1. Point Source Search with 2007 & 2008 data
Production
Limit setting methods
Questions

2. Few errata to the note
keeping track of them on

3. Production Available . data 2007 & 2008, all basic runs MC
mupage, corsika, neutrinos
with and without smearing
run-dependence included (see note)
Scripts to run and submit jobs
(maybe they are useful to someone)
aafit package now has example directory showing how to
read files
convert files to calreal-style (used for Antdst production).
analysis elog #473

4. Limit setting methods Limits for background-only experiments
Often very background-like likelihood ratio
Can give weird limits, when not careful
was major issue in recent 5-line analyses
We investigated the behaviour of 3 limit setting methods.

5. Backup: Integrity check

6. Limit setting methods 'normal' / Neyman limits lowest values
but often exclude zero!
Feldman Cousins
all limits >0
~20 higher limits than Neyman
CLs method
higher limits, of same size as counting experiment.
only get higher limit if the data really look more like signal.
Slight random variations in background-only experiments yield very different limits....even if
none of them resemble signal.
These methods exploit the randomness in the test-
statistic to get exact coverage.
We choose Feldman-Cousins
limits are physical
widely used in neutrino astronomy

7. Questions We answer all questions on the webpage
See also there for the original question (here I will briefly summarize them)
Have to make a selection here. Take the ones we think are most
novel / informative / controversial.
Thanks to everyone asking questions!!

8. Origin of the 2ns additional resolution
Several suggestions received about possible underlying effects (Jurgen, Juande, Maarten)
We will look into them, but:
The approach we take to the systematic is precisely to cover a wide range of possible effects and to constrain them by data.
We have to pause at some point (i.e. now) the hunt for effects and do
an analysis with our current knowledge of the detector.
So we do not plan to test every possible theory on the source of
some additional resolution at this point.
But: We will look into the questions asked so far.
nb: our study into the possible systematic on the angular resolution is not
to be understood as a measurement of the timing resolution.
The goal is to constrain the angular resolution by means of emulating a broad array of possible effects. We do this by smearing the hit times.
But reality is probably more complicated than that (alignment, angular acceptance, line offsets, etc, etc)
the 2ns sample is only a (conservative) 'best default guess' within the broad range of systematically allowed values.

9. T3 Trigger (do we want to discard it?)
(F Schussler)
5% of data
11 events where we expect 3.5
statistics? we think so.
good agreement where statistical issues
can be ruled out and where we cut (-5.4)
nb: this plot is obtained by vetoing the 3N
which is 95% of the events.
tiny discrepancy in 3N simulation can cause huge deviation here
it is amazing the plot is so good.
in any case: even a large error in 5% of data would be easily covered by our systematic
we want to keep T3-only triggered events.

10. Breakdown of the factor 15
(J Brunner)
Our sensitivity is a factor 15 better than the previous
analyses...Which factors make up this factor 15?
Exposure: factor ~3.2 (2.1 in live time, but 2x acceptance in 2008)
Nsig corresponding to limit: factor (5.2 events in 5-line analysis, 2.4 here)
Acceptance/Aeff factor (in 5 line data alone) -> surprise
The quality cut used in the previous 5-line analysis was a factor ~2 less
efficient than ours on E-2 neutrinos

11. Angular resolution in 2007/2008
and how it compares with what we said before
(V. Bertin)
peaks at 0.45 and 0.66 degrees
-> not going to split up in likelihood (proper event-by-event resolution fore-
seen in the future)
2008 MC
(fig 22 in note shows
this for )
with no smearing: 0.35 degrees
(compare to 0.24 in my thesis... differences:
12 full lines, low bg, low thresholds...
we have a loose lambda cut
effect of 2ns smearing factor 1.4 resolution degradation.

12. Point spread function
s. Cecchini & JP Gomes
Linear scale:

13. Point spread function
s. Cecchini & JP Gomes

14. Point spread function Yes! Is it spherical?
s. Cecchini & JP Gomes
Is it spherical?
important, because we assume sphericity in the PE generation
Yes!

15. Bias in acceptance vs ra?
s. Cecchini
showing data with loosened (-5.8) L cut
uniform as far as we can tell

16. Data/mc agreement in 3/10 pe and Ag/Non Ag
Juan Pablo
3 pe
10 pe
Not Ag
= 6% of live-time
data/mc about equaly good in all of them

17. comparison of data and smeared MC samples
(JJ Hernandez)
Q: is there special class of events there?
(see next slide)
A: High likelihood events are caused by low-multiplicity bundles
no further strong correlations found
Statistics in corsika MC is not great here
There is no guarantee that such a simple
smearing can give perfect shape agreement
we still think 2ns is a good 'best guess'
happy to try 2.1 / 2.2, etc.
but this takes time and is not crucial for our point source search, since all reasonably allowed values are in our systematically allowed range.
flat = nice
but what about this decline here?

18. which muons
give high L?
mupage MC

19. data/mc for smeared 2007 MC (JJ Hernandez)
2 ns seems to be a bit too much. What does the
muon plot looks like?
table with neutrino scale factors
for 2007 data.
2 ns smearing
The muon plots tells the same
story. 2ns seems a bit too much here.
It is still better than 1 ns (next slide),
and therefore our best guess for the central value.
again: the systematics range we assign covers all this.

20. data/mc for smeared 2007 MC
(JJ Hernandez)
1 ns smearing

21. Conclusions Many (74) questions received already Thanks
Please send more before next meeting: Aug 15.
No major issues emerged so far (in our opinion)
Still hope to unblind soon