DISCOVERY OF A TRANSIENT GAMMA-RAY COUNTERPART TO FRB 131104

We carried out a search for untriggered (subthreshold) transient gamma-ray counterparts to all FRBs from the frbcat catalog⁵ (Petroff et al. 2016), including all reported bursts from the repeating FRB 121102. We examined gamma-ray data from the Swift BAT (Barthelmy et al. 2005), in near-continuous operation since 2004 November, and from the International Gamma-Ray Astrophysics Laboratory's IBIS imager (Ubertini et al. 2003), in near-continuous operation since 2002 October.

During this time, 2 of 13 non-repeating FRBs and 2 of 17 bursts from FRB 121102 occurred within the BAT field of view; no FRBs occurred within the IBIS field of view. For each FRB with simultaneous BAT coverage, we retrieved the relevant data from the High Energy Astrophysics Science Archive Research Center (HEASARC⁶ ) and searched for sources within 15' of the FRB coordinates. This radius accounts for uncertainty in the positions of both the radio source (typically localized to a single beam with FWHM ≈ 15') and the subthreshold BAT source candidates (having 90%-containment radii r₉₀ ≈ 7') that are observed in these data.

We used the heasoft (v. 6.18) software tools and calibration for our high-energy data analyses.⁷ Swift BAT survey data include detector plane histograms (DPHs) of the full-bandpass (15–195 keV) 300 s exposures and scaled detector plane images (DPIs) of the soft-band (15–50 keV) 64 s exposures. We reduced these data using standard procedures, adopting the maximum allowed oversampling parameter of 10, and searched for candidate sources over 15–150 keV using the batcelldetect sliding-cell algorithm. This routine uses local estimates of the background and noise level to identify candidate sources, and then performs a point-spread function (PSF) fit to derive an accurate source position and BAT counts estimate. We estimated uncertainties in source positions (r₉₀) from source significances using the calibration of Baumgartner et al. (2013) (their Equation (7)).

As we are interested in testing the hypothesis of a fixed S_γ:S_GHz fluence ratio for FRBs—and as we are interested in non-repeating sources (as candidate catastrophic events) more than in the known repeating source FRB 121102—we prioritized the search as follows: non-repeating FRBs ordered by decreasing radio fluence, followed by bursts of FRB 121102 ordered by decreasing radio fluence. The results of our search are presented in Table 1.

We identified an untriggered gamma-ray transient candidate with a signal-to-noise of ${ \mathcal S }=4.2\sigma$ in the first search area, that associated with FRB 131104 (Ravi et al. 2015). The transient position is R.A. 06^h44^m3312, decl. −51°11'312 (J2000), with r₉₀ = 68 (Figure 1). It is located near the edge of the BAT field of view, with only 2.9% of BAT detectors illuminated through the coded mask (2.9% coding), which explains its low significance in spite of a relatively bright inferred fluence. Its sky position is well within the search area, 63 from the radio receiver pointing, with 50% of its BAT localization probability within the receiver FWHM (Figure 2). No candidate counterparts are identified for the remaining FRBs with BAT coverage, with results as reported in Table 1.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Since a gamma-ray transient is identified for the highest radio fluence non-repeating FRB in our sample, and since the S_γ:S_GHz constraints for the other FRBs are consistent with the ratio inferred for FRB 131104, this is consistent with our hypothesis and first test, and we adopt a trials factor of one for assessing the significance of the counterpart.

We determine this significance by examining 1429 archived BAT survey pointings with exposure times 200–400 s that were taken over the one-year period 2015 June to 2016 May. On average each of these survey images has 46.3 transient candidates with ${ \mathcal S }\geqslant 4.2\sigma$ at >1% coding; although some may be cosmic sources, for our present purposes we treat them all as noise fluctuations. The density of candidates per unit solid angle varies across the field of view, so we focus on a rectangular region of the BAT image plane, centered on the transient position in tangent plane coordinates. Within this region, which has area 16 deg² (0.36% of the field of view), we find an average of 0.106 ± 0.009 transient candidates per survey pointing. This average density of candidates does not vary systematically with pointing coordinates, exposure time, or date, over the set of BAT pointings. The resulting p-value, corresponding to the number of expected candidates in a single 15' search radius, is p = 0.13%, corresponding to Gaussian confidence level ${ \mathcal C }=3.2\sigma$.

We also estimate the significance of the association by an alternate Bayesian approach that does not require us to define a search area in advance. Instead, we make a point comparison of the probability density for an FRB-associated transient (at the maximum-likelihood position of the gamma-ray transient) to the probability density for a noise fluctuation. The former will be distributed according to the FRB positional uncertainty (derived below), while the latter will be uniform over the larger rectangular test region. Because the transient is located 1.2σ from the FRB coordinates (distance $d=6\buildrel{\,\prime}\over{.} 25$, $\sigma =5\buildrel{\,\prime}\over{.} 2$), the FRB counterpart probability density is 2.85 × 10⁻³ arcmin⁻²; while the background source density is 1.84 × 10⁻⁶ arcmin⁻². Hence an association is preferred by an odds ratio of 1552:1, corresponding to a Gaussian confidence level of ${ \mathcal C }=3.4\sigma$.

As both metrics exceed the 3σ threshold common for counterpart identification in astrophysics, we consider the transient confirmed and designate it Swift J0644.5−5111, the first non-radio counterpart to any FRB. Its properties are summarized in Tables 1 and 2.

Table 2.  Properties of FRB 131104

Joint	R.A.	06h44m2706
	Decl.	−51°12'540
	r90	578
Radio	UTC	18:03:59
	SGHz	2.33 Jy ms
	DM	779 ± 1 pc cm−3
	zmax	0.55
γ-ray	T90	377 ± 24 s (1σ)
	>100 s (90%-c.l.)
PL	Γ	${1.16}_{-0.78}^{+0.68}$
	Sγ,−6	4.0 ± 1.8
TB	kT	${200}_{-125}^{+\infty }$ keV
	Sγ,−6	3.4 ± 1.5

Note. Radio properties including topocentric burst time (UTC), radio fluence (S_GHz), and dispersion measure (DM) are from Ravi et al. (2015), while the maximum redshift for a consensus cosmology (z_max) is from Murase et al. (2016). γ-ray and joint properties are from this work. Coordinates for the joint radio + gamma-ray localization are J2000. Spectral parameters for power-law (PL) and thermal bremsstrahlung (TB) fits are quoted with 90%-confidence intervals; gamma-ray fluences S_γ,−6 (15–150 keV) are in units of 10⁻⁶ erg cm⁻².

Download table as:  ASCIITypeset image

We first seek to refine the location for FRB 131104 by combining radio and gamma-ray constraints. This requires a quantitative form for the radio localization; since this has not previously been derived in a rigorous fashion, we describe our approach here.

We assume an azimuthally symmetric Gaussian form for the response of the Parkes multibeam receiver #5, which has a quoted FWHM of 15' (Ravi et al. 2015). In parallel, we assume an $N({S}_{\mathrm{GHz}}\gt {S}_{0})\propto {S}_{0}^{-3/2}$ form for the radio fluence distribution of FRBs (Law et al. 2015). Together, these assumptions imply a two-dimensional Gaussian probability distribution for the location of FRB 131104, with $\sigma =5\buildrel{\,\prime}\over{.} 2$ and 90%-containment radius r₉₀ = 112, centered on the receiver pointing coordinates R.A. 06^h44^m1040, decl. −51°16'40'' (J2000). This puts 92% of the BAT localization within the radio-derived r₉₀. Weighting this localization with the r₉₀ = 675 localization of the gamma-ray transient yields a joint localization centered at R.A. 06^h44^m2706, decl. −51°12'540 (J2000), with r₉₀ = 578. This localization is illustrated in Figure 2 and reported in Table 2.

Examination of archival images of the burst and transient localization region (Figure 2) does not reveal any prominent Local Group or low-redshift galaxies, nor bright active galaxies, although as noted by Ravi et al. (2015), the field is near the projected tidal limit of the Carina dwarf spheroidal galaxy (D ≈ 100 kpc) and a projected tidal stream of the Large Magellanic Cloud (D ≈ 50 kpc). The absence of known or candidate flare stars has been noted by Maoz et al. (2015).

A NASA/IPAC Extragalactic Database (NED⁸ ) query targeting the predefined search area for FRB 131104 yields six cataloged galaxies⁹ , the quasar QJ0643−5126 (z = 2.77), and the IRAS source IRAS F06441−5118. The observed density of resolved galaxies and the presence of a known quasar and an IRAS source are not remarkable for a field of this size at this Galactic latitude. All of these cataloged sources lie well outside the joint localization region except for 2MASX J06435104−5110507, which at 6' distance from the center of the joint localization lies just outside its r₉₀.

We generated a spectrum of Swift J0644.5−5111 from the 300 s detection image and fitted spectral models within the xspec environment. The relatively low signal-to-noise admits a broad range of spectral models, including simple power-law and thermal bremsstrahlung models, which we prefer and present in Tables 1 and 2. Using the best-fit power law we derive a fluence of S_γ = 4.0 ± 1.8 × 10⁻⁶ erg cm⁻² (15–150 keV).

This implies a radio to gamma-ray fluence ratio of $\mathrm{log}\eta =5.8\pm 0.2$ for FRB 131104, where η ≡ S_GHz/S_γ is expressed in units of Jy ms erg⁻¹ cm as defined by Tendulkar et al. (2016). Those authors estimate that the SGR 1806−20 giant flare had $\mathrm{log}\eta \lt 5.9$ based on modeling of the sidelobe response of the Parkes multibeam receiver (strong constraint), or alternatively, $\mathrm{log}\eta \lt 7.9$ in an idealized diffraction-limited treatment (weak constraint). Our value for FRB 131104, consistent with the lower limits we derive for FRB 110626 and for two fainter bursts from FRB 121102 (Table 1), is consistent with these upper limits from SGR 1806−20.¹⁰

For a nominal 10 GHz bandpass and the observed flat or inverted spectrum (Ravi et al. 2015), the integrated radio fluence of FRB 131104 is S_radio ∼ 3 × 10⁻¹⁶ erg cm⁻². The gamma-ray counterpart thus increases the energy requirements for this event by a factor of roughly ten billion. (Given its inferred off-center location within the receiver beam, the radio fluence of FRB 131104 may be underestimated.)

We constructed a light curve of Swift J0644.5−5111 using batcelldetect to measure the counts from Swift J0644.5−5111 in each of thirteen 64 s and two 320 s soft-band (15–50 keV) exposures covering the sky position of the transient during and immediately subsequent to the occurrence of FRB 131104. The soft-band light curve, presented in Figure 1 and Table 3, suggests a transient duration (for 90% of the burst fluence) of T₉₀ ≈ 380 s, extending partway into the next 300 s exposure, which maintained the same pointing.

To quantify the likely burst duration, we adopt a simple "step-function" model, assuming a fixed active flux level, a start time Δt = 0 relative to the FRB, and an end time of Δt = T₉₀. (We take this approach because the duration of a step-function light curve more closely corresponds to T₉₀ than to T₁₀₀ for realistic gamma-ray transient light curves.) We make a χ² minimization of this function against the light curve at 64 s resolution, finding a best-fit for T₉₀ = 377 s (χ² = 9.9 for 11 d.o.f.) that extends exactly to the end of the highest-flux sixth time bin, and a 1σ uncertainty range of T₉₀ = 377 ± 24 s (Table 2).

We derive a model-independent lower bound on the burst duration by noting the transient signal-to-noise ${ \mathcal S }\propto {S}_{\gamma }/\sqrt{{T}_{90}}$ for relatively plateau-like light curves. Since the transient has ${ \mathcal S }=4.2\sigma$ over 300 s, for T₉₀ ≤ 125 s it would have ${ \mathcal S }\geqslant 6.5\sigma$, leading to a BAT trigger. This is because the BAT flight software continuously evaluates count rates across the detector in search of excesses (leading to a rate trigger, synthesis of a sky image, and a search for a new point source), and synthesizes sky images at 64 s intervals and multiples thereof during each fixed pointing, in search of new point sources (leading to an image trigger). Due to uncertainty as to the exact shape of the light curve and its timing with respect to BAT integration intervals, we conservatively quote T₉₀ > 100 s as our 90%-confidence lower limit on the burst duration. Since we cannot exclude the possibility of a very extended duration at a very low flux level, we do not set a 90%-confidence upper bound. If the gamma-ray emission did extend to t > 293 s then our quoted fluence will be an underestimate; including the additional fluence seen in the sixth time bin increases the gamma-ray fluence by +66%.

Given its duration (T₉₀ > 100 s), its spectrum, and the absence of a BAT trigger, non-detections of the source by the Fermi GBM, INTEGRAL SPI-ACS, Konus-Wind, and other Interplanetary Network detectors are not further constraining of the counterpart's gamma-ray properties.

Given its observed dispersion measure, DM = 779 ± 1 pc cm⁻³, the maximum distance for FRB 131104 is D ≈ 3.2 Gpc or z ≈ 0.55 (Murase et al. 2016). This distance estimate will hold approximately in the absence of significant plasma density near the source, and will be reduced if there is any substantial quantity of such local plasma. It is also subject to systematic uncertainties regarding the ionized fraction of gas in the intergalactic medium. At the maximum distance, the implied gamma-ray energy output for FRB 131104 is E_γ ≈ 5 × 10⁵¹ erg.

The large inferred gamma-ray energy release for FRB 131104 suggests the possibility of subsequent high-energy afterglow, supernova, or galactic nuclear emissions. We therefore reviewed a set of Swift X-ray and UV/optical observations of the burst position that were taken two days after the original radio discovery, likely in a search for such counterparts (Table 4).

Table 4.  Follow-up Observations of FRB 131104

Instrument	Band	tstart	tstop	Exp.	Cover	Limit	Comments
		(d)	(d)	(s)
Swift XRT	0.3–10 keV	2.00	2.10	4912	97%	4 × 10−14	Excludes brightest 65% of Swift GRB afterglows
Swift UVOT	U	2.00	2.07	157	75%	32 μJy	$U\gt 19.4$ mag
"	B	2.00	2.07	157	"	59	B > 19.6 mag
"	V	2.01	2.08	157	"	132	V > 18.6 mag
"	UVM2	2.01	2.10	3432	"	4.3	⋯
"	UVW1	2.00	2.14	322	"	17	⋯
"	UVW2	2.01	2.08	629	"	8.1	⋯
VLT X-shooter	R	2.48	2.55	720	1%	4.5 μJy	Targeting X1 and X2; not used for transient search

Note. Coverage fractions ("Cover") are quoted for the FRB 131104 radio localization 90%-confidence region (r₉₀), requiring achieved depth of >90% of the exposure time for Swift UVOT and >50% of the exposure time for Swift XRT; coverage fractions for the joint localization 90%-confidence region are 96% (Swift XRT), 92% (Swift UVOT), and 0% (VLT X-shooter), respectively. The Swift XRT X-ray limit is in units of erg cm⁻² s⁻¹. UVOT images were searched for new sources by comparison to archival images; absence of deep pre-imaging forestalls a similar search with X-shooter data, which in any case do not provide coverage of the joint localization region.

Download table as:  ASCIITypeset image

We processed the Swift X-ray Telescope (XRT; Burrows et al. 2005) follow-up data (Swift ObsID 00033033001) with xrtpipeline, which provided a clean event list, image, and exposure map for the observations. The XRT data cover 97% of the FRB 131104 radio localization (and 96% of the joint localization) to better than half the nominal 4900s depth, and were taken in two extended exposures between t + 2.0 days and t + 2.1 days post-burst. We identified sources using the wavdetect routine from the ciao software package¹¹ , running against a range of scales, setting the single-trial source threshold at p = 10⁻⁶, and fixing the PSF FWHM at 12''.

This analysis yields two high-confidence X-ray sources (Figure 2), which we designate Swift J064339.9−512042 or X1 (located well outside the joint localization) and Swift J064409.6−511853 or X2 (located 49'' beyond the joint localization r₉₀, on the edge of the 95%-containment region). For each of these sources, we produced refined positions and r₉₀ estimates, optimally extracted source and background counts, and X-ray spectral fits and flux estimates using the routines of Evans et al. (2009) via the UK Swift Science Data Centre.¹²

These established that Swift J064339.9−512042 or X1 (significance ${ \mathcal S }=5.2\sigma ;$ 17 counts compared to 1.3 expected from background), is located at R.A. 06^h43^m3996, decl. −51°20'429 (J2000), with r₉₀ = 31, and has an estimated X-ray flux of ${1.4}_{-0.7}^{+1.2}\times {10}^{-13}$ erg cm⁻² s⁻¹ (0.3–10 keV, 90%-confidence bounds); while Swift J064409.6−511853 or X2 (${ \mathcal S }=3.2\sigma ;$ 9 counts compared to 0.8 expected from background), is located at R.A. 06^h44^m0965, decl. −51°18'536 (J2000), with r₉₀ = 56, and has an estimated X-ray flux of ${8.0}_{-4.9}^{+13.5}\times {10}^{-14}$ erg cm⁻² s⁻¹. In separate analyses, we confirmed that the arrival times and radial distributions of counts for both sources are consistent with a non-variable point-source nature for each.

Given that X1 is located well outside the joint localization, and that the z = 1.525 redshift for X2 (derived below) is beyond the horizon for FRB 131104, we do not consider either source to be a viable counterpart. We conclude instead that we have established an upper limit of 4 × 10⁻¹⁴ erg cm⁻² s⁻¹ (0.3–10 keV) on the flux of any X-ray counterpart at t + 2 days post-burst.

To determine a constraint on GRB afterglow-like counterparts, we considered a library of 192 Swift-detected GRB X-ray afterglows analyzed and modeled by Racusin et al. (2009). Using the power-law decays of these afterglows we interpolated or extrapolated the observed X-ray behavior to t + 2 days post-burst, finding that 65% (125 of 192) of these afterglows have inferred fluxes above our limit. Hence we conclude that the XRT observations exclude association of FRB 131104 with the brightest 65% of Swift-type X-ray afterglows.

We processed Swift UV/Optical Telescope (UVOT; Roming et al. 2005) data with uvotskycorr and uvotimsum, which yields aspect-corrected coadded images covering 75% of the radio localization (and 92% of the joint localization) to better than 90% of the nominal exposures in each of six UVOT filters: U (157 s exposure), B (157 s), V (157 s), UVM2 (3432s), UVW1 (322 s), and UVW2 (629 s). We coadd these six images in a search for new or anomalously bright sources by comparison to archival images; none are found within the UVOT area. Using uvotsource, we derive per-filter flux density upper limits for source-free regions of the image as listed in Table 4; these serve as upper limits for any new point source in the UVOT region.

Review of the Swift GRB Table¹³ shows that there are no examples of Swift burst afterglows detected by UVOT and undetected by XRT; hence we prefer the XRT constraint (excluding the brightest 65% of Swift afterglows) for afterglow-like counterparts.

With respect to supernova (SN) limits, pervasive line-blanketing by metals suppresses SN optical/UV flux at λ < 4000 Å, so that the V > 18.6 mag and B > 19.6 mag limits (Table 4) are the most useful for SN constraints. Observations at t + 2 days post-explosion are sub-optimal for catching associated SNe, as maximum light is achieved after two to three weeks, depending on SN type and explosive energy. Tabulations of SN light curves suggest that optical magnitudes at t + 2 days are ΔM_V ≈ 2 mag fainter than peak for type Ibc supernovae (Drout et al. 2011), and are likely even further suppressed for type Ia events (Firth et al. 2015), although explosion times for Ia events are not well constrained. As a result, the UVOT limits (ignoring unobserved portions of the joint localization, the 10% probability of the source lying outside the joint localization, and any possible extinction in the host galaxy) are not constraining for the nominal distance D = 3.2 Gpc, as they require any associated SN to have peak absolute magnitude M_V > −23.9 mag (M_B > −22.9 mag). UVOT limits become constraining at closer distances; at D = 320 Mpc the limits of M_V > −18.9 mag (M_B > −17.9 mag) at peak exclude the brightest ≈10% of type Ibc supernovae (Drout et al. 2011) and the brightest ≈60% of type Ia supernovae (Ashall et al. 2016).

Unfortunately, the deep southern sky was not being optically surveyed in an unbiased fashion for transients and supernovae during the 2013–14 southern summer season. In particular, the Catalina Real-Time Transient Survey (Drake et al. 2009) had ceased use of its Siding Springs Observatory 1 m facility in July; the La Silla-QUEST Supernova Survey (LSQ; Walker et al. 2015) was using a drift-scan approach not suited (nor applied) to deep polar regions δ ≲ −40°; and the Optical Gravitational Lensing Experiment Real-Time Transient Search (OGLE-IV; Wyrzykowski et al. 2014) was focused on transient discovery and variable star studies within and near the Small and Large Magellanic Clouds, without extending as far as the location of FRB 131104. Hence it is not possible to set quantitative limits on the peak magnitude of any associated supernova to FRB 131104 beyond what can be determined from the UVOT observations.

We retrieved X-shooter (Vernet et al. 2011) follow-up observations of the two Swift X-ray sources from the ESO Data Archive.¹⁴ Each spectroscopic observation was preceded by four 180 s R-band acquisition images. We coadded the undithered exposures for each target, yielding two 15 × 15 images with FWHM ≈ 07 seeing, reaching to roughly R ≈ 22 mag as determined by photometry of unsaturated stars from the USNO-B2.0 catalog (Figure 2). The image of the region surrounding Swift J064339.9−512042 recovers a likely point-source optical counterpart seen in archival data, which has R ≈ 19.0 mag, coordinates R.A. 06^h43^m3984, decl. −51°20'460, and r₉₀ ≈ 03. The image of the region surrounding Swift J064409.6−511853 reveals a single candidate point-source optical counterpart with R ≈ 20.8 mag, coordinates R.A. 06^h44^m0950, decl. −51°18'540, and r₉₀ ≈ 03 (Figure 2).

Each of the candidate optical counterparts to the two Swift X-ray sources was observed in two dithered 600 s exposures through a 1'' slit, yielding spectra from each of the three X-shooter spectrographs (UVB, VIS, NIR). We retrieved the Phase 3 data products¹⁵ , which provide the wavelength-calibrated, rectified, and coadded two-dimensional spectral images and extracted (one-dimensional) spectra for each spectrograph, as processed by the X-shooter Pipeline (Modigliani et al. 2010). We examined the two-dimensional and extracted spectra to determine the nature of each object.

The likely counterpart to Swift J064339.9−512042 (X1) exhibits broad, redshifted emission lines of Mg ii (2798 Å) and Hα with z ≈ 0.38 and Δv ≈ 3500 km s⁻¹ (FWHM), demonstrating its nature as a quasar. A weaker, broad Hβ emission line at this redshift is also present. The redshift is confirmed and refined via narrow emission lines of [O ii] (3727 Å), [O iii] (4959 Å), [O iii] (5007 Å), and [N ii] (6584 Å), as well as an absorption feature due to the Ca ii (3935 Å, 3970 Å) doublet. These yield a refined redshift of z = 0.383, X-ray luminosity L_X ≈ 7.2 × 10⁴³ erg s⁻¹, and absolute magnitude M_R ≈ −22.6 mag. This spectroscopic identification confirms the source as the optical counterpart to Swift J064339.9−512042.

The likely counterpart to Swift J064409.6−511853 (X2) exhibits broad, redshifted emission lines of C iv (1549 Å), Mg ii (2798 Å), Hβ, and Hα with z ≈ 1.53 and Δv ≈ 3000 km s⁻¹ (FWHM), demonstrating its nature as a quasar. The redshift is confirmed via narrow emission lines of [O ii] (3727 Å), [O iii] (4959 Å), and [O iii] (5007 Å), which yield a refined redshift of z = 1.525, X-ray luminosity L_X ≈ 1.2 × 10⁴⁵ erg s⁻¹, and absolute magnitude M_R ≈ −24.5 mag. This spectroscopic identification confirms the source as the optical counterpart to Swift J064409.6−511853.

Given our findings, and the known existence of repeating FRB sources, we initiated a search for gamma-ray activity in the directions of all known FRBs regardless of relative timing. We established that none of the 1050 triggered Swift GRBs have positions consistent with any of the known FRBs. (We note here the previously published search for Swift and Fermi counterparts to the repeating FRB 121102; Scholz et al. 2016.)

We then examined the complete set of Swift BAT subthreshold events from the Swift archive at HEASARC. We extracted all events that occurred within 2° of any FRB location, excluding FRB 010621, as it lies close to a bright X-ray source. Data run from 2004 December to 2016 February, excluding 2011 January (we identified unresolved quality issues with data files for this month), giving a total of 134 months. We searched for an excess of subthreshold events within 15' of each FRB position by comparison to the larger region around the FRB (and beyond 15'; Table 5). Estimated on-axis equivalent exposure times for each FRB position, derived from the exposure map in Baumgartner et al. (2013), are listed. We find no statistically significant excess for any of the examined positions, and set a 3σ upper limit on the number of such subthreshold events of <14 for the typical such position (corresponding to rates per Ms of on-axis equivalent exposure of <0.73 Ms⁻¹), and <11 for FRB 121102 (<0.75 Ms⁻¹). A stacked search over all 15 non-repeater FRB positions gives a limit of <29 excess subthreshold events in all or <2 per FRB position (<0.11 Ms⁻¹).