Young rocky planets are often covered in a global ocean of molten rock. Water released from that melt forms a thick steam atmosphere that acts like a blanket. It traps infrared heat and slows solidification to roughly a million years. When the surface is hot enough, above about 2000 K, the steam also splits into hydrogen and oxygen. That produces oxygen with no biology involved, a known source of false "life" signals in exoplanet spectra. Every detailed model of this phase treats the whole planet as a single vertical column with one surface temperature. Two of the key relationships are strongly nonlinear in temperature: heat escaping to space and the amount of water that splits. So the average over a planet with hot and cold regions need not equal the answer for a planet at the average temperature. This note proposes to measure that gap, find the regimes where it matters, and give a cheap rule for predicting it in advance.
The core idea fits in two sentences. Below about 1800 K the outgoing infrared flux of a steam atmosphere is pinned at a nearly constant floor, and above that it climbs steeply. Water splitting is negligible below about 2000 K and dominant near 4000 K. A planet with a hot spot or a hot dayside should therefore cool faster and hold more abiotic oxygen than its globally averaged twin. The sign and rough size of the effect follow from the curvature of the single-column curves, which we compute once with a full spectral code.
Core Claim
The quantitative core is a Jensen-gap estimate. For any column quantity $f(T)$, such as outgoing infrared flux or oxygen mole fraction as a function of surface temperature $T$, the area-weighted planetary mean differs from the single-column value:
$$\Delta_f = \langle f(T)\rangle - f(\langle T\rangle) \approx \tfrac{1}{2} f''(\langle T\rangle)\,\sigma_T^2 ,$$
Here $\langle\cdot\rangle$ is the area-weighted average over the sphere, $f''$ is the second derivative of the column curve, and $\sigma_T$ is the standard deviation of surface temperature across the planet. The form is a second-order Taylor expansion about the mean, so it should hold for modest contrasts and fail near sharp thresholds.
The claim has three parts. The gap is largest near the 1800–2200 K transition, where the flux curve bends from its floor and dissociation switches on. Grey models, which use one wavelength-averaged opacity, have no flux floor, so they get the curvature and hence the gap wrong. And the gap is small for an Earth-like planet at 1 AU during the hottest phase. There, internal heat flux dwarfs sunlight, so stellar day-night contrast barely matters.
That last point limits the claim, and we state it up front. The effect should matter only in specific regimes. The first is close-in, tidally locked planets, where sunlight is comparable to the roughly 282 W/m² floor. The second is spatially localized post-impact heating. Work on Mercury shows that basin-forming impacts create strongly heterogeneous thermal anomalies at depth (Impact-induced changes in source depth and volume of magmatism on Mercury).
Relation to Prior Work
The motivating and method-supplying work is the Katyal et al. series of 1D coupled models.
- Coupled atmosphere-interior modeling during Earth's magma ocean phase couples a convecting molten mantle, water solubility, the line-by-line infrared code GARLIC, and the NASA chemical-equilibrium solver CEA in a 66-layer column. It finds the flux floor of 282 ± 2 W/m² below 1800 K, about 1 Myr to solidify to the rheology front, up to about 4.7% O₂, and a 1–6% flux reduction from dissociation products.
- Evolution and Spectral Response of a Steam Atmosphere for Early Earth with a Coupled Climate–Interior Model reports up to about 16% O₂ and about 33% H₂ at 4000 K, with an effective absorbing height near 100 km.
We reuse their column physics exactly and change only one assumption, horizontal uniformity. The two O₂ figures from the same group differ by a factor of three, so reproducing both is our first calibration task.
The constraining work concerns grey radiation. Evolution of steam atmosphere with a coupled interior-atmospheric model argues that grey schemes miss spectral-window physics and so misjudge cooling. Redox state and interior structure control on the long-term habitability of stagnant-lid planets runs more than 280,000 grey-atmosphere simulations and notes that grey transfer miscalculates extreme climates. Together these rule out a grey scheme as the column model, but they make it the natural ablation.
The spherical-grid literature plays a smaller, specific role.
- ARROW uses circular "ring" positional encodings to remove the artificial seam where 180°W meets 180°E on a latitude-longitude grid.
- PanoSwin wraps attention windows across the same seam and the poles.
Both target smooth fields: weather variables in ARROW, RGB pixels in PanoSwin. We borrow the wrap-around treatment for a numerical transport solver. We then test a consequence neither paper could: does threshold chemistry amplify a small seam error into a spurious oxygen feature?
Research Program
The work has three components.
1. Non-grey column table. We run GARLIC with CEA on convective-adiabatic steam columns. Because each column is set by surface temperature and water inventory, a two-dimensional table spanning 1500–4000 K and 0.1–3 Earth oceans fully specifies the global-mean problem. The table stores outgoing flux, the composition profile, and the emission spectrum, with and without dissociation, plus a grey counterpart. It must reproduce the 282 ± 2 W/m² floor and the 1–6% dissociation effect before anything else proceeds.
2. Spherical surface energy-balance model. This 2D model evolves surface temperature on a latitude-longitude grid. Each cell exchanges heat with the column table (outgoing flux), with the star, and with the convecting mantle. Horizontal heat transport is a parameterized diffusivity, swept from negligible (independent columns) to very large (the 1D limit). Irradiation follows either a fast rotator or a tidally locked substellar pattern, and impact hot spots enter as initial thermal anomalies. The longitude boundary is periodic and the poles are handled by a cap scheme. The resulting cell-to-cell energy flux is fed into a Katyal-style mantle cooling model.
3. Oxygen bookkeeping under two limits. In the first, composition stays in local chemical equilibrium everywhere, an upper bound on hot-spot oxygen that recombines in cold regions. In the second, composition freezes at a quench temperature as gas moves from hot to cold regions, a bound on how much oxygen survives. Real kinetics lie between these limits. We report both and do not claim one.
A neural emulator is deliberately not the main tool, for reasons given below. It becomes necessary only if nightside columns depart strongly from adiabats, for example through temperature inversions, so that the 2D table no longer covers them.
Baselines
The naive reference is the single-column model at the area-weighted mean temperature, i.e. a faithful reproduction of Katyal et al.. It is also the strongest existing method, because it is the only published treatment with line-by-line transfer and dissociation coupled to interior cooling. Its reported targets are about 1 Myr to reach roughly 1650 K, a 282 ± 2 W/m² floor, and 4.7% or 16% peak O₂ depending on setup.
For fairness, the spherical model and this baseline share the same column table, mantle model, and water inventory. The only difference is horizontal structure. The spherical model "beats" the baseline only if two things hold:
- it changes solidification time, disk-integrated flux, or global O₂ inventory by more than the baseline's own sensitivity to its uncertain inputs (we take the spread between the 4.7% and 16% setups as that scale);
- the change survives the strongest-transport limit converging back to the baseline.
The grey ablation reruns the identical spherical model with a grey column table. It isolates the key ingredient, the curvature created by the non-grey flux floor. If grey and non-grey runs show the same gap, the claim that spectral structure drives it is false.
The seam ablation replaces the periodic longitude boundary with a wall (no-flux) boundary. This tests whether seam handling matters for threshold chemistry beyond what ARROW reports for smooth fields; ARROW reports a 9.3% average RMSE gain on WeatherBench from its full method, not from the seam fix alone. The test is to place a hot spot straddling the seam and measure spurious O₂ there.
The bracketing baselines are zero transport and infinite transport, which bound every intermediate result.
No reported result was found on Lacuna for magma-ocean cooling or abiotic oxygen on a spherical domain, so there is no published spherical number to beat.
Why not just…?
Why not just keep the 1D global-mean column? This is the real threat. For an Earth analogue at 1 AU during the hot phase, we expect it to be adequate, because internal flux overwhelms sunlight and keeps $\sigma_T$ small. The Jensen estimate is designed to settle the question cheaply. If $\tfrac{1}{2}f''\sigma_T^2$ stays below a few percent across plausible contrasts, and the full spherical runs agree, we will report that the 1D approach is vindicated outside the tidally locked and impact regimes. That null result is a useful outcome, not a failure.
Why not just use a grey scheme on the sphere, since it is cheap? Cost is not the bottleneck here; the non-grey table is precomputed. Grey transfer lacks the flux floor that both Katyal et al. and Baumeister et al. identify as essential. The grey ablation directly measures how wrong the horizontal gap becomes without it.
Why not just train a neural emulator of GARLIC and CEA? For adiabatic columns, the physics is a smooth function of two variables, and an interpolated table is exact to grid resolution. Off-the-shelf emulators also generalize poorly to new regimes. On ClimART, MLP test RMSE rose from under 1.0 on present-day climate to over 100 on future climates. An emulator earns its place only if non-adiabatic columns appear, and we will check whether they do before building one.
Why not just run a full 3D general circulation model? None exists in the file for 2000–4000 K dissociating steam, and building one is a multi-year effort. The energy-balance model brackets the answer between the no-transport and infinite-transport limits. If even the zero-transport bound shows a negligible gap, a 3D model is unnecessary. If the bounds differ greatly, that difference justifies the 3D investment.
Proposed Experiments
Calibration. Reproduce the single-column floor (282 ± 2 W/m²), the roughly 1 Myr solidification time, and the 1–6% dissociation flux reduction. Identify which setup yields 4.7% versus 16% O₂.
First decisive check. For each regime (rotator at 1 AU, tidally locked close-in planet, impact hot spot), compare the Jensen-gap prediction with the full spherical result for flux, solidification time, and global O₂. Agreement means the cheap curvature rule can be handed to anyone running 1D models. Disagreement near thresholds tells us where second-order reasoning breaks.
Second decisive check. The non-grey gap should exceed the grey gap most strongly in the 1800–2200 K band, where the floor bends. If it does not, the mechanism we propose is wrong.
Third check. In the seam ablation, a hot spot straddling the boundary should produce a measurable spurious O₂ artifact under the wall boundary that vanishes under periodic wrapping.
Observable test. Compute disk-integrated emission spectra for each regime and ask whether the spherical O₂ and OH features differ from the 1D spectrum by more than the spread between the two published 1D setups. This states, concretely, when a globally averaged model would misread an abiotic oxygen signal.
References
- Impact-induced changes in source depth and volume of magmatism on Mercury and their observational signatures
- Coupled atmosphere-interior modeling during Earth's magma ocean phase
- Evolution and Spectral Response of a Steam Atmosphere for Early Earth with a Coupled Climate–Interior Model
- Evolution of steam atmosphere with a coupled interior-atmospheric model
- Redox state and interior structure control on the long-term habitability of stagnant-lid planets
- ARROW: An Adaptive Rollout and Routing Method for Global Weather Forecasting
- PanoSwin: a Pano-style Swin Transformer for Panorama Understanding
- ClimART: A Benchmark Dataset for Emulating Atmospheric Radiative Transfer in Weather and Climate Models