Ray Tracing, Transport, and Synthetic Observations

This chapter constructs a synthetic observable as a traceable chain from observer to spacetime, ray, plasma, transfer, image, spectrum, polarization, and comparison metric. Its central discipline is separation: coordinate geometry from local measurements, ray error from transfer error, plasma assumptions from spacetime effects, and synthetic products from calibrated observations. Every stage retains enough metadata to be varied independently.

Helpful background. Formulations of General Relativity supplies metrics and tetrads; Differential Equations for Relativity supplies adaptive and stiff integration; Affine and Variational Geodesics supplies null trajectories; Black-Hole Astrophysics supplies emission and accretion models; and Computational Relativity supplies the shared evidence framework.

Enter this chapter

Start with Null Geodesic Integration, which treats nullness, Kerr constants, reversibility, charts, and event classification as numerical tests. Observer Cameras and Local Tetrads then maps pixels and measured frequencies into coordinate initial data without promoting coordinate axes to observables.

Numerical Covariant Radiative Transfer evolves invariant intensity and optical depth along verified rays. Synchrotron and Thermal Emission Models exposes the electron-distribution and unit-scaling assumptions that turn fluid data into coefficients. Polarized Ray Tracing separates parallel-transported basis rotation from Faraday rotation and conversion. Synthetic-Observation Validation closes the chain with analytic, cross-code, shared-snapshot, and instrument-domain comparisons.

Primary accretion modeling and instrument-specific inference remain outside this computational treatment. A page may specify their interface, but it does not infer a black-hole parameter from an image.

One pipeline, six independent controls

The pipeline can be summarized as

$$\begin{aligned} \text{camera}&\longrightarrow(x^\mu,p_\mu) \longrightarrow\text{ray},\\ \text{ray and plasma}&\longrightarrow(j_\nu,\alpha_\nu,\rho_\nu) \longrightarrow(I,Q,U,V),\\ \text{Stokes image}&\longrightarrow\text{synthetic data product}. \end{aligned}$$

Each arrow has a distinct refinement variable. Geodesic tolerance controls ray integration; metric and fluid grids control interpolation; transfer steps control optical depth and Faraday depth; image pixels and frequency bins control sampling; snapshot cadence controls time averaging; Fourier and instrument grids control derived products. Refining only the final image cannot reveal an upstream plateau.

Pihajoki and collaborators present a general-purpose geometric and polarized-transfer architecture, including numerical tests, in Pihajoki et al. 2018, §§2–4 and §5. The useful lesson is architectural rather than package specific: positions, covectors, frames, coefficients, and observables must transform consistently across charts and local bases.

A first calculation through the chapter

Place a static emitter at Schwarzschild radius $6GM$ and a distant static camera. The camera page gives redshift $g_{\mathrm{red}}=\sqrt{2/3}$. If there is no intervening absorption, the transfer page gives $I_{\nu,\mathrm{obs}}=g_{\mathrm{red}}^3 I_{\nu,\mathrm{em}}(\nu_{\mathrm{obs}}/g_{\mathrm{red}})$. The redshift changes both which emitted frequency is sampled and the observed specific intensity. Lensing then determines which emitting patch reaches each pixel and its solid angle.

This small example separates geometry, frequency, and brightness without a fluid simulation. Continue with the homogeneous emitting slab to add absorption, the pure Faraday solution to add polarization, and the Gaussian image transform to add an ideal interferometric observable. Each has an analytic answer before more complicated source models enter.

Benchmarks before source models

The chapter uses a cumulative benchmark ladder. Flat rays and weak Schwarzschild bending test signs and leading scale. Kerr first integrals test null and symmetry conservation in strong curvature. A homogeneous slab tests emission, absorption, thin and thick limits, while analytic Faraday cases test the full Stokes matrix. Known screen rotations test spin-two polarization conventions.

Only after these analytic tests should codes compare a common smooth source and then an identical versioned GRMHD snapshot. Prather and collaborators define a polarized cross-code suite with analytic and shared-snapshot cases in Prather et al. 2023, §§3.1–3.3 and §§4.1–4.4. Their reported residuals are conditional reproduction targets under identical inputs, not universal tolerances.

For a future executable record, pin metric and snapshot checksums, software revisions, camera configuration, coefficient prescriptions, basis and Stokes conventions, precision, integrator tolerances, grids, masks, and comparison metrics. This specification is not treated as evidence that the calculation has been implemented or run.

From numerical agreement to observational meaning

A synthetic image can be a numerical dataset; it is not a measured observation. If measured data enter, record the official release, calibration pipeline, selections, frequency band, response, likelihood, priors, and evidence date. The observational pages in this chapter use an evidence cutoff of 2026-08-18 and distinguish deterministic residuals from statistical uncertainty.

Systematic variations must include coordinates and tetrads, camera radius and projection, integrator and interpolation controls, electron prescription, optical and Faraday depth, image resolution, time sampling, and post-processing. Predeclare any image alignment or flux normalization. Optimizing these transformations after inspecting the result can hide handedness, origin, or absolute-calibration errors.

The scientific boundary is equally important. A converged difference between two emission prescriptions is a model sensitivity. A converged difference between two spacetimes is not automatically a spacetime signature unless fluid, emission, camera, and selection dependence are controlled. Cross-code agreement shows implementation consistency under common assumptions; it does not validate those assumptions against nature.

Review the chapter

Two ray-tracing codes agree on a polarized image after using the same metric and fluid snapshot. Both violate the homogeneous thick-slab limit by the same $5\%$, while pixel refinement changes the image by less than $0.1\%$. What conclusion follows?

Solution

The cross-code agreement does not establish correct transfer. The shared slab failure points to a common coefficient convention, stiff-step treatment, or formula error, and pixel refinement probes a different scale. Correct the analytic benchmark first, refine transfer and fluid interpolation independently, and only then repeat the image comparison. Agreement between two implementations is not independent evidence when both share the same faulty assumption.

References

• Pihajoki, Pauli, Matias Mannerkoski, Joonas Nättilä, and Peter H. Johansson. 2018. “General-Purpose Ray Tracing and Polarized Radiative Transfer in General Relativity.” The Astrophysical Journal 863, 8. arXiv.
• Prather, Cora, et al. 2023. “Comparison of Polarized Radiative Transfer Codes Used by the Event Horizon Telescope Collaboration.” The Astrophysical Journal 950, 35. arXiv.