The FAST Galactic Plane Pulsar Snapshot Survey. II. Discovery of 76 Galactic Rotating Radio Transients and the Enigma of RRATs

Time-domain astronomical observations reveal more and more Galactic or extragalactic transients in many electromagnetic wave bands, rested on the new development of high sensitivity detectors. The detection of the Gamma-Ray bursts (see review by Fishman & Meegan 1995), Fast Radio Bursts (FRB, Lorimer et al. 2007) and extrasolar planets (e.g., Mayor & Queloz 1995; Butler & Marcy 1996) were the most successful. Radio pulsar detection, even in the first discovery (Hewish et al. 1968), has been working in the time domain. It started with a high time resolution of seconds, later milliseconds (e.g., Burke-Spolaor & Bailes 2010), and nowadays microseconds or even to nanoseconds (e.g., Hankins et al. 2003). For a bright pulsar, each individual pulse from every pulsar period can be directly observed if a radio telescope is sensitive enough. For a weak pulsar, signals have to be integrated over many periods, so that a mean pulse profile is obtained. Because of the interstellar medium, the pulses must be dedispersed to diminish the delay between frequency channels in an observational radio band.

Most known pulsars were discovered through searching in the Fourier domain by using the packages such as PRESTO ⁹ (Ransom 2011) and SIGPROC ¹⁰ (Lorimer 2011), or using the fast-folding algorithm (Staelin 1969; Kondratiev et al. 2009; Cameron et al. 2017; Parent et al. 2018) in the time domain. The enhanced red noise from the observation system caused by the fluctuations of the receiver gain, system temperature and radio frequency interference (RFI) affects the detection of long-period pulsars in the Fourier domain. Cordes & McLaughlin (2003) developed a new single-pulse search technology, finding enhance points in dedispered data series. When the signal-to-noise ratio (S/N) is much larger than a given threshold, then radio transients can be found as an isolated distribution in the image of dispersion measurement (DM) versus time. By using this technology, McLaughlin et al. (2006) discovered the first rotating radio transients (RRATs), and Lorimer et al. (2007) discovered the first FRB. This technology is effective for searching for bright radio transients, and has been used in many pulsar surveys (Cordes et al. 2006; Hessels et al. 2008; Deneva et al. 2009, 2016; Burke-Spolaor & Bailes 2010; Keane et al. 2010, 2011, 2018; Burke-Spolaor et al. 2011; Coenen et al. 2014; Stovall et al. 2014; Karako-Argaman et al. 2015).

As defined by Burke-Spolaor & Bailes (2010), RRATs are often considered as a special group of pulsars that emit bright pulses sporadically, making them difficult to find in normal periodic search methods. They are usually discovered by a single-pulse search method. Up to now, more than 160 RRATs have been discovered, see the list in the RRATalog ¹¹ and also a recent summary by Abhishek Malusare et al. (2022), plus more new RRATs recently discovered (Patel et al. 2018; Tyul'bashev et al. 2018a, 2018b, 2022; Good et al. 2021; Han et al. 2021; Bezuidenhout et al. 2022; Dong et al. 2022). The number of known RRATs claimed in literature (Abhishek Malusare et al. 2022) is about or more than five percent of the known pulsars. It is quite possible that many RRATs or long-period pulsars are missed in some pulsar surveys, if the single-pulse search has not properly been carried out for surveys (see Keane et al. 2010). The discovery of more RRATs can help to understand the neutron star population in the Galaxy.

To understand why the pulses radiate very occasionally, sensitive observations and detailed statistics of long-term observations are desired (Bhattacharyya et al. 2018). However, only a small fraction of RRATs have been well studied. The energy distribution or peak flux distribution of single pulses from a very sensitive observation could give the answer to how an RRAT behaves. For a generally very weak pulsar, it could appear as an RRAT if only a few bright pulses are detectable for a given sensitivity. The energy of normal pulses of a pulsar always follows a log-normal distribution, while abnormal bright pulses should show an extra component at the high energy end as if they are giant pulses. This has been verified by some RRATs, such as PSR J1846−0257 ¹² (Mickaliger et al. 2018). For other RRATs, Cui et al. (2017) and Mickaliger et al. (2018) did not find the tail beyond the log-normal distribution, indicating that the pulses are normal but just very nulling (Burke-Spolaor & Bailes 2010), so that they radiate pulses by the same emission mechanism as pulsars. Then, the concerns are turned to why RRATs are very nulling. Some RRATs, e.g., PSR J0941−39 (Burke-Spolaor & Bailes 2010) and PSR J1752+2359 (Sun et al. 2021), can have two emission states, one with occasional bright pulses and one with a very nulling state or normal pulsar state. Lu et al. (2019) observed three RRATs by using the Five-hundred-meter Aperture Spherical radio Telescope (FAST, Nan et al. 2011), and they detected many bursts for PSRs J1538+2345 and J1854+0306, while PSR J1913+1330 radiates pulses continuously but clearly shows a nulling nature. Based on the FAST observation of PSR J0628+0909, Hsu et al. (2023) concluded that this known RRAT has weak single pulses as a normal pulsar with transient-like strong pulses, similar to PSR J0659+1414 (B0656+14, Weltevrede et al. 2006).

To understand the sporadic behavior of RRATs and the switching of different emission states, the polarization observations are desired not only for bright single pulses but also for weaker normal pulses. Previously available measurements were made for single pulses of PSR J1819−1458 by Karastergiou et al. (2009), of 17 RRATs by Caleb et al. (2019), of PSR J1752+2359 by Sun et al. (2021) and of PSR J1905+0849g by Han et al. (2021). Mean polarization profiles of only a few RRATs have been measured (Karastergiou et al. 2009), which in fact is the key to understanding the radiation mechanism and radiation geometry.

Timing is fundamental for a rotating neutron star to get the period derivative measured, and hence the characteristic age and magnetic field derived. For RRATs, it is challenging to do timing, because only a small number of strong pulses can be detected in a limited observation time. The initial discovery position of detection is often poorly given. Single pulses probably emerge randomly in the averaged pulse emission window. A mean pulse profile can therefore rarely be obtained because of the sparseness of pulses, and the timing residual by using these individual pulses could be as large as a hundred milliseconds (Bhattacharyya et al. 2018). The available timing results for several RRATs in the P-Pdot diagram (McLaughlin et al. 2009; Keane et al. 2011; Cui et al. 2017; Bhattacharyya et al. 2018) show that long-period RRATs look like either magnetars if they have a large period derivative, or simply aged pulsars near the death line if they have small period derivative, or other emission mechanisms such as the interaction between vacuum neutron star magnetosphere and high-energy photon from the cosmic gamma-ray background (Istomin & Sobyanin 2011a, 2011b). Statistical analysis of populations of known RRATs by Abhishek Malusare et al. (2022) shows no correlation with nulling pulsars, nor tending to be older and closer to the death line. It is very surprising that a glitch detected from PSR J1819-1458 (Lyne et al. 2009), which indicates that some RRATs may be just young pulsars with merely bright single pulses being detectable.

The FAST (Nan et al. 2011) has a larger collecting area with an aperture diameter of 300 m. Mounted with a 19-beam L-band receiver with a system temperature of 22 K (Jiang et al. 2020), it has a great sensitivity in studying the universe in radio waves, such as the discovery of weak pulsars (Han et al. 2021; Pan et al. 2021; P. Wang et al. 2021), and the discovery of new FRBs (Zhu et al. 2020; Niu et al. 2021, 2022) or study the details of known pulsars (Wang et al. 2023) and FRBs (Luo et al. 2020; H. Xu et al. 2022). At present, the FAST Galactic Plane Pulsar Snapshot (GPPS) survey ¹³ (Han et al. 2021) is hunting pulsars in the Galactic plane, and has discovered more than 500 pulsars already. In the first paper by Han et al. (2021), the GPPS survey strategy is introduced and the discoveries of 201 pulsars and 1 RRAT have been presented, together with parameter improvements for 64 previously known pulsars. In this second paper, we report the discoveries of 76 Galactic transient sources from our single-pulse search, and also present FAST observations of previously known RRATs. In Section 2, we briefly summarize the GPPS survey observations, and introduce our single pulse search module and the data processing. All the newly discovered results are presented in Section 3. During the GPPS surveys, the covers with known pulsars have been observed for system verification in every observation session. With such large a mount of data, together with targeted observations of some RRATs in several normal applied FAST projects (see details below), 59 previously known RRATs have been observed by FAST as presented in Section 4, which can tell the RRAT enigma. A summary and discussions are given in Section 5.

2.1. The GPPS Survey and other Observations

2.2. Data Analysis

In principle, the single-pulse search simply has three steps: (1) dedisperse data, (2) pick out pulses above a threshold of significance over the system noise, and (3) sift candidates. For example, in the single pulse search pipeline on PRESTO (Ransom 2011), the first step is to use the prepsubband to dedisperse data, and then the second step is to find the outstanding data points in each dedisperse data based on the software single_pulse_search.py. Finally, the classifier RRATtrap ¹⁴ is used for candidate sifting, which is a single-pulse sifting algorithm used in the Green Bank North Celestial Cap (GBNCC) survey and Low Frequency Array (LOFAR) survey, developed by Karako-Argaman et al. (2015) and updated by Patel et al. (2018).

After a huge number of outstanding data points are found in dedispersed data, a skillful job is the step for candidates sifting. Many software packages have been developed for the "classifier." In the package Clusterrank ¹⁵ used in Arecibo 327 MHz Drift Pulsar Survey (AO327), Devine et al. (2016) developed a machine-learning classifier for identifying and classifying dedispered pulse groups in single-pulse search outputs. Agarwal et al. (2020) developed a new classifier to recognize the original achromatic waterfall image and the diagram of the occurrence DM-time of candidates. Other similar packages have been developed by Men et al. (2019) and Zhang et al. (2021) as well. Michilli et al. (2018) introduce the machine-learning classifier L-SPS ¹⁶ (based on SPS, ¹⁷ Michilli & Hessels 2018) to discriminate astrophysical signals and RFI for the LOTAAS. All these classifiers are basically working on the bright points in the DM-time image, which is output by single_pulse_search.py or other similar codes. The parameters for such a single pulse search must be set differently for different projects by different telescopes, and they directly affect candidate classification, otherwise narrow, weak or scattered pulses could be mis-eliminated. In general, these classifiers are hard to work in real time.

With the developments of the Graphics Processing Unit (GPU) and Artificial Intelligence (AI), the single-pulse search has adopted these technologies to improve the computation efficiency. For example, Petroff et al. (2014) searched for fast radio transients by using HEIMDALL ¹⁸ based on GPU devices for dedispersion and candidates clustering, which can process data in real time and directly produces a list of candidates for multi-beam surveys. Magro et al. (2011) developed a GPU-based real-time transient search machine.

2.2.1. Single Pulse Search in the GPPS Survey

The single-pulse search module has been developed (see Figure 1) and applied to the GPPS survey data.

Zoom In Zoom Out Reset image size

The original data of 2048 channels covers the frequency range of 1000–1500 MHz. The channels within 31.25 MHz from each edge of the band are discarded due to their low gain. The original data of other channels for polarization channels XX and YY are scaled according to their root mean square (rms) and added together for the total power, and then deposited into the data1j repository for pulsar searching. The details for data preparation are described in the first paper (Han et al. 2021). The single pulse search module is taking these fits files from the data1j repository.

We have developed a source code to do quick dedispersion and generate DM-time images in GPU. The process includes three steps: loading data, dedispersion in the range from 3 pc cm⁻³ to 1000 pc cm⁻³ with a step of 1.0 pc cm⁻³, and plotting DM-time images for every 4 s of data. This process consumes only about 0.25 s of computation time in one node of the FAST computing cluster. In addition to recognizing the pulse signal with a high DM, mostly for searching FRBs, images for three other DM ranges are plotted and radio transients are searched for independently: 900 – 1900 pc cm⁻³, 1800 – 2800 pc cm⁻³ and 2700 – 3700 pc cm⁻³. Because single pulses also rarely have a width less than one millisecond, after loading the data, we down-sample data from 49.152 μs to 4 × 49.152 = 196.608 μs and also combine every four frequency channels together. This not only improves the detection efficiency of single pulses, but also makes the subsequent data processing fast. After the data are dedispersed, in order to suppress the noise level and preserve the pulse features, the "fifth-order median filter" is applied to the dedispersed data, in which the median of in a sliding window of 5 × 5 data points in the two-dimensional Time-DM image is taken to replace the original point value. This filtering process can effectively smooth out strong RFI emerging in only one or two data bins and preserve the original features of the image. Because some transient pulses have a steep spectrum or an inverse spectrum, we also make two more DM-time images independently for data in the upper half and lower half of the band, i.e., channels below and above 1250 MHz, and plot an image every 1.5 s. This operation ensures an effective detection of narrow and weak pulses with a steep spectrum, especially for repeating FRBs (Zhou et al. 2022).

After the above-mentioned DM-time images are generated for a beam or even a cover, we used the YOLO object detection technology to recognize the outstanding points, which is developed in the Darknet ¹⁹ neural network framework for searching the distinctive features in the DM-time images. At the beginning of this single pulse search for the GPPS survey, the yolov3.cfg network file of YOLOv3 ²⁰ (Redmon & Farhadi 2018) version was used. In 2021, we switched to YOLOv4 (Bochkovskiy et al. 2020) and the new yolov4_csp.cfg ²¹ network file (C.-Y. Wang et al. 2021). After the huge number of images sifted by this AI technology, the left images of single-pulse candidates are manually examined, and the final burst candidates are listed with the DM and time of arrival (TOA, in MJD) for further verification observations and period-finding (see below).

Most of our single-pulse searches of the GPPS survey have been carried out by using the GPU node of the computing cluster in the FAST data center. For data of 76 beams from one cover of the GPPS snapshot observation of 21 minutes, we organize the multitask parallel computing in the GeForce RTX 2080Ti GPU*4 with the program of GNU Parallel (Tange 2018). The processing costs about 9 minutes for the single-pulse search in the DM range from 3 to 1000 pc cm⁻³.

2.2.2. Sensitivity and Pulse Parameters

For any pulse detected from the single pulse searches of the GPPS survey data, some parameters must be determined.

The first is the DM value estimated by using the method given by Zhu et al. (2020), in which the most refined pulse structure is preserved when turning DM values, and the uncertainty is given as being the half-width of the Gaussian function fitted for the curve of $\sum {\left(\tfrac{d}{{dt}}\right)}^{2}$ over DMs. If a number of pulses are detected with the almost the same DM, they probably come from the same source. The final DM is determined by the brightest pulses.

Second is the time of arrival (TOA) of a pulse. We dedispered all the candidates and get the time of arrival (TOA, in MJD) at the pulse peak. This TOA must be converted to the time at the solar barycentric center using the DE438 ephemeris.

For a single detected pulse, we have to get other necessary parameters, such as the observed pulse width (W_obs) and the signal-to-noise ratio (R = S/N). Supposing the pulse in n bins has an equivalent width of W_obs (or simply written as W in units of ms in tables or figures), the R = S/N of the pulse detection can be estimated from the total energy of n on-pulse bins (∑S_i ) divided by the standard deviation (σ) of a nearby off-pulse range in the dedispersed data series, as

$\begin{eqnarray}&&R={\rm{S}}/{\rm{N}}=\displaystyle \frac{\sum {S}_{i}}{\sigma \cdot \sqrt{n}}.\end{eqnarray} \tag{ 1 }$

Note that the observed pulse width is a combination of the intrinsic pulse width with other factors. In addition to the sampling time of the dedispersed data t_bin = 196.608 μs, for a high DM pulse, the scattering broadening time τ_s has to be taken into account, which can be estimated as the following (e.g., Karako-Argaman et al. 2015):

$\begin{eqnarray}&&\mathrm{log}({\tau }_{s})\approx a+b\cdot \mathrm{log}(\mathrm{DM})+c\cdot {\mathrm{log}}^{2}(\mathrm{DM})-\alpha \cdot \mathrm{log}(\nu ),\end{eqnarray} \tag{ 2 }$

here a = −3.59, b = 0.129, c = 1.02 and α = −4.4. The dispersion time inside one frequency channel with a bandwidth BW_chan = 0.97656248 MHz:

$\begin{eqnarray}&&{\rm{\Delta }}{t}_{\mathrm{chan}}=4.7\,\mu {\rm{s}}\cdot \displaystyle \frac{{\mathrm{BW}}_{\mathrm{chan}}}{24.414\,\mathrm{kHz}}\cdot {\left(\displaystyle \frac{\nu }{350\,\mathrm{MHz}}\right)}^{-3}\cdot \displaystyle \frac{\mathrm{DM}}{{\mathrm{cm}}^{-3}\,\mathrm{pc}},\end{eqnarray} \tag{ 3 }$

must be concerned. Therefore the observed pulse width W_obs could be estimated from the intrinsic pulse width W_intrinsic with these broadening, as being

$\begin{eqnarray}&&{W}_{\mathrm{obs}}=\sqrt{{W}_{\mathrm{intrinsic}}^{2}+{t}_{\mathrm{bin}}^{2}+{\tau }_{s}^{2}+{\rm{\Delta }}{t}_{\mathrm{chan}}^{2}}.\end{eqnarray} \tag{ 4 }$

The sensitivity of the single-pulse detection at a given DM can be estimated from the pulse width W_obs and σ of the dedispersed data. The σ of the data flow can be related to the system noise by

$\begin{eqnarray}&&\sigma =\displaystyle \frac{{T}_{\mathrm{sys}}}{{G}_{0}\sqrt{{n}_{p}\cdot {t}_{\mathrm{bin}}\cdot \mathrm{BW}}},\end{eqnarray} \tag{ 5 }$

where T_sys is the system noise temperature in units of K, 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_bin is the time of a down-sampled bin, BW = 437.5 MHz is the frequency bandwidth (MHz) after removing the upper and lower sidebands of 31.25 MHz.

The capability of the single pulse search must therefore consider the observed pulse width and the system noise of σ. For pulses with a high DM and an intrinsic pulse width W_intrinsic, one can estimate the minimum flux density of detected pulse at a signal-to-noise ratio of ${\rm{S}}/{\rm{N}}\gt {R}_{\min }$ (Cordes & McLaughlin 2003; Deneva et al. 2016) as being:

$\begin{eqnarray}&&{S}_{1250}={R}_{\min }\displaystyle \frac{{W}_{\mathrm{obs}}}{{W}_{\mathrm{intrinsic}}}\displaystyle \frac{{T}_{\mathrm{sys}}}{{G}_{0}\cdot \sqrt{{n}_{p}\cdot \mathrm{BW}\cdot {W}_{\mathrm{obs}}}}.\end{eqnarray} \tag{ 6 }$

Taking the minimum detection threshold of ${R}_{\min }$ = 7, one can get the sensitivity curves for different DM values as shown in Figure 2.

Zoom In Zoom Out Reset image size

The fluence of each single pulse, F in units of Jy s at 1250 MHz, is simply the integration of the pulse over the width (t_bin, in units of s) as being F = ∑S_i · t_bin. The method for estimating the mean flux densities of pulsars with periods estimated is consistent with that described in Section 4 of Paper I (Han et al. 2021).

2.2.3. Period Finding

A possible period for detected pulses with the almost same DM has always been searched. If a period is found, the bright pulses obviously come from a pulsar or an RRAT.

For a sequence of TOAs of pulses with almost the same DM, one can get N(N −1)/2 differences for those TOAs. We test a number of periods in a range, and check the residual rms for these TOAs and TOA differences for a given period of P. One would get complete noises for a series of residual rms if there is no period and periodic derivative. Nevertheless, if there is a period, the very distinct minimum of residual rms could be found, and the residual is then taken as the uncertainty of the so-found period. The data are then folded with the best DM and the period P, and the average pulse profile is obtained. We wrote a small program "PF" for the purpose, and obtained the best periods from TOA sequences for many sources. After we got this job done, we found that the method is very coincident with that reported by Keane et al. (2010).

For bright pulses and the mean profile of a pulsar or an RRAT, we get their polarization profile by using the package PSRCHIVE (Hotan et al. 2004), after the polarization data are calibrated and the Faraday rotation is corrected (Wang et al. 2023).

After processing the GPPS survey data obtained so far, we discover 76 Galactic transient sources, which are all taken as RRATs since they are discovered in single-pulse searches rather than the periodicity search, see Table 1 for the list.

Among them, we detect 105 pulses from 26 transient sources. Merely a few pulses from each source, as discussed in Section 3.1, are not enough for the "PF" to find a period, though we believe they should be RRATs. We simply take these sources as the first subclass of newly discovered RRATs with a period to be found. For the other 50 sources, based on the discovered observation or follow-up observations, we classify them into three categories. The first is the prototypes of RRATs, in short "proto-RRATs", from which non-consecutive single pulses are detected in the available observations but a period can be identified from these sporadic bright pulses. In the other periods rather than these with sparse pulses, no significant pulsed emission is detected from any individual period, or only very weak emission is detected from the averaged for many periods. We discovered 16 such proto-RRATs. The second category is extremely nulling pulsars, which have no emission detected in the most (≳90%) of observation duration, but some pulses are detected consecutively for some periods. We discover 10 such extremely nulling pulsars. The last category is weak pulsars with sparse strong pulses. They mostly stay in a very weak emission state, but occasionally have strong single pulses. The two emission states are very clearly shown in the individual pulses. We discovered 24 weak pulsars with sparse strong pulses.

In the following, we discuss each kind of newly discovered source first, and then present the polarization observation results for strong pulses or the averaged profiles of detected pulses.

3.1. Galactic Transient Sources with Merely Few Pulses

Very few pulses have been detected by FAST for 26 transient sources, as listed in the first part of Table 1 and shown in Figure 3 for examples and Figure A1 for pulses of all detected transient sources in the Appendix. Mostly 1 or 2 pulses were detected in 5- or 15-minute observations, and sometimes even non-pulses are detected in the following-up observations for 15 or even 60 minutes. Note that the transient sources listed in Table 1 at least have two pulses, so they have "repeating radio bursts" with almost the same DM, and the DM in Table 1 is obtained from the strongest pulse. If only one pulse is detected in the first observation, we have to catch at least another one in the following observations. J1853+0209g, for example, one pulse was detected on 2020 August 12, while no pulse is detected in many sessions such as the 15-minute observation on 2021 June 27, the 60-minute observation on 2021 October 14, one 30-minute observation on 2022 June 2, and one 50-minute observation on 2022 November 12. Fortunately, one pulse is detected in the 60-minute observation on 2022 August 24. Therefore, two single pulses have been detected by FAST observations of in total 290 minutes. For J1855−0154g we get 3 pulses from the 5-minute survey pointing on 2021 September 7, but none from the following-up observations for 15 minutes on 2021 October 4, and for 60 minutes on 2022 November 16. For all these transient sources, the pulse detection rate is very low. It is very hard to get a period from the TOAs of so few pulses from all observations of each source.

Zoom In Zoom Out Reset image size

For these transients, we cannot do much but present the two-dimensional dynamic spectra of the dedispersed pulses, see Figures 3 and also Figure A1 in the Appendix. The pulse morphology of these transient sources is rich. In addition to the single peak pulse, we do see that some pulses have two peaks, such as J1828−0003g at 2022 August 3, and 2022 November 14, J1853+0353 on 2021 June 24, or one sharp peak plus a weak peak, e.g., J1847−0046g on 2020 June 6. Pulses detected from one source in different epochs are often different in shape. Some pulses have a very wide profile, such as J1853+0209g, reaching hundreds of milliseconds.

In Table A1 in the Appendix, we list parameters for each pulse, including the TOA of the pulse peak, S/N of dedispersed pulse R, pulse width in millisecond, and the fluence discussed in Section 2.2.2. We will discuss the polarization of some strong pulses in Section 3.5.

The DM values of the pulses from these transient sources are all smaller than that the estimated upper limits given by the Galactic electron distribution models (Cordes & Lazio 2002; Yao et al. 2017), except for J2030+3833g who has a high DM of 417 ± 6 pc cm⁻³ which is near the DM upper limit of 402 pc cm⁻³ and 432 pc cm⁻³ given by NE2001 (Cordes & Lazio 2002) and YMW2016 (Yao et al. 2017), respectively. Therefore we conclude that all these transient sources are inside our Milky Way.

3.2. Newly Discovered Proto-RRATs

Based on the individual pulses detected by the single-pulse module from the survey data and also from the longer follow-up observations, we get a period from the TOA value of pulses for 50 transient sources (see Section 2.2.3), see the second part of Table 1 for the list. We therefore confirm that these 16 sources are characterized as proto-RRATs with occasionally emitted pulses. With a newly found period, the detected pulses can be well aligned, see an example in Figure 4. Such plots for newly discovered RRATs in the GPPS survey are shown in Figure A2 in the Appendix. In general, we can get much better average profiles for these detected pulses over 3σ, but cannot detect significant emissions from other periods though they are all added together. Such proto-RRATs cannot be detected via general normal pulsar periodicity search, and their integrated profiles over all periods in an observation session do not have a good S/N.

Zoom In Zoom Out Reset image size

For some proto-RRATs with several observation sessions, the pulse detect rate is somehow a constant, such as about 24 pulses per hour for J2005+3156g. For some RRATs, the pulse strengths and the detection rates vary a lot, see N_d /N_p and T_obs in Table 1 for J1905+0558g as examples. For some RRATs, however, a period is obtained fortunately from few detected pulses. For example, both J1857+0229g and J1924+1446g have only one pulse detected in the 5 minutes survey data, and their periods are obtained from only nine pulses detected from the expensive 1-hour observations. No such fortune is available for many transient sources discussed above, such as J1850−0004g, J1855+0033g and J1916+1142Ag.

3.3. Newly Discovered Extremely Nulling Pulsars

In addition to detecting sparse pulses from RRATs discussed above, the single pulse search module can also pick up successive pulses for several periods but merely emerge in a small fraction of observation time, so that they cannot be picked out via the periodicity pulsar search. No detection of the pulsed flux for other periods than bright pulses implies their nature of very nulling. We get 10 new extremely nulling pulsars from the single-pulse searching, as listed in the third part of Table 1. Figure 5 shows one example, while all newly discovered very nulling pulsars are shown in Figure A3 in the Appendix.

Zoom In Zoom Out Reset image size

Such extremely nulling pulsars have an advantage for period determination. Catching the emission active episode sometimes is not easy for some objects, so there is no detection of any pulses (Np = 0) in some observation sessions for e.g., PSR J1845−0008g and J1855+0240g is possible, see Table 1. PSR J1845−0008g is a particularly interesting nulling pulsar. We detect pulses in three sessions, but not the other three. For PSR J1855+0240g, successive pulses are merely detected in a short episode during the 15-minute observation.

The mean flux density 〈S〉 of these pulsars in Table 1, calculated from the averaged flux over all periods in the observation session, should be taken as the upper limit.

3.4. Weak Pulsars with Sparse Strong Pulses

For 24 weak pulsars listed in the last part of Table 1, the single pulse search module can also pick up sparse strong pulses from the 5-minute snapshot observations. The normal pulsar searching cannot pick up these objects due to the low S/N. They may be detected as a pulsar with the same DM and period from the follow-up 15-minute tracking observations. Figure 6 shows one example, and plots for all the 24 weak pulsars are presented in Figure A4 in the Appendix.

Zoom In Zoom Out Reset image size

Some newly discovered pulsars are remarkable in some aspects. PSR J1849+0619g (GPPS0522) is discovered via the single pulse search, one pulse in G38.56+3.30-M14P1 on 2022 October 1 and another one pulse in J184950+062304-M04 on 2022 November 11 with the same DM of 110 ± 1 pc cm⁻³. Together with the 4 pulses discovered in a recent verification observation on 2023 February 28, the period is found from this observation that is about 2.011230 s via the careful period search at this DM. PSR J1859+0239Bg (GPPS0526) is discovered by only one pulse on 2022 May 29, in the data for timing PSR J1859+0239Ag (GPPS0091, i.e., PSR J1859+0239g reported in the Paper I Han et al. 2021). This new pulsar has only one pulse detected in each of the three sessions, 2022 May 29, 2022 December 27, and 2023 January 25. Its period was found via the periodicity searching on 2023 January 25, though with a very weak S/N (see Figure A4 in the Appendix), but not confirmed by the folded profiles with the same period on 2021 December 26, 2023 January 2, and 2023 January 25. PSR J1938+1748g has a period of 7.106 s, while the folded pulse-width is 8.7 ms which means an extremely narrow pulse with a duty cycle of only about 0.082% or 029 in the rotation phase longitude. PSR J1956+2911g is also a source with a narrow pulse width of only 3.7 ms, or 036 in the rotation phase longitude. PSR J1921+1227g has a low mean flux density of about 0.28 μJy, which is probably one of the weakest pulsars. The pulse-stacks of PSR J1843-0051g and J1856+0029g show mode-changing with bright wider pulses detected only in some episodes and weaker narrow pulses in other periods. They are similar to the case of the known pulsar PSR J1938+2213 with an occasionally bright mode, see Section 5. PSR J1927+1849g was first picked up as 2 single pulses from the single pulse search in the 5-minute observation, and the two tracking observations show a wide profile with these sparse narrow bright pulses.

3.5. Polarization of New Sources

Fortunately, polarization data are recorded for all tracking observations for verification of the detected single pulses. Following Han et al. (2021) and Wang et al. (2022), we calibrate the polarization data, and obtain the polarization profiles for some strong single pulses of the transient sources (see Figure 7) and the mean polarization profiles of bright pulses of the newly discovered proto-RRATs, extremely nulling pulsars and weak pulsars (see Figures 8–10). We derive the polarization properties and also the Faraday rotation measures (RMs) of the bright pulses for the transient sources, as listed in Table 2, and of the mean for bright pulses of RRATs, nulling pulsars and weak pulsars as listed in Table 3. The new measurements of RMs can be used for revealing the interstellar magnetic fields (Han 2017; Han et al. 2018; J. Xu et al. 2022).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 2. Polarization Properties of Individual Pulses for the Newly Discovered Transient Sources

Name	ObsDate:No.	L/I	V/I	∣V∣/I	RM
		(%)	(%)	(%)	(rad m−2)
J1828−0003g	20221114:B1	40.5	−4.1	18.4	5.7(36)
J1850−0004g	20200415:B1	48.0	−13.2	16.4	−286(10)
	20200902:B1	49.1	−30.7	29.7	−317(9)
	20220608:B1	73.4	12.3	13.1	−298(1)
	20220608:B3	44.9	−15.9	15.4	−302(9)
	20221106:B1	38.2	−3.9	0.0	−286(7)
J1853+0209g	20220824:B1	57.4	−1.6	2.7	−27(10)
J1853+0353g	20210624:B1	63.4	2.3	14.7	381(13)
J1855−0211g	20221205:B1	31.5	1.8	12.0	715.9(46)
	20221205:B2	61.1	12.1	10.0	711.8(44)
	20221205:B3	31.3	−2.2	13.5	712.1(37)
	20230305:B1	50.3	4.9	27.9	725.3(14)
J1855+0033g	20210328:B2	53.5	17.5	15.3	381(9)
J1859+0832g	20221107:B1	80.1	−8.8	8.5	1004(2)
J1918+0342g	20211202:B1	57.4	−8.4	20.2	208(5)
J1921+1629g	20211004:B1	52.7	18.5	18.1	362(4)
	20211004:B2	52.5	17.7	16.6	368(3)
J1924+1734g	20211005:B3	51.4	7.5	10.9	108(6)
J1934+2341g	20210624:B5	58.5	0.4	1.4	133(3)
J2005+3154g	20211009:B2	54.5	−70.4	70.2	−132(9)

Download table as:  ASCIITypeset image

Table 3. Polarization Properties of Newly Discovered RRATs

Name	ObsDate	W50	L/I	V/I	∣V∣/I	RM
		(ms)	(%)	(%)	(%)	(rad m−2)
RRATs with a good period identified
J1857+0229g	20201108	5.7	37.4	8.2	6.6	34(12)
J1904+0621g	20210318	9.6	22.8	4.4	7.2	−238(25)
J1905+0156g	20210113		42.7	10.4	19.8	−84(5)
J1905+0558g	20210627	6.6	25.0	−19.8	18.0	431(20)
J1908+0911g	20221107	18.9	57.0	−0.7	5.0	73(2)
J1917+0834g	20201123	37.2	51.0	0.0	6.8	192(3)
	20210113	22.9	53.3	−6.8	8.5	180(4)
J1921+1006g	20221107	19.6	38.5	36.0	34.5	489(8)
J1948+2314g	20221107	14.4	35.6	6.0	7.5	70(4)
J2005+3156g	20211009	12.6	21.5	1.5	3.3	−283(6)
Extremely nulling pulsars
J1828−0038g	20220828	8.3	31.6	6.4	6.6	33(2)
	20221114	12.4	24.9	0.8	6.5	36.7(9)
J1842+0114g	20210317	60.7	33.0	4.9	10.6	119(5)
J1845−0008g	20210626	14.9	18.9	5.0	2.8	24(9)
J1855+0240g	20210626	16.7	33.1	−6.6	9.3	99(6)
J1858−0113g	20220403	13.5	18.9	−25.6	25.5	673(3)
J1911+1017g	20201123	70.5	38.0	−19.1	20.0	261(14)
J1921+0851g	20210624	14.0	16.7	6.3	8.2	381.3(8)
J1935+1901g	20220329	14.0	31.1	5.4	2.1	82(8)
Newly discovered weak pulsars with occasionally strong pulses
J1900+0732g	20210626	6.7	35.3	1.9	5.3	372(17)
J1906+0335g	20210627	6.3	37.5	26.4	26.4	203(4)
	20210707	6.3	27.4	19.5	19.8	205(4)

Download table as:  ASCIITypeset image

Some features of these polarization profiles of transient sources are remarkable. Pulse No.1 of J1850-0004g observed on 2022 June 8 is highly linearly polarized, with the polarization angles (PAs) clearly sweep up. So does the pulse on August 24, 2022, of J1853+0209g and pulse No.1 on November 7, 2022, of J1859+0832g. The PAs in part of the profiles sweep down are seen in the pulse No.1 on November 14, 2022, of J1828−0003g, pulse No.3 on October 5, 2021, of J1924+1734g and pulse No.5 on August 24, 2021, of J1934+2341g. Such PA sweeping is a typical feature of pulsar signals produced in pulsar magnetosphere. Therefore we trust that the pulses of these transient sources are produced by neutron stars, and they are probably just bright pulses of undetectable weaker pulsars, see more discussion of FAST observations of previously known RRATs below.

The polarized profile of the pulse No.2 on October 9, 2022, of J2005+3154g has a highly circular polarization of V/I = −0.702.

As seen in Tables 2 and 3, for pulses detected from one source or observed at different days, their RMs are consistent with each other within uncertainties. The largest RM value is detected for the pulse No.1 on November 7, 2022, of J1859+0832g, which is 1004 ± 2 rad m⁻² for the pulse with a DM of 259 ± 2 pc cm⁻³. The result is understandable that this object is located near the Galactic plane in the spiral arm tangent where the interstellar magnetic fields could be very strong (Han 2017; Han et al. 2018; Xu et al. 2022).

There is no question that the polarization profiles of the mean bright pulses of RRATs, nulling pulsars and weak pulsars are more or less similar to polarization of normal pulsars (Wang et al. 2023).

Sensitive observations of previously known RRATs can help to understand the enigma of RRATs.

In the visible sky area of FAST, there are more than 60 previously known RRATs. Since these RRATs were discovered via detection of sparse strong pulses by other telescopes (McLaughlin et al. 2006; Hessels et al. 2008; Deneva et al. 2009, 2016; Burke-Spolaor & Bailes 2010; Keane et al. 2010, 2011, 2018; Burke-Spolaor et al. 2011; Coenen et al. 2014; Stovall et al. 2014; Karako-Argaman et al. 2015; Patel et al. 2018; Tyul'bashev et al. 2018a, 2018b; Good et al. 2021; Dong et al. 2022), very sensitive observations by FAST may reveal possible weaker pulses from these rotating neutron stars. Some known RRATs have basic parameters (period, DM and position) with a poor accuracy, and our new observations with the FAST snapshot observations can improve these parameters (see Table 4), as done for other pulsars in the first paper (see Han et al. 2021).

With the observations of the GPPS survey and also the project for the known pulsar positioning (PT2021_0051, PI: J. Xu) and the projects for known RRATs studies (PT2021_0151, PI: D.J. Zhou), we have got 59 RRATs observed by chance or by targeting observations, and detected 48 RRATS (see the list in Table 4). 11 RRATs have been observed but not detected (see Table 5), among which eight RRATs could not be detected in targeted observations and the other 3 could not be detected in the snapshot observations for the GPPS survey. It is possible that these RRATs were not active when they were observed by FAST, e.g., for J1911+00 only 0.31 pulses hr⁻¹ in the original discovery paper. Some RRATs were discovered at a very low-frequency band, and do not have enough flux at the L-band for the FAST to detect, e.g., J0534+3407. Otherwise, non-detection could be caused by the very large uncertain position.

For 48 known RRATs we detect, their parameters are presented in Table 4, and period, DM or position accuracy has been improved for 28 RRATs. The period of an RRAT is obtained either from the TOAs of single pulses or even via a normal periodic search for the survey data (see Table 4). The spin periods of J0156+0358, J1857+0719, and J1905+0414 are obtained for the first time here by our FAST observations. The periods of J1838+0414 and J1924+1006 were 3.640 s and 5.281 s, respectively, in the discovery paper for the PALFA, ²² however, we detect the periods of 1.330681 s and 4.619757 s by analyzing the TOA of the detected single pulses in our FAST data with the period analysis program PF. The DM values of these RRATs were estimated by using some bright single pulses, and we get DMs of 10 RRATs improved. J1945+2357 (J1946+24) was reported to have a DM of 96 pc cm⁻³ (Deneva et al. 2009), but our FAST observations give a much better determined DM as being 87.53 ± 0.18 pc cm⁻³. For some RRATs, their position can be constrained by the FAST observations to a half beam size of FAST (see discussion in Sections 4 and 5 of Han et al. 2021), and we get positions of 19 RRATs, significantly improved from these in references. Their new names according to the new coordinates are given in the first column in Table 4.

We emphasize that most of the previously known RRATs are very bright for the FAST, as indicated by the high detection percentage of individual pulses over the 3σ criteria as listed in column (10). Nevertheless, for 1 known RRAT, we detect only 4 pulses, and for other 4 known RRATs, we detect only 1 pulse each. PSR J0534+3425 was first discovered at 111 MHz (Tyul'bashev et al. 2018a), and we detect 4 pulses in a 5-minute tracking observation of the GPPS survey, but no pulse was detected in the following 15 monitoring observations. For these 4 pulses, only signals below 1180 MHz are detected in our observations. Considering the low DM of about 24.7 pc cm⁻³, the detection in a partial band is likely to be caused by interstellar scintillation. No period can be found from our FAST observation yet. PSR J0550+0948 was reported by Deneva et al. (2016) with P = 1.745 s and DM = 86.6 pc cm⁻³. Only one weak pulse is detected by FAST with S/N = 17.2 during the snapshot observation on December 15, 2021, around a candidate position in the beam of J0550+0900-M02P3. PSRs J0550+0948 (Deneva et al. 2016), J0625+1254 (Patel et al. 2018), J0640+0744 (Tyul'bashev et al. 2018a) and J1859+0759 (Patel et al. 2018) have no previously known period. In our FAST observation, only one pulse is detected in a 5-minute observation for each source with S/N = 17.2, S/N = 32.8, S/N = 25.0 and S/N = 15.0, respectively. For these sources, we cannot get a period from our data even after folding for J0550+0948. The dynamic spectra for these eight pulses are shown in Figure 11.

Zoom In Zoom Out Reset image size

For the other 43 known RRATs detected by FAST, the folded data according to the period and DM show that 25 of them are just normal pulsars (see Figure 12), 5 of them are very nulling pulsars (see Figure 13), and others are pulsars with sparse strong pulses (see Figure 14). In the following, we discuss such these cases in detail.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

4.1. Known RRATs as Normal Pulsars

4.2. Known RRATs as Extremely Nulling Pulsars and Pulsars with Sparse Strong Pulses

Five previously known RRATs appear as extremely nulling pulsars in the FAST observation, as shown in Figure 13 and Figure A6 in Appendix. For example, the PSR J0103+5354 and J1839−0141 have only one, J1717+0305 and J1928+1725 have two brief continuous radiation episodes, and J1720+0040 have several interspersed emission fractions in our observations. Outside the active episodes, no detectable pulse is found in FAST data.

We have got the other 13 known RRATs observed by the FAST, and they generally appear as many sparse strong pulses as shown in Figure 14 and also Figure A7 in Appendix, much more pulses than in their original discovery papers. They "must" be RRATs in the view of other telescopes. In our FAST observations, we can get them easily detected not only by the single-pulse search module, but most of them can be picked out through normal pulsar searches.

Some objects are worth noting. PSR J1720+0040 is an interesting RRAT. We detected 7 pulses during a 305 s observation (see Figure A6), and we fortunately can get its period as being 3.356875 s. We can determine the position according to the position of the snapshot beam with significant pulse signal detection. The occasional brightened state, for example, has been recently detected for PSR J1938+2213 (Chandler 2003; Sun et al. 2022), which is shown in Figure 15. The single pulses in the bright state are suddenly 100 times brighter. If it is observed by a small telescope, probably only the brightened part would be observable, it could be shown as an extremely nulling pulsar, like PSR J1839−0141 (Lu et al. 2019) and PSR B0823+26 (Sobey et al. 2015) which have a "bright" mode that regularly emits bright pulses and a "quiet" mode that occasionally emits weak pulses. PSR J1946+1449 (J1946+14) (Deneva et al. 2016) is also shown RRAT-like pattern that emission is intermittent, and we detect few bright pulses. If they are observed by a not-so-sensitive telescope, they must be classified as RRATs as being pulsars with sparse strong pulses.

J0623+1536 is a known RRAT J0623+15 discovered in the PALFA survey (Patel et al. 2018) with the previously given DM of 92.5 ± 1.6 cm⁻³ pc. In the 15-minute observation on 2023 February 13, ten single pulses have been detected. We analyze and get a DM of 92.77 ± 0.24 cm⁻³ pc and a period of about 2.638545(18) s. Interestingly, the single pulse at the period No. 330 shows the fine structures with frequency drifting in the dynamic spectrum, while other pulses are very normal without any drifting, see Figure 16. The dynamic spectrum of this anomalous pulse shows well-dedispersed structures with the DM of 92.77 cm⁻³ pc. Such a phenomenon has been identified for some bursts of FRBs (Hessels et al. 2019; Zhou et al. 2022) but rarely seen in pulsars. The previously known case is the emission drifting structure of PSR J0953+0755 at low frequency (Bilous et al. 2022).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

4.3. Polarization of Known RRATs

When these known RRATs were observed by FAST, the polarization signals are recorded. We use the standard polarization calibration procedure (Wang et al. 2023) and process all data. We obtain the polarization profiles of these known RRATs as shown in Figures 17–19. To maximize the S/N, we have to ignore these periods with undetectable signal, i.e., only work on the periods with a signal of S/N > 3. We extract polarization information from the Faraday RMs as listed in Table 4.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As one can see from these polarization profiles, in general, they are not unusual compared to normal pulsar profiles. For these RRATs, which appear as normal pulsars in the FAST observations, the PA curves of J1849+0106, J1853+0427 and J1919+1745 show a smooth variation with an "S" shape. Some pulsars show a steep polarization angle sweep at the profile center with a reversal of circular polarization, such as J0302+2252 and J1856+0912. The J1850+1532 has a highly circular polarized component. Orthogonal polarized modes have been detected in the PA curve of J1538+2345.

For five RRATs as extremely nulling pulsars, it is very interesting that the extremely highly linear polarization of J1928+1725 with a flat PA curve (see Figure 18). For pulsars with sparse strong pulses, because of the limited periods accumulated, it is hard to say the profiles in Figure 19 are stabilized. Interestingly the RM value for J1905+0414, which has a large RM value of 1090 ± 3 rad m⁻². Combined with the DM of this RRAT 383.0 pc cm⁻³, the averaged magnetic fields in the line of sight to this distant RRAT is 1.232 × 1090/383 = 3.5 μG. This seems to be fine (Han 2017) for a pulsar at a spiral arm tangent at (l, b) = (38246, −1079). J1720+0040 and J1848+1516 also have reversal of circular polarization.

4.4. Sparse Strong Pulses and Polarization

For previously known RRATs, it is important to verify if these sparse strong pulses are very outstanding from the energy distribution of normal pulses, and if they have very different polarization from normal pulses.

For known RRATs with a sufficient number (>300) of single pulses observed by FAST so that a well-defined average pulse profile has been obtained, we obtain the energy distribution of their single pulses. For each pulse, we sum the individual pulse energy in the on-pulse range, and then obtain the average energy of all single pulses. The energy of all individual pulses is then normalized by the averaged energy. As shown in Figure 20, the pulse energy distributions of the known RRATs rarely show any points with a scaled energy greater than 10 if they are believed as normal pulses, except for J0302+2252 and J1915+0639 which seems to be very unusual with two more peaks in the distribution in addition to the main component for nulling pulses around zero. For RRATs as extremely nulling pulsars, the distribution has a component for nulling together with a few periods with a high-energy tail for emission. For RRATs with sparse strong pulses, J0628+0909, J1848+1515 and J1905+0414 and also J1913+1330, we do see these sparse pulses with unusual energy.

Zoom In Zoom Out Reset image size

Figure 21 compares the polarization profiles for the normal weak pulses and these sparsely strong pulses, in the two sides of the normalized energy distributions, see Figure 20. In principle their number distributions should be fitted with a log-normal energy distribution. However, such fitting cannot be done well since there is only a small number of bright pulses available. The sparse strong pulses can therefore be distinguished from the weak pulses in the distribution roughly by the lowest valley. We find that the polarization profiles from these small number of strong pulses in the high-energy tail have the same or the orthogonal polarized modes for their polarization angle distributions for weaker pulses, except for the case of PSR J1915+0639, which is hard to understand. This indicates that mostly there is no difference between polarization profiles for sparse strong pulses and mean polarization profiles of normal pulses.

Zoom In Zoom Out Reset image size

We developed an efficient single pulse searching module for generating DM-time images and target detection with AI, so that it is computationally fast. Two processing strategies, the full bandwidth and the two halves of observation band of FAST are processed independently for the single pulse search, which can improve the detection of narrow or steep-spectrum pulses. We apply this new single-pulse search module to the FAST GPPS survey, and discovered 76 new RRATs in our Galaxy. Among them, 26 sources have only several pulses detected by FAST, so that their rotation period cannot be derived from the limited duration of FAST tracking observations. We have got 16 sources identified as proto-RRATs, which have sparse strong pulses with a recognized period. 10 sources of them are apparently extreme nulling pulsars, and the other 24 sources are weak pulsars with sparse strong pulses.

For many newly discovered transient sources, the occasional strong emission state can help them to be found if the pulses are strong enough over the detection threshold. This has some implications. One is there must be a large neutron star population in our Galaxy to uncover but is mostly undetectable. Second, single-pulse searching is a necessary step to catch them. Third, the occasionally brightened emission state of these neutron stars must be related to some physical conditions, which must be understood.

Notice that these 76 transient objects here, including 12 objects discovered by the single pulse search that reported in the Paper I (Han et al. 2021), are about 15% of newly discovered pulsars. In the other words, at least 15% of neutron stars would not be detected from the traditional periodicity signal search. Therefore, the single pulse search is important in revealing a large population of unknown RRATs, i.e., these pulsars with sporadic strong pulses.

The RRAT detection percentage of the FAST GPPS survey is much larger than the percentage of the known RRATs compared to all known pulsars, which is only 5% for all known RRATs, including extremely nulling pulsars and weak pulsars with sparse strong pulses. The detection of RRATs by FAST data leads us to find pulsars at the sub-μJy level. For some of them, we cannot get enough pulses to figure out the period yet. Finding their period by multiple observations or by increasing the duration of FAST observation is very necessary.

Figure 22 shows the distribution for the period and mean flux density of newly discovered transient sources in this paper compared to values for the known RRATs and the other pulsars with a period of larger than 0.1 s in the first FAST GPPS paper (Han et al. 2021). The mean flux densities of these newly discovered sources are much lower than that of the known RRATs and even the GPPS pulsars. Most of them have a mean flux density of few μJy, and J1921+1227g even has the lowest mean flux density of about 0.28 μJy. This again demonstrates the population of weak radiate neutron stars inside the Milky Way to be discovered. For the probability density functions (PDFs) and cumulative distribution functions (CDFs), there is no significant difference in period distribution between these transient sources and the previously known RRATs. The difference between the FAST GPPS pulsars and these RRATs is prominent, which apparently caused by different selection effects from the search method. Any pulsars discovered by the normal periodical search are not classified as RRATs. On the other hand, the single pulse detection also has limitations. The identification of extremely narrow transient (<1 ms, like the ultranarrow burst of FRB 20191107B, Gupta et al. 2022) or even narrower band radiation in FRB, has to be improved in the future. The detection efficiency of signals at the boundary of the DM-time images has to be improved.

Zoom In Zoom Out Reset image size

Extremely sensitive observations are fundamental to understand the transient sources, especially RRATs. Observations of the known RRATs by using FAST could detect more weak pulses, in addition to previously detected sparse strong pulses. In our FAST GPPS survey area, a large number of known RRATs have the FAST data recorded already. Together with other FAST projects, we have got 48 known RRATs detected by FAST. Four of them have only one pulse detected in our FAST observations and another one has four pulses detected in a 5-minute observation session but not confirmed in the verification observation. Twenty-five known RRATs are simply normal pulsars in the FAST observations, five of them are extremely nulling pulsars, and 13 objects are weaker pulsars with sparse strong pulses. Polarization observations show no unusual difference of these strong pulses from weaker pulses.

We therefore conclude that most RRATs are simply weaker pulsars with sparse strong pulses or very nulling pulsars. The sensitivity of the telescope and long observations are the key to understanding the enigma of RRATs.

This project, as one of five key projects, is being carried out by using FAST, a Chinese national mega-science facility built and operated by the National Astronomical Observatories, Chinese Academy of Sciences. J. L. Han is supported by the National Natural Science Foundation of China (NSFC, Nos. 11988101 and 11833009) and the Key Research Program of the Chinese Academy of Sciences (grant No. QYZDJ-SSW-SLH021); D. J. Zhou is supported by the Cultivation Project for the FAST scientific Payoff and Research Achievement of CAMS-CAS. H. G. Wang, C. Wang, P. F. Wang and X. P. You are supported by NSFC No. 12133004; P. F. Wang, C. Wang and H. G. Wang are also partially supported by the National SKA program of China No. 2020SKA0120200. In addition, C. Wang is also partially supported by NSFC No. U1731120; P. F. Wang is also partially supported by the NSFC No. 11873058; H. G. Wang is also partially supported by the Guangzhou Science and Technology Project No. 202102010466; Jun Xu is partially supported by NSFC No. U2031115.

The FAST GPPS survey is a key science project of FAST led by J. L. Han. D. J. Zhou realized the single pulse search module and related software under the supervision of J. L. Han, and processed all data presented in this paper. J. L. Han coordinated the teamwork and coordinated computational resources. and was also in charge of writing this paper. Jun Xu observed some RRATs in his project and contributed the data of some known RRATs. Chen Wang designed the survey observation plan for the GPPS survey, and fed all targets for each observation session and verification observations. P. F. Wang realized the polarization processing pipeline which is used in this paper. Tao Wang initialized and realized parts of the data preparing module. P. F. Wang and Jun Xu made fundamental contributions to the construction and maintenance on the computation platform. W. C. Jing, Xue Chen, Yi Yan and W. Q. Su joined many group discussions and commented on the results of this paper. H. Q. Gan, P. Jiang and J. H. Sun ensured the success of FAST observations. Other people jointly proposed or monitored the project. All authors contributed to the finalization of the paper.

Original FAST observational data will be open sources according to the FAST data 1 yr protection policy. The folded and calibrated data for sources in this paper can be obtained from authors with kind requests. Pulsar profile data presented and the source codes in this paper are available on the webpage of the GPPS survey http://zmtt.bao.ac.cn/GPPS/.

The single pulse search module for the GPPS survey uncovers many radio transient sources in the sky.

Table A1 list parameters for each pulse, including the TOA of the pulse peak, S/N of dedispersed pulse R, pulse width in millisecond, and the fluence discussed in Section 2.2.2.

Figure A1 shows all water-fall plots for 26 radio transients with only a few pulses detected for each source, including the 3 example plots in Figure 3.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure A2 show plots for 16 RRATs discovered in the GPPS survey, while Figure 4 is only one example.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure A3 show plots for 10 extremely nulling pulsars as discovered in the GPPS survey, while Figure 5 is only one example.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure A4 show plots for 24 weak pulsars with sparse strong pulses discovered in the GPPS survey, while Figure 6 is only one example.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure A5 show plots for 25 previously known RRATs shown as normal pulsars in the FAST observations, while Figure 12 has just two examples.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure A6 show plots for five previously known RRATs shown as extremely nulling pulsars in the FAST observations, while Figure 13 has just two examples.

Zoom In Zoom Out Reset image size

Figure A7 show plots for 13 previously known RRATs shown as pulsars with sparse strong pulses in the FAST observations, while Figure 14 has just examples respectively.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size