Identification of GRB precursors in Fermi-GBM bursts
IIdentification of GRB precursors in Fermi-GBM bursts
Paul Coppin, ∗ Krijn D. de Vries, and Nick van Eijndhoven
Department of Physics and AstronomyVrije Universiteit Brussel, Pleinlaan 2, Elsene, Belgium (Dated: April 8, 2020)We present an analysis of more than 11 years of Fermi-GBM data in which 217 Gamma-Ray Bursts(GRBs) are found for which their main burst is preceded by a precursor flash. We find that shortGRBs ( < ∼
10 times less likely to produce a precursor than long GRBs. The quiescent timeprofile is well described by a double Gaussian distribution, indicating that the observed precursorshave two distinct physical progenitors. The light curves of the identified precursor GRBs are publiclyavailable in an online catalog ( https://icecube.wisc.edu/~grbweb_public/Precursors.html ) Introduction — Gamma-ray bursts (GRBs) are cat-aclysmic transient cosmic events characterized by theemission of one or multiple flashes of gamma radiation.They are the most powerful outbursts of electromagneticradiation in our universe and a possible source of (ultra)high-energy cosmic rays [1, 2]. The duration of GRBscan be described using a bi-modal distribution indicatingthe existence of two progenitor source classes. In general,bursts lasting longer than 2 s are related to the collapse ofa super-massive star, as confirmed by the observation oftype-Ic supernovae in coincidence with long GRBs [2, 3].Short bursts, lasting less than 2 s, are believed to oc-cur when two co-orbiting neutron stars collide. Evidencefor this model was recently obtained by the detection ofgravitational waves from a binary neutron star mergerfollowed by a short GRB [4, 5].The main outburst of gamma radiation, called theprompt phase, is followed by an afterglow stage in whichthe ejected matter collides with the surrounding medium.Thanks to multi-wavelength observations, ranging fromX-ray to radio, the physical processes related to this af-terglow emission are well understood. Apart from theprompt and afterglow phases, there is a third emissionphase, called the GRB precursor. Precursors are typ-ically defined as relatively dim gamma-ray flashes thatoccur before the prompt emission. Previous studies [6–16], comprising GRBs up to the year 2014, found thatprecursor flashes occur in a subset of both long and shortGRBs. The fraction of bursts in which a precursor is ob-served strongly depends on the method and criteria usedto define a precursor and typically ranges from 3% to20%.Numerous models have been proposed to explain pre-cursor flashes in both long and short GRBs. Theseinclude photospheric emission [17–19], pre-burst jets,[20, 21] and interactions between magnetized neutronstars [22, 23]. Currently, there is no consensus on theorigin of precursor flashes and, most likely, more thanone model will be needed to explain all observed precur-sors. Given that precursors only occur in a subset of allGRBs, an extensive study is thus required to uncover thephysical origin of GRB precursors.We performed an automated search that identifies pre- cursor flashes observed by the Fermi-GBM detector [24].Out of a total sample of 2364 GRBs, 244 precursors wereidentified originating from 217 GRBs, of which 158 arenewly identified GRBs with precursor emission.In this letter, we present the details of our selectionand show that short GRBs are ∼
10 times less likely toproduce a precursor than long GRBs. We performed ananalysis on the quiescent time profile, given by the timesbetween the precursor flash and the main burst. Theincreased statistics from our search allowed us to identifya novel feature in the quiescent time profile, which is welldescribed by a double Gaussian distribution, indicatingthat the observed precursors have two distinct physicalprogenitors. To allow for follow-up studies, searching forcoincidences with other astrophysical messengers, such asneutrinos and gravitational waves, the obtained results ofour analysis are presented in the supplemental materialand have been made available via an online tool [25].
Data — The Fermi Gamma-ray Space Telescope is cur-rently the most efficient GRB detection satellite in or-bit. Its two main instruments are the Large Area Tele-scope (LAT) and the Gamma-ray Burst Monitor (GBM).Whereas LAT has a sky coverage of 20%, GBM continu-ously observes the full region of the sky not occulted byEarth. On average, the GBM and LAT detect 240 and18 GRBs per year, respectively [26–28]. In this study weanalyzed 2684 GRBs, using all GBM recorded bursts upto the year 2020.The GBM telescope is composed of 12 sodium io-dide (NaI) and two bismuth germanate (BGO) detectors.Trigger and localization information is provided by theNaI detectors, which are sensitive to gamma-rays of 8keV to 1 MeV. The BGO detectors, which will not beused in this analysis, are sensitive from 200 keV to 40MeV and serve to cover the energy gap with the LAT[27].The GBM burst data was obtained from the FermiScience Support Center [24] and provides the raw pho-ton counts as a function of time and energy for each ofthe 14 detectors. Time-Tagged Event (TTE) data pro-vides the highest temporal resolution of 2 µ s. Since Au-gust 2010, TTE data is available over a time window[ t tr −
135 s , t tr + 300 s], where t tr is the GBM detector a r X i v : . [ a s t r o - ph . H E ] A p r −
20 0 20 40 60 80
Time (s) R a t e ( H z ) GRB190114C − −
500 0 500 100010 FIG. 1. Illustration of the Bayesian block light curve of GRB 190114C. Two dim precursors (yellow) are observed 5.57 s and2.85 s before the onset of the prompt emission (red). The time range displayed in this figure corresponds to the grey shadedarea in the inset image, which displays the full light curve. trigger time. Before August 2010, TTE data is only avail-able starting 30 s before t tr , but again up to 300 s after t tr . CTIME data is provided over a 2000 s time win-dow centered around t tr , but has a coarser nominal res-olution of 0.256 s. To allow the detection of very shortemission periods, we have used TTE data whenever avail-able. CTIME data was used to extend the examined timewindow to 1000 s before and after the trigger time.During normal operation, the Fermi telescope func-tions in a sky survey mode [29]. To allow full sky monitor-ing of the LAT telescope, the orientation of the detectorcontinuously changes. As such, the background rates ofthe GBM detectors vary with time. A linear backgroundrate approximation can however still be used over periodsof time less than 100 s, as the period of the oscillatorymotion of the spacecraft is on the order of 3 hours [27, 29]. Method — For every burst, we select the GBM NaIdetectors that were triggered by the GRB. If more thanthree detectors were triggered, only the three triggereddetectors which were pointing closest to the burst loca-tion are used. The data analysis is two-fold. An initialanalysis on raw time data is performed to characterizethe background, allowing to capture global fluctuations.Subsequently, a Bayesian Block (BB) algorithm [30] isused to select the physical signal regions.Our analysis aims at identifying all emission periods inwhich gamma-ray activity is observed from the detectedGRBs. This is achieved by constructing background sub-tracted light curves. For more than 90% of the identifiedbursts, a stable background fit is first obtained between800 s and 1000 s before t tr , marking the start time of theanalysis interval. The end time of the analysis interval isset 50 s past the end of the Fermi T90 interval, defined asthe central time window that contains 90% of the fluenceof the GRB. For bursts whose T90 exceeds 250 s, thisend time is extended to 20% of the T90 time beyond theT90 interval. One final consideration is that a minority of all bursts have one or more gaps in their light curvesdue to missing data. For those bursts, we only examinethe continuous data taking period containing t tr . Thischoice is motivated by the observation that for less than1% of all bursts, additional data is available at earliertimes.We automated the selection of background time in-tervals in which no increased gamma-ray activity is ob-served. Our selection is therefore fully reproducible andbased on physically motivated parameters. Backgroundtimes are selected based on the requirement that the ratedoes not undergo a sudden increase. For this purpose, weuse an algorithm similar to the Fermi-GBM online trig-ger [27], which compares the observed rate to a predictionbased on a fit to the rate at earlier times. The rate inthe identified background intervals is then extrapolatedto intermediate, possible signal, regions. As such, we ob-tain an estimate for the background rate over the fulllight curve. A more detailed description of this methodis provided in the supplemental material.Having characterized the background rate, we proceedby rebinning the data using the Bayesian Block (BB)algorithm [30]. The BB algorithm was specifically de-signed to identify localized structures, such as bursts, inGRB light curves. It optimizes both the number of binsand the location of the bin edges by maximizing a fitnessfunction. For every selected GBM detector, we constructa BB light curve. In addition, a single BB light curvebased on a combination of the photon counts of the se-lected detectors is also constructed for every burst. Thesecombined light curves contain the largest statistics andwill thus serve as the basis for our selection. To illus-trate the BB procedure, the light curve of GRB 190114Cis displayed in Fig. 1. Analysis — To quantify the physical signal, a back-ground subtraction procedure is applied. The back-ground rate is integrated over each bin to estimate the
Treshold rate r th (Hz) − − − F r a c t i o n o f G R B s Detected before t tr −
500 sNot detected within 5 s of t tr FIG. 2. Relative number of GRBs that are not detectedwithin 5 s of the trigger time t tr (full orange line) as a func-tion of the threshold rate r th . The same relation is shown forbursts that are detected more than 500 s before the triggertime (blue dashed line), where few to no precursors are ex-pected [10]. The shaded grey bands show the 1 σ statisticaluncertainty for both curves. In our analysis, the thresholdrate is set equal to 30 Hz, indicated by the vertical line. total number of background events N b . Subtracting N b from the observed count and dividing by the duration,we obtain an estimate for the signal rate r s . A thresh-old condition r s > r th is then imposed to tag those binsthat potentially contain a physical signal. In addition,we impose the requirement that the excess is separatelyobserved by two or more detectors.The threshold rate r th is based on the trade-off of min-imizing the number of false positives, whilst maximizingthe sensitivity of the search. Ideally, every event trig-gering the GBM detector should also be selected by ouranalysis. To estimate the loss of sensitivity as a func-tion of r th , we therefore consider the fraction of GRBsin which, following our selection criteria, no excess is ob-served within 5 s of the GBM trigger time t tr . Thisquantity is shown as the full orange line in Fig. 2 andshows a slow but steady increase as a function of r th .To estimate the false positive rate, we consider thenumber of GRBs which, following our criteria, yielded anemission episode which could be identified as a precursorin the period from 1000 s to 500 s before the Fermi-GBMtrigger time t tr . This time window is based on a previousstudy presented in [10], which, using a sample of 956Fermi-GBM observed bursts, found only a single burst inwhich a precursor event occurred more than 500 s before t tr . For our analysis, the fraction of GRBs which yieldeda signal in this time range is displayed by the dashed blueline in Fig. 2 as a function of the threshold rate r th . Aplateau is reached at r th >
30 Hz. We therefore set r th =30 Hz, as this corresponds to the minimal value for whichthe estimated false positive rate approaches the plateauat ∼ r f of 1 . × − Hz. Considering that, on rare occasions, precursors do occur more than 500 s before the promptemission [31], this approach is expected to result in aconservative estimate of the true false positive rate.We define an emission episode as a continuous periodof increased emission in the background subtracted lightcurve. If a GRB has two or more emission episodes, weverify that the intermediate quiescent periods containenough statistics to ensure that the rate has droppedback to the background level. Quiescent periods forwhich the Poisson uncertainty on the average backgroundrate exceeds 5% are disregarded. For a typical burst [32],this corresponds to a lower limit on the allowed durationof the quiescent period of ∼ . Results — Applying our signal selection method onall 2684 bursts, we find 320 GRBs that were triggeredby Fermi-GBM, but do not show a signal following ourcriteria. In the following, we therefore restrict ourselvesto the 2364 bursts for which a signal is found.Our analysis identified 244 precursor emission episodesspread over 217 GRBs. We thus find that 9% of all GRBshave one or more precursors. Any given burst is observedto have at most 3 precursors. The number of bursts hav-ing 1, 2 and 3 precursors corresponds to 192, 23 and 2,respectively. Considering the combined time precedingthe prompt emission of all GRBs, equal to 2 . · s,the expected number of false positives in our analysis is36.1 ± Quiescent period ∆ t Q (s) N u m b e r o f o cc u rr e n ce s Double Gaussian fitSingle Gaussian fit All GRBsShort GRBs
FIG. 3. Distribution of the quiescent time between two sub-sequent emission episodes. The data is found to be well de-scribed by the sum of two Gaussian functions, with a reduced χ -value of 1.23 for the best fit parameters. A single compo-nent Gaussian fit results in a higher reduced χ -value of 8.87.The 4 short GRBs are indicated in yellow. the prompt emission. All 4 short GRBs have a precursorthat is shorter in duration than the prompt phase andtheir quiescent times are consistent with one another upto a factor ∼
3. While limited in statistics, we note thatthe time intervals between the onset of the precursor andprompt emission are smaller than the 1 . t Q ∼ χ -value of 1.23 assuming Poissonerrors on the number of events. The two Gaussian distri-butions peak at 0 .
57 s and 27 s and have a weight of 12%and 88%, respectively. Performing a single componentGaussian fit, thus simplifying our model by neglectingthe excess at ∆ t Q < ∼ χ -value increases to 8.87.The leftmost component of the double Gaussian fit inFig. 3 could have several origins. A first contribution isfound from the precursors of short GRBs, though theycan only account for ∼
15% of the observed excess. Asecond contribution could come from bursts whose ob-served flux drops below the observable limit in betweendifferent peaks of the prompt phase, falsely identifyinga precursor emission during the prompt phase. Figure 4illustrates that bursts with ∆ t Q < .
32 s before the start of the prompt emission.We thus observe that a subclass of long GRBs do havephysical precursors occurring within two seconds of theprompt emission. Given the short time scale, this could Total fluence N γ . . . . d N / d l n ( N γ ) ∆ t Q > t Q < FIG. 4. Probability distribution of the total fluence of longGRBs in which we identified precursors. The normalizationis taken such that the y-values sum to unity. Bursts for whichthe quiescent times in between emission episodes is less than2 s are observed to have a lower average fluence. indicate that the precursors of those bursts originate froma different physical mechanism.A related study of quiescent times was performedin [16], where a strong linear correlation between the du-ration of the quiescent time ∆ t Q and that of the subse-quent emission episode ∆ t sub was found. However, dueto lack of data, redshift effects, which could naturally in-duce such a linear correlation, were not considered. Toprobe possible redshift effects, we determined the corre-lation between ∆ t Q and ∆ t sub for the 21 bursts in our se-lection with known redshift z , and apply a correction forredshift. The obtained Pearson correlation factor is 34%.To determine the significance of this value, we composeda test statistic distribution by calculating the correlationcoefficient between random combinations of the quies-cent times and secondary emission episodes. Based onthis distribution, we obtain a p -value of 7.1%. No signif-icant linear correlation is thus observed between the du-ration of the quiescent time following precursor episodes∆ t Q and the duration of the secondary emission episode∆ t sub .Quiescent times also provide an independent probeto investigate potential differences between the pre-cursor and prompt emission. Previous studies gener-ally found that precursor emission exhibits the samespectral properties and evolution as prompt emission[6, 7, 9, 11, 14, 34]. In the case of long GRBs, thisobservation can be embedded in a model in which theprecursor and prompt emission correspond to differentshells of matter falling onto the central engine [2, 12, 14].A parameter that could be sensitive to this hypothesis isthe quiescent time in between emission episodes. Figure 5shows the distribution of the quiescent time between twoprecursors (red) and between precursor and prompt emis-sion (blue). To quantify their resemblance, we test if thetwo data sets were drawn from the same parent distribu- ∆ t Q (s) . . . . . d N / d l n ( ∆ t Q ) Between precursor & promptBetween two precursors
FIG. 5. Probability distribution of the quiescent time ∆ t Q between two precursors (red) and between a precursor andthe prompt emission (blue). The normalization is taken suchthat the y-values sum to unity. tion using a two-sample Kolmogorov-Smirnov test. Themaximal separation between the cumulative distributionsis 28.7%, resulting in a p -value of 3.0%.One object in our selection is of special interest, GRB190114C, a particularly bright burst that occurred on the14th of January 2019 [35]. GRB 190114C/bn190114873is the first GRB from which TeV photons have been de-tected, as observed by the MAGIC telescope in La Palma[36]. Our analysis identified two faint precursors occur-ring 5.57 s and 2.85 s before the start of the promptemission and lasting 1 .
94 s and 1 .
54 s, respectively. Thedetailed light curve of this burst is shown in Fig. 1.
Conclusion—
By applying a fully automated precur-sor search on the light curves of 2364 GRBs, we identifieda total of 244 precursors spread over 217 bursts. Onlyfour of those precursors occurred for short GRBs. Wethus find that the fraction of long and short GRBs withone or more precursors equals 10.5% and 1.2%, respec-tively. All precursors for short GRBs occurred within 2 sbefore the start of the prompt emission. A notable longGRB for which we found two precursors is the extremelybright GRB 190114C. This burst was preceded by twodim precursors, indicating that gamma-ray productionwas already ongoing 5 . ∗ [email protected][1] A. M. Hillas, The origin of ultra-high-energy cosmic rays,ARA&A , 425 (1984).[2] B. Zhang, The Physics of Gamma-Ray Bursts (Cam-bridge University Press, 2018).[3] S. E. Woosley, R. G. Eastman, and B. P. Schmidt,Gamma-Ray Bursts and Type IC Supernova SN1998BW, ApJ , 788 (1999), arXiv:astro-ph/9806299.[4] LIGO Scientific Collaboration and Virgo Collaboration,GW170817: Observation of Gravitational Waves froma Binary Neutron Star Inspiral, Phys. Rev. Lett. ,161101 (2017), arXiv:1710.05832.[5] LIGO Scientific Collaboration and Virgo Collaboration,Fermi Gamma-ray Burst Monitor, and INTEGRAL,Gravitational Waves and Gamma-Rays from a BinaryNeutron Star Merger: GW170817 and GRB 170817A,ApJ , L13 (2017), arXiv:1710.05834.[6] E. Troja, S. Rosswog, and N. Gehrels, Precursors of ShortGamma-Ray Bursts, ApJ , 1711 (2010).[7] T. M. Koshut, C. Kouveliotou, W. S. Paciesas, J. vanParadijs, G. N. Pendleton, M. S. Briggs, G. J. Fishman,and C. A. Meegan, Gamma-Ray Burst Precursor Activityas Observed with BATSE, ApJ , 145 (1995).[8] D. Lazzati, Precursor activity in bright, long BATSEgamma-ray bursts, MNRAS , 722 (2005).[9] D. Burlon, G. Ghirlanda, G. Ghisellini, D. Lazzati,L. Nava, M. Nardini, and A. Celotti, Precursors in SwiftGamma Ray Bursts with Redshift, ApJ , L19 (2008),arXiv:0806.3076.[10] M. Charisi, S. M´arka, and I. Bartos, Catalogue of isolatedemission episodes in gamma-ray bursts from Fermi, Swiftand BATSE, MNRAS , 2624 (2015), arXiv:1409.2491.[11] Y.-D. Hu et al. , Internal Energy Dissipation of Gamma-Ray Bursts Observed with Swift: Precursors, PromptGamma-Rays, Extended Emission, and Late X-RayFlares, ApJ , 145 (2014), arXiv:1405.5949.[12] F. Nappo, G. Ghisellini, G. Ghirlanda, A. Melandri,L. Nava, and D. Burlon, Afterglows from precursorsin gamma-ray bursts. Application to the optical af-terglow of GRB 091024, MNRAS , 1625 (2014),arXiv:1405.3981.[13] P. Coppin and N. van Eijndhoven, IceCube Search forHigh-Energy Neutrinos Produced in the Precursor Stagesof Gamma-ray Bursts, in
ICRC2019 , Vol. 36, p. 859,arXiv:1908.06653.[14] D. Burlon, G. Ghirlanda, G. Ghisellini, J. Greiner,and A. Celotti, Time resolved spectral behavior ofbright BATSE precursors, A&A , 569 (2009),arXiv:0907.5203.[15] G. Calderone, P. Romano, V. Mangano, and
G. Cusumano, Search for precursor activity inSwift/BAT gamma-ray bursts., in
The 7th INTE-GRAL Workshop (2008) p. 18.[16] E. Ramirez-Ruiz and A. Merloni, Quiescent times ingamma-ray bursts - I. An observed correlation betweenthe durations of subsequent emission episodes, MNRAS , L25 (2001).[17] B. Paczynski, Gamma-ray bursters at cosmological dis-tances, ApJ , L43 (1986).[18] A. M. Beloborodov and P. M´esz´aros, Photospheric Emis-sion of Gamma-Ray Bursts, Space Sci. Rev. , 87(2017), arXiv:1701.04523.[19] M. Lyutikov and V. V. Usov, Precursors of Gamma-RayBursts: A Clue to the Burster’s Nature, ApJ , L129(2000), arXiv:astro-ph/0007291.[20] S. Razzaque, P. M´esz´aros, and E. Waxman, Neutrino to-mography of gamma ray bursts and massive stellar col-lapses, Phys. Rev. D , 083001 (2003).[21] P. M´esz´aros and E. Waxman, TeV Neutrinos from Suc-cessful and Choked Gamma-Ray Bursts, Phys. Rev. Lett. , 171102 (2001).[22] B. D. Metzger and C. Zivancev, Pair fireball precur-sors of neutron star mergers, MNRAS , 4435 (2016),arXiv:1605.01060.[23] H. Tong, C. Yu, and L. Huang, A magnetically drivenorigin for the low luminosity GRB 170817A associatedwith GW170817, Res. Astron. Astrophys. , 067 (2018),arXiv:1711.06593.[24] https://heasarc.gsfc.nasa.gov/FTP/fermi/data/gbm/bursts .[25] https://icecube.wisc.edu/~grbweb_public/Precursors.html .[26] Fermi-LAT collaboration, A Decade of Gamma-RayBursts Observed by Fermi-LAT: The Second GRB Cat-alog, ApJ , 52 (2019), arXiv:1906.11403.[27] C. Meegan et al. , The Fermi Gamma-ray Burst Monitor,ApJ , 791 (2009), arXiv:0908.0450.[28] A. von Kienlin et al. , The 4th Fermi-GBM Gamma-RayBurst Catalog: A Decade of Data, , arXiv:2002.11460(2020).[29] Fermi-LAT collaboration, The Large Area Telescope onthe Fermi Gamma-Ray Space Telescope Mission, ApJ , 1071 (2009), arXiv:0902.1089.[30] J. D. Scargle, J. P. Norris, B. Jackson, and J. Chi-ang, Studies in Astronomical Time Series Analysis. VI.Bayesian Block Representations, ApJ , 167 (2013),arXiv:1207.5578.[31] See, for instance, the case of GRB 091024 [12].[32] Assuming a GRB for which two detectors contribute tothe combined light curve and each detector has a back-ground rate of 1000 Hz.[33] The precursor emission our analysis identified in the shortburst bn130504314 was disregarded after a visual inspec-tion of the light curve revealed that it corresponded tothe prompt emission.[34] K. L. Page et al. , GRB 061121: Broadband Spectral Evo-lution through the Prompt and Afterglow Phases of aBright Burst, ApJ , 1125 (2007), arXiv:0704.1609.[35] M. Ajello, , et al. , Fermi and Swift Observations ofGRB 190114C: Tracing the Evolution of High-EnergyEmission from Prompt to Afterglow, ApJ , 9 (2020),arXiv:1909.10605.[36] V. A. Acciari et al. (MAGIC), Teraelectronvolt emissionfrom the γ -ray burst GRB 190114C, Nature , 455 (2019). SUPPLEMENTARY MATERIAL
Background characterization — During normal op-eration, the Fermi telescope functions in a sky surveymode [29]. This implies that the orientation of the space-craft continuously changes to allow the LAT telescope tomonitor the entire sky. A downside to this mode of oper-ation is that the background rates of the GBM detectorsare changing with time. A linear approximation can stillbe used over periods of time less than 100 s, as the periodof the oscillatory motion of the spacecraft is on the orderof 3 hours [27, 29].During previous searches, the time ranges used to es-timate the background rate were generally set by hand[7–9]. Since we plan to examine a time range of 2000 sfor over 2000 bursts, this would become a very demand-ing endeavor. Therefore, we automated the selection oftime intervals in which no increased gamma-ray activityis observed. This method has the added advantage thatthe selection is fully reproducible and based on physicallymotivated parameters.The tagging of background time intervals is illustratedin Fig. 6 and based on the assumption that the observedrate can be predicted using the rate at earlier times. Topredict the background rate at an arbitrary time t , weperform a linear fit to the data in the time interval [ t −
30 s , t −
10 s]. By extrapolating the fit to time t , weobtain a prediction r p for the background rate at time t . This prediction is then compared to the true rate r t found at time t , averaged over 2.5 s. As long as thetrue rate is within a 3 σ Poisson upper-fluctuation of thepredicted background rate, i.e. r t < r p + 3 · (cid:114) r p . , (1)the time t is tagged as background. The next pointin time t = ( t + 1 s) is then subjected to the sameprocedure, until a time t n is found for which Eq. (1) nolonger holds. To determine the length of the possible re-gion of interest, we then proceed by verifying if the RMSof r p − r t , averaged over a 10 s period centered around t n + 25 s is within 1.5 σ of the Poisson expectation. Byimmediately advancing 25 s, we aim to overshoot theperiod with excess emission while still ensuring that thebackground can be well approximated by a linear extrap-olation. If the RMS exceeds 1.5 σ , t n + 25 s is labeled asnon-background and added to the analysis region. If, onthe other hand, the RMS is sufficiently low, a new back-ground interval is started at t n + 25 s.Using the approach outlined above, background re-gions are identified in each of the light curves. The back-ground rate is then set equal to the true rate, averaged − − −
250 0 250 500 750
Time (s) R a t e ( H z ) Background fitPredicted rate r p True rate, 0.256 s binsTrue rate r t , averaged over 2.5 s FIG. 6. Characterization of the rate of GRB trigger bn150422703 and the GBM detector labeled nb. The true rate averaged over2.5 s (black) is compared to a prediction based on prior data points (orange). Time intervals in which these two distributionsmatch are used to characterize the background rate (red). over 2.5 s, in these background intervals. In intermediateregions of interest containing possible signal, a linear in-terpolation is used based on the last and first point of theadjacent background intervals. Figure 6 displays a visu-alization of this procedure for GRB trigger bn150422703and the GBM detector labeled nb.
Precursor catalog — To enable follow up studies,we provide a complete list of the emission times of allprecursors identified by our analysis. Our catalog pro-vides the start time of the prompt emission in UTC, thestart time of the precursor emission with respect to theonset of the prompt emission, and the duration of theprecursor emission. An electronic version of this tablecan be downloaded from https://icecube.wisc.edu/~grbweb_public/Precursors.html . Table I: Temporal properties of the identified precursors. For every GRB, we provide the start time of the promptemission t prompt , the start time of the precursor emission with respect to t prompt and the duration of the precursoremission. To access this table in a digital format, please visit https://icecube.wisc.edu/~grbweb_public/Precursors.html . GRB t prompt (UTC) t precursor (s) Duration (s) bn080723557 13:22:55.412 -34.284 28.319bn080807993 23:50:44.177 -11.612 1.032bn080816503 12:04:39.495 -21.823 1.823bn080818579 13:54:44.589 -20.361 5.596bn080830368 08:50:22.699 -8.559 5.112bn081003644 15:27:27.738 -11.363 4.320bn081121858 20:35:31.671 -8.498 7.855bn090101758 18:13:07.574 -86.950 6.082bn090113778 18:40:38.870 -0.475 0.150bn090117335 08:02:26.183 -24.653 1.296bn090131090 02:09:43.196 -22.324 12.445bn090309767 18:25:41.699 -36.134 6.122bn090326633 15:10:16.566 -583.057 0.256bn090326633 15:10:16.566 -580.753 5.376bn090419997 23:55:38.751 -37.348 23.251bn090425377 09:04:14.740 -44.805 2.705bn090428441 10:34:37.862 -26.762 18.048bn090502777 18:40:11.917 -37.539 3.065bn090510016 00:23:00.368 -0.420 0.024bn090602564 13:32:22.296 -1.242 0.683bn090610723 17:22:58.385 -90.937 6.686bn090618353 08:29:16.651 -50.628 28.946bn090720710 17:02:57.665 -0.776 0.264bn090810659 15:50:40.542 -94.594 43.262bn090811696 16:41:54.351 -4.958 1.583bn090814950 22:48:30.233 -43.778 18.577bn090815946 22:44:41.956 -179.466 12.722bn090820509 12:13:25.368 -8.951 4.124bn090907017 00:24:10.767 -1.967 1.664bn090929190 04:33:04.488 -0.571 0.122bn091109895 21:28:49.421 -9.606 2.788bn100116897 21:32:19.006 -83.382 6.319bn100130729 17:30:19.867 -65.378 23.215bn100204566 13:34:36.243 -16.948 15.677bn100323542 13:01:32.005 -54.935 9.109bn100326402 09:37:30.596 -55.808 32.512bn100424876 21:03:54.875 -123.791 2.521bn100517154 03:42:30.304 -22.365 1.362bn100619015 00:22:24.001 -77.870 9.918bn100625891 21:22:58.362 -15.645 4.029bn100709602 14:28:25.731 -56.254 16.328bn100718160 03:50:13.287 -25.036 6.090bn100718160 03:50:13.287 -7.415 6.808bn100730463 11:06:50.220 -41.808 12.805bn100730463 11:06:50.220 -18.243 0.001bn100827455 10:55:49.710 -0.442 0.079bn100923844 20:15:31.462 -24.128 4.019bn101030664 15:56:24.411 -69.697 31.744bn101224578 13:53:30.861 -33.455 10.658bn101227536 12:51:49.785 -3.895 3.646bn110102788 18:55:41.740 -67.434 25.256bn110227229 05:30:09.611 -111.145 21.120bn110428338 08:07:18.821 -70.448 42.874bn110428338 08:07:18.821 -18.748 13.398bn110528624 14:59:12.297 -217.477 11.264bn110528624 14:59:12.297 -35.653 13.654bn110528624 14:59:12.297 -21.303 13.839 Continued on next page
Table I continued: Temporal properties of the identified precursors.
GRB t prompt (UTC) t precursor (s) Duration (s)(s) Duration (s)
GRB t prompt (UTC) t precursor (s) Duration (s)(s) Duration (s) bn110725236 05:39:57.932 -16.720 7.619bn110729142 03:30:47.288 -342.504 52.731bn110729142 03:30:47.288 -185.188 51.556bn110825102 02:26:58.702 -7.864 0.814bn110903111 02:42:41.553 -187.466 22.062bn110904124 02:58:55.085 -44.632 7.665bn110909116 02:47:01.914 -4.433 1.670bn110926107 02:34:30.183 -45.717 3.110bn111010709 17:01:07.319 -34.749 31.018bn111015427 10:15:22.011 -25.770 17.144bn111228657 15:45:16.506 -46.111 10.496bn111228657 15:45:16.506 -32.543 11.776bn111230683 16:23:06.415 -11.301 4.631bn111230819 19:39:41.521 -9.814 1.304bn111230819 19:39:41.521 -8.120 4.234bn120118709 17:00:24.779 -6.498 5.475bn120308588 14:06:05.511 -21.363 3.092bn120319983 23:35:18.709 -17.629 5.551bn120412920 22:05:51.344 -71.057 5.502bn120504945 22:40:07.713 -1.369 0.799bn120513531 12:44:14.932 -15.008 1.330bn120530121 02:54:31.969 -50.475 7.974bn120611108 02:35:54.181 -8.321 6.602bn120710100 02:25:09.865 -113.086 4.857bn120711115 02:45:52.633 -61.735 4.838bn120716712 17:08:00.170 -176.365 5.383bn120819048 01:09:20.076 -60.316 7.618bn120819048 01:09:20.076 -30.405 1.638bn121005340 08:10:54.001 -101.730 38.794bn121029350 08:24:27.774 -11.090 8.798bn121031949 22:50:21.029 -191.769 38.495bn121113544 13:03:25.589 -45.362 31.652bn121125356 08:32:50.026 -29.374 20.325bn121217313 07:29:53.089 -714.103 65.792bn130104721 17:18:12.706 -5.969 3.898bn130106995 23:52:56.117 -33.325 17.558bn130208684 16:24:43.858 -21.975 5.099bn130209961 23:03:46.502 -5.102 4.597bn130219775 18:36:47.745 -56.310 20.260bn130310840 20:09:45.591 -4.755 1.194bn130318456 10:57:50.305 -82.735 6.897bn130320560 13:29:06.051 -159.315 42.085bn130404840 20:10:25.030 -21.354 8.355bn130418844 20:16:08.506 -87.313 16.452bn130504314 07:32:36.037 -32.672 0.464bn130623130 03:07:03.470 -26.744 1.821bn130720582 13:59:14.940 -146.139 115.366bn130813791 18:59:18.842 -5.810 1.680bn130815660 15:51:22.993 -31.482 6.925bn130818941 22:34:29.441 -70.463 8.706bn130919173 04:09:40.924 -0.686 0.236bn131014513 12:18:34.911 -20.917 2.089bn131108024 00:34:43.981 -2.395 1.815bn140104731 17:34:01.991 -120.439 66.204bn140104731 17:34:01.991 -24.501 1.459bn140108721 17:19:53.720 -71.900 11.570bn140126815 19:33:40.215 -62.234 20.478bn140126815 19:33:40.215 -24.368 14.110bn140304849 20:25:37.760 -189.609 30.654bn140329295 07:04:57.833 -19.534 0.630 Continued on next page Table I continued: Temporal properties of the identified precursors.
GRB t prompt (UTC) t precursor (s) Duration (s)(s) Duration (s)
GRB t prompt (UTC) t precursor (s) Duration (s)(s) Duration (s) bn140404030 00:43:22.825 -71.917 7.657bn140512814 19:33:23.687 -98.421 11.788bn140621827 19:50:14.988 -4.111 0.718bn140628704 16:54:21.456 -66.005 4.910bn140709051 01:13:51.906 -11.597 5.700bn140714268 06:27:35.035 -109.468 27.544bn140716436 10:29:26.513 -89.084 2.218bn140818229 05:31:17.613 -69.604 10.233bn140824606 14:34:24.964 -73.928 12.933bn140825328 07:53:42.446 -59.289 11.821bn140825328 07:53:42.446 -38.258 3.215bn140917512 12:17:10.292 -4.434 3.940bn141029134 03:14:24.675 -66.449 3.739bn141029134 03:14:24.675 -41.574 6.940bn141102536 12:51:40.471 -1.269 0.088bn150126868 20:51:32.131 -55.037 13.019bn150127398 09:32:49.909 -6.512 5.747bn150226545 13:08:44.224 -202.152 1.028bn150226545 13:08:44.224 -155.467 7.878bn150226545 13:08:44.224 -41.188 16.158bn150330828 19:53:59.254 -98.194 11.512bn150416773 18:33:22.811 -824.534 42.496bn150422703 16:52:31.997 -468.581 15.616bn150506398 09:33:46.679 -116.285 27.791bn150508945 22:40:36.620 -102.265 15.712bn150512432 10:23:46.759 -86.467 43.029bn150512432 10:23:46.759 -28.593 20.212bn150522433 10:24:07.264 -19.511 7.822bn150523396 09:30:14.993 -28.370 19.748bn150627183 04:23:22.017 -458.665 3.072bn150702998 23:56:45.108 -6.691 2.490bn150703149 03:33:54.082 -13.280 0.008bn150830128 03:04:38.646 -14.638 14.021bn151027166 04:00:00.254 -96.360 40.571bn151030999 23:59:47.634 -88.314 17.686bn151211672 16:07:28.188 -151.405 26.022bn160131174 04:12:52.609 -179.691 44.007bn160201883 21:11:44.177 -1.590 0.968bn160215773 18:36:08.605 -109.239 44.645bn160219673 16:11:34.712 -110.393 12.546bn160223072 01:45:54.364 -95.615 10.496bn160225809 19:25:09.731 -48.115 23.215bn160512199 04:45:57.662 -56.663 9.377bn160519012 00:18:55.054 -83.260 3.345bn160519012 00:18:55.054 -65.164 17.101bn160523919 22:04:13.977 -38.410 5.424bn160625945 22:43:14.090 -178.317 2.418bn160724444 10:40:02.521 -7.324 1.790bn160821857 20:36:22.642 -117.067 31.832bn160825799 19:10:50.313 -1.449 0.599bn160908136 03:16:48.679 -87.733 6.845bn160912521 12:31:42.840 -57.422 36.635bn160912521 12:31:42.840 -17.072 5.193bn160919613 14:43:36.685 -24.729 0.498bn160919613 14:43:36.685 -15.527 0.761bn161105417 10:01:18.575 -30.217 12.749bn161111197 04:44:50.633 -102.555 11.125bn161117066 01:37:14.177 -103.474 77.027bn161119633 15:11:02.131 -10.916 7.666bn161129300 07:11:45.292 -5.373 0.040 Continued on next page Table I continued: Temporal properties of the identified precursors.
GRB t prompt (UTC) t precursor (s) Duration (s)(s) Duration (s)
GRB t prompt (UTC) t precursor (s) Duration (s)(s) Duration (s) bn170109137 03:21:41.186 -245.940 18.163bn170109137 03:21:41.186 -217.040 6.377bn170115662 15:54:01.580 -95.287 18.563bn170209048 01:09:05.007 -28.188 8.228bn170302719 17:15:41.992 -22.294 12.259bn170323775 18:36:31.186 -12.963 12.697bn170402961 23:03:40.777 -15.936 1.501bn170402961 23:03:40.777 -12.442 0.230bn170416583 14:00:34.758 -35.298 12.494bn170514152 03:38:43.989 -5.895 0.678bn170514180 04:19:54.177 -79.666 35.908bn170830069 01:38:59.546 -19.395 5.987bn170831179 04:18:03.061 -73.621 8.547bn170831179 04:18:03.061 -43.400 6.309bn170923188 04:31:15.015 -10.012 1.018bn171004857 20:33:34.433 -2.263 1.378bn171102107 02:34:03.231 -29.516 10.393bn171112868 20:50:13.004 -198.952 8.192bn171112868 20:50:13.004 -43.928 9.502bn171120556 13:20:33.596 -31.460 4.221bn171211844 20:17:18.932 -82.541 12.393bn180124392 09:23:59.613 -4.987 0.611bn180126095 02:16:29.991 -820.685 11.776bn180307073 01:44:35.183 -39.275 23.342bn180411519 12:28:28.650 -54.086 26.673bn180416340 08:10:01.701 -36.541 10.291bn180426549 13:10:59.907 -13.182 5.544bn180618724 17:22:55.701 -61.611 26.238bn180620354 08:29:22.735 -72.842 5.855bn180710062 01:29:21.269 -49.933 13.542bn180720598 14:21:26.039 -29.189 10.000bn180728728 17:29:11.437 -15.219 10.040bn180822423 10:08:32.522 -5.898 2.803bn180822562 13:30:29.570 -128.070 7.513bn180822562 13:30:29.570 -118.178 6.344bn180906988 23:42:36.388 -2.471 1.039bn180929453 10:52:35.121 -1.456 0.606bn181007385 09:14:19.608 -23.373 3.996bn181008877 21:04:29.161 -131.183 27.879bn181119606 14:32:19.202 -2.566 1.798bn181122381 09:09:04.964 -1.937 0.299bn181203880 21:06:37.705 -6.482 0.870bn181222279 06:42:52.975 -79.631 40.808bn190114873 20:57:02.490 -5.573 1.942bn190114873 20:57:02.490 -2.854 1.537bn190205938 22:31:11.876 -40.086 9.198bn190228973 23:21:30.204 -15.148 7.989bn190310398 09:33:20.756 -49.157 4.120bn190315512 12:17:39.138 -366.193 6.912bn190323879 21:05:17.600 -893.855 26.624bn190324947 22:44:18.392 -17.146 2.474bn190326314 07:32:13.823 -27.769 1.672bn190326314 07:32:13.823 -18.099 2.115bn190610750 18:00:04.042 -14.819 1.160bn190611950 22:48:51.696 -62.594 20.082bn190719624 15:00:01.045 -86.830 1.579bn190806675 16:12:34.836 -1.664 1.188bn190828542 12:59:58.210 -46.588 38.536bn190829830 19:56:40.582 -47.965 5.565bn190901890 21:21:37.555 -63.144 20.014 Continued on next page Table I continued: Temporal properties of the identified precursors.
GRB t prompt (UTC) t precursor (s) Duration (s)(s) Duration (s)