2.1. Improved localisation of FRB 20171020A

The localisation region of FRB 20171020A presented in M18 was based on signal-to-noise ratio (SNR) measurements of the detection in adjacent ASKAP beams. A Bayesian algorithm, with flat priors on Right Ascension and Declination, and a prior on fluence $P(F) \propto F^{-1}$ (i.e., flat in log-space), was used to find posterior probabilities for the FRB arrival direction as per the method in Shannon et al. (Reference Shannon2018). The results were then fitted with an elliptical function to obtain characteristic localisation bounds.

Here we present two updates to this method. Firstly, the FRB fluence distribution is now known to closely follow a Euclidean distribution, $P(F) \propto F^{-5/2}$ (James et al. Reference James, Ekers, Macquart, Bannister and Shannon2019), arguing for a steeper prior on F. However, such a prior is agnostic of DM, which is a proxy for distance, and the result is obtained by only averaging over distance. For FRBs constrained to a certain distance range (e.g., through DM cuts), the distribution of F will approach that of the intrinsic FRB luminosity function, which is flatter—at least until any cutoff due to a maximum FRB energy. Such a result has been found recently by the CHIME/FRB Collaboration et al. (2021). We therefore use a prior on F of $P(F) \propto F^{-2.1}$ , where $-2.1$ is the differential power-law slope of the FRB luminosity function from James et al. (Reference James2022a). Secondly, we present results without the elliptical fit, that is, using the full Bayesian posterior map, as shown in Figure 1. From this map, we define a localisation probability $p(s) = 1-{\rm C.L.}$ , where C.L. is the confidence limit at which the FRB is found (thus an FRB on the $1 \sigma$ contour has $p(s) \approx 0.32$ ).

The effect of our updated prior on p(F) is that intrinsically dimmer FRBs are preferred, making it less likely that FRB 20171020A was observed far from the beam centre. This shifts the localisation region some 20 $^\prime$ to the south-west and emphasises the importance of the choice of prior in FRB localisations. The resulting 68% confidence region is well fit by an ellipse, centred at RA = $22^{\rm h} 14^{\rm m} 32^{\rm s}$ , Dec = $-19^{\circ} 48^{\prime} 54^{\prime\prime}$ (J2000), with semi-major axis $0.28^{\circ}$ oriented $35^{\circ}$ east from north, and semi-minor axis $0.2^{\circ}$ .

Figure 1. Full Bayesian posterior map with 1,2,3-sigma contours overlaid on a DSS2 Red image. The candidate galaxies (out to z < 0.08) are circled in yellow and blue (where blue indicates ESO 601-G036) and numbered indicating the most likely host, in descending order, for FRB 20171020A (refer to Table 1 for additional details). The $2 \sigma$ error ellipse from M18 is shown in orange for comparison.

We therefore update the list of candidate host galaxies from M18 to include all those within the $3 \sigma$ localisation and $z<0.08$ (the maximum redshift considered in M18). Using the NASA/IPAC Extragalactic Database (NED), we find five galaxies matching these criteria. Searching WISE $\times$ SCOSPZ (Bilicki et al. Reference Bilicki2016), which is deeper than NED and limited primarily by SuperCOSMOS completeness to optical magnitudes $R<19.5$ and $B<21$ , there are 32 such galaxies, including five co-located with those listed in NED. Two of these, however, are located at small ‘islands’ of probability almost one degree to the north-east due to fluctuations in the MCMC fitting and are discounted. The remaining 30 candidates are listed in Table 1. While the most likely candidate, ESO 601-G036, is displaced to the $1\,\sigma$ contour, we note that the only other candidate host galaxy discussed by M18, WISEJ221621.59-191829.9 at $z=0.024$ , is now excluded from the 3-sigma localisation region. As this method of localisation is not enough to uniquely identify a candidate galaxy, we proceed to consider other factors.

2.2. Likely redshift of FRB 20171020A

The DM of an FRB is a vital clue to its likely redshift due to the baryonic content of the Universe, as demonstrated by Macquart et al. (Reference Macquart2020). However, there is significant scatter about the ‘Macquart relation’ between DM and z—due to different host galaxy contributions and inhomogeneities in the cosmic web—leading to a broad distribution of DMs for a given redshift, $p({\rm DM}|z)$ . However, $p({\rm DM}|z)$ has a robust lower bound due to the minimum density of voids and the (at least zero) contribution from host galaxies, leading to the reciprocal $p(z|{\rm DM})$ having a robust upper bound. Thus, the scatter in $p(z|{\rm DM})$ is smaller for low DMs. This probability is also a function of the properties of the detecting instrument, the intrinsic FRB population distribution and luminosity function (James et al. Reference James2022a), and the SNR and time width w of the burst.

For FRB 20171020A, we have accurate measurements of its DM (114.1 pc cm $^{-3}$ ) and SNR (19.5) from Shannon et al. (Reference Shannon2018), and a suitable limit on w (<0.58 ms, effectively w = 0; Qiu et al. Reference Qiu2020), but no suitably reliable estimate of the detected beam position. These values do allow us to estimate $p(z|{\rm DM, SNR,} w)$ . The measured SNR is important because bright FRBs are on-average closer than dimmer ones (Shannon et al. Reference Shannon2018), while wider burst widths also require closer proximity to compensate for the increased noise over their duration.

We estimate $p(z|{\rm DM, SNR}, w)$ using the formulation from James et al. (Reference James2022b),

(1) \begin{eqnarray}p(z|{\rm DM, SNR}, w) & \propto & \Phi(z) \frac{d V(z)}{d \Omega dz} p({\rm DM} |z) \\&& \int dB \Omega(B) \frac{dp(s E_{\rm th}(B,w,{\rm DM},z))}{ds}, \nonumber\end{eqnarray}

where $\Phi(z)$ is the FRB source evolution function; V(z) is the comoving volume per solid angle per redshift interval per proper time; $p({\rm DM}|z)$ is the intrinsic probability distribution of dispersion measure as a function of redshift (including cosmological and host components); $\Omega(B)$ is the solid angle viewed at beam sensitivity B; and $p(s E_{\rm th}(B,w,{\rm DM},z))$ reflects the FRB luminosity function, that is, it is the fraction of all FRB bursts from redshift z emitted with energy above $s E_{\rm th}(B,w,{\rm DM},z)$ , where $s \equiv {\rm SNR} / {\rm SNR_{\rm th}}$ , and $\rm SNR_{\rm th}$ is the FRB detection threshold. The energy threshold $E_{\rm th}$ is dependent upon B, w, DM, and z, where w is the measured burst width. The constant of proportionality in Equation (1) is the inverse of the expression integrated over redshift, which we leave out for brevity. We use best-fit parameter values from James et al. (Reference James2022a) and refer readers to James et al. (Reference James2022b) for further details of the definitions of these functions.

Equation (1) is plotted in Figure 2, calculated for the best-fit FRB population parameters of James et al. (Reference James2022a). Since the best-fit slope of the luminosity function is relatively steep, a nearby host galaxy for FRB 20171020A is strongly preferred.

Table 1. Candidate host galaxies for FRB 20171020A. Column (1) indicates their rank, in order of most to least likely host (see labels on Figure 1); (2) WISE $\times$ SCOSPZ designation; (3) spatial localisation probability; (4) photometric (with the exception of the H i redshi_ of galaxy 1, taken from Meyer et al. Reference Meyer2004), corrected for the North-South asymmetry in photometric z derived from ANNz (Bilicki et al. Reference Bilicki2016); (5) probability of that redshift according to the analysis of Section 2.2; (6) number density of galaxies with that redshift per square degree per $0.001$ interval in z in WISE $\times$ SCOSPZ; (7) redshift probability per galaxy, according to Equation (2); (8) r-band magnitude, corrected for North-South inconsistencies in R-band data due to the different passbands of the UKST in the South and POSS-II in the North (Bilicki et al. Reference Bilicki2016); (9) probability according to the PATH analysis of Section 2.3; (10) joint probability used to rank the candidates and normalised to unity over the candidates; and (11) apparent associations from the NASA/IPAC Extragalactic Database (NED).

In order to use p(z) as a Bayesian prior on the probability of a candidate galaxy being the true host, we take p(z) as the probability of an FRB being detected from a given redshift range z to $z +dz$ . p(z) is then the sum of the per galaxy prior probability, $p^\prime(z)$ , over all n(z) galaxies in that redshift range, that is, $p(z) = \sum_i^{n(z)} p^\prime(z)$ . Thus, any calculation of $p^\prime(z)$ should account for the increasing number of galaxies in the greater volume of the Universe probed in the interval $(z,z+dz)$ at larger redshifts.

To account for possible incompleteness in the WISE $\times$ SCOSPZ catalogue, we construct a histogram with linear bins of dz and correct p(z) by the number of galaxies in each bin n(z), giving:

(2) \begin{eqnarray}p^\prime(z) & = & \frac{p(z)}{n(z)}. \end{eqnarray}

Table 1 lists these corrected probabilities $p^\prime(z)$ and the joint probability $p(s) p^\prime(z)$ , which we use for ranking the candidates. We normalise all probabilities to sum to unity, which is equivalent to stating that the probability of the true host galaxy not being listed in the catalogues employed is zero. ESO 601-G036 is found to be the most likely host, with 98% confidence. While other galaxies have spatial probabilities p(s) up to four times higher than ESO 601-G036, the probability of the host galaxy p(z) lying at the redshift of ESO 601-G036 is at least four times higher than that from other galaxies. The greatest driver of the result however is the very low number density n(z) of galaxies at ESO 601-G036’s low redshift on the sky. At 0.68 per square degree, it would be an extraordinary coincidence for such a galaxy to be located in the localisation region purely by chance. We note however that this analysis does not consider the stellar shred ESO 601-G037 as a separate object, and cannot say anything about the relative likelihoods of it and ESO 601-G036 as host objects.

2.3. Comparison with PATH methodology

The Probabilistic Association of Transients to their Hosts (‘PATH’; Aggarwal et al. Reference Aggarwal2021) methodology has recently been formulated and applied to the identification of the host galaxies of FRBs. It uses a Bayesian framework with the specific formulation incorporating priors on the r-band magnitude $m_r$ of the host galaxy, the offset distribution for an FRB from its putative host, and instrumental localisation uncertainties.

Our analysis in Sections 2.1 & 2.2 is equivalent to a PATH analysis. Given that the instrumental localisation uncertainty is so large, the offset distribution of an FRB from its host can be ignored. The standard prior on the r-band magnitude distribution of FRB host galaxies used in PATH is a flat prior on an FRB originating in a galaxy with a given $m_r$ , so that the prior for any individual galaxy is inversely proportional to the spatial number density of such galaxies on the sky. This method is the equivalent to our use of p(z) normalised by the volumetric density of galaxies from the WISE $\times$ SCOSPZ catalogue.

It is interesting to compare the statistical power of our use of a prior on redshift z compared to a prior on magnitude $m_r$ in this case. Taking the corrected r-band magnitudes from the WISE $\times$ SCOSPZ catalogue, we calculate the prior $p(m_r)$ according to the prescription of PATH, that is, the priors are inversely proportional to the number densities from Driver et al. (Reference Driver2016) and list the values in Table 1. Using $p(m_r)$ alone clearly makes ESO 601-G036 the mostly likely host galaxy; however, the association only has a 42% confidence, whereas using p(z) alone yields a 97% confidence in the association. We caution that hosts derived using p(z) should not then be used as input for FRB population analyses since p(z) is itself derived from such an analysis (James et al. Reference James2022b) and would result in a circular argument.

Ideally, $p(m_r)$ and p(z) would be combined into a joint statistic since p(z) alone ignores the fact that larger galaxies are more likely to be FRB hosts (Bhandari et al. Reference Bhandari2022). As they are correlated however—close galaxies are on-average brighter—this would be a non-trivial exercise, which we leave to a future work.

Figure 2. Likelihood of FRB 20171020A coming from redshift z, given its observed DM, SNR, and width, for best-fit FRB population parameters from James et al. (Reference James2022a). The red dashed line shows the redshift of ESO 601-G036, z = 0.00867.

2.4. Limits on repetition

Our increased confidence in the association of FRB 20171020A with ESO 601-G036 makes it the closest apparently non-repeating FRB. An extensive follow-up programme with the Murriyang (Parkes) radio telescope and the Robert C. Byrd Green Bank Telescope (GBT) has failed to find any repetitions from the source, in addition to the $\sim$ 1150 h that ASKAP has spent observing it, albeit mostly with a single antenna (James et al. Reference James2020). Those authors use these non-detections to place a 90% confidence level upper limit on the repetition rate of bursts above $10^{39}$ erg in energy of 0.181–0.057 per day, assuming a differential power-law spectrum $dN(E)/dE \propto E^\gamma$ , with $-0.7 \ge \gamma \ge -1.1$ . However, this was a conservative upper limit calculated assuming a maximum redshift $z_{\rm max}$ of 0.0636.

The effective rate of visible bursts would be expected to scale (in a locally Euclidean Universe) as

(3) \begin{eqnarray}R & \propto & \left( \frac{z}{z_{\rm max}}\right)^{-2 \gamma}\end{eqnarray}

where $z=0.008672$ . Thus our 90% upper limits on the intrinsic rate of bursts will decrease by the same factor to 0.011–0.0007 bursts day $^{-1}$ above $10^{39}$ erg. By contrast, the rate of approximately one burst per hour observed from FRB 20121102A by the Five-hundred-metre Aperture Spherical radio Telescope (FAST; Li et al. Reference Li2021) corresponds to a rate of approximately six per day when accounting for the $\sim$ 25% active duty cycle of that FRB (Cruces et al. Reference Cruces2021). Accordingly, FRB 20121102A is at least 545–8450 times more active than FRB 20171020A. Any model treating all FRBs as being intrinsically repeating sources must therefore account for at least 3–4 orders of magnitude difference in their rates.

3.1. HI observations

Previous single-dish neutral hydrogen (H i) observations from HIPASS (Figure 3) show that the spatial extent of the gas in and around ESO 601-G036 and the nearby stellar shred ESO 601-G037 could easily fit within the Australia Telescope Compact Array (ATCA) field of view at 1.4 GHz with a single pointing. In an attempt to resolve any H i associated with ESO 601-G037 from that of ESO 601-G036, we acquired 80 h of observing time during semester 2019APR, under project code C3288. Preliminary test observations (project code CX417) indicated that the H i component of the system was readily detectable by ATCA. Different array configurations were used to sample various spatial scales of the uv plane, as summarised in Table 2.

Table 2. ATCA observations of ESO 601-G036.

Figure 3. Hi total intensity contours from the HIPASS survey of ESO 601-G036 and neighbouring gas-rich galaxies superimposed on archival optical r-band Pan-STARRS1 (Chambers et al. Reference Chambers2016) (left) and GALEX UV (inset at right) images. The HIPASS 15.5 arcmin beam is shown in the bottom left corner.

The observations used the ATCA Compact Array Broadband Backend (CABB; Wilson et al. Reference Wilson2011), 1 M–0.5 k correlator mode, and two zoom bands. Each zoom band had a bandwidth of 8.5 MHz divided into 17409 channels, for a spectral resolution of 0.5 kHz. Both zoom bands were centred at 1408.25 MHz. This redundancy helped to mitigate the effects of possible hardware issues occurring in either band, such as periodic correlator block dropouts during observations that would stop data recording for a specific frequency range in the affected zoom band.

Each observing run began with 10 min on the primary and bandpass calibrator PKS 1934-638, followed by a loop of alternating 3 min on the phase calibrator PKS 2135-209 and 40 min on the science target ESO 601-G036. Every four loops, PKS 1934-638 was observed for another 5 min before continuing the loop. Any correlator block dropouts were noted and the alternate band was used for processing and imaging.

The raw data were edited, calibrated, and imaged using miriad (Sault, Teuben, & Wright Reference Sault, Teuben, Wright, Shaw, Payne and Hayes1995) version 1.5 and its standard library of tasks. Each night of observations was processed individually, in a uniform manner, that incorporated visual inspection and manual outlier excision. During the imaging stage, the processed data were incorporated together to make each image cube. At this point, the 0.5 kHz channels were combined for a coarser resolution of 18.7 kHz (4 km s $^{-1}$ ). Different weighting schemes and imaging parameters were tested to determine optimal settings for the intended science.

For the most sensitive image cube we used natural weighting and did not include any data from ATCA antenna 6 (i.e., using only baselines $<$ 3 km). The shortest baselines (with uv range $<$ 0.25 k $\lambda$ ) had strong radio frequency interference (RFI) that caused a large scale ripple across the entire field of view and were also not used for the naturally weighted image cube. For a higher spatial resolution cube, we included all the processed data and used a Briggs robustness of 0.5. The parameters for these two cubes are summarised in Table 3. The root mean square (RMS) noise value for both cubes are comparable as the robust weighting incorporates additional baselines and is not as affected by the RFI in the shorter baselines.

Table 3. ATCA H i processing and image cube details.

Figure 4 shows channel maps of the system using a 3-channel average for conciseness. Both weighting schemes are shown as natural weighting captures the diffuse H i and the robust weighting shows more detail of the higher density H i peaks. Moment maps showing the total H i intensity of the system as well as the velocity field are presented in Figure 5. The ATCA cubes have an elongated synthesised beam, due to the array’s east-west configuration, which hindered clear separation of the H i of ESO 601-G036 from ESO 601-G037. Nevertheless, a gaseous tail extending to the south-west of the stellar component of the system is clearly detected for the first time. The diffuse nature of the H i tail is evident as it is barely detectable in the higher spatial resolution image.

A combined spectral profile incorporating all detected H i within the immediate spatial and spectral vicinity of this system was measured both manually and using software tools such as mbspect in miriad and is presented in Figure 6 along with profiles for ESO 601-G036 (combined with ESO 601-G037) and the gaseous tail. Overall, the combined ATCA spectrum is in good agreement with the slightly noisier, yet higher sensitivity HIPASS spectrum.

The spectra for the individual features were measured manually since there is no clear spatial or spectral separation between ESO 601-G036 and its H i tail. As such, these plots represent an estimate of the flux contribution from each feature and no error bars have been included. Table 4 summarises the H i properties measured from the ATCA data.

Figure 4. Channel maps of ATCA H i cubes using a 3-channel average superimposed on a GALEX near-UV image. The cyan contour—at 1.5 mJy beam $^{-1}$ —is from the naturally weighted cube with the gaseous tail clearly visible between 2604–2616 km s $^{-1}$ . Yellow contours—at 1.5, 3 mJy beam $^{-1}$ —show the robust weighting. At least a portion of the H i in the system, around 2568–2592 km s $^{-1}$ , appears to follow the arc-like stellar structure of ESO 601-G037. The two different synthesised beams are shown in their respective colours in the bottom left corner of each panel.

Figure 5. H i moment maps of ESO 601-G036 showing a gaseous tail extending to the south-west of the stellar component. The synthesised beam is shown at lower left in each case. (a) Total intensity H i contours, with natural weighting, superimposed on a GALEX UV image. Contours are at (0.2, 2, 6) $\times$ $10^{20}$ atoms cm $^{-2}$ . (b) Total intensity H i contours, with robust weighting, at (1.9, 5, 14) $\times$ $10^{20}$ atoms cm $^{-2}$ . (c) Velocity field map of the naturally weighted cube.

3.2. Radio continuum observations

Broadband radio continuum observations of ESO 601-G036 spanning 2.1–21.2 GHz were made with ATCA as summarised in Table 2. PKS 1934-638 was observed as the primary flux calibrator and the bandpass calibrator at frequencies below 10 GHz. Observations at 16.7 and 21.2 GHz central frequencies used PKS 1921-293 as the bandpass calibrator. PKS 2155-152 was used as the phase calibrator at all frequencies.

The radio continuum data were also reduced following standard flagging, calibration, and imaging tasks in miriad. To retain sensitivity to the diffuse radio emission without heavily degrading spatial resolution, a Briggs robustness of 0.5 was used at 5.5 and 9.0 GHz, and 0.0 at 2.1, 16.7 and 21.2 GHz in the imaging step.

An overlay of the radio emission detected at each frequency is shown in Figure 7. Extended radio emission was detected for ESO 601-G036 at all frequencies up to 16.7 GHz. No continuum emission was detected at 21.2 GHz. There is a hint of emission within the spatial vicinity of ESO 601-G037; however, since it is brightest at 16.7 GHz and not detected at 9.0 GHz, the peaks are likely from noise in the continuum data.

To measure the flux density across the wide range of observed frequencies and synthesised beam sizes, images were first convolved to match the lowest resolution image at 2.1 GHz. The flux density was then extracted at each frequency using the miriad task imfit over the same aperture defined by the source size at 2.1 GHz (a Gaussian of 37.8 $\times$ 19.7 arcsec and position angle of $-18.3^{\circ}$ ). A summary of the measured flux densities along with the RMS and native resolution of each image are given in Table 5. A spectral index of $\alpha=-0.9$ is measured by fitting a simple power law to the flux densities using the scipy package (Virtanen et al. Reference Virtanen2020). This steep spectral index, combined with the clearly resolved nature of the continuum emission confirms that the radio emission is dominated by synchrotron radiation from star formation processes, consistent with what was presented in M18.

3.3. Optical imaging and photometry

After initial localisation of FRB 20171020A, as part of a preliminary FRB follow-up strategy, optical imaging observations centred on ESO 601-G036 were taken on 2018 July 15 with Gemini-South (programme GS-2018A-Q-205) using the GMOS HaC filter (6590–6650 Å) and the broadband r-filter (5620–6980 Å). We obtained $3\times 180$ s exposures with the HaC filter, and $1\times 180$ s exposure with the r-filter. The data was processed using the Data Reduction for Astronomy from Gemini Observatory North and South (DRAGONS) packageFootnote a using standard procedures. The resulting images are shown in Figure 8 and show an arc-like structure of ESO 601-G037 that corresponds to the galaxy’s shape revealed in the GALEX image (Figure 3). The HaC imaging enable positioning of the X-shooter spectroscopy slits (see Section 3.4).

The deeper Gemini r-band image was used as the reference image to determine flux-carrying pixels in the Pan-STARRS1 images (Chambers et al. Reference Chambers2016), taking advantage of data availability and self-consistent multi-band photometry provided by the latter dataset. After registering all images—including the Gemini r-band image and the Pan-STARRS1 g, i, z, and y images—to the Pan-STARRS1 r-band image, the detection mask and petrosain elliptical apertures were obtained from the Gemini image. Using the petrosain elliptical aperture, fluxes were derived for ESO 601-G036 from Pan-STARRS1 images and are presented in Table 6. For ESO601-G037, the Gemini detection mask was used to derive fluxes. The different methods were used because ESO 601-G036 has a relatively regular galactic morphology, while ESO601-G037 is a distorted stellar shred. During this process, a standard photometric procedure following Wang et al. (Reference Wang2021) was utilised to conduct background subtraction, source finding, segmentation, and characterisation of photometry and morphological properties. The weight maps of Pan-STARRS1 were used to derive photometric uncertainties. Foreground extinction was corrected with the reddening map of Schlafly & Finkbeiner (Reference Schlafly and Finkbeiner2011). The (g-r) colour-based r-band stellar-mass-to-light ratio from Zibetti, Charlot, & Rix (Reference Zibetti, Charlot and Rix2009) was then used to estimate stellar masses, $M_\star$ , which are also listed in Table 6.

Table 4. H i properties measured from the ATCA observations. (1) Sources considered; (2) central H i velocity; (3) H i line width at 50% of peak flux; (4) H i line width at 20% of peak flux; (5) peak H i flux; (6) integrated H i flux; and (7) H i mass.

Figure 6. ATCA H i spectra of ESO 601-G036 (and ESO 601-G037), the gaseous tail, and all sources combined. For comparison, the HIPASS spectrum—which has higher sensitivity to diffuse emission—has also been included.

Figure 7. Optical r-band image of ESO 601-G036 from Pan-STARRS1 with 2.1 GHz (blue; 70, 90, 110, 130, 150 $\unicode{x03BC}$ Jy), 5.5 GHz (cyan; 30, 40, 50, 60, 70 $\unicode{x03BC}$ Jy), 9.0 GHz (yellow; 20, 25, 30, 35 $\unicode{x03BC}$ Jy) and 16.7 GHz (red; 25, 30, 40 $\unicode{x03BC}$ Jy) continuum contours overlaid. The synthesised beam from each frequency is shown in the bottom left corner. Most observations were carried out using a E-W array leading to an elongated beam in the N-S direction. The 16.7 GHz data was observed with a hybrid array resulting in a different orientation of the beam.

3.4. Optical spectroscopy

M18 used the X-shooter spectrograph (Vernet et al. Reference Vernet2011) mounted on UT2 (Kueyen) of the European Southern Observatory’s Very Large Telescope to effectively filter out other potential host galaxies, by virtue of the fact that only one was bright (and thus potentially close) enough to yield a redshift with just 5–6 min of exposure time. The same instrument was also used that night to obtain spectroscopy at the locations of three H ii region complexes in and around the ESO 601-G036 system, to probe the ionised gas phase kinematics and metallicity. Complex A lies close to the nucleus of ESO 601-G036. Complex B is $\sim$ 10 ${^{\prime\prime}}$ to the south-east along the major axis, near the edge of the optical disc. Complex C corresponds to the bright knots of star formation along the northern edge of ESO 601-G037, closest to ESO 601-G036. These locations are marked in Figure 8b.

Slit widths of 1 $.\!\!^{\prime\prime}$ 0 (UVB) or 0 $.\!\!^{\prime\prime}$ 9 (VIS/NIR arms) and exposure times of 300 s (VIS) or 360 s (UVB/NIR arms) were used, but since diffuse gas emission tended to fill the 11 ${^{\prime\prime}}$ slit length at these locations, matching observations of blank sky regions up to 35 ${^{\prime\prime}}$ away were also obtained. Observations of the white dwarf EG 274 (Hamuy et al. Reference Hamuy1992) and the B9.5 V star HD 123247 were used for relative spectrophotometric calibration and telluric feature removal, respectively.

The echellograms for each X-shooter arm were processed using ESOReflex (Freudling et al. Reference Freudling2013), including debiasing, flatfielding, rectification, as well as wavelength and instrumental response calibration. Although automatic extraction of a point source from the centre of the 2D (slit position, wavelength) image is performed, this is not optimal for our purposes. Instead, the apall task within V2.16.1 of iraf was used to extract a spectrum of consistent spatial width in all arms about the H $\alpha$ peak, separately for each H ii Complex (Table 7). Gaussian fitting with the splot task yielded the centroids and continuum-subtracted integrated fluxes of key emission lines. The fluxes have been corrected for interstellar reddening by assuming the intrinsic Balmer decrement for Case B recombination appropriate to a $T_{\rm e} = 10^4$ K H ii region, that is, I(H $\alpha$ )/I(H $\unicode{x03B2}$ ) = 2.86, and the interstellar extinction curve of Cardelli, Clayton, & Mathis (Reference Cardelli, Clayton and Mathis1989). The corrected fluxes are presented in Table 7, along with the measured reddening parameter $c({\rm H}\unicode{x03B2})$ and mean velocity for each complex.

Table 5. ATCA radio continuum properties of ESO 601-G036. No continuum source is detected at 21.2 GHz so a 3 $\sigma$ upper limit is given.

Figure 8. Gemini optical images of ESO 601-G036 and ESO 601-G037. (a) r-band, scaled to bring out the arc-like morphology of ESO 601-G037. (b) HaC, scaled to show the H ii region complexes. X-shooter slit positions for ESO 601-G036 (A and B) and ESO 601-G037 (C) are also shown. Each slit is 11 $^{\prime\prime}$ long by 1 $^{\prime\prime}$ wide.

Traditionally, the combined strength of the forbidden oxygen transitions can be used as an oxygen abundance diagnostic, via the $R_{23}$ parameter:

$R_{23}$ = ([O ii] $\lambda\lambda3726,3729$ + [O iii] $\lambda\lambda4959,5007$ )/H $\unicode{x03B2}$

and its relationship to O/H. However $R_{23}$ is known to be sensitive to both the ionisation parameter and to temperature, with the latter resulting in an ambiguity between an ‘upper’ and a ‘lower’ abundance branch (Kewley & Ellison Reference Kewley and Ellison2008). In order to break this ambiguity, the prescription laid out by Poetrodjojo et al. (Reference Poetrodjojo2018) has been followed, which employs the N2O2 diagnostic:

N2O2 = log([N ii] $\lambda6583$ /[O ii] $\lambda\lambda3726,3729$ )

and the criterion that for N2O2 $<-1.2$ (as is the case for all three complexes, Table 7) the abundance falls on the ‘lower’ $R_{23}$ branch. Together with the value of the O32 diagnostic:

O32 = log([O iii] $\lambda5007$ /[O ii] $\lambda\lambda3726,3729$ ),

this initial metallicity estimate can place limits on the ionisation parameter q using Figure 5 of Kobulnicky & Kewley (Reference Kobulnicky and Kewley2004). The oxygen abundance 12+log(O/H) is then given by the lower branch relation from Equation (16) of Kobulnicky & Kewley (Reference Kobulnicky and Kewley2004), with the uncertainty shown in Table 7 dominated by the range in q. These abundances are found to be consistent with the M18 estimations.

3.5. X-ray observations

We observed ESO 601-G036 with the Neil Gehrels Swift observatory (Swift, Gehrels et al. Reference Gehrels2004) on 2021 May 19, for 1.7 ks (ObsID 00014317001). The observation was performed with the X-ray Telescope (XRT, Burrows et al. Reference Burrows2005) in photon-counting mode and the Ultraviolet/Optical Telescope (UVOT, Roming et al. Reference Roming2005) with the UVW2 filter. The XRT data was reduced using the xrtpipeline v 0.13.6, (HEASoft 6.29; Blackburn et al. Reference Blackburn, Shaw, Payne and Hayes1999; NASA High Energy Astrophysics Science Archive Research Center (Heasarc) 2014).

We do not detect any significant X-ray emission at the location of ESO 601-G036. Using Poisson statistics for low-count experiments (e.g., Gehrels Reference Gehrels1986; Kraft, Burrows, & Nousek Reference Kraft, Burrows and Nousek1991), we calculate a 3 $\sigma$ upper limit of $<$ $6.8 \times 10^{-3}$ ct s^–1 in the 0.3–10 keV band. Assuming a standard X-ray power-law spectrum ( $F\propto E^{-\alpha}$ ) with a photon index $\alpha = 2$ and the Galactic contribution to the hydrogen column density in the direction of the FRB, $N_{H,MW}=1.9 \times 10^{20}\,\mathrm{cm^{-2}}$ (HI4PI Collaboration et al. 2016), we estimate an upper limit of $F_X \lesssim 2.5 \times 10^{-13}\,\mathrm{erg\ s}^{-1}\,\mathrm{cm}^{-2}$ on the unabsorbed X-ray flux in the 0.3–10.0 keV band. This corresponds to an upper limit of $L_X \lesssim 4.1 \times 10^{40}\,\mathrm{erg\,s}^{-1}$ on the X-ray luminosity. In comparison with the currently cataloged sample of ultra-luminous X-ray sources (ULXs), this limit is less than the observed luminosity of a small fraction of the brightest ULXs (Kovlakas et al. Reference Kovlakas2020), which have been proposed as FRB progenitors (Sridhar et al. Reference Sridhar2021; Sridhar & Metzger Reference Sridhar and Metzger2022).

3.6. Spectral energy distribution

Using archival GALEX, Pan-STARRS1, VISTA, and WISE data, we fit the SED of ESO 601-G036 using the ProSpect SED fitting code (Robotham et al. Reference Robotham2020). We follow the implementation of ProSpect outlined in Thorne et al. (Reference Thorne2021). stellar templates, assume a Bruzual & Charlot (Reference Bruzual and Charlot2003) initial mass function, and model the dust attenuation and re-emission using the Charlot & Fall (Reference Charlot and Fall2000) and Dale et al. (Reference Dale2014) models respectively while assuming energy balance. We adopt a skewed Normal star formation history parameterisation and an evolving metallicity history where the metallicity growth is mapped linearly to the stellar mass growth and the final metallicity is modelled as a free parameter. We use the same priors as presented in Table 2 of Thorne et al. (Reference Thorne2021) and include a 10% error floor across all bands to account for offsets between facilities and instruments.

Figure 9 shows the input photometry, best-fitting SED, and resulting star formation and metallicity histories for ESO 601-G036. From the ProSpect fit, we recover a stellar mass of $\log_{10}\!(M_\star / M_\odot) = 8.64^{+0.03}_{-0.15}$ (which is consistent to the mass calculated using the g-r colour method in Section 3.3) and a $\text{SFR} =$ $0.09 \pm 0.01\,{\rm M}_\odot\,\text{yr}^{-1}$ (which is consistent to the previously reported value in M18). These values place ESO 601-G036 on the star-forming main sequence at $z\sim 0$ , derived by Thorne et al. (Reference Thorne2021).

In addition, we fit available GALEX and Pan-STARRS1 photometry for ESO 601-G037 using the same ProSpect implementation as described above. ESO 601-G037 is undetected in the infrared imaging and so we limit our fitting to the available ultraviolet (UV) and optical data. We recover a stellar mass of $\log_{10}\!(M_\star / M_\odot) = 7.82^{+0.04}_{-0.8}$ (which is also comparable to the optical-only estimate) and a $\text{SFR} = 0.013 \pm 0.001\,{\rm M}_\odot\,\text{yr}^{-1}$ . The star formation history for ESO 601-G037 has a very similar shape to that of ESO 601-G036 (indicating constant star formation over the last $\sim$ 5 Gyr), with the former having a lower normalisation due to its lower mass.

Table 6. Stellar properties from Pan-STARRS1 photometry.

Table 7. H ii region emission line properties measured from the X-shooter observations.

Figure 9. The upper panel shows the input photometry (green points) and resulting best-fitting SED (black), with the contribution from unattenuated stars (blue), attenuated stars (red), and dust emission (orange) shown. The lower left panel shows the resulting best-fitting star formation history as the black line while the grey lines show the star formation histories of the rest of the posterior. The lower right panel shows the resulting metallicity history with lines as per the star formation history panel.

4.1. Signs of recent interaction

The ATCA H i observations have revealed a $\sim$ 3 arcmin long gaseous tail extending towards the south-west of ESO 601-G036. This tail is fairly diffuse, contains $\sim$ 10% of the Hi mass of the system, and spans a velocity width of $W_{20} = 48 \pm 2$ km s $^{-1}$ corresponding to the receding side of ESO 601-G036 (Figure 6). As seen with other nearby FRB host galaxies (Michałowski Reference Michałowski2021; Hsu et al. Reference Hsu2023), the spectral profile of ESO 601-G036 is fairly asymmetric, indicating turbulent motion in and around the galaxy (Glowacki et al. 2022). It is likely that a disturbance has caused some gas from the eastern side of the galaxy to move towards the west, ultimately forming the H i tail. Such disturbance could be an interloping galaxy—with a central velocity closer to v $_c$ $\sim$ 2600 km s $^{-1}$ —which may have also been stripped of at least some of its H i, to contribute to the gaseous tail (see Bournaud et al. Reference Bournaud, Duc, Amram, Combes and Gach2004; Lee-Waddell et al. Reference Lee-Waddell2012; Lelli et al. Reference Lelli2015 for other examples). Considering other H i -rich galaxies in the region are located over 1.4 deg ( $\sim$ 900 kpc in projection) from ESO 601-G036 (see Figure 3), the most obvious and possibly only interaction companion would be ESO 601-G037.

ESO 601-G037 has a pronounced arc-like stellar structure (see Figure 8) that resembles a tidal tail (Barnes & Hernquist Reference Barnes and Hernquist1992; Paudel et al. Reference Paudel, Smith, Yoon, Calderón-Castillo and Duc2018). This stellar shred, located to the south of ESO 601-G036, is about an order of magnitude less massive but has a comparable specific SFR. The optical redshift obtained for ESO 601-G037 (Table 7, H ii complex C) confirms it to be part of the ESO 601-G036 system, since it is within the velocity range observed for ESO 601-G036 and there does appear to be some H i tracing the arc-like structure (see Figure 4). ESO 601-G037’s higher optical velocity (2587 km s $^{-1}$ ) is broadly consistent with the H i tail ( $v_c = 2613 \pm 2$ km s $^{-1}$ and $W_{20} = 48 \pm 2$ km s $^{-1}$ ). Furthermore, the oxygen line ratios of complex C are indicative of a less-enriched ISM, and hence a younger stellar population than complexes A and B in the main body of ESO 601-G036, suggesting that star formation in ESO 601-G037 has been delayed relative to ESO 601-G036.

Overall, there are both kinematic and chemical arguments for ESO 601-G037 being a satellite companion that is undergoing a minor merger event with ESO 601-G036, similar to the host system of FRB 20180916B (Kaur et al. Reference Kaur, Kanekar and Prochaska2022). The current spatial offset of ESO 601-G037 from the H i tail (as much as $\sim$ 40 kpc) suggests that there has been at least one close passage affecting the H i in the system and possibly another encounter transforming the stellar morphology of ESO 601-G037 (Karera et al. Reference Karera2022).

Figure 7 shows the radio continuum contours. Due to the elongation of the beam, the extension of the radio emission to the south of the galaxy may not be truly from the interaction between ESO 601-G036 and ESO 601-G037. Nevertheless, even accounting for the beam size, there does appear to be a general offset of the continuum emission from the optical disc of ESO 601-G036, which is also consistent with the assumed interaction scenario.

4.2. Other properties of ESO 601-G036

ESO 601-G036 is one of only a few FRB host galaxies that is detected in the radio continuum. The extended nature of the emission combined with the steep spectral index indicates that the emission is likely driven by star formation in the host galaxy, similar to what is seen is a small number of other FRB host galaxies (FRB 20191001A, Bhandari et al. Reference Bhandari2020b; FRB 20190608B, Bhandari et al. Reference Bhandari2020a) and the host galaxy of the repeating FRB 20201124A (Fong et al. Reference Fong2021; Piro et al. Reference Piro2021; Ravi et al. Reference Ravi2022). We find no evidence for any compact, persistent radio source such as that detected in repeating FRBs FRB 20121102A (Marcote et al. Reference Marcote2017) and FRB 20190520B (Niu et al. Reference Niu2022).

Using the measured spectral index, we calculate a 1.4 GHz radio luminosity for ESO 601-G036 of 1.89 $\times$ 10 $^{20}$ W Hz $^{-1}$ . Following the 1.4 GHz luminosity-to-SFR relation given in Davies et al. (Reference Davies2017), we compute a SFR of 0.18 $\substack{+2.54 \\ -0.17}$ ${\rm M}_\odot\,\text{yr}^{-1}$ . The relatively large uncertainities in the 1.4 GHz-derived SFR are due to the scatter in the 1.4 GHz-SFR relation from Davies et al. (Reference Davies2017) and demonstrates that this method should only be used to provide an order-of-magnitude estimate of the SFR. Nonetheless, it is consistent with the SFR estimated from SED fitting, despite the methods probing different timescales; the 1.4 GHz radio luminosity traces synchrotron emission from core-collapse SNe, whereas the SED fitting captures emission from hot young stars in the UV, right through to dust heating in the infrared by older, low-mass stars.

4.3. Constraining an FRB progenitor model

Minor mergers can cause increased star formation activity (Hernquist & Mihos Reference Hernquist and Mihos1995; Lambas et al. Reference Lambas, Alonso, Mesa and O’Mill2012; Kaviraj Reference Kaviraj2014). ESO 601-G036 is clearly undergoing some form of interaction event, likely caused by the merging of ESO 601-G037, which aligns with the conclusions of other radio studies of FRB host galaxies that show evidence for recent interactions and/or significant spectral asymmetries (Michałowski Reference Michałowski2021; Kaur et al. Reference Kaur, Kanekar and Prochaska2022; Hsu et al. Reference Hsu2023). This finding is also consistent with FRB progenitor models predicting a tight temporal correlation with star formation; that is, emission from young magnetars formed from the collapse of massive stars.

Our limits on repetition from FRB 20171020A suggest that it was formed either from a cataclysmic merger, or is a significantly older object with correspondingly rarer and/or weaker emission. We have found that star-forming activity in ESO 601-G036 and ESO 601-G037 has been ongoing over the past $\sim$ 5 Gyr (lower left panel of Figure 9), which is consistent with both of these scenarios.

Nevertheless, recent H i observations of the host galaxy of the non-repeating FRB 20211127I do not show evidence for any significant asymmetry (Glowacki et al. Reference Glowacki2023). This particular host galaxy is similar in size and morphology to ESO 601-G036, possibly suggesting that the two bursts could have a common origin mechanism despite their host galaxies having quite different interaction histories. A larger sample size of H i -rich FRB host galaxies is required to draw any firm conclusions on the origin of FRBs in such systems.