0. Elisa Prandini Padova University MAPSES – Lecce 23 November 2011
It’s a kind of MAGIC! Come i raggi gamma ad altissima energia diventano segnale
PART 2
Elisa Prandini
Padova University
MAPSES – Lecce
23 November 2011

1. Summary Intro: VHE g-rays Astrophysics Observation Technique
Instruments and data taking
From raw data to shower images
Background Rejection
Signal extraction
Results:
Sky Maps
Integral and differential fluxes

2. Steps of the Analysis of CT data
Pixel signal extraction
Raw signal
Image cleaning
Cleaned signal
Image parameterization
Stereo- reconstruction
Event parameter reconst
Background rejection
Background estimation
Detection
Source detection
Spectrum
Sky maps
Spectrum / light curve
Spectrum Unfolding

3. Characterization of the events
Once we have obtained the shower images, the next step is to obtain the characteristics of the primary particle which originated the shower
Nature of the particle (Background rejection)
Primary Direction
Primary Energy

4. Random Forest
Reloaded

5. Monte Carlo Simulation
Random Forests
Based on
Monte Carlo Simulation
+ real data
How it works: The growing of a tree
elements:
- parameters (~10)
trees (100)
In each tree:
nodes
Possible parameters in each node (trials, ~3)  RANDOM
best separator and value per node (via Gini index)
final node size (1-5)
each event is classified!
training
this is a tree
Application to real data:
- We apply each tree ( from training) to them
- Each event is univocally classified (with a local hadronness)
 Final hadronness is an average among all the values!

6. Primary direction

7. Helpful to “detect the signal”
Primary direction
The ellipse major axis points to the center of the camera
With many telescopes:
The intersection of the axes is related to the incoming direction
Helpful to “detect the signal”

8. The incoming direction

9. Reconstruction of the incoming direction
Geometrical reconstruction: for more than 1 telescope
Efficient for
δ  30 deg
In Plan ┴ direction M1
Impact
parameter M2 Δδ
Reconstructed direction
Reconstructed core impact point M1 M2
Monte Carlo independent

10. Energy Reconstruction

11. Energy reconstruction
Basic fact: Energy ~ Image size
Methods:
A parameterization:
Energy = f(size, impact, zenith,…)
Look-up tables
Optimized decision trees
(Random Forest)
Based on
Monte Carlo Simulation

12. Energy reconstruction
Energy resolution:
20% at 100 GeV, down to 15% around 1 TeV
Energy Resolution:
E(est) - E(true) / E(true)
Energy Bias:
E(est) - E(true) / E(est)
Big low energies.
Solved with unfolding
Estimated with
Monte Carlo
(Gammas)

13. Signal Extraction

14. What is the aim of our analysis?
Detection of the Signal
Populate the VHE sky
Localize the emission
Characterize the emission
Energy
Time
Sky Map
Spectrum
Lightcurve

15. The analysis
Data acquisition
Calibration
Image cleaning and Hillas Parameters calculation
Hadronness and energy reconstruction
Former steps
File  Set of Events Event  Set of Parameters characterizing the shower Remember: event  is a shower (induced by a CR)

16. Characteristics of a g-like event
Therefore if we plot the parameter related to the reconstructed incoming direction, the gamma-like events will have it close to the telescopes pointing direction
To discriminate gamma-ray induced images from hadrons induced images, we use the square of the parameter q, the angle between the pointing direction (camera center) and the (reconstructed) incoming direction.
And what about the hadrons?
They are randomly distributed!

17. Detection: the Theta2 plot!
Where is the signal???

18. ?
We have a problem…
The number of background events is ~ 104 times the number of gamma Events
We have to reduce the number of bkg events:
HADRONNESS PARAMETER!
We apply a cut and reject the events that are likely hadrons
(MC data)

19. The Detection Source data “Signal” Background data
Now we are ready to perform our detection plot (also called theta-square plot)
VERITAS Collaboration, V. A. Acciari et al, ApJ 715 (2010) L49
Source data
“Signal”
Background data

20. The Background We need a background in order to estimate the signal
Old technique: observe the source (ON data) and a region of the sky with the same characteristics but without a source (OFF data)
PROBLEMS - time consuming!
- different observations conditions (weather, hardware)
SOLUTION: find an observing mode which allows to collect ON and OFF data simultaneously!
Wobble mode: the telescopes point to a region 0.4 deg offset from the source (and the background can be extracted)

21. Not always there is a detection of course…
The Detection
Some theta2 plots:
VERITAS Collaboration, V. A. Acciari et al, ApJ 715 (2010) L49
H.E.S.S. Collaboration, F. Aharonian et al. A&A 442 (2005)
Not always there is a detection of course…

22. The Significance
We need a tool to say if our observation is physically relevant
We use the significance of the signal
significance is a measure of the likelihood that pure background fluctuations have produced the observed excess (i.e., assuming no signal)
The common rule is that a source is detected if the significance of the signal exceeds 5 sigma!
Li, T., and Ma, Y. 1983, ApJ, 272, 317

23. Examples A five sigma signal A 10 sigma signal A four sigma signal
Mkn 180
MAGIC
Examples
A four sigma signal
A five sigma signal
A 10 sigma signal
PKS
MAGIC PG
H.E.S.S.

24. Standard Example: the Crab Nebula
MAGIC observations of the Crab Nebula:
E>300 GeV
60 GeV < E < 100 GeV
The energy range considered!
WHICH IS THE DIFFERENCE?

25. Images and Energy
At low energies the characterization of a gamma-like is much more difficult!!!

26. The analysis can continue
Therefore:
The data analysis of IACTs telescopes is non trivial…
The first result to look for is a signal (is there a gamma ray source or not?):
The tool is the theta2 plot, a plot of the parameter theta2, that discriminates between hadrons (our background) and gamma-rays induced showers using the incoming direction
The signal is quantified through its Significance:
< 5 sigma  no signal or more statistics needed
> 5 sigma  there is a signal!
The analysis can continue

27. Physical Results
Sky Maps

28. Second step: the localization
Important: in general IACTs don’t operate in scan mode but in pointing mode!
Moreover our resolution is… ~ 0.1 degrees
Extended sources: are galactic and very large regions
Extragalactic objects, for the moment, are point-like!
In order to localize the emission, we perform the so-called sky map, that is a bi-dimensional map of the reconstructed incoming directions of the primary gamma rays

29. The sky map For each event we reconstruct the incoming direction
Is the same parameter that we have used in the detection plot!
The background is modeled
Remember: our PSF is ~ 0.1 degree…
We use it:
As a check tool
In few cases: extended emission or multiple sources in the field

30. Example: the Crab Nebula
Find the difference!

31. Example: the Crab Nebula
The angular resolution is strictly related to the Energy range!

32. Extended sources
Some galactic sources are extended enough to be mapped nicely at TeV energies
SNR HESS J
Very interesting studies on acceleration sites!
H.E.S.S. collaboration,
A. Abramowski et al. A&A 531 (2011) A81

33. NGC 1275 region
Clear signal from the head tail RG IC 310 at E > 400 GeV
And below this energy?
E > 400 GeV
If we go down to 150 GeV, a signal from the RG NGC becomes significant!
E > 150 GeV

34. The galactic scan
HESS Coll. ICRC 2009 Proceedings

35. So, in the lucky case: Spectrum Light curve
We have detected a significant signal
We have checked that the emission is coming from the direction that we expect (if not, go back to point 1, changing the true source position)
Now?
Spectrum
Light curve

36. Physical Results
The Spectrum

37. The differential energy spectrum
The spectrum is essential in our study
Allows the characterization of the emission
A comparison is possible (also between different experiments and energy thresholds!)
What is it?
The number of VHE photons per area and time
VERITAS Coll. APJ Letters, 709, L163-L167 (2010)

38. Definitions Related concepts:
-ray flux: rate of -rays per unit area ( to their direction)
units: [L-2] [T-1] (e.g. cm-2 s-1)
Needed ingredients: a number of -rays, a collection area and an observation time
Related concepts:
Differential energy spectrum: flux per interval in -ray energy (cm-2 s-1 TeV-1)
Integral flux: integrated in a given energy range (cm-2 s-1), e.g. :
Light curve: time evolution of integral flux:  vs. t
Courtesy of A. Moralejo

39. Differential energy spectrum: the observables
Number of detected -rays: obtained from the observed excess
Effective collection area
Estimated energy of the events
Effective observation time: not equal to the elapsed time!

40. Number of g-rays per energy
We have a “discretization”  dN/dE becomes the number of excess per energy interval
How do we estimate this quantity?
Through the theta2 plot (per energy interval)!

41. Effective observation time
The effective observation time is not equal to the elapsed time between the beginning and end of the observations:
there may be gaps in the data taking (e.g. between runs)
there is a dead time after the recording of each event
Useful quantity: t, the time difference between the arrival time of an event and the next one
In a Poisson process, t follows an exponential
probability of observing n events in time t, given event rate 
probability that the next event comes after time t
Courtesy of A. Moralejo

42. Calculation of effective observation time
Distribution of the time differences:
In case of fixed dead time d, the distribution is still exponential with slope   The true rate of events  (i.e. before the detector) can be obtained from an exponential fit to the distribution
And teff = Nd,0 / 
Courtesy of A. Moralejo

43. Example effective time
l  slope
N  intercept
And teff = Nd,0 / 
Log scale!

44. Effective Area Order of magnitude? Aeff ~105 m2
Estimated from MC (gamma) data!

45. Effective Area Order of magnitude  Aeff ~105 m2
It’s roughly the Cherenkov light pool
Depends on the Zenith angle of the observations:
We estimate it with MC data

46. Differential energy spectra
Numbers of bins is of course variable and set by the analyzer
Errors are larger at high energies… why?
Usually fitted with power laws

47. The unfolding
Before publishing our spectrum, there is still one thing that we can do:
Use the MC data to calculate the errors that we perform and apply a correction to the data
limited acceptance and finite resolution
systematic distortions
reconstructed energy is not true energy!
How?
 With a matrix (correlation matrix) correlating the true Energy (from simulations) to the reconstructed Energy (estimated through RF)

48. Crab Nebula Spectrum
…and SED

49. Other examples Extremely variable objects
Extremely distant objects (z=0.536)

50. Physical Results
The lightcurve

51. Integral Flux Differential (energy bins)  integral (energy threshold)
If studied as a function of the time we make a light curve
How can we build a light curve?
Roughly speaking: we always have the same ingredients, but integral in energy:
Theta2 plot above Eth
Integral effective area above Eth
The time… is the same!
Time evolution
studies
PKS
H.E.S.S. collaboration, A. Abramowski et al. A&A. 520 (2010) A83

52. LC Examples
3C 279
HESS PKS

53. … we are only one piece! 1ES 2344+514 1ES 2344+514
VERITAS Coll., Acciari et al., ApJ 738 (2011) 169

54. By the way:
you can find the fits files of our final results at:

55. EBL: The gamma ray horizon
Last Comment
EBL: The gamma ray horizon

56. VHE photons absorption by the Extragalactic Background Light
VHE photon + diffuse light
 electron-positron pairs production
VHEEBL  e+e-
Absorption:
dF/dEobs= (dF/dEem) e-t

57. VHE photons absorption by the Extragalactic Background Light
EBL SED
Hauser and Dwek (2001)
VHE photon + diffuse light
 electron-positron pairs production
VHEEBL  e+e-

58. VHE photons absorption by the Extragalactic Background Light
Dominguez et al. (2011)
VHE photon + diffuse light
 electron-positron pairs production
VHEEBL  e+e-
Large uncertainties!

59. g-g opacity Absorption: dF/dEobs= (dF/dEem) e-t
Our range of observations
z = 1
z = 0.5
z = 0.3
z = 0.1
z = 0.03
z = 0.01
z = 0.003
EBL Model
Franceschini et al. (2008)

60. g-g opacity Absorption: dF/dEobs= (dF/dEem) e-t Strong suppression
z = 1
z = 0.5
z = 0.3
z = 0.1
z = 0.03
z = 0.01
z = 0.003
EBL Model
Franceschini et al. (2008)

61. EBL absorption effect
EBL model:
Franceschini et al. (2008)

62. Current “limit”
The FSRQ 3C 279 at redshift 0.536

63. Conclusions Thank you! The analysis of Cherenkov data is non trivial…
…but it is worth!!!
Probably there are still large margins of improvements:
New analysis techniques
More powerful instruments (CTA)
The present and future is multi-instrument!
Thank you!

64. Exercise… how many Crab gammas in 1 hour?
dN/dE = (3.3±0.11)×10−11E−2.57±0.05cm−2s−1TeV−1

65. Model analysis: A Global reconstruction method
An alternative to the use of image parameterization
MC template
Analytic model (based on MC) gives the expected signal in each pixels as a function of E, Direction & Impact
A fit of the MC templates on the real data reconstructs at same time the E, direction, and nature (gamma/hadron)
This method developed by CAT and then by HESS is time consuming but provides the best results (for telescope arrays).

66. Reconstruction of the incoming direction
DISP method: Developed for single telescope data
DISP can be determined with:
- A parameterization:
Optimized decision trees
(Random Forest)
DISP
reconst. direction
major axis
centroid
Possible confusion with
symmetric direction
Image asymmetry and time
gradient help the distinction
All methods are based
on Monte Carlo Simulation