5D Calorimetry

Friday 15 Sept 2023, 09:30 → 11:30 America/Los_Angeles

Zoom:

https://stanford.zoom.us/j/92728532486pwd=eHBpcDQ0RjdnZlBjQmRHYW1WRjdFUT09

Password: 059507
 

09:30 → 09:50 t0 reconstruction

• t0 resolution and efficiency
• Applications:
  o t0 for HGTD
  o PU suppression
  o LLP search (displaced photons/jets)

09:50 → 10:10 single pion studies

• time evolution/structure of pion showers
• response vs time
• neutron component identification

Speakers: Doyeong Kim, Zahra Farazpay (Loisiana Tech)

HFSF2023 FollowUp - 10 (2).pdf

Doyeong

Summary of presented slides

1) slides 2~5

A bug fix in the averaged cell time plots.

2) slides 6~18

Cell time with the leading cell

Due to the significantly poor timing resolution below 10 GeV cells (slide 11), using only the leading cell in each pion results in considerably better cell time resolution. Also, double peak effects are observed due to the two different cell time resolution regions, depending on the cell energy (slide 17).

3) slides 19~24

Investigating the source of the non-Gaussian tail in cell time distributions.
The center (Gaussian) and tails (non-Gaussian) of cell time distributions were draw independently. It was found that each LAr EM barrel layer has a distinct contribution to the cell time (slide 21).

4) slides 25~27

Cell time is measured in bins of pion energy using the 𝜋- sample. After fixing the mentioned bug, the cell time resolution improves as the pion energy increases.

 

Discussions  & Action items

Checking E resolution for validation - unfinished from last week

slides 16&17

• When I averaged more than one cell, the time distribution of low energy cells became shifted. (Compare the left plot on slide 16, which has a mean of 0.09 ns, with the left plot on slide 17, which has a mean of 0.46 ns.)
• The plots on the right of each slide have comparable RMS, indicating that there is no gain by using more cells. It is likely that applying a reco cell E > 10GeV cut could already select one cell most of the time. (This needs to be confirmed.)

Comprehensive 

• We couldn’t see the enhancement of timing resolution when we plotted averaged cell time distributions. Why?
  • To avoid assuming that the errors are identical, it is necessary to use a weighted average.
  • The best optimal way to do this average is
    t_mean = (Sum(cell weight* cell time)) / Sum(cell weight)
    Where cell weight = 1/RMS^2
  • For this, we would like to have a RMS map.
    Plots, RMS as a function of pion E (overlay all curves for each layer)
  • Also, Mean as a function of pion E (overlay all curves for each layer)
    Here, I will see the shift in the low energy pion bins. This can ultimately be used in the following calculation to remove the double peak, which was seen on slide 17 making everything center at 0.
    t_mean = (Sum(cell weight* (cell time - cell time mean))) / Sum(cell weight)  -> this is tentative action item
  • Another easier but less accurate trial:
    t_mean = (Sum(cell energy * cell time)) / Sum(cell energy)
    Assumption: cell energy is proportional to the error. Therefore, energy is capturing their correlation but missing the granularity in layers.
     

 

 

10:10 → 10:30 Space-time structure of hadronic showers

• Response vs time
• overlapping showers
• time-aware GNN PFlow reconstruction

10:30 → 10:50 Higgs Factory calorimeter timing studies