Continuum-Fitting Spin Inference

Continuum fitting measures an apparent emitting area from the thermal X-ray spectrum of a stellar-mass black-hole disk. Spin enters only after that area is converted to a physical inner radius and the inner radius is identified with the Kerr innermost stable circular orbit. The method is therefore strongest in a deliberately narrow parameter regime: an optically thick, geometrically thin, disk-dominated state with independently constrained mass, distance, and viewing inclination. Outside it, an excellent count-spectrum fit can still yield a precise answer to the wrong physical model.

Required background. Probability and Inference for Gravity supplies response-space likelihoods, nuisance-parameter marginalization, model comparison, and posterior predictive checks.

Helpful background. Black-Hole Geometry and Mechanics supplies the Kerr ISCO and photon transfer; Relativistic Matter and Compact Objects supplies disk thermodynamics and atmosphere physics; and Black-Hole Astrophysics supplies accretion-state and binary context.

From thermal area to the Kerr ISCO

A useful Newtonian scaling isolates the inverse problem. If the color temperature is $T_{\rm col}=f_{\rm col}T_{\rm eff}$, then a disk-dominated bolometric flux behaves approximately as

$$F_{\rm bol}\propto \frac{R_{\rm in}^2\cos i}{D^2} \left(\frac{T_{\rm col}}{f_{\rm col}}\right)^4, \qquad R_{\rm in}\propto \frac{D f_{\rm col}^2}{\sqrt{\cos i}}\, \frac{\sqrt{F_{\rm bol}}}{T_{\rm col}^2}.$$

This relation already exposes why distance $D$, inclination $i$, and atmosphere hardening $f_{\rm col}$ are not optional corrections. A relativistic model replaces it with an integral of the atmosphere intensity over geodesics,

$$F_E^{\rm src}(\Theta)= \int_{\rm image} g^3 I_{E/g}^{\rm atm} \!\left[r,\mu_{\rm em}; M,\dot M,a_*,f_{\rm col},\eta_{\rm atm}\right]d\Omega,$$

where $g=E_{\rm obs}/E_{\rm em}$ includes gravitational and Doppler shifts, and $\Theta$ contains disk, binary, and calibration parameters. The Novikov–Thorne surface-flux profile supplies the radial effective temperature. Disk-to-observer ray tracing supplies the geometric transfer, including lensing and frequency shifts, while an atmosphere model supplies spectral hardening and the angular emission law, such as limb darkening. Returning radiation requires a further disk-to-disk transfer calculation and a prescription for absorption, scattering, and re-emission. Reynolds gives the physical and practical construction in Reynolds 2021, §§ 3.3.1–3.3.2.

After applying the inner-boundary and atmosphere corrections, continuum fitting sets $R_{\rm in}=r_{\rm ISCO}(a_*)GM/c^2$. For $-1\leq a_*\leq1$,

$$r_{\rm ISCO}(a_*) =3+Z_2-\operatorname{sgn}(a_*) \sqrt{(3-Z_1)(3+Z_1+2Z_2)},$$ $$Z_1=1+(1-a_*^2)^{1/3} \left[(1+a_*)^{1/3}+(1-a_*)^{1/3}\right], \qquad Z_2=\sqrt{3a_*^2+Z_1^2}.$$

The corresponding physical prograde ISCO shrinks from $6GM/c^2$ at zero spin toward $GM/c^2$ at the extremal limit. The mapping is nonlinear: the same fractional radius error produces a much more asymmetric spin interval near extremality.

The count-space inference

An observation supplies source and background counts, not $F_E^{\rm src}$. For detector channel $j$ and epoch $k$,

$$\lambda_{jk}= t_k\int R_{jk}(E)\, A(E;N_H,\eta_{\rm abs}) \frac{F_E^{\rm disk}(\Theta_k)+F_E^{\rm Compt}(\Theta_k)}{E}\,dE +B_{jk}, \qquad C_{jk}\sim{\rm Poisson}(\lambda_{jk}).$$

Here $R_{jk}$ is the release-specific redistribution times effective area, $A$ is absorption, $F_E^{\rm Compt}$ treats scattered disk photons consistently, and $B_{jk}$ is the expected background count. The transfer equation defines $F_E$ as energy flux per unit energy; division by photon energy $E$ converts it to photon number flux before the count response is applied. Use consistent energy units in that conversion. Software spectra already expressed in photons per unit area, time, and energy require no second division by $E$. Accepted epochs share $M$, $D$, $i$, $a_*$, and ideally one physical inner radius, while accretion rate and weak coronal parameters may vary. Mass, distance, binary inclination, and any disk-alignment constraint enter as a joint external likelihood, not independent Gaussian constants.

For a concrete reproducible case, use the public RXTE/PCA LMC X-3 observations selected by Steiner et al., whose reduction, flux calibration, spectral selection, and radius test are specified in Steiner et al. 2010, §§ 2–4. Record every ObsID; HEASoft 6.36 and XSPEC 12.15.1; the RXTE CALDB files actually resolved; detector layer and gain epoch; good-time, background, dead-time, energy-band, and systematic-error choices; absorption; and exact atmosphere and transfer model versions. The archive’s mission-long products are discovery aids, not substitutes for extracting response-matched spectra.

Selection must precede the spin fit. A defensible teaching selection requires a disk-dominated spectrum, a declared Compton-scattered fraction, good response coverage near the thermal peak, and a thin-disk luminosity window such as $0.05\lt L/L_{\rm Edd}\lt0.30$. Those numerical cuts are method choices, not laws of nature. Report excluded transition and hard-state epochs so that radius constancy is not manufactured by hiding failures. Use a Poisson source-plus-background likelihood where counts require it, propagate calibration covariance, and check predicted counts and residual structure rather than only a reduced $\chi^2$.

Results and limitations through 2026-08-18

The evidence does not establish that every thermal inner radius is an ISCO. It supports a conditional method whose unresolved question is the accuracy of the disk-to-ISCO identification for each source, state, and model family.

Evidence or result	Parameter regime and discriminant	Scope or limitation
Novikov–Thorne atmosphere plus Kerr transfer predicts a spin-dependent thermal area	Thin, radiatively efficient, aligned disk; test the full response-folded spectrum	Inner stress, hardening, thickness, wind loss, limb law, and returning radiation are model inputs
Selected RXTE spectra of LMC X-3 show a nearly constant fitted inner radius across large luminosity variation	Preselected soft spectra and a declared luminosity window; compare radius epoch by epoch	Constancy supports a stable edge but does not by itself prove that the edge equals the ISCO
GRMHD synthetic spectra quantify bias from departures from the analytic disk	Compare fitted and injected spin across inclination, luminosity, thickness, and magnetic state	Existing simulations cover a finite parameter regime and depend on cooling, atmosphere, and radiative-transfer closure
Transition, strongly Comptonized, truncated, or wind-dominated spectra can fail radius constancy	Retain excluded and null epochs and predict their count spectra	A good fit after flexible continuum adjustment can conceal method breakdown
Independent mass, parallax, inclination, and alignment data narrow the radius scale	Refit rather than fixing external quantities	Shared priors can create apparent cross-epoch or cross-method agreement

The competing positions are not simply “continuum fitting works” versus “it fails.” They include a low-stress thin disk reaching the ISCO; a finite-stress or magnetized inner disk with a modest bias; a thick or truncated disk; an atmosphere-driven color correction; misalignment between binary and inner disk; and response or distance errors. GRMHD spectra provide a controlled challenge, but their answer depends on the simulated luminosity and inclination, as shown in Kulkarni et al. 2011, §§ 2.2–3.2. Reynolds summarizes the corresponding systematic boundary in Reynolds 2021, § 3.3.3.

The interpretation could change after a revised dynamical mass, parallax, binary inclination, or inner-disk alignment constraint; after a response or atmosphere revision changes the inferred radius; or after simulations or observations demonstrate truncation, thickness, winds, or inner stress in the selected state.

Adversarial checks and reproduction

Radius before spin. Fit $R_{\rm in}$ freely across epochs before enforcing the Kerr mapping. A model that cannot reproduce radius constancy in its advertised state should not return a combined spin posterior.

Leave out the most constraining spectrum. Repeat the hierarchy without the highest-signal epoch. A large shift can reveal calibration leverage or continuum misspecification that a narrow formal interval hides.

Vary physical families, not one parameter at a time. Repeat with alternative atmosphere, inner torque, disk thickness, alignment, returning-radiation, and Comptonization treatments. Their differences are model dependence and should not be folded into counting error.

A truthful reproduction retrieves the named RXTE event or standard-mode products from HEASARC, freezes the ObsID list before fitting, regenerates background and responses with the pinned release, forward-folds every accepted epoch, and publishes the spectra, responses, selection table, configuration, external likelihood samples, and posterior-predictive residuals. It must also state whether proprietary atmosphere tables or unreleased calibration covariance prevent exact independent recovery.

Exercises

At fixed $M$, $T_{\rm col}$, and flux, suppose the adopted distance rises by $5\%$, the hardening factor rises by $3\%$, and the inclination changes from $60^\circ$ to $63^\circ$. Using the scaling above, estimate the fractional change in $R_{\rm in}$. Would the inferred prograde spin rise or fall if the radius is identified with the ISCO?

Solution

The radius ratio is

$$\frac{R_{\rm in,new}}{R_{\rm in,old}} =1.05(1.03)^2 \sqrt{\frac{\cos60^\circ}{\cos63^\circ}} \simeq1.05\times1.0609\times1.049 \simeq1.168.$$

Thus the inferred radius grows by about $16.8\%$. On the prograde Kerr branch, $r_{\rm ISCO}$ decreases as $a_*$ increases, so identifying the larger radius with the ISCO moves the inferred spin downward. This exercise also shows why varying $D$, $f_{\rm col}$, and $i$ separately and then adding symmetric errors can fail: their nonlinear product creates an asymmetric radius and spin posterior.

References

• HEASARC. n.d. “HEASoft: A Unified Release of the FTOOLS and X-Ray Analysis Software Packages,” version 6.36. Official software page.
• Kulkarni, Akshay K., et al. 2011. “Measuring Black Hole Spin by the Continuum-Fitting Method: Effect of Deviations from the Novikov–Thorne Disc Model.” Monthly Notices of the Royal Astronomical Society 414:1183–1194. DOI.
• Reynolds, Christopher S. 2021. “Observational Constraints on Black Hole Spin.” Annual Review of Astronomy and Astrophysics 59:117–154. DOI.
• Steiner, James F., et al. 2010. “The Constant Inner-Disk Radius of LMC X-3: A Basis for Measuring Black Hole Spin.” The Astrophysical Journal Letters 718:L117–L121. DOI.