360 likes | 463 Vues
A detailed summary of Ecal calibration progress and future prospects for Ecal data from 2010. Includes data analysis results, iterations, job processes, and optimization methods discussed in slides from May and June 2010.
E N D
Ecal Calibration 2k+10 Dasha Savrina, Victor Egorychev & Vanya Belyaev
Outline Slides from 18 May 2k+10 Slides from 8 June 2k+10 Prospects Vanya Belyaev
Kali • Use 80M of real data 2k+10 • Another ~70M in pipeline. Should we add them? • First pass of Kali run (prepare fmDST) using • LHCBCOND.db_new by Olivier with all known corrections and coefficients • CVS HEAD of Calo/CaloReco + Corrections.py from Olivier • Prs+Ecal corrections are applied “by-hand” at the second run by Kali • takes less than 15 minutes to get 85-90% jobs from Grid (A bit longer during working hours) • And few hours to get the remaining 10-15% Vanya Belyaev
Starting point After 9 primary iterations l= (98.98±5.91)% Vanya Belyaev
Re-reconstruct After 9 primary iterations l*= (99.49±2.19)% l= (98.51±6.81)% Vanya Belyaev
Inner Zone Vanya Belyaev
Middle Zone Vanya Belyaev
Outer Zone Vanya Belyaev
All Zones Vanya Belyaev
One more re-reconstruction ? l*= (99.98±0.50)% 917 cells > 0.5% 231 cells > 1% 14 cells > 3% 3 cells > 5% Vanya Belyaev
(technical) numbers It scales nicely with number of of cores • fmDST reprocessing: • O(50) Grid jobs, less than 10 minutes each • NTuple processing • 1 primary iteration O(2 hours) @ CAF, 7 cores • Histo projection: O(¾hour) • Histo fitting: O(1¼hour) • 1 “long night” for ~8-9 iterations • The whole cycle around 2-3 days … Vanya Belyaev
Summary Part-I Next steps? Add more data? One more re-reconstruction round? Check with h0 ? • Ecal calibration constants have been obtained from 2k+10 data • “Iteration convergency/stability”: rms = 0.3% • Clear improvement in p0 mass resolution has been observed • Note: p0 width is not a parameter for calibration! • The mass of p0 is fine • Tiny bias for outer zone: • b? Vanya Belyaev
Compare with Barcelona Vanya Belyaev
Outer area: 1-l1/l2 Pattern??? x↔ y flip! Vanya Belyaev
Middle area: 1-l1/l2 x↔ y flip! Vanya Belyaev
Inner area: 1-l1/l2 Pattern x↔ y flip! Vanya Belyaev
Next 14 slides • Test 7 variables ( 2 slides per variable) • Try to find the correlation between difference in Spain & Russian constants and the variables • 3 plots per variable: • Slide 1: the four superimposed distributions of variable: |1-l1/l2|<2% 2%<|1-l1/l2|<5% 5%<|1-l1/l2|<10% 10%<|1-l1/l2| • Slide 2, profile plots: • Left: <variable> as function of 1-l1/l2 • Right: <|1-l1/l2|> as function of variable Vanya Belyaev
#LLp0 |1-l1/l2|<2% 2%<|1-l1/l2|<5% 5%<|1-l1/l2|<10% 10%<|1-l1/l2| no dependency? Number of reconstructed p0 in max(Eprs1, Eprs2)<10 MeV category Vanya Belyaev
#LLp0 Profiles no dependency? <NLL> <|1-l1/l2|> 1-l1/l2 log10 NLL <NLL> Vanya Belyaev
#LGp0 |1-l1/l2|<2% 2%<|1-l1/l2|<5% 5%<|1-l1/l2|<10% 10%<|1-l1/l2| some dependency Number of reconstructed p0 in Eprs1<10 MeV,Eprs2>10 MeV category Vanya Belyaev
#LGp0 Profiles some dependency <NLG> <|1-l1/l2|> 1-l1/l2 log10 NLG Vanya Belyaev
#GGp0 |1-l1/l2|<2% 2%<|1-l1/l2|<5% 5%<|1-l1/l2|<10% 10%<|1-l1/l2| clear dependency! Number of reconstructed p0 in min(Eprs1, Eprs2)>10 MeV category Vanya Belyaev
#GGp0 Profiles clear dependency! <NGG> <|1-l1/l2|> 1-l1/l2 log10 NGG Vanya Belyaev
#Sp0 |1-l1/l2|<2% 2%<|1-l1/l2|<5% 5%<|1-l1/l2|<10% 10%<|1-l1/l2| clear dependency! Total Number of reconstructed p0 Vanya Belyaev
#Sp0 Profiles clear dependency! <NS> <|1-l1/l2|> 1-l1/l2 log10 NS Vanya Belyaev
#entries |1-l1/l2|<2% 2%<|1-l1/l2|<5% 5%<|1-l1/l2|<10% 10%<|1-l1/l2| no dependency??? Total Number of entries in p0 histograms (mainly background) Vanya Belyaev
#Sentries Profiles no dependency??? <NSentries> <|1-l1/l2|> 1-l1/l2 log10 NSentries Vanya Belyaev
sp0 |1-l1/l2|<2% 2%<|1-l1/l2|<5% 5%<|1-l1/l2|<10% 10%<|1-l1/l2| clear dependency p0 width (sigma) from the fit Vanya Belyaev
sp0 Profiles clear dependency <sp0> <|1-l1/l2|> 1-l1/l2 sp0 Vanya Belyaev
smassp0 |1-l1/l2|<2% 2%<|1-l1/l2|<5% 5%<|1-l1/l2|<10% 10%<|1-l1/l2| Very clear dependency!!! Error in p0 mass from the fit Vanya Belyaev
smassp0 Profiles Very clear dependency!!! <smassp0> <|1-l1/l2|> 1-l1/l2 log10 smassp0 Vanya Belyaev
Summary-Part-II • Difference for innermost and outermost cells • Moderate dependency on p0 statistics • “No” dependency on number of entries • Clear dependency on p0 width • The most clear dependency on error in p0 mass • “obvious” • Good sign: “error is reliable” • For small error estimate both methods gives the same result No miracles… Vanya Belyaev
Prospects • Re-run Kali on Reco04 data • ~150M stripped events are available • Dasha has prepared fmDST • To be analysed • We would like to get there of three coefficients • For whole sample, for the first half and for the second half • The comparison of these three sets will give us some hints about the internal precision of the method for 79-80M statistics Vanya Belyaev
What to do with border cells? • We know from MC that for innermost belt and outermost belt we are systematically wrong (~2-10%) • Apply this correction factors from MC for these cells • *AND* exclude these cells from the definition of Ecalfiducial volume for physics studies • This should preserve the goodness of reconstructed phtoons within fiducialvolume Vanya Belyaev
Overall Summary This is average <per-cell> The p0-weighted average is much-much-much better This is not in a contradiction with the hypothesis that numbers from Barcelona are “better” • Ecal is calibrated at some reasonably level • Probably not worse than 2% • MC predictions • agreement between Spain & Russia • For some cells there is disagreement between Spain & Russia • For these cells we expect from MC disagreement between iterative-p0 calibration and true calibration Vanya Belyaev
Summary in short: she works Vanya Belyaev