Chapter V: Impact of data scouting
5.1 Dijet resonance searches with Run II scouting data
The first search for new physics conducted with the Run II data scouting framework was a low-mass dijet resonance search. The search was carried out on the first 12.9 fb−1 of data collected in 2016 [103]. A follow-up analysis extended the results to include a total of 27 fb−1 [104]. The search was performed using the same strategy as past CMS dijet searches, using a fit to a smooth parameterized functional form. The latest iteration of the search, performed on the 2017 CMS dataset, is in preparation as of this writing.
Trigger turn-on measurements
In the 2016 dataset, the available triggers for the dijet resonance search are the calo- scouting trigger, which selects events withHT > 250 GeV; and theHT PF-scouting trigger, which requiresHT >410 GeV.
The efficiency of the HT > 250 GeV path is measured as a function of the dijet mass, yielding a turn-on curve for the trigger. It is computed using the dedicated L1-only and minimum-bias trigger paths, as described in Section 4.4. The turn-on curve is illustrated in Figure 5.1.
The PF-scouting trigger efficiency is measured in the same way. It is found that the trigger becomes fully efficient for dijet masses between 600 and 700 GeV.
Based on the measured trigger efficiencies, it is decided to perform the 2016 dijet search using the calo-scouting dataset collected using the HT > 250 GeV trigger path. This choice yields sensitivity to signal masses in the range 0.6-1.6 TeV. Using calo jets instead of PF jets is not found to have any significant negative impact on the analysis sensitivity.
Figure 5.1: Efficiency of the HT > 250 GeV calo-scouting trigger on 2016 data.
The efficiency is measured with respect to the dijet invariant mass and is calculated using dedicated L1-only and minimum-bias scouting trigger paths [105].
For the 2017 search, the scouting trigger efficiency is measured using a different technique. A trigger is added to the scouting monitor stream that selects events having a muon with pT > 50 GeV. The set of events collected by this path is used as an unbiased reference sample to measure theHT trigger efficiency. This method is more convenient than measuring the efficiency using looserHT triggers, because there is no confounding effect from the turn-on of the L1 trigger. The turn-on curve from the 2017 measurement is shown in Figure 5.2 for the calo-scouting trigger.
The trigger reaches 99% efficiency at a dijet mass of 350 GeV.
Figure 5.2: Efficiency of the HT > 250 GeV calo-scouting trigger on 2017 data.
The efficiency is measured with respect to the dijet invariant mass and is calculated with reference to a sample of events collected with a muon trigger [106].
Comparison of HLT reconstructed objects with standard physics objects The scouting monitor dataset described in Section 4.4 makes it convenient to di- rectly compare the momenta of HLT calo jets in the scouting data with those of PF jets reconstructed offline. The percent difference between the pT of an HLT calo jet and that of the corresponding PF jet, measured in bins of HLT jet pT, is shown on the left-hand side of Figure 5.3. The bias is no larger than 4% and decreases with increasing jetpT.
The resolution of the dijet invariant mass is measured for HLT calo jets and com- pared with that for offline reconstructed PF jets. The measured resolution values are displayed on the right-hand side of Figure 5.3. It is seen that the resolution for HLT calo jets is 1-2% worse than that of offline PF jets.
Results and impact
The dijet mass spectrum obtained using the full 2016 scouting dataset is shown in Figure 5.4. The parametric functional form fits the background well, and no
Figure 5.3: Left: average percent difference between HLT calo jet pT and offline reconstructed PF jetpT, measured in bins of HLT calo jet pT. The bias is parame- terized with the smooth functional form shown in red. Right: resolution of the dijet mass in HLT (blue) and offline reconstructed (red) events [106].
excess over the smooth background shape is observed. Limits are set on a variety of theoretical models for new physics using the results of the search. Among these are new limits on the production of a Z’ resonance decaying to quarks, which improve on those shown in Figure 4.2. The limit is shown as a function of the Z’ mass in Figure 5.5.
Historical aside: dijet scouting and the 750 GeV diphoton excess
Interest in hadronic resonance searches below 1 TeV was spurred in late 2015 by the joint announcement by ATLAS and CMS of excesses in the diphoton mass spec- trum [107, 108]. ATLAS and CMS observed local excesses with significances of 3.9 and 2.6 standard deviations, respectively, at a diphoton mass of approximately 750 GeV. The global significances of the excesses were 2.1 and 1.2 standard deviations, respectively. Despite the low significance of the CMS excess, the announcement generated a large amount of attention from the theory community.
It was noted [109, 110] that the 8 TeV dijet search using data scouting [100] placed important constraints on the production of a new state at 750 GeV (see Figure 5.6).
This is because a strongly-produced resonance should generically exhibit decays to final states with jets. Data scouting therefore drew interest as a way to investigate the possible new particle (see Figure 5.7). If the diphoton excess were confirmed on
Figure 5.4: Dijet mass spectrum obtained using scouting data collected in 2016, with the fit to a parameterized functional background shape overlaid in red. The bottom panel shows the significance of the difference between the data and the fit in each bin [104].
the 2016 dataset, the discovery could be corroborated in the hadronic decay channel using the scouting dataset.
The CMS and ATLAS diphoton resonance searches were repeated with the data collected in the first half of the 2016 LHC run [112, 113]. Both searches returned null results, as did the dijet scouting search, which was made public at the same time [103]. This indicates that the excesses in the 2015 data had merely been sta- tistical fluctuations. However, the incident highlights the role that data scouting can play in searching for signatures of new physics that are not accessible through standard physics analyses.
Figure 5.5: Limits on the coupling of a leptophobic Z’ resonance decaying to quarks, computed using the results of the 2016 CMS dijet resonance search. In the region to the left of the dashed gray line, the limit is computed using the dataset collected with the calo-scouting trigger [104].
Impact on dark matter limits
The 2016 dijet search yielded strong limits on simplified models of dark matter, as illustrated in Figure 5.8. The simplified models contain a dark matter particle and a heavy mediator that couples to quarks. Because the dijet search is sensitive to direct production of the mediator in LHC collisions, the limits obtained are relatively insensitive to the mass of the dark matter particle. At mediator masses below 1.6 TeV, the exclusion limits are driven by the data scouting part of the dijet analysis.