Script to produce bundled PDF set

[Zaharid]

We should be able to make an LHAPDF set with combined PDF+as uncertainty. This is done by taking a normal PDF set and adding the central values of two alpha_s variation fits.

[Zaharid]

@enocera @juanrojochacon @scarrazza Do you have somewhere a specification of how these things should be labelled in LHAPDF (e.g. what are the relevant "error strings")

[juanrojochacon]

yes I will send the info later

[juanrojochacon]

Hi @Zaharid if you look in LHAPDF in one of the bundled sets
https://lhapdfsets.web.cern.ch/current/PDF4LHC15_nlo_mc_pdfas/PDF4LHC15_nlo_mc_pdfas.info
you see that the error string is
ErrorType: replicas+as
This is also the case for our bundled sets
https://lhapdfsets.web.cern.ch/current/NNPDF30_nnlo_nf_5_pdfas/NNPDF30_nnlo_nf_5_pdfas.info
and this one for the Hessian bundled sets
https://lhapdfsets.web.cern.ch/current/NNPDF31_nlo_hessian_pdfas/NNPDF31_nlo_hessian_pdfas.info

Hope this helps!

[Zaharid]

OK, this is the header of one of the extra replicas, which we should propagate:

https://lhapdfsets.web.cern.ch/current/PDF4LHC15_nlo_mc_pdfas/PDF4LHC15_nlo_mc_pdfas_0101.dat

I guess asking for an actual specification would be too much?

[juanrojochacon]

not sure I see what is your question here?

[Zaharid]

A piece of documentation saying "this is how to bundle the as sets". Anyway should be sufficient to look at these files.

@siranipour I guess in practice one should

• pull replica0 from the varied pdfs and copy the AlphaS_MZ and AlphaS_Val keys from the info file to the replica header.
• change the error type key in the new info file.
• And that's it?

[juanrojochacon]

yes, pretty much. As far as I can tell this is how the previous iteration of bundled PDF sets was constructed.

[Zaharid]

@juanrojochacon @enocera Incidentally is there some code around that implements this error specification so we can test something? I know nnpdf doesn't.

[siranipour]

Do we want a script or will a simple validphys action suffice?

[enocera]

Do we want a script or will a simple validphys action suffice?

I think that a vp action will perfectly do the job.

[enocera]

@juanrojochacon @enocera Incidentally is there some code around that implements this error specification so we can test something? I know nnpdf doesn't.

I'll leave @juanrojochacon to reply to this question.

[juanrojochacon]

Not as far as I can tell. But if you uploaded the bundled sets I can write simple code that carries out the test: just verify for a bunch of x and Q points that the error that the bundled set provides is the same as the MC error of the unbundled set added in quadrature to the alphas variations

[juanrojochacon]

I can test this with bundled sets that already exists

[juanrojochacon]

Hi @Zaharid I have written a simple script that does the test and applies it to the NNPDF3.1 bundled tests, and it works fine

bundled mean, std = 29.679044826171154 0.3474663407070872
orig 0.118 mean, std = 29.679044826171154 0.2090380038092176
orig 0.116 mean, std = 29.934210076088956 0.2515457024389879
orig 0.120 mean, std = 29.37910321898288 0.19770199075984504
Combination by hand, mean, std = 29.679044826171154 0.34746624690477296

in that I reproduce doing by hand what should be the result.

Once the LHAPDF grids for the bundled NNPDF4.0 fits are in the server, I will polish the code and cross-check them