ctools

It should be checked whether such an analysis can actually be emulated using the existing tools, running for example csphagen on On and Off runs and combining the results for the analysis. If possible, one could simple write a cscript that performs the necessary steps.

What I have in mind is running csphagen twice, once on the ON observations and once on the OFF observations, and then using the ON PHA file of the OFF observations as OFF PHA file for the analysis. Maybe some fiddling with the files is needed, and the XML file has probably to be created “by hand”, and these steps, as well as the csphagen runs, could be integerated into a new cscript. Probably also some code is needed to determine the ON region for the OFF observations.

After discussion with Marianne it looks like the alpha factors need to take into account optical efficiency, that is only possible to do via the (IRF) background model.
Therefore, it seems that this requires some modification at the level of GCTAOnOffObservation.
Below I outline my proposal.

• modify the private methods set and compute_alpha so that they always take in input two observations and two model containers (one for On and one for Off)
• add one more constructor with two observations and two model containers; in the two constructors the observations and model containers will either be set to be the same (same observation for On and Off) or to two different ones.

• add the option OFF to the parameter bkgmethod; when this is invoked, query for an Off observation container (must be same size as On observation container)
• internally the script will have a new member variable self.obs_off to store the dedicated Off observations, set to None unless the method OFF is used
• add dedicated method to retrieve the sky region in the Off observation that correspond to the same camera region as the On region in the On observation, call this method in set_background_regions (implemented in csphagen to keep the GCTAOnOffObservation methods as general as possible)
• implement in process_observation the case bkgmethod=OFF

The new constructor for GCTAOnOffObservation needs to be added also in the Python extension GCTAOnOffObservation.i

Looks good. I’m however wondering why you need two model containers? The source model in the model container is only used for ARF computation. For the background models we can use the observation ID to assign a given background model to a given observation.

I think it’s correct that the Off spectrum exposure should be used in N_bgd. I probably used on the On spectrum exposure since in the current code On and Off spectra have the same exposure.

I had forgotten that energy dispersion is disabled by default in simulations and 3D analyses, but is always enabled in On/Off analyses. I repeated my test enabling it everywhere.

Now the results from unbinned analysis and all the flavours of On-Off analyses are almost identical.

I looked into the GammaLib code. In GCTAOnOffObservation::compute_alpha I see:

        // Get CTA observation pointing direction, zenith, and azimuth
        GCTAPointing obsonpnt  = obs_on.pointing();
        GCTAPointing obsoffpnt = obs_on.pointing();

        // Get CTA observation pointing direction, zenith, and azimuth
        GCTAPointing obsonpnt  = obs_on.pointing();
        GCTAPointing obsoffpnt = obs_off.pointing();

Yes, absolutely, thank you for catching this! I’m actually puzzled that this did not show up in the test I have described above

While checking why the error you found did not show up in my test (still not understood), I found another problem (missing statement, would have ignored exclusion regions for the OFF method). I pushed a fix to my branch

Tibaldo Luigi wrote:

While checking why the error you found did not show up in my test (still not understood), I found another problem (missing statement, would have ignored exclusion regions for the OFF method). I pushed a fix to my branch

On the GammaLib side or the ctools side?

ctools (in csphagen)

Tibaldo Luigi wrote:

ctools (in csphagen)

The if statement that you added at 15:38?

        # ... otherwise store region in container
        if is_valid:
            regions.append(region)

Yes, thanks