[petlenz]

Closes #96, closes #97, closes #98, closes #99, closes #103. Part of the uvwmaterial port (#94); stacked on #127 (which sits on #145).

Each Eshelby tensor is a material, and a pure producer. As in uvwmaterial (eshelby_tensor_solid_sphere, _cylinder_x1, _cylindrical_ellipsoid, _general_geometry) every shape is its own node of the graph:

• eshelby_sphere, eshelby_cylinder (direction), eshelby_spheroid (aspect_ratio = length over diameter, direction), eshelby_ellipsoid (semi_axes along the global axes).
• Every node consumes Poisson's ratio (nu_source, nu_property, default value) and publishes the tensor as the output tensor. Nothing else: no interface, no tensor(nu) for callers off the graph. A composite consumes tensor like any other property; the engine re-evaluates the node when ν changes, so a curing matrix is followed and a constant one costs one evaluation. The common plumbing lives once in the CRTP detail::eshelby_tensor_base; a shape is ~20 lines (the callback is bound from the most-derived constructor, UBSan-clean).
• Registered for Dim = 3 only (if constexpr), so a 2D registration still compiles.

Producers of ν: poisson_ratio_from_moduli takes K and G as parameters and computes ν once at construction (output value, no callback, like constant_scalar) — the constant case. isotropic_moduli consumes a tangent (tangent_source, tangent_property) and produces K, G, poisson_ratio per update, rejecting a tangent off its isotropic projection by more than tolerance — the case of a ν that follows a tangent.

core/eshelby.h keeps the closed forms (Mura ch. 11) as free functions behind the nodes: sphere(nu) from the projectors, elliptic_cylinder(nu, a, b, axis), cylinder(nu, axis | direction), ellipsoid(nu, a1, a2, a3) with I_i = (4π/3) a₁a₂a₃ R_D from Boost.Math (ellint_rd; an installed Boost ≥ 1.74 or the standalone boostorg/math repository fetched at boost-1.83.0 — a header-only dependency; like Eigen it is exported only when installed). The tensors are assembled on the eigenprojectors eᵢ⊗eᵢ (detail::orthotropic(A, s) with otimes/otimesu/otimesl) instead of per component, and equal axes are detected at √ε so the difference rule (I_j − I_i)/(a_i² − a_j²) never runs into its cancellation window (continuity into the spheroid tested at δ = 1e-9, 1e-11, 1e-13).

Tests: test_eshelby (8, the closed forms); test_eshelby_materials (8, the nodes: each shape against its closed form at the consumed ν, a changing ν re-evaluates the node, ν from an isotropic tangent through isotropic_moduli, invalid geometry rejected at construction, all four from JSON); test_scalar_producers (4: ν fixed at construction, non-positive moduli rejected, the K-G-tangent-moduli round trip, JSON); test_isotropize (5).

796/796 tests pass (GCC 14, reports on).

JSON input

New material types: poisson_ratio_from_moduli, isotropic_moduli, eshelby_sphere, eshelby_cylinder, eshelby_spheroid, eshelby_ellipsoid. A json_model document that loads (the FromJson test of the PR):

{"materials": [
  {"type": "poisson_ratio_from_moduli", "name": "nu", "K": 100, "G": 40},
  {"type": "eshelby_sphere", "name": "sphere", "nu_source": "nu"},
  {"type": "eshelby_cylinder", "name": "cylinder", "nu_source": "nu", "direction": [0, 0, 1]},
  {"type": "eshelby_spheroid", "name": "spheroid", "nu_source": "nu", "aspect_ratio": 10, "direction": [1, 0, 0]},
  {"type": "eshelby_ellipsoid", "name": "ellipsoid", "nu_source": "nu", "semi_axes": [1, 2, 3]}
]}

When the matrix tangent is itself a product (cure, temperature, a composite), isotropic_moduli turns it into K, G and ν; Eshelby nodes then consume "nu_source": "moduli", "nu_property": "poisson_ratio":

{"materials": [
  {"type": "constant_scalar", "name": "K", "value": 3333.33},
  {"type": "constant_scalar", "name": "G", "value": 1153.85},
  {"type": "isotropic_tangent", "name": "stiffness", "K_source": "K", "G_source": "G"},
  {"type": "isotropic_moduli", "name": "moduli", "tangent_source": "stiffness"},
  {"type": "eshelby_sphere", "name": "sphere", "nu_source": "moduli", "nu_property": "poisson_ratio"}
]}

Outputs: value (ν), K/G/poisson_ratio, tensor (rank 4).

[petlenz]

Review findings (design, tmech expressions instead of component loops, modern C++, comments, tests, correctness). Severity: must = wrong result, hang, build break or unreachable feature; should = design, duplication, accuracy; nit = style.

Eshelby — needs changes

• [should] eshelby.h:118 equal threshold 1e-12 leaves a cancellation window: ellipsoid(ν,1,1+δ,2) vs spheroid 3.9e-5 at δ=1e-12, 8.6e-7 at 1e-10. Use sqrt(eps); test δ=1e-11.
• [should] :29-45, :66-78 per-index assembly, shear block repeated four times → S = Σ A_ij otimes(E_i,E_j) + Σ_{i≠j} s_ij (otimesu(E_i,E_j)+otimesl(E_i,E_j)) with E_i = otimes(e_i,e_i); one detail::orthotropic(A, s) serves from_integrals and elliptic_cylinder.
• [should] semi-axis validation / infinite-axis parsing duplicated (:98-108, :139-148) → detail::semi_axes; std::size_t axis + run-time range check + ambiguous overload pair → enum class axis.
• [should] header must state: ν is the matrix Poisson ratio; S has minor symmetries only.
• [nit] cyclic convention of elliptic_cylinder undocumented; <limits> unused; test helper trace computes S_iikk; spheroid reference is uvwmaterial's own transcription (cite Mura 11.28/11.29); no ν<0, no float.

[petlenz]

Refactored as requested: each Eshelby tensor is its own material (eshelby_sphere, eshelby_cylinder, eshelby_spheroid, eshelby_ellipsoid), wired to the matrix by name and publishing the output tensor; the PR body describes the design. The consumers (inclusion, mean_field_composite in #131, short_fibre_composite in #133, the curing test in #137, the converter in #139) follow as the stack is rebased.

Review findings addressed in the same commit:

• should, per-index assembly: detail::orthotropic(A, s) on the eigenprojectors, shared by the ellipsoid and the elliptic cylinder.
• should, near-equal axes: threshold √ε; continuity tested at δ = 1e-9, 1e-11, 1e-13 to 1e-8.
• should, semi-axis parsing duplicated: detail::infinite_axis.
• should, header doc: ν is the matrix Poisson's ratio, minor symmetries only.
• Carlson integrals: Boost.Math ellint_rd (dependency added here, where it is first used).

Kept: std::size_t axis for the free cylinder functions (now internal to the nodes); the depolarisation tensors stay in #129.