Post on 01-Apr-2015
transcript
Data/Monte Carlo disagreement for Rustem’s Signal Fluctuation
variable in the Far Detector(Working Title)
Philip RodriguesOxford MINOS Group Meeting
15 Jan 08
2
Low/High variable
• Rationale: muon tracks small variation between planes, non-muon tracks larger variation
• Construction:1. Exclude first 30% of track planes (to veto shower)2. Find window around reco’d track: ±4 strips, ±40ns3. Take all strips in this window (track and non-track)4. Sort these strips by PH5. Find mean PH of lower half, mean PH of upper half6. Low/High = mean of lower half / mean of upper half
3
ND Recap
• Problems in ND investigated quite thoroughly:– Uncalibrated spectrometer– Low PH afterpulsing
• Use only calorimeter strips > 100 sigcor solves problem:
From docdb 4025 (Rustem)
4
Far Detector
• Very similar problem appears in FD: more low PH hits in data• Ad hoc 175 sigcor (~2.5 pe) cut helps quite a lot:
• But why? We think we understand the FD• It pays to understand this:
– concerns that it may affect other aspects of the analysis (docdb 4024, slide 3)
5
But why?
• Why has this never come up before?– The low/high variable is doubly sensitive to data/MC disagreements– Non reco’d strips (not track or shower) are used in its construction– Taking low/high ratio amplifies data/MC disagreement
• So, unlikely to affect anything else
• Candidates:– Noise: seems unlikely. My noise studies show good data/MC
agreement– Crosstalk: seems unlikely. Lots of test stand studies– Afterpulsing: seems unlikely. FD timing very different to ND– Real Physics?: seems unlikely. We know what tracks do
6
Noise?
• No
Dogwood noise No noise Double noise
7
Noise (default)
• No
8
More Less ad hoc cuts
• Change strip and time windows• 1 strip window helps – something happening around
the track
Tim
e window
meaningless in
FD
?
9
Transverse variable
• ~ (PH in track) / (PH around track + PH in track)
• Timing cut makes “halo” narrower:– Early/late hits are mostly around track
10
Mini-conclusion
• Should only use track hits for this variable (no window):– The physics rationale depends only on the track itself (cf
transverse variable)– Track modelling is better than detector modelling(?)
• I plan to recommend this to Rustem/the CC group
11
Crosstalk
• Reco FD MC with optical and charge xtalk +10%
Default xtalk +10%
• Not quite perfect, but major improvement• Plots with +20% and +13% upcoming
12
Being sure
• How can we be certain it’s crosstalk?– xtalk-tagging code somewhere – knows about PMT pixels, etc– Plot strip distributions in time and space to look for physics– Learn about xtalk– Any other ideas?
13
Conclusions
• Starting to understand data/MC disagreement
• Crosstalk seems a likely candidate• Need to convince a skeptical public