---
title: "Telling source repetition from path effects with timing statistics"
description: "Compare waiting times, observing windows and periodic folding with what a plasma lens can do, and find the one regime where the two are degenerate."
canonical_url: "https://sandboxphysics.com/stories/fast-radio-burst-echoes/repetition-statistics-source-or-path/"
language: "en"
page_type: "story"
alternate_languages:
  zh-CN: "https://sandboxphysics.com/zh/stories/fast-radio-burst-echoes/repetition-statistics-source-or-path/"
last_modified: "2026-11-14"
---

[Fast radio bursts: repeats, echoes and plasma lenses](https://sandboxphysics.com/stories/fast-radio-burst-echoes/)

# Telling source repetition from path effects with timing statistics

Sandbox Physics · 2026-11-14

The repeats of FRB 20180916B come from the source. Its 104 catalogued bursts are spaced from 36 seconds to 113 days, they cluster more than random bursts seen through the same observing windows, and they fold on a 16.35-day cycle. A single plasma lens can add at most two echoes at fixed delays, and an echo a day late would need gas so dense that it absorbs the burst. The path can still decide when bursts are bright enough to see. Within one burst, sub-bursts milliseconds apart are where a lens and the source genuinely look alike.

Before this chapter, read [images, gains and delays of a Gaussian plasma lens](https://sandboxphysics.com/stories/fast-radio-burst-echoes/gaussian-plasma-lens-images/). You also need the exponential distribution and the idea of a likelihood.

## Waiting times and the observing window

Take the arrival times of FRB 20180916B from the second CHIME/FRB catalogue (release 20251231). Drop rows flagged as non-nominal telescope operation and keep one event per catalogue name: 104 events between MJD 58410 and 60126. The 103 gaps between them run from 36.3 s to 112.7 days, with a median of 14.9 days.

CHIME is a transit telescope. It sees this source only while the sky carries it through the beam. In our data, 95 percent of arrivals lie within 0.42 hours of transit, so each day offers a window of about 0.84 hours. That window shapes the gaps: 22 are shorter than an hour (two bursts in the same transit), 21 lie between an hour and a day, which in practice means close to one sidereal day of 23.93 hours, and 60 are longer than a day. Any statistic of the gaps mixes the source with the telescope.

## Clustering: the Weibull shape

If bursts arrived independently at a constant rate $r$, the gaps would follow the exponential distribution $r\,e^{-r\delta}$. Oppermann, Yu and Pen proposed the Weibull distribution, which adds a shape parameter $k$:

$$
W(\delta\mid k,r)=\frac{k}{\delta}\left[\delta\,r\,\Gamma(1+1/k)\right]^k\exp\left\{-\left[\delta\,r\,\Gamma(1+1/k)\right]^k\right\}.
$$

It reduces to the exponential for $k=1$; $k<1$ means bursts come in clusters, with more very short and very long gaps than chance gives. [On the non-Poissonian repetition pattern of FRB121102, Eq. 2](https://arxiv.org/abs/1705.04881v2)

Fitting this density to all 103 gaps gives $k=0.42$. That number is not the source's clustering on its own, because the windows also make short and long gaps. Oppermann and colleagues handle windows exactly in their likelihood; we use a simpler check. We simulate random bursts at a constant rate, seen only through the same 0.84-hour daily windows inside the published 5.4-day activity windows, and fit them the same way. Over 200 trials, $k$ falls between 0.60 and 0.84. The observed 0.42 lies below that range, so the bursts are more clustered than the windows alone explain.

## Folding on 16.35 days

The CHIME/FRB Collaboration found a period of $16.35\pm0.15$ days, with phase zero at MJD 58369.18 and activity within $\pm2.7$ days of phase 0.5. [Periodic activity from a fast radio burst source](https://arxiv.org/abs/2001.10275v5) For an arrival at time $t$, the phase is

$$
\phi=\frac{t-t_0}{P}\bmod1 .
$$

Folded this way, 91 percent of the catalogue events fall inside the published window. The period's uncertainty matters here. A period error $\sigma_P$ shifts the phase at time $t$ by $(t-t_0)\,\sigma_P/P^2$. At the last event that is $1756\times0.15/16.35^2=0.99$ cycles, so the 2020 ephemeris alone cannot phase these later bursts. A Rayleigh scan, which measures how strongly the phases bunch,

$$
Z=\frac{2}{N}\left|\sum_{i=1}^{N}e^{2\pi i\,t_i/P}\right|^2,
$$

peaks at 16.33 days over the range 15.5 to 17.5 days. With the same phase zero, 97 percent of events then fall within 2.7 days of phase 0.5. Catalogue times are topocentric and referenced to 400 MHz, while the paper used barycentric times at infinite frequency. The differences are minutes, small against a 16-day cycle, so this fold is a consistency check of the published period.

## What a lens would need

The previous chapter showed that a Gaussian lens gives at most three images. Their delays are fixed for a fixed geometry and change only as the lens drifts. So a lens can add at most two echoes to each burst; it cannot produce 104 bursts with gaps spread over a factor of $3\times10^5$.

Could a lens at least make the longest gaps? The longest delay with gain near one comes from images near $u^2=1/2$, where the bracket in the gain is exactly one and

$$
t\approx t_{\rm DM0}\,\frac{\alpha}{2e}\qquad(\alpha\gg1).
$$

Geometric optics needs the lens to be much wider than the Fresnel scale; take $a=3r_{\rm F}$. Fixing the delay then fixes the column $\mathrm{DM}_\ell$ and the density $n_e=\mathrm{DM}_\ell/a$. Free-free absorption follows Eqs. 25 and 26 of Cordes and colleagues,

$$
\mathrm{EM}=a\,n_e^2,\qquad \tau_{\rm ff}\approx3.4\times10^{-7}\,\nu_{\rm GHz}^{-2.1}\,T_4^{-1.32}\,\mathrm{EM},
$$

with $\mathrm{EM}$ in $\mathrm{pc\,cm^{-6}}$ and $T_4$ the temperature in units of $10^4\,\mathrm K$. [Lensing of fast radio bursts by plasma structures in host galaxies, Eqs. 15, 25 and 26](https://arxiv.org/abs/1703.06580v1)

**Worked example.** A one-day echo at 600 MHz, with the lens 1 kpc from the source and $T_4=1$, needs $a=0.031\,\mathrm{AU}$, $\mathrm{DM}_\ell=2.05\,\mathrm{pc\,cm^{-3}}$ and $n_e=1.35\times10^7\,\mathrm{cm^{-3}}$. Then $\mathrm{EM}=2.8\times10^7\,\mathrm{pc\,cm^{-6}}$ and $\tau_{\rm ff}=27.5$, so $e^{-27.5}\approx10^{-12}$ of the echo gets out. With the lens 1 pc from the source, $\tau_{\rm ff}=871$. Because $\tau_{\rm ff}\propto D^{-1/2}$ at fixed delay, even 10 kpc leaves $\tau_{\rm ff}\approx9$. For the median gap of 14.9 days the optical depths are 412 and 13000. Only the shortest gap, 36 s, survives absorption ($\tau_{\rm ff}=0.012$ at 1 kpc), and it would need $\alpha\approx4\times10^5$ with the source $1.7\times10^5$ lens radii off axis.

## Where the path does matter

A lens moving across the line of sight changes the gain with time. For their example lens, Cordes and colleagues find a gain spike of about a day as the line of sight approaches, a trough of low gain lasting months, then a second spike. A survey with a fixed fluence threshold would record more bursts during spikes and fewer in troughs. Lensing can therefore modulate when bursts are detected, without making the bursts.

Absorption can do the same job. Lyutikov, Barkov and Giannios propose that the 16-day cycle of FRB 20180916B is orbital: a neutron star circles an O or B star, and the companion's wind absorbs the bursts except near one orbital phase. [FRB-periodicity: mild pulsars in tight O/B-star binaries](https://arxiv.org/abs/2002.01920v5) That is a model; it predicts phase-dependent absorption and dispersion that later data can test.

## The degenerate case: sub-bursts

Inside one event, the catalogue lists sub-bursts 0.10 to 68 ms after the first. Those offsets overlap lens delays exactly. Sub-bursts from repeaters usually drift down in frequency; lens echoes do too when the dispersive delay dominates, because it grows as $\nu^{-2}$. Three tests can separate them:

- A lens echo's delay scales as $\nu^{-2}$ across the band; an intrinsic sub-burst need not.
- Each lens copy crosses a different column, so its apparent dispersion measure differs, by up to $\mathrm{DM}_\ell$.
- Hessels and colleagues note that a single dominant lens is needed to keep the drift direction fixed, and that a changing lens geometry should make the drift change rate and sign. [FRB 121102 bursts show complex time-frequency structure, Sec. 4.3](https://arxiv.org/abs/1811.10748v1)

## Lab tasks

1. In [the plasma-lens echo Lab](https://sandboxphysics.com/labs/frb-plasma-lens-echoes/), open the delay chart. Drag the source until the latest echo is as late as you can make it, and compare it with the observed sub-burst offsets and repeat gaps.
2. Raise the lens column to its maximum and read the lens-maximum line. How far is it from the shortest observed repeat gap?
3. Choose the caustic preset and then move the lens slowly. Describe how the summed gain changes, and connect it to a burst rate that rises and falls.

## Exercises

1. Show that the Weibull density reduces to $r\,e^{-r\delta}$ when $k=1$.
2. During an active window CHIME detects about 0.9 bursts per hour from this source. If bursts were independent, what is the chance that the next burst comes within 36 s?
3. Show that an image at $u^2=1/2$ has gain one, and that its delay tends to $t_{\rm DM0}\,\alpha/(2e)$ for large $\alpha$.
4. Using $\tau_{\rm ff}\propto D^{-1/2}$, estimate the effective distance at which a one-day echo would have $\tau_{\rm ff}=1$.

Answers

1. With $k=1$, $\Gamma(2)=1$, so $W=\delta^{-1}(\delta r)\,e^{-\delta r}=r\,e^{-r\delta}$.
2. For a Poisson process the gap is exponential, so $P(\delta<36\,\mathrm s)=1-e^{-0.9\times0.01}=0.009$, about 1 percent. Several of the 22 sub-hour gaps are short, which already hints at clustering.
3. At $u^2=1/2$ the factor $1-2u^2$ vanishes, so the bracket is exactly 1 and $G=1$. The delay is $t_{\rm DM0}\,e^{-1/2}(1+\tfrac12\alpha e^{-1/2})=t_{\rm DM0}(e^{-1/2}+\alpha/2e)$, which tends to $t_{\rm DM0}\,\alpha/(2e)$.
4. At 1 kpc $\tau_{\rm ff}=27.5$. Setting $27.5\,(D/1\,\mathrm{kpc})^{-1/2}=1$ gives $D\approx760\,\mathrm{kpc}$. A lens in the host galaxy cannot be that far from the source, so a day-long echo stays absorbed.

Next: [search one CHIME burst for a lens echo and test R3's repeats](https://sandboxphysics.com/stories/fast-radio-burst-echoes/fit-a-chime-burst-and-r3-repeats/).

## Try it in the Lab

[FRB echoes through a plasma lens](https://sandboxphysics.com/labs/frb-plasma-lens-echoes/): Find the lens setting with the longest echo and compare it with the observed gaps.

## Sources

1. [On the non-Poissonian repetition pattern of FRB121102, Eq. 2](https://arxiv.org/abs/1705.04881v2)
2. [Periodic activity from a fast radio burst source](https://arxiv.org/abs/2001.10275v5)
3. [Lensing of fast radio bursts by plasma structures in host galaxies, Eqs. 15, 25 and 26](https://arxiv.org/abs/1703.06580v1)
4. [FRB-periodicity: mild pulsars in tight O/B-star binaries](https://arxiv.org/abs/2002.01920v5)
5. [FRB 121102 bursts show complex time-frequency structure, Sec. 4.3](https://arxiv.org/abs/1811.10748v1)

## Keep reading

Fast radio bursts: repeats, echoes and plasma lenses · Reading path

### From one burst to source-or-path tests

1. [Article **Is a repeating radio flash the source repeating, or its path?**](https://sandboxphysics.com/stories/fast-radio-burst-echoes/repeat-or-echo/)
2. [Article **How a clump of plasma delays and bends a radio burst**](https://sandboxphysics.com/stories/fast-radio-burst-echoes/plasma-delay-and-bending/)
3. [Article **Images, gains and delays of a Gaussian plasma lens**](https://sandboxphysics.com/stories/fast-radio-burst-echoes/gaussian-plasma-lens-images/)
4. [You are here **Telling source repetition from path effects with timing statistics**](https://sandboxphysics.com/stories/fast-radio-burst-echoes/repetition-statistics-source-or-path/)
5. [Article **Search one CHIME burst for a lens echo and test R3's repeats**](https://sandboxphysics.com/stories/fast-radio-burst-echoes/fit-a-chime-burst-and-r3-repeats/)
6. [Article **How to read the CHIME paper on the 16-day FRB cycle**](https://sandboxphysics.com/stories/fast-radio-burst-echoes/reading-the-chime-r3-periodicity-paper/)
