Muon-induced neutrons: how deep should we go to get rid of them?

1. Muon-induced neutrons: how deep should we go to get rid of them? V. A. Kudryavtsev Department of Physics and Astronomy University of Sheffield Vitaly Kudryavtsev

2. Outline • Introduction: what are muon-induced neutrons and why are they dangerous for sensitive experiments? • Simulations of muons. • Simulations of neutrons from cosmic-ray muons: methods, codes. • What was done in the paper by Mei&Hime? • Conclusions. Vitaly Kudryavtsev

3. Production of neutrons by muons • Mechanisms: • Photonuclear interactions of muons (muon inelastic scattering, photonuclear disintegration). • Production of neutrons by hadrons in muon-induced hadronic cascades (showers) - consequence of muon inelastic scattering. • Production of neutrons by photons in muon-induced electromagnetic cascades - consequence of bremsstrahlung, pair production and knock-on electron production. • Capture of stopping negative muons by a nucleus followed by disintegration of a nucleus - significant only at small depths with a large fraction of low-energy (stopping) muons. • The neutron yield from each mechanism depends on the muon energy and target material: • Capture is important only for stopping (low-energy) muons - small depths; at large depths the fraction of low-energy muons decreases. • Cascades contribute more as the muon energy increases. • Neutron yield from electromagnetic cascades is higher for high Z materials (most electromagnetic interactions of muons are proportional to Z(Z+1)/A). • Neutron yield from hadronic cascades also increases with the increase of A but slower than that from e.m. cascades. • All this should be taken into account in simulations. Vitaly Kudryavtsev

4. Muon simulations • It is difficult to achieve good accuracy without knowledge of the rock composition and proper Monte Carlo. • Rock composition (<Z>, <A>) is as important as the mean density, because muon cross-sections (energy losses) for bremsstrahlung and pair production are proportional to Z(Z+1)/A, whereas for ionisation to Z/A, for inelastic scattering to A0.9. • Measurements of the total muon flux are important for normalization and for checking Monte Carlo. • Without proper measurements and simulations the accuracy of muon flux/spectra/angular distribution estimates can be not better than 30% or even worse for large depths. • With proper normalization and accurate Monte Carlo it can be improved to a few percent: muon transport can be done to 1% accuracy assuming that the cross-sections are known to better precision. • Monte Carlo for muon propagation: MUSIC code - Antonioli et al., Astrop. Phys. 7 (1997) 357, Kudryavtsev et al. Phys. Lett. B 471 (1999) 251. • If the transport has been done for a certain type of rock, then MUSUN code for muon simulations underground can be used which uses the results of muon propagation (done for standard rock, Gran Sasso, Boulby etc.). Vitaly Kudryavtsev

5. Flux estimates and their accuracies D.-M. Mei and A. Hime, Phys. Rev. D 73 (2006) 053004. • Data from different types of rock are fitted with exactly the same function. • In reality the difference between standard and Gran Sasso rock (for the same density) is 5% at 3 km w.e., 10% at 5 km w.e. and 15% at 7 km w.e. • Claim is made that the fit (not simulations) is accurate to within 5%!!! Vitaly Kudryavtsev

6. Flux estimates and their accuracies • Without careful selection of data for comparison, even accurate Monte Carlo may differ from some experimental results by as much as 40%. • This does not mean that the simulation are inaccurate. It may mean that the conversion from the measured muon rate in a particular rock to the muon intensity in the standard rock is not done accurately. Tang et al., hep-ph/0604078, to be published in Phys. Rev. D Vitaly Kudryavtsev

7. Muon energy spectra Table from V. A. Kudryavtsev et al., NIMA, 505 (2003) 688. • An accurate Monte Carlo simulations are needed to achieve the precision ~(1-3)%. • All parameterizations are usually derived for standard rock and do not work to a few % accuracy for other rocks. • Different parameterizations sometimes do not agree with each other (mean energies of 253 GeV and 278 GeV for LNGS as in Mei&Hime) and were not checked against more accurate Monte Carlo codes or experimental data (not easily available). Taken by Mei and Hime, PRD 73 (2006) 053004 from earlier papers. Vitaly Kudryavtsev

8. Total neutron yield • Usually expressed as the average number of neutrons produced by a muon crossing 1 g/cm2 of target material. • It is measured or simulated by counting neutrons produced by many muons along long tracks. • Dark matter experiments should worry about fast neutrons ( > 0.5-1 MeV) but we cannot measure the yield of fast neutrons because their energy degrades during transport in matter. • The flux of fast neutrons can be measured by scintillators or dark matter detectors but is possible to do only at certain (quite big) distances from the muon track. At small distances neutron signal may coincide with the muon signal is they are detected by the same detector. • Delayed signal from neutron capture is usually used to detect neutrons. It provides information about neutrons with all energies which were thermalised and captured. • Total neutron yield is used to check Monte Carlo codes which can then be used to simulate fast neutrons. Vitaly Kudryavtsev

9. Muon-induced neutrons: A-dependence GEANT4 Coll. NIMA, 506 (2003) 250. Fasso et al. Proc. MC2000 Conf., p. 159; p. 995. A-dependence of neutron production rate - GEANT4: Araujo et al., FLUKA: Kudryavtsevet al. FLUKA gives twice as many neutrons compared to GEANT4 in most materials tested. Contribution of different processes in various materials - GEANT4: Araujo et al. NIMA 545 (2005), 398. Vitaly Kudryavtsev

10. Muon-induced neutrons: processes • Contribution of different processes: real photonuclear disintegration dominates in GEANT4 at all energies and for (almost) all materials. FLUKA: Wang et al. PRD, 64 (2001) 013012 GEANT4: Araujo et al. NIMA 545 (2005), 398 Vitaly Kudryavtsev

11. Muon-induced neutrons: GEANT4 vs FLUKA From Mei&Hime. The authors do not like the inconsistency between LVD data and other experiments. They do not trust FLUKA model either. Corrections to LVD result and FLUKA model have been suggested. Araujo et al., NIMA 545 (2005), 398. FLUKA (Paper 1) - Kudryavtsev et al. NIMA, 505 (2003) 688. FLUKA (Paper 2) - Wang et al. PRD, 64 (2001) 013012. Vitaly Kudryavtsev

12. Corrections to LVD data? LVD Collaboration, hep-ex/9905047; Proc. of ICRC. LVD data at > 2 m • Left: Energy spectrum of neutron energy depositions (proton recoils) - arb. units. • Can be measured only at large distances from the muon track or cascade. • Cannot be (and was not) used to determine the total neutron yield. • Right: neutron spectrum, converted from proton spectrum using known quenching. > 1 m Filled circles -FLUKA > 2 m Kudryavtsev et al., NIMA, 505 (2003) 688. Measured energy Neutron energy Vitaly Kudryavtsev

13. Corrections to LVD data? • Mei&Hime suggestion: • ‘Neutron spectrum’ published in hep-ex/9905047, was not corrected for proton quenching factor - correct, but this was not a neutron spectrum but spectrum of ‘measured energies’. • As proper neutron spectrum should take into account the quenching of protons, whereas the spectrum in hep-ex/9905047 did not, then the absolute neutron yield as calculated from this spectrum, was also wrong - this is the wrong statement since absolute neutron yield was not (and cannot) be determined from neutron (or proton recoil) spectra: these spectra can be measured only at large distances from the muon track, otherwise the energy detected is not the neutron energy but muon or cascade energy, but neutron flux is small at large distances from the muon track, so the efficiency is small (and not known). Neutrons close to the track give higher contribution to the total neutron yield which was measured by the LVD experiment. Vitaly Kudryavtsev

14. LVD data analysis • Size of a scintillator - 1.511 m3. • Trigger at about 5 MeV for inner scintillators; trigger pulse is mainly from a muon or muon-induced cascade, rarely from a neutron at large distance from the track. • Time gate of ~600 s to record a 2.2 MeV photon from neutron capture on hydrogen (threshold within the gate at about 0.5 MeV). • Fit to the time delay distribution gives the rate of random coincidences and the neutron rate which then can be converted into the neutron yield per muon per unit of muon track length. • The result does not depend on the accuracy of the measurements of neutron spectrum. Similar analysis has been used in the LSD experiment - higher neutron yield found. Vitaly Kudryavtsev

15. Are all these measurements correct? • Assumption made by Mei&Hime about LVD data is not correct. • This does not mean that LVD data are absolutely correct. • The main disadvantage of all experiments at large depths (not only LVD) is the absence of proper and accurate Monte Carlo simulations. • All experiments should model the detectors and neutron production, propagation and detection in order to estimate the detection efficiency, contribution of different materials etc. • It is very difficult to estimate the neutron yield without detailed Monte Carlo. We are simulating not exactly what has been measured. • LVD and LSD are similar detectors (LVD is much bigger) and used similar analyses to estimate the neutron yield. The discrepancy between them is a real puzzle but can be explained (if Monte Carlo is done) by: • Different energy thresholds; • Different contribution from iron and/or rock; • … Vitaly Kudryavtsev

16. Modifications to FLUKA • Mei&Hime suggested to increase the neutron multiplicity to match the higher measured neutron yields with the reference to two papers where such increase in multiplicity (compared to FLUKA predictions) has been observed: • D.-M. Mei, PhD thesis, University of Alabama (2003) - cannot be checked but probably no measurements of multiplicity. • Bezrukov et al. Sov. J. Nucl. Phys., 17 (1973) 51 - no FLUKA simulations there; only measurements. • Higher multiplicity means that either each neutron should have smaller energy or there should be more energy transferred to neutrons (to secondary particles) - the model has changed (not FLUKA model anymore). Vitaly Kudryavtsev

17. Results for different depths and materials • Amazing fit results for different depths (different rocks) despite a large spread of neutron yields for different materials (graph on the left). Vitaly Kudryavtsev

18. Neutron energy spectra From V. A. Kudryavtsev et al., NIMA, 505 (2003) 688. • Neutron energy spectra are different in different materials. • Typically higher neutron yield result in softer spectrum (less high energy neutrons). • Should be checked for all our labs and other materials. CH2 NaCl Vitaly Kudryavtsev

19. Neutron energy spectra: GEANT4 CH2 Pb • Neutron production spectra from 270 GeV muons in CH2 (left)and lead (right) - GEANT4: Araujo and Kudryavtsev (talk at IDM2004). • Energy spectra are different for different materials but the main difference is at low energies: fast neutron yield (> 10 MeV) is not much different for CH2 and lead. • Neutron flux at >1 MeV in NaCl is 50% lower than in Gran Sasso rock (FLUKA: Carson et al., Wulandari et al., hep-ex/0401032). At low energies the difference is bigger. Vitaly Kudryavtsev

20. Fast neutron fluxes Neutron fluxes coming from rock at different energies in units of 10-9 cm-2 s-1. All results are from simulations with FLUKA. Vitaly Kudryavtsev

21. Energy spectrum and lateral distribution Neutron production spectrum - GEANT4: Araujo et al., FLUKA: Wang et al.; data - LVD: LVD Collab., Proc. 26 ICRC (1999), vol. 2, p. 44; hep-ex/9905047. LVD and FLUKA spectra are in arbitrary units and are normalised to the GEANT4 simulations. Neutron lateral distribution (from muon track) - GEANT4, FLUKA and LVD data. Simulations did not include detector specific features. LVD data are in arbitrary units normalised to the simulations. See also Menghetti et al. Proc. IDM2004, for more LVD data and another analysis technique. Vitaly Kudryavtsev

22. Muon-induced neutrons: problems • Differential cross-section of neutron production in thin targets for 190 GeV muons (En>10 MeV). Upper (thick) histograms - GEANT4; dashed line - FLUKA (Araujo et al.); data - NA55 (Chazal et al. NIMA, 490 (2002) 334). • Other data for lead (Bergamasco et al. Nuovo Cim. A, 13 (1973) 403; Gorshkov et al. Sov. J. Nucl. Phys., 18 (1974) 57) are old and controversial but also show significantly higher neutron production compared with simulations. Mei&Hime claimed an agreement with Chazal et al. results. Vitaly Kudryavtsev

23. Is there anything else on the market? • Galbiati and Beacom, hep-ph/0504227: FLUKA calculation of the (n,p) reactions on carbon and oxygen (oil and water-based detectors) and comparison with KamLAND measurements (Phys. Rev. Lett., 94 (2005) 081801) - good agreement found: 63.6 events/kt/day (calculated) vs ~60 events/kt/day (measured) - relevant to high-energy neutrons from muons En>10 MeV. • Ahmad, Brown University - PhD Thesis, 2002 (http:www.sno.phy.queensu.ca/sno/publications.html): Muon-induced neutron rate at SNO - measurements 10.030.64 n/day/kt (Nn<15) or 27.130.64 n/day/kt (all Nn) translated to 3.610-4 n/muon/(g/cm2) or 9.710-4 n/muon/(g/cm2). FLUKA gives 3.410-4 n/muon/(g/cm2) (Kudryavtsev) - 1/3 of the highest reported value; in agreement with the smaller one. • CDMS (Akerib et al., astro-ph/0507190): may be less neutrons than expected - 0 measured, about 2.80.5 conservative estimate. More data and simulations are coming. • More data (with simulations) are welcome to check neutron production by muons in different materials (especially with high A): LVD, SNO, KamLAND, Borexino CTF? Vitaly Kudryavtsev

24. salt m lead n xenon C10H20 air Geometry • Site specific (detector specific) simulation. • Sampling muons according to angular distribution and energy spectrum at specific site, propagating muons, generating secondaries, propagating secondaries, everything is detected. • Sampling neutrons (as Mei&Hime): where? Around the detector? Or on sides only? How do you know where the tracks of muons and secondaries are and whether they will be detected by the target or veto? Vitaly Kudryavtsev

25. Geometry and neutron angular distribution • Distributions peaked at 0o relative to the muon track. How the direction of a neutron is sampled if muon track is not known? From Mei&Hime 2006 From Wang et al. 2001 Vitaly Kudryavtsev

26. Multiplicity distribution • Total neutron yield includes different neutron multiplicities. • How can neutron multiplicity per event be taken into account if neutron production my muons or in a cascade is not simulated properly? From Mei&Hime 2006 From Wang et al. 2001 Vitaly Kudryavtsev

27. Neutron spectra in the lab: FLUKA The figure on the left shows that many neutrons are produced in the lead shielding between the rock and the detector but the enhancement in the neutron flux is mainly at low energies. To suppress this flux we have to put the neutron shielding CH2between the lead and the detector (see figure on the right). Red points - neutron spectrum at rock/cavern boundary. Blue points - neutron spectrum after 30 cm of lead and 50 cm of CH2 shielding. Red points - neutron spectrum at rock/cavern boundary. Blue points - neutron spectrum after 30 cm of lead shielding. Carson et al. Astropart. Phys. 21 (2004) 667; Kudryavtsev et al. NIMA, 505 (2003) 688). Vitaly Kudryavtsev

28. Spectra in the laboratory Neutron spectra in the lab before and after shielding - GEANT4: Araujo et al.; also Araujo&Kudryavtsev, talk at IDM2004; FLUKA: Kudryavtsev et al. - good agreement for all energies of interest (within 50%). Vitaly Kudryavtsev

29. Events in xenon detector Energy spectrum of all events - GEANT4 and FLUKA: Araujo et al. Nuclear recoil event rate as a function of measured energy (quenching = 0.2 for xenon) - GEANT4 and FLUKA: Araujo et al. - good agreement (within 30%). Only <10% of nuclear recoil events contain nuclear recoils only; others have larger energy deposition from other particles. To estimate the background from nuclear recoils only, all particles should be produced, propagated and detected. Vitaly Kudryavtsev

30. Events in xenon detector From Araujo et al.NIMA 545 (2005), 398; hep-ex/0411026. Neutron signal will be in time coincidence with ‘electromagnetic’ signal (fast neutrons) - no big delays. If there is an active veto (100 keV threshold, 4 geometry) - no signal in anticoincidence with veto - a limit of <0.5 ev/year. Vitaly Kudryavtsev

31. Events in Ge • Nuclear recoil spectra in Ge: flat?! This has to be checked! • Result without muon propagation and secondary particle production. • No explanation of the neutron sampling procedure: angle, position, neutron multiplicity per event. • Very hard neutron spectra (other FLUKA and GEANT simulations do not show this). Needs to be checked for other sites! • Neutrons from rock only. • Only fast neutrons >10 MeV? From Mei&Hime, 2006 Simulations for 1 kg of Ge and 0.2 kg of Si: this set-up will never reach 10-10 pb because of the small mass! Vitaly Kudryavtsev

32. Summary • Data and Monte Carlo: • FLUKA and GEANT4 agree within a factor of 2 (even better for some materials and energies). • Most experimental data (although with large uncertainties) are also in agreement with simulations within a factor of 2; some data show significantly higher neutron yield in heavy materials. • Mei&Hime paper: • Too optimistic about muon simulation precision (no accurate Monte Carlo). • Assumption about LVD results is wrong (included in the paper despite my explanation given to Andrew in Seattle in June 2005 - prior to preprint publication). • Too little information about the procedure for neutron sampling (see above). • Neutron spectra are very hard - in contradiction with other FLUKA and GEANT4 simulations. • Only fast neutrons from rock. • Small mass of Ge. • Conclusions are not justified. Vitaly Kudryavtsev

33. Summary • Designing dark matter detector: • 40 g/cm2 of CH2 (after lead) suppress the muon-induced neutron flux down to a few events/year/tonne. • Self-vetoing is important: most neutron events have accompanying particles - gammas, electrons, muons (not so important if only neutrons from rock are considered but we need to consider all neutrons and all details of a detector - veto etc.). • (3÷5)10-10 ((1÷2)10-10) pbis reachable at~2.5 (~3.0) km w. e.even without active veto (tested for a large xenon detector and NaCl as a rock). However, active veto is good for diagnostic and against neutrons (gammas) from the detector components. • Plenty of work to do for ILIAS labs to prove that they are well suited for high sensitivity experiments: to check neutron spectra in various rocks and other materials, different targets for experiments etc. The work has started but needs more efforts. • Accurate measurements of muon-induced neutron flux (with detailed Monte Carlo)? Vitaly Kudryavtsev