ctools

I used 0.04 degress binning.
I tried also with a binning of 0.02 degrees and 0.01 degrees. In these cases there are a lot of pixels with 0 counts, so I smoothed a little the images with a Tophat2D kernel of 0.06 degrees.
The following plots show the unsmoothed and smoothed maps with these binnings obtained with a model containing a source with a flux 100 times of the Crab flux (case 3 in the description).

The structures in residuals are still evident, with a negative residual at the center of the source and positive structure around it. In the simulation and in ctmodel I am using caldb=“prod2”, irf=“South_0.5h”.

Knödlseder Jürgen wrote:

I’m wondering whether there is really a problem there or whether what you see is simply related to the fact that there are more events at the position of the Crab.

What do you mean?
With smaller pixels the dip is smaller, but there is still the “ring” structure around the source. I also tried to move the source away from the center of the ROI, I generated events with a different seed and used a smoothing with different radii. The small dip and the ring excess are still there.
Some other plots with these attempts are attached, but the results do not change much.
(The details of the attached maps are: pixel size = 0.005 deg, source offset from te center of the ROI = 0.5 deg, smoothing radius = 0.02deg / 0.05 deg)

I don’t understand what I am doing wrong in producing the residual maps. The pixels are much smaller than the PSF (which is roughly 0.02-0.05 above 1 TeV right?) and the smoothing radii are compatible. I don’t expect to see these structures in the residual map. Do you think this effect might be due to the smoothing?

This is the same smoothed map as in my comment #4, but using 48 bins between 0.1 and 100 TeV (16 bins/decade)

The effect seems to be reduced.

I also produced a residual map in terms of percentage (using the SUBDIV algorithm in csresmap). The following plots show the (unsmoothed) residual map zoomed on the region where the source is (radius of 0.15° around the source) and the relative distribution of pixels. A lot of pixels seem to have values larger than 1%.

Following the discussion at the last ctool call, I tested the residual map using a custom energy binning generated with csebins.
The result is not very different from previous cases.
In the following plots I compare the case in which I used 40 bins equally spaced in log scale with the case in which csebins is used. For the record, the custom ebin file generated by csebins contains 33 energy bins.