14/06/11 Jet physics meetingV.Kostyukhin 1 Flavour fractions in di-jet system V.Kostyukhin C.Lapoire M.Lehmacher Bonn
14/06/11 Jet physics meetingV.Kostyukhin 2 Some theory 1) Heavy flavour pair creation Heavy flavour production can be approximately described by 3 mechanisms 2) Flavour exitation heavy flavour from proton sea or alternatively from initial state showers 3) Gluon splitting g QQ in final state parton showers
14/06/11 Jet physics meetingV.Kostyukhin 3 Some theory PYTHIA predictions for inclusive b-jet production (b-quark p >5GeV) Similar picture for charm
14/06/11 Jet physics meetingV.Kostyukhin 4 Some theory Only flavour pair creation mechanism (1) makes back-to-back jet pairs with identical (heavy) flavours. Mainly LO process. Flavour exitation (2) and gluon splitting (3) produce back-to-back heavy+light jet pairs. NLO processes. Even flavour pair creation process contribute to heavy+light back-to-back pair in NLO, see below QQ pairs – LO dominant Q+light pairs – NLO dominant
14/06/11 Jet physics meetingV.Kostyukhin 5 Di-jet flavours Di-jet ( 2 leading jets in event back-to-back) analysis model includes 6 fractions (full set): UU (light+light) ~85% CC (charm+charm) ~1% BB (beaty+beaty) ~0.6% BC (beauty+charm) ~0.3% BU (beauty+light) ~4% CU (charm+light) ~10% PYTHIA predictions
14/06/11 Jet physics meetingV.Kostyukhin 6 Di-jet flavours analysis To distinguish flavours the kinematical variables from secondary vertices in jet are used. Many variables were considered. Optimisation included Highest sensitivity to jet flavour content Minimal jet p dependence Stability with respect to detector effects The final (minimal) choice includes 2 variables: “Product” variable“Boost” variable
14/06/11 Jet physics meetingV.Kostyukhin 7 Di-jet flavours analysis To simplify statistical description and template construction both variables are transformed to be in [0.,1.] range. “Boost” is the only variable which has the extreme values for charm, not for beauty! The beauty here is between light and charm. Should facilitate charm separation. “Product” variable“Boost” variable
14/06/11 Jet physics meetingV.Kostyukhin 8 Di-jet analysis model Variable definitions (8 in total): fBB – fraction of b-jet pairs in 2-jet sample fCC – fraction of c-jet pairs in 2-jet sample fBU – fraction of b-jet plus u-jet in 2-jet sample fCU – fraction of c-jet plus u-jet in 2-jet sample fBC – fraction of b-jet plus c-jet in 2-jet sample fUU=1.-fBB-fBC-fBU-fCC-fCU - not independent v b – probability to reconstruct secondary vertex in b-jet v c – probability to reconstruct secondary vertex in c-jet v u – probability to get fake secondary vertex in u-jet Templates from MC (boost case): B(b) – secondary vertex boost distribution for b-jet C(b) – secondary vertex boost distribution for c-jet U(b) – secondary vertex boost distribution for u-jet
14/06/11 Jet physics meetingV.Kostyukhin 9 Di-jet analysis model Case of 2 reconstructed secondary vertices in di-jet event : Probability: 2-dim probability density function for the fit :
14/06/11 Jet physics meetingV.Kostyukhin 10 Di-jet analysis model Case of single reconstructed secondary vertex in di-jet event : Probability: 1-dim probability density function for the fit :
14/06/11 Jet physics meetingV.Kostyukhin 11 Di-jet analysis model 2 fitting procedures are used in analysis: 1. and B are templated separately and parametrised with b-splines (1D fit) 2.Joint & B distributions are used as templates (2D fit) 2D fit has better statistical accuracy due to explicit correlation treatment, but they may be wrong on data so 1D fit is less biased. More important is template construction from JX Monte Carlo data samples influence. Splitting of single process into subsamples (JX) results in highly nonuniform errors in templates. E.g. few events from J0(low p sample) with huge weights fall into several template bins. Then these bins get much higher errors and shifted(!!!) in some cases mean. It’s not known a priori how to deal correctly with such bins. The 2 fitting methods use completely different strategies – 1D approach washes out such shifts due to b-spline smoothing, 2D fit accepts them. 1D and 2D analysis procedures are far from 100% correlated and then they are used simultaneously for data fit
14/06/11 Jet physics meetingV.Kostyukhin 12 Fast simulation model Heavy flavour fractions are small fully simulated jet statistics is not enough for validation of analysis properties (bias, error estimations, etc.). Fast simulation model is developed All reconstructed secondary vertices in fully simulated jet events are collected into database as a function of jet p , and flavour. Generation procedure: 1.Jet pair is created according to jet p , distributions taken from data. 2.Jet flavours are chosen according to the model fractions fUU,fBB,fCC,fBC,fCU,fBU 3.From flavour content one decides whether secondary vertex is “reconstructed” in each jet according to model efficiencies v u,v c,v b. 4.If SV is “reconstructed” – its parameters are taken from database according to jet p , (randomly chosen from nearby region). 5.Finally the recorded SV parameters are smeared with detector resolution to avoid the repetition of exactly the same numbers in generated events due to a single vertex in database chosen several times.
14/06/11 Jet physics meetingV.Kostyukhin 13 Fast simulation model Fast simulation demonstrated that 2010 statistics is not enough for reliable simultaneous estimation of all 8 model parameters. Then it was chosen to fix 2 SV reconstruction efficiencies on MC values. Beauty and charm vertex reconstruction efficiencies are chosen to be fixed because they are most precisely predicted by Monte Carlo. In our analysis of 2010 data we fit simultaneously 6 model parameters (v c,v b are fixed): light jet fake vertex probability (v u ). five flavour fractions (fBB,fBC,fCC,fCU,fBU) Both 1D and 2D models with 6 parameters demonstrate correct behavior with fast simulation (next slides…)
14/06/11 Jet physics meetingV.Kostyukhin 14 Fast simulation model 2D fitting model performance with fast simulation (200 tries)
14/06/11 Jet physics meetingV.Kostyukhin 15 Fast simulation model 2D fitting model pulls with fast simulation (200 tries)
14/06/11 Jet physics meetingV.Kostyukhin 16 Data selection Jets are selected in ID volume | |<2.1 to guarantee the performance of vertex reconstruction. Usual jet cleaning cuts are applied on data. Anti-kt R=0.4 jets are used. Analysis is done in leading jet p bins. They are chosen to match ATLAS single jet trigger thresholds. o The A-D 2010 periods are used for [40,60] and [60,80] bins with L1 jet trigger. For other bins E-I 2010 periods with EF triggers are used. o Subleading jet p is also restricted in analysis to decrease systematic due to p dependence of templates.
14/06/11 Jet physics meetingV.Kostyukhin 17 Vertex reconstruction efficiencies The efficiencies are obtained as weighted average over all JX PYTHIA samples.
14/06/11 Jet physics meetingV.Kostyukhin 18 Vertex asymmetry The amount of reconstructed secondary vertices in leading and subleading jets is DIFFERENT in di-jet event. One of the reasons is the semileptonic decays of heavy flavours. Jet energy disappears with neutrino, what automatically makes the heavy flavour jet subleading. Qualitatively described by MC BUT DISAGREE(!!!) with data quantitatively
14/06/11 Jet physics meetingV.Kostyukhin 19 Vertex asymmetry The SV asymmetry is coming from mixed heavy+light jet pairs. Reason of data-MC discrepancy here could be either bad description of semileptonic decays in PYTHIA or bad description of gluon splitting (also causes jet energy loss) Asymmetry is added to the fitting model. The exact reason is unclear (doesn’t seem PYTHIA problem, but…) then for baseline result the asymmetry is fixed on MC values. Changes due to free asymmetry parameter in the are taken as systematics.
14/06/11 Jet physics meetingV.Kostyukhin 20 Data fit quality Example: data description in [80,120] GeV bin with 2D fit
14/06/11 Jet physics meetingV.Kostyukhin 21 Fit results Data in each p bin are fitted with 1D and 2D methods. Fit results are combined with error dependent weights. Statistical error in bin is calculated from 1D and 2D errors assuming 100% correlation between them. First – probability to get fake vertex in light jet : ̶̶̶ data fit ̶̶̶ Monte Carlo (PYTHIA) Leading jet p GeV
14/06/11 Jet physics meetingV.Kostyukhin 22 Fit results Fitted di-jet flavour fractions compared with PYTHIA predictions ̶̶̶ data fit ̶̶̶ Monte Carlo (PYTHIA) Leading jet p GeV fUU fraction is estimated from others fUU=1-fBB-fBC-fCC-fBU-fCU
14/06/11 Jet physics meetingV.Kostyukhin 23 Fit results Fitted di-jet flavour fractions compared with other generators. MC boxes represent statistical errors. At particle jet level MC band should be much more narrow.
14/06/11 Jet physics meetingV.Kostyukhin 24 Unfolding to particle jet level Exists, but requires some final polishing. That’s why not presented here. However changes in flavour fractions are very small just because they are ratios !
14/06/11 Jet physics meetingV.Kostyukhin 25 Systematics Systematic due to SV asymmetry discrepancy between data and MC. Is checked by leaving free the b-jet asymmetry in the fit.
14/06/11 Jet physics meetingV.Kostyukhin 26 Systematics Systematic due to template shape description. Done by using templates made from inclusive jets. No di-jet selection, much bigger statistics, different production mechanisms. Different MC statistics produces additional statistical fluctuations, so the average shift for all p bins is taken as systematic of give model parameter. To be completed with another MC generators…
14/06/11 Jet physics meetingV.Kostyukhin 27 Systematics Systematic due to charm and beauty vertex reconstruction efficiencies. Can be estimated from data using the tight link between these efficiencies and fake vertex probability in light jet. Due to vertex reconstruction algorithm they are controlled by single parameter. Then data-MC difference in fake vertex probability can be translated to v c /v b uncertainties. They are estimated to be 1.1% for charm and 2% for beauty To be completed with another MC generators… Changes in results due to simultaneous change of v c by 1.1% and v b by 2%
14/06/11 Jet physics meetingV.Kostyukhin 28 Status 1.The measurements are done and now they are under final polishing. 2.Systematic needs to be completed with different MC generators. Corresponding MC samples are being processed right now. 3.Backup note should be finished soon.