Content uploaded by David M Palmer
Author content
All content in this area was uploaded by David M Palmer on May 15, 2017
Content may be subject to copyright.
arXiv:1501.04435v1 [astro-ph.HE] 19 Jan 2015
Mon. Not. R. Astron. Soc. 000, 1–18 (2015) Printed 25 February 2017 (MN L
A
T
E
X style file v2.2)
Swift follow-up of IceCube triggers, and implications for
the Advanced-LIGO era
P.A. Evans1⋆, J.P. Osborne1, J.A. Kennea2, M. Smith3, D.M. Palmer4,
N. Gehrels5, J.M. Gelbord6,7, A. Homeier8, M. Voge8, N.L. Strotjohann8,9,
D.F. Cowen10, S. B¨oser11, M. Kowalski8,9, A. Stasik9
1Department of Physics and Astronomy, University of Leicester, Leicester, LE1 7RH, UK
2Department of Astronomy and Astrophysics, Pennsylvania State University, 525 Davey Lab, University Park, PA 16802, USA
3Jet Propulsion Laboratory, 4800 Oak Grove Drive, Pasadena, CA 91109, USA
4Los Alamos National Laboratory, B244, Los Alamos, NM 87545, USA
5NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA
6Spectral Sciences Inc., 4 Fourth Ave., Burlington, MA 01803, USA
7Eureka Scientific Inc., 2452 Delmer St. Suite 100, Oakland, CA 94602, USA
8Physikalisches Institut, Universit¨at Bonn, Nußallee 12, D-53115 Bonn, Germany
9DESY, D-15735 Zeuthen, Germany
10Dept. of Astronomy and Astrophysics, Pennsylvania State University, University Park, PA 16802, USA
11Institute of Physics, University of Mainz, Staudinger Weg 7, D-55099 Mainz, Germany
Accepted 2015 January 17. Received 2015 January 12; in original form 2014 October 24
ABSTRACT
Between 2011 March and 2014 August Swift responded to 20 triggers from the IceCube
neutrino observatory, observing the IceCube 50% confidence error circle in X-rays,
typically within 5 hours of the trigger. No confirmed counterpart has been detected.
We describe the Swift follow up strategy and data analysis and present the results of
the campaign. We discuss the challenges of distinguishing the X-ray counterpart to
a neutrino trigger from serendipitous uncatalogued X-ray sources in the error circle,
and consider the implications of our results for future strategies for multi-messenger
astronomy, with particular reference to the follow up of gravitational wave triggers
from the advanced-era detectors.
Key words: methods: observational – x-rays: general – gamma-ray burst: general –
neutrinos – gravitational waves
1 INTRODUCTION
For biological reasons, astronomy has been a science car-
ried out using electromagnetic (EM) radiation, and indeed
until comparatively recently was limited to that portion of
the EM spectrum to which our eyes are sensitive, and the
atmosphere transparent. This has changed over the last cen-
tury and at the present time observatories exist collecting
data from the longest wavelengths (e.g. LOFAR1) to the
shortest (e.g. HESS, Hinton et al. 2004). Today, a growing
area of astronomical research does not use EM radation at
all, but probes other messengers, such as neutrinos, gravi-
tational waves or cosmic rays. The detection and identifica-
tion of astrophysical sources of these messengers is difficult,
⋆pae9@leicester.ac.uk
1http://www.lofar.org
and an ideal scenario is to combine detections of non-EM
messengers with EM signals, to provide a multi-messenger
dataset. To date, the only object outside of our solar sys-
tem to be detected in this way is the supernova SN1987a
(Kunkel et al. 1987), which, in addition to its EM discovery
and observations, was also detected as a neutrino emitter
(Alekseev et al. 1987; Bionta et al. 1987; Hirata et al. 1987,
1988). While extra-solar cosmic rays (e.g. Abraham et al.
2010) and PeV neutrinos (Aartsen et al. 2013) have been
detected, the origin of these is still unclear and they have
not been reliably coupled with a known EM object.
Neutrinos are expected from various sources, such as
supernovae (SNe, e.g. Kachelrieß et al. 2005; Abbasi et al.
2014) and gamma-ray bursts (GRBs, e.g. Asano & M´esz´aros
2014), but targeted searches – retrospectively searching the
neutrino data corresponding to the time and location of EM
detections of these phenomena – have so far yielded null
c
2015 RAS
2Evans et al.
results from both the IceCube (Abbasi et al. 2011, 2012)
and ANTARES (Adri´an-Mart´ınez et al. 2013) observatories.
Similarly, gravitational waves are expected to arise from a
range of phenomena, particularly the merger of two neutron
stars in a short GRB. Targeted searches for gravitational
waves from short GRBs have also, so far, failed to produce
any detections (e.g. Abadie et al. 2010b).
Effort has also been expended to search for EM coun-
terparts to non-EM triggers. Because EM facilities tend to
have narrow fields of view, the likelihood of a non-EM trig-
ger being contemporaneously observed by an EM telescope
are very low, therefore the EM data have to be collected af-
ter the non-EM trigger. The error regions from neutrino or
gravitational wave facilities are on the scale of degrees, thus
it often requires multiple pointings to collect the necessary
EM data. It is also not clear when is the optimal time to
search for the counterpart, as the relative timescales of EM
and non-EM radiation depends on the physical source of the
emission. For example, for supernovae the neutrino signal
precedes the EM signal by many days. An optimal follow-
up facility would, therefore, have a large (ideally all-sky)
field of view, and high level of sensitivity. Due to the high
rate of transient events in the universe, multiwavelength ca-
pabilities are also desirable, for example to help distinguish
rapidly between GRBs and flare stars.
The Swift satellite (Gehrels et al. 2004) arguably pro-
vides the best existing facility for the EM follow up of non-
EM triggers, at least in X-rays. Although the X-ray telescope
(XRT; Burrows et al. 2005) has only a modest field of view
(radius ∼0.2◦), the Swift spacecraft is capable of rapid slew-
ing, and has the ability to ‘tile’ regions on the sky, so as to
cover a large error region in a single spacecraft orbit. The
XRT is sensitive to 5×10−13 erg cm −2s−1in 1 ks (0.3–10
keV), and can localise sources to a 90% confidence radius
of 3.5 arcsec (improving to 1.4 arcsec for brighter sources;
Goad et al. 2007; Evans et al. 2009).
Evans et al. (2012) reported on Swift follow up of
two gravitational wave triggers from the LIGO-Virgo
(Abbott et al. 2009; Accadia et al. 2012) facilities. No X-
ray counterpart to the gravitational triggers could be found,
and indeed it transpired that neither of the gravitational
wave triggers was in fact real (one was a subthreshold noise
event, the other an artificial signal introduced to the data
as a blind test of the detection algorithms). In this work,
we report on the search with Swift-XRT for X-ray coun-
terparts to 20 neutrino-doublet triggers from the IceCube
facility (Achterberg et al. 2006), and discuss the challenges
related to idenfitying the EM counterpart. A neutrino dou-
blet (or multiplet) was defined as two or more neutrinos
detected within 100 s of each other, and with an angular
separation of at most 3.5◦; more details about this is given
in a companion paper (Aartsen et al., in preparation).
The Swift follow-up observations began as soon as pos-
sible after the neutrino trigger, implicitly assuming that
the X-ray emission from the astrophysical neutrino source
is temporally coincident with (or only a few hours after)
the neutrino emission. We consider two ways of identifying
the X-ray counterpart: by its brightness compared to refer-
ence catalogues, or by its temporal variability (in particular,
whether it shows signs of fading, as may be expected follow-
ing some form of outburst).
We did not set the threshold at which Swift will re-
Figure 1. An example exposure map of a 7-tile Swift -XRT ob-
servation of an IceCube trigger. This observation was taken with
the on-board tiling, so the exposure in each field has been built up
over multiple spacecraft orbits; the pointing is slightly different
on each orbit, hence the blurring round the edges of the fields.
The black lines and dots are the bad columns and pixels on the
CCD.
spond to a neutrino trigger based on theoretical predictions
of neutrino flux (which are highly uncertain due to the lack
of observational constraint), instead we set it such that Ice-
Cube would be expected to produce roughly six spurious
(i.e. non-astrophysical) triggers per year, which represents a
compromise between sensitivity to astrophysical neutrinos,
and the value of Swift’s observing time. The companion pa-
per (Aartsen et al., in preparation) will discuss the expected
rate of doublet triggers from the background and from astro-
physical objects, and consider the lack of neutrino triplets
during this experiment.
This paper is organised as follows. In Section 2 we de-
scribe the follow-up observing strategy employed by Swift,
and in Section 3 we overview the data analysis techniques. In
Section 4 we consider the sources detected, and attempt to
identify if either of these is likely to be the counterpart to the
neutrino trigger, which we expect to be a source undergoing
some form of outburst. Finally, in Section 5 we consider the
implications of our findings for future EM follow-up of non-
EM triggers, in particular, the expected gravitational wave
triggers from the Advanced LIGO-VIRGO facility.
Throughout the paper we have assumed a cosmology
with H0= 71 km s−1Mpc−1, Ωm= 0.27,Ωvac = 0.73.
Unless otherwise stated, all quoted errors are at the 90%
confidence level, and upper limits at the 3-σ(=99.7%) con-
fidence level.
2SWIFT ’S OBSERVING STRATEGY
Following IceCube triggers, high-priority Target of Oppor-
tunity (ToO) requests were submitted to Swift. Due to the
c
2015 RAS, MNRAS 000, 1–18
Swift follow-up of IceCube triggers 3
Table 1. Details of the 20 IceCube triggers followed-up by Swift as of 2014 September 03
Trigger # Error radius Trigger time Delay 1Tiling type
(50% conf) (UT) (hours)
1 0.7◦2011-03-26 21:53:41 1.4 Manual
2 0.7◦2011-09-27 12:23:29 7.5 Manual
3 0.3◦2011-10-24 02:41:11 10.0 Automatic
4 0.7◦2011-12-06 01:40:15 1.7 Automatic
5 1.1◦2012-01-17 22:01:34 4.8 Automatic
6 0.5◦2012-02-08 00:14:29 1.4 Automatic
7 1.8◦2012-03-03 16:47:22 1.2 Automatic
8 0.7◦2012-08-29 22:49:59 2.5 Automatic
9 0.9◦2012-09-17 18:08:03 1.6 Automatic
10 0.8◦2012-10-23 04:46:15 5.7 Automatic
11 0.4◦2012-12-21 02:17:24 3.7 Automatic
12 1.4◦2013-01-15 11:08:46 1.5 Automatic
13 0.7◦2013-02-13 18:47:42 1.9 Automatic
14 1.2◦2013-03-08 22:15:49 17.0 Automatic
15 0.9◦2013-03-27 19:54:12 1.5 Automatic
16 1.3◦2013-05-17 15:57:03 0.7 Automatic
17 0.8◦2014-01-08 05:29:08 9.0 Automatic
18 0.7◦2014-01-17 04:02:10 1.8 Automatic
19 0.3◦2014-02-26 19:04:14 3.2 Automatic
20 0.7◦2014-08-29 13:54:49 1.1 Automatic
1The time between the possible neutrino event detected by IceCube and the start of the first observation with the Swift-XRT.
efficient and flexible operation of Swift, observations were
able to begin rapidly once the ToO was received: the median
time from IceCube trigger to the first Swift observation was
1.8 hours. The IceCube 50% error radius is typically >0.5◦,
whereas the Swift -XRT has a field of view of radius of 0.2◦,
therefore it was necessary to observe the error region in a se-
ries of seven overlapping ‘tiles’: an example exposure map is
shown in Fig. 1. Initially this tiling had to be performed by
manually commanding seven separate observations as Swift
Automatic Targets2; each tile was consequently observed on
a separate spacecraft orbit. Under this system, all of the
requested exposure in a given tile (typically 1–2 ks) was
gathered in a single spacecraft pointing3; however, for each
successive field the delay between the trigger and the ob-
servation increased by ∼96 min (Swift’s orbital period). On
2011 August 10 the software on-board Swift was modified
to support automatic tiling. In this system, which was used
from trigger #3 onwards (Table 1), a single Automatic Tar-
get is uploaded, and Swift automatically divides the visibil-
ity window of the IceCube trigger location in each spacecraft
orbit between the 7 tiles. This is repeated on each orbit until
the requested exposure time has been gathered for each tile4.
Under this system, all tiles are usually observed in the first
visibility window after the observation is uploaded, however
the individual exposures are short, so for a given tile to ac-
cumulate its full exposure time takes longer. This strategy
2That is, the observations were not in the pre-planned science
timeline, and overrode targets which were. The times of the ob-
servations were set automatically by the on-board software.
3XRT can observe a single target for a maximum of 2.7 ks per
96-min spacecraft orbit.
4The minimum time permitted for a continuous exposure is 60 s;
if the observing window is not sufficient for all tiles to be observed
for at least this duration on a given orbit, then the next orbit will
begin with the first unobserved tile.
is better than the manual-upload approach, since it allows
Swift to cover the entire search box promptly, increasing
the chances of detecting a bright, rapidly-fading counter-
part (e.g. a GRB afterglow), without significantly reducing
the likelihood of finding a slowly-fading counterpart. Details
of the 20 IceCube triggers are given in Table 1.
3 DATA ANALYSIS
The XRT data were automatically analysed at the United
Kingdom Swift Science Data Centre (UKSSDC) at the
University of Leicester, using heasoft v6.15.1, and the
Swift caldb b20090130 u20111031 x20131220 m20140221.
The data were first processed using xrtpipeline, then a
series of source detection and analysis routines were ap-
plied. In a previous work, similar to this but relating to
gravitational wave triggers (Evans et al. 2012), we noted the
need for a source detection system that was optimised for
fainter sources. Since then such a system has been devel-
oped (Evans et al. 2014), and we used it in this work. It
consists of the following steps: filtering the data, creating
images and exposure maps5, and locating and characterising
sources. The latter step is an iterative process which uses a
combination of sliding-cell detection, background modelling,
source PSF-fitting and likelihood tests to detect and localise
sources. It also provides a quality flag for each source, which
indicates the probability of the source being spurious. 0.3%
of sources flagged as Good are spurious; this rises to 1% when
the Good and Reasonable sources are considered (Reasonable
5This step differed in one detail from Evans et al. (2014): in that
work images could not exceed 1000 ×1000 pixels in size. We have
overcome this limitation so a single image covering all 7 tiles and
being 1900 ×1900 pixels (1.25◦×1.25◦) in size is used.
c
2015 RAS, MNRAS 000, 1–18
4Evans et al.
10−3 0.01
10−3
0.01
Count rate from source detection
Count rate from the light curve
Figure 2. The mean source count-rates determined from the
source detection plotted against the values derived from source
light curves, where both are available. The two are in good agree-
ment below 0.01 ct s−1, confirming that using the brightness es-
timates from source detection is a safe approach for the faintest
sources for which light curves could not be produced.
sources on their own have a 7% false positive rate), and 10%
when all sources (Good,Reasonable and Poor) are included
(Poor sources on their own have a 35% false positive rate,
so should be viewed with caution). Full details of this pro-
cedure are given in Evans et al. (2014), particularly sections
3.4 and 7 and fig. 3.
The astrometric accuracy of Swift-XRT positions, de-
termined using the on-board star trackers, is 3.5 arcsec
(90% confidence; Moretti et al. 2007). This can sometimes
be improved either by using the UV/optical telescope on
Swift as a super star-tracker (so-called ‘position enhance-
ment’; Goad et al. 2007; Evans et al. 2009), or by aligning
detected XRT sources with 2MASS (Skrutskie et al. 2006)
objects (Butler 2007; Evans et al. 2014). We tried both of
these methods for each position, although in most cases the
source was too faint for the former method to work (i.e.
the XRT source could not be detected in an exposure cor-
responding to a single UVOT image), and there were too
few XRT/2MASS matches for the latter method to offer a
more accurate astrometric solution than the star-trackers on
Swift.
We also built light curves and spectra of each source us-
ing the tools developed by Evans et al. (2007, 2009). These
provide fully validated products in which all necessary cor-
rections (e.g. for pile up – where multiple photons impact
a single pixel in a single CCD frame, and are consequently
miscounted as a single event – the presence of dead columns
on the XRT CCD, and exposure variation across the im-
age) have been applied. For most objects there were too few
counts to produce more than a single light-curve bin, thus
the ‘light curve’ is really simply a single brightness measure-
ment. For those objects with sufficient counts, we binned
the light curve such that there were at least 15 counts in
each bin. The spectra were automatically fitted in xspec
(Arnaud 1996) with an absorbed power-law model, using
two tbabs absorption components (with redshift fixed at
0). One of these was fixed at the Galactic value of NH
from Willingale et al. (2013), the other was free to vary.
The spectral fit was used to determine the energy conver-
sion factor (ECF; conversion from the XRT count rate into
an observed 0.3–10 keV flux) for each individual object, and
also, via pimms, a conversion from the measured XRT count-
rate to the equivalent in the Rosat PSPC (which had an
energy coverage of 0.1–2.4 keV, Pfeffermann et al. 1987).
The latter conversion allows us to compare the detected
sources with the Rosat all-sky survey (RASS; Voges et al.
1999): this is the most sensitive all-sky X-ray survey to date
and we use it as a reference catalogue (see Section 4.1),
along with the XMM Slew Survey (XSS; Saxton et al. 2008;
Warwick, Saxton & Read 2012) which covers ∼2/3 of the
sky in the 0.2–10 keV band.
For some sources, a light curve or spectrum could not
be produced (or the spectrum could not be fitted) as the
source was too faint, or too near to the edge of the field
of view. Where the light curve could not be produced, the
mean count-rate was taken from the output of the source de-
tection procedure (which includes corrections for PSF losses,
pile up etc.). These were designed as indicators, and are not
fully calibrated. Therefore, to verify their accuracy we com-
pared the mean count-rate from the light curve with that
from the source detection routine for all sources where both
were available. As Fig. 2 shows, the two are in reasonable
agreement.
For objects where the automatic spectrum or fit could
not be produced, we still require both an ECF and a con-
version between the XRT count rate and an equivalent
Rosat PSPC rate, to compare the source brightness with
the XSS and RASS respectively. In these cases we used a
standard AGN spectral model: an absorbed power-law with
NH=3×1020 cm−2, and Γ = 1.7. This yields a 0.3–10 keV
ECF of 4.03×10−11 erg cm−2XRT ct−1, and a conversion
of 1.497 PSPC counts per XRT count.
In Table 2 we list all of the X-ray objects detected in
our follow-up observations, giving their positions, the mean
0.3–10 keV flux, and various parameters which are useful
in identifying whether the source is related to the neutrino
trigger (described in Section 4.1). The sources for which a
spectrum could be constructed and fitted are shown in Ta-
ble 3 along with details of the spectral fit. A small number
of sources have unconstrained spectral fit parameters, or ex-
treme values which are artefacts of a poor quality spectrum,
rather than indicating extreme physical properties. However,
since the ECF and PSPC conversion for those sources are
compatible with the range seen for the other sources, we do
not revert to the standard values given above. One object,
source 2 in the field of trigger 1, has a very high ECF which
is driven by the fact that the spectrum is very hard, how-
ever we have not reverted to the standard AGN spectrum
for this, as it is clearly inappropriate. The lack of soft emis-
sion from this object means that we do not expect Rosat to
have seen it; a fact which would be unclear if we used the
standardised spectrum: that is only used for objects where
we had no XRT spectral fit at all.
c
2015 RAS, MNRAS 000, 1–18
Swift follow-up of IceCube triggers 5
Table 2. Sources detected in the XRT follow up of the IceCube triggers.
IceCube Src # Location ErraFlagbExposure C (B)cObs FluxdUpper LimiteNf
seren Pg
const Cat?h
Field (J2000) (′′) (ks) (×10−13 erg cm −2s−1) (×10−13 erg cm −2s−1)
(0.3–10 keV) (0.3–10 keV)
1 1 03h31m05.s1 +13◦56′00′′ 3.2∗G 3.5 16 (0.8) 1.81 (±0.48) 12 2.7
1 2 03h30m58.s7 +13◦57′06′′ 5.2 G 3.5 12 (0.8) 1940 (±640) §3.7
1 3¶03h31m01.s8 +14◦15′06′′ 6.1∗G 1.8 8 (0.5) 1.97 (±0.84) 6.9 3.1
1 4 03h29m47.s9 +14◦26′37′′ 4.4∗G 1.5 8 (0.5) 3.8 (±1.6) 33 2.1
3 1¶14h59m06.s8 +59◦31′09′′ 4.3∗G 1.3 17 (0.5) 5.1 (±1.3) 0.7 Y
3 2 14h59m05.s1 +60◦08′43′′ 3.7∗G 1.8 10 (0.6) 3.1 (±1.2) 18 2.3
3 3¶15h01m06.s8 +59◦59′43′′ 5.7 G 2.1 7 (0.8) 1.79 (±0.76) 7.6 2.6
4 1¶22h15m02.s5 +57◦02′43′′ 3.5∗G 0.4 16 (0.1) 9.7 (±2.7); 9.3 (±2.1) 15 0.1 0.78
4 2 22h15m03.s2 +57◦02′16′′ 4.0∗G 0.4 10 (0.1) 4.1 (±1.4); 4.03 (±0.85) 4.8 0.1 0.92
5 1¶08h31m36.s4 +18◦20′55′′ 4.3∗R 0.2 15 (3.5) 4.0 (±1.7) 12 0.3
5 2¶08h31m10.s8 +18◦33′01′′ 3.2∗G 0.3 14 (3.6) 6.3 (±1.7) 13 0.3
5 3¶08h31m09.s6 +18◦25′37′′ 3.2∗G 0.8 11 (4.4) 3.8 (+2.1, -1.7)♯10 0.3
5 4¶08h32m12.s8 +18◦37′02′′ 3.2∗G 0.7 15 (3.8) 6.5 (±2.4) 14 0.3
5 5¶08h31m38.s0 +18◦37′55′′ 3.2∗R 0.2 7 (2.7) 8.0 (+5.6, -4.3)♯17 0.2
5 6¶08h30m47.s0 +18◦57′09′′ 5.7 G 0.6 6 (0.2) 4.8 (+2.3, -1.8)♯18 0.4 Y
5 7 08h30m28.s9 +18◦46′11′′ 5.8 R 0.4 11 (3.0) 6.6 (+3.1, -2.5)♯5.4 0.2
5 8¶08h30m28.s8 +18◦46′10′′ 4.0∗R 0.3 9 (4.7) 7.5 (+5.7, -4.6)♯6.3 0.2
5 9¶08h30m04.s3 +18◦52′31′′ 4.7∗R 0.2 13 (4.3) 19.1 (+8.6, -7.1)♯8.9 0.09
6 1¶01h27m33.s8 +10◦24′09′′ 5.0∗G 1.4 12 (0.5) 4.2 (±1.4) 13 1.0
6 2 01h25m51.s4 +09◦59′24′′ 5.5 G 1.6 10 (0.5) 2.21 (±0.82) 44 1.7
6 3 01h24m25.s8 +10◦09′32′′ 5.5 G 1.6 9 (0.6) 2.9 (±1.3) 29 3.0
6 4 01h24m45.s2 +10◦07′08′′ 7.0 G 1.6 9 (0.4) 3.2 (±1.2) 32 2.1
6 5¶01h24m51.s5 +10◦10′33′′ 6.1 G 1.7 7 (0.4) 1.98 (±0.90) 13 3.1
6 6¶01h25m23.s1 +10◦03′23′′ 8.4 G 1.5 9 (0.5) 2.3 (±1.0) 2.5 Y
6 7 01h26m56.s8 +10◦12′59′′ 5.7 G 2.0 7 (0.6) 1.39 (±0.63) 38 3.5
6 8¶01h26m12.s4 +10◦26′27′′ 10.4 G 1.7 7 (0.6) 2.4 (±1.2) 11 2.3
6 9¶01h25m52.s7 +10◦22′14′′ 7.7 R 2.8 7 (0.9) 0.93 (±0.50) 9.2 1.4
6 10 01h24m51.s1 +10◦33′06′′ 4.3∗G 1.5 21 (0.5) 4.8 (±2.4); 4.2 (±1.0) 17 0.5 0.71
6 11 01h26m11.s4 +10◦23′37′′ 6.2 G 2.6 9 (0.9) 0.84 (±0.40) 9.0 1.3
7 1 04h08m41.s0 +03◦34′37′′ 1.5∗G 12.9 318 (12.6) 11.3 (±2.1); 5.43 (±0.36) 0.2 5.6×10−4Y
aradius, 90% confidence.
bG=Good, R=Reasonable, P=Poor; see text for details.
cNumber of counts found in the region used to detect the source, with the number expected from the background in parentheses.
dThis is the observed (i.e. absorbed) flux. For objects with more than one light curve bin, the peak flux and mean flux are given.
e3σupper limit, from the RASS unless otherwise indicated. If no value is given, this is because the source is a known X-ray emitter, so an upper limit is unnecessary.
fThe number of sources at least as bright as this one, expected in the XRT field of view. See Section 4.2 for details.
gThe probability that the source is constant, from a χ2test. Only available if the light curve contained at least 2 bins.
hWhether a catalogued source was found which matched this object; details are given in Table 4
∗Position enhanced by UVOT astrometry.
†Position derived from XRT-2MASS astrometry.
‡Upper limit obtained from the XMM Slew Survey.
¶No XRT spectral fit was available, generic AGN spectrum assumed.
♯No XRT light curve was available: the brightness was estimated by the source detection routine.
§This was a very hard source to which Rosat was insensitive, therefore a RASS limit is uninformative.
c
2015 RAS, MNRAS 000, 1–18
6Evans et al.
Table 2 –continued
IceCube Src # Location ErraFlagbExposure C (B)cObs FluxdUpper LimiteNf
seren Pg
const Cat?h
Field (J2000) (′′) (ks) (×10−13 erg cm −2s−1) (×10−13 erg cm −2s−1)
(0.3–10 keV) (0.3–10 keV)
7 2 04h09m25.s0 +03◦12′37′′ 6.2 G 2.6 10 (0.8) 1.66 (±0.67) 12 2.8
7 3¶04h09m42.s9 +03◦19′20′′ 4.8 G 2.1 6 (0.6) 2.04 (±0.98) 15 2.9 Y
7 4 04h09m27.s0 +03◦34′07′′ 1.5∗G 18.2 404 (15.7) 22.3 (±5.8); 7.91 (±0.49) 12 0.3 2.6×10−2
7 5 04h10m40.s7 +03◦19′19′′ 5.5 G 1.7 9 (0.6) 1.26 (±0.47) 4.2 2.1
7 6 04h09m53.s2 +03◦18′48′′ 7.7 G 2.0 7 (0.8) 1.02 (±0.49) 16 3.0
7 7 04h08m42.s9 +03◦39′01′′ 4.2 G 14.3 46 (4.7) 1.83 (±0.36); 1.57 (±0.26) 35 3.0 0.10
7 8 04h09m27.s8 +03◦30′16′′ 4.3 G 19.0 27 (6.1) 0.94 (±0.41); 0.92 (±0.25) 112 5.1 0.95
7 9 04h09m16.s7 +03◦40′19′′ 5.4 G 17.6 23 (5.7) 1.20 (±0.43); 1.05 (±0.31) 44 5.8 0.46
7 10 04h09m58.s5 +03◦27′42′′ 5.3 G 9.5 26 (3.9) 0.74 (±0.19); 0.70 (±0.17) 7.0 2.4 0.43
7 11 04h08m29.s6 +03◦33′03′′ 5.2 G 4.8 13 (2.0) 1.32 (±0.53) 25 2.2
7 12 04h09m03.s5 +03◦29′36′′ 5.9 G 17.3 16 (6.1) 0.29 (±0.13) 5.6 6.3
7 13 04h09m21.s0 +03◦38′04′′ 5.5 G 18.3 22 (5.8) 0.80 (±0.40) 104 1.2
7 14 04h09m08.s9 +03◦31′36′′ 5.3 G 19.1 17 (6.2) 0.53 (±0.23) 22 8.4
7 15¶04h09m36.s7 +03◦31′59′′ 5.2 G 17.8 14 (5.6) 0.44 (±0.22) 9.9 7.5
7 16¶04h09m54.s8 +03◦33′47′′ 7.0 P 16.6 16 (5.7) 0.30 (±0.16); 0.28 (±0.12) 6.2 13.6 0.69
7 17 04h08m43.s0 +03◦33′50′′ 5.9 R 14.8 23 (5.1) 1.59 (±0.42); 1.32 (±0.31) 175 3.5 0.12
7 18 04h09m12.s3 +03◦42′34′′ 5.0 G 17.2 19 (5.8) 0.33 (±0.18); 0.33 (±0.13) 14 6.8 0.98
7 19 04h09m40.s6 +03◦46′28′′ 7.0 P 7.3 11 (2.3) 0.42 (±0.21) 9.6 5.6
7 20 04h09m24.s9 +03◦46′25′′ 10.8 G 9.8 11 (3.2) 0.75 (±0.45); 0.61 (±0.24) 12 4.7 0.58
7 21¶04h09m59.s2 +03◦44′11′′ 9.7 R 6.2 12 (2.1) 0.71 (±0.34) 11 5.6
7 22 04h09m16.s7 +03◦29′38′′ 5.3 G 18.4 15 (6.1) 0.54 (+0.26, -0.21)♯39 17.5
7 23 04h10m10.s3 +03◦33′34′′ 6.8 R 8.0 10 (3.0) 0.65 (±0.39); 0.39 (±0.15) 6.5 2.8 0.12
7 24¶04h08m28.s1 +03◦37′55′′ 8.8 R 4.1 7 (1.2) 0.66 (+0.35, -0.27)♯13 4.0
7 25¶04h09m43.s6 +03◦28′35′′ 5.9 P 17.7 10 (6.2) 0.122 (+0.091, -0.074)♯7.9 16.6
7 26¶04h10m00.s1 +03◦36′23′′ 6.2 P 15.1 6 (1.4) 0.23 (+0.12, -0.10)♯7.8 13.9
7 27¶04h09m30.s9 +03◦29′53′′ 5.4 R 19.0 12 (6.1) 0.34 (±0.20) 8.7 9.9
7 28¶04h09m21.s7 +03◦27′40′′ 6.5 P 17.0 12 (5.7) 0.143 (+0.099, -0.080)♯12 22.0
7 29¶04h09m16.s1 +03◦39′06′′ 5.1 R 18.1 11 (5.7) 0.131 (+0.091, -0.074)♯7.1 20.1
7 30¶04h08m54.s9 +03◦27′31′′ 8.1 P 15.1 6 (1.1) 0.139 (+0.078, -0.064)♯10 18.5
7 31¶04h10m01.s5 +03◦30′24′′ 7.1 P 13.0 13 (4.7) 0.52 (±0.39); 0.33 (±0.16) 10 12.5 0.31
7 32¶04h09m06.s0 +03◦37′12′′ 5.3 P 19.8 6 (2.1) 0.105 (+0.082, -0.067)♯8.9 2.86
9 1 04h03m51.s5 +05◦10′47′′ 4.7†G 1.8 10 (0.2) 3.19 (±0.86) 16 1.1
9 2¶04h03m06.s3 +04◦44′13′′ 7.1 G 0.5 8 (0.1) 8.4 (±3.4) 0.3 Y
9 3 04h03m43.s1 +05◦16′28′′ 5.9†G 1.6 8 (0.7) 0.95 (±0.44) 10 3.2
9 4¶04h04m28.s5 +05◦18′44′′ 4.3†G 1.5 15 (0.5) 4.9 (±1.3) 0.8 Y
9 5¶04h03m09.s6 +05◦27′27′′ 5.9†G 1.5 6 (0.3) 1.70 (±0.85) 10 3.4
10 1¶22h39m49.s1 +06◦06′00′′ 5.5 G 0.3 10 (0.2) 14.3 (±5.1) 0.1 Y
10 2¶22h40m32.s6 +06◦39′59′′ 5.0 G 1.3 13 (0.3) 4.6 (+1.5, -1.3)♯18 0.8
10 3¶22h38m49.s7 +06◦31′10′′ 6.5 G 0.9 6 (0.4) 3.1 (+1.6, -1.2)♯13 1.0
11 1¶14h14m58.s2 +07◦41′49′′ 5.8 G 1.7 10 (0.7) 2.66 (±0.98) 7.0 1.1
c
2015 RAS, MNRAS 000, 1–18
Swift follow-up of IceCube triggers 7
Table 2 –continued
IceCube Src # Location ErraFlagbExposure C (B)cObs FluxdUpper LimiteNf
seren Pg
const Cat?h
Field (J2000) (′′) (ks) (×10−13 erg cm −2s−1) (×10−13 erg cm −2s−1)
(0.3–10 keV) (0.3–10 keV)
11 2¶14h15m47.s1 +08◦08′10′′ 3.3∗G 1.3 28 (0.8) 9.5 (±1.8) 0.1 Y
12 1¶23h47m47.s9 +20◦32′02′′ 7.6 G 0.4 6 (0.1) 7.1 (+3.4, -2.6)♯5.2 0.4
13 1 10h57m46.s9 +36◦15′39′′ 6.1 G 0.8 9 (0.3) 8.3 (±2.8) 16 0.3 Y
13 2¶10h58m17.s1 +35◦30′30′′ 6.8 G 1.7 9 (0.6) 2.40 (+0.96, -0.77)♯19 2.3
13 3¶10h57m02.s8 +35◦35′22′′ 8.6 G 2.5 10 (1.0) 1.72 (+0.68, -0.55)♯10 3.4 Y
13 4¶10h56m31.s9 +35◦41′52′′ 5.8 G 1.3 7 (0.4) 2.54 (+1.16, -0.90)♯10 2.1 Y
13 5¶10h58m33.s3 +36◦12′32′′ 8.9 G 1.4 7 (0.8) 2.12 (+1.03, -0.80)♯10 2.9 Y
14 1 09h38m14.s9 +02◦00′23′′ 4.4 G 1.5 20 (0.5) 7.3 (±3.6); 5.0 (±1.3) 0.5 0.26 Y
14 2¶09h37m08.s6 +01◦25′50′′ 6.4 G 1.5 7 (0.4) 2.06 (±0.94) 2.9 Y
14 3¶09h37m28.s4 +01◦35′12′′ 6.1 G 2.1 8 (0.9) 3.3 (±1.7) 7.9 1.4
14 4 09h39m09.s2 +01◦44′38′′ 5.1 G 1.4 14 (0.6) 4.4 (±1.2) 9.6 0.8 Y
14 5¶09h37m46.s5 +02◦08′04′′ 5.9 G 1.6 7 (0.4) 2.03 (+0.93, -0.72)♯12 2.9
15 1 18h57m53.s8 +02◦40′08′′ 5.2 G 2.4 20 (1.3) 13.0 (±7.3); 9.7 (±2.6) 1.3 0.47 Y
15 2¶18h57m41.s1 +02◦42′07′′ 4.2†G 2.1 13 (1.2) 2.93 (±0.99) 1.7 Y
15 3¶18h58m09.s8 +02◦21′27′′ 4.8 G 0.5 12 (0.2) 12.4 (±8.9) 0.2 Y
15 4¶18h59m35.s8 +02◦42′01′′ 7.3 G 1.6 7 (0.8) 2.13 (±0.94) 7.2 2.7
16 1¶12h47m06.s6 +15◦12′37′′ 4.2†G 1.4 11 (0.7) 3.6 (+1.3, -1.1)♯12 1.1 Y
16 2¶12h45m46.s9 +14◦36′27′′ 5.7†G 0.7 9 (0.4) 5.9 (+2.3, -1.9)♯11 0.6
16 3¶12h45m37.s6 +14◦48′57′′ 4.8†G 0.8 10 (0.4) 5.6 (+2.1, -1.7)♯15 0.6 Y
16 4¶12h45m35.s6 +14◦40′11′′ 5.3†G 0.8 10 (0.4) 6.0 (+2.2, -1.8)♯13 0.6 Y
16 5¶12h45m37.s9 +14◦56′36′′ 5.1†G 1.5 10 (0.6) 2.97 (+1.11, -0.90)♯11 1.3 Y
18 1 19h55m03.s6 +45◦37′04′′ 5.3 G 1.6 9 (0.6) 3.4 (±1.2) 6.4 1.6
18 2¶19h58m25.s8 +45◦23′29′′ 11.0 G 1.5 8 (0.5) 2.4 (±1.1) 2.4 Y
18 3¶19h58m01.s7 +45◦18′06′′ 5.6 G 1.6 7 (0.7) 2.4 (±1.0) 2.4 Y
18 4¶19h54m45.s1 +45◦43′43′′ 5.1 G 1.4 7 (0.6) 2.7 (±1.1) 8.8 1.9
18 5¶19h59m28.s8 +45◦17′43′′ 7.8 G 1.5 8 (0.6) 2.5 (±1.1) 8.4 2.5
18 6¶19h58m07.s6 +45◦35′49′′ 4.9 G 1.6 10 (0.5) 2.43 (±0.97) 7.9 2.3
18 7¶19h58m39.s9 +46◦01′38′′ 7.0 G 1.8 7 (0.6) 1.54 (±0.78) 6.6 3.5
18 8 19h58m41.s5 +45◦34′11′′ 9.7 P 3.0 5 (0.2) 0.74 (±0.41) 9.6 0.04
19 1 10h21m00.s7 +16◦25′54′′ 4.7 G 1.3 24 (0.6) 5.5 (±1.3); 5.1 (±1.1) 0.3 0.33 Y
19 2 10h21m28.s6 +17◦12′44′′ 4.7 G 2.4 16 (1.0) 1.97 (±0.52) 12 1.3 Y
19 3 10h22m49.s1 +16◦53′45′′ 4.9 G 1.4 17 (0.5) 7.4 (±1.9) 26 0.6
19 4 10h21m51.s5 +16◦34′32′′ 6.1 G 1.3 7 (0.5) 4.6 (±2.0) 44 2.1
19 5¶10h21m18.s2 +17◦12′29′′ 5.2 G 1.3 8 (0.4) 3.2 (±1.2) 24 1.5 Y
19 6¶10h19m35.s0 +16◦52′54′′ 7.3 G 1.5 9 (0.6) 2.71 (+1.09, -0.87)♯16 1.9
19 7¶10h21m48.s5 +17◦03′57′′ 8.7 R 2.4 7 (0.9) 1.19 (+0.59, -0.46)♯8.3 1.1
20 1 17h49m35.s9 +04◦22′29′′ 3.2∗G 1.6 21 (1.2) 3.3 (±1.4); 3.15 (±0.74) 10 0.6 0.88 Y
20 2 17h50m16.s7 +04◦22′07′′ 5.3 G 1.6 9 (1.4) 2.85 (±0.98) 1.3 Y
20 3¶17h47m47.s0 +04◦53′37′′ 5.7 G 0.9 7 (0.5) 3.0 (±1.3) 14 1.5
c
2015 RAS, MNRAS 000, 1–18
8Evans et al.
Table 3. Details of the X-ray sources with spectra
IceCube Source # Na
H,Gal Nb
H,intr Photon Index ECFcPSPCd
Field (×1020 cm−2) (×1020 cm−2)×10−11 erg cm−2ct−1
1 1 27.77 <44.1 1.95(+1.32, -0.70) 3.4 1.11
1 2 27.70 <260 <0.2 4600 0.01
1 4 26.44 <87.9 1.11(+1.40, -0.95) 6.1 0.69
3 2 0.93 <34.0 1.06(+1.15, -0.80) 5.7 1.18
4 2 92.67 <37.6 11.03(+2.81, -0.72) 1.5 1.74
5 7 3.35 <411 2.4(+2.2, -1.3) 2.4 1.43
6 2 7.45 821(+1130, -790) 7.4(+26.5, -6.8) 3.0 0.31
6 3 6.64 <899.4 1.1(+1.9, -1.0) 5.7 0.85
6 4 6.82 <48.1 1.27(+1.01, -0.75) 4.9 0.98
6 7 8.23 <1004 3.8(+3.2, -4.4) 3.4 0.28
6 10 6.54 <46 2.38(+2.05, -0.90) 2.6 1.45
6 11 7.54 <69.4 1.72(+2.26, -0.85) 3.0 1.22
7 1 21.64 <6.2 3.39(+0.36, -0.19) 1.9 1.65
7 2 21.43 <56.0 1.32(+1.53, -0.80) 4.1 0.84
7 4 21.20 <12.5 1.96(+0.26, -0.20) 3.6 1.15
7 5 20.62 <116 5.2(+12.9, -2.2) 2.0 1.81
7 6 21.13 <502 3.0(+7.0, -2.7) 2.3 1.01
7 7 21.46 <85 1.64(+1.04, -0.57) 4.1 0.86
7 8 21.31 240 (+546, -218) 2.05(+0.93, -0.85) 6.7 0.24
7 9 21.27 <74 0.67(+1.14, -0.70) 7.8 0.52
7 10 20.99 <39.9 2.30(+1.30, -0.55) 2.3 1.34
7 11 21.72 <44.7 1.43(+1.25, -0.84) 3.7 0.90
7 12 21.48 <42.1 2.6(+2.1, -1.0) 2.5 1.44
7 13 21.24 <355 1.1(+2.0, -1.1) 10.0 0.35
7 14 21.43 <344 2.2(+2.9, -1.0) 5.1 0.48
7 17 21.62 <305 1.24(+2.82, -0.61) 7.0 0.55
7 18 21.30 <99.5 1.37(+2.08, -0.92) 4.3 0.87
7 19 21.24 <95 1.8(+1.7, -1.5) 3.2 1.08
7 20 21.26 <61.0 1.50(+1.51, -0.79) 4.0 0.94
7 22 21.38 <89.1 1.2(+1.9, -1.1) 9.8 0.81
7 23 20.97 541(1950, -225.0) 10.1(+82.8, -7.0) 1.7 1.04
7 30 21.56 39050(+10000, -8000) >50 2.6 1.50
9 1 19.15 <200 2.6(+3.4, -1.9) 3.4 1.08
9 3 19.01 <497 5.4(+4.5, -4.8) 2.0 1.10
13 1 2.75 <28.8 1.6(+1.3, -1.0) 3.9 1.49
aThe Galactic absorption value, taken from Willingale et al. (2013).
bThe instrinsic absorption, fitted to the spectrum.
cThe conversion from XRT measured count-rate to 0.3–10 keV observed flux.
dThe conversion from XRT measured count-rate to that expected in the Rosat PSPC.
c
2015 RAS, MNRAS 000, 1–18
Swift follow-up of IceCube triggers 9
Table 3 –continued
IceCube Source # Na
H,Gal Nb
H,intr Photon Index ECFcPSPCd
Field (×1020 cm−2) (×1020 cm−2)×10−11 erg cm−2ct−1
14 1 3.44 <21.4 2.37(+1.44, -0.54) 3.1 2.10
14 4 3.89 <34.0 1.91(+2.13, -0.64) 3.7 1.56
15 1 144.00 <1312.4 1.4(+3.3, -2.2) 11.4 0.06
18 1 26.20 <54.7 1.63(+1.27, -0.88) 4.3 0.96
18 5 32.40 37200(6300, -13200) e4.4 1.50
18 8 30.30 <476.9 1.3(+3.4, -2.0) 4.1 0.76
19 1 3.11 <18.6 2.40(+1.49, -0.39) 2.4 2.27
19 2 2.90 <93.5 2.70(+4.96, -0.88) 2.2 1.68
19 3 2.90 <31.8 1.33(+1.05, -0.56) 4.9 1.18
19 4 3.02 <274.9 0.59(+1.42, -0.92) 7.2 0.65
20 1 16.55 <31.9 2.84(+1.47, -0.78) 2.1 1.55
20 2 16.30 <79.9 1.36(+1.15, -0.96) 4.9 0.91
eThe photon index was unconstrained.
c
2015 RAS, MNRAS 000, 1–18
10 Evans et al.
4 IDENTIFYING POTENTIAL
COUNTERPARTS TO THE NEUTRINO
TRIGGER
Each 7-tile observation with Swift-XRT covers ∼0.8 square
degrees, with typical exposures of 1–2 ks per tile. In such
exposure times, our detection limit is 6–10×10−13 erg cm
−2s−1[corresponding to 90% completeness; see Evans et al.
(2014), fig. 14]. This is significantly more sensitive than the
RASS, which covers 92% of the sky down to 0.1 ct s−1in the
PSPC6(Voges et al. 1999), which corresponds to 2.8×10−12
erg cm −2s−1in the 0.3–10 keV band covered by XRT (as-
suming the canonical AGN spectrum described above). For
absorbed sources (i.e. with little flux in the Rosat band-
pass) the increase in sensitivity of Swift-XRT over Rosat
is even greater. The XSS provides hard-band coverage, be-
ing sensitive to ∼3×10−12 erg cm −2s−1in the 2–10 keV
band (Warwick, Saxton & Read 2012), however it only cov-
ers ∼2/3 of the sky. Because of the sensitivity of the XRT
compared to these catalogues, and the low spatial cover-
age of deeper catalogues like 1CSC (Evans et al. 2010)7,
3XMMi-DR4 (Watson et al., in preparation) and 1SXPS
(Evans et al. 2014), we expect to discover uncatalogued X-
ray sources that are serendipitously present in the IceCube
error region. We therefore need to be able to identify the
true X-ray counterpart to the neutrino trigger from among
the unrelated objects detected in the field of view. The first
step is to remove any sources which are already known X-ray
emitters (and are not in outburst at the time of the Swift
observations). We searched the X-ray Master catalogue8for
any catalogued X-ray object with a position agreeing with
our XRT position at the 3-σlevel (including any system-
atic errors on the catalogued positions). For all observations
after 2012 October9we also searched for matches in the
1SXPS catalogue. The sources with catalogue matches are
indicated in Table 2, and details of the matches are given in
Table 4. In all cases where a match was found, the XRT flux
was consistent with (at the 2-σlevel) or occasionally slightly
lower than the catalogued flux, therefore these sources were
all rejected as possible counterparts.
For the remaining, uncatalogued sources we performed
two tests to identify the counterpart: variability tests, and
serendipity likelihoods.
4.1 Variability
The simplest test of whether an uncatalogued X-ray source
is likely to be related to the neutrino trigger is to ask whether
the source brightness is such that is should have been seen
previously, i.e. the measured brightness is significantly above
the sensitivity limit of any instruments which have previ-
ously observed the source location without discovering the
6For individual fields the RASS limit may be deeper.
7No relation.
8http://heasarc.gsfc.nasa.gov/W3Browse/master-
catalog/xray.html
http://ledas-www.star.le.ac.uk/arnie5/arnie5.php?action=basic
&catname=xcoll
9i.e. the dates covered by the 1SXPS catalogue; the obser-
vataions of IceCube fields prior to this date are in the 1SXPS
catalogue, therefore it cannot be used as a reference in those cases.
104105
0.01
0.02
0.05
Count Rate (0.3−10 keV) (s−1)
Time since MET=352486041.0 (s)
red: PC Swift/XRT data of trigger 7 − Source 4
Figure 3. Source 4 of the 7th IceCube trigger. This source mo-
mentarily peaked above the RASS upper limit for its location, and
was fading; however further observations showed the brightness
to have levelled off. This is probably a background AGN.
source. If this is the case then the source is in a bright state
compared to previous observations, and is thus a good can-
didate to be the counterpart.
For each uncatalogued source, we therefore derived
a 3-σupper limit on the Rosat PSPC count rate us-
ing images, background maps and exposure maps ob-
tained from ftp://ftp.xray.mpe.mpg.de/ftp/rosat/archive,
and the Bayesian method of Kraft, Burrows & Nousek
(1991). Where available, we also obtained 3-σupper lim-
its on the 0.2–12 keV flux from the XSS via the service at
http://xmm.esac.esa.int/UpperLimitsServer10 In no cases
was the mean XRT brightness above these upper limits at
the 1-σlevel, although in one source (discussed below) the
flux did briefly peak above the RASS upper limit. The upper
limits for each detected source are given in Table 2.
A second test is to see if the source is variable dur-
ing the Swift observations, particularly whether it is show-
ing significant signs of fading. Such behaviour is expected
from explosive transients such as GRBs or SNe; Tidal Dis-
ruption Events (TDEs) also fade, although they may un-
dergo a period of ∼constant flux for some time before fad-
ing (Lodato & Rossi 2011). Following Evans et al. (2014) we
calculated the Pearson’s χ2for each source where we were
able to produce a light curve (i.e. with more than one bin),
using as a model a constant source with count-rate set to the
mean detected rate for that source. The results of this test
are shown in Table 2. The drawback to this method is that
it requires binned data, therefore one could argue that the
K-S test would be a better indicator of variability. However,
many objects show flickering behaviour in X-rays (e.g. CVs)
which is not indicative of ourbursting. We therefore require
not only a measure of variability, but also a light curve that
we can inspect, after being prompted by that measure. This
10 This service also provides upper limits derived from pointed
XMM observations, however with the exception of the two
sources, which have been previously detected by XMM none of
the objects found by XRT have been in the field of pointed XMM
observations.
c
2015 RAS, MNRAS 000, 1–18
Swift follow-up of IceCube triggers 11
Table 4. Details of the X-ray sources with catalogue matches
IceCube Source # SIMBAD match X-ray match X-ray fluxa
Field (arcsec) (arcsec) (×10−13 erg cm −2s−1)
3 1 2MASS J14590700+5931128 ( 2.9′′ ) 1RXS J145906.5+593105 (8.2′′ ) 6.9±2.4
5 6 [VV2010c] J083046.9+185707 ( 2.0′′ )
6 6 1WGA J0125.4+1003 (55.6′′) 4.23 ±0.67
7 1 TYC 76-1038-1 ( 2.9′′ ) 1RXS J040840.9+033448 (9.7′′) 11.9±2.3
7 3 HD 26292 ( 2.5′′ )
9 2 V* V1296 Tau ( 3.7′′) 1RXS J040307.5+044427 (22.4′′ ) 8.9±3.2
9 4 TYC 79-810-1 ( 2.7′′) 1RXS J040428.2+051854 (10.9′′) 12.5±3.6
10 1 [VV2000] J223949.1+060613 ( 6.8′′) 1RXS J223949.7+060610 (13.0′′) 6.0±2.5
11 2 TYC 902-318-1 ( 4.3′′) 1RXS J141546.6+080755 (19.6′′) 17.9±4.6
13 1 MCG+06-24-038 ( 2.0′′ )
13 3 USNO-A2.0 1200-06616808 ( 3.7′′)
13 4 SDSS J105631.92+354152.8 ( 0.8′′)
13 5 SDSS J105833.38+361228.3 ( 4.2′′)
14 1 2MASX J09381483+0200236 ( 1.2′′ ) 1RXS J093815.6+020026 (10.8′′) 5.0±1.6
14 2 2MASS J09370850+0125440 ( 6.2′′) 1RXS J093708.2+012620 (31.1′′ ) 5.5±2.2
14 4 SDSS J093909.42+014433.5 ( 5.7′′)
15 1 1SXPS J185753.6+024012 (5.0′′) 9.4±1.6
15 2 BD+02 3740 ( 4.1′′) 1SXPS J185741.3+024208 (2.6′′) 2.40 ±0.45
15 3 1SXPS J185809.7+022131 (3.9′′) 5.4 (+1.4, -1.2)
16 1 LBQS 1244+1529 ( 2.9′′ )
16 3 2MASS J12453751+1448572 ( 1.5′′)
16 4 LBQS 1243+1456 ( 1.3′′ )
16 5 [VV2006] J124537.7+145635 ( 4.2′′)
18 2 1RXS J195826.1+452321 ( 8.7′′ ) 3XMM J195825.7+452325 (3.3′′ ) 1.467 ±0.048
18 3 2MASS J19580115+4518060 ( 6.1′′) 3XMM J195801.1+451805 (6.3′′) 0.637 ±0.017
19 1 SDSS J102100.35+162554.0 ( 5.0′′) XMMSL1 J102100.3+162550 (5.5′′ ) 37.8±9.6
19 2 LP 430-7 ( 3.3′′)
19 5 SDSS J102118.34+171227.7 ( 3.7′′)
20 1 HD 162178 ( 8.9′′)
20 2 XMMSL1 J175016.8+042202 (5.9′′) 35±11
aThe catalogued 0.3–10 keV flux. For Rosat objects, this is the catalogued count-rate, converted to 0.3–10 keV flux using the
conversion from Table 2; for other sources it is the catalogued flux, which is already in the 0.3–10 keV band (1SXPS sources)
or the similar 0.2–12 keV band (XMM sources).
implicitly demands binned data, hence our choice of the χ2
test.
Only one uncatalogued object (out of 14 for which we
can probe variability) has a probability of <10% of be-
ing constant, source 4 of field 7; this is the same source
which briefly peaked above the RASS upper limit. In the
initial observations, the light curve of this source comprised
2 bins, with evidence for fading at the 2.3-σlevel. We there-
fore observed the source again repeatedly for a week, but
no signs of continued fading were seen (Fig. 3). This source,
which is spatially coincident with the USNO-B1 source 0935-
0049746, is therefore consistent in behaviour with a back-
ground AGN, from which variations of factors of ∼2 are
not uncommon; there is no evidence for a powerful flare
that could have produced a high neutrino flux (this would
have yielded a much stronger EM flare than that observed).
Detailed examination of the IceCube data showed that the
event was fully consistent with detection of two neutrinos
(i.e. not cosmic-ray-induced muons), but the significance was
at the level at which ∼4 false positives are expected per
year. As a final check of this source we re-observed it with
Swift-XRT on 2014 September 17. In this observation the
source was still easily detected, with a count-rate of ∼0.03
ct s−1, consistent with the previous observations and further
arguing against its being related to the IceCube trigger.
One other source (source 1 in field 7) also showed strong
signs of variability (Pconstant = 6 ×10−4), however this was
a catalogued X-ray source, 1RXS J040840.9+033448, which
SIMBAD11 lists as a rotationally variable star, and it was a
factor of two fainter in our observations than in the Rosat
catalogue, therefore we do not consider this to be a probable
counterpart to the neutrino trigger.
4.2 Probability of serendipity
If we can identify a given source as being unlikely to be
serendipitously present in the field of view then it is, con-
versely, a strong candidate to be the counterpart to the
neutrino trigger and worthy of further observation. A sim-
ple metric to use to quantify this likelihood is the source
brightness. To determine the probability of serendipity for
a given source, we first determined the minimum exposure
necessary to detect the source: for our detection system we
require at least 6 counts, therefore the minimum exposure is
11 http://simbad.u-strasbg.fr/simbad/
c
2015 RAS, MNRAS 000, 1–18
12 Evans et al.
6/R (where Ris the XRT count-rate of the source). We
then measured the sky area, A, in our tiled observation
which was observed with at least this exposure (account-
ing for vignetting, overlaps etc.). We used the log N-log S
brightness distribution of extra-galactic sources, calculated
by Mateos et al. (2008) (based on the data in the 2XMM
catalogue, Watson et al. 2009) to determine the expected
sky density of sources, Dat least as bright as the detected
object in question. The number of expected serendipitous
sources in our field of view, at least as bright as the de-
tected object, is A·D. This was then multiplied by the
completeness of our detection system for the source bright-
ness and exposure time to account for the fact that not all
such sources will have been detected. This yields the num-
ber of serendipirous sources at least as bright as the detected
source that we would expect to detect in our observations.
This value is given, for each source, in Table 2.
Only one uncatalogued source (field 18, source 8) was
found with a low probability of serendipity. This source is
faint, but lies in a region covered by three of the tiled point-
ings and therefore with a deeper exposure than most of the
image. The sky area covered with such an exposure is very
low, which is driving the low probability of serendipity. How-
ever, the source is flagged as Poor by the detection system,
meaning that it has a 35% probability of being a spurious
detection, and the image does not show a strong clustering
of events as expected from a point-like source, therefore we
think this source is unlikely to be real.
The log N-log Scurve of Mateos et al. (2008) was de-
rived only for Galactic latitudes |b|>20◦, whereas some of
the IceCube triggers were at lower Galactic latitudes. We
therefore used the 1SXPS catalogue (Evans et al. 2014) to
investigate the Galactic dependence of the X-ray source den-
sity. For each field in the 1SXPS catalogue we calculated
the density of sources brighter than xct s−1, for values of
x= 1 ×10−4,3×10−4,1×10−3,3×10−3,1×10−2,3×
10−2,1×10−1,3×10−1. We did this using the healpix li-
braries (G´orski et al. 2005) with nside=8192, which results
in pixels of area 0.18 square arcmin. Since Swift observes
many fields multiple times, for each field in 1SXPS we deter-
mined for each healpix pixel which observation or stacked
image contained the most exposure. Only that dataset and
sources detected in it, were considered. For field M, the den-
sity of sources brighter than xis:
ΩM,>x =X
N
1
ANCN
(1)
where the sum is over all Nsources in the field, ANis the
area (in the healpix map) covered by the observation in
which source Nwas detected, and CNis the completeness
of the 1SXPS catalogue for sources at least as bright as
source N(i.e. for each source we detect, there are 1/CN
actual sources). The overall density as a function of Galactic
latitude can then be found simply as:
Ωb,>s =PMΩM,>s AM
PMAM
(2)
where the sum is over all fields Mwith Galactic latitude b.
We compared our results for |b|>20◦with Mateos et al.
(2008), and found that at XRT count rates below ∼2×10−3
ct s−1(equivalent to 0.3–10 keV flux ∼6.3×10−14 erg cm −2
s−1) they were in good agreement. At higher count-rates we
−50 0 50
0.1
1
10
100
Source density (/sq deg)
Galactic latitude
Top−bottom: 1e−4,3e−4, 1e−3,3e−3 etc
Figure 4. The density of X-ray objects above a given flux as a
function of Galactic latitude. For bright sources there is a slight
increase in the density towards the Galactic centre, and for fainter
sources there is a slight deficit, however the effect is small. For
a typical AGN spectrum 1 XRT count corresponds to 4×10−11
erg cm−2. From top to bottom, the lines correspond to 10−4,3×
10−4,10−3,3×10−3,10−2,3×10−2,10−1,3×10−1counts per
second.
predicted slightly more sources than Mateos et al. (2008),
probably because we did not remove from our source list
the objects which were the target of the Swift observations,
whereas Mateos et al. (2008) did, meaning our results are
slightly biased. Nonetheless, we can use our results to give
an indication of the Galactic latitude dependence of source
density, so we split the data into bins of 5◦in Galactic lati-
tude; the source density as a function of latitude and count-
rate is shown in Fig. 4. There is a small effect seen towards
the Galactic plane: at most source brightnesses there is a
slight reduction in source density at |b|<10◦, presumably
related to the increased absorption in the Galactic plane,
however this reduction is less than ∼10%. At the bright-
est fluxes there is an increase of ∼40% in the density of
sources at |b|<10◦, however this is likely to be effected by
pointed observations of transient objects, i.e. this does not
accurately reflect the density of bright serendipitous sources
in the Galactic plane. Based on this analysis, we believe that
it is safe to use the extra-Galactic log N-log Sdistribution
of Mateos et al. (2008) for all of the IceCube triggers, re-
gardless of Galactic latitude.
5 DISCUSSION AND IMPLICATIONS FOR
MULTI-MESSENGER ASTRONOMY
We have reported on Swift-XRT X-ray follow up of 20 neu-
trino triggers from the IceCube observatory. Although we
have found 109 X-ray sources, none of them have been iden-
tified as a likely counterpart to an IceCube trigger. Con-
versely, only the 30 objects listed in Table 4 were known
prior to the Swift observations, and only 16 of them had
been detected in X-rays previously, therefore it is very dif-
ficult for us to rule out with any degree of confidence that
we have detected an electromagnetic (EM) counterpart to
the neutrino trigger, even though we cannot identify it. To
investigate further we therefore need to consider what we
c
2015 RAS, MNRAS 000, 1–18
Swift follow-up of IceCube triggers 13
expect the counterparts to look like in our follow-up obser-
vations. Also, IceCube continues to send neutrino triggers
to Swift and the field of multi-messenger astronomy at large
is growing rapidly: for example, advanced LIGO (aLIGO)
is expected to be commissioned in early 2015 (Harry et al.
2010). Therefore, it is necessary to consider whether lessons
can be learned from the IceCube followup presented in this
paper, for future EM follow up on non-EM triggers. Note
that we will focus on how to improve our ability to iden-
tify the counterpart from among the sources detected in an
observation: we do not consider the related issue of how op-
timally to observe a large (and potentially disjoint) error
region; for a discussion of that sub ject, see Kanner et al.
(2012).
Although there are a range of phenomena that can pro-
duce neutrino triggers, in this discussion we focus on GRBs
(which are also a potential source of gravitational waves),
since these have been well studied in the X-ray domain by
Swift and, as the brightest known transients in the universe,
they represent an upper bound on the detectability of the
EM counterparts to neutrino triggers.
5.1 How bright will the X-ray counterpart be?
In our analysis (Section 4.1) we looked for, and failed to
find, any uncatalogued sources in our dataset which were so
bright that they would have been catalogued if they were
persistent. The lack of such sources may indicate that we
did not find the counterpart to the IceCube trigger, or sim-
ply that we observed too late after the trigger, when the
source had faded below the limit of the existing catalogues.
How bright the EM counterpart to a non-EM trigger is at
a given time depends on the nature of the emitting object.
Here we consider GRBs (Klebesadel, Strong & Olson 1973),
since they are expected to be sources of neutrinos and gravi-
tational waves. GRBs are the brightest known EM transients
in the universe, and, although they fade fairly rapidly, they
give us a reasonable upper limit on the brightness we may
expect from the X-ray counterpart to a non-EM trigger.
The top panel of Fig. 5 shows the distribution of the X-
ray flux of GRB afterglows observed with XRT, at a range
of times since the trigger. These were derived using the
live XRT GRB catalogue12 (Evans et al. 2009) from which
we used the light-curve fits for all GRBs with at least 5
light-curve bins (i.e. where the fit was reasonably well con-
strained), and an ECF of 4.1×1011 erg cm−2ct−1, which is
the mean value in that catalogue. About 60% of GRBs are
expected to be above the the typical RASS/XSS sensitiv-
ity limit of ∼3×10−12 erg cm −2s−1(0.3–10 keV) at one
hour after the trigger, this falls to about 15% by eight hours.
Comparing this with the delay between the IceCube trigger
and the start of the Swift observations (Table 1) we would
expect that in ∼8 of cases we should have found an uncat-
alogued source above the RASS/XSS limit, if the neutrino
triggers were related to GRBs. The lack of any such object
rules out the idea that all 20 triggers arose from GRBs with
>99% (i.e. 3-σ) confidence. However, the companion paper
to this one (Aartsen et al., in preparation) shows that many
(or all) of the neutrino triggers could have been spurious;
12 http://www.swift.ac.uk/xrt live cat
10−18 10−17 10−16 10−15 10−14 10−13 10−12 10−11 10−10 10−9 10−8
0
50
100
Cumulative percentage
0.3−10 keV flux (erg cm−2 s−1)
10−18 10−17 10−16 10−15 10−14 10−13 10−12 10−11 10−10 10−9 10−8
50
100
Cumulative percentage
0.3−10 keV flux (erg cm−2 s−1)
10−18 10−17 10−16 10−15 10−14 10−13 10−12 10−11 10−10 10−9 10−8
0
50
100
Cumulative percentage
0.3−10 keV flux (erg cm−2 s−1)
Figure 5. The cumulative distribution of the brightness of GRB
afterglows detected by the Swift-XRT. The solid line showed the
flux 1 hour after the trigger, the dashed line is 1.8 hours after
the trigger, and the dot-dashed and dotted lines are for 4 and
8 hours after the trigger, respectively. The brightness is taken
by evaluating the best-fitting model at these times, which may
involve extrapolating past the time a GRB ceases to be detected
by Swift. The top panel shows all GRBs, the centre panel shows
only the short GRBs. The bottom panel is as the centre panel, but
with the GRBs shifted to be at a distance of 200 Mpc [assuming
the redshifts given in Rowlinson et al. (2013)].
c
2015 RAS, MNRAS 000, 1–18
14 Evans et al.
Figure 6. The sensitivity of Swift -XRT (black lines), and the
expected number of serendipitous source expected per XRT field
above this limit (red lines), as a function of exposure time. The
solid lines correspond to the 50% completeness level, and the
dashed lines the 90% completeness level. Note that the two y-
axes do not correspond with each other, but are only related via
the x-axis and plotted data.
The green horizontal line shows the sensitivity limit of the RASS
Voges et al. (1999), corresponding to 0.1 PSPC ct s−1(converted
to 0.3–10 keV flux assuming the canonical AGN spectrum de-
scribed in the text), at which level the RASS covers 92% of the
sky. The XMM slew survey 2–10 keV band limit is at a similar
level (3×10−12 erg cm −2s−1; Warwick, Saxton & Read 2012).
if even half of the triggers were spurious, this significance
drops to below 3-σ.
The lack of bright sources does not mean that we did not
detect a GRB afterglow: in more than half of the triggers,
by the time Swift observed, the afterglow would have faded
below the RASS/XSS limit. However, the ability to identify
an afterglow at these lower fluxes is hampered by the density
of expected (uncatalogued) sources, as illustrated in Fig. 6.
This shows the level (black) at which XRT is 50% and 90%
complete (Evans et al. 2014), and the expected number of
serendipitous sources (red) per XRT field of view above these
levels (Section 4.2) as a function of exposure time. The green
line corresponds to the typical RASS/XSS limit. The XRT
90% completeness level reaches the RASS and XSS limits in
an exposure of ∼350 s; and we expect ∼0.01 serendipitous
sources per XRT field with fluxes above this limit. That is,
in a 7-tile observation such as those reported in this paper,
any detected source below the flux limit set by the exist-
ing large-area catalogues, will have a probability of being
serendipitous of ≥0.07, i.e. we cannot expect to identify the
counterpart with even 2-σconfidence.
It is impossible therefore, for us to identify the coun-
terpart to the neutrino triggers reported in this paper based
on the source flux at detection, and in any future follow-up
of astrophysical neutrinos, we would expect at best 50% of
GRB afterglows to be identified in this way.
While neutrinos are expected from all GRBs, a prime
candidate for the sources of gravitational waves are nearby
short GRBs, which arise from the merging of two neutron
stars. The middle panel of Fig. 5 shows the flux distribu-
tion of the short GRBs detected by the Swift-XRT: they are
much fainter than long GRBs and we are unlikely to observe
any before they fall below the limits of existing catalogues.
However, the horizon distance of aLIGO is around 200 Mpc
(Abadie et al. 2010a), whereas the the average short GRB
redshift in the Swift sample is 0.72 (Rowlinson et al. 2013),
corresponding to a luminosity distance of ∼4000 Mpc. Thus
on-axis short GRBs detected by aLIGO should be a factor
of ∼400 brighter than those detected by Swift, although the
time-axis of the light curve is compressed by the reduced
time dilation, which shortens any plateau phase. In the bot-
tom panel of Fig. 5, we have shifted the XRT afterglows
from the redshifts given in Rowlinson et al. (2013) to 200
Mpc (z= 0.045). In this case ∼80% of short GRBs would
be above the RASS limit one hour after the trigger, and 50%
would still be that bright at eight hours. These results are
less optimistic than those reported by Kanner et al. (2012),
however they used only short GRBs with known redshift
(giving a smaller sample), whereas we have included short
GRBs with no known redshift, assigning to them the mean
short GRB redshift of 0.72. It should also be noted, that in
ten years of operation, Swift has not yet detected a short
GRB less than 500 Mpc away (GRB 061201, z= 0.111
Berger 2006), and indeed no short GRB thousands of times
brighter than the typical Swift short GRBs has been re-
ported in over twenty years of observations by various fa-
cilities. This tells us that nearby short GRBs, which may
trigger aLIGO, are extremely rare.
5.1.1 Increasing the sensitivity
Our ability to identify a counterpart by its brightness would
be enhanced if we had a more sensitive reference catalogue.
For example, Fig. 6 shows that if Swift-XRT had conducted
a 2 ks observation of a field prior to an IceCube trigger, then
the list of known sources at that location would be 90% com-
plete down to a flux 5 times below the RASS limit; for hard
or absorbed sources the increase in sensitivity is significantly
more pronounced. At such levels, 95% (50%) of the Swift-
detected GRBs would be bright enough to be confirmed as
new (non-serendipitous) sources in an observation at 1 (8)
hours after the trigger.
To pre-image the entire sky with Swift -XRT, at 2-ks
per field, is clearly not practical (it would require around
18 years of observing time!), although some subset of the
sky, for example, corresponding to the galaxies deemed most
likely to yield a short GRB that aLIGO would detect, could
potentially be observed. The forthcoming eRosita mission,
expected to launch in 2016, will produce an all-sky survey in
the 0.2–10 keV band which will be a factor of 30 more sensi-
tive than the RASS (Cappelluti et al. 2011). This will pro-
vide a valuable resource for identifying new sources in Swift-
XRT observations of non-EM triggers. In the meantime, cat-
alogues such as the 1CSC (Evans et al. 2010), 3XMMi-DR4
(Watson et al., in preparation) and 1SXPS (Evans et al.
2014) could be used when available, but their sky coverage
is very limited.
5.2 Identifying counterparts by fading light curves
Transient events by definition fade over time. However, in
our follow-up observations, only 19 (out of 109) sources were
c
2015 RAS, MNRAS 000, 1–18
Swift follow-up of IceCube triggers 15
100 1000200 500
100
1000
200
500
2000
5000
Required exposure time (s)
Initial exposure time (s)
Figure 7. The minimum second-observation exposure time
needed to detect the fading of a source as a function of the expo-
sure time in the initial image. The source brightness in the first
observation is that at which our detection system is 50% com-
plete in the initial exposure time. If the source is detectable in
the second observation, the black (red) solid line shows the min-
imum time needed to detect the source in the second image, and
measure the count-rate as 2 (3) σfainter than in the initial im-
age. If the source is not detectable in the second image, the black
dotted line shows the exposure time needed to obtain a 3-σupper
limit inconsistent with the initial brightness if the source is only
just below the detection threshold; the black dashed line is that
needed if the source has faded away completely. See Section 5.2
for full details.
bright enough (or observed for long enough) to yield two
or more light curve bins, and 12 of these occured in the
field of trigger #7, which was observed for an unusually
long time to allow us to rule out the possible counterpart
in that field (Section 4.1). Also, not all transient sources
fade on the timescale of a single observation. GRBs, for
example, tend to fade quickly, but many show a ‘plateau’
phase where there is little or no decay (Zhang et al. 2006;
Nousek et al. 2006); tidal disruption events similarly begin
with a period of roughly constant flux before beginning their
decay (Lodato & Rossi 2011). To discover whether a source
is fading it is therefore necessary to perform repeat observa-
tions, and if the type of transient is not known, then several
such observations are needed, with increasing delays since
the trigger to account for different progenitor types. While
this strategy was not employed for the neutrino triggers re-
ported in this paper, such an approach should be considered
in the future, given the lack of identified counterparts in this
work.
In order to confirm fading, these extra observations need
to be of sufficient exposure for us to either measure the
count-rate accurately enough to confirm a decay with some
predetermined level of significance, or find an upper limit
below the level of the previous detection. To accurately de-
termine how much exposure is needed requires knowledge of
the source light curve. As a generic approach, we can de-
termine the count-rate R2in the second observation which
allows us to confirm fading in the shortest possible exposure
time13. Since the optimal value of R2is not a priori obvi-
13 This is a function of the background level and the size of region
ous, we stepped it over the range 0.001R1– 0.99R1, finding
the value which gave the minimum exposure time needed to
detect fading at the 2-σand 3-σlevels. These are shown in
Fig. 7; where we plot the initial exposure (E1) on the x-axis,
and take R1as the count rate at which the source detection
algorithm is 50% complete in that exposure time. If R2was
below the detection threshold in the second observation, we
found the exposure time that would be needed to give a
3-σupper limit on the count-rate that was below the 1-σ
lower bound on R1. We did this for two cases: where the R2
was just below the detection threshold (i.e. the source con-
tributed 5 counts to XRT dataset), and where the source
had completely vanished (it contributed no photons); these
are plotted as the dotted lines in Fig. 7.
From this we see that a typical repeat observation will
need to be at least a factor of two longer than the initial
exposure; unless prior knowledge of the source type means
that we expect the source to have faded away completely by
the time of the second observation. However, this does not
mean that the total observing time needs to be doubled. The
goal of the follow-up observations is to determine whether
any of the uncatalogued sources detected in the first obser-
vation have faded, not to search for new sources. Therefore,
only those fields containing uncatalogued sources need to
be re-observed. We suggest that a modest investment of ob-
serving time spent on observations of the newly-discovered
X-ray sources would significantly increase the likelihood of
identifying an X-ray counterpart to a non-EM trigger using
Swift – provided that the source was detected in the initial
observation.
We can determine with more confidence the exposure
time needed to determine fading if we know in advance what
the source light curve looks like. One of the main candidates
to provide a gravitational wave + EM detection are short
GRBs, and thanks to Swift we have some idea of their X-ray
light curve morphology. We took from the short GRBs listed
in Rowlinson et al. (2013) those which triggered Swift and
for which the live XRT GRB catalogue contains light curve
fits (i.e. at least 2 light curve bins, excluding upper limits).
Then, assuming an initial exposure of 500 s and 1000 s, start-
ing 1, 2 and 4 hours after the GRB trigger, we determined
how many of those GRBs would be detected by Swift, and
at what brightness. We then determined how long a second
observation would have to be in order to confirm that the
source had faded at the 3-σlevel, as a function of the delay
between the first and second observations. We considered a
source to have faded either if it was detectable, and had a
measured count-rate inconsistent with the initial rate at the
3-σlevel, or if it was undetected with a 3-σupper limit in-
consistent with that rate. The results are shown in the top
panels of Fig. 8, in which we show the exposure time needed
to detect fading in 50% and 90% of the bursts as a func-
tion of the time between observations. Note that “50% of
the bursts” means 50% of the short GRBs that would have
been detected in the initial observation. The shorter that
observation is, or the longer after the trigger it begins, the
over which source counts are accumulated. We set the background
to 10−6ct s−1pixel−1, the mean value from the 1SXPS catalogue
(Evans et al. 2014), and set the region size to be that used by the
XRT auto-analysis software for the brightness of the source in the
initial observation, see table 1 of Evans et al. (2007).
c
2015 RAS, MNRAS 000, 1–18
16 Evans et al.
105
2×1045×1042×1055×105
10
100
1000
104
105
Exposure needed to confirm fading (s)
Time between first and second observation (s)
Original distance
Initial obs 500 s starting:
Black: 1 hr (11 GRBs)
Red: 2 hr (9 GRBs)
Blue: 4 hr (6 GRBs)
hrs after the trigger
105
2×1045×1042×1055×105
10
100
1000
104
105
Exposure needed to confirm fading (s)
Time between first and second observation (s)
Original distance
Initial obs 1000 s starting:
Black: 1 hr (12 GRBs)
Red: 2 hr (11 GRBs)
Blue: 4 hr (9 GRBs)
hrs after the trigger
105
2×1045×1042×1055×105
10
100
1000
104
105
Exposure needed to confirm fading (s)
Time between first and second observation (s)
Distance adjusted to 200 Mpc
Initial obs 500 s starting:
Black: 1 hr (21 GRBs)
Red: 2 hr (21 GRBs)
Blue: 4 hr (17 GRBs)
hrs after the trigger
105
2×1045×1042×1055×105
10
100
1000
104
105
Exposure needed to confirm fading (s)
Time between first and second observation (s)
Distance adjusted to 200 Mpc
Initial obs 1000 s starting:
Black: 1 hr (21 GRBs)
Red: 2 hr (21 GRBs)
Blue: 4 hr (19 GRBs)
hrs after the trigger
Figure 8. The exposure time necessary in a follow-up observation of a short GRB needed to confirm fading at the 3-σlevel, as a
function of the delay between the initial and follow-up observations. The solid (dotted) lines are the exposure needed for 50% (90%) of
the short GRBs detected in the initial observation to be confirmed fading at the 3-σlevel. The black, red and blue lines are for an initial
observation beginning 1, 2 and 4 hours after the GRB trigger respectively. The numbers in the legend indicate the number of GRBs
corresponding to 100% for each dataset, i.e. the number of GRBs that would have been detected in the initial observation. This is from
a total of 43 GRBs which are input to the simulations.
Left: For an initial observation of 500 s. Right: For an initial observation of 1000 s. In the bottom panels we have adjusted the light
curves of the short GRBs as if the GRB was at 200 Mpc, using the redshifts and GRB list given in Rowlinson et al. (2013). In the top
panels, the GRB light curves as detected by Swift were used.
The point where the lines stop decyaing reflects corresponds to the point at which the 50th or 90th percentile becomes too faint to
detect, i.e. to confirm fading we require an upper limit below the level of the initial detection. This level, and hence the time needed to
generate the upper limit, does not depend on the time between observations.
fewer the bursts detected in the initial observation. There-
fore (for example) the black curve in the top-left panel of
Fig. 8 was compiled from 11 GRBs, whereas the red curve
in the same panel was compiled from 9 GRBs; the sample
from Rowlinson et al. (2013) contains 43 bursts14. This ex-
plains why the necessary exposure is surprisingly short at
some times: only the brightest bursts are detectable, but for
these bright bursts, it is relatively simple to identify fading.
For some GRBs, we were not able to confirm fading with
3-σconfidence if the delay between observations was too
short, with the second observataion’s exposure extending to
14 Since we can’t measure fading in objects we don’t detect, and
we are interested in how to detect fading, not how to detect the
GRB in the first place, this approach is the most informative.
a maximum of 105s; for the purposes of plotting, we set
the necessary exposure time in these objects to be 106s (i.e.
off the scale). For shorter delay times it is sometimes not
possible to determine fading in even 50% of cases; the only
solution is to wait longer before performing the second ob-
servation. Note also that for simplicity we assumed that the
second observation was a continuous one, i.e. we ignored the
fact that Swift can observe a given target for a maximum of
2.7 ks every ∼5.7 ks orbit.
As noted earlier (Section 5.1), the mean redshift of Swift
short GRBs is ∼0.7, whereas for aLIGO the maximum dis-
tance to a short GRB is expected to be 200 Mpc (z∼0.045).
Therefore we repeated the above calculations, first shifting
the GRB to be at 200 Mpc. We used the redshifts from
Rowlinson et al. (2013), i.e. reverting to the mean (z= 0.72)
where it is not known. The results of this are shown in the
c
2015 RAS, MNRAS 000, 1–18
Swift follow-up of IceCube triggers 17
bottom panels of Fig. 8. Not surprisingly, fading is much
easier to detect when the GRB is nearby; indeed, if we can
detect a GRB at 200 Mpc within 4 hours of the trigger, then
we should be able to confirm fading at most 8 hours later.
6 CONCLUSIONS
We have used the Swift-X-ray telescope to observe 20 Ice-
Cube neutrino doublet triggers, covering the IceCube error
circle in 7 tiled observations. We find 109 X-ray sources,
only 16 of which had been detected in X-rays before these
observations were taken. However, none of the uncatalogued
sources are bright enough to be distinguished from the
serendipitous sources expected in a 7-tile XRT observation.
Given the behaviour of GRB afterglows observed by the
Swift-XRT, this lack of bright counterparts allows us to rule
out with >99% confidence that the neutrino triggers arise
exclusively from GRBs. In only one case could we identify
signs of fading in these single observations, and follow-up
observations showed that this was not a transient object.
Considering the wider question of using Swift -XRT to
detect the counterpart to non-EM triggers, such as PeV neu-
trinos or Gravitational Wave triggers from the Advanced-
LIGO facility, we have shown that a deeper all-sky X-ray
catalogue than the RASS, such as that which will be cre-
ated by eRosita, will make it easier to identify a counterpart
simply from its flux in the initial observation. However, a
better and more immediately available approach will be to
probe source variability, in particular, searching for fading.
Re-observing every uncatalogued X-ray source detected by
XRT in the initial follow-up observations, with at least twice
the initial exposure time, should enable us to identify any
source which has faded at the 3-σlevel.
For a short GRB at a distance of 200 Mpc – as aLIGO
expects to discover – with its jet pointed towards Earth,
an initial observation within a few hours of the trigger, fol-
lowed by a second observation of 2-ks, 8 hours after the first,
should be sufficient to confirm fading in almost all cases (if
the GRB was detected in the initial observation), provided
the afterglows of such GRBs are of the same luminosity and
morphology distributions as the sample of short GRBs de-
tected to date by Swift.
While we have demonstrated techniques to maximise
the potential for multi-messenger astronomy with Swift,
a sensitive, all-sky (or at least large field-of-view) X-ray
imager, ideally with some form of gamma-ray detector to
rapidly distinguish GRBs from other X-ray transients, would
be the ideal facility with which to locate the high-energy
counterparts to non-EM-detected transients.
ACKNOWLEDGEMENTS
This work made use of data supplied by the UK Swift Sci-
ence Data Centre at the University of Leicester. PAE and
JPO acknowledge UK Space Agency support. JMG grate-
fully acknowledges the support from NASA under award
NNH13CH61C. We thank the anonymous referee for their
helpful and constructive feedback on the manuscript.
REFERENCES
Aartsen M. G. et al., 2013, Physical Review Letters, 111,
021103
Abadie J. et al., 2010a, Class. Quantum Gravity, 27, 173001
Abadie J. et al., 2010b, Astrophys. J., 715, 1453
Abbasi R. et al., 2014, Astron. Astrophys., 563, C1
Abbasi R. et al., 2012, Nature, 484, 351
Abbasi R. et al., 2011, Physical Review Letters, 106, 141101
Abbott B. P. et al., 2009, Reports on Progress in Physics,
72, 076901
Abraham J. et al., 2010, Physics Letters B, 685, 239
Accadia T. et al., 2012, Journal of Instrumentation, 7, 3012
Achterberg A. et al., 2006, Astroparticle Physics, 26, 155,
cited By (since 1996)198
Adri´an-Mart´ınez S. et al., 2013, Astron. Astrophys., 559,
A9
Alekseev E. N., Alekseeva L. N., Volchenko V. I.,
Krivosheina I. V., 1987, Soviet Journal of Experimental
and Theoretical Physics Letters, 45, 589
Arnaud K. A., 1996, in Astronomical Society of the Pacific
Conference Series, Vol. 101, Astronomical Data Analysis
Software and Systems V, Jacoby G. H., Barnes J., eds.,
p. 17
Asano K., M´esz´aros P., 2014, ApJ, 785, 54
Berger E., 2006, GRB Coordinates Network, 5952, 1
Bionta R. M., Blewitt G., Bratton C. B., Casper D., Ciocio
A., 1987, Physical Review Letters, 58, 1494
Burrows D. N. et al., 2005, Space Sci. Rev., 120, 165
Butler N. R., 2007, AJ, 133, 1027
Cappelluti N. et al., 2011, Memorie della Societa Astro-
nomica Italiana Supplementi, 17, 159
Evans I. N. et al., 2010, ApJS, 189, 37
Evans P. A. et al., 2009, MNRAS, 397, 1177
Evans P. A. et al., 2007, A&A, 469, 379
Evans P. A. et al., 2012, ApJS, 203, 28
Evans P. A. et al., 2014, ApJS, 210, 8
Gehrels N. et al., 2004, ApJ, 611, 1005
Goad M. R. et al., 2007, A&A, 476, 1401
G´orski K. M., Hivon E., Banday A. J., Wandelt B. D.,
Hansen F. K., Reinecke M., Bartelmann M., 2005, ApJ,
622, 759
Harry G. M. et al., 2010, Classical and Quantum Gravity,
27, 084006
Hinton J. A. et al., 2004, New A Rev., 48, 331
Hirata K., Kajita T., Koshiba M., Nakahata M., Oyama
Y., 1987, Physical Review Letters, 58, 1490
Hirata K. S. et al., 1988, Phys. Rev. D, 38, 448
Kachelrieß M., Tom`as R., Buras R., Janka H.-T., Marek
A., Rampp M., 2005, Phys. Rev. D, 71, 063003
Kanner J., Camp J., Racusin J., Gehrels N., White D.,
2012, ApJ, 759, 22
Klebesadel R. W., Strong I. B., Olson R. A., 1973, ApJ,
182, L85+
Kraft R. P., Burrows D. N., Nousek J. A., 1991, ApJ, 374,
344
Kunkel W. et al., 1987, IAU Circ., 4316, 1
Lodato G., Rossi E. M., 2011, MNRAS, 410, 359
Mateos S. et al., 2008, A&A, 492, 51
Moretti A. et al., 2007, in Society of Photo-Optical In-
strumentation Engineers (SPIE) Conference Series, Vol.
6688, Society of Photo-Optical Instrumentation Engineers
c
2015 RAS, MNRAS 000, 1–18
18 Evans et al.
(SPIE) Conference Series
Nousek J. A. et al., 2006, ApJ, 642, 389
Pfeffermann E. et al., 1987, in Society of Photo-Optical
Instrumentation Engineers (SPIE) Conference Series, Vol.
733, Soft X-ray optics and technology, p. 519
Rowlinson A., O’Brien P. T., Metzger B. D., Tanvir N. R.,
Levan A. J., 2013, MNRAS, 430, 1061
Saxton R. D., Read A. M., Esquej P., Freyberg M. J., Al-
tieri B., Bermejo D., 2008, A&A, 480, 611
Skrutskie M. F. et al., 2006, AJ, 131, 1163
Voges W. et al., 1999, A&A, 349, 389
Warwick R. S., Saxton R. D., Read A. M., 2012, A&A, 548,
A99
Watson M. G. et al., 2009, A&A, 493, 339
Willingale R., Starling R. L. C., Beardmore A. P., Tanvir
N. R., O’Brien P. T., 2013, MNRAS, 431, 394
Zhang B., Fan Y. Z., Dyks J., Kobayashi S., M´esz´aros P.,
Burrows D. N., Nousek J. A., Gehrels N., 2006, ApJ, 642,
354
c
2015 RAS, MNRAS 000, 1–18