Citron Lab

CMS searches

We work on several searches for exotic signatures of dark sectors using the CMS detector at CERN. We focus on long lived particles that travel for a measurable distance through the detector before decaying. I am particularly interested in search strategies that use non-standard reconstruction techniques with low level detector information. We also make use of machine learning to combat challenging backgrounds. This allows us to push thresholds as low as possible without being swamped by backgrounds. Finally, making sure CMS “triggers” events with LLPs is particularly challenging. My group designed and implemented several complementary dedicated trigger selections. We also make use of exotic scouting and parking trigger “streams” for analysis.

I am interested in the sensitivity that can be achieved by future colliders and we are currently working on estimating sensitivity to LLPs at the FCC using similar strategies to those carried out for CMS.

Non-standard reconstruction

CMS cross section showing LLP decay that can be reconstructed with timing signature
Displaced decays occurring near CMS ECAL can be reconstructed using timing

Searching for decays of LLPs can be very challenging because the CMS detector was not originally designed to be able to reconstruct particles that decay far from the collisional point. I helped to figure out new ways of gaining sensitivity to these signatures by using low level CMS detector information to reconstruct from scratch. This includes using calorimeter timing to search for delayed jets and using clusters of hits in the muon system to search for decays occurring inside. These techniques have helped provide substantial new sensitivity to a range of models.

Machine learning

ROC curves for HNL displaced jet tagger
ROC curve for the HNL search tagger

The current frontier in LLP searches is pushing sensitivity to lower energy, highly challenging signatures. These face huge backgrounds from SM processes that must be rejected. I work on several searches that exploit Machine Learning techniques to efficiently reject background processes while maintaining signal. This includes using a DNN to tag displaced jets produced in Heavy Neutral Lepton decays and using a BDT to search for low mass dark QCD models.

Triggering on LLPs

Turn on curves for signal models triggered by standard and dedicated triggers.
Trigger efficiency gain from dedicated delayed jet triggers

Before any collisional events can be reconstructed they must be recorded by the CMS trigger system. This must rapidly decide only ~1/40,000 events to save with very limited information. I have worked on efforts to include exotic reconstruction in the CMS trigger selection to greatly increase acceptance for LLP models. I also work on searches using the specialized “scouting” and “parking” data streams to significantly reduce thresholds.