Moving film.
Conservation passes.
Spatial accuracy needs work.
Actual energy-driven particles coupled to conservative relative drainage. This report separates successful bookkeeping from unresolved physical-field accuracy.
What changed
The old normal-face viscosity misses Cartesian shear. The new opt-in metric integrates full symmetric velocity gradients on geometric face pyramids, adds shared-face consistency correction to control alternating velocities, and applies no-slip samples on actual outer faces. Physical viscosity is unchanged; no fitted penalty coefficient is introduced. Legacy output agrees exactly with the old control.
Spatial checks
Relative L2 comparisons use independent clipped Voronoi cells and exact polygon intersections. Background thickness error stays small; thickness change, velocity and displacement expose the unresolved differences. All pairs below share their time step and20us horizon.
| Candidate / reference | h | Change | Velocity | Displacement | Gate |
|---|---|---|---|---|---|
| closed_l100 / closed_u100 | 0.10% | 40.57% | 35.02% | 34.80% | FAIL |
| open_l50 / open_u50 | 0.05% | 19.32% | 19.14% | 18.47% | FAIL |
| closed_l50_small / closed_u50_time | 0.05% | 18.05% | 18.01% | 17.33% | FAIL |
Recorded particles / uniform50um
White-background view of actual saved particle centers. Colors show thickness change (blue thinning, red thickening). No interpolated states, prescribed motion, or displacement amplification. The actual displacement is below1um.
Evidence and limits
Domain1.2x0.8mm, initial thickness6um; SDS SI material constants; insoluble monolayer, no gravity or DLVO. The optional500um-curvature boundary is a fixed-potential bath with a resolution-dependent pinned proxy interface. It is not a resolved dynamic meniscus. No remeshing, adsorption, long-time convergence or experimental validation is claimed.
The kinetic-removal ledger includes donor mixing and removal at pinned supports, and is checked separately for nonincrease. No complete temporal comparison passed: the finer-step reference failed. Each job had a600s limit, with the main batch capped at55minutes; actual batches took minutes.
Next decisive experiments
- Record drainage gradient/step histories at the two failed steps to distinguish floating-point stagnation from nonlinear failure.
- Recover a reproducible matched time-refinement pair without relaxing conservation or field gates.
- Test boundary and coarse-region resolution separately before any long-time or dynamic-meniscus run.
Original v1 failures and repairs remain preserved. Constitutive checks are numerical evidence, not a novelty or experimental validation claim.