Time integration and restart

Steady workflows

RANS and CHT use a positive config.max_iterations ceiling. This limits work; it does not guarantee convergence. Set convergence requirements for the engineering quantities you will report, and inspect histories after the run.

URANS

The public transient request requires:

  • simulation_type="urans" and architecture="fluxcore";

  • an owned, compatible restart_artifact_id;

  • max_iterations=null explicitly;

  • a positive time_step in seconds and positive n_physical_steps.

The physical time horizon is time_step × n_physical_steps. Choose it from the phenomenon of interest. Inner iterations resolve each physical step; increasing their limit is different from extending physical duration.

Control

Meaning / current conditional default

bdf_order

Physical time order, 1 or 2; resolved default 2

bdf_startup_steps

Startup steps; at least one for order 2

inner_iterations

Maximum inner iterations per physical step; resolved default 100

inner_chunk_size

Must not exceed inner iterations; resolved default at most 25

inner_residual_drop

Requested inner residual reduction; inspect achieved behaviour

inner_cfl

Positive value or -1 for the service default

transient_predictor

previous or extrapolate

transient_washout_steps

Initial transient excluded from the sampling allocation

transient_sample_steps

Must sum with washout steps to the total physical steps

iterations_per_output

Physical-step output cadence for transient work; 0 omits intermediate exports

Do not send transient controls with a steady request. The schema defaults and runtime validation define the exact allowable ranges.

Restart lineage

Use the saved restart artifact produced by the service. Confirm its compatibility with the mesh, model and intended continuation. A restart is not a generic volume image or a surface result. Never substitute a local path or stale artifact from a different mesh. After a timeout, inspect the existing job before deciding that a new restart run is needed.

Request shape

Replace the identifiers and boundary conditions with verified inputs. The time step and sampling allocation below illustrate the contract; choose them from your physical time scales and convergence evidence.

{
  "mesh_artifact_id": "REPLACE_WITH_MESH_ARTIFACT_ID",
  "restart_artifact_id": "REPLACE_WITH_RESTART_ARTIFACT_ID",
  "config": {
    "architecture": "fluxcore",
    "simulation_type": "urans",
    "turbulence_model": "k_omega_sst",
    "wall_treatment": "wall_functions",
    "airspeed": 30,
    "boundary_conditions": {
      "farfield": {"type": "farfield"},
      "body": {"type": "wall"}
    },
    "max_iterations": null,
    "time_step": 0.001,
    "n_physical_steps": 1000,
    "transient_washout_steps": 200,
    "transient_sample_steps": 800,
    "iterations_per_output": 100,
    "export_solution_fields": true
  }
}