GW190521: Evidence for an IMBH Merger

• GW190521 is a gravitational-wave event detected on 2019-05-21, characterized by a short burst lasting ~0.1 s with only a few cycles in the 30–80 Hz band.
• Detection pipelines like PyCBC Live and coherent WaveBurst confirmed the event with high signal-to-noise ratios and extremely low false-alarm rates.
• Bayesian inference estimates source-frame masses around 85 and 66 solar masses, underscoring its significance for intermediate-mass black hole formation and hierarchical merger scenarios.

GW190521 is the gravitational-wave transient detected on 2019-05-21 at 03:02:29 UTC by the Advanced LIGO Hanford (H1), Advanced LIGO Livingston (L1), and Advanced Virgo (V1) network during the first half of the O3 observing run. Under the standard quasi-circular binary-black-hole interpretation developed in the discovery analyses, it is a very short, very massive merger with source-frame component masses of about $85^{+21}_{-14}\,M_\odot$ and $66^{+17}_{-18}\,M_\odot$, producing a remnant of $142^{+28}_{-16}\,M_\odot$. Because a remnant near $142\,M_\odot$ lies in the conventional intermediate-mass-black-hole range $10^2\!-\!10^5\,M_\odot$, GW190521 was presented as the first strong gravitational-wave evidence for an intermediate-mass black hole. The event is also unusual because the observed signal lasted only about $0.1$ s in band and contained only about four cycles in the $30$–$80$ Hz range, so much of the inference is driven by merger and ringdown rather than a long inspiral (Collaboration et al., 2020, Szczepanczyk et al., 2020, Collaboration et al., 2020).

1. Observation and waveform morphology

GW190521 was publicly released in low latency within about 6 minutes as S190521g. Its morphology immediately set it apart from earlier LIGO/Virgo binary-black-hole events: the signal appeared as a short transient with fewer than 4 cycles in the $30$–$80$ Hz band, peaking near $66^{+17}_{-18}\,M_\odot$0–$66^{+17}_{-18}\,M_\odot$1 Hz, and showing an almost symmetric morphology rather than a textbook chirp (Szczepanczyk et al., 2020, Collaboration et al., 2020).

That morphology follows from the system’s large total mass. For very massive binaries, the merger occurs at low frequency, below or near the most contaminated low-frequency part of the detector band. The literature summarized here repeatedly emphasizes that intermediate-mass-black-hole binaries are expected to merge below $66^{+17}_{-18}\,M_\odot$2 Hz, while detector sensitivity is limited below $66^{+17}_{-18}\,M_\odot$3 Hz, so the instruments primarily observe the late inspiral, merger, and ringdown. A plausible implication is that GW190521 sits close to the regime in which generic burst morphology and compact-binary morphology become harder to separate cleanly.

The event’s mass scale was already suggested by its low-frequency content. In the coherent WaveBurst reconstruction, the peak signal frequency was about $66^{+17}_{-18}\,M_\odot$4 Hz; under the rough equal-mass estimate quoted in that analysis, a binary merging at that frequency should have a detector-frame total mass exceeding $66^{+17}_{-18}\,M_\odot$5 (Szczepanczyk et al., 2020). This is consistent with the later source-frame inference of a total mass around $66^{+17}_{-18}\,M_\odot$6 once cosmological redshift is taken into account through the standard relation $66^{+17}_{-18}\,M_\odot$7 (Collaboration et al., 2020).

2. Detection pipelines and significance

GW190521 was established as a confident detection by several search pipelines with different assumptions. In low latency, PyCBC Live found the trigger with network SNR 14.5 and false-alarm rate 1 in 8 yr, while coherent WaveBurst (cWB), configured for intermediate-mass-black-hole searches, found network SNR 15.0 and FAR < 1 in 28 yr. Other low-latency matched-filter pipelines, SPIIR and GstLAL, also identified consistent candidates (Collaboration et al., 2020).

The decisive significance came from offline analyses. The offline cWB search, using improved calibration and data-quality information, found GW190521 with network SNR 14.7 and estimated the background by time-shifting the two LIGO detectors against each other. The effective background corresponded to 9800 yr of observation and contained only two louder events, both consistent with accidental coincidence of glitches, yielding a false-alarm rate of 1 in 4900 yr. This is the key detection-significance result in both the Letter and the cWB-focused follow-up (Collaboration et al., 2020, Szczepanczyk et al., 2020).

The cWB analysis is methodologically important because it is a coherent excess-power search for short-duration transients and does not assume a compact-binary template bank. It operates on whitened data in a wavelet time-frequency representation, clusters excess-power pixels, and reconstructs coherent transients across the detector network with a constrained maximum-likelihood method. The event passed signal-independent coherence and morphology tests: cWB rejects events with $66^{+17}_{-18}\,M_\odot$8 or $66^{+17}_{-18}\,M_\odot$9, rejects blip-like events with $142^{+28}_{-16}\,M_\odot$0, and GW190521 had $142^{+28}_{-16}\,M_\odot$1 (Szczepanczyk et al., 2020).

This minimal-assumption detection mattered because the signal sat in a difficult regime for inspiral-dominated matched filtering. The cWB paper explicitly argues that the method does not depend on detailed waveform features such as higher-order modes, high mass ratios, misaligned spins, eccentricity, or deviations from general relativity, and is therefore especially useful for very massive binaries whose observable signal is short (Szczepanczyk et al., 2020).

3. Baseline source inference under the quasi-circular binary-black-hole interpretation

Under the working assumption that GW190521 was a quasi-circular binary black hole merger, Bayesian parameter estimation was performed on 8 s of data using NRSur7dq4, SEOBNRv4PHM, and IMRPhenomPv3HM. The Letter quotes the NRSur7dq4 results because it was the only model calibrated to precessing numerical simulations, although some posterior support extended outside its formal calibration region (Collaboration et al., 2020).

The standard source-property summary is:

Quantity	Inference
Component masses	$142^{+28}_{-16}\,M_\odot$2 and $142^{+28}_{-16}\,M_\odot$3
Total source-frame mass	$142^{+28}_{-16}\,M_\odot$4
Remnant	$142^{+28}_{-16}\,M_\odot$5
Distance and redshift	$142^{+28}_{-16}\,M_\odot$6
Merger rate of similar systems	$142^{+28}_{-16}\,M_\odot$7

These values were accompanied by a source-frame mass ratio $142^{+28}_{-16}\,M_\odot$8, chirp mass $142^{+28}_{-16}\,M_\odot$9, and radiated energy $142\,M_\odot$0 in the broader properties paper (Collaboration et al., 2020, Collaboration et al., 2020).

Spin information was informative but not decisive. The quoted effective aligned and precession parameters were $142\,M_\odot$1 and $142\,M_\odot$2, with $142\,M_\odot$3 and $142\,M_\odot$4, which the discovery Letter explicitly describes as only mild preferences (Collaboration et al., 2020). The properties paper similarly reports mild evidence for spin-induced orbital precession and notes that higher modes are slightly disfavored by $142\,M_\odot$5, although higher-mode-capable models still improve distance and inclination constraints (Collaboration et al., 2020).

The astrophysical significance of the baseline inference is tied to pair instability. The Letter quantified $142\,M_\odot$6, implying a 99.68% probability that the primary lies above $142\,M_\odot$7, and the companion properties paper states that the probability that at least one component lies in the $142\,M_\odot$8 gap is 99.0% with NRSur7dq4 (Collaboration et al., 2020, Collaboration et al., 2020). This is why GW190521 was presented not only as an IMBH-forming merger, but also as a challenge to ordinary isolated stellar-collapse channels.

4. Minimal-assumption reconstruction, ringdown, and waveform consistency

A central result of the cWB follow-up is that the template-free reconstruction is consistent with a binary-black-hole interpretation. Rather than comparing the observed cWB waveform directly to clean theoretical templates, the authors injected posterior samples from LALInference into nearby off-source data, passed them through the same cWB pipeline, and constructed confidence intervals that include posterior variability, detector noise, and cWB reconstruction error. Against a maximum-likelihood waveform, the network overlaps were 0.89 for NRSur7dq4 and 0.85 for SEOBNRv4, with corresponding $142\,M_\odot$9-values 7.9% and 1.0%; the measured interpretation was that NRSur7dq4 is consistent with the cWB reconstruction, while SEOBNRv4 is less consistent but not ruled out (Szczepanczyk et al., 2020).

The same paper also showed that template-free sky localization remained compatible with template-based localization for this unusual signal. In low latency, cWB estimated a 90% sky area of 1474 deg$10^2\!-\!10^5\,M_\odot$0, broadly consistent with the 1163 deg$10^2\!-\!10^5\,M_\odot$1 Bayestar localization. In follow-up with calibrated data, the cWB and Bayestar areas became 1653 deg$10^2\!-\!10^5\,M_\odot$2 and 765 deg$10^2\!-\!10^5\,M_\odot$3, respectively (Szczepanczyk et al., 2020).

Ringdown-only analyses provided an additional consistency check. Modeling the late-time signal as a single damped sinusoid starting 12.7 ms after the peak of the complex strain, the discovery Letter found $10^2\!-\!10^5\,M_\odot$4 with $10^2\!-\!10^5\,M_\odot$5. Imposing Kerr perturbation-theory relations for the $10^2\!-\!10^5\,M_\odot$6 mode gave $10^2\!-\!10^5\,M_\odot$7 and $10^2\!-\!10^5\,M_\odot$8, consistent with the full inspiral-merger-ringdown inference (Collaboration et al., 2020).

Later ringdown work argued that the apparent tension between earlier ringdown analyses and full IMR analyses was largely a mode-content problem. In that reinterpretation, ringdown models including both the fundamental $10^2\!-\!10^5\,M_\odot$9 and $0.1$0 quasinormal modes, and possibly a subdominant $0.1$1 mode, become fully consistent with NRSur7dq4 over a range of start times. That study proposed a precessional interpretation in which misalignment between the orbital angular momentum and the remnant spin can rotate inspiral-merger perturbations into strong $0.1$2 ringdown content, but it explicitly did not rule out eccentricity or other explanations (Siegel et al., 2023).

A complementary time-domain study of the quasi-circular interpretation reached a related conclusion by dissecting the observed cycles. It found that precession inference hinges on a quiet portion of the pre-merger data that is suppressed relative to the merger-ringdown, and that neither pre-merger nor post-merger data alone are the sole driver; rather, their combination is. In that picture, precession emerges as a mechanism to accommodate the lack of a stronger pre-merger signal in light of the observed post-merger (Miller et al., 2023).

5. Formation channels and astrophysical scenarios

The literature on GW190521 does not converge on a single formation channel. Instead, it uses the event as a stress test for several mechanisms capable of producing very massive black holes, black holes in or near the pair-instability gap, or both.

Hierarchical merger scenarios are among the most direct responses to the mass-gap problem. A backward-inference study that conditioned on the hierarchical-merger hypothesis found that the heaviest parental black hole of GW190521 would have mass $0.1$3, with 78% probability of lying in the $0.1$4 gap and 43% probability of lying above $0.1$5. It also found that the heaviest great-grandparent would be roughly $0.1$6, suggesting a possible ancestral chain from ordinary stellar-origin black holes to the observed system (Barrera et al., 2022).

Dense stellar environments provide several concrete realizations of that idea. In young massive star clusters, $0.1$7 binary-single scattering experiments with post-Newtonian dynamics found that about 0.17% of all simulated BHB mergers match simultaneously the component masses, $0.1$8, $0.1$9, remnant mass, and remnant spin of GW190521 within the $30$0 credible intervals under a fiducial low-spin assumption for first-generation black holes. Those matches consisted of 7 first-generation exchanged binaries and 5 second-generation BHBs, with an estimated merger-rate density $30$1 (Dall'Amico et al., 2021). In globular-cluster $30$2-body work, one object underwent a chain of seven binary mergers within 6 Gyr, reaching a final mass of $30$3, and the authors estimated that IMBH formation through hierarchical mergers occurs with probability $30$4 over the lifetime of a median-mass globular cluster in the no-recoil limit (Anagnostou et al., 2020).

Other studies asked whether direct stellar-origin formation could still work. One low- and zero-metallicity stellar-evolution analysis concluded that models with $30$5 to $30$6 have three properties favoring higher black-hole masses—lower post-main-sequence mass loss, more compact stars that disfavor binary stripping, and possible H–He shell interactions that reduce the CO core mass—and argued that direct-collapse/fallback black holes up to $30$7–$30$8 are supported, with uncertainties potentially pushing the limit to $30$9 (Farrell et al., 2020). A different isolated-binary study focused on the IMRI-like interpretation rather than the near-equal-mass interpretation and found that GW190521-like systems with $80$0 and $80$1 can form in metal-poor environments with $80$2, with predicted merger-rate density $80$3 within $80$4 (Cui et al., 2023).

More exotic channels were also explored. In the primordial black hole scenario, GW190521 was found to be difficult to explain without PBH accretion, because the required abundance would conflict with current bounds; with efficient pre-reionization accretion, the event becomes viable and GW190521-like mergers can occur at roughly the observed rate (Luca et al., 2020). Another study proposed that the components could be the central black holes of two ultra-dwarf galaxies of stellar mass $80$5, arguing that the ultra-dwarf galaxy major-merger rate at $80$6 can match the GW190521-like rate after a likely delay of $80$7 Gyr, though with major uncertainties in black-hole occupation fraction and low-mass-galaxy dynamics (Palmese et al., 2020).

6. Competing waveform interpretations, multimessenger claims, and continuing controversy

Because the observed signal is so short, GW190521 has also become a case study in posterior multimodality and waveform dependence. A high-mass-ratio reinterpretation argued that if one uses priors uniform in mass ratio and source-frame total mass, the posterior shifts toward an IMRI-like solution with $80$8 while a prior uniform in source-frame component masses yields a bimodal posterior with a low-mass-ratio mode and a high-mass-ratio mode, the latter centered near $80$9, $30$0. That analysis argued that the maximum-likelihood point lies in the high-$30$1 mode and that earlier sampling and prior choices may have suppressed it (Nitz et al., 2020).

A later detailed reanalysis with fourth-generation phenomenological waveform models confirmed that the posterior is genuinely multimodal, but substantially reduced the support for the most extreme unequal-mass solution. In that study, the robust structure was a near-equal-mass mode plus a secondary mode near $30$2, while the strong $30$3 interpretation was not retained as a significant mode. The same paper argued that support for both component masses being outside the pair-instability gap is drastically reduced relative to the earlier high-$30$4 claim (Estellés et al., 2021).

Dynamical alternatives beyond quasicircular inspiral remain active. An eccentric analysis using SEOBNRE found that the data prefer a signal with $30$5 over a non-precessing, quasi-circular signal, with $30$6, but also found that a moderately eccentric nonspinning GW190521-like binary can be mistaken for a quasi-circular precessing binary, and vice versa; the authors therefore concluded that one cannot confidently determine whether the source was precessing or eccentric, although either interpretation supports a dynamical origin (Romero-Shaw et al., 2020). A separate hyperbolic-capture study argued that a nonspinning dynamical-capture waveform is favored over a quasi-circular precessing-spin model by Bayes factors larger than 4300 to 1, with preferred source-frame masses $30$7 and $30$8, though that claim depends on a restricted waveform family and uncertain astrophysical prior odds (Gamba et al., 2021).

The proposed electromagnetic counterpart ZTF19abanrhr in AGN J124942.3+344929 is similarly unsettled. One quasi-circular reanalysis using NRSur7dq4 and a prior uniform in $30$9 inferred $80$0, $80$1, $80$2, and argued for a common-source odds ratio $80$3, which would imply a bright-siren estimate $80$4 when combined with GW170817 (Bustillo et al., 2021). A later GWTC-2.1-based analysis argued for even stronger association, with $80$5, corresponding to odds of about $80$6 under prior odds $80$7, and obtained $80$8 from GW190521 alone or $80$9 when using GW170817 as prior information (Morton et al., 2023). By contrast, the phenomenological reanalysis reported only mild coincidence odds, $66^{+17}_{-18}\,M_\odot$00–$66^{+17}_{-18}\,M_\odot$01, not enough for a confident association (Estellés et al., 2021). The contrast across these studies directly reflects the event’s dependence on waveform choice, prior choice, and posterior release.

The enduring legacy of GW190521 is therefore not only its status as the heaviest binary-black-hole merger detected to date and the first strong gravitational-wave evidence for an intermediate-mass black hole, but also its role as a boundary case for gravitational-wave inference. Across matched filtering, weakly modeled reconstruction, ringdown spectroscopy, hierarchical-growth studies, eccentric and hyperbolic alternatives, AGN-counterpart analyses, and multiband forecasts, the event repeatedly appears as a system in which the shortness of the observed waveform makes modeling assumptions especially consequential (Collaboration et al., 2020, Szczepanczyk et al., 2020, Nakano et al., 2021).