A Repeating Fast Radio Burst
L. G. Spitler
1
, P. Scholz
2
, J. W. T. Hessels
3,4
, S. Bogdanov
5
, A. Brazier
6,7
, F. Camilo
5,8
, S. Chatterjee
6
,
J. M. Cordes
6
, F. Crawford
9
, J. Deneva
10
, R. D. Ferdman
2
, P. C. C. Freire
1
, V. M. Kaspi
2
, P. Lazarus
1
,
R. Lynch
11,12
, E. C. Madsen
2
, M. A. McLaughlin
12
, C. Patel
2
, S. M. Ransom
13
, A. Seymour
14
,
I. H. Stairs
15,2
, B. W. Stappers
16
, J. van Leeuwen
3,4
& W. W. Zhu
1
Published online by Nature on 2016 March 2. DOI: 10.1038/nature17168
1
Max-Planck-Institut f
¨
ur Radioastronomie, Auf dem H
¨
ugel 69, B-53121 Bonn, Germany
2
Department of Physics and McGill Space Institute, McGill University, 3600 University St., Mon-
treal, QC H3A 2T8, Canada
3
ASTRON, Netherlands Institute for Radio Astronomy, Postbus 2, 7990 AA, Dwingeloo, The
Netherlands
4
Anton Pannekoek Institute for Astronomy, University of Amsterdam, Science Park 904, 1098 XH
Amsterdam, The Netherlands
5
Columbia Astrophysics Laboratory, Columbia University, New York, NY 10027, USA
6
Department of Astronomy and Space Sciences, Cornell University, Ithaca, NY 14853, USA
7
Cornell Center for Advanced Computing, Cornell University, Ithaca, NY 14853, USA
8
Square Kilometre Array South Africa, Pinelands, 7405, South Africa
9
Department of Physics and Astronomy, Franklin and Marshall College, Lancaster, PA 17604-
3003, USA
10
Naval Research Laboratory, 4555 Overlook Ave SW, Washington, DC 20375, USA
11
National Radio Astronomy Observatory, PO Box 2, Green Bank, WV, 24944, USA
12
Department of Physics and Astronomy, West Virginia University, Morgantown, WV 26506, USA
13
National Radio Astronomy Observatory, Charlottesville, VA 22903, USA
14
Arecibo Observatory, HC3 Box 53995, Arecibo, PR 00612, USA
15
Department of Physics and Astronomy, University of British Columbia, 6224 Agricultural Road
Vancouver, BC V6T 1Z1, Canada
16
Jodrell Bank Centre for Astrophysics, School of Physics and Astronomy, University of Manch-
ester, Manchester, M13 9PL, UK
Fast Radio Bursts are millisecond-duration astronomical radio pulses of unknown physical
origin that appear to come from extragalactic distances
1–8
. Previous follow-up observations
have failed to find additional bursts at the same dispersion measures (i.e. integrated col-
umn density of free electrons between source and telescope) and sky position as the original
detections
9
. The apparent non-repeating nature of the fast radio bursts has led several au-
thors to hypothesise that they originate in cataclysmic astrophysical events
10
. Here we report
the detection of ten additional bursts from the direction of FRB 121102, using the 305-m
Arecibo telescope. These new bursts have dispersion measures and sky positions consistent
with the original burst
4
. This unambiguously identifies FRB 121102 as repeating and demon-
strates that its source survives the energetic events that cause the bursts. Additionally, the
1
arXiv:1603.00581v1 [astro-ph.HE] 2 Mar 2016
bursts from FRB 121102 show a wide range of spectral shapes that appear to be predom-
inantly intrinsic to the source and which vary on timescales of minutes or shorter. While
there may be multiple physical origins for the population of fast radio bursts, the repeat
bursts with high dispersion measure and variable spectra specifically seen from FRB 121102
support models that propose an origin in a young, highly magnetised, extragalactic neutron
star
11, 12
.
2
FRB 121102 was discovered
4
in the PALFA survey, a deep search of the Galactic plane
at 1.4 GHz for radio pulsars and fast radio bursts (FRBs) using the 305-m William E. Gordon
Telescope at the Arecibo Observatory and the 7-beam Arecibo L-band Feed Array (ALFA)
13, 14
.
The observed dispersion measure (DM) of the burst is roughly three times the maximum value
expected along this line of sight in the NE2001 model
15
of Galactic electron density, i.e. β
DM
≡
DM
FRB
/DM
Gal
Max
∼ 3, suggesting an extragalactic origin.
Initial Arecibo follow-up observations were limited in both dwell time and sky coverage
and resulted in no detection of additional bursts
4
. In 2015 May and June we carried out more
extensive follow-up using Arecibo, covering a ∼ 9
0
radius with a grid of six ALFA pointings
around the then-best sky position of FRB 121102 (Figure 1 and Extended Data Table 1 and 2). As
described in the Methods, high-time-resolution, total intensity spectra were recorded, and the data
were processed using standard radio-frequency interference (RFI) excision, dispersion removal,
and single-pulse-search algorithms implemented in the PRESTO software suite and associated
data reduction pipelines
14, 16, 17
.
We detected 10 additional bursts from FRB 121102 in these observations. The burst prop-
erties, and those of the initial FRB 121102 burst, are listed in Table 1. The burst intensities are
shown in Figure 2. No other periodic or single-pulse signals of a plausible astrophysical origin
were detected at any other DM. Until the source’s physical nature is clear, we continue to refer to
it as FRB 121102 and label each burst chronologically starting with the original detection.
The ten newly detected bursts were observed exclusively in two adjacent sky positions of the
telescope pointing grid located ∼ 1.3
0
apart (Figure 1 and Extended Data Table 1). The unweighted
average J2000 position from the centres of these two beams is right ascension α = 05
h
31
m
58
s
,
declination δ = +33
d
08
m
04
s
, with an uncertainty radius of ∼ 3
0
. The corresponding Galactic
longitude and latitude are l = 174
◦
.89, b = −0.
◦
23. This more accurate position is 3.7
0
from the
beam centre of the discovery burst
4
, meaning that FRB 121102 Burst 1 was detected well off axis,
as originally concluded.
The measured DMs of all 11 bursts are consistent to within the uncertainties, and the dis-
persion indices (dispersive delay ∆t ∝ ν
−ξ
) match the ξ = 2.0 value expected for radio waves
traveling through a cold, ionised medium. This is strong evidence that a single astronomical source
is responsible for the events. In addition, the ∼ 0.002 DM index uncertainty we calculate for Burst
11 (see Methods) is slightly less than that reported for FRB 110523
8
, making this the most precise
determination of dispersion index for any FRB thus far. The upper bound on the dispersion index
is identical to that of FRB 110523
8
, and hence, following the same arguments used there, Burst 11
provides a similar lower limit of 10 AU for the size of the dispersive region.
The 11 bursts have peak flux densities gS
1400
∼ 0.02 − 0.3 Jy at 1.4 GHz, where g is the
antenna gain at the source’s unknown location in the beam normalised to unit amplitude on the
beam axis. The other known FRBs typically have order-of-magnitude higher peak flux densities
of gS
1400
∼ 0.2 − 2 Jy. The wide range of flux densities seen at Arecibo, some near the detection
threshold, suggests that weaker bursts are also produced, likely at a higher rate. The rate of burst
3
detections is ∼3 hr
−1
for bursts with gS
1400
& 20 mJy over all observations in which an ALFA
beam was within 3.5
0
of the improved position. We note, however, that the bursts appear to cluster
in time with some observing sessions showing multiple bright bursts and others showing none.
The observed burst full widths at half maximum are w
50
= 2.8− 8.7 ms, which are consistent
with the w
50
= 1.3 − 23.4 ms widths seen from other FRBs. No clear evidence for scatter broad-
ening was seen in any of the bursts. Bursts 8 and 10 show double-peaked profiles, which has also
been seen in FRB 121002
7
. Furthermore, the morphologies of Bursts 8 and 10 evolve smoothly
with frequency.
Within our observing band (1.214 − 1.537 GHz) the burst spectra are remarkably variable.
Some are brighter toward higher frequencies, as in the initial discovery, Burst 1, while others
are brighter toward lower frequencies. The spectra of Bursts 8 and 10 are not monotonic. The
detections of Bursts 6–11 exclusively in Beam 0 of the ALFA receiver (see Extended Data Table
1) means that the bursts must have been detected in the main beam and not in a side-lobe. While
the frequency-dependent shape of the main beam attenuates the bursts’ intrinsic spectra at higher
frequencies if the source is off-axis
4
, this bias is either not large enough or in the wrong direction
to cause the observed spectral variability of Bursts 6–11. Given our improved position, Burst 1
is consistent with its detection in a side-lobe, which, unlike in the main beam, could have caused
attenuation of the spectrum at lower frequencies. This spectral volatility is reflected by the wide
range of spectral indices α ∼ −10 to +14 obtained from fitting a power-law model (S
ν
∝ ν
α
,
where S
ν
is the flux density at frequency ν) to burst spectra (Table 1).
There is no evidence for fine-scale diffractive interstellar scintillation, most likely because it
is unresolved by our limited spectral resolution. In principle, the spectra could be strongly modu-
lated if the source is multiply imaged by refraction in the interstellar medium
18
or by gravitational
lensing. However, the splitting angle between sub-images required to produce spectral structure
across our band ( 1 milli-arcsecond) is much smaller than the expected diffraction angle from
interstellar plasma scattering. The fine-scaled diffraction structure in the spectrum will therefore
wash out the oscillation. Lastly, positive spectral indices could also be explained by free-free ab-
sorption at the source
19
, but this is ruled out by the large spectral differences among bursts. We
therefore conclude that the spectral shapes and variations are likely to be predominantly intrinsic
to the source.
An analysis of the arrival times of the bursts did not reveal any statistically significant pe-
riodicity (see Methods). If the source has a long period (& 1 s), then it is likely emitting at a
wide range of rotational phases, which is not uncommon for magnetars
20
, making a convincing
period determination difficult. Due to the small number of detected bursts, we are not sensitive to
periodicities much shorter than ∼ 100 ms.
Repeat bursts rule out models involving cataclysmic events – such as merging neutron stars
21
or collapsing supra-massive neutron stars
10
. Bursts from Galactic flare stars have been proposed
as a model for FRBs with the DM excess originating in the stellar corona
22
. However, temporal
density variations in the corona should produce bursts with varying DMs, which we do not observe.
4
Planets orbiting in a magnetised pulsar wind may produce a millisecond-duration burst once per
orbital period
23
; however, the observed intra-session separations of our bursts (23 to 572 s) are
too short to correspond to orbital periods. Repeated powerful radiative bursts are associated with
magnetars, and indeed giant flares from the latter have been suggested as a FRB source
12, 19, 24
.
However, no Galactic magnetar has been seen to emit more than a single giant flare in over four
decades of monitoring, arguing against a magnetar giant flare origin for FRB 121102. Magnetars
have been observed to exhibit repeating bright radio pulses
20
, however not yet at the energy scale
implied if FRB 121102 is more than several hundred kpc away.
Giant pulse emission from an extragalactic pulsar remains a plausible model
11
. The most
prominent giant pulses are from the Crab pulsar, which has a large spin-down energy loss rate.
Spectral indices calculated from wideband measurements of giant pulses from the Crab pulsar
25
have a broad distribution ranging from α ∼ −15 to +10, as well as frequency “fringes” – i.e.
a banded structure to the emission brightness as a function of frequency
26
. These fringes have
characteristic widths of a few hundred MHz, and we speculate that – given our 322-MHz ob-
serving bandwidth – a similar phenomenon could create the spectral variability we have seen in
FRB 121102. The double-peaked nature of some FRB 121102 bursts is also possible in the giant
pulse model
7
, and the evolution of these burst morphologies with frequency could imply rapid
spectral variation between consecutive (sub-)pulses only milliseconds apart.
The low Galactic latitude and relatively small β
DM
of FRB 121102 compared with other
FRBs raises the question whether it is genuinely extragalactic in origin (see also Methods). How-
ever, no Hα or HII regions are seen in archival data along the line of sight to FRB 121102, as might
be expected for an intervening ionised nebula
4
that can give β
DM
1. Furthermore, a detailed
multi-wavelength investigation, which searched for a compact nebula in a sky region that includes
the refined position presented here, concluded that FRB 121102’s high DM cannot be explained
by unmodelled Galactic structure along the line of sight and that FRB 121102 must therefore be
extragalactic
19
. Conclusively establishing that FRB 121102 is extragalactic will require arcsecond
localisation and association with a host galaxy. The repeating nature of the bursts facilitates such
localisation with a radio interferometer.
While the FRB 121102 bursts share many similarities to the FRBs detected using the Parkes
1–3, 5–7
and Green Bank
8
telescopes, it is unclear whether FRB 121102 is representative of all FRBs. The
10 bursts from FRB 121102 in 2015 were detected near the best-known position in 3 hrs of obser-
vations. In contrast, follow-up observations of the Parkes FRBs, again using the Parkes telescope,
range in total time per direction from a few hours
9
to almost 100 hours
1
and have found no ad-
ditional bursts. Arecibo’s > 10× higher sensitivity may allow detection of a broader range of
the burst-energy distribution of FRBs, thus increasing the chances of detecting repeated bursts;
for example, of the 11 bursts from FRB 121102, Parkes may have been capable of detecting only
Bursts 8 or 11. More sensitive observations of the Parkes FRBs may therefore show that they also
sporadically repeat.
Alternatively, FRB 121102 may be fundamentally different from the FRBs detected at Parkes
and Green Bank. As was the case for supernovae and gamma-ray bursts, multiple astrophysical
5
