sPHENIX tracking discussion

[Michael Joseph Peters <mjpeters AT mit.edu>]

Correction: the last email should say "All combinations of two reco modules and one truth module have either reco TPC seeding or reco matching included..." instead of "reco TPC seeding or reco silicon seeding."

---

From: Michael Joseph Peters
Sent: Thursday, December 1, 2022 3:58 PM
To: sphenix-tracking-l AT lists.bnl.gov
Subject: Re: TPC Seeding Efficiency Comparison

 

Hi all,

I spent the last couple days running through all combinations of the switches use_truth_tpc_seeding, use_truth_silicon_seeding, and use_truth_silicon_matching. Here are the results in which only one module was turned to reco, along with full truth (red) and full reco (black) efficiency for reference:

The blue points (reco matching, but truth TPC and Si seeding) were run after the build changed, and appear to have notably fewer statistics; looking into this later.

Anyway, from this plot it's fairly clear that the reco silicon seeding alone doesn't produce the odd behavior on its own. Both the reco TPC seeding and the reco matching each produce the odd behavior by themselves, but their respective inefficiencies apparently don't combine straightforwardly.

Separated for clarity, here are all the combinations of two reco modules and one truth module, again with full truth and full reco references:

All combinations of two reco modules and one truth module have either reco TPC seeding or reco silicon seeding included, so all of them produce the odd behavior. The depths of the trough seem to be fairly similar, though those involving reco TPC seeding appear to be somewhat deeper. The difference in high-pT behavior is also much clearer with two reco modules than with two truth modules.

So, overall conclusions:

- Both the TPC seeding and the matching seem to contribute to the odd behavior we're seeing, and

- The interaction between TPC seeding and matching is complicated, in terms of how they each contribute to the overall tracking efficiency.

Thanks,

Michael Peters

---

From: Anthony Frawley <afrawley AT fsu.edu>
Sent: Wednesday, November 30, 2022 10:51 AM
To: Michael Joseph Peters; sphenix-tracking-l AT lists.bnl.gov
Subject: Re: TPC Seeding Efficiency Comparison

 

Hi Michael,

OK, good, that absolves the Acts track fit. The silicon seeding or the silicon-TPC  matching are the other candidates.

Thanks

Tony

---

From: Michael Joseph Peters <mjpeters AT mit.edu>
Sent: Wednesday, November 30, 2022 10:39 AM
To: Anthony Frawley <afrawley AT fsu.edu>; sphenix-tracking-l AT lists.bnl.gov <sphenix-tracking-l AT lists.bnl.gov>
Subject: Re: TPC Seeding Efficiency Comparison

 

Hi,

The full truth seeding does not have this behavior, at least:

I'm running the other possibilities now, will update once they finish.

Thanks,

Michael Peters

---

From: Anthony Frawley <afrawley AT fsu.edu>
Sent: Wednesday, November 30, 2022 10:14 AM
To: sphenix-tracking-l AT lists.bnl.gov; Michael Joseph Peters
Subject: Re: TPC Seeding Efficiency Comparison

 

Hello Michael,

Very interesting! The question is: where does the inefficiency come in? It could be:

  -- At the silicon seeding stage.

  -- At the silicon matching stage.

  -- At the Acts fitting stage.

If you use truth TPC seeding, the silicon-TPC matching is still done normally. It would be useful to try both of these:

  -- Add the truth silicon matching flag to truth TPC seeding (skips the silicon-TPC matcher).

  -- Using full truth seeding instead (skips the seeders and silicon-TPC matcher completely).

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: Tuesday, November 29, 2022 10:32 PM
To: sphenix-tracking-l AT lists.bnl.gov <sphenix-tracking-l AT lists.bnl.gov>
Subject: [Sphenix-tracking-l] TPC Seeding Efficiency Comparison

 

Hi all,

Finally changed over to logarithmic binning, and ran all-new reco jobs for both a current-build CA seeding reference and a TPC-truth-seeding reference within the last few hours. Here is the resulting efficiency comparison:

There are two strange things I see here:

1) The drop at low pT occurs even with truth TPC seeding; and

2) CA seeding appears to somehow outperform truth TPC seeding at high pT.

Are either of these things understood? 

For details, the black points were produced with the default G4_Tracking.C and the red were produced using use_truth_tpc_seeding = true and use_truth_vertexing = true. (The truth vertexing was a holdover from an earlier test that I forgot to change; I don't think it would affect the efficiency much at all, but I'll run jobs tomorrow to verify.) Reco tracks were selected using nhits>40&&quality<10&&nmaps>0.

Thanks,

Michael Peters