SEARCH FOR POINT SOURCES OF HIGH-ENERGY NEUTRINOS WITH AMANDA
J. Ahrens,
1
X. Bai,
2
G. Barouch,
3
S. W. Barwick,
4
R. C. Bay,
5
T. Becka,
1
K.H. Becker,
6
D. Bertrand,
7
F. Binon,
7
A. Biron,
8
S. Boeser,
8
O. Botner,
9
A. Bouchta,
8,10
O. Bouhali,
7
T. Burgess,
11
S. Carius,
12
T. Castermans,
13
D. Chirkin,
5,6
J. Conrad,
9
J. Cooley,
3
D. F. Cowen,
14
A. Davour,
9
C. De Clercq,
15
T. DeYoung,
3
P. Desiati,
3
J.P. Dewulf,
7
P. Doksus,
3
J. Edsjo
¨
,
11
P. Ekstro
¨
m,
11
T. Feser,
1
T. K. Gaisser,
2
M. Gaug,
8
L. Gerhardt,
4
A. Goldschmidt,
16
A. Hallgren,
9
F. Halzen,
3
K. Hanson,
3
R. Hardtke,
3
T. Hauschildt,
8
M. Hellwig,
1
P. Herquet,
13
G. C. Hill,
3
P. O. Hulth,
11
K. Hultqvist,
11
S. Hundertmark,
4
J. Jacobsen,
16
A. Karle,
3
J. Kim,
4
L. Ko
¨
pke,
1
M. Kowalski,
8
K. Kuehn,
4
J. I. Lamoureux,
16
H. Leich,
8
M. Leuthold,
8
P. Lindahl,
12
J. Madsen,
17
P. Marciniewski,
9
H. Matis,
16
C. P. McParland,
16
T. C. Miller,
2,18
Y. Minaeva,
11
P. Miocinovic
´
,
5
P. C. Mock,
4,19
R. Morse,
3
T. Neunho
¨
ffer,
1
P. Niessen,
15
D. R. Nygren,
16
H. O
¨
gelman,
3
P. Olbrechts,
15
C. Pe
´
rez de los Heros,
9
A. C. Pohl,
12
P. B. Price,
5
G. T. Przybylski,
4
K. Rawlins,
3
E. Resconi,
8
W. Rhode,
6
M. Ribordy,
8
S. Richter,
3
J. Rodrı
´
guez Martino,
11
P. Romenesko,
3
D. Ross,
4
H.G. Sander,
1
T. Schmidt,
8
D. Schneider,
3
R. Schwarz,
3
A. Silvestri,
4
M. Solarz,
5
G. M. Spiczak,
17
C. Spiering,
8
D. Steele,
3
P. Steffen,
8
R. G. Stokstad,
16
K.H. Sulanke,
8
I. Taboada,
20
L. Thollander,
11
S. Tilav,
2
C. Walck,
11
C. Weinheimer,
1
C. H. Wiebusch,
8,10
C. Wiedemann,
11
R. Wischnewski,
8
H. Wissing,
8
K. Woschnagg,
5
W. Wu,
4
G. Yodh,
4
and S. Young
4
(The AMANDA Collaboration)
Received 2002 July 26; accepted 2002 October 2
ABSTRACT
This paper describes the search for astronomical sources of high-energy neutrinos using the
AMANDA-B10 detector, an array of 302 photomultiplier tubes used for the detection of Cerenkov light
from upward-traveling neutrino-induced muons, buried deep in ice at the South Pole. The absolute pointing
accuracy and angular resolution were studied by using coincident events between the AMANDA detector
and two independent telescopes on the surface, the GASP air Cerenkov telescope and the SPASE extensive
air shower array. Using data collected from 1997 April to October (130.1 days of live time), a general survey
of the northern hemisphere revealed no statistically significant excess of events from any direction. The
sensitivity for a flux of muon neutrinos is based on the effective detection area for through-going muons.
Averaged over the northern sky, the effective detection area exceeds 10,000 m
2
for
E
l
?
10 TeV. Neutrinos
generated in the atmosphere by cosmic-ray interactions were used to verify the predicted performance of the
detector. For a source with a differential energy spectrum proportional to
E
?
2
?
and declination larger than
+40
?
, we obtain
E
2
ð
dN
?
=
dE
Þ?
10
?
6
GeV cm
?
2
s
?
1
for an energy threshold of 10 GeV.
Subject headings:
neutrinos — surveys
On-line material:
color figures
1
Institute of Physics, University of Mainz, Staudinger Weg 7, D-55099 Mainz, Germany.
2
Bartol Research Institute, University of Delaware, Newark, DE 19716.
3
Department of Physics, University of Wisconsin, Madison, WI 53706.
4
Department of Physics and Astronomy, University of California, Irvine, CA 92697.
5
Department of Physics, University of California, Berkeley, CA 94720.
6
Fachbereich 8 Physik, BUGH Wuppertal, D-42097 Wuppertal, Germany.
7
Universite
´
Libre de Bruxelles, Science Faculty CP 230, Boulevard du Triomphe, B-1050 Brussels, Belgium.
8
DESY-Zeuthen, D-15735 Zeuthen, Germany.
9
Division of High Energy Physics, Uppsala University, S-75121 Uppsala, Sweden.
10
Current address: CERN, CH-1211 Geneva 23, Switzerland.
11
Fysikum, Stockholm University, S-11385 Stockholm, Sweden.
12
Department of Technology, University of Kalmar, S-39182 Kalmar, Sweden.
13
University of Mons-Hainaut, Mons, Belgium.
14
Department of Physics, Pennsylvania State University, University Park, PA 16802.
15
Vrije Universiteit Brussel, Dienst ELEM, B-1050 Brussels, Belgium.
16
Lawrence Berkeley National Laboratory, Berkeley, CA 94720.
17
Department of Physics, University of Wisconsin, River Falls, WI 54022.
18
Current address: Johns Hopkins University, Applied Physics Laboratory, Laurel, MD 20723.
19
Current address: Optical Networks Research, JDS Uniphase, 100 Willowbrook Road, Freehold, NJ 07728-2879.
20
Departamento Fı
´
sica, University Simo
´
n Bolı
´
var, Caracas, Venezuela.
The Astrophysical Journal
, 583:1040–1057, 2003 February 1
#
2003. The American Astronomical Society. All rights reserved. Printed in U.S.A.
E
1040
1. INTRODUCTION
Nature provides precious few information carriers from
the deep recesses of space, and it is imperative to develop
techniques to exploit each one. Throughout history, the
photon messenger has made vital contributions to the
understanding of the observable universe. In this paper, we
present results from a new generation of telescopes designed
to detect a very different kind of information carrier, high-
energy neutrinos (where
E
?
>
1 TeV). The search for
astronomical sources of high-energy neutrinos is one of the
central missions of the Antarctic Muon and Neutrino
Detector Array (AMANDA; Wischnewski et al. 1999). In
this paper, we describe a general search for continuous emis-
sion from a spatially localized direction in the northern
sky,
21
restricted to declinations greater than +5
?
. The search
technique is conceptually simple: a point source is identified
by a statistically significant enhancement over expected
background fluctuations from a particular direction.
Expected background is readily obtained experimentally
from off-source sky bins within the same band of declina-
tion. In contrast, unresolved, or diffuse, signals are charac-
terized by an isotropic distribution, and backgrounds are
estimated by detector simulation programs. The most favor-
able flux predictions for point sources are several orders of
magnitude lower than the most optimistic predictions for
diffusely distributed sources. However, atmospheric neu-
trino background is diffusely distributed as well, so the level
of intrinsic background in the diffuse search is also several
orders of magnitude higher. While signal-to-noise ratio con-
siderations favor the search for diffuse emission over point-
source searches, the interpretation of a diffusely distributed
signal is more ambiguous. Thus, the search for point sources
complements the search for diffuse sources. The latter
search is described in Hill & Leuthold (2001).
22
The more
specific searches for point emission from gamma-ray
bursters (Hardtke & Barouch 2001)
23
and quasi-pointlike
emission from Galactic dark matter trapped in the core of
the Earth (Ahrens et al. 2002b) are presented in separate
papers since these analyses were optimized for different flux
spectra and different background characteristics.
2. MOTIVATION
The origin of cosmic rays is one of the oldest puzzles in
particle astrophysics. Shocks from Galactic supernovae are
widely believed to accelerate cosmic rays to
?
10
15
eV, while
the sources of cosmic rays at the most extreme energies are
not known. Plausible models of particle acceleration exist
for many classes of Galactic and extragalactic objects, but
supporting evidence for any model is largely circumstantial.
The observation of high-energy neutrinos from point
sources would unequivocally confirm the hadronic nature
of such accelerators. Unfortunately, the predicted neutrino
fluxes from Galactic and extragalactic point sources are
too low to be detected with AMANDA-B10, although
uncertainties in the model parameters lead to considerable
variation in the flux predictions.
Supernova remnants (SNRs) are one of the few classes of
Galactic sites that have the capability to supply sufficient
power to accelerate Galactic cosmic rays. The diffusive
shock mechanism naturally produces a power-law spectrum
of
dN
=
dE
/
E
?
2
:
1
, which is consistent with the deduced
spectral index of cosmic rays.
Recent observations of TeV gamma rays from plerions
such as the Crab Nebula and SNRs provide direct evidence
for particle acceleration to high energies. However, these
observations do not provide compelling evidence for
hadronic
acceleration because of an unfortunate ambiguity:
it is possible (and even probable) that electrons are solely
responsible for the high-energy gamma-ray production. But
if SNRs are the accelerators of Galactic cosmic rays, they
must also accelerate hadrons. A class of models exploits this
idea by suggesting that both protons and electrons are accel-
erated by the supernova shock. Pions, both neutral and
charged, are produced in the nuclear collisions between pro-
tons and ambient material (a cosmic equivalent of a ‘‘ beam
dump ’’ commonly used by terrestrial accelerators) and then
decay to high-energy gamma rays and neutrinos.
While the notion of particle acceleration by supernova
shocks provides a credible and largely consistent picture,
not all observations neatly fit this scheme. Alternative sites
for cosmic-ray acceleration may emerge from a detailed
study of the neutrino sky. For example, Galactic microqua-
sars, a subclass of X-ray binary systems that exhibit relativ-
istic radio jets, have been identified as possible sources of
high-energy neutrino emission (Levinson & Waxman 2001)
and potential sources of the highest energy cosmic rays. If
they accelerate cosmic rays to high energies, then their dense
environment creates suitable conditions for an efficient
beam dump.
Turning to extragalactic sources, active galactic nuclei
(AGNs) are among the most luminous objects in the uni-
verse and promising sources of neutrinos. In these models,
high-energy neutrino fluxes are generated near the central
engine or in the jets of radio-loud AGNs (e.g., blazars, a
class of objects where the jet intersects the line of sight of the
observer). The fact that gamma-ray emission has been
detected (Cantanese & Weekes 1999) from nearby blazars
Mrk 421 and Mrk 501 provides strong evidence for particle
acceleration to high energies.
24
The time-averaged energy
spectrum from Mrk 501 during 1997 is consistent with an
unbroken power-law energy spectrum up to 10 TeV
(Weekes 2001; Konopelko et al. 1999). Beam dump models
of neutrino production predict comparable fluxes of
gamma-ray photons and neutrinos. However, gamma-ray
photons at TeV energies may interact with material or pho-
ton fields in the source or interact with the diffuse infrared
background photons during their flight. Because of this
reprocessing, the measured energy spectrum for gamma-ray
photons may not trace the energy spectrum of the source.
Consequently, it is possible for the ratio of neutrino flux to
gamma-ray flux from a given source to exceed unity. Con-
straints on this ratio are discussed in
x
7.
Recently, it has been argued (Buckley et al. 1998a, 1998b)
that the rapid time variability of high-energy photon
21
Although the flux limits reported in this paper are computed assuming
continuous emission, upper bounds could be generated for periodic or
episodic emission as well.
22
Available at http://area51.berkeley.edu/manuscripts/
20010604xx-diffuse-sens.pdf.
23
Available at http://area51.berkeley.edu/manuscripts/
20010606xx-grbdoc.pdf.
24
The review by Cantanese & Weekes (1999) presents a current list of
detected very high energy gamma-ray sources.
SEARCH FOR HIGH-ENERGY NEUTRINO SOURCES 1041
emission from AGN blazars and the correlated variation
between the X-ray and TeV regimes disfavor hadronic
acceleration models for this particular class of objects, but
others have shown that rapid and correlated variability can
be accommodated by modest extensions to the existing
hadronic acceleration models (Rachen 2000; Dermer 1999).
The vigorous debate suggests that high-energy neutrino
detectors can play a central role in deciphering the accelera-
tion mechanism.
Figure 1 provides a survey of model predictions for
fluxes of high-energy neutrinos. The models were selected
to highlight the variation in spectral characteristics. The
line labeled 3C 273 is representative of several recent neu-
trino flux predictions, e.g., a similar differential flux is
predicted for microquasars (Distefano et al. 2002) in the
region of sky visible to AMANDA and the flat-spectrum
radio quasar 3C 279 (Dermer & Atoyan 2001). The figure
also shows the differential neutrino flux limit for an
assumed source spectrum proportional to
E
?
2
for
AMANDA-B10 and the anticipated corresponding sensi-
tivity of AMANDA-II and IceCube (Karle et al. 2003).
The AMANDA-B10 result, the subject of this paper, is
valid for declinations greater than +40
?
. Many theoreti-
cal models of potential astronomical neutrino sources
predict a very hard energy spectrum, approximately
E
?
2
(Learned & Mannheim 2000), which leads to a most
probable energy for a detected neutrino well above 1
TeV (typically 10–30 TeV). This high energy is a conse-
quence of three facts: the neutrino cross section for weak
interactions increases with neutrino energy, and the prop-
agation length of the secondary muon increases and the
effective detection area increases as light emission along
the muon track increases with energy.
Even though the cosmic-ray puzzle provides a powerful
motivation to explore the sky for neutrino emission, not all
high-energy neutrino sources need to contribute to the
cosmic-ray flux. In particular, a powerful accelerator may
be surrounded by too much material to emit high-energy
photons or cosmic rays (they would interact and cascade
down to lower energies), but this accelerator could be dis-
covered by exploiting the neutrino messenger. Several
models of such ‘‘ hidden ’’ sources have appeared in the liter-
ature. For example, the predicted flux of neutrinos from
pre-AGN objects (Berezinsky & Dokuchaev 2001) leads to
a muon detection rate of
?
10 yr
?
1
km
?
2
.
3. DESCRIPTION OF THE AMANDA DETECTOR
The AMANDA telescope is located below the surface of
the Antarctic ice sheet at the geographic South Pole. The
neutrino detection technique relies on the detection of Cer-
enkov light from upward-traveling neutrino-induced
muons. Figure 2 shows the current configuration of the
AMANDA detector. The shallow array, AMANDA-A,
was deployed to depths between 800 and 1000 m in an
exploratory phase of the project. The deeper array of 10
strings, referred to as AMANDA-B10, was deployed during
the austral summers between 1995 and 1997, to depths
between 1500 and 2000 m. At this depth, the optical proper-
ties are suitable for track reconstruction (Woschnagg et al.
10
10
10
8
10
6
10
4
10
2
10
3
10
4
10
5
10
6
E
2
(dN/dE) [GeVcm
2
s
1
]
E
ν
[GeV]
AMANDAII (1 yr l.t.)
AMANDAB10 (this work)
IceCube
3C273
Crab
AGN Core
Mkn 501 (
ν
=
γ
)
Atm.
ν
Fig.
1.
—Representative survey of
?
þ
?
flux predictions from cosmic
accelerators of high-energy neutrinos. The AMANDA-B10 result is pre-
sented here. The dashed horizontal lines give preliminary estimates of the
minimum detectable flux by AMANDA-II after 1 yr of live time (Barwick
et al. 2001) and IceCube (Spiering 2001). The atmospheric neutrino fluxes
(Agrawal et al. 1996) are appropriate for a circular patch of 1
?
(
lower curve
)
and 3
?
radius. The curves do not include the normalization uncertainty,
possibly 30% in magnitude (Gaisser 2002; T. K. Gaisser 2002, private com-
munication). Models: 3C 273 (Nellen, Mannheim, & Biermann 1993), Crab
model I (Bednarek & Protheroe 1999), AGN core (Stecker & Salamon
1996), and Mrk 501 assuming neutrino spectrum is identical to observed
gamma spectrum during flaring phase (Weekes 2001). [
See the electronic
edition of the Journal for a color version of this figure.
]
120 m
snow layer
optical module (OM)
housing
pressur
e
Optical
Module
silicon gel
HV divider
light diffuser bal
l
60 m
AMANDA as of 2000
zoomed in on one
(true scaling)
200 m
Eiffel Tower as comparison
Depth
surface
50 m
1000 m
2350 m
2000 m
1500 m
810 m
1150 m
AMANDAA (top)
zoomed in on
AMANDAB10 (bottom)
AMANDAA
AMANDAB10
main cable
PMT
Fig.
2.
—Schematic view of the AMANDA neutrino telescope. This
paper describes an analysis of data taken in 1997 with AMANDA-B10, the
10 inner strings shown in the expanded view in the center. Each dot
represents one optical module in the array. [
See the electronic edition of the
Journal for a color version of this figure.
]
1042 AHRENS ET AL. Vol. 583
1999).
25
The strings are arranged in a circular pattern when
viewed from the surface. The instrumented volume of
AMANDA-B10 forms a vertical cylinder with a diameter of
120 m. Most electronics are housed on the surface in a
research facility located within a kilometer of the Amundsen-
Scott South Pole Station. The detector was commissioned in
1997 February (Wischnewski et al. 1999; Barwick et al. 2000)
and expanded by adding nine strings of optical modules
(OMs) between 1997 December and 2000 January. The com-
posite array of 19 strings forms the AMANDA-II array,
which was commissioned in 2000 February.
AMANDA-B10 consists of 302 OMs arranged on 10 ver-
tical strings. Each OM contains an 8 inch (21 cm) diameter
photomultiplier tube (PMT) controlled by passive elec-
tronics and housed in a glass pressure vessel. The OMs are
connected to the surface by dedicated electrical cables,
which supply high voltage and carry the anode signals from
the PMTs. For each event, the amplitudes and arrival times
of the pulses from the OMs are digitized by peak analog-to-
digital converter and time-to-digital converter (TDC) val-
ues. The TDCs are capable of measuring eight distinct
pulses per channel. The precision of the arrival time mea-
surement is 5 ns. Details of deployment, timing resolution,
and detector operation can be found in Andres et al. (2000,
2001). Readout of the entire array was triggered by a major-
ity logic system, which demanded that at least 16 OMs pro-
duce signals, or ‘‘ hits,’’ within a time window of 2.2
l
s. This
window takes into account the rather large time variation
introduced by the large geometric size of the detector and
the cable propagation delays. Random signals from the
OMs (or ‘‘ noise ’’) were observed at a rate of 300 Hz on the
inner four strings and 1.5 kHz for OMs on the outer six
strings, the difference being due to different levels of radio-
active potassium in the glass pressure vessels. On average,
random noise contributed one count per event to the major-
ity trigger.
Optical absorption and scattering properties of the glacial
ice that encapsulates the AMANDA detector have been
studied using light sources buried with the strings and
Cerenkov light from atmospheric muons. These studies
(Woschnagg et al. 1999) confirm that the ice is not homoge-
neous but consists of horizontal strata correlated with cli-
matological events in the past, such as ice ages.
26
Variations
in the concentration of insoluble impurities between the
strata produce a strong modulation of the optical proper-
ties. The absorption length, averaged over depth within the
AMANDA-B10 array, is 110 m at a wavelength of 400 nm,
and the average effective scattering length is approximately
20 m.
The detection of neutrinos relies on the observation of
Cerenkov photons generated by muons created in charged-
current interactions. At the energies of interest, muons typi-
cally propagate for distances in excess of several kilometers
(e.g., a muon with
E
¼
10 TeV will travel 8 km in water).
Therefore, neutrino interactions outside the instrumented
volume can be inferred by the presence of a muon, providing
a method to extend the volume of the detector beyond the
instrumented boundary of the array. The average angle
between the muon direction and the parent neutrino direc-
tion,
h
?
?
l
i
, is approximately 0
=
65
=
ð
E
?
=
TeV
Þ
0
:
48
for
E
?
less
than 100 TeV (Oppelt 2001).
27
However, nearly independ-
ent of the muon energy, the precision of the measured muon
direction in AMANDA-B10 is approximately 4
?
(see
x
6),
which dominates the angular uncertainty in the neutrino
direction.
The AMANDA-B10 data analyzed here were collected
between 1997 April and October. Once construction was
completed in 1997 February, calibration and data manage-
ment activities continued until April. Operations ceased
between 1997 late October and 1998 February, because of
the beginning of construction of the AMANDA-II array.
Furthermore, limitations in the data acquisition and archiv-
ing system during that first year of operation reduced the
total live time to approximately 130.1 days.
4. SIMULATIONS
Astronomical signals are unlikely to produce more than a
few tens of upgoing neutrino events per year in AMANDA-
B10. Data are therefore overwhelmingly dominated by two
types of background: downgoing atmospheric muons gener-
ate essentially all of the recorded events, and atmospheric
neutrinos contribute a few tens of events per day. The point-
source search relies on a good understanding of both signal
and background through simulations based on Monte
Carlo techniques.
Atmospheric muon events are generated from the mea-
sured flux of cosmic rays (Wiebel-Sooth & Biermann 1998).
Two different air shower simulation packages were used to
assess systematic uncertainty: BASIEV (Bosiev et al. 1989)
and CORSIKA (Version 5.6; Heck et al. 1998, 1999) using
the QGSJET hadronic interaction model. CORSIKA was
modified to include the curved geometry of the Earth and
atmosphere to provide a more accurate description of the
flux at zenith angles close to the horizon. Most characteris-
tics of the events generated with BASIEV were found to be
similar to those from the more accurate but computation-
ally more intensive CORSIKA simulation. The density pro-
file of the atmosphere was modified for polar conditions,
but no attempt was made to replicate the small seasonal var-
iations of the trigger rate (Bouchta et al. 1999).
28
Muon
tracking from the surface to the detector was handled by the
muon propagation program MUDEDX (Lohmann, Kopp,
& Voss 1985; W. Lohmann, R. Kopp, & R. Voss 1995, pri-
vate communication [MUDEDX Version 2.02]), and the
energy-loss characteristics were compared against two addi-
tional propagation programs that are available for general
use: PROPMU (Lipari & Stanev 1991) and MMC (Rhode
& Chirkin 2001). Integral lateral distributions of muons at
the depth of AMANDA were simulated for proton and iron
showers (X. Bai et al. 2003, in preparation) and used for ver-
ification of the detector performance as described below.
The propagation of upward-traveling muons from neu-
trino interactions was treated differently from that of down-
going atmospheric muons because the energies of signal
neutrinos were expected to extend to much higher energies.
25
Available at http://krusty.physics.utah.edu/~icrc1999/root/vol2/
h4_1_15.pdf.
26
See P. B. Price, K. Woschnagg, & D. Chirkin (2000), AMANDA
public manuscript 20000201, available at http://amanda.berkeley.edu/
manuscipts.
27
Available at http://www-zeuthen.desy.de/~apohl/files/doktor/ps.gz.
28
Available at http://krusty.physics.utah.edu/~icrc1999/root/vol2/
h3_2_11.pdf.
No. 2, 2003 SEARCH FOR HIGH-ENERGY NEUTRINO SOURCES 1043
Neutrinos were tracked through the Earth and allowed to
interact in the ice within or near the instrumented volume of
the detector or in the bedrock below (Hill 1997). Muons
with energies above 10
5
:
5
GeV at production were propa-
gated using PROPMU until they reached the rock-ice boun-
dary and then propagated through the ice in exactly the
same way as downgoing atmospheric muons. For energies
below 10
5
:
5
GeV at the production vertex, muons were
tracked with MUDEDX. The background fluxes from at-
mospheric
?
l
and
?
l
(Agrawal et al. 1996) were included
(AMANDA cannot differentiate the charge sign of the
muon).
In addition to background from atmospheric muons and
muon neutrinos, the detection efficiency for atmospheric
electron neutrinos has been simulated. At the relevant ener-
gies, the flux of
?
e
is far smaller than for
?
l
, so the back-
ground contribution is small a priori. Furthermore, the
topology of electron neutrino events, reflecting the electro-
magnetic shower generated by the secondary electron, is
spherical rather than linear, and this characteristic has been
exploited to further increase the rejection. Simulations show
that the detection rate of atmospheric
?
e
in the point-source
analysis is only 0.3% of
?
l
(Young 2001)
29
and therefore
negligible in this context. We note that a separate analysis
was devised to search for electron neutrinos and conse-
quently achieved a much larger sensitivity (Ahrens et al.
2002c).
An overview of the simulation of the detector response is
given in Hundertmark (1999). The depth dependence of the
optical properties of the bulk ice (Woschnagg et al. 1999)
30
is included along with a realistic treatment of trigger condi-
tions, intrinsic noise rates, and hardware-dependent pulse
shapes. The linearity and saturation of the photomultiplier
response is included. The angle-dependent sensitivity of the
optical module was computed from a convolution of PMT
quantum and collection efficiencies as estimated by the man-
ufacturer, detection efficiency, wavelength-dependent trans-
parency through the pressure housing and buffer gel, and
obscuration by cables and mechanical support hardware.
The relative angular dependence of the sensitivity of the
photomultiplier tubes was measured in the laboratory. The
local optical properties of the refrozen ice were included in
the photon tables that describe the probability and the tim-
ing characteristics of photon propagation (Ahrens et al.
2002a).
Figure 3 presents the differential distribution of the multi-
plicity of optical modules, or channels, participating in an
event,
N
ch
, for the full detector simulation and for experi-
mental data after known detector-related artifacts were
removed. The integrated rates differ by less than 25%, which
is within the systematic uncertainties associated with the
flux of the primary cosmic rays (Gaisser 2002; T. K. Gaisser
2002, private communication) and uncertainties associated
with the absolute sensitivity of the optical module (see
x
8).
The agreement in the shape of the distribution demonstrates
a good understanding of the overall sensitivity of the array
for the most common events that trigger AMANDA-B10.
As the inset of Figure 3 shows, the largest values of
N
ch
are produced by events with more than one muon. This
information provides indirect evidence that the response of
AMANDA to single high-energy muons is correctly mod-
eled, by the following argument. Multimuon bundles that
reach AMANDA mainly consist of muons below the critical
energy of 600 GeV, which implies that energy loss due to
ionization is near minimum. The Cerenkov light production
from muons well above the critical energy is dominated by
electromagnetic showers, and the total light from a muon
with
E
l
is approximately equal to the light of a bundle of
N
muons, where
N
?
E
l
=
E
crit
. Therefore, the light production
by a multimuon event can be related to the light production
by a single-muon event. For example, the average energy
loss per unit length for a muon with energy 10
13
eV is
approximately a factor of 15 larger than for a single muon,
as long as the energy is below 600 GeV. There are several
modest limitations to this line of reasoning. One is that the
lateral distribution of multimuon events generates Ceren-
kov photons over a much larger cylinder than a single
muon. Another difference is that multimuon events deposit
Cerenkov photons more uniformly than the equivalent
high-energy muon, for which energy loss is dominated by
occasional pair production and bremsstrahlung. However,
optical scattering by the ice mitigates the effects of nonuni-
form photon generation. Simulations show that bundles of
20 muons generate an
N
ch
distribution similar to that of
single muons with an energy of 10 TeV. The correlation
between muon multiplicity and
N
ch
multiplicity is shown in
Figure 4. Note that bias introduced by the majority logic in
the trigger prevents the muon multiplicity from converging
to zero as
N
ch
approaches zero.
29
Available at http://area51.berkeley.edu/manuscripts/
20010702xx-young-main.pdf.
30
Available at http://krusty.physics.utah.edu/~icrc1999/
proceedings.html.
N
ch
Differential Rate [Hz]
experiment (99 Hz)
simulation (77 Hz)
N
ch
Relative Frequency
single
μ
multi
μ
10
3
10
2
10
1
1
0 20 40 60 80 100 120 140
10
5
10
4
10
3
10
2
10
1
0 50 100
Fig.
3.
—Differential distribution of observed (
solid curve
) and predicted
(
dashed curve
) trigger rates as a function of the event multiplicity
N
ch
(i.e.,
the number of optical modules that participate in each event). The inte-
grated rates are given in parentheses. Note that
N
ch
extends below the
majority logic threshold of 16 because of removal of data caused by experi-
mental artifacts.
Inset
: Relative contribution to the trigger rates from single
muons (
solid curve
) and multiple-muon bundles (
dashed curve
) that traverse
the fiducial volume of the array. [
See the electronic edition of the Journal for
a color version of this figure.
]
1044 AHRENS ET AL. Vol. 583
5. ANALYSIS PROCEDURE
The analysis procedure exploits two essential characteris-
tics of the signal to simplify the analysis relative to atmo-
spheric neutrino measurements. First, the sources are
assumed to be point sources in the sky, so only events within
a restricted angular region are considered. Second, we use
the topological and directional characteristics of the spec-
trally hard neutrino signal to help reject poorly recon-
structed atmospheric muons (i.e., downward-traveling
muons reconstructed as upward traveling) and atmospheric
neutrinos, both of which have softer spectra. Unlike many
neutrino detectors, the effective sensitivity of AMANDA
varies dramatically as a function of the background-
rejection requirements. By concentrating on harder spectra,
the effective area of the detector can be increased by relaxing
the background-rejection criteria. Since the point-source
analysis tolerates a larger background (
B
) contamination in
the final data sample, the analysis procedure optimizes on
signal to noise (
S
=
ffiffiffiffi
B
p
) rather than signal purity (
S
=
B
).
Prior to track reconstruction and event selection, experi-
mental data were selected from runs that exhibited no
abnormal behavior, and various instrumentally induced
artifacts were removed. Once the data in the runs were certi-
fied, individual OMs in the array were examined to insure
proper operation. OM channels with hardware malfunction
(
?
15% of OMs), such as pickup from unusually large exter-
nal noise sources or fluctuations in the response of the
amplifier electronics, were rejected. Approximately 85% of
OMs remained after deselection. Occasional signals induced
by cross talk in the electrical cables or surface electronics
exhibited characteristic behavior and could be removed by
straightforward restrictions on pulse amplitude and width.
Noise signals generated internally by the photomultiplier
tubes were readily removed if their time of arrival occurred
earlier than 5 ms prior to the formation of the event trigger.
The reconstruction programs stochastically account for
PMT noise within the event duration.
After this initial data cleaning, a number of event recon-
struction techniques (Andres et al. 2000) are applied to the
data. The most sophisticated technique relies on a search in
multiparameter space to find the maximum likelihood for a
track hypothesis given the recorded hits. After reconstruc-
tion is completed, events are selected according to a set of
criteria that retain only the highest quality events that pos-
sess topological and directional information consistent with
those expected for upgoing neutrino-induced muons. In a
first step, the data sample of 1
:
05
?
10
9
events at trigger
level is reduced to a more manageable size by two filtering
stages. Most events in data are readily identified as due to
downward-traveling muons by computationally fast recon-
struction routines. Removing these events reduces the data
approximately by a factor of 10
3
.
An iterative analysis procedure was developed to maxi-
mize
S
=
ffiffiffiffi
B
p
for a simulated signal with an energy spectrum
proportional to
E
?
2
. It ignored the absolute time of the
event, which helped to minimize bias from potential sources
in the data. In this optimization the background was deter-
mined from experimental data by assuming that the fraction
of signal events in the data sample is negligible. After the fil-
tering stages, cuts were applied sequentially on a set of selec-
tion variables, with several variables included more than
once. The specific value for each cut after stage 2 was chosen
to retain
e
80% of the signal. At each stage, given this
constraint on signal efficiency, the same cut was made on
data for the variable with the largest rejection power,
R
¼
?
sig
=?
bgr
, where
?
¼
N
pass
=
N
0
, and
N
0
and
N
pass
are the
numbers of events before and after the application of the
selection cut, respectively. The signal-to-noise ratio was
then computed as a function of zenith angle to ensure that
the acceptance of AMANDA-B10 remained as large as pos-
sible near the horizon. The effective areas for detection of
background and signal needed for this computation were
determined from simulations (as described in
x
7). After
each stage, this procedure was repeated on the remaining
variables.
Besides restrictions on the reconstructed zenith angle
h
,
the most effective selection criteria impose a threshold on
the number
N
dir
of only slightly scattered, or ‘‘ direct,’’ pho-
tons (i.e., photons that travel between the reconstructed
track and the OM in nearly a straight line) and the track
length
L
dir
over which these photons are detected. Further-
more, the analysis requires a minimum goodness-of-fit value
from the maximum likelihood procedure. Other criteria
evaluate the topological distribution of the photon emission
using variables that describe the granularity of the light pat-
tern along the trajectory and a related observable that
assesses the sphericity of the photon pattern.
31
Table 1
shows the selection variables and cuts used in this analysis,
including a brief technical description of the two filtering
stages. The selection variables were introduced previously
(Ahrens et al. 2002a), and a complete description is also
available (Young 2001). Also shown in the table are efficien-
cies and rejection factors at each stage of the analysis for
experimental data, simulated background, and simulated
signal, averaged over all angles.
31
Muons normally generate a linear distribution of Cerenkov
photons, whereas electromagnetic cascades initiated by pair production or
bremsstrahlung produce spherically symmetric distributions.
N
ch
N
μ
0
5
10
15
20
25
30
0 20 40 60 80 100 120
Fig.
4.
—Muon multiplicity
N
l
vs. OM multiplicity
N
ch
from a full
detector simulation. The area of the boxes is linearly proportional to num-
ber of events. The average values (
dots
) show the correlation between these
two quantities, and the vertical error bars show the statistical uncertainty.
No. 2, 2003 SEARCH FOR HIGH-ENERGY NEUTRINO SOURCES 1045
The simulated background from atmospheric muons and
neutrinos is compared to data at all stages of the analysis to
establish confidence in the simulation. Figure 5 shows the
comparison at stage 4, which is sufficiently early in the anal-
ysis to retain high statistical precision. All distributions are
equally normalized because the optimized cut value depends
on the fraction of events that remain (
bottom panel
).
Figure 6 shows the final stage (13) of the analysis procedure.
We note that the signal sensitivity determined from simula-
tions is quite robust against small deviations between the
simulated event distribution and the actual response of the
detector (Young 2001).
Because of the large experimental data sample, precision
studies of detector performance are possible from the most
common events in the sample to extremely rare compo-
nents. The predicted and experimental sample sizes are com-
pared through all stages of the analysis to establish the
absolute calibration of detector sensitivity. This method
assumes that signal from astronomical sources contributes
negligibly to the sample. Figure 7 shows that the simulated
background and data agree to within a factor of 2 at all
stages of the analysis even though the size of the event sam-
ples varies by 6 orders of magnitude. The relative rates and
the agreements in shape of the selection variable distribu-
tions provide evidence that the background generators and
detector simulation programs are an adequate description
of the detector physics, including detector response.
Because of the optimization on signal to noise, all stages of
the analysis produce event samples that are dominated by
poorly reconstructed downgoing atmospheric muon events.
Diffuse backgrounds from atmospheric neutrinos become
noticeable only after rejecting most of the poorly recon-
structed atmospheric muons, but they never dominate the
event sample.
Obviously, atmospheric muon data are an imperfect tool
to investigate the sensitivity of the detector to neutrino-
induced muons, because of the downgoing nature of the
events and differences in the energy and multiplicity distri-
butions. This concern is addressed by the measurement of
atmospheric neutrinos (Andres et al. 2001; Ahrens et al.
2002a), which were used to verify the basic operational sen-
sitivity of AMANDA-B10 to a known neutrino signal. In
TABLE 1
Description of Variables Used in Analysis
Stage Selection Cut
?
data
?
bgr
?
sig
R
data
R
bgr
0..................... Trigger 1 1 1
. . . ...
1..................... Filter 1: 0.0193 0.0190 0.433 22.4 22.8
1a...................
?
ð
1
Þ
>
50
?
... . . . ... . . . ...
1b...................
?
ð
2
Þ
>
80
?
... . . . ... . . . ...
1c ...................
N
ð
2
Þ
dir
ð
b
Þ
>
2
... . . . ... . . . ...
2..................... Filter 2: 0.0232 0.0283 0.538 23.2 19.0
2a...................
?
ð
2
Þ
>
90
?
... . . . ... . . . ...
2b...................
?
0
:
43
<
S
ð
2
Þ
mrl
<
0
:
3
... . . . ... . . . ...
2c ...................
L
ð
2
Þ
dir
ð
c
Þ
>
75 m
... . . . ... . . . ...
2d...................
L
ð
2
Þ
=
L
ð
1
Þ
<
4
?
10
?
6
... . . . ... . . . ...
2e ...................
?
ð
5
Þ
>
90
?
... . . . ... . . . ...
3..................... cos
?
ð
5
Þ
<
?
0
:
1 0.867 0.861 0.925 1.07 1.07
4.....................
P
ð
5
Þ
up
=
P
ð
5
Þ
down
>
9
:
2 0.212 0.178 0.817 3.85 4.59
5.....................
L
ð
5
Þ
=
L
ð
4
Þ
<
1
:
02 0.370 0.334 0.947 2.56 2.84
6.....................
?
0
:
21
<
S
ð
3
Þ
Phit
<
0
:
33 0.458 0.408 0.927 2.02 2.27
7.....................
L
ð
5
Þ
dir
ð
c
Þ
>
100 m 0.638 0.657 0.932 1.46 1.42
8.....................
N
ð
5
Þ
dir
ð
c
Þ
?
N
ð
4
Þ
dir
ð
c
Þ
>
3 0.737 0.710 0.912 1.24 1.28
9.....................
?
0
:
25
<
S
ð
5
Þ
mrl
<
0
:
26 0.711 0.671 0.912 1.28 1.36
10...................
L
ð
5
Þ
dir
ð
b
Þ
>
40 m 0.744 0.698 0.955 1.28 1.37
11...................
P
ð
5
Þ
=
P
ð
4
Þ
>
9
:
2 0.581 0.562 0.921 1.59 1.64
12...................
L
ð
3
Þ
<
4
:
9 0.515 0.466 0.906 1.76 1.94
13...................
N
ð
5
Þ
dir
ð
c
Þ
>
9 0.659 0.616 0.963 1.46 1.56
Notes.
—Description of selection criteria applied to reconstruction variables in the data
reduction procedure. For additional information on the filters and selection variables, consult
Young (2001). The numerical identification (superscript in parentheses) refers to the reconstruc-
tion algorithm of the event: (1) line fit used as first guess for likelihood fit; (2) maximum likelihood
method for muon track; (3) hit probability reconstruction based on radial distribution of OMs
that detect photons; (4) maximum likelihood method assuming cascade event; (5) iterative applica-
tion of maximum likelihood for muon track. Direct hits are photons that arrive within (b)
[
?
15, +25] ns or (c) [
?
15, +75] ns of the unscattered time of flight between track and optical mod-
ule. The reduced likelihood parameter
L
is
?
log
ð
P
Þ
divided by the number of degrees of freedom,
where
P
is the maximized probability. Passing efficiencies (
?
) relative to the prior stage are shown
at each stage for experimental data, simulated background, and simulated signal. Rejection
factors (
R
) for experimental data and simulated background are shown for each stage in the two
columns farthest to the right.
1046 AHRENS ET AL. Vol. 583
this analysis, the relative agreement between the measured
and predicted event rates is 30%, which is consistent with
uncertainties in the measured flux of cosmic-ray primaries,
theoretical uncertainty in the interaction models, and sys-
tematic uncertainties in the modeling of the detector
response. However, because of the steeply falling energy
spectrum for atmospheric neutrinos, the mean energy of the
muons induced by charged-current interactions is close to
the energy threshold of the detector, which implies that they
cannot be used to reliably probe the high-energy response of
AMANDA-B10.
6. POINTING RESOLUTION AND
POINT-SOURCE SEARCH
The final stage of the data analysis procedure yields a
sample of 815 events (as is evident from Fig. 7, atmospheric
neutrinos contribute about 25% of the events to the simu-
lated background). Visual inspection of the distribution in
the sky of the final event sample, shown in Figure 8, reveals
no obvious clustering. The increase in event density near the
horizon (decl
:
¼
0
?
) is due to the zenith dependence of the
atmospheric muon component of the background. In order
to perform a quantitative search for possible sources of
high-energy neutrinos in the northern hemisphere, the sky
was divided into nonoverlapping angular bins of approxi-
mately equal solid-angle coverage. A point source would
then be revealed by a statistically significant clustering of
events within a particular angular bin. The optimal bin size
and shape depend on the space-angle resolution of the
detector, which can be expressed in terms of a point-spread
function. The space-angle deviation
?
between the true
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
5 1015 20253035
data
signal
background
N
(5)
dir(c)
Normalized Units
Efficiency
N
(5)
dir(c)
0
0.2
0.4
0.6
0.8
1
1.2
5 1015 20253035
Fig.
6.
—
Top
: Similar to Fig. 5, but for analysis stage 13, which involves
the number of direct hits in a time window of
?
15 to +75 ns.
Bottom
:
Passing efficiencies—defined as integrated sums, from given value to
infinity, of distributions shown in the top panel—as a function of cut value.
The vertical lines indicate the cut applied in the analysis. [
See the electronic
edition of the Journal for a color version of this figure.
]
Analysis Stage
No. of events
1997 AMANDAB10 data
atmospheric muons
atmospheric neutrinos
total background
10
2
10
3
10
4
10
5
10
6
10
7
10
8
10
9
01 2 3 45 67 89 10 11 12 13
Fig.
7.
—Number of events remaining in the sample as the selection
criteria in the 13 analysis stages listed in Table 1 are applied sequentially.
The 1997 AMANDA-B10 data (
circles
) are compared to simulated back-
ground from atmospheric muons generated by cosmic-ray interactions
(
squares
) and from muons induced by atmospheric neutrinos (
asterisks
).
Also shown is the sum of atmospheric muon and neutrino backgrounds
(
triangles
). No normalization was applied, but systematic uncertainty at the
trigger level (analysis stage 0) may be as large as
?
30%. [
See the electronic
edition of the Journal for a color version of this figure.
]
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0 5 10 15 20 25 30
data
signal
background
P
(5)
up
/P
(5)
down
Normalized Units
Efficiency
P
(5)
up
/P
(5)
down
0
0.2
0.4
0.6
0.8
1
1.2
0 5 10 15 20 25 30
Fig.
5.
—
Top
: Equally normalized distributions for experimental data
(
circles
) and simulated background (
asterisks
) and signal (
triangles
) for
stage 4 of the point-source analysis, which compares the best likelihood for
an upgoing track hypothesis with the likelihood for a track in the opposite
(i.e., downgoing) direction.
Bottom
: Passing efficiencies as a function of cut
value. The efficiency is obtained from the integrated sums of distributions
shown in the top panel from given value to infinity. The vertical lines
indicate the cut applied in the analysis. [
See the electronic edition of the Jour-
nal for a color version of this figure.
]
No. 2, 2003 SEARCH FOR HIGH-ENERGY NEUTRINO SOURCES 1047
angular coordinates of a muon (
?
true
;
?
true
) and the recon-
structed coordinates (
?
rec
;
?
rec
) is given by
cos
?
¼
cos
?
rec
cos
?
true
þ
sin
?
rec
sin
?
true
cos
ð
?
rec
?
?
true
Þ
:
ð
1
Þ
Figure 9 shows the distribution of
?
and the corresponding
point-spread function, computed from the
?
-distribution
by dividing by the appropriate solid angle, for the simulated
sample of upward-traveling muons generated by neutrinos
with an
E
?
2
energy spectrum. A median value of
?
¼
3
=
9,
averaged over positive declinations, is achieved. A function
involving the sum of two Gaussian distributions was fitted
to the point-spread function. It yields an amplitude ratio of
A
2
=
A
1
¼
0
:
25, indicating the importance of the second com-
ponent related to the degrading angular resolution at large
muon energies (Young 2001). Given this point-spread func-
tion and the relatively small number of background events
shown in Figure 8, the optimal slice in zenith angle is 11
=
25
(Young 2001). For azimuth angle, a weak optimum occurs
for a width of 12
?
for the declination band closest to the
horizon. These angular dimensions of the bins were chosen
to maximize the signal to noise.
Two studies were performed to check the predicted space-
angle resolution and absolute pointing accuracy. The first
uses AMANDA events that were also tagged by the GASP
air Cerenkov telescope (Barbagli et al. 1993). GASP deter-
mines the direction of the air shower, and AMANDA meas-
ures the direction of the penetrating muon component. At
AMANDA depths, these events are almost entirely single
muons. Unfortunately, the duty cycle of operation is low, so
the sample size is relatively small. To improve the statistical
accuracy of the angular resolution studies, a second method
based on extensive air showers was developed. This method
utilized events that triggered both the SPASE array and
AMANDA (X. Bai et al. 2003, in preparation). SPASE
responds to the electron and photon content of the shower
front that reaches the surface. Since the direction of muons
within the air shower event is nearly perpendicular to the
shower front, the difference between the direction of the air
shower and the reconstructed muon direction can be used to
deduce the angular resolution of AMANDA. SPASE meas-
ures the direction of an air shower with a pointing resolution
of approximately 1
?
–2
?
(Dickinson et al. 2000; depending
on shower size), which is small enough to calibrate the
AMANDA pointing resolution.
In this study, AMANDA data were analyzed using the
procedure outlined in Table 1, with the exception that
angle-dependent cuts were inverted to account for the
downgoing direction of travel of SPASE/AMANDA coin-
cidence events (e.g., the cut at stage 3 was changed to
cos
?
ð
5
Þ
>
0
:
1). The absolute pointing accuracy is character-
ized by the average of
D
?
, the difference between the true
and the reconstructed zenith angle. Because of the excellent
zenith-angle resolution of SPASE, SPASE/AMANDA
coincidence data were used to deduce
D
?
using the
reconstructed zenith angles of both detectors,
D
?
¼
?
AMANDA
?
?
SPASE
. Figure 10 shows the measured zenith-
angle resolution using SPASE/AMANDA coincidence
events, together with the resolution obtained for a SPASE/
AMANDA simulation of air showers initiated by protons
and iron nuclei. Iron primaries produce a larger fraction of
coincidence events with more than one muon penetrating to
AMANDA depths, which accounts for the small difference
between protons and iron nuclei. The coincidence data sup-
port the predicted angular resolution and show that the
angular offset is small compared to the angular dimensions
of the sky bins. These conclusions are nearly independent of
the choice of cosmic-ray primary.
Also shown in Figure 10 is the expected angular resolu-
tion as function of declination for single muons with ener-
gies of 0.1 and 4 TeV within the detector volume. These
muon energies were chosen to be representative of the aver-
age muon energy that were initiated by atmospheric neutri-
nos and by a source with a differential energy spectrum
proportional to
E
?
2
. The predicted value for the absolute
offset is less than 1
=
5, which is consistent with results
obtained by additional study of the SPASE/AMANDA
8
4
9
1
3
2
6
5
7
10
1
Mkn 501
2
Mkn 421
3
NGC4151
4
1ES2344
5
3C66A
6
1ES1959+650
7
Crab Nebula
8
Cassiopeia A
9
Cygnus X3
10
Geminga
Fig.
8.
—Sky plot of 815 events obtained from the point-source analysis.
Horizontal coordinates are right ascension, and vertical coordinates are
declination. Also shown are the sky coordinates for 10 potential high-
energy neutrino sources. [
See the electronic edition of the Journal for a color
version of this figure.
]
0
0.05
0.1
0.15
0.2
0.25
0 2468 10 12 14 16 18 20
mean
RMS
median
= 5.00
°
= 3.82
°
= 3.94
°
Ψ
[degrees]
Arbitrary Units
Ψ
[degrees]
Point spread function
A
1
= 0.33
σ
1
= 1.90
°
A
2
= 0.08
σ
2
= 4.45
°
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 2468 10 12 14 16 18 20
Fig.
9.
—
Top
: Distribution of space angle
?
between true and recon-
structed muon track direction for simulated signal with neutrino energy
spectrum proportional to
E
?
2
, averaged over direction.
Bottom
: Point-
spread function for signal in AMANDA-B10, deduced from the space-
angle distribution in the top panel. A function composed of the sum of two
Gaussians was used to characterize this distribution.
1048 AHRENS ET AL. Vol. 583
and GASP/AMANDA coincidence events (Rawlins 2001;
X. Bai et al. 2003, in preparation).
32
The offset is due to bias
in the AMANDA event reconstruction, which tends to pro-
duce more vertical events. This effect is most evident from
the high-energy muon simulation, which shows that the
angular offset changes sign for positive and negative decli-
nations. Since the absolute offset is substantially smaller
than the angular resolution and small compared to the size
of the search bin, the offset has minimal impact on the signal
efficiency. For a zenith (declination) offset of 1
=
5, 6% of the
signal events are shifted to the neighboring bin. The bottom
panel of the figure also shows that the angular resolution for
upward-traveling events is slightly better than for events
traveling in the downward direction, presumably because of
the asymmetry in the response of the photomultiplier tubes,
which are oriented toward the center of the Earth.
To obtain approximately equal solid-angle coverage for
all bins, the northern sky is divided into 154 nonoverlapping
bins, using the calculated optimal declination slice (11
=
25)
and a varying number of bins in azimuth for the resulting
eight declination bands—from 30 near the horizon to three
near +90
?
declination. Each angular bin is then tested for
an excess of events by computing the significance,
?
¼?
log
10
ð
P
Þ
;
ð
2
Þ
where
P
¼
X
1
n
¼
N
0
e
?
l
l
n
n
!
ð
3
Þ
is the probability of the bin containing at least the
observed number of events
N
0
, assuming that fluctuations
are described by a Poisson distribution. The expected
mean number of events
l
is obtained by taking the aver-
age of the number of events in all other bins in the same
declination slice. The polar location of AMANDA
assures equal sky coverage for all declinations, independ-
ent of time gaps in the collection of data. Figure 11
shows the distribution of significance for the experimental
data and for random fluctuation of the background
events. This noise estimate is obtained by randomizing
the right ascension coordinate of the data events, then
recalculating
?
for each bin. A point-source candidate is
identified by a large observed value of significance with a
large ratio to the significance expected from random fluc-
tuations of background. To avoid the statistical problem
of a potential source near a bin boundary distributing
signal between two adjacent bins, the procedure was
repeated with the grid shifted by one-half of a bin in
both declination and azimuth. The largest value of signif-
icance,
?
¼
1
:
85, appears in the bottom panel. Taking
into account the 154 bins in the sky and the two versions
of the sky grid, there is a 40% probability that the most
significant sky bin is produced by random fluctuation of
background. Therefore, the distribution of significance
shows no evidence of a source.
Another approach was also investigated, using the angu-
lar correlation function between event pairs to avoid the
problem of a source near a bin boundary, but this alternate
approach did not reveal sources either (Young 2001).
32
Available at http://area51.berkeley.edu/manuscripts/
20011002xx-kaths_thesis.pdf.
3
2
1
0
1
2
3
4
80 60 40 20 0 20 40 60 80
signal (
E
μ
= 4 TeV)
signal (
E
μ
= 100 GeV)
SPASE/AMANDA coincidence data
proton simulation
iron simulation
Declination [degrees]
Δθ
[degrees]
Declination [degrees]
Ψ
median
[degrees]
0
1
2
3
4
5
6
7
8
80 60 40 20 0 20 40 60 80
Fig.
10.
—Offset in reconstructed zenith angle (
top
) and median space
angle (
bottom
) as a function of declination. Positive declination
corresponds to upward-traveling events in the AMANDA array. SPASE/
AMANDA coincidence data (
squares
) are compared to expectation assum-
ing that the cosmic-ray elemental composition is entirely protons (
circles
)
or iron nuclei (
triangles
). Also shown is the expectation for signal (i.e.,
neutrino-induced muons) with two different energies within the detector.
[
See the electronic edition of the Journal for a color version of this figure.
]
10
2
10
1
1
10
10
2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Significance (
ξ
)
No. of bins
data
background
nominal grid
shifted grid
Significance (
ξ
)
No. of bins
10
2
10
1
1
10
10
2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Fig.
11.
—Distribution of significance for the 154 sky bins. Data
(
squares
) are compared to expectation from randomized background
(
circles
). The statistical uncertainty of the randomized background is negli-
gible. The top panel shows the results for the nominal sky grid. The sky grid
used in the bottom panel has been shifted from the nominal sky grid by
one-half of a bin in both right ascension and declination. [
See the electronic
edition of the Journal for a color version of this figure.
]
No. 2, 2003 SEARCH FOR HIGH-ENERGY NEUTRINO SOURCES 1049
7. FLUX LIMITS
The absence of a detected source translates into an upper
limit on the high-energy neutrino flux. Neutrino flux and
neutrino-induced muon flux limits depend on the effective
area of the detector
A
eff
for a muon with energy
E
l
. The
effective area, obtained by dividing the signal rate by the
incident flux, is determined from simulations by
A
eff
ð
E
l
Þ¼
f
ev
ð
E
l
Þ
A
GEN
;
ð
4
Þ
where
A
GEN
is the cross-sectional area of the cylinder in the
simulation that contains all neutrino interaction vertices
and
f
ev
is the fraction of generated muon events that survive
the 13-stage data analysis procedure. As an example, results
are shown in Figure 12 for muon vertices located near the
detector. The effective area is computed as a function of
declination for muons with energies of 1, 10, and 100 TeV.
These effective areas can be contrasted to the geometric
cross section of the instrumented volume of deep ice, which
spans from 1
:
1
?
10
4
m
2
in the vertical direction to
6
:
1
?
10
4
m
2
in the horizontal direction.
The
average
effective area, folding in the source spectra,
detector response, and the probability of a neutrino-nucleon
interaction, is given by
A
i
eff
¼
N
A
?
ice
A
GEN
"
Z
E
max
?
E
min
?
?
?
l
ð
E
?
Þ
S
ð
E
?
Þ
?h
R
ð
E
?
;
E
min
l
Þi
f
ev
d
?
?
dE
?
dE
?
#
?
Z
E
max
?
E
min
?
d
?
i
dE
?
dE
?
!
?
1
;
ð
5
Þ
where the index
i
denotes either
l
or
?
,and
E
min
?
¼
10 GeV
and
E
max
?
¼
10
7
GeV define the energy range in the simula-
tion, which safely brackets the interval of interest for most
theoretical models. The other variables are
N
A
, Avogadro’s
number,
?
ice
, the molar density of nucleons in ice, and
?
?
l
,
the charged-current cross section for
?
l
(or
?
l
) interactions.
The term
S
ð
E
?
Þ
accounts for neutrino absorption in the
Earth, and
h
R
ð
E
?
;
E
min
l
Þi
is the average propagation length
for a muon created by a neutrino with energy
E
?
, corrected
for the energy threshold of the detector. In this calculation,
the neutrino interaction vertices are located randomly
within a volume that is large compared to the instrumented
volume of the detector. The propagation programs properly
account for muon energy losses.
The muon differential flux is related to the neutrino differ-
ential flux by
d
?
l
dE
?
¼
?
?
l
ð
E
?
Þ
S
ð
E
?
Þh
R
ð
E
?
;
E
min
l
Þi
d
?
?
dE
?
:
ð
6
Þ
The energy-averaged muon effective area is presented in
Figure 13 for different spectral indexes. The energy response
of AMANDA is described by the distribution of
E
l
at the
detector, which depends on the spectral index of the neu-
trino spectrum. Figure 14 shows the distribution for differ-
ential spectra proportional to
E
?
2
and
E
?
3
. For
E
?
2
, the
most probable muon energy (mode) is 5 TeV, and the cen-
tral 90% of the muon events are within the energy interval
80 GeV to 200 TeV. For a spectral index of 3, appropriate
for atmospheric neutrinos, the most probable detected
energy of the muon is much lower.
If neutrino emission from a point source is described by a
differential energy spectrum, the energy-averaged flux limit
is calculated from
?
limit
i
¼
l
s
N
0
;
N
bgr
??
T
live
?
bin
A
i
eff
:
ð
7
Þ
Declination [degrees]
A
μ
eff
[m
2
]
E
μ
= 100 TeV
E
μ
= 10 TeV
E
μ
= 1 TeV
0
10000
20000
30000
40000
50000
0 102030405060708090
Fig.
12.
—AMANDA-B10 effective area for muon detection as a func-
tion of declination (90
?
is vertically up) for three different muon energies at
the detector. The vertical error bars are statistical. [
See the electronic edition
of the Journal for a color version of this figure.
]
E
ν
2
E
ν
2.3
E
ν
2.5
E
ν
2.7
E
ν
3
Input spectrum:
Declination [degrees]
A
–
μ
eff
[m
2
]
0
2000
4000
6000
8000
10000
12000
0 102030 40506070 8090
Fig.
13.
—AMANDA-B10 average effective area for muon detection as
a function of declination (90
?
is vertically up) for assumed differential
neutrino spectral indexes between 2 and 3. The vertical error bars indicate
the uncertainty obtained from studies described in
x
8. [
See the electronic
edition of the Journal for a color version of this figure.
]
1050 AHRENS ET AL. Vol. 583
The quantity
l
s
ð
N
0
;
N
bgr
Þ
is the upper limit on the number
of signal events at the 90% confidence level, calculated fol-
lowing the unified procedure of Feldman & Cousins (1998),
where
N
0
is the observed number of events in a potential
source bin and
N
bgr
is the expected number of background
events. For the binning technique employed in our point-
source search,
N
bgr
is determined by averaging the number
of observed events over the declination band, excluding the
bin being considered. The efficiency factor
?
bin
accounts for
the finite angular resolution and the possible noncentral
location within the bin of a potential source. The factor
T
live
is the operational live time of the detector. The resulting
muon and neutrino flux limits are shown in Figures 15 and
16, respectively, for various assumed spectral indexes
between 2 and 3.
Figure 17 shows a comparison of the AMANDA flux lim-
its with a representative sample of neutrino detectors
located in the Northern Hemisphere. The AMANDA-B10
detector approaches its maximum sensitivity for declina-
tions greater than +30
?
, which complements the sky regions
covered by neutrino detectors such as MACRO (Montaruli
et al. 1999; Ambrosio et al. 2001; Perrone et al. 2001) and
Super-Kamiokande (Matsuno et al. 2001).
33
With only
130.1 days of detector live time, the muon flux limits for pos-
itive declinations approach those achieved for the southern
sky. The AMANDA flux limits were calculated for
E
?
>
10
GeV, while both Super-Kamiokande and MACRO present
fluxes for
E
l
>
1–2 GeV, but for relatively hard differential
neutrino spectra, such as
E
?
2
?
, the impact of energy thresh-
old on muon flux limits is modest (Biron 2002).
34
In addition to the general search for a point source, a
number of potential sources of particular interest were
investigated by performing the significance test while cen-
tering the search bin on their sky coordinates. The result-
ing flux limits are presented in Table 2. As one
interesting example, we compare the AMANDA limit on
log
10
(E
μ
/GeV)
Normalized Units
E
ν
2
input spectrum
E
ν
3
input spectrum
mode
E
μ
= 5 TeV
mode
E
μ
= 80 GeV
0
0.02
0.04
0.06
0.08
0.1
0.12
12 345 67
Fig.
14.
—Neutrino-induced muon energy distribution at the detector.
The differential neutrino spectra used as input for the simulation are pro-
portional to
E
?
2
(
solid curve
) and
E
?
3
(
dashed curve
), respectively. The
most probable energy for each distribution is shown.
E
ν
2
input spectrum
E
ν
2.3
input spectrum
E
ν
2.5
input spectrum
E
ν
2.7
input spectrum
E
ν
3
input spectrum
Declination [degrees]
Φ
ν
limit
[cm
2
s
1
]
10
7
10
6
10
5
10
4
10
3
0 102030 405060 708090
Fig.
15.
—Neutrino flux limit (90% CL) for various spectral indexes. The
results are shown as a function of declination, averaged over right
ascension. Power-law exponent refers to the differential neutrino energy
spectrum. The vertical error bars indicate the uncertainty obtained from
systematic studies [
See the electronic edition of the Journal for a color version
of this figure.
]
E
ν
2
input spectrum
E
ν
2.3
input spectrum
E
ν
2.5
input spectrum
E
ν
2.7
input spectrum
E
ν
3
input spectrum
Declination [degrees]
Φ
μ
limit
[cm
2
s
1
]
10
14
10
13
10
12
0 102030 405060 708090
Fig.
16.
—Neutrino-induced muon flux limit (90% CL) for various
spectral indexes. The results are shown as a function of declination, aver-
aged over right ascension. Note that the power-law exponent refers to the
differential neutrino energy spectrum. [
See the electronic edition of the Jour-
nal for a color version of this figure.
]
33
Available at http://www.copernicus.org/icrc/papers/ici7384_p.pdf.
34
Available at http://area51.berkeley.edu/manuscripts.
No. 2, 2003 SEARCH FOR HIGH-ENERGY NEUTRINO SOURCES 1051
the neutrino flux from Mrk 501 to the observed gamma-
ray flux and this flux corrected for intergalactic absorp-
tion by infrared photons in Figure 18. Assuming that the
neutrino energy spectrum is proportional to the inferred
gamma-ray spectrum at the source, the AMANDA limit
constrains the proportionality factor. For example, the
ratio of the flux of neutrinos to the flux of gamma rays
must be less than 10 if the source spectrum of Konopelko
et al. (1999) is assumed. Recent work (de Jager & Stecker
2002) suggests that the ratio may be smaller.
8. IMPACT OF SYSTEMATIC UNCERTAINTIES
ON FLUX LIMITS
In the absence of a well-understood source of high-energy
neutrinos, the sensitivity of the AMANDA-B10 detector, as
expressed in terms of the effective area and angular resolu-
tion, had to be estimated from detector simulations. The
required input relies on knowledge of detector performance
extracted from, e.g., laboratory measurements of the indi-
vidual components, in situ measurements of the optical
Fig.
17.
—Upper limit on the muon flux (90% CL) as a function of decli-
nation. The solid curve is the AMANDA-B10 limit, averaged over right
ascension, and the region delineated by the long-dashed curves provides a
guide to the statistical fluctuation within the declination interval (see Table
4). The band defined by the short-dashed curves indicates the range of limits
presented by MACRO (Montaruli et al. 1999; Ambrosio et al. 2001;
Perrone et al. 2001) and Super-Kamiokande (Matsuno et al. 2001). [
See the
electronic edition of the Journal for a color version of this figure.
]
TABLE 2
Muon and Neutrino Flux Limits on Selected Point Sources
Source Model
N
0
N
bgr
D
E
(TeV)
?
limit
?
(10
?
8
cm
?
2
s
?
1
)
?
limit
l
(10
?
15
cm
?
2
s
?
1
)
Mrk 501 ....................... 1 7 3.5 0.3–20 86.0 38.9
Mrk 501 ....................... 2 7 3.5 1–1000 9.5 14.6
Mrk 421 ....................... 3 4 3.7 1–1000 11.2 9.7
NGC 4151.................... 3 5 3.6 1–1000 12.9 10.9
NGC 4151.................... 4 5 3.6 60–2500 0.0042 5.6
1ES 2344...................... 5 5 2.9 1–400 12.5 10.3
3C 66A......................... 5 3 3.5 0.8–250 7.2 6.6
1ES 1959+650.............. 5 4 1.7 0.8–250 13.2 9.7
Crab Nebula ................ 5 2 5.6 1–1000 4.2 5.0
Cas A........................... 5 3 2.2 1.8–1000 9.8 7.6
Cyg X-3 ....................... 5 2 3.4 1–1000 4.9 4.6
Geminga ...................... 5 4 7.1 1.8–1000 6.8 9.1
Notes.
—Muon and neutrino flux limits on selected sources for
E
?
>
10 GeV. The term
N
0
is the number
of observed events in the search bin, and
N
bgr
is the expected background. The energy interval
D
E
contains
90% of the neutrino events, and the flux limits are corrected for systematic uncertainty (see
x
8). Representa-
tive survey of models (second column): (1) neutrino spectrum identical to measured photon spectrum
(Aharonian et al. 1999); (2)
d
?
?
=
dE
/
E
?
1
:
92
?
; (3) Szabo & Protheroe 1992; (4) Stecker et al. 1991; (5)
d
?
?
=
dE
/
E
?
2
?
.
log
10
(
E
γ
/ TeV)
log
10
(
E
2
d
N
/d
E
/ TeV m
2
s
1
)
Markarian 501,
z
= 0.0336
AMANDA flux limit
observed gamma flux (HEGRA)
corrected gamma flux (Konopelko, et al.)
10
8
6
4
2
0
1 0.5 0 0.5 1 1.5 2
Fig.
18.
—Time-averaged spectrum of gamma rays from Mrk 501
observed in 1997 (Aharonian et al. 1999; Krennrich et al. 1999) and cor-
rected for intergalactic absorption by the diffuse infrared background
(Konopelko et al. 1999). Gamma-ray flux is compared to the AMANDA
neutrino limit assuming an energy dependence on the neutrino flux
proportional to
E
?
2
.[
See the electronic edition of the Journal for a color
version of this figure.
]
1052 AHRENS ET AL. Vol. 583
properties of the ice, and calibration studies. Consequently,
the predicted sensitivity is affected by uncertainty in this
information. Table 3 lists the dominant contributions to
systematic uncertainties. The uncertainty in the right
column is defined as the variation
jð
A
j
eff
?
A
nom
eff
Þ
=
ð
A
j
eff
þ
A
nom
eff
Þj
of the effective area from its value determined
with the nominal set of input parameters,
A
nom
eff
, given by the
area,
A
j
eff
, obtained by varying the specified parameter
(index
j
) by its estimated uncertainty.
The most significant component is generated from the
uncertainty in the angular dependence of the OM sensitiv-
ity. It arises mainly from a lack of detailed understanding of
the physics governing the refreezing process in the water col-
umn required to be melted for the deployment of OMs. A
local increase in scattering from air bubbles trapped in the
vicinity of the OM translates into a modulation of its angle-
dependent acceptance. This effect is difficult to disentangle
from the intrinsic angular dependence of the OM sensitivity,
which was measured in the laboratory. An event sample
highly enriched in atmospheric muons was used to investi-
gate the in situ angle dependence of the OM sensitivity
(Ahrens et al. 2002a). The modification to the angular sensi-
tivity leads to a 25% uncertainty in the effective area.
Since the angle-integrated sensitivity of the OM is a
poorly constrained parameter in this analysis, we also inves-
tigated the impact of varying the absolute sensitivity of the
OM. It was parameterized by a wavelength-dependent func-
tion that included the PMT quantum efficiency, the OM col-
lection efficiency, obscuration by nearby cables, and
absorption properties of the glass pressure vessel and cou-
pling gel. We obtain a fractional uncertainty of 0.15 in the
effective area after reducing the absolute OM sensitivity by
15%, a value consistent with the atmospheric neutrino
results. Further reduction is inconsistent with observed
experimental trigger rates.
As mentioned in
x
4, two muon propagation routines
were employed to show that systematic variations in the
effective area for signal (i.e., upward-traveling) muons
were between 5% and 10%. This is much less than
observed for studies of atmospheric muons, presumably
because of the much weaker angular dependence of the
average path length. Possible uncertainties in timing and
position calibration of individual OMs are included by
varying these parameters to the largest extent allowed by
the imprecision of the calibration procedures. The effec-
tive area changes by 10%. A conservative estimate of the
variation in sensitivity introduced by uncertainties in the
depth-dependent optical properties and their approximate
treatment in the detector simulation is obtained by sub-
stituting the nominal bulk ice model, containing a param-
eterization of the measured dust strata, with a
homogeneous ice model. The impact on effective area is
less than 15%.
The impact of the most dominant systematic uncertain-
ties on the average effective area is shown in Figure 19,
where the systematic variations have been applied one at a
time, with the exception of the muon propagation curve,
which also includes the variation of the angular OM
sensitivity.
The variation of the detector sensitivity due to systematic
uncertainties was studied by adjusting the physical parame-
ters in the detector simulation. The parameters were
adjusted according to the known or estimated uncertainties
listed in Table 3, which are assumed to bound the true val-
ues of the parameters. Systematic uncertainty was included
according to the prescription of Conrad et al. (2003), which
is an extension of the method of Cousins & Highland
(1992). The calculations assumed that the distribution of
systematic uncertainty was flat and bounded by the maxi-
mum and minimum values for a given declination bin,
found in Figures 20 and 21.
The solid curves in Figures 20 and 21 indicate the flux lim-
its after adjusting for systematic uncertainty. They are valid
for declination greater than +5
?
. The limits including sys-
tematic uncertainties are about 25% worse than those
obtained from the simulation with nominal input parame-
ters. Finally, the flux limits change by less than 6% because
of the effects of zenith offset and the variation in
?
median
due
to declination (Young 2001). These small effects were not
taken into consideration in the limit calculations.
9. DISCUSSION
The previous sections have shown that AMANDA-B10
has unprecedented sensitivity to high-energy neutrinos and
possesses the necessary angular response and background
TABLE 3
Systematic Uncertainty in AMANDAB10 Effective Area
Source of Systematic Uncertainty
Error in
A
l
eff
(%)
Angular dependence of OM sensitivity............................
?
25
Absolute OM sensitivity..................................................
?
15
Muon propagation..........................................................
?
10
Calibration (timing and geometry)..................................
?
10
Hardware simplifications in detector simulation .............
?
<
10
Optical properties of bulk ice ..........................................
?
15
Declination [degrees]
A
–
μ
eff
[m
2
]
nominal (layered ice)
homogeneous ice
absolute OM sensitivity
angular OM sensitivity
muon propagation
0
2000
4000
6000
8000
10000
12000
0 102030405060708090
Fig.
19.
—Comparison of average muon effective area for differential
neutrino signal proportional to
E
?
2
. See text for explanation of legend.
Statistical uncertainty is indicated by vertical lines, unless it is obscured by
the symbol. [
See the electronic edition of the Journal for a color version of this
figure.
]
No. 2, 2003 SEARCH FOR HIGH-ENERGY NEUTRINO SOURCES 1053
rejection to search for point emission of these particles from
astronomical objects; i.e., it is a novel telescope that detects
the neutrino messenger. The sensitivity and angular
response were determined by simulation. The reliability of
these programs was established by utilizing the known sig-
nals generated by (downgoing) atmospheric muons and
(upgoing) atmospheric neutrinos. The angular response was
confirmed by the study of air shower events that triggered
both AMANDA-B10 and SPASE. Systematic uncertainty
in the analysis procedure was also addressed.
The search for point sources of high-energy neutrinos
revealed no candidates. A set of event selection criteria was
determined by optimizing the signal-to-noise ratio for a sig-
nal with a hard energy spectrum, yet this analysis retains
reasonable sensitivity for softer spectra. The upper limits on
muon flux for all search bins in the northern hemisphere are
presented in Table 4.
The neutrino flux limits in Figure 15 are inferred from the
assumption of a power-law energy spectrum. This proce-
dure is reliable if the mean energy of the neutrino-induced
muon is compatible with the energy response of the detec-
tor. For example, Figure 14 shows that
E
l
at the detector
brackets the interval between 0.1 and 10
3
TeV for source
spectra proportional to
E
?
2
. Two lines of evidence show
that the simulated energy response of the detector is valid
over this interval. First, the agreement between the detected
and expected rates of atmospheric neutrinos shows that the
response of AMANDA is being correctly modeled in the
sub-TeV region. Second, the tails of the
N
ch
-distribution are
sensitive to brighter events within AMANDA, which
are roughly equivalent to single muons with energy above 1
TeV. We know of no reason to doubt the predicted energy
response for
E
l
<
10
3
TeV. Evaluation and calibration of
the energy response beyond 10
3
TeV remain an ongoing
activity.
Not all model predictions are well characterized by
power-law energy spectra. Therefore, Table 2 shows the
results for a selection of models in the literature. The
inferred limits on neutrino flux apply to point sources with
continuous emission (or episodic emission averaged over
the time interval of data collection) and power-law energy
spectra with a fixed spectral index. The limits presented here
for sources at large positive declination complement exist-
ing data, so that comparable limits now exist for the entire
sky.
During 1997, the TeV gamma-ray emission of two
nearby AGN blazars (Mrk 421 and 501) were observed
to exhibit episodic flaring. If neutrino emission follows
the same time variability, then it may be possible to
improve the signal-to-noise ratio by eliminating the peri-
ods of relatively low output. Multiple detections of Mrk
501 from several air Cerenkov instruments allowed nearly
continuous monitoring, including periods when the Moon
was shining. However, monitoring by multiple instru-
ments extended only from March to late August. Because
of uncertainties in the details of the time dependence of
the gamma-ray emission,
neutrino
flux limits are not
greatly improved by restricting the analysis to high-flux
periods of gamma-ray emission.
While this paper describes an analysis dedicated to the
search for point sources, another strategy was developed
based on the event selection of the atmospheric neutrino
analysis (Biron 2002). The results of this complementary
analysis are consistent with the results presented here. The
absolute efficiency was extracted by comparing to the known
flux from atmospheric neutrinos. Moreover, the second
analysis was subject to different systematic uncertainties.
The method based on the atmospheric neutrino analysis
retained a smaller event sample of 369 events, of which
?
270 are expected from atmospheric neutrinos. The cut
selections produce an implicit optimization on more vertical
Declination [degrees]
Φ
μ
limit
[cm
2
s
1
]
nominal (layered ice)
homogeneous ice
absolute OM sensitivity
angular OM sensitivity
muon propagation
corrected
10
14
0 10 2030 405060 7080 90
Fig.
20.
—Comparison of muon flux calculations for differential neutrino
signal proportional to
E
?
2
. See text for explanation of legend. Statistical
uncertainty is small and not shown. The solid curve (corrected) includes
systematic uncertainty and indicates the final result of this work. [
See the
electronic edition of the Journal for a color version of this figure.
]
Declination [degrees]
Φ
ν
limit
[cm
2
s
1
]
nominal (layered ice)
homogeneous ice
absolute OM sensitivity
angular OM sensitivity
muon propagation
corrected
10
7
0 10 2030 405060 7080 90
Fig.
21.
—Comparison of neutrino flux calculations for differential
neutrino signal proportional to
E
?
2
. See text for explanation of legend.
Statistical uncertainty is small and not shown. The solid curve (corrected)
includes systematic uncertainty and indicates the final result of this work.
[
See the electronic edition of the Journal for a color version of this figure.
]
1054 AHRENS ET AL. Vol. 583
events and/or softer energy spectra. Figure 22 compares the
average effective area of the two analyses for an assumed
differential spectra proportional to
E
?
2
. The best flux limits
for soft spectra are obtained by atmospheric neutrino analy-
sis, but the neutrino and muon flux limits for either analysis
are much larger than obtained for an assumed power law of
an index of
?
2.0.
While the flux limits for any particular source or direction
in the northern hemisphere can be extracted from this analy-
sis (see Table 4), flux limits, both integral and pseudodiffer-
ential, for a preselected list of 62 sources have been reported
(Biron 2002). These include all known TeV gamma-ray
blazars, nearby QSOs, and Galactic TeV gamma-ray
sources in the northern hemisphere. The list also includes
TABLE 4
Muon Flux Limits
Decl.
(deg)
R.A.
(hours)
?
limit
l
(10
?
15
cm
?
2
s
?
1
)
Decl.
(deg)
R.A.
(hours)
?
limit
l
(10
?
15
cm
?
2
s
?
1
)
Decl.
(deg)
R.A.
(hours)
?
limit
l
(10
?
15
cm
?
2
s
?
1
)
85........... 4.0 2.5 39........... 7.8 6.1 17........... 7.9 9.0
12.0 6.5 8.9 2,5 8.7 9.0
20.0 10.5 9.9 15.5 9.5 19.9
73........... 1.3 6.9 11.0 13.5 10.3 14.9
4.0 4.2 12.0 13.5 11.2 42.6
6.7 2.1 13.0 8.0 12.0 11.9
9.3 6.9 14.1 8.0 12.8 34.1
12.0 13.7 15.1 8.0 13.7 24.6
14.7 6.9 16.2 10.6 14.5 42.6
17.3 9.6 17.2 15.5 15.3 8.9
20.0 4.2 18.3 3.9 16.1 11.9
22.7 4.2 19.3 6.1 17.0 14.9
62........... 0.9 7.9 20.3 10.6 17.8 8.9
2.6 5.5 21.4 3.9 18.6 42.6
4.3 12.1 22.4 3.9 19.4 19.9
6.0 3.4 23.5 6.1 20.3 19.9
7.7 7.9 28........... 0.4 20.5 21.1 48.0
9.4 5.5 1.3 20.5 21.9 14.9
11.1 7.9 2.2 7.1 22.8 34.1
12.9 1.9 3.1 13.9 23.6 29.1
14.6 5.5 4.0 24.0 6............. 0.4 33.6
16.3 3.5 4.9 13.9 1.2 64.9
18.0 3.5 5.8 7.1 2.0 127.2
19.7 7.9 6.7 20.5 2.8 64.9
21.4 7.9 7.6 13.9 3.6 47.8
23.1 9.9 8.4 20.5 4.4 33.6
51........... 0.6 7.7 9.3 7.1 5.2 96.1
1.9 10.0 10.2 24.0 6.0 64.9
3.2 2.2 11.1 20.5 6.8 78.0
4.4 7.7 12.0 24.0 7.6 78.0
5.7 5.9 12.9 5.4 8.4 47.8
6.9 10.0 13.8 13.9 9.2 64.9
8.2 2.2 14.7 5.4 10.0 64.9
9.5 10.0 15.6 20.5 10.8 127.2
10.7 5.9 16.4 16.7 11.6 47.9
12.0 1.6 17.3 5.4 12.4 33.6
13.3 10.0 18.2 16.7 13.2 143.5
14.5 5.9 19.1 7.1 14.0 24.9
15.8 10.0 20.0 10.2 14.8 47.8
17.1 7.7 20.9 7.1 15.6 33.6
18.3 14.1 21.8 13.9 16.4 64.9
19.6 3.9 22.7 10.2 17.2 47.8
20.8 5.9 23.6 7.1 18.0 47.8
22.1 12.4 17........... 0.4 48.0 18.8 112.4
23.4 7.7 1.2 34.1 19.6 96.1
39........... 0.5 10.6 2.1 14.9 20.4 33.6
1.6 15.5 2.9 51.9 21.2 96.1
2.6 8.0 3.7 19.9 22.0 24.9
3.7 13.5 4.6 24.6 22.8 64.9
4.7 3.9 5.4 14.9 23.6 78.0
5.7 15.5 6.2 4.5
6.8 6.1 7.0 9.0
Notes.
—Neutrino-induced muon flux upper limits for source spectra proportional to
E
?
2
. The impact of systematic uncertainty is included. Angu-
lar coordinates refer to the center of the search bin.
No. 2, 2003 SEARCH FOR HIGH-ENERGY NEUTRINO SOURCES 1055
microquasars, the five most luminous AGNs in wavelength
bands that span across MeV, X-ray, infrared, and radio
bands. Of particular interest are radio galaxies with strong
emission at GHz frequencies. We have also investigated BL
Lac objects that are close to the arrival directions of the very
highest energy cosmic rays (Tinyakov & Tkachev 2001) and
the 10 reported cosmic-ray doublets at extreme energies
(Uchihori et al. 2000).
10. FUTURE
The technique employed in this paper optimized the selec-
tion criteria on signal to noise. Because of the relatively
large number of sky bins and the relatively low number of
events in any individual bin, the analysis procedure pro-
duced event samples that are dominated by poorly recon-
structed atmospheric muons rather than upward-traveling
atmospheric neutrino background. However, as the back-
ground rejection of downgoing events improves with
larger detectors, such as AMANDA-II, this trend may not
continue.
AMANDA-II, completed in 2000 January, surrounds the
B10 core with nine additional strings, more than doubling
the number of optical modules. For this broader configura-
tion, the effective area for neutrino-induced muons remains
relatively constant over the entire hemisphere (Barwick et
al. 2001).
35
Consequently, AMANDA-II is expected to
achieve a factor of 5 improvement in sensitivity for nearly
horizontal events compared to AMANDA-B10 (Wischnew-
ski et al. 2001).
36
The greater statistical sample of atmo-
spheric neutrinos will allow better tests of the detector
simulation programs, especially near the horizon. With the
data already collected on tape, AMANDA-II can observe
(or exclude) neutrino fluxes that are approximately 1 order
of magnitude below the limits presented here, as shown in
Figure 1.
This research was supported by the following agencies:
the US National Science Foundation Office of Polar Pro-
grams and Physics Division; the University of Wisconsin
Alumni Research Foundation; the US Department of
Energy; the Swedish Natural Science Research Council; the
Swedish Polar Research Secretariat; the Knut and Alice
Wallenberg Foundation, Sweden; the German Ministry for
Education and Research; the US National Energy Research
Scientific Computing Center (supported by the Office of
Energy Research of the US Department of Energy); the
University of California, Irvine, AENEAS Supercomputer
Facility; and the Deutsche Forschungsgemeinschaft
(DFG). C. Pe
´
rez de los Heros received support from the EU
fourth framework of Training and Mobility of Researchers,
and D. F. Cowen acknowledges the support of the NSF
CAREER program. P. Desiati was supported by the
Koerber Foundation.
REFERENCES
Agrawal, V., Gaisser, T., Lipari, P., & Stanev, T. 1996, Phys. Rev. D, 53,
1314
Aharonian, F. A., et al. 1999, A&A, 349, 11
Ahrens, J., et al. (AMANDA Collaboration). 2002a, Phys. Rev. D, 66,
012005
———. 2002b, Phys. Rev. D, 66, 032006
———. 2002c, Phys. Rev. D, submitted (astro-ph/0206487)
Ambrosio, M., et al. (MACRO Collaboration). 2001, ApJ, 546, 1038
Andres, E., et al. 2000, Astropart. Phys., 13, 1
———. 2001, Nature, 410, 441
Barbagli, G., et al. 1993, Nucl. Phys. B (Proc. Suppl.), 32, 156
Barwick, S. W., et al. (AMANDA Collaboration). 2000, Nucl. Phys. B
(Proc. Suppl.), 87, 402
———. 2001, Proc. 27th Int. Cosmic-Ray Conf. (Hamburg), 2, 1101
Bednarek, W., & Protheroe, R. J. 1999, MNRAS, 302, 373
Berezinsky, V. S., & Dokuchaev, V. I. 2001, Astropart. Phys., 15, 87
Biron, A. 2002, Ph.D. thesis, Humboldt Univ. Berlin
Bouchta, A., et al. (AMANDA Collaboration). 1999, Proc. 26th Int.
Cosmic-Ray Conf. (Salt Lake City), 2, 108
Boziev, S. N. et al. 1989, preprint (INR preprint P-0630)
Buckley, J. H., et al. 1998a, Adv. Space Res., 21, 101
Buckley, J. H., et al. 1998b, Science, 279, 676
Catanese, M., & Weekes, T. C. 1999, PASP, 111, 1193
Conrad, J., Botner, O., Hallgren, A., & Pe
´
rez de los Heros, C. 2003, Phys.
Rev. D, in press (hep-ex/0202013)
Cousins, R. D., & Highland, V. L. 1992, Nucl. Instrum. Methods Phys.
Res. A, 320, 331
de Jager, O. C., & Stecker, F. W. 2002, ApJ, 566, 738
Dermer, C. D. 1999, Astropart. Phys., 11, 1
Dermer, C. D., & Atoyan, A. 2001, Phys. Rev. Lett., 87, 221102
Dickinson, J. E., et al. 2000, Nucl. Instrum. Methods Phys. Res. A, 440,
95
Distefano, C., Guetta, D., Waxman, E., & Levinson, A. 2002, ApJ, 575, 378
Feldman, G. J., & Cousins, R. D. 1998, Phys. Rev. D, 57, 3873
Gaisser, T. K. 2002, Astropart. Phys., 16, 285
Hardtke, R., & Barouch, G. (for the AMANDA Collaboration). 2001,
Proc. 27th Int. Cosmic-Ray Conf. (Hamburg), 2, 1121
Heck, D., et al. 1998, Rep. FZKA 6019 (Forschungszentrum Karlsruhe)
———. 1999, in Simulations and Analysis Methods for Large Neutrino
Telescopes, ed. C. Spiering (DESY-PROC-1999-01; Zeuthen: DESY),
228
Hill, G. 1997, Astropart. Phys., 6, 215
Hill, G., & Leuthold, M. (for the AMANDA Collaboration). 2001, Proc.
27th Int. Cosmic-Ray Conf. (Hamburg), 2, 1113
Hundertmark, S., et al. (AMANDA Collaboration). 1999, Proc. 26th Int.
Cosmic-Ray Conf. (Salt Lake City), 2, 12
Karle, A., et al. 2003, Proc. XXth Int. Conf. Neutrino Physics and Astro-
physics (Amsterdam: North-Holland), in press (astro-ph
=
0209556)
Konopelko, A. K., Kirk, J. G., Stecker, F. W., & Mastichiadis, A. 1999,
ApJ, 518, L13
Fig.
22.
—Comparison of average muon effective area as a function of
declinations for differential neutrino signal proportional to
E
?
2
. The
dashed curve is from Biron (2002). Neither curve includes the effect of
systematic uncertainties. [
See the electronic edition of the Journal for a color
version of this figure.
]
35
Available at http://area51.berkeley.edu/manuscripts/
20010601xx-ICRC_AmIIv3.pdf.
36
Available at http://area51.berkeley.edu/manuscripts/
20010602xx-icrc2001_aii.pdf.
1056 AHRENS ET AL. Vol. 583
Krennrich, F., et al. 1999, ApJ, 511, 149
Learned, J. G., & Mannheim, K. 2000, Annu. Rev. Nucl. Part. Sci., 50, 679
Levinson, A., & Waxman, E. 2001, Phys. Rev. Lett., 87, 171101
Lipari, P., & Stanev, T. 1991, Phys. Rev. D, 44, 3543
Lohmann, W., Kopp, R., & Voss, R. 1985, preprint (CERN preprint
CERN-85-03)
Matsuno, S., et al. (Super-Kamiokande Collaboration). 2001, Proc. 27th
Int. Cosmic-Ray Conf. (Hamburg), 2, 1065
Montaruli, T., et al. (MACRO Collaboration). 1999, Proc. 26th Int.
Cosmic-Ray Conf. (Salt Lake City), 2, 213
Nellen, L., Mannheim, K., & Biermann, P. L. 1993, Phys. Rev. D, 47, 5270
Oppelt, A. 2001, Ph.D. thesis, Univ. Me
´
diterrane
´
e Aix-Marseille II
Perrone, L., et al. (MACRO Collaboration). 2001, Proc. 27th Int. Cosmic-
Ray Conf. (Hamburg), 2, 1073
Rachen, J. P. 2000, preprint (astro-ph/0003282)
Rawlins, K. 2001, Ph.D. thesis, Univ. Wisconsin, Madison
Rhode, W., & Chirkin, D. 2001, Proc. 27th Int. Cosmic-Ray Conf.
(Hamburg), 3, 1017
Spiering, C. 2001, Proc. 27th Int. Cosmic-Ray Conf. (Hamburg), 2, 1242
Stecker, F. W., Done, C., Salamon, M. H., & Sommers, P. 1991, Phys. Rev.
Lett., 66, 2697 (erratum 69, 2738 [1992])
Stecker, F. W., & Salamon, M. H. 1996, Space Sci. Rev., 75, 341
Szabo, A. P., & Protheroe, R. J. 1992, in Proc. Workshop on High Energy
Neutrino Astrophysics, ed. V. J. Stenger, J. G. Learned, S. Pakvasa, &
X. Tata (Singapore: World Scientific), 24
Tinyakov, P. G., & Tkachev, I. I. 2001, J. Exp. Theor. Phys. Lett., 74, 445
Uchihori, Y., et al. 2000, Astropart. Phys., 13, 151
Weekes, T. 2001, AIP Conf. Proc. 558, High Energy Gamma-Ray Astron-
omy, ed. F. A. Aharonian & H. Vo
¨
lk (Melville: AIP), 15
Wiebel-Sooth, B., & Biermann, P. L. 1998, in Landolt-Bo
¨
rnstein New
Series, Group VI, Vol. 3C (Berlin: Springer), 37
Wischnewski, R., et al. (AMANDA Collaboration). 1999a, Nucl. Phys. B
(Proc. Suppl.), 75 (1–2), 412
———. 2001, Proc. 27th Int. Cosmic-Ray Conf. (Hamburg), 2, 1105
Woschnagg, K., et al. (AMANDA Collaboration). 1999, Proc. 26th Int.
Cosmic-Ray Conf. (Salt Lake City), 2, 200
Young, S. 2001, Ph.D. thesis, Univ. California, Irvine
No. 2, 2003 SEARCH FOR HIGH-ENERGY NEUTRINO SOURCES 1057
Back to top