Proposal for a one-dimensional, unbinned, spectrum format

[cosimoNigro]

Hello all,

the format currently foresees binned spectral data, the so-called OGIP files, in turn based on NASA's definition.

The OGIP format provides all the input necessary for a classical one-dimensional binned spectral analysis.

Me and @javierrico would like to ask whether there might be interest in adding an unbinned one-dimensional spectral format.

Unbinned one-dimensional likelihoods have been (and are being) used - for example - in dark matter searches, see the likelihood definition in Aleksić et al. (2012) and its later usage in Aleksić et al. (2014). Additionally, any point like analysis (e.g. extragalactic) can profit of this technique.

In my opinion it will not be a big effort to introduce such definition in the format. Prototypical 1D unbinned spectral data should contain:

1. an ON event list, containing all the events in the source region with their reconstructed energies;

2. an OFF event list, containing all the events in the background region with their reconstructed energies;

3. A column, or a HEADER keyword, with the ratio between the ON and OFF region exposures.

[ alternatively to 1) and 2) a single Event list with reconstructed energies and the ON / OFF columns could be used; ]

4. unbinned effective area as a function of true energy, at the offset of the observation, corrected with the PSF (if full enclosure IRFs are provided);

5. unbinned energy migration, at the offset of the observation (alternatively resolution and bias as a function of true energy).

From the point of view of the science tools, I believe gammapy and ctools already produce data with such informations, they are simply not stored. Such data should be the intermediate products obtained from DL3 right after the signal / background separation in SpectrumExtraction and csphagen, respectively. They could be dumped in a FITS file before being binned to produce the OGIP files.

Any thoughts?

Thank you.