Model

Model#

A Model is the symbolic PDE: state, parameters, closures and boundary conditions, derived from a vertical ansatz. It has no mesh and no numerics.

What you write: a model family with its options.

from zoomy_core.model.models import SME
model = SME(level=0, dimension=3, closures=[...], boundary_conditions=...)

choice	options (default)
family (`zoomy_core.model.models`)	`SME` (shallow moments; `level=0` is shallow water), `VAM` (non-hydrostatic moments, pressure modes `P_i`), `MLSWE` / `MLSME` / `MLVAM` (multilayer), `KESME`, `QRKESME`, `ElderSME`
`level`	number of velocity moments above the depth average (`SME`: 2)
`dimension`	spatial dimension including the vertical: 2 = (x, z), 3 = (x, y, z)
`closures`	stress pieces, one per stress component: `Newtonian`, `NavierSlip`, `StressFree`, `ManningFriction`, `RoughWall`, `Bingham`, `KEpsilonViscosity`, … (`[]`: unclosed)
`boundary_conditions`	per mesh tag (below)
`small_slope`	`True`: the small-slope frame (the standard shallow model)

Boundary conditions (zoomy_core.model.boundary_conditions), each on one mesh tag, collected in BoundaryConditions([...]) or passed as a list:

class	meaning
`Wall` / `FromModel(definition="wall")`	reflecting (free-slip) wall; `FromModel` takes the model’s own definition
`Dirichlet(tag, on=field, value=v)`	prescribed value of a field
`Flux(tag, on=field, gradient=g)`	prescribed normal gradient
`Extrapolation(tag)`	zero gradient (transmissive)
`InflowOutflow(tag, prescribe_fields={...})`	prescribe some fields, extrapolate the rest
`Periodic(tag, periodic_to_physical_tag=partner)`	periodic seam
`Lambda(tag, prescribe_fields={...})`	face values from a user function
`Coupled(tag)`	preCICE interface (see Backends)

Every BC feeds three channels, the data structures its value goes to:

channel	what	consumed by
`face`	the state on the boundary face	the numerical flux (Riemann problem)
`ghost`	the value at the mirror image of the inner cell centre; by default from the face value (`2 face - Q` for a face value)	gradients, limiters, derivative aux, diffusion
`normal_gradient`	a prescribed dQ/dn on the face (Neumann data; 0 unless set)	diffusion, the LSQ boundary closure

Any BC can take any channel from another BC, on the fields of that BC’s on: Wall(tag="w", ghost=Dirichlet(on="momentum", value=0)) (reflecting Riemann state, no-slip stencils), Wall(tag="w", normal_gradient=Flux(on="h", gradient=2.0)).

A new BC subclasses BoundaryCondition and implements only what differs: compute_boundary_condition (the face value; set boundary_value_kind = BoundaryValueKind.face when it is a true face value), compute_ghost_state (only if the stencils need something else), compute_boundary_gradient (Neumann data). Wall is the template: about 20 lines building the reflected state from normal and the momentum groups. The kernels are symbolic, so every backend gets them.

Model parameters are param parameters, like the solver settings (Solvers): the class declares them, the constructor’s arguments set them, and a misspelt name gets the closest one (SME(levl=1): TypeError: SME has no setting 'levl'; did you mean 'level'?).

Connects to: SystemModel.from_model(model) (SystemModel). model.describe() shows what was built. Models without a vertical ansatz (the compressible Euler equations, zoomy_core.systemmodel.euler.compressible_euler) are built directly as a SystemModel.

API: zoomy_core.model.