IC40 untriggered flare search – Q&A sample Mike Baker PS Call – April 19 2010. The goal of the untriggered flare analyis is to provide an all-sky search for transient phenomena. To start here is a scrambled skymap with 0.5 degree binning. The color scale is -log10 of the estimated
IC40 untriggered flare search –
PS Call – April 19 2010
The goal of the untriggered flare analyis is to provide an all-sky search
for transient phenomena. To start here is a scrambled skymap
with 0.5 degree binning. The color scale is -log10 of the estimated
Next I have the best-fit mean of the Gaussian for all points on the sky,
including under-fluctuations (best-fit n_src=0).
The same fit parameters for flares tend to be found in a radius of about 2 degrees.
We can also look at the duration of the most significant flare found (in log10(days)), here also plotted including under-fluctations.
Due to the method of finding flares very long widths are returned when no interesting flare is found (here in red).
I have a plot of the signal-ness of the events (including the energy term) as a function of distance to the tested spot. Left is for data, where the black line is the median weight. It tells me that the an event two degrees away from my source will have an average S/B of ~10^3.
The right plot was made with E^-2 spectrum signal injected from the source. A zoomed-in on the interesting region of this plot is available on the wiki.
Levent asked about how well we'd be able to reconstruct the flare event parameters if we do see something. I did a quick check to see how much that will depend on the flare itself vs the number of signal events.
Here I inject exactly 2 E^-2 spectrum events and check what the reconstructed parameters are as a function of the true flare width. With only 2 injected events, a few percent of trials one of the events is mis-reconstructed and the flare is missed.
It looks like flare widths are reconstructed within around +0.2/-0.3 of the log10(width) for 5 events, and +0.2/-0.5 of the log10(width) for 2 events.
Here is the same set of plots for exactly 5 E^-2 signal events injected.
I also had a look at what the DP and AUL looks like converted into fluxes, to try to compare to GRB analyses. Alexander also asked about how this would perform with something like the “naked-eye” GRB. I still need to think about this a bit, but it's very interesting!
One final point on the periodic search, I did a bit more investigation comparing the case where we use marginalization vs. not. I find that it leads to a greater number of under-fluctuations from scrambled samples (see arrows below). This effect is large enough that it skews the averaging in the AUL to higher values than we would expect from the time-integrated search.
– Large crosses are the result for exactly n=0,1,2... injected events (size reflects error in Y), the dotted line is the Poisson average.
– Blue lines mark the result for the AUL in each case.