A Radiation Exchange Factor Transformation with Proven Convergence, Non-Negativity, and Energy Conservation

TL;DR

This work introduces a matrix-based exchange factor transformation for coupled radiative transfer with mixed boundary conditions. It converts a first-interaction exchange-factor matrix $\mathbf{F}$ into an absorption $\mathbf{A}$ and a multiple reflection-scattering $\mathbf{R}$ through an analytically traced Neumann-series pathway, with rigorous proofs of convergence, non-negativity, and exact energy conservation. The method reveals a discrepancy in Noble's Hottel zonal-method formulation, which the new transformation avoids, and it is validated against diffusion theory and Crosbie & Schrenker results while handling general geometries and complex boundaries. Practically, it offers a scalable, analytic approach for medium-scale problems and a pathway to large-scale problems when reflection-scattering is sparse, enabling accurate radiative transfer analyses across transparent to highly scattering media. The framework emphasizes physical correctness, numerical precision control, and flexibility via Monte Carlo-derived or analytic exchange factors, with potential extensions to domain decomposition and non-grey spectra.

Abstract

This paper presents a matrix-based exchange factor transformation for solving coupled mixed boundary condition radiative transfer problems on general domains. The method applies to participating media ranging from transparent to absorbing, emitting, and scattering, with boundaries ranging from absorbing to reflecting. Given a first-interaction exchange factor matrix $\mathbf{F}$, the transformation produces an absorption matrix $\mathbf{A}$ and a multiple reflection-scattering matrix $\mathbf{R}$ through a Neumann series that analytically traces all reflection-scattering paths to steady state. The paper establishes rigorous conditions under which the method guarantees convergence, non-negative radiation, and exact energy conservation to machine precision. A comparison with Noble's matrix formulation of Hottel's zonal method reveals a previously unidentified discrepancy in that classical approach; the proposed transformation eliminates this discrepancy. The method is validated against the diffusion approximation in the high-extinction limit and against results of Crosbie and Schrenker for pure and partial scattering cases. The method is applicable to medium-scale general reflecting-scattering problems and scales to large problems when negligible reflection-scattering and high extinction ensure matrix sparsity.

Figures (12)

• Figure 1: Normalized emissive power along the short centreline in a 1000 m by 1 m geometry (approximately one-dimensional), calculated from the proposed transformation, compared to the diffusion approximation between infinite parallel plates Daun2021. The medium is absorbing-emitting non-scattering of extinction $\beta=100$ and the domain has one hot wall, and the remaning non-reflecting non-emitting, all $\varepsilon_w=1$, with the volume divided into $3\times 51$ elements, and an $\mathbf{F}$ from $10^9$ ray samples.
• Figure 2: Non-dimensional source function $i_{\mathrm{V},i}/i_\mathrm{S}$ for the centre line perpendicular to the incident Lambertian intensity onto a two-dimensional square geometry as a function of relative optical depth, based on an estimated $\mathbf{F}$ from ray tracing $10^{9}$ rays in total with no smoothing applied, in a $21\times 21$ geometry, compared to the solutions of Crosbie and Schrenker Crosbie1984.
• Figure 3: Comparison of Noble's matrix formulation Noble1975 of Hottel's Zonal Method (left) to the proposed exchange factor transformation (right), showing the magnitude of predicted source fluxes of each element in a $21\times 21$ unit square enclosure of a medium in radiative equilibrium with constant extinction $\beta=1$ and varying albedo $\omega$.
• Figure 4: Root mean square error of the total radiant power in a unit square $21\times 21$ enclosure with incident Lambertian intensity onto one side, from an emissive power corresponding to 1000 K, comparing cases with total number of samples of $10^4$, $10^5$, $10^6$, $10^7$, $10^8$ to a reference solution obtained using $10^{10}$ samples.
• Figure 5: Ratio of RMS relative output uncertainty $(\sigma/\mu)_\mathrm{out}$ (calculated from $\mathbf{j}$) to RMS relative input uncertainty $(\sigma/\mu)_\mathrm{in}$ (calculated from $\mathbf{F}$) as a function of the number of ray samples used for obtaining $\mathbf{F}$ (x-axis) and of the number of domain divisions per dimensions (y-axis) in a two-dimensional square geometry. Estimates or results equal to zero were excluded from the calculations.

Theorems & Definitions (28)

• Theorem 3.1: Convergence and Irreducibility of the Steady State Path
• proof
• Remark 3.2
• Theorem 3.3: Numerical Stability of the Steady State Path
• proof
• Theorem 4.1: Spectral Radius of Combined Absorption and Reflection-Scattering
• proof
• Theorem 4.2: Spectral Radius of Absorption
• proof
• Theorem 4.3: Upper Bound on Spectral Radius of Reflection-Scattering