Main

ASKAP J1935+2148 was serendipitously discovered during a target of opportunity observation of the gamma-ray burst GRB 221009A with the Australian Square Kilometre Array Pathfinder (ASKAP) telescope. Bright pulses of radio emission from ASKAP J1935+2148 were seen on 15 October 2022 using a fast-imaging technique on images with 10 s integration times (Methods). ASKAP J1935+2148 is located at right ascension (J2000) 19 h 35 min 05.126 s ± 1.5″ and declination (J2000) +21° 48′ 41.047″′ ± 1.5″, which is coincidentally 5.6′ from the magnetar SGR 1935+2154 and sits on the edge of the supernova remnant in which SGR 1935+2154 is centred. The observation lasted ~6 h, revealing 4 bright pulses lasting 10–50 s in the images, with the brightest peak pulse flux density measuring 119 mJy. Inspection of the light curves of the pulses revealed a tentative period of ~54 min.

In addition to a weak detection in an archival ASKAP observation, the source was consistently detected in four follow-up observations. A summary of all observations with ASKAP is presented in Extended Data Table 1. Overall, the pulses are visible across the whole bandpass-corrected observing band of 288 MHz, leading to a spectral index estimation of α ≈ + 0.4 ± 0.3 at 887.5 MHz. However, a dispersion measure (DM) constraint/estimate was not possible due to the coarse time resolution of 10 s. We quantify the pulses to be >90% linearly polarized—implying strongly ordered magnetic fields, with a rotation measure (RM) of +159.3 ± 0.3 rad m−2 calculated using the RM synthesis method1. In comparison, the RM and DM of SGR 1935+2154 are approximately +107 rad m−2 and approximately 330 pc cm−3, respectively2. The RM of ASKAP J1935+2148 is consistent with the contribution from the smoothed Galactic foreground3 and with those of nearby pulsars (https://www.atnf.csiro.au/research/pulsar/psrcat/), precluding the presence of a substantial RM imparted at the source.

Following the discovery, we conducted simultaneous beamformed and imaging follow-up observations at 1,284 MHz with the MeerKAT radio interferometer (Methods). Two pulses were detected in both the beamformed and imaging data in two independent observations (Extended Data Table 1). The initial estimate of period allowed us to predict the times of arrival (ToAs) of future pulses at the same rotational phase of ASKAP J1935+2148, and the MeerKAT pulses are observed to arrive within 319 ms of the predicted times (that is, within 10−4 of a period). The ToAs of all the ASKAP and MeerKAT detections were used to determine a phase-connected timing solution with a period P of 3,225.313 ± 0.002 s (Methods and Fig. 1), and an upper limit on the period derivative, \(\dot{P}\) of ≲(1.2 ± 1.5) × 10−10 s s−1 with a 1σ error. The location of ASKAP J1935+2148 in the \(P-\dot{P}\) parameter space, which is frequently used to classify different sorts of pulsars, is consistent with other known long-period sources (Extended Data Fig. 1). ASKAP J1935+2148 is seen to reside in the pulsar death valley where detectable radio signals are not expected, challenging currently accepted theories of radio emission via spin-down (Extended Data Fig. 1). The radio properties of ASKAP J1935+2148 are presented in Table 1.

Fig. 1: Light curves of the ASKAP and MeerKAT detections.
figure 1

The y axis is the pulse number. The peak flux densities of these detections are reported in Extended Data Table 1. The different colours represent the dates of the observations. Pulse detections within an observation represent consecutive rotations of the source. Although different in terms of radio properties, the MeerKAT detections appear to arrive in phase with the ASKAP detections.

Full size image
Table 1 Measured and derived radio quantities for ASKAP J1935+2148 from the ASKAP and MeerKAT observing campaigns
Full size table

A first single pulse was detected by the MeerTRAP real-time detection system (Methods) on 3 February 2023 with a DM of 145.8 ± 3.5 pc cm−3 and a width of ~370 ms (Extended Data Table 1), which is ~135 times narrower than the brightest ASKAP pulse. Contrary to the ASKAP duty cycle of 1.5%, the narrow width of the MeerKAT pulse results in a duty cycle of only 0.01%. The average DM inferred distance based on the NE20014 and YMW165 Galactic electron density models places ASKAP J1935+2148 at a distance of 4.85 kpc (Table 1). The detection is accompanied by weak pre- and post-cursor pulses, as seen in Fig. 2. The data recorded to disk with the PTUSE backend (Methods) did not reveal a broader underlying emission envelope similar to the wide pulse widths seen in the ASKAP detections. The corresponding MeerKAT 2-s-resolution image (the shortest possible timescale) revealed a single 9 mJy detection, which is ~26 times fainter than the brightest ASKAP pulse. The pulse was localized to right ascension (J2000) 19 h 35 min 05.175 s ± 0.3″ and declination (J2000) +21° 48′ 41.504″ ± 0.6″, which is consistent with the ASKAP coordinates. Throughout the paper, the flux densities quoted for the MeerKAT data are from the images as the beamformed data are only polarization and not flux calibrated. Unlike the ASKAP pulses, our MeerKAT detection revealed a substantial circular polarization fraction exceeding 70%, coupled with a linear polarization fraction of ~40%. We did not find evidence for Faraday conversion (Methods). In addition to being spatially coincident, the measured RM of +159.8 ± 0.3 rad m−2 agrees with that measured for the ASKAP detections, giving us added confidence that despite the drastically different pulse widths, the ASKAP and MeerKAT detections were produced by the same object.

Fig. 2: The dynamic spectra and polarization pulse profiles of ASKAP J1935+2148 from the MeerKAT beamformed data.
figure 2

Left: a bright detection on 3 February 2023. Right: a weaker detection on 8 May 2023. The data have a time resolution of 2.4 ms and are de-dispersed to a DM of 145.8 pc cm−3and corrected for an RM of +159.3 rad m−2. Top: the polarization position angle (PA; for values of linear polarization greater than three times the off-pulse noise), which is observed to be flat across the main pulse profile in the left panel. The insufficient signal-to-noise ratio during the detection on 8 May 2023 prevented robust measurements of polarization position angles. Middle: the Stokes parameter pulse profiles for ASKAP J1935+2148 at 1,284 MHz where black represents the total intensity, magenta represents linear polarization and blue represents circular polarization. The flux density is in arbitrary units as the data are not flux calibrated. The arrows indicate the positions of the pre- and post-cursor bursts for the detection on 3 February 2023 (left). Bottom: the dynamic spectra where the backwards sweeping striations across the observing band in the left panel correspond to ~50 Hz radio-frequency interference.

Full size image

The second MeerKAT detection was made on 8 May 2023 with a flux density of 2.9 mJy averaged over 2 s (Extended Data Table 1). This burst was also ~370 ms wide and highly circularly polarized with no broader emission envelope, but the lack of sufficient signal to noise precluded a reliable RM estimation. Both MeerKAT pulses are visible across the whole 856 MHz band with a spectral index estimate of α ≈ −1.2 ± 0.1 at 1,284 MHz. Such variations in spectral indices have been observed in both pulsars and radio magnetars showing different emission states6,7. Combining all the ASKAP and MeerKAT observations (Extended Data Table 1), we see that the source is not detectable in every single observation indicating intermittency, potential nulling where the pulsed emission temporarily ceases or becomes undetectable, or drastic variations in flux density. Using epochs 1 to 17 in Extended Data Table 1, we estimate the source to be in a quenched or quiescent state ~40–50% of the time, at MeerKAT and ASKAP.

Observationally, ASKAP J1935+2148 appears to show three emission states:

  1. (1)

    The strong pulse mode consisting of 15 bright, tens-of-seconds-wide and highly linearly polarized pulses as seen with ASKAP

  2. (2)

    The weak pulse mode characterized by two faint, hundreds-of-milliseconds-wide and highly circularly polarized pulses as seen with MeerKAT

  3. (3)

    The completely nulling or quiescent mode as seen with both telescopes

We consider two possible scenarios for the observed differences in properties of the ASKAP and MeerKAT bursts.

In the first scenario, the ASKAP pulses could span only a small fraction (that is a few hundreds of milliseconds) of the shortest possible time resolution of 10 s. This scenario would imply that the source produces only subsecond-duration pulses. In Fig. 1, we see the flux densities of detections from 10 s ASKAP images to gradually rise and fall, resulting in an almost Gaussian-like pulse profile. This distribution of the flux densities makes it unlikely for the burst to be composed of several consecutive millisecond-duration pulses. However, it remains possible for subsecond-timescale structure to be superimposed on the broader emission envelope.

In the second scenario, it is likely that there are different emission modes at play. The source was undetectable with ASKAP in all follow-up observations post 5 November 2022, until the first MeerKAT observation on 3 February 2023 with its five times better sensitivity. The MeerKAT pulses that would have been undetectable at ASKAP are analogous to the ‘quiet’ pulse mode in PSR B0823+26, and the ‘dwarf pulse’ mode in PSR B2111+468. The pulses in these modes are generally undetectable in lower-sensitivity and/or low time-resolution observations such as with ASKAP. These weak pulses potentially exist between the nulling or quenched states of ASKAP J1935+2148. The location of the source at the edge of the supernova remnant in Extended Data Fig. 2 makes it is difficult to determine exactly what background emission to subtract. Therefore, we are unable to confirm the presence of persistent continuum radio emission that might be indicative of a wind nebula in either the MeerKAT or ASKAP data (Methods).

Collectively, the pulse widths, spectra and polarization properties of the ASKAP and MeerKAT detections suggest different physical coherent processes even though they occur at roughly the same rotational phase. Coherent radio emission from rotating neutron stars is efficiently generated by the creation of electron–positron pairs in the magnetosphere. The rotational spin-down creates an electric potential at the polar cap, causing pair production. Such charged plasma can emit radio waves that can be attributed to curvature radiation and inverse Compton scattering, and diverse magnetic-field configurations in emission models, including dipolar, multipolar and twisted fields along with vacuum gaps and space-charge-limited flows9,10. Magnetically powered neutron stars, however, generate coherent radio emission through decaying magnetic fields11. Extended Data Fig. 3 shows the manifestation of the physics underlying coherent and incoherent emitters, and indicates a coherent emission mechanism (brightness temperatures between 1014 K and 1016 K) being responsible for both the ASKAP and MeerKAT detections of ASKAP J1935+2148.

When comparing with other known long-period sources, ASKAP J1935+2148 appears to be similar to GLEAM-X J162759.5−523504.312 and GPM J1839−1013 albeit with a period that is four times longer but with a duty cycle not too dissimilar. GLEAM-X J162759.5−523504.3 was active for only three months, while GPM J1839−10 has remained active for over three decades12,13. Despite searches across radio data from the Giant Metrewave Radio Telescope (GMRT), Very Large Array (VLA) and VLA Low-band Ionosphere and Transient Experiment (VLITE) spanning 2013 to 2023, no pulses from ASKAP J1935+2148 were detected. We note that ASKAP J1935+2148 shares similarities with the Galactic Centre radio transient (GCRT), or ‘Burper’, GCRT J1745−3009. At the time of its discovery, GCRT J1745−3009 showed 10-min-wide pulses with a periodicity of 77 min (ref. 14), but subsequent observations revealed narrower and weaker pulses spanning 2 min (ref. 15). Varying circular polarization was also found in one of the pulses16. The similarities in the periods and the different emission states imply that ASKAP J1935+2148 could be a bridge between GCRT J1745−3009, GLEAM-X J162759.5−523504.3 and GPM J1839−10.

Owing to the proximity of SGR 1935+2154, there are numerous archival high-sensitivity X-ray observations at the position of ASKAP J1935+2148 (Methods and Extended Data Table 2). We focused on observations with the Chandra X-ray Observatory and Neil Gehrels Swift Observatory17. Using a combination of the more sensitive Chandra observations, we did not detect any X-ray source at the location of ASKAP J1935+2148. For an average DM inferred distance of 4.85 kpc (Table 1), this corresponds to a luminosity limit of about \(4\times 1{0}^{30}{d}_{4.85}^{2}\,{{{\rm{erg}}}}\,{{{{\rm{s}}}}}^{-1}\) (Methods) for a blackbody spectral model and a power-law spectral model for non-thermal emission from a neutron star. This is below the X-ray luminosities of most but not all rotation-powered pulsars and magnetars18,19 and is comparable to the X-ray luminosities of other long-period radio transients (≤1032−33 erg s−1)12,13. We also searched for flaring activity in ASKAP J1935+2148 using Swift, with 291 individual visits using the X-ray Telescope20 lasting 5–2,600 s from December 2010 to December 2022 for a total exposure time of 302.4 ks (exposure corrected). We see no sources in the summed dataset at the position of ASKAP J1935+2148 (Methods). The repeated visits with Swift allow us to rule out any flaring behaviour during this period.

Archival 300 s exposures in the J, H and Ks bands (1.2 μm, 1.6 μm and 2.1 μm) with the Very Large Telescope (VLT) using the near-infrared HAWK-I21 imager showed a source within the conservative 1.5″ ASKAP error radius of ASKAP J1935+2148 (Fig. 3). This source with J = 18.4 ± 0.1 mag, H = 17.3 ± 0.1 mag and Ks = 17.1 ± 0.1 mag (Vega) is catalogued as PSO J293.7711+21.8119 in Data Release 2 of the Pan-STARRS1 (Panoramic Survey Telescope and Rapid Response System; PS1) 3π survey22. We compute the chance of finding a source randomly (drawn from the background) in the Ks image, which has the highest source density, with magnitude brighter than or equal to this value to be 5% (that is ~2σ association) given the crowded nature of the field. However, to confidently rule out the association, we obtained a spectrum of PSO J293.7711+21.8119 in the 3,200–10,000 Å wavelength range with the Low Resolution Imaging Spectrometer at the Keck telescope in Hawaii (Extended Data Fig. 4). The calibrated spectrum is a red continuum devoid of discernible emission or absorption lines. The combination of the VLT magnitudes and the spectral characteristics suggests that it is an L/T-dwarf star. Gaia Data Release 1 parallaxes for known L and T dwarfs give J-band absolute magnitudes MJ in the range 10 < MJ < 16 (ref. 23), which implies a distance of less than 0.5 kpc for an apparent magnitude of J = 18.4. Because at 4.85 kpc such a star would be undetectable, we conclude that PSO J293.7711+21.8119 is a foreground star that is unlikely to be associated with ASKAP J1935+2148.

Fig. 3: Three-colour near-infrared image of the field around ASKAP J1935+2148, from VLT/HAWK-I JHKs imaging on 2 April 2015.
figure 3

The image cut-out is 20″ × 30″. The uncertainty in the position of ASKAP J1935+2148 is shown by a green circle of radius 1.5″.

Full size image

The observed period and emission of ASKAP J1935+2148 could be explained by a rotating magnetic white dwarf (MWD) emitting coherent radio emission like a neutron-star pulsar24. MWDs can be either isolated or in interacting binaries. There are ~600 known isolated MWDs with surface dipole magnetic fields up to 109 G and ~200 in interacting binaries with magnetic fields up to a few 108 G (ref. 25). Although radio emission from isolated MWDs has never been detected, despite searches for possible counterparts in large-area radio surveys (for example, refs. 26,27), we now entertain this possibility and derive the parameters that would be required to explain the radio emission of ASKAP J1935+2148. If ASKAP J1935+2148 is an isolated rotation-powered MWD, the measured P and upper limit on \(\dot{P}\) would yield a surface magnetic-field strength and spin-down luminosity of a few 1010 G and a few 1031 erg s−1, respectively, for a dipolar magnetic-field configuration, a magnetic inclination angle of 90° and a moment of inertia of 1050 g cm2. Even though the currently known isolated MWDs have magnetic fields below 109 G, it is theoretically possible for MWDs to have surface fields of up to a few 1013 G (refs. 28,29). In this case, ASKAP J1935+2148 would be the first MWD discovered to possess such a high magnetic field.

The radius of the source can be related to rotational period and magnetic-field strength to estimate the minimum radius of the source30 (Methods and Extended Data Fig. 5). Even under the most conservative assumptions, we can rule out an isolated MWD origin if we presume that the magnetic field cannot exceed 109 G, which is the maximum measured in an MWD. Similar considerations can also be applied to GLEAM-X J162759.5−523504.3 and GPM J1839−10, and we conclude that it is highly unlikely that the radio emission from these sources can be interpreted in terms of an isolated rotation-powered MWD. However, coherent and highly polarized radio emission has been detected in cataclysmic variables26,31, which are close binary systems containing a white dwarf primary accreting matter from a low-mass M-dwarf companion. In all detection cases, the radio emission appears to arise from the lower corona of the magnetically active M-dwarf and is attributed to the electron cyclotron maser instability. The problem here is that the radio luminosities of cataclysmic variables, in the range 1021–1025 erg s−1 (ref. 26), would be too low to explain the emission of ASKAP J1935+2148. Hence, it is also highly unlikely that a cataclysmic variable could be responsible for the radio emission of ASKAP J1935+2148.

Assuming a neutron-star origin, the period and upper limit on the period derivative correspond to a surface magnetic-field strength and spin-down luminosity of a few 1016 G and a few 1026 erg s−1, respectively, for a dipolar magnetic-field configuration, a magnetic inclination angle of 90° and a moment of inertia of 1045 g cm2. It is unclear why a neutron-star magnetar would still possess such a large magnetic field at this stage of its evolution, but explanations have been provided either in terms of the magnetic field’s structure (for example, refs. 32,33) or as due to a fall-back accretion disk (for example, ref. 34). Similar to GLEAM-X J162759.5−523504.312 and GPM J1839−1013, the observed radio luminosity of ASKAP J1935+2148 is much larger than the inferred spin-down luminosity, suggesting that alternative emission mechanisms must be involved to explain the radio emission of these long-period radio transients.

Achieving the observed duty cycle of approximately ~1% in ASKAP J1935+2148 necessitates a high degree of beaming, implying the generation and acceleration of relativistic particles—a phenomenon that is generally more easily accommodated in neutron stars than in white dwarfs. Remarkably, the isolated intermittent pulsar PSR J1107−5907, with P ≈ 253 ms, shows the same three distinct emission states as ASKAP J1935+214835. The emission in both PSR J1107−5907 and ASKAP J1935+2148 alternates between distinct modes, each characterized by unique pulse profiles, polarization properties and, at times, varying intensities. The intricate interplay of magnetic fields, plasma flows and the magnetospheric environment leads to the emergence of these different modes36. Instabilities within the pulsar magnetosphere may trigger transitions between these modes, contributing to the observed switching phenomenon. In addition, changes in the geometry of the magnetic-field configuration and the location of emission sites within the magnetosphere could influence the emitted radiation characteristics. Notably, PSR J1107−5907 is close to the pulsar death line(s)37,38 (Extended Data Fig. 1), beyond which radio emission is expected to cease. All these similarities seem to suggest that a neutron-star-like emission mechanism is at play for ASKAP J1935+2148.

In summary, we report the discovery of the long-period source ASKAP J1935+2148, which is unique compared with other known long-period sources, by manifesting three distinctive emission states reminiscent of mode-switching pulsars. The strong pulse mode shows bright and linearly polarized pulses lasting tens of seconds, the weak pulse mode features faint and circularly polarized pulses lasting hundreds of milliseconds, and the completely nulling or quiescent mode shows an absence of pulses. These diverse emission states offer valuable insights into the magnetospheric processes and emission mechanisms at play within this object, with similarities to the radio pulsars PSR J1107−5907, PSR B0823+26 and PSR B2111+46. Given the brief 3-month activity of GLEAM-X J162759.5−523504.3, ongoing monitoring may unveil emission modes similar to the those observed in ASKAP J1935+2148. We see that radio emission via pair production within dipolar magnetospheres presents considerable challenges24. However, a large magnetic field can power the observed radio emission via the dissipation of energy due to magnetic re-connection events, higher-order magnetic fields and untwisting of field lines due to plastic motion of the crust39,40. It would be prudent to study further whether such processes can persist for long timescales consistent with the long-term emission seen in a few long-period sources. Population-synthesis simulations incorporating various parameters such as masses, radii, beaming fractions and magnetic field show that only a limited number of long-period radio emitters of neutron-star origins are expected to exist in the Galaxy24. Conversely, in the white-dwarf scenario, a sizable population of long-period emitters can be accounted for. Nonetheless, explaining the production of coherent radio emission remains a formidable task in either scenario24,41. While MWDs have been considered to be responsible for the radio emission observed in sources such as GLEAM-X J162759.5−523504.3 and GPM J1839−10, we have ruled out this possibility for ASKAP J1935+2148. Thus, it is much more likely for ASKAP J1935+2148 to be an ultralong period magnetar or neutron star either isolated or in a binary system. Continued monitoring of this source should allow us to determine whether additional periods are present and the possible existence of a companion star.

Methods

ASKAP

The ASKAP array comprises 36 antennas, each equipped with a prime-focus phased array feed (PAF). Each PAF has 188 linearly polarized receiving elements sensitive to frequencies between 0.7 GHz and 1.8 GHz. The signal from each element is channelized to 1 MHz frequency resolution over a usable bandwidth of 288 MHz. The standard ASKAP hardware correlator produces visibilities on a 10 s timescale. During the detection of ASKAP J1935+2148, ASKAP was operated in the square_6 × 6 configuration with 1.05° pitch and 887.5 MHz central frequency. The pointing centre was chosen such that ASKAP’s large field of view (~30 deg2) would also encompass the magnetar SGR 1935+2154 known to have produced a burst with fast radio burst like (FRB) energies in 202042,43 and several less energetic bursts since.

The source was found during testing of a fast pulse-detection pipeline. The pipeline subtracts visibilities from neighbouring 10 s integrations and then images the result. The process effectively subtracts out quiescent emission from the field and retains only sources that change dramatically over a single integration. As such sources are rare, most images are effectively dominated by thermal noise and so do not require computationally expensive deconvolution. Once the image is checked for peaks above the noise, it is discarded to minimize storage requirements. The process is repeated for all time integrations and all 36 ASKAP beams. The processing initially found no detections in the first epoch of the ASKAP observation but found a bright pulse in beams 21, 22 and 23 of the second ASKAP epoch.

While the pulse-detection pipeline is reasonably effective for finding bright pulses, it forgoes some sensitivity to weak pulses to minimize resource usage. Once we had discovered ASKAP J1935+2148, more traditional techniques were used to investigate the pulse. A deep model image was derived for the beam and subtracted from the visibility data. The data were then phase shifted to the location of the source and dynamic spectra extracted (averaging over all baselines >200.0 m). This allowed weaker pulses to be detected and also allowed the linear polarization properties to be analysed. This approach was also repeated for MeerKAT observations.

Murriyang

The shortest ASKAP imaging observation (SBID 44918 in Extended Data Table 1), took place at the same time the 64 m Parkes (Murriyang) radio telescope was pointed at the known radio-burst emitting magnetar SGR 1935+2154. This observation was taken using the Ultra Wideband Low receiver system, spanning a bandwidth of 704–4,032 MHz. The position of ASKAP J1935+2148 was well within the 30-arcmin-wide low-frequency part of the Ultra Wideband Low beam. Only one weak pulse (~9σ or 10.7 mJy per beam) was detected from ASKAP J1935+2148 in the ASKAP data but no detection was made in the Parkes data above a signal-to-noise ratio of 8.

MeerKAT

In the observations presented in this work, MeerKAT operated at the L band (0.86-1.71 GHz) in the c856M4k configuration where the correlations were integrated for 2 s before saving to disk. We used PKS J1939−6342 as the primary flux calibrator and bandpass calibrator, and PKS J2011−0644 as the phase calibrator. MeerKAT was also used to simultaneously perform beamformed observations in the incoherent and coherent modes using MeerTRAP. The MeerTRAP backend is the combination of two systems: the Filterbank and Beamforming User Supplied Equipment (FBFUSE), a multiple-beam beamformer44,45, and the Transient User Supplied Equipment (TUSE), a real-time transient detection instrument46. FBFUSE applies the geometric and phase delays (obtained by observing a bright calibrator) before combining the data streams from the dishes into 1 incoherent beam and up to 780 coherent beams recording in total intensity, only. The beams can be placed at any desired location within the primary beam of the array, but are by default tessellated into a circular tiling centred on the boresight position, and spaced so that the response patterns of neighbouring beams intersect at the 25% peak power point. The beams are then sent over the network to TUSE for processing.

The TUSE single-pulse search pipeline searches for total intensity pulses in real time at a sampling time of 306.24 μs, up to a maximum boxcar width of 140 ms in the dispersion measure range of 0–5118.4 pc cm−3 at the L band. Only extracted candidate files are saved to disk for further investigation. Further details on the MeerTRAP backend can be found in refs. 47,48. The Accelerated Pulsar Search User Supply Equipment (APSUSE) backend instrument was used to record total intensity data with 4,096 frequency channels. In addition, full polarization data for the on-source beam were recorded to disk, but not searched in real time, using the Pulsar Timing User Supplied Equipment (PTUSE) backend of the MeerKAT pulsar timing project MeerTime described in ref. 49. These data were recorded in the PTUSE search mode with a sampling time of 38.28 μs in the psrfits format.

To address the impact of baseline variation in the recorded data, we utilized the APSUSE off-source beam as a reference to eliminate the baseline of the on-source beam. Following this process, we conducted a search for single pulses using TransientX (https://github.com/ypmen/TransientX)50, within a DM range of 120–160 pc cm−3 and a maximum pulse width of 1 s. This search yielded the detection of the two pulses in the PTUSE data. Subsequently, we extracted the polarization profiles from the PTUSE data of these two pulses, having removed the baseline based on the off-source APSUSE beam.

Period estimation

We generated ToAs for each of the detections made with the ASKAP and MeerKAT observation in Extended Data Table 1. Manipulation of the data used the tools available in the PSRCHIVE package51. The ToAs for the ASKAP detections are chosen to be the midpoints of the 10 s integrations they were detected in, with the error on the ToA equal to the duration of an non-detection in 10 s integrations immediately preceding, and succeeding the first and last detections of a pulse, respectively. The MeerKAT ToAs were obtained by using PSRCHIVE’s PAT on the beamformed data to estimate the time of the peak flux. Given the variability in the morphology of individual pulses, and presumable jitter in the emission measuring the pulse arrival times, we estimate the uncertainty on the arrival times to be the half-width at half-maximum of the widths of the pulses. The initial timing analysis for both telescopes used the best known period and DM at the time and the position determined from the imaging.

Timing was done using TEMPO252 with the JPL DE436 planetary ephemeris (https://naif.jpl.nasa.gov/pub/naif/JUNO/kernels/spk/de436s.bsp.lbl). The ToAs were fitted using a model including the period P and period derivative \(\dot{P}\). We do not need to fit for position as it is well determined from the imaging as described in previous sections. We also do not fit for DM as this is sufficiently well determined from optimizing the signal to noise of the individual MeerKAT pulses.

Archival radio searches

VLITE

The VLITE53,54 is a commensal instrument on the National Radio Astronomy Observatory’s Karl G. Jansky Very Large Array that records and correlates data across a 64 MHz bandwidth at a central frequency of 340 MHz. Since 20 July 2017, VLITE has been operating on up to 18 antennas during nearly all regular VLA operations, accumulating roughly 6,000 h of data per year. An automated calibration and imaging pipeline53 processes all VLITE data, producing final calibrated visibility datasets and self-calibrated images. These images and associated META data are then passed through the VLITE Database Pipeline55 to populate a Structured Query Language database containing catalogued sources.

Using the VLITE Database Pipeline, we searched for all VLITE datasets that contain the position of ASKAP J1935+2148 within 2.5° from the phase centre of the VLITE observations taken when the VLA was in its A and B configurations. We note that the half-power radius of the VLITE primary-beam response is ~1.25°; however, the system is sensitive to sources well beyond this radius. We identified 124 VLITE datasets observed between 9 November 2017 and 22 August 2023. From these, we selected all observations with a length of at least 15 min, for a total of 26 observations in A configuration (angular resolution ~5″) and 10 observations in B configuration (resolution ~20″). None of these 36 datasets are targeted observations of ASKAP J1935+2148, rather the source position ranges between a radius of 1.1° to 2.2° from the pointing centre of the VLITE observation.

To search for possible 340 MHz emission from ASKAP J1935+2148 in the VLITE data, we first subtracted all known continuum sources from the self-calibrated visibilities of each observation, we then phase shifted the data to the position of ASKAP J1935+2148 using chgcentre56, and finally we made a time series of dirty images of the target at 10 s and 4 s intervals using WSClean56. The primary-beam corrected noise in the 10 s snapshots ranges on average from 41 mJy per beam (when the position of ASKAP J1935+2148 is 1.1° away from the VLITE phase centre) to 130 mJy per beam (at 2.2°). In the 4 s snapshots, the average primary-beam corrected noise ranges between 62 mJy per beam (at 1.1° from the phase centre) and 181 mJy per beam (at 2.2°). No obvious pulses from ASKAP J1935+2148 were detected.

VLA and GMRT

The archives of the VLA and the GMRT were searched for data in which the position of ASKAP J1935+2148 lies within the observation field of view. VLA P-band (~325 MHz) observations of PSR 1937+21, 1.1° away, were made on 25 September 2013, 28/29 November 2013, 30 November/1 December 2013, 1/2 December 2013 and 7 July 2014 with the array in the B and A configurations (~20″ and ~5″ resolutions), respectively. The 25 September 2013 observation was on-source for 42 min except for short calibrator scans intermixed, while the other 2013 observations consisted of three 6 min scans separated by 1 h, and the 2014 observation consisted of ten 6 min scans spread over 4 h. Calibration and imaging were performed using the Astronomical Imaging Processing System (AIPS)57. Amplitude and phase calibrations were both performed using 3C48, as no separate phase calibrator was observed. Imaging was performed with the AIPS task, IMAGR. The field was self-calibrated on a wide-field image using 19 facets to cover the ~3° (full-width at half-maximum) field of view. Ten-second integration snapshot images were made after subtracting the deep image clean components from the UV data using the AIPS task, UVSUB. The root mean square (RMS) noise of the 10 s snapshot images at the location of ASKAP J1935+2148 was typically ~15 mJy per beam after applying a 1.5× correction for primary-beam attenuation.

Two L-band (~1.4 GHz) observations on 28–29 May 2020 (GMRT; 2.5 h; ~2″ resolution) and 27 June 2021 (VLA; 1 h; C configuration; ~10″ resolution) targeted SGR 1935+2154 located only 0.1° away from ASKAP J1935+2148. A similar analysis in AIPS was followed as with the P-band observations except that wide-field imaging was not necessary due to the much smaller ~0.5° (full-width at half-maximum) fields of view. The amplitude and phase calibrators were 3C48 and 1822-096 (GMRT) and 3C286 and J1925+2106 (VLA). The RMS noise of the 5.4 s (GMRT) and 5.0 s (VLA) snapshot images at the location of ASKAP J1935+2148 were both typically ~0.5 mJy per beam after applying a 1.1× correction for primary-beam attenuation. No obvious pulses from ASKAP J1935+2148 were detected.

X-ray searches

For Chandra, we identified 5 Chandra observations for a total of 157.7 ks of exposure as listed in Extended Data Table 2. For all observations, ASKAP J1935+2148 was located on a front-illuminated charge-coupled device (CCD): ACIS-S2 for 21305/21306 and ACIS-S4 for the remainder.

The observations were analysed and combined using CIAO version 4.15.1, with CALDB 4.10.258. We first examined all observations individually for background flares by looking visually at the summed light curves. No flares were identified. We reprocessed the data to level 2 using a consistent calibration database, and reprojected the data to a common tangent point. We then combined the reprojected observations to create an exposure-corrected image. No source was found within 2″ of ASKAP J1935+2148. We also looked at the individual reprojected event files. For each file we computed the number of events within a 2″ radius of ASKAP J1935+2148 along with the background rate determined from all of the counts on the appropriate CCD between 0.3 keV and 10 keV. A total of 6 counts were found near the position of ASKAP J1935+2148, but the mean background rate predicts 3.8 counts, and the chance of getting ≥6 counts is 9.2%. Therefore, we do not consider this a detection and place a 3σ upper limit of 10 counts in 157.7 ks or a count-rate limit of 6.3 × 10−5 counts per second (0.3–10 keV).

On the basis of the observed DM of 145.8 ± 3.5 pc cm−3, we predict an absorbing column density of NH ≈ 4 × 1021 cm−2 (ref. 59). We computed unabsorbed flux limits for two spectral models: a blackbody with kT = 0.3 keV (appropriate for thermal emission from a young pulsar/magnetar), and a power law with index Γ = 2 (appropriate for non-thermal emission from an energetic pulsar/magnetar), following ref. 34. This was done using PIMMS (https://cxc.harvard.edu/toolkit/pimms.jsp), where we assumed a response appropriate for Chandra cycle 22 and used the ACIS-I CCDs in place of the front-illuminated ACIS-S CCDs. With the blackbody model, we infer an unabsorbed flux limit of FBB < 1.3 × 10−15 erg s−1 cm−2, while with the power-law model we infer an unabsorbed flux limit of FPL < 1.7 × 10−15 erg s−1 cm−2. These imply luminosity limits of about \(4\times 1{0}^{30}{d}_{4.85}^{2}\,{{{\rm{erg}}}}\,{{{{\rm{s}}}}}^{-1}\). Overall, the Chandra observations lead to very low limits regarding the time-averaged X-ray flux.

To search for any flaring from ASKAP J1935+2148, we used extensive observations from Swift, with 291 individual visits using the X-ray Telescope20 lasting 5–2,600 s from December 2010 to December 2022 for a total exposure time of 302.4 ks (exposure corrected). We combined the individual barycentred exposures using HEADAS60 on SciServer61 into a single exposure-corrected dataset. We see no sources in the summed dataset at the position of ASKAP J1935+2148: there are 33 counts within 15″ of ASKAP J1935+2148, but the background expectation computed using an annulus from 30″ to 60″ is 35.4 counts, so we estimate a 3σ upper limit of 53.0 counts or a count-rate limit of 1.7 × 10−4 counts per second (0.15–10 keV). Using the same spectral models as above, we limit the unabsorbed flux (0.3–10 keV) to be FBB < 7.4 × 10−15 erg s−1 cm−2 and FPL < 1.1 × 10−14 erg s−1 cm−2. These are much less constraining than the corresponding Chandra limits.

Optical and near-infrared searches

The position of ASKAP J1935+2148 was observed by the VLT using the near-infrared HAWK-I21 imager. There were a number of observations of SGR 1935+2154 that placed ASKAP J1935+2148 near the edge of the field of view; we found the observations on 2 April 2015 to be the most useful. These included 300 s exposures in the J, H and Ks bands (1.2 μm, 1.6 μm and 2.1 μm). However, even these exposures were not ideal, with weightmap values only 20% of the peak at the position of ASKAP J1935+2148. Nonetheless, the collecting area of VLT makes them valuable.

We show a RGB composite of the field around ASKAP J1935+2148 in Fig. 3. It is clear that the source is near the edge of the field, and is barely covered by the H-band image. Coverage in the J and Ks bands is better. There is a source within a 1.5″ radius around ASKAP J1935+2148 and has J = 18.4 ± 0.1 mag, H = 17.3 ± 0.1 mag and Ks = 17.1 ± 0.1 mag (Vega). We compute the chance of finding a source randomly (drawn from the background) in the Ks image with magnitude brighter than or equal to this value is 5% (that is, ~2σ association). This suggests that the association between the near-infrared source and ASKAP J1935+2148 is not statistically significant. Otherwise, we infer 5σ upper limits of J > 21.4 mag, H > 20.5 mag and Ks > 19.8 mag. Aside from the deep VLT pointings, we examined images from Data Release 2 of the PS1 3π survey22. There is a source in the ‘stack’ catalogue that corresponds to the near-infrared source identified above. This source is catalogued as PSO J293.7711+21.8119. The source is not detected in the g, r or i bands, and has detections only in z (22.0 ± 0.3 mag) and y (20.4 ± 0.1). For the other bands and for the rest of the error region, we adopt the standard PS1 stack upper limits g > 23.3, r > 23.2, i > 23.1, z > 22.3 and y > 21.3.

Search for persistent radio emission

To ascertain whether there is an unpulsed radio component that might be attributed to a wind nebula, or perhaps indicate emission of a non-neutron-star origin, we imaged the ASKAP visibility data. In a stacked ASKAP deep image at 887.5 MHz, if we subtract the mean background emission, there is no detection above 3σ with an RMS of 25 μJy per beam. However, the location of the source at the edge of the supernova remnant in Extended Data Fig. 2 makes it is difficult to determine exactly what background emission to subtract. Therefore, we are unable to confirm the presence of potential (possibly faint) persistent continuum radio emission in either the MeerKAT or ASKAP data.

Faraday conversion

We tested whether the large circular polarization fraction seen in the MeerKAT beamformed data could be due to Faraday conversion using a simple phenomenological model62, where the polarization vector is modelled as a series of frequency-dependent rotations on the Poincaré sphere. However, we failed to recover any notable frequency dependence to the circular polarization. This indicates that it is either intrinsic to the emission mechanism or arises from a more complex propagation effect such as the partially coherent addition of linearly polarized modes63.

Model constraints

White dwarf

We examine the potential for the optical/near-infrared data described above to constrain white-dwarf scenarios for the source. We used the synthetic photometry (see https://www.astro.umontreal.ca/~bergeron/CoolingModels/) of refs. 64,65,66 together with the three-dimensional extinction model of ref. 67 to compute distance constraints as a function of effective temperature for hydrogen-atmosphere (DA) white dwarfs with masses 0.6 M⊙ and 1.0 M⊙, representing standard and massive white dwarfs, respectively (using helium-atmosphere DB white dwarfs does not change the conclusions). We did two analyses, one where we modelled the potential near-infrared counterpart and one where we treated the source as non-detections. Note that there are large degeneracies involved: extinction and effective temperature are highly degenerate, and other quantities such as mass (and hence radius) degenerate with distance. When considering the potential near-infrared counterpart as correct, and given the sparse data that would all be on the Rayleigh–Jeans tail, we unsurprisingly found a plausible fit to the VLT data with effective temperature, Teff ≈ 15,000 K and distance ~6 kpc (implying reddening E(B − V) ≈ 2.8). However, the implied radius is ~0.8 R⊙, leading us to conclude that this source cannot be expected by standard white-dwarf models.

Considering only upper limits (so assuming that the source PSO J293.7711+21.8119 is not associated with ASKAP J1935+2148), we find that a white dwarf with Teff < 30,000 K could be present at distances >1 kpc. Despite their greater sensitivity, we found the VLT data generally less constraining than the PS1 data given the range of effective temperatures considered. Given the implied average DM distance of 4.85 kpc from the NE2001 and YMW16 models, we do not consider the limits described here especially constraining. Under the framework of coherent radio emission from pair production, the compactness of the source can be related to the period and magnetic field so that we can estimate the minimum radius of the source30 using

$$R \gtrsim 4\times 1{0}^{9}{\left(\frac{{Q}_{{\mathrm{c}}}}{10}\right)}^{4/17}\,{\left(\frac{P}{1,000\,{{{\rm{s}}}}}\right)}^{13/17}\,{\left(\frac{B}{1{0}^{9}\,{{{\rm{G}}}}}\right)}^{-8/17}\,{{{\rm{cm}}}},$$
(1)

where P is the period in seconds, B is the magnetic field in gauss and Qc = ρc/R is the dimensionless characteristic field curvature radius in which the curvature radius near the polar cap is assumed to be ρc ≈ 10R. The conventional emission model for any compact object to emit dipole radiation assumes the existence of a vacuum gap above the polar cap. To sustain pair production, the potential difference across this gap must be sufficiently large and this is no longer possible beyond the classic death line37. As a result, pair production and consequently, radio emission ceases. The relation above therefore encodes radio death-line physics due to requirements on pair-cascade production and provides a lower limit on the compactness of an object to sustain this emission. A choice of Qc ≫ 10 is commensurate with the expected small polar cap size for a P ≈ 1,000 s rotator. Assuming 10 ≤ Qc ≤ 105 and B = 109 G, we can rule out an isolated magnetic white-dwarf origin for the observed emission as the estimated lower limit on the radius of 0.14 R⊙, even in the case of Qc = 10, is much too large for a white dwarf (Extended Data Fig. 5).

Neutron star

It has been proposed that bright coherent radio bursts can be produced by highly magnetized neutron stars that have attained long rotation periods (few tens to a few thousands of seconds), called ultralong period magnetars. Typically, magnetars have quiescent X-ray luminosities anywhere between 1031 erg s−1 and 1036 erg s−1 (refs. 19,68) regardless of radio emission (typically they are brighter in X-rays following outbursts that lead to radio emission), and so our deep X-ray limits from searching archival X-ray data challenge the magnetar interpretation. However, there are subclasses of magnetars with considerably weaker X-ray emission, <1030 erg s−1, such as the ‘low field’ magnetars whose69 spin-down inferred fields are ~1013 G, but where local X-ray absorption features suggest much higher localized fields70,71. If indeed ASKAP J1935+2148 and similar sources are part of another emerging subclass of magnetars, the quiescent X-ray luminosities (which are attributed to the decay of large-scale dipole magnetic fields) may be lower. If that is the case, it would also explain the location of ASKAP J1935+2148 as magnetars are typically expected to be young objects that lie in the Galactic Plane11. Combining all these observational aspects, ASKAP J1935+2148, is probably part of an older population of magnetars with long spin periods and low X-ray luminosities, but magnetized enough to be able to produce coherent radio emission. It is important that we probe this hitherto unexplored region of the neutron-star parameter space to get a complete picture of the evolution of neutron stars, and this may an important source to do so.