1. Update on LHCb Level-1 trigger
4 February 2019
Federica Legger

2. Summary TDR L1 Post TDR needs L1 redesign L1 bandwidth division
L1 efficiencies for some representative channels
Long-standing L1 open questions
Status and plans
4 February 2019

3. TDR times
LHCb trigger TDR (September 2003)
Aim for highest possible efficiencies for a list of benchmark channels
(tagging not taken into consideration due to the lack of statistics…)
4 February 2019

4. TDR L1 code L1 Generic (B meson decay products)
Massive  high Pt
Long-lived  high IP
L1 Generic (B meson decay products)
Selects events with two high PT tracks in IP window [0.15, 3.] mm.;
Bonus system (specific)
Dimuon
|mmm – mJ/| < 500 MeV
mmm > mB – 500 MeV
Electron
Etmax > 3 GeV
Photon
Courtesy of Thomas Schietinger
4 February 2019

5. Advantages: Disadvantages:
Bonus values are tunable for optimal efficiencies/bandwidth;
Good for TDR!
Trigger summary in just one variable:
Convenient for comparisons
Easy for users
Disadvantages:
Correlation among various subtriggers;
Cut on PT depends on bonus values
Hard to understand why an event has triggered;
An event with some momentum, some dimuon mass, a little bit of photon and electron can trigger…(???)
Not easy to implement parallel subtriggers
example: Thomas’ single muon hack
4 February 2019

6. Post TDR Redesign of L1 code
Some bug corrections and code improvements
tracking and multiple PVs
Need to better understand L1 behaviour & systematics
Unbiased data samples
trigger on tag
Indipendent trigger line
Redesign of L1 code
4 February 2019

7. DC04 L1 code Parallel (and overlapping) subtriggers
GENERIC
SPECIFIC
Parallel (and overlapping) subtriggers
Generic
Single muon
Dimuon
Dimuon (J/Psi)
Electron
Photon
Final decision is OR of single subtrigger decisions
Courtesy of Thomas Schietinger
some PT cut still needed to reduce bandwidth
4 February 2019

8. Generic (pT): ln(PT1) + ln(PT2) > 13.925
For tracks with IP > 0.15 mm
Pile-up veto (IP < 0.15 mm), up to 2 PVs
Single-muon: IP > 0.15 mm, PT > 2.32 GeV
Pile-up veto (IP < 0.1 mm), up to 3 PVs
Dimuon general: mmm > 500 MeV, IPmm > 0.05
Dimuon J/: |mmm – mJ/| < 500 MeV OR
includes B, Z, H, X  mm mmm > mB – 500 MeV
No IP cut
Electron: max. ET(e) > 3.6 GeV AND
ln(PT1) + ln(PT2) > 13.0
Photon: max. ET(g) > 3.1 GeV AND
4 February 2019

9. Overlaps are absorbed in this direction
Bandwidth division
Massive improvements in generic algorithm (multiple PVs!) result in larger bandwidth for other triggers (in particular dimuons)
Before:
Bandwidth (kHz)
Photon
Electron
Dimuon, J/Psi
Dimuon, general
Single-muon
Adjusted for overlap
Generic
29.4 (74.1%)
6.8 (17.1%)
1.5 ( 3.8%)
1.8 ( 4.6%)
3.7 ( 9.4%)
4.3 (10.8%)
3.2 ( 8.0%)
1.2 ( 3.1%)
1.2 ( 2.9%)
2.2 ( 5.5%)
2.5 ( 6.4%)
Now:
Overlaps are absorbed in this direction
4 February 2019
Courtesy of Thomas Schietinger

10. L1 efficiencies
Offline selected
Reconstructible
4 February 2019

11. 4-prong give best generic efficiency!
10-20% improvement!!!
L1 efficiencies
Offline selected
Reconstructible
why different?
Hadrons
trigger
e and g!
4 February 2019
4-prong give best generic efficiency!

12. Implementation: DaVinci (LHCb analysis program) Trg/L1Decision
v12r2;
Trg/L1Decision
v3r1;
Activating L1 decision via option file generates
N-tuple with L1 information
Root macro coming with L1 package creates all kinds of plots to analyze L1 performances
4 February 2019

13. To check L1 performances…
summary
Min Bias
fit
Efficiency vs.
Retention
plots for each
subtrigger
4 February 2019

14. More complicated case…
Choose second
cut first:
Electron
Photon
4 February 2019

15. 4 February 2019

16. Advantages of new L1 code:
Easy to impose bandwidth division
user-steerable;
Single trigger bits accessible to users
Easy to understand why an event triggered;
Allows clean implementation of new trigger lines.
4 February 2019

17. L1 open questions for generic subtrigger:
Multiple PVs treatment
Is current solution the best one?
L1 Generic decision function
S (log PT)
log (S PT)
Weighted PT2?
PT3?
Is 400 MeV/C a good value for tracks with no measured PT?
4 February 2019

18. Control channels: Bd  p+ p- 2 high PT tracks
Bs  Ds K high PT track
Bd  D0(Kp)K* no high PT track
4 February 2019

19. Multiple PVs No big differences # PVs > 2
current solution seems to be the best
# PVs > 2
Not enough statistics at current luminosity
Higher number of PVS
Need some high luminosity data!!
NOW
Is PV1 = highest
multiplicity vertex
enough?
Veto on # PVs?
4 February 2019

20. TDR L1 Generic D Signal Min Bias
Log IPS1 + Log IPS2
Signal
Min Bias D
Log PT1 + Log PT2
L1 Variable = D (2d distance from L1 line)
4 February 2019

21. DC04 L1 Generic Improved vertexing Simpler cut: No need for IPS;
Log PT1 + Log PT2
4 February 2019

22. Do we need a weighted PT2? NOW NOW Log (PT1+PT2)
Log PT1 + Log PT2 = Log(PT1*PT2)
Do we need PT3?
Do we need a weighted PT2?
GENERIC
GENERIC
NOW
NOW
Weight
4 February 2019

23. Default PT
for tracks with no measured PT
NOW
4 February 2019

24. Current solution is OK, but do we really understand it?
Are default Pt = 400 MeV/c and L1 = S log disentangled?
PT1, PT2 dependence not so clear
Try two separate thresholds for PT1 and PT2 and play with it
4 February 2019

25. Status and plans L1 shows good (and unexpected) performances;
Still some open issues;
Start thinking about systematics (high luminosity, beam background) and online issues
4 February 2019