Quasi-Periodic Eruptions in Galactic Nuclei

• Quasi-Periodic Eruptions (QPEs) are repeating soft X-ray bursts from galactic nuclei, characterized by large amplitudes, thermal-like spectra, and recurrence intervals from minutes to days.
• Recent observations by eROSITA and XMM-Newton have revealed diverse burst profiles and timing behaviors, challenging single-parameter models of accretion physics.
• Studies of QPEs offer practical insights into accretion disk dynamics and the transient conditions following tidal disruption events, advancing multi-messenger astronomy.

Quasi-periodic eruptions (QPEs) are repeating soft X-ray bursts from the nuclei of galaxies, observed as large-amplitude flares above a faint quiescent level and recurring on timescales from tens of minutes to several days; the broader literature summarized here also includes systems with mean recurrence times of approximately 48 hours and $\sim 4.5$ days (Arcodia et al., 20 Jun 2025, Nicholl et al., 2024, Hernández-García et al., 9 Apr 2025). Their spectra are generally thermal-like, commonly fitted with single-temperature blackbodies, redshifted blackbodies, or disk-blackbody components, with characteristic peak temperatures of order $kT\sim 100$–$200$ eV in much of the known population, although hotter regimes have been proposed for early super-Eddington post-TDE phases (Suzuguchi et al., 15 May 2025, Suzuguchi et al., 1 Sep 2025). QPEs have emerged as a distinct class of nuclear transients because they combine large amplitudes, soft spectra, recurrence, and host-galaxy environments that differ from classical unobscured AGN, while also admitting multiple physical interpretations, notably EMRI-related disk interactions, episodic mass transfer, and geometrical modulation by Lense–Thirring precession.

1. Discovery history and observational census

The modern observational history of QPEs began with a very small sample. In 2021, only two such sources were known before a blind and systematic eROSITA search yielded two further galaxies, eRO-QPE1 and eRO-QPE2, demonstrating that QPEs also occur in previously quiescent galaxies and not only in nuclei already classified as active (Arcodia et al., 2021). Subsequent blind searches with SRG/eROSITA added eRO-QPE3 and eRO-QPE4, and then eRO-QPE5, the fifth source found through a dedicated blind search in the all-sky survey data (Arcodia et al., 2024, Arcodia et al., 20 Jun 2025).

eRO-QPE5 widened the known parameter space substantially. Its measured eruption duration is $t_{\rm dur}\simeq 0.64\pm0.11$ d, its recurrence time is $t_{\rm recur}\simeq 3.70\pm0.02$ d, its integrated X-ray energy per eruption is $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$ erg, and its estimated black-hole mass is $$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$. With a spectroscopic redshift of $z=0.1155$, it is the most distant QPE source discovered to date (Arcodia et al., 20 Jun 2025).

The census expanded along several independent observational routes. A blind algorithm-assisted search of the XMM-Newton catalogue identified XMMSL1 J024916.6–041244 as a QPE candidate associated with a TDE candidate, with 1.5 QPE-like flares in a 2006 pointed observation and no residual QPEs in a 2021 follow-up (Chakraborty et al., 2021). AT2019qiz became the first confirmed optically selected TDE with subsequent repeating X-ray QPEs (Nicholl et al., 2024). SDSS J133519.91+072807.4 (“Ansky”) then broadened the phenomenology again by showing extreme $\sim 4.5$-day QPEs in a turn-on AGN candidate, implying that QPEs are not linked solely to tidal disruption events but more generally to newly formed accretion flows (Hernández-García et al., 9 Apr 2025).

Detection strategies have evolved accordingly. eROSITA blind searches exploit soft-band sensitivity, a large field of view, and repeated short visits to each field every $\sim 6$ h, with light curves rebinned into “eRO days” and processed by algorithms such as eRebin to flag flare-shaped profiles spanning several consecutive eRO days (Arcodia et al., 20 Jun 2025). XMM-Newton archival searches have used QATS on $kT\sim 100$0 serendipitous light curves (Chakraborty et al., 2021). Machine-learning approaches have also been applied: a neural network trained on 14 variability measures achieved test accuracies of 0.938–0.950 on simulated light curves and, after threshold optimization, accuracy up to 0.984 on observational data (Webbe et al., 2023).

2. Timing, spectral evolution, and burst morphology

QPE light curves are not monolithic. Early source definitions emphasized hour-scale recurrences, amplitudes up to $kT\sim 100$1–100 above quiescence, and soft X-ray confinement below $kT\sim 100$2–2 keV (Arcodia et al., 2022, Arcodia et al., 2021). Later discoveries extended the recurrence scale from the few-hour regime to days. eRO-QPE3 shows $kT\sim 100$3 h recurrence, AT2019qiz has a mean $kT\sim 100$4 h with $kT\sim 100$5 h, eRO-QPE5 repeats every $kT\sim 100$6 d, and Ansky shows a characteristic peak-to-peak recurrence of $kT\sim 100$7 ks, approximately $kT\sim 100$8 d, together with a $kT\sim 100$9 d super-period (Arcodia et al., 2024, Nicholl et al., 2024, Arcodia et al., 20 Jun 2025, Hernández-García et al., 9 Apr 2025).

Burst profiles also vary. eRO-QPE1 established an asymmetric archetype with fast rise and slower exponential decay, whereas eRO-QPE2 showed more symmetric Gaussian-like eruptions (Arcodia et al., 2021). The phenomenology later became more complex: eRO-QPE1 can alternate between isolated bursts and chaotic mixtures of overlapping bursts with very different amplitudes, implying that any common trigger must be able to generate both regular and complex timing behavior (Arcodia et al., 2022). J0249 exhibited nearly two symmetric soft flares separated by $200$0 ks, and eRO-QPE4 showed three eruptions with FWHM $200$1–2.1 ks separated by $200$2 h and $200$3 h (Chakraborty et al., 2021, Arcodia et al., 2024).

A defining property is energy-dependent spectral timing. In eRO-QPE1, lower-energy photons peak later, burst rise and decay broaden toward lower energy, and eruptions were found for the first time to begin earlier at lower energies; the hardness ratio versus count rate traces a pronounced anti-clockwise loop, so that the spectrum is harder on the rise than on the decline at fixed count rate (Arcodia et al., 2022). eRO-QPE3 and eRO-QPE4 further showed that the peak temperature leads the peak luminosity, producing hysteresis in $200$4 versus $200$5 (Arcodia et al., 2024). AT2019qiz displayed an anticlockwise hysteresis loop in the $200$6–$200$7 plane and an emitting radius $200$8 at peak, expanding through the eruption (Nicholl et al., 2024).

Thermal-like spectroscopy recurs across the class. In eRO-QPE3, quiescent emission was fit with $200$9 eV in eRASS1, while XMM-Newton captured bursts peaking at $t_{\rm dur}\simeq 0.64\pm0.11$0 eV; in eRO-QPE4, the quiescent disk had $t_{\rm dur}\simeq 0.64\pm0.11$1 eV and the QPE peak reached $t_{\rm dur}\simeq 0.64\pm0.11$2 eV (Arcodia et al., 2024). J0249 showed a stable cool disk component at $t_{\rm dur}\simeq 0.64\pm0.11$3–82 eV plus a hotter flare component at $t_{\rm dur}\simeq 0.64\pm0.11$4 eV, while AT2019qiz reached $t_{\rm dur}\simeq 0.64\pm0.11$5 eV with $t_{\rm dur}\simeq 0.64\pm0.11$6 erg s$t_{\rm dur}\simeq 0.64\pm0.11$7 (Chakraborty et al., 2021, Nicholl et al., 2024).

A common misconception is that QPEs are strictly periodic and achromatic. The observational record shows the opposite: recurrence times wander, profiles can overlap or change qualitatively, and the spectral evolution within a burst is strongly energy dependent (Arcodia et al., 2022, Pasham et al., 2024).

3. Host galaxies, nuclear environments, and links to TDEs

Many QPE hosts are low-mass galaxies with black holes in the $t_{\rm dur}\simeq 0.64\pm0.11$8–$t_{\rm dur}\simeq 0.64\pm0.11$9 range. Early summaries emphasized nuclei of low-mass galaxies with stellar masses $t_{\rm recur}\simeq 3.70\pm0.02$0 and $t_{\rm recur}\simeq 3.70\pm0.02$1–$t_{\rm recur}\simeq 3.70\pm0.02$2 (Arcodia et al., 2022). This remains characteristic of much of the sample, but the known population now extends to more massive systems, including eRO-QPE4 with $t_{\rm recur}\simeq 3.70\pm0.02$3–$t_{\rm recur}\simeq 3.70\pm0.02$4 and eRO-QPE5 with $t_{\rm recur}\simeq 3.70\pm0.02$5 (Arcodia et al., 2024, Arcodia et al., 20 Jun 2025).

Optical spectroscopy and multiwavelength diagnostics often indicate weak or absent canonical AGN signatures. eRO-QPE1 showed no optical emission lines and was classified as a passive galaxy; eRO-QPE2 was classified as H II/star-forming; eRO-QPE3 and eRO-QPE4 lacked broad optical lines, IR torus signatures, UV/X-ray power-law coronae, and ATCA radio emission at the reported levels (Arcodia et al., 2021, Arcodia et al., 2024). For eRO-QPE5, no previous or concurrent optical-IR transient was found in archival photometric data, and the optical spectrum was described as almost featureless (Arcodia et al., 20 Jun 2025).

A central development in the field has been the strengthening QPE–TDE connection. eRO-QPE3 showed eruptions on top of a decaying quiescent flux that faded from $t_{\rm recur}\simeq 3.70\pm0.02$6 to non-detection over successive eROSITA/XMM epochs, reinforcing the idea that QPEs can ignite in a transient TDE remnant disk (Arcodia et al., 2024). J0249 had already been identified as a TDE candidate, and its long-term X-ray decay was well fit by a $t_{\rm recur}\simeq 3.70\pm0.02$7 law characteristic of TDE fallback (Chakraborty et al., 2021). AT2019qiz then provided the first confirmed optically selected TDE with subsequent repeating X-ray QPEs years later (Nicholl et al., 2024).

At the same time, the host class is not reducible to classical TDE remnants alone. Ansky had remained optically stable for $t_{\rm recur}\simeq 3.70\pm0.02$8 yr, then brightened in 2019 and entered five years of AGN-like stochastic optical variability before the appearance of extreme QPEs. Its phenomenology was therefore interpreted as evidence that QPEs can arise in any newly formed accretion disk, not only in the aftermath of classical TDEs (Hernández-García et al., 9 Apr 2025). A plausible implication is that the decisive condition is the existence of a compact, recently assembled accretion flow rather than a unique trigger class.

4. Population trends and long-term evolution

The growing source sample now permits limited but nontrivial population-level tests. The clearest current empirical correlation is between eruption duration and recurrence time. In the rest frame, a Bayesian linear regression gives

$t_{\rm recur}\simeq 3.70\pm0.02$9

with $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$0, $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$1 when both quantities are in days, and intrinsic scatter $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$2 dex. This corresponds to an approximately constant duty cycle of $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$3 (Arcodia et al., 20 Jun 2025).

By contrast, no significant correlation has been found between either $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$4 or $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$5 and black-hole mass, and neither timescale shows any correlation with the peak QPE temperature $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$6; the fitted slopes are indistinguishable from zero at $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$7, and the Bayesian evidence does not favor linear trends over constant models (Arcodia et al., 20 Jun 2025). This is consistent with earlier source-based arguments that luminosity and recurrence are not simply linked. eRO-QPE3 combined the longest recurrence time then known with the faintest peak luminosity, excluding a predictive $E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$8–$E_{\rm QPE}\simeq (3.4\pm0.7)\times10^{47}$9 correlation in that sample (Arcodia et al., 2024).

Individual sources also evolve on multi-year baselines. Swift and XMM-Newton monitoring of eRO-QPE1 over three years found recurrence times varying between 0.6 and 1.2 days, no detectable secular trend in recurrence, and a steady decline of peak flux from $$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$0 to $$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$1 erg s$$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$2 cm$$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$3 (Pasham et al., 2024). The 3.5-year NICER/XMM campaign then revealed still more structure: complex non-monotonic evolution in energy output and inferred emitting area, disappearance of the QPEs within NICER detectability in October 2023, reappearance by January 2024 at luminosity $$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$4 fainter and temperature $$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$5 cooler than at discovery, and a possible $$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$6-day modulation of timing residuals (Chakraborty et al., 2024).

Long-term amplitude decline has become a recurrent theme. Recent analytical work showed that a post-TDE disk with $$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$7 can reproduce factor $$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$8–10 decreases in QPE amplitude over several years if the first monitored epoch occurs years to a few decades after disruption (Mondek et al., 30 Mar 2026). This suggests that at least part of the secular evolution may be tracing the decay of a transient accretion structure rather than the orbital clock alone.

5. Physical interpretations and theoretical tensions

Theoretical models fall into several broad families. The largest body of work links QPEs to an orbiting stellar-mass object or its debris interacting with an accretion disk around the SMBH. In EMRI disk-crossing models, a compact secondary on an inclined orbit periodically plunges through a rigidly precessing or otherwise evolving disk, shocking gas and producing an expanding, initially optically thick cloud whose soft X-ray emission reproduces fast-rise, slow-decay flares and alternating intervals in some sources (Franchini et al., 2023, Zhou et al., 2024). In a related hydrodynamic picture, QPEs are powered by streams of stellar debris liberated in prior pericenter passages, yielding a scaling $$M_{\rm BH}\simeq (2.8^{+1.7}_{-0.7})\times10^7\,M_\odot$$9; for the small dynamic range in $z=0.1155$0 among known QPEs this gives nearly $z=0.1155$1, closer to the observed duration–recurrence slope (Arcodia et al., 20 Jun 2025).

A second group of models invokes episodic Roche-lobe overflow or mass transfer at pericenter. One variant considers white dwarfs in highly eccentric orbits whose orbital evolution is driven by gravitational radiation and for which mass transfer remains highly stable; this framework was proposed to explain duty cycles, alternating long/short eruptions, and a continuum from HLX-1-like year-long cycles to shorter-period systems (King, 2022). Another identifies hydrogen-deficient post-AGB stars as promising donors, using MESA stellar models to show that their He envelopes can reproduce QPE periods, energetics, and asymmetric light curves while also implying EMRI gravitational-wave emission in the millihertz band (Zhao et al., 2021). A third proposes a main-sequence star on a moderately eccentric orbit, with irradiation feedback from the flare itself augmenting mass transfer and generating the observed jitter in period and duty cycle (Krolik et al., 2022).

A distinct interpretation is geometrical rather than impact-driven. In the Lense–Thirring precession model, a spinning SMBH misaligned with a thick, super-Eddington accretion flow forces the disk and its wind funnel to precess as a solid body. The observer then sees a QPE when the line of sight sweeps into the X-ray-bright funnel. Analytic inversions and GR-RMHD comparisons suggest that periods of hours to days can be obtained for $z=0.1155$2–$z=0.1155$3, $z=0.1155$4–10)$z=0.1155$5 or higher, and moderate spin, while also accommodating wandering recurrence times through changes in $z=0.1155$6 (Middleton et al., 10 Jan 2025).

Accretion-flow instability models remain in contention but face strong source-specific challenges. For eRO-QPE1 and eRO-QPE2, radiation-pressure driven thin-disk limit cycles were found unable to account simultaneously for the observed periods, durations, amplitudes, duty cycles, profiles, and timescales (Arcodia et al., 2021). For AT2019qiz, radiation-pressure instability models were also described as predicting recurrence times $z=0.1155$7 years and therefore inconsistent with the observed $z=0.1155$8 h recurrence (Nicholl et al., 2024). Broader instability-based proposals, including shock-bubble oscillations and disk tearing, can qualitatively reproduce day-scale variability in eRO-QPE1, but require tuned boundary conditions or source-specific assumptions (Pasham et al., 2024).

No single model currently accounts for all observed features across the full class. The present empirical situation is narrower: the duration–recurrence relation disfavors a purely arbitrary phenomenology, the lack of simple $z=0.1155$9 or temperature scaling constrains one-parameter explanations, and the association with TDEs or newly assembled disks places the engine within a transient nuclear accretion environment (Arcodia et al., 20 Jun 2025, Nicholl et al., 2024, Hernández-García et al., 9 Apr 2025).

6. Radio behavior, search methodology, and multi-messenger prospects

Radio observations show that QPEs are not ordinarily accompanied by strong radio jets or flare-driven radio outflows. A compilation of 12 bona-fide QPE sources found compact, weak radio sources in 5/12 systems, no luminous radio jets, no radio variability linked to the X-ray eruptions, and no statistically significant correlations between radio luminosity and QPE X-ray properties, SMBH mass, or host properties (Goodwin et al., 17 Jun 2025). The radio spectra and luminosities were instead found to be consistent with outflows produced by a recent TDE or accretion event. This constrains models in which each X-ray eruption would launch a substantial radio-emitting outflow.

The observational selection function is already shaping the field. eROSITA has proved highly successful because its sensitivity and large field of view, combined with repeated scans of the full sky, deliver many repeated exposures across $\sim 4.5$0 galaxies and enable blind detection of $\sim 4.5$1 QPE candidates (Arcodia et al., 20 Jun 2025). Machine-learning pipelines further reduce the cost of searching large archives, and future soft X-ray missions with wide-area, high-cadence survey modes are explicitly recommended to extend the sample to fainter and shorter-period systems (Webbe et al., 2023, Arcodia et al., 20 Jun 2025).

If the EMRI interpretation is correct in at least a subset of sources, QPEs become a multi-messenger population. The collision paradigm naturally places the system in the LISA/TianQin band once gravitational-wave emission dominates the orbital evolution (Arcodia et al., 20 Jun 2025, Zhao et al., 2021). A dedicated detectability analysis found that X-ray observations are most effective at orbital frequencies up to roughly $\sim 4.5$2 mHz, whereas LISA is chiefly sensitive above about $\sim 4.5$3 mHz; because the sensitivity windows overlap only marginally, the expected yield is at most one joint detection during LISA’s nominal mission lifetime, although extending GW sensitivity into the sub-millihertz regime would raise the joint-detection संभावना by an order of magnitude (Suzuguchi et al., 15 May 2025).

The immediate observational program follows from this. Sensitive wide-area soft X-ray surveys are needed to enlarge the population; coordinated UV/X-ray follow-up of optically selected TDEs is needed to catch nascent disks and possible QPE onset; and continued joint X-ray–gravitational-wave framework development is needed in anticipation of EMRI detections (Arcodia et al., 20 Jun 2025). This suggests that QPEs are not only a transient class but also a potentially precise probe of the structure, evolution, and geometry of newly formed accretion flows in galactic nuclei.