Boosted b-jet tagging
Many searches for new physics relies on identifying top and bottom quarks in the final state. Top quarks have multiple handles for identification (the top mass, the W mass, the constituent b jet, spin correlations, etc), but bottom jets are generally identified via the secondary vertex.
A "track-tag" looks for charged tracks inside a jet which don't originate from the primary vertex. This is strong evidence that the jet originated from a bottom quark, allowing the jet to be tagged as a b jet.
But when quarks become highly boosted:
- Tracks become collimated or overlap, reducing reconstruction efficiency and secondary vertex resolution.
- Light jets start to contain more tracks, and more easily fake a b tag.
- Light jets can contain b quarks (via gluon splitting).
These problems cause the b tagging efficiency to drop as jet energy increases, while the fake rate rise dramatically; this hinders the search for new physics in the TeV regime by severely diminishing sample purity.
The $\mu_x$ tag
My group (me, Z. Sullivan and D. Duffty) has proposed a new type of b tag that discriminates using the angle between a muon and the energetic core of a b jet. This "$\mu_x^{}$" method uses more information from the detector (tracker, calorimeter, muon system) to find b jets. Additionally, the $\mu_x^{}$ tag is much more dependent on the muon's angle than its $p_{T}^{}$, which is measured with much greater accuracy at high $p_{T}^{}.$
The figures below demonstrate that the efficiency and fake rate of $\mu_x^{}$ tagging is flat w.r.t. jet energy and are robust to pileup (dotted line). Thus, the tag will allow LHC searches to carve out previously unreachable regions of parameter space for particles beyond the standard model, including a leptophobic Z' and a charged Higgs boson. Implementing the $\mu_x$ tag at ATLAS or CMS will add a formidable tool to TeV physics at the LHC and beyond.
For much more detailed information about the $\mu_x^{}$ tag, please see the original paper or an improved version. Both papers include links to a Delphes module which implements the tag, available on github. Or you can skim through some of my talks on the subject.