1 Introduction

Gamma-ray bursts (GRBs) are the short and intense pulses of \(\gamma \)-rays, occurring randomly at a rate of \(\sim \)1 event per day. The bi-modality in \(T_{90}\) distribution of GRBs is used to divide these energetic events into two classes: short and long bursts with a boundary at 2 s (Kouveliotou et al. 1993). However, the \(T_{90}\) value relies on the energy range in which the instrument operates and its trigger method. In general, bursts have a lower value of \(T_{90}\) in higher energy channels (Fenimore et al. 1995; Qin & Chen 2013). The \(T_{90}\) value also depends on the sensitivity of the detector and background variations. Further, the observed \(T_{90}\) depends on the redshift; the rest frame duration (\(T_{90}/(1+z)\)) will be lower than the observed. It is challenging to decide the class of the GRB without any redshift information. It is also observed that some GRBs with \(T_{90}\) values favorable to long GRBs have afterglow and host properties similar to the short GRBs (Gal-Yam et al. 2006). On the other hand, other GRBs with \(T_{90}<2\) s show properties identical to long GRBs (Antonelli et al. 2009; Ahumada et al. 2021).

Therefore, it is difficult to classify the GRBs based on \(T_{90}\) alone, particularly for GRBs lying close to the boundary. It is also essential to look for other observational characteristics apart from their \(T_{90}\) information that can distinguish the two classes.

Hardness ratio (HR), the fluence ratio in harder to softer energy bands, can be used to classify GRBs. Short GRBs are found to be harder with large values of HR compared to their long companions (Fishman & Meegan 1995; Tavani 1998). Furthermore, HR is found to be correlated with \(T_{90}\) for the complete sample of GRBs. However, no correlation is noticed between the two for an individual class (Qin et al. 2000).

Another characteristic is the spectral lag (i.e., the delay in the arrival times of low-energy photons to high-energy photons) which can differentiate the two classes. Long GRBs have significant lags (up to a few seconds for some of them) in their light curves in different energy channels. On the other hand, no lag (nearly zero) is observed for short GRBs (Cheng et al. 1995; Yi et al. 2006; Kaneko et al. 2015).

The two classes can also be compared concerning their energetics (\(E_{\gamma ,{\mathrm{iso}}}\)) and luminosities (\(L_{p, {\mathrm{iso}}}\)). Short GRBs have, on average, energies that are smaller than that of long bursts (Ghirlanda et al. 2009). They are located at two different places in \(E_{\gamma , {\mathrm{iso}}} -E_{p}\) plane (Amati plane), following a correlation that is parallel to each other (Amati et al. 2002; Amati 2006). Additionally, the host galaxy properties such as stellar population, star formation rate (SFR), morphology, offset, etc., are different for short and long bursts in general and can provide a clue about the burst progenitor system and hence the class of the GRB (Li et al. 2016).

The sensitivity of detectors affects the duration timescales of the bursts. As a result, GRBs detected by Swift and Fermi have different duration values reported. Figure 1 represents the \(T_{90}\) values as measured by Swift (15–350 keV) and Fermi (50–300 keV). The bursts have different values of \(T_{90}\) reported by Swift and Fermi. Most of them lie in the same class, either short or long burst (first and third quadrant). Sixteen GRBs were detected by Swift and Fermi between 2008 and February 2021, including GRB 210217A (second and fourth quadrant, the shaded grey regions), having different classification provided by two satellites.

Figure 1
figure 1

\(T_{90}\) duration of GRBs detected by both Fermi (50–300 keV) and Swift (15–350 keV). GRBs lying in the first and third quadrant have the same classification from both satellites. However, GRBs lying in the second and fourth quadrant (shaded grey region) have different classifications from both satellites. The magenta triangle represents GRB 210217A.

Full size image

GRB 210217A is one of the recent bursts lying at the boundary of short and long GRBs divide with different burst duration values reported by Swift and Fermi. GRB 210217A was detected by Swift-BAT (Simpson et al. 2021) and Fermi-GBM (Fletcher & Fermi-GBM Team 2021). Swift Burst alert telescope (BAT) reported a \(T_{90}\) of \(4.22\pm 1.15\) s (15–350 keV), suggesting it to be a long GRB (Sakamoto et al. 2021). The sub-threshold detection by Fermi-GBM with a duration of 1.024 s (25–294 keV) suggests that the burst might belong to the short population of GRBs.Footnote 1 This burst was well within the observational capabilities of moderate-sized Indian telescopes. Therefore, we observed the optical afterglow of this burst with the 1.3m Devasthal Fast Optical Telescope (DFOT) and 3.6m Devasthal Optical Telescope (DOT) located in Aryabhatta Research Institute of Observational Sciences (ARIES), Nainital. The study of prompt and afterglow emission of a GRB provides a complete picture of the nature of the GRB.

We performed a detailed analysis to classify GRB 210217A. This paper presents the multi-wavelength analysis of GRB 210217A and discusses various methods for classification. Further, we estimated the burst’s redshift using joint fitting of the spectral energy distribution (SED) obtained from UVOT/XRT afterglow. In Section 2, we describe the discovery and follow-up observations of GRB 210217A. The data acquisition and reduction procedures are explained in Section 3. Section 4 presents the results obtained and a discussion on the possible classification of GRB 210217A. Finally, we summarise the conclusions of this work in Section 5. We used the Hubble parameter \(\mathrm{H}_{0} = 70\) km s\(^{-1}\) Mpc\(^{-1}\), density parameters \(\Omega _{\Lambda }= 0.73\) and \(\Omega _{\rm m}= 0.27\) (Jarosik et al. 2011).

2 GRB 210217A

BAT on-board NASA’s Swift space mission (Barthelmy et al. 2005) triggered GRB 210217A on 17 February 2021 at 23:25:42 UT and provided the location of the source with coordinates RA, and Dec \(=\) 06h 30m 26s, \(+\)68d 42\('\) 53\(''\) (J2000), respectively, with an uncertainty of 3 arc-min (Simpson et al. 2021). The Swift X-ray telescope (XRT) observations 97.8 s after the BAT trigger located an uncataloged X-ray source at RA, and Dec \(=\) 06h 30m 20.82s, \(+\)68d 43\('\) 29.9\(''\) (J2000) with an uncertainty of 3.7 arc-sec. Swift ultra-violet and optical telescope (UVOT) observations at 103 s after the BAT trigger detected a transient in white and U filters.

Several ground-based optical/NIR telescopes started observing the field of GRB 210217A and reported the optical afterglow magnitudes.Footnote 2Fermi-GBM did not trigger the event automatically (Fletcher & Fermi-GBM Team 2021). Still, an automated, blind search for short GRBs (below the onboard triggering threshold) in Fermi-GBM identified a short GRB consistent with the Swift-BAT event in both time and location with a high significance value with a duration of 1.024 s (Fletcher & Fermi-GBM Team 2021).

We also carried out the observations of the optical afterglow of GRB 210217A with 1.3m DFOT located in ARIES at the earliest available opportunity (Kumar et al. 2021). Once the afterglow crosses the limit of 1.3m DFOT, we carried out deep observations with the 3.6m DOT (Dimple et al. 2021). The characteristics of the burst are presented in Table 1.

Table 1 Characteristics of GRB 210217A. \(T_{90}\): duration from Swift-BAT observations; Transient duration: duration of the transient reported by Fermi-GBM; HR: ratio of the fluence in 25–50 keV to that in 15–25 keV; \(t_\mathrm{mvts}\): minimum variability timescale in 15–350 keV; \(E_{p}\): peak energy as reported by Fermi-GBM team (Fletcher & Fermi-GBM Team 2021); z: photometric redshift obtained by fitting UVOT/X-ray SED; \(E_{\gamma ,{\mathrm{iso}}}\): isotropic \(\gamma \)-ray energy calculated using the photometric redshift.
Full size table

3 Data acquisition and reduction

This section narrates the data acquisition and analysis from different space missions and ground-based telescopes as part of the present work. The Fermi-GBM data is not publicly available for GRB 210217A as this burst was detected below the onboard triggering threshold.Footnote 3 For the rest of the work we used the parameters provided by the Fermi-GBM team.Footnote 4\(^,\)Footnote 5

3.1 Swift-BAT

To extract the temporal and spectral properties of GRB 210217A in high energy bands, the raw data of Swift-BAT (Observation Id: 01033264000) is obtained from the online portal of Swift Archive.Footnote 6 We reduced the data utilising the HEASOFT (version-6.25). The inbuilt pipelines batbinevt, bathotpix and batmaskwtevt are used to create detector plane image (DPI) followed by removal of hot pixels and mask weighting. Later, the mask-weighted BAT light curves are extracted using batbinevt pipeline for different energy channels. The light curves in different energy channels are shown in Figure 2. Further, we estimated the \(T_{90}\) duration of light curves in different energy channels and are tabulated in Table 2. The \(T_{90}\) value in the energy range 15–350 keV is \(3.76\pm 0.26\) which is consistent with the value reported in Sakamoto et al. (2021).

To examine the spectral properties of the burst, we extracted the time-averaged BAT spectrum for the total duration of the burst starting from \(T_0-0.065\) to \(T_0+0.489\) s (the start and end time of the pulse are identified using Bayesian binning of the light curve in the energy range 15–350 keV).

Figure 2
figure 2

Background subtracted Swift-BAT count rate light curve of GRB 210217A in different energy channels with a time resolution of 64 ms.

Full size image
Table 2 \(T_{90}\) value of GRB 210217A in different energy channels.
Full size table

We used the batbinevt and batdrmgen pipelines to produce the spectrum and detector response matrix (DRM), respectively. The resultant spectrum is fitted with the power-law and the cutoff power-law using the Multi-Mission Maximum Likelihood framework (3 ML) package (Vianello et al. 2015). The maximum likelihood estimation technique was used for choosing the best fit model. We got the maximum value for likelihood for a single power-law model with photon index (\(\Gamma _\mathrm{BAT}\)) of \(1.99 \pm 0.09\). The fluence is \(6.7 \pm 0.39 \times 10^{-7}\) erg cm\(^{-2}\) in the 15–150 keV band, which is consistent with the value reported by (Sakamoto et al. 2021).

3.2 Swift-XRT and UVOT

We obtained the X-ray flux light curve data in the 0.3–10 keV energy band from Swift-XRT Burst Analyser repositoryFootnote 7 hosted by the University of Leicester (Evans et al. 2009). The flux light curve is then converted to flux density at 5 keV using the relations given by Gehrels et al. (2008) for further analysis

$$\begin{aligned} F_\nu ,x=4.13\times 10^{11} \frac{F_x(2-\Gamma _x)E_0^{1-\Gamma _x}}{E_2^{2-\Gamma _x}-E_1^{2-\Gamma _x}}, \end{aligned}$$
(1)

where \(E_1\) and \(E_2\) are lower and upper bounds of band pass in keV, \(E_0\) is the energy in keV at which flux density is calculated, \(\Gamma _x\) is the X-ray photon index. \(F_x\) is the measured flux in erg cm\(^{-2}\) s\(^{-1}\).

We downloaded the Swift-UVOT data from the online Swift data archive page.Footnote 8 We analyzed the UVOT data using standard pipeline uvotproduct of heasoft software version 6.25 with the latest calibration database. A source region of 5 arc-sec and a background region of 25 arc-sec aperture radius are extracted for the photometric analysis of the burst. We detected a source in the white and U filter. Table 3 shows the magnitudes of the source in these filters.

Table 3 AB magnitudes of the afterglow of GRB 210217A along with the magnitudes reported in the GCN circulars. Magnitudes are not corrected for Galactic extinction. All the upper limits are given with a 3-sigma value.
Full size table

3.3 1.3m DFOT and 3.6m DOT

We started observations with 1.3 DFOT on 18-02-2021 at 17:31:22 UT located at Devasthal observatory of ARIES, India, for the follow-up observations of the optical afterglow of GRB 210217A. We observed 30 images with an exposure of 120 s each in the Bessel R filter and a set of 20 images (120 s exposure) in the Bessel I filter. We detected the optical afterglow of GRB 210217A in the stacked frames within the Swift XRT error circle (left panel of Figure 3). Later, for deep observations, we observed the field with Aries Devasthal Faint Object Spectrograph and Camera (ADFOSC) mounted on the 3.6m DOT around \({\sim }1.7\) days after the burst. We took four consecutive images in r-band with an exposure of 900 s each. We corrected the science images acquired from these telescopes for bias, flat and cosmic rays using Astropy module of Python.

Figure 3
figure 3

The optical finding chart of GRB 210217A was taken with 1.3m DFOT (left) and 3.6m DOT (right) telescopes. The afterglow is visible, and its position is marked with yellow circles.

Full size image

We stacked the images to enhance the signal-to-noise ratio. The source is visible in stacked image (Dimple et al. 2021). The finding chart for the same is shown in the right panel of Figure 3. We used the DAOPHOT package to perform point spread function (PSF) photometry which was calibrated against the Panstarrs catalog resulting in apparent magnitudes listed in Table 3. The magnitudes are converted to flux density after correcting for galactic extinction. Figure 4 shows the multi-band light curve of GRB 210217A constructed using our data and those reported in GCN circulars.

Further, we fitted the X-ray and optical R band light curves of GRB 210217A using a single power law of the form \(F=F_{0}t^{-\alpha }\), where t corresponds to time, and \(\alpha \) is the decay index. Due to the limited number of data points in other optical bands, it is not possible to fit these. Therefore, we overplotted the light curves in other bands using the R band decay index. The light curves with power laws are shown in Figure 4.

Figure 4
figure 4

Multiwavelength afterglow light curves of GRB 210217A. Optical data points are corrected for galactic extinction. The X-ray data (shown in grey) are taken from the Swift-XRT page. The best fit models are also shown with dashed lines.

Full size image

4 Results and discussions

This section presents the results obtained from the multi-band analysis and possible classification scenario of GRB 210217A.

4.1 Photometric redshift

To estimate the photometric redshift of GRB 210217A, we analyzed the joint XRT and UVOT SED in the time interval of \(\sim \)100–200 s. We didn’t observe any spectral evolution during this time interval. Using the XRT data and UVOT magnitudes, we created the SED following the methodology described in Chand et al. (2020) and Gupta et al. (2021). We fitted the SED using power-law and broken power-law models using XSPEC (Arnaud 1996). The Galactic and intrinsic absorber components (phabs and zphabs) are also included from XSPEC models. The Galactic absorption is fixed to NH\(_\mathrm{Gal}=8.69 \times 10^{20}\) cm\(^{-2}\) (Willingale et al. 2013). We further included two dust components using the XSPEC model zdust, one at redshift \(z= 0\) for the galactic dust component and the other for the intrinsic dust component with varying redshift, which provided the redshift information. The Galactic reddening is fixed at 0.0847 (\(E(B-V)\)) conforming to the map of Schlafly & Finkbeiner (2011). The SED is fitted with Milky Way, Large and Small Magellanic Clouds (MW, LMC and SMC) extinction laws (Pei 1992) at the redshift of the burst. Although all these models well explain the SED, a minimum \(\chi ^{2}\) value is obtained for the Milky Way extinction law with a power-law model. The value of the spectral index is \(1.85 \pm 0.13\), and that of photometric redshift is \(0.55^{+0.9}_{-0.4}\) for GRB 210217A. The SED is shown in Figure 5.

Figure 5
figure 5

The SED was obtained using joint Swift UVOT and XRT afterglow data for GRB 210217A. Data points are shown using blue circles. The SED is best fitted with a power-law model (shown with red dashed-dotted line) with Milky Way extinction law.

Full size image

4.2 Hardness ratio and peak energy

We estimated the HR using fluence ratio in two different energy channels: 15–25 keV and 50–100 keV energy bands of Swift-BAT and used it for the classification of GRB 210217A. The HR during \(T_{90}\) is found to be \(0.64 \pm 0.005\). We collected the sample of GRBs from Swift-BAT catalogFootnote 9 and estimated the hardness ratio using the fluence in the same energy window as used for GRB 210217A. We fitted this sample with the Bayesian Gaussian mixture model (BGMM), which is a machine learning algorithm supported by scikit-learn (Pedregosa et al. 2011). Using this algorithm, we estimated the probability of GRB 210217A being a short GRB as 98.2%. The probabilities of the whole sample being a short GRB along with GRB 210217A are shown in Figure 6(a).

Figure 6
figure 6

(a) The spectral hardness and \(T_{90}\) for GRB 210217A (shown with a red square) along with the GRB sample (for short and long GRBs) presented in Goldstein et al. (2017). (b) GRB 210217A in \(E_{p} {\text{-}} T_{90}\) plane with the dataset available from Fermi-GBM catalog. (c) Minimum variability timescale (\(t_{\mathrm{mvts}}\)) as a function of \(T_{90}\) for GRB 210217A along with the short and long GRBs sample studied by Golkhou et al. (2015). The color scale on the right side of the Figure shows the probability of the GRB being a short GRB and the vertical dashed line at 2 s shows the boundary separating the short and long classes of GRB. (d) GRB 210217A in the Amati plane along with the data points for long and short GRBs taken from Minaev & Pozanenko (2020). The solid lines shows the best correlation between \(E_{p,i}-E_{\gamma ,{\mathrm{iso}}}\) along with a 2-sigma correlation presented by the dashed line of respective colors for short (grey) and long class (steel blue). \(E_{p,i}-E_{\gamma ,{\mathrm{iso}}}\) for GRB 210217A is shown with a red square. The red dashed line shows the \(E_{p,i}-E_{\gamma ,{\mathrm{iso}}}\) for the redshift range considering the uncertainty in photometric redshift.

Full size image

As Swift BAT has narrow spectral coverage, we could not determine \(E_{p}\) using spectral fitting. Therefore, we used the \(E_{p}\), peak energy and \(T_{90}\) values provided by Fermi-GBM along with the \(E_{p}-T_{90}\) values from the Fermi-GBM catalog to identify the class of GRB 210217A. We again fitted the distribution with BGMM and found a probability of 96% for GRB 210217A being a short GRB. The probability map is shown in Figure 6(b).

4.3 Minimum variability timescale

High energy light curves of GRBs are highly variable and can be explained by the GRB central engine resulting from internal shocks. Hence, minimum variability timescale (MacLachlan et al. 2013, \(t_{\mathrm{mvts}}\)) gives an idea about the central engine, the source emission radius (\(\mathrm R_\mathrm{c}\)), and the minimum Lorentz factor (Sonbas et al. 2015, \(\Gamma _{\mathrm{min}}\)) of GRBs. In general, the \(t_\mathrm{mvts}\) value for long bursts is larger with an average value of 200 ms than their short counterparts having a mean value of 10 ms, indicating that short GRBs have a more compact central engine (MacLachlan et al. 2012, 2013). We measured the \(t_\mathrm{mvts}\) for GRB 210217A using the method described in Golkhou et al. (2015). The estimated value of \(t_\mathrm{mvts}\) is \(\sim \)0.512 s for GRB 210217A, which we further used for its classification. We collected \(t_\mathrm{mvts}\) values for short and long GRB samples from Golkhou et al. (2015). Using BGMM, we found the probability of GRB 210217A being a short GRB equal to 28% (Figure 6c).

4.4 Spectral lag

We calculated the spectral lag using bat light curves in energy channels; 15–25 keV and 50–100 keV. We estimated the using the cross-correlation function (CCF) and uncertainties in the CCF following the method described in Bernardini et al. (2015). We fitted the CCF with an asymmetric Gaussian function using emcee (Foreman-Mackey et al. 2013) to find its global maximum, which represents the lag in two light curves. We found a positive lag 86\(^{+68}_{-65}\) ms for the burst. The average value of lag for short and long GRBs are \(16.5\pm 7.5\) and \(375.1\pm 69.6\) ms, respectively (Bernardini et al. 2015). The value lies close to the mean value of short GRBs within the error bar. With the intermediate value of spectral lag lying between the mean value of two classes, it is hard to classify the burst.

4.5 Amati correlation

We also used the well-known correlation between the isotropic equivalent energy emitted in the high energy regime, \( E_{\gamma ,{\mathrm{iso}}}\) and \(E_{p}\) (the energy at which \( \nu F_{\nu } \) is maximum) to classify GRB 210217A (Amati et al. 2002; Amati 2006; Minaev & Pozanenko 2020). A power-law well fits this correlation with an index of \( a \simeq 0.4 \); however, the correlation regions are well separated for different classes (Minaev & Pozanenko 2020).

For the redshift range (0.15–1.45) obtained from SED (Section 4.1), we estimated the \( E_{\gamma ,{\mathrm{iso}}}\) using the following Equation (2)

$$\begin{aligned} E_{\gamma ,{\mathrm{iso}}}=K_{\mathrm{bol}}\times \frac{4 \pi d_{L}^2}{1+z}f_\gamma , \end{aligned}$$
(2)

where \(K_{\mathrm{bol}}\) is the bolometric correction factor, \(d_L\) is the luminosity distance and \(f_\gamma \) is the fluence (erg cm\(^{-2}\)). For Swift-BAT energy range (15–150 keV) we used \(K_{\mathrm{bol}} = 5\) (Fong et al. 2015) and the fluence value calculated in Section 3.1. We also estimated the \(E_{p,i}\) (peak energy in the source frame) for the estimated redshift range. As Swift-BAT covers only a small energy range, we used the \(E_{p}\) and burst duration as reported by Fermi-GBM team (Fletcher & Fermi-GBM Team 2021). The red dashed line in the Figure 6(d) shows the \( E_{\gamma ,{\mathrm{iso}}} - E_{p,i}\) for the redshift range of GRB 210217A. GRB 210217A lies in the region in between long and short GRBs.

5 Summary

We report a detailed analysis of GRB 210217A using publicly available multi-wavelength observations, including our observations from our ARIES telescopes. We try to find the true class of GRB 210217A using various methods described in the literature. We calculated the photometric redshift using Swift-XRT/UVOT data. Using this estimated redshift, we calculated the isotropic equivalent energy and peak energy in the source frame. GRB 210217A lies at the boundary between the short and long classes in the Amati plane.

We also calculated the HR, minimum variability timescale, spectral lag and fitted the \(T_{90}\)-HR, \(T_{90}\)-Ep, \(T_{90}\)-mvts distribution using BGMM. We found a probability of GRB 210217A being a short GRB equal to 98.2, 96 and 28% in these cases, respectively. It is hard to conclude if GRB 210217A belongs to the long or the short class as some of the properties belong to long GRBs and others to short.

The host studies can clarify the true class of this burst. The fact that two categories have different progenitors, they originated in different kinds of host galaxies. Long GRBs are located in the star-forming young population of galaxies; the short GRBs belong to the old population of galaxies (Li et al. 2016). Long GRBs are generally located in bright star-forming regions with minimal offsets from the center of the host galaxy on a galactic scale. In contrast, due to significant merger scale times, short GRBs have large offsets from the centre of their galaxies (Bloom et al. 2002; Fong et al. 2013). The host observations are not available for GRB 210217A. However, the host observations can give a clue about the class of GRB 210217A.