Milliarcsecond Localisation of the Hyperactive Repeating FRB 20220912A

We present very-long-baseline interferometry (VLBI) observations of the hyperactive repeating FRB 20220912A using the European VLBI Network (EVN) with an EVN-Lite setup. We detected 150 bursts from FRB 20220912A over two observing epochs in October 2022. Combining the data of these bursts allows us to localise FRB 20220912A to a precision of a few milliarcseconds, corresponding to a transverse scale of less than 10 pc at the distance of the source. The precision of this localisation shows that FRB 20220912A lies closer to the centre of its host galaxy than previously found, although still significantly offset from the host galaxy’s nucleus. On arcsecond scales, FRB 20220912A is coincident with a persistent continuum radio source known from archival observations, however, we find no compact persistent emission on milliarcsecond scales. The persistent radio emission is thus likely to be from star-formation in the host galaxy. This is in contrast to some other active FRBs, such as FRB 20121102A and FRB 20190520B.


INTRODUCTION
Fast radio bursts (FRBs) are flashes of coherent radio emission that have durations of microseconds to seconds (for a recent review see Petroff et al. 2022).Some of them are known to repeat (Spitler et al. 2016).While more than 2 000 unique sources of FRBs have been detected to date (CHIME/FRB Collaboration et al. 2023), less than 50 have been localised to a host galaxy 1 .Precise localisations of FRBs are key to understanding their origins and for using them as astrophysical and cosmological probes.While arcsecond precision ★ E-mail: d.m.hewitt@uva.nl 1 The FRB Community Newlsetter (Volume 04, Issue 12) reported 44 host galaxies https://hosting.astro.cornell.edu/research/frb/news/ is normally sufficient to identify a host galaxy robustly (Eftekhari & Berger 2017), sub-arcsecond localisations are key to identifying the exact galactic and stellar neighbourhoods in which FRB sources reside (e.g., Tendulkar et al. 2021).
Magnetars are widely favoured as the engines powering FRBs, given the high burst rate and millisecond timescales associated with FRBs (e.g., Ravi 2019;Li et al. 2021;Nimmo et al. 2021), as well as the detection of exceptionally bright radio bursts from the Galactic magnetar SGR 1935+2154 that were coincident with an X-ray burst (Bochenek et al. 2020;CHIME/FRB Collaboration et al. 2020b).Nonetheless, the diverse properties and environments of FRB sources suggests that a single magnetar progenitor model may be overly simplistic (e.g., Kirsten et al. 2022a).
FRB sources show a wide range of activity rates, from a few hyperactive repeaters to apparent one-off events that constitute 97% of the currently known sources (CHIME/FRB Collaboration et al. 2023).The lack of obvious bi-modality in the burst rates suggests that oneoff sources may be capable of repeating (CHIME/FRB Collaboration et al. 2023), but statistically significant differences in burst properties between repeaters and non-repeaters suggest they may be distinct (Pleunis et al. 2021).Another possibility is that a single source model is capable of producing multiple burst types (Hewitt et al. 2023;Snelders et al. 2023).Both repeaters and (apparent) non-repeaters have been localised to a wide variety of host galaxies, with no clear distinction in galaxy type (Gordon et al. 2023;Bhardwaj et al. 2023).
There is an overall trend towards star-forming galaxies, with notable exceptions (Sharma et al. 2023).
The global galactic properties of an FRB host are only indirectly informative about the source's nature.More directly, we can study the local environment of FRB sources via time-variable propagation effects (e.g., Michilli et al. 2018) and precision localisation coupled to high-resolution imaging (e.g., Mannings et al. 2021).Ideally, radio localisations should have < 100 mas uncertainty, in order to maximize the degree to which one can zoom-in on their local environment.Thus far, only five repeaters, and as yet no non-repeaters, have been localised to milliarcsecond precision (Marcote et al. 2017(Marcote et al. , 2020;;Kirsten et al. 2022a;Nimmo et al. 2022;Bhandari et al. 2023).
The first detected repeater, FRB 20121102A, was localised to a low-metallicity star-forming dwarf galaxy (Chatterjee et al. 2017;Tendulkar et al. 2017).These observations also showed that FRB 20121102A was spatially consistent with a faint persistent radio source (PRS).Follow-up observations by the EVN (European VLBI Network; Marcote et al. 2017) enabled a milliarcsecond localisation of the bursts and concrete association between the bursts and PRS (a projected linear separation of ≲40 pc).They also showed that the PRS is compact on sub-pc scales, and hence cannot be due to local star formation.Rather, it may be a hyper-nebula powered by the FRB source, or a low-luminosity active galactic nucleus (Marcote et al. 2017).The precision of this FRB 20121102A localisation further enabled characterisation of the local environment using the Hubble Space Telescope (HST; Bassa et al. 2017), which revealed that FRB 20121102A is inside, but slightly off center (∼ 200 pc) from a knot of star-formation in its host galaxy.Together with the source's extreme and highly variable Faraday rotation measure (RM), this supports the case for a young magnetar progenitor (e.g., Metzger et al. 2019).
Thereafter, the repeating, and periodically active (CHIME/FRB Collaboration et al. 2020a), FRB 20180916B was localised by the EVN to a nearby massive spiral galaxy (Marcote et al. 2020).The precision of the EVN localisation allowed for the association of the FRB source with the apex of a relatively large, apparently v-shaped star-formation region, but also ruled out the presence of a PRS, distinguishing it from the other known and localised repeater at the time.Follow-up HST observations showed that FRB 20180916B is located slightly offset (∼ 250 pc) from the nearest knot of star formation -suggesting that it is a neutron star formed by a runaway massive star, or perhaps an older neutron star in a binary system (Tendulkar et al. 2021).
The hyper-active repeater FRB 20201124A (e.g., Lanman et al. 2022), was first localised to arcsecond precision by the Australian Square Kilometre Array Pathfinder (ASKAP; Day et al. 2021), Very Large Array/realfast (VLA; Law et al. 2021), and upgraded Giant Metrewave Radio Telescope (uGMRT; Wharton et al. 2021b).The host galaxy was found to be star-forming, dusty and an orderof-magnitude more massive than the hosts of other repeaters at the time, bridging the gap between the hosts of repeaters and apparent non-repeaters (Ravi et al. 2022).The VLA (in D-configuration) and uGMRT detected unresolved, persistent radio emission at radio frequencies of 3 and 9 GHz (Ricci et al. 2021), and 300 MHz (Wharton et al. 2021a), respectively.Follow-up observations with the VLA (in C-configuration) at 22 GHz, however, resolved this emission, disqualifying it as a compact PRS and showing that the radio emission is more likely due to star formation (Piro et al. 2021).Milliarcseond localisation with the EVN (Nimmo et al. 2022) found no evidence for compact radio emission coincident with the burst position, supporting the notion that the previously detected emission is of extended nature and that FRB 20201124A is embedded in a region of star formation.The milliarcsecond localisation also enabled deeper, high-resolution radio and optical studies with the VLA and HST, respectively, leading to the conclusion that the FRB source formed in situ (Dong et al. 2023).
Using the raw voltage data of three bursts, the Canadian Hydrogen Intensity Mapping Experiment FRB project (CHIME/FRB; CHIME/FRB Collaboration et al. 2018) localised FRB 20200120E to the outskirts of the M81 spiral galaxy complex (at a luminosity distance of 3.6 Mpc) with a 90 per cent confidence interval of ≃ 14 arcmin 2 (Bhardwaj et al. 2021).Follow-up observations by Kirsten et al. (2022a) confirmed that FRB 20200120E is indeed associated with the M81 galactic system and, surprisingly, coincident with a globular cluster.This finding challenged theories that advocate that all FRBs originate from young magnetised neutron stars formed via core collapse SNe (supernovae).If FRB 20200120E is indeed such a magnetised neutron star, alternative formation channels need to be invoked: e.g., formation via binary merger or accretion-induced collapse of a white dwarf (Kremer et al. 2021).
Finally, FRB 20190520B, discovered by the Five-hundred-meter Aperture Spherical Telescope (FAST), was localised to a dwarf host galaxy at a  = 0.241 using the VLA (Niu et al. 2022).VLA observations identified a potential PRS with a flux density of ∼ 200µJy at 3 GHz.Recent observations with the EVN have confirmed the compact PRS nature by constraining the transverse size of the source to be < 9 pc.These observations have also showed that the FRB source and the PRS are consistent with being co-located within ≤ 80 pc -consistent with the hypothesis that a single central engine must power both the bursts and the PRS (Bhandari et al. 2023).
The primary focus of this paper, a hyper-active repeater called FRB 20220912A, was discovered by CHIME/FRB (McKinven & CHIME/FRB Collaboration 2022).CHIME/FRB reported a position of RA (J2000): 347.29(4) • , Dec (J2000): +48.70(3) • (90 per cent uncertainty errors), and a dispersion measure (DM) of 219.46(4) pc cm −3 for this source.The FRB position lies somewhat outside of the Galactic plane:  = 106.1 • ,  = −10.8• .The expected scattering timescale from the Galactic interstellar medium (ISM) along this line-of-sight is a moderate 2.6 µs (at 1 GHz) according to the NE2001 Galactic electron density model (Cordes & Lazio 2002).The Deep Synoptic Array (DSA-110) localised FRB 20220912A to a host galaxy, PSO J347.2702+48.70,at  = 0.0771(1) (Ravi et al. 2023).The host galaxy has a stellar mass of approximately 10 10 M ⊙ and a star-formation rate of ≳ 0.1 M ⊙ yr −1 , making it unremarkable compared to some other known host galaxies of repeaters (Gordon et al. 2023).FRB 20220912A is the most active FRB known to date, with FAST detecting as many as 390 bursts per hour (Zhang et al. 2023).Interestingly, the local environment of the source also appears to be clean, as the RM of the bursts has been stable around zero for a period on the order of months (e.g., Feng et al. 2023;Zhang et al. 2023;Hewitt et al. 2023).
Here we present the interferometric localisation of 150 bursts detected from FRB 20220912A with EVN-Lite observations in October 2022.Section 2 outlines the technical details of our observations.We describe our search pipeline, localisation procedures and the measurement of burst properties in Section 3. Finally, our main conclusions are presented and placed in the context of other FRBs in Section 4.

OBSERVATIONS
We observed FRB 20220912A in three observing runs in October 2022 as part of the ongoing FRB VLBI localisation programme called PRECISE (Pinpointing REpeating ChIme Sources with EVN dishes; PI: Kirsten).The first observation (Epoch 1; EVN project code EK051G; PRECISE code PR249A) was conducted on 22 October 2022 from 00:00-04:46 UT, and utilised an ad hoc array of 11 EVN and eMERLIN dishes: Cambridge, Darnhall, Defford, Effelsberg, Knockin, Jodrell Bank Mark II, Medicina, Noto, Pickmere, Toruń, and Westerbork.The second observation (Epoch 2; EK051H; PR247A), was conducted from 24 October 2022 21:00 UT to 25 October 2022 02:00 UT.Westerbork did not participate in the second observation and the array consisted of the remaining 10 dishes from Epoch 1.The third observing run (Epoch 3; PR248A), was conducted from 26 October 2022 23:00 UT to 27 October 2022 04:30 UT.During this run we used the 11 aforementioned dishes as well as the Onsala 25-m telescope.In the first two observations, we pointed the array to a sky position of RA=23 h 09 m 05.49s Dec=+48 • 42 ′ 25.6 ′′ , which is the position of the initial localisation determined using the DSA-110 (Ravi 2022).We note that this position is 5.8 ′′ offset from the final reported DSA-110 position, RA=23 h 09 m 04.9 s Dec=+48 • 42 ′ 25.4 ′′ (Ravi et al. 2023), but still well within the primary beam of all EVN dishes.In Epoch 3 we pointed to the updated DSA-110 position.Observations for Epoch 1 and 2 were carried out at a central frequency of 1.4 GHz with bandwidth ranging from 64-256 MHz for the different antennas, and we recorded dual-polarization raw voltage data in a circular basis with 2-bit sampling at all the participating stations in VDIF (Whitney et al. 2010) format.The frequency coverage was not identical between individual dishes and is illustrated in Figure 1.For Epoch 3 we observed with a similar set-up but at higher frequencies (4798-5054 MHz).We provide a more concise overview of Epoch 3 as no bursts were detected during this higher-frequency observation (Kirsten et al. 2022b).
For Epoch 1, our observations interleaved target scans of 5.75 min on FRB 20220912A and scans of 1.5 min on a nearby (3.0 • offset) phase calibrator source, J2311+4543, resulting in phase referencing cycles with a duration of approximately 7.25 min.Every fifth iteration we also observed another nearby source, J2314+4518 (0.6 • offset from the phase calibrator), for 3.5 min to be used as an interferometric check source.This check source is used to estimate the absolute astrometric uncertainty and potential amplitude losses that might have been introduced during phase referencing.A 5 min scan was scheduled on J1327+4326 to use as a fringe finder and bandpass calibrator.Finally, the pulsar B2111+46 was also observed for 5 min to verify the integrity of our data for the burst search and single pulse analyses with Effelsberg.A similar strategy was followed in Epoch 2 and 3, but in Epoch 2 the phase calibrator J2311+4543 was also used as a fringe finder and bandpass calibrator, the check source was replaced with J2327+4754 (3.4 • offset from the phase calibrator) and B0329+54 was used as a test pulsar.In Epoch 3 we used J2308+4629 as a phase calibrator, J2327+4911 as a check source again (4.1 • offset from J2327+4911), J2311+4543 as the fringe finder and B2020+28 and B0540+23 as test pulsars.

Search for Bursts
We searched the raw voltage data from Effelsberg for bursts, using the pipeline previously described in detail in Kirsten et al. (2021).
In short, the raw voltage data were converted to Stokes I filterbanks with time and frequency resolutions of 64 µs and 62.5 kHz, respectively, using digifil (van Straten & Bailes 2011).We then used the GPU-accelerated transient detection software Heimdall2 to search a DM range of 169 − 269 pc cm −3 for FRB candidates that are above a signal-to-noise ratio (S/N) of 7. The resulting candidates were classified by the machine learning convolutional neural network FETCH (Agarwal et al. 2020), using the models A and H 3 .All candidates for which these models assigned a > 0.5 probability of the burst being astrophysical in origin were manually inspected.We detected a total of 45 and 105 bursts in the first and second observation, respectively.No bursts were detected in the 5-GHz data from Epoch 3. A sub-sample of the bursts is shown in Figure 2.

Correlation of Interferometric Data
The PRECISE data were correlated in numerous passes at the Joint Institute for VLBI ERIC (JIVE) in the Netherlands (EVN correlation proposal EK051; PI: Kirsten), making use of the software correlator SFXC (Keimpema et al. 2015).For Epoch 1, the first pass was a delaymapping correlation where three bursts and their bracketing phase calibrator scans were used to determine the burst position to an uncertainty of ≈ 1 arcsec (see detailed description in Marcote et al. 2020).The initial DSA-110 localisation (Ravi 2022) was used as the phase centre of the FRB 20220912A target field.The correlation was done with 8 × 32-MHz sub-bands consisting of 64 spectral channels each and an integration time of 2 s for the phase calibrator scans, while the target scans were manually gated according to the width of the bursts and coherently dedispersed to a DM of 219.46 pc cm −3 .In the second pass, all the bursts were coherently dedispersed and correlated at a phase center determined by the delay-mapping.In order to maximise the S/N, gate widths were chosen around the arrival time of each burst.After the interferometric localisation of the FRB 20220912A bursts described in the next sub-section, all target data were then re-correlated in a third and final correlation pass using this position as the phase center, in order to create a deep image to look for persistent radio continuum counterparts.We repeated the procedure for Epoch 2, but without the first delay-mapping pass (as the position was already known), and using the interferometric localisation from Epoch 1 as the phase center.
During these correlations we encountered a few technical issues that required resolution before finalising the analysis.The Earth Orientation Parameters (EOPs) used in the correlation of the EK051H data were not properly updated in the correlator due to a failure of a software that pings NASA's Archive of Space Geodesy Data4 .This was discovered and corrected for the continuum data by re-correlating them with updated EOPs.However, the burst correlation still had outdated EOPs with a large discrepancy, notably in the UT1−UTC values.This initially introduced an offset of ∼ 30 mas in RA for the position of the burst source between EK051G and EK051H.The offset was resolved internally at JIVE by performing an EOP correction on EK051H burst data.The correction applied a phase shift to the visibilities that corresponds to the delay difference that results from the different sets of parameters.These delays are approximated by using the IAU2000A precession and nutation model to calculate the celestial to terrestrial coordinate transformation matrix.The source code for this implementation is available online5 , and this feature will be added to CASA (McMullin et al. 2007;van Bemmel et al. 2022) in a future release.

Burst Localisation
The EVN data were calibrated using standard interferometric techniques in AIPS (Greisen 2003) and DIFMAP (Shepherd et al. 1994).Imaging was performed in DIFMAP and CASA v6.1.
After the correlated visibilities (in FITS-IDI format) were loaded into AIPS, we first applied the calibration table from the EVN AIPS pipeline that contains the parallactic angle correction and a-priori gain correction, using the gain curves and system temperature measurements that the stations recorded during the observations.We also applied the a-priori flagging table and bandpass calibration table.We then flagged the edges of sub-bands (≈ 15 per cent of the channels in total) and manually flagged data from the fringe finder scans that were contaminated by RFI (radio frequency interference).Ionospheric dispersive delays can have a significant impact on the calibration and localisation accuracy at milliarcsecond scales at Lband.To mitigate this we made use of the VLBATECR task in AIPS to correct for these delays.The task makes use of maps from the Jet Propulsion Laboratory of the total electron content (TEC) at the different EVN sites during the observations, and compensates for the dispersive delays accordingly.We used the fringe finder scans (J1327+4326 for Epoch 1 and J2311+4543 for Epoch 2), with Effelsberg as the reference antenna, to remove the phase jumps between sub-bands and phase slopes within sub-bands that are introduced because of the different signal paths for individual sub-bands.Next, a global fringe fit was performed to correct the phases of the entire observation for all calibrator sources as a function of both frequency and time.The solutions were manually inspected, and bad solutions were flagged.
Having applied the aforementioned calibration, we imaged the phase calibrator (J2311+4543) and check sources (J2314+4518 for Epoch 1 and J2327+4754 for Epoch 2) in DIFMAP using a cell size of 1 mas and a natural weighting scheme (synthesized beam sizes of ≈ 30×40 mas).We were able to reproduce the positions of both check sources to a precision of ≲2 mas compared to the expected positions from 5 GHz maps of the sources in the RFC 2023B catalogue 6 .For J2314+4518 we measure a positional offset of Δ  = 0.1 mas and Δ  = 1.0 mas, and for J2327+4754, a positional offset of Δ  = 1.6 mas and Δ  = 1.9 mas.The expected positions of J2314+4518 and J2327+4754 in the 5 GHz maps have uncertainties of 1.06 mas and 1.47 mas, respectively.Taking these uncertainties into account, together with the difference in observing frequency, we conclude that our calibration was successful.We factor in these positional offsets when determining the FRB position and conservatively include the statistical uncertainty on the check source positions in the calculation of the final FRB positional uncertainty.
We also performed self-calibration to further improve our calibration solutions.We first imaged and self-calibrated the phase calibrator in DIFMAP to obtain the best possible model of the source.This model allowed us to improve the phases and amplitudes of the different antennas.The resulting model was imported into AIPS and was used to create a calibration table.Finally, we applied these calibration solutions to the target field of FRB 20220912A and imaged the target (both the continuum data and burst data).We again used a cell size of 1 mas and natural weighting.
We combined the visibilities of the 45 bursts detected in Epoch 1 and 105 bursts detected in Epoch 2 to create the dirty maps shown in Figure 3 with CASA.Taking into account the positional offset of our check sources, for Epoch 1 we find the position of the bursts from FRB 20220912A to be RA (J2000) = 23 h 09 m 04.8990 s ± 3.4 mas, 6 http://astrogeo.org/sol/rfc/rfc_2023b/rfc_2023b_cat.htmlDec (J2000) = 48 • 42 ′ 23.9104 ′′ ± 3.3 mas.For Epoch 2 we find a position of RA (J2000) = 23 h 09 m 04.8987 s ± 3.5 mas, Dec (J2000) = 48 • 42 ′ 23.9053 ′′ ± 3.5 mas.These positions are offset from one another by Δ  = 3.5 mas and Δ  = 5.1 mas, but despite the total offset of 6.2 mas, still consistent with one another to within ∼1 (the synthesized beam sizes are 29 × 40 mas and 24 × 30 mas for Epoch 1 and 2, respectively).The uncertainties we quote take into account multiple factors that are summed in quadrature: the statistical uncertainty derived from the shape and size of the synthesized beam normalized by the S/N (ΔRA = 0.3 mas, ΔDec = 0.3 mas for Epoch 1 and ΔRA = 0.4 mas, ΔDec = 0.3 mas for Epoch 2); the statistical uncertainty on the position of the phase calibrator, J2311+4543 (0.10 mas); an estimate of the uncertainty from phase-referencing due to the angular separation between the phase calibrator and target (∼3 mas; Kirsten et al. 2015); an estimate of the frequency-dependent shift in the phase calibrator position from the International Celestial Reference Frame (ICRF), here conservatively ∼1 mas (Plavin et al. 2022); and the statistical uncertainty on the positions of the interferometric check sources (1.06 and 1.47 mas for Epochs 1 and 2, respectively).A more in-depth per epoch analysis is presented in Appendix B1.
We also imaged a 2 × 2 arcsec 2 area surrounding the position of the bursts to search for a compact radio continuum counterpart (i.e., a PRS). Figure 4 shows these dirty images.The images of the first and second epochs have an RMS of 21 and 24 µJy beam −1 , respectively, while the combined data from both epochs have an RMS of 16 µJy beam −1 .We find no evidence for any persistent radio emission on milliarcsecond scales surrounding FRB 20220912A, ruling out the presence of a PRS above 5 significance ( ≈1.2×10 28 erg s −1 Hz −1 ) in a region of ∼ 2 × 2 arcsec 2 .

Burst Properties
For each burst, we used digifil to create filterbank files from the baseband data recorded by Effelsberg.The time and frequency resolution of these filerbanks were 64 µs and 62.5 kHz, respectively.The bursts were then incoherently dedispersed to a DM of 219.37 pc cm −3 .This DM value was determined by temporally aligning high-S/N, broadband microshots in the dynamic spectra of exceptionally bright bursts detected from FRB 20220912A with the Nançay Radio Telescope (Hewitt et al. 2023).Optimising for structure by using DM_phase (Seymour et al. 2019), with a bandpass filter on the fluctuation frequencies, yields similar results.Since we did not apply coherent dedispersion, we chose enough frequency channels to limit DM smearing in the lowest frequency channel to less than the time resolution of the data.After bandpass correction (subtracting the mean and dividing by the standard deviation of the off-burst noise on a per-channel basis) we applied a static mask at frequencies ranging from 1448-1477 MHz, in addition to manually flagging channels that are contaminated by RFI.
radiometer equation (Cordes & McLaughlin 2003).As is convention for Effelsberg observations, we assume a system temperature and gain of 20 K and 1.54 K Jy −1 , respectively.These values have an uncertainty of approximately 20%, which propagates into the following energy calculations.We calculated the spectral energy density (  ) as: where  is the fluence, Δ is the spectral extent of the burst,  is central observing frequency, and   and  are the luminosity distance (362.4Mpc) and redshift (0.0771) of the host galaxy of FRB 20220912A, respectively (Ravi et al. 2023).These properties, as well as the times of arrival of the bursts, are tabulated in Table A1. Figure 5 shows the normalized distribution of these properties, per epoch.The burst property distributions show little variation between the two epochs.Note that the spectral extent of the burst only considers the observed range, and is consequently often a lower limit as many bursts appear to have emission outside of our observing window.The median values for width and fluence of all bursts detected are 6.6 ms and 47 Jy ms, respectively.

DISCUSSION AND CONCLUSIONS
In this paper we report the detection of 150 bursts from FRB 20220912A using an ad hoc EVN-Lite array of dishes, which allowed us to localise this FRB source to a precision of a few milliarcsecond: RA (J2000) = 23 h 09 m 04.8988 s ± 5 mas, Dec (J2000) = 48 • 42 ′ 23.9078 ′′ ±5 mas.FRB 20220912A is now the sixth repeating FRB source to be localised to milliarcsecond precision using VLBI.We find that FRB 20220912A is significantly closer to the optical centre of its host galaxy, PSO J347.2702+48.70,compared to the earlier localisation presented by Ravi et al. (2023), shown in Figure 6.The transverse offset from the host galaxy center is ≈ 0.8 kpc.The precision of this VLBI localisation corresponds to a physical length of less than 10 pc at the redshift of the source, and this mas-level position will serve future high-resolution IR/optical/UV imaging with HST, James Webb Space Telescope JWST, and the Extremely Large Telescope (ELT), which could reveal star-forming regions or other discrete sources coincident on parsec scales with the position of FRB 20220912A.
FRB 20121102A and FRB 20190520B are the only known repeating FRBs that exhibit a PRS (Marcote et al. 2017;Niu et al. 2022), which may represent a hyper-nebula powered by the burst source (e.g., Sridhar & Metzger 2022).These are also two of the four repeaters from which burst storms have been observed (e.g., Li et al. 2021;Hewitt et al. 2022;Jahns et al. 2023;Niu et al. 2022), the other two being FRB 20200120E (Nimmo et al. 2023) and FRB 20201124A (e.g., Lanman et al. 2022;Zhou et al. 2022).As previously mentioned, in the case of FRB 20201124A there is persistent radio emission attributed to star formation that resolves out at higher angular resolution (e.g., Piro et al. 2021;Nimmo et al. 2022), whereas FRB 20200120E resides in a globular cluster with no signs of persistent radio emission (Kirsten et al. 2022a).
The upper limit we have placed on the presence of a PRS for FRB 20220912A is more than an order-of-magnitude below the luminosity of the PRSs associated with FRB 20121102A and FRB 20190520B.There exists a catalogued continuum radio source, APTF J230904+484222, detected by the Westerbork Synthesis Radio Telescope Aperture Tile In Focus (WSRT-APERTIF) and located at RA (J2000) = 23 h 09 m 04.9 s ± 1.7 arcsec, Dec (J2000) = 48 • 42 ′ 22.3 ′′ ± 2.2 arcsec, which is 1.6 arcsec away from our VLBI position for FRB 20220912A.APTF J230904+484222 has a peak brightness of 0.27±0.04mJy beam −1 at 1355 MHz ( ≈3.9×10 28 erg s −1 Hz −1 ).The contours and centroid of this source are overplotted on the optical image and VLBI position in Figure 6.The centroid position does not coincide with the nucleus of the galaxy.We also explore 4×4 arcsec 2 around this centroid-position for a compact PRS, but the highest peak we find is 0.072 mJy (<5), still more than 3 times fainter than APTF J230904+484222.This continuum radio source thus likely reflects star formation in the host galaxy, similar to the case of FRB 20201124A (Nimmo et al. 2022).Using the 1.4 GHz luminosity-to-SFR relation (Murphy et al. 2011), a star-formation rate of about 2.5 M ⊙ yr −1 is required to explain the observed radio flux density of APTF J230904+484222.This radio luminosity inferred SFR is consistent with the SFR of ≳ 0.1 M ⊙ yr −1 derived from H observations and is ∼3.5 times less than the star-formation rate inferred from radio observations of the host of FRB 20201124A (Dong et al. 2023).
If the PRSs associated with FRB 20121102A and FRB 20190520B are in fact hyper-nebulae, powered by the central active magnetar (Sridhar & Metzger 2022), the lack of a PRS in the case of FRB 20220912A is particularly surprising, given how active this source is.The absence of a PRS is, however, consistent with the stable and near-zero RM that suggests a non-turbulent and clean local environment (e.g., Feng et al. 2023).
We strongly encourage multi-wavelength observations of FRB sources that are outliers in terms of proximity or activity, such as FRB 20220912A and the others FRBs that have been localised to milliarcsecond precision.These observations can be a powerful means of characterizing the local environments of FRBs, and detailed studies such as these complement studies that provide less detailed information for a larger number of sources.Currently, with HST it is possible to compare positions at the ∼ 10 mas level (∼ 10% of the point spread function width), so there remains much to be gained from precision localisations.Scheduled to commence observations within a decade, the ELT will provide 5 mas resolution, enabling an even stronger optical synergy to milliarcsecond localisations in the radio regime.

DATA AVAILABILITY
The EVN observations are available on the EVN Data Archive at JIVE7 under project codes EK051G and EK051H.The relevant code and data products for this work will be uploaded on Zenodo at the time of publication. The time-of-arrival of the burst at the solar system barycenter in TDB, corrected to infinite frequency for a DM of 219.37 pc cm −3 and using a DM constant of 1/(2.41×10−4 ) MHz 2 pc −1 cm 3 s. FWHM of 1D Gaussian fit.
Measured over the spectral extent of the burst.
The estimated uncertainty is approximately 20 per cent due to uncertainty in the system equivalent flux density (SEFD).

Figure 1 .
Figure 1.Our PRECISE observations spanned a total bandwidth of 256 MHz from 1254-1510 MHz.The frequency coverage of each of the EVN dishes in the array is shown in the plot by the horizontal bars, while the dashed vertical lines indicate the edges of the sub-bands.Noto observes from 1350-1606 MHz, but only the range below 1510 MHz, where there is overlap with other stations, is correlated.

Figure 2 .
Figure2.This sub-sample of bursts detected from FRB 20220912A with the Effelsberg dish illustrates the diversity in burst duration and the complex morphology seen in some high-S/N bursts.All bursts have been dedispersed using a DM of 219.37 pc cm −3 .In each thumbnail, the main panel shows the dynamic spectrum of the burst.The top panel shows the frequency-averaged time profile (averaged over the spectral extent of the burst), while the side panel shows the time-averaged frequency spectrum.For visual purposes the bursts have been averaged in time and frequency, and the plotted time and frequency resolutions are shown in the top-right corner of each thumbnail.Horizontal white lines in the dynamic spectrum indicate channels that have been masked due to the presence of RFI.

Figure 3 .Figure 4 .
Figure 3.In the top row, the dirty maps of the EVN 1.4 GHz observation of the combined visibilities of the 45 bursts detected in Epoch 1 and 105 bursts detected Epoch 2 are shown in the left and middle panels, respectively.The combined visibilities of both epochs, i.e. all 150 bursts, are shown on the right.The cyan contours start at 4 times the RMS noise level of each image and increase by factors of 3. In the bottom row the CLEAN images are shown with the synthesized beam displayed as a silver ellipse in the bottom left corner.

Figure 5 .Figure 6 .
Figure 5.The distribution of the temporal width, fluence, and spectral extent of the bursts we detected are shown in dark blue and cyan for Epoch 1 and 2, respectively.Note that the spectral extent distribution only reflects the observed spectral extent of bursts and are thus in many cases lower limits.The histograms have been normalised for each observation, so that the total area equals 1. Vertical lines indicate median values: solid black for Epoch 1 and dashed grey for Epoch 2.

Figure B1 .
Figure B1.The position of the peak S/N of the dirty maps of individual bursts, compared to the best-known FRB position for Epoch 1 (top) and Epoch 2 (bottom).The points have been coloured according to a detection metric that is defined as fluence of the burst divided by the square root of the temporal width of the burst.The uncertainty of individual burst positions is underestimated, since only the peak value of the dirty map is used here for illustrative purposes.The true uncertainty of individual burst positions is more precisely known from taking into account the sidelobes.Bursts that occur during times when the calibration solutions are less robust are excluded from this analysis.