Models of physics beyond the Standard Model often require new particles at the TeV energy scale that couple to quarks and/or gluons and decay to jets. So far no sign of these new resonances was found by LHC experiments. The searches have focused on single production of resonances decaying in two jets, or pair-production in final states with four or more jets.
We want to extend the resonance search to unexplored trijet final states using proton-proton collision data collected by the CMS detector at a center-of-mass energy of 13 TeV. The signal benchmark model is the production of a new resonance (R1) which decays to a quark/gluon and a second resonance (R2), which in turn decays to two quarks/gluons. These particles are predicted in theories that foresee the existence of heavy partners of SM quarks or the existence of extra spatial dimensions.
The experimental signature is defined by the resonance mass ratio R between R2 and R1, a free parameter of the theory. For small values of R, the resonance R2 is produced with large Lorentz boost and its decay products are collimated, forming a single reconstructed jet in the detector; the final state presents two jets (dijet category). For higher values of R, three resolved jets are reconstructed (trijet category). We plan to use the CMS technique of "data scouting", based on fast online event reconstruction and small recorded event sizes, to lower trigger thresholds and gain access to the, otherwise inaccessible, mass region below 1 TeV.
Jet substructure has a central role in the dijet channel. The radiation pattern inside these jets can be exploited to distinguish between the massive jet coming from the R2 decay (signal) and jets in QCD events originating from the hadronization of single quarks/gluons (background). We plan to explore the impact of modern data science technologies, such as deep learning, in jet substructure identification, in order to boost the sensitivity of this search to new physics signals.
This research project fits into possible scenarios where new physics beyond SM is accessible at the LHC energy but not yet observed: the new particles may be weakly-coupled and hidden within large SM backgrounds; the interesting signal events might be rejected by the standard triggers; or, more simply, new physics events might be recorded in the data but there is no dedicated analysis in the corresponding final state. In this context, we focus on the trijet resonances predicted by well-motivated theories beyond the SM, and use this model as a signal benchmark to study a new experimental signature at LHC.
One of the main experimental challenges is related to the use of the data scouting stream. These data has been already employed in previous searches for sub-TeV dijet resonances in CMS [1,2] only by using calorimetric jets reconstructed at HLT which have no jet substructure information associated. In order to study the substructure properties of the jets coming from boosted resonance decays, we will follow a new strategy of using the particle-flow candidates reconstructed at HLT to form the jets employed in the data analysis. Beyond the immediate scientific impact which motivates this proposal, the development of this new technique can also be applied in other new physics searches which benefit from lower trigger thresholds to extend the sensitivity at low-mass. Finally, the data scouting paradigm is expected to become one of the key elements of the trigger and data acquisition strategies at future colliders (such as the High-Luminosity phase of LHC, starting in 2026) in order to deal with the huge amount of data expected.
In the context of this project, we plan to explore the impact of modern data science technologies, such as deep learning. We envision the use of classifiers based on computing-vision techniques (e.g. convolutional neural networks) to boost the sensitivity of this search by tagging boosted resonances decaying to close-by jet pairs. Jet substructure at the LHC has been a particularly active field for machine learning techniques [3] as jets contain O(100) particles whose properties and correlations may be exploited to identify physics signals. The high dimensionality and highly correlated nature of the phase space makes this task an interesting test for machine learning techniques.
[1] [CMS Collaboration], "Search for narrow resonances in dijet final states at sqrt(s)=8 TeV with the novel CMS technique of data scouting'', PRL 117 031802 (2016), arXiv:1604.08907
[2] [CMS Collaboration], "Search for narrow and broad dijet resonances in proton-proton collisions at sqrt(s)= 13 TeV and constraints on dark matter mediators and other new particles", arXiv:1806.00843, Submitted to Journal of High Energy Physics
[3] A. J. Larkoski et al., "Jet Substructure at the Large Hadron Collider: A Review of Recent Advances in Theory and Machine Learning'', arXiv:1709.04464