C. Veelken
Searches for H H production in channels without b-jets have in general smaller signal yields, but are typically less contaminated by backgrounds than those with b-jets. As the sensitivity of searches without b-jets is mainly limited by statistical uncertainties, we expect that their sensitivity will scale better with the integrated luminosity than the b-jets final states. ATLAS has recently investigated both the H H → γγW W∗[226,490] and H H → W W∗W W∗[457] final states, while CMS has studied for the first time H H → τ+τ−τ+τ− [491]. The branching fractions of these channels for SM H H bosons are 9.85 · 10−4, 4.67 · 10−2, and 4.00 · 10−3, respectively. Phenomenological studies of the
H H → γγW W∗and H H → W W∗W W∗channels have been published in Refs. [452] and [452,492–
5.5. H H , other signatures: status and perspectives 145
5.5.1 H H → γγW W∗
Events in the H H → γγW W∗ channel are selected in the final state γγ`νj j, covering 34.3% of the total H H → γγW W∗ signal. The search looks for both SM non-resonant and resonant H H production in the mass range between 260 and 500 GeV [226,490]. The signal is extracted by means of a maximum-likelihood fit to the distribution in mass of the photon pair, mγγ. In the non-resonant analysis and in the search for resonances of mass 400 GeV and higher, the pT of the di-photon system, pTγγ, is required to exceed 100 GeV, in order to reduce backgrounds. Within a mass window centred on mH = 125.09 GeV and of size equal to 2 times the experimental resolution on mγγ, 7 events are observed in the data, in agreement with an expected background of 6.1±2.3 events. In the search for resonances of mass below 400 GeV, where no pTγγ> 100 GeV cut is applied, 33 events are observed in the data, while 24 ± 5.0 events are expected from background processes. The expected signal contribution from SM non-resonant H H production amounts to 3.8 · 10−2(4.6 · 10−2) in case the requirement on pTγγis applied (not applied). The distributions in mγγ, obtained when no cut on
pTγγis applied and with the pγγT > 100 GeV cut applied, are shown in Figure5.19. The event yields, as well as the distributions in mγγ, observed in the data agree with the SM expectation in both cases. As no evidence for a H H signal is observed, the analysis proceeds by setting an upper limit on the
H H signal cross section. The observed (expected) limit on the cross section for non-resonant H H
production with SM kinematics amounts to 230 (160) times the SM prediction. In the corresponding Run 1 analysis, 4 events were observed in the signal mass window of the mγγdistribution, compared to 1.65±0.47 events expected from background processes and 7.2·10−3signal events expected from SM non-resonant H H production, and an observed (expected) upper limit of 1150 (680) times the SM cross section was set.
Figure 5.19: Distribution in mγγobserved in the ATLAS analysis of H H → γγW W∗, compared to the expected contribution from SM single Higgs boson plus SM non-resonant H H production (dash- dotted line) and other backgrounds (dashed line), when no cut on the pTof the di-photon system
is applied (left) and with a cut of pγγT > 100 GeV applied (right) [226].
5.5.2 H H → W W∗W W∗
The ATLAS analysis of H H → W W∗W W∗[457] selects events in a combination of final states with 2, 3, and 4 leptons. In the di-lepton channel, the contamination from background processes is reduced by requiring the two leptons to be of the same charge. The combination of the 2, 3, and 4 lepton final states covers 10.7% of the total H H → W W∗W W∗signal. Similar to the H H → γγW W∗ analysis, the analysis of the H H → W W∗W W∗final states exploits both SM non-resonant and BSM resonant production in the mass range 260 to 500 GeV. In addition, the presence of heavy scalars S of mass 135 < mS< 165 GeV originating from the decay of resonances X of mass 280 < mX< 340 GeV,
X → SS is probed. An automatic optimisation of event selection criteria ("rectangular cuts"), im- plemented in the package TMVA [367], is employed in order to enhance the ratio of signal over background events, before the signal gets extracted by means of a maximum likelihood fit to the event yields in nine event categories. Events selected in the di-lepton channel are analysed in three event categories, containing events with either two electrons (ee), two muons (µµ), or one elec- tron plus one muon (eµ), respectively. In the 3 lepton channel, events containing zero and events containing one or more pairs of leptons of the same flavour and opposite charge are analysed sep- arately. Events selected in the 4 lepton channel are analysed in four event categories, based on the multiplicity of same flavour and opposite charge lepton pairs and the mass of the 4 lepton system. The event yields observed in the data is compared to the SM expectation for the H H signal and for background processes in Figure5.20. The data is in agreement with the SM expectation. The analysis proceeds by setting upper limits on the H H signal cross section. The combined fit of the nine event categories yields an observed (expected) limit on the cross section for non-resonant H H production with SM kinematics of 160 (230) times the SM prediction.
Figure 5.20: Event yields observed in the ATLAS analysis of H H → W W∗W W∗[457], compared to the expected contribution of background processes and to a non-resonant H H signal of SM kine- matics and production rate amounting to 20 times the SM value. The symbol NSFOSdenotes the number of lepton pairs of same-flavour and opposite-charge, while the low and high m4`categories refer to events in which the mass of the 4 lepton system is below and above 180 GeV, respectively.
5.5.3 H H → τ+τ−τ+τ−
The CMS search for H H → τ+τ−τ+τ− [491] is performed in the final state with 2 leptons and 2
τh, corresponding to 31.2% of the total H H → τ+τ−τ+τ− signal. The analysis is performed in
six event categories, based on the flavour of the leptons (ee, µµ, eµ) and on their charge (same- sign, opposite-sign). Events containing pairs of leptons of the same flavour, opposite charge, and mass within the range 70 to 110 GeV are rejected, in order to remove background arising from
Z /γ∗→ `+`−Drell-Yan production. The multi-jet background is estimated from data, while the contribution of other backgrounds is modelled using the MC simulation. The signal extraction is based on a maximum-likelihood fit to the distribution in mass of the 2 leptons plus 2τhsystem
in case of the three event categories containing opposite-sign lepton pairs. In the event categories with same-sign lepton pairs, the small number of background events precludes the usage of a shape analysis and the event yields are instead used as input to the maximum-likelihood fit ("cut and
5.5. H H , other signatures: status and perspectives 147
count" analysis). The analysis is still blinded. Based on the expected signal acceptance and effi- ciency and on the expected background contamination, the analysis is expected to be sensitive to resonant H H signals produced with a cross section of order 10 pb.
5.5.4 Potential improvements
A common feature of the three channels H H → γγW W∗, W W∗W W∗, andτ+τ−τ+τ−is that their sensitivity is limited by small signal yields and sizeable statistical uncertainties with the present data. Significant gains in analysis sensitivity have been achieved in the "established" channels
H H → b ¯bb ¯b, b ¯bγγ, and b ¯bτ+τ−during LHC Runs 1 and 2, thanks to improvements in the analysis
methods (up to a factor 2-3 improvement in sensitivity for the same luminosity). Significant poten- tial exists to likewise improve the sensitivity of the "new" channels H H → γγW W∗, W W∗W W∗, andτ+τ−τ+τ−.
In the H H → γγW W∗channel, potential improvements include the use of multivariate meth- ods to enhance the separation of the H H signal from backgrounds, the reconstruction of the mass of the H H system by means of an algorithm similar to the “High Mass Estimator" (HME) algorithm developed for the analysis of resonant H H production in H H → b ¯bW W∗, described in Ref. [203], the replacement of the pγγT > 100 GeV cut by event categories based on pγγT , and the extension of the analysis to theγγ`ν`ν and γγj j j j final states.
Potential improvements to the sensitivity of the H H → W W∗W W∗channel comprise the sub- stitution of the "rectangular cuts" that are employed for separating the H H signal from backgrounds by more modern multivariate methods such as BDTs or NNs, and by upgrading the analysis from a "cut and count" approach to a shape analysis, based on the output of a BDT or NN. Besides improv- ing the separation of the H H signal from the background, we expect that a shape analysis based on the output of the BDT or NN will have the further benefit of providing useful constraints to the sys- tematic uncertainties, compared to the simple "cut and count" approach. Non-prompt and fake leptons constitute a sizeable source of background in particular in the final state with 2 leptons of the same charge, where it amounts to 30-40% of all backgrounds. We expect significant reductions of this background may be achievable thanks to anticipated improvements in the identification of leptons with multivariate methods in the future.
Potential improvements to the sensitivity of the H H → τ+τ−τ+τ−channel are expected from extending the analysis to cover further final states (4 leptons, 3 leptons plus 1τh, 1 lepton plus 3τh,
4τh) and to determine reducible backgrounds other than multi-jet production from data instead of
from the MC simulation. The latter is expected to not only reduce the systematic uncertainties, but also the statistical uncertainties on the background expectation, as samples of backgrounds with large cross sections, for example Drell-Yan production, have a higher event statistics already in the LHC Run 2 data, compared to the event statistics presently available by MC simulation. Moreover, the current CMS analysis of H H production in the final state with 2 leptons and 2τhneglects the
signal contribution arising from the decays H H → τ+τ−W W∗and W W∗W W∗. We expect these decays to provide a significant contribution to the overall H H signal yield.
A further improvement in the sensitivity of the H H → τ+τ−τ+τ− channel may be achieved by using an algorithm for reconstructing the mass of the H H system, presented at the workshop. The algorithm is based on a dynamical likelihood approach [471,472] and represents an extension of the SVfit algorithm [474,475] that is used in the CMS H H → b ¯bτ+τ−analysis presented in section5.3. Measurements of the energies and momenta of the visible τ decay products and of the missing transverse energy are combined in a probability model for the H H → τ+τ−τ+τ−decay with con- straints on the mass of eachτ+τ−pair to equal mH= 125.09 GeV. Details of the algorithm are given
in Ref. [496]. The algorithm achieves a resolution on mH H, the mass of the H H system, of 22% (7%) in simulated H H → τ+τ−τ+τ− signal events in which the Higgs boson pair originates from the decay of a narrow resonance X of mass mX = 300 (500) GeV and produces a final state with 2 leptons and 2τh. The quoted resolutions include the effect that the algorithm chooses an incorrect
assignment of the 2 leptons and 2τhto the first and second H boson in 13% (2%) of simulated signal
events at mX = 300 (500) GeV, which causes the Higgs mass constraint to be applied to the wrong combinations of leptons andτh, thereby degrading the resolution on mH H. In case the algorithm
could be improved to always choose the correct assignment, the resolution on mH H would im- prove to 4% (6%) for signal events of mX = 300 (500) GeV. Distributions in the ratio of reconstructed to true mass of the H H system are shown in Figure5.21, separately for simulated H H → τ+τ−τ+τ− signal events in which the correct assignment ("correct pairing") is chosen and events in which the incorrect assignment ("spurious pairing") is chosen by the algorithm.
1 true HH /m HH m 2 − 10 1 − 10 1 10 ) true HH /m HH dN/d(m h τ h τ ll → τ τ τ τ → HH → X(300 GeV) Correct pairing Spurious pairing 1 true HH /m HH m 2 − 10 1 − 10 1 10 ) true HH /m HH dN/d(m h τ h τ ll → τ τ τ τ → HH → X(500 GeV) Correct pairing Spurious pairing
Figure 5.21: Distributions in the ratio of reconstructed to true mass of the H H system, mH H/mt r ueH H , in simulated H H → τ+τ−τ+τ− signal events of true H H mass 300 GeV (left) and 500 GeV (right) [496]. The x-axis ranges from 0.2 to 5.
In summary, we expect that the sensitivity of channels without b-jets will increase faster com- pared to the sensitivity of channels with b-jets as more LHC data becomes available in the future and more refined and sophisticated analysis techniques get utilised in the new channels. In our view, it is a worthwhile effort to study the feasibility of these new channels in preparation for the upcoming HL-LHC data-taking period.