Skip to content

A core on the PALEOS liquid iron isentrope

How much heat does a liquid iron core give up as it cools, when its structure comes from Zalmoxis and its temperature from the PALEOS equation of state (EOS)?

The short answer: Aragog keeps the core structure that Zalmoxis computed and puts the convecting core on one isentrope of the PALEOS liquid iron table. An isentrope is a curve of constant specific entropy; for a convecting liquid it is also called the adiabat, the word the Gaussian profile uses. The heat the core gives up per kelvin of cooling at the core-mantle boundary (CMB) is then an integral of the temperature along that isentrope. Aragog tabulates it once per structure, before the run; the solver steps then read the table. The Gaussian profile of the core module needs no such step.

When to use it

The core module has 2 profile sources, set with profile_source in core_module_params (parameters):

  • "gaussian", the default: a closed-form density profile with a constant thermal expansivity \(\alpha\) and heat capacity \(c_p\).
  • "paleos": the density, pressure and gravity of a Zalmoxis core, and the temperature from the PALEOS liquid iron EOS of Attia et al. (2026)1.

Use "paleos" when Zalmoxis solves the structure with PALEOS iron. Then the core structure and the isentrope come from one EOS; by default the heat capacity of the secular term does not (see Limits of the model). The Gaussian profile takes \(\alpha\) and \(c_p\) of the Earth's core, which do not hold for the cores of larger planets.

Where the structure comes from

Zalmoxis integrates the structure of the planet from the centre outward. For the core it returns, at each of its nodes, the radius \(r\) (m), the enclosed mass \(m\) (kg), the pressure \(P\) (Pa), the density \(\rho\) (kg m\(^{-3}\)) and the gravity \(g\) (m s\(^{-2}\)). These arrays go into an .npz file, which core_profile_file names.

Aragog reads the arrays once and holds them fixed until the next structure solve. Between the nodes it uses cubic Hermite curves. Their slopes come from the physics where possible: \(dm/dr = 4 \pi r^2 \rho\) and \(dP/dr = -\rho g\). The gravitational potential \(\psi\) (J kg\(^{-1}\)) is the integral of \(g\), zero at the CMB.

When the profile is built, Aragog checks the arrays against the mantle mesh and the structure:

  • The last radius must equal the CMB radius of the mesh to 1 m or a relative \(10^{-9}\), whichever is larger.
  • When m_core is given, the mass of the last node must equal it to a relative \(10^{-10}\). PROTEUS passes the core mass of its structure solve.
  • A CMB pressure of the arrays more than 1 % from that of the mesh is logged as a warning.

The isentrope of the core

The liquid core convects, so it is well mixed. A well-mixed liquid has the same specific entropy \(s\) (J kg\(^{-1}\) K\(^{-1}\)) everywhere. At the CMB temperature \(T_\text{cmb}\) this entropy is

\[ s_c = s(P_\text{cmb}, T_\text{cmb}). \]

At each node \(k\), the temperature \(T_{a,k}\) of the isentrope solves \(s(P_k, T_{a,k}) = s_c\). Aragog finds it with Newton steps on the table entropy.

The table is the liquid side of the PALEOS iron EOS, which Zalmoxis writes and core_liquid_table names. It holds the liquid also below the melting curve, where the liquid is metastable. So the isentrope stays liquid at all temperatures of the table, and no latent heat enters it. The inner core is treated separately (below).

Some cells at the top or bottom of a table did not converge when Zalmoxis wrote it, and hold NaN. Aragog drops each pressure row that holds such a cell, logs the new pressure range and uses the rows that are left. A table cut to a narrow temperature range can keep more rows than the full table. On the full default Zalmoxis grid, which ends at 100 TPa, the last full row is at 59.3 TPa (file format).

Aragog solves the isentrope on a grid of CMB temperatures, 10 K apart by default (core_t_cmb_step). The grid covers every CMB temperature whose isentrope stays inside the table. The slopes in radius and in \(T_\text{cmb}\) come from the table derivatives, and the mixed slope from a spline in \(T_\text{cmb}\), so the solver reads a surface \(T_a(r, T_\text{cmb})\) with continuous slopes. Aragog checks the entropy of this surface at 4 sets of points; at each it must equal \(s_c\) to a relative \(10^{-6}\):

  • at the nodes on the grid temperatures;
  • at the nodes between the grid temperatures;
  • at the radial midpoints;
  • at the centres of the grid cells.

What the path refuses says what happens when it does not.

The tabulation takes about 10 s on the default Zalmoxis grid: 8 to 13 s for pure-Fe cores of 1 to 7 \(M_E\) on an Apple M1 Ultra, in 2 runs.

A worked example: for the Zalmoxis core of a 1 Earth-mass planet in the tutorial, a CMB at 4000 K gives a centre at 6111 K.

The secular heat capacity

The secular term is the heat a core gives up as its whole convecting mass cools. At fixed pressure, the change of the specific enthalpy \(h\) is \(dh = T\,ds\). Every parcel shares the change of \(s_c\). At the CMB the pressure does not change, so \(ds_c = c_p\,dT_\text{cmb}/T_\text{cmb}\), with \(c_p\) (J kg\(^{-1}\) K\(^{-1}\)) the heat capacity at the CMB. The heat content \(H\) of the core then changes with \(T_\text{cmb}\) as

\[ C_s = \frac{dH}{dT_\text{cmb}} = \frac{ds_c}{dT_\text{cmb}} \int T_a \, dm = \frac{c_p}{T_\text{cmb}} \int T_a \, dm. \]

For a constant \(c_p\) and the Gaussian adiabat this is the secular capacity of the Gaussian profile.

By default \(c_p\) is 840 J kg\(^{-1}\) K\(^{-1}\) (core_cp_override), not the table value. Attia et al. (2026), p. 5, set the heat capacity at constant volume of the PALEOS liquid to \(3nR\), with \(n\) atoms per formula unit and \(R\) the gas constant, so it holds no electronic heat capacity. In the worked example above, the table gives \(c_p\) = 506 J kg\(^{-1}\) K\(^{-1}\) at the CMB at 4000 K. core_cp_override = "table" takes the table value.

In the example, \(C_s\) is \(1.989\times10^{27}\) J K\(^{-1}\) at 4000 K. That is 1.22 times \(c_p\) times the core mass. The factor 1.22 is the mass-weighted mean of \(T_a / T_\text{cmb}\).

Aragog uses only changes of \(H\). CoreEnergyBudget.heat_content counts \(H\) from the lowest tabulated temperature, so on a tabulated profile it holds a constant that depends on the structure and the table. A difference on one structure is exact; a difference between 2 structures, for example before and after a structure update, is not a heat.

Why the core energy balance closes in each solver call

The core energy balance of a solver call compares 2 numbers: the heat the core loses, and the heat that crosses the CMB. Each call writes both to its output (step_dE_core_J and step_dE_F_cmb_J); the tests compare them, and a caller can compare them too. Aragog itself does not stop on a mismatch.

Aragog tabulates \(C_s\) at the grid temperatures, with its derivative, and reads it as a cubic in \(T_\text{cmb}\). The heat change of a call is the exact integral of that same cubic over the temperature change of the call. So the solver integrates and books one function, and the balance closes to the solver tolerance.

With an inner core, the latent and gravitational heat of a call is a quadrature of their capacities over the temperature change of the call. These capacities have a kink where the inner-core boundary crosses a core node, so Aragog cuts the quadrature into pieces at freeze-out and at those temperatures.

The inner core

The inner core starts to grow when the isentrope meets the melting curve at the centre. Aragog finds this onset temperature as the \(T_\text{cmb}\) whose isentrope has its centre on the melting curve. When no isentrope of the grid meets it, the onset is the end of the grid: the lowest grid temperature when the centre stays liquid on the whole grid, the highest when the centre is frozen on the whole grid. In both cases the value is a grid end, not a nucleation temperature. The latent heat per unit mass is \(T_\text{icb}\,\Delta s\), with \(T_\text{icb}\) the temperature at the inner-core boundary and \(\Delta s = 172.8\) J kg\(^{-1}\) K\(^{-1}\) (ds_fusion), or the constant latent_heat when it is set. The latent and gravitational terms of the budget are otherwise unchanged.

The constant \(\Delta s\) lies 10.5 % above the ab initio 1.05 \(k_B\) per atom at 300 GPa of Gonzรกlez-Cataldo & Militzer (2023)2, with \(k_B\) the Boltzmann constant.

What the path refuses

  • A core whose pressures leave the pressure range of the table that core_liquid_table names stops the build, with an error that names both ranges; give a table that covers the core. The range comes from the table: 45 GPa to 3 TPa for the cut paleos_iron_liquid_cut_3tpa.npz in tests/validation/data/.
  • A cell that did not converge (NaN) inside the table stops the read.
  • The density of the arrays must integrate to the node mass \(m\) at the CMB within \(10^{-2}\). The stored Zalmoxis cores in tests/validation/data/ miss by at most \(5\times10^{-6}\) with one phase, and by \(1.08\times10^{-3}\) with an inner-core boundary, from its density jump; a density 5 % too large misses by \(5\times10^{-2}\). A structure whose pressure bends between nodes so far that the isentrope leaves the table stops with an error that points to the core arrays.
  • The CMB temperature must lie on the grid of isentropes: every CMB temperature whose isentrope stays inside the table at every node. Outside it, or at a NaN temperature, the solver and CoreModule stop with an error that names the temperature.
  • The grid holds at most 2,000,000 cells, nodes times temperatures; make core_t_cmb_step larger for a larger grid.
  • An isentrope that misses \(s_c\) by more than a relative \(10^{-6}\) stops the build with an error that names where: between the grid temperatures (make core_t_cmb_step smaller), between the radial nodes (give the core arrays more nodes), or inside the grid cells (either).
  • stratification = true and capacity_mode = "legacy" are refused: the heat content of a stratified core depends on the CMB heat flow, not on \(T_\text{cmb}\) alone.
  • The dynamo diagnostics, the entropy margin and the field strength, are not computed on this path. Their entropy budget (CoreEntropyBudget) takes a constant \(\alpha\) and \(c_p\) and refuses a tabulated profile.

Limits of the model

  • The Gaussian profile keys are accepted and unused (parameters). The secular heat capacity comes from core_cp_override (840 J kg\(^{-1}\) K\(^{-1}\) by default) and the isentrope shape from the PALEOS \(c_p\), so the two are not from one EOS.
  • The liquid of the PALEOS table is calibrated between 8000 and 14,000 K and between 50 and 1300 GPa (Attia et al. (2026), p. 5). The CoreModule diagnostic core_eos_outside_calibration gives the core mass fraction outside this window; the run continues.
  • The isentrope is that of pure liquid iron, also for a density with light elements.
  • The profile reads only r, m, P, rho and g of the core arrays, not phase or M_ic. It takes the density as given, solid or liquid, and finds its own inner-core boundary where the liquid isentrope meets the melting curve.
  • A fully solid core keeps its temperature on the liquid isentrope of its \(T_\text{cmb}\). Aragog models no heat conduction in the solid core.
  • The entropy of fusion is one constant at all pressures. The ab initio value falls from 1.05 \(k_B\) per atom at 300 GPa to 0.8 \(k_B\) per atom at 1000 GPa (Gonzรกlez-Cataldo & Militzer (2023), p. 9), so the constant is 1.45 times the value at 1000 GPa.
  • Above about 2.4 TPa the alpha column of the PALEOS liquid iron table departs from its density and entropy columns. These 2 agree: with \(\alpha = -(\partial \ln\rho/\partial T)_P\) from the density, the Maxwell relation \((\partial s/\partial P)_T = -\alpha/\rho\) holds to 1 % up to 5 TPa. With the alpha column it misses by \(10^{-4}\) at 2.4 TPa and \(2\times10^{-2}\) at 3 TPa. From 3.8 TPa the density gives a negative \(\alpha\); the alpha and nabla_ad columns stay positive, alpha within 0.01 % of its floor of \(10^{-7}\) K\(^{-1}\) in 569 cells from 4.5 TPa.
  • Aragog builds the isentrope from the entropy column. In a core whose centre pressure is above 3.8 TPa, the temperature therefore rises outward from the centre up to about 3.75 TPa. The stored pure-Fe 10 \(M_E\) core (centre 4.7 TPa) is 48 K hotter there than at its centre for \(T_\text{cmb} = 3000\) K, and 99 K hotter for 6000 K. A pure-Fe 7 \(M_E\) core (centre 3.1 TPa) is at the edge of this region, and its temperature still falls outward. Aragog builds these cores with no warning.
  • On the cut paleos_iron_liquid_cut_3tpa.npz the spline keeps the Maxwell relation with the alpha column to \(10^{-5}\) at the nodes from 50 GPa to 2.1 TPa and to \(1.3\times10^{-5}\) between the nodes below 2 TPa; at 45 GPa, the lowest pressure of the cut, it misses by \(3.6\times10^{-5}\).

Where this is checked. Tabulated core on the Validation tab.


  1. M. Attia, T. Lichtenberg, E. Jungovรก, M. Sastre, PALEOS: Multiphase equations of state and mass-radius relations for exoplanet interiors, Astronomy & Astrophysics, 713, A225, 2026. arXiv:2605.03741. ↩

  2. F. Gonzรกlez-Cataldo, B. Militzer, Ab initio determination of iron melting at terapascal pressures and Super-Earths core crystallization, Physical Review Research, 5, 033194, 2023. SciX. ↩