Convergence and result qualification

Separate job completion, numerical convergence, engineering adequacy and published outputs. A run can reach its iteration limit with useful diagnostic histories but without meeting the required convergence standard.

Read the five primitives and residuals to interpret the core continuity, Ux, Uy, Uz and energy curves and the additional turbulence/thermal channels. Residuals measure equation imbalance, not the primitive field values themselves.

Define the standard

Use convergence_criteria to state a run’s assessment criteria. The current standard examines the final 20% of iterations, tests force-coefficient spread against 1% of its mean with a 0.001 allowance near zero, and checks for rising residuals. Read the returned criteria and reasons; required histories may be absent or insufficient.

Optional controls include a fixed trailing window, selected forces/residuals, allowed residual rise, required residual reduction and a final residual threshold. These assess evidence; they do not make a physically unsuitable setup valid.

Assess saved evidence

assess_convergence evaluates an existing run against a requested standard using saved histories. It starts no new solver run and does not overwrite the saved verdict. Report the resulting status, reasons, window and thresholds.

Review the residual chart and engineering-output histories together. Also inspect mass/energy balance, field plausibility, mesh sensitivity and, for transient work, time-step and sampling sensitivity. A steady coefficient can hide a localized problem; a small residual is not a universal error bound.

Retain the normalized request, mesh identity, wall-resolution evidence and analysis definitions. Present predictions or reduced models with their actual evidence label instead of describing them as a new CFD solution.

Exact criteria: ConvergenceCriteria. Results workflow: API outputs.