Simultaneous X-Ray, Gamma-Ray, and Radio Observations of the Repeating Fast Radio Burst FRB 121102

The 110 m Robert C. Byrd GBT observed FRB 121102 on 2016 September 16, 18, November 26, and 2017 January 11 during periods that overlapped with either XMM-Newton or Chandra observations (Table 1; Figure 1). FRB 121102 was observed with the S-band receiver at a center frequency of 2 GHz and a bandwidth of 800 MHz, of which about 600 MHz is usable due to receiver roll-off and the masking of spectral channels containing radio frequency interference (RFI). We used the Green Bank Ultimate Pulsar Processing Instrument (GUPPI; DuPlain et al. 2008) in coherent de-dispersion mode where each of the spectral channels were corrected for dispersion in real time to DM = 557 pc cm⁻³. These coherently de-dispersed observations therefore do not suffer from intra-channel DM smearing, as the correction is performed before Stokes parameters are formed. This mode provides full Stokes parameters, 512 spectral channels (each 1.56 MHz wide), and a time resolution of 10.24 μs.

The 100 m Effelsberg telescope performed a three hour observation of FRB 121102 on 2016 September 16, simultaneous with both the GBT and XMM-Newton observations (Table 1). Data were recorded at 1.4 GHz with an observing configuration that was identical to what is used for the HTRU-N pulsar and fast transient survey. Details can be found in Barr et al. (2013). FRB 121102 was observed with the central pixel of the 7-beam L-band receiver and data were recorded with the PFFTS pulsar search mode backends. The data cover a frequency range of 1210–1510 MHz with 512 frequency channels and have a time resolution of 54.613 μs. Note, unlike the GBT and Arecibo observations, these data were not coherently de-dispersed and so suffer from ∼1 ms of intra-channel DM smearing at the DM of FRB 121102.

The 305 m William E. Gordon Telescope at Arecibo Observatory observed FRB 121102 on 2017 January 12 simultaneously with GBT and Chandra observations (Table 1). We used the single-pixel L-wide receiver with the Puerto Rican Ultimate Pulsar Processing Instrument (PUPPI) backend. This setup provided 800 MHz of bandwidth centered at 1380 MHz, of which about 600 MHz is usable due to receiver roll-off and RFI excision. As with GUPPI, the data were coherently de-dispersed to DM = 557 pc cm⁻³ in real time with 512 spectral channels (each 1.56 MHz wide) and a time resolution of 10.24 μs.

Two XMM-Newton Director's Discretionary Time (DDT) observations were performed on 2016 September 16 and 18 (Table 1). These DDT observations were scheduled in response to a period of high radio burst activity, with several bursts detected per hour (Chatterjee et al. 2017; Law et al. 2017). We used the EPIC/pn camera in Small Window mode (5.7 ms time resolution) and the EPIC/MOS cameras in Timing mode (1.75 ms time resolution). The Timing mode observations provide only one dimension of spatial information, which results in a high background (see Section 3.2).

Unfortunately, the pn-mode observations have a deadtime fraction of 29%. This means that for every 5.7 ms frame of the pn observation, the telescope is blind to X-ray photons for 1.65 ms. Though this time resolution is helpful in resolving bursts when a significant number of counts are detected, the deadtime is detrimental when placing a limit following a non-detection. We therefore do not use the pn-mode data below when placing fluence limits for putative X-ray bursts.

Standard tools from the XMM-Newton Science Analysis System (SAS) version 16.0 and HEASoft version 6.19 were used to reduce the data. For each observation, the raw Observation Data Files (ODF) level data were downloaded and were pre-processed using the SAS tools emproc and epproc. Data were filtered so that single-quadruple events with energies between 0.1–12 keV (pn) and 0.2–15 keV (MOS) were retained, and standard "FLAG" filtering was applied. The light curves were then inspected for soft proton flares and none were found. Event arrival times were then corrected to the SSB. For the pn, we extracted source events from an 18'' radius (80% encircled energy) source region centered on the position of FRB 121102. For the MOS cameras, we extracted events from a 20 pixel (22'') wide strip centered on the position of the source.

The Chandra X-ray Observatory targeted FRB 121102 on 2016 November 26 (ObsID 19286) and 2017 January 11 (ObsID 19287) using the front-illuminated ACIS-I instrument for 20 ks in both instances. The detector was operated in VFAINT mode with the entire array read out using a 3.2 s frame time. These observations were part of a joint Chandra/GBT Cycle 18 project to obtain contemporaneous data with the two telescopes.

The resulting Chandra data sets were analyzed using CIAO¹⁷ version 4.8.2 (Fruscione et al. 2006) following standard procedures recommended by the Chandra X-ray Center. We extracted events from a 1'' radius region (95% encircled energy) centered on the position of FRB 121102 and corrected the photon arrival times to the SSB using the aforementioned source position.

The radio observations from GBT, Effelsberg, and Arecibo were searched for bursts using standard tools in the PRESTO¹⁸ (Ransom 2001) software package. For the purposes of searching, the data were downsampled by a factor of 8 (to a time resolution of 81.92 μs) for Arecibo and GBT observations and a factor of 16 (874 μs resolution) for Effelsberg observations. We first used 'rfifind' to identify frequency and time blocks contaminated by RFI, which were masked in the subsequent search. The data were then de-dispersed in a DM range of 507–607 pc cm⁻³ with step size 0.5 pc cm⁻³ for GBT, 535–585 pc cm⁻³ with a step size of 1 pc cm⁻³ for Effelsberg, and 527–626 pc cm⁻³ with a step size of 1 pc cm⁻³ for Arecibo. Burst candidates were identified in a boxcar matched filtering search for pulse widths up to 20 ms and S/N > 5 using 'single_pulse_search.py'. Due to the effects of RFI, our search is only complete in the Arecibo observation to S/N $\gtrsim$ 13 and to S/N $\gtrsim$ 7 for Effelsberg and GBT.

Four astrophysical bursts at a DM consistent with that of FRB 121102 were found in the GBT observations on 2016 September 16 and 18 at times that overlapped with the simultaneous XMM-Newton observations (labeled, in this work, as bursts GBT 1–4). Radio bursts at times outside of the simultaneous X-ray coverage are not presented here. Five more bursts were found in the GBT observation on 2017 January 11 that overlapped with the Chandra observation (bursts GBT 5–7, 9, and 10). Three bursts were found in the search of the 2017 January 11 Arecibo observation (AO 1, 2, and 4). During one of the detected GBT bursts (GBT 9) a coincident burst (AO 3) was found in the Arecibo observation when searching specifically at the time of burst GBT 9, which was otherwise missed due to RFI. Similarly, burst GBT 8 was found at the time of burst AO 1. The remainder of the GBT- and Arecibo-detected bursts were not detected in the corresponding other telescope. To summarize, a total of 12 radio bursts were found with two co-detections (GBT 8/AO 1 and GBT 9/AO 3).

In the Effelsberg observation on 2016 September 16, all events with S/N > 7 can be attributed to RFI. Because these observations overlap with the times of GBT-detected bursts, the three second windows around the arrival times of the coincident GBT bursts were searched manually, using a range of downsample factors, but no bursts were identified. These contemporaneous non-detections, as well as those during the 2017 January 11 Arecibo and GBT observations that did not result in co-detections, are likely due to the narrow-band nature of FRB 121102 radio bursts (see Scholz et al. 2016; Spitler et al. 2016).

The GBT-detected radio bursts are shown in Figure 2, and those detected at Arecibo are shown in Figure 3. We show only the time series for each burst and defer a full spectral analysis of the bursts to a future work. For each burst, we measured the peak flux density, fluence, and burst width (Gaussian FWHM) using an identical procedure to Scholz et al. (2016). These values are shown in Table 2. We note that bursts GBT 1 and 2 arrived ∼40 ms apart. This is the minimum separation reported thus far for radio bursts from FRB 121102. This does not necessarily imply an upper limit to an underlying periodicity of $\lt 40$ ms, as we cannot exclude the possibility that this is a single wide burst with multiple peaks, or if the source is a rotating object that multiple bursts were emitted from during a single rotation. We defer a more detailed analysis of the arrival times to a future work with a much larger sample of bursts. The measured radio burst arrival times were corrected to the SSB and corrected for the dispersive delay to infinite frequency using DM = 559 pc cm⁻³ (Scholz et al. 2016; which is more accurately measured than the DM = 557 pc cm⁻³ used during data collection, see Section 2) and are therefore directly comparable to the SSB-referenced arrival times of the X-ray photons.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For each detected radio burst, we searched nearby in time for X-ray photons that could be due to X-ray bursts. In the 2016 September 16 XMM-Newton observation, the closest photon to Bursts 1 and 2 was 5.8 s away. The false alarm probability for an event to arrive in such a window given the 0.5–10 keV background count rate of 0.01 counts s⁻¹ is 25%. In the 2016 September 18 XMM-Newton observation, no photons were closer than 0.7 s from a radio burst (background count rate 0.05 counts s⁻¹, false alarm probability 13%). In the 2017 January 11 Chandra observation, a single photon was detected at the source position and was 893 s away from the closest radio burst (background count rate 5 × 10⁻⁵ counts s⁻¹, false alarm probability 42%). Given the high probability that these are background events, we have no reason to think they are related to FRB 121102. For each observation, the total number of X-ray counts detected within the source extraction region of FRB 121102 was consistent with the background count rate.

For both XMM-Newton and Chandra observations, we also performed a similar exercise as above with the raw event lists (Level 1 for Chandra and ODF-level for XMM-Newton) prior to applying any standard filtering in case a bright X-ray burst was flagged as a cosmic ray. There was no evidence for any X-ray events in excess of the unfiltered background count rate.

We place a limit on the number of X-ray counts from FRB 121102 using the Bayesian method of Kraft et al. (1991). For a putative X-ray burst at the time of a detected radio burst, we can place an upper limit of 14.4 counts at a 5σ confidence limit in 0.5–10 keV. This limit is independent of duration up to the time of the nearest detected photon (see above), because the background is negligible. For an X-ray burst arriving outside of this window at any time during the observation, the background rate and trials factor depend on the assumed duration. So, we assume a duration of 5 ms, similar to that of the radio bursts, which leads to a 0.5–10 keV count limit of 32.3 counts during the XMM-Newton observations and 33.8 counts for Chandra (5σ confidence).

A count limit, ${\mu }_{\mathrm{lim}}$ can be translated to a fluence limit by dividing by a spectrally averaged effective area for the instrument, A, and multiplying by an average photon energy, E. We can also stack individual limits, under the assumption that X-ray bursts with similar spectra are emitted at the times of every radio burst. In the case where zero counts are detected in each detector (i.e., XMM-Newton/MOS and Chandra/ACIS) and the background count rate is negligible, the count rate limit is simply ${\mu }_{\mathrm{lim}}=-\mathrm{log}(1-\mathrm{CL})$, where CL is the desired confidence level (in our case 0.9999994 for 5σ), and the fluence limit takes the form,

$$\begin{eqnarray}&&{F}_{\mathrm{lim}}=\displaystyle \frac{{\mu }_{\mathrm{lim}}}{{\displaystyle \sum }_{i}\tfrac{{N}_{i}{A}_{i}}{{E}_{i}}}.\end{eqnarray} \tag{ 1 }$$

Here, the sum is over each instrument, which have each observed simultaneously during N_i radio bursts. We read the effective area of each telescope as a function of energy from the ancillary response files for each telescope.¹⁹ Using these effective area curves, in Figure 5, we plot the limiting burst energy as a function of photon energy. Here, we derive a model-independent limit where there is an equal probability of a source photon occurring across the entire band. The 5σ confidence upper limit on the 0.5–10 keV fluence for a single X-ray burst at the time of one of the detected radio bursts is 2 × 10⁻¹⁰ erg cm⁻² for observations simultaneous with XMM-Newton and 5 × 10⁻¹⁰ erg cm⁻² for Chandra, or 3 × 10⁴⁶ erg and 6 × 10⁴⁶ erg at the luminosity distance of FRB 121102, respectively. If we additionally assume that X-ray bursts of similar fluence are emitted at the time of every radio burst (i.e., stacked as per Equation (1)), the upper limit at the time of the bursts becomes 3 × 10⁻¹¹ erg cm⁻² (4 × 10⁴⁵ erg at the source distance). The limit for an X-ray burst arriving at any time during the X-ray observations is 5 × 10⁻¹⁰ erg cm⁻² for XMM-Newton and 1 × 10⁻⁹ erg cm⁻² for Chandra for an assumed duration of 5 ms (which correspond to energy limits of 6 × 10⁴⁶ and 1 × 10⁴⁷ erg).

We also searched data from the Fermi Gamma-ray Burst Monitor (GBM; Meegan et al. 2009) for gamma-ray burst counterparts during the radio bursts presented in this work. FRB 121102 was only visible to GBM during the 2016 September XMM-Newton/GBT observations, and so our limit applies only to those four bursts. In an analysis similar to what has been done for previous FRB 121102 radio bursts (Younes et al. 2016), we used the Time Tagged Event GBM data in the energy range 10–100 keV and searched for excess counts in 1 and 5 ms bins in 2 s windows centered on the arrival time of the four radio bursts. We find no signals that are not attributable to Poisson fluctuations from the background count rate at a 5σ confidence level. Taking into account the effective area of Fermi/GBM at 10–100 keV, the background count rate, and a typical photon energy of 40 keV, we place an upper limit of 1 × 10⁻⁸ erg cm⁻² for each burst and 4 × 10⁻⁹ erg cm⁻² if we assume a gamma-ray burst is emitted at the time of each radio burst, which, at the measured luminosity distance, corresponds to a 10–100 keV burst energy limit of 5 × 10⁴⁷ erg.

In order to probe more deeply for faint emission from a persistent X-ray source at the location of FRB 121102 than previous limits (Scholz et al. 2016; Chatterjee et al. 2017), we produced a summed image of all the Chandra data to date. We co-added, aspect-corrected, and exposure-corrected the two ACIS-I exposures from our 2016 November and 2017 January observations along with the ACIS-S exposure previously presented in Scholz et al. (2016) using the merge_obs script in CIAO. In the combined 80 ks exposure, only two events are registered within an aperture of radius $1^{\prime\prime}$ centered on the position of FRB 121102 (see Figure 4), entirely consistent with being due to background emission. We measure a 0.5–10 keV background count rate in a $40^{\prime\prime}$ radius region away from the source to be 0.20 counts s⁻¹ sq. arcsec⁻¹. Using the number of detected counts and measured background rate, we place a count rate limit using the Bayesian method of Kraft et al. (1991) and translate it to a flux using the same method as in the burst case (Section 3.2). Assuming a photoelectrically absorbed power-law source spectrum with a spectral index of ${\rm{\Gamma }}=2$ and a hydrogen column density of N_H $\sim \ 1.7\times {10}^{22}$ cm⁻² (as in Chatterjee et al. 2017), the 5σ upper limit on any persistent 0.5–10 keV X-ray emission from FRB 121102 or the host galaxy is 4 × 10⁻¹⁵ erg cm⁻² s⁻¹. As the XMM-Newton data presented in this work were included for the persistent X-ray source limit in Chatterjee et al. (2017), their value of 5 × 10⁻¹⁵ erg cm⁻² s⁻¹ is still valid for the XMM-Newton images.

Zoom In Zoom Out Reset image size

Potential sources of X-ray bursts that accompany FRBs can have different underlying spectra. Here, we explore the effect of different source spectra on the fluence limits. To generate the assumed source spectra, we use XSPEC v12.9.0n, using abundances from Wilms et al. (2000) and photoelectric cross-sections from Verner et al. (1996) for N_H.

We initially estimate the N_H to the source from the DM–N_H relation of He et al. (2013). We take the DM contribution from our Galaxy to be 188 pc cm⁻³ (from the NE2001 model of Cordes & Lazio 2002). The DM of the host has been estimated to be $55\lesssim {\mathrm{DM}}_{\mathrm{host}}\lesssim 225$ pc cm⁻³ (Tendulkar et al. 2017), so we use the average value of 140 pc cm⁻³. We assume that the IGM does not have a significant contribution to the N_H, as it is expected to be nearly fully ionized and thus provides negligible X-ray absorption (e.g., Behar et al. 2011; Starling et al. 2013). This Galactic plus host DM of 328 pc cm⁻³ corresponds to N_H ∼ 1 × 10²² cm⁻².

However, such a determination only holds in environments similar to our Galaxy. Photoelectric absorption and dispersion are dominated by separate components of the ISM, namely atomic metals and free electrons, respectively, and their ratios could be significantly different in other environments. To illustrate the effect of excess X-ray absorption, we also consider a case where the N_H is two orders of magnitude higher at ∼1 × 10²⁴ cm⁻². At this N_H nearly all of the 0.5–10 keV X-ray flux is absorbed. This situation may be possible in a supernova remnant (SNR) in the first few decades following the supernova, where the ratio of atomic metals to free electrons could be high (Metzger et al. 2017).

For the spectra of the bursts, we assume a few fiducial models. We take a blackbody spectrum with ${kT}=10\,\mathrm{keV}$ as a model similar to those observed in magnetar hard X-ray bursts (e.g., Lin et al. 2012; An et al. 2014). We take a cutoff power-law with index ${\rm{\Gamma }}=0.5$ and cutoff energy of 500 keV as a spectrum typical of a magnetar giant flare (Mazets et al. 2005; Palmer et al. 2005, for SGR 1806−20). Finally, we use a power-law model with index ${\rm{\Gamma }}=2$ as an example soft spectrum to contrast with the harder magnetar models.

In Table 3, we give the fluence limits for each of these models and the implied limit on their unabsorbed emitted energy both in the 0.5–10 keV band and extrapolated to the 10 keV–1 MeV gamma-ray band. Note that the energy limits are highly dependent on the amount of absorption and the gamma-ray energy is heavily dependent on the assumed spectrum, as it is extrapolated outside of the 0.5–10 keV band. In Figure 5, we plot each model normalized to its fluence upper limit.

Zoom In Zoom Out Reset image size

It is clear that assumptions on the underlying spectral model change the implication for the X-ray–gamma-ray luminosity. If the X-ray absorption is increased, the energy limits for the models in Table 3 increase by 1–2 orders of magnitude. Furthermore, a burst that primarily emits energy outside of the 0.5–10 keV soft X-ray band, similar to the magnetar burst and flare models in Table 3, can be much more luminous than in a soft model and be undetectable by Chandra and XMM-Newton.

When searching for X-rays at the time of radio bursts and placing an associated limit, we are assuming that an X-ray burst is emitted at the same time as each radio burst and that periods of radio burst activity are correlated with X-ray activity. However, if the episodic detection of radio bursts is not intrinsic but due to amplification of the intrinsic emission from lensing by the intervening medium (e.g., Cordes et al. 2017), the detection of radio bursts and X-ray bursts may not be strongly correlated. If the radio bursts are externally amplified, their intrinsic energies could be lower by $\lesssim {10}^{2}$ than what is implied from their detection (Cordes et al. 2017). So, if the intrinsic source of FRBs also produces X-ray emission with a fluence ratio ${F}_{R}/{F}_{X}$, the expected X-ray emission could be up to two orders of magnitude lower if the radio burst is extrinsically amplified.

For all of the known FRBs at the time, Tendulkar et al. (2016) placed a limit on the fluence ratio defined as $\eta ={F}_{1.4\mathrm{GHz}}/{F}_{\gamma }$, where ${F}_{1.4\mathrm{GHz}}$ is the radio fluence at a frequency of 1.4 GHz and F_γ is the gamma-ray fluence. The most constraining limit is for FRB 010724 with $\eta \gt 8\times {10}^{8}$ Jy ms erg⁻¹ cm². If we assume our fiducial giant flare model (i.e., cutoff power-law with ${\rm{\Gamma }}=0.5$ and cutoff energy of 500 keV) with N_H = 10²² cm⁻², our gamma-ray fluence limit is 1 × 10⁻¹⁰ erg cm⁻² for a typical photon energy of 20 keV (as in Tendulkar et al. 2016). If we further assume that the 1.4 GHz fluence is approximately the same as the 2 GHz fluence, we can place a lower limit on the ratio of radio-to-gamma-ray fluence of $\eta \gt 6\times {10}^{9}$ Jy ms erg⁻¹ cm² for FRB 121102. We can also compare to our Fermi/GBM limits, which, though less constraining, do not rely on an extrapolation from soft X-rays into gamma-ray wavelengths. Using this limit the corresponding fluence ratio limit is $\eta \gt 2\,\times {10}^{8}$ Jy ms erg⁻¹ cm².

A gamma-ray burst counterpart to FRB 131104 with energy of 5 × 10⁵¹ erg has been claimed by DeLaunay et al. (2016), though it has been contested by Shannon & Ravi (2017). The implied radio-to-gamma-ray fluence ratio from the claimed detection is $\eta =6\times {10}^{5}$ Jy ms erg⁻¹ cm². DeLaunay et al. (2016) also searched for Swift/BAT sub-threshold events at any time Swift/BAT was pointed toward the source for 16 FRBs, including FRB 121102, and concluded that there is no evidence for repeated gamma-ray emission from those FRBs above Swift/BAT sensitivities. We can nevertheless compare our limits to an event similar to that claimed for FRB 131104. Our limits clearly rule out an event of that magnitude at any time during XMM-Newton or Chandra observations of FRB 121102. Further, such an event is clearly ruled out from the Fermi/GBM limits, both at the time of the bursts and at any time while FRB 121102 is visible to GBM and actively emitting radio bursts.

Models of FRBs from magnetars predict a small ratio between the fluence of radio and high-energy emission. Because the radio is a small fraction of the total emitted energy in these models, high-energy emission may be detectable. Based on analogies to solar flares, Lyutikov (2002) estimates a ratio of 10⁻⁴. Lyubarsky (2014) predicts ${10}^{-5}\mbox{--}{10}^{-6}$ based on a model of a synchrotron maser produced from a magnetized shock during magnetar activity. Our X-ray limit implies a radio-to-X-ray ratio of $\gt {10}^{-6}\mbox{--}{10}^{-8}$, depending on the spectral model and the absorbing column, which is close to the range where we can begin constraining these models.

We can also compare our limits to the most luminous magnetar giant flare emitted in our Galaxy, that of SGR 1806−20. The 2004 giant flare had a spectrum similar to that of our canonical giant flare in Table 3, a gamma-ray luminosity of ∼10⁴⁷ erg s⁻¹, and a duration of ∼100 ms (Mazets et al. 2005; Palmer et al. 2005). For the N_H $=1\times {10}^{22}$ cm⁻² case, this corresponds to a 0.5–10 keV fluence of ∼5 × 10⁻¹² erg cm⁻² at the luminosity distance of FRB 121102. This is approximately an order of magnitude lower than our corresponding extrapolated limit.

Chatterjee et al. (2017) showed that FRB 121102 is associated with a ∼0.2 mJy persistent radio source and Marcote et al. (2017) further showed, using EVN observations, that the persistent source is compact to a projected size of 0.7 pc and consistent with being coincident with the source of the FRB 121102 bursts. Two possible scenarios are considered in Marcote et al. (2017) for the persistent source: an extragalactic neutron star embedded in an SNR, perhaps producing a PWN. Alternatively, an AGN origin is considered. Here, we explore how the limit on X-ray emission from a persistent source at the location of FRB 121102 informs these scenarios.

At the luminosity distance of FRB 121102, 972 Mpc (Tendulkar et al. 2017), no nebula similar to those in our Galaxy would be visible by several orders of magnitude (e.g., the Crab Nebula would have a 0.5–10 keV X-ray flux of ∼1 × 10⁻¹⁹ erg s⁻¹ cm⁻², well below the sensitivities of current X-ray detectors). However, given the possibility that the radio bursts of FRB 121102 originate from a young neutron star (e.g., Cordes & Wasserman 2016; Katz 2016b; Lyutikov et al. 2016), a nebula resulting from either the supernova that produced the neutron star or a wind driven by either the rotational (PWN) or magnetic (magnetar wind nebula) energy of the young neutron star is an attractive model for the persistent radio source (Metzger et al. 2017).

Given the existence of a luminous persistent radio source at such a distance, one might also expect an exceptionally luminous X-ray source. Taking the Crab Nebula and scaling its X-ray flux by the ratio between its radio luminosity and the radio luminosity of the persistent counterpart to FRB 121102 (a factor of 4 × 10⁵), we arrive at a 0.5–10 keV X-ray flux of ∼5 × 10⁻¹⁴ erg s⁻¹ cm⁻². This is over an order of magnitude brighter than the 5σ limit that we placed in Section 3.4. We can therefore confidently rule out a scaled version of the Crab Nebula. However, such a nebula, powered by a young ($\lt 100$ years) pulsar or magnetar, does not have an analogue in our Galaxy and may therefore have properties different from that of the Crab. Furthermore, in the case that the nebula is a bright X-ray emitter, the soft X-rays may be absorbed by the supernova ejecta (Metzger et al. 2017). For example, assuming a Crab-like X-ray spectrum, our hypothetical "scaled-Crab" nebula would have an absorbed 0.5–10 keV X-ray flux below our 5σ limit for N_H $\gtrsim \ 5\times {10}^{23}$ cm⁻².

The persistent source coincident with FRB 121102 could also plausibly be a massive black hole, resembling the properties observed in low-luminosity active galactic nuclei (LLAGNs). It was shown by Marcote et al. (2017) that the observed persistent radio emission cannot be explained by either a stellar-mass black hole (such as in X-ray binaries) or an intermediate-mass black hole. On the other hand, the mass of a super-massive black hole is constrained by the stellar mass of the dwarf galaxy (Tendulkar et al. 2017). We would thus expect a black hole mass in the range $\sim {10}^{5}\mbox{--}{10}^{7}\ {M}_{\odot }$.

Considering our upper limit on the X-ray emission (which implies a luminosity of $\lesssim {10}^{41}\ \mathrm{erg}\ {{\rm{s}}}^{-1}$) and the radio flux density at 5 GHz (Chatterjee et al. 2017; Marcote et al. 2017), we infer a ratio between the radio and X-ray luminosities (Terashima & Wilson 2003) of $\mathrm{log}{R}_{{\rm{X}}}\gtrsim -2.7$, consistent with the ratio observed in radio-loud AGNs (Ho 2008). This ratio is also consistent with the values of ${R}_{{\rm{X}}}\sim -2$ observed in radio-loud LLAGNs (Paragi et al. 2012), and so a massive black hole resembling a radio-loud LLAGNs but scaled down in mass remains a plausible model for the persistent radio source.