Events are selected online by combining two different trigger selections to identify b
jets, both using the CSV algorithm. For the first trigger selection, four jets with p
T
>
30 GeV and |η| < 2.4 are required. The latter requirement ensures that the jet lies within
the tracker acceptance. Of those four jets, two are required to have p
T
> 90 GeV and at
least three jets are required to be tagged as b jets. The second trigger selection requires
four jets with p
T
> 45 GeV and at least three of those jets identified as b jets.
Events are selected offline by requiring at least four b tagged jets with p
T
> 30 GeV
and |η| < 2.4. The selected jets are combined randomly into pairs to form two Higgs boson
candidates with masses m
H
1
and m
H
2
. For the LMR, HH candidates are chosen from the
four selected jets such that |m
H
−120 GeV| < 40 GeV for each candidate Higgs boson. For
the MMR the H candidates are selected using ∆R =
√
(∆η)
2
+ (∆φ)
2
less than 1.5, where
∆η and ∆φ are the differences in the pseudorapidities and azimuthal angles (in radians)
of the two jets.
In the two-dimensional space defined by the reconstructed masses of the two Higgs
boson candidates, H
1
and H
2
, a circular signal region (SR) is defined with R < 1, where
R is defined as:
R =
s
m
H
1
− M
r
2
+
m
H
2
− M
r
2
(4.1)
The central mass value (M) is the average of the means of the m
H
1
and m
H
2
distributions
for simulated signal events and the parameter r is set to 20 GeV. The centers of these
circular regions have been determined separately for the LMR and MMR and found to
be 120 and 125 GeV, respectively. If there are multiple HH candidates in an event, the
combination that minimizes R
2
is used.
After these event selection criteria are applied, the dijet invariant mass resolution
for m
H
1
and m
H
2
is approximately 10–13%, depending on the p
T
of the reconstructed
Higgs boson, with a few percent shift in the value of the mass peak, relative to 125 GeV.
The Higgs boson mass resolution is further improved by applying multivariate regression
techniques similar to those used in the searches for SM Higgs bosons decaying to bb in
CMS [39, 40]. The regression estimates a correction that is applied after the standard
CMS jet energy corrections [34, 41], and it is computed for individual b jets to improve
the accuracy of the measured energy with respect to the b quark energy. To this end, a
specialized boosted decision tree [42] is trained on simulated b jets from tt events, with
inputs that include observables related to the jet structure and b tagging information. The
average improvement in the Higgs boson mass resolution, measured with simulated signal
samples, is 6–12%, depending on the p
T
of the reconstructed Higgs boson. The use of the
regression technique increases the sensitivity of the analysis by 5–20% depending on the
mass hypothesis. The regression technique is validated with data samples of Z → (ee, µµ)
events with two b tagged jets and in tt-enriched samples [39]. The cumulative selection
efficiencies of the selection criteria described above for the graviton and radion signal
benchmarks are reported in figure 1.
The reconstructed resonance mass (m
X
), computed as the invariant mass of H
1
and
H
2
, is displayed for simulated signal events with different mass hypotheses in figure 2. In
– 4 –
