A new field of astronomy has arisen with the detection of gravitational waves (GWs). To date, the Laser Interferometer Gravitational-wave Observatory (LIGO)1 and Virgo2 have observed twelve merging binary black holes (BBHs)3,4,5,6,7, two binary neutron star mergers8,9, and a putative neutron star-black hole merger10. These detections are allowing us to understand the nature of compact objects, their populations11, and their formation channels12,13. These observations have also put to test general relativity (GR)14,15,16 in the strong-field regime for the first time, so far confirming its predictions17,18. Despite this groundbreaking achievement, LIGO and Virgo have not yet reached the sensitivity to observe in exquisite detail the merger and the relaxation of the highly distorted black holes (BHs) left behind by BBHs, when dynamical gravity reaches its ultimate expression. As the sensitivity of LIGO and Virgo increases and future generation detectors, such as the Laser Interferometer Space Antenna (LISA) and the Einstein Telescope (ET), become operational, GWs will provide us with an unprecedented view of highly distorted BH horizons, allowing us to test in further detail fundamental aspects of GR like the “no-hair” theorem15,19 and explore the quantum properties of BHs20.

Such studies will rely on a deep understanding of how GW signals encode the dynamical properties of the source. In anticipation of future large signal-to-noise ratio detections, it is important to investigate how GWs reflect not only the “common” properties of BBHs (e.g., BH masses and spins, orbital eccentricity, and orientation) but also other fundamental aspects that can be inferred from the morphology of the signal. For instance, all current observations show a rather simple “chirp” morphology4, consisting of a monotonic increase of both frequency and amplitude. Initially, both quantities increase slowly, reflecting the low frequency and tightening of the orbit as the two BHs approach each other21. Just before the merger, the two BHs reach speeds comparable to that of light, leading to a rapid rise of both frequency and amplitude21,22. Once the BHs merge, a highly distorted BH settles into a Kerr BH radiating exponentially decaying ringdown emission3,19,23,24. However, this simple morphology describes the GW signal only when the binary has comparable masses or it is viewed face-on so that the emission is vastly dominated by the so-called quadrupole mode.

In contrast, asymmetric BBHs also show strong GW emission in sub-dominant higher-order modes during the merger and ringdown stages25, allowing for GWs with non-trivial morphology complexity26,27,28 that may unveil new features of the post-merger dynamics. The connection between the horizon dynamics and the GW emission has been widely studied in two main ways. The first consists of finding correlations between far-field signals and fields near the horizon, revealing close connections between the horizon geometry, the GW flux, and strong-field phenomena such as the anti-kick29,30,31,32. The second approach has focused on the systematic development of analytical tools that can probe and explain the geometrodynamics of space-time causing such correlations and relate it with the generation of GWs33,34,35,36,37. None of these studies discuss a direct link to the GW strain observable by detectors. In this work, we follow the approach of29,30,31,32 to correlate a concrete observable feature of the GW strain to a geometrical property of the final BH horizon.

Using numerical relativity simulations, we show that multiple post-merger frequency peaks (or chirps) can be measured near the orbital plane of unequal-mass binaries We show that these correlate to the line-of-sight passage of strongly emitting regions of large mean curvature gradient and locally extremal Gaussian curvature, present on the dynamical apparent horizon of the final black hole (BH), which cluster around a “cusp”-like defect on it. Conversely, frequency minima correlate to the passage of the “smoother” opposite region of the horizon, where curvature gradients are smaller.

Results and discussion

Post-merger chirps

Figure 1a depicts the different stages of a BBH. Figure 1b shows the GW strain time series and time-frequency maps recorded by observers in different locations around a numerically simulated BBH with a mass ratio q = m1/m2 = 3. We perform our numerical simulations using the MAYA code, based on the EINSTEINTOOLKIT38 (for details, please see the “Methods” section). We use geometrical units in terms of the total mass of the binary M = m1 + m2 with the speed of light set to unity. The green vertical line denotes the formation of the common apparent horizon. The “face-on” panel shows the signal observed face-on, showing the “vanilla” chirp structure consistent with current observations4,5. The different panels in Fig. 1b show the signals recorded at different positions on the orbital plane.

Fig. 1: Post-merger chirps.
figure 1

a The different stages of a binary black-hole coalescence. b, c The strain time series h(t) of extracted from binary black-hole simulations (white background) and the corresponding time-frequency maps (black background). The green vertical line denotes the instant at which the final common horizon is first found in our simulations. All waveforms include the most dominant modes (m) = {(2, ± 1), (2, ± 2), (3, ± 2), (3, ± 3), (4, ± 3), (4, ± 4)}. b The case of mass ratio q = 3 for different viewing angles (from left to right): face-on, kick-off, kick-on, and 55° away from the kick direction measured in the direction of the orbit, for which the double-chirp feature is most prominent. c The signals emitted by binaries with a corresponding mass ratio q = 1.1–10 in this last direction. The double-chirp feature becomes more pronounced as the mass ratio q increases.

Some time after the common horizon forms, these signals show a clear drop in the frequency, followed by a secondary peak, or post-merger chirp. Depending on the location of the observer, the chirps occur at different times and involve different peak frequencies and intensities. It is illustrative to compare the signals observed in the direction of the recoil of the final BH26,28, (kick-on) to those observed in the opposite direction (kick-off). While the kick-on observer records a secondary chirp with larger amplitude and peak frequency than the first, the converse is measured by the kick-off observer. We find that the secondary chirp is more intense compared to the first one in a direction 55° away from the final recoil (or kick) direction, measured in the direction of the original orbit. This signal is shown in the rightmost panel, showing a clear “double-chirp”. Notably, we find that this is also true for varying mass ratios. The corresponding signals are shown in Fig. 1c for binaries with mass ratios q = 1.1–10. These also make evident that the double-chirp signature becomes more pronounced as the binary becomes more asymmetric. We acknowledge that a similar non-trivial post-merger emission, visible in the time-domain, was shown by González et al.26, in terms of the Newman–Penrose scalar Ψ4. However, its frequency content, which is the departing point of our study and the reason behind the double-chirp name, was not shown.

Analytically, these complex and observer-dependent post-merger waveform morphologies can be explained by the asymmetric interaction of the different quasi-normal emission modes beyond the quadrupole one in different directions around the binary. These modes are triggered during the merger and ringdown of asymmetric binaries and have a larger impact on highly inclined binaries25,28,39. However, the clarity of the double-chirp signature suggests a connection to some underlying post-merger feature, similar to how the increase of the frequency during the inspiral is connected to the increasing frequency of the binary. In the following, we argue that this feature is the existence of regions of locally extremal curvature that are distributed non-uniformly on the dynamical apparent horizon of the final BH. While three of these regions cluster around a global curvature maximum that we denote as the “cusp” forming a “trident”, a fourth one sits on the opposite or “back” side of the horizon. These coincide with regions of maximal GW emission. As the final BH relaxes, this structure rotates pointing to all observers on the orbital plane while fading away. We show that after the cusp (back) of the horizon crosses the line-of-sight, frequency peaks (minima) are recorded at a time consistent with the GW travel time determined by the distance to the observer.

Emission profile far from the source

To make a first connection between the time-frequency morphology of the signal and the structure of the GW emission, the left panel of Fig. 2a shows a snapshot of the GW emission in the orbital plane of our q = 3 binary at a time of 52.3M after the merger after the waves have traveled far from the source. We represent the GW emission by the absolute value of the Newman–Penrose scalar Ψ440, related to the GW strain h by \({\Psi }_{4}(t)=\frac{{d}^{2}{h}^{* }(t)}{{\mathrm{d}}{t}^{2}}\). As one goes around the final BH, it is clear that the recorded signal will depend on the viewing angle. This can be, in fact, observed in Fig. 2b, which showcases the time-frequency maps recorded by the kick-on and kick-off observers highlighted in the left panel. The radial separation of \(\Delta R \sim {\mathcal{O}}(20M)\) between the wave trains reaching each observer in the left panel is consistent with the time delay \(\Delta t \sim {\mathcal{O}}(20M)\) between frequency peaks shown in the right one. Following the kick-on and kick-off directions, it is also evident that the intensity of the GW fronts is different in each direction. While two wave trains of similar intensity reach the kick-on observer, the second train reaching the kick-off observer has a much lower intensity. Consistently, the kick-on observer records two chirps of similar intensity plus a weak third one, while the second chirp is barely visible for the kick-off observer.

Fig. 2: Relation between post-merger chirps and arriving gravitational-wave trains.
figure 2

a A snapshot of the gravitational waves in the orbital plane at a time of 52.3M after the merger, for the case of a q = 3 binary. This is expressed in terms of the absolute value of the Newman–Penrose scalar Ψ4. We highlight the kick-on and kick-off directions and the separation \(\Delta t \sim {\mathcal{O}}(20M)\) of the wave trains traveling towards each observer. b The corresponding Ψ4 time-frequency maps recorded by the kick-on and kick-off observers. The time elapsed \(\Delta t \sim {\mathcal{O}}(20M)\) between frequency peaks (or chirps) is consistent with the separation of the arriving wave trains. Two strong wave trains reach the kick-on observer, which translates into a prominent double-chirp. In contrast, the second front traveling toward the kick-off observer is much weaker, translating into a weak second chirp.

The near-horizon region

After making a first connection between the post-merger chirps and the emission profile far from the source, we now zoom in near the source to investigate the connection to the post-merger dynamics. The panels e–h in Fig. 3 show the structure of Ψ4 near the horizon at four selected times tframe throughout its evolution. In the first frame (Fig. 3e), soon after the horizon forms, Ψ4 shows a clearly asymmetric pattern. Three arms (a central one, most prominent, and two surrounding weaker ones) cluster on one side of the horizon forming a “trident” while another arm is present on the opposite side (or “back”). We note that this structure does not form abruptly at merger but arises smoothly from the pattern it had when the BHs were approaching each other. The inset shows that the central arm sits on a “cusp” defect present on the horizon. The other three frames show how this structure rotates and fades away as the final BH evolves. As this happens, the three arms (in particular the central one) and the back of the BH cross the line of sight of every observer multiple times.

Fig. 3: Relation between post-merger chirps and the black-hole cusp passage through the line-of-sight.
figure 3

a, b The time-frequency maps and time series of the gravitational-wave strain h, respectively, observed in the kick-on and kick-off directions, extracted at a distance rext = 75M from the source. c, d The same for the Newman–Penrose scalar Ψ4. The top time axis shows the retarded time tframe − rext. The four vertical lines denote retarded times tframe − rext = tframe,i, with the times tframe,i corresponding to the four simulation snapshots shown in eh. The bottom time axis has been shifted so that t = 0 denotes the time at which the (2, 2) emission mode has its amplitude peak, as it is common in GW data analysis. eh The corresponding four frames of the absolute value of Ψ4 in the orbital plane of a q = 3 binary, at the times tframe,i highlighted in ad. Bright yellow regions denote large values of Ψ4 while dark purple regions denote zeros. The black arrow points in the kick-on direction. The inset highlights the initially asymmetric shape of the final black hole. The three-arm (or “trident”) structure of Ψ4 present on the bottom side of the horizon has its most prominent arm aligned with a “cusp” defect on it.

We now draw our attention to the observed signals. The panels a–d in Fig. 3 show the time-frequency maps and time series for the GW strain h (panels a and b) and Ψ4 (panels c and d) measured in the kick-on and kick-off directions at retarded times tframe − rext, with rext denoting the distance to the source, also known as the GW extraction radius. The vertical lines denote the retarded times that correspond to the four frames on Fig. 3b. As the three Ψ4 arms, and in particular the central, most prominent one, cross the line-of-sight, a frequency peak is measured at a time rext later. This is also noticeable in the time-domain plots in terms of a short instantaneous signal wavelength, consistent with the short separation of the arriving wave-fronts shown in Fig. 2a Similarly, a frequency minimum is observed rext after the back of the horizon crosses the line of sight. This is reflected in the time-domain signals by a larger instantaneous wavelength, consistent with the large separation of \({\mathcal{O}}(20M)\) of the arriving wave trains shown in Fig. 2a This way, we establish a time-connection between the line-of-sight passage of the three Ψ4 arms (in particular the most prominent one) and the back arm and the respective observation of post-merger frequency peaks and minima.

Connecting the horizon geometry with post-merger chirps

Next, we connect the arm structure of Ψ4 present on the dynamical apparent horizon with its curvature. To this, we measure Ψ4 on the intersection of the horizon with the original orbital plane of the binary (i.e., on the final horizon equator), together with the mean curvature H and the Gaussian curvature K. Figure 4 shows the absolute values of the gradient of the mean curvature dH/ds and Ψ4 as a function of the azimuthal angle φ at a time 5M after the common horizon has formed. The parameter s denotes the arc-length along the horizon’s equator (for a detailed description of the arc-length parameter and curvature quantities, please see “Methods” section) and the inset panel shows the corresponding simulation frame. A tight correlation is obvious. In particular, all four Ψ4 maxima, i.e., the arms highlighted in the inset, clearly correspond to local maxima of dH/ds. The three tight Ψ4 arms correspond to three nearby regions of large dH/ds spanning barely ~2 radians, with the maximum of dH/ds (the cusp) matching the maximum of Ψ4, i.e., the “central” arm.

Fig. 4: Relation between the gravitational-wave emission and mean curvature of the apparent horizon.
figure 4

The main panel shows the absolute values of the gradient of the mean curvature dH/ds and the Newman–Penrose scalar Ψ4 measured on the intersection of the orbital plane and the final common apparent horizon of a q = 3 binary as a function of the azimuthal angle ϕ, measured 5M after the formation of of the common apparent horizon (corresponding to tframe = 82.4 in e–h of Fig. 3). The parameter s denotes the arc-length along the equator. The inset shows the corresponding simulation snapshot for Ψ4 measured on the orbital plane.

  Figure 5a shows that the Ψ4 arms also match regions of locally extremal Gaussian curvature K which, unlike H, is intrinsic to the apparent horizon and coordinate independent. In Supplementary Fig. 1 and Supplementary Notes I, we show that the same relations hold for all the mass ratios shown in Fig. 1. Moreover, while the location of the apparent horizon itself depends on the gauge choice, we show that these relations also hold in an alternative gauge. Figure 5b–d and those in Supplementary Fig. 2 (and Supplementary Notes II) show that this relation is well preserved throughout the evolution of the final black hole, as the central arm points to the different observers in the orbital plane and the Ψ4 structure fades away. We observe that a slight degradation of this correlation occurs at very late times of 30M after the common horizon has formed when the emission is very weak and more prone to be affected by numerical artefacts. Despite this, the three maxima and minima of Ψ4 sitting around the central arm are well co-located with those of K. Also, while here we have used dH/ds to facilitate a visual comparison with the simulation frames, we show that Ψ4 actually follows  −dH/ds.

Fig. 5: Relation between the gravitational-wave emission and Gaussian curvature of the apparent horizon.
figure 5

Values of the Newman–Penrose scalar Ψ4 and the Gaussian curvature K on the intersection of the orbital plane and the final common apparent horizon of a q = 3 binary as a function of the azimuthal angle ϕ at times 5M, 10M, 17.3M, and 35.4M after the formation of the common apparent horizon. The x axis has been shifted by a value ϕcusp so that the maximum of Ψ4 (i.e., its central arm) is at ϕ − ϕcusp = 0.

Finally, even if not clearly visible Figs. 3e–h and 5 makes also evident that the highly asymmetric structure of the horizon understood as the “clustering” of three Ψ4 arms and the corresponding extremal curvature regions, is maintained during BH evolution, as these always span an angle that barely exceeds  ~2.5 radians. This way, the passage of these Ψ4 arms by the line-of-sight corresponds to the passage of three regions of extremal mean curvature gradient and extremal Gaussian curvature on the horizon, with the central arm corresponding to the cusp.


The observation of the merger and ringdown stages of BH mergers grants access to the strongest regime of gravity, in which space-time shows its ultimate phenomenology through the dynamical evolution of highly distorted BH horizons. Connections between these dynamics and the observed GWs have been proposed and widely investigated29,30,31,32,33,34,35,36,37. However, no explicit examples of concrete observable features in the GWs have been described to date. In this work, we propose the first such connection. We have shown that non-trivial features consisting on multiple frequency peaks in the post-merger GW emission of edge-on, asymmetric BH mergers, are linked to the presence of large curvature regions in the dynamical apparent horizon of the final BH that is asymmetrically distributed.

First, we have shown that an asymmetric emission pattern forms around the final common horizon, with three arms clustering one side and one on the other. Second, we have shown that frequency peaks, which we refer to as post-merger chirps, are observed in the post-merger signal as the three arms cross the line of sight, after a time consistent with the GW travel time. Last, we show that these arms coincide with locally extremal values of the Gaussian curvature and the mean curvature gradient of the dynamical apparent horizon, with the strongest arm sitting on its largest curvature region, which we call cusp. While post-merger chirps may resemble the signature of BH echoes20, these are not, as we are considering standard BHs. For all the mass ratios we have considered, we find this feature is more prominent on the orbital plane of the binary, 55° from the final kick direction measured in the direction of the original orbit (or that of the final spin). Nevertheless, we assume that this will be subject to change if spinning BHs were considered41,42.

Finally, in the Supplementary Fig. 3 and Supplementary Notes III, we show that Advanced LIGO detectors working at their design sensitivity may observe the post-merger chirp signature for the case of a correctly oriented copy of the BBH GW1707294,43, suggesting that such observation may be feasible before the arrival of the next generation of GW detectors.


Time-frequency maps

The continuous wavelet transform of a function g(t) is,

$$W(a,b)=\frac{1}{\sqrt{a}}\mathop{\int}\limits_{-\infty }^{\infty }g(t)\psi \left(\frac{t-b}{a}\right)\,{\rm{d}}t,$$

where ψ, the “mother wavelet”, is a continuous function in time and frequency, evaluated at scales a > 0 and translations b. The scale parameter dilates the mother wavelet ψ, providing a range of time-frequency resolutions, while the translations provide time localization of the signal power, g2(t). By selecting for ψ a function which is compact in the time- and frequency-domains, together with a judicious choice of scales appropriate for the problem at hand, one can use the continuous wavelet transform to resolve substructure in the signal g(t) of particular physical interest. A complete description of wavelet analysis may be found in ref. 44.

The mother wavelet used in our decomposition is the Morlet wavelet, a Gaussian-modulated sinusoid with minimal compactness in the time- and frequency-domains: \(\psi (x)=\frac{1}{{\pi }^{4}}\exp \left(2\pi i{f}_{0}x\right)-\exp \left(-2{\pi }^{2}{f}_{0}^{2}\right)\exp \left({x}^{2}/2\right)\), where \(x=\frac{t-b}{a}\) and f0 is the frequency of the sinusoid, the center frequency of the wavelet.

As described, the continuous wavelet transform is expressed in terms of mother wavelet dilation scales a. Our interest lies in the frequency content of the signal, however. It may be shown, by considering the response to a sinusoid of frequency f, that the maximum of the mother wavelet W(ab) lies at a = f0/f, so that the focii of the time-frequency response of the signal to the wavelet transform are centered at frequencies f = f0/a. For discretized time series data, f = Fsfc/a, where Fs is the sample frequency of the data. We determine empirically, via visual inspection, that a mother wavelet center frequency f0 = 0.4, and scales a [1. . 128] yield a time-frequency resolution which allows us to resolve the pertinent substructure of the chirping features reported in this work. We have used the pyCWT software library available at to perform these decompositions numerically.

Higher modes of the gravitational-wave emission

The complex GW strain emitted in a direction (ιϕ) on the sky of a BBH can be written as a superposition of different GW modes h,m(ιϕt) as45

$$h(t)={h}_{+}(t)-{\rm{i}}{h}_{\times }(t)=\sum _{\ell \ge 2}\mathop{\sum }\limits_{m = -\ell }^{m = \ell }{Y}_{\ell ,m}^{-2}(\iota ,\phi ){h}_{\ell ,m}(t).$$

Here, h+ and h× denote the two GW polarizations, the Y,m’s are spin −2 weighted spherical harmonics, and (ιϕ) are the polar and azimuthal angles of a spherical coordinate system centered on the binary. This is chosen so that, ι = 0 (face-on) denotes the direction of the orbital angular momentum, while the orbital plane of the binary is located at ι = π/2 (edge-on). For face-on binaries, the quadrupolar (m) = (2, ± 2) modes vastly dominate during all the stages of the binary45. As consequence, the frequency of the resulting GW is, to a good approximation, twice the orbital one and the observed signal shows a canonical single chirp morphology. Secondary GW modes are triggered during the merger and ringdown stages and are more visible near the orbital plane of the binary23,27,28.

Computational methodology

We performed the simulations for this study using our MAYA code46,47,48,49, a branch of the EINSTEINTOOLKIT38. This code was also used produce the Georgia Tech catalog of gravitational waveforms50. We evolve the BSSN formulation of the Einstein equations51 using the moving puncture gauge condition52,53 for binary BH systems. The MAYA code is built upon the CACTUS54 code, using KRANC55 for code generation and CARPET56 for mesh refinement. We extract gravitational waveforms from the simulation data using the Newman–Penrose scalar Ψ457, which we calculate using the WEYLSCAL4 thorn of the EINSTEINTOOLKIT. We use the AHFINDERDIRECT58 thorn with minor modifications (see below) to locate and analyze apparent horizons.

Calculating curvatures on the apparent horizon

Consider an apparent horizon surface \({\mathcal{S}}\) (with 2-metric γab) in a spacelike hypersurface Σt (with 3-metric gij and associated covariant derivative operator i). The surface \({\mathcal{S}}\) can be defined by the outward-pointing unit normal to the surface ni, whose divergence is equal to the mean curvature of the surface H = ini. H is calculated at run time by the AHFINDERDIRECT thorn because it is an extrinsic quantity and therefore more difficult to calculate in post-processing, as it depends on the metric and its derivatives in the neighborhood of horizon. On the other hand, the Gaussian curvature K of \({\mathcal{S}}\) is intrinsic to the surface and can therefore be calculated using only the metric induced on the horizon γij = gij − ninj and its derivatives. We have modified the AHFINDERDIRECT thorn to output γij on \({\mathcal{S}}\) for this purpose.

On a two-dimensional surface, the Riemann tensor has only one independent component and is therefore completely determined by the Ricci scalar. The Gaussian curvature of \({\mathcal{S}}\) is half of the Ricci scalar and is related to components of the Riemann tensor by Rabcd = K(γacγdb − γadγcb), where the indices a, b, c, and d denote the polar (θ) and azimuthal (ϕ) angles on the horizon surface. Specifically, we use the Rθϕθϕ component of the Riemann tensor in our calculations. Furthermore, by calculating K in the orbital plane of the binary we significantly reduce the complexity of this expression by taking advantage of the symmetry properties of γab.

We compute derivatives with respect to the arc-length parameter s on the intersection of the orbital plane and the apparent horizon. This is related to the azimuthal angle ϕ by

$$\frac{{\mathrm{d}}s}{{\mathrm{d}}\phi }=\sqrt{{\gamma }_{ij}\frac{{\mathrm{d}}{x}^{i}}{{\mathrm{d}}\phi }\frac{{\mathrm{d}}{x}^{j}}{{\mathrm{d}}\phi }}=\sqrt{{\gamma }_{rr}\ {({\partial }_{\phi }R)}^{2}+{\gamma }_{\phi \phi }+2{\gamma }_{r\phi }\left({\partial }_{\phi }R\right)}\ .$$

Here, R denotes the coordinate radius of the horizon. The spherical components of γij are related to the cartesian component’s output by the AHFINDERDIRECT thorn by:

$${\gamma }_{rr} ={\cos }^{2}(\phi ){\gamma }_{xx}+{\sin }^{2}(\phi ){\gamma }_{yy}+\sin (2\phi ){\gamma }_{xy}\\ {\gamma }_{\phi \phi } ={r}^{2}\left[{\sin }^{2}(\phi ){\gamma }_{xx}+{\cos }^{2}(\phi ){\gamma }_{yy}-\sin (2\phi ){\gamma }_{xy}\right]\\ {\gamma }_{r\phi } =\left[\frac{1}{2}\sin (2\phi )({\gamma }_{xx}-{\gamma }_{yy})+\cos (2\phi ){\gamma }_{xy}\right].$$