Characterizing the Fast Radio Burst Host Galaxy Population and its Connection to Transients in the Local and Extragalactic Universe

FRB20180301A, a burst originally detected in the Breakthrough Listen project at the Parkes radio telescope (Price et al. 2019), was observed to emit repeating radio pulses in sensitive follow-up observations with the Five-hundred-meter Aperture Spherical radio Telescope (FAST) telescope using the 19-beam receiver centered at 1.25 GHz (Luo et al. 2020). Fifteen repeating bursts, with fluences ranging from 0.03 to 0.4 Jy ms, were detected from the source of FRB20180301A in a total effective observation time of ∼12 hr at FAST.

We performed follow-up observations of FRB20180301A using Director's Discretionary Time (DDT) with 40 hr scheduled under VLA/19B-351, using the VLA in its C array configuration at the L band, spanning 1–2 GHz. We used the L16f5DC-realfast correlator mode, which enables a fast sampling time of 10 ms. We observed the field of FRB20180301A centered at RA(J2000): 06:12:54.96 and decl.(J2000): +04:38:43.60 covering a localization uncertainty region of at 2.6' at 1.25 GHz obtained from detection at FAST. The source was observed at nine epochs, each with an observation time of 2 hr in the period 2020 February–May. Each observation had an on-source time of ∼1.5 hr and was searched for fast transients by the Realfast system. The details of the Realfast search procedure can be found in Law et al. (2018, 2020). We detected a repeat burst from FRB20180301A on 2020 May 28 at UTC 19:14:47.310 in the last epoch. It was detected in the real-time system with an image signal-to-noise ratio (S/N) of 10.8 at a DM of 517.45 pc cm⁻³. The deep learning-based classifier fetch (Agarwal et al. 2020) reported an astrophysical probability of 99% for this burst. We used the recorded visibilities of the burst data to refine our estimate of the burst properties in an offline reprocessing. We reran the search with a finer DM grid at 0.1% fractional sensitivity loss on a subband of the data. This subband was manually identified to consist of burst signal. The refined search led to an improvement in detection significance to ∼19 at a DM of 536 pc cm⁻³. The top panel of Figure 1 shows the burst profile and dedispersed spectrogram. This figure also shows a weaker component ∼50 ms after the burst that is present in the lower part of the band. This could either be a component of the main burst, or another burst from this FRB. We used the visibilities from the main component to determine the localization of this FRB.

Zoom In Zoom Out Reset image size

We used burstfit ²⁶ (Aggarwal et al. 2021a) to model the spectro-temporal properties of FRB20180301A using its spectrogram. Following the method described in Aggarwal et al. (2021a), we modeled the spectrum and the pulse using a Gaussian, and fit for the following parameters simultaneously: arrival time of peak, Gaussian FWHM of pulse, peak of the spectrum, FWHM of the spectrum, fluence and DM. We used scipy.curve_fit followed by Markov Chain Monte-Carlo methods to obtain the posterior distribution of all the fit parameters. The resulting burst properties are presented in Table 1.

Table 1. Measured and Derived Properties of the New FRBs Presented in This Work

Properties	FRB20180301A	FRB20191228A	FRB20200906A
Arrival time a (UT)	2020-05-28 19:14:47.310	2019-12-28 09:16:16.444	2020-09-06 21:40:50.923
S/N	19.0 b	22.9	19.2
DM (pc cm−3)	${536}_{-13}^{+8}$	297.5 ± 0.05	577.8 ± 0.02
DMISM NE2001 (pc cm−3)	152	33	36
DMISM YMW16 (pc cm−3)	254	20	38
DMcosmic c (pc cm−3)	289	210	324
R.A. (J2000)	06h12m5444(±001 ± 062)	22h57m4330(±034 ± 083)	03h33m5908(±010 ± 034)
Decl. (J2000)	+04d40'15.8(±001 ± 06)	−29d35'38.7(±03 ± 08)	−14d04'59.5(±01 ± 06)
Fluence (Jy ms)	${4.9}_{-0.4}^{+0.5}$	${40}_{-40}^{+100}$	${59}_{-10}^{+25}$
Pulse width (ms)	${7}_{-3}^{+2}$	2.3 ± 0.6 d	6.0 ± 0.6
Spectral energy density e (erg Hz−1)	6.9 × 1030	6.4 × 1031	1.1 × 1032
Persistent source,	<1.8 × 1022	<3.4 × 1021	<4.3 × 1021
radio luminosity (W Hz−1)	1.5 GHz	6.5 GHz	6 GHz
Host galaxy
Redshift f	0.3304	0.2432	0.3688
u − r g (rest-frame)	0.90 ± 0.11	2.13 ± 0.74	1.22 ± 0.11
Mr g (rest-frame)	−20.18 ± 0.07	−18.26 ± 0.05	−21.49 ± 0.05
Galactic E(B − V) g	0.46	0.02	0.05
M* g (1010 M⊙)	0.23 ± 0.06	0.54 ± 0.60	1.33 ± 0.37
SFR f (M⊙ yr−1)	1.93 ± 0.58	0.50 ± 0.15	0.48 ± 0.14
log(sSFR) (yr−1)	−9.08	−10.03	−10.44
Metallicity f , h	8.70	8.48	8.76
Projected offset from galaxy center (kpc)	10.8 ± 3.0	5.7 ± 3.3	5.9 ± 2.0
Effective radius (kpc)	5.80 ± 0.20	1.78 ± 0.06	7.58 ± 0.06

Notes. 1σ uncertainties are quoted for these measurements.

^a Arrival time of a repeating burst from FRB20180301A source and one-off ASKAP/CRAFT FRBs. ^b This is the coherent S/N of the FRB detected in the VLA image. ^c Estimated using the Macquart relation. ^d FRB20191228A shows a scattering tail of 6.1(6) ms. The deconvolved width is 2.3(6) ms. ^e The energies are derived assuming a flat spectrum for FRBs (α = 0) and zero k-correction. ^f These measurements are derived from spectroscopy. ^g These properties are derived using CIGALE SED fitting. ^h In units of 12 + log[O/H].

Download table as:  ASCIITypeset image

2.1.1. Spatial Localization

2.1.2. VLA Radio Continuum Image

2.1.3. Astrometry

2.1.4. Optical Follow-up

In archival Panoramic Survey Telescope and Rapid Response System (Pan-STARRS) images we identify a faint galaxy, PSO J093.2268+04.6703, as the putative host of FRB20180301A. We acquired deep (0.5–0.7 hr in each band) follow-up imaging of the FRB field with the Nordic Optical Telescope (NOT) using the Sloan filters u, g, r, i, and z on nights between UT 2021 October 26 and December 14, followed by additional deep (36 × 100 s) Gemini r-band observations on UT 2021 January 25. The Gemini image was astrometrically calibrated to match the Gaia DR2 catalog (Gaia Collaboration et al. 2018; Lindegren et al. 2018), with a relative accuracy of 02. We note that there is an additional nearby source to the south of the identified host galaxy in the field of FRB20180301A. The morphology of this source is consistent with being point-like, so we classify it as a Galactic foreground star.

Observations of the host with the Keck/DEep Imaging Multi-Object Spectrograph (DEIMOS) on UT 2020 September 17 determined the galaxy's redshift z = 0.3304 based on nebular emission lines such as [N ii] and Hα. The measured redshift was found to be consistent with the redshift range (z = 0.13–0.35) estimated by extragalactic DM in Luo et al. (2020). We also performed near-IR (NIR) observations of the galaxy in the JHK bands using the MMT telescope on UT 2021 February 27 and 28. The photometric measurements derived from MMT imaging, along with those from NOT, were used by CIGALE (Noll et al. 2009) for spectral energy distribution (SED) fitting to estimate the stellar mass of the galaxy (see the Appendix for details). The deep r-band Gemini imaging was used for the probabilistic host galaxy association analysis. We find a probabilistic association of transients to their hosts (PATH; Aggarwal et al. 2021b) probability of 99.9% for the association of FRB20180301A with the given galaxy. The host is presented in Figure 2 and properties are listed in Table 1.

Zoom In Zoom Out Reset image size

FRB20191228A was detected on 2019 December 28 at UTC 09:16:16.44440 during CRAFT observations conducted with ASKAP, with 28 antennas in a 336 MHz band centered on 1271.5 MHz. The burst had a detection S/N of 23 at a DM = 297.5(5) pc cm⁻³. The real-time detection in online incoherent-sum data triggered a download of 3.1 s of voltages around the FRB. The offline correlation of voltages across 28 antennas and their interferometric analysis led to an initial localization of FRB20191228A.

On 2020 January 31, we used the Australia Telescope Compact Array (ATCA) in the frequency range of 1.1–3.1 GHz to perform astrometric observations of background sources near the position of FRB20191228A, to refine the position of the burst. In addition to four target background sources in the field of FRB20191228A (J2258−2929, J2255−2937, J2258−2955, and J2259−2957), we observed three gain calibrators (PKS2254−367, PKS2337−334, and PKS2255−282). For each target background source, we obtained three position estimates—one for each gain calibrator—by applying phase-calibration solutions derived from that calibrator. We observed a scatter of ∼80 mas in R.A. and ∼200 mas in decl. in the position of the background sources depending on which gain calibrator was used. To account for this systematic uncertainty to the ATCA positions, for each background source we took a weighted mean of the three positions to estimate the probable location, and added the average residual offset from this mean position in quadrature to our estimate of the absolute positional uncertainty for the source.

The positions of target sources obtained from the ATCA radio image were compared with those obtained from the 3.1 s ASKAP data, following the method described in Day et al. (2020), to astrometrically register the FRB image frame to that of the International Celestial Reference Frame (ICRF3; Gordon 2018) and estimate the accuracy of this registration. J2258−2929 and J2259−2957 were excluded from the field source comparison due to low S/N in the ASKAP image for the latter, and the former being resolved.

Finally, we computed a weighted mean systematic image frame offset using the method described in Day et al. (2021). We found positional offsets of 0410 ± 0830 and −0856 ± 0823 in R.A. and decl., respectively. After accounting for astrometric shifts in the image frame as well as statistical and systematic errors, the final FRB position is R.A.(J2000): 22^h57^m4330(±034 ± 083) and decl.(J2000): −29^d35'38.7(±03 ± 08).

3.1.1. Follow-up Observations

FRB20200906A was discovered in the incoherent sum of seven ASKAP antennas on UTC 2020 September 6 21:40:50.923 during CRAFT observations at 864.5 MHz. The burst had an optimal S/N of 19.2 at a DM of 577.8(2)pc cm⁻³ in the low-time-resolution search data stream. Voltages spanning 3.1 s around the FRB were downloaded, cross-correlated and imaged offline to obtain a preliminary position of the burst. The astrometric registration was performed by comparing background sources in the 3.1 s ASKAP image with their NRAO VLA Sky Survey (NVSS) radio source catalog (Condon et al. 1998) counterparts. We cross-matched seven sources and computed a weighted mean systematic image frame offset using the method described in Day et al. (2021). We found positional offsets of 205 ± 034 and 051 ± 055 in R.A. and decl., respectively. The final burst position after correcting for the astrometric shifts in the image frame is R.A.(J2000): 03^h33^m5908(±010 ± 034) and decl.(J2000): −14^d04'59.5(±01 ± 06), where statistical and systematic uncertainties in R.A. and decl. are quoted respectively.

3.2.1. Follow-up Observations

We use the data published in Heintz et al. (2020) and Mannings et al. (2020), and measurements derived for the hosts in this work. We also update the measurements published for a subset of FRB hosts (see the Appendix). For FRB20171020A, FRB20201124A, and FRB20200120E, we use the data published in Mahony et al. (2018), Fong et al. (2021), and Bhardwaj et al. (2021), respectively (see Table A1 and A3).

We considered only the FRB host galaxies for which the PATH posterior probability is greater than 90% (see last column of Table A2). This includes all three of the new FRBs presented here which have posterior probability P(O∣x)> 0.999. The hosts of FRB20190614D (Law et al. 2020), FRB20181112A (Prochaska et al. 2019), and FRB20190523A (Ravi et al. 2019) have low probabilities of P(O∣x) of 0.58, 0.67, and 0.82, respectively, and are therefore excluded. Also, FRB20191001A has two nearby candidate hosts at a common redshift (Bhandari et al. 2020) that yield PATH posterior probabilities P(O∣x) = 0.6 and 0.4. We proceeded by adopting the galaxy with smaller angular separation from the FRB as the host, which is akin to adopting a stronger prior on the projected offset than adopted by Aggarwal et al. (2021b). Future associations will refine this assumption. We also included the host of FRB20200120E (Bhardwaj et al. 2021; Kirsten et al. 2021) and the candidate host for FRB20171020A, because of their low probability of chance association (F. North-Hickey et al. 2021, in preparation). Thus, we obtained a sample of 16 confident host associations for our analysis, of which 10 are hosts of apparently nonrepeating FRBs and six host repeating FRBs (see Table A3).

The spectroscopy and photometry of all FRB hosts are analyzed with the pPXF (Cappellari 2017) and CIGALE (Noll et al. 2009) software packages by fitting a set of stellar population models and star formation history to the spectra and SED, respectively (see Heintz et al. (2020) for details). The star formation rates (SFRs) are derived from the dust-corrected Hα line flux measurements (Kennicutt 1998), adopting an initial mass function (IMF) from Chabrier (2003).

We used the PRism MUlti-object Survey (PRIMUS) data as an underlying sample of field galaxies (Moustakas et al. 2013). We restricted the redshift range to be z < 0.6 to match the redshifts of the FRB hosts (see Figure 4). We used a K-correct synthesized rest-frame SDSS absolute magnitude of ∼108,000 galaxies and stellar masses/SFRs of ∼31,200 galaxies. We note that the PRIMUS sample lacks good stellar mass and SFR measurements of field galaxies for z < 0.2. All stellar masses and SFRs assume a universal (Chabrier 2003) IMF and are derived using the SED-modeling code iSEDfit, designed to extract the physical properties of galaxies (Moustakas et al.2013).

Zoom In Zoom Out Reset image size

Figure 5 shows our comparison of the color−magnitude and SFR−M_* distributions of the FRB host galaxies with those of the general population of galaxies at redshift <0.6. The left panel of Figure 5 presents the current SFR as a function of M_*. The hosts of repeating and one-off bursts are distinguished by symbol shape, and the color represents the four redshift bins. We also present the redshift evolution of the boundary separating the star-forming and quiescent galaxies (Moustakas et al. 2013). Most of the FRB hosts lie in or around the star-forming cloud of galaxies, but are offset from the star-forming main sequence for galaxies with similar stellar masses. We caution here that since a significant fraction of the FRB hosts show low-ionization nuclear emission-line region (LINER)-like emission (see Section 4.4), the derived SFR should in these cases only be considered as upper limits since the total line emission may not reflect solely that of star formation. This would further offset the FRB hosts from the star-forming main sequence.

Zoom In Zoom Out Reset image size

The right panel of Figure 5 shows a color–magnitude diagram and provides information about the overall stellar populations in these galaxies. The late-type galaxies with ongoing star formation and, therefore, young stellar populations lie in the "blue cloud", i.e., blue galaxies (Strateva et al. 2001), while massive early-type galaxies live in the "red and dead" zone characterized by very low star formation and hence older stellar populations, i.e., red galaxies. The host galaxies of the FRBs appear to lie on the luminous side of the absolute magnitude distribution, mainly near the "blue cloud" and "green valley" region, where galaxies are expected to be transitioning between star-forming and quiescent systems (Martin et al. 2007). We observe a dearth of red galaxies in our current sample of FRB host galaxies.

In Figure 6 we again present the color–magnitude diagram, but with the background PRIMUS galaxy sample weighted by their stellar masses and divided into redshift bins of 0.0 < z < 0.3 and 0.3 < z < 0.6. This is a good approximation of where most stars are in the local universe. We show that the majority of FRB hosts do not trace massive red galaxies and also do not seem to align with the color–magnitude diagram space of the peak of stellar mass-weighted blue galaxies which tend to redder color, particularly for the low-redshift bin. To quantify this trend, we examined the null hypothesis that FRB hosts (both repeating and nonrepeating population) track stellar mass. The galaxy stellar mass function (GSMF) of low-z galaxies, Φ(M_*)ΔM, was weighted by stellar mass and compared to the observed FRB host mass distribution. We used the double Schechter function to model GSMF for all, star-forming, and passive galaxies in the redshift range 0.2 < z < 0.5 in the COSMOS field (Davidzon et al. 2017). Our Kolmogorov–Smirnov (KS) tests comparing the cumulative mass distributions of all FRB hosts, together and separately for the two FRB populations, with the mass-weighted stellar mass distribution of field galaxies yield a p-value P_KS < 0.05 (see Table 2). Therefore, we reject the null hypothesis that FRB hosts directly track stellar mass with more than 95% confidence, consistent with the findings of Heintz et al. (2020). We note that the redshift range of FRB hosts is broader than the range used for the stellar mass-weighted mass function. We replicated the above analysis using a subset of FRBs in the range 0.2 < z < 0.5 to investigate if the differences in the redshift had an effect on our conclusions. We obtained identical results, indicating that the effect of redshift evolution is not substantial.

Zoom In Zoom Out Reset image size

Furthermore, we tested the null hypothesis that FRB hosts track SFRs. We used the star formation rate distribution function derived from the UV and IR luminosity Schechter functions for the local universe using Galaxy Evolution Explorer (GALEX; Bothwell et al. 2011) data. We then computed the star formation rate volume density distribution function, which is given by Φ(Ψ) × Ψ, where Ψ is the star formation rate in M_⊙ yr⁻¹. The p-values for the KS test performed between the cumulative star formation rate volume density distribution function and the SFRs of FRB hosts (both repeating and nonrepeating population) in our sample are presented in Table 2. We find the p-values for all and nonrepeating subset to be <10⁻³. (Considering the SFRs as upper limits due to possible LINER emission contamination (see Section 4.4) would make the discrepancy even more significant.) However, when comparing with the repeating host population, we find a p-value slightly higher than our significance level. As a result, while we reject the null hypothesis with greater than 95% confidence for all FRB hosts, we are unable to reject it with the same level of confidence for the repeating host population. We note a possible caveat of comparing the SFRs derived using the UV and IR luminosity for the GALEX sample with those derived using Hα emission-line luminosity for the FRB host sample. Nevertheless, Lee et al. (2009) showed a coarse agreement between the far-UV and Hα SFRs, where the SFRs agree to within about a factor of two for the majority of galaxies with SFR ≥ 0.01 M_⊙ yr⁻¹.

Zoom In Zoom Out Reset image size

Finally, we investigated the specific star formation rates (sSFRs) of FRB hosts in our sample with those of sSFR functions derived from the star-forming sample of galaxies defined by a main sequence using a color–color selection in the COSMOS and GOODS surveys in the redshift range 0.2–0.4 (Ilbert et al. 2015). The sSFR function can be modeled as a log-normal or a double exponential profile for a given stellar mass bin range (Equations (2) and (3) of Ilbert et al. 2015). We computed the function over four mass bins ranging from $9.5\lt \mathrm{log}(M)\lt 11.5$ and combined them together. The weighted cumulative sum of this sSFR function is compared to the cumulative distribution of sSFRs of FRB hosts in the same mass range. The results are presented in Table 2 and cumulative distributions are presented in Figure 7. We observed that FRB hosts do not follow the sSFRs of star-forming galaxies and thus reject the null hypothesis with more than 95% confidence.

Thus, we conclude that FRB hosts have lower M_*, SFR, and sSFR than randomly selected field galaxies weighted by M_*, SFR, or sSFR.

To identify the dominant source of ionization in FRB host galaxies, we plot their nebular emission-line ratios, namely [N ii]/Hα and [O iii]/Hβ, in a Baldwin–Phillips–Terlevich (BPT) diagram (Baldwin et al. 1981) in Figure 8. This diagram can also be used to distinguish between star-forming (SF) galaxies, LINER galaxies, and active galactic nuclei (AGNs; see Kewley et al. 2001 for more details). For comparison, we show the distribution of nearby (0.02 < z < 0.4) emission-line (>5σ significance) galaxies from the Sloan Digital Sky Survey (SDSS) including the standard demarcation lines between SF, AGN, and LINER galaxies (Kauffmann et al. 2003; Cid Fernandes et al. 2010). We performed a 2D KS test to compare the host population of FRBs with that of underlying SF, AGN, and LINER galaxies, and found that they are not drawn from a specific class (P_KS < 10⁻³), which remains consistent with the findings of Heintz et al. (2020). The majority of FRB hosts occupy the star-forming and LINER region of the BPT diagram. We note that all three hosts of repeating FRBs lie in the star-forming region, and the host of repeating FRB20121102A remains an outlier as compared to the hosts of other FRBs (Li et al. 2019).

Zoom In Zoom Out Reset image size

Studies of extragalactic transients often use their physical offsets from the center of their hosts, and locations within their host galaxies, to aid in inferring the nature of their progenitors. In the past, such investigations including other global properties for a small sample of FRB hosts have been conducted (Bhandari et al. 2020; Heintz et al. 2020; Li & Zhang 2020; Mannings et al. 2020; Safarzadeh et al. 2020; Bochenek et al. 2021). The majority of them suggest that galaxies hosting gamma-ray bursts (SGRBs) and CCSNe are similar to galaxies hosting FRBs. Additionally, the progenitor scenarios of LGRBs and SLSNe have been disfavored for the majority of FRBs. Recently, ultra-luminous X-ray sources (ULX)-like binaries were proposed as a possible progenitor for periodically active FRBs (Sridhar et al. 2021). Motivated by such studies, we performed two-sample 1D KS tests comparing the distributions of various global host properties of the most updated sample of FRBs together and separated into two populations (repeating and nonrepeating) with the host galaxy properties of other transients (see Figure 9). We also included Type Ia supernovae—a proxy for the AIC of WD stars, which have been suggested as possible FRB progenitors (Margalit et al. 2019; Kirsten et al. 2021). Table 3 represents the data from the literature that has been used in this study for all these transients. We used the method described in Bochenek et al. (2020) to correct for redshift evolution by scaling the stellar masses and SFRs of host galaxies of FRBs and other transients to be statistically representative of z = 0 galaxies.

While the distribution of projected physical offsets for ULXs and Type Ia supernovae are statistically consistent with those of FRBs, their host galaxies are often more massive and star-forming, i.e., a mean of log(M_*) =10.4 M_⊙ and 10.8 M_⊙, SFR = 6.2 M_⊙ yr⁻¹ and 11.7 M_⊙ yr⁻¹ for the hosts of Type Ia SN and ULXs, respectively. A possible caveat is that the ULX sample in the HECATE and the Chandra X-ray catalog is biased against low-mass galaxies due to various selection effects (Sridhar et al. 2021). However, we observe the sSFRs for ULX hosts to be similar to those of FRB hosts. Also, the majority of the properties of the hosts of SGRBs and CCSNe are very similar to FRB hosts, suggesting that the host population of FRB progenitors shares the same/similar characteristics to those of these transients. A KS test shows that the SFR distribution of LGRB hosts and SLSNe hosts are consistent with FRB hosts. However, we note that such events are common in dwarf galaxies with high sSFRs, as evident from the left-middle and right-bottom panels of Figure 9, which is not consistent with the overall FRB host population. Interestingly, while the results of our KS tests allow us to statistically rule out LGRB scenarios for all FRBs combined (P_KS < 0.05), we are unable to reject the hypothesis that the repeating FRB and LGRB hosts are drawn from the same continuous distribution. When comparing with SLSNe hosts, we found the physical offsets of the repeating FRBs (P_KSR = 0.37), stellar mass (P_KSR = 0.06), and r-band magnitude (P_KSR = 0.06) of their hosts to agree with that of SLSNe hosts. The p-values from our KS tests are presented in Table 4.

The prevailing view currently associates FRBs with magnetars, whose extreme magnetic fields provide a reservoir of energy to produce FRBs (Metzger 2018; Lyutikov & Rafat 2019). With the detection of an FRB-like burst from SGR1935 + 2154, the association of some (low-luminosity) FRBs with magnetars has been observationally confirmed (Bochenek et al. 2020; CHIME/FRB Collaboration 2020). Following the analysis of Chrimes et al. (2021), we compared the host-normalized offset distributions of low-mass X-ray binaries (LMXBs), high-mass X-ray binaries (HMXBs), pulsars and magnetars in the Milky Way with the updated FRB host sample—including all repeating and nonrepeating FRBs in the left panel of Figure 10. Based on our KS test, the offset distribution of only LMXBs in the Milky Way is consistent with that observed for FRBs in their host galaxies (95% confidence).

Zoom In Zoom Out Reset image size

Recently, a repeating FRB20200120E originally associated with the galaxy M81 (Bhardwaj et al. 2021) has been precisely localized, by the EVN network, to a globular cluster (GC) system [PR95] 30244 in M81 (Kirsten et al. 2021). Motivated by this finding, we compared the host-normalized offset distributions of FRBs in our sample with globular clusters in late-type spirals and early-type elliptical galaxies in the right panel of Figure 10.

For the GCs in late-type galaxies, we used a sample of 340 high-quality GC candidates (of which 74 were confirmed using spectroscopy) associated with M81, identified in HST imaging (Nantais & Huchra 2010; Nantais et al. 2010). We note that the available data is dominated by a GC disk population with the majority of GCs within 10 kpc from the center of M81. We also used a sample of 390 GCs associated with the galaxy M31 (both disk and halo population) identified in images from the Wide Field Camera (WFCAM) on the United Kingdom Infrared Telescope and from the SDSS (Peacock et al. 2010). For early-type galaxies, we used the available GC data associated with an isolated elliptical galaxy NGC 821, extending up to 50 kpc from the center of the galaxy (Spitler et al. 2008). The sample consists of 306 GCs identified in the new imaging from the 3.5 m Wisconsin Indiana Yale NOAO (WIYN) Mini-Mosaic imager, supplemented with the HST WFPC2 images. We also used a sample of 1828 GC candidates (of which 270 are confirmed using spectroscopy from Keck/DEIMOS) associated with the L_* elliptical galaxy NGC 4278, identified in HST/Advanced Camera for Surveys and wide-field Subaru/Suprime-Cam imaging (Usher et al. 2013). These selected galaxies have stellar masses and SFRs in the range M_⋆ = 10^10.5–10^11.5 M_⊙ and SFR = 0.01–1 M_⊙ yr⁻¹, overlapping with the SFRs and high-mass end distribution of FRB host galaxies (Tamm et al. 2012; Forbes et al. 2016; Rahmani et al. 2016; Kokusho et al. 2017; Bhardwaj et al. 2021).

We compared the host-normalized offset distribution of FRBs in their hosts with that of globular clusters in selected galaxies and found that they are not consistent with being drawn from the same underlying distribution (95% confidence) except for M81 (P_KS = 0.604). This could be due to the fact that the M81 globular cluster data is incomplete and dominated by disk population.

A possible caveat in the above analysis is that the number of GCs scale with galaxy stellar mass, and our FRB host population spans four orders of magnitude in the stellar mass, i.e., M_⋆ = 10⁸–10¹¹ M_⊙. We also repeated the KS test by selecting FRB hosts which had M_⋆ > 10¹⁰ M_⊙ and found similar results, however with lower confidence of 80%. Furthermore, we note that our FRB host sample currently shows a deficiency of elliptical galaxies. A subset of FRBs in our sample were found to originate from or near the spiral arms of their hosts. Whether they are linked to the disk population of GCs is presently unknown and we are limited by the sensitivity of current telescopes. A much larger sample of precisely localized FRBs with high spatial observations of nearby hosts are needed to pursue this further.