Fast radio bursts are millisecond-duration radio pulses whose frequency drifts during the burst — sweeping upward or downward across the band. The drift rate varies from burst to burst, but the variation was assumed to reflect a single emission mechanism with a range of parameters. More drift or less drift, from the same source region.
This paper finds two distinct drift rate populations in FRB 20240114A, a hyperactive repeater observed by FAST. Among 233 upward-drifting burst clusters, unsupervised machine learning identifies a subpopulation of 45 bursts with mean drift rates 2.5× higher than the rest (246 vs 98 MHz/ms). The bimodality is statistically unambiguous: Gaussian mixture modeling yields a BIC improvement of 297, with a gap between modes exceeding 11σ.
The fast-drifting subpopulation differs systematically: 7% lower peak frequencies, 29% shorter durations, and distinct clustering in multi-dimensional feature space. The differences are not just in drift rate — the bursts look different in multiple independent dimensions simultaneously.
The interpretation is geometric: two spatially separated emission regions in the magnetosphere, each producing upward-drifting bursts but at different characteristic rates. The magnetar has two voices, and they speak at different speeds. The burst-to-burst variation that previously looked like noise in a single mechanism was actually two mechanisms superimposed.