Deep Synoptic Array Science: First FRB and Host Galaxy Catalog

DSA-110 is built to discover, localize, and characterize FRBs. The first FRB localization by the instrument demonstrated the techniques used for the current sample (Ravi et al. 2023a). A full description of the DSA-110 interferometer and FRB search system will be presented in V. Ravi et al. (2024, in preparation).

FRBs described here were detected during science commissioning with an array of sixty-three, 4.65 m antennas distributed over a 2.5 km area at the Owens Valley Radio Observatory. ⁵ The FRB search system uses 48 antennas arranged as a linear array to form 256 coherent fan beams spaced by 1', thus spanning a little beyond the primary beamwidth of 3.4° (FWHM). DSA-110 is a transit telescope with motorized control over antenna elevation. During all observations described here, the array was pointed at a decl. of 716.

The digital and software systems search a continual stream of calibrated power beams in real time for FRBs. The FRB search is designed to avoid and remove radio frequency interference (RFI) by pointing away from known RFI sources, automatically suspending the search during RFI storms and data cleaning. The search system is optimally sensitive to temporal widths from 0.26–8.32 ms, and FRB-like dispersion measures (DMs) of 50–1500 pc cm⁻³. For the fixed decl. of the observations, the instrument observed Galactic latitudes from 8°–46°, which corresponds to a Milky Way DM contribution ranging from 133–36 pc cm⁻³ (Cordes & Lazio 2002).

The search system is designed to automatically identify FRBs based on several criteria. The significance of power excess in any beam must be greater than 8.5 and not associated with strong terrestrial interference, and have a DM greater than both 50 pc cm⁻³ and 0.75 times the Galactic DM contribution. For each candidate FRB, the system saves metadata, and channelized voltage data from the entire array (48 search antennas and 15 outrigger antennas). The voltage data can fully reproduce the real-time search data, as well as enable fine localization, high time and frequency resolution, and full polarimetry. Verification of each candidate is ultimately reliant on a successful interferometric localization.

Table 1 and Figure 1 present the first sample of 11 FRBs detected by DSA-110 and robustly associated with counterparts detected by Pan-STARRS1 (Section 2.3). These events were identified during science commissioning from 2022 January to October. Analysis of this sample is presented in Section 3.1. A more detailed analysis of the burst properties, including full polarimetry, is presented in Sherman et al. (2023, 2024) and G. Chen et al. (2024, in preparation). DSA-110 FRB detections, including data to aid in reproducing this analysis, are available on the DSA-110 archive ⁶ and in the FRB software and data repository (Prochaska et al. 2023).

Zoom In Zoom Out Reset image size

Figure 2 shows Pan-STARRS1 (PS1; Chambers et al. 2016) gri imaging of the likely FRB host galaxies. PS1 provides images and catalogs for the entire sky north of a decl. of −30° (the "3pi Steradian Survey"). Multi-epoch image stacks reach a depth of 21–23 mag (5σ) with a median seeing from 13–10 in the g, r, i, z, and y bands.

Zoom In Zoom Out Reset image size

DSA-110 localizes FRBs with a precision better than roughly ±2'' at 90% confidence, which corresponds to 3.8 kpc at a redshift of 0.1. This resolution is similar to the half-light radius of a typical galaxy (e.g., Shen et al. 2003). Thus, in many cases, DSA-110 FRBs can be associated with individual galaxies at the characteristic distance of the FRB population. However, by requiring robust association with PS1, we introduce an optical magnitude limit to the FRB host sample. Selection requires that the host be detected in the r and i bands above the 5σ detection limits of 23.2 and 23.1 mag, respectively.

Table 2 summarizes the properties of the host galaxies used to associate them with the FRBs. We use astropath (Aggarwal et al. 2021) to calculate the probability of association for each FRB with its host galaxy. The Bayesian framework estimates the probability of association for all nearby galaxies from the FRB position and error, as well as galaxy magnitude, location, and size. Galaxies with a probability of association greater than 0.9 are considered robust, given the conservative assumptions, and thus included in this sample.

Table 2. Parameters of DSA-110 FRB Host Galaxies Listed in Order of Discovery

FRB Name	Host Name	Host R.A.	Host Decl.	mr	re	Source	Phost
		(J2000)	(J2000)	(mag)	(arcsec)
20220207C	PSO J310.1977+72.8826	20h40m47.s419	+72°52'5790	16.15 ± 0.01	8.356	P	0.99
20220307B	PSO J350.8747+72.1918	23h23m29.s925	+72°11'3080	20.45 ± 0.05	1.393	P	0.99
20220310F	PSO J134.7211+73.4910	08h58m53.s054	+73°29'2745	21.20 ± 0.08	1.909	P	0.99
20220319D	IRAS 02044+7048	02h08m40.s812	+71°02'0965	13.25 ± 0.01	11.764	P	0.99
20220418A	PSO J219.1065+70.0953	14h36m25.s584	+70°0'4360	22.22 ± 0.20	1.290	L	0.97
20220506D	PSO J318.0447+72.8272	21h12m10.s727	+72°49'3776	19.91 ± 0.16	2.425	P	0.98
20220509G	PSO J282.6748+70.2427	18h50m41.s916	+70°14'3395	16.51 ± 0.01	3.938	L	0.99
20220825A	PSO J311.9820+72.5848	20h47m55.s6	+72°35'065	19.36 ± 0.16	1.935	P	1.0
20220914A	PSO J282.0585+73.3363	18h4m13.s958	+73°20'1070	20.05 ± 0.04	1.278	L	0.97
20220920A	PSO J240.2568+70.9180	16h01m01.s634	+70°55'0505	19.46 ± 0.03	2.116	L	0.99
20221012A	PSO J280.8014+70.5242	18h43m12.s3	+70°31'272	19.67 ± 0.03	3.096	L	1.0

Note. The columns include information used to identify and associate the galaxy to the FRB, including r-band magnitude and effective radius, r_e. P_host is the astropath posterior probability of association of the galaxy to the FRB. The field "source" describes what catalog was used to search for potential counterparts, with "P" for Pan-STARRS1 and "L" for Legacy Survey. P_host refers to the probability that the FRB is associated with this galaxy (see Section 2.3).

Download table as:  ASCIITypeset image

We use the adopted priors in Aggarwal et al. (2021), defined as an exponential FRB angular offset distribution and a probability of association that scales inversely to the number density of galaxies at a given magnitude ("exp" and "inverse," respectively). We deviate from the adopted set of astropath priors by using a nonzero prior on undetected host galaxies, P(U). Seebeck et al. (2021) analyzed the false-positive and false-negative association rates for simulated FRB host galaxies. At the depth of PS1 r-band imaging, a reliable and accurate probability of association is found for P(U) = 0.2. All 11 FRBs are covered by Pan-STARRS1 DR2, but six of the FRBs are covered by the deeper Legacy Survey DR9 (Dey et al. 2019). For FRBs covered by the Legacy Survey, we use P(U) = 0.1.

For each FRB, we build a galaxy catalog from PS1 stack images or Legacy catalogs by selecting resolved sources within 30'' (117 kpc at the median redshift of this sample; see Table 3). For PS1, pointlike sources are removed if they are in the PS1-PSC catalog with a "ps_score" less than 0.83 (Tachibana & Miller 2018). No compactness filter was used for Legacy Survey sources, but all FRB host galaxies are resolved, with R_e > 1''. For PS1, we use the r-band Kron radius to represent the galaxy's size, which is within 20% of the half-light radius under realistic scenarios (Graham & Driver 2005). We find a probability of association much greater than 0.9 for all FRBs presented in Table 2.

Table 3. Measured and Derived Properties of the DSA-110 FRB Host Galaxies with FRBs Listed in Order of Discovery

FRB Name	Redshift	Mr	g − r	log(M*)	SFR100Myr	tm	LPRS
		(mag)	(mag)	(M☉)	(M☉ yr−1)	(Gyr)	(erg s−1 Hz−1)
20220207C	0.043040	$-{20.20}_{-0.00}^{+0.00}$	${0.30}_{-0.00}^{+0.01}$	${9.91}_{-0.02}^{+0.03}$	${1.16}_{-0.08}^{+0.10}$	${6.53}_{-0.72}^{+0.84}$	<2.3 × 1028
20220307B	0.248123	$-{19.97}_{-0.00}^{+0.00}$	${0.85}_{-0.01}^{+0.01}$	${10.04}_{-0.01}^{+0.02}$	${2.71}_{-0.61}^{+0.76}$	${0.65}_{-0.04}^{+0.04}$	<1.0 × 1030
20220310F	0.477958	$-{20.79}_{-0.03}^{+0.03}$	${0.99}_{-0.05}^{+0.06}$	${9.97}_{-0.05}^{+0.05}$	${4.25}_{-1.67}^{+2.45}$	${2.13}_{-0.60}^{+1.06}$	<4.6 × 1030
20220319D	0.011228	$-{20.23}_{-0.01}^{+0.01}$	${0.46}_{-0.02}^{+0.03}$	${10.16}_{-0.06}^{+0.03}$	${0.21}_{-0.05}^{+0.12}$	${6.30}_{-1.82}^{+1.51}$	<1.8 × 1026 a
20220418A	0.622000	$-{20.88}_{-0.01}^{+0.01}$	${0.64}_{-0.03}^{+0.04}$	${10.31}_{-0.05}^{+0.06}$	${29.41}_{-10.70}^{+18.28}$	${4.62}_{-0.98}^{+1.20}$	<8.6 × 1030
20220506D	0.30039	$-{20.63}_{-0.01}^{+0.01}$	${1.50}_{-0.03}^{+0.03}$	${10.29}_{-0.07}^{+0.06}$	${5.08}_{-0.47}^{+0.31}$	${0.07}_{-0.01}^{+0.02}$	<1.5 × 1030
20220509G	0.089400	$-{21.34}_{-0.00}^{+0.00}$	${0.78}_{-0.00}^{+0.00}$	${10.79}_{-0.01}^{+0.01}$	${0.23}_{-0.03}^{+0.04}$	${9.03}_{-0.42}^{+0.46}$	<1.1 × 1029
20220825A	0.241397	$-{19.46}_{-0.02}^{+0.07}$	${0.87}_{-0.16}^{+0.06}$	${9.95}_{-0.15}^{+0.17}$	${1.62}_{-0.31}^{+0.47}$	${3.46}_{-1.02}^{+1.49}$	<9.4 × 1029
20220914A	0.113900	$-{18.31}_{-0.02}^{+0.02}$	${0.52}_{-0.02}^{+0.03}$	${9.48}_{-0.09}^{+0.07}$	${0.72}_{-0.15}^{+0.20}$	${7.50}_{-1.50}^{+1.86}$	<1.8 × 1029
20220920A	0.158239	$-{19.96}_{-0.01}^{+0.02}$	${0.48}_{-0.01}^{+0.02}$	${9.81}_{-0.04}^{+0.08}$	${4.10}_{-1.00}^{+1.09}$	${3.86}_{-0.89}^{+1.56}$	<3.6 × 1029
20221012A	0.284669	$-{20.94}_{-0.01}^{+0.01}$	${1.50}_{-0.01}^{+0.01}$	${10.99}_{-0.02}^{+0.02}$	${0.15}_{-0.03}^{+0.04}$	${6.75}_{-0.41}^{+0.56}$	<1.4 × 1030

Note. t_m refers to the mass-weighted age of the galaxy's stellar population. See the Appendix for model fit plots. Limit on persistent radio counterparts derived from the VLA Sky Survey at 3 GHz, except where noted.

^a Dedicated VLA follow-up at 1.4 and 5 GHz.

Download table as:  ASCIITypeset image

Appendix Table 4 summarizes the photometric and spectroscopic measurements made and used in the modeling of host galaxies. We use catalog photometry from wide-area optical/IR surveys PS1, Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006), and ALLWISE (Cutri et al. 2021) surveys. We also acquired IR photometric data in J and H bands with the Wide Field Infrared Camera (WIRC; Wilson et al. 2003) observing instrument mounted on the Palomar 200 inch Telescope. These observations were acquired on 2022 August 16, and were reduced by a custom pipeline following the standard data reduction pipeline procedures. We largely follow the analysis framework established in Sharma et al. (2023) to characterize the host galaxies of these FRBs (including the removal of a star on the host galaxy of 20220509G). We convert the Vega magnitudes to AB magnitudes and correct for galactic dust extinction.

With a reliable host identified, we observed each galaxy with optical/IR spectroscopic instruments. Detection of line emission allows accurate redshift measurement and line flux estimates to better understand the nature of ongoing star formation and nuclear activity, and to model the ages of the stellar populations in a galaxy. The age of the stellar population of an FRB host galaxy tells us about how star formation creates FRB sources, a crucial test of origin models. Appendix Table 4 summarizes the spectroscopic instruments used for this FRB sample.

We observed with the Low-Resolution Imaging Spectrometer on the Keck I telescope (Keck I/LRIS Oke et al. 1995) on May 26, July 21, and October 17 of 2022. The light was dispersed using a 300/5000 grism to measure a detector-frame wavelength range from 3450–6700 Å. The spectra were reduced with the standard lpipe software (Perley 2019) and calibrated using observations of the standard star BD+28 4211. The spectrum was scaled to match the PS1 photometry to account for slit losses.

We observed with the Double Spectrograph (DBSP; Oke & Gunn 1982) mounted on the Palomar Hale 200 inch telescope on 2022 June 1 and December 12, and 2023 May 28. Data were reduced following procedures described in Ravi et al. (2022).

We measure the spectroscopic redshift and the line fluxes of the host galaxies using the penalized PiXel-Fitting software (pPXF; Cappellari 2017, 2023) by jointly fitting the stellar continuum and nebular emission using the MILES stellar library (Sánchez-Blázquez et al. 2006). The line fluxes are shown in Appendix Table 5. A redshift could be reliably identified for all hosts by the detection of at least three emission lines. For one FRB, 20220509G, we measured the redshift by modeling absorption lines.

Two FRBs have been associated with PRS and another has been associated with a known high-energy source (a magnetar). We searched for new examples of these associations by matching FRB positions to wide-area radio and X-ray catalogs. The focus was on compact counterparts, which are simpler to characterize and associate with pointlike FRB emission. Radio catalogs with coverage of this decl. range include VLASS (Lacy et al. 2020), NVSS (Condon et al. 1998), TGSS (Intema et al. 2017), GB6 (Gregory et al. 1996), WeNSS (Rengelink et al. 1997), VLSSr (Lane et al. 2014), and LoTSS (Shimwell et al. 2019). At high energies, we searched the Chandra Source Catalog 2 ("csc21_snapshot_scs"; Evans et al. 2020) and XMM/Newton 4XMM-DR11 (Webb et al. 2020). No counterpart was found for any of our samples in these catalogs. Radio limits are fairly heterogeneous and range from approximately 1 mJy to tens of millijansky over frequencies from 120 MHz to 5 GHz. X-ray limits. Most FRBs had no coverage in high-energy catalogs.

As the nearest nonrepeating FRB, the environment and counterparts to 20220319D can be probed especially well. We observed 20220319D with the Karl G. Jansky Very Large Array (VLA) under program 22A-490 to search for PRS and repeat bursts. We observed with antennas in the A configuration in the 1.4 and 5 GHz bands. The correlator was configured for continuum observing and a commensal 10 ms transient search with the realfast instrument (Law et al. 2018). We detected no burst or persistent counterpart and the continuum imaging sensitivity was approximately 21 μJy beam⁻¹ in both bands (1σ, dual-polarization, robust weighting).

The radio bursts themselves can be used to constrain both intrinsic properties and propagation effects. Bursts are characterized by DM, intrinsic width and structure, scintillation, scattering, as well as polarimetric properties. For the total-intensity burst properties considered here, we find that the DSA-110 sample is typical of the nonrepeating population identified previously. Figure 3 shows the extragalactic DM compared to burst fluence for DSA-110 and other FRB discoveries. The new sample has a median fluence of ∼5 Jy ms, compared to ∼20 Jy ms for the reference sample of localized FRBs with data available in Prochaska et al. (2023). This reflects the improved sensitivity of DSA-110 relative to other localization instruments (dominated by ASKAP).

Zoom In Zoom Out Reset image size

Figure 3 also shows the luminosity and timescale distribution for Galactic and extragalactic millisecond transients (see Nimmo et al. 2022). The present sample is somewhat shorter and more luminous than the bulk of previous detections. This offset reflects the sensitivity and relatively fast sampling time of the DSA-110 search system. We note that the requirement for a PS1 host biases against very distant hosts and high FRB luminosities. Future FRB samples from DSA-110 with dedicated follow-up optical observing should be expected to show higher burst luminosities.

Of special interest is whether bursts repeat or have a complex substructure within individual bursts. Using the burst classification defined in Pleunis et al. (2021), we characterize eight bursts as broad and three as complex. These terms are synonymous with having single or multiple millisecond-scale components in time, respectively. We note that the relatively fast time resolution and relatively narrow bandwidth of DSA-110 make it easier for DSA-110 to detect and identify bursts as complex, while limiting the ability to see narrow or downward drifting bursts. A more complete discussion of burst classification with the inclusion of polarimetric properties is presented in Sherman et al. (2024), and a detailed analysis of the total-intensity properties will be presented in G. Chen et al. (2024, in preparation).

None of the bursts were observed to repeat on timescales longer than a second. Based on the observing pattern during the science commissioning period, the FRB locations were nominally observed for 150 hr through the 2022 observing year. ⁷ If the three multicomponent bursts were treated as repeat bursts, then the wait times range from 0.5–3 ms. This is consistent with the first peak of the wait time distribution for active repeating FRBs such as 121102 (Li et al. 2021).

We obtain the stellar properties of our set of host galaxies using the stellar population synthesis modeling software Prospector (Johnson et al. 2021). We follow the spectral energy distribution (SED) modeling approach described in Sharma et al. (2023), which includes a discussion of spectral processing, model parameters, and priors. In brief, we simultaneously fit a model to the spectrophotometric data. The redshift is initialized to the value measured with pPXF and given a uniform prior width of 1%. We use a continuity nonparametric star formation history with seven bins and the Chabrier initial mass function. We add an active galactic nucleus (AGN) component to the model whenever data above 2 μm is available. We sample from the posterior using the ensemble sampler emcee (Foreman-Mackey et al. 2013).

Generally, the photometric and flux-calibrated spectroscopic observations are jointly fit when possible. Spectroscopy toward host galaxies of 20220207C, 20220506D, and 20220319D had poor quality, so channels were selected around line emission for inclusion in SED modeling. For 20220418A, we use Legacy Survey data instead of PS1 in our SED modeling, due to the better signal-to-noise ratio (S/N) in Legacy data. The resulting SED fits for all the host galaxies, along with their constrained nonparametric star formation histories, are displayed in Appendix Figures 8–9, and 10–11, respectively. The set of derived host properties is summarized in Table 3 and plotted in Figure 4. Best-fit line fluxes measured by pPXF are shown in Appendix Table 5. We note that the recent star formation history is not well constrained in a few cases. We suspect that this is primarily due to the lack of UV photometry in our SED modeling.

PRS have been robustly associated with two FRBs (Tendulkar et al. 2017; Niu et al. 2022) and have been a valuable driver of models for the physics of FRBs (e.g., Margalit & Metzger 2018). Law et al. (2022) present a phenomenological definition of a PRS as a compact (subparsec), luminous (L_r > 10²⁹ erg s⁻¹ Hz⁻¹) radio source that is spatially coincident with an FRB. Given this definition, they were able to provide a more homogenous analysis of the PRS population and how they may be causally connected to FRBs. Note that this luminosity limit is defined phenomenologically to exclude astrophysical foregrounds and that nondetections do not necessarily rule out the physical object that produces PRS-like emission.

We find no persistent radio counterparts associated with these DSA-110 FRBs. The VLASS 3σ limit is 0.5 mJy (at 3 GHz). Using the PRS luminosity threshold proposed previously, we can exclude the presence of a PRS in FRBs 20220207C and 20220319D. FRBs 20220509G, 20220506D, and 20220914A are slightly above the limit, but still fainter than the two confirmed PRS.

We reviewed recently published FRB localizations to build on the sample of known PRS and upper limits (Caleb et al. 2023; Gordon et al. 2023; Ravi et al. 2023a; Ryder et al. 2023; Driessen et al. 2024). Among new localizations from ASKAP and MeerKAT, we find that nonrepeating FRBs 20211212A, 20211127I, 20210405I, and 20210410D all have significant constraints on PRS counterparts. The FRBs presented in Rajwade et al. (2022) have less robust associations with host galaxies and Chibueze et al. (2022) include a PRS candidate that may be attributed to star formation in its host galaxy, so they are not counted here.

Considering these new PRS constraints from DSA-110, MeerKAT, and ASKAP, we find 15 nonrepeating FRBs with upper limits on PRS counterparts. The measurements toward repeating FRBs remain unchanged, with two detections out of six useful measurements. Repeating the analysis of Law et al. (2022), we lower the 90% upper limit on the fraction of nonrepeating FRBs with a PRS at 0.14, and find an 8% chance that the repeater and nonrepeater PRS fraction are consistent with each other. We conclude that there is weak evidence for repeating FRBs to be associated with PRS. We encourage more radio imaging of localized FRBs to understand whether PRS and repetition have a causal relationship.

Analysis of a population of FRB host galaxies can provide insights into formation and evolution channels. We use the present DSA-110 sample of localized FRBs to first explore the characteristics of the host galaxies. In Section 4.2 below, we focus on the star formation histories as probes of the formation of FRB progenitors via the delay time distribution (DTD).

The basic properties of the host galaxies of the DSA-110 sample, and the FRB sightlines through these galaxies, are not unusual with respect to the existing sample of localized FRBs (e.g., Gordon et al. 2023). The left panel of Figure 5 shows the flux ratios of nebular emission lines observed from the DSA-110 hosts in a Baldwin–Phillips–Terlevich (BPT) diagram (Baldwin et al. 1981). Most measurements and constraints on the line ratios are consistent with nebular emission driven by star formation activity, and we see no compelling evidence for AGN contributions. We note that 20220319D is more consistent with line ratios seen in LINER galaxies, although that line ratio was measured directly on the nucleus of the (well resolved) galaxy (Ravi et al. 2023b). Despite line ratios consistent with ionization by star formation, we found that SED modeling was improved by the inclusion of AGN terms; IR colors are particularly sensitive to the presence of AGN (Leja et al. 2018).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The right panel of Figure 5 compares the published and present sample of FRBs to a model of DM from the intergalactic medium. The median redshift of the DSA-110 sample is 0.24. A typical analysis assumes DM = DM_ISM,MW + DM_halo,MW + DM_IGM + DM_host/(1 + z). This equation explains the DM-redshift relationship ("Macquart relation") when DM_IGM is evaluated based on the mean baryon density over the path length to the FRB redshift, z. There are four DSA-110 FRBs with more than 200 pc cm⁻³ excess DM (20220307B, 20220825A, 20220914A, 20221012A). In the case of 20220914A, this is directly attributed to the intracluster gas associated with the host galaxy (Connor et al. 2023). Contributions from intervening structures (e.g., filaments, groups) for other DSA-110 FRBs will be discussed in more detail in L. Connor et al. (2024, in preparation).

Useful morphological information is only accessible for a handful of DSA-110 hosts. As discussed by Ravi et al. (2023b), the host of 20220319D is a face-on barred spiral and the FRB is significantly offset. The disk-dominated host galaxy of 20220207C is viewed nearly edge-on, and a significant DM_IGM + DM_host/(1 + z) = 170 pc cm⁻³ is observed (assuming DM_halo,MW = 10 pc cm⁻³) despite the low redshift of 0.043. Detailed morphological analysis will be presented in K. Sharma et al. (2024, in preparation).

Important insight can be gained from an analysis of the total stellar mass and recent star formation rates (SFRs) of the FRB host galaxy sample. Figure 4 shows the stellar mass and SFR for both the published and DSA-110 FRB host galaxies in three redshift bins. The SFRs are averaged over the last 100 Myr from a nonparametric star formation history analysis. For reference, we used the largest field galaxy sample modeled using similar techniques (COSMOS; Leja et al. 2020), following Gordon et al. (2023), and the published FRB hosts were also analyzed using similar techniques. The DSA-110 sample spans 2 orders of magnitude in both stellar mass and SFR, and 3 orders of magnitude in specific SFR. This diversity of host properties is consistent with the published FRB sample.

The DSA-110 sample is subject to a selection effect on optical magnitude. For example, using kcorrect (Blanton & Roweis 2007), we estimate the characteristic galaxy that is detectable in PS1 (5σ stack limits). For galaxies with mass-to-light ratios from 0.7–1, we find minimum stellar masses, ${M}_{\odot ,\min }$ of 8 × 10⁸ M_⊙ at z = 0.1, 9 × 10⁹ M_⊙ at z = 0.3, and 4 × 10¹⁰ M_⊙ at z = 0.6. This selection effect is clear in Figure 4. While the minimum stellar mass of a DSA-110 host excludes dwarf galaxies, the present sample is sensitive to the star-forming main sequence and quiescent galaxies in all redshift bins.

Figure 4 shows the distribution of mass and SFR for published and DSA-110 host galaxies. Early analysis of FRB hosts (Heintz et al. 2020; Bhandari et al. 2022) seemed to distinguish FRB hosts from the star-forming main sequence and quiescent galaxy population (Schawinski et al. 2014). However, newer FRB host modeling analysis that uses nonparametric star formation histories shows no such distinction (Leja et al. 2019; Gordon et al. 2023). Figure 4 confirms that FRB host galaxies are associated with star-forming and quiescent hosts, which dominate the field galaxy population by number and stellar mass.

A new result from the DSA-110 FRB host galaxy sample is the identification of massive hosts with low specific SFRs (as parameterized in, e.g., Speagle et al. 2014). Two of the 11 hosts in our sample (20220509G, 20221012A) are massive, low-SFR galaxies with high mass-weighted stellar ages. However, the significance of the quiescent hosts is only meaningful when considered relative to the incidence of star-forming hosts. The fact that the DSA-110 hosts are a magnitude-limited sample means that we can test this significance statistically.

Figure 6 shows the cumulative distribution of the hosts in stellar mass, in comparison with the cumulative distributions of stellar mass and SFR of the background galaxy population. The background galaxy population was obtained from the COSMOS sample modeled with Prospector and Leja et al. (2020) using selections identical to the PS1 magnitude limits for the FRB host sample. Roughly half of the star formation in the Universe is contributed by galaxies that contribute only 15%–30% of its stellar mass (for z < 0.7). That distinction, and how it evolves with redshift, allows us to associate FRB formation with specific environments by comparing mass distributions. If FRB occurrence is tied to either current or the accumulation of past star formation, we expect the stellar mass distribution to trace that of star formation or stellar mass, respectively. Despite the presence of massive, quiescent hosts, the DSA-110 hosts tend to follow the cumulative SFR distribution in both redshift bins. This is consistent with previous FRB host distributions and FRB rate analyses that attribute FRB formation to recent star formation (Bochenek et al. 2021; James et al. 2022; Gordon et al. 2023).

Zoom In Zoom Out Reset image size

More detailed insight can be gained by associating the occurrence of FRBs with specific stellar population ages via a DTD, as described below.

FRBs and other transients are produced by sources that are born during periods of star formation. Galaxies that are actively forming stars are associated with short-lived sources (e.g., core-collapse SNe; Cappellaro et al. 2015), while early-type galaxies are more likely to be associated with long-lived transient progenitors (Zheng & Ramirez-Ruiz 2007). The DTD describes the time between the formation of stars and transient events and has been used to study short GRBs (Zevin et al. 2022), Type Ia supernovae (SNe Ia; Wiseman et al. 2021), and core-collapse SNe (Castrillo et al. 2021).

The DTD is used to statistically infer transient origins. For example, the minimum delay time for core-collapse SNe was initially assumed to be tied to the lifetime of massive stars (t_d ≈ 20–50 Myr). However, DTD analysis now supports theoretical modeling that shows binary evolution introduces a tail of ∼15% late explosions (up to 200 Myr; Sana et al. 2012; Castrillo et al. 2021). Compact object binaries are expected to inspiral as they emit gravitational radiation, producing a DTD power-law slope of β = −1. For comparison, short GRBs have a delay time power-law slope of $\beta =-{1.83}_{-0.39}^{+0.35}$ (from 200 Myr to 8 Gyr; Zevin et al. 2022) and for SNe Ia $\beta =-{1.13}_{-0.06}^{+0.04}$ (assuming ${t}_{\min }=40\,\mathrm{Myr};$ Wiseman et al. 2021).

For the present host galaxy sample, we model the DTD as a power law with a minimum, maximum, and slope for the FRB time relative to the star formation history of each galaxy (Appendix Figures 10 and 11; first described in Sharma et al. 2023). The probability of detecting an FRB is assumed to follow a Poisson distribution, such that for a given host galaxy i and a star formation history posterior sample j, the expected rate of FRBs is ${\dot{n}}_{i}^{j}$ at redshift z_i ^j . FRB repetition seems to violate the assumption that the delay time is calculated for a single, cataclysmic event. However, FRB source models predict activity on relatively short timescales (<10⁶ yr; Margalit et al. 2020; Kremer et al. 2023) or with rapid decay power law (< −2; Yang & Zhang 2021) compared to the typical delay times and slopes. In this case, the chance of discovering an FRB is a reasonable estimate of the maximum of its burst rate (modulated by the potential for obscuration by surrounding material). Therefore, we argue that the discovery date can be treated as a one-off event for DTD purposes.

Figure 7 shows the DTD of the current FRB host galaxy sample compared to that of other transient classes. The minimum and maximum delay time is constrained to be ${110}_{-20}^{+250}$ Myr and ${6.7}_{-4.15}^{+4.56}$ Gyr, respectively, and a slope of $\beta =-{1.93}_{-0.71}^{+1.10}$. The posterior distribution requires some chance of long delay times, which supports the conclusions of Sharma et al. (2023). While the FRB host sample is small, it looks similar to that of the short GRB DTD under our assumption of a power-law DTD.

Zoom In Zoom Out Reset image size

An important caveat to this analysis is that we assume a single, stationary Poisson process for the entire FRB population. In fact, there may be multiple kinds of FRB. For these reasons, the slope of the DTD power law should not be directly compared to that of other DTDs. The minimum and maximum of the distribution are likely robust to these caveats. However, we expect the estimate of ${t}_{\min }$ to be affected by the lack of data constraining recent star formation history for some hosts.

Despite these caveats, the DTD analysis provides useful context to the binary framing of whether FRB hosts trace star formation or stellar mass. We find FRBs can occur with either a short (∼100 Myr) or long (≳2 Gyr) delay from star formation. This could be interpreted as a single FRB population with a power law spanning a range of delay times or multiple populations and an overall rate dominated by short delay times. Some modeling of the larger sample of FRB DMs and redshifts show that the rate evolves with redshift similarly to the mean SFR of the Universe, which is consistent with the association of FRBs to mostly star-forming hosts in the present and previous samples (Lu & Piro 2019; Bochenek et al. 2021; James et al. 2022). Alternatively, analysis of the DM-implied distance distribution of CHIME/FRB discoveries suggests formation via a delayed channel, which is consistent with the wide DTD measured here (Hashimoto et al. 2020; Zhang & Zhang 2022). It should eventually be possible to resolve this tension through deviations of the evolution of the FRB rate from that of star formation, which scales as (1 + z)^2.7 in the Local Universe. Specifically, long delays shift the rate peak to later times and flatten its evolution. This is evident in other transient populations with wide DTDs, such as SNe Ia and short GRBs (Greggio et al. 2008; Zevin et al. 2022).