verify_mesh_independence

Refines the mesh step by step and reports whether the answer stopped changing.

Group: Check the answer.

Parameters

Parameter	Type	Required	Default
solver	string	yes
input_template	string	yes
resolution	number	yes
refinement_factor	number	no	2.0
levels	integer	no	1
parameter_kind	string	no	'divisions'
field	string	no	''
probe_points	string	no	''
rel_tol	number	no	0.01
job_name	string	no	''
np	integer	no	1
critic_approved	boolean	no	False

What the model reads

The text below is the tool's own description, exactly as the AI model receives it.

Show the full description

Heuristic mesh-independence study for problems WITHOUT an exact
solution: re-run the SAME problem at successively refined
resolutions and accept it as converged only if ALL monitored
quantities stop changing materially.

MMS convergence tests need a manufactured exact solution; real
application problems have none. This tool automates the
established recourse: halve the discretisation length (once by
default, more via `levels`), then compare (a) a volume-weighted
global L2 norm and the global max of the primary field, (b) the
field value at probe points (auto-chosen from the mesh — field
hotspot, domain centre, off-centre interior points — or supplied
explicitly), and (c) any scalar QoIs the script writes to
results_summary.json. Verdict: CONVERGED only if every monitored
quantity changes by less than `rel_tol` on the finest refinement
step; otherwise NOT CONVERGED, with all numbers in the report.

The input template must contain the placeholder __RESOLUTION__
where the characteristic discretisation parameter goes, e.g.
`nx = __RESOLUTION__`. For Python-scripted solvers (fenics,
ngsolve, skfem, dune) the template is the solve script itself; for
compiled/file-input solvers (fourc, dealii, kratos, febio) it is a
generator script that writes the input file, exactly as in
run_with_generator. The solve must write the primary field as
nodal data in a VTU/VTK/VTP result file.

IMPORTANT — this tool checks discretisation convergence only. It
does not validate the model physics; have the MANDATORY critic
review the setup and pass critic_approved=True as with the run
tools.

Args:
    solver: Backend name (any registered backend).
    input_template: Solve/generator script containing __RESOLUTION__.
    resolution: Coarsest value of the discretisation parameter.
    refinement_factor: Refinement per level (default 2 = halving h).
    levels: Number of refinements (default 1; runs levels+1 cases).
    parameter_kind: 'divisions' (parameter counts elements; refining
        multiplies) or 'size' (parameter is h; refining divides).
    field: Field name to monitor (default: auto-select from result).
    probe_points: Optional JSON list of probe coordinates, e.g.
        "[[0.5, 0.5], [0.25, 0.75]]" (default: auto from the mesh).
    rel_tol: Acceptance threshold on relative change (default 0.01).
    job_name: Optional study directory name.
    np: MPI processes per run.
    critic_approved: True only after the critic approved the setup.