Unfolding atmospheric neutrino spectrum with IC9 data (second update)

1
Unfolding atmospheric neutrino spectrum with IC9
data (second update)
4 / 4 / 2007
2
Upgrades

• Use of the energy estimate of Rime
• Smearing matrix build with a distribution closer
  to what we expect to measure

3
Reminder How to choose the best variable?

• Criteria
• As much linear as possible with energy
• With the least spread
• Possible candidates nchan, tot_charge, energy
  estimator

4
log10 of Total charge
delta
log10 Charge (pe)
log10 Charge (pe)
log10 Enu GeV
log10 Enu GeV
RMS 0.42
RMS is related with the quality of the fit
deviation from fit / delta
Y-projection
log10 Enu GeV
5
log10 of estimator
log10 estimator
log10 estimator
log10 Nchan
log10 Enu GeV
log10 Enu GeV
log10 Nchan
deviation from fit / delta
log10 Nchan
log10 Enu GeV
6
MC samples

• Neutrino events 50 x 106 (?-2)
• (dataset 437)
• CORSIKA 1000 x 106
• (dataset 296)
• Coincident muons 3000 x 106
• (dataset 394)

(files had to be reprocess to include the energy
estimator)
7
Spectral index for simulation

• It is convenient to generate the MC sample in
  such a way that generated/expected is flat (more
  efficient smearing matrix calculation, smoother
  distribution to unfold)

dataset 437 (?-2)
dataset 390 (?-1)
With ?-2, we avoid to generate too many useless
high energy events
8
Generated / expected
dataset 437 (?-2)
dataset 390 (?-1)
Flatter ratio between generated and
expected Still, statistics could be not enough
(factor 10 more)
9
Processing

• First guess with line fit
• The result is used to feed the LLH
• Quality cuts are applied after processing

10
Ndir ? 8, Ldir gt 200 m, ? gt 92?
atmospheric neutrinos 671 corsika events
13 coincident muons 2.6
purity 98
11
Unfolding initial distribution

• In unfolding problems, it is usually very useful
  to have an idea of the distribution we want to
  unfold.
• BUT the robustness of our method against our
  choice of the initial distribution has to be
  checked.

12
Smearing matrix
Maybe more statistics are still needed
13
Measured distribution
(background from single and double atmospheric
muons already included)
14
Unfolded spectrum
Using Singular Value Decomposition
15
Deviation from the true spectrum
16
New initial spectrum

• We have to make things harder to the algorithm,
  to check its robustness

17
The unfolded spectrum follows well the true
spectrum A more systematic way of robustness
check is underway
18
Difference between true and unfolded
19
Summary and Prospects

• Energy estimate shows better behavior as energy
  correlated variable
• The new MC (gamma-2) offers flatter generate /
  expected distribution
• Good quality of unfolding even when the initial
  spectrum is different from the true one
• To do list
• More statistics for smearing matrix
• Combine different variables (?)
• Continue on optimization of cuts
• Tuning of regularization constant and other
  internal parameters