Allow fitting with perturbative charm (or any other basis)

Created by: scarlehoff

There were a few places where the flavours were fixed (which assumed charm was fitted), this basically propagates the information from the runcard, hoping that at the end it will be rotated to whatever is the correct thing wanted by the fktable.

Two things are missing here,

1. An actual basis (from a quick look, all basis in pdfbases.py are for fitted charm right?, so it needs to be added)
2. An actual check (at the moment I am acting as if the fktables would be just the same meaning that combination 1-2 for theory 53 is the same as for theory 65. Is that the case? (i.e., is any reshuffling of the PDF output needed from one theory to the other?)

In any case, these were the basic points I mentioned in #1002 (closed) blocking the perturbative charm fit from the n3fit side.

requested review from @enocera

requested review from @enocera

added n3fit label

Created by: scarlehoff

I guess the perturbative charm basis would be the one used here, if so I think this code should work. If (2.) above is true what I am doing is just setting to 0 the charm (the same as we do for the photon right now) and trusting the fktable not to ask about it.

I guess the perturbative charm basis would be the one used here,

That's correct. It is the current flavour basis with no T15.

if so I think this code should work. If (2.) above is true what I am doing is just setting to 0 the charm (the same as we do for the photon right now) and trusting the fktable not to ask about it.

The answers to point 2. above should be respectively

Is that the case?

Yes

(i.e., is any reshuffling of the PDF output needed from one theory to the other?)

No

Created by: scarlehoff

Given the comment above and that (as expected) once n3fit arrives to the fktable everything is (length of the) basis-independent, I think this is ready for review.

Created by: Zaharid

I think we need a check to make sure that the basis is consistent with the theory. AFAICT anything that is not matching basis and theory is an avoidable mistake in the fit.

There is the flag theoryid.get_description()["IC"] that tells is whether a theory is fitted charm or not.

Created by: scarlehoff

Well, clearly both the test and check I added work :)

Maybe we need a theory 162-like with perturbative charm.

@scarlehoff I just wanted to say that this PR can now be completely tested by using theory 211 (perturbative charm, NNLO) and theory 212 (perturbative charm, NLO). Both theories are available through vp-get theoryID. Are you willing to test this PR in conjunction with those theories? Thanks.

Created by: scarlehoff

Sure, I can do that. I'll run NNLO first, same dataset as NNPDF40_nnlo_as_0118, right?

Yes.

Created by: scarlehoff

The fit "works" but there's something funny going on https://vp.nnpdf.science/d8zgwtKMRcytQZ4zIlfqJQ== looking at the training and validation chi2s (very high with respect to the central value) there might be a bug in the reporting of the chi2 within n3fit (so the chi2 is not computed correctly and the fit is not stopping at the right point).

I'll look into this.

Created by: scarlehoff

I think the problem with the chi2 during the fit was due to using a IC t0, didn't know it was such a big effect. I've ran a fit iterating preprocessing and t0 and should be ready later today.

Created by: scarlehoff

Yes, once iterated the fit works as expected: https://vp.nnpdf.science/EeRrB0nWToOo_NVwZuV9OA==

Don't know whether the big difference for T15 is pointing to a problem on my definition of the basis or whether it is expected. All other differences happen at low-x.

The fit looks OK to me. Isn't the difference in T15 driven by the difference in charm, which of course has a completely different behaviour if it is not fitted?

approved this merge request

Created by: tgiani

Review: Approved

Merged by: scarrazza at 2021-01-13 15:45:53 UTC

closed

merged manually