Faint Repetitions from a Bright Fast Radio Burst Source
Pravir Kumar, R. M. Shannon, Stefan Osłowski, Hao Qiu, Shivani Bhandari, Wael Farah, Chris Flynn, Matthew Kerr, D.R. Lorimer, J.-P. Macquart, Cherry Ng, C. J. Phillips, Danny C. Price, Renée Spiewak
DD RAFT VERSION D ECEMBER
10, 2019Typeset using L A TEX twocolumn style in AASTeX63
Faint Repetitions from a Bright Fast Radio Burst Source P RAVIR K UMAR , R. M. S
HANNON , S TEFAN O SŁOWSKI , H AO Q IU ,
2, 3 S HIVANI B HANDARI , W AEL F ARAH , C HRIS F LYNN , M ATTHEW K ERR , D.R. L
ORIMER ,
5, 6
J.-P. M
ACQUART , C HERRY N G , C. J. P
HILLIPS , D ANNY
C. P
RICE ,
1, 9
AND R ENÉE S PIEWAK
1, 101
Centre for Astrophysics and Supercomputing, Swinburne University of Technology, P.O. Box 218, Hawthorn, VIC 3122, Australia Sydney Institute for Astronomy, School of Physics, University of Sydney, Sydney, NSW 2006, Australia Australia Telescope National Facility, CSIRO Astronomy and Space Science, P.O. Box 76, Epping, NSW 1710, Australia Space Science Division, Naval Research Laboratory, Washington, DC 20375, USA Department of Physics and Astronomy, West Virginia University, Morgantown, WV 26506, USA Center for Gravitational Waves and Cosmology, West Virginia University, Chestnut Ridge Research Building, Morgantown, WV 26505, USA International Centre for Radio Astronomy Research, Curtin University, Bentley, WA 6102, Australia Dunlap Institute for Astronomy and Astrophysics, University of Toronto, 50 St. George Street, Toronto, ON M5S 3H4, Canada Department of Astronomy, University of California Berkeley, Berkeley CA 94720, USA ARC Centre of Excellence for Gravitational Wave Discovery (OzGrav), Australia
ABSTRACT
We report the detection of repeat bursts from the source of FRB 171019, one of the brightest fast radio bursts(FRBs) detected in the Australian Square Kilometre Array Pathfinder (ASKAP) fly’s eye survey. Two burstsfrom the source were detected with the Green Bank Telescope in observations centered at 820 MHz. Therepetitions are a factor of ∼
590 fainter than the ASKAP-discovered burst. All three bursts from this sourceshow no evidence of scattering and have consistent pulse widths. The pulse spectra show modulation that couldbe evidence for either steep spectra or patchy emission. The two repetitions were the only ones found in anobserving campaign for this FRB totaling hr, which also included ASKAP and the 64-m Parkes radiotelescope, over a range of frequencies (720–2000 MHz) at epochs spanning two years. The inferred scaling ofrepetition rate with fluence of this source agrees with the other repeating source, FRB 121102. The detectionof faint pulses from FRB 171019 shows that at least some FRBs selected from bright samples will repeat iffollow-up observations are conducted with more sensitive telescopes.
Keywords:
Radio transient sources (2008), Transient sources (1851), Fast radio bursts
1. Introduction
We are now starting to unravel the enigmatic astrophys-ical phenomenon of fast radio bursts (FRBs), millisecond-duration transient events first discovered over a decade ago(Lorimer et al. 2007). The observed dispersion measures(DMs) of FRBs significantly exceed the expected contribu-tion from the Milky Way (Thornton et al. 2013), suggest-ing extragalactic origins. The localization of several burstssources (Chatterjee et al. 2017; Bannister et al. 2019a; Raviet al. 2019) unequivocally places them at cosmological dis-tances; nevertheless, their physical origin has yet to be deter-mined.
Corresponding author: Pravir [email protected]
There are currently about 100 FRB sources published (FR-BCAT ; Petroff et al. 2016), most of which have only beendetected once. The repeat bursts from FRB 121102 (Spitleret al. 2016) enabled precise localization of the burst sourceand the identification of its host galaxy (Chatterjee et al.2017; Tendulkar et al. 2017). The existence of repetitionsruled out cataclysmic progenitor scenarios for the origin of itsemission. Since its discovery, more than 100 bursts (Zhanget al. 2018; Hessels et al. 2019) have been detected fromthis source in a broad range of frequencies, from as highas 8 GHz (Gajjar et al. 2018) to as low as 600 MHz (Jose-phy et al. 2019). The discovery of second repeating source,FRB 180814 (CHIME/FRB Collaboration et al. 2019b), withproperties similar to FRB 121102, strengthened evidence forthe existence of a substantial population of repeating FRB a r X i v : . [ a s t r o - ph . H E ] D ec K UMAR ET AL . Table 1.
Details of FRB 171019 follow-up observationsTelescope Receiver Gain T sys
Central Frequency Bandwidth Beam FWHM Sensitivity a Obs. Time(K Jy − ) (K) (MHz) (MHz) ( (cid:48) ) (Jy ms) (hr)ASKAP PAF 0.1 50 1297.5 336 60 51.8 986.6Parkes Multibeam 0.7 23 1382 340 14 1.10 12.4GBT Prime Focus 1 2.0 20 820 200 15 0.27 9.7GBT L-band 2.0 20 1500 800 9 0.13 0.9 a The limiting fluence for a pulse width of 5 ms and S/N threshold of 7.5 σ for GBT, 9.5 σ for ASKAP and 10 σ for Parkes, asdiscussed in Section 2. sources. Recently the Canadian Hydrogen Intensity Map-ping Experiment (CHIME) telescope reported detection ofeight new repeating FRB sources (CHIME/FRB Collabora-tion et al. 2019c).The localization of the ostensibly one-off (single pulse de-tection, which has not been shown to repeat) FRB 180924to a position 4 kiloparsecs from the center of a luminousgalaxy at a redshift of z = 0 . (Bannister et al. 2019a),enabled the first comparison of burst host galaxies. The mas-sive ( ∼ M (cid:12) ) host galaxy of FRB 180924 is in stark con-trast with the low-mass ( ∼ M (cid:12) ), low-metallicity dwarfgalaxy of the repeating source FRB 121102 (Tendulkar et al.2017), thus raising questions whether there are multiple FRBformation channels. Recently, another burst (FRB 190523)has also been localized to (cid:48)(cid:48) × (cid:48)(cid:48) uncertainty, and asso-ciated with a massive ( ∼ M (cid:12) ) host galaxy (Ravi et al.2019), partially based on the agreement between the burstDM ( . − ) and the galaxy redshift ( z = 0 . ).One of the most exciting open questions is the relation-ship between the repeating and one-off FRB sources. It isnot clear whether all FRBs repeat. Are there two (or more)classes of FRBs, or are the one-off FRBs simply the mostenergetic bursts from repeating sources? The absence of re-peat bursts even after hundreds of hours of follow-up (Raviet al. 2015, 2016) and the diversity in properties (e.g., tempo-ral structure and polarization) of one-off FRBs could be ev-idence for multiple populations of FRBs (Caleb et al. 2018;James 2019). However, in a recent analysis, Ravi (2019) hassuggested that the volumetric rate of one-off FRBs is incon-sistent with the rate of all possible cataclysmic FRB progen-itors and concludes that most FRBs are repeating sources.Among the strongest constraints on FRB repetition so farcome from Shannon et al. (2018) with the discovery of 20FRBs in the first Commensal Real-time ASKAP Fast Tran-sient (CRAFT ; Macquart et al. 2010) survey. The surveywas conducted using a “fly’s eye” configuration to maximize https://astronomy.curtin.edu.au/research/craft/ sky coverage at a Galactic latitude of | b | = 50 ± anda central frequency of 1.3 GHz. The survey produced a well-sampled population of FRBs and established a relationshipbetween burst dispersion and observed luminosity. The meanspectral index for these bursts ( α ≈ − . , where E ν ∝ ν α )is found to be similar to that of the normal pulsar population(Macquart et al. 2019). A key feature of the survey was thatit revisited the same positions hundreds of times over its du-ration, producing ∼ ∼ times greater than ASKAP, assuming the luminosity distribu-tion follows a power-law where, above some luminosity L ,the number of detections N ( > L ) ∝ L α assuming α = − (Connor & Petroff 2018). To complement the ASKAP selffollow-up, we have also been conducting sensitive monitor-ing campaigns of ASKAP detections with the 64-m Parkesradio telescope and the 110-m Robert C. Byrd Green BankTelescope (GBT). The arcminute localization of FRBs, madepossible by the multi-beam detection (Bannister et al. 2017)using ASKAP’s phased-array feed (PAF) enabled the follow-up of FRB fields with large aperture telescopes.In this Letter, we report the discovery of repetitions fromFRB 171019, one of the brightest bursts found in the ASKAPfly’s-eye survey. The burst was ∼ ms wide with a measuredfluence of Jy ms (Shannon et al. 2018). The observedDM was 460 pc cm − , a factor of 11 in excess of the NE2001model (Cordes & Lazio 2002) prediction along that line ofsight. In Section 2, we describe the observational campaignsfor this FRB. In Section 3, we present the properties of the Analysis of the entire campaign is ongoing and will be reported elsewhere.
ETECTION OF REPETITIONS FROM
FRB 171019 3 - - GBT 820GBT L-bandASKAPParkes Multibeam - - - - - - - - - - - - Figure 1.
Timeline of follow-up observations of FRB 171019. Each row represents a set of observations from a given radio telescope. Observations with burstsare encircled with red. The first repeat burst is found in be observation dated 2018 July 20 and the second one on 2019 June 9. repeat pulses. In Section 4, we discuss the implications forthe FRB population as a whole.
2. Observations and Data Processing
We searched for repeat pulses from FRB 171019 usingASKAP, Parkes, and the GBT. The observing details for allthree telescopes used are summarized in Table 1. Each tele-scope was pointed at the position of FRB 171019 reportedin Shannon et al. (2018), i.e., R.A. = h m s and decl.= − ◦ (cid:48) (cid:48)(cid:48) (J2000.0 epoch). This position was obtainedwith (cid:48) × (cid:48) uncertainty (90% confidence) as described inBannister et al. (2017). As such, the positional uncertaintywas well within the full-width at half maximum (FWHM) ofthe follow-up telescopes. Figure 1 shows a timeline of theradio observations of FRB 171019. ASKAP follow-up was conducted in fly’s eye configura-tion with each antenna pointing at a different position in thesky, and the survey regularly revisiting the same positions(Shannon et al. 2018). FRB searches are performed in near-real-time using
FREDDA (Bannister et al. 2019b), a GPU-based implementation of the fast dispersion measure trans-form algorithm (FDMT; Zackay & Ofek 2017). For a de-scription of the detection methods and search pipeline, seeBannister et al. (2017). We found no other astrophysicalevents at similar DMs of FRB 171019 exceeding a thresh-old signal-to-noise ratio (S/N) of 9.5 (which corresponds to afluence sensitivity of 52 Jy ms for a pulse duration of 5 ms)in 987 hr of observations.
At Parkes, we used the 20-cm multibeam receiver tosearch for bursts from FRB 171019, using the Berkeley-Parkes Swinburne Recorder (BPSR) mode of the HI-Pulsarsystem to record full-stokes spectra with 64 µ s time and390 kHz frequency resolution (Staveley-Smith et al. 1996;Price et al. 2016). The search process (Osłowski et al. 2019) was similar to that of the SUrvey for Pulsars and Extragalac-tic Radio Bursts project “Fast” pipeline (SUPERB; detailsin Keith et al. 2010; Keane et al. 2018). The online pipelinestored the 8-bit data stream from all 13 beams in a ring bufferover the bandwidth of 340 MHz centered at 1382 MHz. Thedata were then searched using Heimdall (Barsdell et al.2012) up to a maximum DM of 4096 pc cm − with a tol-erance (S/N loss tolerance between each DM trial) of 20 %.The transient pipeline sorts candidate FRB events from ra-dio interference using the methods detailed in Bhandari et al.(2018). The pipeline searched for bursts above a thresholdS/N of 10, thus sensitive up to a fluence of . Jy ms for aburst of width similar to FRB 171019. No bursts were foundin all the 12.4 hr of observations at the dispersion measure ofFRB 171019.
The GBT observations were obtained with the Prime Fo-cus 1 (centered at 820 MHz) and L-band receivers (details inTable 1), and data recorded with the Green Bank UltimatePulsar Processing Instrument (GUPPI; DuPlain et al. 2008).Each pointing was sampled with a time resolution of 81.92 µ s and 2048 frequency channels (512 channels for the L-band receiver), and written to a PSRFITS format file withfull-Stokes parameters.To search the GBT data for bursts, we first converted thePSRFITS data to total intensity SIGPROC filterbank format.The dynamic spectra were then normalized to remove the re-ceiver bandpass by scaling each channel to a mean of zeroand standard deviation of unity. Using the PRESTO (Ran-som 2001) tool rfifind and the median absolute devia-tion statistics, we identified bad channels affected by radiofrequency interference (RFI). The resulting data were thensearched using Heimdall for dispersed pulses. We per-formed two searches: a narrow search within the DM range http://sigproc.sourceforge.net https://github.com/scottransom/presto K UMAR ET AL .
150 100 50 0 50 100Time (ms)725750775800825850875900 F r e q u e n c y ( M H z ) F l u x ( J y ) GBT-1820 MHz
200 100 0 100 200Time (ms)725750775800825850875900 F r e q u e n c y ( M H z ) F l u x ( J y ) GBT-2820 MHz
150 100 50 0 50 100 150Time (ms)1150120012501300135014001450 F r e q u e n c y ( M H z ) F l u x ( J y ) ASKAP1.3 GHz
Figure 2.
Dynamic spectra for both repeat bursts detected at GBT and ASKAP FRB 171019 dedispersed at their optimal DM. From the left:repeat burst 1 (resolution = 1.31 ms), repeat burst 2 (resolution = 2.62 ms), and ASKAP FRB 171019 (resolution = 1.26 ms). For each burst,the top panel shows the flux density averaged over frequency channels.
Table 2.
Properties of detected bursts. Bursts properties calculated for full bandwidth appear in numbered rows and for lowerhalf band in row next to them in chronological order.No. Telescope TOA a Fluence b Gaussian FWHM Integrated Spectral Index d DM e (MJD) (Jy ms) (ms) S/N c ( pc cm − )0 ASKAP 58045.56061371(2) 219 ± ± − ± ± ±
12 5.2 ± − ± ± ± − ± ± ± ± − ± ± ± − ± ± ± ± − ± a Burst time of arrival is referenced at the highest frequency (1464 MHz for ASKAP and 920 MHz for GBT). The ASKAPburst arrival time is measured in TAI, while GBT burst arrival times are in UTC. Uncertainties are in parentheses. b SEFD curve of GBT-820 MHz is taken from https://science.nrao.edu/facilities/gbt/proposing/GBTpg.pdf. Fluence errorranges correspond to an uncertainty of one in S/N. For ASKAP burst, fluence is taken from Shannon et al. (2018). c S/N is the signal-to noise ratio calculated with width of the pulse as twice the Gaussian FWHM. d These are forced fits based on the assumption of a power-law spectrum. For GBT-1 spectrum, the fit obtained in the frequencyrange (820, 750 Mhz) of the lower half band is − ± e DM for ASKAP burst has been corrected from the value in Shannon et al. (2018) to account for an identified 1 MHz offsetin frequency labeling. of 446 to 474 pc cm − over 220 trials using a tolerance of 1%and then a wider search in a DM range of 0 to 2000 pc cm − with a tolerance of 5%. Candidates satisfying the followingcriteria were retained for further analysis: S/N ≥ . (7.5 forthe wider search), pulse width ≤ . ms and members Number of individual boxcar/DM trials clustered into a candidate. ≥ . For the L-band data, we also apply a minimum thresh-old for pulse width (0.65 ms) to mitigate false-positives pro-duced by spurious narrow-band short-duration candidates.We used deep neural network trained models, as developedby Agarwal et al. (2019) to perform the FRB/RFI binary All 11 trained models are taken from https://github.com/devanshkv/fetch
ETECTION OF REPETITIONS FROM
FRB 171019 5
725 750 775 800 825 850 875 900 (MHz)7.55.02.50.02.55.07.5 E ( J y m s ) OnpulseOffpulsesmoothbest-fit
725 750 775 800 825 850 875 900 (MHz)10.07.55.02.50.02.55.07.5 E ( J y m s ) OnpulseOffpulsesmoothbest-fit E ( J y m s ) OnpulseOffpulsesmoothbest-fit
150 100 50 0 50 100 150Lag (MHz)0.50.00.51.0 A u t o c o rr e l a t i o n a m p li t u d e GBT-1820 MHz
150 100 50 0 50 100 150Lag (MHz)0.500.250.000.250.500.751.00 A u t o c o rr e l a t i o n a m p li t u d e GBT-2820 MHz
150 100 50 0 50 100 150Lag (MHz)0.250.000.250.500.751.00 A u t o c o rr e l a t i o n a m p li t u d e ASKAP1.3 GHz
Figure 3.
Burst spectra and autocorrelation functions. From left: GBT-1 ( ∆ ν = 0.39 MHz), GBT-2 ( ∆ ν = 0.39 MHz), ASKAP detection ( ∆ ν = 1 MHz). Upper panels: burst spectra. Red lines are smoothed spectra (using a Gaussian kernel with standard deviation of 4 MHz). Bluedashed lines are best-fit power law model. Gray lines are off-pulse baseline spectra, and are offset from zero for clarity. Horizontal lines showzero power for both the on- and off-pulse spectra.
Bottom panels : autocorrelation function of the time-averaged spectrum of bursts. The zerolag value, which is associated with self noise present in spectrum, has been removed. classification of the candidates. Following their prescription,we created dedispersed frequency-time and DM-time imagedata for each candidate, which were then classified using keras (Chollet et al. 2015) with the
TensorFlow (Abadiet al. 2016) backend. We took the union of all the 11 modelpredictions and visually inspected each one of the resultingFRB candidates to identify astrophysical pulses. We foundtwo bursts (hereafter GBT-1 and GBT-2) at similar DM tothat of FRB 171019 in the observations.
We also conducted a search for periodicity in the GBT datausing Fourier domain searching with the PRESTO routine accelsearch , as well as time domain searching using theFast Folding Algorithm (FFA ) package riptide . Beforesearching, frequency channels and time blocks significantlyaffected by RFI were identified using rfifind and masked.The data were corrected for dispersion over 240 trial DMsevenly spaced from 400 to 520 pc cm − , generating a timeseries at each trial. We used dedisp (Barsdell et al. 2012),a GPU-accelerated package, to create time series. The FFA-based periodicity search was carried out to find long-periodsignals, where we searched periods ranging from 0.2 to 10 s.We detected no significant periodic astrophysical signal inthe data above a S/N threshold of 10 (chosen to minimize thenumber of false-positive candidates). Based on https://bitbucket.org/vmorello/riptide
3. The Repeat bursts
The two repeat bursts were detected in 820 MHz GBT ob-servations 9 and 20 months after the initial ASKAP detec-tion, and are marked with red circles in Figure 1. The dy-namic spectra of the bursts are shown in Figure 2, along withthe original detection at ASKAP. To measure the width ofthe bursts, we fit the frequency-averaged pulse profile with aGaussian model and report the FWHM ; both bursts are ap-proximately 4.5 ms in duration. The residual after subtract-ing the best-fit model from pulse profiles appears to be white,thus there is no underlying temporal sub-structure in the dy-namic spectrum of either repeat burst. The burst durationsare well in excess of the maximum DM smearing across achannel for the GBT data, which is 1.0 ms. For reference, wealso calculate the properties of the ASKAP detection. Thetime resolution for ASKAP data is 1.26 ms with a maximumDM smearing of 2.66 ms present within a channel. All threebursts are visible in the lower half of the band but not de-tected in the top half. Thus, the lower sub-band fluences arelarger than the full-band averaged values. The burst proper-ties obtained from the full band as well as from the lower halfof the band are listed in Table 2. The spectral structures ofthe bursts are described in Section 3.3. The measured FWHM values are consistent with the W50 estimates (widthat 50% of pulse peak). K UMAR ET AL .To obtain scattering timescales and burst DMs, we per-form multi sub-band modeling of the burst pulse profiles us-ing the nested sampling method
Dynesty (Speagle 2019)implemented in the parameter estimation code
Bilby (Ash-ton et al. 2019). We model each of the pulse profiles to bea Gaussian convolved with an exponential pulse-broadeningfunction. The broadening time τ is assumed to vary withfrequency with a fixed index, τ ∝ ν − . We model both inter-channel dispersion delay (which causes the pulse to arrive atdifferent times in different sub-bands) and intrachannel dis-persion smearing (which increases the pulse width in quadra-ture with an intrinsic width). Based on the ratio of Bayesianevidence between models with and without scattering, weconclude that the data do not support presence of scattering.For the ASKAP pulse, we limit the scattering timescale tobe < . ms, at a reference frequency of 1 GHz. For therepeat bursts, we group the lower half band of the data intofour sub-bands to perform the analysis; we limit the scatter-ing time scales (referenced to 1 GHz) to be < . ms and < . ms for GBT-1 and GBT-2 respectively. In contrast,the optimized DMs of the bursts shown in Table 2 suggestthat the repetitions have a different apparent DM than thehigher-frequency ASKAP detection. We extracted the GBT/GUPPI data for the detected repeatbursts using dspsr (van Straten & Bailes 2011) producinga full-Stokes archive file. We found no evidence for linearor circular polarization in the pulse data. It is possible thatthe non-detection of linear polarization is the result of Fara-day rotation of the burst through magnetized plasma. Wesearched for Faraday rotation using the PSRCHIVE (Hotanet al. 2004; van Straten et al. 2012) rmfit routine in therange | RM | ≤ × rad m − (this is the rotation measure(RM) at which the polarization position angle rotates by oneradian in one frequency channel at the center of the band), butno significant RM was found. We note that no polarizationcalibration procedures were conducted during GBT observa-tions. For the ASKAP burst, only the total intensity data wereretained; hence, no polarimetric properties could be derivedfrom this burst. The spectrum for each burst shown in Figure 3 is formedby integrating the signal over the time samples within twicethe measured FWHM of the frequency-averaged pulse. Theamplitude of each spectrum was then scaled to fluence, usingthe system equivalent flux density (SEFD) and the radiometerequation. Modest changes to the window do not significantlyaffect estimates of fluence. All three bursts show lower flu-ences at higher frequencies. One possibility is that the burstshave steep spectra. We characterize this by fitting a power-law model E ν ∝ ν α . Spectral indices, α obtained from the fits to individual spectra are in Table 2. All three bursts showsteep spectra in the observed bands with α ranging from − −
8. While both the ASKAP burst and the GBT-2 spectrais extremely steep in the lower half band as well, the GBT-1spectrum is nearly flat.Off-axis attenuation is unlikely to significantly change thefluences or spectral indices of the repetitions. Based on theposterior distribution from the ASKAP multi-beam localiza-tion in Shannon et al. (2018), the median correction to thefluence results in an increase of 8%, and is < with 90%confidence. The median spectral index correction is − . ,and with 95% confidence is less than < − . . This analysisassumes the GBT beam can be modeled as a Gaussian withan FWHM (beamwidth) of (cid:48) at 820 MHz (Table 1). Wetherefore rule out any primary beam offset as the cause of theobserved steep spectra for the GBT pulses.The spectral modulation in the bursts could be intrinsic tothe emission or due to the propagation effects. To charac-terize this, we calculate the autocorrelation function (ACF)of the burst spectra (Farah et al. 2018) as shown in Figure3. We fit the ACF with Gaussian component models usinga non-linear optimization approach (Newville et al. 2016)to find the frequency scales of characteristic modulation inspectra. We detect two characteristic frequency scales in theASKAP spectrum of band extent 13 and 147 MHz. For theGBT-1 spectrum, the ACF can be best described with a sin-gle component (100 MHz), which is the total bandwidth overwhich the pulse is visible. We observe a bright spike in thespectrum (at ∼ MHz), but its width is comparable to thechannel width. It is unclear if this is astrophysical or RFI.For the GBT-2, apart from the frequency scale of 82 MHz,we also see marginal evidence for a second component (7MHz wide). However, because the second component is notpresent in an analysis of the lower half of the band where theburst is bright, it is most likely due to RFI or noise fluctua-tions. We also estimate the amplitude of the spectra variabil-ity using the square of the modulation index m , by comput-ing the mean-normalized spectral autocovariance (Macquartet al. 2019) from the spectrum of bursts. The estimated val-ues of m for the three bursts are 2.4, 1.1, and 1.9 respec-tively. We use Bayesian methodology to characterize the repeti-tion statistics of FRB 171019, given the detection of pulseswith ASKAP at 1.3 GHz and the GBT at 820 MHz, and thenon-detections with the GBT at 1.5 GHz and Parkes at 1.3GHz. We assume that the cumulative burst rate above a fidu-cial fluence S at a frequency ν is R ( > S, ν ) = R (cid:18) SS ( ν ) (cid:19) γ , (1) ETECTION OF REPETITIONS FROM
FRB 171019 7 . ρ A10 -6 -5 -4 -3 - - R (hr -1 ) β D - - γ B . ρ C-2 -1 0 γ E -10 -5 0 5 10 . β ρ F . . . . ρ G Figure 4.
Posterior distributions for burst rate parameters. Pan-els A, C, and F show the one-dimensional marginalized distribu-tions for R , γ , and β , with peak probability densities normalizedto unity. Panels B, D, and E show the two-dimensional distributions(normalized again such that peak probability density are unity), withgrayscale shown in panel G. The rate R has been scaled to ASKAPsensitivities and frequencies ( Jy ms; see Table 1). where R is the rate of bursts above fluence S ( ν ) = S (cid:18) νν (cid:19) β (2)at a frequency ν .We assume that the burst event rate in a survey i of to-tal integration time T i will follow Poisson statistics with rateparameter λ i = T i R ( > S i , ν i ) , where ν i is the observing fre-quency of the survey and S i is the survey sensitivity. In thiscase we can infer the parameters in the survey R , β , and γ using the likelihood L = N s (cid:89) i =1 n i ! e − λ i ( λ i ) n i , (3)where n i is the number of bursts found in survey i = 1 to N s .We sample the posterior distribution using the multinest algorithm (Feroz et al. 2009) assuming uni-form priors on β and γ ( − < β, γ < ), and logarithmicpriors on R between − and hr − , where the referencefrequency ν = 1 . GHz and sensitivity S = 52 Jy ms. Wedo not take into account the spectral index obtained for bursts(Table 2) in this repetition analysis, which allows for an in-dependent estimation of the spectral index. The posterior distribution is shown in Figure 4. We find that the slope ofthe burst intensity distribution is consistent with a power-lawdistribution with an index between − . (cid:46) γ (cid:46) . The valuedepends strongly on the spectral dependence of the burstemission rate β . The inferred steep values of β ( β (cid:28) − . ,with the lower prior acceptable) are consistent with the ob-served spectra (in the case, the spectrum is attributed to asteep power-law process), but inconsistent with the ASKAPpopulation overall (Macquart et al. 2019). The observedshallow values of γ are consistent with observations of thefirst repeating FRB 121102 (Law et al. 2017).
4. Discussion and Conclusions
The bursts in FRB 171019 extend over the range of 219Jy ms to 0.37 Jy ms, a fluence range of ∼ ∼ L ∼ × erg s − to L ∼ × erg s − , nearly 3 orders in magnitude. Modelsfor burst emission need to account for this wide range.We find evidence for variations in the apparent DMs ofthe pulses. It is unclear whether the difference is genuineDM variation or due to non-dispersive effects as has beenobserved in FRB 121102 (Hessels et al. 2019). We note thatthis discrepancy in apparent DM can also be due to the differ-ent volumes of the medium being probed by the ASKAP andthe GBT. All three bursts are temporally resolved with simi-lar widths. We note that the pulse width of the GBT-2 is lessreliable when measured in the whole band due to the pres-ence of RFI in the upper half of the band. However, takingthe DM smearing and sampling time into account, the intrin-sic width of all bursts are consistent within uncertainties. Wefind no evidence for sub-structure in the pulse profile as seenin other FRBs (Farah et al. 2018; Hessels et al. 2019). Wewould be insensitive to any sub-structure narrower than ∼ UMAR ET AL . ( ν GBT /ν ASKAP ) − ≈ smaller. The widest structures inthe ASKAP burst (width approximately half the band) wouldbe observed to be ∼ MHz wide in the GBT spectrum.However, we only see evidence for structures much widerthan this in the GBT observations. We do not find any con-clusive evidence of diffractive scintillation in repeat bursts.All of the bursts from FRB 171019 are only visible in lowerhalf of their respective bands, which could be evidence ofan extremely steep spectrum. This argument is also con-sistent with the non-detection of repetitions at Parkes andGBT L-band receivers. If we assume this steep spectrum( ∼ − ) to be the case, it provides a very natural way to un-derstand the detection of repetitions from this source in thecontext of all the non-detections (Shannon et al. 2018; James2019) from other ASKAP FRBs (assuming a non-negligiblefraction are repeaters). It would make the repeat bursts atleast a factor of ( ν ASKAP /ν GBT ) ≈ fainter at the cen-ter frequency of ASKAP. In that scenario, the fluence dis-crepancy between ASKAP and the GBT detection is actu-ally > , assuming a constant spectral index that makesFRB 171019 special within the ASKAP population of flatter-spectrum FRBs (Macquart et al. 2019). However, we are cau-tious not to over-interpret this result, as there are not manyphysical mechanisms to produce such a steep spectrum. Itis quite possible that the spectrum is similar to patchy emis-sion, seen in the other repeater FRB sources (Michilli et al.2018; CHIME/FRB Collaboration et al. 2019c). This is ev-ident as the GBT-1 spectrum is nearly flat in the lower halfof the band. Also, the ASKAP detection has a large spectralmodulation that can not be explained by scintillation. In thisscenario, the power-law model might not be the correct ap-proach for the spectral index measurement (Sokolowski et al.2018).Another possibility is FRBs having stochastic patchy ormodulated emission in different parts of the frequency bandfor an individual burst but, when ensemble-averaged, pro-duce steep spectra as observed in ASKAP one-off FRBs sam-ple (Shannon et al. 2018; Macquart et al. 2019). This wouldbe tested with further detections.The other published repeating burst sources (CHIME/FRBCollaboration et al. 2019b; Hessels et al. 2019) share com-mon features such as spectra variability, sub-structures intheir dynamic spectrum, and sub-components in pulse pro-file. We do not observe any of these features in all threebursts. A coherently dedispersed detection from FRB 171019with high time resolution will provide more informationon these distinctions. The bursts from the FRB 171019 source are fainter at higher frequencies, which is not thecase with many of the bursts from the FRB 121102 source,where bursts have been reported brighter at higher frequen-cies (Gourdji et al. 2019). The first detection of FRB 171019comes from a different sample of bright FRBs (Shannon et al.2018) than the CHIME detections (CHIME/FRB Collabora-tion et al. 2019a) and FRB 121102. The host galaxy of alocalized burst (FRB 180924; Bannister et al. 2019a) fromthe ASKAP population also originates from a galaxy signif-icantly different to that of FRB 121102. It will be interest-ing to see if all repeating FRBs have similar environmentsas of FRB 121102. If not, it could be indicative of a differ-ent channel for producing repeat burst sources. The detec-tion of further repetitions from this source and localizationto a host galaxy will be key to understanding the nature ofFRB 171019 and its relation to other repeating burst sources. Acknowledgments
We thank C. James for useful discussions. P.K.,R.M.S., S.O., and R.S. acknowledge support throughAustralian Research Council (ARC) grant FL150100148.R.M.S. and J.P.M. acknowledge support through ARC grantDP180100857. R.M.S. also acknowledges support throughARC grant CE170100004. D.R.L. was supported by the Na-tional Science Foundation through the award number OIA-1458952. This work was performed on the OzSTAR na-tional facility at Swinburne University of Technology. OzS-TAR is funded by Swinburne University of Technology andthe National Collaborative Research Infrastructure Strategy(NCRIS). Work at NRL is supported by NASA. The GreenBank Observatory is a facility of the National Science Foun-dation operated under cooperative agreement by AssociatedUniversities, Inc. The Parkes radio telescope is part of theAustralia Telescope National Facility which is funded by theAustralian Government for operation as a National Facilitymanaged by CSIRO. The Australian SKA Pathfinder is partof the Australia Telescope National Facility which is man-aged by CSIRO. Operation of ASKAP is funded by the Aus-tralian Government with support from the National Collab-orative Research Infrastructure Strategy. ASKAP uses theresources of the Pawsey Supercomputing Centre. Establish-ment of ASKAP, the Murchison Radio-astronomy Observa-tory and the Pawsey Supercomputing Centre are initiatives ofthe Australian Government, with support from the Govern-ment of Western Australia and the Science and Industry En-dowment Fund. We acknowledge the Wajarri Yamatji peopleas the traditional owners of the Observatory site. We note that CHIME has recently detected a repeat burst from this source(Patel & CHIME/FRB Collaboration 2019).
ETECTION OF REPETITIONS FROM
FRB 171019 9