1. Analysis of the MAGIC data
1 - Description of the images by the
Hillas parameters
2 - What do we expect to see
3 - g/h separation and signal
extraction by using the Supercuts
ON-OFF classes
4 - Suggestions to the analyzers

2. 1 - Description of the images by the Hillas parameters
Parameters first proposed by M. Hillas in 1985
(Proceedings 19th ICRC, Vol 3, pag )
Hillas params lead to the 1st significant (9 sigmas) detection with an IACT (Whipple); the Crab nebula (Weekes et al, Astrophysical Journal, Vol 342, pag , 1989)
Also 1st significant detection (6 sigmas) of an extragalactic source: Mkn 421 (Punch et al, Nature, Vol 358, pag , 1992)
More about Hillas param can be
found at TDAS 02-03

3. Calculation of the Hillas parameters

4. 1 - Description of the images by the Hillas parameters
00 ≤ ALPHA ≤ 900
More about Hillas param can be
found at TDAS 02-03

5. 2 - What do we expect to see
We can obtain this info by playing with the Monte Carlo simulations
I used the “standard MAGIC MC”
- Same conversion efficiency for inner/outer pixels
- Optical point spread function is produced
by a 2D Gaussian with 0.7 cm sigma
- Ped RMS 7 photons
- etc, etc…

6. 2 - What do we expect to see
2.1 - Distribution of the basic Hillas parameters
Size > 2000 photons

7. 2 - What do we expect to see
2.1 - Distribution of the basic Hillas parameters
Size > 1000 photons

8. 2 - What do we expect to see
2.1 - Distribution of the basic Hillas parameters
No SIZE cut

9. 2 - What do we expect to see
2.2 - Relation SIZE-ENERGY
Clear correlation between SIZE parameter and E
Gammas
Protons
Helium nuclei
At a given E, gamma images have (in average) a larger light content
As E decreases the probability for Proton and He showers to trigger also decreases; see change of slopes in profile histograms

10. The quality of the cuts is usually quantified (in MC) by using the Q factor
This is a natural definition; directly related
to the significance of the detected signal
E > 150 GeV
Q  8
E > 30 GeV
Q  3
(values from TDAS 03-01)
However, sometimes people (also) use it to quantify the hadron rejection of the telescope, and this is not FEAR;
the hadron rejection due to the trigger and image cleaning is NOT included.

11. 2 - What do we expect to see
2.2 - Relation SIZE-ENERGY
Clear correlation between SIZE parameter and E
Gammas
Protons
Helium nuclei
At a given E, gamma images have (in average) a larger light content in the camera
As E decreases the probability for Proton and He showers to trigger also decreases; see change of slopes in profile histograms
At very low E (< 300 GeV)
contribution of isolated muons become important

12. 2 - What do we expect to see
2.2 - Relation SIZE-ENERGY
Fraction of Cherenkov light produced by muons
(in proton showers that trigger the telescope) increases as the E of the primary proton decreases

13. 2 - What do we expect to see
2.2 - Relation SIZE-ENERGY
Gamma images
Proton images with
90% light from muons;
ISOLATED MUONS
Rest of proton images
Mean SIZE of images from isolated muons is almost constant at 1000 photons
Thing to be considered when decreasing Eth of the analysis

14. 2 - What do we expect to see
2.2 - Dependence on the shower characteristics
Gammas
Protons
Helium nuclei
Strong dependence of the parameters on the energy of the primary
Soft dependence on the impact parameter, except for gammas

15. 2 - What do we expect to see
2.3 - Dependence on the SIZE and DIST
Gammas
Protons
Helium nuclei
Strong dependence of the parameters on the SIZE
Soft dependence on the DIST, except at DIST > 0.90
Consequence of the limited trigger region of the camera of MAGIC. At large DIST values (>10), only EXTENDED images can trigger the telescope; thus sample is biased to extended images

16. 2 - What do we expect to see
2.3 - Dependence on the SIZE and DIST
SIZE > 2000 photons
Bias to extended images produced by the limited trigger region is washed out when dealing ONLY with large images (SIZE > 2000 ph).
Thing to be considered when decreasing Eth of the analysis

17. DIST parameter of the showers (Standard MC)
After Cuts
( = 57% ;  = 7%)
Before cuts
(Filter cuts with Size > 2000 photons)

18. Impact parameter of the showers (Standard MC)
After Cuts
( = 57% ;  = 7%)
Before cuts
(Filter cuts with Size > 2000 photons)
Random Forest also keeps about 50% of gammas
with impact > 125 m

19. Impact and DIST parameter After Dyn cuts with fixed Dist cut (0. 60-1
Impact After Cuts
( = 50% ;  = 6%)
DIST After Cuts
( = 50% ;  = 6%)
Events with large impact parameters CANNOT be removed simply by cutting in DIST

20. Explanation for the difference with respect to “old” Cherenkov telescopes
Coulomb scattering and Cherenkov angle
200m
300m
400m

21. Large collecting mirror High sensitivity and FINE PIXELIZED camera
Explanation for the difference with respect to “old” Cherenkov telescopes
80m
160m
Large collecting mirror
High sensitivity and
FINE PIXELIZED camera
Gamma/hadron separation possible even at large (>150m) impact parameters

22. Consequences of this effect
Energy resolution will worsen (somewhat) due to
the larger fluctuations in the light yield
Head-tail asymetry will move towards tail for
hadron showers
We see that in MC and experimental data
Additional contribution might come from
aberration of the images as one moves away from
camera center
Reduction in the effectiveness of the
LENGTH/WIDTH cut

23. A personal remark concerning this effect
I was strongly criticized in Tuorla (June 2004)
when presenting this effect.
I DID NOT create these guys.
Do not blame me for them !!!!!
Gammas with large impact parameters are included in all the MAGIC sensitivity and E resolution calculations since 1.5 years (new MC)
I just realized they were there…
They might not be that bad if we can handle the situation… they increase our collection area by 2

24. 2 - What do we expect to see
2.4 - Dependence of ALPHA on the SIZE CUT
ALPHA reconstruction worsens as E decreases
<E>  50 GeV
<E>  100 GeV
<E>  180 GeV
<E>  350 GeV
At SIZE > 2000 photons: fraction of gamma events in ALPHA range is 7.5% of the events contained < 150

25. 3 - g/h separation and signal extraction by
using the Supercuts ON-OFF classes
a) Definition of set of INITIAL SUPERCUTS on a chosen set of parameters
Significance, Nex, ratio of efficiencies, signal/background or a function F(Significance, Nex, signal/background,)…
e) Variation of the SUPERCUTS
3.1 - Basic algorithm
b) Application of the set of cuts to the data
c) Extraction of the gamma signal
d) Estimation of the “quality” (Qestimator) of the signal
f) Continuation with step b) as long as Qestimator is not optimum

26. 3.2- Implementation in the MARS environment
Introduction
Done by Wolfgang almost a year ago (only ON data used). Structure follows steps of the analysis of Daniel Kranich
New classes done by myself to use ON-OFF data
(finished at mid February)
Basic tools (polynomial fits, Minuit interface for the
minimization… ) are the same, and were/are working
reliably.
A small set of new functionalities were added
Possibility to optimize in different bins in zenith angle,
and combine (a posteriori) the results
Storage of all info (alpha plots, normalization factors,
shower parameters, cuts… ) in a root file
Usage or static/dynamical cuts (use/not use theta, dist…)

27. 3.2- Implementation in the MARS environment
Working principle
Within class MFindSupercutsONOFF
1) INITIAL SET of CUTS are given (by user) and stored in container MSupercuts
Cuts are computed (in case of dynamical cuts) and applied to data through class MSupercutsCalc
3) Signal extraction and significance calculation using class MHFindSignificanceONOFF
Non is obtained by counting, Noff by fitting with a second order palimony
ON-OFF normalization computed after cuts
in background region (defined by user). But
possibility to normalize ON-OFF before cuts
(Data padding needed)
Significance calculation: LiMa (1984) formula 17

28. 3.2- Implementation in the MARS environment
Working principle
4) Parameters are modified to maximize significance using the class MMinuitInterface
Final (optimized) parameters are written into container MSupercuts, and then applied to the data (MFindSupercutsONOFF) to produce (MHFindSignificanceONOFF) the alpha plots and compute final significance and Nex
6) Alpha Plots stored in postscript files and inside a root file together with other information (normalization factors, significance, Nex, hillas parameters, cuts…) FOR A LATER STUDY
Class MFindSupeructsONOFFThetaloop used to run
MFindSupercutsONOFF separately in different ZENITH angle
bins (eventually also SIZE bins), and combining results (if desired)
at the end.
Besides, this is also the class defining names of root data files,
histograms were to store info (Signfinicance, Nex…) and saving
all those guys into single root file

29. 3.2- Implementation in the MARS environment
How to use these classes
Only 3 steps required
1) Download current Mars version from the CVS
2) Add the line “mtemp/mmpi/SupercutsONOFFClasses” in the SUBDIR field of the general Makefile, and compile it
3) Define some variables in a silly macro specifying how you want to run the analysis

30. 4 - Suggestions to the analyzers
4.1 - Data Reconstruction
Spend time INSPECTING calibration and pedestal runs.
If they are bad you will spoil the data (which might be good)
CheckConversionFactorsEvolution2.C
TestCalibrationEvolutionForPixelSoftId.C
TestCalibrationParameterForAllPixels.C
Use always the closest (possible) calibration and (specially) pedestal run to calibrate and clean the data
Inspect images in the camera display using different image cleanings. It is also interesting to apply some
“simple” cuts to select the fraction of the data to be displayed
SeeMagicCalibratedEventsBeforeAfterCuts2.C
After computing the image parameters, inspect their distributions AND compare these distributions for ON-OFF
MacroToPlotHillasBeforeAfterCuts.C

32. 4 - Suggestions to the analyzers
4.2 - Supercuts optimization
Compare ALWAYS the results obtained for the TRAIN and the TEST sample. They MUST be statistically compatible
TRAIN
TEST

33. 4 - Suggestions to the analyzers
4.2 - Supercuts optimization
Compare ALWAYS the results obtained for the TRAIN and the TEST sample. They MUST be statistically compatible
Default when optimizing cuts with these classes
Try ALWAYS to understand the SUPERCUTS applied from a physics point of view.
SCDynamicalSupercutsApplied.C

35. 4 - Suggestions to the analyzers
4.2 - Supercuts optimization
Compare ALWAYS the results obtained for the TRAIN and the TEST sample. They MUST be statistically compatible
Default when optimizing cuts with these classes
Try ALWAYS to understand the SUPERCUTS applied from a physics point of view.
SCDynamicalSupercutsApplied.C
Compare with Monte Carlo simulations.
a) Compare parameter distributions before/after cuts
b) efficiency of the cuts, Q factor calculation,
collection area calculation
ComputeCollectionArea.C

36. 4 - Suggestions to the analyzers
4.3 - Some other suggestions
Transfer knowledge to the collaboration and (specially) to the other members of the group
I feel partially responsible for a non (totally) successful
transfer of the know how… sorry… a bit stressed in the last stage of my thesis…
Be very CRITIC; specially SELF-CRITIC
If you find some UNEXPECTED behaviour (calibration, Hillas distribution, cuts applied…), STOP and SPEND time trying to UNDERSTAND it.
Most probably a problem in the analysis, but may be a new feature/discovery. In BOTH cases you gain
It always bring good results in a medium-long time

37. 4.3.2 - Be very CRITIC; specially SELF-CRITIC
From my personal experience
Analysis
Noticed problems in pedestals, calibration, image cleaning procedures in both software and quality of data at the source hunting period (end February beginning March)
On Sat Feb 28, :01:46 I sent an (to the muc-group) announcing the detection (7-10 sigmas) of the Crab (Jan 27). The day after I noticed that the data was biased by a low cleaning level, which allowed some noisy pixels to show up in many images biasing the alpha values of the ON data towards low values. Is this effect familiar to you ?????
I rectified the next day.
- In the next days we got the 5.5 sigmas detection of
Crab January 27th, first gamma source detection with MAGIC

38. 4.3.2 - Be very CRITIC; specially SELF-CRITIC
From my personal experience
Analysis
Discrepancies with other groups in several issues, which at end it was shown that the analysis performed at Munich was right:
calibration with MExtSignal1 and MExtSignal2,
light curve of Mkn 421 at February
Big Barcelona flare (6 Hz) of Mkn421 in April (Daniel, Hendrik) ….
Noticing about some new/unexpected features in MC data:
50% of triggered proton showers below 100 GeV are muons
Bias to extended images produced by limited trigger region
Large impact parameters of half of the accepted gammas

39. 4.3.2 - Be very CRITIC; specially SELF-CRITIC
From my personal experience
Hardware
Discovery of the MAGIC coating. It was the result of a very
small (2-3%) NON-UNDERSTOOD increase in QE at visible wavelengths when working with a WLS

40. Increase in the U.V.

41. Sensitivity increased by a scattering surface
attached to the PhC

42. 4.3.2 - Be very CRITIC; specially SELF-CRITIC
From my personal experience
Hardware
Discovery of the MAGIC coating. It was the result of a very
small (2-3%) NON-UNDERSTOOD increase in QE at visible wavelengths when working with a WLS
Low collection efficiency of the PMTs in the PhC periphery
Discovery of the remaining of the Sb pill inside the ET9116 PMTs. This Sb grains could produce a short in the PhC-D1, preventing the PMT to work.
Finding correlation between VCSEL noise and bias current, as well as identifying the fluctuations in the light produced by the bias current itself as the leading term in the VCSEL noise.
Finally VCSEL drivers could be used…

43. 4.3.2 - Be very CRITIC; specially SELF-CRITIC
From the history of physics.
Many Nobel Prizes are the result of the careful
study of UNEXPECTED effects
Radiactivity discovery Becquerel 1897
Pulsar discovery in 1967
Binary pulsar discovery
CMB discovery