Comment by Vincent Le Brun on 2023-03-28 12:07:40:
try to make it simple : a given % of the +/- 4 sigma support should be covered by the samples actually used, whatever their location.

I would put 50%, even if it is a little more restrictive that the current rule for lines on the border, but on Euclid, 1 pixel might represent the 30% of the 1/2 support, while it will be harder with a global 50% rule

Comment by Didier Vibert on 2023-03-28 12:31:08:
Vincent Le Brun wrote in #note-1:
> try to make it simple : a given % of the +/- 4 sigma support should be covered by the samples actually used, whatever their location

yes this may be the way to go. But then, currently the amazed library has no knowledge of the sampling, and of the integrated spectral range of one sample. It assumes it is a dirac sampling of a convolved signal where the pixel integration is included in the LSF. Shall we had it as a parameter ?

Then, strictly speaking one sample is enough to fit the amplitude if we know the redshift which may be constrained by other lines/features. And the flux variance is supposed to contain the uncertainty due to the number of samples, if the estimated variance on the flux includes the uncertainty on the redshift (and width), which is not the case currently, the estimated flux-variance currently assumes a fixed known width and position...

>
> I would put 50%, even if it is a little more restrictive that the current rule for lines on the border, but on Euclid, 1 pixel might represent the 30% of the 1/2 support, while it will be harder with a global 50% rule

different conditions could also be set for redshift determination and linemeas.