FAST Observations of an Extremely Active Episode of FRB 20201124A: I. Burst Morphology

An independent data analysis for searching single pulses is carried out by the first author of this paper. The discovered bursts are verified by members of other groups in the author list, as are also reported in other papers of this series.

Before the single pulse search, all small FITS files, of which each consists of 12.88 s observation data, are merged to a single FITS file in the chronological order. To reduce the file size, 256 out of 4096 frequency channels at each edge of the L-band that have a very low gain are discarded, thus a 31.25 MHz band on each side of the band is cut off. The power of XX and YY of other frequency channels are then combined, and the channel number and sampling time are reduced by a factor of 8 and 4, respectively. The total power for a total of 448 (rather than 512, reduced by a factor of 8) frequency channels with a sampling time of 49.152 × 4 μs (i.e., downsampling by a factor of 4) are prepared for the following single pulse detection. If a pulse is detected, the original raw data for all channels are analyzed for pulse properties.

Specifically for FRB 20201124A, we search pulses in a few steps. First of all, we dedisperse data in the range of 3–1000 cm⁻³ pc in steps of 1.0 cm⁻³ pc to form two-dimensional images on the time versus DM. An artificial intelligence (AI) approach is carried out in the GNU Parallel (Tange 2020) to identify dedispersed pulses. Some frequency channels always with strong Radio frequency interference (RFI) are discarded during this process. To improve the pulse detection, we divide 448 channels into two parts, the upper half and lower half of the total band, and perform the single pulse search separately. We also search pulses in the upper and lower quarter of 448 channels. Any pulse candidates with signal-to-noise ratio (S/N) > 7 are manually viewed and checked further in the two-dimensional frequency-time water-fall plots.

In an earlier FAST paper reporting the previous active episode of the source (Xu et al. 2022) and that in Paper II (Zhang et al. 2022), the waiting time distribution of dedispersed emission peaks from FRB 20201124A exhibits two wide peaks, one from a few ms to a few tens of ms, and the other from a few seconds to a few tens of seconds. The first peak may indicate several emission components occurring in one burst, while the second peak probably stands for the interval between two adjacent independent bursts. The valley between the two peaks of the waiting time distribution can be used to distinguish independent bursts. Based on the waiting time distribution, we take the separation of 400 ms between any of two emission components as indication for the another independent burst. This allows us to better characterize and compare the emission morphology of the emission peaks within the same emission episodes. More specifically, we adopt the following definitions for the terminology:

• 1.  
  A burst is defined as all emission components with the adjacent peak separation not longer than 400 ms, according to the waiting time distribution of the emission peaks (Xu et al. 2022, Niu et al. 2022b).
• 2.  
  A sub-burst stands for one of several more or less connected components in the frequency-time waterfall plots (see Figure 2) generally within a few tens of milliseconds, and it has a distinguished peak in the dedispersed burst profile.
• 3.  
  A cluster-burst is defined as a collection of several somehow independent bursts with a separation less than 400 ms, but there is no bridge emission in between since the burst intensities come back to the system noise level.

We note that the definitions of the bursts are somewhat different in other papers in the series. This is because different papers have different scientific purposes and the lead authors defined their "bursts" for the convenience of conducting their respective analyses. Therefore, the numbers of "bursts" reported in different papers are somewhat different, even though the same data set has been analyzed.

Zoom In Zoom Out Reset image size

We obtained 29, 57, 169, and 369 bursts in the first 4 days, respectively, and in total 624 bursts from FRB 20201124A during this active episode. A full list of the bursts (see Table A1) together with plots for each bursts (see Figures A4–A13) are presented in the Appendix. The total numbers of detected bursts in different observing sessions are listed in Table 1 and presented in Figure 1. The burst rate rapidly increases on 20210927 and 20210928, reaching 381.7 bursts per hour (hr⁻¹) and about one magnitude higher than 45.8 bursts per hour as reported in the April-May observation sessions (Xu et al. 2022). The highest burst rate for FRB 20201124A on 2021 September 28, is nearly four times of the previously known highest rate of FRBs, which is 122 hr⁻¹ for FRB 20121102A (Li et al. 2021). It is worth mentioning here that our most inclusive definition of bursts has reduced the total number of bursts, which means that the burst rate is even higher if the same burst definition is adopted as Li et al. 2021 or Xu et al. (2022). More intriguing is that the FRB source was suddenly quenched after 20210928 so that no plausible bursts were detected in many following days. It is the first time to see such a dramatic change in the burst rate, and it is puzzling how an FRB source increases the burst rate exponentially, rushes to a peak rate and then is suddenly quenched.

The bursts of FRB 20201124A detected by the FAST show various dynamic spectra structures. They are detected first with a given DM value. However, when the detailed structures are studied, a proper DM is the key to determine the physical properties of the bursts (see Figure 2).

Following the approach for other FRBs in the literature (CHIME/FRB Collaboration et al. 2019b; Marthi et al. 2022; Xu et al. 2022), we first use the package DM_phase ¹¹ to find the best DM for every burst. This handy program can find the best frequency alignment of de-dispersed sub-burst features.

For bursts of FRB 20201124A, however, the outcomes of DM_phase need to be taken with caution. It gives a very reasonable result for some bursts, but not for all, mainly because of the complicated burst properties of FRB 20201124A. After examining the determined DMs and the waterfall plots for all bursts, we found that it is hard to make a good and uniform standard to determine the DMs for various bursts. Figure 2 shows some examples of the dilemma. The burst No. 33 on 20210927 shows a more reasonable DM determined from the upper half-band by using DM_phase, while it is reversed for the bursts No. 35_A and 144_A on 20210927. If one takes the leading component of the burst No.35_A as an independent burst (see Figure 2), DM_phase would derive a DM that is so large that this burst appears to have its lower part distorted toward earlier time in the water-fall plot, rather than simply vertically aligned. We get many similar cases. This fact suggests that the bursts of FRB 20201124A almost always have an intrinsic downward frequency drifting.

For this session of FRB 20201124A, we have to determine the DMs only for a small number of selected bursts which we feel confident in the DM results from DM_phase, having either a significant inter-structural gap as No. 15 on 20210925 in Figure 3 or a sharp leading edge as No. 228 on 20210928. With some well-determined DMs of a small fraction of bursts in Figure 4, we see the insignificant DM variation in about one hour each day, and the averaged DM values are consistent with each other on the first three days. A slightly smaller averaged DM value and a weaker trend of DM increasing is seen on the last day 20210928.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In the following analyses, the averaged DM for each day is adopted for data analyses. There is no question that a different DM would cause a different drifting rate, but we do not have a better choice at present.

We measured the observational parameters of 624 bursts, such as the TOA expressed in MJD for the peak of the each sub-burst, the emission peak frequency (ν₀, in MHz), the sub-burst emission bandwidth (BW_e, in MHz), the sub-burst width (W_sb, in ms), the detection S/N, and the fluence of a bursts or a sub-burst (F_ν , in mJy ms). All these parameters are listed in Table A1 in the Appendix.

For each burst, we dedisperse the data by using the average DM value each day and obtain a burst profile first from an obvious frequency range manually selected, because most bursts have emission merely in some parts of the observation band of 500 MHz of the FAST L-band receiver, see examples in Figure 3 and plots for all bursts in the Appendix. Guided by the primary burst profile, the observed burst energy over a frequency range is then fitted by a Gaussian function, so that the emission peak frequency ν₀ and the emission bandwidth BW_e are determined. Here the BW_e is defined as the full frequency width at the 10% of emission peak (FWTM) obtained from the fitted Gaussian function. The burst profile is finally obtained from the integrated data of these frequency channels. Note that the fluctuations of emission strength over the frequency channels are not caused by random noises but by the scintillation which we will discuss later. For bursts brightening at an edge of the observed band, it is difficult to have ν₀ and BW_e well determined, so we may get a ν₀ outside the range of the FAST band, or we may take the frequency boundary of the L-band receiver as the replacement. For a burst with many sub-bursts, we take the emission band to cover all these sub-bursts to get the burst profiles. The such determined emission bandwidth leads to a more accurate estimation of burst energy, because many bursts of FRB 20201124A have emission in the limited band only, rather than in all 500 MHz band of the FAST L-band 19-beam receiver.

For each sub-burst, the TOA is defined as the arrival time of the emission peak at the infinity frequency. The peak is obtained from the Gaussian function fitted to the de-dispersed burst profile. There are 44, 111, 441 and 865 sub-burst peaks in the first 4 days, respectively, and in total 1459 peaks. The TOA is then converted to the solar barycentric center using the DE438 ephemeris.

The sub-burst width W_sb is defined as the full width of 10% of maximum (FWTM) measured from a fitted Gaussian function to the dedispersed profile of a sub-burst. The burst width W_b is then defined as the overlapping width of these fitted multi-Gaussian functions for all sub-bursts with the 10% of maximum of two Gaussian functions for the two outermost components. The signal-to-noise ratio (S/N) is calculated from the summed energy of the fitted Gaussian profile relative to the standard deviation (σ) obtained from the nearby off-pulse ranges with a similar width of a burst.

The fluence of a burst or sub-bursts, F_ν in units of mJy ms, is estimated from the above-defined sub-burst parameters and the system characteristics, via

$$\begin{eqnarray}&&{F}_{\nu }=\displaystyle \frac{\sum {S}_{{\rm{i}}}\,\times \,{T}_{\mathrm{sys}}\,\times \,{t}_{\mathrm{samp}}\,\times \,{10}^{3}}{\sigma \,\times \,{G}_{0}\,\times \,\sqrt{{n}_{{\rm{p}}}\,\times \,{t}_{\mathrm{samp}}\,\times \,{\mathrm{BW}}_{{\rm{e}}}}}\end{eqnarray} \tag{ 1 }$$

here ∑ S_i is the summed value of on-burst bins, T_sys = 25 K is the system noise temperature, and G₀ = 16.1 K Jy⁻¹ is the effective gain of the telescope (Jiang et al. 2020), n_p = 2 is the number of polarization summed, t_samp is sampling time (s), and BW_e is emission frequency bandwidth (MHz) obtained above.

After obtaining the burst parameters, i.e., emission peak frequency ν₀ and the emission bandwidth BW_e, sub-burst width W_sb and fluence F_ν of every sub-burst of all bursts, we perform a statistics quantitatively for these parameters in Figure 5. Due to the limitation of the FAST observation bandwidth, some fitted ν₀ for the sub-bursts emerging near the band edges have the best fitted value outside the FAST frequency range or just near 1080 or 1420 MHz but with a BW_e < 150 MHz. These account for 21% of the total sub-bursts. The fitted parameters for these 21% sub-bursts are unusual and have a large uncertainty and are therefore dropped. The remaining approximately 79% of data as indicated by the blue symbols in Figure 5 are used for the statistical analysis.

Zoom In Zoom Out Reset image size

First of all, we notice that there are two peaks in the histogram of the ν₀ distribution, which are at 1091.9 MHz and 1327.9 MHz as the Gaussian fitting gives, and the standard deviation σ of the Gaussian distribution is 62.0 MHz and 113.0 MHz, respectively. The number of sub-bursts distributed in the lower-frequency component is larger, suggesting that this FRB source preferably emits at the lower part of the FAST band.

One may question if the two peaks of the ν₀ distribution in Figure 5 are caused by removing the strong Radio frequency interference (RFI) around 1250 MHz within the observation bandwidth of FAST. Because RFI appears mostly in a few tens of MHz in the central part of the observation band, the emission strength of the bursts from channels in the two sides of the RFI bands is not affected. Because the emission bandwidth is as wide as more than 200 MHz, discarding some channels with RFI does not affect the fitting results for both ν₀ and BW_e. This has also been verified by simulations (not presented in this paper).

The emission bandwidth BW_e in Figure 5 follows a log-normal distribution with a peak at about 277 MHz. The borderlines of the full width half maximum of the emission bandwidth are at about 193 MHz and 399 MHz, narrower than the FAST observed bandwidth.

The sub-burst width W_p in Figure 5 also follows a log-normal distribution concentrated in the range of 1 to 30 ms, with a peak at about ${7.4}_{-3.0}^{+5.0}$ ms. We notice that sub-burst width is wider for the bursts emerging at the lower part of the band. We divide the band into three sub-bands centered at 1104.2 MHz, 1250.0 MHz and 1395.8 MHz, and found that sub-burst widths have median values of W_sb of 9.8 ± 4.3 ms, 8.8 ± 4.2 ms and 7.1 ± 3.9 ms, respectively, for the sub-bursts peaking in these three sub-bands. On the other hand, the emission bandwidth becomes wider for the sub-bursts emerging at the higher part of the band, i,e, BW_e = 270.4 ± 108.3 MHz, 362.8 ± 131.2 MHz and 367.3 ± 148.2 MHz for the three sub-bands. The correlation coefficient between ν₀ and W_sb is −0.25 and that between ν₀ and BW_e is 0.36.

The distribution of sub-burst specific fluence F_ν has a dip near the peak but can be roughly fitted with two Gaussian functions in the logarithmic scale, peaking at logF_ν = 2.2 ± 0.3 and logF_ν = 2.9 ± 0.3. Details of the energy distribution of FRB 20201124A are discussed in Paper II (Zhang et al. 2022).

An important feature of FRB 20201124A is the time-frequency drifting pattern clearly shown in most bursts, not only for bursts with multiple components, but also for single component bursts. In the literature (Gajjar et al. 2018; CHIME/FRB Collaboration et al. 2019a; Hessels et al. 2019; Hilmarsson et al. 2021), frequency drifting was always discussed for the bursts with multiple components and the authors defined the shift of sub-bursts in the 2D frequency and time domain. Our observations of FRB 20201124A in Figures A1–A4 show that even the bursts with a single component also have a clear drifting pattern when the front edge of burst is aligned with the average DM, as shown in the example presented in Figure 6.

Zoom In Zoom Out Reset image size

First we look at drifting of single component bursts. We cut the emission band of the burst BW_e to three sub-bands, then get three averaged burst profiles from each sub-band, and obtain the TOAs of sub-burst peaks with Gaussian fitting. The drifting rate, R_d in MHz/ms, is obtained as the slope of the least square fit to TOAs for the three sub-bands, as demonstrated in Figure 6. The drifting rate R_d depends on the implemented DM value for the burst, and the data in Table 2 are obtained with the average DM of the day. The uncertainty of R_d depends on the uncertainty of the central frequency and the TOAs in each sub-band. In principle, we can measure R_d for each sub-burst. Nevertheless, to examine how the bursts drift, we choose only 31 single-component bursts with BW_e > 250 MHz and burst detection S/N > 50. Their drift rates R_d are listed in Table 2 and plotted against the emission peak frequency in Figure 7.

Zoom In Zoom Out Reset image size

For multi-component bursts, we selected 12 bursts with at least three sub-bursts. Each sub-burst is taken as a component to obtain the central emission frequency ν₀ and the TOA of the frequency scrunched profiles, e.g., the multi-component burst No. 15 on 20210925 in the left panel of Figure 3. The precursor and post-cursor near the band edges are excluded. We confidently obtain the frequency drifting rate for these 12 multi-component bursts listed in Table 2 and their values are also plotted against the emission peak frequency in Figure 7.

As listed in Table 2 and seen in Figure 7, all confidently and quantitatively derived frequency drifting rates have negative values, which indicates the downward frequency drifting of bursts. The drifting rate is plainly understandable in the frequency-time waterfall plot, though the exact values depend on the implemented DM value. As seen in Table 2, the drifting rates for very carefully selected single-component bursts are in the range between −166 and −28 MHz ms⁻¹, and their average value is −61 ± 9 MHz ms⁻¹. The drifting rates for multi-component bursts represent the average delay of several components in different parts of the frequency band, hence they have much lower values than those of single-component bursts. The drifting rates for multi-component bursts are in the range from −29 to −5 MHz ms⁻¹, with an average value of −21 ± 4 MHz ms⁻¹. This is only about one third of the drifting rate for single-component bursts. The bursts emerging in the higher half band tend to have a larger drifting rate than those detected only at the lower part of frequency band.

With the implemented daily-average DM to all the bursts, one can systematically study the burst morphology. One immediate observation is that the bursts show a diverse morphology. The primary features of the bursts are the frequency drift and the limited emission band. The number of sub-bursts, i.e., the component of burst profiles, is another interesting feature. We therefore can classify the detected bursts from FRB 20201124A based on these key features.

First of all, based on the frequency drifting, more than half of 624 bursts have shown a downward frequency drifting feature in the waterfall plots. Only a few cases show an upward drifting feature. The remaining bursts are either complex, show no drifting with the given DM, or have no evidence for drifting due to the limited emission bandwidth caught by FAST. Second, only a small fraction of bursts, which we categorize as "wide-band", are detected in the entire FAST observation band. Most of the bursts are detected only in a narrow part of the FAST L-band of 1.0–1.5 GHz, either in the high frequency part, the middle part, or the lower part. Therefore, the sub-classes can be distinguished according to the emission band where the bursts appear. Moreover, for the downward drifting bursts, the sub-classes can be further grouped according to the number of burst components. See Table 3 for the number of bursts in each class or sub-class. Examples are shown in Figure 8 for the dynamic spectrum of frequency downward-drifting bursts with one (in the left column panels), two (in the middle column panels) and multiple components, with their emission seen in wide-band, high-frequency, middle-frequency band, or low-frequency part of the FAST band (from top to bottom).

Zoom In Zoom Out Reset image size

In the following, we briefly discuss each class and sub-class of bursts.

3.3.1. One-component Bursts with Downward Frequency Drifting: D1-W, D1-H, D1-M and D1-L

3.3.2. Two-component Bursts with Downward Frequency Drifting: D2-W, D2-H, D2-M and D2-L

3.3.3. Multi-component Bursts with Downward Frequency Drifting: Dm-W, Dm-H, Dm-M and Dm-L

3.3.4. Upward Frequency Drifting Bursts

Three bursts have the second or the third component appearing at a higher frequency band than the first one, so that they look like an upward frequency drifting as shown in Figure 9. One of them have emission in the higher part of the band, denoted as subclass U-H, and two of them have emission in the lower part of the band, denoted as subclass U-L.

Zoom In Zoom Out Reset image size

3.3.5. Bursts with no Evidence for Drifting

3.3.6. No Drifting Bursts

3.3.7. Complex

Frequency drifting structures have been observed in many FRB bursts, e.g., FRB 121102, FRB 180814.J0422+73, FRB 180916.J0158+65 and FRB 190711 (Gajjar et al. 2018; Michilli et al. 2018; CHIME/FRB Collaboration et al. 2019a, 2019b; Hessels et al. 2019; Josephy et al. 2019; Caleb et al. 2020; Chawla et al. 2020; Day et al. 2020; Fonseca et al. 2020; Pastor-Marazuela et al. 2021a; Chamma et al. 2021; Hilmarsson et al. 2021; Pastor-Marazuela et al. 2021b; Platts et al. 2021). The drifting rate varies in a wide range for each FRB, and it may be related to other parameters. For example, Chamma et al. (2021) found that the drifting rate is inversely correlated with the widths of the sub-bursts. Our results of sub-burst width of 7.4 ms and drifting rate of multi-component burst R_d = −21 MHz ms⁻¹ of FRB 20201124A are slightly higher but consistent with the result of Hilmarsson et al. (2021). Wang et al. (2022) investigated the relationship between the drifting rate and the burst emission frequency and found an anti-correlation. Our mean drifting rate R_d = −21 MHz ms⁻¹ at 1250 MHz, as well as the drifting rate of bursts with multi-component, which is shown in Figure 10 with parameters that have been converted to the rest frame of the host galaxy, is consistent with the predication given by Wang et al. (2022).

Zoom In Zoom Out Reset image size

Frequency drifting and narrow-band emission features as shown from FRB 20201124A have been also observed from other radio sources. For example, PSR J0953+0755 has emission in a narrow band at a low frequency band of 18–30 MHz and shows a sub-pulse frequency drifting structure (Ulyanov et al. 2016). Bilous et al. (2022) observed the sub-pulse drifting structure of PSR J0953+0755 at several tens of MHz. The giant pulses of Crab (Thulasiram & Lin 2021) often show narrow-band emission at various center frequencies with much pulse broadening at relatively lower frequencies which are not caused by scattering. These features are somewhat similar to the drifting behavior of the FRB 20201124A bursts.

Scintillation is a distinct feature of FRB 20201124A. In order to test if the scintillation bandwidth changes within one hour observation, we calculate the scintillation bandwidths of strong bursts on 2021 September 28 using the autocorrelation function (ACF) method (Cordes et al. 1986) independently in three sub-bands: the lower band of 1031.74–1141.11 MHz, the medium band of 1195.31–1304.69 MHz, and the higher band of 1359.86–1469.24 MHz. The results are shown in Figure 11. The mean values of scintillation bandwidths are 0.456 ± 0.035 MHz, 0.772 ± 0.057 MHz, and 1.325 ± 0.285 MHz for the three sub-bands, respectively. Obviously, the scintillation bandwidths increase with the observation frequency. The variations of the calculated scintillation bandwidths are smaller than the width of one frequency channel (0.122070 MHz) in the lower and middle sub-bands. There is no systematic variation of scintillation bandwidth with time in one-hour sessions of FAST observations.

Zoom In Zoom Out Reset image size

To get a clear relation between scintillation bandwidths and the observing frequencies, we integrated the energy of all bursts for each frequency channels, and then analyzed scintillation in a car-box sub-band of 1/4 band. The results are shown in Figure 12 for the center frequency ν and scintillation bandwidth BW_sc, which can be fitted by a power-law function ν_s = a ν^γ , where γ = 3.0 is the power-law index.

Zoom In Zoom Out Reset image size

The scintillation bandwidth increases from about 0.5 MHz at 1.05 GHz to 1.4 MHz at 1.45 GHz in our observation frequency, with a mean scintillation bandwidth of about 0.94 MHz. The power-law index γ = 3.0 ± 0.2 is smaller than the value of 3.5 ± 0.01 obtained by Main et al. (2022), or 4.9 reported by Xu et al. (2022), or the theoretical value of 4.0 or 4.4 of Kolmogorov spectrum. The corresponding scattering timescale of about 0.31 μs is consistent with the results of Main et al. (2022), which is much smaller than the sampling time of 49.152 μs of our observation. Therefore scattering has an almost negligible effect on the morphological study for the bursts of FRB 20201124A. For the large number of bursts we observed, the properties shown in this paper should be intrinsic to the FRB 20201124A source.

We emphasize at the end of this paper that the sensitive observations by a large radio telescope such as the FAST is very important to detect rich features of bursts and reveal their properties. The classification scheme and other results presented in this paper are hard to be achieved without such sensitive observations. We also realize that multi-epoch wide-band observations are fundamental to understand the environment and physical mechanisms of FRB emission.