Late-Time Tails, Branch Cuts, and Spectral Representations

Why does a black-hole signal continue after its readily visible ringing has faded? Curvature backscatter and spectral thresholds can produce a tail that outlives each isolated damped resonance. Changing the outer boundary can instead give exponential decay, a stationary remainder, or long-lived trapped waves. This chapter connects these possibilities to retarded response. Every decay statement specifies the field and angular projection, initial data, observation region, and time coordinate. The focus is linear response; nonlinear stability and detector inference require additional arguments.

Helpful background. Black-Hole Geometry and Mechanics supplies the Schwarzschild, Kerr, de Sitter, and anti-de Sitter exteriors; PDEs and Green Functions for Gravitation supplies retarded resolvents and endpoint asymptotics; Formulations of General Relativity identifies the evolved field component; and Causality and Global Structure distinguishes fixed radius, horizons, and null infinity.

The spectral question behind a late-time signal

Fix a stationary background, an angular projection, initial data, and an observation radius. Write the resulting field component as

$$X(t,r)=\frac{1}{2\pi} \int_{\mathbb R+i\gamma}e^{-i\omega t} \widetilde X(\omega,r)\,d\omega .$$

Here $\widetilde X$ includes the Green function’s contraction with the data, and $\gamma$ lies above the response’s growth bound. Lowering the contour requires analytic continuation and estimates on the remaining contour. The factor $e^{-i\omega t}$ alone does not make the large-frequency arc vanish at every positive time: the radial solutions and source transform can grow on that arc. Prompt propagation, pole residues, cuts, and contour remainders must be kept until their omission is justified. Compare the existing contour diagram with Leaver’s explicit Schwarzschild decomposition (Leaver 1986, § II B, author PDF pp. 7–8, Eqs. 22–24).

Analytic structure	Typical inverse-transform signature	Information still required
Simple pole $\omega_j$	$e^{-i\omega_j t}$	Excitation coefficient, component, and observation region
Zero-frequency cut with density proportional to $\sigma^q$	$t^{-(q+1)}$ under the endpoint bounds below	Nonzero coefficient, data moment, and observation limit
Thresholds $\omega=\pm\mathsf m$	Oscillation near $\mathsf m$ with an inverse-power envelope	Intermediate or ultimate regime and fixed-harmonic hypotheses
Simple zero-frequency pole	Constant or stationary contribution	Projection removed or retained; higher-order poles can give polynomial factors
Resonances with damping tending to zero	Regularity-dependent slow summed decay	Boundary condition, trapping, and angular regularity

This table does not determine the answer from singularity locations alone. A residue can vanish for the chosen data; a cut can dominate every isolated damped pole; and an infinite angular sum needs estimates uniform in its angular label. An upper bound on a norm is also different from a nonzero leading asymptotic term for a particular field.

Route through the six calculations

For an analytic route, read Price’s law and the branch-cut calculation together, then change either the angular basis on Kerr or the mass threshold. For a boundary-comparison route, first work through the endpoint example below, then compare the de Sitter and AdS pages. Return to the first two pages to explain why their conclusions differ from asymptotically flat decay.

Page	Central question	Output to carry forward
Price-Law Tails in Schwarzschild	Why does curvature backscatter outlive isolated quasinormal modes?	Fixed-radius, horizon, and null-infinity powers for declared scalar data
Branch-Cut Contributions to Retarded Response	How does a frequency-domain discontinuity become a time-domain power?	Cut orientation, endpoint coefficient, and Watson-lemma power counting
Kerr Mode Coupling and Tail Exponents	Which angular mode controls a rotating tail?	Separate spherical projections, spheroidal Green modes, spin, data, and location
Massive Fields and Oscillatory Tails	What changes when propagation has thresholds at $\omega=\pm\mathsf m$?	Intermediate oscillations, quasi-bound poles, and the precise very-late theorem
de Sitter Exponential Decay and Spectral Gaps	Which resonance controls approach to the de Sitter late-time state?	Two-horizon short-range scattering, a nonzero gap, and the scalar zero mode
AdS Recurrence, Reflecting Boundaries, and Tails	How does timelike infinity change the spectral problem?	Boundary-dependent normal modes, horizon damping, stable trapping, and nonlinear seams

Check your preparation with two questions: does taking $r\to\infty$ at fixed $t$ follow an outgoing light ray, and can a list of poles determine a response without its excitation coefficients? If either answer is unclear, review null infinity and source-dependent excitation. The first requires fixed retarded time; the second requires the data and the non-pole pieces as well as the poles.

From a branch endpoint to a time-domain power

Place a downward cut at $\omega=-i\sigma$, $\sigma\gt0$, and define its east-minus-west discontinuity. At the fixed observation point, suppress $r$ and write

$$\begin{aligned} \Delta\widetilde X(-i\sigma) &=\widetilde X(0^+-i\sigma)-\widetilde X(0^--i\sigma),\\ \rho(\sigma)&=\frac{\Delta\widetilde X(-i\sigma)}{2\pi i},\\ X_{\mathrm{cut}}(t)&=\int_0^\infty e^{-\sigma t}\rho(\sigma)\,d\sigma . \end{aligned}$$

The factor $1/(2\pi i)=-i/(2\pi)$ fixes the orientation, not just the power. Equivalently, for the Laplace variable $s=-i\omega$, these are the lower minus upper banks at $s=-\sigma$. We keep $\omega$ as the argument of $\widetilde X$ throughout. The density includes the chosen data and need not be positive.

Here is a sufficient endpoint estimate. Suppose that, for some $q\gt-1$, $\delta\gt0$, and $\tau_b\gt0$,

$$\begin{gathered} \rho(\sigma)=C\sigma^q+O(\sigma^{q+\delta}) \quad(\sigma\to0^+),\\ \int_0^\infty e^{-\tau_b\sigma}\lvert\rho(\sigma)\rvert\,d\sigma\lt\infty . \end{gathered}$$

Split the integral at a small fixed $\sigma_0\gt0$. On its first part, set $y=\sigma t$. The leading monomial has the integral

$$\int_0^\infty e^{-\sigma t}\sigma^q\,d\sigma =t^{-q-1}\int_0^\infty e^{-y}y^q\,dy =\Gamma(q+1)t^{-q-1}.$$

The local remainder is bounded by a constant times $\Gamma(q+\delta+1)t^{-q-\delta-1}$. On the second part, extract $e^{-(t-\tau_b)\sigma_0}$ and use the weighted absolute integral. Extending the leading monomial to infinity also adds only an exponentially small term. Thus

$$X_{\mathrm{cut}}(t) =C\Gamma(q+1)t^{-q-1}+O(t^{-q-\delta-1}).$$

The condition $q\gt-1$ ensures integrability at the endpoint. The constants carry the necessary dimensions: $\rho$ has units of field times time and $C$ has units of field times time to the power $q+1$. A nonzero $C$ gives the stated leading power. If $C=0$, only the remainder bound survives; the next nonzero endpoint term must be found. The estimate concerns this cut piece, not every other term in the inverse transform.

An exact example of cancellation

Use dimensionless $T=t/L$ and $z=L\sigma$, with a fixed time scale $L\gt0$. Two elementary densities and their responses are

$$\rho_a(z)=z^2e^{-az},\qquad h_a(T)=\int_0^\infty e^{-zT}\rho_a(z)\,dz =\frac{2}{(T+a)^3},\qquad a=1,2.$$

These are dimensionless densities. With the dimensionless Laplace variable $\zeta=-iL\omega$, their frequency-domain realization is the Stieltjes transform

$$H_a(\zeta)=\int_0^\infty\frac{\rho_a(y)}{\zeta+y}\,dy, \qquad \operatorname{Re}\zeta\gt0.$$

Indeed, integrating $e^{-\zeta T}h_a(T)$ first in $T$ gives this expression. Continuing to $\zeta=-z$, the distributional identity

$$\frac{1}{y-z-i0}-\frac{1}{y-z+i0}=2\pi i\delta(y-z)$$

gives the lower-minus-upper jump $2\pi i\rho_a(z)$. Thus the toy has the same normalized cut construction. Both densities begin with $z^2$, so both responses begin with $2T^{-3}$. Their difference has a different leading term:

$$\begin{aligned} \rho_1(z)-\rho_2(z)&=z^3+O(z^4),\\ h_1(T)-h_2(T) &=\frac{2}{(T+1)^3}-\frac{2}{(T+2)^3} =6T^{-4}+O(T^{-5}). \end{aligned}$$

Nothing moved the branch endpoint. Cancellation removed its leading coefficient. This is an elementary Laplace-transform model, not a Schwarzschild evolution: it illustrates what cancellation does after the physical low-frequency coefficient has been derived. In a linear field problem that coefficient is a functional of the initial data, so special data can remove a leading tail.

Why the observation limit changes Price’s law

For a massless scalar on Schwarzschild, matching the long-range radial solutions and contracting with generic compact data gives an endpoint power $q=2\ell+2$ at fixed exterior radius. The preceding integral then explains the power $2\ell+3$; it does not supply that radial matching calculation. The radiation field instead takes the limit of $r\Phi_\ell$ at fixed retarded time $u=t-r_*$ before letting $u$ grow. That limit is not obtained by sending $r$ to infinity in a fixed-radius late-time remainder estimate.

For a fixed spherical harmonic of a massless scalar with smooth compactly supported Cauchy data, the generic Schwarzschild comparison is

Observation	Field quantity	Leading power when its coefficient is nonzero
Fixed exterior areal radius	$\Phi_\ell(t,r)$	$t^{-(2\ell+3)}$
Future null infinity	$\lim_{r\to\infty}r\Phi_\ell(u,r)$ at fixed $u$	$u^{-(\ell+2)}$

For $\ell\ge1$, precise asymptotic theorems identify the relevant data functional as a time-inverted Newman–Penrose charge: an asymptotic conserved quantity of a time integral of the solution (Angelopoulos, Aretakis, and Gajic 2023, Theorems 2.1 and 2.3, preprint PDF pp. 11–12, Eqs. 2.2 and 2.5). The monopole gives $t^{-3}$ and $u^{-2}$ under the corresponding nonvanishing condition (Angelopoulos, Aretakis, and Gajic 2018, Theorem 1 and Remark 1.4, preprint PDF pp. 8–10). The Price-law page explains the radial matching and data distinctions. Compact support does not by itself imply a nonzero leading coefficient. Noncompact initially static exterior data are also different from compact, momentarily stationary data.

Kerr preserves stationarity and axisymmetry but not each spherical $\ell$. Low-frequency spheroidal harmonics approach spherical harmonics with frequency-dependent mixing. Therefore one must say whether the reported quantity is a fixed spheroidal Green-function mode, a spherical projection of the evolved field, or the full field after lower allowed modes have been excited. Spin weight, azimuthal number, parity, special data cancellations, and the horizon, fixed-radius, or null-infinity limit all affect the answer.

Thresholds and boundary conditions

A scalar mass $\mathsf m\gt0$ introduces thresholds at $\omega=\pm\mathsf m$. An endpoint’s location supplies a carrier frequency, but not its envelope exponent: long-range scattering also affects the threshold phase. The massive-field calculation distinguishes intermediate behavior, quasi-bound poles, and the very-late $t_*^{-5/6}$ oscillatory profile. On a fixed subextremal Schwarzschild or Reissner–Nordström exterior, the latter theorem applies to smooth compact data in a fixed finite set of angular harmonics. It is uniform on bounded radial regions including the horizon, using a horizon-regular time $t_*$; its leading data functional can vanish. Its error is not controlled uniformly in an unrestricted angular sum, and the theorem is not a Kerr or nonlinear result (Pasqualotto, Shlapentokh-Rothman, and Van de Moortel 2024, Theorem I and Remarks 1.1–1.5, PDF p. 2).

Changing the asymptotics changes the analytic problem more radically. For a massless minimally coupled scalar on a fixed nondegenerate Schwarzschild–de Sitter exterior, both radial ends are horizons and the potential is exponentially short-range. For smooth compact data, the field approaches a data-dependent spatial constant locally, with an exponentially decaying remainder. Only the monopole has this zero-resonance contribution; its coefficient can vanish. The higher angular sectors have no such constant. The rigorous statement is a cutoff-energy estimate with additional regularity; local pointwise decay follows by commuting derivatives and Sobolev embedding (Bony and Häfner 2008, Proposition 1.2, Theorem 1.3 and Remark 1.4, preprint PDF pp. 4–6). This does not assert a decay rate uniform as the cosmological constant goes to zero or the two horizons coalesce. Nor does the scalar result establish the same zero mode for a massive field or gravitational master variable.

In AdS, timelike infinity requires a boundary condition. For example, a spherical massless scalar on pure global AdS$_4$, regular at the center with standard reflecting falloff, has linear frequencies

$$\omega_n L=3+2n,\qquad n=0,1,2,\ldots .$$

Here $L$ is the AdS radius and time is physical time. Its real linear field returns after $2\pi L$ in a norm in which the mode sum converges: each cosine and sine has completed an integer number of cycles (Bizoń and Rostworowski 2011, PDF p. 3, Eqs. 11–13). Equal frequency spacing more generally guarantees only common-phase alignment of the positive-frequency part, not exact return of an arbitrary real field. Different masses or boundary conditions need separate checks.

In a stable AdS black-hole sector the horizon absorbs energy, so damped resonances replace undamped normal modes; exact recurrence is generally lost. High-angular-momentum trapping can nevertheless make decay very slow. For a concrete upper-bound example, the massless Klein–Gordon equation on four-dimensional Kerr–AdS with $|a|\lt L$ and $r_+^2\gt|a|L$, homogeneous fast reflecting falloff and finite higher-order data energy satisfies a logarithmic decay estimate in the specified weighted energy norm (Holzegel and Smulevici 2013, Theorem 1.1, PDF p. 4). This is not an asymptotic equality for each individual solution. General angular sums, boundary conditions, and the nonlinear competition among collapse and time-dependent states are treated on the AdS comparison page.

How to state and test a tail law

Before writing $\Phi\sim t^{-p}$, record the following:

1. Field and component: scalar, electromagnetic, metric master field, or a normalized Teukolsky variable.
2. Angular label: fixed spherical multipole, fixed spheroidal mode, or the full field after mode coupling.
3. Data: compact or noncompact, generic or momentarily stationary, and which leading conserved or asymptotic moments vanish.
4. Observation: fixed area radius, future event horizon, future null infinity, or a threshold scaling region.
5. Coordinate: $t$, horizon-regular advanced time $v$, retarded time $u$, or a stated hyperboloidal time.
6. Regime: intermediate, asymptotic, near-extremal, small-mass, or small-black-hole approximation.
7. Boundary domain: outgoing, horizon regular, reflecting Dirichlet, alternative, or Robin.

For a black-hole numerical check, evolve at least two data profiles with the same declared asymptotic class, extract at two fixed radii and the relevant characteristic boundary, and refine both the grid and the fit window. Plot the local power

$$p_{\mathrm{loc}}(t) =-\frac{d\log(\lvert X(t)\rvert/X_0)}{d\log(t/L)}, \qquad X_0\gt0,\quad L\gt0,$$

where $X_0$ has the field’s units and $L$ has time units. These fixed reference scales do not affect the derivative. Evaluate it away from zero crossings. A plateau that shifts when the outer boundary, angular truncation, or fit window changes is not an asymptotic exponent. For oscillatory tails, fit the envelope and instantaneous phase separately.

Review the chapter

How late is a local-power plateau? For the exact functions $h_a(T)$ above, calculate the local power. How large must $T$ be for it to be within one percent of its limiting value? Explain why subtracting the two signals changes the limiting power even though each has the same endpoint location.

Solution

$$p_a(T)=-T\frac{h_a'(T)}{h_a(T)} =\frac{3T}{T+a},\qquad \frac{3-p_a(T)}{3}=\frac{a}{T+a}.$$

Thus one-percent accuracy requires $T\ge99a$: $99$ for $h_1$ and $198$ for $h_2$. A finite-time slope need not yet equal the final exponent. For the difference, expand $e^{-z}-e^{-2z}=z+O(z^2)$ at the endpoint; multiplication by $z^2$ raises the density power from two to three. The Gamma integral then gives $\Gamma(4)T^{-4}=6T^{-4}$, as the exact subtraction confirms. These numbers test the elementary model, not the onset time of a Schwarzschild tail.

What does a mass threshold determine? Write $\omega=\mathsf m-\varepsilon$ near the positive threshold. Which part of the inverse-transform phase fixes the carrier? When can the negative threshold be represented by the complex conjugate contribution? Does this argument determine the envelope power?

Solution

$$e^{-i\omega t}=e^{-i\mathsf m t}e^{i\varepsilon t}.$$

The first factor supplies the carrier. For a real field with compatible real initial data, the full opposite-frequency contributions pair by complex conjugation. A complex fixed-azimuthal projection need not pair with itself; its partner may have opposite azimuthal number. If the slowly varying complex amplitude is $A(t)$, a conjugate pair contributes $2\operatorname{Re}[e^{-i\mathsf m t}A(t)]$. Determining $A(t)$ still requires the threshold discontinuity, long-range phase, and any relevant saddle point. Moving the endpoint alone therefore fixes neither the intermediate nor the very-late envelope power.

Does equal spacing imply recurrence? Consider a linear spectrum $\omega_n=(\beta+2n)/L$ with $\beta\gt0$. Compare the positive-frequency mode factors after $\pi L$. For a real field with both $n=0$ and $n=1$ excited, explain why integer $\beta$ permits return after $2\pi L$, but irrational $\beta$ prevents a generic exact nonzero return time.

Solution

$$e^{-i\omega_n\pi L}=e^{-i\pi\beta}.$$

This common phase aligns the positive-frequency modes, but the negative-frequency modes acquire its conjugate. The real field is not generically unchanged. Returning the full field and its time derivative requires both excited mode phases to be one. For a return time $t_R$, write

$$\frac{\beta t_R}{L}=2\pi k_0,\qquad \frac{(\beta+2)t_R}{L}=2\pi k_1, \qquad k_0,k_1\in\mathbb Z.$$

Subtracting gives $t_R/L=\pi(k_1-k_0)$, hence $\beta(k_1-k_0)=2k_0$. Irrational $\beta$ allows only $t_R=0$. For integer $\beta$, $t_R=2\pi L$ makes every mode phase one. The spherical massless AdS$_4$ example has $\beta=3$. Approximate recurrence, single-mode periodicity, and nonlinear evolution are different questions.

After the analytic route, you should be able to turn a derived endpoint coefficient into a decay law and recognize data cancellations. After the boundary route, you should be able to distinguish a tail, a zero-mode remainder, a decay bound, and a recurrence. Continue to boundedness and decay for the estimates behind stability, or to nonflat boundary conditions to formulate the corresponding evolution problem precisely.

References

• Angelopoulos, Yannis, Stefanos Aretakis, and Dejan Gajic. “Late-Time Asymptotics for the Wave Equation on Spherically Symmetric, Stationary Spacetimes.” Advances in Mathematics 323 (2018): 529–621. arXiv. Open PDF.
• Angelopoulos, Yannis, Stefanos Aretakis, and Dejan Gajic. “Price’s Law and Precise Late-Time Asymptotics for Subextremal Reissner–Nordström Black Holes.” Annales Henri Poincaré 24 (2023): 3215–3287. DOI. Open PDF.
• Bizoń, Piotr, and Andrzej Rostworowski. “On Weakly Turbulent Instability of Anti-de Sitter Space.” Physical Review Letters 107 (2011): 031102. DOI. Open PDF.
• Bony, Jean-François, and Dietrich Häfner. “Decay and Non-Decay of the Local Energy for the Wave Equation on the De Sitter–Schwarzschild Metric.” Communications in Mathematical Physics 282 (2008): 697–719. DOI. Open PDF.
• Holzegel, Gustav, and Jacques Smulevici. “Decay Properties of Klein–Gordon Fields on Kerr–AdS Spacetimes.” Communications on Pure and Applied Mathematics 66 (2013): 1751–1802. DOI. Open PDF.
• Leaver, Edward W. “Spectral Decomposition of the Perturbation Response of the Schwarzschild Geometry.” Physical Review D 34 (1986): 384–408. DOI. Open PDF.
• Pasqualotto, Federico, Yakov Shlapentokh-Rothman, and Maxime Van de Moortel. “The Asymptotics of Massive Fields on Stationary Spherically Symmetric Black Holes for All Angular Momenta.” arXiv preprint arXiv:2303.17767 (2024), version 2, 5 January 2024. arXiv.