I was thinking more about the validation plots in !63242 (merged). In particular, I took another look at the
MuGirl muon distributions. The vast majority of the built muons are complete garbage as they do not have hits in the middle or outer station:
Another glance at the MuonCombined CPU footprint, the 2 most expensive algorithms are the MuonInDetToMuonSystemExtensionAlg and MuGirl. The former is the beginning of the inside->out reconstruction chain. Inner detector tracks are roughly matched to segments or hits in the MS and then properly extrapolated through the stations. It does not matter whether the hits are in the inner, middle or outer station. As long as at one matching is successful, the track is considered as candidate. The obtained track parameters are then used segment seeding in MuGirl.
Fortunately, the algorithm has a kill switch to ignore anything from the NSW. We can save some extra CPU by activating the switch. Either via a postExec in the current mc and data reprocessing campaigns
--postExec 'RAWtoALL:cfg.getEventAlgo(\"MuonInDetToMuonSystemExtensionAlg\").noNSWforSeeding = True;cfg.getEventAlgo(\"MuonInDetToMuonSystemExtensionAlg_LRT\").noNSWforSeeding = True;'
and/or by accepting this MR. The test on 50k dimuon particle gun events demonstrates that the impact is largest in in the transition region and at super high eta:
In terms of efficiency against pt, losses are only visible below 4.5 GeV.
None of the lost muons passes the medium working point
or could be potentially resurrected by the LowPt / Loose muon selection
The full set of validation plots is given here.
I should also mention that these super low-pt muons are swamed by pile-up fakes. @hpang checked the muon origin in 5M JPsi events:
|Algorithm||Needed time in 1k events [ms]||with postExec||Saving|
So let's make our little friend here happy and let's save some CPU
The philosopher of the day: