X-Ray Spectral Tests of Strong Gravity

X-ray spectra can probe redshift, Doppler beaming, light bending, and the location of emitting matter near a compact object. They do not isolate the metric: the measured counts also depend on disk atmosphere, ionization, density, corona, absorption, background, detector response, and source-state selection. A spacetime comparison is credible only when these layers are fitted and challenged together. The assessment here is current through 2026-08-19.

Helpful background. Weak-Field Gravity supplies the far-field and redshift checks; Probability and Statistical Inference supplies forward-folded likelihoods; Black-Hole Geometry and Mechanics supplies the Kerr photon and orbital structure; and Black-Hole Astrophysics supplies disks, coronae, absorption, and source states.

Relativistic transfer into detector counts

Along a transparent ray, $I_\nu/\nu^3$ is invariant. Let $I_e(E_e)$ denote emitted energy specific intensity per unit photon energy. If $g=E_o/E_e$ is the redshift factor, the observed energy flux density is

$$F_o(E_o)=\int_{\rm image}g^3 I_e\!\left(\frac{E_o}{g},r,\mu_e;\eta_{\rm src}\right)d\Omega.$$

The metric and observer tetrad determine rays, $g$, emission angle $\mu_e$, lensing, and the transfer from disk to image. The source parameters $\eta_{\rm src}$ determine local emissivity, absorption, ionization, density, electron distribution, and disk thickness. A deformation that moves the innermost stable circular orbit can be degenerate with spin or an inner edge that does not coincide with that orbit.

The redshift has an observer-independent definition once the emitter and observer are specified: $g=(-k_\mu u_o^\mu)/(-k_\mu u_e^\mu)$, where $k^\mu$ is the photon wavevector and $u_e,u_o$ are their four-velocities. As a check, emission at a single rest energy $E_0$, $I_e(E_e)=I_0\delta(E_e-E_0)$, contributes $g^4I_0\delta(E_o-gE_0)$ to the integrand. The extra $g$ comes from transforming the delta function; a broad line mixes many such shifted contributions across the disk.

Detectors count photons. Their incident photon flux density is $N_o(E)=F_o(E)/E$, with units such as photons per square centimetre per second per keV. For channel $i$,

$$\lambda_i=t_{\rm exp}\int dE\, R_i(E)A_{\rm eff}(E) \left[N_{\rm cont}(E)+N_{\rm refl}(E)\right]+B_i,$$

where $R_i(E)$ is the dimensionless probability of redistribution into channel $i$, $A_{\rm eff}(E)$ is effective area, and $B_i$ is expected background counts over exposure $t_{\rm exp}$. Absorption along the path is included in the incident $N$ spectra. The product has units of counts; inserting an energy flux in its place would weight photons by an extra factor of $E$. Some supplied response matrices already include effective area, so it must be applied only once. This photon-response convention is documented by HEASARC, XSPEC manual, “The XSPEC implementation,” response and model-spectrum subsections.

A response matrix, ancillary response, calibration release, observation identifier, extraction region, grouping rule, pileup treatment, dead time, and background region belong to the result. Reusing unfolded points in a new metric fit loses the response likelihood.

Continuum fitting adds black-hole mass, distance, inclination, accretion rate, color correction, torque, returning radiation, and atmosphere. Reflection adds coronal spectrum and geometry, disk emissivity, iron abundance, ionization $\xi=4\pi F_X/n$, density, thickness, absorption, and distant reprocessing. Reynolds details both the statistical power and the disk, corona, absorption, and calibration systematics in Reynolds 2021, §§3.2.1–3.2.4 and §§3.3.1–3.3.3.

Selection and model comparison

A versioned analysis starts from event files and named calibration database versions, not a published best-fit spectrum. It records good-time intervals, source state, flux and pileup thresholds, energy band, detector modules, extraction masks, background, absorption model, and whether multiple epochs were selected because they appeared disk dominated or reflection dominated. Such cuts condition the population and can favor data that agree with a preferred spectral family.

For Poisson counts, a normalized likelihood is

$$\mathcal L(\vartheta,\eta)= \prod_i\frac{\lambda_i(\vartheta,\eta)^{n_i} e^{-\lambda_i(\vartheta,\eta)}}{n_i!},$$

extended with background and calibration nuisance terms. The tested spacetime parameters $\vartheta$ require priors whose measure is stated; spin–deformation contours can change under a different coordinate range or regularity boundary. Posterior predictive checks should compare counts and residual correlations across detectors and epochs. Covariance introduced by shared effective-area or atomic-model calibration cannot be removed by multiplying nominally independent spectra.

Reproducibility requires observation identifiers, exact event and response files or archive queries, calibration version, spectral-extraction software and commands, model code and commit, atomic tables, priors, sampler settings, convergence diagnostics, and posterior samples. This page reports no new spectral fit. Numerical constraints in older reviews are examples tied to their data and models, not current bounds.

Results and limitations through 2026-08-19

Evidence or position	Result supported	Limitation or scope
Broad reflection features and relativistic transfer fits	Show that inner-disk motion, redshift, and illumination shape the spectrum	Absorption, density, ionization, corona, disk thickness, and calibration remain coupled
Thermal continuum fits in selected disk-dominated states	Constrain an apparent inner scale under atmosphere and disk assumptions	Mass, distance, inclination, color correction, stress, and inner-edge assumptions dominate mapping
Joint epochs, detectors, or spectral methods	Can test whether one geometry predicts different source states	Shared atomic physics and source-model families prevent automatic independence

The conservative position is that X-ray data constrain spacetime parameters conditionally within validated disk–corona families. A competing interpretation may regard a tight deformation posterior as a metric test, but evidence against that claim arises when alternate absorption, density, thickness, or coronal geometry moves the result. Agreement between continuum and reflection helps only if their shared mass, inclination, disk physics, and calibration covariance are retained. The result remains model-dependent and does not establish a unique non-Kerr geometry.

The adversarial check generates synthetic spectra with withheld disk–corona and spacetime combinations, folds them through alternate responses, and asks whether the analysis recovers a GR injection with calibrated coverage. A putative deformation should persist across source states, detector modules, absorption families, disk thicknesses, and coronal prescriptions while predicting held-out energy channels or epochs.

What could change the conclusion. The interpretation could change after a superseding event or calibration release, a material atomic or transfer-model correction, or a blinded synthetic challenge demonstrates that alternate disk, corona, absorption, or detector responses change deformation recovery.

Exercises

1. Energy flux versus photon counts. Two narrow incident lines at $E$ and $2E$ carry equal integrated energy flux. An ideal detector has the same effective area and efficiency for both. What is their count ratio?

Solution

Photon number flux is energy flux divided by photon energy. The second line therefore supplies half as many photons, so its expected count is half that of the first. Equal energy flux does not mean equal photon counts. Folding $F_E$ instead of $N_E=F_E/E$ through a photon response would miss precisely this factor.

2. Response inversion. Why can fitting an unfolded flux spectrum with ordinary least squares give a misleading spacetime constraint even when the plotted error bars look Gaussian?

Solution

Unfolding uses an assumed spectral model to invert a non-diagonal detector response, so the displayed flux points and errors are model dependent and correlated. Ordinary least squares then treats them as independent measurements and can count the original model assumption twice. Therefore one should predict incident photons for each spacetime and source model, fold them through the response, and evaluate the count likelihood with background and calibration nuisances. This preserves Poisson information and exposes response–geometry degeneracies.

References

• High Energy Astrophysics Science Archive Research Center (n.d.), XSPEC User’s Guide, “The XSPEC implementation,” response and model-spectrum definitions (accessed 2026-09-24). NASA Goddard Space Flight Center. Official manual.
• Reynolds, C. S. (2021), “Observational Constraints on Black Hole Spin,” Annual Review of Astronomy and Astrophysics 59, 117–154. DOI.

Further reading

• Bambi, C. (2017), “Testing Black Hole Candidates with Electromagnetic Radiation,” Reviews of Modern Physics 89, 025001. DOI.
• Psaltis, D. (2008), “Probes and Tests of Strong-Field Gravity with Observations in the Electromagnetic Spectrum,” Living Reviews in Relativity 11, article 9. DOI.