Revealing the Temporally Stable Bimodal Energy Distribution of FRB 20121102A with a Tripled Burst Set from AI Detections

TL;DR

Addressing incompleteness in repeating FRB burst catalogs caused by dedispersion and SNR thresholds, the paper introduces EDEN, an end-to-end dedispersion-agnostic AI detector that treats FRB detection as image recognition on dynamic spectra and employs Teacher-Student learning and Positive-Unlabeled training. When applied to FAST L-band data for FRB 20121102A, EDEN recovered 5,927 bursts, yielding a total isotropic energy of $5.94\times10^{41}$ erg, nearly doubling previous estimates and tripling the burst count. The enlarged dataset reveals that the bimodal energy distribution persists over time, indicating multiple stable emission mechanisms and a lack of temporal evolution linked to burst rate. Narrow-band, faint bursts detected by EDEN fill previously missed regions of parameter space, refining constraints on the magnetar energy budget and demonstrating a scalable AI-based detection approach for transient radio phenomena.

Abstract

Active repeating Fast Radio Bursts (FRBs), with their large number of bursts, burst energy distribution, and their potential energy evolution, offer critical insights into the FRBs emission mechanisms. Traditional pipelines search for bursts through conducting dedispersion trials and looking for signals above certain fluence thresholds, both of which could result in missing weak and narrow-band bursts. In order to improve the completeness of the burst set, we develop an End-to-end DedispersE-agnostic Nonparametric AI model (EDEN), which directly detect bursts from dynamic spectrum and is the first detection pipeline that operates without attempting dedispersion. We apply EDEN to archival FAST L-band observations during the extreme active phase of the repeating source FRB 20121102A, resulting in the largest burst set for any FRB to date, which contains 5,927 individual bursts, tripling the original burst set. The much enhanced completeness enables a refined analysis of the temporal behavior of energy distribution, revealing that the bimodal energy distribution remains stable over time. It is rather an intrinsic feature of the emission mechanisms than a consequence of co-evolving with burst rate.

Figures (11)

• Figure 1: Overview of the proposed model EDEN for FRB detection. Panel (a): The pipeline for training the model, which incorporates Teacher-Student Learning and Positive-Unlabeled Learning to enhance the diversity of detectable signals. Panel (b): A comparison between EDEN and the conventional Heimdall method for FRB detection. EDEN outperforms Heimdall in all aspects, particularly in terms of speed, precision, and the detection of high-time width (TW) and low-bandwidth (BW) signals.
• Figure 1: Analysis of the corresponding fraction of magnetar energy and total energy of the bursts. Panel (a): The fraction of magnetar energy corresponding to the total energy of various repeating FRB sources observed by FAST. Four sources are included, represented by blue bars (for data from the literature) and orange bars (for this work). The stacked bars for FRB 20201124A highlight the combined contributions from two observation periods. The total isotropic burst energy derived in this study accounts for 26% of the available energy of a typical magnetar, surpassing the corresponding energy obtained from other sources (including the sum of the two observation periods of FRB 20201124A). The values used in panel (a) are as follows: 26.2%, 0.4%, 24.4%, 20.8%. Panel (b): Cumulative energy distribution for two burst sets of FRB 20121102A. Panel (c): The distribution of residuals between the cumulative energy curves shown in panel (b). The black dashed line represents the cumulative residuals, while the red solid line represents the relative residuals per energy bin.
• Figure 2: Temporal energy distribution of the bursts. The top panel compares burst counts between the newly detected bursts and the previously detected 1652 bursts. The burst count for each epoch and the cumulative burst count for the two burst sets are shown separately in deep blue and green, with light blue-shaded regions indicating periods without observation. The middle panel displays the burst energy distribution, where blue dots represent the energy of the original 1652 bursts, red dots represent the energy of the newly detected bursts in each observing session, and the blue contour represents the two-dimensional KDE of the burst distribution. Pearson correlation coefficient analysis between subsets of bursts is shown with black lines and with shaded area being the 68% confidence interval. The vertical dashed line indicates the MJD at which the slope of the PCC curve is zero, marking the point where the datasets on either side exhibit the largest morphological difference. The right panel presents a histogram of isotropic burst energy, distinguishing bursts detected before (dark blue) and after (light blue) MJD 58740.
• Figure 2: Burst rate distribution of energy for FRB 20121102A bursts detected in a deep search. The histogram illustrates the burst rate distribution, with the dashed blue line representing the Log-Normal fit, the solid green line indicating the Cauchy fit, and the solid yellow line showing the power-law fit in the energy range $1\times 10^{38} \leq E \leq 8\times 10^{39}$. The solid red line denotes the combined Log-Normal and Cauchy fit. The vertical dashed red line marks the 90% completeness threshold. The upper subplot presents the cumulative count of burst events over the specified time period.
• Figure 3: Distribution of the bandwidth-to-central-frequency ratio and fluence for the bursts. In the left panel, new detections are represented by red dots, while the detections reported in Ref. Li_2021 are represented by blue dots. Three contour lines for each burst set partition the probability mass function into four regions. The density contours reveal two distinct clusters: one in the lower-left corner, predominantly consisting of new detections, characterized by lower fluence and narrower bandwidths, and another in the upper-right region, primarily composed of previously reported bursts, characterized by higher fluence and broader bandwidths. The right panel presents histograms for the two burst sets. Both panels show clustering in the narrow-bandwidth region, emphasizing the morphological differences between the two sets of detected bursts.