Adding / Prototyping new IRF parametrizations

[maxnoe]

We discussed a lot over the last couple of years, but due to the amount of coordination and pure work, no concrete proposal or implementation thereof has emerged.

I'd like to discuss a bit, what on the Gammapy side would be necessary to implement new IRF parametrizations.

For now, let's assume we only talk about new parametrizations of the same components we currently have (effective area, energy dispersion, psf, background), not different approaches to factorizing the IRF itself.

What would e.g. adding a 3D effective area to Gammapy look like that fully supports:

• Use in analysis
• Use for simulation of event lists
• more?

Some questions, things to think about?

• Can we make the IRF API more general to make adding more IRF versions easier (as might also be required for new / different formats)?

E.g. currently, some makers deal with internal IRF specific things themselves (e.g. computing offset from a coordinate if irf.has_offset_axis) instead of having a generic method that just takes energy and 2d direction.

• What would be necessary to protoype / add support for a 2d PSF (tables or parametrized) around the source position?

• Can we come to a point where e.g. people or packages using Gammapy can implement a new class MyPSF(PSF) without needing to change Gammapy itself?

@adonath @registerrier @bkhelifi