You are on page 1of 4


Studying shower to shower fluctuations with simulations

Patricia M. Hansen∗, Ricardo A. Vazquez†, Jaime Alvarez-Muñiz†

∗ Dep. de Fı́sica, IFLP CONICET Facultad de Ciencias Exactas-Univ. Nacional de La Plata C.C. 67, 1900, Argentina
† Depto. de Fı́sica de Partı́culas & Instituto Galego de Fı́sica de Altas Enerxı́as,
Universidade de Santiago de Compostela, 15782 Santiago de Compostela, Spain.

Abstract. The study of shower to shower fluc- ∼ 1011 , being almost impossible to store all the neces-
tuations is hampered by the thinning algorithm, sary data for such amount of particles. EAS simulations
which introduces artificial fluctuations and distorts at the highest energies are only possible because they use
the probability distribution. We consider the effect a statistical sampling algorithm called thinning. Thinning
of the thinning in the calculation of the shower to consists on propagating only a small fraction of the
shower fluctuation for the electromagnetic and the total number of particles in the shower and to assign
muonic component. We also report a complete study statistical weights to the sampled particles to compensate
of the parameters that are affected by the thinning for the rejected ones. However that thinning procedure
and semi-analytical estimations of this fluctuation. introduces artificial fluctuations in the parameters of
Keywords: Fluctuation simulation thinning the showers. For this reason the study of fluctuations
using Monte Carlo simulations is quite difficult because
I. I NTRODUCTION they are strongly affected by the thinning in a not very
well controlled manner. An incorrect assumption on
Extensive air showers (EAS) have been studied over the fluctuation in the shower observables can lead to
the last 70 years [1]. They result from the interaction in systematic uncertainties on their measurement.
the atmosphere of high-energy particles arriving from In this work we study the shower to shower fluctua-
space. The products of these collisions are a set of tions using Monte Carlo simulations of shower develop-
secondary particles sharing the energy of the primary ment. In particular we study how the thinning algorithm
particle initiating the shower. These secondaries move affects the calculation of the fluctuations and we present
through the atmosphere and interact again generating a method to approximately separate the fluctuations
new secondaries. The process continues and the number introduced by thinning from those due to the intrinsic
of particles increases, until their energies are too low to fluctuations of the physical processes that occur in the
continue with the particle generation. In this cascading shower. This paper is organized as follows: In section
phenomena, the complexity and the poor knowledge of II we describe the shower simulations and the thinning
the hadronic interactions at very high energy make the algorithm. In section III we describe the fluctuations and
study of the development of the shower quite difficult. classify them. Our results are reported in section IV. We
The secondary particles can be detected with ground summarize our conclusions in section V.
based detectors and correlated with the primary particle.
Measurements of the electron and muon density, the II. E XTENSIVE AIR SHOWER SIMULATIONS
arrival time of the particles, and the depth at which We have simulated EAS using the Aires code [2],
the shower has the maximum number of particles pro- [3] which makes use of a thinning algorithm [5], in
vides valuable information on the primary energy, arrival which the weights assigned to the sampled particles are
direction and mass composition of the primary parti- adjusted in such a way that both the total energy and
cle. The physical interpretation of these observables is the average number of particles are guaranteed to be
however not easy because even primary particles with conserved.
the same energy, composition and direction produce Before the simulation starts the user sets the relative
secondaries having characteristics that vary from shower thinning level Rth . The thinning energy Eth – the energy
to shower. This feature is called the “shower to shower below which the thinning algorithm is applied – is
fluctuations”. An understanding of the shower to shower obtained as Eth =Rth × Ep , where Ep is the energy
fluctuations will help to improve the measurement of the of the primary particle initiating the shower. For ultra
composition of the primaries and refine the measure- high energy cosmic ray shower simulations convenient
ments. values for the relative thinning are Rth = 10−5 − 10−9 ,
Besides this, in the case of simulated data, the number the actual choice depends on the calculation to be per-
of particles that are produced in an EAS at ultra high formed. The Rth chosen affects both the simulation CPU
energy is too large as to make impossible the propagation time and the size of its output, both typically behaving
of all the secondaries. For instance, the total number linearly with Rth
. If for instance Rth is increased by
of particles in an EAS initiated by a 10 EeV proton is a factor of 10 the simulation speeds up and the size

fluctuation in the depth where the first interaction occurs ticles (W (HADRONIC)) are defined. HANSEN et al.6 0.1 0. as is customary. and will not be considered in this work. (1) In the case of real showers.. The default value of the weight factor shower. zenith angle 0◦ and relative will not be considered further here. which are detector dependent values unless otherwise indicated. fluctuations in the simulated showers (see more details • Artificial fluctuations: Induced by the thinning and about the thinning algorithm in the Aires Manual [3] ). AEH = . 10−6 and 10−7 (10−7 for proton showers On the other hand. (a) (b) Fig. Number of showers. of the output file is reduced by about the same factor.8 Thinning = 10 + 0. taken from rmin =912 m to rmax =1092 m where the .08 0. we will use these default source of fluctuations. inelasticity. different weight “fluctuations in the first interaction” as those due to the factors for electromagnetic (W (EM)) and heavy par. In Aires. The black solid line corresponds to simulation using thinning level Rth = 10−5 . R ESULTS III.14 Sigma (e + e ) Thinning = 10 -6 - -5 0. etc. 0. These affect the interpretation of to the depth of the first interaction point (see below).04 0. properties may emerge when accounting for the fluctu- cal weight factor can be controlled by the user.. we will not consider the effect of unthinning [7]. and primary energy 10 EeV. thinning 10−5 .3 0. 1 Sigma (e + e ) - Electrons and positrons at ground. SHOWER TO SHOWER FLUCTUATIONS WITH SIMULATIONS Electrons and positrons at 1000 m from the shower core. etc. the thinning algorithm is further comple. Here we refer to optimize the procedure of sampling. The detector response introduces an additional W (EM) is 12. keeping in mind that other fluctuations arise in the W (HADRONIC) is specified indirectly through the first interaction. 1 we show the relative RMS (σ/average) of Rather generally we can make a simple classification the total number of electrons at ground and for electrons of fluctuations in shower development: falling in a ring at a distance r =1000 m from the shower • Physical fluctuations: Due to fluctuations in the core. sampling fluctuations.5 0.7 -7 Thinning = 10 0. The red dotted line corresponds to simulations with Rth = 10−6 . The ring was interaction and of the secondary interactions. of the first tion of the number of showers simulated.06 0.9 Thinning = 10 -5 + 0. such as those induced by multiplicity ratio (see [3]) and inelasticity which we do not include under this W (EM) definition. fluctuations are enlarged W (HADRONIC) due to the response of the detector and to the fact that the The default value of this ratio in the simulation is detector typically samples a small fraction of the whole AEH = 88. The “physical fluctuations” can be further splitted into mented with an “extended thinning algorithm” which those due to the first interaction and those induced in the was designed to guarantee that the statistical weights secondary interactions. 1. IV. The sampling We have simulated proton and iron showers with fluctuation is a statistically well-known problem. This splitting is are always smaller than a certain positive number.1 0.4 0. Relative RMS (σ/Average) of the total number of electrons (left) in the ground (left) and in a ring at r = 1000 m around the shower axis (right) versus the shower number. The parameter only. In this paper. an external parameter.02 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Number of showers. called statisti.12 Thinning = 10 -6 Thinning = 10 -7 0. To study the fluctuations due thinning procedures.2 P. multiplicity. In this work depth. In both cases the relative RMS is plotted as a func- depth. Experimental fluctuations: Fluctuations due to the • but the price to pay is an enhancement of the artifical response of the detector.M. C LASSIFICATION OF FLUCTUATIONS In Fig. simulated showers are affected only). The blue dashed line corresponds to simulation with Rth = 10−7 . To ations in the first interaction point [6]. Simulations at other zenith angles and energies by artificial fluctuations due to the thinning and un- have also been performed. mainly motivated by the fact that “universal” shower To ensure this. un-thinning (re-sampling) processes. we Monte Carlo simulations and make them difficult to be have also simulated showers with fixed first interaction usable in the generation of artificial events.2 0.

Also since evaluation of the RMS can be done with less than about w̄ ∝ Rth and s ∝ Rth −1 as obtained above. Since −1 ring. i. (2) with those obtained in the simulation. A and with the aid of our simulations we have studied their good agreement is seen for muons and electrons in 10 behaviour with the thinning level and with the distance EeV proton showers simulated with two relative thinning to the shower core. by muon decay which typically carry a smaller weight. as expected. fluctuations due to physical processes in the shower.7 g cm2 ). i. Monte Carlo simulations of shower development are • N̄e : The average number of entries (sampled parti. support to our identification of σphys with the true • Ω: The RMS of the distribution of weights of the physical fluctuations. N̄e → N̄ • N̄ : Average total number of particles (note that and σ → s as expected. easy to understand. it would be equivalent to w particles entering inside N̄e ∝ Rth and Ω ∝ Rth the fluctuations induced by the same ring. Also a thinning factor Rth > 10−5 is not sufficient σphys ∼ constant. We also plot the contribution of σth and σphys to the total σ as predicted The average weight w̄ for electrons and muons is by Eq. im- interaction. physical fluctuations. In Eq. In this case w̄ → 1. Using these parameters it is possible to obtain an analitycal expresion for the fluctuations in the total V. artificial (thinning) fluctuations at least for Rth > 10−7 . and the fluctuations are enlarged. especially for fluctuations in the 2 ring. and are in fact much from the shower core the electrons are mainly produced smaller than the physical fluctuations. Finally. In the case of the electromagnetic the total fluctuations. N̄ = w̄N̄e ). For this purpose we define: expected but it is remarkable that it was obtained from • w̄: The average weight of entries (sampled particles the Monte Carlo simulation.PROCEEDINGS OF THE 31st ICRC. and simulations in which we The relative RMS of the weight Ω/w̄ is found to be have fixed the first interaction depth of the primary almost independent of Rth or equivalently Ω ∝ Rth for particle (proton or iron) to the mean interaction depth small values of Rth .e. The average number of entries is −1 of that primary in the QGSJET [4] model (for proton found to be N̄e ∝ Rth as expected since the thinning the mean free path is ∼ 45 g cm−2 at 10 EeV and for algorithm is made in such a way that N̄ = w̄N̄e (so iron ∼ 10. (2) to shower fluctuations. The plots also clearly display that no reliable – as expected for fluctuations due to thinning. we have also studied the importance of the In the case of the muonic component higher energy fluctuation in the first interaction depth. 2 and 3 we compare the fluctuations predicted • σ: The RMS of the total number of particles. point is allowed to vary. they are the • s: The RMS of the number of entries. (2). by Eq. we would also lose or gain 2 σth ∝ Rth . performed) clearly Ω → 0 and the fluctuations are only cles). both simulations are of comparable value. If we consider an entry (a sampled Each of the two terms in the formula for σ 2 has in particle in the shower) of weight w falling inside the fact a different behaviour with the thinning level. Moreover. so that if due to a physical fluctuation thinning behave as: we lose or gain that entry. This was of course ring around the shower axis. we have: 20 simulated showers. We have seen that for thinning . (2) σphys contains the the fluctuations when the whole ground is considered. levels of Rth = 10−5 and 10−7 . It can be seen that for Rth = 10−5 the found to be simply proportional to the thinning level fluctuations due to thinning are a significant fraction of w̄ ∝ Rth . ŁÓDŹ 2009 3 radii correspond to a logarithmically symmetric interval The proof only assumes that the probability for a sam- around 1000 m to compensate for the steep decrease of pled particle to have a given weight is independent of the density of particles with r. (2). The result is that the fluctuations in that the average number of particles is independent Rth ). (4) to evaluate the RMS. and also that the fluctuations in the ring have a stronger while σth contains the fluctuations associated to the dependence with the thinning level Rth used.e. the fluctuations associated to physical processes do We have further studied the particle fluctuations in a not depend on the thinning level. if Rth → 0 (when full entries. (3) w particles inside the ring. In Figs. This is thinning procedure [9]. It can be seen that the the probability of a shower to have a given number fluctuations in the ring around 1000 m are larger than of sampled particles [8]. which clearly Finally for s/N̄e we obtain that the relative fluctuation shows that we can not see the fluctuation due to the first is approximately independent of the thinning level. due to the second term in Eq. because it is obscured by the much larger plying that s ∝ Rth−1 . while lower energy simulations in which the depth of the first interaction muons arrive far from the core and carry larger weigths. however for Rth = 10−7 these are component w̄ decreases with r because at large distances very much reduced as expected. and therefore gives a strong in the shower) falling in the ring. For this purpose muons are typically produced closer to the shower axis we have performed two types of simulations: Regular and typically carry smaller weights. It can be proven that: We study how the artificial fluctuations introduced by thinning affect the calculation of the physical shower σ 2 = σth 2 2 + σphys = N̄e Ω2 + w̄2 s2 . C ONCLUSIONS number of particles.

We thank Centro de tuations introduced by the thinning procedure. Schmidt. Vazquez and J. Procs. 689 (2000). 30. Pavlov. Zas for discussions and [5] A. manner allows to separate the physical shower to shower R EFERENCES fluctuations from the artificial fluctuations induced by thinning. G.A. Astropart. Ostapchenko.0 (2002). Watson.6 2.3750 [astro-ph] Galicia (PGIDIT 06 PXIB 206184 PR) and Consellerı́a [7] P. . Same as in Fig. T. 27th ICRC (Hamburg) 1 237 (2001). Acad. Supercomputación de Galicia (CESGA) for computer We have obtained a semi-analytical formula which resources. AIRES User’s Manual and Reference simulations for an arbitrary thinning level smaller than Guide. J. 27th Int. (2) (red dashed line). N. Risse. S. and A. I. SHOWER TO SHOWER FLUCTUATIONS WITH SIMULATIONS LDF LDF 2 × 10-1 Sigma of (e ++e-) 7 × 10-1 6 × 10-1 5 × 10-1 4 × 10-1 3 × 10-1 10-1 2 × 10-1 Sigma of (e +e-) + 10-1 σ MC σ MC σ with the formula σ with the formula σ th σ th σ phy 2 × 10-2 σ phy 2 × 10 -2 10-2 10-2 2 2. [8] P. 2001. Phys 72. Cosmic Ray Conference.4 2. 2.4 2. Nucl. 39 (1981). ical shower to shower fluctuations from Monte Carlo [3] S.2 2. J. 4 P. 2 for the total number of muons. Hansen. (Physics) We thank V. Nagano and A. 270 (2008). comments.).8 3 2 2. HANSEN et al.A. Hillas. Germany. J. Bull. CPAN) and Feder Funds.8 3 log (r/m) log (r/m) 10 10 (a) (b) Fig. levels Rth = 10−5 − 10−6 . Fiz. Rev.8 3 2 2. ACKNOWLEDGMENTS (1993). was supported by Juan de la [6] F. Also shown are the contributions to the fluctuations due to thinning σth (green dotted line) and that due to physical processes in the shower σphys (blue dash-dotted line) as predicted by Eq. de Educación (Grupos de Referencia Competitivos – [9] M. Spain. S. of Mod. Right panel: Same as in left panel for 10 EeV proton showers with Rth = 10−7 . 56 (3) 346 1993. www. R. Russ.8 3 log (r/m) log (r/m) 10 10 (a) (b) Fig. 56. the determination of shower Ciencia e Innovación (FPA 2007-65114 and Consolider to shower fluctuation is hampered by the artificial fluc. Ministerio de Hamburg. S.6. Ostapchenko and A.M. Yad.4 2. S. Proc. Sciutto. Sciutto. L. Astropart. version 2. Kalmykov and S. and fluctuations predicted by Eq. Moreover Eq. Phys. (2). Canoa. At. Proc of the Paris Workshop on Cascade simulations.6 2. Rodriguez–Fernandez. Ave.6 2. This work was made possible by the partial gives a good description of the fluctuations in the support of ALFA-EC funds. in the framework of the total number of particles.2 2. (2) allows to obtain the phys. Billoir. Phys.V.4 2. Consolider Xunta de Galicia 2006/51). M. Phys. e-Print: arXiv:0712.unlp. 3. 105 VI. Sci.2 2. [1] M. Left panel: Fluctuations (σ) of the total number of electrons at ground as obtained in Monte Carlo simulations of 10 EeV proton showers with Rth = 10−5 (black solid line). P. [2] S. 58 1966 1994. J. and which in an approximate HELEN Project. H. [4] N.2 2. N. I. also thank Xunta de 29 (2008) 355. Kalmykov. M. Valiño and E.. available electronically at Rth ∼ 10−5 . Cazon.A. Linsley and A. Tarutina.A-M and R.fisica. p. 2 × 10-1 Sigma of (µ++µ -) Sigma of (µ++µ -) σ MC σ with the formula σ th 10-1 σ phy σ MC 2 × 10-1 σ with the formula σ th σ phy 2 × 10-2 10-1 10-2 2 2. Hillas (eds. Chou. Cierva grant.6 2. Alvarez-Muñiz in preparation. N. et al.