Hit density / pT threshold dependence

Masahiko Saito

1

Based on Qallse framework: https://github.com/derlin/hepqpr-qallse

Configuration

• Compared tracking time/performance between three models
• Mp, D0, Doublet-based
• Dataset
• No pT requirement
• Phi boundary = 0.15 ~ 1.05
• min. N of hits in tracks >= 5
• variable hit density
• Tracking
• Use the same datasets for each model
• tplet_max_curv : 8e-4, 1e-3, 2e-3, 3e-3, 5e-3 [1/mm]
• Corresponds: 120 MeV ~ 0.75 GeV
• 2-try for each point.

​

​

2

3

Hit density dependence

tplet_max_curv = 5e-3 (pT > 120 MeV)

​

4

5

6

7

8

9

10

11

pT threshold dependence

Hit density = 20 %

​

12

Main time consuming part is a creation of triplet, and the pT(curvature) selection is applied after a triplet creation. Therefore, the model building (triplet/quadruplet creation) time is roughly independent of pT.

13

Low efficiency is due to pT threshold

14

Many missings are due to pT threshold

15

Minimum energy - ideal energy

Need to improve the energy function of doublet-based model...

• Calculated using true doublets

• solution by neal

16

Effect of phi slice

in doublet-building

17

• Find hit-combination from limited phi cones
• Default : +- 2*pi / 53
• Low momentum track bends over the the finding region

18

• Dataset
• No pT requirement
• min. N of hits in tracks >= 5
• density = 5%
• Four experiments for each “N of phi slices”

Default

Small nPhiSlice is better, but it takes much longer time (next page).

19

20

Bonus

21

• X-axis
• Used in the TrackML score calculation
• This variable changes hit-weight

• Y-axis
• Used in the track reconstruction
• Given as max_curvature [mm].
• Translated to GeV assuming 2 Tesla

Recal (Efficiency)

• Dataset
• variable pT requirement
• min. N of hits in tracks >= 5
• density = 5%