Appearance
Seam studies (M6)
These pages are the investigations behind milestone M6 (issue #68): the prototypes, operator variants and ablations that decided how the static 2:1 seams were changed. The verification cases they build on, and that freeze the outcome, are in seam accuracy and divergence.
The phases of #68, in the order the evidence led:
- B (prototypes, NumPy, no ADAM run): B1 confirms in 2-D that the mean restriction and the reflux each cap the order at 2; B3 picks the shift-inward footprint for the 6-point restriction; B2 gives the go for a sensor-gated reflux (D-M6-4).
- C (restriction):
[amr] seam_restriction = lagrange6,weno6, measured on AO and AS. - D (coarse-to-fine fill): D1 adds the quintic fill, which reaches the scheme's order on smooth flows; D2 adds two limited fills and shows that the shock-crossing defect is not the fill's.
- E (reflux): E0 shows the shock-crossing defect is a capturing error, no seam operator's; E2 measures
[numerics] reflux_policy = troubled(D-M6-2: the default staysalways). - G (divergence): G2 keeps the GLM damping length frozen (D-M5-9); G3a finds that the reflux causes #78; G3b fixes it by not refluxing
under EGLM.
The defaults are unchanged by every study except G3b: mean restriction, tricubic fill, reflux_policy = always. No study has a figure yet: they are listed in #85. The prototypes run from src/tests/flume/verification/amr-proto/check.sh (about 5 min; --legs b1,b2,b3 selects the prototypes, --long the finer ladders).
| Case | ID | What it verifies |
|---|---|---|
| Quintic coarse-to-fine fill | D1 | the 6-point fill brings smooth flows to sixth order; unlimited, it worsens a shock entering a finer region |
| Limited coarse-to-fine fills | D2 | weno6 and quintic-mp keep the order; no fill controls the shock-crossing defect |
| Shock-crossing defect is a capturing error | E0 | neither the restriction, the fill/restriction consistency nor the reflux causes the defect |
| Reflux policy: troubled against always | E2 (E1) | the sensor-gated reflux recovers the high order with shocks unchanged, at a conservation cost |
| #78 is the reflux, not the restricted ghosts | G3a | EGLM's |
| GLM damping length on a seam | G2 | every damping length keeps the seam divergence bounded; strong damping doubles its floor |
| EGLM reflux leaves B alone | G3b | the fix of #78: no reflux of |
| 2:1 caps in a 2-D prototype | B1 | the mean restriction and the reflux each cap the order at 2 in 2-D; the corners are not the cause |
| Sensor-gated reflux prototype | B2 | a troubled reflux recovers order 5 and keeps the shocks: go for Phase E |
| Footprint of a high-order restriction | B3 | shift-inward and two-pass are equivalent with the quintic prolongation; shift-inward chosen |