Appearance
Inter-realm seams between misaligned blocks
| Cases | Issues | Type | Backends | Figure | Regression |
|---|---|---|---|---|---|
rmf-2realm-fv-pulse-blocks | #51, #54 | C | CPU, FNL | no | — |
Case and scope
The 1:1 mirror pulse of inter-realm 1:1 seams, on the finite-volume path, with the seam blocks of the two realms not lined up: one realm carries the same cells in smaller blocks, or in a single block. Each block face of the fine side (realm 2, realm_b) is scattered into every register face it overlaps. On a mirror seam the fluxes of the two sides are equal, so every register face must close: the printed mismatch
What the case catches:
- a fine face scattered into the wrong register face, or into too few (#51);
- a seam-peer lookup that fails when one realm is a single block: the lookup then walks up from a level-1 code to the root, and aborted with error −111 before #54.
Models and numerics
| Item | Choice | Code |
|---|---|---|
| Equations | Maxwell, SI, no source | models |
| Space | centred finite volumes of order 6 (fv_centered) | numerics |
| Time | SSP Runge–Kutta (5,4), CFL 0.5, β cadence | adam_rk_object.F90 |
| Register | fine skins scattered over the overlapping register faces (maps%seam_overlap) | accumulate_fine_overlaps, adam_flux_register_object.F90; adam_maps_object.F90 |
| Peer lookup | closest block of the peer tree (get_closest_block) | adam_tree_object.F90 |
Initial and boundary conditions
| Item | Value |
|---|---|
| Realms | mirror seam at |
| Default realm grid | 8 blocks of |
small1, small2 | realm 1 (resp. 2) on 64 blocks of iu_ref_levels = 2): the same cells |
one1, one2 | realm 1 (resp. 2) on one block of iu_ref_levels = 0, max_level kept 1) |
| Reference | the single-realm FV pulse: rmf-amr-fd-pulse on fv_centered, marker off, 8 blocks of |
| Initial condition | Gaussian pulse along sigma = 0.01, B0 = 10 |
| Boundaries | Neumann on the outer faces |
| Steps, ranks | 20, 2 |
Input files
| File | Role |
|---|---|
src/tests/prism/regression/rmf-2realm-fv-pulse-blocks/input.ini, realm_{1,2}.ini | the manifest and the realms |
src/tests/prism/regression/rmf-2realm-fv-pulse-blocks/check.sh | builds the reference and the four legs (--build, --np N; PRISM_EXE) |
src/tests/flume/verification/multirealm/multirealm_oracle.py | the union of the realms against the single realm, cell by cell |
bash
cd src/tests/prism/regression/rmf-2realm-fv-pulse-blocks && ./check.sh
PRISM_EXE=$PWD/../../../../../exe/adam_prism_fnl ./check.sh # FNLResults
Each leg asserts: the run reaches 100 % with no error/abort/NaN; the reflux line is printed and its maximum is exactly 0; the union of the realms matches the single-realm FV pulse within
| Leg | Register lines (face × step) | max mismatch | Fields against the single realm | Backends |
|---|---|---|---|---|
small1 (register side on small blocks: each register face holds | 320 | 0 | CPU, FNL | |
small2 (each realm-2 face scattered into 4 register faces) | 80 | 0 | CPU, FNL | |
one1 | 20 | 0 | CPU, FNL | |
one2 | 80 | 0 | CPU, FNL |
Related pages: inter-realm 1:1 seams on aligned blocks; static AMR for the register on intra-realm 2:1 faces.
Findings
- The register closes exactly on every face and step, whichever side carries the small blocks.
- The fields are compared within
, not bitwise: PRISM FV is not invariant under a change of block layout at round-off. The single realm alone on 4 blocks per axis instead of 2 differs by . The register check is the exact one. - A realm made of one block runs: the peer lookup reaches the root without aborting.