Multiband and Multimessenger Black-Hole Astrophysics

Black holes are not observed through a single transparent channel. Their gravity, matter, radiation, and environments are filtered into images, spectra, time series, gravitational waves, neutrinos, cosmic rays, and selected catalogs. The purpose of multimessenger inference is to ask whether those products admit one causal source model—not merely whether their plotted summaries look similar. This chapter develops that discipline from single-source cross-band synthesis to selection-aware population inference. Instrument-specific gravitational-wave likelihood methods remain outside its scope.

Helpful background. Black-Hole Geometry and Mechanics supplies mass, spin, and horizon scales; Relativistic Matter and Compact Objects supplies source populations and relativistic matter; Black-Hole Astrophysics supplies accretion, jets, and tidal phenomena; Gravitational-Wave Astronomy supplies event inference and detector selection; Relativistic Cosmology supplies redshift, distance, and source-frame rates.

Enter this chapter

The six pages progress from combining measurements of one evolving source to comparing entire selected populations.

1. Cross-Band Spectral and Timing Synthesis constructs a latent source-state model for nonsimultaneous spectra, light curves, and lags. Cadence, censoring, calibration, and transfer functions enter before a correlation is interpreted.
2. Electromagnetic Counterparts of Black-Hole Mergers turns spatial, temporal, and energetic coincidence into association odds against a measured transient background. It distinguishes an interesting candidate from a confirmed common source.
3. Neutrinos, Cosmic Rays, and Black-Hole Jets follows nonthermal energy from acceleration through hadronic interactions, opacity, cascades, beaming, and detector exposure. Energetic closure and population limits constrain interpretations that a positional match alone cannot.
4. Tidal Events Across Messengers builds the causal delay chain from disruption through fallback, circularization, accretion, outflow, and messenger production. It treats upper limits and follow-up completeness as part of the evidence.
5. Imaging, Timing, and Orbital Probes Together combines complementary scalings with mass, distance, inclination, metric, and source-state parameters represented at the correct shared or local level.
6. Population Synthesis Across Surveys and Detectors completes the argument with a marked point-process likelihood, catalog overlap, posterior recycling, selection injections, and a rerunnable synthetic recovery test.

The order matters. A population model cannot repair a misspecified single-source transfer function, and a precise joint posterior cannot repair double-counted data.

One causal architecture

A useful decomposition separates five layers. Population or source parameters $\Theta$ generate a latent physical history $\lambda(t)$; messenger-specific physics maps that history into emitted signals; propagation changes those signals; instruments produce calibrated data $d_k$; and selection rules decide whether an event enters catalog $C_k$. Schematically,

$$p(\{d_k,C_k\}\mid\Theta) = \int p(\lambda\mid\Theta) \prod_k p(d_k\mid\lambda,\phi_k)\, p(C_k\mid d_k,\psi_k)\, p(\phi_k,\psi_k)\,d\lambda\,d\phi_k\,d\psi_k .$$

Here $\phi_k$ contains calibration and transfer nuisance parameters, while $\psi_k$ contains cadence, threshold, and follow-up selection. The product is justified only after conditional independences have been argued. A triggered observation, shared calibration, reused distance prior, or overlapping event joins branches of the graph and must be represented explicitly.

Three distinctions prevent most false synthesis. First, shared physics is not shared source state: mass and distance may persist across epochs while density, magnetic flux, obscuration, and transient geometry change. Second, simultaneity is not identity: a common time window raises association odds only relative to the background rate and sky localization. Third, catalog agreement is not population agreement: observed distributions are the astrophysical distribution multiplied by different selection functions.

This architecture also clarifies why null results matter. A gamma-ray nondetection can bound hadronic jet power only after opacity and survey exposure are modeled. An absent electromagnetic counterpart constrains the detectable counterpart fraction, not automatically the fraction of mergers in gas. A missing population in a sensitive part of parameter space is informative; a missing population outside exposure is not.

What observations established by 2026-08-18

The common evidence cutoff is 2026-08-18; results after that date are out of scope. Durable inputs include released timestamps, calibrated strain or visibility products, spectra, localization maps, detector response, survey cadence, and published selection cuts. Mutable conclusions include source association, virial or emission-model masses, jet composition, unique formation channels, and the claim that one population explains several catalogs.

For example, the EHT Sagittarius A* release measures a horizon-scale ring while explicitly conditioning comparisons on stellar-orbit information and variable-source modeling Event Horizon Telescope Collaboration 2022, §§3–7, Tables 1–2. GWTC-5.0 population inference uses a stated event threshold, posterior products, and detectable injections rather than every public candidate indiscriminately LIGO–Virgo–KAGRA 2026, §§2.1–3 and Appendix A.2. IceTracks-DR2 publishes fourteen years of track data with binned responses and effective areas, but it does not identify a unique black-hole source population by itself IceCube Collaboration 2026, public-release “Introduction” and “Data release”.

Every claim should therefore carry a status appropriate to its inferential distance from the data. “The detector recorded this calibrated product” is different from “this object caused the event,” which is different again from “this formation channel dominates the population.” The pages preserve those levels rather than averaging them into a single confidence label.

A working method for joint inference

Begin by freezing each dataset version, calibration, time and astrometric frame, source/detector-frame convention, cosmology, and selection rule. Then draw the causal graph and mark each parameter as global, source-specific, epoch-specific, or instrument-specific. Only after that should likelihood factors be multiplied.

Next, test identifiability with synthetic data. Inject a known source or population through the complete cadence, response, selection, and overlap logic; recover it with the analysis code; and vary numerical resolution until the result is stable. Successful structural validation is not a substitute for this scientific benchmark.

Fit each data product separately before the joint model. Posterior-predict the other products while holding them out, and verify that the joint fit improves prediction rather than merely narrows a marginal. For population work, reproduce each catalog’s rate under its own exposure and check the predicted number of cross-catalog overlaps. Selection-corrected hierarchical inference is essential because the detected distribution need not equal the underlying one Mandel, Farr, and Gair 2019, §§2–4, pp. 1087–1092.

Finally, report statistical uncertainty separately from calibration, transfer-model, population, selection, and association systematics. Preserve competing physical positions, state the parameter regimes where each remains viable, and explain how future data or a failed prediction could change the interpretation.

Check your synthesis

An imaging analysis used a Gaussian prior on $M/D$ derived from stellar-orbit data. A researcher multiplies the published imaging posterior by the original orbit likelihood and by a flare-timing posterior from a different epoch, forcing one immutable plasma state. Identify the two formulation errors and give a defensible joint model.

Solution

The orbit information is counted twice: it entered the imaging posterior as a prior and is multiplied again as a likelihood. Recover an imaging likelihood by dividing out the documented prior where that operation is valid, or preferably refit the calibrated imaging data without that orbit prior. Include the orbit likelihood once.

The common-state assumption is also unjustified. A defensible model shares persistent parameters such as $M$, $D$, and any metric parameters, but gives the imaging and flare epochs separate density, temperature, magnetic, emissivity, and transient-pattern variables. The flare period constrains spin only after marginalizing over emitting radius and pattern speed. Fit the three data products with those dependencies, retain calibration covariance, and perform leave-one-probe-out predictions. Narrower constraints count as genuine complementarity only if held-out imaging, orbit, and timing observables are predicted successfully.

References

• Event Horizon Telescope Collaboration et al. (2022), “First Sagittarius A* Event Horizon Telescope Results. I. The Shadow of the Supermassive Black Hole in the Center of the Milky Way,” The Astrophysical Journal Letters 930, L12, doi:10.3847/2041-8213/ac6674.
• IceCube Collaboration et al. (2026), “IceCube Second Track Data Release IceTracks-DR2: Data from 2008–2022 for Neutrino Source Searches,” arXiv:2605.19040v1, arXiv; public data release, doi:10.7910/DVN/MMIIZA.
• LIGO Scientific Collaboration, Virgo Collaboration, and KAGRA Collaboration (2026), “GWTC-5.0: Population Properties of Merging Compact Binaries,” arXiv:2605.27226, arXiv; data release version 6, doi:10.7935/bk00-6a89.
• Mandel, I., Farr, W. M., and Gair, J. R. (2019), “Extracting Distribution Parameters from Multiple Uncertain Observations with Selection Biases,” Monthly Notices of the Royal Astronomical Society 486, 1086–1093, doi:10.1093/mnras/stz896.