1
Comparisons of data/mc using down-going muons
Jon Dumm, Chad Finley, Teresa Montaruli
UW-MadisonApril 26, 2007
2
Outline
• Low level– Time differences between NN DOMs,
Occupancies, Nchan, Nstring
• Reconstruction level– Zenith, Azimuth, residuals, Quality cuts (Ndir,
Ldir, paraboloid sigma)
3
Data sets
• Down-going muons (~0.5M) with high quality reconstructions– 34-fold muon-llh, gulliver, paraboloid
• Data for comparison (~ 1 hr livetime)– IC9 minbias data from June 2006– Corsika Simulation V01-09-06, datasets 296, 394
• Both single and coincident muons
• Cuts – Trigger level (Cleaned)– Hard Cuts
• Sigma<3 deg, Ndir>9• ~11% signal efficiency (E^-2), ~95% background rejection
4
Time difference between NN DOMs
Known problem: wrong LC time window!
Exp LC window
Sim LC window
(ns)
5
Zoomed in on time difference
This plot tells us about hole ice properties – local scattering near DOMs.
(ns)
6
Occupancy – trigger
Structure washed out in sim
Ratio of Exp / Sim
Not enough light near bottom of IC in sim
Format for remainder of talk:
-Data vs corsika+doublemu = total sim
-No normalization. Real rates as given.
Frequency each DomID is hit
0 ~ 1450m 60 ~ 2450m
7
Nchan – trigger
Cut at Nchan<46 for blindness
Number of DOMs hit in an event
Difference gets worse at higher Nchan
8
Nchan – hard cutsHard cuts = Sigma<3, Ndir>9
Even with cuts, the difference at high Nchan does not quite go away
Number of DOMs hit in an event
9
Nstring – trigger Number of strings hit in an event
Similar to the difference at high Nchan but worse!
10
Nstring – hard cutsHard cuts = Sigma<3, Ndir>9
Number of strings hit in an event
11
Zenith - trigger
The rates of mis-reconstructed events are underestimated by simulation
Reconstructed zenith given by paraboloid
Ideally, we need to find a way to oversample these fakes to save CPU time
12
Zenith – hard cutsHard cuts = Sigma<3, Ndir>9
In order to test background rejection, may need weighted corsika sample near horizon
Reconstructed zenith given by paraboloid
13
Azimuth - triggerReconstructed azimuth given by paraboloid
Structure is from having only 9 strings
14
Azimuth – hard cutsHard cuts = Sigma<3, Ndir>9
Reconstructed azimuth given by paraboloid
15
Time ResidualTime Residual =
(Observed time – expected time)
given Cherenkov cone and track
Remember, simulation LC window at 500 ns instead of 1000ns
16
Time Residual at two depths
DomID 5 ~1600m DomID 45 ~ 2300m
Keep in mind, there is an LC time window problem after 500 ns
17
Ndir - triggerDirect hit time window:
-15 ns <residual time <+75 ns
1 hit per DOM (first hits)
Difficult to hope for agreement without agreement in Nchan
18
Ndir – hard cutsHard cuts = Sigma<3, Ndir>9
Direct hit time window:
-15 ns <residual time <+75 ns
1 hit per DOM (first hits)
19
Ldir - triggerLength of direct hits along track
μ
Ldir
Good agreement, but not ideal for IC9 since the detector is asymmetric
Direct hit
20
Ldir – hard cutsHard cuts = Sigma<3, Ndir>9
Length of direct hits along track
μ
Ldir
21
Paraboloid Sigma - triggerParaboloid samples the likelihood space around the track and fits it to a paraboloid. Sigma is the circularized width of this paraboloid.
θ,φ
L Sigma
There have since been further improvements in paraboloid for higher efficiency
22
Paraboloid Sigma – hard cuts
Hard cuts = Sigma<3, Ndir>9
θ,φ
L Sigma
Paraboloid samples the likelihood space around the track and fits it to a paraboloid. Sigma is the circularized width of this paraboloid.
23
The End
• We have some confidence in our quality cuts for IC9 analysis
• Fix LC bug and reprocess
• We need to standardize these comparisons for all to see– avoid making it too long and painful to be
useful