Stop writing a fabricated key into every generated Kratos DEM deck

[alhermann]

One file. Deliberately minimal so it can land ahead of the full consolidation in #50, like #52 did.

What main does today

Every Kratos DEM deck OASiS generates contains PARTICLE_FRICTION — in the mdpa and in MaterialsDEM.json.

The key does not exist. Measured with grep -r -a -F against the full 28-application Kratos build: zero occurrences. The real keys are STATIC_FRICTION and DYNAMIC_FRICTION, per contact pair, in material_relations — a different place in a different schema, not a rename.

It also emitted the FEM materials schema, plus problem_data / solver_type / time_stepping (DEM reads none of them), DEM_timestep_safety_factor (the real key is DeltaTimeSafetyFactor, itself dead code), and a mesh filename DEM never opens.

The deck had never run, and a green gate said otherwise

catalog_template_executes records kratos::dem::2d_rc=0. But that gate runs the template, which only writes files and exits. It never runs the input.py the template writes. Run that:

RuntimeError: Error: Getting a value that does not exist. entry string : strategy
  kratos/sources/kratos_parameters.cpp:426

So the gate was green on the half that cannot fail. (A general screen for this shape found exactly one affected template across all 255 in the catalog — this one.)

After

kratos:dem:2d   template_rc=0   input.py_rc=0   both run

A 2-D granular settling problem: 64 CylinderParticle2D discs onto a floor between two walls, ~9 s. Physical, not merely exit-0 — all 64 particles survive, the pile comes to rest at max |v_y| = 0.005 m/s, and y_min sits one radius above the floor with a 0.1%-of-radius Hertzian overlap.

The emitted input.py now also refuses to exit 0 when the model has quietly lost its particles. That matters: Kratos DEM deletes escaped particles silently. Measured on this case, dt=2e-4 loses all 16 and dt=1e-4 loses 12 of 16 — both at exit code 0.

Scope

PARTICLE_FRICTION still appears once in the file, in a comment stating it does not exist. The generated deck writes STATIC_FRICTION / DYNAMIC_FRICTION.

The knowledge entries, fixtures and execution ledger for this work (151 rows, 151 pass, 150 discriminate; coverage 73.0% → 80.9%) sit on feature/kratos-generator and are not in this PR.

[Copilot]

Pull request overview

Updates the Kratos DEM 2D generator so it stops emitting the non-existent PARTICLE_FRICTION key and instead writes a runnable DEMApplication deck using the correct DEM schemas (flat ProjectParametersDEM.json, and MaterialsDEM.json with material_relations contact-pair friction).

Changes:

• Rewrites the DEM deck generator to emit the correct DEMApplication file set and naming conventions (<problem_name>DEM.mdpa, <problem_name>DEM_FEM_boundary.mdpa, MaterialsDEM.json, ProjectParametersDEM.json, input.py).
• Fixes friction/material modeling by moving friction to contact-pair keys (STATIC_FRICTION / DYNAMIC_FRICTION) under material_relations (removing fabricated PARTICLE_FRICTION from generated decks).
• Improves execution robustness by generating an input.py that summarizes results and exits non-zero if particles are silently lost.

Suppressed comments (2)

src/backends/kratos/generators/dem.py:97

• If dt_override is 0/negative (or the Rayleigh estimate produces NaN/inf due to invalid material params), the generated script will crash at T_end / dt with a ZeroDivisionError/ValueError. Add a small validation step so failures are explicit and actionable.

dt = float(dt_override) if dt_override is not None else dt_safety * dt_rayleigh
n_steps = max(1, int(round(T_end / dt)))

src/backends/kratos/generators/dem.py:104

• The y-fit is validated, but x-fit is not: for small domain_x (or large radius), x0 can be < radius and the initial particle centers will start inside/through the left wall. Since this is a deck generator, it's better to fail fast with a clear message when the particle block cannot fit in x within [radius, domain_x - radius].

spacing = 2.2 * radius
cols = max(1, int(math.floor((domain_x - 2.0 * spacing) / spacing)))
cols = min(cols, max(1, int(math.ceil(math.sqrt(n_particles)))))
rows = int(math.ceil(n_particles / cols))
block_w = (cols - 1) * spacing