A Distant Fast Radio Burst Associated with Its Host Galaxy by the Very Large Array

In 2018, the VLA and CHIME/FRB teams began collaborating to use the VLA for follow-up of repeating FRBs found by CHIME/FRB. We have carried out two approved projects: VLA/18B-405 and VLA/19A-331. We targeted FRB 180916.J0158+65 and FRB 190303.J1353+48 for 40 hr scheduled under VLA/18B-405 and FRB 180814 for 39 hr scheduled under VLA/19A-331; this paper focuses on the second project.

We observed using the L-band system of the VLA, spanning 1–2 GHz , in 20 separate observations. We observed a field centered at (R.A., decl.) [J2000] = (04^h22^m22^s, +73^d40^m00^s), the approximate position of FRB 180814. The nominal field of view of the VLA antennas at L-band is $\sim 30^{\prime}$ (FWHM at 1.4 GHz), but the realfast system is configured to image a field two times wider than that. The first seven observations were performed in 2018 December, in the C-configuration of the VLA, with maximum baselines $\sim 3$ km and a resolution of $\sim 14^{\prime\prime}$ at 1.4 GHz. Thirteen later observations were performed in February through July of 2019, in the B- or BnA-configurations of the VLA, with maximum baselines $\sim 11$ km in length and a resolution of $\sim 4\buildrel{\prime\prime}\over{.} 5$ at 1.4 GHz. Each observation had an on-source time of around 1.5 hr that was searched by the realfast system. The detection reported here is the strongest FRB-like event found in this campaign and is the focus of the analysis presented.

The observations used a commensal correlator mode that generated visibilities with an integration time of 5 ms to be searched by realfast. The same data were also used to generate and save the standard visibility data product to the NRAO archive with a sampling time of 3 s, for all observations in 2019 June and July (nine of them). Prior to that, all visibilities were saved to the archive at their full time resolution, resulting in large data sets (of order 1.5 TB). Both fast and slow visibilities were made in 16 64-channel spectral windows, with each channel set to a width of 1 MHz. Taking typical interference flagging into account, the usable bandwidth is 600 MHz.

The fast-sampled visibilities were distributed to a dedicated GPU cluster using vysmaw (Pokorny et al. 2018) and searched with rfpipe (Law 2017). After applying available online calibrations, the search pipeline dedispersed and integrated visibilities in time before forming images. Calibration solutions were derived from ∼minute-long scans and are stable in time (less than 5° change from the mean value). Images were generated with a simple, custom algorithm that uses natural weighting and a pillbox gridding scheme. The search used 215 DM values from 0 to 1000 $\mathrm{pc}\,{\mathrm{cm}}^{-3}$ and four temporal widths from 5 to 40 ms, which is inclusive of the known properties of FRB 180814 (CHIME/FRB Collaboration et al. 2019).

For the B-configuration observations, each image had 2048 × 2048 pixels with a pixel size of 17, covering a field of view of 1°. The C-configuration images were 512 × 512 pixels with a pixel size of roughly 68. The nominal 1σ sensitivity in a single 5 ms integration is 6 mJy beam⁻¹. All candidates detected with significance greater than 7.5σ trigger the recording of 2–3 s of fast-sampled visibilities and a visualization of the candidate. Each candidate is classified by fetch, a convolutional neural network for radio transients (Agarwal et al. 2020). Finally, realfast team members review the visualizations of the real-time analysis to either remove data corrupted by interference or identify candidates for more refined offline analysis.

On 2019 June 14 (UT), the realfast system detected a candidate transient in the FRB 180814 field. The real-time detection system reported a candidate with image significance of 8.0σ and $\mathrm{DM}=959\,\mathrm{pc}\,{\mathrm{cm}}^{-3}$, far in excess of the expected DM contribution of the Milky Way (83.5 $\mathrm{pc}\,{\mathrm{cm}}^{-3}$; Cordes & Lazio 2002). However, the DM of FRB 180814 is 189.4 pc cm⁻³; no FRB has shown changes in DM of more than a few pc cm⁻³ (Gajjar et al. 2018), so the candidate FRB is likely unrelated to the CHIME FRB.

The real-time candidate analysis revealed multiple signatures consistent with an astrophysical source. First, the spectrum (Figure 1, right panel) shows emission over a range of frequencies spanning at least 50 MHz and the image shows a compact source. Most sources of interference tend to have circular polarization, narrow spectral extent, or are spatially incoherent (i.e., radio frequency interference in the near-field of the array). Second, the fetch FRB classification system reported an astrophysical probability of 99.9%. Third, there is a weak prior expectation for blindly detected astrophysical events to be detected where the antenna sensitivity is highest. The candidate was detected roughly 9' away from the pointing center, where the antenna has roughly 80% of its nominal sensitivity; only 10% of the image has this sensitivity or higher.

Zoom In Zoom Out Reset image size

The realfast search system was starting to receive visibilities from the VLA correlator during the burst. This is seen in Figure 1, which shows that the mean of all recorded visibilities during the burst (phased toward the event) is noisier at early times and at higher frequencies. Visibilities for each baseline, polarization, and spectral window (64 channels) are distributed separately such that the fraction of data grows to 100% over a few hundred milliseconds as the system turns on.

Traditional fast transient surveys measure event significance based on a noise estimate that is local in time (e.g., a standard deviation of a time series). Our interferometric search measures significance in a single image, so the noise estimate is made simultaneously. The Appendix describes how the visibility domain search can be thought of as a time-domain search that allows for more accurate noise estimates.

In our initial analysis of the candidate, we confirmed that the event significance was not affected by different flagging algorithms or calibration solutions from a calibrator observation a few minutes after the event. We also confirmed that removing an antenna from the 27-antenna array reduced the detection significance by roughly 5% ($\approx 1/27$ antennas). With confidence in the quality of data, we proceeded to more carefully quantify the event significance.

We used the raw, saved visibilities to rerun the search with a larger image (8192 × 8192 pixels) and finer DM grid. This optimized search improved the detection significance slightly to a signal-to-noise ratio (S/N) of 8.27 at $\mathrm{DM}=959.19$ $\mathrm{pc}\,{\mathrm{cm}}^{-3}$. Using the same refinement procedure on other candidates typically does not reproduce the initial detection. Noise-like events are expected to be sensitive to the image gridding parameters, so we ignore all events that cannot be reproduced in larger images. We use these refined properties for visualizations and all further analysis.

Figure 2 shows the cumulative distribution of event significance for all events seen in this campaign. The FRB search pipeline automatically applies flags for bad calibration, antenna state, missing data, and interference. We visually inspected the 263 candidates detected above 7.5σ in observations of this field and removed those affected by unflagged interference to get a sample of 31 candidates.

Zoom In Zoom Out Reset image size

Figure 2 also shows an independent estimate of the ideal event rate significance distribution for the array and correlator configuration used to find this candidate. The ideal cumulative event rate assumes that each pixel imaged has a brightness that is drawn from a stationary Gaussian distribution. The number of independent pixels searched is ${\left({N}_{\mathrm{pix}}/{O}_{\mathrm{pix}}\right)}^{2}\times {N}_{\mathrm{int}}\ast ({N}_{\mathrm{DM}}/{O}_{\mathrm{DM}})$, where ${N}_{\mathrm{pix}}$ is the width of an image in pixels, ${N}_{\mathrm{int}}$ number of integrations (at all time widths), ${N}_{\mathrm{DM}}$ is the number of DM trials, and ${O}_{\mathrm{pix}/\mathrm{DM}}$ are the oversampling of the synthesized beam and dispersion sensitivity function, respectively. Both images and DMs are oversampled to maintain uniform sensitivity to all locations and DMs. The search we run here uses ${O}_{\mathrm{pix}}=2.5$ and ${O}_{\mathrm{DM}}=3$. In this configuration, we have 8.4 hr of observing time and 5 × 10¹⁴ independent pixels. The candidate S/N of 8.27 corresponds to a false-alarm rate (FAR) of once in 250 hr. The measured and ideal distributions are independent and in rough agreement, which shows that the significance distribution approximately follows a Gaussian distribution and that this candidate is an outlier.

The FRB search pipeline also uses spectral brightness fluctuations to distinguish candidate events from noise (Law et al. 2017; The CHIME/FRB Collaboration et al. 2019). The Kalman detector (B. Zackay 2020, in preparation) is a method to estimate the statistical significance of FRB spectral variations for an assumed noise model and signal smoothness. For a given noise and signal model, we can marginalize the detection statistic over all matched filters, weighted by their prior probability. This prior probability is defined by a random walk with one free parameter, the coherence bandwidth. We calculated the Kalman score on the candidate FRB, using logarithmic spaced options for the smoothing scale, but found no significant change in the total confidence for the candidate FRB (other FRBs do show some improvement; B. Zackay 2020, in preparation). We conclude that the candidate FRB spectrum is consistent with a constant flux density.

The real-time FRB search software makes several assumptions to improve computational efficiency, and as a result images that are used within it are not optimal. To address this, we used the stored raw, dedispersed visibilities to reimage the burst data with a combination of CASA (McMullin et al. 2007) and AIPS (Greisen 2003).¹⁷ Here, we describe a unified calibration and imaging procedure used in both fast and deep imaging. This procedure allows us to quantify the systematic error in the FRB localization.

Prior to reimaging the burst data, we reduced all of the data taken in 2019 June and July for a deep image of the field. Nine data sets during B configuration were included in this analysis. We excluded C configuration data, as it has poorer spatial resolution. We also excluded early B configuration data recorded at the fast sampling rate, as it was computationally expensive to include in the deep imaging analysis.

We started by applying the calibration and flagging tables for each observation, which were provided by the VLA calibration pipeline. For all observations, the flux density scale was set with an observation of the calibrator source 3C 147, and at these frequencies is accurate to 1%–2% (Perley & Butler 2017). Bandpass and delay calibrations were also determined by the 3C 147 observation. Complex gain (amplitude and phase) fluctuations over time were calibrated with observations of the calibrator source J0410+7656 every 30 minutes. We then exported the calibrated visibilities from CASA and imported them into AIPS. After further RFI flagging, we averaged in time (to 9 s) and frequency (to 4 MHz channels) to reduce the computational load for the imaging.

We used faceted imaging in AIPS to image beyond the first null of the antenna primary beam response (11 width). A total of 73 separate fields, each 1024 × 1024 pixels (with 05 pixel size), and 250 CLEAN boxes were used to image and clean the area. After cleaning, the 73 images of the fields were combined together, and that result was used to self-calibrate (Cornwell & Fomalont 1999) the visibilities on a 1 minute timescale. The imaging and self-calibration were then repeated using this self-calibrated data set, on a 9 s timescale—essentially self-calibrating every visibility. A final image was then made, and a primary beam correction performed on it, based on Perley (2016). This is the final deep image used for further analysis. The synthesized beam in this final deep image is 36 × 28 at a position angle of 79° (north through east). The image has a 1σ sensitivity of 3.6 μJy beam${}^{-1}$, consistent with expectations for the total on-source time and flagging.

For the reimaging of the burst data, we first copied the VLA calibration pipeline tables (calibration and flagging) from the full June 14th observation, and ran a modified version of the procedure to reapply these tables. Calibration tables from the three spectral windows (384 MHz of bandwidth) with valid, uncorrupted data were applied. The synthesized beam in this final burst image is 103 × 42 at a position angle of 67°. It is significantly worse than the resolution of the deep image because of the drastically reduced amount of data that went into it.

The deep and fast radio images were exported to CASA format for source detection and modeling. The source detected by the realfast system (using rfpipe) is also detected in the burst image. We fit an ellipse to that source to measure the centroid location, peak flux density, and their 1σ uncertainties (see Table 1). The localization precision is approximately 1/$10\mathrm{th}$ of the synthesized beam diameter, which is typical for sources of this significance observed with the VLA (Becker et al. 1995).

Table 1.  Measured Properties of FRB 20190614D with 1σ Errors

Time (MJD, @2.0 GHz)	58648.05071771
R.A. (J2000)	4h20m1813
Decl. (J2000)	+73d42m243
R.A. (J2000, deg)	65.07552
Decl. (J2000, deg)	73.70674
Centroid ellipse ('', '', °)	08, 04, 67
S/N ratio${}_{\mathrm{image}}$	8.27
DM${}_{\mathrm{obs}}$ ($\mathrm{pc}\,{\mathrm{cm}}^{-3}$)	959.2 ± 5
DM${}_{\mathrm{MW}}$ ($\mathrm{pc}\,{\mathrm{cm}}^{-3}$)	83.5
Peak flux density (mJy)	124 ± 14
Fluence (Jy ms)	0.62 ± 0.07
Deep limit (μJy beam−1)	<11

Note. The centroid ellipse is defined with the major and minor axes and orientation (east of north). Deep limit refers to the flux density limit on 1.4 GHz radio counterparts in a deep image of the FRB field. The Milky Way DM estimate is calculated from Cordes & Lazio (2002).

Download table as:  ASCIITypeset image

We then searched the deep image to determine whether there is persistent radio emission associated with the candidate FRB. We find no such associated persistent radio emission at the location of the candidate FRB, to a 3σ limit of 11 μJy (see Figure 3).

Zoom In Zoom Out Reset image size

We tested the astrometric precision by associating compact radio sources with optical sources in the Pan-STARRS DR2 catalog (Chambers et al. 2016). We ran the aegean source finding package (Hancock et al. 2018) and identified 270 compact radio sources with a flux density greater than 100 μJy ($\gt 25\sigma$). Of these, 102 had optical counterparts within 3'' and $\mathrm{nDetections}=5$. No systematic offset is found between the radio and optical sources; the standard deviation of the radio/optical offsets is 02. We note that given the resolution of the radio image (36 × 28), we expect the astrometric accuracy to be of the order of 01 for these brighter sources (a few percent of the synthesized beamwidth).

The CHIME/FRB system, operating in its commissioning phase, has observed the sky position of FRB 20190614D for a total of 153 hr during the interval from 2018 August 28 to 2019 September 30. The large exposure is due to the circumpolar nature of the source and is split between 88 hr for the upper transit and 65 hr for the lower transit. The average duration of the upper and lower transits is 17 and 13 minutes, respectively, during which the source is within the FWHM region of the synthesized beams at 600 MHz. We searched through all low-significance events that were detected by the CHIME/FRB system in the abovementioned observing time. No significant event or excess event rate was found to be consistent with the location and DM of FRB 20190614D, so there is no evidence for repetition from this FRB.

To determine CHIME/FRB sensitivity to FRB 20190614D, we follow the methods detailed in Josephy et al. (2019). The sensitivity of CHIME/FRB varies with observing epoch, position along transit, and burst spectral shape. We used a Monte Carlo simulation with 10⁶ realizations to generate fluence thresholds for different detection scenarios within the quoted exposure. These simulations define a set of relative sensitivities, which are tied to a flux density scale using beam-formed, bandpass-corrected observations. As a reference, we use a burst from FRB 180814.J0422+73 detected on 2018 November 11 with an S/N ratio of 9.8σ, fluence of 2.3 ± 0.8 Jy ms, and a Gaussian spectral fit with center frequency of 524 MHz and FWHM of 72 MHz. Figure 4 shows the fluence threshold distribution is 90% complete at 3.8 Jy ms. The distribution is valid for the upper, more-sensitive transit; we estimate the lower transit to be approximately a factor of 4 less sensitive (CHIME/FRB Collaboration et al. 2019).

Zoom In Zoom Out Reset image size

We considered the significance of this candidate high enough to trigger observations designed to find an optical counterpart. On UT 2019 July 2, we observed the field surrounding FRB 20190614D with the Gemini Multi-Object Spectrograph (GMOS) on the Gemini-N telescope. We obtained a series of $8\times 300\,{\rm{s}}$ image exposures in the r-band. These data were reduced with standard procedures using the Gemini's pyraf package, and the images were registered using Pan-STARRS DR1 astrometric standards (Flewelling et al. 2016). We performed photometry on these images using DoPhot (Schechter et al. 1993) and calibrated the image using Pan-STARRS r-band calibrators.

On UT 2019 September 25, we obtained a series of 4 × 600 s images with the Low Resolution Imaging Spectrograph (LRIS) on the Keck I telescope in V and I bands. These data were reduced using a custom-built pipeline used for transient searches and based on the photpipe imaging and reduction package (Rest et al. 2005). Following standard procedures, we removed bias and flattened our images using bias and dome flat-field exposures obtained on the same night and in the same instrumental configuration. We registered the images using Pan-STARRS astrometric standards and combined the individual exposures with SWarp (Bertin et al. 2002). We performed point-spread function photometry on the final stacked images with DoPhot and calibrated these data using Pan-STARRS grizy calibrators transformed to VI using the bandpass transformations described in Tonry et al. (2012).

On UT 2019 November 26, we obtained an additional set of $18\times 200\,{\rm{s}}$ z-band images of the FRB field with the Alhambra Faint Object Spectrograph and Camera on the Nordic Optical Telescope (NOT). The images were processed with standard procedures and astrometrically calibrated to the Gaia-DR2 reference frame.

On UT 2020 March 9, we also obtained a set of $4\times 300\,{\rm{s}}$ images (each one coming from $5\times 60\,{\rm{s}}$ co-adds) in the near-infrared J-band using the Near InfraRed Imager and spectrograph (NIRI; Hodapp et al. 2003) on the Gemini-N telescope. The images were reduced with standard procedures using the dragons¹⁸ package and were astrometrically calibrated to the Gaia-DR2 reference frame. A photometric calibration was derived using Two Micron All Sky Survey sources in the image.

Figure 5 shows the VrI images centered on the radio localization of the candidate FRB. All optical images were registered in the Pan-STARRS DR1 astrometric frame, and so the uncertainty in their relative alignment is given by the precision of the original alignment solutions. We estimate a registration precision of $\approx 0\buildrel{\prime\prime}\over{.} 06$ (1σ) for each image.

Zoom In Zoom Out Reset image size

There are two optical sources that are plausibly associated with the radio source. The brighter source is J042017.85+734222.8, referred to as Source A, and approximately 1'' north of that is J042017.86+734224.5, referred to as source B. The 1σ radio localization region overlaps with source B, but the 2σ (90% confidence interval) radio localization region overlaps with source A. We consider both sources to be potentially associated with the event. Final VrIzJ photometry of the candidate FRB hosts was obtained using a 1'' aperture centered at the locations described in Table 2 and corrected for Galactic extinction.

Table 2.  Optical Candidates

		Source A		Source B
Quantity	Unit	Value	Error	Value	Error
R.A. (J2000)	deg	65.07380	0.00005	65.0745	0.0001
Decl. (J2000)	deg	73.70636	0.00005	73.7068	0.0001
V	mag	25.42	0.25	24.58	0.16
r	mag	23.25	0.15	23.94	0.24
I	mag	22.83	0.10	23.74	0.18
z	mag	23.18	0.30	22.53	999.
J	mag	22.56	0.20	23.66	999.
${z}_{\mathrm{phot}}$		0.63	0.12	0.60	0.17
${\mathrm{log}}_{10}\,{M}_{* }$	${M}_{\odot }$	9.6		8.8
u − r	mag	2.1		0.8

Note. This AB photometry has been corrected for Galactic extinction. A 999.9 value for photometric error indicates a 3σ upper limit. Estimates for M* and u − r are based on the photometric redshift and bear great uncertainty.

Download table as:  ASCIITypeset image

With the photometry of the galaxies as inputs, we have used the software package Eazy (Brammer et al. 2008) to estimate photometric redshifts for the two sources closest to FRB 20190614D. We find ${z}_{\mathrm{phot}}=0.63\pm 0.12$ (68% confidence interval) for source A, and ${z}_{\mathrm{phot}}=0.60\pm 0.17$ (68% confidence interval) for source B. Figure 6 shows the redshift posterior distributions for sources A and B and their best-fitting templates. The best-fitting template for source A is a relatively quiescent galaxy with weak emission features whereas source B, which exhibits a bluer color, is best fit with a star-forming template. The spectral energy distribution (SED) templates that were fitted to the data, agree well with the color difference that we observe in the source. Testing multiple sets of SED templates, we consistently find source A to be quiescent, and similar in shape to what is shown above, as well as source B consistently being fit to star-forming, bluer templates. We also note that the u − r rest-frame colors from the CIGALE analysis detailed below, are consistent with the eazy outputs.

Zoom In Zoom Out Reset image size

On UT 2019 September 29, we obtained a series of long-slit spectra ($1^{\prime\prime}$ wide) of sources A and B with LRIS configured to cover wavelengths $\lambda \approx 3200\mbox{--}6800$Å with the blue camera and its 300/5000 grism and $\lambda \approx 6720\mbox{--}9090$Å with the red camera using the 831/8200 grating. These data were reduced and calibrated with the PypeIt software package (Prochaska et al. 2019a). While we detect a very faint trace of continuum emission from source A, there is no obvious emission or absorption feature to establish a spectroscopic redshift. This is consistent with it being an early-type galaxy with low or negligible star formation and correspondingly weak nebular emission. We did not identify any significant flux from source B.

To roughly estimate the stellar mass and rest-frame color of each candidate host galaxy, we performed an SED analysis of the measured photometry (Table 2). This analysis, using the CIGALE software package (Noll et al. 2009), also requires the source redshift; we adopted the posterior-weighted photometric redshift from the eazy analysis (Table 2). For the SEDs constructed by CIGALE, we adopt a delayed-exponential star formation history model with no late burst population, a Chabrier initial mass function, and the Calzetti et al. (2000) dust extinction model. Because of the applied extinction corrections, changes in these assumptions would produce similar results. Consistent with the eazy analysis, the best-fitting SEDs were quiescent for source A and star-forming for source B. In Table 2, we report estimates for the stellar mass and rest-frame u − r color with the latter reflective of the inferred star-forming properties of each galaxy. We caution that the stellar mass, especially, bears great uncertainty due to the uncertain redshifts of each source.

The chance of randomly associating a point on the sky with a galaxy has previously been studied in the context of gamma-ray bursts. The chance association has an empirically defined functional form parameterized by an association between tolerance and survey depth (Bloom et al. 2002).¹⁹ Following the same approach we estimate chance association probabilities of ${P}_{\mathrm{ch},{\rm{A}}}=7.1\times {10}^{-2}$ and ${P}_{\mathrm{ch},{\rm{B}}}=6.9\times {10}^{-2}$ for the two galaxies to be unrelated to FRB 20190614D based on the r-band detections of Galaxies A (r = 23.24 mag) and B (r = 23.93 mag). The maximum "search radius" ${r}_{\mathrm{ch}}$ used for these estimates take into account the half-light radii ${R}_{1/2}$ of the host candidates and the distance to the galaxy centroids (${d}_{\mathrm{gal},{\rm{A}}}=2\buildrel{\prime\prime}\over{.} 22$ and ${d}_{\mathrm{gal},{\rm{B}}}=1\buildrel{\prime\prime}\over{.} 07$) as ${r}_{\mathrm{ch}}=\sqrt{{d}_{\mathrm{gal}}^{2}+4{R}_{1/2}^{2}}$. We also use the background flux as a limit on the presence of a galaxy below the detection limit of $r\gt 25.1$ mag. At this limit, the chance association probability of an unrelated galaxy to be located only within the error region of the FRB position is ${P}_{\mathrm{ch},\mathrm{undet}}=1.08\times {10}^{-2}$. The probability that source A and source B are unrelated to the FRB is therefore small and the expectation for an even fainter host galaxy counterpart within the error region is even smaller.

We also used the methods of Eftekhari & Berger (2017) to calculate the chance association probabilities.²⁰ Using this approach, we estimate the chance coincidence probabilities of ${P}_{\mathrm{ch},{\rm{A}}}=2.6\times {10}^{-2}$ and ${P}_{\mathrm{ch},{\rm{B}}}=2.7\times {10}^{-2}$ for Galaxies A and B, respectively. Although, Eftekhari & Berger (2017) followed a similar procedure to Bloom et al. (2002), the estimates using their methods are smaller because they used a more recent estimate of r-band number counts of galaxies (Driver et al. 2016) to calculate the number density of galaxies above any given limiting magnitude. We use the more conservative and more widely used estimates obtained using the formalism of Bloom et al. (2002) to calculate the significance of the FRB candidate.

Under the assumption that an FRB should reside in a galaxy, we can use the host galaxy association to improve the confidence in the significance of the candidate event. The radio signal alone has been characterized by an S/N of 8.27 and an FAR of $4\times {10}^{-3}\,{\mathrm{hr}}^{-1}$. If we assume that false positives are randomly distributed in the field, then the association of the radio source to a host galaxy improves the confidence in the significance of the FRB candidate as ${\mathrm{FAR}}_{\mathrm{assoc}}=\mathrm{FAR}\cdot {P}_{\mathrm{ch},\det }$. According to this relation, we find ${\mathrm{FAR}}_{\mathrm{assoc}}=3\times {10}^{-4}\,{\mathrm{hr}}^{-1}$.

Given that the association of a false positive with a host galaxy is unlikely, we conclude that the FRB candidate is an astrophysical event. We used the Transient Name Server²¹ to name the event FRB 20190614D. This naming convention is consistent with a new standard developed by several groups in the FRB community. The common convention used prior to this change is suitable as a shorthand and is "FRB 190614."

The identification of a specific FRB host galaxy can be critical for both estimating the likely host DM contribution to the total observed DM, and for identifying trends in FRB host galaxy types and environments, which can in turn help discriminate between FRB origin models. FRB 20190614D is offset from optical counterpart(s), the components of which appear to be galaxies that differ in their mass, color, and type.

Given the total extent of the optical counterparts and the color differences, it is most likely that they are two distinct galaxies. However, the photometric redshifts of the two galaxies are consistent with each other and they are close in projection. Assuming a distance of z = 0.6 (6.8 kpc/''), the galaxy centers are separated by 13.6 kpc in projection. Assuming that the two galaxies are located at the same redshift, they are likely an interacting pair in which source B may be a star-forming dwarf satellite of source A. If instead we do not assume they are interacting, this projected separation can occur by chance in galaxies of this magnitude about 10% of the time. In this case, one galaxy is the foreground object to the other.

While the optical data do not directly indicate which galaxy might be in the foreground, an analysis of dispersion does provide some hints. The net observed DM is a sum of contributions from the Milky Way's interstellar medium and its halo, diffuse contributions from IGM plasma, any intervening galaxies and their related circumgalactic media, any cluster plasma, and any host galaxy or FRB-engine contribution (Simha et al. 2020). Any distant contribution is cosmologically redshifted, causing the rest-frame DM contribution to scale by ${\left(1+z\right)}^{-1}$ (e.g., Yao et al. 2017).

Regarding local contributors to DM, for contribution from the interstellar medium of the Milky Way, we adopt the value of 83.5 $\mathrm{pc}\,{\mathrm{cm}}^{-3}$ predicted by Cordes & Lazio (2002). Generally, contributions from the Milky Way's ionized halo are taken to be 30–80 $\mathrm{pc}\,{\mathrm{cm}}^{-3};$ here we adopt the model of Prochaska & Zheng (2019), which predicts a halo contribution of 64 $\mathrm{pc}\,{\mathrm{cm}}^{-3}$ for this sightline. Given these local contributions, we arrive at a representative extragalactic contribution of ${\mathrm{DM}}_{x}=812\pm 25\,\mathrm{pc}\,{\mathrm{cm}}^{-3}$, which encompasses all nonlocal contributions.

We can use the scaling of DM with redshift (known as the Macquart relation; Macquart et al. 2020) to estimate a maximum possible redshift for our FRB. To do this, we attribute all of ${\mathrm{DM}}_{x}$ to an IGM that is devoid of cluster and galaxy group halos. There are various models that predict the ionization and elemental make-up of the IGM as a function of redshift; most of these provide results in the same range (e.g., Yao et al. 2017; Pol et al. 2019; Prochaska & Zheng 2019), predicting maximum redshifts in the $z=1.1\mbox{--}1.3$ range. These values are well beyond the photometric redshifts we measured for both candidate hosts, implying that there are other significant contributors to DM than the IGM for these sight lines.

Figure 7 shows the expected probability distribution of nonlocal DM components using the model of Prochaska & Zheng (2019). The distribution assumes an FRB located at z = 0.6 and uses a parameterized model for halos from individual galaxies, groups, and clusters.²² In this formulation, the mean and 68% range of this distribution gives $\mathrm{DM}={734}_{-107}^{+42}\,\mathrm{pc}\,{\mathrm{cm}}^{-3}$. The estimated ${\mathrm{DM}}_{x}$ for FRB 20190614D is not consistent with the predicted 68% range at this redshift. Note that the peak ${\mathrm{DM}}_{x}$ probability lies at a higher value than the mean; however, even if we use that as a reference point for indicating potential host contributions, this minimal difference does not change the conclusion or analysis presented below.

Zoom In Zoom Out Reset image size

There are a few uncertainties in the comparison of measured to expected DM. First, this estimate ignores the host DM, both from the galaxy halo and interstellar medium. Any such component would push the predicted distribution to higher DM, making it more consistent with the observed value. However, this correction term is diminished from the rest-frame value by $1+{z}_{{\rm{c}}}$, where ${z}_{{\rm{c}}}$ is the host redshift. For z = 0.6, a rest-frame host contribution would be roughly $57\,\mathrm{pc}\,{\mathrm{cm}}^{-3}$ to make the 68% interval of the prediction consistent with the observed value. Second, the DM estimate for the Milky Way tends to be underestimated because the models do not include all small-scale contributions (e.g., Hα features) of roughly ∼10%–20% of the total DM column. Given this, the predicted DM is marginally consistent with expectations at the best-fit photometric redshift. This tension could be resolved if the FRB host galaxy were more distant than the second, nonhost galaxy of the pair.

While the above constraints do not provide conclusions about which of the two sources the FRB is likely associated with (either galaxy or an interacting pair could provide sufficient host DM contribution), it is worth making the simple note that the FRB's location appears to be closer in projection to source B, which we found to have a likely bluer stellar component and potentially more star formation. Potential links with star-forming regions in galaxies have been noted for several past FRBs, particularly the localized repeating FRBs (e.g., Tendulkar et al. 2017; Prochaska et al. 2019b; Marcote et al. 2020).