Clarify that imported subgrids have to be provided as convolution with `f` and not `xf`

[felixhekhorn]

We should document more prominently that when providing ImportSubgridV1, i.e. when not using the filling methods, the convolution is with f(x) and not x*f(x).

yes, Eq. (2.8) of the paper says so and I think I could dig out the relevant line here:

pineappl/pineappl/src/convolutions.rs

Line 226 in f67347c

xfx(pid, x, mu2) / x

However, grid consumers have to specify xfxs , e.g. here:

pineappl/pineappl_py/src/grid.rs

Line 395 in f67347c

xfxs: Vec<Py<PyAny>>,

This may confuse new users who are both grid producers and grid consumers (to test the former).

The convention is fine (and set in stone now), it is a matter of documentation.

cc @samiyrj

[cschwan]

This was never documented, because filling subgrids yourself was never encouraged to do and the user should always use Grid::fill. However, I suppose there are programs that want to perform the interpolation themselves and it is needed now.

[felixhekhorn]

I see your point, but this can and is happening for generators for which PineAPPL comes as an add-on, when the main code (with interpolation) is already in place - as I think it was happening for yadism (which is surely a relevant grid provider)

[Radonirinaunimi]

Note that this is always the case for predictions for which Coefficient Functions are computed (DIS, SIA, SIDIS, etc.).

[cschwan]

I see your point, [..]

I wasn't trying to make a point 😄, I only wanted to point out the reason for the status quo. I agree that this is a (missing) documentation problem that should be fixed.

[Radonirinaunimi]

Please have a look at 4e3f09c. I tried to document this extensively in that commit given that this might be a confusing aspect.