Echo and Near-Horizon Searches

Postmerger “echoes” are delayed structures sought after the ordinary black-hole response. They are not a single universal prediction: their delay, phase, damping, frequency dependence, and even existence depend on the proposed near-horizon boundary condition and exterior potential. A credible search must therefore separate a reproducible signal from detector glitches, spectral leakage, template flexibility, and the trials incurred while scanning many delays and morphologies.

Helpful background. Ringdown Likelihoods and Start-Time Dependence provides the finite-window noise and truncation controls. Quasinormal Boundary Conditions and Resonances supplies the standard absorbing-horizon problem against which a modified boundary is compared.

From a modified boundary to delayed response

In a one-dimensional radial picture, the exterior curvature potential partially reflects waves near its barrier, while a hypothetical inner surface at tortoise coordinate $r_*^{\,0}$ reflects some incident radiation. A wave packet can then make repeated round trips. The leading geometric delay is approximately

$$\Delta t_{\rm echo} \simeq \frac{2}{c} \left( r_*^{\,{\rm barrier}}-r_*^{\,0} \right),$$

up to frequency-dependent scattering and source effects. Here $r_*$ has units of length and the delay is measured in the asymptotic time coordinate of the source spacetime; cosmological redshift multiplies it by $1+z$ at the detector. Because $r_*$ grows logarithmically near a classical horizon, a microscopic proper displacement need not imply a microscopic observed delay.

A schematic frequency-domain train can be written

$$\widetilde h_{\rm echo}(f) = \widetilde h_{\rm seed}(f) \frac{ \mathcal R_{\rm in}(f)\mathcal T_{\rm b}(f) e^{2\pi i f\Delta t} }{ 1-\mathcal R_{\rm in}(f)\mathcal R_{\rm b}(f) e^{2\pi i f\Delta t} },$$

Here $\mathcal R_{\rm in}$ is the inner reflectivity and $\mathcal R_{\rm b},\mathcal T_{\rm b}$ describe the barrier in a common wave-amplitude normalization. The positive delay phase follows from $\widetilde h(f)=\int h(t)e^{+2\pi ift}dt$: delaying a waveform by $\Delta t$ multiplies its transform by $e^{+2\pi if\Delta t}$. All frequencies and delays in the expression must use the same frame.

The denominator sums repeated round trips, $1+q+q^2+\cdots=1/(1-q)$, where $q=\mathcal R_{\rm in}\mathcal R_{\rm b}e^{2\pi if\Delta t}$. A decaying train requires $|q|\lt1$ on the frequencies used. Perfect trapping or net superradiant amplification needs a separate causal stability analysis, not an assumed convergent train. Identical, equally spaced, geometrically damped copies further require approximately frequency-independent round-trip gain; dispersive reflection changes each packet’s shape.

The ordinary prompt and quasinormal response must also be modeled. Otherwise a flexible echo component can absorb late merger–ringdown mismatch. Nonlinear mode coupling, mode mixing, precession, memory, or an evolving remnant can produce postpeak structure without a reflective surface.

A search statistic with an honest background

For parameters $\lambda$ describing delay, phase, damping, and morphology, a matched-filter search may use a likelihood ratio

$$\Lambda(\lambda) = \frac{p(d\mid h_{\rm ordinary}+h_{\rm echo}(\lambda))} {p(d\mid h_{\rm ordinary})}.$$

Its maximum over a bank, $\Lambda_{\max}$, is not distributed like the statistic at a preselected $\lambda$. The false-alarm probability is

$$p_{\rm FA} = \Pr\!\left( \max_{\lambda\in\mathcal B}\Lambda_{\rm noise}(\lambda) \geq \Lambda_{\max}^{\rm obs} \right),$$

where $\mathcal B$ is the complete search, including event selection, delay windows, waveform variants, data-conditioning choices, and any analyst tuning. Off-source stretches and time slides should reproduce the same nonstationarity and veto logic as the on-source analysis.

Bayesian evidence also pays a trials penalty through prior volume only if the prior and model were fixed before inspecting the candidate. Trying many undocumented priors or templates and reporting the largest Bayes factor reintroduces a hidden look-elsewhere effect.

What a persuasive echo candidate would require

• Coherent arrival times, phases, and amplitudes across multiple detectors under the known antenna responses.
• Stability under calibration uncertainty, power-spectrum estimation, gating, line removal, and modest changes to the analysis window.
• A background measured with the complete pipeline and adequate tail statistics.
• Recovery efficiency from injections spanning the claimed morphology, including injections into real neighboring noise.
• Null injections containing only the best available merger–ringdown waveform, so ordinary model error is not mislabeled.
• Independent implementations and, for a population claim, a predicted mass-and-spin scaling of the delay across events.

A residual plot is diagnostic, not a detection statistic. A peak after whitening can be created by the whitening filter itself, and visually repeated features do not supply a calibrated false-alarm probability.

Frontier evidence through 2026-08-18

Evidence stream	Result	Scientific boundary
Initial compact-binary echo claim	A phenomenological template search reported tentative postmerger support in early events Abedi, Dykaar, and Afshordi 2017, §§III–V.	The statistic, background, template choices, and a posteriori trials became central points of dispute.
Independent reanalysis	A reproduction with wider prior and background studies found substantially lower significance and emphasized analysis-choice sensitivity Westerweck et al. 2018, §§IV–VI.	This constrains that claim and template family; it does not rule out every near-horizon response.
Bayesian O1 search	Explicit signal-versus-noise model comparisons calibrated against empirical backgrounds found no compelling evidence in the events and template family analyzed Lo, Li, and Weinstein 2019, §§II–IV.	Null results map to model- and prior-dependent sensitivity, not to the absence of all exotic compact objects.
O3 template search	A 34-event search using a repeated-waveform template and a perturbation-theory-motivated template found both p-value distributions consistent with noise Uchikata et al. 2023, §§III–V, Fig. 2, and Tables I–III.	The null statement applies to those two template families, search ranges, and background constructions; it is not a universal exclusion of near-horizon response.
GWTC-5.0 program status	The official catalog paper does not report a new echo analysis; it states that earlier post-ringdown echo searches were discontinued because stronger constraints were not expected LVK 2026, §1, p. 3.	Omission is neither a fresh null result nor evidence for a signal. The empirical status must be read from the searches that were actually performed.

The appropriate present statement is therefore a null, model-bounded observational status—not a discovery and not a universal exclusion. Refresh it after a new official catalog/TGR release, revised calibration, an independently reproduced candidate, or a search with materially broader injection-calibrated coverage.

Exercises

1. A search evaluates $N=100$ effectively independent templates. Each has single-template tail probability $p_1=10^{-3}$ for exceeding the observed statistic. Compute the global false-alarm probability.

Solution

The probability that none exceeds the threshold is $(1-p_1)^N$, so

$$p_{\rm global} = 1-(1-p_1)^N = 1-0.999^{100} \simeq 0.0952.$$

The local $10^{-3}$ tail becomes a roughly $9.5\%$ global probability. Correlated templates reduce the effective $N$, which must be calibrated rather than guessed.

2. In a simplified cavity, the round-trip amplitude multiplier is $q=\lvert\mathcal R_{\rm in}\mathcal R_{\rm b}\rvert=0.6$. What is the amplitude of the fourth returning packet relative to the first?

Solution

If the first return has amplitude $A_1$, each additional round trip multiplies it by $q$. Thus

$$\frac{A_4}{A_1}=q^3=0.6^3=0.216.$$

This geometric law follows only from the simplified frequency-independent cavity. Dispersive reflectivities change both shape and phase from return to return.

References

• Abedi, J., Dykaar, H., and Afshordi, N. (2017). “Echoes from the Abyss: Tentative evidence for Planck-scale structure at black hole horizons.” Physical Review D 96, 082004. DOI.
• LIGO Scientific Collaboration, Virgo Collaboration, and KAGRA Collaboration (2026). “GWTC-5.0: Tests of General Relativity.” arXiv:2607.19293.
• Lo, R. K. L., Li, T. G. F., and Weinstein, A. J. (2019). “Template-based gravitational-wave echoes search using Bayesian model selection.” Physical Review D 99, 084052. DOI.
• Uchikata, N., Narikawa, T., Nakano, H., Sago, N., Tagoshi, H., and Tanaka, T. (2023). “Searching for Gravitational Wave Echoes from Black Hole Binary Events in the Third Observing Run of LIGO, Virgo, and KAGRA Collaborations.” arXiv:2309.01894.
• Westerweck, J., Nielsen, A., Fischer-Birnholtz, O., Cabero, M., Capano, C., Dent, T., Krishnan, B., Meadors, G., and Nitz, A. H. (2018). “Low significance of evidence for black hole echoes in gravitational wave data.” Physical Review D 97, 124037. DOI.