QPEs: Repeating X-ray Flares in Galactic Nuclei

• Quasi-periodic eruptions (QPEs) are repeating soft X-ray flares from galactic nuclei defined by hours-to-days recurrence, thermal-dominated spectra, and hysteretic temperature–luminosity loops.
• Observations show diverse profiles with fast-rise/slow-decay asymmetries and durations ranging from minutes to days, probing accretion disk dynamics and post-TDE processes.
• Dynamical models—including EMRI-driven disk crossings, Roche-lobe overflow, and precessing slim disks—explain QPE timing, shock energetics, and links to newly formed compact accretion flows.

Searching arXiv for recent QPE papers to ground the article in the current literature. Quasi-periodic eruptions (QPEs) are a recently discovered class of repeating X-ray flares from galactic nuclei, typically emerging from otherwise quiescent or only weakly active systems. Across the currently studied sample, QPEs are characterized by large-amplitude soft X-ray outbursts recurring on timescales from hours to days, with thermal-dominated spectra and a stable or slowly evolving quiescent component between flares. The phenomenon is now studied in connection with extreme mass-ratio inspirals (EMRIs), tidal disruption event (TDE) remnants, Roche-lobe overflow, and precessing super-Eddington flows, and it has become a timing-based probe of accretion-disk structure, orbital dynamics deep in galactic nuclei, and the conditions under which newly formed compact accretion flows appear (Suzuguchi et al., 1 Sep 2025, Arcodia et al., 20 Jun 2025).

1. Observational phenomenology

Since the first detection in GSN 069 in 2019, roughly ten systems or candidates have been found, and later population summaries describe 12 bona-fide QPE sources. In the canonical phenomenology, recurrence intervals span from a few hours up to days, durations are typically $\sim10^3$–$10^5\,{\rm s}$, duty cycles are of order $10\,\%$, and peak X-ray luminosities are $L_X\sim10^{41}$–$10^{43}\,{\rm erg\,s^{-1}}$. Their spectra are extremely soft compared with most active nuclei, with thermal fits yielding $kT\sim100$–$200\,{\rm eV}$, while population summaries of peak flare temperatures quote $kT\sim30$–$120\,{\rm eV}$ (Suzuguchi et al., 1 Sep 2025, Arcodia et al., 20 Jun 2025).

The class is observationally diverse. eRO-QPE1 has a mean period $P=18.5\,{\rm h}\pm2.7\,{\rm h}$, burst duration $10^5\,{\rm s}$0, and strongly asymmetric fast-rise/slow-decay profiles; eRO-QPE2 has $10^5\,{\rm s}$1, $10^5\,{\rm s}$2, and a nearly symmetric Gaussian-like profile (Arcodia et al., 2021). XMMSL1 J024916.6–041244 showed two symmetric soft X-ray flares separated by $10^5\,{\rm s}$3, with harder sub-bands peaking earlier and remaining briefer, extending the class into a TDE-candidate host (Chakraborty et al., 2021). eRO-QPE3 and eRO-QPE4 preserved the characteristic thermal-like quiescent and eruptive spectra, with eruptions showing a harder rise than decay and hysteretic temperature–luminosity evolution (Arcodia et al., 2024).

Energy-dependent evolution is now a defining empirical constraint in at least some sources. In eRO-QPE1, eruptions peak later and are broader at lower energies, and they also start earlier at lower energies; the hardness-ratio versus count-rate plane shows an anti-clockwise loop, implying that the rise is harder than the decay at a fixed total count rate (Arcodia et al., 2022). AT2019qiz and SDSS J133519.91+072807.4 (“Ansky”) likewise show hysteresis in luminosity–temperature space, together with blackbody-radius evolution during individual eruptions (Nicholl et al., 2024, Hernández-García et al., 9 Apr 2025).

The observed parameter space has expanded substantially. eRO-QPE5, at $10^5\,{\rm s}$4, has $10^5\,{\rm s}$5, $10^5\,{\rm s}$6, and integrated energy $10^5\,{\rm s}$7, placing it at the high end of the known population in duration, recurrence time, and black-hole mass (Arcodia et al., 20 Jun 2025). Ansky extends the class to extreme $10^5\,{\rm s}$8-day QPEs with $10^5\,{\rm s}$9-day durations, a $10\,\%$0-day super-period, and very large integrated energies, demonstrating that QPE phenomenology is not confined to the short-period, few-hour regime (Hernández-García et al., 9 Apr 2025).

2. QPEs, tidal disruption events, and newly formed accretion flows

A major development in the field is the growing empirical connection between QPEs and TDEs. Recent observations confirm that some QPEs occur in galactic centers that experienced a TDE a few years earlier, and AT2019qiz provides the first spectroscopically confirmed TDE with later repeating QPEs. In that system, nine eruptions were detected between December 2023 and March 2024 with a mean recurrence time $10\,\%$1, flare durations of $10\,\%$2–$10\,\%$3, a peak blackbody temperature $10\,\%$4, and a quiescent disk component described by a color-corrected multi-temperature disk with $10\,\%$5 (Nicholl et al., 2024).

Other sources support the same association by indirect means. eRO-QPE3 shows eruptions superposed on a decaying quiescent flux, providing further evidence for a preceding TDE. XMMSL1 J024916.6–041244 had previously been identified as a TDE candidate after an $10\,\%$6 flux rise over a ROSAT upper limit followed by an approximate $10\,\%$7 decay over a decade, and the later disappearance of its QPEs suggests strong source evolution on post-TDE timescales (Arcodia et al., 2024, Chakraborty et al., 2021). Long-term declines in eruption amplitudes in GSN 069 and eRO-QPE1 are also discussed within the context of fading post-TDE disks (Pasham et al., 2024).

This TDE connection is significant because it links QPE timing directly to disk formation. In the EMRI + TDE disk picture, a secondary object on a bound orbit repeatedly punctures a remnant accretion disk created by a preceding TDE, generating shocks and expanding optically thick ejecta. Early in a TDE, however, the disk is expected to be in a super-Eddington slim-disk state with $10\,\%$8, and the predicted QPEs differ strongly from the currently known population: durations of $10\,\%$9–$L_X\sim10^{41}$0, temperatures of $L_X\sim10^{41}$1–$L_X\sim10^{41}$2, and duty cycles typically $L_X\sim10^{41}$3 (Suzuguchi et al., 1 Sep 2025). In the Strubbe & Quataert slim-disk model, the fallback and disk structure evolve as

$L_X\sim10^{41}$4

while in a self-similar viscously spreading slim disk the surface density follows

$L_X\sim10^{41}$5

The slim-disk phase persists until $L_X\sim10^{41}$6 at $L_X\sim10^{41}$7–$L_X\sim10^{41}$8, after which a standard thin disk is expected (Suzuguchi et al., 1 Sep 2025).

Ansky broadens the discussion beyond classical TDEs. Its optical evolution is much shallower than a canonical TDE decline, it lacks broad lines, and it appears consistent with a turn-on AGN candidate in which QPEs emerged only after soft X-rays appeared. This suggests that QPEs are not linked solely to TDEs but more generally to newly formed compact accretion flows (Hernández-García et al., 9 Apr 2025).

3. Dynamical and radiative models

The most extensively developed model class is the EMRI plus disk-interaction framework. In this picture, a stellar or compact secondary on an inclined orbit around the SMBH crosses an accretion disk once or twice per orbit, shocks the gas, ejects an optically thick cloud or “plasma ball,” and produces a soft X-ray flare as the ejecta expand, cool, and radiate. The recurrence time is then set by the orbital dynamics:

$L_X\sim10^{41}$9

or, for two disk crossings per orbit,

$10^{43}\,{\rm erg\,s^{-1}}$0

This framework has been worked out in analytic and semi-analytic form by several groups, including Franchini et al. and the orbit-inference analysis of GSN 069 (Zhou et al., 2024, Franchini et al., 2023).

Within this class, the shock energetics depend on the nature of the secondary. For a star of radius $10^{43}\,{\rm erg\,s^{-1}}$1, the shocked mass scales as $10^{43}\,{\rm erg\,s^{-1}}$2, whereas for a stellar-mass black hole the energy deposition is described through gravitational drag in the supersonic limit. Franchini et al. model the flare as emission from an adiabatically expanding, initially optically thick gas cloud expelled from the disk plane at each impact; by varying $10^{43}\,{\rm erg\,s^{-1}}$3, $10^{43}\,{\rm erg\,s^{-1}}$4, $10^{43}\,{\rm erg\,s^{-1}}$5, disk mass, and inclinations, the model reproduces the range of recurrence times, amplitudes, and duty cycles observed in GSN 069, eRO-QPE1, eRO-QPE2, and RX J1301.9+2747 (Franchini et al., 2023). In the early slim-disk phase of a TDE, the same disk-crossing physics predicts much shorter, harder QPEs:

$10^{43}\,{\rm erg\,s^{-1}}$6

because the disk is thicker and the breakout conditions differ from those in the later thin-disk regime (Suzuguchi et al., 1 Sep 2025).

A second model family invokes periodic mass transfer rather than disk puncture. Zhao et al. propose Roche-lobe overflow from hydrogen-deficient post-AGB stars on eccentric EMRI orbits, where low-density helium envelopes are stripped near periapsis and the rise and decay times reflect dynamical and viscous timescales, respectively (Zhao et al., 2021). Krolik & Linial instead consider a main-sequence star on a moderately eccentric orbit, with periodic Roche overflow, magnetic-stress-aided angular-momentum loss, mildly relativistic shocks near the SMBH, and flare-enhanced irradiation feedback that modulates both mass transfer and timing jitter (Krolik et al., 2022). King argues that all known QPEs fit naturally into a picture of accretion from white dwarfs on highly eccentric, gravitational-wave-decaying orbits, with stable mass transfer at pericenter and alternating long-short eruptions arising from the donor’s forced oscillatory response (King, 2022).

A third line of work dispenses with a secondary perturber entirely. In the Lense–Thirring precession model of super-Eddington flows, a tilted, thick disk plus wind around a spinning SMBH precesses as a solid body, and the observed X-ray flux increases when the wind-cone is oriented at lower inclination. The precession period is written as

$10^{43}\,{\rm erg\,s^{-1}}$7

with $10^{43}\,{\rm erg\,s^{-1}}$8 obtained from the ratio of integrated Lense–Thirring torque to total angular momentum of the flow. For $10^{43}\,{\rm erg\,s^{-1}}$9, $kT\sim100$0–$kT\sim100$1, and $kT\sim100$2–$kT\sim100$3, the predicted periods are of order hours to days, comparable to QPE recurrence times (Middleton et al., 10 Jan 2025).

The main controversy is therefore not whether QPEs are periodic accretion phenomena, but which clock dominates: an orbiting secondary, episodic mass transfer, rigid or differential precession of a tilted flow, or some combination. Several observational papers argue that simple radiation-pressure instability implementations struggle with the observed fast-rise/slow-decay profiles, duty cycles, and quiescent luminosities of at least some sources, especially eRO-QPE1 and eRO-QPE2 (Arcodia et al., 2021, Nicholl et al., 2024, Pasham et al., 2024). That does not eliminate all disk-based mechanisms, but it narrows the viable parameter space.

4. Timing diagnostics, orbital inference, and population scalings

QPE timing is unusually constraining because it is often possible to map recurrence directly onto orbital or precessional dynamics. In AT2019qiz, taking $kT\sim100$4 and $kT\sim100$5 yields $kT\sim100$6 if one flare is produced per orbit, while two disk crossings per orbit imply $kT\sim100$7 (Nicholl et al., 2024). In the toy-model treatment of post-TDE disks perturbed by an orbiting star, the recurrence time scales as

$kT\sim100$8

placing hour-to-day QPEs at radii of tens to thousands of gravitational radii depending on SMBH mass (Mondek et al., 30 Mar 2026).

The most explicit orbital inference so far comes from GSN 069. By fitting individual flare start times with either a plasma-ball model or a phenomenological rise-and-decay template, and then fitting the resulting crossing times under Schwarzschild geodesic motion, one finds at $kT\sim100$9

$200\,{\rm eV}$0

with $200\,{\rm eV}$1 at $200\,{\rm eV}$2 confidence level (Zhou et al., 2024). The low eccentricity and $200\,{\rm eV}$3 are consistent with the wet EMRI formation channel and are described as incompatible with classical loss-cone capture or Hills-binary-disruption alternatives (Zhou et al., 2024).

Long-baseline monitoring adds a second layer of diagnostics. In eRO-QPE1, recurrence times vary between approximately $200\,{\rm eV}$4 and $200\,{\rm eV}$5 days but show no detectable secular drift over three years, with a linear regression slope consistent with zero within $200\,{\rm eV}$6 and $200\,{\rm eV}$7 (Pasham et al., 2024). On shorter baselines, an O–C analysis of 92 eruptions reveals a possible sinusoidal timing modulation with a super-period of approximately $200\,{\rm eV}$8 days. Interpreting this as rigid nodal precession of a misaligned TDE-fed disk yields joint constraints on SMBH spin and disk surface-density slope through

$200\,{\rm eV}$9

where $kT\sim30$0 is the angular-momentum-weighted Lense–Thirring precession frequency (Chakraborty et al., 2024).

At the population level, the best-established scaling is between duration and recurrence time. For nine bona-fide QPE sources, the fit

$kT\sim30$1

gives

$kT\sim30$2

with intrinsic scatter $kT\sim30$3 (Arcodia et al., 20 Jun 2025). No significant correlation is found between either $kT\sim30$4 or $kT\sim30$5 and $kT\sim30$6, nor between QPE peak temperature and timing properties (Arcodia et al., 20 Jun 2025). This is important because simple star–disk collision models predict $kT\sim30$7 at fixed opacity, stellar size, and SMBH mass, whereas models powered by stellar-debris streams around the orbiter predict a slope closer to unity; the observed slope is therefore described as favoring the latter (Arcodia et al., 20 Jun 2025).

5. Secular evolution, source states, and amplitude changes

QPEs are not stationary clocks. Multi-year monitoring of eRO-QPE1 shows peak fluxes declining from $kT\sim30$8 to $kT\sim30$9 over three years, corresponding to a decrease in $120\,{\rm eV}$0 from $120\,{\rm eV}$1 to $120\,{\rm eV}$2 (Pasham et al., 2024). Over 3.5 years, the source displayed non-monotonic evolution in inferred blackbody radius and energy output, disappearance within NICER detectability in October 2023, and reappearance by January 2024 at a luminosity $120\,{\rm eV}$3 fainter and temperature $120\,{\rm eV}$4 cooler than at discovery (Chakraborty et al., 2024). Such behavior is consistent with substantial evolution in the underlying accretion flow or impact geometry.

Toy-model calculations for a post-TDE disk with $120\,{\rm eV}$5 and surface density $120\,{\rm eV}$6 provide a compact amplitude law,

$120\,{\rm eV}$7

where $120\,{\rm eV}$8 encodes stellar ablation and $120\,{\rm eV}$9 captures orbit geometry (Mondek et al., 30 Mar 2026). In that treatment, amplitude ratios of $P=18.5\,{\rm h}\pm2.7\,{\rm h}$0–$P=18.5\,{\rm h}\pm2.7\,{\rm h}$1 over a few years arise naturally if the first monitored epoch occurs years to decades after the TDE, and stellar ablation steepens the decline for more massive main-sequence stars (Mondek et al., 30 Mar 2026). This suggests that at least part of the long-term QPE evolution can be driven by secular decay of a TDE-born disk rather than rapid orbital decay.

AT2019qiz provides a direct example of a post-TDE system entering a QPE phase years after disruption, while Ansky shows that similarly structured eruptions can arise in a newly accreting massive black hole without a classical TDE signature. Ansky exhibited a $P=18.5\,{\rm h}\pm2.7\,{\rm h}$2 optical brightening in December 2019, soft X-rays appeared in February 2024, and from May to July 2024 a NICER campaign confirmed QPEs recurring every $P=18.5\,{\rm h}\pm2.7\,{\rm h}$3 days with a $P=18.5\,{\rm h}\pm2.7\,{\rm h}$4-day cascade-like super-period (Hernández-García et al., 9 Apr 2025). The implication drawn in that work is that QPEs are associated not exclusively with TDE aftermaths but with newly formed, compact accretion flows more broadly (Hernández-García et al., 9 Apr 2025).

6. Multiwavelength properties, searches, and multimessenger prospects

Outside the soft X-ray band, QPEs are comparatively quiet. A compilation of 12 bona-fide QPE sources finds compact, weak radio sources in 5/12 systems, no signatures of strong AGN activity via luminous radio jets, no radio variability correlated with individual X-ray QPEs, and no significant correlations between radio luminosity and X-ray QPE properties such as flare duration, recurrence time, or integrated energy (Goodwin et al., 17 Jun 2025). The detected radio luminosities, compactness, and spectral indices are instead described as consistent with outflows from a recent TDE or related accretion event, including in sources that otherwise look AGN-like (Goodwin et al., 17 Jun 2025). This is a useful constraint on models that would launch a powerful radio-emitting outflow on every X-ray cycle.

Systematic discovery methods are also improving. A neural-network search over simulated and archival XMM-Newton light curves used 14 time-domain variability measures, achieved test accuracies above $P=18.5\,{\rm h}\pm2.7\,{\rm h}$5 on simulated data and, after threshold optimization, up to $P=18.5\,{\rm h}\pm2.7\,{\rm h}$6 on observational data, then scanned 83,531 4XMM detections. The pipeline recovered known QPE sources but found no new bona fide QPEs after manual vetting, with major contaminants coming from stellar flares, X-ray binaries, and background artifacts (Webbe et al., 2023). This indicates that QPEs remain intrinsically rare or are difficult to recover in broad-band archival products without energy-resolved screening.

The multimessenger case is promising but observationally challenging. In EMRI plus disk-crossing models with compact secondaries, the relevant GW frequencies are typically $P=18.5\,{\rm h}\pm2.7\,{\rm h}$7–$P=18.5\,{\rm h}\pm2.7\,{\rm h}$8, only marginally overlapping LISA’s optimal band. Under a minimal flare-emission model, one predicts at most one joint X-ray and GW detection during LISA’s nominal mission lifetime, whereas extending GW sensitivity below $P=18.5\,{\rm h}\pm2.7\,{\rm h}$9 would increase the prospects by an order of magnitude (Suzuguchi et al., 15 May 2025). QPEs therefore remain plausible EM counterparts to EMRIs, but the detectability window is narrow.

The observational strategy that follows from current models is correspondingly specific. For early post-TDE slim-disk QPEs, the advocated program is rapid X-ray follow-up of optically discovered TDEs starting within weeks of disruption, high-cadence monitoring with exposures of order $10^5\,{\rm s}$00 every few hours over months, and broad-band coverage up to $10^5\,{\rm s}$01–$10^5\,{\rm s}$02 (Suzuguchi et al., 1 Sep 2025). A detection in that phase would directly constrain when and how debris circularizes and forms a radiatively efficient accretion flow. More broadly, continued long-term monitoring of established QPE sources is required to discriminate among orbital evolution, precession, disk depletion, and irradiation-feedback interpretations of the observed period wander, amplitude decay, and intermittent source disappearance (Chakraborty et al., 2024, Pasham et al., 2024).

QPEs now occupy a distinct position among nuclear transients: they are repeating, spectrally soft, strongly structured in time, increasingly linked to post-TDE or otherwise newly formed accretion flows, and sufficiently regular to permit dynamical inference. The present literature does not yet support a single universal engine, but it has already established QPEs as a sensitive probe of disk formation, EMRI dynamics, and the transient accretion states of low-mass galactic nuclei (Suzuguchi et al., 1 Sep 2025, Zhou et al., 2024, Arcodia et al., 20 Jun 2025).