86
3 Jet Substructure at the LHC
400 GeV. Instead, in ATLAS τ 32 is calculated from trimmed jets, which results in less
discrimination power when used as sole tagging variable compared to ungroomed τ 32 .
However, groomed τ 32 can still lead to considerable improvements when combined
with other variables [523].
ATLAS and CMS have studied various top taggers also for their stability with
respect to the number of pileup interactions [459, 524]. Single variables and their
combinations are studied and compared with shower deconstruction, CMSTT, HTT,
and an improved version thereof with shrinking cone size (HTTv2) [234].
Figure 3.10 (right) shows a comparison based on simulation of the single variable
performance in CMS, where signal jets are generated through a heavy resonance
decaying to tt and background jets are taken from QCD multijet production. Note
that for this study reconstructed jets are matched to a generated parton, and the
distance between the top quark and its decay products must be less than 0.6 or 0.8
for a reconstructed R = 0.8 and 1.5 jet, to ensure that the top quark decay products
are fully merged and reconstructed in a single jet. The best single variable in terms
of efficiency versus background rejection is the discriminator log χ , calculated with
shower deconstruction. The second best variables are the N -subjettiness ratio τ 32 at
low efficiency and the jet mass calculated with the HTTv2 at high efficiency values.
The individual groomed jet masses show similar performance, and CMS moved to
using the soft drop mass due to its beneficial theoretical properties [24]. The default
for CMS analyses of 13 TeV data was chosen to be the soft drop jet mass combined
with τ 32 for top tagging at high p T . Generally, at high boost, the combination of a
groomed mass with τ 32 leads to a large gain in background rejection.
The CMS study [459] also investigated combining single variables with more
complex taggers. Combining shower deconstruction with the soft drop mass, τ 32 ,
and subjet b tagging can lead to improvements. However, the efficiency and misidentification rate for this combination were found not to be stable as a function of jet
p T (the combined algorithms were studied using working points corresponding to a
background efficiency of 30%). At low boosts, the dedicated HTTv2 shows the best
performance. In this kinematic region, using groomed τ 32 , obtained by using the set
of particles from the soft drop jet instead of the original jet, helps to improve the
performance.
ATLAS has commissioned a top tagger with little complexity for use by physics
analyses of 13 TeV data. The rationale behind this approach was the potential benefit
of having an efficient top tagger with well-understood efficiency and associated
systematic uncertainties for use in early analyses. The supported top tagger makes
use of anti-k T R = 1.0 trimmed jets, but with a parameter of R sub = 0.2 instead of
0.3. Candidate top jets are required to satisfy a calibrated mass window requirement
and a p T -dependent, one-sided cut on τ 32 [524]. The variable τ 32 has been chosen
since it shows the best background rejection in combination with a small correlation
with m jet , a reduced p T -dependence, and good performance across a large range in
p T . A comparison of this two-variable tagger, and more generally a tagger based on
the jet mass and an angular variable like τ 32 , and more complex taggers shows that
the background rejection can be improved by about 50% when using the HTT or
shower deconstruction [525].
3 Jet Substructure at the LHC
400 GeV. Instead, in ATLAS τ 32 is calculated from trimmed jets, which results in less
discrimination power when used as sole tagging variable compared to ungroomed τ 32 .
However, groomed τ 32 can still lead to considerable improvements when combined
with other variables [523].
ATLAS and CMS have studied various top taggers also for their stability with
respect to the number of pileup interactions [459, 524]. Single variables and their
combinations are studied and compared with shower deconstruction, CMSTT, HTT,
and an improved version thereof with shrinking cone size (HTTv2) [234].
Figure 3.10 (right) shows a comparison based on simulation of the single variable
performance in CMS, where signal jets are generated through a heavy resonance
decaying to tt and background jets are taken from QCD multijet production. Note
that for this study reconstructed jets are matched to a generated parton, and the
distance between the top quark and its decay products must be less than 0.6 or 0.8
for a reconstructed R = 0.8 and 1.5 jet, to ensure that the top quark decay products
are fully merged and reconstructed in a single jet. The best single variable in terms
of efficiency versus background rejection is the discriminator log χ , calculated with
shower deconstruction. The second best variables are the N -subjettiness ratio τ 32 at
low efficiency and the jet mass calculated with the HTTv2 at high efficiency values.
The individual groomed jet masses show similar performance, and CMS moved to
using the soft drop mass due to its beneficial theoretical properties [24]. The default
for CMS analyses of 13 TeV data was chosen to be the soft drop jet mass combined
with τ 32 for top tagging at high p T . Generally, at high boost, the combination of a
groomed mass with τ 32 leads to a large gain in background rejection.
The CMS study [459] also investigated combining single variables with more
complex taggers. Combining shower deconstruction with the soft drop mass, τ 32 ,
and subjet b tagging can lead to improvements. However, the efficiency and misidentification rate for this combination were found not to be stable as a function of jet
p T (the combined algorithms were studied using working points corresponding to a
background efficiency of 30%). At low boosts, the dedicated HTTv2 shows the best
performance. In this kinematic region, using groomed τ 32 , obtained by using the set
of particles from the soft drop jet instead of the original jet, helps to improve the
performance.
ATLAS has commissioned a top tagger with little complexity for use by physics
analyses of 13 TeV data. The rationale behind this approach was the potential benefit
of having an efficient top tagger with well-understood efficiency and associated
systematic uncertainties for use in early analyses. The supported top tagger makes
use of anti-k T R = 1.0 trimmed jets, but with a parameter of R sub = 0.2 instead of
0.3. Candidate top jets are required to satisfy a calibrated mass window requirement
and a p T -dependent, one-sided cut on τ 32 [524]. The variable τ 32 has been chosen
since it shows the best background rejection in combination with a small correlation
with m jet , a reduced p T -dependence, and good performance across a large range in
p T . A comparison of this two-variable tagger, and more generally a tagger based on
the jet mass and an angular variable like τ 32 , and more complex taggers shows that
the background rejection can be improved by about 50% when using the HTT or
shower deconstruction [525].