ARTICLE IN PRESS
Nuclear Instruments and Methods in Physics Research A 588 (2008) 171–175
www.elsevier.com/locate/nima
Multiparametric topological analysis (MTA) for the study of the
primary CR composition: Performances with Auger simulated data
D. D’Ursoa,b,, M. Ambrosioa, C. Aramoa, F. Guarinoa,b, L. Valorea,b,
for the Pierre Auger Collaborationc
a
Sezione INFN di Napoli, Italy
Università di Napoli ‘‘Federico II’’, Italy
c
Pierre Auger Observatory: Av. San Martin Norte 304, (5613) Malargüe, Argentina
b
Available online 12 January 2008
Abstract
We describe the application of a multiparametric analysis to estimate the UHE Cosmic Rays composition. The proposed method,
MTA (Multiparametric Topological Analysis), is based on the study of the correlations among different shower observables.
This technique is designed to fully exploit the complementarity of Auger fluorescence and ground array data.
In the present work, we report the results of the application to Conex showers, fully simulated through the Auger detector, using only
parameters describing the longitudinal development of air showers as recorded by fluorescence detector for hybrid data.
r 2008 Elsevier B.V. All rights reserved.
Keywords: Ultra high energy cosmic rays; Auger observatory; Mass compostion; Multiparametric analysis
1. Introduction
The understanding of the nature of the ultra high energy
cosmic rays (UHECR) is a crucial point towards the
determination of their origin, acceleration and propagation mechanism. In the energy range of the knee (about
1015 eV) is widely accepted that cosmic rays have a
galactic origin and their most probable source are the
supernova remnants (SNR). Above 1018 eV all observed
cosmic rays are presumed to come from extragalactic
sources, because there are no known galactic sources able
to produce particles up to such energies and they cannot be
contained in our galaxy long enough to be accelerated.
The energy at which the transition between galactic and
extragalactic cosmic rays occurs is still unknown and only
a detailed knowledge of the particle spectrum and composition could allow to distinguish among different models
(see Ref. [1] for further details).
Designed to solve the UHECR puzzle, the Pierre Auger
Observatory [2] collects data with unprecedented precision
Corresponding author at: Sezione INFN di Napoli, Italy.
E-mail address: domenico.durso@na.infn.it (D. D’Urso).
0168-9002/$ - see front matter r 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.nima.2008.01.034
and statistics. The Observatory is a ‘‘hybrid detector’’
composed by an array of 1600 water Cherenkov particle
detectors (tanks), distributed over an area of 3000 km2
(Surface Detector, SD) overlooked by 24 fluorescence
telescopes (Fluorescence Detector, FD). The combined use
of these two detection techniques provides cross-calibration and a better event reconstruction accuracy. Moreover,
the combined use of FD and SD observables will help to
put tighter constraints on hadronic interaction models. A
description of Auger hybrid performances is given in Ref.
[2]. Nowdays, the Auger Observatory has recorded more
events above 1018 eV than all the previous cosmic ray
experiments on UHECR.
The most popular method developed so far to infer the
primary composition from fluorescence experiments makes
use of the depth of the maximum of the cascade
development ðX Max Þ and derives the observed mean mass
composition as a function of the primary energy comparing
the measured shower maxima with the Monte Carlo
predictions (Elongation Rate analysis). Within the
Auger Collaboration such studies are performed on hybrid
events, in which at least one SD station is involved,
with a consequent improvement of the reconstruction
ARTICLE IN PRESS
172
D. D’Urso et al. / Nuclear Instruments and Methods in Physics Research A 588 (2008) 171–175
accuracy with respect to pure FD data (shower maximum can be measured with an accuracy better than
40 g cm2 [3]).
In surface array experiments, informations on primary
composition are usually extracted from signal rise time,
time curvature of the shower front, muon content and
azimuthal signal asymmetry. This kind of analysis is in
progress on Auger SD data alone [4].
However the sensitivity of these methods is limited, due
to the strong fluctuations of the physical processes which
smear out the informations and produce an overlapping of
the observed features of showers initiated by different
primaries. Such unsatisfactory situation can improve if
more parameters are taken into account, exploiting the
unique future of Auger hybrid data, that contain informations on the longitudinal profile and lateral distribution of
a shower at the same time. This type of arguments
triggered our efforts to find methods which make use of
a larger amount of informations and which are capable to
allow the identification (at least in terms of probabilities) of
the cascade origin also for individual showers.
A comparison of the performances of such methods
applied to the study of the longitudinal development,
recently published by our group [5], indicates that
multiparametric topological analysis (MTA) analysis
performances are very promising and nearly comparable
with those of neural networks. Because of the simplicity
of the basic algorithms we decided to use MTA for this
study.
2. The MTA method
Our aim is to infer the UHECR composition using
the full information available in the hybrid Auger data.
Any shower observable cannot be unambiguously connected to the primary mass: distributions are always mixed
also at fixed energy and we can never identify the
individual primary mass without uncertainty. However,
using a multiparametric approach, comparing the correlation of a set of discriminant variables extracted from real
events with the corresponding quantities derived from
simulated ones of known energy and composition, it is
possible to define a probability for each individual shower
to originate from a nucleus of a given mass. Once a set of
measured shower parameters sensitive to primary composition is chosen, MTA method [5] relies on a topological
analysis of their correlations. Therefore, application to
Auger data requires a full and detailed simulation of
showers in both SD and FD detectors. CORSIKA [6]
showers are ideal for this purpose, but too heavy both in
terms of CPU and storage requirements, to reproduce real
data with reasonable statistics. Two solutions are possible:
the use of CORSIKA with non optimal thinning, or the use
of parametric or semianalytical approaches to shower
development.
For the study of the longitudinal profile of atmospheric
cascades is now available a fast Monte Carlo program,
Conex [7], based on a semianalytical approach. The distribution of particles at ground is not provided, nevertheless,
tank trigger simulation is performed using a parametrized ‘‘Lateral Trigger Probability’’ function [8] and
the time of the station with the highest signal is calculated
(a detailed description of the hybrid detector simulation
program is given in Ref. [9]). This information is sufficient
to operate a hybrid geometry reconstruction [2]. A more
complete fast simulator, SENECA [10], which also
provides reliable informations of particles at ground, will
be soon available.
For the time being, we restrict the analysis to the
use of only FD parameters for hybrid events by means of
Conex simulated showers. Simulated data have been
produced by the Conex MonteCarlo, version 2r1.4b, fully
processed with the Auger simulation-reconstruction software, using:
(i) QGSJET-II-3 [11] as high energy hadronic interaction
model;
(ii) energy spectrum from 1017:5 eV to 1020:2 eV, according
to dN=dE / E 3 , for a total of 2 106 events;
(iii) zenith distribution from 0 to 60 , according to
dN=ðd cos yÞ / cos y;
(iv) uniform azimuth distribution.
Quality cuts to simulated data, as described in Ref. [3],
have been applied.
3. Definition of the parameter space
Once a set of observables is chosen, it defines a
parameter space, which is divided into cells whose dimensions are related to the experimental accuracy. The set of
shower reconstructed parameters provides the association
of the event to one of the cells of the space. For each
primary mass and for each energy bin, a set of simulated
events, fully processed by the detector simulation and
reconstruction codes, is used to populate the parameter
space.
Restricting to the case of shower quantities measured by
the FD, the space is defined by a choice of the four
parameters describing the fit with a Gaisser–Hillas function
to the shower profile:
ðX Max X 0 Þ=l
dE
dE
X X0
¼
eðX Max X 0 Þ=l
(1)
dX dX Max X Max X 0
where dE=dX and dE=dX Max are the energy deposit at the
depth X and at the shower maximum, respectively.
As an example, in Fig. 1 a parameter space built with
two parameters from the fit, X Max and l is shown. The
parameter space is divided in cells with dimensions 20 and
5 g cm2 , respectively, and is populated with Conex
simulated showers induced by the two limit masses, proton
(triangles) and iron (circles).
ARTICLE IN PRESS
D. D’Urso et al. / Nuclear Instruments and Methods in Physics Research A 588 (2008) 171–175
In the near future, the composition analysis will be
extended to observables measured by the ground array
detector.
Considering only two primary masses, protons and
iron nuclei, in each cell, which in the most general
n-dimensional case is defined by ðh1 . . . hn Þ, one can define
ðh1 ...hn Þ
the total number of showers N tot
, the total number of
showers induced by protons and iron nuclei, N Pðh1 ...hn Þ and
ðh1 ...hn Þ
N Fe
, respectively, and then derive the associated
frequencies:
ðh1 ...hn Þ
pPðh1 ...hn Þ ¼ N Pðh1 ...hn Þ =N tot
ðh1 ...hn Þ
ðh1 ...hn Þ
ðh1 ...hn Þ
pFe
¼ N Fe
=N tot
(2)
which can be interpreted as the probability for a real
shower falling into the cell ðh1 . . . hn Þ to be initiated by
protons or iron primary nuclei.
QGSJET-II E = 1019.1 eV
Proton
173
4. Average mixing probabilities
In the case of a data set of M showers, the sample may be
seen as composed by a mixture of M pP proton showers
and M pFe iron induced showers, where
pj ¼
M
X
ðh1 ...hn Þm
pj
=M
(3)
m¼1
and the sum is performed over all the experimental data,
with (h1 . . . hn Þm indicating the cell interested in by the mth
event. A second set of showers is used to compute the
mixing probabilities Pi!j that an event of mass i is
identified as primary j. The mean Pi!j are obtained by
computing pj for samples of pure primary composition i.
In Fig. 2 average mixing probabilities Pi!j computed for
the 2D example (Fig. 1) for proton and iron are shown.
The mean probabilities that the primary mass is correctly
identified are of the order of 80%.
5. Determination of the primary mass composition
Iron
1200
The mixing probabilities Pi!j can be used for the
reconstruction of the primary mass composition as the
coefficients in the system of linear equations:
1100
XMax [g/cm2]
1000
900
N 0P ¼ N P PP!P þ N Fe PFe!P
800
N 0Fe ¼ N P PP!Fe þ N Fe PFe!Fe
700
600
500
400
20
30
40
50
60
70
80
90
100
λ [g/cm2]
Fig. 1. X Max vs. l parameter space populated by Conex proton (triangles)
and iron (circles) induced showers.
where N P and N Fe are the true values, which are altered to
N 0P and N 0Fe , due to misclassification.
Observed composition is also altered by different triggerreconstruction–selection efficiencies for different primaries,
P and Fe for proton and iron. This effect is compensated
by applying a correction given by the efficiency ratio to the
composition values obtained solving the system. In Fig. 3
the ratio of the overall efficiencies for protons P and irons
Fe is shown as a function of the energy obtained applying
the analysis cuts described in Ref. [3]. Defined Di ¼ N i =N
and if DP and DFe are the solutions, the corrected
Proton
Iron
1
1
0.8
0.8
Probability
Probability
(4)
0.6
0.4
0.2
0.6
0.4
0.2
0
0
Proton
Iron
Proton
Iron
Fig. 2. Average mixing probabilities estimated for the data in Fig. 1: the mean probability that a proton is correctly identified as proton or that is
misidentified as iron (left pad); that one that an iron nucleus is misidentified as proton or correctly classified as iron (right pad).
ARTICLE IN PRESS
D. D’Urso et al. / Nuclear Instruments and Methods in Physics Research A 588 (2008) 171–175
174
Proton - Iron Efficiency Ratio
Reconstructed Composition
1.6
1
1.4
0.8
Composition
∈P /∈Fe
1.2
1
0.8
0.6
0.4
0.6
0.2
0.4
17.5
18
18.5
19
log10 Energy [eV]
19.5
0
20
17.5
Fig. 3. Ratio of the overall efficiencies for protons and irons as a function
of the energy.
abundancies DP and DFe are given by
DP
D P =Fe
¼ P
DFe
DFe
18.5
19
log10 Energy [eV]
19.5
20
Fig. 4. MTA mass composition (full markers protons, empty markers
irons) as a function of the energy for two proton–iron mixing 90–10%
(circles), 70–30% (triangles) compared with the expected values (solid and
dashed lines).
(5)
Elongation Rate Comparison
with the conditions
850
DP þ DFe ¼ 1
DP þ DFe ¼ 1.
The technique has been tested on simulated samples of
known mass composition, in energy bins of 0:2 log10
ðE½eVÞ. The MTA analysis has been applied to 3parameters X Max l X 0 . The parameter space has been
divided in cells whose dimensions are 20, 5 and 50 g cm2
for X Max ; l and X 0 , respectively.
In Fig. 4 MTA composition results (full markers
protons, empty markers irons) are shown as a function of
the energy for two proton–iron mixing 90–10% (circles),
70–30% (triangles) compared with the expected values
(solid and dashed lines).
MTA sensitivity to mass composition has been compared
with the usual elongation rate analysis performed on the same
data: for different simulated composition mixture, the elongation rate computed in the usual way has been confronted with
an MTA-elongation rate obtained weighting the mean X Max of
pure data samples by the MTA abundancies. Let hX Max ðE i ÞiP
and hX Max ðE i ÞiFe be the mean values of the X Max distributions
of pure proton and iron samples, DP ðE i Þ and DFe ðE i Þ the
MTA percentage for the two masses in the ith energy bin, then
it is possible to compute the mean X Max at the energy E i
(6)
XMax [g/cm2]
800
6. Method performances
hX Max ðE i Þi ¼ hX Max ðE i ÞiP DP ðE i Þ
þ hX Max ðE i ÞiFe DFe ðE i Þ.
18
750
700
Expected ER
MTA ER 90%P
MTA ER 80%P
MTA ER 70%P
MTA ER 60%P
650
600
17.5
18
18.5
19
log10 Energy [eV]
19.5
20
Fig. 5. Comparison between the standard elongation rate and that one
obtained from MTA results for samples with different proton percentage:
90% (full squares), 80% (empty squares), 70% (full circles) and 60%
(empty circles).
In Fig. 5 is shown the elongation rate obtained for samples
with 90% (full squares), 80% (empty squares), 70% (full
circles) and 60% (empty circles) of protons, compared with
the expected values (thin lines). Elongation rate curves
reproduce the expected ones with high accuracy over the
whole energy range. Curves corresponding to proton compositions different at level of 10% are easily distinguishable.
The test has shown that, once an hadronic model is
chosen, in this case QGSJET-II, MTA applied to FD only
ARTICLE IN PRESS
D. D’Urso et al. / Nuclear Instruments and Methods in Physics Research A 588 (2008) 171–175
parameters has a sensitivity to the input primary mass
better than 5%, in the whole energy range and at each
tested proton–iron mixing. The use of both FD and SD
observables in the analysis will give better results. It is clear
that the dominant uncertainty is related to the present
indetermination of hadronic models. Hopefully, Auger
data will help to constrain the models.
References
[1] T. Stanev, astro-ph/0611633; D. Allard, et al., Astropart. Phys. 27
(2007) 61.
[2] B. Dawson, Pierre Auger Collaboration, in: Proceedings of the 30th
ICRC, Mérida, 2007.
[3] M. Unger, Pierre Auger Collaboration, in: Proceedings of the 30th
ICRC, Mérida, 2007.
175
[4] M.D. Healy, Pierre Auger Collaboration, in: Proceedings of the 30th
ICRC, Mérida, 2007.
[5] M. Ambrosio, et al., Astropart. Phys. 24 (2005) 355;
M. Ambrosio, et al., in: Proceedings of the CRIS 2004, Nucl. Phys. B
(Proc. Suppl.) 136 (2004), astro-ph/0410179;
M. Ambrosio, et al., in: Proceedings of Vulcano Workshop 2004,
astro-ph/0410184.
[6] D. Heck, J. Knapp, J.N. Capdevielle, G. Schatz, T. Thouw, Rep.
FZKA 6019 (1998) Forshungszentrum Karlsruhe.
[7] T. Pierog, et al., Nucl. Phys. B (Proc. Suppl.) 151 (2006), in:
Proceedings of the 13th International Symposium on Very HighEnergy Cosmic Ray Interactions at the NESTOR Institute, Pylos,
Greece, 6–12 September 2004, astro-ph/0411260.
[8] E. Parizot, Pierre Auger Collaboration, in: Proceedings of the 29th
ICRC, Pune, vol. 7, 2005, pp. 71–74.
[9] L. Prado, et al., Nucl. Instr. and Meth. 545 (2005) 632.
[10] H.J. Drescher, G. Farrar, Phys. Rev. D 67 (2003).
[11] S. Ostapchenko, Nucl. Phys. B (Proc. Suppl.) 151 (2006) 143.