Add a basis with c & cbar

[scarlehoff]

Which also requires ensuring that the rotation to the fktable basis and sum rules work.

• Add the c-cbar basis
• FKTable rotation (see Add a basis with c & cbar #1627 (comment))
• Sum rules
• Check that old FK Tables are not optimizing out the ccbar asymmetry
• Check that pineappl FK Tables are not optimizing out the ccbar asymmetry

@felixhekhorn I added you to this issue just because of the last point ofc, let us now and if it is we just do a theory 4XY with the asymmetry.

Although I wonder whether the fktables in theory 200 would allow for it (I'll have a look at that)

[tgiani]

thank you @scarlehoff I m happy to work on this

[tgiani]

in the FK tables I think there is already V15, which now is taken equal to V, no?

[felixhekhorn]

• Check that old FK Tables are not optimizing out the ccbar asymmetry

• the relevant line in pineko is: https://github.com/NNPDF/pineko/blob/f7164fa8df0bf78a2cde71dd30fede24779d4e47/src/pineko/evolve.py#L182
• which calls this pineappl function https://github.com/NNPDF/pineappl/blob/58158e6d97f8e6a8a5b52d27743dd7b9b1924a06/pineappl/src/fk_table.rs#L291
• however, the problem is that the argument of that function is currently constructed by investigating various fields in the runcard https://github.com/NNPDF/pineko/blob/f7164fa8df0bf78a2cde71dd30fede24779d4e47/src/pineko/theory_card.py#L46 which makes it impossible at the moment to generate Nf4Ind which is the setting we need. We need to adjust that function.

[scarlehoff]

in the FK tables I think there is already V15, which now is taken equal to V, no?

Yes. The value is allowed. I'm wondering whether in Th 200 they are optimized by making V = V + V15.

[tgiani]

from the n3fit point of view, in order to add a parameterization for c-cbar, is it enough to use nodes_per_layer: [25, 20, 9] in the runcard and then just change the FK table rotation to account for V15?

[scarlehoff]

I've removed my previous msg. No, unfortunately in pdfbases.py we implicitly assume that the maximum number of flavours that are going to be fitted is 8.

So we need to scratch this function:

nnpdf/validphys2/src/validphys/pdfbases.py

Line 656 in 94b80c3

def fitbasis_to_NN31IC(flav_info, fitbasis):

And then address this TODO from 3 years ago

nnpdf/n3fit/src/n3fit/layers/rotations.py

Line 64 in 94b80c3

# TODO: Generate a rotation matrix in the input and just do tf.tensordot in call

which didn't matter until now.

[Radonirinaunimi]

And then address this TODO from 3 years ago

nnpdf/n3fit/src/n3fit/layers/rotations.py

Line 64 in 94b80c3

# TODO: Generate a rotation matrix in the input and just do tf.tensordot in call

which didn't matter until now.

Oh, if you remember @scarlehoff we discussed about this a bit at some point in the context of nuclear PDFs (where only 6 flavours are active) and I sort of recall I had a go on this.

[tgiani]

right yes I remember this function. But it s ok, just need to change it with the new basis definition before doing the rotation to the FK basis I would say

[scarlehoff]

Yes. It shouldn't be a problem. It's just not as ready as I thought :P

Oh, if you remember @scarlehoff we discussed about this a bit at some point in the context of nuclear PDFs (where only 6 flavours are active) and I sort of recall I had a go on this.

Right. But in principle this is the opposite scenario. Actually in this case the "relevant" part is in the vp side so the two changes should not be incompatible I'd say.

[tgiani]

I ll have a go and open a PR to iterate with you and @Radonirinaunimi

[scarlehoff]

In the meantime, I had a look at theory 200 and there is V15 in the fktables so we can use th 200 for now.

[scarlehoff]

Closed with #1629

For the fktables with ccbar, that's pineko's job NNPDF/pineko#64