Kerr Mode Coupling and Tail Exponents

On a rotating black hole, an initially quadrupolar scalar pulse can generate a monopole. The monopole may dominate the late field even while a spherical quadrupole projection decays faster. Projecting that same field onto a fixed spheroidal shape can put the monopole back into the measured signal. Thus a tail exponent belongs to specified initial data, a field variable, a projection, and an observation limit—not to an angular label alone. We derive this distinction for a linear massless scalar on subextremal Kerr, then compare it with separated Teukolsky Green functions.

Required background. PDEs and Green Functions for Gravitation supplies retarded evolution, projections, and asymptotic estimates.

Helpful background. Black-Hole Geometry and Mechanics supplies Kerr and its horizon; Formulations of General Relativity distinguishes scalar and Teukolsky variables; Causality and Global Structure supplies the observation limits; Black-Hole Perturbation Theory supplies the separated equations.

Scalar coupling in the time domain

Let $\mu=GM$, $a=J/M$, $0\lt\lvert a\rvert\lt\mu$ and $r\gt r_+=\mu+\sqrt{\mu^2-a^2}$. First consider an axisymmetric scalar, $\partial_\phi\Phi=0$, in Boyer–Lindquist coordinates. Define

$$B=r^2+a^2,\qquad \Delta=r^2-2\mu r+a^2, \qquad \frac{dx}{dr}=\frac{B}{\Delta}.$$

Here $x=r_*$ is the tortoise coordinate, with an arbitrary fixed additive constant. Multiplying $\Box\Phi=0$ by $-\Sigma$, where $\Sigma=r^2+a^2\cos^2\theta$, gives

$$\left(\frac{B^2}{\Delta}-a^2\sin^2\theta\right)\partial_t^2\Phi -\partial_r(\Delta\partial_r\Phi)-\Delta_{\mathbb S^2}\Phi=0.$$

The angular dependence multiplying the acceleration produces spherical-mode coupling. It is present in the linear equation; it is not nonlinear coupling between perturbations. See Barack and Ori 1999, Eqs. (1)–(3), preprint p. 2, PDF, with their geometrized mass replaced by $\mu$.

Use real, unit-normalized harmonics $Y_{j0}=\sqrt{(2j+1)/(4\pi)}P_j(\cos\theta)$ in the even sector $j=0,2,4,\ldots$. Expand $\Psi=\sqrt B\,\Phi=\sum_j\psi_j(t,x)Y_{j0}$, and define

$$C_{jk}=\int_{\mathbb S^2}Y_{j0}\cos^2\theta\,Y_{k0}\,d\Omega, \qquad \Lambda_{jk}=j(j+1)\delta_{jk},\qquad \beta=\frac{\Delta}{B^2}.$$

The radial system is

$$\begin{aligned} \mathsf M(x)\,\partial_t^2\boldsymbol\psi &=\partial_x^2\boldsymbol\psi-(U I+\beta\Lambda)\boldsymbol\psi,\\ \mathsf M&=I-a^2\beta(I-C),\\ U&=\frac{(\sqrt B)_{xx}}{\sqrt B} =\frac{\Delta(2\mu r^3+a^2r^2-4\mu a^2r+a^4)}{B^4}. \end{aligned}$$

Multiplication by $\cos\theta$ changes $j$ by one, so $C$ connects $j$ to $j$ and $j\pm2$. In particular, $C_{02}=2/(3\sqrt5)\ne0$: a quadrupole can excite a monopole while preserving axisymmetry and reflection parity. The positive kinetic matrix $\mathsf M$ must be inverted as a whole to obtain the acceleration. Its quadratic form lies between $(1-a^2\beta)\|\boldsymbol z\|^2$ and $\|\boldsymbol z\|^2$. At $a=0$, independent scalar Schwarzschild equations return, with potential $(1-2\mu/r)[2\mu/r^3+j(j+1)/r^2]$.

The infinite angular system is exact; truncating it is an approximation. When forming $C$ at a finite $j_{\max}$, retain the intermediate $j_{\max}+1$ path before projecting back. Squaring an already truncated $\cos\theta$ matrix loses a real contribution.

Three meanings of an angular mode

The spherical coefficient $\psi_j(t,x)$ is an instantaneous projection onto a fixed function. A scalar spheroidal harmonic instead solves

$$[-\Delta_{\mathbb S^2}-c^2\cos^2\theta]S_{L0}(c) =A_{L0}(c)S_{L0}(c),\qquad c=a\omega,$$

with unit norm and continuation from $S_{L0}(0)=Y_{L0}$. For real $c$, choose its phase continuously and write $S_{L0}(c)=\sum_j b_{jL}(c)Y_{j0}$, where $\sum_j b_{jL}^2=1$. First-order perturbation in $c^2$ gives

$$S_{20}(c) =Y_{20}-\frac{c^2}{9\sqrt5}Y_{00} +\frac{2c^2}{49\sqrt5}Y_{40}+O(c^4).$$

For example, the monopole coefficient is $-c^2C_{02}/[2(2+1)-0]$. The Spin-Weighted Spheroidal Harmonics page develops the general spin-weighted problem.

Choose a fixed real number $c_0$. The time-domain projection

$$\chi_L(t,x;c_0) =\int_{\mathbb S^2}\Psi(t,x,\theta)S_{L0}(c_0,\theta)\,d\Omega =\sum_j b_{jL}(c_0)\psi_j(t,x)$$

is a fixed angular filter, not separation of a broadband evolution. If $\psi_0\sim A_0t^{-3}$ and $\psi_2\sim A_2t^{-5}$, with $A_0b_{02}(c_0)\ne0$, then $\chi_2\sim b_{02}(c_0)A_0t^{-3}$. One solution has different projected powers because one projection contains a monopole.

Angular prescription	What is fixed?	What it measures
Spherical $\psi_j(t,x)$	$Y_{j0}$	One spherical coefficient at each time
Fixed-shape $\chi_L(t,x;c_0)$	$S_{L0}(c_0)$	A fixed mixture of spherical coefficients
Separated frequency mode	$L,m$ while $S_{Lm}(a\omega)$ varies inside the integral	A separated resolvent channel before reconstruction

In the last row, the scalar $L=0$ channel has a $Y_{20}$ admixture of order $(a\omega)^2$. Its radial jump starts at frequency power two; extracting that admixture produces power four and hence a $t^{-5}$ contribution when its zero-frequency source coupling is nonzero. A source-side projection can supply additional powers or cancel this term. A fixed filter instead supplies a constant coefficient and adds no frequency power. See Branch-Cut Contributions to Retarded Response for the endpoint inversion and the distinction between a resolvent and its jump.

Initial data determine which leading terms survive

Let $\boldsymbol h=\boldsymbol\psi(0,x)$, $\boldsymbol g=\partial_t\boldsymbol\psi(0,x)$, and $\mathsf K=-\partial_x^2+U I+\beta\Lambda$. The Laplace-transformed Cauchy problem is

$$(\zeta^2\mathsf M+\mathsf K)\widehat{\boldsymbol\psi} =\mathsf M\boldsymbol g+\zeta\mathsf M\boldsymbol h, \qquad \operatorname{Re}\zeta\gt0.$$

Here $\zeta$ denotes the Laplace variable; $s$ below denotes spin weight. Even pure quadrupolar velocity has a monopole in its initial source: $(\mathsf M\boldsymbol g)_0=a^2\beta C_{02}g_2$. It is not the same preparation as pure quadrupolar canonical momentum $\boldsymbol\Pi=\mathsf M\boldsymbol g$.

For smooth axisymmetric data compactly supported away from both ends of a Boyer–Lindquist $t=0$ slice, the time-inverted monopole Newman–Penrose charge is

$$\mathcal I=\frac{\mu}{4\pi} \int_{r_+}^{\infty}\!\!dr\int_{\mathbb S^2}\!\!d\Omega\, \left(\frac{B^2}{\Delta}-a^2\sin^2\theta\right) \partial_t\Phi(0,r,\theta).$$

This is a conserved asymptotic charge of a time integral of $\Phi$, not an initial spherical monopole coefficient. It follows from the initial-data formula in Angelopoulos, Aretakis, and Gajic 2021, § 1.1.2, p. 6, and Corollary 9.12, p. 77, PDF. The horizon term is zero here, and $n(\Phi)d\mu_\Sigma$ gives the displayed weighted velocity measure. Here $n$ is the future unit normal and $d\mu_\Sigma$ the induced volume measure on the initial slice. Compact support and finite propagation let the slice be bent to their horizon-crossing, hyperboloidal form through regions where the solution remains zero, while agreeing with the original slice on the support. An arbitrary later slice through an initially empty region would not justify this transfer.

For $\Phi(0)=0$ and $\partial_t\Phi(0,r,\theta)=F(r)Y_{20}(\theta)$, angular integration gives

$$\mathcal I=\frac{\mu a^2}{3\sqrt{5\pi}}\int_{r_+}^{\infty}F(r)\,dr.$$

Thus nonnegative, nonzero compact $F$ gives $\mathcal I\ne0$. The compact-data scalar asymptotics then have a leading monopole of order $t^{-3}$ and a higher-angular sector of order $t^{-5}$, led by $Y_{20}$. At fixed $r$, rescaling by $\sqrt B$ changes amplitudes but not powers. The radiation field $\lim_{r\to\infty}r\Phi$, at fixed retarded time $u=t-r_*$, has a leading monopole of order $u^{-2}$ and higher-angular terms of order $u^{-4}$; the latter coefficients can involve additional charges and cancellations. These statements use Angelopoulos, Aretakis, and Gajic 2021, Theorem 1.1, pp. 4–5, and § 1.1.4, pp. 7–8, PDF, for smooth compact scalar data on subextremal Kerr. They are not a theorem for all spin weights or every individual higher multipole.

If instead $\partial_t\Phi(0)=0$ with compact initial displacement, then $\mathcal I=0$: these particular leading terms vanish. An allowed coupling channel alone does not guarantee a nonzero coefficient; radial cancellations can also make the integral zero. The next nonzero power needs a separate calculation.

Barack and Ori’s older generic hierarchy has its own preparation: a short outgoing characteristic pulse initially far out, generic low-multipole content, and no initial static multipole. Its extracted spherical mode has finite-radius exponent

$$p_{jm}=j+\lvert m\rvert+3+\epsilon_{jm},\qquad \epsilon_{jm}= \begin{cases} 0,&j+m\text{ even},\\ 1,&j+m\text{ odd}. \end{cases}$$

Their pure-quadrupole discussion directly identifies an induced monopole at null infinity; it does not make arbitrary compact Cauchy data generic. See Barack and Ori 1999, data on p. 2 and Eqs. (9), (15)–(16) on pp. 3–4, PDF.

Separated Green functions and observation limits

Holding a spheroidal label fixed inside the frequency integral is useful even though it is not an instantaneous fixed-shape projection. For a scalar, the small-frequency radial jump of a separated $(L,m)$ channel scales as $\sigma^{2L+2}$. Inversion gives $t^{-2L-3}$ at finite source and field radii. The first logarithmic correction is of order $t^{-2L-5}\log(t/\mu)$, two inverse powers later. See Casals, Kavanagh, and Ottewill 2016, § VIII, p. 20, and § IX, p. 22, PDF.

For the integer-spin extension, let $\mathcal G_{Lm}$ denote the frequency-integrated contribution

$$\mathcal G_{Lm} =\int_{\mathbb R}d\omega\,e^{-i\omega t} g_{Lm}(r,r';\omega)\, {}_sS_{Lm}(\theta,a\omega)\,{}_sS_{Lm}(\theta',a\omega).$$

Here $g_{Lm}$ is the radial Teukolsky Green function; on the real axis the angular normalization is $2\pi\int_0^\pi{}_sS_{Lm}^2\sin\theta\,d\theta=1$. The retarded prescription is understood. This follows Casals et al. 2026, Eqs. (2.1)–(2.9), p. 4, PDF: the full Green function also includes $-2\Delta(r')^s e^{im\phi}$ and the angular sums. Continue the angular product analytically; inserting pointwise complex conjugation at complex $\omega$ would destroy analyticity. See Casals et al. 2026, § II D, p. 7, PDF.

Let $u=t-r_*$, $v=t+r_*$ and $\Omega_H=a/(r_+^2+a^2)$. A regular ingoing azimuth is $\widetilde\phi=\phi+\int a\,dr/\Delta$. The corotating label $\varphi_H=\widetilde\phi-\Omega_Hv$ is constant along a horizon generator. It differs from the source’s $\phi-\Omega_Ht$ by a finite radial function, changing a constant phase but not the late-$v$ power. For the branch-cut contribution $\delta\mathcal G$, the generic nonzero leading terms are:

Observation limit	Normalized quantity	Time dependence
Fixed $r,r'\gt r_+$	$\delta\mathcal G_{Lm}$	$t^{-2L-3}$
Future horizon, fixed $\varphi_H$	$\Delta(r)^s e^{im\phi}\delta\mathcal G_{Lm}$	$e^{im\Omega_Hv}v^{-2L-3-b}$
Future null infinity, fixed angles	$r^{2s+1}\delta\mathcal G_{Lm}$	$u^{-L+s-2}$

Here $b=1$ for $s\gt0,m=0$ and $b=0$ otherwise; the source radius $r'\gt r_+$ remains finite in every row. These are the quantities in Casals et al. 2026, Eqs. (7.2), (7.6), and (7.8), pp. 20 and 22–23, PDF. The horizon phase describes sampling along generators, not instability. A zero leading angular, radial, source or observation coefficient requires the next term. The boundary rescalings are essential: none of these rows is an unqualified power for strain, a metric component or raw $\Psi_4$. Extremal Kerr is outside this comparison.

Testing the projected powers

Fix both data functions, the slice, and the angular filter before comparing numerical powers. A finite calculation of the scalar system above uses $a/\mu=0.7$, $c_0=0.3$, $R=r/\mu$, $X=x/\mu$ and $T=t/\mu$. For this calculation the tortoise origin is specified by

$$\begin{aligned} X={}&R+\frac{2R_+}{R_+-R_-}\log\frac{R-R_+}{2}\\ &-\frac{2R_-}{R_+-R_-}\log\frac{R-R_-}{2}, \qquad R_\pm=1\pm\sqrt{1-0.7^2}. \end{aligned}$$

Both preparations have zero initial displacement. The coordinate velocity of $q=\sqrt{R^2+0.7^2}\,\Phi$ is $b(X)Z/N_Z$, with $Z=Y_{20}$ or $Z=S_{20}(0.3)$, and

$$b(X)= \begin{cases} \displaystyle\exp\!\left[-\left(\frac{X-40}{6}\right)^2 +1-\frac{1}{1-((X-40)/36)^2}\right], &\lvert X-40\rvert\lt36,\\ 0,&\text{otherwise}. \end{cases}$$

If $\boldsymbol z$ is the spherical-coefficient vector of $Z$, choose $N_Z^2=\int b^2\boldsymbol z^T\mathsf M\boldsymbol z\,dX$; each preparation then has initial reduced energy $1/2$. Calling $c_0/a$ a basis frequency does not make either pulse monochromatic. Both solutions are extracted at $R=10,30$ and projected onto both bases.

For a signed signal $f(T)$, define

$$p_{\mathrm{local}}(T)=-\frac{T f'(T)}{f(T)}.$$

It is undefined at a zero. Its numerical sensitivity also depends on the derivative, not just on the field. Suppose the computed field and derivative are $f+\delta f$ and $f'+\delta v$. Their inferred local power $\widetilde p$ differs from the exact local power $p$ by the identity

$$\widetilde p-p =-T\frac{f\,\delta v-f'\,\delta f}{f(f+\delta f)}.$$

Both denominators must be nonzero. A small relative field error does not control the term $T\delta v/f$: two waveforms can look alike while their local powers disagree strongly. Compare the signed fields and the evolved time derivatives separately, before interpreting the ratio. Differences between numerical runs measure sensitivity; they are not automatically bounds on either error.

A different, finite-window diagnostic is

$$p_{[T_1,T_2]}= -\frac{\log\lvert f(T_2)/f(T_1)\rvert}{\log(T_2/T_1)}.$$

Use it only for a resolved signal with no zero crossing in the window. It is a logarithmic average of the local power, not its value at the midpoint and not an asymptotic theorem.

The calculation uses even degrees through $j_{\max}=6$, fourth-order centered interior radial differences and second-order velocity-Verlet time stepping, in binary64 arithmetic. The fine radial run has $\Delta X=0.125$, $\Delta T=0.025$, $-650\le X\le700$ and $0\le T\le1000$. The table reports its finite-window slopes on $[700,1000]$. Rescaling between $q$ and $\Phi$ at fixed radius leaves these slopes unchanged. A slope is withheld if the unit-spaced samples cross zero or fall below $\max(10^{-18},10^{-12}\max_T\lvert f(T)\rvert)$. This censoring rule is a numerical diagnostic, not a proved error estimate.

Initial velocity shape	$R$	Spherical $j=0$	Spherical $j=2$	Fixed-shape $L=2$
$Y_{20}$	10	3.02	7.02	Zero crossing; no slope assigned
$Y_{20}$	30	3.02	7.04	7.95
$S_{20}(0.3)$	10	3.02	7.01	3.09
$S_{20}(0.3)$	30	3.03	7.03	3.62

The monopole trends toward power three. The spherical quadrupole is still dominated by a faster transient, while the fixed filter can combine terms of opposite sign. In particular, 7.95 is a finite-time cancellation effect, not a new universal exponent; 3.62 is not yet the limiting value three.

Independent radial spacings $0.5,0.25,0.125$ at fixed $\Delta T=0.025$ give observed order 4.018 in the joint spherical $j=0,2$ waveform norm over $[0,200]$. Independent time steps $0.05,0.025,0.0125$ at fixed $\Delta X=0.25$ give order 2.000 in that same interval. The radial order falls to 3.379 over $[300,500]$: small late signals must not inherit the early-time convergence claim automatically. Changing $j_{\max}$ to 8 and enlarging the domain to $[-750,800]$ provide separate checks. Across the radial, time and angular comparisons, the largest componentwise relative sampled-$L^2$ waveform difference over $[500,1000]$ is $3.1\times10^{-5}$; the domain comparison gives identical sampled fields. These are differences between finite runs, not rigorous error bounds on the continuum tail or on the logarithmic derivative.

In the $a=0$ control, a pure spherical quadrupole generates no other spherical modes; its two finite-window slopes are 7.02 and 7.04, consistent with the compact-velocity Schwarzschild power seven. The ultimate spherical-quadrupole $t^{-5}$ tail is not resolved by these runs. The nominal machine epsilon is $2.22\times10^{-16}$; an independent higher-precision long-time evolution is still needed. Signals near the numerical floor and sign crossings are retained rather than fitted. A plateau in one discretization is not a measured asymptotic tail. The horizon and null-infinity rows above remain analytic comparisons, not outputs of this finite-radius grid.

Tail results and open questions through 2026-09-05

The evidence cutoff is 2026-09-05. A component-independent exponent without a data and observation prescription is not a well-posed universal claim. Research questions concern specified problems, especially higher-spin data reconstruction and controlled remainders. In particular, distinguish a disagreement about the same Cauchy problem from a change of angular projection or observation limit.

Source	Evidence	Boundary of the conclusion
Barack–Ori 1999	Scalar spherical coupling and three observation limits for a stated outgoing-pulse setup	Generic angular content is not every pure Cauchy preparation; convergence of the full expansion was not proved there
Casals–Kavanagh–Ottewill 2016	Scalar separated Green-function powers and high-order logarithms, with numerical frequency-integral comparisons	A Cauchy response still needs source weighting and angular reconstruction
Angelopoulos–Aretakis–Gajic 2021, published 2023	Rigorous compact-data scalar asymptotics for the full field and low angular projections	Vanishing charges remove leading terms; this is not a Maxwell or gravitational theorem
Casals–Kavanagh–Neef–Ottewill 2026	Integer-spin separated Green-function powers, explicit gravitational coefficients and selected numerical checks	July 2026 preprint; source convolution, component conversion and independent replication remain separate tasks

The selected 2021 preprint has inconsistent quadrupole prefactors between its local and global asymptotic formulas; compare Angelopoulos, Aretakis, and Gajic 2021, Theorem 1.1, p. 4, and Proposition 12.7, p. 95, PDF. The powers and initial-charge condition used here agree. Whether the published journal version corrects this discrepancy has not been verified here. A numerical prefactor should not be imported from the introductory theorem without checking the detailed expression and the field equation.

The conclusions could change with a generic-data gravitational tail theorem, a converged high-dynamic-range evolution, a correction to a branch coefficient or theorem prefactor, or a result changing the slicing or extraction prescription. Full nonlinear decay belongs to Mathematical Relativity; observational fitting additionally requires a source and detector model.

Exercises

1. Which pure quadrupole? Compare compact axisymmetric data with (a) zero displacement and positive quadrupolar coordinate velocity, (b) quadrupolar displacement and zero coordinate velocity, and (c) zero displacement and pure quadrupolar canonical momentum. Must all three have a nonzero leading monopole tail?

Solution

No. In (a), the weighted angular integral gives $\mathcal I=\mu a^2(3\sqrt{5\pi})^{-1}\int F(r)\,dr\ne0$. In (b), the compact-data charge vanishes because its velocity and horizon terms vanish. In (c), $\mathsf M\boldsymbol g$ has no monopole, so its angular integral in the charge is zero. Inverting $\mathsf M$ means that (c) does not have pure quadrupolar coordinate velocity. Vanishing charge removes the stated leading term without determining the next power.

2. A slowly decaying angular filter. Suppose $\psi_0=A_0t^{-3}+o(t^{-3})$, $\psi_2=A_2t^{-5}+o(t^{-5})$, and higher modes are smaller. Why can $\chi_2(t;c_0)$ decay as $t^{-3}$ without contradicting a fixed-separated-$L=2$ Green-function term of order $t^{-7}$?

Solution

For fixed small nonzero $c_0$, $b_{02}(c_0)=-c_0^2/(9\sqrt5)+O(c_0^4)$, so $\chi_2=b_{02}(c_0)A_0t^{-3}+o(t^{-3})$ when this coefficient is nonzero. The separated calculation instead uses $S_{20}(a\omega)$ at each frequency before integration and convolution with the initial source. Its label is not the fixed filter. At $c_0=0$ exactly, $\chi_2=\psi_2$; the onset of monopole dominance is not uniform as $c_0\to0$.

3. A finite-time slope is not an exponent theorem. Let $f(T)=A T^{-p}(1+\alpha/T)$, with $A\ne0$. Find the local power and explain why one apparent plateau is insufficient.

Solution

Away from $T+\alpha=0$,

$$p_{\mathrm{local}}(T)=p+\frac{\alpha}{T+\alpha}.$$

The slope drifts toward $p$, and a zero creates a singular spike. Numerical errors divided by a tiny signal can produce another apparent plateau. Compare signed waveforms and slopes across resolution, precision, boundary and time-window changes.

References

• Angelopoulos, Yannis, Stefanos Aretakis, and Dejan Gajic. “Late-Time Tails and Mode Coupling of Linear Waves on Kerr Spacetimes.” Advances in Mathematics 417 (2023): 108939. DOI. Open PDF: arXiv:2102.11884v1 (2021).
• Barack, Leor, and Amos Ori. “Late-Time Decay of Scalar Perturbations Outside Rotating Black Holes.” Physical Review Letters 82 (1999): 4388–4391. DOI. Open PDF: arXiv:gr-qc/9902082v2.
• Casals, Marc, Chris Kavanagh, and Adrian C. Ottewill. “High-Order Tail in Kerr Spacetime.” Physical Review D 94 (2016): 124053. DOI. Open PDF: arXiv:1608.05392v1.
• Casals, Marc, Chris Kavanagh, Jakob Neef, and Adrian Ottewill. “High-Order Gravitational Late-Time Tails in Kerr Spacetime.” arXiv:2607.12151v1, 13 July 2026. Preprint, Open PDF.