how to add size of energy bin to differential flux points about gamma-astro-data-formats

[cdeil]

You can add e_min and e_max to denote the reco energy bin in which the differential flux measurement was made. At least that's what I did for gamma-cat. One example: https://github.com/gammapy/gamma-cat/blob/master/input/data/2005/2005A%2526A...437L...7A/tev-000039-sed.ecsv

Format spec: http://gamma-astro-data-formats.readthedocs.io/en/latest/spectra/flux_points/index.html#energy-columns

@woodmd - I'm assigning this issue to you. Either close, or if you have time, you could also add this to one of the example files in the spec and explicitly state that using e_min and e_max for differential spectra is allowed?

[woodmd]

The intention was that e_min and e_max could apply to both integral and differential representations of the flux measurement. I can add some text to the spec to make this more explicit.

[cdeil]

@woodmd - Thanks!

The way I'm using e_min, e_max is as follows:

• For differential spectral points, it's the reco energy band in which the spectral point was measured
• For integral spectral points, the assumption is that that measurement has taken care of energy dispersion somehow and the quoted integral flux is for a true energy band.

I don't know if this way to fill and interpret e_min, e_max, with different semantics for differential and integral spectral points, is what you intended or use it yourself. But @woodmd, it's a non-obvious point that probably others will wonder about too as they do measurements where energy dispersion has a big effect (e.g. pulsar spectra from TeV instruments), so it's good if there's a note on that point in the spec.

[woodmd]

I've opened #87 which includes some text to clarify that e_min and e_max can apply to differential quantities.

Regarding reco vs. true energy my intention was that SEDs would be used only for measurements in true energy since this is generally what you are interested in when you are comparing to a measurement from another instrument or a theoretical model. If you distribute measurements in reconstructed energy it's not clear to me how one would use those without also providing additional information about the instrument response (e.g. an energy dispersion matrix). I also don't see any particular reason to make a distinction between differential and integral measurements with respect to the influence of energy dispersion.

To the extent that there may be large uncertainties from energy dispersion I think a better way to handle this would be as an additional systematic error on the normalization in each bin (perhaps we could define a new column for this). To give a concrete example, until fairly recently most LAT spectral analyses neglected energy dispersion entirely (for instance all spectral measurements from the 3FGL, 2FHL, and 3FHL) but we still treat these as measurements in "true" energy even though technically the measurements were performed in bins of reconstructed energy. This approach is valid as long as we account for the effect of neglecting dispersion in the total systematic error budget.

Low energy data points for both the LAT and TeV instruments can be somewhat tricky since it can be difficult to derive a fully model-independent measurement of the spectrum due to the strong migration effects -- i.e. the influence of the global model can become much larger than the statistical error. This was part of the motivation for the reference spectrum columns which allow one to define the global spectral shape that was used for the SED extraction. It's perhaps not an ideal solution so we could think of creating a new SED sub-format that would allow you to carry around the migration matrix along with flux values.

[cdeil]

I think this issue was fully addressed by @woodmd in #87 . Closing it now.
@GernotMaier - Please re-open or file a new one if something is still missing or unclear.