SoLID Ecal Weekly 20230608

Update from Tuesday Analysis/Sim Meeting

• Mike on tracking:
  • looks like back/forward tracking can be added to level0 tree but then it needs to be passed onto level 1

• Spencer update on TS3 study, suggestion:
  • for data: use individual shower calibrated by MIP, not showersum directly
  • for data, run 4680 had SC-A.and.SC-D and SC-A did not work well, should try the newer run
  • for sim, separate the 4 particle types and weigh each by the cross-section.

• Darren/Zhiwen is working with Kishan on AI/ML
  • How does AI/ML PID compare with classical PID for beam test sim?
  • How well does the trained ML PID work for beam test data?
• High-Level Question to be answered by our study:
  • Is AI/ML PID complimentary to classical method (thus assist classical method), or superior?

• Update on prescale in data:
  • (Jixie, W. Gu): CODA3.0 uses power(2,n-1)+1, so n=6 means 33 and for every 33 events, 1 is recorded.

Discussion on Cooking with Updated Tracking

• Presentation by Mike: Tracking_With_Target_Cut
  • track back projected to beamline (figure shows in direction perpendicular to 18-deg line, i.e. something like "ytar") shows wide distribution and an offset of target center that is larger than possible in reality (even accounting for pointing of our setup).
  • Suspect this is due to GEM chamber position not perfectly known. Note that optimization won't give us the correct alignment, as Xinzhan has shown. Could do an estimate of how much offset in ytarg a 1cm-misalignment of one of the GEM chambers would cause.
  • If we include Jimmy's measurement, things could improve.
  • Once Xinzhan comes back, focus will be to correct efficiency. Note that the current thinking is that there is no way we can get the correct y (or z)-targ without survey of GEMs anyways. However, forward-projection to detectors should still be okay and good enough for PID study.

Discussion on Cooking with Edge-Finding

• Ye presented the principle of edge-finding, presentation: Ye's slides
• Carter is working on setting up his code on ifarm.

Update on simulation (Ye)

• with the PS understood, we now have very good agreement in ECal ShowerSum spectrum between different runs (different PS, trigger setup, and threshold). All should be within 15% of the simulation
• the end point of electron sim has a sharp drop off but the data show slower drop off. Not sure what is causing the difference, anything missing in the simulation? The comparison used Run 4780. Points to check:
  • How much is pileup?
  • simulation threshold should account for Shower block normalization (different scaling factor)
  • Is the shashlyk length accurate?
  • The MIP calibration is accurate to what level?
  • Peak-finding for individual shower block may not result in a soft-ware sum the same as the peak-finding of the hardware sum directly.

Update from Darren + Spencer

• Spencer studied TS3 from SC-C.and.SC-D.and.ShowerSum runs -- SC-C and SC-D show low-integral events below threshold and need some study. ShowerSum spectrum looks okay (with a low threshold).
• Darren showed AI/ML on ECal and Cherenkov: SULI_Week1_Slides
  • For Cherenkov we need to add position cut
  • For detector signal of events identified as electron, or pion, etc.. need to plot all elements of the confusion matrix individually (rather than plotting all "mis-identified" particles together).
  • One potential "deliverable" is comparison between classical and AI/ML methods for beam test simulation and also data.