5.3 Vector-Like Quarks
155
of which two are consistent with coming from a Z boson decay. The highest sensitivity
to T T production is obtained from the signal regions with two large-R jets and three
leptons, whereas for B B production with B(B → Zb) = 1, the signal region with
one large-R jet is better than the trilepton region and comparable in sensitivity to the
selection with two large-R jets. The analysis by CMS follows a different strategy,
being less inclusive in its jet substructure selection. A total of 16 signal regions are
defined, based on the number of small-R b-tagged jets and large-R V -, H - or ttagged jets. In order to maintain signal efficiency at low m VLQ , also resolved events
are considered. Doublets and triplets of small-R jets are built, which do not overlap
with tagged large-R jets. Based on the mass of these pairings, V , H or t candidates
are built. For m VLQ < 1.2 TeV the resolved and substructure taggers are equally
efficient in identifying signal events. For higher masses, the jet substructure taggers
result in a more efficient signal selection. Note that this strategy for tagging resolved
decays only works in the Z → channel, where multijet and tt backgrounds are
small. The ATLAS and CMS analyses achieve comparable sensitivity to B B and T T
production, and exclude VLQs below masses between 1 and 1.34 TeV.
The presence of a number of highly boosted vector bosons or top quarks facilitates searches for VLQ pair production in all-hadronic final states with low p
miss
T .
Optimised jet substructure taggers are the key for achieving the best possible sensitivity, which depends on an efficient suppression of the large multijet background
and at the same time high signal efficiency. Because of the large number of possible decay cascades from VLQ pair production, and the need for a dedicated and
thorough background estimation for each measured distribution, it is advantageous
to design an inclusive all-hadronic analysis. ATLAS and CMS have both performed
a search for T T and B B production in the all-hadronic final state, using machine
learning jet substructure taggers to suppress the multijet background. In the ATLAS
analysis [885], VR jets are used (see Sect. 2.3.2), reclustered from calibrated small-R
jets with R = 0.4. The VR jets are clustered using ρ = 315 GeV, R min = 0.4 and
R max = 1.2, a parameter choice which is a compromise between the values for V
decays (ρ ≈ 200 GeV) and for t decays (ρ ≈ 600 GeV), offering the possibility to
reconstruct all relevant objects in the final state. Once VR jets have been clustered,
trimming is applied, removing subjets of the VR jet if their p T is less than 5%
of the VR jet p T . The VR jets are required to have p T > 150 GeV, m jet > 40 GeV
and |η| < 2.5. The subjets of the VR jet are used to train a DNN-based multi-class
tagger, able to discriminate between V , H , t and background jets. The advantage of
the small-R jet reclustering into VR jets is that the small-R jets have already been
calibrated, leading to a precise momentum and mass reconstruction of VR jets, with
uncertainties derived from the small-R jet uncertainties. These uncertainties can be
propagated through the DNN to obtain uncertainties on the tagging efficiency for signal jets. The disadvantage is that the minimum distance between decay products that
can be resolved is governed by the resolution parameter of the small-R jets, which
is 0.4 in this case. At high p T , this is too coarse to resolve the substructure of highly
boosted hadronic decays, leading to a decreasing efficiency of the derived multi-class
tagger. Twelve signal regions are defined based on the multiplicity of b-tagged smallR jets, V -, H -, and t-tagged VR jets, designed to cover all possible VLQ decays.
155
of which two are consistent with coming from a Z boson decay. The highest sensitivity
to T T production is obtained from the signal regions with two large-R jets and three
leptons, whereas for B B production with B(B → Zb) = 1, the signal region with
one large-R jet is better than the trilepton region and comparable in sensitivity to the
selection with two large-R jets. The analysis by CMS follows a different strategy,
being less inclusive in its jet substructure selection. A total of 16 signal regions are
defined, based on the number of small-R b-tagged jets and large-R V -, H - or ttagged jets. In order to maintain signal efficiency at low m VLQ , also resolved events
are considered. Doublets and triplets of small-R jets are built, which do not overlap
with tagged large-R jets. Based on the mass of these pairings, V , H or t candidates
are built. For m VLQ < 1.2 TeV the resolved and substructure taggers are equally
efficient in identifying signal events. For higher masses, the jet substructure taggers
result in a more efficient signal selection. Note that this strategy for tagging resolved
decays only works in the Z → channel, where multijet and tt backgrounds are
small. The ATLAS and CMS analyses achieve comparable sensitivity to B B and T T
production, and exclude VLQs below masses between 1 and 1.34 TeV.
The presence of a number of highly boosted vector bosons or top quarks facilitates searches for VLQ pair production in all-hadronic final states with low p
miss
T .
Optimised jet substructure taggers are the key for achieving the best possible sensitivity, which depends on an efficient suppression of the large multijet background
and at the same time high signal efficiency. Because of the large number of possible decay cascades from VLQ pair production, and the need for a dedicated and
thorough background estimation for each measured distribution, it is advantageous
to design an inclusive all-hadronic analysis. ATLAS and CMS have both performed
a search for T T and B B production in the all-hadronic final state, using machine
learning jet substructure taggers to suppress the multijet background. In the ATLAS
analysis [885], VR jets are used (see Sect. 2.3.2), reclustered from calibrated small-R
jets with R = 0.4. The VR jets are clustered using ρ = 315 GeV, R min = 0.4 and
R max = 1.2, a parameter choice which is a compromise between the values for V
decays (ρ ≈ 200 GeV) and for t decays (ρ ≈ 600 GeV), offering the possibility to
reconstruct all relevant objects in the final state. Once VR jets have been clustered,
trimming is applied, removing subjets of the VR jet if their p T is less than 5%
of the VR jet p T . The VR jets are required to have p T > 150 GeV, m jet > 40 GeV
and |η| < 2.5. The subjets of the VR jet are used to train a DNN-based multi-class
tagger, able to discriminate between V , H , t and background jets. The advantage of
the small-R jet reclustering into VR jets is that the small-R jets have already been
calibrated, leading to a precise momentum and mass reconstruction of VR jets, with
uncertainties derived from the small-R jet uncertainties. These uncertainties can be
propagated through the DNN to obtain uncertainties on the tagging efficiency for signal jets. The disadvantage is that the minimum distance between decay products that
can be resolved is governed by the resolution parameter of the small-R jets, which
is 0.4 in this case. At high p T , this is too coarse to resolve the substructure of highly
boosted hadronic decays, leading to a decreasing efficiency of the derived multi-class
tagger. Twelve signal regions are defined based on the multiplicity of b-tagged smallR jets, V -, H -, and t-tagged VR jets, designed to cover all possible VLQ decays.