Repository navigation
Viscoelastic history: timestep from the wall strain rate, and a conformation health check #768
Description
On the confined cylinder (Oldroyd-B, beta 0.59) every stress history loses positive-definiteness of the conformation c = tau*/G + I in the first step at dt 0.4 (Courant 1 on the far-field mesh): the wall strain rate is about 7, so the local lambda|D| is about 3 and the explicit stretching term over one step is far outside its range. The nodal history then runs away and hangs (with or without DEVSS); the IP history holds the defect bounded and gives out at Wi 0.6. At dt 0.04 the IP history is admissible everywhere to Wi 0.6 and the drag lands at 119.3 / 114.4 (references 120.6 / 117.8) on the coarse mesh.
Two things the model should provide:
max_elastic_timestep(safety=0.3): safety / max over the domain of the strain-rate magnitude, so a run can takedt = min(courant_dt, model.max_elastic_timestep()). The Courant number on the far-field mesh is the wrong criterion for the stretching term.conformation_min_eigenvalue(): the smallest eigenvalue of tau*/G + I on the carried polymer stress (nodal or integration-point storage), with its location and the fraction of points below zero. This is the health line that separates a preconditioner failure from a lost conformation; the per-step check that found the problem lives only in a run script so far (~/+Simulations/stress_transport/cylinder_ob/ob_cylinder.py, CONF lines).
Both with hard-baseline tests on a simple shear box (one BDF-1 step from rest at dt = lambda gives min eigenvalue 1 - lambda/2 exactly).
Underworld development team with AI support from Claude Code
Activity
Hold the integration-point stress history below Courant one, and watc…
lmoresi commentedon Sep 20, 2026
Implemented on feature/stress-transport-ddt (71bb82d):
IntegrationPointSemiLagrangian.store_smoothing(alpha = c·h², c ≈ 0.03–0.07),ViscoElasticPlasticFlowModel.max_elastic_timestep(safety)andconformation_min_eigenvalue(), hard-baseline tests, and docs/developer/subsystems/stress-transport.md. Closes with the branch PR.Underworld development team with AI support from Claude Code
lmoresi commentedon Oct 8, 2026
Fixed by #833, merged to
development(f4733d4f).Both halves this issue asked for are there:
max_elastic_timestep(safety)on the viscoelastic model — the safety factor over the largest strain rate, which is the timestep-from-the-wall-strain-rate half;conformation_min_eigenvalue()— the health line that tells a lost preconditioner from a lost problem.Both trace-back histories expose
carried_tensors()so the check can read what they hold.The motivating measurement, kept here: on the confined cylinder every history lost positive-definiteness of
tau*/G + Iin the first step at Courant one on the far-field mesh, because the wall shear rate is ten times the far-field one and the explicit stretching term cannot spandt*gammadot ~ 3. A far-field Courant number is the wrong guide when the wall sets the rate — which is whatmax_elastic_timestepnow answers.Hard baselines rather than a method comparison: one BDF-1 step from rest at
dt = lambdagives eigenvalues1 -+ 1/2exactly, and a wall four times faster gives-1everywhere.Closing manually:
Closeslines here fire whendevelopmentreachesmain.lmoresi commentedon Oct 12, 2026
Reopened and labelled
fixed-in-development, under the policy now in CLAUDE.md:The fix stands and the evidence above is unchanged — this is a bookkeeping correction, not a reversal. Closing it on merge was wrong because
developmenthas not been released, so anyone running the release still has the defect; the label says where the fix is and the release sweep closes it.