Moving-facet implicit contact candidate — 2026-10-16

Status: stacked experimental serial slice after the fixed-facet J2 candidate. It is not a general die-forming or periodic-contact release claim.

Design and ownership

• Model retains the existing RigidContactPair and RigidBody with a PrescribedRigidMotionSchedule. The ordinary static Step admits only its normalized 0-to-1 form without an independent physical-time amplitude; the Step amplitude remains the sole load-path owner.
• Search retains one reference-geometry triangle BVH. Each Newton trial maps current slave points into the rigid reference frame, queries the BVH, then maps closest points and normals back to the current pose. It does not reuse a previous gap, normal, or force.
• Operator evaluates the penalty potential and the fixed-current-pose displacement residual/tangent on a smooth planar facet branch. The generalized tool dual contains the contact resultant and moment about the current rigid reference point, conjugate to prescribed translation and axis-angle coordinate along the proportional motion path.
• Procedure records trapezoidal force--coordinate work only on accepted increments. State snapshots and restores the tool factor with displacement, material trial/commit, and the accepted energy ledger. A failed increment cannot advance pose or work. Checkpoint identity includes the motion schedule through the contact pair.
• Result reports force, moment, potential, tool work, and their provenance; energy residual is a discrete accounting check, not a general first-law proof for arbitrary contact and plasticity.

Local evidence and admission boundary

The moving coplanar triangulated tool agrees with the existing analytical moving-plane J2 route for translation and simultaneous z rotation through 0.2 rad on the bounded planar fixture: displacement, committed material state, generalized force, penalty potential, and accepted tool work agree within the regression tolerances. The tool-work dual also agrees with a fixed-state finite difference of potential with respect to the prescribed motion factor at 0.01 and 0.2 rad. A serial checkpoint at both rotation magnitudes resumes to the uninterrupted state, while an intentionally failed energy update restores the prior accepted pose, displacement, and plastic history. The public example completes with --tool-travel -0.006 --tool-rotation 0.01 and reports nonzero tool work. An active free tool edge fails closed during the global solve and preserves the prior accepted displacement, material state and pose. These are local plane-limit checks, not independent curved-tool validation or physical-time simulation.

A bounded 0.2-rad moving-plane fixture was also scanned without changing its penalty or motion path. With two cells per axis, 3/6/12 accepted increments gave tool work 1.486774 / 1.508690 / 1.514850 and contact potential 0.636862 / 0.636358 / 0.636349. With six increments, 1/2/3/4 cells per axis gave tool work 1.766163 / 1.508690 / 1.409196 / 1.363561 and contact potential 0.923614 / 0.636358 / 0.543069 / 0.502516. Differences between successive levels shrink for these integral observables, but the spatial series is not yet grid-independent. The discrete energy residual is not monotone under the 3/6/12-step scan, so this is not an energy-convergence claim. No pointwise maximum is used as a universal accuracy certificate. At the fixed two-cell/six-increment resolution, penalties 250/500/1000/2000 gave tool work 1.138341 / 1.508690 / 1.841833 / 2.108082 and contact potential 0.688932 / 0.636358 / 0.517584 / 0.375544. This sensitivity is expected for a penalty approximation and must not be mistaken for an algorithmic performance comparison or a penalty-independent solution.

One frictionless prescribed rigid triangle tool, serial 3D blocked displacement, reference-area quadrature, and active projection on planar interior/coplanar seams only. No free/noncoplanar edge transition, friction, multiple pairs, contact/strong-BC overlap, periodic MPC, or distributed implicit assembly. A non-normalized independent motion schedule is rejected rather than silently remapped to the static Step; both capability discovery and model.step(...) report the same stable diagnostic. Full installed-wheel and remote CI gates must still pass for this candidate.

At source commit 515d3468, the macOS arm64 CPython 3.11 wheel (sha256:d8b6b06e7555f06153eebff11df7fa85bf3f0d9abbc713c0db82bc94c75b4f69) and sdist (sha256:92866bdfc5aba1a9e61c8b1b75b5ec86f43fcfd846b44de3c316fe7d93c526b5) were built from clean source. The isolated installed-wheel release_gate.py --dist ... --smoke --require-platform macos completed successfully. This is local platform evidence, not a release tag or proof on Windows/Linux. The remote PR gate remains a separate promotion decision.

Next review gates

1. Compare a non-planar smooth tool against an independent structural reference, including force and moment, mesh/penalty/increment sensitivity, and facet-switch stability. A planar-limit equivalence is insufficient.
2. Implement reviewed edge/vertex or smooth-surface transition derivatives before claiming general finite sliding on a tessellated die.
3. Reduce both material-plus-contact residual and tangent through periodic MPC, with identical line-search and reaction semantics. The present affine J2 route assembles a material-only full-space form before T^T R and T^T K T, whereas the strong-boundary J2 contact route adds the external contact operator before Newton. A reviewed additive full-space operator seam, trial-pose refresh and joint acceptance transaction are required; merely admitting a contact parameter to the affine Step would silently omit contact from line search and reactions. The provider capability probe and Step lowerer now both reject any registered boundary model on this affine J2 path with AFM-J2-CONTACT-MPC; the material-only periodic regression still passes.
4. Establish distributed owner/ghost assembly and portable restart evidence.

The accepted-path trapezoidal work choice is consistent with the contact energy accounting discussed in the Abaqus contact-energy example. That reference also warns that contact energy balance depends on increment size and contact transitions; its published numbers are not AgentFEM evidence.