5.2 Resonances Coupling to Third Generation Quarks
141
magnitude larger than the irreducible background from tt production. The same is
observed for semi-merged events, which result in a similar sensitivity as fully merged
events for resonance masses of about 1 TeV. For masses of 1.5 TeV and higher, the
signal efficiency for the semi-merged selection is lower by a factor of four and more.
Since later analyses in the all-hadronic channel focus on high resonance masses,
semi-merged events are not considered any longer.
The sensitivity of the analysis could be largely improved by a follow-up analysis,
using four times more data recorded at 8 TeV [805]. The same techniques as in
the 7 TeV result are employed, particularly the CMSTT is used to select two ttagged jets [805]. Relative to the 7 TeV result, the sensitivity improves because of
the increased
√
s and the larger dataset, while the ratio of reducible multijet to the
irreducible tt background is the same. Advanced top tagging algorithms allowed
for an improvement on these results using the same data [806]. The performance
of the CMSTT is enhanced by using subjet b tagging and a selection of τ 32 < 0.7.
In addition, the HEPTopTagger, applied on CA jets with R = 1.5, is used to gain
sensitivity to low mass resonances in the all-hadronic channel. Events failing the t
tagging selection based on the CMSTT, are tested for two HEPTopTagger t-tagged
jets. The use of subjet b tagging and τ 32 makes the background estimation for multijet
production more complicated. While the overall strategy is the same, correlations in
the variables m tt , p T , τ 32 and the subjet b tagging discriminant are observed. The
mistag probability in a given bin of m tt is therefore parametrised by the three variables
p T , τ 32 and the value of the subjet b tagging discriminator. For events selected with
the HEPTopTagger, the background estimation is obtained from a sideband region
where the jet mass and pairwise subjet masses are inverted. The use of better top
tagging techniques results in a large suppression of the multijet background, as shown
in Fig. 5.7. This background constitutes only about 30% of the total background for
events with two b-tagged subjets, and increases to about 70% for one b-tagged subjet.
The better background suppression improves the 95% CL upper cross section limits
on the production of a narrow resonance by nearly 50% for masses between 1 and
2 TeV.
In analyses of 13 TeV data, CMS uses the soft drop mass, τ 32 and subjet b tagging
applied on the soft drop subjets to identify t jets. First results in the all-hadronic
channel have been obtained on a dataset corresponding to 2.6 fb
−1 [807]. The same
algorithm is used in the latest CMS analysis in this channel using 35.9 fb
−1 [808].
The algorithm shows a very similar performance as the CMSTT with τ 32 and subjet
b tagging. The multijet background is obtained from sideband regions, similar to
the approach developed for the 7 and 8 TeV analyses. The only difference is that the
mistag rate is parametrised as function of p instead of p T . This has been shown to
give more stable results because the Lorentz boost is roughly constant in a given
interval of p, whereas it depends on the rapidity when considering an interval in
p T . Events are classified into six categories based on the number of subjet b tags
and the rapidity difference between the jets. This allows to retain signal efficiency at
high masses, where the SM background decreases rapidly. The six categories have
different background compositions, which allows to constrain the tt background in
a simultaneous fit to data. This is exploited in a combination with analyses in the
141
magnitude larger than the irreducible background from tt production. The same is
observed for semi-merged events, which result in a similar sensitivity as fully merged
events for resonance masses of about 1 TeV. For masses of 1.5 TeV and higher, the
signal efficiency for the semi-merged selection is lower by a factor of four and more.
Since later analyses in the all-hadronic channel focus on high resonance masses,
semi-merged events are not considered any longer.
The sensitivity of the analysis could be largely improved by a follow-up analysis,
using four times more data recorded at 8 TeV [805]. The same techniques as in
the 7 TeV result are employed, particularly the CMSTT is used to select two ttagged jets [805]. Relative to the 7 TeV result, the sensitivity improves because of
the increased
√
s and the larger dataset, while the ratio of reducible multijet to the
irreducible tt background is the same. Advanced top tagging algorithms allowed
for an improvement on these results using the same data [806]. The performance
of the CMSTT is enhanced by using subjet b tagging and a selection of τ 32 < 0.7.
In addition, the HEPTopTagger, applied on CA jets with R = 1.5, is used to gain
sensitivity to low mass resonances in the all-hadronic channel. Events failing the t
tagging selection based on the CMSTT, are tested for two HEPTopTagger t-tagged
jets. The use of subjet b tagging and τ 32 makes the background estimation for multijet
production more complicated. While the overall strategy is the same, correlations in
the variables m tt , p T , τ 32 and the subjet b tagging discriminant are observed. The
mistag probability in a given bin of m tt is therefore parametrised by the three variables
p T , τ 32 and the value of the subjet b tagging discriminator. For events selected with
the HEPTopTagger, the background estimation is obtained from a sideband region
where the jet mass and pairwise subjet masses are inverted. The use of better top
tagging techniques results in a large suppression of the multijet background, as shown
in Fig. 5.7. This background constitutes only about 30% of the total background for
events with two b-tagged subjets, and increases to about 70% for one b-tagged subjet.
The better background suppression improves the 95% CL upper cross section limits
on the production of a narrow resonance by nearly 50% for masses between 1 and
2 TeV.
In analyses of 13 TeV data, CMS uses the soft drop mass, τ 32 and subjet b tagging
applied on the soft drop subjets to identify t jets. First results in the all-hadronic
channel have been obtained on a dataset corresponding to 2.6 fb
−1 [807]. The same
algorithm is used in the latest CMS analysis in this channel using 35.9 fb
−1 [808].
The algorithm shows a very similar performance as the CMSTT with τ 32 and subjet
b tagging. The multijet background is obtained from sideband regions, similar to
the approach developed for the 7 and 8 TeV analyses. The only difference is that the
mistag rate is parametrised as function of p instead of p T . This has been shown to
give more stable results because the Lorentz boost is roughly constant in a given
interval of p, whereas it depends on the rapidity when considering an interval in
p T . Events are classified into six categories based on the number of subjet b tags
and the rapidity difference between the jets. This allows to retain signal efficiency at
high masses, where the SM background decreases rapidly. The six categories have
different background compositions, which allows to constrain the tt background in
a simultaneous fit to data. This is exploited in a combination with analyses in the