4.2 Measurements Using Jet Substructure
113
While the H boson couplings to vector bosons and third generation quarks and
leptons have been measured, its couplings to quarks and leptons of the second generation have not been experimentally accessible so far. For H → μμ, a significance of three standard deviations has been reached with 137 fb
−1 of data [616].
The coupling to charm quarks is even more difficult to establish experimentally,
even though the branching fraction B(H → cc) is about a factor of 100 larger than
B(H → μμ). However, with a value of 0.0288, B(H → cc) is about 20 times
smaller than B(H → bb), and has much higher background levels at hadron colliders. A first attempt to study H → cc at the LHC has been made by ATLAS in VH
production [617]. The key to this analysis is a modification of the b tagging algorithm
to target c quark jets, while rejecting b jets. Charm jet identification is particularly
challenging as c hadrons have shorter lifetimes and decay to fewer charged particles
than b hadrons. A c-jet efficiency of 41% is achieved for background efficiencies of
25% for b jets and 5% for light quark or gluon jets. In this analysis, the resolved
topology has been targeted, without considering boosted H production. Using the
distribution of the dijet mass of c-tagged jets, observed (expected) upper limits at
95% confidence level (CL) on the SM signal strength modifier of 110 (150
+80
−40 ) have
been obtained. A recent analysis by CMS obtains a similar performance for small-R
jet c tagging, where a working point corresponding to a c jet efficiency of 28% has
been chosen with misidentification rates of 15% for b jets and 4% for light quark
or gluon jets [618]. A multivariate discriminant using c tagging information and
kinematic variables of the VH system in resolved final states as input is used for
the H → cc signal extraction. The observed (expected) upper limit on the signal
strength modifier for SM VH production with H → cc is 75 (38
+16
−11 ). In the same
publication, CMS analyses boosted H → cc production with p T > 200 GeV, where
the signal is extracted via a binned maximum likelihood fit to the soft-drop mass
distribution of large-R jets. Since the H production cross section for p T > 200 GeV
amounts to approximately 5% of the total cross section only, it is paramount to
choose R appropriately to keep the signal efficiency as high as possible. To meet this
goal, anti-k T jets with R = 1.5 are used (see Sect. 2.2.4). A deep neural network is
trained for the identification of H → cc decays, exploiting information related to jet
substructure, flavour, and pileup simultaneously. The use of an adversarial training
procedure results in a discriminator which is largely decorrelated from the jet mass,
while preserving most of the method’s discriminating power. The achieved cc tagging performance can be expressed in efficiencies for three working points of 23,
35, and 46% for cc, with misidentification rates of 9, 17, and 27% for b jets and 1,
2.5, and 5% for light quark and gluon jets, respectively. The sensitivity is further
improved with an event selection based on a multivariate discriminator derived from
kinematic variables of the high- p T VH system, uncorrelated to the signal jet mass and
cc tagging discriminant. The statistical analysis uses distributions in the soft drop jet
mass in nine categories (three lepton channels and three cc tagger working points).
It results in an observed (expected) upper limit of 71 (49
+24
−15 ). It is a remarkable
achievement that the boosted analysis has a comparable sensitivity as the resolved
one, considering the smallness of the cross section in the boosted fiducial region.
A significant fraction of events overlap between the resolved and boosted channels,
113
While the H boson couplings to vector bosons and third generation quarks and
leptons have been measured, its couplings to quarks and leptons of the second generation have not been experimentally accessible so far. For H → μμ, a significance of three standard deviations has been reached with 137 fb
−1 of data [616].
The coupling to charm quarks is even more difficult to establish experimentally,
even though the branching fraction B(H → cc) is about a factor of 100 larger than
B(H → μμ). However, with a value of 0.0288, B(H → cc) is about 20 times
smaller than B(H → bb), and has much higher background levels at hadron colliders. A first attempt to study H → cc at the LHC has been made by ATLAS in VH
production [617]. The key to this analysis is a modification of the b tagging algorithm
to target c quark jets, while rejecting b jets. Charm jet identification is particularly
challenging as c hadrons have shorter lifetimes and decay to fewer charged particles
than b hadrons. A c-jet efficiency of 41% is achieved for background efficiencies of
25% for b jets and 5% for light quark or gluon jets. In this analysis, the resolved
topology has been targeted, without considering boosted H production. Using the
distribution of the dijet mass of c-tagged jets, observed (expected) upper limits at
95% confidence level (CL) on the SM signal strength modifier of 110 (150
+80
−40 ) have
been obtained. A recent analysis by CMS obtains a similar performance for small-R
jet c tagging, where a working point corresponding to a c jet efficiency of 28% has
been chosen with misidentification rates of 15% for b jets and 4% for light quark
or gluon jets [618]. A multivariate discriminant using c tagging information and
kinematic variables of the VH system in resolved final states as input is used for
the H → cc signal extraction. The observed (expected) upper limit on the signal
strength modifier for SM VH production with H → cc is 75 (38
+16
−11 ). In the same
publication, CMS analyses boosted H → cc production with p T > 200 GeV, where
the signal is extracted via a binned maximum likelihood fit to the soft-drop mass
distribution of large-R jets. Since the H production cross section for p T > 200 GeV
amounts to approximately 5% of the total cross section only, it is paramount to
choose R appropriately to keep the signal efficiency as high as possible. To meet this
goal, anti-k T jets with R = 1.5 are used (see Sect. 2.2.4). A deep neural network is
trained for the identification of H → cc decays, exploiting information related to jet
substructure, flavour, and pileup simultaneously. The use of an adversarial training
procedure results in a discriminator which is largely decorrelated from the jet mass,
while preserving most of the method’s discriminating power. The achieved cc tagging performance can be expressed in efficiencies for three working points of 23,
35, and 46% for cc, with misidentification rates of 9, 17, and 27% for b jets and 1,
2.5, and 5% for light quark and gluon jets, respectively. The sensitivity is further
improved with an event selection based on a multivariate discriminator derived from
kinematic variables of the high- p T VH system, uncorrelated to the signal jet mass and
cc tagging discriminant. The statistical analysis uses distributions in the soft drop jet
mass in nine categories (three lepton channels and three cc tagger working points).
It results in an observed (expected) upper limit of 71 (49
+24
−15 ). It is a remarkable
achievement that the boosted analysis has a comparable sensitivity as the resolved
one, considering the smallness of the cross section in the boosted fiducial region.
A significant fraction of events overlap between the resolved and boosted channels,