account for efficiency loss in conversion

[volodymyrss]

No description provided.

[volodymyrss]

there is some confusion between efficiency and effective area. xspec uses "efficiency" basically for effective area (consider plot effi). But ISGRI treats them very differently.

Efficiency in ISGRI is part of the spectral response matrix, applied to all spectra and images before fitting.
Efficiency is a function of reconstructed photon energy, is generally <1, and depends on the part of the detector and on time.
The purpose of this separation of response is to make all pixels and time have approximately the same response.
This way, when combined (still taking into account response evolution explicitly), they provide maximal information.

In addition, efficiency includes dependency on source position. In principle, source position and detection part are related, but due to complexity of coded mask process, this relation is quite non-trivial.
Which is why, technically, the location-dependency and pixel-dependence are treated separately.

Because if these corrections, counts as reported by ISGRI (maps and images equally) are distantly related to actual detector counts. Furthermore, this relation depends on sky position, time (long-term and short variations), and most of all energy.

As a result, in some energy ranges, where efficiency is low, and actual number of counts is low, reported number of counts can be large, and have very large uncertainty.

For example, 20-40 keV at in 2020 would include 20-30 keV range with very low efficiency, and, effectively, unknown number of counts.
Hence it not generally possible to constrain counts in broad 20-40 keV energy range, even if there is relevant sensitivity in 30-40 keV, since it is not known what is the contribution of 20-30 keV.

This is an extreme case, but in all cases when efficiency is not constant within energy range, further assumptions on source properties are needed.

For example, ii_shadow_build assumes "Crab spectrum" (-2) to compute broadband efficiency, applicable to all images, lightcurves, spectra. This is not particularly relevant for narrow-bin spectra, but images and light curves, if computed in ranges including extreme variations of efficiency can be affected very strongly by this assumption.

Since low-energy efficiency cutoff grows with time very substantially, 20-40 keV does not have this issue in 2002, but it has very strongly in 2020.

Which is why, for simplicity, it is advisable to use narrower energy ranges in analysis of the whole mission (e.g. 35-40 keV), and use sub-ranges for special treatment.

Alternatively, it is also possible to customize this model dependency assumed in efficiency. In the first approximation, position-dependence is not critical (but it's effect is to be verified).