update
play

Update Nick Amin Nov. 14, 2019 Overview Last update These slides - PowerPoint PPT Presentation

Update Nick Amin Nov. 14, 2019 Overview Last update These slides L1 selection Misc. checks of muon chi2, track isolation Signal MC "scan" Start looking into GP regression 2 L1 selection Finalize L1


  1. Update Nick Amin 
 Nov. 14, 2019

  2. Overview ⚫ Last update ⚫ These slides • L1 selection • Misc. checks of muon chi2, track isolation • Signal MC "scan" • Start looking into GP regression 2

  3. L1 selection ⚫ Finalize L1 selection • OR of ‣ L1_DoubleMu4p5_SQ_OS_dR_Max1p2 ‣ L1_DoubleMu0er1p4_SQ_OS_dR_Max1p4 ‣ L1_DoubleMu_15_7 • Removed L1_DoubleMu_15_5_SQ because it was not part of HLT inputs, even though it was active/ unprescaled • Not updated yet (next set of babies) 3

  4. Plot dump ⚫ http://uaf-10.t2.ucsd.edu/~namin/dump/plots/plots_split_v4/ ⚫ Plot dump of various variables split by • nothing ("splitnone_") • near or away from pixel material ("splitpixel_") • DV rho ("splitrho_") • eta of dimuon system ("spliteta_") • pT of dimuon system ("splitpt_") • mass of dimuon system ("splitmass_") • dphi(mu1,mu2) ("splitdphimumu_") ⚫ Selection • ==2 OS muons • ==1 DV • DV rho > 1cm • cos( 𝛦𝜚 (dimuon,DV))>0 • max(DV x error,DV y error)<0.05cm • DV z error<0.1cm • # valid muon hits > 0 for both muons 4

  5. Muon chi2 ⚫ Plot muon chi2 for subleading muons for 1<m 𝜈𝜈 <4 separated into on- and o ff -j/psi • Comparing DY and j/psi so they look the same (o ff and on) ⚫ Also include signal MC, and data on-j/psi with p T , 𝜃 reweighted to MC 5

  6. TrackIso == 1 ⚫ Track isolation computation based on the hltIter2L3MuonMergedNoVtx track collection • Calculated within a cone of 𝛦 R<0.3 for tracks within dz<0.2 and d0<0.1 • Exclude 𝛦 R<0.01 tracks • Exclude leading track if it has p T >2 GeV and 𝛦 R<0.025 ⚫ Stored ScoutingTrack collection from hltIterL3MuonAndMuonFromL1MergedNoVtx ⚫ Isolating the peak around 1, I couldn’t find a track close to the subleading muon (in 𝛦 R and p T ) in the stored collection. • However, sanity cuts on DV errors eliminate the majority of the peak at 1 max(DV x error, DV y error) DV z error 6

  7. Signal MC ⚫ Generated 100k events for a handful of lifetimes and masses (still H(125) → Z d Z d → 2 𝜈 +2X) • E ffi ciency of gen-level fiducial denominator requirement wrt full sample is ~64% • Reco*trigger e ffi ciency below 7

  8. GP regression ⚫ Other groups looking to use gaussian process regression fits • Non-parametric, learns distribution over space of functions rather than space of parameters ⚫ Still need to understand how (poisson) error is encoded, and certain input parameters ⚫ But, out of the box, we can get smooth fits when blanking out resonances in order to subtract out background ⚫ Left • In data, fit invariant mass distributions around J/psi in bins of DV 𝜍 ⚫ Right • Plot background-subtracted J/psi yields as a function of displacement • Overlay yields from a JpsiToMuMu MC sample (note, scouting collections are included up to AODSIM) • Seems like prompt Jpsi MC 8

  9. Backup 9

  10. L1 in full 2018 ⚫ Repeat L1 checks from previous set of slides, but fired unprescaled with full 2018 data (4B events) instead of just L1_DoubleMu0 0 0 L1_DoubleMu0_Mass_Min1 0 0 50M events in 2018C L1_DoubleMu0_OQ 0 0 • The seeds I’ve been using since last time L1_DoubleMu0_SQ_OS 0 0 L1_DoubleMu0er1p5_SQ 0 0 (DoubleMu4p5_SQ_OS_dR_Max1p2 || L1_DoubleMu0er2p0_SQ_OS_dR_Max1p4 0 0 DoubleMu_15_5_SQ || DoubleMu_15_7) L1_DoubleMu0er2p0_SQ_dR_Max1p4 0 0 L1_DoubleMu_12_5 0 0 are still in the unprescaled=1 set. Good. L1_DoubleMu_15_7_Mass_Min1 0 0 ⚫ EXO-19-018 AN has L1_ZeroBias 6.2417E-05 0 L1_DoubleMu0er1p5_SQ_OS 0.00138888 0 L1_DoubleMu4_SQ_OS 0.001443 0 In 2018, the DST path collected the total integrated luminosity of 61.3 fb − 1 . The L1 seeds L1_DoubleMu0_SQ 0.002448 0 L1 DoubleMu 12 5, L1 DoubleMu 12 8, L1 DoubleMu 13 6, L1 DoubleMu 15 5 were inactive L1_DoubleMu4p5_SQ_OS 0.00408753 0 and only L1 DoubleMu 15 7 was present. The following never pre-scaled and never disabled L1 seeds are considered only: L1_DoubleMu4p5er2p0_SQ_OS 0.0107283 0 • L1 DoubleMu4p5 SQ OS dR Max1p2 OR L1_DoubleMu0er1p5_SQ_dR_Max1p4 0.0689305 0.155734 • L1 DoubleMu4p5er2p0 SQ OS Mass7to18 OR L1_DoubleMu10_SQ 0.0295695 0.29647 • L1 DoubleMu 15 7 L1_DoubleMu4_SQ_OS_dR_Max1p2 0.471659 0.732179 L1_DoubleMu9_SQ 0.110944 0.919628 L1_DoubleMu_15_7_SQ 0.117701 0.919628 ⚫ I don’t understand the claim about L1_DoubleMu0er1p5_SQ_OS_dR_Max1p4 0.395483 0.999874 L1_DoubleMu18er2p1 0.0187654 1 DoubleMu_15_5 being inactive, since I see it L1_DoubleMu_15_7 0.132635 1 was firing and unprescaled for the whole dataset L1_DoubleMu_15_5_SQ 0.168136 1 L1_DoubleMu4p5er2p0_SQ_OS_Mass7to18 0.176602 1 L1_DoubleMu0er1p4_SQ_OS_dR_Max1p4 0.327869 1 L1_DoubleMu4p5_SQ_OS_dR_Max1p2 0.550969 1 10

  11. Pixel volumes ⚫ Consider only events with a displacement >1cm ⚫ Separate into "near" and "away" based on simple rectangular rho-z regions to cut out pixel layers 11

  12. Pixel hits before vertex ⚫ Try to look for pixel hits between the beamspot and the DV ⚫ No hitpatterns, so start with a simple case looking at the displacement and nValidPixelHits for nearly transverse muons in the barrel • Plot distribution of valid pixel hits for leading muon with DV rho within one of 3 bins between the first 3 layers • Leading muon | 𝜃 |<0.5, DV |z| < 20cm, cos( 𝜚 )>0.95 ⚫ MC peaks where we expect ⚫ Removing low p T muons cleans up data a little bit # valid pixel hits MC 8-10cm 4-6cm 1-2cm data data p T >25GeV overlaps? 12

  13. L1 triggers (MC e ffi ciency) ⚫ Sample • MC with c 𝜐 =50mm and m Zd =20GeV • Gen-level denominator requirements: ≥ 2 muons with p T >4GeV, | 𝜃 |<2.4, vertex displacement 𝜍 <11cm ⚫ For the L1 seeds marked in blue on the previous slide, calculate the e ffi ciency of each seed wrt the denominator above • For the 3 seeds with 𝜃 requirements, also add a column with | 𝜃 |<1.4/1.5/2.1 in the denominator instead ⚫ Best combination of 4 L1s is (DoubleMu0er1p5_SQ_OS_dR_Max1p4 || DoubleMu9_SQ || DoubleMu_15_5_SQ || DoubleMu_15_7) which gives an e ffi ciency of 88.7% • If we remove DoubleMu9 and DoubleMu10, then we get (DoubleMu0er1p5_SQ_OS_dR_Max1p4 || DoubleMu0_15_5_SQ || DoubleMu0_15_7) and an e ffi ciency of 87.6% • If we also remove eta-restricted seeds, then we get (DoubleMu4p5_SQ_OS_dR_Max1p2 || DoubleMu_15_5_SQ || DoubleMu_15_7) and an e ffi ciency of 86.7% e ff e ff w/ | 𝜃 | DoubleMu18 er2p1 0.342 0.378 DoubleMu0 er1p4 _SQ_OS_dR_Max1p4 0.463 0.753 DoubleMu0 er1p5 _SQ_OS_dR_Max1p4 0.497 0.751 "SQ" uses tighter DoubleMu4p5_SQ_OS_dR_Max1p2 0.621 quality requirements from SingleMuon seeds DoubleMu10_SQ 0.647 (L1 details here and DoubleMu9_SQ 0.682 here) DoubleMu_15_7_SQ 0.762 DoubleMu_15_7 0.782 DoubleMu_15_5_SQ 0.842 13

  14. Gen cutflows ⚫ MC sample • H(125) → ZdZd → 2 𝜈 +2X • c 𝜐 =50mm and m Zd =20GeV • SM Z BRs — 98.2% (1.8%) of events are 2 𝜈 (4 𝜈 ) ⚫ At least 2 gen muons must pass each cut yield fraction wrt N-1 cut inclusive cumulative yield fraction wrt all 1.00 cut inclusive p T >4GeV 0.81 𝜍 <11cm 0.61 p T >5GeV 0.76 p T >4GeV 0.49 | 𝜃 |<2.1 0.70 | 𝜃 |<2.4 0.38 | 𝜃 |<2.4 0.77 𝜍 <11cm 0.61 14

  15. Track isolation vs displacement 15

Recommend


More recommend