Skip to content

Surface lid

How does Aragog follow a crust that is much thinner than the top cell of the mesh?

Short answer. With subgrid_lid = true, Aragog carries the crust as two extra states: its thickness \(\delta\) [m] and its energy per area \(E_\mathrm{lid}\) [J/m²]. The temperature across the crust follows a profile from the textbook solution for a freezing lava layer, so the crust conducts the right heat from its first metres on. The top cell keeps only the melt below the crust, and new crust forms at the base by the balance of the heat conducted up through the crust and the heat the melt delivers.

Why the mesh alone cannot do it

A magma ocean need not freeze from its surface. The adiabat is steeper than the solidus, so the two meet first at depth and solidification is expected to begin at the bottom, the more so the larger the pressure range of the planet. An insulating atmosphere can keep the surface above its solidus for much of the solidification, and a quench crust on a planet-sized body is then unlikely (Elkins-Tanton (2012)2 section 4.1).

A crust can still form at the surface when the surface loses heat faster than the melt below delivers it, for example under a thin atmosphere. The lid treats that crust; it does not decide where in the mantle freezing starts.

That crust starts metres thick, while the top cell of a 100-node mantle is 29 to 35 km thick. A cell that thick has one temperature: it is either all melt or all solid. While the crust is thin, neither choice is right. As melt, the cell radiates from a hot surface; as solid, it blocks the heat of the whole cell. The heat the planet loses then jumps between two wrong values as the top cell freezes and thaws. A mesh fine enough to resolve a metre-thick crust needs metre-thick cells, which makes the solver steps tiny.

The two states and the profile

The crust runs from the surface, at temperature \(T_s\), down to its base at depth \(\delta\), at temperature \(T_b\). Material joins the crust at its base, so \(T_b\) is the temperature at which the melt becomes rigid: the solidus above the rheological transition of the melt, and the lid-base temperature of the solid rheology below it. The base is never warmer than the solidus and never warmer than the melt below the crust: \(T_b \le T_f\), with \(T_f\) the temperature of the top cell.

Between the surface and the base, the temperature follows one member of a family of profiles. With \(z\) the depth below the surface,

\[T(z) = T_s + (T_b - T_s)\, f_s(z / \delta), \qquad f_s(x) = \frac{\operatorname{erf}(\sqrt{s}\, x)}{\operatorname{erf}\sqrt{s}},\]

and \(f_s(x) = x\) for \(s = 0\) (with the imaginary error function for \(s < 0\)). The energy \(E_\mathrm{lid}\) is the heat content of the crust: the heat content per volume \(H(T)\) of the solid, taken from the equation of state at the pressure of the top cell, integrated over the profile. On the grey-body surface the balance below fixes \(T_s\) for each shape, and under a fixed surface temperature \(T_s\) is the boundary value, so one shape \(s\) gives the stored energy.

A growing crust has \(s > 0\), a steady crust has the straight line \(s = 0\), and a thinning crust has \(s < 0\). This family contains the exact solution for a crust that freezes over melt, \(\delta = 2\lambda\sqrt{\kappa t}\) with \(s = \lambda^2\) (Turcotte and Schubert (2014)5 section 4.18, Eqs. 4.136 and 4.141, pp. 197-198), so the lid grows exactly as that solution.

The surface and the base

The surface temperature follows from the balance of the conducted heat and the grey-body radiation,

\[\varepsilon \sigma \left(T_s^4 - T_\mathrm{eq}^4\right) = k\, (T_b - T_s)\, \frac{f_s'(0)}{\delta},\]

with \(\varepsilon\) the emissivity, \(\sigma\) the Stefan-Boltzmann constant, \(T_\mathrm{eq}\) the equilibrium temperature and \(k\) the conductivity of the crust. The left side is \(F_s\), the heat the planet loses [W/m²].

At the base the crust conducts \(q_b = k (T_b - T_s) f_s'(1) / \delta\) upward, and the melt delivers \(F_b\). The difference freezes new crust:

\[\left(W(S_f) - W(S_b)\right) \frac{d\delta}{dt} = q_b - F_b, \qquad W(S) = \int_{S_\mathrm{sol}}^{S} \rho\, T\, dS'.\]

Here \(W\) [J/m³] is the heat content per volume at the top-cell pressure, counted from the solidus entropy \(S_\mathrm{sol}\). It is the measure of the heat content the solver reports.

The entropy equation of the mesh has the capacitance \(\rho T\) with the two phases blended near the solidus and the liquidus; there it differs from the table \(\rho T\) in \(W\) by less than 0.1 %. The crust holds heat in the same measure, as a function of its temperature, wherever the temperature rises with entropy. Some tables keep the temperature at the solidus over a band of entropy below \(S_\mathrm{sol}\); a crust profile at the solidus temperature then holds the heat of the lower end of that band, not \(W(S_\mathrm{sol}) = 0\).

\(S_f\) is the entropy of the melt and \(S_b\) that of the crust at its base. The left side is the heat released when melt joins the crust, latent heat of the melt fraction included; a melt at the base entropy releases none. The density sits inside the integral because it changes along the way from \(S_b\) to \(S_f\). The melt delivers \(F_b\), the larger of the heat flux into the top cell from below and the flux of a boundary layer at the critical Rayleigh number lid_ra_crit, and at least 0. With energy.convection off the melt does not convect, so \(F_b\) is the flux into the top cell from below, or 0.

The base at the melt temperature

Why. The lid-base temperature of the solid rheology can lie above the temperature of the melt below the crust, for example a fixed lid_base_temperature of 1300 K over a top cell at 1200 K. Material at the melt temperature cannot join a crust whose base is warmer: the jump \(W(S_f) - W(S_b)\) would turn negative and the crust would grow from heat it does not get.

What. Below the rheological transition the base entropy is capped at the melt, \(S_b = \min(S_\mathrm{rh}, S_\mathrm{sol}, S_f)\), with \(S_\mathrm{rh}\) the entropy of the rheological lid-base temperature. At the cap the jump is 0, and the balance above has no finite growth rate. The heat jump per mass, \((W(S_f) - W(S_b)) / \rho_b\) with \(\rho_b\) the density at the base, therefore has a floor LID_DH_MIN of 10³ J/kg, a numerical help of the solver. At the cap the crust grows at \(d\delta/dt = (q_b - F_b) / (\rho_b \times 10^3\ \mathrm{J/kg})\).

Why it matters. At the cap the growth rate depends on the floor and not on a latent heat, so the crust thickness and the surface flux there depend on the floor and on the mesh. The Validation page gives the size of that dependence.

The base at the solidus

Why. A top cell on a coarse mesh can freeze whole while the mush below it still moves, and it then blocks the heat of the column: the surface flux falls far below that of a fine mesh. On a fine mesh the eddy-diffusivity floor of the solver carries the heat through the mush to the crust base, while the boundary-layer law for \(F_b\) gives a flux many orders of magnitude smaller, because the mush near the solidus is about 10²² Pa s.

What. With lid_mush_base = true the crust base lies at the solidus, \(S_b = S_\mathrm{sol}\), and the melt delivers the steady flux of the floor through the fluid part of the top cell,

\[F_K = \frac{\Phi(S_f)}{(h - \delta)/2}, \qquad \Phi(S) = \int_{S_\mathrm{sol}}^{S} \rho\, T\, \kappa_\mathrm{floor}\, f(\phi)\, dS',\]

with \(h\) the top-cell width, \(\kappa_\mathrm{floor}\) the setting kappah_floor, and \(f(\phi)\) its weight in the melt fraction, 0 at the solidus. \(\Phi\) is the Kirchhoff integral, the exact steady flux times distance for a point value of \(S\). The solver holds a cell mean: for \(\Phi \propto (S - S_\mathrm{sol})^n\) this gives \(F_K\) low by the factor \(2 (n/(n+1))^n\), 0.89 at \(n = 2\) and at least \(2/e = 0.74\). The base takes \(F_K\) plus the boundary-layer flux.

The top cell must stay above the solidus, because the jump \(W(S_f) - W(S_b)\) vanishes there. A weight \(g\) rises smoothly from 0 at \(S_\mathrm{sol} + 1\) J/kg/K to 1 at \(S_\mathrm{sol} + 6\) J/kg/K. At \(g = 0\) the base takes exactly the heat that enters the top cell from below and the lid acts with weight 1, so the entropy of the cell holds. This needs the heat jump of the base, \((W(S_f) - W(S_\mathrm{sol})) / \rho_b\), at or above the floor LID_DH_MIN = 1000 J/kg of the Stefan rate: below it the rate runs at the floor while the cell gives the lid the true jump, so the cell cools while the lid grows and warms while it thins. The band is about \(S_\mathrm{sol}\) to \(S_\mathrm{sol} + 1000/T_\mathrm{sol}\) J/kg/K, so a lid that starts there is refused (see the settings below). The base flux is

\[F_b = F_\mathrm{in}\frac{A_\mathrm{in}}{A} + g \left(F_K + q_\mathrm{conv} - F_\mathrm{in}\frac{A_\mathrm{in}}{A}\right),\]

with \(A_\mathrm{in}\) and \(A\) the areas of the lower and upper faces of the cell. The lid hands over to the mesh when it reaches 0.7 of the top-cell width.

Why it matters. \(F_K\) scales with \(\kappa_\mathrm{floor}\), and the fine mesh it reproduces scales with it too. The lid therefore follows a mesh solution that depends on the floor. This does not test the floor as physics: the floor is a numerical setting in the mush, not a measured transport law. A run with a molten start and the solid rheology without this setting keeps the base of the previous section, so set it for such a run.

The lid under an atmosphere

Why. A coupled run gives the interior a flux that holds for a whole coupling step. A crust at the solidus takes up a flux difference with a small heat jump, so under a held flux a thin crust grows far past its steady thickness in one step. The atmosphere then returns a much smaller flux, the crust melts back, and the flux jumps from step to step.

What. With outer_boundary_condition = 4 and lid_mush_base, the caller sets lid_surface_temperature to the surface temperature of the atmosphere before each call. The lid holds that surface through the call and loses

\[F_s = k\, (T_b - T_\mathrm{surf})\, \frac{f_s'(0)}{\delta},\]

so its flux falls as it thickens within the step. The part \(1 - w\) of the surface without a lid keeps the flux value of the boundary. The weight is \(w = 1 - g\,(1 - w_\mathrm{s})\), with \(g\) the floor weight of the base at the solidus and \(w_\mathrm{s}\) the existence weight of the crust section for the steady crust under the atmosphere surface \(T_\mathrm{s}\) (the held surface, or without one the surface of the lid), \(\delta_\mathrm{eq} = k (T_b - T_\mathrm{s}) / F_b\). \(w_\mathrm{s}\) is 0 when \(T_\mathrm{s}\) lies at or above \(T_b\), or, without a held surface, when the prescribed flux is 0 or below; otherwise it is 1 when \(F_b \le 0\). For \(F_b > 0\) it falls to 0 as \(T_\mathrm{s}\) rises to \(T_b\), so the weight has no step at the onset: \(w_\mathrm{s} = 0.5\) where \(T_b - T_\mathrm{s} = F_b\,\delta_\mathrm{on} / k\). At \(g = 0\) the lid therefore acts with weight 1 under any atmosphere, and its base takes the flux from below. Under a held surface the top cell then holds its entropy above the heat-jump band while the lid lies between its stops; a lid that thins to its melt-back stop, under a held surface or under a prescribed flux below the flux from below, stays there and the top cell warms by the difference. The surface of the lid stays at least 1 K below \(T_b\), because a profile without a temperature drop has no shape. Without lid_surface_temperature the lid takes the flux value itself.

The caller can also set lid_surface_flux_cap, the flux that the last atmosphere balance gave the lid. A lid that thins under a held surface conducts more heat than that balance took from it, so the planet would lose the excess in the call. Where the held surface would conduct more than the cap \(C\), the lid surface carries \(C\) at a warmer surface instead. With lid_surface_flux_slope \(F'\) set as well, it carries \(C + F' (T_s - T_\mathrm{held})\), the flux the atmosphere gives at the warmer surface to first order in \(T_s\); \(F' = 0\) is the cap alone. Below, \(C\) stands for that flux. Its base does not take up \(C\) at once: after a start transient of the profile, the thickness changes at about \((C - F_b) / (W(S_f) - H(T_s))\), the rate of a straight profile (0.97 to 1.00 of it in a measured melt-back call), with \(H(T_s)\) the heat content per volume of the crust at its surface temperature, in the measure of \(W\). This rate does not depend on the length of the call. The cap can also act while the lid grows: the held lid conducts \(k f_s'(0) (T_b - T_s) / \delta\), and \(f_s'(0)\) rises with the shape parameter \(s\), so a profile whose \(s\) rises within the call can conduct more than at its start. Below the cap the lid is the held state.

For its next solve the atmosphere needs the conductance from the melt to its surface. The output lid_conductance gives \(k f_s'(0) / \delta\), with the slope at least 0.1; PROTEUS blends it in parallel with its skin, capped at the skin conductance. With the mush base, on every outer boundary, the output T_magma is the melt temperature at the crust base where the lid acts: \(w T_b + (1 - w) T_\mathrm{node}\), with \(T_\mathrm{node}\) the top basic node. Outgassing reads this value, so it runs at the melt below a crust and takes no account of the crust as a barrier to gas.

Why it matters. The surface flux of a call is not the flux of the atmosphere: it starts from it and follows the crust. The planet loses the integrated flux of the call, and PROTEUS books that as F_int, with the difference from F_atm in lid_F_diff.

Where this is checked. The lid surface under an atmosphere checks the Robin surface, and the weight of the lid at its onset checks the onset weight, both against closed forms on NumPy and JAX.

Energy

The heat content of the crust follows the equation of state across the crust, so its heat capacity and density change from the base to the surface as they do in the mesh. A crust that took them at its base would hold too much heat under a cold surface: on the PALEOS table at 10 MPa the heat capacity of the solid at 255 K is 55 % of its value at 1300 K.

The top cell keeps its melt part: its volume shrinks by the crust volume, area times \(\delta\), and the heat \(W(S_f)\) of each unit of that volume passes to the crust. The flux out of the melt follows from the energy of the melt and the crust together, so the energy of the mantle changes only by the heat that crosses the surface and the core boundary. The energy diagnostics count the crust, and their solver residual reports this balance for every call. Radiogenic and tidal heat act on the melt part of the top cell and on the crust mass, and the reported source powers state that heating.

When the crust exists, and when it goes

A hot, fast-convecting melt carries more heat than a crust at \(T_b\) can radiate. No crust can then form. The lid weight \(w\) [0 to 1] compares the thickness of a crust in steady state, \(\delta_\mathrm{eq} = k (T_b - T_s) / F_b\), with lid_delta_on: below it the lid is hidden (\(w = 0\)) and the surface is the mesh surface, above it the lid acts (\(w = 1\)). A hidden lid relaxes to a thin phantom lid of thickness lid_delta_lo.

When the crust is resolved by the mesh, the lid hands over to the mesh: the solver merges the crust into the top cell with its energy, sets both lid states to zero and continues. A zero lid stays zero. Which condition starts the hand-over depends on the solid rheology.

With the solid rheology and stress_closure_mode = 'lid', without the mush base. When the mantle near the surface is solid, a crust that has grown thicker than a few top cells is resolved by the mesh. The lid hands over when three conditions hold together: the base of the cold region (the deepest point where the column rises through the lid-base isotherm of the solid rheology) lies below the face of lid_handover_cells top cells, the stagnant-lid state of the solid rheology is active (its weight w_active above 0.5), and the top cell is below the rheological transition (the melt ramp \(w_\phi\) of the lid base below 0.5).

Why all three: in a molten or mushy mantle the solid rheology has no stagnant lid, so its isotherm marks no crust and can lie hundreds of kilometres deep.

Without the solid rheology. The lid base is the solidus. When the top cell cools to its solidus, \(S_f \le S_\mathrm{sol}\), the jump \(W(S_f) - W(S_b)\) is 0: no melt is left below the crust to freeze, and the freezing front has passed through the top cell. From there the mesh carries the cold top cell, and the lid hands over. A top cell at or below the solidus at the start of a call hands over before the call runs; a top cell that cools through the solidus during a call hands over at that time.

With the base at the solidus. With lid_mush_base the lid hands over when it reaches 0.7 of the top-cell width, with any closure of the solid rheology; lid_handover_cells is then not used, although the mesh check of that field still applies.

With the solid rheology and another closure, without the mush base. The lid does not hand over. The base then lies at the rheological temperature, at or below the solidus, and a top cell can freeze through while melt stays under a thin resolved crust.

Where the lid is valid

Why. The lid replaces a crust the mesh cannot resolve, so it is right only where its answer has been compared with a mesh that does resolve that crust.

What. The Validation page compares the lid with fine meshes for two starts: a solid shell below the solidus and a shell in the mush that freezes from its surface. For a molten shell without the solid rheology it compares only the hand-over at the solidus and the coarse mesh after it; the lid itself has no measured error before the hand-over. A molten start with the solid rheology on, with either closure, is not validated: no run resolves its crust yet, so no converged reference exists.

Where the lid base sits at the melt temperature, a hot shell with the solid rheology and the 'lid' closure gives a crust that changes by up to 13 % from 200 to 400 nodes from 3 × 10³ to 3 × 10⁵ years; later its thickness depends on the mesh and on the floor (see mesh and floor).

Why it matters. Use the lid for a molten start with the solid rheology only with this limit in mind; its heat loss there has no measured error.

A worked example

A melt layer at \(T_b\) = 1400 K freezes under a surface held at 300 K, with \(c_p\) = 1200 J/kg/K and a latent heat of 3.9 × 10⁵ J/kg, so the Stefan number \(c_p (T_b - T_s) / L\) is 3.385 and \(\lambda\) = 0.948. With \(\kappa\) = 1.1 × 10⁻⁶ m²/s, the crust is \(2 \lambda \sqrt{\kappa t}\) = 11 m thick after one year and 353 m after a thousand years. Both are far below a 30 km top cell.

On a constant-property column

The stagnant-lid benchmark rows of Korenaga (2009)3 Table 1 and Deschamps and Vilella (2021)1 Table 1 use constant material properties (phase_mixed.const_properties = true): \(T(S) = T_\mathrm{ref} \exp((S - S_\mathrm{ref}) / c_p)\) with constant density, heat capacity and conductivity, no melt and no solidus. The lid runs there in one mode only: no entropy EOS attached (entropy_eos=None), the solid rheology on, stress_closure_mode = 'lid', whose stagnant-lid state gives the lid base, and core_bc = 'quasi_steady'.

Every other combination with constant properties is refused at setup by name, also a run that attaches an EOS, because the lid would then take other physics than the column.

In this mode the lid uses closed forms instead of the tables. The heat content per volume is \(W(S) = \rho c_p T(S)\), zero at 0 K; the Stefan jump \(W(S_f) - W(S_b) = \rho c_p (T_f - T_b)\) does not depend on that zero. With no solidus the lid base is \(T_b = \min(T_\mathrm{rh}, T_f)\), the stagnant-lid isotherm capped at the fluid, for any width of the rheological transition. The hand-over merge solves \(T' = (V_N W(S_f) + A (E - W(S_f)\, \delta)) / (V_N \rho c_p)\) in one step, and the lid conducts with the conductivity of the column. The step ledger and the lid step checks run as with the tables; the cell heat content is \(\rho c_p (T(S_1) - T(S_0))\) per volume.

Under a fixed surface temperature

With outer_boundary_condition = 5 the surface of the lid is held at the boundary temperature \(T_\mathrm{surf}\) instead of the grey-body balance. The shape of the profile follows from the lid energy at that surface temperature, the steady crust of the existence weight is \(\delta_\mathrm{eq} = k (T_b - T_\mathrm{surf}) / F_b\), and there is no lid where the base is not hotter than the surface. The surface flux is \(F_\mathrm{top} = w F_s + (1 - w) F_\mathrm{grid}\), with \(w\) the existence weight, \(F_s\) the conduction through the lid and \(F_\mathrm{grid}\) the conduction of the top cell to the fixed surface over the half cell, so the second term takes over as the weight falls.

Settings and limits

  • subgrid_lid (bool, default false): the switch. Only with the grey-body surface, outer_boundary_condition = 1, a fixed surface temperature, outer_boundary_condition = 5, or outer_boundary_condition = 4 with lid_mush_base. It is refused at setup with core_bc = 'gradient' or 'bower2018', and with param_utbl = true, because the lid replaces the surface flux that the correction acts on. With const_properties only in the mode of On a constant-property column.
  • lid_delta_on [m] (default 1): must be finite, positive and below 0.8 of the top-cell width, and with lid_mush_base below 0.7 of it; a larger value is refused at solver setup with the width named.
  • lid_delta_lo [m] (default lid_delta_on / 2): must lie in (0, lid_delta_on).
  • lid_handover_cells (integer ≥ 1, default 3): must leave at least one cell below the hand-over face.
  • lid_ra_crit (default 450): positive and finite.
  • lid_mush_base (bool, default false): the base at the solidus, with the floor flux. It needs subgrid_lid, the solid rheology, kappah_floor > 0 (its base flux is the floor flux) and the EOS tables (it is refused in the constant-property mode), lid_delta_on below 0.7 of the top-cell width, where the lid hands over, and a solidus of at least 1000 K at the top-cell pressure (LID_DH_MIN / LID_FLOOR_GAP; below it the heat-jump band of the lid start reaches past the floor gap). The SPIDER and PALEOS MgSiO3 tables have 1439 K and 1421 K at the surface. A phantom start with a live lid below \(S_\mathrm{sol} + 1\) J/kg/K is refused with an error, so a run that starts solid must run without lid_mush_base. The step check refuses an accepted state with the top cell below \(S_\mathrm{sol} + 0.5\) J/kg/K, and a set or carried lid starts down to that value, or down to the top of the heat-jump band where that lies higher. Every start with a live lid (phantom, set, carried or after a remap) is also refused where the heat jump of the base is below LID_DH_MIN (the band of the base at the solidus); on a top cell with the solidus near 1440 K that band reaches about \(S_\mathrm{sol} + 0.7\) J/kg/K. For a solidus below 2000 K the band reaches above the 0.5 J/kg/K of the step check, so a state that a call accepted can be refused at the next start.
  • lid_surface_temperature [K] (default none): the held surface of the lid. Finite and positive; only with outer_boundary_condition = 4 and lid_mush_base.
  • lid_surface_flux_cap [W/m²] (default none): the largest flux of the lid surface in a call. Finite and positive; only with lid_surface_temperature, outer_boundary_condition = 4 and lid_mush_base.
  • lid_surface_flux_slope [W/m²/K] (default none): the slope \(F'\) of the atmosphere flux with the surface temperature, used above the cap. Finite and 0 or above; only with lid_surface_flux_cap.
  • The crust stops growing at 0.8 of the top-cell width (one logged warning per run) and stops thinning at lid_delta_lo. The integrator can leave a crust at the growth stop slightly above it; the step check accepts up to 10⁻³ above the stop and records the largest excess of each call as lid_cap_excess on the solution. Below the lower stop it accepts the solver precision, \(\sqrt{N}\,(\mathrm{atol} + \mathrm{rtol}\,\delta_\mathrm{lo})\) with \(N\) the number of solver states and the tolerances of the call: the weighted error test of a step lets one state carry that local error.
  • The heat jump per mass, \((W(S_f) - W(S_b)) / \rho_b\) with \(\rho_b\) the density at the lid base, has a floor of 10³ J/kg (one logged warning when it acts).
  • The floors cost solver steps. While the top cell freezes and thaws, the lid forms and vanishes at its phantom thickness and passes the weight floor and the lower stop many times.
  • Without the mush base, the lid hands over with the solid rheology and stress_closure_mode = 'lid' and without the solid rheology; with lid_mush_base it hands over at 0.7 of the top-cell width, with any closure. With the solid rheology, another closure and no mush base it never hands over. A molten start with the solid rheology is not validated (see Where the lid is valid).
  • The lid takes its base temperature, its heat content and its density at the pressure of the top-cell centre, not at the pressure of its base. On a coarse mesh the two differ: on the 10-node mantle of the tutorial the top-cell centre lies at 7.8 GPa, where the solidus is 2149 K, against 1439 K at the surface. Use a mesh whose top cell is thin enough that the solidus changes little across it.
  • A coupled run with a prescribed surface flux (outer_boundary_condition = 4) can use the lid only with lid_mush_base; see The lid under an atmosphere. Without it the lid sets the surface flux itself through the grey-body balance, and the setting is refused.
  • The lid has no buoyancy test: it stays at the surface until the hand-over, whatever the density contrast with the fluid below. A crust denser than the melt below it is gravitationally unstable if the two have a similar composition (Miyazaki and Stevenson (2024)4 p. 1). That study finds that on small bodies the foundering of the whole crust is often improbable, but this is not modelled here, and a crust that founders in a run is not followed.

    On the tables of the Validation page, at its top-cell pressure of 55.7 MPa, the sign is the opposite: the melt at the liquidus is denser than the solid at the solidus (3603 against 3041 kg/m³ on the extended PALEOS table, a ratio of 1.185). This sign is a property of those tables and the opposite of what that study takes for silicates of one composition, so do not read it as a measured result. The density sign in a run comes from the table, not from a check in the lid.

  • At the growth stop the lid base lies deeper than the node of the top cell, by \(\delta - h/2\) with \(h\) the top-cell width. The base has the temperature of that node, so the column below is warmer than on a mesh that resolves the crust, by about \((\delta - h/2)(T_b - T_s)/\delta\), with the mean temperature gradient of the lid. After a hand-over the surface flux lies above a run without the lid and falls back over millions of years, which fits this excess leaving through the surface. This size is an estimate from the geometry: it matches the runs of the Validation page, but no run exists with the lid base moved, so the cause is not shown.

  • The melt flux \(F_b\) at the lid base is the flux per area at the lower face of the top cell, not at the lid base. The bias in the growth rate and the weight is about \(2h/R\), with \(h\) the top-cell width and \(R\) the planet radius: 1.1 % for a 35 km top cell of the Earth, 0.09 % on the shell of the Validation page. The energy budget is not affected, because the flux out of the melt follows from conservation.
  • With the solid rheology off, the lid checks and the EOS table guard accept an entropy up to 10⁻³ J/kg/K past a table end; in that band the heat content continues linearly. With the solid rheology on, its own check refuses a state outside the table at the first step.
  • Below the first knot of the heat table the lid continues the heat content with the slope at that knot; one logged warning says so. On the JAX path the check reads only the last state of a call, as the growth-stop warning does. The SPIDER test tables start near 320 K, so a lid surface can lie below them under a 255 K sky.
  • The lid runs on the NumPy and on the JAX right-hand side. A JAX factory for a run with the lid must be built with lid_params=solver.jax_lid_params(), called again at each build, because the held surface temperature, the cap and the slope of the call enter those parameters; a factory built with the lid on for a run with it off, or the reverse, is refused before the solve, but a factory with stale values of the held surface, the cap or the slope is not detected. The constant-property mode and the fixed surface temperature run on the NumPy right-hand side only: jax_lid_params() refuses both by name.
  • Inside the lid, the depth of the lid-base isotherm that the hand-over root reads interpolates linearly between the surface and the lid base, not along the profile of the lid. For a curved profile this changes the depth value only: the lid is at most 0.8 of the top cell thick and the hand-over face lies at least one cell deep, so a crossing inside the lid gives the root the same sign either way.

Where this is checked. Surface lid compares the lid with the freezing lava layer, a cooling half-space and a steady crust, and with a fine mesh for a solid shell and for a shell that freezes from the mush, which tests the Stefan jump. It also lists the solver tests of the energy budget, the Stefan jump and the hand-over.

Restarting a run

A new solver takes the lid from set_initial_lid(delta, E_lid), in either order with set_initial_entropy; the lid states are the last two entries of the solver state. Without it, a new solver starts the phantom lid, so a run that resumes must pass its lid on. A lid that has handed over resumes with set_initial_lid(0.0, 0.0, handover_time=t), where t is the hand-over time in years; without the time, lid_handover_time in the output is NaN.

A lid set after set_initial_entropy is written into that start and checked when the solve starts, against the top cell of the start: a state the start refuses in the other order is refused there too, with the solver left as it was.

A solver that is used again continues from its previous solution, unless reset has changed the number of nodes: the phantom lid then starts, and a logged warning says so. A start with a live lid clears the hand-over time. A reset can make the top cell thinner than the lid it carries: in PROTEUS runs with structure updates the top cell narrows by up to a factor 1.36 at one update, as it solidifies, so a lid near the hand-over can lie at up to about 0.96 of the new width.

  • With lid_mush_base a set or carried lid at or above 0.7 of the top-cell width hands over at the start of the next solve. The merge keeps the heat of the lid and the top cell to round-off (relative difference below \(10^{-15}\) in the tests). A lid is refused, with its thickness, the width and the largest thickness the merge can take, where the merged top cell or the top basic node (extrapolated from the cells) would leave the EOS table or come within eos_floor_margin of its lower end; with eos_floor_margin > 0 that node is the one the EOS floor watches. A lid at the hand-over over a top node that already lies outside that band is refused with the node named. In the solver tests (a 380 km top cell at 3050 J/kg/K on the SPIDER tables) that limit is 1.17, 1.06 and 0.98 top-cell widths for a lid surface at 1800, 1000 and 300 K, set by the top node.
  • Without the mush base a set or carried lid above 0.8 of the width (the growth stop) is refused at the start with both thicknesses named.

  1. F. Deschamps, K. Vilella, Scaling laws for mixed-heated stagnant-lid convection and application to Europa, Journal of Geophysical Research: Planets, 126, e2021JE006963, 2021. SciX. ↩

  2. L. T. Elkins-Tanton, Magma oceans in the inner solar system, Annual Review of Earth and Planetary Sciences, 40, 113–139, 2012. SciX. ↩

  3. J. Korenaga, Scaling of stagnant-lid convection with Arrhenius rheology and the effects of mantle melting, Geophysical Journal International, 179, 154–170, 2009. SciX. ↩

  4. Y. Miyazaki, D. J. Stevenson, The stability of a dense crust situated on small bodies, The Planetary Science Journal, 5, 192, 2024. SciX. ↩

  5. D. L. Turcotte, G. Schubert, Geodynamics, 3rd edition, Cambridge University Press, 2014. SciX. ↩