Parameter file reference

Every MILET run is fully specified by a single TOML file in cases/. The loader (load_config_toml in src/io/TOMLLoader.jl) parses the file, deep-merges it onto cases/_defaults.toml, and builds the immutable Config struct (src/core/Config.jl) that every module reads. A minimal case therefore only states what differs from the defaults.

julia --project=. main.jl cases/my_case.toml

This page documents every key the loader reads, grouped by section, with type, default, and units. The Default column shows the value shipped in cases/_defaults.toml; keys marked (loader) are not present in the defaults file and fall back to the hard-coded default inside load_config_toml.

Loading semantics and precedence

The loader applies three deterministic transformations (src/io/TOMLLoader.jl):

1. Deep merge (_deep_merge): the user file is merged onto cases/_defaults.toml, with user values winning on every key collision, recursively through nested tables.

2. Section flattening (_flatten_sections): all [section] tables are flattened into one namespace. Keys are promoted bare ([time] ntot → ntot); for the sections in _PREFIX_SECTIONS (output, checkpoint, free_surface, amr, anisotropic_heating, boundary_velocity, boundary_temperature, solver, gravity, twophase) a prefixed alias is added as well ([output] dir → output_dir, [solver] type → solver_type), which is the name the loader actually reads. You may write either form.

3. [model]-last precedence: sections are flattened in sorted alphabetical order, with the legacy flat [model] block flattened last, so an explicit key in [model] always beats a section-promoted value. This rule is deterministic by construction — earlier versions relied on Dict iteration order, which could silently load a 3D case as 2D depending on hash order.

The array-of-tables ([[rock]], [[phase]], [[melt]], [[layer]], [[init_T]]) are replaced wholesale, never merged: if your file defines any [[rock]] entry, the defaults' rock list is discarded entirely (materials are case-specific).

Sectioned layout for nondimensional cases Benchmarks that override many defaults (Blankenbach, SolCx) must use the sectioned layout — see the header comment in cases/blankenbach_1a.toml. A flat layout that relies on bare-key promotion can lose to defaults during flattening. :::

[model]

Key	Type	Default	Meaning
case	string	"subduction"	Case identifier, dispatched by select_case (src/io/TOMLLoader.jl) to the initial rock-type / temperature closures. Known names: subduction, subduction_init, extension, falling_block, plume, plume3d, rt3d, busse, blankenbach, solcx, chunk. Unknown names fall back to the data-driven [[layer]] / [[init_T]] path.

Any other key placed in [model] is treated as a flat top-level key and, by the precedence rule above, overrides the same key from any section.

[mesh]

Key	Type	Default	Units	Meaning
dim	int	2	—	2 or 3
xsize	float	4.0e6	m	Domain extent in x (required for box geometry)
ysize	float	1.4e6	m	Domain extent in y; y grows downward (y = 0 is the surface)
zsize	float	0.0	m	Domain extent in z (3D only)
nx, ny, nz	int	101, 36, 1	—	Grid lines per axis (cells = n − 1)
mnx, mny, mnz	int	4, 4, 1	—	Markers per cell per axis
geometry	string	"box"	—	box | chunk | sphere_shell. Non-box geometries derive xsize/ysize from r_outer − r_inner if absent
r_inner	float	3.481e6	m	Inner radius (chunk/shell)
r_outer	float	6.371e6	m	Outer radius (chunk/shell)
chunk_theta_extent	float	1.0472 (π/3)	rad	Angular extent θ of the chunk
chunk_phi_extent	float	1.0472	rad	Angular extent φ (3D chunk)

[gravity]

Key	Type	Default	Units	Meaning
gx, gy, gz	float	0.0, 9.81, 0.0	m/s²	Constant Cartesian gravity. gy > 0 points down (y-down convention)
model	string	"vertical_const"	—	vertical_const | radial (gravity_fn in src/physics/Gravity.jl; radial points toward the origin)
g_mag	float	9.81	m/s²	Magnitude for radial gravity

[time]

Key	Type	Default	Units	Meaning
maxxystep	float	0.5	—	Courant fraction for both the advection and the thermal-diffusion limit (compute_dt, src/time_stepping/CFL.jl)
maxtkstep	float	50.0	K	Maximum temperature change per step
maxtmstep	float	1.5e13	s	Hard upper bound on Δt (≈ 500 kyr)
ntot	int	100	—	Total number of time steps
pinit	float	0.0	Pa	Reference pressure offset
t_end_myr	float	0.0 (loader)	Myr	Stop when simulated time exceeds this; 0 disables (steps then end only at ntot)

The time step is $\mathrm{\Delta }t=max(min(Ch/v_{max},C{h}^2\rho c_p/k,\mathtt{\text{maxtmstep}}),\mathtt{\text{maxtmstep}}\cdot {10}^{-8})$ with $C$ = maxxystep. With two-phase flow enabled, the melt segregation speed (max_melt_segregation_speed) is added to the CFL constraint.

[rheology]

Global limits applied after the per-rock clamps (the effective bound is clamp(η, max(eta_min, rock.eta_min), min(eta_max, rock.eta_max)) — _compute_viscosity_core in src/materials/Rheology.jl).

Key	Type	Default	Units	Meaning
eta_min	float	1e19	Pa·s	Global lower viscosity clamp
eta_max	float	1e25	Pa·s	Global upper viscosity clamp
str_min	float	1e-18	1/s	Minimum strain-rate invariant (regularizes plastic viscosity and strain accumulation)
eii_seed	float	0.0	1/s	Initial ${\dot{\epsilon }}_{II}$ for the first Picard iteration so power-law creep engages on step 1
body_force_full	bool	false	—	false: Boussinesq perturbation buoyancy $(\rho -\overline{\rho })g$; true: full $\rho g$ body force
formulation	string	"boussinesq"	—	boussinesq | ala | tala — anelastic forms change the continuity equation to $\mathrm{\nabla }\cdot ({\rho }_{\mathrm{a}\mathrm{d}\mathrm{i}}u)=0$ (src/physics/Compressibility.jl)
rho_ref_ala	float	3300.0	kg/m³	Reference surface density for the ALA adiabatic profile
H_scale	float	2.9e6	m	Density scale height for the ALA profile
grain_exponent	float	0.0	—	Legacy grain-size modifier on the eta0 path: $\eta \ast =(d/d_{\mathrm{r}\mathrm{e}\mathrm{f}}{)}^m$; 0 = off (the calibrated grain-size law uses diff_* instead)
strain_gate_plastic	bool	true (loader)	—	Accumulate weakening strain only weighted by the per-cell yield fraction (see Softening & healing); false = legacy total-strain accumulation

[temperature]

Key	Type	Default	Units	Meaning
T_top	float	273.0	K	Surface Dirichlet temperature
T_bot	float	1623.0	K	Bottom Dirichlet temperature
T_ref_density	float	298.15	K	Reference temperature for the density expansion ${\rho }_0(1-\alpha (T-T_{\mathrm{r}\mathrm{e}\mathrm{f}}))$
rho_ref_heat	float	3300.0	kg/m³	Reference ρ in the $\rho c_p$ heat-capacity term
marker_T_mode	string	"pic_increment"	—	pic_increment: markers carry T and advect it; the FE solve contributes only the diffusion + sources increment (I2ELVIS/PIC pattern, validated on Blankenbach). overwrite: the grid solution replaces marker T each step — with the default diffusion-only heat operator temperature does not advect; only use deliberately (e.g. with do_supg_heat = true)
heat_stabilisation	string	"none"	—	none | supg | entropy — stabilisation for FE temperature advection (see Heat transport)
H_const	float	0.0	W/m³	Uniform constant volumetric heating added to every cell

[physics] — toggles

All booleans, read by run_case (src/time_stepping/Driver.jl):

Key	Default	Meaning
do_stokes	true	Solve momentum each step
do_heat	true	Solve the energy equation each step
do_advect	true	Advect markers (RK4, advect_markers!)
do_adiabatic	false	Adiabatic heating $H=\alpha T{\rho }_{\mathrm{r}\mathrm{e}\mathrm{f}}g{u}_y$
do_shear_heating	true	Viscous dissipation $H=4\eta {\dot{\epsilon }}_{II}^2$
do_friction	true	Drucker–Prager plastic yield + strain weakening
do_phase_change	false	Per-marker phase transitions ([[phase]])
do_melting	false	Katz-lite melting + extraction ([[melt]])
do_free_surface	true	Sticky-air free-surface handling
do_radiogenic	true	Radiogenic heating $H=\rho h_r$ per rock
do_supg_heat	false	FE advection of T with SUPG stabilisation (instead of marker advection)
do_composition_fe	false	Additional smooth Q1 composition field, advected on the FE grid
buoyancy_from_T	false (loader)	Buoyancy directly from the FE temperature field, $\rho g=-{\rho }_0\alpha Tg$ (used by nondimensional convection benchmarks)
do_subgrid_diffusion	false (loader)	Subgrid-scale marker–grid temperature relaxation (Gerya & Yuen 2003, apply_subgrid_diffusion!)
subgrid_d	1.0 (loader)	Subgrid diffusion strength coefficient d
do_volatile	false (loader)	Multi-species volatile transport — part of a separate study; disabled in this release
n_comp_fields	0 (loader)	Number of independent FE composition fields when do_composition_fe

[boundary_velocity]

Per-wall mechanical BC type; walls are left, right, surface (y = 0), deep (y = ysize), and front/back in 3D.

Key	Type	Default	Meaning
bc_left, bc_right, bc_surface, bc_deep, bc_front, bc_back	string	"prescribed"	prescribed | free_slip | traction
free_slip_penalty	float	1000.0	Penalty factor γ in the weak no-penetration term ${\int }_{\mathrm{\Gamma }}\gamma (u\cdot n)(v\cdot n)d\mathrm{\Gamma }$
t_traction_surface_n / _t	float	0.0	Normal / tangential traction (Pa) when bc_surface = "traction"
t_traction_deep_n / _t	float	0.0	As above for the deep wall
t_traction_left_n / _t	float	0.0	As above for the left wall
t_traction_right_n / _t	float	0.0	As above for the right wall

• prescribed — strong Dirichlet from the per-wall closures bc_left(cfg), bc_right(cfg), … (src/boundary_conditions/Velocity.jl); for the subduction case these inject the plate velocities vx_plate / vx_plate_right within the plate-depth windows.

• free_slip — only $u\cdot n=0$ enforced weakly (penalty term added in build_free_slip_bcs, src/physics/Stokes.jl); tangential motion free.

• traction — Neumann $\sigma \cdot n=t$ via build_traction_bcs.

[boundary_temperature]

Key	Type	Default	Units	Meaning
bcT_surface, bcT_deep	string	"dirichlet"	—	dirichlet | flux | insulating | robin
bcT_left, bcT_right, bcT_front, bcT_back	string	"insulating" (loader)	—	Lateral wall T BCs (insulating = natural zero-flux)
q_surface, q_deep	float	0.0	W/m²	Prescribed flux when type = flux ($-k\mathrm{\nabla }T\cdot n=q$)
h_robin	float	0.0	W/m²/K	Robin coefficient: $-k\mathrm{\nabla }T\cdot n=h(T-T_{\mathrm{a}\mathrm{m}\mathrm{b}})$
T_amb	float	298.15	K	Ambient temperature for the Robin BC

Flux and Robin contributions are assembled by build_heat_flux_bcs and assemble_heat (src/physics/Heat.jl).

[free_surface]

Key	Type	Default	Meaning
do_ale_free_surface	bool	false	ALE vertical mesh stretching following the rock–air interface (Kaus et al. 2010; src/physics/ALEFreeSurface.jl)
fs_stabilisation	float	0.5	Implicit damping θ in the Kaus 2010 stability factor (0 = none, 1 = strong)
surface_smooth	float	0.0	Horizontal Laplacian smoothing of the marker surface (smooth_air_interface!)
air_rock_id	int	1	Which [[rock]] index represents sticky air

See Free surface.

[anisotropic_heating]

Fabric-aligned anisotropic shear heating (anisotropic_shear_heating, src/physics/AdvancedHeating.jl).

Key	Type	Default	Meaning
fabric_nx, fabric_ny	float	0.0	Fabric direction components; (0, 0) disables
fabric_back_stress	float	1.0	Back-stress amplification factor

[amr]

Adaptive mesh refinement (src/solvers/AMR.jl). Marker loops fall back to serial Gridap point location on adapted meshes — AMR currently slows runs substantially (see AMR & ALE).

Key	Type	Default	Meaning
do_amr	bool	false	Enable adaptive refinement
amr_every	int	10	Adapt every N steps
amr_refine_fraction	float	0.3	Top fraction of cells (by indicator) refined
amr_coarsen_fraction	float	0.0 (loader)	Bottom fraction coarsened
amr_max_level	int	2 (loader)	Refinement-level cap
amr_criterion	string	"T_gradient"	T_gradient | composition | strain_rate | viscosity | user (amr_indicator)

[fe]

Key	Type	Default	Meaning
fe_order_v	int	2	Velocity polynomial order (Q2, Taylor–Hood)
fe_order_p	int	1	Pressure order (Q1)
fe_order_T	int	1	Temperature order (Q1)
n_picard	int	3	Max nonlinear (Picard) iterations per step
picard_tol	float	1e-3	Relative convergence $\Vert \mathrm{\Delta }u\Vert /\Vert u\Vert$
nonlinear_method	string	"picard"	picard | newton (damped Picard update) | full_newton (AD Jacobian, newton_stokes!)
newton_damping	float	0.5	Under-relaxation α ∈ (0, 1] for newton; α = 1 recovers Picard

See Time stepping and picard_stokes! (src/time_stepping/PicardLoop.jl).

[solver]

Key	Type	Default	Meaning
type	string	"mumps"	mumps (cached direct LU) | gmg (alias vcycle; FGMRES + geometric multigrid) | auto (gmg above 400k DOFs, mumps below — resolve_solver_type) | petsc | bamgs
petsc_preset	string	"lu_mumps"	lu_mumps | fgmres_gamg | bcgs_ilu | fgmres_fieldsplit | custom (petsc_options in src/solvers/PETScSolvers.jl)
petsc_options	string	""	Raw PETSc CLI string when petsc_preset = "custom"
krylov_tol	float	1e-7	Relative residual tolerance for the Krylov solve
krylov_maxit	int	200	Maximum Krylov iterations
krylov_restart	int	30	FGMRES restart length (gmg path)
gmg_levels	int	3	Multigrid depth (auto-clamped to what the mesh allows)
gmg_cheb_degree	int	3	Chebyshev smoother degree per pre/post sweep
gmg_cycles	int	1	V-cycles per preconditioner application
gmg_coarse_dofs	int	2000	Stop coarsening below this many velocity DOFs

The iterative path is right-preconditioned FGMRES on the monolithic saddle-point system with a block preconditioner: Galerkin geometric multigrid on the velocity block, η-weighted lumped pressure mass on the Schur complement (see Linear solvers and Geometric multigrid).

[checkpoint]

Key	Type	Default	Meaning
do_checkpoint	bool	false (loader)	Master enable for periodic checkpoints
every	int	0	Checkpoint every N steps (checkpoint_every; both do_checkpoint = true and every > 0 are required)
restart	bool	false	Resume from the latest checkpoint found in the output directory (load_latest_checkpoint, src/io/Checkpoint.jl)

[output]

Key	Type	Default	Meaning
dir	string	"output"	Output directory (output_dir)
every	int	10	Write VTU snapshots + PVD collections every N steps
diagnostics_every	int	1	Full marker/DOF diagnostics (O(n_markers) scans, isfinite health checks in record_step!) every N steps; the cheap per-steprun.log line is always written. Raise (e.g. 25) for large runs
projection_mode	string	"host"	Marker→cell projection: host (host-cell mean, sharp) | bilinear (2×2 stencil scatter, smoothed)

[subduction]

Only used when case = "subduction" (wall plate driving + slab geometry, rock_type_subduction / bc_left / bc_right).

Key	Type	Default	Units	Meaning
trench_x	float	2.0e6	m	Initial trench position
plate_thickness	float	8.0e4	m	Oceanic plate thickness
plate_age	float	4.0e7	yr	Plate age for the half-space-cooling geotherm
continent_thickness	float	1.2e5	m	Overriding continental plate thickness
air_thick	float	4.0e4	m	Sticky-air layer thickness
x_plate_left, x_plate_right	float	2.94e5, 3.706e6	m	Plate-velocity window bounds on the side walls
vx_plate	float	9.51173e-10	m/s	Left-wall plate velocity (≈ 3 cm/yr)
vx_plate_right	float	−9.51173e-10	m/s	Right-wall plate velocity
vx_plate_cm_yr, vx_plate_right_cm_yr	float	—	cm/yr	Readable alternatives; converted to m/s by the loader and taking precedence over the m/s keys

[twophase]

McKenzie two-phase flow — porous melt migration through a compacting viscous matrix. Three-field $(u_s,p_f,p_c)$ solve plus marker porosity transport; see Two-phase flow, doc/design/two_phase_flow.md and src/physics/TwoPhase.jl.

Key	Type	Default	Units	Meaning
do_two_phase	bool	false	—	Enable the 3-field two-phase solve
k0	float	1e-12	m²	Reference permeability at φ₀: $k_{\varphi }=k_0((\varphi +{\varphi }_{min})/{\varphi }_0{)}^n$
phi0	float	0.01	—	Reference porosity
n_perm	float	3.0	—	Permeability exponent n
mu_fluid	float	1.0	Pa·s	Melt viscosity μ_f
rho_fluid	float	2800.0	kg/m³	Melt density (mixture buoyancy $\overline{\rho }=(1-\varphi ){\rho }_s+\varphi {\rho }_f$)
alpha_phi	float	27.0	—	Matrix melt-weakening exponent: ${\eta }_{\varphi }={\eta }_s{e}^{-{\alpha }_{\varphi }\varphi }$ (Mei et al. 2002)
phi_min	float	1e-4	—	Regularization porosity for the φ → 0 limit (recovers single-phase Stokes, enforced by test/smoke/test_twophase.jl)
phi_max	float	0.5	—	Cap on marker porosity
pf_surface0	bool	false	—	Homogeneous Dirichlet $p_f=0$ on the surface (melt escapes freely)
phi_init	float	0.0	—	Uniform initial marker porosity

[[rock]] — material parameters

One entry per rock type; rocks are referenced by 1-based index from [[layer]], [[phase]], [[melt]], and air_rock_id. Each key also accepts the legacy I3ELVIS shorthand (table at the end); canonical names win when both are present.

Viscous creep

Key	Default	Units	Meaning
eta0	1e22	Pa·s	Pre-exponential viscosity for the Arrhenius path: $\eta ={\eta }_0\exp ((Q+VP)/(nRT))$, n = 1 diffusion-like, n > 1 dislocation
activation_energy	0.0	J/mol	Q
activation_volume	0.0	J/(mol·bar)	V (note the bar-based convention on the eta0 path — see src/materials/Rheology.jl)
stress_exponent	1.0	—	n
ref_yield_stress	0.0	Pa	${\sigma }_{\mathrm{r}\mathrm{e}\mathrm{f}}$ for the smooth power-law blend $\eta ={\eta }_{\mathrm{d}\mathrm{i}\mathrm{f}\mathrm{f}}/(1+x)$, $x=(2\eta {\dot{\epsilon }}_{II}/{\sigma }_{\mathrm{r}\mathrm{e}\mathrm{f}}{)}^{1-1/n}$
disl_prof	""	—	Named dislocation-creep law from the catalog in src/materials/FlowLaws.jl (e.g. "Dry_Olivine-Ranalli_1995", "Dry_Olivine_disl-Hirth_Kohlstedt_2003"); overrides creep_A, activation_energy, activation_volume, stress_exponent
creep_A	0.0	MPa⁻ⁿ s⁻¹	Explicit dislocation-creep pre-exponential, ${\dot{\epsilon }}_{II}=A{\sigma }_{II}^n\exp (-(E+PV)/RT)$; > 0 switches from the eta0 path to dislocation_viscosity

Diffusion creep (grain-size-sensitive)

$${\dot{\epsilon }}_{II}=A{\sigma }_{II}{d}^{-m}\exp (-(E+PV)/RT)$$

, lab convention A in MPa⁻¹ µm^m s⁻¹ (Hirth & Kohlstedt 2003); acts in parallel with dislocation creep (compliances add). diff_A = 0 (default) disables the mechanism. See Grain-size evolution.

Key	Default	Units	Meaning
diff_prof	""	—	Named diffusion-creep law ("Dry_Olivine_diff-Hirth_Kohlstedt_2003", "Wet_Olivine_diff-Hirth_Kohlstedt_2003"); overrides the explicit keys below
diff_A	0.0	MPa⁻¹ µm^m s⁻¹	Pre-exponential
diff_m	3.0	—	Grain-size exponent m
diff_E	375e3	J/mol	Activation energy
diff_V	6e-6	m³/mol	Activation volume

Plastic yield (Drucker–Prager + strain weakening)

$${\tau }_y=min(C({\epsilon }_p)+\sin \phi ({\epsilon }_p)P,{\sigma }_{y,max})$$

, with C and sin φ ramping linearly over accumulated plastic strain ${\epsilon }_p\in [{\epsilon }_0,{\epsilon }_f]$.

Key	Default	Units	Meaning
cohesion_init, cohesion_weak	0.0	Pa	C₀, C_f
friction_init, friction_weak	0.0	—	sin φ₀, sin φ_f
strain_thresh_init, strain_thresh_weak	0.0, 0.1	—	ε₀, ε_f bracketing the weakening ramp
yield_stress_max	5.0e8	Pa	Absolute Drucker–Prager yield cap (Kaus 2010, Glerum et al. 2018); prevents unbounded plastic stress at depth. ≤ 0 disables

Viscous softening & healing

See Softening & healing and doc/design/softening_healing.md.

Key	Default	Units	Meaning
visc_soft_factor	1.0	—	Viscous strain softening (fabric/CPO): the aggregate viscous viscosity ramps 1 → factor over [strain_thresh_init, strain_thresh_weak]; 1.0 = off
heal_tau0	0.0	s	Healing (annealing) time at heal_Tref: ${\epsilon }_p\leftarrow {\epsilon }_p{e}^{-\mathrm{\Delta }t/\tau (T)}$ with $\tau ={\tau }_0\exp (\frac{E_h}{R}(\frac{1}{T}-\frac{1}{T_{\mathrm{r}\mathrm{e}\mathrm{f}}}))$; 0 = no healing
heal_E	3e5	J/mol	Healing activation energy
heal_Tref	1000.0	K	Healing reference temperature

Grain-size evolution

Austin–Evans (2007) paleowattmeter, integrated per marker (update_grain_size, src/materials/GrainSize.jl): $\dot{d}=\frac{G_0}{p{d}^{p-1}}{e}^{-(E_g+P{V}_g)/RT}-\frac{\lambda }{c\gamma }{d}^2{\mathrm{\Psi }}_{\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{l}}$. grain_G0 = 0 (default) freezes the grain size.

Key	Default	Units	Meaning
grain_G0	0.0	m^p/s	Normal-growth prefactor
grain_p	3.0	—	Growth exponent
grain_Eg	350e3	J/mol	Growth activation energy
grain_Vg	8e-6	m³/mol	Growth activation volume
grain_lambda	0.1	—	Fraction of dislocation work stored as grain-boundary surface energy
grain_gamma	1.0	J/m²	Grain-boundary energy
grain_c	9.4248 (3π)	—	Geometric constant
grain_init	1e-3	m	Initial marker grain size

Density, thermal, elastic, misc

Key	Default	Units	Meaning
reference_density	3300.0	kg/m³	ρ₀
thermal_expansion	3e-5	1/K	α
compressibility	0.0	1/Pa	β
specific_heat	1000.0	J/kg/K	c_p
thermal_conductivity	3.0	W/m/K	k
radiogenic_heating	0.0	W/m³	h_r
shear_modulus	0.0	Pa	G — Maxwell visco-elasticity; 0 = pure visco-plastic
immobile	false	—	Marker never advects (sticky-air anchor)
eta_min, eta_max	1e17, 1e26	Pa·s	Per-rock viscosity clamps

Volatile species keys

The loader also accepts per-species volatile parameters (sulfur_*, graphite_*, carbonatite_*, water_*, each with _capacity (wt%), _release_T (K), _release_frac (per step), _solidus_drop (K); legacy volatile_* keys map to carbonatite). The volatile transport module is part of a separate study and is disabled in this release (do_volatile = false).

[[phase]] — phase transitions

Applied per marker by apply_phase_change! (src/materials/PhaseChange.jl) when do_phase_change = true.

Key	Default	Units	Meaning
rock_from	(required)	—	Rock id the transition applies to
rock_to	(required)	—	Rock id after crossing (same id → only Δρ + latent)
transition	1	—	1 = Clapeyron P–T line; 2 = depth-triggered; 3 = divariant rate-limited eclogitization
T_ref	0.0	K	Reference point on the transition line
P_ref	0.0	Pa	Reference pressure. Mode 2 reinterprets P_ref as the trigger depth in metres (a marker crosses when y > P_ref)
clapeyron	0.0	K/Pa (mode 1); Pa/K (mode 3)	Mode 1: $T_{\mathrm{l}\mathrm{o}\mathrm{c}\mathrm{u}\mathrm{s}}(P)=T_{\mathrm{r}\mathrm{e}\mathrm{f}}+c(P-P_{\mathrm{r}\mathrm{e}\mathrm{f}})$, positive = exothermic. Mode 3: $P_c(T)=P_{\mathrm{r}\mathrm{e}\mathrm{f}}+c(T-T_{\mathrm{r}\mathrm{e}\mathrm{f}})$
drho	0.0	kg/m³	Density jump across the transition (accumulated on the marker phase_drho)
latent	0.0	J/kg	Latent heat (positive = endothermic, absorbed)
band_width	0.0	Pa	Mode 3: divariant band width — $X_{\mathrm{e}\mathrm{q}}=\mathrm{clamp}((P-P_c)/\mathrm{\Delta }P+\frac{1}{2},0,1)$
react_tau_ref	0.0	s	Mode 3: reaction time at react_Tref; 0 = instantaneous equilibrium
react_Q	0.0	J/mol	Mode 3: reaction activation energy (Arrhenius τ)
react_Tref	973.15	K	Mode 3: reference T for react_tau_ref
weaken_decades	0.0	—	Mode 3: reaction softening $\eta \leftarrow \eta {10}^{-wX}$ with X the reaction progress

See Phase changes.

[[melt]] — melting laws (Katz lite)

Linear melt fraction between a quadratic-in-P solidus and liquidus (melt_fraction, apply_melting! in src/materials/Melting.jl); enabled by do_melting = true.

Key	Default	Units	Meaning
rock_id	(required)	—	Rock the law applies to
T_sol0	1373.0	K	Solidus at P = 0: $T_{\mathrm{s}\mathrm{o}\mathrm{l}}(P)=T_{\mathrm{s}\mathrm{o}\mathrm{l},0}+a_{\mathrm{s}\mathrm{o}\mathrm{l}}P+b_{\mathrm{s}\mathrm{o}\mathrm{l}}{P}^2$ (P in Pa)
a_sol, b_sol	0.0	K/Pa, K/Pa²	Solidus pressure coefficients
T_liq0	1973.0	K	Liquidus at P = 0
a_liq, b_liq	0.0	K/Pa, K/Pa²	Liquidus pressure coefficients
extract_threshold	0.05	—	F above which melt is extracted and the marker switches rock type
density_melt	2800.0	kg/m³	Melt density for the two-phase ρ mixture
latent	L_MELT (4e5)	J/kg	Latent heat of fusion
nu_factor	25.0	—	Melt weakening: $\eta \leftarrow \eta e^{-\nu F}$ (Mei et al. 2002)
rock_after_extract	rock_id	—	Depleted-residue rock id after extraction

With do_two_phase = true, a melt-fraction increase becomes marker porosity (percolating melt) instead of instant extraction (melt_to_porosity!). See Melting.

[[layer]] — initial geometry (data-driven)

The initial rock-type field is built from [[layer]] entries (rock_type_from_layers, src/initialization/Geometry.jl) whenever the list is non-empty; later layers override earlier ones at overlapping points.

Key	Default	Units	Meaning
name	"unnamed"	—	Label (diagnostics only)
rock_id	(required)	—	1-based index into [[rock]]
shape	"background"	—	background | rectangle | polygon | bent_band
x_min, x_max, y_min, y_max	±Inf	m	Bounds for rectangle
vertices	[]	m	List of [x, y] pairs: closed polygon (polygon) or polyline (bent_band)
thickness	0.0	m	bent_band: band of this normal thickness extending downward from the polyline

[[init_T]] — initial temperature regions (data-driven)

Used by initial_T_from_regions (src/initialization/Temperature.jl); same shape vocabulary and later-overrides-earlier semantics as [[layer]].

Key	Default	Units	Meaning
name	"unnamed"	—	Label
shape	"background"	—	background | rectangle | polygon | bent_band
profile	"constant"	—	constant | linear | hsc | gaussian
x_min, x_max, y_min, y_max	±Inf	m	Region bounds (rectangle)
vertices	[]	m	Polygon / polyline vertices
thickness	0.0	m	bent_band thickness
T_value	0.0	K	constant: T everywhere in the region
T_top, T_bot	273.0, 1623.0	K	linear: $T=T_{\mathrm{t}\mathrm{o}\mathrm{p}}+(T_{\mathrm{b}\mathrm{o}\mathrm{t}}-T_{\mathrm{t}\mathrm{o}\mathrm{p}})y_{\mathrm{l}\mathrm{o}\mathrm{c}\mathrm{a}\mathrm{l}}/Y$
T_surf, T_mantle	273.0, 1593.0	K	hsc half-space cooling: $T=T_s+(T_m-T_s)\mathrm{erf}(y_{\mathrm{l}\mathrm{o}\mathrm{c}\mathrm{a}\mathrm{l}}/2\sqrt{\kappa t})$
plate_age_yr	50e6	yr	hsc cooling age t
kappa	1e-6	m²/s	hsc thermal diffusivity
xc, yc	0.0	m	gaussian anomaly centre
r0	100e3	m	gaussian radius
dT	0.0	K	gaussian amplitude: $T=T_{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}}+\mathrm{\Delta }T{e}^{-r^2/r_0^2}$
T_base	1593.0	K	gaussian background T

y_local is depth within the region (depends on shape — see the header of src/initialization/Temperature.jl).

Legacy / shorthand rock keys

Accepted for backwards compatibility; canonical names win when both appear.

Canonical	Legacy	Canonical	Legacy
eta0	nu0	eta_min / eta_max	nu_min / nu_max
activation_energy	dh	reference_density	ro0
activation_volume	dv	thermal_expansion	alpha
ref_yield_stress	ss	compressibility	beta
stress_exponent	mm	specific_heat	cp
cohesion_init / cohesion_weak	a0 / a1	thermal_conductivity	kt
friction_init / friction_weak	b0 / b1	radiogenic_heating	ht
strain_thresh_init / strain_thresh_weak	e0 / e1	shear_modulus	G

Minimal example

[mesh]
xsize = 1_000_000.0
ysize = 1_000_000.0
nx    = 81
ny    = 81

[physics]
do_heat = false

[[rock]]
eta0 = 1e21
reference_density = 3300.0

[[layer]]
rock_id = 1
shape   = "background"

Everything else inherits from cases/_defaults.toml. For complete worked examples see the examples by tectonic regime and the case files under cases/; for the function-level API see the function reference.