QF Solver 0.2.5a0 - Lot 5A contact implementation report¶

Scope: implementation blockers for 025-G05 only.

Implementation base commit: d6ede9d8c3cf01ea6d381ff84441ad2067482095

The implementation snapshot above is historical. The controlled G05 replay was generated later on qualified source SHA a3ab8de707ffc88fc5e39e4f999eb872c9223b73 with dirty=false; its evidence manifest is results/vnv_0_2_5/g05_latest/evidence_manifest.json.

Gate decision: 025-G05 = PASS for the bounded contract below.

This report retains the implementation history and records the subsequent controlled closure. It does not qualify general contact.

1. Implementation result¶

The existing contact contribution remains on the common sparse penalty path:

R_total = R_internal - R_external + R_contact
K_total = K_material + K_geometric + K_contact

No second Newton driver was introduced. The implementation adds a bounded slave-node patch representation while retaining the legacy slave_node input. Each node in slave_nodes is expanded into a stateless contribution against the selected triangulated master facet. This is an explicit node-patch approximation, not a mortar or segment-to-segment formulation.

The updated-search path recomputes the selected facet, projection, normal and tangential basis from the current trial configuration. The sparse contact tangent is assembled together with the other global contributions.

2. Evidence produced¶

Requirement	Internal observation	Result
A - multi-facet plane	Two slave nodes, two master facets, both active, sparse tangent non-empty	`PASS_INTERNAL_RESEARCH`
B - updated normal	Master facet node displaced in `UZ`; normal becomes `[0, -0.4472136, 0.8944272]` and remains unit length	`PASS_INTERNAL_RESEARCH`
C - finite sliding	One slave node selects facets `[0, 1, 2]` across a connected three-facet strip; two switches observed	`PASS_INTERNAL_RESEARCH`
D - open/close/recontact	Existing common-driver contact path retains the bounded open/close/recontact tests	`PASS_INTERNAL_RESEARCH`
E - transition rollback	Committed facet `1`, failed trial facet `2`, rollback facet `1`, retry facet `2`; deterministic replay	`PASS_INTERNAL_RESEARCH`
F - common Newton	Existing contact composition uses the common nonlinear residual/tangent/Newton path	`PASS_INTERNAL_RESEARCH`
G - penalty sensitivity	Existing bounded penalty sweep remains available	`PASS_INTERNAL_RESEARCH`
H - local tangent FD	Existing fixed-active contact tangent error is approximately `6.0e-9`	`PASS_INTERNAL_RESEARCH`

The new direct benchmark observations are:

surface patch:
  slave_surface_mode = node_patch_to_faceted_surface
  slave_node_count = 2
  selected_face_indices = [0, 1]
  active_contacts = [0, 1]
  gaps = [-0.1, -0.1]
  tangent_nnz = 21

three-facet sliding:
  selected_face_indices = [0, 1, 2]
  face_switch_count = 2
  facet_count = 3
  gap = -0.1 on each position

facet-transition rollback:
  committed_face = 1
  failed_trial_face = 2
  rollback_face = 1
  retry_face = 2
  clean_retry = true

3. Tests executed¶

The targeted regression command was executed with the authoritative source tree on PYTHONPATH:

python -m pytest tests/unit/test_contact_finite_sliding.py \
  tests/unit/test_nonlinear_contact_composition.py \
  tests/unit/test_frictionless_contact.py \
  tests/unit/test_large_solver_and_serialization_edges.py \
  tests/unit/test_schema_entity_validation_paths.py -q

Result: 55 passed in 2.86 s.

The run covers backward-compatible serialization, schema validation, sparse assembly, updated search, common Newton diagnostics, patch expansion, three facet traversal, updated normals and facet-transition replay. No full regression or coverage campaign was launched for this implementation slice.

4. Files changed¶

src/solveur/contact/entities.py
src/solveur/contact/support.py
src/solveur/contact/solver.py
src/solveur/core/model.py
src/solveur/io/contact_schema.py
src/solveur/io/model_writer.py
src/solveur/mesh/validation.py
src/solveur/verification/robustness_contact.py
tests/unit/test_contact_finite_sliding.py
docs/verification/0_2_5/0_2_5_vnv_matrix.md
docs/verification/0_2_5/0_2_5_known_limitations.md

The document registry also records this report and the pre-existing Lot 4 implementation report so controlled-document tests can enumerate every Markdown source consistently.

5. Explicit limitations¶

The surface representation is a bounded collection of slave nodes against a triangulated master surface. It is not a continuum surface-to-surface, mortar or segment-to-segment formulation.
The finite-sliding projection is a bounded closest-point-on-triangle approximation when the orthogonal projection leaves a facet. It can be nonsmooth at active-set and facet transitions; a classical smooth FD tangent claim does not apply exactly at those kinks.
The master search is local to the explicitly supplied faceted surface. A generalized broad-phase search, self-contact and arbitrary deformable surface pairing are not included.
Friction is unchanged and remains outside this Lot 5A implementation.
The current external correlation is bounded to compatible Code_Aster normal contact histories; CalculiX is supporting pre-contact evidence only.
The contact claim is qualified only for the bounded node/patch-to-faceted- surface contract. General surface-to-surface contact, friction, impact, self-contact and unrestricted large sliding remain outside scope.

6. Status¶

IMPLEMENTATION STATUS = COMPLETE
SURFACE_TO_SURFACE = PARTIAL
FINITE_SLIDING = PASS
UPDATED_NORMALS = PASS
FACET_TRANSITIONS = PASS
ROLLBACK = PASS
FD_TANGENT = PASS
TESTS = 55 passed in 2.86 s
READY_FOR_G05_VNV = COMPLETE
G05 = PASS

The implementation blockers addressed here are complete within the bounded node-patch/faceted-surface scope. The controlled closure used the current internal contact suite, bounded penalty characterization and the archived Code_Aster additional-model campaigns. It does not promote the implementation to a general surface-contact claim.

Owner closure record¶

Field	Value
Gate	`025-G05`
Owner decision	`APPROVED`
Qualified source SHA	`a3ab8de707ffc88fc5e39e4f999eb872c9223b73`
Source worktree	`CLEAN`
Contract lowered	`NO`
Internal targeted tests	`82 passed / 2 skipped`
Code_Aster	`PASS_EXTERNAL_CORRELATION` on 768 and 9,984 TET4 confirmations
CalculiX	supporting SHOULD evidence; not release-blocking

The resulting qualified scope is 025-REQ-018 through 025-REQ-021: common sparse frictionless penalty contact, bounded open/close/recontact, opt-in updated finite-sliding projection over an explicitly supplied triangulated master surface, penetration/enforcement sensitivity, and exact transaction replay. This closure does not close G06 or promote G07.