[NN b-Tagger Public Homepage] | [tt Cross-Section Internal Page] | [Higgs Search Public Page] |
---|
tt
Production Cross-Section Measurement Using the Neural Network b-Tagger
H. Bachacou1, Y. Kusakabe2, P. Lujan1, T. Masubuchi3, M. McFarlane1, A. Taffard4, W. Yao1
1LBNL, 2Waseda,3Tsukuba,4UIUC
Introduction
Using SecVtx, and the NN b-tagger to reduce background, we ascertain a cross-section of 8.5 ± 0.6 (stat) ± 1.0 (syst), which is consistent with the measurement 8.4 ± 0.6 (stat.) ± 1.1 (syst.) determined with SecVtx alone. This offers evidence the NN b-tagger is behaving appropriately and usefully. Events for this measurement are taken from the lepton + jets channel (695 pb-1). Please refer to the NN b-tagger public homepage for more information relevant to this analysis.
Event Selection
Electrons: ET > 20GeV, isolated
Muons: pT > 20GeV/c, isolated
Jets: ET > 15GeV, |η| < 2.0, (for the actual measurement, >3 jets are required, but all jet multiplicities are examined.)
Vetoed: Dilepton events, Z0, photon conversions, cosmic rays
At least one SecVtx tag passing the NN b-tagger
Relevant Backgrounds
Mistags (light jets identified as b by SecVtx)
W + heavy flavor
Non-W (QCD)
WW, WZ, ZZ, Z→ττ, single top
NN Operating Point
b-l network cut at 0.182, b-c network cut at 0.242; result: 90% b efficiency, 50% c efficiency, 35% l efficiency.
Method
All W + jet events passing the NN b-tagger are fed into the b-l neural network so that their combined output distribution is fitted according to the distributions for known b, c and l jets, to estimate the constituent fractions thereof. This process is repeated using the b-c network. We then input the W + jet events into a prescription called "Method 2", which also calculates fractional jet composition and was used in the standard cross-section measurement using SecVtx. The fractions calculated with the neural networks and Method 2 agree.
To obtain the measurement reported here, the W + jets events passing the NN b-tagger are inputted into Method 2 to yield the background decomposition below.
Background Estimates | ||||
---|---|---|---|---|
Njets | 1 jet | 2 jets | 3 jets | >4 jets |
Pretag | 68183 | 10647 | 1713 | 468 |
Mistag | 100.1 ± 20.6 | 41.7 ± 8.6 | 12.3 ± 2.5 | 3.0 ± 0.6 |
Wbb | 179.2 ± 55.4 | 100.4 ± 29.7 | 22.1 ± 5.8 | 4.1 ± 1.1 |
Wcc | 37.3 ± 10.7 | 28.2 ± 8.5 | 5.6 ± 1.7 | 1.12 ± 0.35 |
Wc | 137.4 ± 34.6 | 26.6 ± 6.8 | 2.7 ± 0.7 | 0.4 ± 0.1 |
Single Top | 13.0 ± 1.3 | 20.7 ± 2.1 | 4.5 ± 0.5 | 0.79 ± 0.09 |
Diboson | 7.3 ± 1.2 | 11.9 ± 1.9 | 3.0 ± 0.5 | 0.81 ± 0.15 |
Non-W (QCD) | 59.6 ± 10.0 | 29.5 ± 5.1 | 8.8 ± 1.7 | 3.96 ± 0.85 |
Total Bkg | 533.9 ± 79.2 | 259.2 ± 41.0 | 58.9 ± 8.2 | 14.2 ± 1.9 |
Top | 6.6 ± 0.9 | 48.1 ± 6.7 | 105.2 ± 14.7 | 130.2 ± 18.2 |
Top + Bkg | 540.5 ± 79.2 | 307.3 ± 41.5 | 164.1 ± 16.8 | 144.4 ± 18.3 |
Data | 623 | 332 | 166 | 145 |
Method 2 output totals are scaled to data totals for the W + 3, 4 jet bins. The W + 1, 2 jet bins, which serve as a control, agree within one one standard deviation. If this were not the case, the 3 and 4 jet bins could not be trusted. Plots representing the table above are below:
Without NN b-tagging | After NN b-tagging |
---|---|
![]() |
![]() |
σtt = | Nobs - Nbg |
εtt × L |