AdS Recurrence, Reflecting Boundaries, and Tails

Anti-de Sitter infinity is timelike, so evolution is not defined until one chooses a boundary condition there. Under the standard fast reflecting condition, a scalar on pure global AdS has a discrete real normal-mode spectrum. Its positive-frequency sector has exact common-phase rephasings; return of an arbitrary real field requires an additional phase condition. Adding a black-hole horizon changes the same cavity into a lossy one: frequencies become complex, stable trapping can produce long-lived perturbations at large angular momentum, and exact recurrence is generally lost. These facts do not select one universal “AdS tail.” The field, dimension, black-hole size and rotation, boundary condition, data regularity, observation, and linear or nonlinear regime must all be named.

Required background. PDEs and Green Functions for Gravitation supplies boundary-value problems, normal modes, resonances, and energy estimates.

Helpful background. Black-Hole Geometry and Mechanics supplies AdS horizons and surface gravity; Formulations of General Relativity separates fixed-field and backreacting systems; Causality and Global Structure explains timelike conformal infinity; and Black-Hole Perturbation Theory supplies quasinormal boundary conditions.

Timelike infinity is part of the evolution problem

In global AdS$_{d+1}$, one convenient form of the metric is

$$ds^2=\frac{L^2}{\cos^2\rho} \left(-dt^2+d\rho^2+\sin^2\rho\,d\Omega_{d-1}^2\right), \qquad 0\le\rho<\frac{\pi}{2},$$

where $t$ is dimensionless and physical time is $Lt$. For a scalar of mass $m$, the two asymptotic powers are

$$\Phi\sim A(t,\Omega)(\cos\rho)^{\Delta_-} +B(t,\Omega)(\cos\rho)^{\Delta_+}, \qquad \Delta_\pm=\frac d2\pm \sqrt{\frac{d^2}{4}+m^2L^2}.$$

Standard reflecting, or Dirichlet, quantization sets the slower coefficient $A$ to zero. In the Breitenlohner–Freedman window an alternative or a Robin condition may also be admissible. Those choices change the spectrum and can change stability; “AdS boundary conditions” is therefore not a complete assumption. Here, reflecting means that the chosen homogeneous condition makes the flux of the relevant linear energy through conformal infinity vanish.

For a scalar regular at the center of pure global AdS with standard quantization,

$$\omega_{n\ell}L=\Delta_++\ell+2n, \qquad n=0,1,2,\ldots .$$

For a superposition containing only positive frequencies, every mode at fixed $\ell$ acquires the same phase after a physical time $\pi L$; across all $\ell$, the common rephasing time is $2\pi L$. The positive-frequency field then returns up to $e^{-2\pi i\Delta_+}$, so its local quadratic observables invariant under a common complex phase recur. A general complex or real solution also contains negative frequencies, which acquire the opposite phase. Those two sectors need not rephase together. The total conserved linear energy is constant at every time and is not, by itself, evidence of recurrence. A real massless scalar, for which $\Delta_+=d$ is an integer, returns as a field after $2\pi L$. This is a linear spectral statement for pure AdS, not a prediction that arbitrary nonlinear or black-hole data must recur (Green et al. 2015, §§ I–II, pp. 1–7).

A horizon turns normal modes into damped resonances

For Schwarzschild–AdS$_{d+1}$, now let $t$ denote physical boundary time (the preceding global-AdS coordinate was dimensionless):

$$F(r)=1+\frac{r^2}{L^2}-\frac{\mu}{r^{d-2}}, \qquad ds^2=-Fdt^2+F^{-1}dr^2+r^2d\Omega_{d-1}^2.$$

The radial problem now combines an ingoing condition at the future horizon with the specified reflecting condition at infinity. The normal frequencies move to quasinormal frequencies

$$\omega_j=\Omega_j-i\Gamma_j, \qquad \Gamma_j>0$$

for a linearly stable sector. Horizon absorption makes mode amplitudes $e^{-\Gamma_j t}$, so unequal $\Gamma_j$ prevent an exact rephasing. A small black hole may produce long echo trains because a pulse crosses the cavity many times before appreciable absorption, but an echo is not an exact recurrence.

At high angular momentum, reflection at the AdS boundary and the centrifugal barrier can keep a wave far from the absorbing horizon. This stable trapping makes uniform decay slow, but approximate modes and exact resonances are different objects.

For a precise upper-bound example, take the massless Klein–Gordon equation on four-dimensional Kerr–AdS with $|a|\lt L$ and $r_+^2\gt |a|L$, with homogeneous fast reflecting falloff. For this estimate measure lengths and times in units of $L$. On the horizon-penetrating stationary slices $\Sigma_{t^*}$, define

$$\begin{aligned} e_1[\Phi]&=r^{-2}|\partial_{t^*}\Phi|^2 +r^2|\partial_r\Phi|^2+|\nabla_S\Phi|^2+|\Phi|^2,\\ E_1(t^*)&=\int_{\Sigma_{t^*}}e_1[\Phi]r^2\,dr\,d\Omega,\\ e_2[\Phi]&=e_1[\Phi]+e_1[\partial_{t^*}\Phi] +\sum_{i=1}^3e_1[\Omega_i\Phi]+r^4|\partial_r^2\Phi|^2,\\ \mathcal E_2&=\int_{\Sigma_0}e_2[\Phi]r^2\,dr\,d\Omega. \end{aligned}$$

Here $d\Omega=\sin\theta\,d\theta\,d\phi$, the radial derivative holds $t^*$ fixed, $\nabla_S$ uses the induced metric on the two-surfaces of constant $(t^*,r)$, and $\Omega_i$ are the three standard angular rotation vector fields. This is a positive weighted norm, not simply the conserved Killing energy. With finite $\mathcal E_2$, the nondegenerate energy satisfies

$$E_1(t^*)\leq \frac{C\mathcal E_2}{(\log t^*)^2},\qquad t^*\geq2.$$

The constant depends on the fixed background, not on the initial datum. The spatial $H^1_{\mathrm{AdS}}$ field norm therefore decays at most as $C\mathcal E_2^{1/2}/\log t^*$. The energy estimate is proved in Holzegel and Smulevici 2013, Proposition 11.1, printed p. 38, Eqs. (139)–(140), PDF; the slicing and equivalent higher-order norms are specified in §§2.8–3, printed pp. 10–11. These are uniform upper bounds, not asymptotic equalities for every solution. The paper also treats nonzero masses with its additional mass-dependent parameter restrictions.

Separately, exponentially accurate, real-frequency quasimodes give long-lived approximate solutions and obstruct any faster uniform decay bound in the fixed energy hierarchy Holzegel and Smulevici 2014, Theorem 1.2 and Corollary 1.1, preprint pp. 4–5, PDF. That construction is not itself a proof of an exact quasinormal-frequency law such as $-\operatorname{Im}\omega_\ell\sim e^{-c\ell}$.

The mechanism can be seen in a schematic mode sum. Suppose $\Gamma_\ell\sim e^{-c\ell}$ and

$$\mathcal A(t)\sim \sum_{\ell\ge0}a_\ell e^{-t e^{-c\ell}}.$$

At time $t$, modes with $\ell\lesssim c^{-1}\log t$ have substantially decayed, whereas higher modes remain. If finite differentiability gives only $a_\ell\sim\ell^{-p}$ with $p\gt1$ so that this schematic amplitude sum converges, the surviving sum can scale as a negative power of $\log t$. If analytic data instead give $a_\ell\sim e^{-\sigma\ell}$, the same crossover can yield a power $t^{-\sigma/c}$. Thus the summed tail can depend on data regularity even though every individual quasinormal mode decays exponentially.

AdS decay and recurrence results through 2026-08-12

The following results answer different questions and should not be collapsed into a single recurrence verdict.

Evidence	System and boundary condition	Trapping and regime	Supported conclusion	Limitation
Exact linear spectrum	Scalar on pure global AdS; standard reflecting condition	No horizon; linear	Commensurate real modes give exact rephasings	Robin data, multiple fields, or nonlinear shifts can spoil commensurability
Analytic energy estimate	Klein–Gordon on four-dimensional Kerr–AdS; homogeneous reflecting condition	Stable trapping; fixed background	Uniform field decay no slower than the stated inverse-logarithmic estimate under the theorem’s hypotheses	It is not an exact asymptotic for each datum or fixed $\ell$
Nonlinear numerical evolution	Spherical Einstein–massless-scalar system in AdS$_4$; reflecting boundary	Resonant pure-AdS spectrum; fully backreacting	Narrow Gaussian families show repeated focusing and apparent-horizon formation, motivating weak-turbulence instability	Finite families, time, and resolution do not prove collapse for every small generic datum
Weakly nonlinear two-time analysis	Spherical Einstein–scalar AdS$_4$; reflecting boundary	Mode coupling on slow time $\tau=\varepsilon^2t$	Quasiperiodic equilibria, stability islands, and recurrence scales coexist with cascading solutions	Leading-order approximation and finite mode truncation fail before actual collapse
Fully nonlinear simulation	Slowly rotating Kerr–AdS with a scalar and reflecting boundary	Stable trapping; nonaxisymmetric backreaction	Generic simulated perturbations approach a time-dependent nonaxisymmetric black hole rather than Kerr–AdS	Numerical evidence in a specified model, not a general nonlinear theorem
Spectral prediction plus nonlinear simulation	Gravitational perturbations of Schwarzschild–AdS$_5$ with an $SO(3)$ isometry and normalizable boundary data	QNM accumulation; smooth analytic data	A late power law with oscillatory modulation is observed, with no instability during the resolved evolution	Dimension, symmetry, regularity, and finite runtime are essential

The pure-AdS collapse evidence comes from four-dimensional, spherically symmetric Einstein–massless-scalar evolutions; it supports a conjectured resonant cascade for broad data classes, not a generic theorem (Bizoń and Rostworowski 2011, full article, pp. 1–4). The countervailing recurrence and stability-island evidence is derived in the small-amplitude two-time framework and explicitly includes both collapsing and noncollapsing examples (Green et al. 2015, §§ III–V, pp. 7–15).

Two later calculations sharpen the black-hole side without making it universal. Fully nonlinear Einstein–scalar simulations find persistent stable trapping and a nonaxisymmetric time-dependent endpoint for slowly rotating Kerr–AdS (Figueras and Rossi 2025, §§ 3–5). For smooth gravitational data with an $SO(3)$ isometry on Schwarzschild–AdS$_5$, a spectral calculation and long nonlinear evolutions instead find an oscillatory power-law regime and no instability within the simulated window (Crump and Santos 2026, §§ 2.3–4, pp. 5–17). Different dimension, rotation, symmetry, regularity, and nonlinear channel make these complementary rather than contradictory results.

Recurrence, slow decay, and nonlinear transfer are distinct

Three questions are often conflated. First, does the linear spectrum have real commensurate frequencies? Pure AdS with a fixed reflecting condition can answer yes. Second, does a black-hole perturbation decay uniformly? Stable trapping can preclude a uniform exponential decay estimate and allow inverse-logarithmic or regularity-dependent summed decay. Third, does nonlinear mode coupling form a horizon, approach a quasiperiodic island, or settle to another black hole? That depends on the full Einstein–matter system, data family, amplitude, symmetry, and accessible time scale.

Rotation adds another distinction. Outside a suitable Hawking–Reall regime, superradiant amplification at the horizon and reflection at infinity can produce growing modes rather than decay or recurrence. Even within nominally slow rotation, nonlinear stable-trapping evidence must be stated for the specific model tested. A claim about pure-AdS scalar recurrence cannot decide either case.

A discriminating comparison

To isolate the effect of the horizon, use the same scalar mass, dimension, angular data, normalization, and reflecting boundary condition in pure global AdS and in a small Schwarzschild–AdS black hole. Compute real normal frequencies in the first problem and complex quasinormal frequencies in the second. Evolve the same boundary-compatible wave packet and record:

1. the field and energy at the same areal radius after each round-trip time;
2. the vanishing boundary flux and, in the black-hole run, the accumulated future-horizon flux;
3. projections onto successively larger $(n,\ell)$ ranges;
4. convergence under radial, angular, and time-step refinement; and
5. sensitivity of late-time fits to data regularity and fit window.

Pure AdS should rephase according to its commensurate spectrum. The black-hole signal should lose energy by precisely the integrated horizon flux and show phase and amplitude shifts consistent with $\operatorname{Im}\omega_j<0$. Failure to close that energy balance is a numerical failure, not evidence for or against recurrence.

Exercise: why weak damping destroys exact recurrence

For a positive-frequency superposition at fixed $\ell$, let pure-AdS modes have $\omega_n=(\Delta+\ell+2n)/L$. Show that they rephase after $\pi L$. Then replace them by $\omega_n-i\Gamma_n$ with positive, nonidentical $\Gamma_n$. Can a generic superposition recur exactly at a nonzero time?

Solution

At $T=\pi L$,

$$e^{-i\omega_nT} =e^{-i\pi(\Delta+\ell+2n)} =e^{-i\pi(\Delta+\ell)},$$

which is independent of $n$. Every positive-frequency coefficient therefore acquires one common phase. Phase-invariant local quadratic observables of that complex positive-frequency field recur. Negative-frequency modes instead acquire $e^{+i\pi(\Delta+\ell)}$, so the conclusion does not apply without qualification to arbitrary two-sign data. For example, when $\Delta$ is an integer, a real field returns after $2\pi L$; a generic noninteger $\Delta$ need not do so. Constancy of the total energy alone establishes none of these local recurrence statements. With a horizon, the positive-frequency factor becomes

$$e^{-i\omega_nT}e^{-\Gamma_nT}.$$

For exact recurrence up to a common factor, all $e^{-\Gamma_nT}$ would have to be equal. Distinct positive $\Gamma_n$ violate that condition for every $T>0$. Very small damping can make a finite-time echo look nearly recurrent, but it cannot restore exact recurrence for a generic superposition.

What remains open

As of 2026-08-12, it remains unresolved which regularity classes and generic small-data sets lead to nonlinear stability, weak turbulence, or collapse for the major asymptotically AdS black-hole families. No boundary-condition-free or dimension-independent recurrence law is settled. Further progress could come from a nonlinear stability or instability theorem for a black-hole family, a converged evolution spanning the predicted long trapping time, a classification change for admissible boundary conditions, or a new result on resonance accumulation.

The scope here is the spectral and boundary mechanism behind linear decay, echoes, and oscillatory or slowly decaying response. Full nonlinear decay theorems are coordinated with mathematical relativity, and observational late-time fitting requires an observation model.

References

• Bizoń, Piotr, and Andrzej Rostworowski. “On Weakly Turbulent Instability of Anti-de Sitter Space.” Physical Review Letters 107 (2011): 031102. DOI. Open PDF
• Crump, John R. V., and Jorge E. Santos. “Tails from the Bulk: Gravitational Decay in AdS$_5$.” Journal of High Energy Physics 2026 (2026): 140. DOI. Open PDF
• Figueras, Pau, and Lorenzo Rossi. “Non-Linear Instability of Slowly Rotating Kerr–AdS Black Holes.” Journal of High Energy Physics 2025 (2025): 107. DOI. Open PDF
• Green, Stephen R., Antoine Maillard, Luis Lehner, and Steven L. Liebling. “Islands of Stability and Recurrence Times in AdS.” Physical Review D 92 (2015): 084001. 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
• Holzegel, Gustav, and Jacques Smulevici. “Quasimodes and a Lower Bound on the Uniform Energy Decay Rate for Kerr–AdS Spacetimes.” Analysis & PDE 7 (2014): 1057–1090. DOI. Open PDF.