Ocean neutral transport: sub-Riemannian geometry and hypoelliptic diffusion

Transport and mixing of tracers in the ocean is thought to be preferentially along neutral planes defined by the potential temperature and salinity fields. This gives rise to a conceptual model of ocean transport in which water parcel trajectories are everywhere neutral, that is, tangent to the neutral planes. Because the distribution of neutral planes is not integrable, neutral transport, while locally two dimensional, is globally three dimensional. We describe this form of transport, building on its connection with contact and sub-Riemannian geometry. We discuss a Lie-bracket interpretation of local dianeutral transport, the quantitative meaning of helicity and the implications of the accessibility theorem. We compute sub-Riemnanian geodesics for climatological neutral planes and put forward the use of the associated Carnot–Carathéodory distance as a diagnostic of the strong anisotropy of neutral transport. We propose a stochastic toy model of neutral transport which represents motion along neutral planes by a Brownian motion. The corresponding diffusion process is degenerate and not (strongly) elliptic. The non-integrability of the neutral planes however ensures that the diffusion is hypoelliptic. As a result, trajectories are not confined to surfaces but visit the entire three-dimensional ocean. The short-time behaviour is qualitatively different from that obtained with a non-degenerate highly anisotropic diffusion. We examine both short- and long-time behaviours using Monte Carlo simulations. The simulations provide an estimate for the time scale of ocean vertical transport implied by the constraint of neutrality.

💡 Research Summary

The paper investigates the consequences of imposing the “neutral‑plane” constraint on oceanic tracer transport. Neutral planes are defined by the joint fields of potential temperature (θ) and salinity (S) through the equation of state ρ = R(θ,S,p). The authors introduce the dianeutral vector n ∝ Rθ∇θ + RS∇S and its dual 1‑form η (η(u)=0 expresses the neutral constraint). This constraint forces parcel velocities to lie in a two‑dimensional distribution Δ of planes at every point in the ocean.

A key observation is that Δ is not integrable: the Frobenius condition η∧dη = 0 fails because the helicity H = ∗(η∧dη) is non‑zero for realistic oceanic equations of state. Physically, this means that there are no globally defined neutral surfaces; parcels cannot remain confined to a single surface, even though locally their motion is two‑dimensional. The non‑integrability is quantified by H, which is proportional to the Jacobian determinant |∂(θ,S,p)/∂(x₁,x₂,x₃)| and is small but finite in the ocean.

The authors recast the problem in the language of contact and sub‑Riemannian geometry. The distribution Δ together with the ambient Euclidean metric makes the ocean a three‑dimensional contact manifold, and neutral trajectories become Legendrian curves. The natural distance on such a manifold is the Carnot–Carathéodory (CC) distance, defined as the infimum of the lengths of admissible (neutral) curves joining two points. Using climatological θ‑S fields, the authors compute sub‑Riemannian geodesics and the associated CC distances. They demonstrate that the CC distance between points at different depths can be orders of magnitude larger than the Euclidean distance, providing a quantitative measure of the strong anisotropy of neutral transport.

To capture the stochastic aspect of tracer dispersion, the paper introduces a degenerate diffusion model: a Brownian motion constrained to the neutral planes. The diffusion tensor is singular, having n as its null direction, so diffusion occurs only in the horizontal (neutral) directions with diffusivity κ. Because the distribution Δ satisfies Hörmander’s bracket‑generating condition (the Lie brackets of the two spanning vector fields generate the missing vertical direction), the resulting diffusion operator is hypoelliptic. Consequently, the probability density becomes smooth in all three dimensions despite the degeneracy, and particle trajectories eventually explore the full ocean volume.

Monte‑Carlo simulations of the stochastic model reveal two distinct regimes. At short times (tens of years) the dispersion is dominated by the strong horizontal anisotropy; vertical spread is negligible. At longer times (hundreds of years) the non‑integrability allows the particles to drift vertically, leading to a crossover to isotropic‑like spreading. The authors compare this behavior with that of a non‑degenerate but highly anisotropic diffusion tensor and find qualitative differences: the hypoelliptic model predicts a slower, more realistic vertical mixing time scale. From the simulations they estimate that a parcel released at depth would, on average, require on the order of 10⁸ seconds (several hundred years) to reach the mixed layer under the neutral‑constraint alone.

The paper also discusses the Lie‑bracket interpretation of dianeutral transport. Two independent neutral vector fields v₁ and v₂ satisfy η(v_i)=0, and their Lie bracket