Rapid radio bursts: machine learning suggests a population of sources that emit only once

By applying simulation-based inference to the second CHIME catalogue (3,641 bursts), two researchers have concluded that there are indeed ‘unique’ sources, distinct from repetitive sources.

Canada's CHIME radio telescope has released a second catalogue of 3,641 fast radio bursts (FRBs). Of these, 981 come from 83 repeating sources, but only 2.3% of the detected sources have been seen more than once. It remains to be seen whether all sources would eventually repeat themselves with enough observation, or whether there is a distinct population of point transmitters.

Eight models put to the test

Sloane Sirota and Duncan R. Lorimer compared eight population models to the catalog using simulation inference, a machine learning method that does not require explicit likelihood functions. Half of the models included truly unique sources, and they are the ones that the data clearly favors. An independent random forest classifier confirms this result, and these models replicate the catalog's hyperactive repeating sources on their own without having been trained on them.

The authors set upper limits of 36,000 repetitive sources and 19 unique sources per day observable in the CHIME sky. They predict that the share of repetitive sources will fall below 2% within five years, and then to 0.5% as the catalog fills up with unique detections.