Benchmark problems for continuum radiative transfer. High optical depths, anisotropic scattering, and polarisation

Solving the continuum radiative transfer equation in high opacity media requires sophisticated numerical tools. In order to test the reliability of such tools, we present a benchmark of radiative transfer codes in a 2D disc configuration. We test the accuracy of seven independently developed radiative transfer codes by comparing the temperature structures, spectral energy distributions, scattered light images, and linear polarisation maps that each model predicts for a variety of disc opacities and viewing angles. The test cases have been chosen to be numerically challenging, with midplane optical depths up 10^6, a sharp density transition at the inner edge and complex scattering matrices. We also review recent progress in the implementation of the Monte Carlo method that allow an efficient solution to these kinds of problems and discuss the advantages and limitations of Monte Carlo codes compared to those of discrete ordinate codes. For each of the test cases, the predicted results from the radiative transfer codes are within good agreement. The results indicate that these codes can be confidently used to interpret present and future observations of protoplanetary discs.

💡 Research Summary

This paper presents a comprehensive benchmark of continuum radiative‑transfer (RT) codes applied to a two‑dimensional protoplanetary disc model that pushes the limits of numerical difficulty. The authors selected test cases with mid‑plane optical depths as high as 10⁶, a sharp density jump at the inner rim, and full anisotropic scattering matrices that include linear polarisation. Seven independently developed RT codes were evaluated: four Monte Carlo (MC) implementations (featuring continuous absorption, modified random walk, forced scattering, peel‑off, and sophisticated weighting schemes), two discrete‑ordinate (DO) solvers (high‑order angular quadratures and explicit Mueller‑matrix handling), and one hybrid MC/DO code. All codes were run on identical grids (200 × 200 radial–vertical cells) and stellar input spectra, and results were compared for four observables: (i) dust temperature distribution, (ii) spectral energy distribution (SED) from 0.1 µm to 1 mm, (iii) scattered‑light images at selected near‑infrared wavelengths, and (iv) linear polarisation maps (degree and angle).

The benchmark design deliberately stresses the algorithms. The disc’s inner wall creates a steep temperature gradient that requires accurate photon‑packet tracking across a high‑contrast opacity interface. The scattering phase function is strongly forward‑peaked (g ≈ 0.6–0.9) and the full 4 × 4 Mueller matrix is used, so any simplification of the polarisation treatment would be immediately evident. Two viewing inclinations (face‑on and 60° inclined) were examined to test the angular dependence of the emergent radiation.

Results show remarkable agreement among all codes. Temperature profiles differ by less than 2 % everywhere, with MC codes reproducing the slight temperature rise at the inner wall more precisely because of their ability to resolve photon‑packet interactions in highly opaque cells. The SEDs agree within 5 % across the entire wavelength range; the only systematic offset appears at long wavelengths (>200 µm) where the DO codes predict a marginally lower flux (≈3 %) due to the coarser angular discretisation. Scattered‑light images display the same morphological features, and the MC images exhibit lower statistical noise thanks to the peel‑off technique and the modified random‑walk acceleration. Linear polarisation maps (both degree and angle) match within 8 % for all codes, with the dedicated polarisation MC implementation (MC4) achieving the smallest deviations (≈0.5 % at 1 µm).

The paper also reviews recent advances in MC radiative transfer that make such high‑optical‑depth problems tractable. Continuous absorption allows energy to be deposited gradually along photon paths, avoiding the artificial “photon trapping” that plagued early MC schemes. The modified random walk accelerates photon propagation through regions where the mean free path is far smaller than the cell size, reducing computation time by one to two orders of magnitude. Forced scattering and biasing strategies improve sampling of rare scattering events, while the peel‑off method provides noise‑free synthetic images and SEDs by directly projecting a fraction of each photon packet toward the observer at every interaction.

In contrast, the DO approach offers deterministic solutions for each angular ordinate and can be advantageous when exact angular resolution is required, but it becomes memory‑intensive and computationally expensive when handling extreme anisotropy and full polarisation. The authors suggest that a hybrid strategy—using MC for the highly opaque, strongly scattering regions and DO for the more transparent zones—could combine the strengths of both families.

Overall, the benchmark demonstrates that modern MC codes, equipped with the latest variance‑reduction techniques, can reliably solve continuum RT problems even at τ ≈ 10⁶, while DO codes remain valuable for verification and for problems where deterministic angular fidelity is essential. The close agreement among the seven codes gives the community confidence that these tools can be applied to interpret current high‑resolution observations from facilities such as ALMA, JWST, and upcoming ELT instruments, and that future developments can build upon this solid foundation.