Discovery of ASKAP J173608.2–321635 as a Highly Polarized Transient Point Source with the Australian SKA Pathfinder

ASKAP J173608.2−321635 was first discovered as a compact radio source in a transients search of VAST-P1 data (Project Code AS107) using the VAST transient detection pipeline (Figure 1). It was detected in the adjacent fields 1724−31A and 1752−31A, observed 13 times between 2019 April 28 and 2020 August 29. The VAST-P1 survey incorporates the Rapid ASKAP Continuum Survey (RACS, Project Code AS110; McConnell et al. 2020) as its first epoch. Both RACS and VAST-P1 were conducted at a central frequency of 888 MHz with a bandwidth of 288 MHz and they shared the same tiling footprints. The integration time for RACS was 15 minutes while that for VAST-P1 was 12 minutes, achieving an rms noise of 0.36 mJy beam⁻¹ and 0.40 mJy beam⁻¹ for regions near the GC, respectively. Details of these survey observations and data reduction are given by McConnell et al. (2020) and Murphy et al. (2021).

Zoom In Zoom Out Reset image size

Figure 2 shows the full radio lightcurve of ASKAP J173608.2−321635, as well as the fractional circular polarization. Other than variability, ASKAP J173608.2−321635 was highly circularly polarized with a fractional polarization ranging from 20% to 30% in VAST-P1 bright detections (see Figure 2, lower left panel).

Zoom In Zoom Out Reset image size

There were four additional ASKAP observations that cover our source (Table 1). These observations were calibrated using PKS B1934–638 for both the flux density scale and the instrumental bandpass. All observations were processed using standard procedures in the ASKAPsoft package (Guzman et al. 2019). We note that there was a ∼50 mJy detection in a 10-hour observation at 943 MHz on 2020 November 1. However, the systematic error is high due to the source being located near the edge of the beam.

To check for any shorter timescale variability we imaged the source using data from the 2020 November 1 ASKAP observation with an integration time of 15 min (resulting in 40 images in total). This lightcurve showed a relatively low modulation index (standard deviation divided by the mean) of ∼13%, and had a reduced χ² relative to a constant model (a measure of the significance of the variability, see, e.g., Swinbank et al. 2015) of 1.6 for 39 degrees of freedom (40 observations minus one parameter for the mean). Overall we did not see any evidence for hour-scale variability (Figure 3).

Zoom In Zoom Out Reset image size

Motivated by the possibility that ASKAP J173608.2−321635 is a pulsar, we conducted follow-up observations with the 64 m Parkes telescope of ASKAP J173608.2−321635 on 2020 April 20 and 2020 July 29 using the pulsar searching mode with the Ultra-Wideband Low (UWL) receiver (Hobbs et al. 2020), which provides simultaneous frequency coverage from 704 to 4032 MHz. Each observation was 30 minutes with 32 μs time resolution and high frequency resolution (1024 channels per 128 MHz subband). We used Presto (Ransom 2001) to perform a standard pulsar search. We found no candidates in a search of dispersion measures (DMs) spanning 0–3000 pc cm⁻³, corresponding to 25 kpc based on the YMW16 electron-density model (Yao et al. 2017, hereafter YMW16) or about two times the highest DM for pulsars discovered to date (e.g., Shannon & Johnston 2013), period <25 s and accelerations up to ∼20 m s⁻² (assuming a pulsation period of 1 ms). We also found no single pulse above a signal-to-noise ratio (S/N) of 8 using the single pulse search procedure for Presto. However, the lack of simultaneous imaging meant we cannot determine whether the source was radio-loud during these observations. These nondetections (with an upper limit of ∼0.05 mJy, assuming the duty cycle of the pulsar (W/P) to be 10%) therefore do not rule out the presence of a pulsar.

To simultaneously search for pulsed and continuum emission from ASKAP J173608.2−321635, we observed it using the MeerKAT radio telescope with a central frequency of 1.28 GHz and a two-week cadence starting from 2020 November 19 (project code DDT-20201005-DK-01). Each observation had 12 minutes on the target, achieving an rms noise of 40 μJy beam⁻¹. Imaging and pulsar searching were performed simultaneously in all MeerKAT observations. We used PKS J1830–3602 for bandpass, flux density scale and phase calibration. We reduced the image data using Oxkat ²⁰ (v1.0; Heywood 2020), where the Common Astronomy Software Applications (CASA; McMullin et al. 2007) package and Tricolour ²¹ were used for measurement sets splitting, cross calibration, self-calibration, and flagging, and Wsclean (Offringa et al. 2014) was used for continuum imaging.

We did not detect any source to a 5σ limit of 0.04 mJy in the first five epochs. However, we detected a source in our observation on 2021 February 7 at a flux density of 5.67 ± 0.04 mJy but did not detect any pulsations. The best-fit position of the source is: (J2000) R.A. 17^h36^m0819 ± 003, decl. $-32^\circ 16^{\prime} 35\buildrel{\prime\prime}\over{.} 0\pm 0\buildrel{\prime\prime}\over{.} 3$ with Galactic coordinates l, b = (35608, − 004) based on the MeerKAT detection, where the uncertainties are based on a comparison of the positions of field sources to their RACS matches. We imaged the source with an integration time of 16 s (resulting in 40 images in total). The lightcurve showed a relatively low modulation index of ∼4% and had a reduced χ² of 0.8 for 39 degrees of freedom, with no evidence for minute-scale variability (Figure 3).

The source was moderately circularly polarized (V/I = +8%) and had a steep radio spectrum within the bandpass (α = −2.7 ± 0.1 ²² , where S_ν ∝ ν^α ). We also found the source to be highly linearly polarized (∣L∣/I ∼ 80%) with a moderately low Faraday rotation measure (RM) of −11.2 ± 0.8 rad m⁻². The source also exhibited depolarization behavior toward lower frequencies: the fractional total polarization is nearly 100% at 1.6 GHz but only ∼20% at 0.9 GHz (Figure 4). We performed further tests to verify the polarization and RM variability, as discussed in the Appendix.

Zoom In Zoom Out Reset image size

Further radio observations showed a very rapid decline with an exponential timescale of ∼26 hr (Figure 2 inset). Our ASKAP observation 20 hr after the first MeerKAT detection gave a flux density of 2.4 ± 0.3 mJy at 1.3 GHz. Two further MeerKAT observations over the following days demonstrated that the source continued to fade exponentially, while the spectral shape remained similar (α = −3.4 ± 0.3). We found the source was still highly linearly polarized in these observations, although the RM changed significantly, from −11.2 ± 0.8 rad m⁻² on 2021 February 7 to −63.3 ± 1.5 rad m⁻² on 2021 February 9. The ionosphere usually contributes to Faraday rotation of order ∼ 1 rad m⁻² (Sotomayor-Beltran et al. 2013), which can potentially cause RM variations between epochs. We used IonFR ²³ to model the ionospheric Faraday depth at the dates of the observations. The ionospheric Faraday rotation is +0.65 ± 0.05 rad m⁻² and +0.75 ± 0.06 rad m⁻² on 2021 February 7 and 2021 February 9, respectively. The corrected RM of the source is therefore −11.8 ± 0.8 rad m⁻² and −64.0 ± 1.5 rad m⁻² on these days, after ionospheric RM corrections. The intrinsic polarization angle was consistent between the epochs (see justifications in the Appendix).

We also obtained three 12 minute observations in the ultra-high-frequency band (UHF; 544–1088 MHz) with MeerKAT, about 100 hours after the first MeerKAT detection. There was no detection in these single observations, but there was a ∼5σ detection when all three were summed coherently (see blue diamonds in Figure 2). This UHF-band detection is a factor of two higher than what we expected from the exponential decay (we corrected the UHF-band detection to 1.3 GHz assuming a spectral index of α = −2.7), suggesting that the spectrum may have steepened to α < −4 or that the decay slowed.

During imaging observations with MeerKAT, the FBFUSE (Filterbanking Beamformer User Supplied Equipment; Barr 2017) instrument was used to produce high-time-resolution Stokes-I beams to enable pulsar and fast-transient searching. At both the L and UHF band, FBFUSE was configured to produce a tiling pattern of seven coherent beams with the central beam positioned at (J2000) R.A. 17^h36^m0820, decl. $-32^\circ 16^{\prime} 33\buildrel{\prime\prime}\over{.} 0$. The beams were arranged in a close-packed hexagonal grid with an overlap at their 70% power points (see Chen et al. 2021 for detail of FBFUSE beam tiling). At the L band FBFUSE produced 4096-channel data covering the 856 MHz band with a time resolution of 76.56 μs. At the UHF band the instrument produced 4096-channel data covering the 544 MHz band with a time resolution of 120.47 μs.

Data streams from FBFUSE were recorded to disk on the Accelerated Pulsar Search User Supplied Equipment (APSUSE; Barr 2017) cluster. The data were dedispersed to dispersion measures in the range 0–2000 pc cm⁻³ at the UHF band and 0–3000 pc cm⁻³ at the L band, with the different maximum DMs chosen to have roughly constant scattering timescales between the two bands. The resultant trials were searched for periodicities up to 10 s using the GPU-accelerated Peasoup ²⁴ software with the resultant candidates folded modulo the detected periodicities using PulsarX. ²⁵ To retain sensitivity to binary systems, the data were time-domain resampled (Johnston & Kulkarni 1991) to constant acceleration values between −150 and 150 m s⁻² before searching. Folded candidate signals were inspected by eye. No significant pulsed emission was detected above a signal-to-noise threshold of 9.

The MeerTRAP real-time single-pulse pipeline running on the Transients User Supplied Equipment (TUSE) instrument (B. W. Stappers et al. 2021, in preparation) was run in parallel with all of the MeerKAT observations. It operated on the same central beam that the pulsar search described above with a time resolution of 306.24 μs for the L-band observations and 361.4 μs for the UHF-band observations. Single pulses that are greater than an S/N limit of 8 were searched for overdispersion measures from 23–5000 pc cm⁻³ in the L band and 23–1500 pc cm⁻³ in the UHF band over a range of widths from the time resolution up to 196 ms and 231 ms for the two frequencies, respectively. No astrophysical pulses were detected above the S/N threshold.

After our ASKAP and MeerKAT monitoring observations ended, we observed ASKAP J173608.2−321635 with the Australia Telescope Compact Array (ATCA) in three bands (centered at 2.1 GHz, 5.5 GHz, and 9.0 GHz) for 80 minutes each on 2021 April 25 (project code: C3431). The observation was calibrated using PKS B1934−638 for the flux density scale and the instrumental bandpass. PMN J1733−3722 was used for phase calibration.

We used Miriad (Sault et al. 1995) to perform the data calibration and Casa to perform the continuum imaging. We detected a source with a flux density of 4.41 ± 0.14 mJy at 2.1 GHz. We did not find any detection at 5.5 GHz or 9.0 GHz, which places 3σ upper limits of 78 μJy beam⁻¹ and 60 μJy beam⁻¹ at 5.5 GHz and 9.0 GHz, respectively. The nondetection at higher frequency (5.5 GHz) constrains the spectral index to be α < −4.2. We measured the spectral index to be α = −5.6 ± 0.1 across the L-band (2.1 GHz) bandpass (Figure 5), which is consistent with the constraints from the nondetection at 5.5 GHz. The source was moderately circularly polarized, with V/I ∼ +6%, which is consistent with the MeerKAT observation that fractional circular polarization is lower at higher frequencies (Figure 4).

Zoom In Zoom Out Reset image size

We identified archival observations covering ASKAP J173608.2−321635 with the Neil Gehrels Swift Observatory (Swift; Gehrels et al. 2004), restricting observations to those using the X-Ray Telescope (XRT; Burrows et al. 2005) in the photon counting mode. We used four observations between 2012 February 1 and 2012 September 9, with a summed exposure time of 2.3 ks. There was no source within 15'' of ASKAP J173608.2−321635, and we determine a 95% count-rate upper limit of 8.4 × 10⁻⁴ s⁻¹ (over the default energy range of 0.2–10 keV).

Following the MeerKAT detections of ASKAP J173608.2−321635, we were awarded Director's Discretionary Time observations with Swift (observation IDs 00014071001 and 00014071002 ). We obtained 1.7 ks on 2021 February 10.95 and another 0.8 ks on 2021 February 11.35. There was one count within 15'' of ASKAP J173608.2−321635, but this is consistent with the background (mean expectation with 15'' of 0.3 counts). So, we set an upper limit of 1.2 × 10⁻³ s⁻¹ (0.2–10 keV). We estimated the upper limit of H i column density for the position of our source based on the H i 4π survey (HI4PI Collaboration et al. 2016) using the HEASARC web-based PIMMS to be 1.59 × 10²² cm⁻² (through the entire Galaxy). Assuming a power-law photon index of Γ = 2.0 (Hyman et al. 2021), the nondetection in Swift observations yields an upper limit on the unabsorbed flux (0.3–8 keV) of 2.0 × 10⁻¹³ erg cm⁻² s⁻¹. The upper limit for the X-ray luminosity at a distance of d is $\sim 2.4\times {10}^{33}{\left(d/10\,\mathrm{kpc}\right)}^{2}$ erg s⁻².

Finally, we were awarded Director's Discretionary Time with the Chandra X-Ray Observatory. We used the back-illuminated ACIS-S3 detector with the thin filter, and the 1/8 subarray to maintain subsecond temporal resolution. ASKAP J173608.2−321635 was observed on 2021 February 17.61 for 25.1 ks (observation ID 24966). We filtered the data to 0.3–10 keV. There are 0 events within 1'', and based on the observed background rate we set a 95% upper limit of 1.0 × 10⁻⁴ s⁻¹. Likewise, we estimate the upper limit of the X-ray luminosity (0.3–8 keV) based on the Chandra nondetection to be $\sim 5.0\times {10}^{31}{\left(d/10\,\mathrm{kpc}\right)}^{2}$ erg s⁻¹.

We searched for near-IR counterparts in the VISTA Variable in the Via Lactea Survey (VVV; Minniti et al. 2010). There is no counterpart visible in the VVV DR2 catalog. We find 3σ upper limits of J > 19.25, H > 17.65 mag, and K_s > 16.70 mag from VVV within a 25 radius (corresponding to a 5σ positional error).

We observed the source using Gemini Flamingos-2 in the J-band (1.2 μm) for 40 minutes on 2021 April 28 and 2021 April 29, and in the K_s band (2.15 μm) for 18.5 minutes on 2021 May 24 (project code GS-2021A-FT-210). We used Gemini Dragons (Labrie et al. 2019) to reduce the data and Sextractor (Bertin & Arnouts 1996) to perform the photometry.

We used the VVV catalog as both astrometric and photometric references to correct the Gemini data. For astrometry, we used 340 sources that we identified as not blended or badly saturated for the J band and 96 sources for the K_s band. The uncertainty is ≈ 0.15''in each coordinate. For photometry, we used fewer sources to avoid sources that showed signs of saturation or nonlinearity. We used 270 sources in the J band and 90 sources in the K_s band. We estimated zero-point uncertainties of 0.02 mag for the J band and 0.04 mag for the K_s band. The seeing of both observations was ∼0 7.

There is a faint source within 25 of the radio position with J = 20.8 ± 0.2 mag and K_s = 17.6 ± 0.1 mag. This infrared source is just within the 5σ error circle of the radio position (Figure 6), therefore we consider it unlikely to be associated with the radio source, but we examine this in more detail in Section 3.1. Finally, just to the south of that source is a fainter source visible in both J and K_s bands but with magnitudes at or fainter than our 3σ limit. We are unable to measure its properties reliably, but given the density of such sources in the image we do not believe the association to be significant.

Zoom In Zoom Out Reset image size

This source was not detected in previous radio surveys including the quick look images from the Karl G. Jansky Very Large Array Sky Survey (VLASS; Lacy et al. 2020), the TIFR GMRT Sky Survey (TGSS; Intema et al. 2017), the GaLactic, and Extragalactic All-sky MWA (GLEAM; Wayth et al. 2015; Hurley-Walker et al. 2017), the NRAO VLA Sky Survey (NVSS; Condon et al. 1998), and the second epoch Molonglo Galactic Plane Survey (MGPS-2; Murphy et al. 2007). These limits are included in Table 1. We have also searched for any archival VLA and ATCA data but did not find any other observation that covers our source.

Low-mass flare stars and chromospherically-active binaries such as RS CVns often show polarized flares (e.g., Mutel et al. 1987; Zic et al. 2019). We show the color–magnitude diagram for sources within the field of our Gemini observation in Figure 8, with additional sources from VVV. We investigate the possibility that the Gemini source in Figure 6 is a nearby cool dwarf associated with ASKAP J173608.2−321635 (RS CVns would be far brighter, e.g., Driessen et al. 2020). According to Pecaut & Mamajek (2013) ²⁸ cool dwarfs (spectral type of M/L/T/Y) have typical colors of J − K_s from −1.0 to 1.0 and absolute magnitudes in the K_s band of ${M}_{{K}_{s}}\gtrsim 6$. For a cool dwarf with observed color of J − K_s ≈ 3.0, it would need at least an extinction in the V band of A_V ≈ 12 mag (we use the extinction coefficients in Yuan et al. 2013). With an average extinction of A_V /d ≈ 1.8 mag kpc⁻¹ (Whittet 1992), it requires a source at a distance of ∼7 kpc, which implies the magnitude in the K_s band would be ≳21 mag (including the effect of extinction). As our source is about 3 magnitudes brighter than this limit, it is hard for this source to be a cool dwarf: more likely is a more distant red giant branch/red clump star. In general it does not stand out at all compared to the surrounding population, suggesting that it is not a unique object. We come to a similar but less robust conclusion about the fainter object in Figure 6, which we cannot measure reliably (also see Kaplan et al. 2008).

Zoom In Zoom Out Reset image size

The high radio flux density of ASKAP J173608.2−321635, together with nondetections at X-ray and near-IR wavelengths, also makes a stellar interpretation unlikely. X-ray and radio luminosities for various types of active stars are typically correlated (the Güdel-Benz relation; Güdel & Benz 1993; Driessen et al. 2020). In contrast, ASKAP J173608.2−321635 has an X-ray upper limit too low by at least two orders of magnitude. Even for ultracool dwarfs (known to be radio over-luminous relative to their X-ray luminosity; e.g., Williams et al. 2014), the X-ray limit of our source is lower than most of the ultracool dwarfs (Figure 9). ²⁹

Zoom In Zoom Out Reset image size

Similarly, based on the brightest possible object that we cannot rule out in infrared (excluding the object in Figure 6), we measure J > 20.8 mag from our Gemini observation. Empirically, we can examine different types of active stars with circularly polarized emission (Figure 10). The vast majority of stars across different types (L/T dwarfs, magnetic CVs, and radio flux-limited samples) have radio-to-near-IR flux ratios of ≲1. For the radio-discovered T dwarf BDR J1750 + 3809, the ratio is near 10. Except for the youngest, most energetic pulsars, this ratio is typically ≳10³ (e.g., Zyuzin et al. 2016). ASKAP J173608.2−321635 itself has a ratio >10³, depending on the radio state.

Zoom In Zoom Out Reset image size

We can do the same analysis a different way, based on the ratio of radio to bolometric flux from ultracool dwarfs. We determined a lower limit on the distance of a stellar/substellar counterpart (spectral type from late L to mid-M) to be ∼150–1400 pc based on the observed population of ultracool dwarfs (Reid et al. 2008). At this distance, we would expect low extinction, about 0.5 mag in the J band. Based on this lower limit on the distance (applying the extinction correction), we calculated upper limits on the radio flux density at 888 MHz to be <0.3–0.6 μJy, assuming L_radio/L_bol = 10⁻⁷ (which is the typical value for M dwarfs, see Berger et al. 2010). The ratio of radio luminosity to bolometric luminosity for later L dwarfs can be as high as 10⁻⁵: for BDR J1750 + 3809, it can reach 2 × 10⁻⁵. The limit would give an expected radio flux density of <120 μJy. Even for a slightly beamed emission (such as for Jupiter, Burningham et al. 2016), the expected flux density would be ≲0.9 mJy. This is considerably lower than our measured values of ∼10 mJy, suggesting that ASKAP J173608.2−321635 is either a star with an extreme near-IR to radio ratio or another kind of source entirely.

To summarize, we excluded ASKAP J173608.2−321635 as a star based on the following:

• 1.  
  Compared to its color (J − K_s ), the IR source that we detect is too bright in the K_s band (Figure 8).
• 2.  
  The ratio of X-ray luminosity to radio luminosity is too low for stars (Figure 9).
• 3.  
  The source is too bright in radio compared to the J band (Figure 10).

Though we found no pulsations in our data, the high degree of polarization and steep spectrum suggest the source may be a pulsar. We can use our MeerKAT observations to constrain the pulsar-like properties of ASKAP J173608.2−321635. The expected signal-to-noise ratio of a pulsar at the beam center can be estimated as (Lorimer & Kramer 2012):

$$\begin{eqnarray}&&{\rm{S}}/{{\rm{N}}}_{\exp }=\displaystyle \frac{{SG}\sqrt{{N}_{\mathrm{pol}}{\tau }_{\mathrm{obs}}{\rm{\Delta }}\nu }}{{T}_{\mathrm{sys}}\beta }\sqrt{\displaystyle \frac{P-W}{W}},\end{eqnarray} \tag{ 1 }$$

where S is the flux density of the pulsar, G = 2.8 K Jy⁻¹ is the gain of the MeerKAT telescope, N_pol = 2 is the number of polarizations recorded, τ_obs = 700 s is the length of the observation, Δν ∼ 856 MHz is the bandwidth, T_sys ∼ 40 K is the system temperature (which includes the sky temperature in this direction), β is a correction factor due to downsampling, W is the pulse width of the pulsar, and P is the period of pulsar. The effective pulse width is a combination of its intrinsic pulse width, pulse broadening due to dispersion, and scattering:

$$\begin{eqnarray}&&{W}_{{\rm{e}}}=\sqrt{{W}^{2}+\delta {t}_{\mathrm{disp}}^{2}+\delta {t}_{\mathrm{scat}}^{2}},\end{eqnarray} \tag{ 2 }$$

where W_e is the effective pulse width, $\delta {t}_{\mathrm{disp}}=8.3\,\times {10}^{6}\mathrm{DM}\,{\nu }_{\mathrm{MHz}}^{-3}\,\delta \nu \,\mathrm{ms}$ (δ ν is the channel bandwidth in units of MHz) is the smearing time due to dispersion across a channel observed at frequency ν_MHz, and δ t_scat is the smearing time due to scattering. We considered scattering as a function of DM based on Bhat et al. (2004); note that this is consistent with the very high degree of scattering from Camilo et al. (2021).

A wide effective pulse width can reduce the pulsation S/N. For example, Hyman et al. (2021) argues that C1709–3918 and C1748–2827, with steep spectra, 10%–20% circular polarization, but no pulsations detected, may be pulsars with scatter-broadened pulses. At the most conservative, if ASKAP J173608.2−321635 is a pulsar with 1 ms pulsation period, considering the effect of dispersion and scattering, the nondetection in our MeerKAT pulsar search with S/N = 9 threshold suggests the duty cycle (W/P) of the pulsar would be >99% for a source with a DM below 200 pc cm⁻³ (∼3 kpc based on YMW16). For longer pulse periods we would have a duty cycle limit of >99% at DMs up to ≲1000 pc cm⁻³ (∼6 kpc based on YMW16). Compared to pulsars in ATNF pulsar catalog, the highest duty cycle is ∼80% (see Figure 11). However, at the highest DMs considered in our search (up to 3000 pc cm⁻³), we would not be sensitive to even the longest period pulsars with typical scattering behavior. Observing at higher frequencies can help us minimize the effect of scattering. We will also employ fast folding algorithms (Staelin 1969) to search for longer periods when they become available for MeerKAT data.

Zoom In Zoom Out Reset image size

An alternative way to smear pulsations would be through orbital acceleration in a tight binary (e.g., Maan et al. 2018; de Gasperin et al. 2018). Our MeerKAT searches were shorter than the Parkes observations, so most binary orbits would not be too smeared out. Based on the range of accelerations searched, we exclude pulsars in a binary system with orbital period P_B ≳ 5 hr (assuming circular edge-on orbit, pulsar mass of 1.4 M_⊙, and companion mass of 0.1 M_⊙).

The decline in flux seen in our MeerKAT detections (lower right panel of Figure 2) is a factor of >10 faster than the initial detections seen with ASKAP (upper right panel in Figure 2), with several intermediate values between the "high" state and nondetections. This suggests that what we see is not "on versus off" behavior, like might be expected for a standard intermittent pulsar (Kramer et al. 2006; Lyne 2009).

These intermediate flux levels may also rule out effects such as random sampling of eclipses from a "black widow" (e.g., Fruchter et al. 1988) or "redback" (e.g., Roberts 2013) system, where radio pulses can be periodically eclipsed when the companion wind's obscures the line of sight, and this can both smear out pulsations (e.g., Stappers et al. 1996) and block the continuum flux (Broderick et al. 2016; Polzin et al. 2020). Typical orbital periods for those are <10 hr, so samples days or weeks apart would be very unlikely to end up during the short (<1 hr) ingress/egress periods. Some systems have been observed to have more complex flux density/eclipse variations (e.g., Polzin et al. 2020), but still generally not the large degree of continuum flux variability seen here.

Similarly, the precession of a pulsar will result in emission that comes and goes with a timescale of hours (e.g., Zhu & Xu 2006). The multiple detections with fading behavior over 50 hr in 2021 February and multiple nondetections over three months make eclipsing and precession unlikely interpretations. Hence, we conclude the observed emission is unlikely to be due to common pulsar-related origins.

Magnetars are neutron stars with extreme strong magnetic fields (up to ∼10¹⁵ G; Duncan & Thompson 1992; Kaspi & Beloborodov 2017). There are 31 known magnetars and magnetar candidates to date ³⁰ (Olausen & Kaspi 2014), but only five are detected in the radio as pulsars (Camilo et al. 2006, 2007; Levin et al. 2010; Eatough et al. 2013; Shannon & Johnston 2013; Rea et al. 2013; Karuppusamy et al. 2020; Lower et al. 2020). All the radio detections of magnetars happened during periods of X-ray outburst (Kaspi & Beloborodov 2017; Esposito et al. 2021), and faded eventually. Magnetars with confirmed radio pulsations show large pulse-to-pulse variability, including pulse morphology (Kaspi & Beloborodov 2017) and polarization (e.g., Dai et al. 2019). The persistent X-ray luminosity for these radio magnetars is typically ∼10³³ erg s⁻¹ (Rea et al. 2012), and can reach as high as ∼10³⁶ erg s⁻¹ during an outburst (e.g., Rea & Esposito 2011). Our upper limit based on the Chandra observation is comparable to the persistent luminosity of radio magnetars but much lower than those during outbursts (Figure 12). All radio magnetars show very high degrees of polarization, but their flat radio spectra (Shannon & Johnston 2013), in contrast to what we see for ASKAP J173608.2−321635, makes a magnetar an unlikely interpretation (although see Pearlman et al. 2018). Similarly, the rotation period of magnetars is typically ∼1–10 s (Kaspi & Beloborodov 2017), and that range is excluded based on our MeerKAT searches for most sources (DM ≲ 1000 pc cm⁻³, corresponding to ≲6 kpc based on YMW16; Figure 11). As we discussed earlier, pulsations can be smeared out due to scattering, but the GC magnetar PSR J1745−2900 has scattering of only 1.3 s at 1 GHz, considerably lower than that expected from DM models (Spitler et al. 2014; Pearlman et al. 2018), so we may actually be sensitive to higher DMs than Figure 11 implies. Regardless, higher radio frequency observations may help to rule out or confirm a magnetar origin. We noted that our search did not exclude sources with extreme long period, such as an ultra long period magnetar 1E 161348-5055.1 (with a rotation period of 6.67 hr, De Luca et al. 2006). Further monitoring observations may help us find such periodic activity.

Zoom In Zoom Out Reset image size

We now consider whether ASKAP J173608.2−321635 could be an X-ray binary or extragalactic transient. The polarization and extremely steep spectrum are inconsistent with expectations for emission produced by a steady jet (α ∼ 0 at these radio frequencies) such as from low-mass X-ray binaries, or optically thin ejecta (α ∼ −0.7) such as gamma-ray bursts (e.g., Fender 2006). The short timescale (∼days) of our source also rules out sources such as supernovae (∼years; e.g., Dubner & Giacani 2015) and tidal disruption events (∼months; e.g., Gezari 2021).

The large variability (∼100×) is inconsistent with standard diffractive scintillation, which has a modulation index of order unity (e.g., Cordes & Lazio 1991; Narayan 1992) for compact sources. Refractive scintillation will produce even less variability, particularly due to the proximity of ASKAP J173608.2−321635 to the GC. While the total electron column density is unclear due to the unknown source distance, the expected variability due to refractive scintillation at 900 MHz ranges from a few tens of percent if it is nearby, to as little as 2% if it is more distance (Walker 1998; Cordes & Lazio 2002).

Intraday variables (IDVs) similarly have typical modulation of up to a factor of two (e.g., Quirrenbach et al. 1992), although it can be a slightly higher (e.g., Dennett-Thorpe & de Bruyn 2000). The linearly polarized flux can vary with higher amplitude and on faster timescales, but the polarization fraction is <10% (Kraus et al. 2003), so inconsistent with ASKAP J173608.2−321635.

However, we consider whether the observed emission could be caused by other forms of extrinsic variability, or a combination of both extrinsic and intrinsic effects. For example, one could invoke a compact radio source undergoing an extreme scattering event (ESE; e.g., Fiedler et al. 1987; Bannister et al. 2016), in which the emission is lensed by plasma in the intervening medium; however, this does not explain the high circular polarization we observe. In this case, the lightcurve variability would be caused by propagation effects, while the change in polarization between the two periods of detectability would be intrinsic to the source.

Similar variability could be caused by gravitational lensing or plasma lensing. In general, gravitational lensing is achromatic while plasma lensing is highly chromatic (e.g., Wagner & Er 2020), but this is only true when the source is unresolved by the lens. If the source has finite size with spectral variations across it, then even gravitational lensing can have a chromatic effect as different regions are magnified/demagnified. For instance, a star with an active region could have different parts of that region magnified, which could increase the radio flux relative to other bands (Section 3.1) and give rise to highly polarized emission. However, it might still be difficult to explain multiple lensing events with similar magnifications. Further multiwavelength searches during bright states and better characterization of the lightcurve could help resolve this scenario.

As the source is located only 4 degrees from the GC, we consider whether it could be another GCRT.

The GCRT sources share some properties with ASKAP J173608.2−321635. GCRT J1742–3001 has a spectral index of ≲−2 and GCRT J1745–3009 has a spectral index varying from −4 to −13, while that for our source varies from −2.7 to −5.6. Both our source and GCRT J1745–3009 are highly polarized. GCRT J1745–3009 was ∼100% circularly polarized at 325 MHz (Roy et al. 2010). Our source was found to be 100% linearly polarized at ∼1.6 GHz and as high as ∼40% circularly polarized at ∼0.9 GHz. The source would have a flux density of ∼0.25 Jy extrapolated to 300 MHz, which is comparable to that for GCRT J1745–3009 (∼0.3 Jy). There is no X-ray detection for any of the GCRT sources when they are radio-bright.

However, some properties of our source are different from those of the GCRTs. GCRT J1745–3009 is thought to be a coherent emitter based on the very rapid variability (∼10 min), while our source shows no rapid variability and therefore may not emit coherently. The variability timescale for GCRT J1742–3001 is of order of one month, comparable with the initial "flare" detected in ASKAP but much longer than the timescale for the latest detections. GCRT J1745–3009 varies on much faster timescales: it emits flare-like emission for about 10 minutes out of a 77 minute period at a relative constant flux density. Hyman et al. (2007) showed that GCRT J1745–3009 has been detected in three different states. We have detected ASKAP J173608.2−321635 in two significant observation states so far (bright for a week versus fast fading). In general the sparse observations of ASKAP J173608.2−321635 and the GCRTs limits conclusions based on their temporal properties, and it is not even clear that all of the GCRTs share a common origin. Further monitoring will help resolve this.