Phase D: emit consistent tangent dσ/dε as a rank-4 tensor property

[petlenz]

What

NumSimMaterialTarget now emits the consistent tangent dσ/dε as a rank-4 tmech::tensor property, computed via cas::diff(stress, strain).

Why this is the exact consistent tangent in scope (not just a partial): the rate-leaf guard forbids the rate from depending on strain, so the integrated state is independent of the current strain (dα/dε ≡ 0). Therefore dσ/dε = ∂σ/∂ε + ∂σ/∂α·(dα/dε) = ∂σ/∂ε — a single cas::diff, no J⁻¹ solve, no rank-4 inverse. (This mirrors the static_assert in AlgorithmicTangentPass that pins the same fact for the standalone path.)

Reuses the proven tangent machinery: cas::diff(σ,ε) → rank-4, rendered by the same tensor() emit pipeline; folded into the tensor-output path so the ctor/callback/member emission handles it uniformly.

Minor symmetry

roles::Strain carries a symmetric tensor space, so cas::diff returns the minor-symmetric rank-4 identity P_sym = ½(I⊗ᵘI + I⊗ˡI) — the emitted RHS contains both otimesu and otimesl, as a stress-strain tangent must.

Verified end-to-end (gcc)

Generated Viscoelastic (σ=α·ε) emits dstress_dstrain; driven through the real graph, dσ/dε = α·P_sym checks out component-wise: C(0,0,0,0)=α, C(0,1,0,1)=α/2 (minor-symmetric), C(0,0,1,1)=0. (Tensor e2e is GCC-gated per numsim-core#16, same as the stress path; the emit shape is verified compiler-independently by the unit test.)

Tests

• Unit: EmitsConsistentTangent (rank-4 property + callback + minor-symmetric otimesu/otimesl), RejectsTangentOfMissingStress.
• e2e: the tensor-stress test now also reads + verifies the rank-4 tangent property.

Scope (honest)

This covers the strain-uncoupled consistent tangent (dα/dε=0). The strain-coupled case (rate depending on ε ⇒ the residual contract + solver-assembled blocks ∂R/∂ε, ∂σ/∂x and J⁻¹) remains future work, gated on the residual contract + strain-coupling support.

274 tests pass locally; warning-gated.

🤖 Generated with Claude Code