make sure that alphas and Q2 grids agree in `alpha_s_bundle_pdf`

[RoyStegeman]

The Q2grid of the alphas variation members in NNPDF40_an3lo_as_01180_mhou_pdfas is different from the Q2grid (and number of points in the alphas grid) in the .info file. This means that alphas cannot be computed for the error members.

To avoid this in the future, it should be checked that these grids are consistent.

The error can be reproduced as follows:

>>> import lhapdf
>>> pdf = lhapdf.mkPDF("NNPDF40_an3lo_as_01180_mhou_pdfas/101")
>>> pdf.alphasQ2(10)

Which then raises the error

RuntimeError: AlphaS value and Q interpolation arrays are differently sized

[scarlehoff]

What do you mean by that?

Is this related to this bug? NNPDF/eko#391 (the ekos prepared before the fix was merged would produce a wrong-ish info file)

[RoyStegeman]

No, for NNPDF40_an3lo_as_01180_mhou_pdfas between mc and mb for members 101 and 102 the Q2 grid is
1.5100000E+00 1.6500000E+00 1.7874390E+00 1.9429050E+00 2.1193750E+00 2.3204100E+00 2.5502940E+00 2.8142020E+00 3.1184120E+00 3.4705780E+00 3.8800750E+00 4.3584520E+00 4.9200000E+00

while for members 0-100 it is
1.510000e+00 1.657384e+00 1.827949e+00 2.026319e+00 2.258232e+00 2.530851e+00 2.853170e+00 3.236569e+00 3.695538e+00 4.248669e+00 4.920000e+00

The AlphaS_Qs in the .info file is the same as the one for members 0-100, but obviously disagrees with what is there for members 101 and 102

Not sure why this happened originally, but it's something we can check for when combining multiple PDF sets into one.

[scarlehoff]

@RoyStegeman is this now fixed? (I would hope so given the many alpha_s datasets to be released :___ )

[RoyStegeman]

I don't think this has been addressed.

The issue here was because in the _pdfas PDFs that we release, we combine different fits into one PDF. Specifically we added rep0 of fits at as=0.117 and as=0.119 as reps 101 and 102 to a standard fit at as=0.118. So if these different grids have different evolution settings that causes an inconsistency.

This doesn't have anything to do with the alpha_s sets to be released since they are still just all individual fits.