NICER Observation of the Temporal and Spectral Evolution of Swift J1818.0$-$1607: a Missing Link between Magnetars and Rotation Powered Pulsars

The BAT triggered an alert at 21:16:47.328 UTC on 2020 March 12 for a short burst located at R.A.$=18^{h}\,18^{m}\,00.2^{s}$ and Decl.$=-16^{\circ}\,07\arcmin\,52.3\arcsec$, which was refined with prompt observation with X-Ray Telescope onboard Swift. The on-board trigger occurred on an 8 ms timescale in the 15–50 keV band with a rate significance of 24$\sigma$. The on-board calculated location was R.A.$=$18h 17m 53s and Decl.$=-16^{\circ}$ 06′ 05″ (Evans et al., 2020). HEAsoft version 6.27 (HEASARC, 2014) and Swift BAT CALDB (version 20171016) were used for the BAT data analysis. We use the raw counts (non-mask-weighted) data for the temporal analysis to maximize the signal-to-noise ratio of the light curve. For spectral analysis, the standard mask-weighted analysis is performed.

NICER is a non-imaging soft X-ray telescope onboard the International Space Station. It has absolute timing uncertainty better than $300$ ns. After the Swift BAT detection of the short burst, a series of follow-up observations was began at 01:38 UTC on 2020 March 13 with NICER. Through 2020 June 23, we have monitored this source with NICER for a total exposure of $\sim 102$ ks. All observations used for the present analysis are listed in Table 1. The basic data processing was carried out with NICERDAS version 7 in HEAsoft 6.27.2 and NICER calibration database version 20200202. We created cleaned event files by applying the standard calibration and filtering tool nicerl2 to the unfiltered data. We performed barycentric correction using barycorr with the JPL solar system ephemeris DE405 and the refined source position (Evans et al., 2020).

We created non-mask-weighted light curves around the BAT trigger time (Figure 1). No significant emission above 100 keV is seen by BAT, which confirms the soft nature of the burst. A clear pulse with a fast rise and an exponentially decaying tail is observed. We used the battblocks tool to estimate the T₉₀ duration, which is defined as the time interval where the integrated photon counts increase from 5% to 95% of the total counts, as $10\pm 4$ ms (15–350 keV). The burst profile can be well fit with the QDP¹¹1https://heasarc.gsfc.nasa.gov/ftools/others/qdp/qdp.html BURS model (a linear rise followed by an exponential decay), where the rising duration is $4.6\pm 0.4$ ms and the e-folding time of the decay is $2.4\pm 0.5$ ms. The T₉₀ duration of this burst is on the short end of typical short bursts of SGRs. We note that a burst-like feature can be marginally seen at $T_{\rm{burst}}-0.015$ s ( where T_burst is the time of the peak) but it is likely an instrumental effect.

We then extracted the BAT spectrum of this burst from the interval between $T_{\rm{burst}}-2$ ms and $T_{\rm{burst}}+12$ ms (the T₁₀₀ duration). Because of the low statistics of the spectrum, we generated a 16 channel spectrum instead of the standard 80 channels with batbinevt. We fit the spectrum with three models, including a blackbody, a bremsstrahlung, and a power law. The spectral analysis was performed with XSPEC v12.11.0. The best fit model is a blackbody with statistics of $\chi^{2}/\mathrm{dof}=5.67/11$ where dof denotes the degrees of freedom. The $\chi^{2}/\mathrm{dof}$ of the bremsstrahlung and the power-law models are 13.4/11 and 19.9/11, respectively. The temperature of the blackbody model is $8.4\pm 0.7$ keV with a radius of $2.4\pm 0.5$ km. The average 15–150 keV flux is $(6.1\pm 0.8)\times 10^{-7}$ erg s^-1 cm^-2 and the fluence of this burst is ($8.5\pm 1.1)\times 10^{-9}$ erg cm^-2. The isotropic equivalent luminosity is $\sim 3.1\times 10^{39}$ erg s^-1 and for a total energy of $\sim 4.3\times 10^{37}$ erg. The uncertainties in this paper denote the 68% confidence intervals unless stated otherwise.

We selected NICER events in the energy range of 2–8 keV for timing analysis. The average count rate between 2020 March 13 and June 23 was $\sim 1.1$ counts s^-1. We first found a coherent pulsation with a frequency of 0.733417(4) Hz ($P\approx 1.36$ s) and a single-peaked pulse profile in the 2020 March 13 data set (ObsID 3201060101, Enoto et al., 2020). Due to limited visibility, Swift J1818 could not be observed with NICER from March 15 – 17. Then NICER performed a series of monitoring observations with a cadence of roughly one day until MJD 58948 (2020 April 9). This allowed us to track the spin period evolution through a phase-coherent analysis. After MJD 58953 (2020 April 14), we revised the cadence down to $\sim$ 2–5 days, resulting in ambiguities of cycle counts during large gaps. Therefore, we did not perform the phase-coherent analysis afterward.

We used the pulse profile smoothed by the local polynomial regression method (Cleveland & Loader, 1996) as the template to calculate phase shifts and corresponding times of pulse arrival (TOAs) for the phase-coherent analysis. In each observation, we divided the time into a few segments such that each contained 1000 photons, and calculated TOAs using the maximum likelihood analysis method described in Livingstone et al. (2009). We found that the spin frequency $\nu$ and the spin-down rate $\dot{\nu}$ of Swift J1818 is highly variable. At least two timing anomalies in the first month of NICER monitoring were observed. The first one, consistent with a traditional spin-up glitch, occurred at MJD 58928.56 (2020 March 20) with a size of $\Delta\nu=(2.7\pm 0.1)\times 10^{-6}$ Hz ($\Delta\nu/\nu=(3.7\pm 0.2)\times 10^{-6}$) and $\Delta\dot{\nu}=(5.1\pm 0.5)\times 10^{-12}$ s^-2 ($\Delta\dot{\nu}/\dot{\nu}=-0.12\pm 0.01$). The second one can be described as an anti-glitch with $\Delta\nu=(-5.28\pm 0.01)\times 10^{-6}$ Hz ($\Delta\nu/\nu=(-7.20\pm 0.01)\times 10^{-6}$) and $\Delta\dot{\nu}=(4.69\pm 0.05)\times 10^{-11}$ s^-2 ($\Delta\dot{\nu}/\dot{\nu}=-0.91\pm 0.01$) at MJD 58934.81 (2020 March 26). However, we could not rule out the possibility that the frequency evolves continuously and dramatically instead of an abrupt jump due to a gap in coverage at the epoch of the timing anomaly. We divided the observations before MJD 58948 (2020 April 9) into three segments according to these two timing discontinuities and fit TOAs with second-order or third-order polynomials individually. The timing solutions for individual segments are summarized in Table 2.

To obtain the evolution of $\nu$ and $\dot{\nu}$, we used the technique described in Dib & Kaspi (2014) by choosing a window that contains 8–15 consecutive TOAs and fitting their phases with a second-order polynomial. Similar to the moving average technique, we moved the window with a step adaptively equal to a separation of 2–4 consecutive TOAs over the entire segments. For data beyond MJD 58948 (2020 April 9), which is noted as segment 4, we did not perform phase-coherent analysis spanning the entire segment due to the ambiguity of cycle counts in a few large gaps. The result is shown in Figure 2. Since the onset of the outburst, the source shows a high level of timing noise, in which $\dot{\nu}$ significantly changes on a timescale of a few days. We derived a long-term $\dot{\nu}=(-2.48\pm 0.03)\times 10^{-11}$ s^-2 ($\dot{P}=(4.61\pm 0.06)\times 10^{-11}$ s s^-1) by fitting the spin frequency evolution over the entire time span with a first-order polynomial function. This results in a characteristic age of $\tau_{c}=470$ yr by assuming a braking index of 3 and rapid spin at birth. The surface equatorial magnetic field can be inferred as $B=2.5\times 10^{14}$ G and the $L_{\rm{sd}}$ is estimated as $7.2\times 10^{35}$ erg s^-1. However, the dramatic changes in the timing behavior make it difficult to conclusively characterize the long-term timing properties at the current stage. If we consider the timing solution in individual epochs, the derived $\tau_{c}$ could be in a wide range of 270–1300 yr, and $L_{\rm{sd}}$ could be in the range of $(2.6$–$13)\times 10^{35}$ erg s^-1.

To see whether the pulse profile is energy-dependent, we created energy-resolved pulse profiles in six bands: 0.5–2, 2–3, 3–4, 4–5, 5–6, and 6–8 keV (see the right panel of Figure 2). The pulsation cannot be seen below 2 keV due to heavy interstellar absorption. The 2–8 keV folded light curve shows a broad asymmetric peak. No significant energy dependence of the pulse shape is seen across 2–8 keV. We calculated the rms pulse fraction (PF, see definition in Dib et al., 2009; An et al., 2015) in these energy bands and found that the PF increases with energy. The background estimated from nibackgen3C50²²2https://heasarc.gsfc.nasa.gov/docs/nicer/tools/nicer_bkg_est_tools.html is subtracted from the pulse profile. To test whether the pulse profile changes following glitches, we created time-resolved pulse profiles. We divided the X-ray observations into thirteen time bins to ensure each one has effective exposure time $\gtrsim 5000$ s: MJD 58920–58928, 58929–58932, 58932–58935, 58935–58938, 58938–58941, 58942–58948, 58953–58960, 58962–58969, 58971–58974, 58987–58994, 58998–59005, 59008–59015, and 59017–59024. We did not observe any significant changes in the shape of the pulse profile accompanying either timing discontinuity. The PF increased from 0.34(1) to $\sim 0.43$ in the first $\sim 15$ days from the onset of the outburst and fluctuated around 0.43 except for an extreme value during MJD 59008–59015 (see Figure 3). We found that the background level estimated with nibackgen3C50 is extremely high on MJD 59014. This could result in an over-estimate of the PF if the background in this observation is not accurately estimated.

In addition to tracking the evolution of the timing behavior and the variability of the pulse profile, we searched for SGR-like X-ray bursts. We created light curves with a bin size of 1/256 s. For each time bin, we calculated the mean count rate in the surrounding 20 seconds. The probability of the photon distribution in each time bin can be evaluated with the Poisson distribution. During the NICER campaign, we identified 21 bursts with a detection significance higher than 5$\sigma$. We also emplyed other time bin sizes to reconfirm the detection of bursts. Their occurrence times are indicated in Figure 2. Four of them are clustered near the first glitch and three of them are near MJD 58972. The candidate anti-glitch likely occurred during the observational gap that prevented us from probing the association between radiative events and the anti-glitch. We noticed that the light curve observed in the last GTI of MJD 58944 (ObsID 3556011701) exhibited many burst-like structures with a timescale much longer than that of regular bursts and accompanied an enhancement of the baseline level with a duration of $\sim 300$ s. After carefully examining the photon distribution in different detectors and the timing/spectral behavior during this period we suggest it was caused by a particle flaring episode, and was not intrinsic to Swift J1818.

To monitor the long-term spectral evolution of Swift J1818 with NICER we extracted X-ray spectra from the burst-free times using XSELECT. We grouped each spectrum to have 50 counts per channel using grppha and used XSPEC version 12.10.0c (Arnaud, 1996) to fit the spectra. We created background files for each observation using the nibackgen3C50 tool. We used the response and ancillary response files currently in the NICER calibration database. Note that we removed the data from Focal Plane Modules 14 and 34 and used an adjusted ancillary response file accordingly.

The very large column density along the line of sight to Swift J1818 causes background counts to dominate below 2 keV. Also, above 7 keV the signal-to-noise ratio of the source decreases significantly, therefore we only perform our fits in the 2$-$7 keV band. We modeled the X-ray spectra with an absorbed blackbody model. To determine the amount of Hydrogen column density ($N_{\rm{H}}$), we used the tbabs model with ISM abundances (Wilms et al., 2000).

To model the spectral evolution, we allowed all the parameters of the models of individual data sets to be free except for the $N_{\rm{H}}$ which was kept linked. Such a fit results in a $\chi^{2}$ = 2527.72 for 2452 dof. The resulting hydrogen column density is found to be $(11.2\pm 0.2)\times 10^{22}$ cm^-2, much higher than that predicted from the dispersion measure (DM) of $\sim 2\times 10^{22}$ cm^-2 if we adopt a linear relationship of $N_{\rm{H}}$ (10²⁰ cm^-2)$=0.30_{-0.09}^{+0.13}$ DM (pc cm^-3) (He et al., 2013; Karuppusamy et al., 2020). This is consistent with other sources near the Galactic center and suggests that a significant part of X-ray absorption could be contributed by molecular clouds rather than neutral hydrogen atoms (Baumgartner & Mushotzky, 2006; Willingale et al., 2013). We show the best fit models together with the first and the last X-ray spectra in Figure 3 and the best-fit parameters in Table 1. We further show the evolution of the inferred spectral parameters in Figure 3. The flux decayed to roughly half of the initial value. The temperature, however, does not show a clear variability and seems to agree within the statistical uncertainties of the individual measurements. We tested linking this parameter throughout all the observations. Such a fit yield an average kT=$1.05\pm 0.01$ and a $\chi^{2}$ = 2619.63 for 2497 dof. The apparent radius shows a decrease of $\sim 60$% regardless of whether the temperature is linked or not.

The quiescent emission from Swift J1818 is not detectable in archival X-ray observations. We used the deepest XMM-Newton observation (ObsID: 0800910101) to estimate the bolometric 3$\sigma$ upper limit of the quiescent luminosity as $7\times 10^{32}$–$1.7\times 10^{34}$ erg s^-1. This range contains the uncertainty in the distance of $4.8$–$8.1$ kpc, and the possible blackbody temperature range of $kT=0.3$–$0.5$ keV. Using this value, we can derive a limit to the luminosity increase as a factor of $>$10. Note that the spectral evolution shown in Figure 3 indicates that the surface temperature of the source did not change significantly. The flux decay is dominated by the decrease in the apparent emitting radius, which decreases by about 30%. This finding supports a scenario where the outburst decay is caused by a shrinking hotspot due to the untwisting of magnetic field loops (Beloborodov, 2013).

The post-outburst timing behavior is largely erratic, consistent with the large torque variations observed from magnetars and high-$B$ RPPs during outburst epochs (Dib et al., 2009; Archibald et al., 2016, 2017). During the first two weeks of our monitoring campaign, the source showed a large spin-up glitch and a likely spin-down glitch. No gradual recovery is observed as shown in regular RPPs (Espinoza et al., 2011) and even radiatively-silent glitches in magnetars (Dib & Kaspi, 2014). The sizes of these two glitches are extremely large even compared with those in other magnetars (Dib & Kaspi, 2014). This implies a substantial change in the kinetic energy that could be released in the form of electromagnetic waves. However, similar to 70% of glitches in magnetars, we did not see any radiative change accompanying the spin-up glitch of Swift J1818 except for possible clustering of short bursts (Janssen & Stappers, 2006; Dib & Kaspi, 2014; Kaspi et al., 2014). The lack of radiative variability may imply a recovery that is dictated by processes internal to the NS. The more erratic changes observed during magnetar outbursts, including the one we observe for Swift J1818, may point to variations dominated by external processes, likely close to the light cylinder where particle outflow could exert large torques on the star. This requires a coupling between the inner-crust, where the glitch occurs, and the external dipolar field lines. This condition may be achieved in high-$B$ sources (Harding et al., 1999; Thompson et al., 2000).

The spin-down glitches are rarely seen in magnetars and never observed in regular pulsars. The first confirmed anti-glitch was observed in 1E 2259+586 with a size of $\Delta\nu=-4.5\times 10^{-8}$ Hz and $\Delta\dot{\nu}=-2.7\times 10^{-14}$ s^-2, occurring at the onset of a radiative outburst (Archibald et al., 2013). Similar behavior has been observed in SGR 1900+14 during a burst active epoch although a gradual slow down remained a possible explanation due to a $\sim 80$-day gap (Woods et al., 1999). For Swift J1818, we do not detect any additional enhancement in the persistent emission coincident with the epoch of the anti-glitch. Recently, 1E 2259+586 has shown a radiatively-quiet anti-glitch with a sudden spin-down amplitude $\delta_{\rm nu}=-8.1\times 10^{-8}$ Hz (Younes et al., 2020a). However, unlike 1E 2259+586, the anti-glitch in Swift J1818 was very close to the start of a major outburst. A radiative change may have occurred with this anti-glitch but was insufficient to present itself above the high persistent flux during the outburst. The mechanism for radiatively silent anti-glitches remains unclear. It is possibly originated from the magnetosphere, but the particle acceleration is too weak to trigger radiative events under extra loading of plasma (Harding et al., 1999). An alternative explanation is that the anti-glitch is caused by the coupling of a superfluid component with a rotation frequency lower than the rest of the NS. It is difficult to interpret why the detached superfluid component spins down much faster than the rest of the NS before the anti-glitch.

It remains possible that the entire temporal evolution is caused by a series of rapid and non-instantaneous torque variability similar to the post-outburst behavior in SGR 1806$-$20 and 1E 1048$-$5937 (Woods et al., 2007; Dib et al., 2009; Dib & Kaspi, 2014). Right after the giant flare on 2004, the $\dot{\nu}$ of SGR 1806$-$20 changed rapidly between $\sim-2\times 10^{-12}$ s^-2 and $\sim-1.5\times 10^{-11}$ s^-2. Similarly, after the onset of the outburst in 2009, the $\dot{\nu}$ of 1E 1048$-$5937 oscillated between $\sim-2\times 10^{-13}$ s^-2 and $\sim-2.8\times 10^{-12}$ s^-2 for $\sim 500$ days. The post-outburst $\dot{\nu}$ of Swift J1818 varied between $\sim-4\times 10^{-12}$ s^-2 and $\sim-6\times 10^{-11}$ s^-2, which has a similar relative amplitude compared to that of SGR1806$-$20 and 1E 1048$-$5937 with a much larger size. This suggests that Swift J1818 is one of the noisiest sources among magnetars and high-$B$ RPPs.

The post-outburst spectral evolution shows that although the observed trend in the unabsorbed flux is not completely monotonic, it decreases by about 55% in 102 days, from 3.79 to 1.27 $\times 10^{-11}$ erg s^-1 cm^-2. Using the spectral parameters we also calculated the 0.3$-$10.0 keV thermal luminosity of Swift J1818. We fit the decaying trend in the flux with both the plateau-decay model (see equation 12 in Enoto et al., 2017) and the double exponential model (see equation 1 in Coti Zelati et al., 2018).

Since NICER is a non-imaging instrument, uncertainties in the estimated background spectrum may have significant effects, especially at low flux levels. We noticed a $\sim 4$% systematic uncertainty by analyzing several background fields observed in 2020. Additionally, Esposito et al. (2020) reported the detection of a dust scattering halo that contributes 2% of the source flux. To take both effects into account, we introduce a 5% systematic uncertainty in our fits. With these considerations, these models provide broadly acceptable fits to data. Note that simpler models, including an exponential or a power-law decay, result in substantially worse fits.

The plateau-decay model assumes that the luminosity first shows a plateau-like slow decay phase which is then followed by a power-law like decay after a certain timescale. It is characterized by the luminosity at the onset of the outburst, $L_{0}$, the timescale of the plateau, $\tau_{0}$ and the power-law index, $p$ (see equation 12 in Enoto et al., 2017). We obtained a $\chi^{2}$=166 with 43 dof for the plateau-decay model. The best fit parameters and their 1$\sigma$ uncertainties are L₀=$(1.92\pm 0.06)\times 10^{35}$ erg s^-1, $\tau_{0}$=23${}_{-7}^{+11}$ days, and $p$=0.51$\pm$0.08. Before this study, only SGR 0501$+$4516, SGR 0418$+$5729, and Swift J1822$-$1606 have detectable $\tau_{0}$ of 15.9, 42.9 and 11.2 days, respectively (Enoto et al., 2017). These $\tau_{0}$ measurements are similar to the value we get for Swift J1818. However, for all of these sources, the slopes of the decay are significantly larger than what we infer, indicating a much faster decline for the other sources compared to Swift J1818. Similarly, the decreasing trend in the luminosity can also be modeled by a double exponential decay function following equation 1 in Coti Zelati et al. (2018). We found the normalization constants of individual components best match the data for A=1.51${}_{-0.08}^{+0.06}\times$10³⁵ erg s^-1 and B=0.49$\pm 0.09\times$10³⁵ erg s^-1. The e-folding times are $\tau_{1}$=157$\pm 13$ days and $\tau_{2}$=9${}_{\pm}2$ days. The fit results in a $\chi^{2}$=161 with 42 dof. This model shows that the luminosity decrease has two components: one showing a rapid decay and another long term decay trend. The best fit values for the e-folding times are in agreement with similar results from Coti Zelati et al. (2018), especially the values found for SGR 0501+4516.

Note that especially at the late stages of the decay, the persistent flux of the source shows significant fluctuations. The fact that at least in some of these observations we also detect short bursts may imply that low-level activity is quasi-continuous during the outburst decay, which could be affecting the apparent persistent flux level.

The broadband PF of the pulse profile of Swift J1818 is $\gtrsim 0.4$. Such a single-peaked pulse profile with a high PF may be difficult to produce with two antipodal hotspots of equal brightness (DeDeo et al., 2001; Hu et al., 2019). Therefore, we suggest that the emission is dominated by a distorted hotspot on the surface of the NS. Emission from hot spots can be highly distorted, with anisotropy governed by the local $B$ field direction when the $B$ field is extremely high. The increase of the PF throughout our observing campaign reinforces the above idea that the outburst evolution is governed by the shrinking of a hotspot on the surface of the magnetar (Rea et al., 2013; Coti Zelati et al., 2015; Mong & Ng, 2018). Moreover, the hotspot could consist of two components with different temperatures. The boundary between the components could be much blurred instead of a sharp discontinuity (DeDeo et al., 2001). These two components may have different shrinking timescales that result in the two timescales of the observed flux decay trend.

A hard power-law spectral component above $\sim$10 keV has been reported from some of the persistently bright magnetars and from the early phases of transient outbursts, where the emission is thought to be radiated from the magnetosphere (see, e.g., Younes et al., 2017b; Archibald et al., 2020). Using the reported correlation between the soft and hard X-ray luminosity of known magnetars (equation 4 of Enoto et al. 2017) with the unabsorbed X-ray flux of Swift J1818, $\sim 3\times 10^{-11}$ erg s cm^-2 observed in the soft band, we would expect the hard power-law flux at $\sim 2\times 10^{-11}$ erg s cm^-2 in the 15–60 keV band with a flat photon index of $\sim 0.8$ (equation 7 of Enoto et al. 2017). However, we did not find any evidence for such a hard power-law component in the soft NICER spectrum. This is consistent with no detection above 15 keV with NuSTAR and INTEGRAL although a hard power-law component can be marginally seen below $\sim$20 keV (Borghese et al., 2020; Esposito et al., 2020). This is in contrast to the prominent hard X-ray radiation in the 2009 outburst from a similar fast spinning and radio-emitting magnetar, 1E 1547.0$-$5408 (Enoto et al., 2010).

Swift J1818 is a transient source showing timing properties between canonical magnetars and high-$B$ RPPs. Observations of low $B$-field magnetars and magnetar-like activity in high-$B$ RPPs hinted that magnetars could represent the high $B$ field tail of a single distribution (Kaspi & McLaughlin, 2005; Ho, 2013). It has been suggested that magnetars have a high quiescent soft thermal luminosity that is powered by the dissipation of the strong magnetic field (Thompson & Duncan, 1995). The magneto-thermal evolution model suggests that the key component is the toroidal magnetic field in the crust (Pons et al., 2009; Perna & Pons, 2011; Viganò et al., 2013). This toroidal field cannot be inferred from the spin down. We plot the thermal luminosity of magnetars, high-$B$ RPPs, and several X-ray RPPs in Figure 4 (a). Several theoretical models with different magnetic field strengths and atmosphere composition are also plotted. Most magnetars have quiescent soft thermal luminosity above $\sim 10^{34}$ erg s^-1 except for transient radio magnetars. Canonical RPPs usually have luminosity lower than $10^{33}$ erg s^-1 and several young high-$B$ RPPs are in between the RPPs and magnetars.

We overlay the estimated upper limit of the quiescent thermal luminosity of Swift J1818 in Figure 4 (a). The luminosity of Swift J1818 during the outburst, which is roughly the same as several bright persistent magnetars, e.g., 4U 0142+61 and 1RXS J170849.0$-$400910, is also plotted for reference. The upper limit of the quiescent luminosity of Swift J1818, although the uncertainty is large, occupies a similar region as the young high-$B$ RPPs J1846$-$0258 and J1119$-$6127, classifying Swift J1818 in the same category. Moreover, the current estimate of $\nu$ and $\dot{\nu}$ may be affected by the glitch and the heavy timing noise similar to several young magnetars. The torque may be larger than the nominal value by $\gtrsim 1$ order of magnitude at this early stage in the outburst. In several other magnetars, the torque decreases to the quiescent level on a timescale of a few months to ten years (Camilo et al., 2016; Younes et al., 2017a; Archibald et al., 2020). Future long-term monitoring of the timing behavior of Swift J1818 is necessary. If we adopted the $\dot{\nu}$ measurement from segment 3, the characteristic age of Swift J1818 could be as high as $\gtrsim 1000$ yr. This value is much older than that derived from the first two segments and comparable to that of PSR J1846$-$0258.

The detection of radio emission and the corresponding spectral index of Swift J1818 provides another hint that Swift J1818 can exhibit features of a regular RPP instead of a canonical magnetar (Karuppusamy et al., 2020; Esposito et al., 2020; Majid et al., 2020). Historically, magnetars are considered radio-silent NSs, where their $L_{\rm{X}}$ are higher than their $L_{\rm{sd}}$. The discovery of radio-emitting magnetars was a milestone that links the magnetars and RPPs. Their $L_{\rm{X}}$ drop to lower than their $L_{\rm{sd}}$ in quiescence. The high-$B$ RPPs J1846$-$0258 and J1119$-$6127 are two important samples to bridge radio-emitting magnetars and RPPs. They show magnetar bursts and X-ray outbursts, but the peak $L_{\rm{X}}$ remains lower than their $L_{\rm{sd}}$. We plotted the $L_{\rm{X}}$ versus the $L_{\rm{sd}}$ in Figure 4 (b) of all magnetars and RPPs together with the luminosity range of Swift J1818 from expected quiescence to the outburst peak. The $L_{\rm{sd}}$ of Swift J1818 is highest among the canonical magnetars and slightly lower than that of PSR J1119$-$6127. Similar to high-$B$ RPPs, the peak $L_{\rm{X}}$ of Swift J1818 remains lower than its $L_{\rm{sd}}$. Moreover, radio-emitting magnetars show intermittent radio emission. On the contrary, the radio emission of PSR J1119$-$6127 shut off during the early stages of its outburst onset (Majid et al., 2017). Swift J1818 shows signs of both magnetar and radio pulsar populations and provides a crucial link between the two populations. Continued radio and X-ray monitoring of Swift J1818 is critical to better understand the nature of this source.