Skip to main content
Back to timeline
arXivSource publication:

RevolIACT would chain ten to fourteen small AI-driven Cherenkov telescopes across longitudes, lifting observable Fermi GBM bursts from 17% at CTAO's two sites to 61% of 4401

Synopsis

The bottleneck for sub-TeV transient astronomy is shifting from sensitivity to time coverage: by the mid-2030s, the only imaging atmospheric Cherenkov telescopes sensitive below 0.1 TeV may be the two CTAO sites, each able to watch a source for only a few hours per night. RevolIACT proposes ten to fourteen compact 5 m stand-alone stations spread in longitude, with a video vision transformer reading the full waveform of each Cherenkov camera so that one station approaches the needed sensitivity. For 2027, fourteen stations cover about 61% of the sky on average versus 17% for the two sites, and could have observed 61% of the 4401 Fermi GBM bursts at trigger versus 17%.

Source-provided article image: RevolIACT: a longitudinally distributed array of compact AI-driven Cherenkov telescopes for the time-domain sub-TeV sky
arXiv

Interpretation

The limiting factor for sub-TeV transient astronomy shifts from the flux sensitivity of a single telescope to time coverage: a station reaching about 15% of the Crab flux in one hour above 0.2–0.3 TeV would suffice for an array to catch most important transients. Current major imaging atmospheric Cherenkov telescope (IACT) systems in stereoscopic mode reach about 1% Crab units in 25 hours, while a single station with the standard analysis falls far short: a 5 m mirror misses the target by a factor of 17. Computed for 2027 in 30-minute steps, fourteen stations cover 60.8% of the sky on average versus 17.3% for the two CTAO sites; of the 4401 Fermi GBM bursts, fourteen stations could have observed 61.2% at trigger versus 16.6% for the two sites, about 3.7 times as many.

A single station becomes viable because a video vision transformer treats the full waveform movie of the Cherenkov camera as a space-time sequence instead of first reducing it to a few image parameters. The standard analysis compresses each event into Hillas image parameters and uses random forests to estimate particle type, energy and direction; with one viewing angle and few photons, those parameters carry too little information. On identical simulated events, a 5 m telescope with LST-1 photomultipliers drops the energy threshold from 0.73 TeV to 0.18 TeV under the transformer analysis and is 17 times more sensitive between 0.4 and 0.63 TeV; its one-hour integral sensitivity above 0.25 TeV is 29% Crab units, and 22% for a 6 m dish.

The array's value lies in stitching together observing windows across longitudes, not in having several telescopes record the same shower. CTAO's two sites span a limited range of longitude and together observe a source only 17% of the time on average; one site sees a celestial-equator source for 11.6% of the year. Near opposition, fourteen stations observe a source around the clock in 18 of 20 tested cases; for a nearby weak burst like GRB 190829A, they could have covered it continuously from trigger to about one day after.

Perspective

The conclusions rest on simulations and catalogue statistics: per-station sensitivity comes from simplified, idealised simulations with a perfect mirror, no electronic noise and a proton-only background, while coverage and burst fractions are computed for 2027 with zenith angles below 60 degrees in astronomical darkness, and the sites are illustrative rather than negotiated host locations.

The transformer analysis has not yet been applied to real data, and domain shift between simulation and the real telescope is the principal risk; the 50% quantum efficiency is simplified as wavelength-independent; the site layout and telescope parameters have not been optimised jointly, and the authors name confirming the performance on real data with a prototype station as the decisive next step.