Interface QED and N3LO, n.2

[niclaurenti]

reopening #311 after rebasing master

[niclaurenti]

Now this is more or less ready. @alecandido I wasn't able to reconcile the qed branches as you asked. I think that the uglier branch is the one happening in

eko/src/eko/evolution_operator/__init__.py

Line 273 in f39b012

if order[1] == 0:

but it is quite complicated to change it since qcd and qed evolution require different ingredients and also the structure is a bit different(in one case we have singlet + non_singlet, in the other also the valence). If you have proposals we can try to restructure it a bit.

Then, whenever we will tag the next eko version, we have to pay attention to the fact that i remover the argument qed from apply_pdf (that information now is contained in the EKO that is passed to the function). The only part that is affected should be
https://github.com/NNPDF/nnpdf/blob/db8b790c1bacf52f71cc3a7fabf0fd9674cbf476/n3fit/src/evolven3fit_new/evolve.py#L178
but maybe some other parts of the NNPDF code are using this function (maybe @andreab1997 knows better)

[giacomomagni]

Looks good to me. to_flavor_basis_tensor still contains a qed=False as default which can be dropped?

[niclaurenti]

Looks good to me. to_flavor_basis_tensor still contains a qed=False as default which can be dropped?

Thanks I missed it. I will fix it asap

[felixhekhorn]

Did you check that the apfel benchmark is still fine? else this seems alright!

[niclaurenti]

Did you check that the apfel benchmark is still fine? else this seems alright!

@felixhekhorn The apfel bench at order = (3, 2) is still good. However, I cannot test order=(4,2) since apfel doesn't implement N3LO (I think)