $S^2$ but is more general: $\Delta^{\mathrm{vec}}$-based smoothing corresponds to the heat kernel alone, whereas the generalized convolution admits arbitrary axisymmetric kernels {citep}[1]. The generalized convolution stands to the ambient convolution precisely as $\Delta^{\mathrm{vec}}$ stands to the scalar Laplacian applied to the Cartesian components, which is diagonalized by the Edmonds basis with the same degree shifts as in {xref}eq. (11)`.

Numerical Results. –

We demonstrate our results using the single-layer shallow water equations (SWE) in a co-rotating frame on a sphere. The model describes the motion of an incompressible layer of fluid with a rigid lower boundary and an upper free surface. The SWE in vector-invariant form are (e.g. [16]):

∂tu+(f+ωre^r)×u=−∇(∣u∣22+gh)+νΔvecu(17)\partial_t\mathbf{u}+(f+\omega_r\hat{\mathbf{e}}_r)\times\mathbf{u}=-\nabla\left(\frac{|\mathbf{u}|^2}{2}+gh\right)+\nu\Delta^{\mathrm{vec}}\mathbf{u} \tag*{(17)}
∂th+∇⋅(hu)=κΔh(18)\partial_t h+\nabla\cdot(h\mathbf{u})=\kappa\Delta h \tag*{(18)}

where u(λ,ϕ)\mathbf{u}(\lambda,\phi) is the velocity tangent to the surface, ωr=(∇×u)⋅e^r\omega_r=(\nabla\times\mathbf{u})\cdot\hat{\mathbf{e}}_r is the scalar vorticity field normal to the surface, h(λ,ϕ)h(\lambda,\phi) is the fluid “height” (or layer thickness) field, f=2Ωsin⁡ϕ e^r\mathbf{f}=2\Omega\sin\phi\,\hat{\mathbf{e}}_r is the Coriolis vector, Ω=7.29×10−5 rad/s\Omega=7.29\times10^{-5}\,\mathrm{rad/s} is the angular velocity, g=9.81 m/s2g = 9.81\ \mathrm{m}/\mathrm{s}^{2} is the gravitational acceleration, ν=4×104 m2/s\nu= 4 \times10^{4}\ \mathrm{m}^{2}/\mathrm{s} is the viscosity, and κ=ν\kappa= \nu is the thickness diffusivity. S2\mathbb{S}^{2} has Earth radius a=aE:=6.37×106ma = a_E := 6.37 \times10^{6}\mathrm{m}. While the tangent velocity u\mathbf{u} is divergent, ∇⋅u≠0\nabla\cdot\mathbf{u} \ne0, the 3D velocity v=u+vre^r\mathbf{v} = \mathbf{u} + v_r\hat{\mathbf{e}}_r,

∇3⋅v=0,(19)\nabla_3\cdot\mathbf{v} = 0, \tag*{(19)}

where ∇3⋅\nabla_3\cdot denotes the 3D divergence in ambient space R3\mathbb{R}^{3}.

We solve eq. (17) with our pseudo-spectral code DiNu-Sphere, a spherical version of DiNuSUR [17] built on MPI-parallel spherical harmonic transforms via libsharp [18]; such spectral solvers are standard in atmospheric research [16]. Time-stepping is 4th-order explicit Runge-Kutta on a Gauss-Legendre grid with nφ=512n_\varphi= 512, nλ=1024n_\lambda= 1024, and triangular truncation nmax⁡=⌊2nφ/3⌋n_{\max} = \lfloor2n_\varphi/3\rfloor (T341).

The initial conditions for the simulation are those of a barotropically unstable zonal (eastward) flow given in [19]. The initial zonal flow u(x,t=0)\mathbf{u}(\mathbf{x},t = 0), which represents a mid-latitude jet in the troposphere, is balanced by an associated height field h(x,t=0)h(\mathbf{x},t = 0). A small-amplitude height field perturbation h′(x,t=0)h'(\mathbf{x},t = 0) is introduced to trigger the instability (see [19]).

Figure 4 shows simulation results. Kinetic energy peaks at approximately day 6 as the instability generates eddying motions. Below, we use the velocity field on day 8, visualized in Figure 4. The radial velocity in Figure 4d, constructed as vr≈−(∇⋅u) δrv_r \approx-(\nabla\cdot\mathbf{u})\,\delta r (see caption), is a proxy for the lateral divergence and shows the gravity waves; the vorticity in Figure 4b highlights the eddies. The 3D divergence in Figure 4c presents one of the benchmarks, cf. eq. (19), against which we compare the divergence of the filtered velocity below.

Simulation results showing speed, radial vorticity, 3D divergence, and radial velocity

Figure 4. Panels 4a–4d show velocity magnitude, radial vorticity, 3D divergence, and radial velocity, respectively. In 4d, vrv_r is constructed to satisfy ∇3⋅v=0\nabla_3\cdot\mathbf{v} = 0: vr(aE)=0v_r(a_E) = 0 and vr(aE+δr)≈−(∇⋅u) δrv_r(a_E + \delta r) \approx-(\nabla\cdot\mathbf{u})\,\delta r on a shell δr=10−6aE\delta r = 10^{-6}a_E above the surface. The embedding 3D Euclidean space is shown in units of aEa_E.

All convolutions below use the kernel in Figure 5a. Since the flow satisfies eq. (19), a commutative filter must return a filtered velocity that is also 3D divergence-free; this is a stringent test because the unfiltered ∇3⋅v\nabla_3\cdot\mathbf{v} is at double-precision round-off by construction, O(10−21) s−1O(10^{-21})\ \mathrm{s}^{-1} (Figure 4c). Figure 5 shows that the generalized convolution passes this test: evaluating G⊛v→\overrightarrow{G\mathbin{\circledast}\mathbf{v}} either via the Edmonds basis in eq. (14) (Figure 5c) or via the Helmholtz scalars in eq. (16) (Figure 5d) yields a divergence of O(10−22−10−20) s−1O(10^{-22}-10^{-20})\ \mathrm{s}^{-1}, i.e., at numerical accuracy. In contrast, the ambient convolution (Figure 5b) produces ∇3⋅(G⊛v)=O(10−2) s−1\nabla_3\cdot(G\circledast\mathbf{v}) = O(10^{-2})\ \mathrm{s}^{-1}: an error more than eighteen orders of magnitude larger, and, importantly, two to three orders of magnitude larger than the flow’s lateral divergence (Figure 4d) and vorticity (Figure 4b). The culprit is the “radial-lateral mixing” by the ambient convolution, as illustrated in Figure 3 and explained spectrally in eqs. (11)–(12). Since ∇3⋅\nabla_3\cdot is evaluated with impermeability, vr(aE)=0v_r(a_E) = 0, imposed at the surface, the leaked radial velocity, reaching ∼0.2 m/s\sim0.2\ \mathrm{m}/\mathrm{s}, appears as a spurious divergence ∼∣e^r⋅(G⊛v)∣/δr\sim|\hat{\mathbf{e}}_r\cdot(G\circledast\mathbf{v})|/\delta r. The componentwise convolution (cf. eq. (2)), which filters the spherical components as scalars, cannot leak into the radial direction; it yields ∇3⋅\nabla_3\cdot errors of O(10−5) s−1O(10^{-5})\ \mathrm{s}^{-1} (not shown), which are still significant and of the order of ∇⋅u\nabla\cdot\mathbf{u} (Figure 4d), though far from the ambient convolution’s catastrophic failure.

Figure 6 probes the (laterally) solenoidal flow via the radial vorticity. Since Ψn,j\mathbf{\Psi}_{n,j} is an eigenfunction of both G⊛G\circledast (eq. (13)) and G⊛→\overrightarrow{G\mathbin{\circledast}}, both convolutions commute with curl: vorticity of filtered velocity equals G⊛ωrG\circledast\omega_r to round-off.

$G\ast\omega_r$$\lvert(\nabla\times\mathrm{Helm})\cdot\hat{\mathbf{e}}_r-G\ast\omega_r\rvert$$\lvert(\nabla\times\mathrm{Edmo})\cdot\hat{\mathbf{e}}_r-G\ast\omega_r\rvert$$\lvert(\nabla\times\mathrm{Amb})\cdot\hat{\mathbf{e}}_r-G\ast\omega_r\rvert$

Figure 6. Ambient convolution is identical to the generalized convolution when applied to a laterally solenoidal vector field, which accounts for the flow’s radial vorticity. Panel (6a) shows the spherical convolution of the radial vorticity as a scalar, the benchmark. Panels (6b),(6c),(6d) calculate the radial vorticity from the filtered vector field, with the vector filtering done via the (6b) Helmholtz scalars, the (6c) Edmonds transformation, and the (6d) ambient convolution. Ambient convolution agrees with (6a) to round-off because the solenoidal basis functions are eigenfunctions of the ambient convolution (eq. (13)).

Figure 7 tests commutation with first-order derivatives using the height gradient ∇h\nabla h, a tangent vector field whose Helmholtz potential is hh itself, so that the generalized convolution satisfies G⊛(∇h)=∇(G⊛h)G\circledast(\nabla h) = \nabla(G\circledast h) (eq. (16)). Figure 7a,Figure 7b probe whether filtering preserves tangency. The generalized convolution (Figure 7a) yields a zero radial component to round-off, with relative error of O(10−12)O(10^{-12}) compared to the gradient magnitude (Figure 7c). In comparison, the ambient convolution (Figure 7b) yields a radial component eleven orders of magnitude larger, reaching ∼20%\sim20\% of the gradient magnitude, which was expected from eq. (5) and illustrated in Figure 3. Figure 7d shows the commutation error of the componentwise convolution in the meridional gradient: the error is O(1)O(1) relative to the true filtered gradient in Figure 7c, i.e., comparable to the signal itself. Note that divergences and gradients such as ∇h\nabla h and ∇⋅(u)\nabla\cdot(\mathbf{u}) appear in the governing equations (17), so commutation errors of this kind directly contaminate any coarse-grained dynamics built on non-commutative filters.

    • *

HA thanks B. Storer, S. Kouhen, and P. H. Sichani for helpful discussions. HA is supported by NSF grants OCE-2123496, OCE-2446475, TI-2449340, PHY-2020249, PHY-2206380, OIA-2535317, DOE grant DE-SC0020229, and NNSA grants DE-NA0004144, DE-NA0004134. Computing time was provided by NERSC and TACC.

Supplementary Material (SM). –

Uniqueness of the commutative convolution.

The generalized convolution (14) is the only linear operator compatible with a given scalar kernel under the commutation requirements. Precisely, let GG be an integrable scalar function on S2\mathbb{S}^{2} that is axisymmetric, let G∗G\ast be the spherical convolution of scalars in eq. (1) (or eq. (10)), and let G1G_{1} be any linear operator on sufficiently smooth vector fields v=vre^r+u\mathbf{v}=v_{r}\hat{\mathbf{e}}_{r}+\mathbf{u} on S2\mathbb{S}^{2} (with u\mathbf{u} tangential) satisfying

e^r⋅(G1v)=G∗vr,(20)\hat{\mathbf{e}}_{r}\cdot(G_{1}\mathbf{v})=G\ast v_{r}, \tag*{(20)}
∇⋅(G1v)t=G∗(∇⋅u),(21)\nabla\cdot(G_{1}\mathbf{v})_{t}=G\ast(\nabla\cdot\mathbf{u}), \tag*{(21)}
e^r⋅[∇×(G1v)t]=G∗ωr,(22)\hat{\mathbf{e}}_{r}\cdot[\nabla\times(G_{1}\mathbf{v})_{t}]=G\ast\omega_{r}, \tag*{(22)}

where (⋅)t(\cdot)_{t} denotes the tangential part. Then G1=G∗⃗G_{1}=G\vec{\ast} on smooth fields.

The proof is as follows. Decompose u=∇Φ+e^r×∇Ψ\mathbf{u}=\nabla\Phi+\hat{\mathbf{e}}_{r}\times\nabla\Psi into Helmholtz scalars, which exist and are unique up to constants on S2\mathbb{S}^{2}. Write G1v=vr′e^r+∇Φ′+e^r×∇Ψ′G_{1}\mathbf{v}=v_{r}^{\prime}\hat{\mathbf{e}}_{r}+\nabla\Phi^{\prime}+\hat{\mathbf{e}}_{r}\times\nabla\Psi^{\prime}. Relation (20) gives vr′=G∗vrv_{r}^{\prime}=G\ast v_{r}. Since ∇⋅u=ΔΦ\nabla\cdot\mathbf{u}=\Delta\Phi and the scalar convolution commutes with the scalar Laplacian (both are diagonal in Yn,jY_{n,j} by eq. (10)), relation (21) yields ΔΦ′=G∗ΔΦ=Δ(G∗Φ)\Delta\Phi^{\prime}=G\ast\Delta\Phi=\Delta(G\ast\Phi), so Φ′−G∗Φ\Phi^{\prime}-G\ast\Phi is harmonic on S2\mathbb{S}^{2}, hence constant, and ∇Φ′=∇(G∗Φ)\nabla\Phi^{\prime}=\nabla(G\ast\Phi). An identical argument applies to relation (22): using ωr=ΔΨ\omega_{r}=\Delta\Psi gives e^r×∇Ψ′=e^r×∇(G∗Ψ)\hat{\mathbf{e}}_{r}\times\nabla\Psi^{\prime}=\hat{\mathbf{e}}_{r}\times\nabla(G\ast\Psi). Therefore G1vG_{1}\mathbf{v} coincides with the right-hand side of eq. (16) for every v\mathbf{v}, i.e. G1=G∗⃗G_{1}=G\vec{\ast}. Conversely, G∗⃗G\vec{\ast} satisfies relations (19) by design [1].

Remark 1: The argument does not assume any specific structure (or kernel representation) of GG, beyond axisymmetry and integrability4. For example, it does not require spatial locality, or positivity, or low-pass character, or preservation of constants. Any linear operator on (sufficiently smooth) vector fields satisfying eq. (19) is necessarily G∗⃗G\vec{\ast}. Examples include spectral (triangular) truncation, isotropic wavelet filters, diffusion, smoothed derivatives (band-pass kernels), Green’s-function inversion, and filters learned from data.

Remark 2: Smoothness of v\mathbf{v} is assumed only so that every quantity in relations (19) is classically defined. We can give up smoothness of v\mathbf{v} if we instead require G1G_{1} to be a continuous linear operator. The argument is as follows. We have for any integrable kernel, ∣G^(n)∣≤∥G∥L1|\hat{G}(n)|\leq\lVert G\rVert_{L^{1}} (for a normalized non-negative kernel, we even have sup⁡n∣G^(n)∣=1\sup_{n}|\hat{G}(n)|=1), so G∗⃗G\vec{\ast} is a bounded operator on vector fields v∈L2(S2)\mathbf{v}\in L^{2}(\mathbb{S}^{2}) (finite-energy fields, possibly rough). Boundedness of G∗⃗G\vec{\ast} implies that it is a continuous operator on L2(S2)L^{2}(\mathbb{S}^{2}) [S2]. Since (i) G1G_{1} and G∗⃗G\vec{\ast} agree on smooth fields by the result above, (ii) smooth fields are dense in L2(S2)L^{2}(\mathbb{S}^{2}) [S2], and (iii) both operators are continuous, we deduce that G1=G∗⃗G_{1}=G\vec{\ast} on all of L2(S2)L^{2}(\mathbb{S}^{2}). This broader setting is relevant for turbulent flows in the infinite Reynolds number limit, whose velocity fields need not be smooth. There, coarse-graining serves as a regularization [S3].

Remark 3: Integrability of GG is not needed for the results on smooth (C∞C^{\infty}) vector fields. The argument uses only that G∗G\ast, as defined by eq. (10), is diagonal in Yn,jY_{n,j} with eigenvalues G^(n)\hat{G}(n) depending only on the degree nn (and independent of jj). This holds for any axisymmetric distribution GG acting on smooth fields (equivalently, any spectral multiplier G^(n)\hat{G}(n) that does not grow faster than algebraically with n→∞n\to\infty). This includes G=δG=\delta (which has G^(n)=1\hat{G}(n)=1 for all nn), high-pass kernels δ−Glowpass\delta-G_{\mathrm{lowpass}}, and differential operators such as Δ\Delta. In this setting, Remark 2 requires only sup⁡n∣G^(n)∣<∞\sup_{n}|\hat{G}(n)|<\infty, which excludes Δ\Delta but still allows for δ\delta.

Remark 4: Replacing conditions (21), (22) with the condition ΔvecG1=0\Delta^{\mathrm{vec}}G_{1}=0 is not sufficient to imply G1=G∗⃗G_{1}=G\vec{\ast}, since Δvec\Delta^{\mathrm{vec}} has the degenerate eigenvalue −n(n+1)/a2-n(n+1)/a^{2} across the radial, potential, and solenoidal components and cannot distinguish them. Even the stronger assumption that operator G1G_{1} is rotation-equivariant (in addition to condition (20)) is not sufficient to show G1=G∗⃗G_{1}=G\vec{\ast}. A counterexample is G1:v↦(G∗vr)e^r+e^r×uG_{1}:\mathbf{v}\mapsto(G\ast v_{r})\hat{\mathbf{e}}_{r}+\hat{\mathbf{e}}_{r}\times\mathbf{u}, which is rotation-equivariant and satisfies condition (20), yet violates relations (21), (22); since G∗⃗G\vec{\ast} satisfies all of relations (19) by design, G1≠G∗⃗G_{1}\neq G\vec{\ast}.

Remark 5: Forthcoming work extends commutative convolutions to arbitrary manifolds [S4].

Supplementary references

[1] Aluie H., GEM Int. J. Geomath., 10 (2019) 9.

[S2] Lieb E. H. and Loss M., Analysis (American Mathematical Society) 2001.

[S3] Eynik G. L., Physica D, 237 (2008) 1956.

[S4] Storer B., Kouhen S., Soltani Tehrani D., Hadi Sichani P., and Aluie H., Commutative coarse-graining on manifolds, (in preparation).

References

  1. [1]Aluie H., GEM Int. J. Geomath., 10 (2019) 9.
  2. [2]Bonev B., Kurth T., Hundt C. et al., Spherical Fourier neural operators: learning stable dynamics on the sphere in proc. of Int. Conf. Mach. Learn. (PMLR) 2023 pp. 2806–2823.arxiv.org/abs/2306.03838
  3. [3]Ghosal S. and Moin P., J. Comput. Phys., 118 (1995) 24.
  4. [4]Jordan S. A., J. Comput. Phys., 148 (1999) 322.
  5. [5]Guillaumin A. P. and Zanna L., J. Adv. Model. Earth Syst., 13 (2021) e2021MS002534.
  6. [6]Pathak J., Subramanian S., Harrington P. et al., arXiv:2202.12114, (2022).
  7. [7]Bretherton C. S., Henn B., Kwa A. et al., J. Adv. Model. Earth Syst., 14 (2022) e2021MS002794.
  8. [8]Bi K., Xie L., Zhang H. et al., Nature, 619 (2023) 533.
  9. [9]Aluie H., Hecht M. and Vallis G. K., J. Phys. Oceanogr., 48 (2018) 225.
  10. [10]Sullivan P. P., McWilliams J. C. and Patton E. G., J. Atmos. Sci., 71 (2014) 4001.
  11. [11]Smith K. M., Hamlington P. E. and Fox-Kemper B., J. Geophys. Res. Oceans, 121 (2016) 908.
  12. [12]Goldstein H., Poole C. P. and Safko J., Classical mechanics 3rd Edition (Addison-Wesley, San Francisco, California) 2002.
  13. [13]Ebin D. G. and Marsden J., Ann. Math., 92 (1970) 102.
  14. [14]Edmonds A. R., Angular Momentum in Quantum Mechanics (Princeton University Press) 1957.
  15. [15]Storer B. A. and Aluie H., J. Open Source Softw., 8 (2023) 4277.
  16. [16]Krishnamurti T. N., Bedi H. S., Hardiker V. M. and Ramaswamy L., An Introduction to Global Spectral Modeling Vol. 35 (Springer) 2006.
  17. [17]Zhao D., Betti R. and Aluie H., J. Fluid Mech., 930 (2022) A29.
  18. [18]Reinecke M. and Seljebotn D. S., Astron. Astrophys., 554 (2013) A112.
  19. [19]Galewsky J., Scott R. K. and Polvani L. M., Tellus A, 56 (2004) 429.

Paper details

vlnc:W_64zh9acq
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