Reproduce 4.0 aN3LO methodology for CF computations

[Radonirinaunimi]

@kamillaurent, @jekoorn, This is the branch that could be used to compute the aN3LO DIS coefficient functions that reproduce the 4.0 predictions. The idea would be to recompute the theories with 4.1 settings but using the 4.0 methodology for the N3LO approximation.

[Radonirinaunimi]

But perhaps before recomputing the theories, let's wait for the Code Meeting this afternoon for the proper action plan.

[jekoorn]

thanks! I agree re code meeting.

[jekoorn]

Which version of yadism should I use for 4.0 methodology?

[Radonirinaunimi]

I don't remember on top of my head the exact released version but this branch has been explicitly checked to reproduce exactly 4.0.

[jekoorn]

Do we want to use the 4.1 theory with PTO set to 3? Or the new cards that we are working on in NNPDF/nnpdf#2494?

[jekoorn]

On a separate issue, ...

EDIT: let's do it #387 🙃

[felixhekhorn]

Let me repeat again what I already said before: we need to be reeeeeaaaally careful with this one. In particular,

• I haven't seen any faithful N3LO fit with 4.1 settings yet, thus we should not worry about reproducing 4.0 yet either.
• we may (or may not) need some hacks here, but we need to be very careful and I'm not convinced that touching src/yadism/coefficient_functions/splitting_functions/__init__.py or src/yadism/esf/esf.py, which are at the very core of Yadism, are really necessary ...

[Radonirinaunimi]

• I haven't seen any faithful N3LO fit with 4.1 settings yet, thus we should not worry about reproducing 4.0 yet either.

• we may (or may not) need some hacks here, but we need to be very careful and I'm not convinced that touching src/yadism/coefficient_functions/splitting_functions/__init__.py or src/yadism/esf/esf.py, which are at the very core of Yadism, are really necessary ...

I agree with this. If one needs to use older version of Yadism, one must use the old cards definition instead. However, these sets of run cards have not been defined at all and thus there is no incentive to "reproduce" 4.0 or to use old Yadism at all.

PS: On the other hand, "faithful aN3LO fits" here is a bit of misstatement because how do you know it is faithful?

[felixhekhorn]

PS: On the other hand, "faithful aN3LO fits" here is a bit of misstatement because how do you know it is faithful?

point take, a bad name 🙃 , what I mean is a full-fledged N3LO fit using all sources of uncertainties (and not just some)

[jekoorn]

must use the old cards definition instead

Fair enough. Even if it entails hard-coding some pto=3 here and there, and putting the CKM values in by hand? I get that we want to be thorough, and not change the core of yadism, but I really don't see the danger here...
Yet this may be naive statement.

I will look into the old runcard definition. Do we have an example somewhere?