On-line recognition of supernova neutrino bursts in the LVD detector
N.Yu.Agafonova, M.Aglietta, P.Antonioli, G.Bari, A.Bonardi, V.V.Boyarkin, G.Bruno, W.Fulgione, P.Galeotti, M.Garbini, P.L.Ghia, P.Giusti, F.Gomez, E.Kemp, V.V.Kuznetsov, V.A.Kuznetsov, A.S.Malguin, H.Menghetti, A.Pesci, R.Persiani, I.A.Pless, A.Porta, V.G.Ryasny, O.G.Ryazhskaya, O.Saavedra, G.Sartorelli, M.Selvi, C.Vigorito, L.Votano, V.F.Yakushev, G.T.Zatsepin, A.Zichichi
aa r X i v : . [ a s t r o - ph ] O c t On-line recognition of supernova neutrinobursts in the LVD detector
N.Yu.Agafonova a , M.Aglietta b , P.Antonioli c , G.Bari c ,A.Bonardi b , V.V.Boyarkin a , G.Bruno f , W.Fulgione b , ∗ , ,P.Galeotti b , M.Garbini c , P.L.Ghia b , f , P.Giusti c , F.Gomez b ,E.Kemp d , V.V.Kuznetsov a , V.A.Kuznetsov a , A.S.Malguin a ,H.Menghetti c , A.Pesci c , R.Persiani c , I.A.Pless e , A.Porta b ,V.G.Ryasny a , O.G.Ryazhskaya a , O.Saavedra b , G.Sartorelli c ,M.Selvi c C.Vigorito b , L.Votano g , V.F.Yakushev a ,G.T.Zatsepin a , A.Zichichi c a Institute for Nuclear Research, Russian Academy of Sciences, Moscow, Russia b Institute of Physics of Interplanetary Space, INAF, Torino, University of Torinoand INFN-Torino, Italy c University of Bologna and INFN-Bologna, Italy d University of Campinas, Campinas, Brazil e Massachusetts Institute of Technology, Cambridge, USA f INFN-LNGS, Assergi, Italy g INFN-LNF, Frascati, Italy
Abstract
In this paper we show the capabilities of the Large Volume Detector (INFN GranSasso National Laboratory) to identify a neutrino burst associated to a supernovaexplosion, in the absence of an ”external trigger”, e.g., an optical observation. Wedescribe how the detector trigger and event selection have been optimized for thispurpose, and we detail the algorithm used for the on-line burst recognition. Theon-line sensitivity of the detector is defined and discussed in terms of supernovadistance and ¯ ν e intensity at the source. Key words:
LVD, Neutrino detection, Supernova core collapse, Burst identification
PACS: ∗ Corresponding author: Walter Fulgione, c/o Laboratori Nazionali del Gran SassoS.S. 17 BIS km. 18.910, 67010 Assergi LAquila - Italy, email: [email protected]
Preprint submitted to Astroparticle Physics 11 November 2018
Introduction
The detection of neutrinos from SN1987A marked the beginning of a newphase of neutrino astrophysics [1,2,3,4]. In spite of the lack of a firmly es-tablished theory of core collapse supernova explosion, the correlated neutrinoemission appears to be well established. However, since this first ν observationwas guided by the optical one, the detector capabilities of identifying a ν burstin the absence of an ”external trigger” should be demonstrated very carefully.In the presence of an electromagnetic counterpart, on the other hand, theprompt identification of the neutrino signal could alert the worldwide networkof observatories allowing study of all aspects of the rare event from its onset.The Large Volume Detector (LVD), in the INFN Gran Sasso National Labo-ratory (Italy), at the depth of 3600 m w.e., is a 1 kt liquid scintillator detectorwhose major purpose is monitoring the Galaxy to study neutrino bursts fromgravitational stellar collapses [5]. Besides interactions with protons and car-bon nuclei in the liquid scintillator, LVD is also sensitive to interactions withthe iron nuclei of the support structure whose total mass is 0.9 kt [6]. Theexperiment has been taking data, under different configurations, since 1992,reaching in 2001 its present and final configuration. Its modularity and rockoverburden, together with the trigger strategy, make this detector particularlysuited to disentangle on-line a cluster of neutrino signals from the background.We will discuss in this paper the LVD performances from the point of view ofthe on-line identification of a neutrino burst: we will describe in section 2 thetrigger of the detector, the event selection and the on-line burst recognition. Insection 3 we will define and discuss the detector sensitivity to neutrino burstswhich, as we will show in section 4, can be expressed in terms of supernovadistance or neutrino intensity at the source. LVD consists of an array of 840 scintillator counters, 1.5 m each. The whole ar-ray is divided in three identical ”towers” with independent high voltage powersupply, trigger and data acquisition (see figure 1). In turn, each tower consistsof 35 ”modules” hosting a cluster of 8 counters. Each counter is viewed fromthe top by three 15 cm photomultiplier tubes (PMTs) FEU49b or FEU125.The charge of the summed PMT signals is digitized by a 12 bit ADC (conver-sion time = 1 µ s) and the arrival time is measured with a relative accuracy of12.5 ns and an absolute one of 100 ns [7].The main neutrino reaction in LVD is ¯ ν e p → e + n, which gives two detectable2ignals: the prompt one due to the e + (visible energy E vis ≃ E ¯ ν e − . e c ) followed by the signal from the np → d γ capture ( E γ = 2 . ≃ µ s,). The trigger logic is optimized for the detec-tion of both products of the inverse beta decay and is based on the three-foldcoincidence of the PMTs of a single counter. Each PMT is discriminated attwo different thresholds resulting in two possible levels of coincidence betweena counter’s PMTs: H and L, corresponding to E H ≃ E L ≃ E vis . The average neutrondetection efficiency, ǫ n , amounts to about 50% for neutrons detected in thesame counter where the positron has been detected [8]. Fig. 1.
LVD schematic view: thebuilding block is a module of 8 scin-tillator counters, 35 modules forma tower. E vis (MeV) (cid:161) H Fig. 2.
Trigger efficiency, averagedover all the scintillator counters,for the H coincidence versus visi-ble energy E vis . One millisecond after the occurrence of a trigger, the memory buffers , con-taining the charge and time information of both H and L signals, are read out.This is performed independently on the three towers, without introducing anydead time. The tower event fragments, if present, are then sent to a centralprocessor which provides event-building requiring time coincidence of the first Each memory buffer, one per module, is able to store up to 512 signals, corre-sponding to 50000 signals in the whole apparatus.
The basis of the search for neutrino bursts is the identification of clustersof signals in fixed time windows. A preliminary step consists in the selectionof good signals detected by good counters; we thus apply to the events twodifferent series of cuts: (A) to the signals, (B) to the counters, namely: (A) Cut on signals (1) The energy of the signals must be in the range E cut ≤ E signal ≤
100 MeV.Two energy thresholds are considered: E cut = 7 MeV, corresponds toan average trigger efficiency > cut = 10 MeV, to an averagetrigger efficiency >
95% (see fig. 2).(2) Events with signals in coincidence in 2 or more counters of the array,within 200 ns, are rejected since they are considered muon candidates .The probability for electron or positron with energy up to 60 MeV totrigger 2 LVD counters is less than 10% [9]. (B) Cut on counters (1) The response to muons of atmospheric origin is used to identify anddiscard not properly working counters. The measurement of muons as-sociated with the CNGS beam [10] provides complementary tests of thedetection capabilities of the detector. A wrong muon counting rate orspectrum can be due to scintillator, PMT or electronics problems (ADCor TDC). Counters rejected for this reason are less than 5% of the to-tal and represents a steady loss of active mass requiring a maintenanceintervention.(2) Counters with a background rate (for E ≥ ≥ · − s − ,during the last two hours of operation, are rejected as noisy. The problemarises from electronics or bad energy calibration and usually regards lessthan 2% of the counters. Also in this case counters involved are almostalways the same, a new energy calibration or a maintenance interventionare required. The typical counting rate distribution for all good counters(after cuts), averaged over 150 days of data taking, is shown in figure 3.(3) Counters that take part too often in a cluster of signals are rejected, andthe cluster redefined. Namely, if m is the cluster multiplicity and N c thenumber of active counters, a counter will be rejected if its multiplicity m i2 This rejection corresponds to a dead time ≃ .
01 %, taking into account that theglobal muon rate in LVD is about 0.1 muon s − . P ∞ k=m i P(k;m/N c ) ≤ · − . This is due tosporadical (of the order of once per month) and local electric noise anddetermines a mass loss ≪ . act , by rejecting counters that are steadily out of order (1 and 2)and/or that are sporadically noisy (3). A snapshot of the detector active massin the last two years is shown in figure 4. In the same period the averagebackground counting rate of the whole array was f bk =0.2 Hz if E cut =7 MeVand f bk =0.03 Hz for E cut =10 MeV. Trigger Rate (10 -2 Hz) N u m b e r o f c o un t e rs Fig. 3.
Counting rate distribution(for E vis ≥ ¯ R = 2 · − s − counter − . M a c t ( t ) Fig. 4.
On-line LVD active massduring the last two year of oper-ation.
At the end of the whole selection procedure the time distribution of the signalsis well described by the Poisson statistics, as can be seen in figure 5, where weshow the difference of the arrival time between successive signals, and in figure6 where we show the fluctuations of the 5 min counting rate, s , with respect tothe local average value (measured during a time window of 40 min) in units ofthe expected error, σ exp , calculated assuming pure Poisson fluctuations. Thetypical experimental distribution, obtained during 100 days of operation, isfitted with a Gaussian one with mean equal zero and σ = 1.01, showing thatthe residual non-Poissonian contribution to the fluctuations σ res = √ σ − For example: in a cluster (m=100, ∆t=20 s) detected when there are 800 activecounters, a counter will be rejected (and the cluster redefined) if 5 or more signalsinside the cluster come from it.
Time difference (s) E n t r i e s Fig. 5.
Distribution of the differenceof the arrival time between succes-sive signals, compared with Pois-son expectations. f E n t r i e s Mean = 0.00 ± ± -4 -2 0 2 4 Fig. 6.
Distribution of the fluctua-tions of the 5 minutes counting rate( f = ( s − ¯ s ) /σ exp ). The superim-posed curve is the free parametersGaussian fit. The statistics repre-sents 100 days of data taking. ν bursts selection algorithm The core of the algorithm for the on-line selection of candidate neutrino burstsis the search for a cluster of H signals within a fixed-duration time window,∆ t . The candidate burst is simply characterized by its multiplicity m , i.e., thenumber of pulses detected in ∆t, and by ∆t itself. All the other characteristicsof the cluster, e.g., detailed time structure, energy spectra, ν flavor content andtopological distribution of signals inside the detector are left to a subsequentindependent analysis. Based on this principle, the LVD data are continuouslyanalyzed by an on-line ”supernova monitor”. In detail, each data period, T ,is scanned through a “sliding window” with duration ∆t = 20 s, that is, itis divided into N = 2 · T ∆ t − ≥ m, due to background, is: F im ( m, f bk , s ) = 8640 · ∞ X k ≥ m P ( k ; 20 · f bk s − ) event · day − (1)6here f bk is the background counting rate of the detector for E vis ≥ E cut , P ( k ; f bk ∆ t ) is the Poisson probability to have clusters of multiplicity k if f bk ∆ t is the average background multiplicity, and 8640 is the number of trialsper day. For example, a cluster with m =10 can be produced by backgroundfluctuations once every 100 years if f bk = 0.03 s − ; to have the same signifi-cance with a higher background, e.g. f bk = 0.2 s − , a cluster with much highermultiplicity, m =22, would be required (see figure 7 where m versus F im isshown, for two different background conditions). For these background rates, m = 10 and m = 22 correspond to the minimum multiplicity, m min , to havean imitation frequency F im < · − yr − .In LVD the search for burst candidates is performed for both energy cuts:7 and 10 MeV. The chosen F im , below which the detected cluster will be anon-line candidate supernova event, is 1 per 100 year working stand-alone whileit is relaxed to 1 per month working in coincidence with other detectors, as inthe SNEWS project [11].In figure 8 we show the distributions of the observed time intervals betweenselected clusters at the imitation frequency F im = 1/day and F im = 1/monthduring 688 days (between July 5 th , 2005 and May 23 rd , 2007) and E cut = 7MeV. The mean rates, derived from Poisson fits, are F obsim = 1 .
24 day − (with m min varying between 13 and 15) and F obsim = 1 .
28 month − ( m min between15 and 18), respectively, meaning that, within 25%, the time behavior of thebackground is consistent with expectations.We can conclude that the background trend is predictable even during long pe-riods of data acquisition and with variable detector conditions, allowing us todefine the significance of a neutrino burst in terms of imitation frequency, F im . Once a candidate cluster (m, 20s) has been identified, the algorithm will searchfor the most probable starting point of the signal within the time window. For allpossible sub-clusters with multiplicity 2 ≤ k ≤ m and duration 0 ≤ ∆ t ≤ s thecorresponding Poisson probability P k ≥ m is calculated and the absolute minimumidentified. The time of the first event of the least probable sub-cluster is assumedas the start time of the signal. im log[Ev/year] -6 -5 -4 -3 -2 -1 0 1 2 m i n i m u m c l u s t e r m u lti p li c it y Fig. 7.
Minimum cluster multiplic-ity m min vs. the imitation fre-quency F im . Triangles correspondto E cut =7 MeV and f bk =0.2 Hz,while squares to E cut =10 MeVand f bk =0.03 Hz. The two verti-cal lines represent the F im thresh-olds of 1 candidate per 100 year(stand-alone) and of 1 per month(SNEWS). Selected at F im =1/daySelected at F im =1/month Delay (days) E n t r i e s Fig. 8.
Distribution of the time in-tervals between observed clusters(histograms) fitted by Poisson laws(lines) for F im = 1 /day (solid) and F im = 1 /month (dashed), during688 days and E cut = 7 M eV . In this section we discuss the sensitivity of the described on-line selection al-gorithm to the recognition of a supernova event. The selection method definesa candidate as any cluster of m ≥ m min signals within a window of ∆ t = 20 s.For a known background rate, m min corresponds to a chosen F im which is setas a threshold. This multiplicity represents the minimum number of neutrinointeractions required to produce a supernova ”alarm”, and contains two terms,one due to the background, f bk ∆ t , and the other due to the neutrino signal.In particular, for LVD, considering only inverse beta decay (IBD) reactions,which are the dominant ones at least in the ”standard” supernova model, andsimply approximating the detector response to E vis = E ¯ ν e − . m min = f bk ∆ t + M act N p ǫ ( E cut ) s Z dt MeV Z E cut +0 . MeV Φ( E ¯ ν e , t ) · σ ( E ¯ ν e ) dE ¯ ν e (2)where: M act is the active mass, N p = 9 .
34 10 is the number of free protonsin a scintillator ton, ǫ ( E cut ) is the trigger efficiency approximated as constant( ǫ = 0 . E cut = 7 MeV and ǫ = 0 .
95 for E cut = 10 MeV, see fig. 2), σ ( E ¯ ν e ) is the IBD cross section[12] and Φ( E ¯ ν e , t ) the differential ¯ ν e intensity8t the detector. The upper limit in the time integral (10 s) corresponds to themaximum unbiased cluster duration.Hence the integral on the right side of (2) is the detector burst sensitivity, S ,in terms of minimum neutrino flux times cross section integrated over ∆ t and∆ E , and is expressed as number of neutrino interactions per target: S E cut = ( m min − f bk ∆ t ) / ( M act · N p · ǫ )The values of S are shown in table 1, for the two LVD thresholds of the im-itation frequency, i.e., F im = 1 per 100 years and F im = 1 per month, twodifferent masses, M act = 1000 t and 330 t, and two values of E cut . As it canbe seen an important improvement can be obtained by increasing the energycut from 7 to 10 MeV. As was shown in figure 7, the minimum cluster multi-plicity, for example at F im = 1 per 100 years, goes from 22 to 10, allowing animprovement of almost a factor of two in the sensitivity S E cut .It must be noted that in the on-line algorithm described so far we have ne-glected the capability of LVD to detect both products of the IBD reaction (seesection 2.1). We can consider the signature of the reaction to build differentburst selection algorithms. For example, we can require that all the H signalsin the cluster are ”signed”, i.e., accompanied by a delayed L one (algorithmIBD-A). In the definition of the imitation frequency (eq. 1) f bk is substitutedby f bk · ¯ P Lbk (being ¯ P Lbk = 0 .
13 the probability for a H signal to be followed,in the same counter, by a L one due to background, averaged over the wholearray) and, in (eq. 2), ǫ E cut by ǫ E cut ǫ n . The sensitivity of the algorithm IBD-A is shown in Table 1: even if the minimum multiplicity is lower, and thebackground rate is reduced, because of the n -capture efficiency ( ǫ n = 0 .
5) theIBD-A method has comparable effectiveness or even less then the on-line one.We can also build several different algorithms, intermediate between the on-line and the IBD-A ones, requiring that only a fraction of the H signals inthe cluster are accompanied by L ones (IBD-B) . However, even if the IBD-Bmethod efficiency results higher (as can be seen in Table 1, where an exam-ple is given), it does not exceed the on-line one enough to justify the loss ofsimplicity of the on-line algorithm and its independence from the model ofsupernova neutrino emission. Moreover, the on-line algorithm is sensitive toall possible neutrino interactions in LVD, both in the liquid scintillator andin the iron structure (that can represent up to 15% of the total number ofinteractions). We reject all the clusters which have a number of ”signed” H signals ≤ k , suchthat: P r = kr =0 P ( r, m, p ) ≤ P , where P ( r, m, p ) is the binomial probability to have rsigned pulses in a cluster of multiplicity m . We choose P = 0 . P = 0 corresponding to the on-line algorithm. Discussion and conclusions
The Large Volume Detector, located in INFN Gran Sasso National Laboratoryhas been designed to detect neutrino bursts from galactic gravitational col-lapses. In this paper we have described how the trigger and the event selectionhave been optimized to recognize on-line a neutrino burst from a supernovaexplosion even in the absence of an ”external trigger”, such as the presence ofan optical counterpart. The fast identification of such a neutrino signal is ofutmost importance especially in view of the LVD participation to the SNEWSproject, which should promptly alert the worldwide network of observatoriesto allow the study of the rare event since its onset. We have discussed theon-line algorithm currently in use at LVD and we have defined its sensitivityS in terms of minimum number of interactions per target unit requested toproduce a burst alarm.If one assumes a model for the neutrino emission and propagation, it is pos-sible to express the sensitivity in terms of physical parameters of the source,such as its distance or the emitted neutrino flux. In particular we will use herethe signal detected by Kamiokande-II and IMB for SN1987A. Following [13](see table 1 herein), [14] and [15], we adopt these values for the astrophysicalparameters of SN1987A: average ¯ ν e energy < E ¯ ν e > =14 MeV; total radiatedenergy E b = 2.4 · erg, assuming energy equipartition; distance D=52 kpcand average non-electron neutrino energy 10% higher than ¯ ν e [16]. Concerningneutrino oscillations (see [6] for a discussion), we consider normal mass hier-archy. We calculate the number of inverse beta decay signals expected from aSN1987A-like event occurring at different distances, for E cut = 7 and E cut = 10MeV, and for two values of the detector active mass, M act = 330 t and M act =1000 t. For example, the number of detected IBD events for such a supernovain the center of the Galaxy (D = 10 kpc) is 230 for E cut = 7 MeV and M act =1000 t. Taking into account Poisson fluctuations in the signal multiplicity, wederive the on-line trigger efficiency as a function of the distance: this is shownin figure 9 (lower scale) for LVD working stand-alone and in the SNEWS.On the other hand, if we fix the distance of this supernova, e.g. at 10 kpc, wecan derive the LVD sensitivity in terms of the minimum neutrino intensity atthe source, varying the total emitted energy. The on-line trigger efficiency, asa function of neutrino luminosity (in terms of percentage of SN1987A one) isshown again in figure 9, upper scale. For example LVD working stand-alone,with 1000 t of active mass and for E cut = 10 MeV, is sensitive (at 90% c.l.) to aneutrino luminosity equivalent to 6% of that of SN1987A placed at a distanceof 10 kpc. 10 istance [kpc]
10 20 30 40 50 60 70 80 90 100 o n - li n e t r i gg e r e ff i c i e n cy
100 25 11 4 2 1 % Distance [kpc]
10 20 30 40 50 60 70 80 90 100 o n - li n e t r i gg e r e ff i c i e n cy
100 25 11 4 2 1 % Fig. 9.
On-line trigger efficiency versus distance (lower scale) and per-centage of SN1987A signal at 10 kpc (upper scale) for E cut =7-10 MeV(light green and dark blue lines, respectively) and M act = 330 t (dotted)and 1000 t (continuous). LVD stand-alone on the left panel and in theSNEWS on the right one.
We can conclude that, without introducing any further check on the timestructure, energy spectra and ν flavor content of the signals in the cluster(which are postponed to the off-line analysis), LVD is able to identify on-line neutrino bursts from gravitational stellar collapses occurring in the wholeGalaxy ( D ≤
20 kpc) with efficiency > ≤ We thank Francesco Vissani for fruitful discussions.
References [1] K.S. Hirata et al. , Phys. Rev. Lett. (1987) 1490 and K.S. Hirata et al. , Phys.Rev. D (1988) 448.[2] R.M. Bionta et al. , Phys. Rev. Lett. (1987) 1494.[3] E.N. Alekseev, L.N. Alekseeva, V.I. Volchenko and I.V. Krivosheina, JETPLett. (1987) 589 and Phys. Lett. B (1988) 209.[4] M. Aglietta et al. Europhys. Lett. (1987) 1315 and V.L. Dadykin et al. , JETPLett. (1987) 593; [Prisma Zh. Eksp. Teor. Fiz. 45 (1987) 464].
5] M. Aglietta et al. , Il Nuovo Cimento A (1992) 1793.[6] N.Yu. Agafonova et al. , Astroparticle Physics (2007) 254-270; [hep-ph/069305].[7] A. Bigongiari, W. Fulgione, D. Passuello, O. Saavedra and G. Trinchero, NIMA (1990) 529.[8] A. Porta, Ph.D. Thesis, University of Torino, 2005.[9] P. Antonioli, W. Fulgione, P. Galeotti and L. Panaro, NIM A (1991) 569.[10] M. Aglietta et al. , NIM A (2004) 96; N.Yu. Agafonova et al. , LNGS/EXP-01/07, submitted to EPJ C.[11] P. Antonioli et al. , New Journal of Physics (2004) 114; [http://snews.bnl.gov].[12] A. Strumia, F. Vissani, Nucl. Phys. B (2005) 294.[13] B. Jegerlehner, F. Neubig and G. Raffelt, Phys. Rev. D , (1996) 1194;[astro-ph/9901111].[14] M.L. Costantini, A. Ianni and F. Vissani, Phys. Rev. D (2004) 043006;[arXiv:astro-ph/0403436].[15] M.L. Costantini, A. Ianni, G. Pagliaroli and F. Vissani JCAP05(2007)014;[astro-ph/0608399(2006)].[16] M.T.Keil, G.G.Raffelt and H.T.Janka, Astrophys. J. (2003) 971;[arXiv:astro-ph/0208035]. able 1Burst sensitivity for different selection algorithms, for two energy thresholds E cut and for two values of imitation frequency; M act is expressed in ton; m min representsthe minimum cluster multiplicity (in parenthesis the minimum number of requested”signed” H signals for the IBD-B algorithm). F im = 1 month − F im = 1 · − year − E cut =7 MeV algorithm M act m min S E cut m min S E cut on-line 1000 18 1 . · −
22 2 . · −
330 10 3 . · −
14 4 . · − IBD-A 1000 7 1 . · −
10 2 . · −
330 5 3 . · − . · − IBD-B 1000 15(5) 1 . · − . · −
330 9(3) 2 . · − . · − E cut =10 MeV on-line 1000 8 8 . · −
10 1 . · −
330 5 1 . · − . · −31