Inclination Diffusion in Relativistic Loss Cones

Abstract: Relativistic capture and tidal disruption around a spinning black hole depend on both the magnitude and direction of the star's angular momentum, yet loss-cone models often assume fixed orbital inclinations by ignoring the associated diffusion. We show that this is not justified: for isotropic two-body relaxation near a small loss threshold, angular-momentum magnitude $L$ and inclination $x = L_{z}/L$ diffuse on comparable timescales, $t_{E} \gg t_{L} \sim t_{x}$. For Kerr capture, retaining inclination diffusion significantly amplifies the prograde--retrograde contrast while leaving the total inclination-integrated flux nearly unchanged. An almost correct integrated flux can hide a badly wrong angular distribution. The three-dimensional diffusion problem nevertheless retains enough angular structure to permit analytic treatment. By representing pericenter removal as a continuous sink, we obtain a closed-form loss flux solution for a nearly linear Kerr tidal-disruption boundary, finding close agreement with phase-resolved calculations. Inclination-dependent loss therefore requires inclination-resolved diffusion even when integrated rates appear robust.

• The paper demonstrates that inclination and angular-momentum magnitude diffuse on comparable timescales near relativistic loss surfaces, so fixed-inclination models omit leading-order transport.
• Kerr capture tests show that independent-slice and rapid-mixing prescriptions preserve total fluxes within a few percent but produce angular errors of up to 24% and 87%, respectively.
• A Legendre-mode reduction and weak-dipole tidal-disruption solution retain analytic tractability, reproducing phase-resolved results to about 7% while revealing the importance of orientation-resolved predictions.

Motivation and central claim

Loss-cone theory governs the supply of stars to relativistic outcomes—horizon capture and tidal disruption events (TDEs)—around massive black holes, and thereby sets the rates that feed TDE and EMRI population models. Classical formulations in a spherical galaxy with a Schwarzschild black hole admit a loss boundary that depends only on energy $E$ and angular-momentum magnitude $L$, so the inclination $x = L_z/L$ can be factored out of the Fokker–Planck problem. Kerr spin destroys this simplification: the capture and disruption thresholds become surfaces $L_{\mathrm{lc}}(E, x)$, and the paper argues that the standard workaround—solving independent loss-cone problems at fixed inclination, or bracketing the answer with "inclination-preserving" and "isotropized" prescriptions—has no dynamical justification (2608.16779).

The central result is a timescale ordering near a small loss surface,

$t_{E} \gg t_{L} \sim t_{x},$

derived geometrically from the fact that a transverse velocity kick at large radius changes the components of $\mathbf{L} = \mathbf{r} \times \mathbf{v}$ parallel and perpendicular to $\mathbf{L}$—magnitude and direction—by comparable amounts, while producing only a small fractional energy change. Freezing $x$ therefore discards a leading-order transport process. The paper then demonstrates, via a Kerr capture stress test and an analytic weak-dipole TDE calculation, that misrepresenting inclination diffusion leaves integrated fluxes nearly correct while badly distorting the angular distribution.

Angular separation of the diffusion tensor

Using coordinates $(E, Y, x)$ with $Y \equiv L^2$, the paper transforms the local diffusion tensor for non-resonant two-body relaxation in an isotropic field-star background. Isotropy forces $L$0 exactly, so the collision operator block-diagonalizes: a coupled $L$1 sector plus an angular sector $L$2, the axisymmetric Laplace–Beltrami operator on the unit sphere. After apsidal averaging (justified by rapid relativistic precession), the angular operator is diagonal in Legendre modes, $L$3, reducing the three-dimensional problem to a hierarchy of two-dimensional $L$4 equations coupled only by the geometry of the loss surface.

At fixed $L$5 and leading order in $L$6, the angular-momentum operator is

$L$7

giving $L$8 and $L$9: for every low-$x = L_z/L$0 mode, magnitude and inclination diffuse on the same asymptotic order as $x = L_z/L$1. The paper is explicit that this separable structure rests on isotropy of the field-star distribution and on fast apsidal phase mixing; non-spherical potentials or vector resonant relaxation (VRR) modify the strength of $x = L_z/L$2—encoded in a diagnostic multiplier $x = L_z/L$3, with $x = L_z/L$4 (independent slices) and $x = L_z/L$5 (rapid mixing) as limiting closures—but do not in general justify freezing $x = L_z/L$6.

Kerr capture stress test

The capture threshold at the marginally bound limit (IBSO) for spin $x = L_z/L$7 is strongly inclination-dependent, with a max/min boundary ratio of 4.71. Solving the phase-resolved Fokker–Planck equation with pericenter reset across $x = L_z/L$8 (where $x = L_z/L$9 is the fullness parameter), the paper finds that both limiting prescriptions fail angularly while succeeding integrally:

Prescription	Angular error $L_{\mathrm{lc}}(E, x)$0	Total-flux error	$L_{\mathrm{lc}}(E, x)$1 at $L_{\mathrm{lc}}(E, x)$2
Independent slices ($L_{\mathrm{lc}}(E, x)$3)	up to 24%	< 3.7%	0.051
Two-body baseline ($L_{\mathrm{lc}}(E, x)$4)	—	—	0.285
Rapid mixing ($L_{\mathrm{lc}}(E, x)$5)	up to 87%	< 2.4%	0.397

The prograde–retrograde hemispheric contrast $L_{\mathrm{lc}}(E, x)$6 varies by nearly an order of magnitude between the frozen and rapidly mixed treatments, while the integrated flux errs by at most a few percent. This is the paper's sharpest quantitative warning: an almost correct integrated flux can conceal a badly wrong angular distribution, which matters because inclination controls Lense–Thirring precession signatures in TDEs and is a directly measurable LISA parameter for EMRIs.

Analytic treatment of a weakly dipolar tidal-disruption boundary

For a solar-type star around a $L_{\mathrm{lc}}(E, x)$7, $L_{\mathrm{lc}}(E, x)$8 Kerr black hole, the disruption boundary is nearly linear in $L_{\mathrm{lc}}(E, x)$9: the dipole coefficient is $t_{E} \gg t_{L} \sim t_{x},$0 against a quadrupole of $t_{E} \gg t_{L} \sim t_{x},$1. The paper therefore isolates the pure-dipole problem $t_{E} \gg t_{L} \sim t_{x},$2 and, extending Broggi's continuous-sink formulation, replaces once-per-orbit removal by a sink $t_{E} \gg t_{L} \sim t_{x},$3, eliminating the radial phase coordinate.

Expanding in $t_{E} \gg t_{L} \sim t_{x},$4, the $t_{E} \gg t_{L} \sim t_{x},$5 deformation sources only the $t_{E} \gg t_{L} \sim t_{x},$6 mode, and the steady-state reaction–diffusion equation admits a closed-form solution in terms of modified Bessel functions. The first-order dipole response $t_{E} \gg t_{L} \sim t_{x},$7 reproduces the phase-resolved endpoint limits exactly: $t_{E} \gg t_{L} \sim t_{x},$8 as $t_{E} \gg t_{L} \sim t_{x},$9 and $\mathbf{L} = \mathbf{r} \times \mathbf{v}$0 as $\mathbf{L} = \mathbf{r} \times \mathbf{v}$1. Since $\mathbf{L} = \mathbf{r} \times \mathbf{v}$2 integrates to zero, the total flux changes only at second order in $\mathbf{L} = \mathbf{r} \times \mathbf{v}$3—consistent with the symmetry that flipping the spin direction cannot alter an isotropically supplied total rate. Across the full range of $\mathbf{L} = \mathbf{r} \times \mathbf{v}$4, the continuous-loss model tracks the phase-resolved baseline to within about 7%, while the independent-slice treatment converges to a qualitatively different empty-cone limit $\mathbf{L} = \mathbf{r} \times \mathbf{v}$5 instead of $\mathbf{L} = \mathbf{r} \times \mathbf{v}$6.

The paper concedes that the continuous sink is not microscopically identical to phase-resolved removal: it attaches a Poisson destruction clock with mean waiting time $\mathbf{L} = \mathbf{r} \times \mathbf{v}$7 to stars inside the loss region, and the two prescriptions differ most at intermediate fullness $\mathbf{L} = \mathbf{r} \times \mathbf{v}$8, where the residual reaches its maximum of roughly 7%.

General structure

The Legendre reduction is shown to be a property of the collisional dynamics rather than of any particular loss prescription. For deterministic pericenter removal, the loss surface couples Legendre modes at each reset through overlap matrices $\mathbf{L} = \mathbf{r} \times \mathbf{v}$9; for continuous loss, the coupling is through sink matrices $\mathbf{L}$0. In both cases the modes evolve autonomously as coupled two-dimensional $\mathbf{L}$1 equations between couplings, so restoring inclination as a dynamical variable does not forfeit analytic tractability.

Limitations and open questions

The quantitative claims rest on several stated assumptions: exact isotropy of the field-star distribution (which produces the block-diagonal tensor and would fail in flattened nuclei or with VRR-driven transport beyond the $\mathbf{L}$2 parametrization); fast apsidal phase mixing relative to collisional evolution; and the marginally bound limit $\mathbf{L}$3 for the capture boundary. The weak-dipole analytic solution is first order in $\mathbf{L}$4 and calibrated against a single representative boundary; strongly inclined or higher-multipole loss surfaces, and the intermediate-$\mathbf{L}$5 regime where the sink and phase-resolved prescriptions diverge, remain to be treated at higher order or numerically. Whether the Legendre hierarchy can be truncated at low $\mathbf{L}$6 for realistic Kerr disruption boundaries is not established in this paper.

Conclusion

The paper establishes that inclination diffusion operates on the same timescale as angular-momentum magnitude diffusion near relativistic loss surfaces and cannot be consistently suppressed or instantaneously mixed. For Kerr capture, both limiting prescriptions distort the prograde–retrograde contrast severely (angular errors up to 24% and 87%) while preserving the integrated flux to a few percent. The exact angular block in the diffusion tensor and the resulting Legendre-mode hierarchy make the extended problem tractable, as demonstrated by a closed-form weak-dipole TDE solution accurate to about 7%. Orientation-dependent relativistic observables therefore require orientation-resolved transport, even when integrated loss rates appear robust.