Skip to content

026-G08 HEX20 CalculiX Rescue

Decision boundary

This document records a targeted correction to the external verification deck. It does not change QF Solver numerical behavior and does not close or promote 026-G08.

Root cause

The C3D20 element continuation line generated by calculix_buckling_025.py started with an empty comma field. CalculiX accepted the process invocation but stopped before producing a buckling factor. The same node mapping and model passed the existing HEX20 static correlation, which isolated the issue to the buckling deck writer.

The fix writes the remaining node numbers directly on the continuation line and is covered by tests/unit/test_calculix_buckling_025.py. No formulation, geometric stiffness, eigenproblem or QF Solver buckling implementation was changed.

Controlled evidence

The machine-readable evidence is qualification/0_2_6/g08_hex20_calculix_rescue_evidence.json.

Mesh QF factor CalculiX factor Relative difference Replays
1 HEX20 cell 12.50453659 12.71666 1.696% 2
2 HEX20 cells 10.98539335 11.21546 2.094% 2

The existing 10% correlation screen was declared before execution and was reused unchanged. Both levels pass that bounded numerical comparison, and both replays reproduce the CalculiX factor within 1e-12 absolute difference.

Mode comparison

CalculiX first-mode vectors were read from the final FRD displacement block and compared with QF's first mode. The arbitrary eigenvector sign was aligned before reporting the cosine and MAC:

Mesh Raw cosine Sign-aligned cosine MAC
1 HEX20 cell -0.935479009 0.935479009 0.875120975
2 HEX20 cells -0.890656829 0.890656829 0.793269586

The G08 contract does not define an external MAC acceptance threshold for this case, so these values are diagnostic evidence, not an invented PASS criterion.

Scope and limitations

  • C3D20 is compared with the same QF HEX20 mesh factory, connectivity mapping, material and nodal dead-load pattern.
  • The study covers the first linearized tangent-instability factor and first mode only.
  • It is numerical external correlation, not physical validation.
  • The historical G08 Owner decision remains PASS_WITH_LIMITATIONS.
  • HEX20 mesh sensitivity remains bounded evidence; no universal high-order convergence claim is made.
  • No independent high-order analytical oracle or comparable Code_Aster route is added.
  • No post-buckling or multi-mode qualification is claimed.