Skip to content

026-G06 Deep J2 Evidence

Status

026-G06 is PASS_WITH_LIMITATIONS by Owner decision. This pack is the controlled evidence package for the bounded closeout.

  • Start SHA: b952520f5db8095e1fa214e451999aaedc672341
  • Execution source SHA: 8bd0f2d8fdce7bf27ffc4c28e6aa26e69288fa63
  • Source dirty state: false
  • Proposed decision: PASS_WITH_LIMITATIONS
  • Owner decision: PASS_WITH_LIMITATIONS
  • Closeout start SHA: 2662c984b64816b3deaf18b0a6bb270141551b40
  • Finite-kinematic J2: RESEARCH_NOT_QUALIFIED

Internal evidence

Evidence Result
Dedicated TET10 cyclic J2 PASS_INTERNAL, 341 nodes, 140 elements, 4 integration points/element
Inverted TET10 rejection EXPECTED_FAILURE / INVALID_ELEMENT, rejected before solve
J2 mesh series PASS_INTERNAL_MESH_REFINEMENT, levels 1/2/4/8, TET4/TET10/HEX8/HEX20
Load-increment sensitivity PASS_INTERNAL, subdivisions 4/8/16, maximum state sensitivity 3.444e-09
Multi-element invariant matrix All four families PASS
Energy and plastic dissipation All four families PASS_INTERNAL_ENERGY
Cyclic paths All four families PASS_INTERNAL_CYCLIC
Adversarial rollback PASS_INTERNAL_ROLLBACK for TET4/TET10/HEX8/HEX20, deterministic rejected increment and clean retry
Consistent tangent checks Targeted tests PASS

The mesh study is deliberately reported as bounded evidence. Its regular unit-block topology and the non-monotone field trends do not justify a universal mesh-convergence claim.

Cross-family proof coverage is explicit: homogeneous constitutive response, yield threshold, shared tangent FD, equilibrium/residuals and rollback are covered in bounded form. Tangent symmetry is not separately assessed, and increment-independence beyond the existing TET4 study is not claimed.

Code_Aster correlation

The pinned Code_Aster 18.1.0 image was executed after the evidence runner was committed. The regular two-cell shared benchmark produced 64/64 PASS checks:

Family Maximum relative error Result
TET4 1.613e-03 PASS
TET10 6.230e-04 PASS
HEX8 2.734e-07 PASS
HEX20 6.339e-04 PASS

The TET10 comparison uses code_aster_5 as an explicit external convention; QF's historical default remains hammer4. This is bounded numerical correlation, not physical validation.

Targeted verification

Command:

python -m pytest tests/unit/test_robustness_tangent_fd.py tests/unit/test_nonlinear_multielement.py tests/unit/test_nonlinear_cyclic.py tests/unit/test_j2_multielement_external.py tests/verification/test_tet10_j2_structural_vnv.py -q

Result: 38 passed, 0 failed in approximately 160.77 seconds.

Registry, anti-forgetting, Ruff, compileall and git diff --check also passed. Full regression was not rerun because no solver or functional FEM code changed.

Limitations and remaining Owner decision

  • Small-strain J2 is qualified within the tested TET4/TET10/HEX8/HEX20 scope.
  • Algorithmic tangent symmetry has not been independently qualified.
  • The mesh series is bounded and does not qualify arbitrary distortion, localization or industrial geometries.
  • Independent increment-refinement evidence is currently limited to TET4 and is not a universal path-independence statement for all families.
  • Rollback evidence covers deterministic rejection before the first accepted increment for all four families; broader material-update and sparse-backend failure matrices remain separate work.
  • Finite-kinematic J2 and coupled nonlinear workflows remain research/experimental.

The generated aggregate artifacts are under the ignored results/g06_depth directory. Their digests are recorded in qualification/0_2_6/g06_depth_evidence.json, together with the exact source and external-image provenance.