Many canonical b-tagging algorithms are based on the estimation of properties of track-in-jet ensemble. With a limited number of reconstructed tracks from B-hadron decay ( ~4 in the low-pt regime) this approach works well for low-pt jets where the number of fragmentation tracks is ~equal to the number of B-hadron tracks. However, with increasing jet-pt the number of B-hadron tracks drops (<3) due to reconstruction inefficiency, while the number of fragmentation tracks grows significantly (>50). This makes the track ensemble based estimations inefficient, resulting in quick b-tagging efficiency drop with jet-pt increase. An alternative approach is a search for (a) single track(s) in a jet which is most probably produced in the B-hadron decay. This solves the track-in-jet multiplicity problem and therefore is efficient in the high-pt domain. The new approach can be tried using an already implemented InDetTrkInJetType classification tool.
The proposed QT is to implement a high-pt b-tagging algorithm based on counting tracks with largest b-hadron probability returned by the InDetTrkInJetType tool.
The performance of the implemented algorithm will be compared with default ATLAS b-tagging algorithms in the high-pt (>=1TeV) domain using different samples.
The track classification BDT in the InDetTrkInJetType tool can be optimised/retrained to increase the new tagger efficiency as well. The InDetTrkInJetType tool will be supported during the task in particular towards an implementation as a general tool in the Flavour tagging infrastructure.
The progress will be reported in the Flavour Tagging algorithms and software meetings and the corresponding software will be made available in gitlab and adequately documented.