AI reads a supernova’s changing light to reconstruct the explosion

A July study trains AI to work backwards from a supernova’s bolometric light curve to the progenitor’s radius, ejected mass, explosion energy and nickel production.[1]

Assembling the light curve

A bolometric light curve estimates luminosity across wavelengths. Producing it requires more than following brightness through one coloured filter. Measurements in different bands have to be placed on a common timeline, converted into physical flux and integrated. Distance is needed to turn received flux into the source’s luminosity.[2]

Matt Nicholl’s Superbol documentation shows the practical distinction. The software can integrate only the observed bands or add blackbody estimates for missing ultraviolet and infrared light. It records the chosen filters and extrapolation methods. Those choices determine what information enters the curve before any neural network sees it.[2]

A radioactive clue in the fading light

The connection between late light and radioactive material has direct observational roots. In 1987, the Solar Maximum Mission detected gamma-ray lines from cobalt-56 in SN 1987A. Cobalt-56 is produced by nickel-56 decay and subsequently becomes iron-56. These observations supported the explanation that radioactive decay powers the supernova’s later optical emission.[3]

The AI study began with 205,000 simpler simulated light curves, then refined its model using 57 hydrodynamical simulations. It tested three observed 1987A-like events. Mean absolute differences from published model estimates were 3.9% for radius, 4.4% for ejected mass and about 4% for energy. These compare modelling results, rather than direct measurements of the vanished stars.[1]

Nickel was less consistent: one event’s estimate differed by 43.8%, with sparse coverage of its radioactive tail. The tested model excludes the first 30 days and has not been validated across all supernova classes.[1]

Report an error or suggest a correction