AT 2024qfm: a luminous fast blue optical transient at a redshift of z = 0.2267 identified by Lasair-ZTF

AT 2024qfm is a luminous fast blue optical transient (LFBOT) discovered at a redshift of z = 0.2267 ± 0.0002, identified in the Zwicky Transient Facility (ZTF) alert stream using a custom filter (FastFinder) in the Lasair broker that flags flux gradients over time. Its ultraviolet-to-optical luminosity and rapid 13-day fade closely match AT 2018cow, the prototype LFBOT. The transient was first detected on 24 July 2024 (MJD 60515.4), with a peak absolute magnitude of Mg ≈ −21.0 and a decline rate of dg/dt ≈ 0.3 mag d−1. Spectroscopy with Gemini/GMOS and NOT/ALFOSC revealed a blue, largely featureless continuum with only narrow host-galaxy nebular emission lines (Hα, Hβ, [O iii], [N ii], [S ii]) at a common redshift of z = 0.2267 ± 0.0002, with no broad transient-intrinsic features. The host galaxy has a stellar mass of log10(M⋆/M⊙) = 10.22+0.04−0.05, a star-formation rate of 1.43+0.45−0.31 M⊙ yr−1, and a mass-weighted age of 7.07+0.84−0.84 Gyr, placing it toward the massive end of known LFBOT hosts with the oldest stellar population. AT 2024qfm is spatially offset by 1.1″ (4.0 kpc) from the host, intermediate between the small offsets of earlier LFBOTs and the large offsets of AT 2022tsd and AT 2023fhn. The rest-frame ugri absolute magnitudes and colours of AT 2024qfm are remarkably similar to AT 2018cow and AT 2020xnd, implying similar effective temperatures and emitting-region radii. The paper also discusses the discovery method: FastFinder flagged AT 2024qfm based on a fast decline (dg/dt = 0.37 ± 0.14 mag d−1) and blue colour, and also flagged another LFBOT candidate, AT 2024kth, at a photometric redshift of z = 0.16 ± 0.06. The authors estimate that these two discoveries would push the LFBOT rate estimates based on only 3 ZTF events up by 66%. Looking to the future, the paper highlights that the Rubin Observatory’s Legacy Survey of Space and Time (LSST) will increase the effective LFBOT survey volume tenfold relative to ZTF, out to z ≲ 0.6–0.8, potentially detecting about 1 LFBOT per month rather than the current rate of just over 1 per year. However, the challenge remains in identifying them early enough to trigger multi-wavelength follow-up, with the latency in confirming LFBOT candidates (e.g., five days for AT 2024qfm from initial detection to spectroscopic confirmation) being a key limitation. The paper concludes that AT 2024qfm is firmly within the LFBOT class, almost identical to AT 2018cow, and that its discovery demonstrates both the potential and limitations of finding fast transients in LSST survey data with the Lasair broker.

Improvements to AI Systems:

1. Adaptive Anomaly Detection for Fast Transients

• The AI can be enhanced to use multi-epoch flux-gradient filters (like FastFinder) that track both rapid decline rates (e.g., >0.3 mag/day) and color evolution (blue) in real-time alert streams.

• Improved system: Automatically flags LFBOT candidates within hours of first detection, reducing latency from 5 days to <1 day by prioritizing objects with high gradient-to-noise ratios and host-galaxy offsets.

2. Photometric Redshift Estimation for Rare Classes

• Train a dedicated regression model on known LFBOTs (AT 2018cow, AT 2020xnd, AT 2024qfm) to estimate redshifts from sparse, multi-band light curves, even without spectroscopy.

• Improved system: Provides real-time redshift probability distributions for fast transients, enabling immediate volume-rate calculations and follow-up prioritization (e.g., distinguishing z 0.16 from z 0.23).

3. Host-Galaxy Property Inference from Offset and Continuum

• Use a neural network that ingests transient position, host photometry, and nebular emission-line ratios to predict stellar mass, star-formation rate, and stellar age—even when the transient outshines the host.

• Improved system: Automatically classifies a transient as LFBOT-like based on host age >5 Gyr and mass >10 10 M⊙, and flags unusual offsets (e.g., 4 kpc) for targeted follow-up.

4. Early Spectroscopic Trigger Optimization

• Implement a reinforcement-learning agent that balances survey cadence, telescope availability, and candidate probability to schedule spectroscopic confirmation within 24–48 hours of detection.

• Improved system: Dynamically re-queues Gemini/NOT-class telescopes for high-value candidates, cutting confirmation latency from 5 days to <2 days, and increasing multi-wavelength coverage (UV, X-ray, radio).

5. Rate Estimation with Small-Number Statistics

• Develop a Bayesian hierarchical model that updates LFBOT volumetric rates using each new discovery (e.g., 3→5 events) while accounting for survey selection biases and redshift completeness.

• Improved system: Provides live, uncertainty-aware rate estimates (e.g., 66% increase) and predicts LSST detection yields (e.g., 1/month) with confidence intervals, aiding survey strategy.

6. Cross-Survey Broker Integration

• Build a federated learning system that shares FastFinder-style filters across brokers (Lasair, ANTARES, Fink) to detect fast-declining, blue transients in real time, using standardized feature vectors.

• Improved system: Enables simultaneous detection in ZTF and LSST streams, with automatic cross-matching to known host galaxies and legacy data, reducing false positives and improving early alerts.