Search for high-energy neutrinos from Einstein Probe detected Fast X-ray Transients using 14 years of IceCube muon track data

Introduction

In 2013, IceCube detected a diffuse neutrino flux observed in the TeV-PeV energy range [1], whose origin is still unknown. The first source for which a statistically significant spatial association was established was the blazar TXS 0506+056 [2]. The second such source with a significance of $4.2\sigma$ was NGC 1068, a Seyfert galaxy [3]. However, these sources can only account for up to 1% of the diffuse astrophysical neutrino flux [3, 4]. Therefore, in order to determine the origin of the diffuse neutrino flux observed in IceCube, searches from a wide range of extragalactic sources have been done to look for a statistically significant excess, most of which have reported null results [5, 6] (and references therein).

In this work, we search for neutrino emission from fast X-ray transients (FXTs). FXTs are short duration bursts of soft X-ray radiation that last from $10-10^{4}$ seconds and are located at cosmological distances [7, 8]. Their spectral characteristics are in between classic Gamma-ray bursts (GRBs) and X-ray flashes [9]. Until recently, they were mostly identified through analysis of archival X-ray data and it was difficult to identify their host galaxies [10]. The recent launch of Einstein Probe (EP) has substantially advanced the field of FXTs. The Wide-field X-ray Telescope (WXT) aboard EP observes the sky in the soft X-ray band covering 0.5-4.0 keV [11]. [12] The Einstein Probe has reported a total of 200 FXTs since its launch two years ago. Based on an analysis of data from the first year of EP, it was shown that only about 20% of FXTs have gamma-ray counterparts [12, 13]. A number of FXTs were found to possess afterglow properties similar to that of GRBs [14]. Type 1c supernovae have been detected through optical afterglow observations of many FXTs [15, 16]. Type 1c (and other types of core collapse supernova) have long been associated with long GRBs [17]. Most recently O’Connor et al. [18] has assembled a catalog of 26 spectroscopically confirmed FXTs from Einstein Probe, and showed that their distribution is consistent with that of long GRBs. Therefore, it was argued that a large fraction of EP detected FXTs have a collapsar origin similar to long GRBs [18]. Subsequently, it was confirmed that the formation rate and luminosity function is similar to long GRBs [14].

A plethora of models have been invoked to explain the origin of FXTs. Possible explanations include stellar flares, supernova shock breakouts, millisecond magnetars, among others [16]. The progenitor models include compact object mergers, tidal disruption events, proto-magnetars, etc [19]. The weak or absent gamma-ray emission observed in most FXTs suggests that the majority of these events are unlikely to originate from standard GRB jets viewed along their axes [16]. However, as predicted in early work, off-axis and choked jets have been shown to reproduce the observed timescales and luminosities of FXTs at the time [20] Most recently, hydrodynamical simulations of a relativistic jet propagating from the progenitor to the photon diffusion radius of the cocoon have been able to explain many of the observed properties of FXTs [16].

Previous searches for neutrinos in coincidence with GRBs have been done by the IceCube collaboration [21, 22, 23, 24]. The main emission mechanism for production of neutrinos from GRBs is the Waxman-Bahcall mechanism where shock-accelerated protons in relativistic GRB jets interact with prompt gamma-ray photons via photo-meson interactions, generating charged pions that decay into high-energy neutrinos [25]. Neutrino production mechanisms have also been proposed from low luminosity as well as untriggered GRBs [26, 27, 28, 29, 30, 31, 32] or GRBs with choked jets [33, 34]. The earliest predictions based on the choked jet model suggested an expected detection of 0.2 neutrino-induced muons from a source at a redshift of $z \sim1.0$ ($D_L \sim10^{28}$ cm), assuming an isotropic equivalent energy of $10^{53}$ ergs [33]. The diffuse all-flavor neutrino flux from the choked jet model has recently been estimated to be $10^{-7}\,\mathrm{GeV}\,\mathrm{cm}^{-2}\,\mathrm{s}^{-1}\,\mathrm{sr}^{-1}$ at 30 TeV [34].

The similarity between FXTs and low-luminosity GRBs, together with the success of models invoking off-axis and choked jets to explain FXTs, motivates a search for high-energy neutrinos from these transients. This search is the primary goal of our work. However, because the Einstein Probe detected FXTs do not have any temporal overlap with the latest publicly available IceCube muon track data, we only carry out time-integrated searches from FXTs using the entire catalog. A temporal search will be deferred to future work with updated IceCube data which overlap the FXT triggers. We also note that there are models for neutrino emission from GRBs before the gamma-ray breakout [34, 35] and also in the afterglow phase [36]. So an untriggered search for FXTs which could pose constraints on some of these models.

This manuscript is structured as follows. The dataset used for the analysis is discussed in Sect. II. The analysis methodology and results can be found in Sect. III and conclusions in Sect. IV.

Dataset

For this analysis we consider muon track data from IceCube’s second data release from April 6, 2008 to May 23, 2022 [37]. This data set has been referred to as IceTracks-DR2 and covers four additional years of data compared to IceTracks-DR1 [38]. The IceTracks-DR2 dataset contains approximately 1.64 million neutrino events. from four detector configurations: IC40, IC59, IC79, and IC86. All IC86 seasons have undergone the same event processing and use the same detector response. Further, data from IC79 to IC86-2018 have been reprocessed using updated detector calibration. The effective areas provided in this data release also incorporate improvements in detector calibration and filtering. A full discussion of the improvements in this catalog compared to IceTracks-DR1 can be found in [37].

The list of FXTs has mainly been obtained from the compilation in [18]. This catalog was obtained by looking at the EP/WFXT transients in the energy range from 0.5-4.0 keV until 29th August 2025, and after culling out the flaring stars [18]. A total of 113 such transients were obtained from the publicly available data. Among these, 26 FXTs had spectroscopic redshifts resulting in a redshift completeness of $\sim23\%$. The spectroscopic redshifts were obtained from public GCN circulars or through dedicated spectroscopic observations with VLT for some FXTs such as EP250302a, EP250821a, and EP250827a[18]. Twelve of these FXTs have a GRB counterparts, whereas no GRB counterparts were detected for the remaining sources. This catalog has been used to estimate the luminosity function and formation rate of FXTs [12, 14]. We also added the recently detected FXT, viz. EP260119a located at a redshift of $z = 5.47$ [9]. A skymap of the 27 FXTs used for this analysis can be found in Figure 1.

Analysis and Results

For this work, we use the unbinned maximum likelihood method [39] which is used in IceCube point source analysis [4]. This analysis was performed using the SkyLLH [40, 41, 42] software package. We describe the procedure below.

For a dataset containing $N$ events, of which $n_s$ are assumed to originate from the source, the likelihood function can be defined as follows:

where $\gamma$ is the spectral index, $B_i$ and $S_i$ denote the background and signal PDF respectively for $i$-th neutrino event. The signal pdf depends on both a spatial and an energy dependent term, of which only the energy dependent term depends on $\gamma$. The background PDF can be extracted from the data. More details about the models used for $B_i$ and $S_i$ can be found in [41] and references therein. We then construct the test statistics defined as follows:

where $\hat{n}_s$ and $\hat{\gamma}$ are the values of $n_s$ and $\gamma$, which maximize the likelihood in Eq. (2). For a large number of signal events, TS asymptotically follows a $\chi^2$ distribution for two degrees of freedom [43]. However, because $n_s$ and $\gamma$ are correlated the distribution of the null hypothesis could lead to deviations from $\chi^2$ distribution for two degrees of freedom [44]. For this purpose we use background trials using the observed data to calculate the $p$-value similar to recent works which have used the same software for point-source searches [45]. The distribution from background trials is obtained using the same SkyLLH software. For a statistically significant detection of $5\sigma$, the local $p$-value for each source has to be smaller than $3.0 \times10^{-7}$ [46, 47]. Accounting for the look-elsewhere effect associated with the 27 FXTs, the local $p$-value should be even smaller if the global $p$-value should exceed $5\sigma$.

For every source the unbinned likelihood ratio is then maximized over the number of signal events $n_s \ge0$ and the spectral index $\gamma$ of a power-law spectrum giving a per-source test statistic TS. The best-fit values can be found in Table I. Fifteen of the 27 positions return a non-zero TS; the other twelve converge to TS = 0, $\hat{n}_s = 0$. The largest values are EP240801a (TS = 4.77, $\hat{n}_s \simeq29$, $\gamma\simeq3.3$), EP250704a (TS=3.74), EP241217b (TS=2.90), EP250821a (TS=2.76) and EP240315a (TS=2.59); all other sources have TS < 2.3. The source EP240801a, which has the largest TS value (TS=4.77) has a local $p$-value of 0.04. Hence, the largest TS value is not significant enough to claim a statistically significant detection.

We therefore derive flux upper limits for each FXT. For this purpose, we carry out trials to obtain the upper limit on neutrino flux by injecting varying number of signal neutrinos [45, 48]. We then calculate the $p$-value for obtaining a value larger than the observed $\hat{n}_s$ for the real data. The 90% confidence level (c.l.) upper limit on the number of signal events $n_s^{90}$ is then given by the value of injected signal for which $p$-value is equal to 0.1. Our limits on $n_s^{90}$ (cf. Table I) are roughly an order of magnitude larger than the predicted muon neutrino rate from choked jet model [33] for $E_{\mathrm{iso}} \sim10^{53}$ ergs and $z = 1.0$. However, since $E_{\mathrm{iso}}$ for most FXTs is less than $10^{53}$ ergs, we cannot rule out this model. However, we would expect more robust constraints from a triggered search.

To convert this limit into a differential neutrino flux limit, we assume a power-law form for the neutrino spectrum:

where $\Phi_0$ is the normalization at $E_0 = 1000$ TeV and $\Gamma$ is the spectral index, for which we assume $\Gamma= 2.0$ for all sources, independent of the best-fit $\gamma$ found for each source. The differential muon neutrino flux limit ($E^2\Phi(E)$) is then calculated by dividing $n_s^{90}$ by exposure and effective area, as outlined in one of our previous works [5]. These limits can also be found in Table I and have values in the range of $O(10^{-10})$ GeV cm$^{-2}$ s$^{-1}$. The most stringent limit is obtained for EP241021a, which is given by $E^2\Phi(E) = 2.1 \times10^{-10}$ GeV cm$^{-2}$ s$^{-1}$ at a pivot energy of 100 TeV.

Conclusions

In this work, we search for steady-state high energy neutrino emission from 27 spectroscopically confirmed FXTs detected between 2024 and 2026 by the Einstein Probe X-ray satellite using the recently released 14 year muon track data from IceCube from 2008-2022. For this purpose, we use the standard unbinned maximum likelihood analysis which has been used for IceCube point source search using the SkyLLH software package. Our results are summarized in Table I. We do not find any statistically significant excess for any FXT. The maximum TS value is observed for EP240801a, whose local $p$-value is 0.04 and hence is not significant even without considering the look-elsewhere effect. We then calculate 90% c.l. differential muon neutrino flux limit at 100 TeV for each of the 27 FXTs. The most stringent upper limit is observed for EP241021a, given by $E^{2}\,dF/dE = 2.1 \times10^{-10}\ \mathrm{GeV}\ \mathrm{cm}^{-2}\ \mathrm{s}^{-1}$, for a $E^{-2}$ spectrum. Our limits are about an order of magnitude compared to some of the early predictions from the choked GRB jet model [33]. In a future work, we will supplement this analysis with a temporal analysis using narrow time windows around the time of the detected FXTs to further increase the sensitivity, once the updated IceCube point source catalog is released.

Acknowledgements

The computational work used for this analysis was supported by the National Supercomputing Mission (NSM), Government of India, through access to the “PARAM SEVA” facility at IIT Hyderabad. The NSM is implemented by the Centre for Development of Advanced Computing (C-DAC) with funding from the Ministry of Electronics and Information Technology (MeitY) and the Department of Science and Technology (DST). We have used OpenAI chatgpt to correct for language and stylistic issues in the manuscript.