Skip to content

WP16 - True 1M DOF qualification

Decision

WP16 is PASS on the official retry executed with the PETSc route selected by WP17-R. The frozen WP14 acceptance criteria were used without change. The historical matrix-free failure remains preserved as historical evidence and is not overwritten.

The official model contains 343,000 nodes, 1,971,054 structured six-tet TET4 elements and 1,029,000 true displacement DOF. It uses the WP14 SI material, fixed x=0 translations, a uniform 1,000,000 N nodal load on x=1, 4,900 load nodes and 14,700 fixed DOF. The retry source SHA is b8db2211536eaaf1c026ddfb7a5843b61b2e3733.

Frozen execution route

The route is PETSc 3.25.1 with MPICH 5.0.1, two MPI ranks, CG, GAMG, AIJ storage, contiguous partitioning and one thread. The PETSc environment is the pinned image qf-solver-large@sha256:d6a1718001fc36772906d1a9505637bbd0a4b7e1d8ccc9afdbcb6f67b7ff6d0e. All relevant PETSc options are explicit and PETSC_OPTIONS is unset. The internal solver target is 1e-10, pre-declared by WP17-R; WP14 acceptance remains rtol=1e-8, atol=0 and max_iterations=10000.

Official replays

Run Setup [s] Solve [s] Reactions [s] Energy/post [s] Total [s] Iterations Peak RSS [bytes] Residual Equilibrium Energy
run1 113.2639 66.3369 3.3231 2.9063 184.0757 431 3,519,639,552 9.704e-11 5.339e-10 8.484e-14
run2 113.4333 67.4453 3.1541 2.7512 185.1151 431 3,520,733,184 9.704e-11 5.339e-10 8.484e-14

Both runs completed the complete model-to-solve-to-post path without timeout, resource-limited status, NaN/Inf, or implicit fallback. The finite-output, SPD/CG, residual, equilibrium and energy checks all pass the WP14 limits.

Raw controlled records are:

  • qualification/0_2_7/wp16_runtime/wp16_retry_run1_raw.json
  • qualification/0_2_7/wp16_runtime/wp16_retry_run2_raw.json

The official index and replay comparison are in qualification/0_2_7/wp16_runtime/wp16_retry_summary.json.

Subscale equivalence

The same PETSc configuration and internal solver target were compared with the existing matrix-free route on the 107,811-DOF subscale. The comparison passes with displacement error 8.341e-15, equilibrium difference 9.279e-10, energy difference 1.598e-12 and matrix-free residual 9.290e-11, all below the unchanged 1e-8 comparison limit.

The raw subscale record is qualification/0_2_7/wp16_runtime/wp16_retry_subscale_raw.json.

Replay policy

The two official replays use the same source SHA, input digest and configuration digest. Their recorded numerical observables and iteration counts are identical; the replay verdict is PASS under the WP14 policy.

Input digest: b65b1cd72a067551490ed5364beb1fcc2d7e55d07a9075f6f7b8899f535d7f92

Configuration digest: e44ef191461ec5e4c6c0c0de31bcda3f674c3e45949513d38a4ce8a067bf9fe6f

Historical negative attempt

The earlier matrix-free/nodal-block-Jacobi attempt remains available at qualification/0_2_7/wp16_runtime/wp16_run1.json. It completed 1,029,000 DOF but recorded equilibrium 3.81975e-8 against the frozen 1e-8 limit. That result is retained as historical evidence and is not used as the active WP16 verdict.

Scope and limitations

  • This is a bounded qualification of the declared TET4 linear-static model, PETSc CG/GAMG route and pinned two-rank environment.
  • The PETSc AIJ route reaches approximately 3.52 GiB peak RSS at 1M DOF.
  • PETSc is unavailable in the normal host environment; the pinned container is required for this evidence.
  • This result does not qualify 3M DOF or any other element family/backend.
  • WP14 acceptance thresholds and existing FEM formulations were not changed.

The machine-readable state is qualification/0_2_7/wp16_state.json; the governing contract is qualification/0_2_7/wp14_execution_contract.json.