
3
3 Signal and background simulation
Each signal model is characterized by key parameters. The bulk graviton model is character-
ized by two free parameters: the mass of the first Kaluza–Klein excitation of a spin-2 boson
(the Kaluza–Klein bulk graviton), and the ratio e
κ=κ√8π/MPl, with κbeing the unknown
curvature scale of the extra dimension and MPl the Planck mass. A scenario with e
κ=0.5,
resulting in resonances with a width smaller than the detector resolution is considered in this
analysis, as motivated in Ref. [59]. The radion model is also characterized by two parame-
ters: rc, the compactification radius, and ΛR, the ultraviolet cutoff of the theory. The scenario
with κrcπ=35 and ΛR=3 TeV [59] is considered in this analysis. The HVT model is char-
acterized in terms of four parameters: the mass of the W′and Z′resonance; a coefficient cF,
which scales the couplings of the additional gauge bosons to fermions; cH, which scales the
couplings to the Higgs boson and longitudinally polarized SM vector bosons; and gV, repre-
senting the typical strength of the new vector boson interaction. Two scenarios are considered
in this analysis: HVT model B, corresponding to gV=3, cH=−0.98, and cF=1.02 [17]; and
HVT model C [17], corresponding to gV=1, cH=1−3, and cF=0. In both scenarios, the
new resonances have a narrow decay width and large branching fraction to vector boson pairs,
while the fermionic couplings are suppressed. In the HVT model C, which has no fermionic
couplings, the resonances would be produced at the LHC exclusively via the VBF mode.
Monte Carlo (MC) simulated events of the radion, bulk graviton, and HVT resonance sig-
nal processes are generated at leading order (LO) in perturbative quantum chromodynamics
(QCD) with MADGRAPH5 aMC@NLO versions 2.4.2 and 2.6.0 [60]. The parton showering and
hadronization is simulated with PYTHIA versions 8.205 and 8.230 [61], for 2016 and 2017–2018
detector conditions, respectively. The NNPDF 3.0 [62] LO parton distribution functions (PDFs)
are used together with the CUETP8M1 [63] and CP5 [64] underlying-event tunes in PYTHIA for
2016 and 2017–2018 conditions, respectively. The signal cross sections are computed at next-to-
LO (NLO) with MADGRAPH5 aMC@NLO with the PDF4LHC15 100 PDF set [62, 65–69].
Simulated event samples of the SM background processes are used to develop the analysis
strategy and create templates for distributions used in the comparison with data. The QCD
multijet production is simulated with three generator configurations: PYTHIA only, the LO
mode of MADGRAPH5 aMC@NLO [70] interfaced with PYTHIA for the parton shower evolution
and matching (MG+PYTHIA8 in the following), and HERWIG++ 2.7.1 [71] with the CUETHS1
tune [63]. Top quark pair (tt), single top quark, and boson pair production are modelled at
NLO with POWHEG v2 [72] interfaced with PYTHIA. The production of W+jets and Z+jets
(V+jets) is simulated at LO with MADGRAPH5 aMC@NLO interfaced with PYTHIA. The same
underlying-event tunes as used in the signal event samples are used in the background event
samples. A correction [73] is applied to the simulated V+jets events to match the pTdistribution
of the vector bosons computed at LO in QCD to the one predicted at NLO in QCD, and another
correction [74] is used to account for NLO electroweak effects at high pT. The NNPDF 3.1 [75]
next-to-NLO (NNLO) PDFs are employed for simulated V+jets events.
All samples are processed through a GEANT4-based [76] simulation of the CMS detector. To
simulate the effect of pileup collisions, additional inelastic events are generated using PYTHIA
and superimposed on the hard-scattering events. The simulated events are weighted to re-
produce the distribution of the number of reconstructed pileup vertices observed in the 2016,
2017, and 2018 data separately. While the detector components and conditions varied across
the three years of data taking, the detector performance relevant to this analysis, in particular
with regard to the mAK8
jj and mAK8
jet scale and resolution, was very similar [77, 78]. We therefore
combine the simulated event samples corresponding to the three years of data taking, weight-