🔭 DRAFT RESEARCH PROPOSAL · TIME-DOMAIN SURVEY · ZTF + RUBIN/LSST · PROPOSE-NOT-BUILD

Machine-Learning Off-Nuclear TDE Search for Wandering and Cluster IMBHs

A host-agnostic tidal-disruption-event classifier tuned one black-hole-mass rung down from the SMBH regime, applied to ZTF archival data and the Rubin/LSST alert stream to search globular clusters, stripped nuclei, and other off-nuclear environments for an IMBH-mass disruption · Working draft · August 2026

Note on scope: this proposal specifies a search design (classifier features, target list, rate expectations) and does not implement the classifier. It adapts a demonstrated method (Stein et al. 2026) to a lower mass regime and a different host population; it is not new instrumentation.

1. Scientific Rationale

1.1 The precedent: an off-nuclear TDE found by dropping the host-galaxy prior

Stein et al. (2026) retrained ZTF's TDE classifier to use only light-curve shape, discarding the host-galaxy centroid information every prior TDE search relied on to reject non-nuclear transients. Pointed at the full ZTF alert stream without a nuclear-offset cut, the reprocessed classifier surfaced TDE 2025abcr: a tidal disruption projected 9.08 ± 0.02 kpc from the nucleus of its host galaxy, evidence for a black hole ejected by a merger or resident in an accreted dwarf satellite. It is the second optically selected off-nuclear TDE and the first found by a search designed to find one. The result matters here for method, not distance: a classifier freed from the nuclear-centroid assumption finds black holes wherever they sit, including permanently off-nuclear populations that were never candidates for a nucleus-anchored search.

1.2 One mass rung down: globular clusters are the permanently off-nuclear population

TDE 2025abcr's black hole is off-nuclear because it wandered there. Globular clusters do not need to wander: ω Cen and systems like it sit tens of kiloparsecs from any galactic nucleus as a matter of formation history, hosting a candidate IMBH population (Papers A–E, this series; Häberle et al. 2024; Bañares-Hernández et al. 2025) an order of magnitude to several orders below the SMBH masses the ZTF classifier was trained on. The same host-agnostic design principle applies without modification: a classifier keyed on light-curve shape rather than host position or host mass can be retuned to the disruption timescales and luminosities expected at the IMBH end and pointed at globular-cluster fields, stripped-nucleus remnants, and extragalactic GC systems the way Stein et al. pointed it at the general ZTF stream.

1.3 The discriminant

IMBH TDEs are not simply fainter, slower SMBH TDEs; the fallback timescale scales with black hole mass, so a solar-type disruption that unfolds over weeks to months around a SMBH compresses toward days around an IMBH, with a correspondingly higher peak temperature for a given peak luminosity (MacLeod et al. 2016). The 2026 IMBH-mass TDE candidate AT 2018cqh gives an observational anchor for the class: a persistently soft X-ray blackbody at ~63 eV, distinct from the harder, longer-lived signatures typical of nuclear SMBH TDEs. A sharper, mass-gated discriminant exists for one debris type specifically: white-dwarf disruptions are only possible below a black hole mass of ~10⁵ M☉ (MacLeod et al. 2016), since above that threshold the WD is swallowed inside the innermost stable circular orbit before tidal forces can unbind it, leaving no optical transient. A white-dwarf-disruption lightcurve (a fast, hot flare, potentially with a thermonuclear component) is therefore a smoking gun for the IMBH mass range specifically; it cannot originate from a nuclear SMBH TDE at all. Cluster IMBH mass estimates in this series (2×10⁴–6×10⁴ M☉, Papers A/E) sit comfortably inside that window.

1.4 Rate expectations: a rate limit, not a promised detection

Paper E's feasibility envelope quotes a per-cluster TDE recurrence horizon of ≳10⁷ yr, consistent with loss-cone rate calculations for IMBHs in evolved globular clusters of ~10⁻⁸–10⁻⁷ yr⁻¹ per cluster for main-sequence disruptions (Stone & Metzger 2016 methodology), with white-dwarf disruptions rarer by a further factor of ~30–100.

Milky Way globular clusters (~150, Harris catalog): N × T × rate MS-star TDEs, 10 yr baseline: 150 × 10 × [1e-8, 1e-7] ≈ [1.5e-5, 1.5e-4] events WD TDEs, 10 yr baseline: 150 × 10 × [1e-10, 3e-9] ≈ [1.5e-7, 4.5e-6] events

Over a ZTF-plus-Rubin decade, the expected yield from the Milky Way's own cluster population is far below one event, and no forced optimism is available here. Extending the search to extragalactic globular-cluster systems (M31's ~500 clusters; the thousands hosted by nearby massive ellipticals) raises the cluster count by one to two orders of magnitude, but optical TDE detectability at those distances is set by survey depth, not cluster count alone, and a rigorous per-system estimate needs the actual Rubin single-visit and coadd depths rather than a scaling guess; this proposal does not attempt one. The accurate framing is a rate limit, not a detection forecast: a null result over a stated survey baseline converts directly into an upper bound on the cluster-IMBH occupation fraction, a quantity that feeds Paper C's decision tree and Paper A's candidate table exactly as a positive detection would, and does so on a timescale the recurrence rate guarantees a search will actually reach, unlike waiting for a single ω Cen event.

2. Search Design

2.1 Classifier adaptation

ComponentAdaptation from Stein et al. (2026)
Host-position priorRetained: drop the nuclear-offset requirement (as in the precedent), but add a positive cross-match against known GC catalogs and stripped-nucleus candidates as the positional feature, replacing "near a nucleus" with "coincident with a dense old stellar system"
Light-curve rise/decay timescaleRe-tuned toward the days-scale fallback expected at 10³–10⁵ M☉ rather than the weeks-to-months SMBH range
Peak color / blackbody temperatureRe-tuned toward the hotter, softer end consistent with AT 2018cqh and IMBH-mass fallback models
White-dwarf-disruption flagNew: a fast-hot subclass tag for candidates whose lightcurve is consistent with WD disruption, since any confirmed instance is a mass-range-defining detection on its own

2.2 Data sources

No claim is made here about any not-yet-released dataset or future data product; both surveys named are producing data now.

2.3 Target prioritization

ω Cen is the first target given its existing candidate-IMBH status and the depth of characterization in this series, followed by other massive, dynamically evolved Milky Way globular clusters and the best-characterized stripped nuclei (ultracompact dwarfs with independently estimated central black hole masses). Extragalactic GC systems are a stretch goal contingent on Rubin coadd depth, not a baseline commitment.

2.4 Follow-up tier

Same as the series' existing proposals: any flagged candidate, and a white-dwarf-disruption flag above all, triggers spectroscopic confirmation to rule out contaminants (stellar flares, supernovae in a coincidentally nearby background source, AGN variability in a chance superposition) before any claim is made.

3. Expected Outcomes

ScenarioOutcomeOCS Implication
No candidates over the survey baselineUpper limit on cluster-IMBH occupation fraction from the null loss-cone rateConsistent with the low expected yield derived above; strengthens the case for waiting on ω Cen's own ≳10⁷ yr recurrence rather than a wide search
Ordinary (main-sequence) off-nuclear TDE in a clusterFirst direct dynamical-timescale confirmation of a cluster IMBH via disruptionIndependent evidence channel alongside pulsar timing and stellar kinematics for the host cluster
White-dwarf-disruption candidate confirmed spectroscopicallyMass-range-defining detection: the host black hole is below ~10⁵ M☉ by the disruption physics itselfDirect IMBH-regime confirmation independent of any dynamical mass modeling

4. Work Plan

YearQMilestoneDeliverable
1Q1–Q2GC and stripped-nucleus catalog cross-match pipeline; classifier feature re-derivationTarget list; adapted feature set
1Q3–Q4Archival ZTF reprocessing over the target listArchival candidate list (if any)
2n/aForward search over the Rubin alert streamRolling candidate alerts
2–3n/aSpectroscopic follow-up of any flagged candidateConfirmed detection or contaminant rejection
3Q3–Q4Occupation-fraction limit from the null result (if no confirmed detection)Submitted to ApJ

5. Budget

ItemCost (USD)Notes
PI (0.2 FTE, 3yr)225,000Time-domain / ML transient classification expertise required
Postdoc (1.0 FTE, 3yr)285,000Classifier adaptation, archival reprocessing, follow-up coordination
Graduate student (1.0 FTE, 3yr)165,000Catalog cross-match pipeline; rate-limit statistics
Spectroscopic follow-up time60,000Director's-discretion / ToO time for candidate confirmation
ZTF data access + HPC70,000Archival reprocessing; Rubin alert-stream ingestion
Travel + publications50,0001 open-access paper
Fringe + overhead (~30%)250,000
Total~$1.11M (3 yr)

6. References

  1. Stein, R., Carney, J., Ward, C., et al. (2026). TDE 2025abcr: A Tidal Disruption Event in the Outskirts of a Massive Galaxy. ApJL, 1006, L57. arXiv:2602.10180
  2. MacLeod, M., Guillochon, J., Ramirez-Ruiz, E., Kasen, D., & Rosswog, S. (2016). Optical Thermonuclear Transients from Tidal Compression of White Dwarfs as Tracers of the Low End of the Massive Black Hole Mass Function. ApJ, 819, 3. arXiv:1508.02399
  3. Fragione, G., Leigh, N. W. C., Ginsburg, I., & Kocsis, B. (2018). Tidal Disruption Events and Gravitational Waves from Intermediate-Mass Black Holes in Evolving Globular Clusters Across Space and Time. ApJ, 867, 119. doi:10.3847/1538-4357/aae486
  4. Stone, N. C. & Metzger, B. D. (2016). Rates of Stellar Tidal Disruption as Probes of the Supermassive Black Hole Mass Function. MNRAS, 455, 859. arXiv:1410.7772
  5. Häberle, M., et al. (2024). Fast-moving stars in ω Cen. Nature, 631, 285. arXiv:2405.06015
  6. Bañares-Hernández, A., et al. (2025). Pulsar timing upper limit on the ω Cen IMBH. A&A, 693, A104.
  7. Gezari, S. (2021). Tidal Disruption Events. ARA&A, 59, 21.
Working draft · August 2026 · ← Return to omegacentauri.me

Relevant tools

Tidal Capture Rate
Loss-cone disruption rates by species
Rubin Alert Throughput
Nightly alert-stream rate budget
IMBH Evidence Dashboard
Live multi-constraint overview