QF Solver 0.2.5a0 J2 qualification report

Scope and decision rule

This report covers 025-G01 only: J2 small-strain qualification on connected TET4, TET10, HEX8 and HEX20 meshes, including constitutive paths, the algorithmic tangent, material transactions, load-step sensitivity, mesh trends, energy accounting, rollback/retry and the bounded Code_Aster correlation. It does not qualify geometric nonlinearity, buckling, arc-length, contact, friction or coupled gates.

The evidence is controlled by the source_sha and dirty fields in the machine-readable manifests. Generated evidence is output produced after checkout and is intentionally not committed into the source revision that it describes.

Controlled evidence

Evidence	Artifact	Result
Internal J2 campaign	results/vnv_0_2_5/g01_latest/summary.json and evidence_manifest.json	PASS_INTERNAL_J2, clean candidate SHA recorded in manifest
Code_Aster correlation	results/vnv_0_2_5/g01_code_aster_latest/summary.json and evidence_manifest.json	PASS_EXTERNAL_CORRELATION, clean candidate SHA recorded in manifest
Targeted regression	J2 constitutive, state, cyclic, sensitivity, multi-element and evidence tests	68 passed

The report records a controlled replay on a clean source revision. The two G01 manifests are the source of truth for the exact candidate SHA, dirty state, artifact digests and commands; the SHA is deliberately not duplicated in this prose so that generated evidence can be regenerated after a documentation-only candidate change. The recorded campaign used the pinned Code_Aster image simvia/code_aster@sha256:4629a21a109309bb97fbdc27d750445cc869e151e2e2ed6290f69539614e4435 and the command documented in the external manifest. TET10 uses the explicit comparison convention tet10_nonlinear_quadrature=code_aster_5; the legacy QF Hammer-4 rule remains the default for existing models.

Measured results

Constitutive and tangent

Elastic, near-yield, traction, compression, shear, reload, shear-cycle and non-proportional states passed. The centered finite-difference tangent sweep used steps 1e-5, 1e-6, 1e-7 and recorded a maximum relative error of 2.1204721119376345e-10, below the existing implementation limit 1e-6.

Four-family connected mesh

Element	DOF	Newton iterations	Max relative residual	Final PEEQ	Plastic dissipation
TET4	15	13	2.39e-08	5.305e-03	2.341e-04
TET10	42	20	1.93e-08	9.991e-01	5.008e+00
HEX8	36	21	3.39e-08	9.649e-02	4.824e-02
HEX20	96	22	7.80e-08	1.149e+00	6.625e+00

All four rows reached positive plastic state, positive internal/external work and passed the existing residual/work checks. The large differences in PEEQ between families are recorded observations of this controlled two-element benchmark; they are not interpreted as a defect without a formulation, quadrature or mesh-convergence diagnosis.

Energy balance

The ten-increment runs satisfied the existing test contract relative_balance_error < 1e-6 and D_p >= 0 for every family. The maximum relative balance errors were:

Element	Relative balance error	Dissipation sign
TET4	1.835e-12	non-negative
TET10	5.490e-09	non-negative
HEX8	8.645e-09	non-negative
HEX20	1.800e-09	non-negative

This is an internal work-energy consistency check, not physical validation.

Mesh and load-step studies

Mesh levels 1/2/4 and load paths 2/4/8 steps were executed for all four families. The runs passed and provide trend evidence. No universal release threshold for coarse-to-refined displacement, reaction, stress, PEEQ or energy change was frozen before this campaign. The Owner has explicitly approved these results as ACCEPTED_BOUNDED_OBSERVATION: they remain bounded observations and are not a universal convergence claim.

For the reference-to-refined load-step comparison, displacement differences were 2.46e-08 (TET4), 7.29e-06 (TET10), 5.03e-05 (HEX8) and 3.67e-05 (HEX20). These are diagnostic values, not release limits.

Rollback and retry

The adversarial run rejected one increment, preserved the committed state, cut the retry increment to 0.5, and converged on the retry path. The final displacement relative difference against the small-step reference was 1.25432459287778e-07; the final PEEQ absolute difference was 3.34527073576896e-09. The exact committed-state transaction invariant passed. A separate Owner decision is still required for any numerical band placed on the reference-difference metrics.

Code_Aster correlation

The common regular two-cell campaign produced 64/64 PASS checks across TET4, TET10, HEX8 and HEX20. Complete four-step histories were compared for loaded displacement, reaction, stress and equivalent plastic strain, with matched integration-point counts. The existing runner tolerance is 5e-3 for each comparable scalar. This is bounded numerical correlation, not physical validation and not an industrial qualification envelope.

Acceptance bands and gate decision

The following table records the observed coarse-to-refined changes and the only technically defensible proposal available without inventing a limit after seeing the results: use a documented asymptotic/convergence criterion for the selected observable and freeze its numerical band before promotion. The numerical band and the reference observable still require Owner approval.

Observable	Coarse-to-refined relative changes (TET4, TET10, HEX8, HEX20)	Proposed threshold	Margin
Displacement	27.5%, 31.7%, 91.3%, 52.5%	ACCEPTED_BOUNDED_OBSERVATION; no universal convergence claim	bounded trend recorded
Reaction	58.5%, 26.3%, 45.0%, 72.9%	ACCEPTED_BOUNDED_OBSERVATION; no universal equilibrium claim	bounded trend recorded
Stress / VM	35.1%, 25.3%, 89.4%, 39.5%	ACCEPTED_BOUNDED_OBSERVATION; field trend only	bounded trend recorded
PEEQ	35.9%, 9.5%, 91.4%, 34.8%	ACCEPTED_BOUNDED_OBSERVATION; localization trend only	bounded trend recorded
Energy	31.3%, 4.1%, 88.5%, 29.1%	ACCEPTED_BOUNDED_OBSERVATION; no universal energy asymptote	bounded trend recorded

These are not failures of the solver by themselves: they are the measured response of the regular one-direction refinement study. They do demonstrate that a universal scalar band cannot be inferred from this campaign alone.

Requirement	Frozen criterion used	Result	Margin	Owner decision
Constitutive paths	finite, physically signed/invariant path checks from existing tests	PASS	n/a	none
Algorithmic tangent	relative FD error < 1e-6	PASS (2.12e-10)	9.998e-7	none
Connected four-family solve	existing residual/work contracts, including residual < 1e-6	PASS	9.220e-7 on max residual	none
Energy	relative balance < 1e-6; D_p >= 0	PASS	9.914e-7 on max imbalance	none
State transaction	rollback leaves committed digest unchanged; commit changes trial state	PASS	exact invariant	none
Rollback reference difference	no universal release band	ACCEPTED_BOUNDED_OBSERVATION	diagnostic only	APPROVED
Load-step sensitivity	no universal release band	ACCEPTED_BOUNDED_OBSERVATION	diagnostic trend	APPROVED
Mesh/PEEQ convergence	no universal release band	ACCEPTED_BOUNDED_OBSERVATION	diagnostic trend	APPROVED
Code_Aster four-family correlation	existing runner tolerance 5e-3 on 64 comparable checks	PASS (64/64)	3.387e-3 on max error	none

G01 status

025-G01 = PASS.

Owner decision study

The Owner-approved classifications below separate an already justified invariant from bounded observations for which no universal release band was frozen before the campaign. The exact source revision is recorded by both final evidence manifests.

Metric	Classification	Basis for decision
Mesh / PEEQ	ACCEPTED_BOUNDED_OBSERVATION	Mesh levels 1/2/4 and four-family trends were executed. The observed coarse-to-refined changes are family-dependent and are accepted as bounded evidence without a universal release band.
Load-step sensitivity	ACCEPTED_BOUNDED_OBSERVATION	The step sweep provides diagnostic trend evidence. The existing tighter 1e-8/2e-2 sensitivity contract remains specific to the established TET4 campaign and is not silently extended to the four-family G01 claim.
Rollback	ACCEPTED_BOUNDED_OBSERVATION	Exact committed-state preservation, cutback and retry are THRESHOLD_JUSTIFIED; the final difference to the small-step reference (1.25432459287778e-07 displacement relative and 3.34527073576896e-09 PEEQ absolute) remains diagnostic.

This treatment does not lower any requirement. It records the available evidence honestly and keeps the broader claim bounded. The final manifests provide the required controlled artifacts for this decision.

G01 OWNER DECISION = APPROVED

CONTRACT LOWERED = NO REMAINING BLOCKER = NONE FOR G01

This is a governance/provenance blocker, not evidence of a failed numerical test. No other functional gate is changed by this report.

Owner decision record

1. The inherited 1e-6 tangent and energy criteria and the existing 5e-3 Code_Aster correlation criterion are accepted for G01.
2. Mesh/load-step trends are accepted as bounded observations with the corresponding claim explicitly limited.
3. Rollback/reference differences remain diagnostic; the exact transaction invariant remains the frozen acceptance criterion.

No unanswered Owner decision remains for G01. The gate closure does not promote universal mesh convergence, universal load-step convergence or physical validation claims.