sPHENIX tracking discussion

[Anthony Frawley <afrawley AT fsu.edu>]

Hello Michael,

Thanks for the report!

The tracking efficiency plots you show here are for tracks with non-zero silicon hits. For that case, your recently achieved 80% reconstruction efficiency in MB + 100 pion events is not out of line with what we saw in the S&C review plots, where we used PHTpcTracker seeding. Those had ~ 90% overall efficiency with ~ 91% silicon matching efficiency, for a combined efficiency of 81%.

Some part (~​5%) of the silicon matching inefficiency is understood as being due to the silicon tracklet not being found by the silicon track seeding, which needs to find 3 MVTX hits or it does not work. The other 5% of thre inefficiency needs more investigation.

The number of TPC clusters per track is more concerning. It is strongly peaked in the high 40's for PHTpcTracker seeding, with almost no tracks having < 35 TPC clusters. The presence in your CA seeder plots of a large number of tracks with 20-30 TPC clusters is the main problem that needs to be attacked now. 

Cheers

Tony

---

From: sPHENIX-tracking-l <sphenix-tracking-l-bounces AT lists.bnl.gov> on behalf of Michael Joseph Peters via sPHENIX-tracking-l <sphenix-tracking-l AT lists.bnl.gov>
Sent: Wednesday, June 9, 2021 2:27 AM
To: sphenix-tracking-l AT lists.bnl.gov <sphenix-tracking-l AT lists.bnl.gov>
Subject: [Sphenix-tracking-l] CA Seeding: current status

 

Hi all,

I was able to fix the issues with the two-iteration CA seeding, so here are the results intended for yesterday. All tests were run using the default G4_Tracking.C settings, with G4Hits files from 

/sphenix/user/frawley/new_macros_april27/macros/detectors/sPHENIX/hits_output/EmbedOut_0_20fm_50kHz_bkg_0_20fm-*.root

In addition to the second iteration, a number of minor refinements were applied to the CA seeding code, including, among other things, a reintroduction of cluster-chain-level seed merging and the application of a cluster cleaner before the Kalman filter (so that bad clusters don't contaminate the initial momentum estimate). These additions lead to something like a 5-10% improvement in efficiency, as shown below, while the second iteration on top of that unfortunately has very little benefit:

A comparison of nhits is shown below; you can see the effect of the cluster cleaner moving 54-hit tracks down a few bins. The red and blue histograms are essentially on top of one another:

There are also more tracks that fail at the ACTS level in the newer versions, likely an effect of pre-KF seed merging producing some weird-shaped tracks (which is the tradeoff for higher efficiency overall).

The quality of the pt and pz estimates does not significantly change between any of the versions.

So we now have a CA seeder that's roughly 80% efficient (instead of 70-75% efficient) above 0.5 GeV in pt in 50kHz pileup. I'll be scrapping the second iteration and getting together a pull request implementing the other changes.

The propagator still does more harm than good, unfortunately, which is why it wasn't included in any of these tests.

Thanks,

Michael Peters