GammaLib

When energy dispersion should be considered, the current GResponse::convolve methods compute the IRF by integrating for each observed energy the model over all relevant true energies. This means that for each evaluation, the correct spectral model will be considered.

An approximation would be to precompute the spatial component of the IRF for a specific fixed spectral model, and to use this energy-independent precomputed quantity for the spectral integration. The advantage of this approximation would be that the precomputed values can be stored in a cache, and the cached values can be used in the spectral integration. The disadvantage would be that the spectrum that is used to determine the number of events in a given bin would differ from the best fitting spectrum, leading to an error in the number of counts and in the spatial response. The question is whether these errors are actually acceptable.