Event Deconstruction: example with top quarks Davison E. Soper � University of Oregon W ork with Michael Spannowsky MIAPP , Munich, August 2014 1
Introduction • One can examine the fine grained structure of events in order to dig out new physics signals. � • Michael Spannowsky and I proposed a general method for subjet analysis: “shower deconstruction.” ( Phys. Rev. D84 ( 2011 ) 074002 ) � • An application is to find top quark jets. ( Phys. Rev. D87 ( 2013 ) 054012 ) � • This works well in ATLAS: ``Performance of shower deconstruction in ATLAS,'' ATLAS - CONF - 2014 - 003. 2
• W e have extended this to “event deconstruction,’’ looking at the relevant features of whole events. ( arXiv:1402.1189 [ hep - ph ]) � • Our example: 3
Our example 4
Event sample 5
Microjets 6
What we would like • We would like to separate signal an background using 7
Why? • Assuming that you believe your Monte Carlo, to get the most signal cross section for a given background cross section by making a cut, your cut should be along a contour line of 8
What we do 9
Event histories a Sudakov factor a decay a decay a QCD splitting a Sudakov factor initial state radiation 10
Sum over event histories 11
distributions for signal and background Z’ 0.7 t t Dijets 0.6 ) [1/0.3] 0.5 χ 0.4 /dlog( 0.3 σ d σ 1/ 0.2 0.1 0 -14 -12 -10 -8 -6 -4 -2 0 log( ) χ 12
Tagging e ffi ciency 13
Comparison to HEP top tagger 6 10 1 / Prob. of accepting fake background Full ED Full ED (only t t bkg.) 5 10 Full ED (only dijet bkg.) Event deconstruction HTT + mass cut HTT (only t t bkg.) + mass cut 4 10 HTT (only dijet bkg.) + mass cut 3 10 HEP top tagger and mass cut 2 10 10 1 0 0.1 0.2 0.3 0.4 0.5 0.6 Efficiency for tagging Z’ 14
Conclusions • Event deconstruction seems to work well. � • For tagging single jets, shower deconstruction also works nicely with Monte Carlo events. � • ATLAS finds that shower deconstruction works well for finding top quark jets in experiment. � • The general method could be helped by contributions from other theorists. 15
Recommend
More recommend
