SP3 transport model =================== The ``SP3`` solver solves a steady-state simplified third-order moment model on the same finite tetrahedral mesh as ``P1``. Its unknowns are two coupled composite moments. In planar one-dimensional geometry these equations coincide with the corresponding P3 moment equations; in three dimensions SP3 is a simplification of P3, not the full spherical-harmonics system. Scattering and bulk equations ----------------------------- The reduced scattering coefficient alone does not determine the higher angular moments needed by SP3. Each material supplies a scalar phase function whose normalized Legendre moments are :math:`g_0=1,g_1,g_2,g_3`. For absorption :math:`\mu_a` and reduced scattering :math:`\mu'_s=\mu_s(1-g_1)`, define .. math:: \mu_l=\mu_a+\mu'_s\frac{1-g_l}{1-g_1},\quad l=1,2,3. The Henyey–Greenstein model requires material ``anisotropy`` :math:`g=g_1` and has :math:`g_l=g^l`. For that specific model, .. math:: \mu_1 &= \mu_a + \mu'_s,\\ \mu_2 &= \mu_a + \mu'_s(1+g),\\ \mu_3 &= \mu_a + \mu'_s(1+g+g^2),\\ D_1 &= (3\mu_1)^{-1},\qquad D_2 = (7\mu_3)^{-1}. The two composite nodal fields :math:`u` and :math:`v` obey .. math:: -\nabla\!\cdot(D_1\nabla u) + \mu_a u - \tfrac23\mu_a v &= 0,\\ -\nabla\!\cdot(D_2\nabla v) - \tfrac23\mu_a u + \tfrac19(4\mu_a+5\mu_2)v &= 0. Each coefficient is evaluated with the local material's phase-function moments at the source wavelength. See :doc:`phase_functions` for the available models. The physical fluence and first angular current are .. math:: \Phi = u - \tfrac23 v,\qquad \mathbf J=-D_1\nabla u. Both composite fields are available as ``result.moments['composite_1']`` and ``result.moments['composite_2']``. ``result.fluence`` contains :math:`\Phi`, and ``result.flux`` contains element-centered :math:`\mathbf J`. Marshak face conditions ----------------------- Let :math:`j_1=-D_1\mathbf n\cdot\nabla u` and :math:`j_2=-D_2\mathbf n\cdot\nabla v`, with :math:`\mathbf n` pointing outward. On a nonreflecting, matched-index face, the incoming radiance is assumed constant over the inward hemisphere. Its partial current is :math:`I_{\mathrm{in}}`. To derive the face conditions, use the local planar angular expansion in :math:`\mu=\mathbf\Omega\cdot\mathbf n`: .. math:: L(\mu)=\frac{1}{4\pi}\left[\Phi+3\mu j_1 +\frac53 v P_2(\mu)+7j_2P_3(\mu)\right]. The second angular moment is :math:`v/3`. For a constant prescribed incoming radiance, the two half-range conditions are .. math:: \int_{\mu<0}(-\mu)L\,d\Omega &= \tfrac14u-\tfrac1{16}v-\tfrac12j_1=I_{\mathrm{in}},\\ \int_{\mu<0}P_3(\mu)L\,d\Omega &= \tfrac1{16}u-\tfrac7{48}v+\tfrac12j_2 =\tfrac14I_{\mathrm{in}}. Solving these two equations for the outward currents gives .. math:: j_1 &= \tfrac12 u-\tfrac18 v-2I_{\mathrm{in}},\\ j_2 &= -\tfrac18 u+\tfrac7{24}v+\tfrac12 I_{\mathrm{in}}. The second source term follows from integrating :math:`(5|\mu|^3-3|\mu|)` over the inward hemisphere. It cannot be inferred from incoming power for an arbitrary angular distribution; the isotropic incoming assumption fixes it. ``UniformSurfaceSource`` and ``SurfaceSource`` use their usual spatial footprints and coupled optical power, now with this explicit angular assumption. A reflecting face has :math:`j_1=j_2=0` and cannot be illuminated. Only ``RobinBoundary(effective_reflectance=0)`` is supported by SP3. A single P1 effective reflectance does not specify the higher angular reflection moments required by an SP3 refractive boundary. Nonzero values are rejected, as are internal refractive-index jumps. No index matching is inferred from the material's numerical refractive index: the user explicitly chooses the matched-index boundary model. On a nonreflecting face the outgoing partial current is :math:`j_1+I_{\mathrm{in}} = u/2-v/8-I_{\mathrm{in}}`. These values are integrated at boundary quadrature points for face powers and detector apertures. Detector angular acceptance continues to use MomentRT's separate Lambertian escape approximation; SP3 does not provide Fresnel-resolved detector optics. Negative fields are retained and reported rather than clipped. Finite-element workflow ----------------------- Multiplying each bulk equation by a nodal test function and integrating by parts gives two diffusion stiffness matrices, three bulk mass terms, and a coupled boundary mass matrix with coefficients :math:`1/2`, :math:`-1/8`, and :math:`7/24`. The boundary load is :math:`(2I_{\mathrm{in}}, -I_{\mathrm{in}}/2)`. The resulting symmetric sparse system has twice as many nodal unknowns as P1. Material coefficients are sampled at tetrahedron centroids; source and detector footprints use boundary triangle quadrature. Spectral sweeps reuse the geometry and quadrature, then assemble and solve a new coupled operator at each wavelength. The test suite independently solves homogeneous and layered planar SP3 systems using a matrix exponential and checks mesh refinement, outgoing power, and energy balance. This validates the implementation against that SP3 problem; it does not validate SP3 against radiative transfer or biological measurements. Check mesh convergence, and compare against Monte Carlo or other transport references for the intended sensor distances and tissue properties. For runnable examples, see :doc:`gallery/transport_models/sp3_layered_anisotropy` for material-specific anisotropy and :doc:`gallery/spectral/sp3_spectral_anisotropy` for a wavelength sweep. Reference: H. K. Kim et al., *Frequency-domain optical tomographic image reconstruction algorithm with the simplified spherical harmonics (SP3) light propagation model*, International Journal of Thermal Sciences 116 (2017), 265–277. `Author-hosted paper `_.