udine 11- 03 - 2004 nicola omodei 1 grb trigger algorithms from dc1 to dc2 nicola omodei riccardo...

1Udine 11- 03 - 2004 Nicola Omodei

GRB Trigger AlgorithmsFrom DC1 to DC2

Nicola Omodei

Riccardo Giannitrapani

Francesco Longo

Monica Brigida


DC1 Closeout status

• Many people involved in GRB detection: I think that the generation of GRB has been a success!

• 4(+1) groups have been working on GRB detections:– David Band

– Jay Norris & Jerry Bonnell

– Riccardo Giannitrapani et al.

– Nicola Omodei

– Tune Kamae et al.


Band’s method• Break up sky in instrument coordinates into regions, and apply rate triggers

to each region. The regions are ~PSF in size (builds in knowledge of the instrument).

• Use two (or more) staggered regions so that the burst will fall in the interior of a region.

• Rate trigger—statistically significant increase in count rate averaged over time and energy bin.


Estimating the Background

• The rate trigger requires an estimate of the background (=non-burst event rate). Typically the background is estimated from the non-burst lightcurve.

• BUT here the event rate is so low that a region’s background estimated only from that region’s lightcurve will be dominated by Poisson noise. The event rate per region is a few×10-2 Hz.

• Band’s current method is to average the background over the FOV, and apportion it to each region proportional to the effective area for that region.


Problem with Background Estimation

• Problem: On short (~100 s) timescales the background is NOT uniform over the FOV. The ridge of emission along the Galactic plane causes many false triggers.

• Solution (not implemented yet): Better model of the background.

Region with false trigger •In the ~6 days of DC1

data,He found 16 bursts and 29 false triggers.•Note that his spatial grids extend to inclination angles of 65º and 70º.•The software He used was all home-grown IDL procedures.


• They used only one N-event sliding window as the first bootstrap step in searching for significant temporal-spatial clustering. Compute Log {Joint (spatial*temporal) likelihood} for tightest cluster in window:

Log(P) = Log{ [1 – cos(di)] / 2 } + Log{ 1 – (1 + Xi) exp(-Xi) }

• Their work is somewhat at 45 to main DC1 purposes. But DC1 set us up with all the equipment necessary to proceed:

• Future emphasis will move to on-board recon problems: highest accuracy real-time triggers & localizations.

Norris’s method


• Very sensitive trigger — incorporates most of the useful information.

• 17 detections: 11 on Day 1; 6 on Days 2-6. Some bright, some dim.

• No false trigger. Formal expectation any detection is false << 10-6/day.

• Additional aspects we will evaluate for on-board implementation:

• Floating threshold; 2-D PSF; spatial clustering (Galactic Plane)


Riccardo’s method• The aim of the Riccardo talk was to present a analysis tool call “R”

• He presented also an application of this tool for GRB searching, based on the quantile analysis.

• He looks for outliers in the distribution of the count rate


Some improvements

• Riccardo also compare the distribution of the counts with the Poisson distribution: The GRB are now really visible!

Outliers


Some other improvements

• Another way to see the outliers is looking at the (smoothed) counts map for (RA, TIME) coordinates (or for (DEC,TIME));

For all photonsFor outliers


Nicola’s method• First algorithm based

on the trigger on the differential count rate (this get rid of the fluctuation of the background due to the galactic plane).

• Very easy and fast algorithm!


The division of the sky

• The same algorithm can be applied separately in sub region of the galactic map. This substantially reduces the background (non-burst events).

5 x 5 array reduces the “background” by a factor 25.

Also faint burst can be detectable.

Direct (70˚ x 36˚) information on the localization.


The 25 lightcurves


Comparing the results

Generated 11 GRBs with spikes with more than 2 photons/secondBurst time David JJ Riccardo Nicola Nph

3000 * * * * 827000 * * 811000 * * * 211500019000 * * 3223000 * * * 1327000 * * * * 353100035000 * * 233900043000 * * * * 10747000 * * 41510005500059000630006700071000 * * * * 7175000 * * * * 7537900083000 * * * * 50

Spectral analysis done with XSPEC (Monica)Light curves visualizedPosition in the sky map visualized

Burst photons

GRB050718i


Common features and diversities• All of us triggered on the counts rate (in different ways): the gamma

background (simulated) is low compared with the burst flux.• Both faint bursts (few tens of photons) and bright bursts (some

hundreds of photons) have been successfully detected.• A big improvement of the burst trigger rate has been reached by

dividing the sky map in smaller region. This procedure represents a big advantage in terms of background “reduction”.

• David divide the sky map using instrument coordinates, maybe this is the reason of so many false triggers.

• Nicola, Jay and Jerry used galactic coordinates: no false trigger.• Nicola developed a simple (and fast) algorithm and detect burst as

much as Jay and Jerry did with more complicated algorithms.• Riccardo pointed out that one of the burst vanishes if the standard

cuts are applied to the data. This means that with with a realistic background which requires a realistic background filter, some burst photons will be killed by the filter.


GRB Trigger, Alert & DC2• On-board vs on-ground trigger algorithms. GBM comparison!• Develop a common interface for the burst alert algorithms• Better simulation of the background (including particles)• Background estimation• The development in other environments (IDL,Matlab,R, ROOT stand alone macros), is

very useful, BUT the key point for the DC2 will be the development of science tools!

On boardOn board recon (filter)

Fast & Low memory consuming

On groundFull recon

High sensitivityNo restriction on memory/time

Data storage

Buffer

Trigger AlgorithmTrigger on the counts rate

Likelihood …

Background estimation

SkyMap segmentation

Outliers


GRB Spectra

EventBin + XSPEC(Francesco tutorial, XSPEC tutorial …..)

Fitting models: power_law / grbm


GRB050720a

1634 countsTstart: 176761Tstop: 176880Ra: 128Dec: 65Flux: 2.9 E-6 erg cm-2 s-1


Power law model

Model: powerlaw<1> Model Fit Model Component Parameter Unit Value par par comp 1 1 1 powerlaw PhoIndex 1.74358 +/- 0.198269E-01 2 2 1 powerlaw norm 6.20041 +/- 1.32153 --------------------------------------------------------------------------- --------------------------------------------------------------------------- Chi-Squared = 849.3165 using 8 PHA bins. Reduced chi-squared = 141.5527 for 6 degrees of freedom Null hypothesis probability = 0.00


powerlaw

ignore **-1e5 1e8-**

Model: powerlaw<1>Model Fit Model Component Param Unit Valuepar par comp1 1 1 powerlaw PhoIndex 2.25854 +/- 0.403036E-012 2 1 powerlaw norm 12150.0 +/- 6275.72----------------------------------------------------------------------------------------------------------------------------------------------Chi-Squared = 36.43491 using 7 PHA bins.Reduced chi-squared = 7.286983 for 5 degrees of freedomNull hypothesis probability = 7.773E-07


• GRB_050718i 75415 / 75474 ; 92 / 57

• Model: powerlaw<1>• Model Fit Model Component Parameter Unit Value• par par comp• 1 1 1 powerlaw PhoIndex 1.79878 +/-

0.280451E-01• 2 2 1 powerlaw norm 18.8932 +/- 5.67197• ---------------------------------------------------------------------------• ---------------------------------------------------------------------------• Chi-Squared = 212.8927 using 7 PHA bins.• Reduced chi-squared = 42.57854 for 5 degrees of

freedom• Null hypothesis probability = 4.905E-44

GRB050718i700 countsTstart: 75415Tstop: 75473Ra: 92Dec: 57Flux: 2.6 E-6 erg cm-2 s-1


GRB050718i powerlaw

• ignore **-1e5 1e8-**

• Model: powerlaw<1>• Model Fit Model Component Parameter Unit Value• par par comp• 1 1 1 powerlaw PhoIndex 2.16394 +/- 0.472441E-

01• 2 2 1 powerlaw norm 2908.87 +/- 1490.38• ---------------------------------------------------------------------------• ---------------------------------------------------------------------------• Chi-Squared = 2.117105 using 7 PHA bins.• Reduced chi-squared = 0.4234209 for 5 degrees of freedom• Null hypothesis probability = 0.833

udine 11- 03 - 2004 nicola omodei 1 grb trigger algorithms from dc1 to dc2 nicola omodei riccardo...

Documents