The Demographics, Stellar Populations, and Star Formation Histories of Fast Radio Burst Host Galaxies: Implications for the Progenitors

We begin with a parent sample of 27 events that have been localized to ≲1''–2'', which is necessary for unambiguous host associations with luminous L* galaxies across the redshift range z ∼ 0.1–1 (Eftekhari & Berger 2017). We then apply the following criteria for inclusion in this work:

• 1.  
  We require a PATH (Probabilistic Association of Transients to their Hosts; Aggarwal et al. 2021) posterior probability ≥90%, following the same convention used in Bhandari et al. (2022b). PATH employs a Bayesian framework to calculate the likelihood that a transient is associated with a galaxy given the transient's localization, the galaxy's position on the sky and angular size, and its apparent magnitude. Combined with prior assumptions on the probability that the host is undetected (i.e., fainter than the flux limit of the imaging observations) and the offset of FRBs from the centers of their hosts, PATH reports the posterior probability of association to every galaxy in the field. Higher probabilities correspond to higher likelihood of association. By requiring the posterior probability to be ≥90%, we construct a sample of high-probability host galaxy associations. We apply PATH to four FRBs—20190520B (using CFHT $r^{\prime}$ archival data; Niu et al. 2022 and Gwyn 2008 for details on the $r^{\prime}$ transmission curve), 20211203C, 20210410D, and 20220105A—for which there are no published PATH probabilities. We otherwise use published PATH probabilities from Aggarwal et al. (2021), James et al. (2022), and Bhandari et al. (2022b). This criterion removes the hosts of FRB 20190614D and FRB 20190523A.
• 2.  
  There is no bright (≲10 AB mag) foreground star within 5' of the FRB position. We employ this criterion to ensure that the observations are not contaminated by scattered light from a nearby bright star; in particular, this can inhibit accurate photometry of the host galaxy.
• 3.  
  There are detections of the host galaxy in at least three photometric bands in the optical/IR or otherwise overlapping the observed spectrum's wavelength coverage. For our methodology used to model the FRB hosts, described in Section 4, the absolute flux calibration of the spectrum depends on the photometry. We have found three photometric bands to be the minimum required in order to obtain a trustworthy model.
• 4.  
  We exclude FRBs with burst spectral energies below 10²⁷ erg Hz⁻¹, to exclude low-energy bursts that would be undetectable over the majority of the redshift range spanned by our sample. Assuming an ASKAP detection limit of 4.4 Jy ms (James et al. 2022), this energy cut corresponds to excluding FRBs that can be seen to a maximum redshift of z ∼ 0.003 by the telescope that contributed the bulk of our sample. This criterion excludes two very nearby FRBs/FRB-like signals: FRB 20200120E and the Galactic source SGR J1935+2154. These sources are significant outliers in luminosity space, as is clear from fluence-redshift diagrams such as the one in Ryder et al. (2022). We note that while all bursts seen to date from both of these sources are excluded by this cut, the highest-energy bursts from the FRB-like SGR J1935+2154 are more energetic than the lowest-energy bursts of FRB 20200120E (Nimmo et al. 2022), making them more comparable to each other than to the higher-z FRB population.
• 5.  
  A spectrum of the host galaxy is available and contains detectable spectral features (i.e., emission and absorption lines) for redshift determination and a signal-to-noise ratio (S/N) ≳3/Å in the continuum. We note that spectroscopy is essential for breaking the known degeneracy in age, dust, and metallicity in SED modeling (e.g., Bell & de Jong 2001; Leja et al. 2017). Additionally, the thousands of data points inherent in a spectrum provide more information to fit than photometry alone. We note that even though there is relative difficulty in detecting high S/N features for quiescent versus star-forming galaxies, no hosts are excluded on this criterion alone.

While these sample criteria are enforced for the majority of the FRBs, we made two exceptions (requiring additional care in our data processing) to ensure that the small sample of repeating FRBs was not further reduced. The repeating FRB 20190520B (Niu et al. 2022) fails the bright star criterion due to its ≈1' proximity to the bright, r ∼ 8 AB mag, variable star V1042 Sco B. In addition to being a repeating FRB, this host contains the second known PRS, making it especially noteworthy. We took appropriate steps during the data collection and reduction phases to ensure the photometry and spectroscopy were not contaminated by extra flux from the star (see Sections 3.1 and 3.2 for details). As mentioned above, we use archival CFHT $r^{\prime}$ (which is close to Sloan $r^{\prime} ;$ Gwyn 2008) data for calculating the PATH probability of this host galaxy. Our analysis assigns the host identified in Niu et al. (2022) a posterior probability of nearly 100%, robustly confirming the host association. The host galaxy of the repeating FRB 20190711A fails the spectroscopic continuum signal requirement; at a decl. of −80°, the source is always at high airmass for most available follow-up resources and is also quite faint, making observation challenging. While we use the existing Gemini South/Gemini Multi-Object Spectrograph (GMOS) spectrum (first reported in Macquart et al. 2020) to determine the redshift, this spectrum does not have a high enough S/N for our subsequent modeling. In this case, we still include this host, and take advantage of the seven photometric data points for modeling.

When applying these criteria to the existing literature and new FRBs obtained from the ASKAP/CRAFT and MeerKAT/MeerTRAP collaborations, we obtain a sample of 23 FRB host galaxies (six repeating FRBs and 17 apparent nonrepeaters). We list the properties of the FRBs, discovery surveys, and optical host magnitudes in Table 1. This compilation is the largest sample of highly secure localized FRB host galaxies to date and allows for a systematic study of their host properties and SFHs.

Once the host for a given FRB is identified, we first search for archival photometry from optical and near-infrared (NIR) surveys: the SDSS (Alam et al. 2015), DECaLS (Dey et al. 2019), the Dark Energy Survey (DES; Abbott et al. 2021), the PS1 3π survey (Chambers et al. 2016), the Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006), the VISTA Hemisphere Survey (VHS; McMahon et al. 2013), and the Wide-field Infrared Survey Explorer (WISE; Wright et al. 2010). For the VHS data, we use a custom script to stack the available, reduced frames. This script implements SWarp (Bertin et al. 2002) to coadd the frames. We then set a default zero-point of 27.5 AB mag using the flux calibration from the European Southern Observatory (ESO) archive data reductions. For the WISE data, for which there is a choice of aperture, we select photometry values from the 'aperture 2' instrumental aperture (825 for W1, W2, and W3, and 165 for W4); these apertures were chosen as a balance between encompassing the entire host galaxy, allowing for the change in point-spread function (PSF) with wavelength, and avoiding flux from unassociated sources. We then convert all values to the AB magnitude system if needed.

For nine observations, particularly for FRBs in crowded regions or for hosts of large angular size, which are not adequately encompassed by the default apertures, we perform manual photometry. We created a custom script that implements the aperture_photometry module of photutils (Bradley et al. 2021). ²⁴ We determined the aperture and annuli sizes by loading the images into SAOImageDS9 (Joye & Mandel 2003) and adjusting the region sizes to ensure the host and background were modeled accurately. In most cases, the zero-points were taken from the header of the images after verifying the values were consistent with nearby photometric standards, but in cases where no zero-point was provided, we performed PSF photometry on all point sources in the image using DoPhot (Schechter et al. 1993) and compared their instrumental magnitudes to SkyMapper DR2 standard stars (Onken et al. 2019) for southern fields (δ < − 30°) or Pan-STARRS DR2 standard stars (Flewelling et al. 2020) for more northern fields (δ > − 30°). We then converted to the AB magnitude system if needed for consistency with the rest of the data. Finally, we corrected the values for Galactic extinction using the Fitzpatrick & Massa (2007) extinction law. We report the data source, filters, photometry, uncertainties, and references in Appendix C.

To complement the existing archival data and fill out the host SEDs, we observed the fields of 10 FRBs with the Goodman High Throughput Spectrograph on the 4 m Southern Astrophysical Research Telescope (SOAR; PIs Fong, Gordon; Clemens et al. 2004); GMOS on the 8 m Gemini South Telescope (PI Tejos; Gimeno et al. 2016); the Low Resolution Imaging Spectrometer (LRIS) on the 10 m Keck I Telescope (PI Nugent; Oke et al. 1995); the MMT and Magellan Infrared Spectrograph (MMIRS) on the 6.5 m MMT (PI Nugent; McLeod et al. 2012); and, the FOcal Reducer and low dispersion Spectrograph 2 (FORS2; Appenzeller et al. 1998) and High Acuity Wide-field K-band Imager (HAWK-I; Pirard et al. 2004) on the 8 m Very Large Telescope (VLT; PIs Macquart, Shannon). Details of these observations, including the observation dates, filter, and corresponding magnitude, are reported in Table 5 in Appendix C. For six FRBs, we duplicate filters used in archival observations to increase the S/N of the detected host, or perform deeper searches if they were previously undetected.

We reduced the new Gemini, Keck, and MMT imaging data with the POTPyRI ²⁵ pipeline. POTPyRI creates master bias, master dark, and master flat frames (depending on the types of calibrations available). These calibrations are applied to the science frames before alignment and stacking. A World Coordinate System (WCS) is calculated then applied by calibrating to the Gaia DR3 catalog (Gaia Collaboration et al.2016, 2023). In the case where automatic WCS alignment fails (i.e., the rms of the astrometric fit is ≥ 05), we perform manual WCS alignment by interactively matching sources detected in the science image with known counterparts in either the Gaia DR3 (for optical images) or 2MASS (for NIR images) catalogs. Once the rms of the astrometric fit is of the order of 02, we calculate the zero-point and proceed with PSF photometry as described above.

For the SOAR data, we use the photpipe pipeline (Rest et al.2005) for reduction. This pipeline performs bias correction and flat-fielding and aligns the WCS against the Gaia DR3 catalog. The science frames are then sky subtracted, stacked, and regridded to a common pixel scale and field center using SWarp (Bertin et al. 2002). PSF photometry is performed on all point sources in the stacked image using a custom version of DoPhot. Finally, the pipeline calculates the zero-point of the final calibrated science image by comparing the instrumental PSF magnitudes to SkyMapper DR2 and Pan-STARRS DR2 standard stars using the same methods described above. We then applied these zero-points to aperture magnitudes obtained using photutils. The VLT data were reduced and further processed using the procedure detailed in Bhandari et al. (2022a).

We show representative images of all FRB hosts in the sample, including new observations and images from the literature (Heintz et al. 2020; Macquart et al. 2020; Mannings et al. 2021; Tendulkar et al. 2021; Bhandari et al. 2022a, 2022b; Caleb et al. 2023), in Figure 1. In total, we present 29 new photometric measurements from our own imaging and 53 measurements from archival imaging. Finally, we collect 114 published photometric values from the literature and archives for 17 FRBs. These values and their references are listed in Table 5 in Appendix C.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For FRB 20190520B, we aligned the SOAR/Goodman field of view so that the bright, nearby star would not land on the detector. We then reduced the data with photpipe following the procedure described above. However, even with the care taken to avoid excess flux from the neighboring star, the background surrounding the host was still significantly greater than the rest of the field. In order to accurately calculate the photometry of the host, we applied a more complex-background model to our photometry code, ²⁶ assuming a spatially varying 1D background that varies with the x, y, and x ∗ y pixel value from the center of the host galaxy. We then placed a 20 aperture around the host galaxy and all pointlike sources of emission close to FRB 20190520B. The resulting values are within 0.1–0.3 AB magnitudes of the values derived from the original photometry script, but we proceed with the complex-background subtracted values in our modeling as these are more representative of the true brightness of the host.

We obtained long-slit optical spectroscopy for the host galaxies of seven FRBs (20180916B, 20210320C, 20210807D, 20211127I, 20211203C, 20211212A, and 20220105A) and one mask observation for the field of FRB 20190520B (with one of the slits centered on the host galaxy center, used in this work). The observations were taken with Keck/LRIS (PIs Miller, Blanchard; Programs O304, O314), Keck/DEIMOS (PI Blanchard; Programs O287, O300), and SOAR/Goodman (PI Fong; Programs SOAR2021A-010, SOAR2021B-002). We designed the observations to span wavelengths of λ ≈ 5000–10000 Å. We list this new spectroscopy, along with the gratings/grisms, and slit widths for each data set in Table 2.

For our subsequent host galaxy modeling (Section 4), we require an S/N ≳3/Å in the continuum across most of the wavelength range. The slits were oriented to cover the center of the host galaxy and the center of the FRB position to search for possible anomalous emission at the FRB location. We manually inspected all of the spectra and found no additional continuum or line-emission detected at the FRB positions.

For the data taken with Keck, we obtained calibrations including arc lamps (Hg, Ne, Ar, Cd, Zn, Kr, Xe, or a subset of these), flat fields, bias frames, and spectrophotometric standard star spectra taken on the same night as the science data. For the SOAR data, we also obtained flat-field and arc lamp spectra (Hg, Ar, and Ne) on the same night as the science data and at a similar airmass. In order to flux calibrate the SOAR data, we used spectrophotometric standards from the SOAR archive ²⁷ that used the same spectral setup and were observed as close to the date of the science observations as possible.

We reduced these data using the Python Spectroscopic Data Reduction Pipeline (PypeIt; Prochaska et al. 2020a, 2020b). PypeIt performs bias-subtraction, flat-fielding, cosmic-ray masking, and wavelength calibration of the raw frames. After the initial processing to generate a calibrated 2D spectral image, the pipeline extracts 1D spectra using the standard Horne algorithm (Horne 1986). At this step, we manually inspected the quality of the calibrated 1D and 2D spectra; in some cases, we implemented minor changes to default parameters to improve the extraction.

For instance, for SOAR/Goodman spectra, we include the flexure parameter spec_method = boxcar in the wavelength calibration module to account for instrumental flexure. In two cases, the host galaxy was too faint to be detected in the individual frames with the default S/N threshold settings and required us to lower snr_thresh ²⁸ in the object finding module to successfully identify the host trace. Finally, for two spectra, very strong emission lines in some of the host galaxies were misinterpreted as cosmic rays and masked in the extraction. If this was found to be the case during the manual inspection, we reran the pipeline with the profile masking turned off (use_2dmodel_mask = False).

After extraction, we apply relative flux calibration using spectrophotometric standard spectra. If multiple exposures were taken, we reduced each exposure separately and coadded the wavelength-calibrated 1D spectra. The final error spectrum generated by the pipeline is a combination of uncertainties propagated from each calibration step and shot noise in the electron counts. We then perform telluric correction on the co-added spectra using the corresponding atmospheric grids for each observatory site and apply correction for Galactic extinction according to the Fitzpatrick & Massa (2007) extinction law.

To measure a preliminary redshift, we use the XSpecGUI viewer from linetools (Prochaska et al. 2017) included in PypeIt to determine the redshift of the host galaxies. This tool overlays the locations of common lines onto the observed spectrum after specifying a fiducial redshift. We adjust the redshift estimate until major features are matched. In particular, we base our initial redshift determination on the Balmer lines (Hα, Hβ), [O iii]λ4959, 5007 emission features, and Ca ii H&K absorption features when available. These serve as input redshifts for our full galaxy modeling (Section 4).

We supplement our new spectra with those of 13 FRB hosts previously published in the literature. These include FRBs 20180301A, 20200430A, 20200906A (Bhandari et al. 2022b), 20180924B (Bannister et al. 2019), 20181112A (Prochaska et al. 2019), 20190102C, 20190608B (Bhandari et al. 2020a), 20190611B, 20190714A, 20191001A (Heintz et al. 2020), 20201124A (Fong et al. 2021), 20210117A (Bhandari et al. 2022a), and 20210410D (Caleb et al. 2023). These fully reduced data sets are publicly available and accessible via the F4 Github repository (FRBs/FRB; Prochaska et al. 2023). In addition, we retrieved archival Gemini North/GMOS spectroscopy for the host of FRB 20121102A (first published in Tendulkar et al. 2017, PI Tendulkar, Program GN-2016B-DD-2) from the Gemini Observatory Archive ²⁹ and reduced the data with PypeIt as described above. As this host is faint, we reduced the S/N threshold so the host trace would be detectable in the individual frames and turned off the profile masking module.

In total, our sample comprises 22 FRB hosts with usable spectroscopy. Details of these observations and those taken from the literature are presented in Table 2. We use all of these spectra in the modeling (Section 4), with the exception of FRB 20190711A, which has insufficient S/N to be included in the fitting. In Figure 2, we show the spectra of all new observations presented in this work with major lines denoted and organized by chemical species (see Section 4 for more details on the normalization process).

Zoom In Zoom Out Reset image size

We now present the inferred stellar population properties of the 23 FRB host galaxies in our sample as a whole. To calculate population medians, we draw 1000 representative samples of log(M_*/M_⊙), log(Z_*/Z_⊙), t_m, A_V,young, and A_V,old, and construct distributions of log(SFR_{0−100 Myr}) and specific star formation rate (sSFR; star formation rate divided by stellar mass; log(sSFR_{0−100 Myr})) as described above for each host galaxy. We chose 1000 draws as this constitutes a representative sample from which deviations in the median and 68% confidence intervals are negligible with increasing numbers of draws. We then combine the distributions for each parameter and derive the medians and 68% credible intervals for the total population.

We find that the FRB host population has a median t_m ≈ 5.12 Gyr (interquartile range of 3.72–6.30 Gyr) and log(M_*/M_⊙) ≈ 9.86 (interquartile range of 9.46–10.33). The two hosts with the lowest log(M_*/M_⊙) are those of the repeating FRB 20121102A and the nonrepeating FRB 20210117A, with ≈8.1 and 8.6, respectively; both fall in the dwarf galaxy class (≲10⁹ M_⊙; Bullock & Boylan-Kolchin 2017). For comparison, these are 1 order of magnitude less massive than the Large Magellanic Cloud with log(M_*/M_⊙) ≈ 9.43 (van der Marel et al. 2002) and closer to that of the Small Magellanic Cloud with log(M_*/M_⊙) ≈ 8.49 (van der Marel et al. 2009). The median SFR_{0−100 Myr} ≈ 1.3 M_⊙ yr⁻¹ (interquartile range of 0.20–4.0 M_⊙ yr⁻¹), while the median log(sSFR_{0−100 Myr}) ≈ − 9.86 yr⁻¹ (interquartile range of −10.69 to −9.17 yr⁻¹). We present the posterior distributions of a selection of stellar population properties and derived properties for the full host sample in Figure 4. We report these results numerically in Table 4.

Zoom In Zoom Out Reset image size

Table 4. FRB Host Galaxy Stellar Population Property Statistics

Population	log(SFR0−100 Myr)	log(sSFR0−100 Myr)	log(MF/M⊙)	log(M*/M⊙)	log(Z*/Z⊙)	tm	AV, young	AV,old
	(M⊙ yr−1)	(yr−1)				(Gyr)	(mag)	(mag)
Full Sample	${0.11}_{-1.31}^{+0.81}$	$-{9.86}_{-1.16}^{+1.26}$	${10.06}_{-0.67}^{+0.55}$	${9.86}_{-0.68}^{+0.55}$	$-{0.63}_{-0.75}^{+0.54}$	${5.12}_{-2.15}^{+1.97}$	${1.13}_{-0.28}^{+0.28}$	${0.23}_{-0.17}^{+0.46}$
Repeaters	$-{0.56}_{-0.95}^{+1.01}$	$-{10.37}_{-0.77}^{+2.39}$	${9.59}_{-1.11}^{+0.75}$	${9.40}_{-1.11}^{+0.73}$	$-{0.96}_{-0.80}^{+0.89}$	${5.31}_{-1.83}^{+1.83}$	${1.11}_{-0.29}^{+0.29}$	${0.30}_{-0.2}^{+0.37}$
Nonrepeaters	${0.26}_{-1.03}^{+0.84}$	$-{9.77}_{-1.06}^{+0.86}$	${10.23}_{-0.63}^{+0.50}$	${10.01}_{-0.55}^{+0.54}$	$-{0.56}_{-0.54}^{+0.48}$	${5.07}_{-2.30}^{+1.99}$	${1.13}_{-0.27}^{+0.28}$	${0.20}_{-0.14}^{+0.51}$

Note. Median and 68% confidence intervals of the stellar population properties for the full sample, repeaters, and nonrepeaters. log(SFR_0−100Myr) is the logarithm of the integrated 0–100 Myr star formation rate. log(sSFR_{0−100 Myr}) is the logarithm of the specific star formation rate. log(M_F/M_⊙) represents total mass formed. log(M_*/M_⊙) is the stellar mass. log(Z_*/Z_⊙) is the stellar metallicity. t_m is the mass-weighted age. Finally, A_{V, young} and A_V,old are the magnitudes of dust extinction for young and old stars, respectively.

Download table as:  ASCIITypeset image

We next compare the host galaxy properties of repeating and nonrepeating FRBs by presenting their five-number summaries derived from the full posterior distributions as described above. These statistics are represented visually by boxplots in Figure 5. We find that their distributions of mass-weighted ages, stellar metallicities, and sSFR values are similar, spanning nearly the full range available to galaxies (although in the latter property, nonrepeaters span a wider range). We also find that the hosts of nonrepeaters tend to have slightly larger stellar masses (population median log(M_*/M_⊙) ≈ 10.01 versus 9.40 for the hosts of repeaters). To test whether the stellar population properties of repeaters and nonrepeaters could originate from the same underlying distribution, we perform an Anderson–Darling (A-D) two-sample test with a chosen cutoff p-value of 0.05 (95% confidence). We find p-values > 0.2 in all properties, with the exception of stellar mass for which we derive a p-value of 0.060 and stellar metallicity for which we derive a p-value of 0.164. Thus, we do not find any evidence that the stellar population properties of repeaters and nonrepeaters are statistically distinct.

Zoom In Zoom Out Reset image size

We also note a few outliers from this analysis. The host of the nonrepeating FRB 20211127I has an sSFR_{0−100 Myr} ≈ 10^−8.29 yr⁻¹, over 1 order of magnitude higher than the next-greatest FRB in the same redshift range—the highly active FRB 20201124A at 10^−9.73 yr⁻¹. At z = 0.0469, FRB 20211127I is a relatively young (≈4 Gyr), nearby spiral galaxy with an age and redshift among the lowest in the sample. This host additionally shows an elevated H i-to-stellar mass ratio and has a slight asymmetry in H i (Glowacki et al. 2023). The other notable outlier is the host age of FRB 20210807D. This is the oldest and most-massive FRB host (${8.36}_{-1.84}^{+2.25}$ Gyr and log(M_*/M_⊙) ≈ 10.97) and is also the only quiescent galaxy in the sample (see Section 6.1) with an sSFR_{0−100 Myr} ≈ 10^−11.41 yr⁻¹, the lowest in the sample.

Heintz et al. (2020) and Bhandari et al. (2022b) were the first FRB host population studies based on 12 and 16 hosts, respectively. They found FRB hosts span log(M_*/M_⊙) ≈ 8 − 10.8, t_m ≈ 0.06–1.6 Gyr, and SFRs ≈0.05–10 M_⊙ yr⁻¹, essentially much of the parameter space expected for galaxies within the redshift ranges of their samples. Their SFRs were primarily derived from Hα emission line measurements, which trace more recent SF (timescales of 10–30 Myr). These previous studies performed stellar population modeling using the photometry-only code CIGALE (Boquien et al. 2019), employing a parametric delayed-τ SFH. One known difference between parametric and nonparametric SFHs is that nonparametric SFHs allow for older, more-massive galaxies (Leja et al. 2019b), essentially giving the galaxies the freedom to form more mass over a longer time period, and are more physically realistic. In our analysis, the strength of the 4000 Å break in the spectrum also drives the older ages, with more considerable breaks implying older stellar populations. Indeed, our new analysis finds significantly older ages by a factor of ≈5 and slightly larger stellar masses compared to previous studies (see Section 6.3 for further discussion). An additional difference between earlier works and this one is the IMF used (Chabrier IMF in Heintz et al. 2020 and Bhandari et al. 2022b versus Kroupa here). However, the expected differences in mass and SFR attributed to the assumed IMF between these models are very small (Conroy 2013).

One of the main context clues for transients and their host galaxies is how they trace active star formation in galaxies. This is particularly important for FRBs given their potential association with magnetars (CHIME/FRB Collaboration et al. 2020a; Bochenek et al. 2020). To systematically classify the degree of star formation in FRB hosts, we use the mass-doubling number criterion from Tacchella et al. (2022) to classify the hosts into star-forming, transitioning (off the main sequence), or quiescent galaxies. From Equation (2) of Tacchella et al. (2022),

$$\begin{eqnarray}&&{ \mathcal D }(z)=\mathrm{sSFR}(z)\times {t}_{{\rm{H}}}(z)\end{eqnarray} \tag{ 1 }$$

where t_H(z) is the age of the Universe at the redshift of the host galaxy. Following their classification, if ${ \mathcal D }(z)\gt 1/3$, the galaxy is star-forming; if $1/3\lt { \mathcal D }(z)\lt 1/20$, the galaxy is transitioning; and, if ${ \mathcal D }(z)\lt 1/20$, the galaxy is quiescent. To determine the classification, we take the distribution of log(sSFR_{0−100 Myr}) (generated from 1000 representative draws of sSFR_{0−100 Myr} and log(M_*/M_⊙)) and 1000 draws of t_m, ensuring the values come from the same models. We then calculate the mass-doubling number for each of the 1000 models and take the mode to determine the most common classification. We find that the large majority of FRB hosts are star-forming (20 hosts), two are transitioning (FRBs 20180916B and 20210410D), and one is quiescent (FRB 20210807D). Two of the transitioning and/or quiescent hosts are associated with apparent nonrepeating FRBs, although the repeating FRB 20180916B is classified as transitioning as well.

We next compare the SFRs and stellar masses of FRB hosts (specifically, the log(SFR_{0–100 Myr})—log(M_*/M_⊙) phase space) to field galaxies from the COSMOS sample (Laigle et al. 2016), as shown in Figure 6. We emphasize that these background galaxies were similarly modeled using Prospector with a nonparametric continuity SFH in Leja et al. (2020), allowing for a direct comparison to our derived properties for FRB hosts. We divide the FRB hosts and COSMOS galaxies into three redshift bins, spanning z < 0.1, 0.1 < z ≤ 0.3, and 0.3 < z ≤ 0.6, to enable a proper comparison as there is redshift evolution in this phase space. The well-known star-forming main sequence of galaxies (SFMS; e.g., Whitaker et al. 2012; Speagle et al. 2014; Leja et al. 2022) is apparent; in general, galaxies below the SFMS are transitioning off or completely quiescent.

Zoom In Zoom Out Reset image size

We find that the majority of FRB hosts trace the SFMS across all redshifts. ³⁰ This demonstrates that FRB hosts are forming stars at a similar rate to field galaxies at a given stellar mass. The one star-forming outlier is the host of the apparent nonrepeating FRB 20211127I, which lies well above the main sequence and is the most active galaxy in our sample. We otherwise note that the classification of an FRB as a repeater or nonrepeater does not appear to have an effect on the placement of star-forming hosts relative to the SFMS.

In previous works on FRB host galaxies, the majority of the FRB hosts were found to be slightly offset from the SFMS, with smaller star-forming rates than field galaxies of similar stellar mass (e.g., Figure 4 of Bhandari et al. 2022b). As discussed in Section 5.1, we find slightly larger stellar masses and otherwise similar SFRs. However, previous studies used the PRIMUS catalog (Moustakas et al. 2013) for their field galaxy comparison. When compared to that of COSMOS, the SFMS of PRIMUS is higher by 1 order of magnitude, effectively resulting in an upward shift in the background galaxy comparison sample and a relative downward shift in the locations of the FRB hosts. The specific interpretation of why the PRIMUS SFMS is significantly higher is beyond the scope of this work, but is likely due to their inclusion of a bursty SFH and lack of IR information. In this work, we use the analysis of the COSMOS data set from Leja et al. (2020) derived using identical methodology to our FRB hosts, and are thus free of inter-code systematics. By performing a direct comparison for the first time and employing a quantitative criterion for degree of star formation, we find that only a few are formally "off" the SFMS. In other words, the primary difference is the field galaxy modeling, as opposed to any large systematic differences in stellar population properties.

We next examine the phase space of sSFR versus mass-weighted age (Figure 7). This comparison is meaningful because for nonparametric SFHs, both of these parameters are moments of the SFH. Thus, this serves as a proxy for comparing the SFHs of the FRB hosts to those of field galaxies. We find the star-forming FRB hosts lie in the densest regions occupied by the majority of the COSMOS galaxies, and that there is no apparent distinction between the host galaxies of repeaters and nonrepeaters in this phase space. This demonstrates that the SFHs of FRB hosts are also not unique among field galaxies (although we note the presence of outliers such as the host of the very star-forming and relatively young FRB 20211127I); we discuss further in Section 6.1.

Zoom In Zoom Out Reset image size

Complementary to comparisons involving recent star formation in hosts, we can also leverage our derived SFHs. In particular, given the higher apparent activity of repeating FRBs, it is useful to examine if their progenitors might depend on the level of SF activity over time. In Figure 8, we show the SFHs of the 23 hosts in our sample over lookback time (t_lookback), color-coded by SFH type. Specifically, we classify the FRB host SFHs into five types: rising (purple), delayed-τ exponentially declining (teal), τ-linear exponentially declining (green), post-starburst (yellow), and rejuvenating (orange). Rising SFHs, which comprise five galaxies in the sample, are typically associated with dwarf and/or irregular galaxies, and naturally are classified by a consistent rise in SF over time (Papovich et al. 2011). The exponentially declining SFHs are characteristic of typical L_* galaxies. The delayed-τ (10 hosts) and τ-linear (five hosts) exponentially declining SFHs are functional forms commonly invoked for parametric SFH modeling (e.g., Carnall et al. 2019). Galaxies of both SFH types form most of their stars at early times and decrease in their activities to the present day, with the delayed-τ model having an important "delay" in the onset of the peak of star formation (contributing to a "rise-and-fall" shape in the SFH). Post-starburst SFH galaxies have undergone a recent epoch of high star formation followed by a quenching event (e.g., Wild et al. 2009; Suess et al. 2020, 2022), and comprise two of the hosts. There is a singular possibly rejuvenating host galaxy, FRB 20210807D. This type of galaxy SFH is fairly rare and is characterized by a recent increase in SF activity after a period of quiescence (e.g., Zhang 2022).

Zoom In Zoom Out Reset image size

Overall, we find that the population of FRB hosts exhibits a diverse range of SFHs, with the majority of FRB hosts (43%) falling into the delayed-τ exponentially declining class in which the peak of SF occurred in the last ∼0.1–1 Gyr; this class includes potentially unexpected hosts like the dwarf-like FRB 20190520B. Notably, we find that the majority of hosts with clear and prolonged rising SFHs (e.g., increase in SFRs over time) are those of repeating FRBs (FRBs 20121102A, 20180301A, 20190711A, and 20201124A); however, the nonrepeating FRB 20211203C shows this SFH type as well. We also find repeating FRB hosts with delayed-τ and τ-linear SFHs. We find evidence of past starburst activity in two host galaxies, both of which originated nonrepeating FRBs, corresponding to an occurrence rate of 9% (2/23 hosts). This is in stark contrast to the rate of post-starburst galaxies in SDSS (0.2%), although we note that the selection criteria for the SDSS sample may underestimate the total fraction of galaxies having undergone a starburst event (French et al. 2018). The SDSS criteria is also based on Balmer absorption lines, whereas ours is from the SFHs. Within the final ∼Gyr of lookback time, most FRB hosts are rising or at their peak SF activity; some exhibit this behavior even in the final ∼100 Myr. Other than these distinctions, we otherwise find no clear patterns or correlations between the host SFHs of repeaters or nonrepeaters.

Given the low luminosity of the first identified FRB host galaxy (FRB 20121102A; Tendulkar et al. 2017), it is worthwhile to compare the luminosities of repeating and nonrepeating FRBs across redshift. We present the distribution of host optical luminosities (ν L_ν ) and redshift in our sample in Figure 9, divided into repeaters and nonrepeaters. We supplement our sample of 23 FRB hosts with identified hosts of seven FRBs from the literature that did not meet the criteria for inclusion in our sample (a combination of low PATH probabilities were detected past the date cutoff of 2022 January, or have burst energetics below our cutoff) but nonetheless have claimed host identifications and redshifts. These include the CHIME FRB 20190425A (Panther et al. 2023), DSA FRB 20190523A (Ravi et al. 2019), realfast FRB 20190614D (Law et al. 2020), CHIME FRB 20190110C (Ibik et al. 2023), CHIME FRB 20200120E (Kirsten et al. 2022), CHIME FRB 20200223B (Ibik et al. 2023), DSA FRB 20220319D (Ravi et al. 2023), DSA FRB 20220509G (Connor et al. 2023; Sharma et al. 2023), ASKAP/CRAFT FRB 20220610A (Ryder et al. 2022), DSA FRB 20220912A (Ravi et al. 2022b), and DSA FRB 20220914A (Connor et al. 2023; Sharma et al. 2023). We also plot a demarcation at ν L_ν = 10⁹ L_⊙ below which a host can be classified as a dwarf galaxy (Figure 9).

Zoom In Zoom Out Reset image size

First, we find that FRB hosts have a median luminosity of ≈6 × 10⁹ L_⊙, and span a wide range from the faintest at ≈2 × 10⁸ L_⊙ (FRB 20121102A) to the most luminous at ≈3 × 10¹⁰ L_⊙ (FRB 20191001A). Next, we find that the hosts of repeaters extend to lower luminosities than those of nonrepeaters. Moreover, no repeating FRB hosts in our sample have ν L_ν ≳ 10¹⁰ L_⊙ while nine nonrepeating FRB hosts (or 53% of the total nonrepeating host population in our sample) have luminosities in this range. This is consistent with the finding that nonrepeaters also exist in galaxies with larger stellar masses (Section 5.1). We also note that for z ≲ 0.6, repeating and nonrepeating FRBs appear to have similar redshift distributions (Figure 9), although only nonrepeating FRBs have been observed at higher redshifts.

Finally, the only repeating FRB whose host falls into the category of dwarf galaxy is FRB 20120112A, while two additional hosts of repeaters, FRBs 20190520B and 20220912A (Ravi et al. 2023a) sit just above the borderline at ≈1.1 × 10⁹ L_⊙. FRB 20190520B has been described as a dwarf galaxy in the literature based on the J-band color (Niu et al. 2022), but our modeling reveals it to be slightly more massive than a canonical dwarf galaxy. Notably, the dwarf galaxy luminosity space is not only occupied by repeating FRB hosts; indeed, two nonrepeating FRB hosts (FRBs 20210117A and 20220319D) have low luminosities of ≈4 × 10⁸ L_⊙ and can be classified as dwarfs (Bhandari et al. 2022a; Ravi et al. 2023). This is again consistent with their low stellar masses as discussed in Section 5.1.

Numerous lines of evidence support the scenario that at least some fraction of the FRB progenitor population is composed of magnetars: the FRBs' coherent emission (Katz 2014), energetics (Margalit et al. 2020b), durations (Nimmo et al. 2022), rates (CHIME/FRB Collaboration et al. 2021), stochastic cycles of strong activity or "burst storms" (Lanman et al. 2022; Marthi et al. 2022), evidence for strongly magnetized local environments via their RMs (Wang et al. 2022), and occasional detection of luminous PRSs colocated with FRBs (Chatterjee et al. 2017; Marcote et al. 2017; Law et al. 2022; Niu et al. 2022; Ravi et al. 2022). The possibility of magnetars as FRB progenitors was strengthened by the repeating FRB-like emission from a known Galactic magnetar SGR 1935+2154 (CHIME/FRB Collaboration et al. 2020a; Bochenek et al. 2020; Zhang et al. 2023). It is therefore natural to consider whether the observed FRB host galaxy population is consistent with all FRBs originating from a magnetar progenitor (although with present data it cannot be ruled out that there are multiple progenitors altogether).

The best-studied pathway to magnetar formation is through the core-collapse supernovae (CCSNe) of recently formed massive stars. Indeed, a young, massive star channel for the origins of Galactic magnetars is supported by observations of core-collapse supernova remnants (see, e.g., Gaensler 2004; Vink 2008; Zhou et al. 2019, 2020). In addition, isochrone dating of Galactic magnetars shows that they occur in stellar populations with a range of main-sequence turnoff masses from 17–50 M_⊙ (implying lifetimes of 5–12 Myr; Muno et al. 2006; Bibby et al. 2008; Davies et al. 2009; Tendulkar et al. 2012).

However, observations of FRB environments as a whole are difficult to reconcile with a single, young production channel for magnetars that is responsible for all observed FRBs. For instance, FRB 20200120E was pinpointed to a ∼9 Gyr old globular cluster environment in M81 (Bhardwaj et al. 2021; Kirsten et al. 2022), signifying that at least some FRBs can originate from delayed channels that do not rely on recent massive star formation. It is still viable that magnetars could be responsible for events that originate in older stellar populations, via close binary evolution or the accretion-induced collapse of a white dwarf (Moriya 2016; Margalit et al. 2019; Kremer et al.2021), although progenitors not involving magnetars could also be at play, especially for repeaters (i.e., a neutron star interacting with a companion; Ioka & Zhang 2020; Lyutikov et al. 2020; or accreting black hole binaries; Sridhar et al. 2021). It has also been predicted that the mergers of two neutron stars (NSs), which can experience potentially long delay times of several gigayears or more, could produce magnetars that may be indefinitely stable to collapse and eventually produce observable FRBs (Totani 2013; Wang et al.2016; Pan et al. 2022). The recent claim of an association between the binary NS merger GW190425 and an FRB 20190425A would be definitive evidence for another such delayed channel (Moroianu et al. 2023; Panther et al. 2023). These pieces of evidence are consistent with the results of Li & Zhang (2020) who found that the current sample of FRB host environments are consistent with magnetars formed through multiple formation channels. Multiwavelength observations of the local environment were also informative for the repeating FRB 20180916B. Due to its relative proximity (149 Mpc), Tendulkar et al. (2021) found that the FRB was 250 pc from the nearest star-forming region, a location inconsistent with a young magnetar had the progenitor been kicked from this region. Kaur et al. (2022) performed H i mapping of the host and larger-scale environment, finding evidence for a past minor galaxy merger. While they find that the progenitor was likely born from a massive star originating in a burst of star formation triggered by the merger, they still conclude that the progenitor likely traveled from the nearest star-forming knot identified in Tendulkar et al. (2021). Studies of the local environments provide unique constraints on progenitor models, but are limited to the closest FRB hosts (z < 0.05). As the majority of hosts are at redshifts higher than this, we must also rely on the properties gleaned from global host studies.

It is thus instructive to examine the FRB host population and its relation to star formation and SFH. On subgalactic scales, high-resolution UV and NIR imaging of a smaller number of FRB hosts have demonstrated that several FRBs occur in or near the spiral arms of their host galaxies, and thus their locations track active star formation within their galaxies (Chittidi et al. 2021; Mannings et al. 2021). Additional studies based on global properties of FRB hosts have found the majority are near to or slightly below the SFMS, in the band typically occupied by galaxies that are transitioning off the main sequence (Heintz et al. 2020; Bhandari et al. 2022b; Ravi et al. 2022). A number of these also occur outside of the "blue cloud" of a color–magnitude diagram, although none to date have formally been categorized as quiescent.

Here, based on uniformly derived properties and an accurate comparison to like-modeled field galaxies, we find that the majority lie on the SFMS (i.e., they are forming stars at similar rates to field galaxies of the same stellar mass). Furthermore, based on quantitative criteria to classify galaxies by degree of star formation (Tacchella et al. 2022), we find that 87% (20/23) of FRB hosts are star-forming, two are transitioning, and one is quiescent. We also find a wide range of SFRs (integrated over the past 100 Myr), spanning 2 × 10⁻² M_⊙ yr⁻¹ − 36 M_⊙ yr⁻¹. Two of the galaxies that are transitioning and quiescent host apparently nonrepeating FRBs (FRBs 20210410D and 20210807D), although we note that the repeating FRB 20180916B host is transitioning as well. Both hosts additionally have older ages of ≈7–8 Gyr.

The SFH can also inform progenitor models of transients. For example, the declining SFHs of so-called Ca-rich transients indicate that core-collapse SNe progenitors are unlikely (Dong et al. 2022), while the post-starburst nature of the hosts of tidal disruption events have been used to exclude O, B, and massive A type stars as likely progenitors (French et al. 2017). In our sample, we find that only a few have monotonically declining SFHs (which signifies that the large majority of the host stellar mass formed early on in the first few gigayears). Instead, examining the last ∼1 Gyr, most FRB hosts are either rising in SF activity or at their peak, while all four FRB hosts with prolonged rises are repeaters; see Section 6.2 for a further discussion. We also find a few hosts with evidence of past starburst activity ∼100 Myr, commensurate with H i mapping studies, which have shown evidence for minor mergers (Michałowski 2021; Kaur et al. 2022). While a complete analysis to determine probable progenitor rates would require constructing mass build-up histories and assumptions on delay-time distributions, our results indicate that most FRB hosts were fairly active over the last ≲0.1–1 Gyr.

Finally we note that there are three hosts in our sample for which diffuse radio emission has been detected and attributed to star formation (Bhandari et al. 2020a, 2020b; Fong et al. 2021; Ravi et al. 2022b), and in one case, possibly more localized to the burst site (Piro et al. 2021). While these hosts are not distinct in their SFHs, we find that they all have generally larger SFRs from their SEDs, and two have higher extinction values (in particular, A_V,old). While many bursts in our sample lack constraining radio observations, it would be useful to compare if bursts with more recent SF activity from their SFHs also have detectable radio emission.

Taken together, FRB hosts thus originate from galaxies of different levels of star formation activity, but have a clear preference for star-forming galaxies representative of those from the field. Our demographics and SFHs also imply that while the majority of FRB progenitor systems are unlikely to come from highly evolved stellar populations (e.g., white dwarfs; Liebert et al. 1988; Wood 1992) or other transient events with long delay times (such as merging neutron stars and black holes; e.g., Zevin et al. 2022), the existence of at least a couple of known hosts in less-active environments leaves open this possibility. Already, more FRBs are being found in these types of environments. Recent analysis from Sharma et al. (2023) presented the first quiescent, elliptical FRB host galaxy association (FRB 20220509G), which is additionally part of the galaxy cluster Abell 2311. However, the connection between most FRBs and active star formation supports the young magnetar model theory, where magnetars are formed through the core-collapse supernovae of recently formed massive stars in the past 100 Myr, for the majority of FRBs.

As the number of distinct FRB sources now exceeds 600, ³¹ various properties have been proposed to distinguish repeating and nonrepeating FRBs. For instance, repeaters first exhibited both a downward drift in frequency known as the "sad-trombone" effect (Hessels et al. 2019) and high linear polarization and/or no circular polarization (Nimmo et al. 2021). However, as more apparent nonrepeating FRBs were discovered, examples in this class also shared some of these properties (see Petroff et al. 2022; Zhang et al. 2023 for a more detailed history). Recently, based on a sample of 18 repeating and 474 nonrepeating CHIME FRBs, it was proposed that the populations may be distinct in the duration-bandwidth phase space. Specifically, repeating FRBs exhibit longer intrinsic durations (after de-dispersion) and narrower bandwidths (∼100–200 MHz) while apparent nonrepeaters have shorter durations and wider bandwidths (e.g., Figure 5 of Pleunis et al. 2021). Pleunis et al. (2021) suggested this could be due to a propagation effect, a result of beaming (Connor et al. 2020), different types of bursts from the same source, or factors intrinsic to the populations. The CHIME/FRB Collaboration et al. (2023) again performed a similar analysis with a larger sample of 25 repeaters and found a distinction in DM between the two populations (although they additionally found the burst rates between repeaters and nonrepeaters are not distinctly bimodal, implying a portion of the nonrepeaters may be eventually observed to repeat.) Thus, it is still unclear if the observed distinctions between repeaters and nonrepeaters are intrinsic to the objects themselves or not.

Now we have leveraged new surveys that have enabled a consistent stream of well-localized events (both repeating and nonrepeating) and thus robust host associations. From our sample of 23 FRB host galaxies, we find that repeaters and nonrepeaters share largely similar distributions of stellar population properties (e.g., mass-weighted age, stellar mass, stellar metallicity, SFH, and ongoing SF), with no statistically significant differences.

To test the statistical power of our sample, we simulate larger samples of FRBs with one major assumption that the current posterior distributions for repeaters and nonrepeaters are representative of the true distributions. We double and triple the current sample while maintaining the same ratio of repeaters to nonrepeaters (i.e., 12 repeaters and 34 nonrepeaters for the doubled sample). For each stellar population property, we draw random samples from the total posterior distributions of the nonrepeating and repeating FRBs. We then perform an A-D test in which the null hypothesis is that the classes are from the same parent distribution in that stellar population property. A p-value of 0.05 indicates the null hypothesis can be rejected. We then repeat this process 5000 times to obtain a distribution of A-D test results. We choose 5000 tests because this is the point at which the percentage of statistically significant results are not dependent on the number of tests run. We find that with twice (three times) the current sample size, the null hypothesis is rejected for 84% (96%) of the tests in stellar mass, demonstrating that increasing the sample sizes by a modest amount produces statistically distinct distributions in this property. Indeed, even an increase in the sample size by 50% results in a majority of A-D tests with p < 0.05. Performing this same exercise for all other properties, we find that SFR and stellar metallicity could greatly benefit from sample sizes two and three times the current sample, respectively. Meanwhile, all other properties have distributions in which the repeating and nonrepeating hosts remain statistically indistinguishable. We caution that these projections assume that the current posterior distributions of repeaters and nonrepeaters are representative of the true parent distributions. Ultimately, continued follow-up and stellar population modeling of new, robustly associated FRB host galaxies will be required to uncover the true population statistics.

While our current sample is not large enough to reveal statistically significant distinctions (if they do exist), it still shows a few noteworthy distinctions. For example, the repeating FRB host population as a whole tends to exist in hosts with lower stellar masses than those of nonrepeating FRBs. Similarly, only nonrepeating FRBs have been found in galaxies with optical luminosities ν L_ν ≳ 10¹⁰ L_⊙. Additionally, two of the FRBs in the environments with the least star formation activity are apparent nonrepeaters (FRBs 20210410D and 20210807D). Finally, the majority of galaxies with clear and prolonged rises in their SFHs host repeating FRBs (FRBs 20121102A, 20180301A, 20190711A, and 20201124A), indicating a heightened level of recent SF activity. While expected for low-mass galaxies like the hosts of FRBs 20121102A and 20190711A (Papovich et al. 2011; Leitner 2012), this is particularly surprising for more-massive galaxies like the hosts of FRBs 20201124A and possibly 20180301A. It is plausible that such distinctions in SFH could be enhanced in the context of a larger sample. Overall, with a sample of six repeaters and 17 nonrepeaters, our findings are consistent with two possibilities: the progenitor is the same for both populations, or they have distinct progenitors but do not strongly select on any single galaxy property given the still small number of FRB hosts.

Finally, among the highly secure FRB host sample, we do not find significant differences in the redshift distribution of repeating and nonrepeating FRBs. However, when including bursts detected after our date cutoff or with less secure host associations, there are three nonrepeating FRB host galaxies identified at z ≳ 0.6 while no repeating FRB hosts in this regime are known. Actual interpretation is muddied by selection biases based on different discovery experiments with varied observational biases and inconsistent criteria for host galaxy follow-up. However, one might expect that repeating FRBs are easier to identify at low-z, as several bursts (presumably some with lower luminosities) must be detected, and we cannot rule out that apparent nonrepeaters also have repeat bursts at lower fluences than the detected burst. Moreover, it might be more difficult to discover FRBs in star-forming environments, which could potentially harbor foreground columns of ionized gas associated with the host galaxy. The increased dispersion contributions result in larger detected pulse widths, which would result in a reduced sensitivity to repeating bursts. If these excess columns contain turbulent gas (as would be expected), scatter-broadening of the emission would also reduce search sensitivity to repeating FRBs.

With the advantage of a large sample of uniformly modeled hosts, we briefly investigate if repeating FRBs with unique burst properties correlate to unique host properties. FRB 20121102A has a PRS (Chatterjee et al. 2017; Marcote et al. 2017), large and variable RM (Michilli et al. 2018), high burst rate (Li et al. 2021), and a potential 160 day activity period (Rajwade et al. 2020). Additionally, its RM is decreasing whereas its DM is increasing (Hilmarsson et al. 2021). FRB 20190520B is the second known host to feature a PRS (Niu et al. 2022) and has a high and rapidly varying RM (Anna-Thomas et al. 2023). Both hosts are notable for having lower stellar masses, optical luminosities, and gas-phase metallicities than the bulk of the population; FRB 20121102A also has a rise in SFH within the last ∼0.1 Gyr. As a counter-example, Bhandari et al. (2022a) recently found the nonrepeating FRB 20210117A to be in a dwarf galaxy, and indeed this is one of the lowest-mass galaxies in our sample. However, this burst lacks a PRS, and its environment was not found to be highly magnetized. There may be evidence for the sad-trombone effect commonly associated with repeaters, but it was not detected as strongly as in confirmed repeaters. The repeating FRB 20180916B has a quasiperiodic ≈5 day activity window every 16.4 days (CHIME/FRB Collaboration et al. 2020b), and resides in an old (∼8 Gyr) transitioning host with a lesser degree of star formation than most repeaters. However, as discussed above, it is possible the merger had some effect on local star formation at the site of the FRB despite its overall low global SFR and declining SFH. Finally, the highly active repeating FRB 20201124A undergoes sudden high activity rates (Lanman et al. 2022), showing 1863 bursts in 82 hr over 54 days, dramatic RM variations, circular polarization (Kumar et al. 2022; Xu et al. 2022), and has the widest mean burst width of any repeater (Marthi et al. 2022). It also has a rising SFH, but is otherwise unremarkable in terms of its galaxy properties.

At face value, there is no strong connection between outliers in FRB burst properties and outliers in host properties. Furthermore, FRB hosts that are outliers in their stellar population properties, such as FRBs 20210807D and 20211127I in regard to age and SFR, respectively, are unremarkable in their burst properties (R. Shannon 2023, in preparation). The detailed interpretation of the connection between FRB burst properties and host properties may be a fruitful path toward constraining progenitor models.

To fully contextualize the stellar population properties we derive in this work, it is important to compare to previous studies of FRB host galaxies, as this can inform if there are any systematic biases in stellar population modeling and methodologies. While stellar population properties in some form exist for 16 of the hosts in our sample of 23 (e.g., Tendulkar et al. 2017; Bannister et al. 2019; Prochaska et al. 2019; Heintz et al. 2020; Marcote et al. 2020; Fong et al. 2021; Bhandari et al. 2022a; Ravi et al. 2022; Bhandari et al. 2022b; Niu et al. 2022), our study presents uniform modeling, assumes nonparametric SFHs (an assumption that can percolate to systematic offsets in other stellar population properties; e.g., Leja et al. 2019b), and derives full posterior distributions in each property allowing for a realistic estimate of uncertainties, which we make use of in our population distributions. In addition, while previous papers generally derived SFR and metallicity from nebular emission lines, we use the full SED.

For the 14 hosts that were also modeled in these previous compilations, we compare their stellar population properties. For SFR, while those derived from Hα and the SED may track different timescales, we do not find that there are any systematic differences. We derive slightly larger stellar masses by ∼0.1–0.5 dex (an expected byproduct of nonparametric versus parametric SFHs; Leja et al. 2019b). We also find systematically older mass-weighted ages than Heintz et al. (2020) by a factor of ≈5 (previously ranging from 0.06–1.6 Gyr, versus 2.47–8.36 Gyr here). This is not surprising, as nonparametric SFH modeling is known to result in older ages than the parametric SFH assumption (Leja et al. 2019b) as the flexible SFH gives galaxies more time to form mass. Since the nonparametric SFH is a more physically realistic assumption than parametric SFHs (the method employed by the majority of the literature), our ages can be considered more representative of the average age of a star in the galaxy. Furthermore, the placement of the FRB hosts in relation to field galaxies in Figure 7 shows that the ages we derive for the FRB hosts are consistent with the field galaxy population given the hosts' sSFRs. In fact, if the hosts have younger ages by 1 order of magnitude, as found in previous studies, they would occupy a phase space not probed by any galaxies in the COSMOS sample. As FRB hosts appear to track typical galaxies in the Universe in their stellar population properties, significantly younger ages than these galaxies would be highly unusual.

Finally, we emphasize the importance of a direct comparison field galaxy sample for an unambiguous interpretation of results. As discussed in Section 6.1, our finding that most FRB hosts trace the main sequence of star-forming galaxies (as opposed to offset from it) is primarily due to the background field galaxy catalog used. As each SED modeling code uses different frameworks and libraries for modeling the stellar populations, which introduces their own systematic uncertainties, one must somehow account for the systematic uncertainties in the codes' assumptions, which is difficult to quantify. Instead, our work demonstrates the importance of modeling hosts in the same manner as the field galaxy population to perform direct comparisons.