Search for Gravitational Waves Associated with Fast Radio Bursts Detected by CHIME/FRB during the LIGO–Virgo Observing Run O3a

We search for gravitational-wave ( GW ) transients associated with fast radio bursts ( FRBs ) detected by the Canadian Hydrogen Intensity Mapping Experiment Fast Radio Burst Project, during the ﬁ rst part of the third observing run of Advanced LIGO and Advanced Virgo ( 2019 April 1 15:00 UTC – 2019 October 1 15:00 UTC ) . Triggers from 22 FRBs were analyzed with a search that targets both binary neutron star ( BNS ) and neutron star – black hole ( NSBH ) mergers. A targeted search for generic GW transients was conducted on 40 FRBs. We ﬁ nd no signi ﬁ cant evidence for a GW association in either search. Given the large uncertainties in the distances of our FRB sample, we are unable to exclude the possibility of a GW association. Assessing the volumetric event rates of both FRB and binary mergers, an association is limited to 15% of the FRB population for BNS mergers or 1% for NSBH mergers. We report 90% con ﬁ dence lower bounds on the distance to each FRB for a range of GW progenitor models and set upper limits on the energy emitted through GWs for a range of emission scenarios. We ﬁ nd values of order 10 51 – 10 57 erg for models with central GW frequencies in the range 70 – 3560 Hz. At the sensitivity of this search, we ﬁ nd these limits to be above the predicted GW emissions for the models considered. We also ﬁ nd no signi ﬁ cant coincident detection of GWs with the repeater, FRB 20200120E, which is the closest known extragalactic FRB.


Introduction
Fast radio bursts (FRBs) are millisecond duration radio pulses that have been observed out to cosmological distances, several with inferred redshifts greater than unity (Lorimer et al. 2007;Cordes & Chatterjee 2019;Petroff et al. 2019).Although intensely studied for more than a decade, the emission mechanisms and progenitor populations of FRBs are still one of the outstanding questions in astronomy.
Some FRBs have been shown to repeat (Amiri et al. 2019a; CHIME/FRB Collaboration et al. 2019;Kumar et al. 2019), and the recent association of an FRB with the Galactic magnetar SGR 1935+2154 proves that magnetars can produce FRBs (Bochenek et al. 2020;CHIME/FRB Collaboration et al. 2020).Alternative progenitors and mechanisms to produce nonrepeating FRBs are still credible and have so far not been ruled out (Zhang 2020a).Data currently suggest that both repeating and nonrepeating classes of FRBs have dispersion measures (DMs), a quantity equal to the integral of the free electron density along the line of sight, and sky locations consistent with being drawn from the same population.However, the two classes have been shown to differ in their intrinsic temporal widths and spectral bandwidths (CHIME/ FRB Collaboration et al. 2021).Whether genuine nonrepeating sources have a different origin compared to that of their repeating cousins is an unresolved question.
The first discovery of an FRB was made over a decade ago by Parkes 64 m radio telescope (Lorimer et al. 2007).This burst, FRB 010724 or FRB 20010724A, known as the Lorimer burst, first indicated an extragalactic origin for FRBs through its observed DM.This burst had a DM of 375 pc cm −3 , far in excess of the likely Galactic DM contribution along the line of sight (of order 45 pc cm −3 for this event), supporting an extragalactic origin.The precise localizations of FRB host galaxies have since unambiguously confirmed an extragalactic hypothesis (Chatterjee et al. 2017;Bannister et al. 2019;Heintz et al. 2020;Li & Zhang 2020), and constraints on the progenitor population are starting to be understood (e.g., Bhandari et al. 2020).The inferred cosmological distances for many FRBs have shown that these transients have extreme luminosities by radio standards, of the order 10 38 -10 46 erg s −1 (Zhang 2018).
The FRB detection rate has greatly increased since the Canadian Hydrogen Intensity Mapping Experiment (CHIME) instrument (Bandura et al. 2014;Newburgh et al. 2014; CHIME/FRB Collaboration 2020; see) 317 began its commissioning phase in late 2018, and its first FRB observation run shortly after.The CHIME radio telescope observes in the frequency range 400-800 MHz and consists of four 20 × 100 m cylindrical parabolical reflectors.Its large collecting area and wide field of view (≈200 deg 2 ) make it a valuable survey instrument for radio transients.FRB detection for this instrument has been led by the CHIME/FRB project (CHIME/FRB Collaboration et al. 2018), which published its first sample of 13 FRBs during its early commissioning phase, despite operating at a lower sensitivity and field of view than design specifications (Amiri et al. 2019b). 314Deceased, August 2020. 315McGill Space Institute Fellow. 316 FRQNT Postdoctoral Fellow.Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
317 https://chime-experiment.ca/ The CHIME/FRB project recently published a catalog of 535 FRBs detected during their first year of operation; this includes 62 bursts from 18 previously identified repeating sources (CHIME/FRB Collaboration et al. 2021).This is the first large collection, (100 s), of FRBs from a homogeneous survey and represents a significant milestone in this area of study.The CHIME/FRB data is supportive of different propagation or emission mechanisms between repeaters and nonrepeaters; however, it is still not clear whether all FRBs do repeat (Ravi 2019), and significantly, the FRB emission mechanism remains unknown.There presently exist many competing FRB emission theories (Platts et al. 2019), some of which predict the accompaniment of a time-varying mass quadrupole moment, and, thus, the emission of gravitational waves (GWs).
A number of studies have looked at the possibility of GW emission associated with FRBs indirectly, using radio observations to search for coherent FRB-like emissions associated with short, hard gamma-ray bursts (GRBs; Anderson et al. 2018;Rowlinson & Anderson 2019;Bouwhuis et al. 2020;Gourdji et al. 2020;Rowlinson et al. 2021).
The identification of an FRB within the sensitive reach of GW interferometric detectors could provide conclusive proof of an association or constrain the parameters of the emission mechanisms for a given FRB.The increased population of detected FRBs from the CHIME/FRB Project therefore offers a unique chance of achieving this endeavor.
The first search for GW counterparts to transient radio sources was conducted by Abbott et al. (2016).This used a minimally modeled coherent search (X-Pipeline) ±2 minutes around the detection time of 6 Parkes FRBs using GW data from GEO600 (Grote 2010) and initial Virgo (Accadia et al. 2012).No GW coincidences were found, but this study provided a useful framework for future searches using improved GW sensitivities.
In this paper, we present the second targeted GW follow-up of FRBs using bursts detected by CHIME/FRB during the first part of the third observing run of Advanced LIGO and Advanced Virgo (O3a; Aasi et al. 2015;Acernese et al. 2015), which took place between 2019 April 1 15:00 UTC and 2019 October 1 15:00 UTC.This search uses both a generic GW transient search and a modeled search targeting coalescing binary systems.
The organization of this paper is as follows: in Section 2, we describe the motivation of this study by discussing possible GW counterparts to FRBs.We introduce the CHIME/FRB data sample in Section 3 and in Section 4 discuss the GW search methods employed; this includes an overview of both of the pipelines used in our analysis.Section 5 provides the results of the GW analysis of the FRB sample.In Section 6, we report results of a gravitational-wave analysis of the repeater, FRB 20200120E, which is the closest known extragalactic FRB.Finally, in Section 7, we summarize the astrophysical implications of our results and discuss future GW searches for FRB counterparts at greater GW sensitivities.

Proposed Gravitational-wave Counterparts to FRBs
This section will review some of the more popular models of nonrepeating and repeating FRBs that could provide plausible GW counterparts and could therefore be constrained or confirmed through GW searches.(An online theory catalog tracks new FRB models; see). 318s the millisecond durations of FRBs indicate compact emission regions, many models of nonrepeating FRBs have suggested cataclysmic events, including coalescing compact objects.As will be discussed below, the fraction of the energy budget emitted by proposed FRB emission models is comparatively small compared to (10 52 ) erg emitted in GWs (e.g., Abbott et al. 2017c) but high by radio standards.
A number of studies have investigated the possibility of FRB-like emissions from binary neutron star (BNS) coalescence around the time of merger (see review in Platts et al. 2019).During this phase, the magnetic fields of the neutron stars (NSs) are synchronized to binary rotation, and a coherent radiation could be generated due to magnetic braking.The mechanism requires magnetic fields of the order 10 12 -10 13 Gauss and could lead to energy-loss rates of the order 10 45 erg s −1 .The predicted FRB pulse widths are consistent with the timescale of the orbital period of the BNS just prior to coalescence (Totani 2013).Wang et al. (2016) considered that an FRB could be produced during the final stages of a BNS inspiral through magnetic reconnection due to the interaction of a toroidal magnetic field, produced as the NS magnetospheres approach each other.The predicted energy-loss rates are order 10 42 erg s −1 assuming magnetic fields of the order 10 12 Gauss.One should note, the dynamic ejecta launched shortly after the final merger would produce significant opacity over a large solid angle, thus screening an FRB-type signal via absorption (Yamasaki et al. 2018).Zhang (2020b) has also entertained the idea that similar interactions between the two NS magnetospheres could produce repeating FRB-like coherent radio emissions decades or centuries before the final plunge.
Other mechanisms to produce prompt coherent radio emission on millisecond timescales include excitation of the circumbinary plasma by GWs (Moortgat & Kuijpers 2005), from dynamically generated magnetic fields post-merger (Pshirkov & Postnov 2010), or from the collision of a GRB forward shock with the surrounding medium (Usov & Katz 2000;Sagiv & Waxman 2002).
Mergers of significant fractions of BNSs are likely to give rise to millisecond magnetars (Gao et al. 2016;Margalit et al. 2019), although this is highly dependent on the unknown nuclear equation of state (see Sarin & Lasky 2021, for a review).If the remnant NS mass is greater than the maximum nonrotating mass, it can survive for hundreds to thousands of seconds before collapsing to form a BH (Ravi & Lasky 2014).As the magnetic field lines snap as they cross the black hole (BH) horizon, an outwardly directed magnetic shock would dissipate as a short, intense radio burst (Falcke & Rezzolla 2014;Zhang 2014).The energy in the magnetic shock can be estimated as (10 47 ) erg, which is more than sufficient to support an FRB emission.This model has been motivated by the observation of relatively long-lived X-ray plateaus following short gamma-ray bursts that exhibit an abrupt decay phase, commonly interpreted as the collapse of the nascent NS to a BH (Troja et al. 2007;Lyons et al. 2010;Rowlinson et al. 2010Rowlinson et al. , 2013)).Such collapses are expected to occur  5 × 10 4 s after the merger (Ravi & Lasky 2014).
The detection of the intense millisecond duration radio associated with the Galactic magnetar SGR 1935+2154 (CHIME/FRB Collaboration et al. 2020) has provided significant evidence to an FRB-magnetar connection (Popov & Postnov 2013).It is known that the energy stored in rotational kinetic energy and the magnetic field of a millisecond pulsar is ample to power a repeating FRB (Metzger et al. 2017).In terms of the energy, Margalit et al. (2020) used the energyloss rates of repeater FRB 20121102A to estimate an energy budget for repeaters at least 10 47 -10 49 erg.This lower limit is based on the so-far observed pulses and without consideration of beaming, so could increase with further monitoring of this source (Petroff et al. 2022).
Resonant oscillation modes in the core and crust of magnetars have been suggested to cause quasi-periodic oscillations observed in the X-ray tails of giant flares.If the process by which FRBs are created also excites nonradial modes in the magnetars, then GWs could simultaneously be produced (e.g., Levin & van Hoven 2011;Quitzow-James et al. 2017).
The stellar oscillation mode that couples strongest to GW emission is the fundamental f-mode.The frequency of this mode depends on the equation of state; however, analyses of the tidal deformability of GW170817 are consistent with NS f-mode frequencies typically being around 2 kHz (Abbott et al. 2017b(Abbott et al. , 2017d(Abbott et al. , 2018;;Wen et al. 2019).This is above the most sensitive frequency of the Advanced LIGO/Virgo observatories.
Other modes such as gravity g-modes (here the restoring force is buoyancy) and Rossbyr-modes (where the restoring force is the Coriolis force) emit at frequencies closer to the most sensitive range for Advanced LIGO/Virgo; however, these modes couple more weakly to gravitational modes, and are therefore not likely to be detectable in association with an FRB.

The CHIME/FRB Sample
The CHIME/FRB data sample provided for this analysis consists of 338 bursts observed within O3a out of 806 total bursts.Out of this sample, 168 bursts have been published in the first CHIME/FRB catalog (CHIME/FRB Collaboration et al. 2021).Within the sample of 338 bursts, only events overlapping with up-time of at least one of the three GW observatories were considered for analysis.Within this subsample, the selection of bursts that were analyzed was based on the inferred distance to each burst.This selection will be described at the end of this section, after the calculation of the inferred distance is described.
The data for each FRB include localization information, a topocentric arrival time, and a measure of the total DM.For each burst, a transient name server (TNS; see https://www.wis-tns.org)designation was also provided.The TNS naming convention takes the form FRB YYYYMMDDLLL with the year (YYYY), month (MM), and day (DD) information in UTC and LLL a unique, sequential letter code starting withAup to Z, then from AA to AZ, up to ZZZ, inorder of detection on any given day.
The arrival time at the CHIME instrument's location (topocentric) at 400 MHz was converted to a dedispersed arrival time using the DM value associated with each event.This time was used as the central event time around which each GW search was conducted.
The localization information of each FRB is in the form of up to 5 disjoint error regions of varied morphology centered around the region with the highest signal-to-noise ratio (S/N); each separate localization island has a central value and a 95% confidence uncertainty region.An example is shown in Figure 1.
The localization regions are reported in the sample as a list of 5 R.A. (R.A.) values, 5 95% confidence uncertainty region sizes for the R.A. values, 5 decl.(decl.)values, and 5 95% confidence uncertainty region sizes for the decl.values.The different approaches to these localization data adopted by the generic transient and modeled search pipelines will be described in Section 4.
To determine a measure of the luminosity distance of each FRB, we employ the Macquart relation (Macquart et al. 2020).This relation maps the redshift to the quantity DM IGM , which is the DM contribution from extragalactic gas along the line of sight; this can be obtained after all other contributions are subtracted.Taking into account all contributions to the total DM, the quantity DM T , a measure of redshift can therefore be determined by solving the following: where DM MW is the Milky Way contribution to the DM along the line of sight, DM halo is the contribution from the Milky Way halo, and DM host is the contribution from the host galaxy, which is corrected by the cosmic expansion factor.The estimates of z are then converted to a luminosity distance assuming a "flat-Λ" cosmology with the cosmological parameters Ω m = 0.31, Ω Λ = 0.69, and H 0 = 67.8km s −1 Mpc −1 (Planck Collaboration et al. 2016).
To determine redshift values for each FRB, we employ the Bayesian Markov Chain Monte Carlo (MCMC) sampling framework described in Bhardwaj et al. (2021b) with a posterior distribution defined by the following: q is the likelihood distribution of the observed quantity DM T,O given the parameters q, ( ˆ) p q are the prior distributions on q, and  is the Bayesian evidence; this latter factor enters Equation (2) as a normalization factor independent of the model parameters and can be ignored if one is only interested in the posterior distribution rather than model selection.We assume a Gaussian likelihood function provided as the following: with σ the uncertainty on DM T,O for each burst and DM T given by Equation (1) (Rafiei-Ravandi et al. 2021).
For the Milky Way contribution DM MW , there is no consensus between the two popular models of Cordes & Lazio (2002), Yao et al. (2017).Therefore, we follow Bhardwaj et al. (2021b) and assume a Gaussian prior based around the minimum of DM MW from these two models along the line of sight; a standard deviation of 20% of this value is also used.
The contribution DM halo has been estimated in a number of studies but is quite uncertain.For example, Yamasaki & Totani (2020) found values of DM halo ∼ 30-245 pc cm −3 using a two component model.Studies by Dolag et al. (2015) found values between DM halo ∼ 30-50 pc cm −3 based on cosmological simulation, and Prochaska & Zheng (2019) estimated values between 30 and 80 pc cm −3 .To take account of the large uncertainty in this quantity, we follow Bhardwaj et al. (2021b) and assume a Gaussian prior such that, at 3σ, DM halo has a value 0 or 80 pc cm −3 .
The prior on DM IGM assumes the parameterization Δ = DM IGM /〈DM IGM 〉 with the denominator obtained through the Macquart relation.This takes the form provided in Macquart et al. (2020): with σ DM = 0.2z −0.5 , and [α, β] = 3; the value of C is determined by requiring that 〈Δ〉 = 1.The form of this model is motivated by the requirement that the DM distribution approaches a Gaussian at small σ DM in accordance with the Gaussianity of large scale structure.It also incorporates a skew at large σ DM to reflect the possibility of overdensities along the line of sight.
Finally, for a prior on DM host , we adopt a lognormal distribution with median e μ = 68.2, and logarithmic width parameter σ host = 0.88 as in Macquart et al. (2020).
The quantities outlined above have a large range of uncertainty, and there could be additional contributions, e.g., circumburst material.As a result, redshift values calculated from DMs are generally taken as upper limits.We perform MCMC sampling using the emcee package (Foreman-Mackey et al. 2013) based on an affine-invariant sampling algorithm (Goodman & Weare 2010) using 256 walkers of 20,000 samples.The inferred values of z, and thereby luminosity distance, and their 90% credible intervals are thus determined for each FRB, based on the observed values of DM T , R.A. and decl., the estimated DM MW along the line of sight, and the priors on other DM contributions described above.
Given the large uncertainties in the distances of FRBs, we based our analysis and results on the 90% credible intervals inferred for the CHIME/FRB sample of bursts.However, for illustration, we show in Figure 2 the distribution of the median distances of the total sample of 338 FRBs that occurred during O3a.The plot shows that most events seem to occur within 1700 Mpc (z ∼ 0.3) and 6000 Mpc (z ∼ 0.9).The closest events in the distribution include a significant number of repeating FRBs.Due to the relatively limited range of the GW detectors, in selecting which bursts to analyze, we first downselected the sample to all bursts from the closest 10% of CHIME/FRB nonrepeating bursts that have GW detector network data available for analysis (if the recent CHIME/FRB catalog of 535 bursts is representative of the FRB population, at least around 11% of FRBs repeat).Within this selection, a coherent analysis using modeled waveforms was then conducted on a smaller subset of the closest 22 nonrepeating events for which data were available from at least one interferometric GW detector, and a generic transient coherent analysis was conducted on a subset of FRBs for which data were available from at least two interferometric GW detectors.The further downselection to the final set of analyses reported was based on two considerations.For some events, the systematic noise in the detector was too significant near the time of the burst for one or both of our two searches, and these events were then excluded.Finally, as each search requires significant person power and computational resources, we performed searches on the remaining subset of events in order of increasing distance, until we reached a point of diminishing returns caused by the reduced overlap between the effective detection range of the GW detection network and the inferred distance to each FRB event.These considerations yielded a sample of 34 nonrepeating FRBs that were analyzed by one or both types of analysis.Using the same considerations for selection, we analyzed a total of 11 repeated bursts from the closest 3 repeating sources: FRB 20180916B (7 repeat events during O3A), FRB 20180814A (2 repeat events), and FRB20190303A (2 events).The lower and upper 90% limits of the credible Figure 2. The distribution of inferred median distances for the CHIME/FRB data sample based on the MCMC analysis of Section 3; there is a large uncertainty in these distances; thus, this distribution should be taken as only an approximate representation.The distribution peaks between 1700 Mpc (z ∼ 0.3) and 6000 Mpc (z ∼ 0.9).The closest nonrepeating event analyzed in our sample was FRB 20190425A for which we inferred a median distance of 133 Mpc and a range [13-386] Mpc at 90% confidence; the most distant was FRB 20190601C with a median inferred distance of 914 Mpc within a range  Mpc.
intervals on the luminosity distances to each of the nonrepeating FRBs analyzed are included in the tables in Section 5.

Search Methods
Here, we will provide a description of the two targeted search methods used in this paper.These are the same methods applied to search for GW events coincident with GRBs that occurred during the first (Abbott et al. 2017a), second (Abbott et al. 2019a), and third (Abbott et al. 2021) Advanced LIGO and Advanced Virgo observing runs.In Section 4.1, we describe the modeled search method that aims to uncover subthreshold GW signals emitted by BNS and neutron starblack hole (NSBH) binaries (PyGRB; Harry & Fairhurst 2011;Williamson et al. 2014), highlighting choices in analysis configuration that are unique to the follow-up of FRB events.In Section 4.2, we discuss the search for generic GW transients (X-Pipeline; Sutton et al. 2010;Was et al. 2012).

PyGRB-Modeled Search for Binary Mergers
The modeled search for GWs associated with FRB events makes use of the PyGRB data analysis pipeline (Harry & Fairhurst 2011;Williamson et al. 2014), and the search is configured to be similar to the search for GW signals coincident with GRBs in O3a (Abbott et al. 2021).This is a coherent matched-filtering pipeline that compares the GW detector network data with a bank of pregenerated waveforms, including the inspiral of BNS and NSBH binaries.PyGRB uses the PyCBC (Nitz et al. 2020) open-source framework for distribution of the analysis of the GW data across large computing clusters, and also relies on several elements of the LALSuite software library (LIGO Scientific Collaboration 2018).
The PyGRB analysis searches the combined detector data in the range 30-1000 Hz.A set of coherent data streams is formed by combining the data from the detectors, using a sample of sky positions in the region reported for the FRB event that is being studied.These data streams are then compared using matched filtering to the same predefined bank of waveform templates (Owen & Sathyaprakash 1999) used in the search for GWs associated with GRBs events in O3a (Abbott et al. 2021).The bank is created with a hybrid of geometric and stochastic template placement methods across target search space (Harry et al. 2008(Harry et al. , 2014;;Brown et al. 2012;Capano et al. 2016;Dal Canton & Harry 2017), using a phenomenological inspiralmerger-ringdown waveform model for non-precessing pointparticle binaries (IMRPhenomD; Husa et al. 2016;Khan et al. 2016).This bank of templates is designed to cover binary masses in the range [1.0, 2.8]M ☉ for NSs, and [1.0, 25.0]M ☉ for BHs.The bank also allows for aligned-spin, zeroeccentricity BNS and NSBH, with dimensionless spins in the range [0, 0.05] for NSs and [0, 0.998] for BHs.
Coherent matched filtering can be susceptible to loud transient noise in the detector data and can produce a high S/N (Nitz et al. 2017).To combat this, the analysis performs additional tests on each point of high S/N data, which we also refer to as triggers.These tests can either remove the trigger or reweight the S/N using a χ 2 test.This latter test determines how well the data agrees with the template over the whole template duration.Such cuts and reweighting significantly improve the ability of the search to distinguish a GW from many types of transient noise, thus improving the significance of real GW triggers.The final reweighted S/N of each candidate event is used as the measure of its relative significance, or ranking statistic, within the search.
The PyGRB analysis searches for GW inspiral events that merge within 12 s of the dedispersed event time of each FRB, with an asymmetric on-source window starting 10 s before the FRB event and ending 2 s after the event.The search window is chosen to strike a balance between maximizing the possible progenitor models through a wider window or maximizing the sensitivity of the search by using a narrower window.In this search, we seek a GW signal with a merger time close to the time of the FRB, assuming the FRB results from the interaction of the two binary components.
The sensitivity of the search is governed by the comparison between the most significant event in the on-source window and the most significant event in equivalent trial searches of 12 s windows in the surrounding data, known as the off-source trials.These off-source trials form the background data for the search, and if a sufficient number of background trials are conducted, this allows the search to determine the significance of any candidate events in the on-source window to the level needed to make a confident detection statement by computing a false-alarm probability.
If multiple detectors are available, then additional effective background data can be produced by combining the data from the detectors with an intentional misalignment in time of at least the light-travel time across the network to ensure any detected events cannot possibly be true coherent GW candidates (Williamson et al. 2014).This can be repeated for multiple possible time shifts, and in this search, these time shifts are set to match the on-source window length of 12 s.This produces fewer time shifts than a 6 s on-source window, as used in previous searches for GW associated with GRB events such as Abbott et al. (2021).This again impacts the effective significance of any detected events, because the amount of background data used by the search is limited by the amount of coherently analyzable data for all detectors in the network that surrounds the target time.Thus, a search is only conducted if a minimum of 30 minutes of data is available.
In the results section, we report the effective range of each search conducted as a 90% exclusion distance, D 90 .This is calculated by first creating a set of simulated GW signals to inject into the off-source data, then attempting to find these injected signals with the standard search pipeline.The signals are injected with amplitudes appropriate for a distribution of distances between their simulated origin and the detectors, and the D 90 distance is defined as the distance within which 90% of the injected simulated signals are recovered with a ranking statistic greater than the loudest on-source event.
Mirroring the approach taken in the O3a search for GW events associated with GRB detections (Abbott et al. 2021), the injected signals include BNS systems with dimensionless spins in the range −0.4 to 0.4, taken from observed pulsar spins (Hessels et al. 2006), and are distributed uniformly in spin and with random orientations.The injections also include alignedspin NSBH binaries, and NSBH binaries with generically oriented spins up to 0.98, motivated by X-ray binary observations (e.g., Özel et al. 2010;Kreidberg et al. 2012;Miller & Miller 2014).The simulated signals are intentionally generated using different GW signal models than those used in the matched-filtering template bank, to approximate the target search space difference between the approximate templates used and the true GW signals.In particular, the injected waveforms are identical to those used in the equivalent O3a GRB event follow-up analysis (Abbott et al. 2021).Precessing BNS signals are simulated using the TaylorT2 time-domain, post-Newtonian inspiral approximant (SpinTaylorT2; Sathyaprakash & Dhurandhar 1991;Blanchet et al. 1996;Mikoczi et al. 2005;Arun et al. 2009;Bohé et al. 2013Bohé et al. , 2015;;Mishra et al. 2016), while NSBH injected waveforms are generated assuming a point-particle effective-one-body model tuned to numerical simulations, which can allow for precession effects from misaligned spins (SEOBNRv3; Pan et al. 2014;Taracchini et al. 2014;Babak et al. 2017).Again, identical to the injections used in Abbott et al. (2021), NS masses for the injections are taken between 1 M ☉ and 3 M ☉ from a normal distribution centered at 1.4 M ☉ with a standard deviation of 0.2 M ☉ (Kiziltan et al. 2013) and 0.4 M ☉ for BNS and NSBH systems, respectively.BH masses are taken to be between 3 M ☉ and 25 M ☉ from a normal distribution centered at 10 M ☉ with a standard deviation of 6 M ☉ .
Although this PyGRB follow-up of FRB events mirrors the search conducted for GWs associated with GRB events in O3a (Abbott et al. 2021) where appropriate, there were several differences in the choices of analysis parameters for the FRB analysis.The first major difference has been noted above, wherein a 12 s on-source window is used, which is double that of the GRB analysis.This does reduce the significance of any detected signals, but has the benefit of allowing for more progenitor models where the EM emission occurs further in time from the peak of the GW emission.
Another significant change was the method of determining the area of sky over which to search for the GW signals.The FRB data sample contains multiple localizations for each event, each with their own R.A. and decl.uncertainties.This effectively creates multiple patches on the sky where the source could potentially reside.The effective GW network localization capability results in 90% credible regions for detections on the order of ≈10-10,000 deg 2 , with an average of order 100 deg 2 .In contrast, the multiple O3a FRB sample localizations spanned only the order 1 deg 2 in total (Abbott et al. 2020a).The sensitivity of the search also did not vary significantly over the sky localizations, and so the final set of sky positions considered by the analysis was one circular patch on the sky with a size large enough to ensure coverage over all possible provided FRB localizations.This circular region is centered on the median of the provided R.A. and decl.values, with a radius scaled to match either the largest position error provided or the largest R.A. or decl.separation between the 5 localization points, using whichever is greater.Within this patch, the sky is sampled by creating a circular grid of sky positions such that the time-delay between grid points is kept below 0.5 s (Williamson et al. 2014).This ensures coverage of the possible sky location of the source.For each sky position, the timestream data from each GW detector are combined with the appropriately different time offsets required to form a coherent stream of data for that point on the grid.These multiple coherent time streams are finally each considered in the search.

X-Pipeline-Unmodeled Search for Generic Transients
The search for generic transients is performed with the coherent analysis algorithm X-Pipeline (Sutton et al. 2010;Was et al. 2012).This targeted search uses the sky localization and time window for each CHIME/FRB trigger to identify consistent excess power that is coherent across the network of GW detectors.We use different search parameters in our searches for repeating and nonrepeating FRB sources.
There are a number of differences between our generic transient search on nonrepeated sources and those previously conducted on GRBs (Abbott et al. 2017a(Abbott et al. , 2019a(Abbott et al. , 2021)).As in GRB searches, the on-source time window is chosen to start 600 s before the trigger, but is extended from 60 s seconds post trigger to 120 s to allow for the possibility of GW emissions delayed relative to the FRB emission.This on-source window is also longer than the ±120 s window employed in the previous FRB search (Abbott et al. 2016).The extended window allows for a greater number of noncompact binary coalescence (CBC) sources than those considered in GRB searches and possible GW emissions from magnetars, given the recent FRB-magnetar association (CHIME/FRB Collaboration et al. 2020).
The broadband search for FRBs with X-Pipeline covers the range 32 Hz up to 2 kHz, the upper range being higher than the GRB search (20-500 Hz) in order to include GW emissions from oscillation modes of NSs that are likely to occur above 1 kHz, specifically f-modes (Wen et al. 2019;Ho et al. 2020).We note that above 300 Hz a ∝f 2 frequency dependence in energy (see later Equation (5)) combined with the ∝f 1 of the noise power spectral density of the detector increases the GW energy required to enable a confident detection as ∝f 3 .Although including high-frequency data increases the computational cost, including this data allows us to set limits on a wider variety of signal models.
X-Pipeline processes the on-source data around each FRB trigger by combining the GW data coherently after the data is whitened by dividing by each detector's amplitude spectrum (Abbott et al. 2020b).The coherent combination is formed by taking into account the antenna response and noise level of each detector to generate a series of time-frequency maps.The maps show the temporal evolution of the spectral properties of the signal and allow searches for clusters of pixels with excess energy significantly greater than one would expect from background noise.These clusters are referred to as events.
Events are given a ranking statistic based on energy and are subjected to coherent consistency tests based on the signal correlations between data in different detectors.This allows X-Pipeline to veto events that have properties similar to the noise background.
The surviving event with the largest ranking statistic is taken to be the best candidate for a GW detection.Its significance is quantified as the probability for the background alone to produce such an event.This is done by comparing the S/N of the trigger within the 720 s on-source to the distribution of the S/Ns of the loudest triggers in the off-source trials.The offsource data are set to consist of at least 1.5 hr of coincident data from at least two detectors around the trigger time.This window is small enough to select data where the detectors should be in a similar state of operation as during the on-source interval, and large enough so that, through artificial timeshifting, probabilities can be estimated at the subpercent level.
We quantify the sensitivity of the generic transient search by injecting simulated signals into off-source data and recovering them.We account for calibration errors by jittering the amplitude and arrival time of the injections according to a Gaussian distribution representative of the typical calibration uncertainties expected in O3a.We compute the percentage of injections that have a significance higher than the best event candidate and determine the amplitude at which this percentage is above 90%; this value sets the upper limit.
As discussed in Section 3, the localization information for each FRB is in the form of up to 5 noncontiguous or overlapping error regions of varied morphology.Occasionally, these islands can be dominated by the uncertainty of a single island.The sky position errors can span a few degrees or more in R.A.This could result in a temporal shift causing a GW signal to be rejected by a coherent consistency test (Was et al. 2012).For each island, we set up a circular grid around the central location of the island, with overlapping grid points discarded.A coherent data stream is formed from the GW detector data with an appropriate time offset for each point on the grid.These data streams are then analyzed.The grid positions are large enough to cover the error radius and dense enough to ensure that a maximum timing delay error, set as 1.25 × 10 −4 s, is within 25% of the signal period at our frequency upper limit of 2000 Hz.This is 4 times finer than GRB searches that typically analyze data up to a frequency cutoff of 500 Hz.Using this grid approach, the antenna responses change only slightly over sky position; of the order of a few percent over a few degrees (Aasi et al. 2014).The responses are known to change rapidly near a null of the response; in such a case, they are already negligible.
A particular difference between this search and other searches focused on GRBs is the increased number of simulated waveform types used in this study.Given the uncertainty in plausible GW emissions, we consider a larger range of generic burst scenarios, using an extended set of those used in both GRB and magnetar searches (Abbott et al. 2019b(Abbott et al. , 2021)).Also, as we have no knowledge on whether or not FRBs are beamed along the rotation axis of the progenitor, all of our signal models correspond to elliptical and random polarization.
The waveforms chosen to cover the search parameter space are from 3 families that have different morphological characteristics: binary signals, generic burst-like signals, and accretion disk instability (ADI) models.X-Pipeline is equally adept at detecting signals whose frequency decreases with time (ADI) and signals whose frequency increases with time (CBC models; Abadie et al. 2012;Abbott et al. 2017a).This paper reports the results for CBCs when obtained using the dedicated modeled search (described in Section 4.1), so we will limit our discussions here to only the latter two waveform families.
The generic burst-type waveforms are described in Table 1, where we list the most important parameters (see also Abbott et al. 2019c).In all cases, to determine exclusion distances for this model family, we assume an optimistic emission of energy in GWs of E GW = 10 −2 M ☉ c 2 (Abbott et al. 2021).The waveforms in this family aim to capture the general characteristics of a burst of GW energy: Sine-Gaussian.These signals have been used previously to represent the GWs from stellar collapses.The models are defined in Equation (1) of Abbott et al. (2017a) with a Q factor of 9 and varying central frequency as shown in Table 1.They can also model f-modes in the core of a canonical NS.We therefore also include them in the search over repeating sources, and include sine-Gaussian (SG) waveforms at additional frequencies listed in Table 1.In order to better constrain some models, we also include circularly polarized SG chirplets at the frequencies nearest the f-mode range (1600 and 1995 Hz) in the search over repeated sources.Ringdowns (DS2P).These signals capture the form of damped sinusoids (DS2P) at a frequency of 1500 Hz and decay constants of 100 and 200 ms.White noise bursts (WNB).These signals mimic broad bursts of uncorrelated white noise, time-shaped by a Gaussian envelope.We use two models band-limited within frequencies of 100-200 and 100-1000 Hz, and with time constants of 11 and 100 ms.
Following the predictions from oscillation modes for NS starquakes (Li et al. 2019;Wen et al. 2019), the first two waveforms in this family (SG and DS2P) have been used in the search for GWs associated with magnetar bursts (Abbott et al. 2019b(Abbott et al. , 2022)).
We also consider a range of models.These are long-lasting waveforms, which are modeled to represent the GW emissions from instabilities in a magnetically suspended torus around a rapidly spinning BH.The model specifics and parameters used to generate the five types of ADI signals, designated ADI-A to Notes.Models and their parameters have been chosen to cover as large a parameter space as possible.For all models, the central frequencies are shown.We note that WNB models are defined by an additional frequency bandwidth; this parameter is shown in parenthesis.For the SG and WNB waveforms, the duration parameter scales the width of the Gaussian envelope; for the DS2P models, this parameter defines the decay time constant.An asterisk ( * ) denotes waveforms used in the repeaters search only; c denotes waveforms with a circular polarization.
ADI-E, are the same used in the previous searches (see Table 1 of Abbott et al. 2017a).
The version of X-Pipeline used in this analysis has a new feature named autogating.This feature increases the sensitivity of the longer-duration (10 s) signals, previously limited by loud background noise transients (Abbott et al. 2021).This technique gates the whitened data from a single detector if the average energy over a 1 s window exceeds a user-specified threshold.To minimize the possibility of a loud GW transient being gated, this procedure is canceled if the average energy at the same time in any other detector exceeds the threshold.

X-pipeline Search on Repeating FRBs
A subset of 11 of the FRBs that we analyze has been identified to repeat.Repeating FRBs are possibly caused by a process distinct from those that produce singular FRBs; most notably they are unlikely to be associated with CBC events.We therefore only run the X-Pipeline generic transient search on these events, and we choose the parameters to provide maximal sensitivity to the GW transients that would most probably be produced by flaring magnetars.
This search is similar to that for GW events associated with magnetars during the third observing run of Advanced LIGO and Advanced Virgo (O3; R. Abbott et al. 2023, in preparation).The frequency band of the search ranges from 50 to 4000 Hz, which encapsulates the NS f-mode frequency band, but excludes the lowest frequencies where nonstationary noise could potentially pollute the search statistics.The search spans 8 s of time centered within 1 s of the arrival time of the FRB to ensure optimal sensitivity at the event time.Injected waveforms are chosen to reasonably model the f-modes of a canonical NS as described in Kokkotas et al. (2001).This includes a series of SG chirplets with a qualtiy (Q) factor of 9 and varying center frequencies as shown in Table 1.We also neglect to use the autogating algorithm for noise transients as described above, as its tendency is also to gate fast injections such as SG.We also inject WNBs to estimate the sensitivity at broadband frequency ranges.

RAVEN Coincident Analysis
To perform a wider sweep of the O3a data, we also looked for coincidences between these CHIME/FRB events and existing GW candidates using the tools of the Rapid, onsource VOEvent Coincidence Monitor (RAVEN; Urban 2016; Cho 2019) to query the Gravitational-Wave Candidate Event Database (GraceDB; Pace et al. 2012).This query to GraceDB tests whether any GW candidates were found by any of the modeled or generic transient low-latency GW search pipelines within a time window around the FRB events.The queries used the same on-source search windows as our modeled and generic transient searches, with [−10, +2 s] and [−600, +120 s] windows around the FRB triggers, respectively.We then computed the joint false-alarm rate of any coincident GW candidate within these windows using the overall rate of FRB events in the CHIME/FRB sample calculated across the full span of the O3a observing run and the false-alarm rate of the GW candidate.The joint false-alarm rates were compared against thresholds of around 6 and 1 yr −1 for modeled and generic transient searches respectively.This analysis, although not as sensitive as a targeted search, is a strategy that allows us to perform a broad search across O3a data for possible coincidences missed by our analysis.

Analysis Subsample
We performed two different searches: for nonrepeating FRBs, a PyGRB modeled search was completed on a total of 22 FRB events; and an X-Pipeline search for generic transient signals was completed on a total of 29 nonrepeaters and 11 repeating FRBs.

The False-alarm Probability (p-value) Distribution
The searches conducted for GW counterparts returned no likely GW signals in association with any of the analyzed repeating or nonrepeating FRB events.
The most significant events found by the PyGRB search and the X-Pipeline search had p-values of 3.74 × 10 −2 and 1.90 × 10 −2 , respectively.For the X-Pipeline analysis of the repeating FRBs, the lowest p-value was 1.3 × 10 −1 , corresponding to the repeat FRB 20190702B of burst FRB 20190303A, for which we analyzed 2 burst events.
The cumulative p-value distributions from both search methods are shown in Figures 3 and 4. In both figures, the dashed lines indicate the expected background distribution under the no-signal hypothesis, and the dotted lines indicate the 90% confidence band around the no-signal hypothesis.

Exclusion Distance Results
Figure 5 shows the cumulative 90% exclusion distances for the 22 FRBs followed up with the modeled search.The lowest exclusion distances, of order 40 Mpc, were obtained for FRBs that occurred during times in which only Virgo data were available.For each of the three simulated signal classes considered in the modeled search, we quote the median of the D 90 results in the top row of Table 2; we see values of the order of 190 Mpc for BNS and around 260 Mpc (350 Mpc) for NSBH with generic (aligned) spins.
Figure 6 provides the cumulative 90% exclusion distances for 29 nonrepeating FRBs considered in the generic transient search.This plot shows three representative burst models: ADI-A, SG-C, and a WNB-C; the latter two have central frequencies of 145 and 550 Hz respectively.Based on a standard E GW ∼ 10 −2 M ☉ c 2 of emitted GW energy, there is a noticeable offset between the SG and the other two GW burst models.For the ADI-A waveform model, this is due to the energy of the former being distributed over a longer signal duration, of the order ∼40 s; for the WNB-C model, this effect is due to a significant portion of its energy content being at higher frequency where detector performance is more comparatively limited.
The lower rows of Table 2 show the median of the D 90 estimates for all other waveforms considered by the generic transient search.We see that SG models spanning central frequencies 70-2000 Hz have corresponding median values of D 90 in the range 78-0.5 Mpc; the latter models' performance diminished at higher frequency through a detector response.This is also clearly evident for the DS2P ringdown models, which are more likely to encounter a transient burst of noise than SG models due to their longer durations.Similarly, the median D 90 values for the higher-frequency WNB models are in comparison with the lower-frequency models (WNB-A and WNB-B).These median D 90 values of the 150 and 550 Hz models differ by around a factor of at least 4. Overall, the median D 90 varies within a range approaching 2 orders of magnitude, reflecting the wide range of models used in the analysis.
In comparison with D 90 values obtained in the O3a GRB paper (Abbott et al. 2021), the values in Table 2 are almost Figure 5. Cumulative histograms of the 90% confidence exclusion distances, D 90 , for the 22 CHIME/FRB bursts followed up by the modeled search.The blue line shows generically spinning BNS models, the orange line shows generically spinning NSBH models, and the thick green line shows alignedspin NSBH models.We define D 90 as the distance within which 90% of the simulated GW signals injected into the off-source data were recovered with a significance greater than the most significant on-source trigger.systematically a factor of 2 smaller for the SG and ADI models used in that study.We find that this is a result of the sky locations surveyed by CHIME corresponding with a region of weak sensitivity for the Virgo interferometric detector, due to their relative locations on the surface of the Earth.The average antenna responses for the LIGO Hanford (H1) and LIGO Livingston (L1) detectors are of order 0.72 and 0.65 respectively; the same metric for the V1 instrument is 0.28.This has a severe effect when V1 is one of only two detectors in a network, a situation that has occurred 55% of the time for the generic transient analysis of nonrepeating FRBs.Looking ahead, this type of sensitivity bias will be a feature of future searches for CHIME/FRB triggers, as well as surveys by other facilities, depending on their location on the Earth.
In Table 3, we present the exclusion distances achieved for each of the FRBs analyzed in our joint analysis.For the modeled search, we quote values from each of the 3 classes of compact binary progenitor models considered.For the generic transient search, we present values of D 90 for a representative sample of SG, ADI, DS2P, and WNB models.We also provide information relating to the times and positions of these events as well as values of the DM, and the inferred 90% credible intervals on the luminosity distance.Table 3 allows comparison of the inferred luminosity distances of each FRB with the D 90 value for different searches.
Figure 7 compares the D 90 values for the BNS and NSBH (with generic spin) emission models with the 90% credible intervals on D L inferred by the MCMC analysis.The plot shows the FRB sample in order of increasing distance.No event can be fully excluded from any of the models we have considered for this search, because there is still a sufficient region of space from which the FRB events could have originated that is outside the detection range of the searches performed.

RAVEN Analysis Results
As described in Section 4.3, two RAVEN coincidence searches were completed with differing time windows, [−600 s, +120 s] for the generic transient search and [−10 s, +2 s] for the modeled search.The generic transient search found 8 coincidences, and the modeled search found 1 coincidence.However, none of these were of sufficient significance, as determined by the computed joint false-alarm rate from the two samples, to be distinguished from random coincidences.All of the FRBs in these coincidences had distances that were well beyond the values of D 90 obtained, with the exception being FRB 20190518E, a repeat of burst FRB 20190518A, with 9 episodes occurring during O3a.Of these 9 repeating episodes, 7 were also analyzed using our generic transient search method, as described earlier.Again, none of the repeating episodes returned a significant false-alarm probability, with the minimum p-value across the search of repeating FRB events equal to 1.3 × 10 −1 .

Upper Limits on GW Energy
A measure of the inferred distance to an FRB source also allows one to place constraints on the energy carried in a burst of GWs.The GW energy, E GW , emitted by an elliptically polarized GW burst signal can be related to the root-sumsquare signal amplitude h rss and the central frequency of the source, f 0 , through the following (Sutton 2013): where D L is the luminosity distance to the source.As the DMs of FRBs provide a measure of the maximum distance, one can use Equation (5) to place 90% upper limits on the GW energy emitted by each FRB source, E GW 90% .This estimate, calculated using h rss 90% , the 90% detection upper limit on the root-sumsquared GW amplitude, is highly dependent on the detector sensitivity and antenna factors at the time of the FRB as well as the central frequency of the simulated waveform injections.
Tables A1 and A2 provide the upper limits on E GW 90% for SG models and DS2P or WNB GW burst models respectively.These limits assume that the FRB distances are at the lower limits of their inferred distance ranges.Given a large range of models, and since this quantity scales as h f rss 2 0 2 , one would expect the lower-frequency models to provide the most constraining limits.For SG models, the most constraining estimate was 2.5 × 10 50 erg for the 70 Hz SG-A model, and for the highest-frequency model considered, SG-H at 1995 Hz, the upper limit was 7.9 × 10 54 erg.These values were obtained for the closest inferred burst in the sample, FRB 20190425A.The same burst yielded upper limit values in the range 4.8 − 470 × 10 50 erg for the WNB model.The DS2P model gave the best constraints, 5.8 − 6.4 × 10 54 erg, for FRB 20190531B.
For completeness, in Tables A3 and A4, we also provide less constraining limits on E GW 90% based on the upper credible intervals on the distance of each FRB.
Table 4 lists the repeating bursts that were analyzed in the generic transient search.The most sensitive counterpart to a repeating FRB was for CHIME/FRB event FRB20190825A.The SG injection centered at 1600 Hz (which most closely models an f-mode) was recovered 90% of the time at   h rss = 2.62 × 10 −22 .The distance to this event is 148.1-149.9Mpc.This corresponds to an energy upper limit range of 5.83 × 10 55 to 5.98 × 10 55 erg.These estimates are well above predictions of the GW emissions through the NSʼs fundamental f-mode discussed in Section 2.

The M81 Repeater FRB 20200120E
A repeater, FRB 20200120E, which was discovered by CHIME/FRB on 2020 January 20, overlaps with the second part of the third observing run of Advanced LIGO and Advanced Virgo (O3b), which took place between 2019 October 1 15:00 UTC and 2020 March 27 15:00 UTC.This burst is at 3.6 Mpc, the closest extragalactic FRB so far discovered (Bhardwaj et al. 2021a).This event was shown to be conclusively associated with a globular cluster in the M81 galactic system (Kirsten et al. 2022), which supports the possibility that it was formed from an evolved stellar population such as a compact binary system.Due to the proximity and significance of this burst, we discuss it in this paper, despite it being discovered after O3a.The burst FRB 20200120E was shown to repeat at least 4 times.Two of the repeats occurred after O3b; another episode, despite being consistent with the localization of the other associated bursts, had no intensity data saved.Therefore, we discuss here only the initial burst FRB 20200120E, for which GW data exists.
At the time of FRB 20200120E, only H1 data were available; thus, a generic transient search was not conducted.Likewise, since this is a repeating event, it does not pass our criteria for conducting a modeled search.Due to these restrictions, only a RAVEN coincidence search was conducted within a [−6000, +6000] s time window.No coincidences were found with sufficient significance as determined by the coincident falsealarm rate.Given the relative close proximity of this burst, further repeat emissions will be of interest for GW follow-up during the fourth observing run of Advanced LIGO, Advanced Virgo, and Kagra (O4; Abbott et al. 2020a) when constraints on the energy emitted in GWs will be of order 10 50 at around 500 Hz.

Conclusions
We performed a targeted search for GWs associated with FRBs detected by the CHIME/FRB project during O3a.As the sources of nonrepeating FRBs are currently not known, we ran both a modeled search for BNS and NSBH signals (Harry & Fairhurst 2011;Williamson et al. 2014) and a generic transient search for generic GW transient signals (Sutton et al. 2010;Was et al. 2012).
Our searches found no significant GW event candidates in association with the analyzed FRBs.We set 90% confidence lower bounds on the distances to FRB progenitors for several different emission models.Additionally, we present 90% credible intervals on the luminosity distance, D L , inferred from the DM measurement of each FRB source.
The D L information can be used to test models based on the simulated injections used for calculating the D 90 values of each FRB.However, the significant uncertainties in the relative contributions to the total DM for each FRB produce relatively wide credible intervals for the D L posteriors.We find no FRB event can be fully excluded from any of the models we have considered due to some posterior support on D L existing for the FRB outside the detection range of the analyses performed.
The results, however, as illustrated in Figure 7, indicate that the GW network's detection range is advancing into cosmological volumes where FRB emissions are expected.This is encouraging as we look forward to future GW searches at higher sensitivity.Furthermore, the redshifts obtained from the ongoing efforts to localize host galaxies (there are currently 18 FRBs with an associated host galaxy; see)319 could significantly improve the chances of constraining progenitor populations (Heintz et al. 2020;Bhandari et al. 2022).
The distance estimates for each FRB allowed us to place 90% upper limits on the GW energy emitted by each FRB source, E GW 90% .For each nonrepeating FRB analyzed with a generic transient search, we provided limits on E GW 90% for a range of emission models.Repeating FRBs were also analyzed to determine 90% upper limits on the energy emitted through GWs.For the most sensitive repeating FRB analysis in our sample, we find an energy upper limit range of 5.83 × 10 54 to 5.98 × 10 55 erg, well above the predictions for GW emissions from the fundamental f-modes of NSs.Based on Equation (5), an FRB event such as that associated with SGR 1935+2154 occurring during O3a would have allowed the search to probe the more optimistic of these estimates allowing limits, E GW ∼ 10 47 erg, assuming a generic burst waveform emitting at roughly 1 kHz at 10 kpc.
We also analyzed the repeater, FRB 20200120E, discovered on 2020 January 20 during O3b.A RAVEN (Urban 2016; Cho 2019) coincidence search for any previously detected compact binary coalescence GW events was conducted within a [−6000, +6000] s time window around the first burst of this repeater.No coincidences were found with sufficient significance to be distinguished from random coincidences, as determined by the computed joint false-alarm rate from the two samples.
A comparison of the expected volumetric rates is one avenue to yield insights on possible associations between two transient source populations.An analysis of the most recent Gravitational Wave Transient Catalog 3 (GWTC-3 The LIGO Notes.The TNS name is provided in the first column.The network column lists the GW detector network used: LIGO Hanford (H1), LIGO Livingston (L1), and Virgo (V1).The total DM for each FRB is listed in the DM column, and the 90% credible intervals on the luminosity distance are provided in columns D L -low and D L -high.11 total events were analyzed for the three different FRB repeaters considered.For FRB 20190518A and its associated repeats, we list only the distance of Marcote et al. (2020) obtained by galaxy localization.
Scientific Collaboration et al. 2021Collaboration et al. , 2023) has inferred merger rates in the ranges 10-1700 Gpc −3 yr −1 for BNS and 7.8-140 Gpc −3 yr −1 for NSBH populations.These estimates are significantly lower than estimates of the FRB rate 3.5 10 2.4 5.7 4 -+ Gpc −3 yr −1 provided by Luo et al. (2020) for sources above 10 42 erg s −1 .Based on these numbers, the percentage of FRBs that could possibly be associated with BNSs is at most 15%, and for NSBH sources, the percentage is 1%.As noted by Luo et al. (2020), if BNS and/or NSBH sources were only associated with FRBs from the high end of the luminosity function (>10 43 erg s −1 ), such rates could be comparable.
However, there are a number of unknown factors that complicate reconciling the GW and FRB source populations; these include the proportion of FRBs that may repeat or the possible effects of beaming (Ravi 2019;Connor et al. 2020).
Probing the local population of FRBs through targeted searches, the strategy adopted in this study can constrain associations between GW sources and FRBs.The distance uncertainties in the FRB sample are a particular obstacle, and ongoing efforts to identify FRB host galaxies (Chatterjee et al. 2017) could provide valuable prior information for FRBs discovered within the BNS and/or NSBH detection range of future searches.CHIME/FRB is deploying a set of Outrigger telescopes located at sufficient distances to allow autonomous very-longbaseline interferometry on CHIME/FRB detected bursts (Cassanelli et al. 2022;Mena-Parra et al. 2022).This development promises subarcsecond localizations on hundreds of FRBs per year allowing host galaxy identification and redshift determination through optical follow-ups or through cross matching of positional data with photometric galaxy surveys (Shin et al. 2023).The resulting sample of FRBs at low redshift will be a significant development for GW detection networks, particularly as the sensitive volume increases with future observation runs and should allow targeted searches to obtain statistical evidence toward supporting or ruling out GW-FRB associations.We would like to thank all of the essential workers who put their health at risk during the COVID-19 pandemic, without whom we would not have been able to complete this work.

Appendix Tables of Upper Limits on the Energy Emitted through
GWs for the Generic Transient Search This section provides the supplemental tables containing the upper limits on the energy emitted through GWs for the generic transient search for different waveform models and luminosity distance estimates.Tables A1 and A2 provide the upper limits on E GW 90% for SG models and DS2P or WNB GW burst models respectively.These limits assume that the FRB distances are at the lower limits of their inferred distance ranges.Tables A3 and  A4 provide less constraining limits on E GW 90% based on the upper credible intervals on the distance of each FRB.Note.The distances represent the lower bounds of 90% credible intervals from the MCMC inference described in Section 3. Note.The distances represent the lower bounds of 90% credible intervals from the MCMC inference described in Section 3.

Figure 1 .
Figure 1.An example of a CHIME localization confidence interval plot for the closest nonrepeating burst in our sample, FRB 190425A.The plot shows 4 localization islands and is centered at the beam with the highest S/N.

Figure 3 .
Figure 3.The cumulative distribution of p-values for the loudest on-source events for the modeled search in O3a around CHIME/FRB data.The dashed line indicates an expected uniform distribution of p-values under a no-signal hypothesis, with the corresponding 90% confidence band shown by the dotted lines.

Figure 4 .
Figure 4.The cumulative distribution of p-values for the loudest events from the generic transient search for transient GWs associated with 29 nonrepeating CHIME/FRB bursts.The dashed line represents the expected distribution under the no-signal hypothesis, with the 90% bands shown as dotted lines.

Figure 6 .
Figure 6.Cumulative histograms of the 90% confidence exclusion distances, D 90 , for SG model (C) (orange line), accretion disk instability signal model (A) (blue line), and white noise burst model (C) (green, thick line).The quantity has the same definition as described in Figure 5.

Notes.
The TNS name is provided in the first column.The network column lists the GW detector network used: LIGO Hanford (H1), LIGO Livingston (L1), and Virgo (V1).The total DM for each FRB is listed in the DM column, and the 90% credible intervals on the luminosity distance of each burst are provided in columns D L -low and D L -high.Where the generic transient search (Section 4.2) and the modeled search (Section 4.1) used a different Interferometer network, the network used by the generic transient search is shown in parentheses.The last 8 columns show the 90% confidence exclusion distances for each FRB (D 90 ) for the following emission scenarios: BNS, generic and aligned-spin NSBH from the modeled search, and from the generic transient search, SG-C, SG-F, ADI-A, DS2P-A, and WNB-C; for the latter 5 types of GW bursts, we assume a total radiated energy E GW = 10 −2 M e c 2 .

Figure 7 .
Figure 7. Lower limits on the 90% confidence level exclusion distances for BNS (lower bar), generic spin NSBH (middle bar), and aligned-spin NSBH (upper bar) progenitor systems are shown as found by the modeled search.These are compared to the 90% credible intervals (whisker plot) on the D L posterior determined by the MCMC method for the FRBs considered in this study.
Institute for Astronomy and Astrophysics at the University of Toronto.Additional support was provided by the Canadian Institute for Advanced Research (CIFAR), McGill University and the McGill Space Institute via the Trottier Family Foundation, and the University of British Columbia.The Dunlap Institute is funded through an endowment established by the David Dunlap family and the University of Toronto.Research at Perimeter Institute is supported by the Government of Canada through Industry Canada and by the Province of Ontario through the Ministry of Research & Innovation.The National Radio Astronomy Observatory is a facility of the National Science Foundation (NSF) operated under cooperative agreement by Associated Universities, Inc. FRB research at UBCis supported by an NSERC Discovery Grant and by the Canadian Institute for Advanced Research.The CHIME/FRB baseband system is funded in part by a CFI John R. Evans Leaders Fund award to IHS.

Table 1
The Main Parameters of the Waveform Injections Used for the Generic

Table 2
Median Values for the 90% Confidence Level Exclusion Distances, D 90 Note.Modeled search results are shown for three classes of BNS progenitor model, and generic transient search results are shown for models described in Table1.

Table 3
Details of the FRB Sample and the 90% Exclusion Distances for Each of the Events Considered in This Analysis

Table 4
Details of the 3 Repeating FRBs Analyzed in the Generic Transient Search and Their Various Repeating Episodes

Table A1 The
Upper Limits on the Energy Emitted through GWs in erg for the Generic Transient Search Using the SG Waveforms Described in Table1

Table A2
The Upper Limits on the Energy Emitted through GWs in erg for the Generic Transient Search Using the DS2P and WNB Waveforms Described in Table1

Table A3
As for Table A1 but with Distances Based on the Upper Bounds of 90% Credible Intervals on the Luminosity Distance Table A4 As for Table A2 but with Distances Based on the Upper Bounds of 90% Credible Intervals on the Luminosity Distance