dis+jets implementation

[t7phy]

No description provided.

[scarlehoff]

Haven't tested them yet, only had a look through the metadata, some general comments.

[scarlehoff]

@enocera one question about the generic DIS cuts that we have in the code (I have a weak recollection of having talk about this or something similar), should they also apply by default to these datasets?

If yes, maybe it is good to add the x to the kinematics of DIS+j since it is used by some of the cuts?

(the q2 cut should work by itself)

[enocera]

The DIS+jet data sets are NOT subject to the generic cuts for DIS we have in the code (certainly not to the x cut; for Q, I'd say that all the data points are above the nominal cut, see Table 2.2 in the NNPDF4.0 paper. So, even if the Q2 cut is there and even if it works, it should have no effect). However, we may want to have a cut in pT, which, I hope, can be implemented in filters.yaml. I need to check the HERA paper to retrieve the optimal value for the cut in pT.

[enocera]

The kinematic cut to be implemented is (pT^2+Q^2)>100 GeV2 (that is, one should retain points with pT^2+Q^2 larger than 100 GeV2).

[enocera]

Using the aforementioned cut, you should see the reduction of data points summarised in Table 1 of https://arxiv.org/pdf/2112.01120.pdf.

[t7phy]

@enocera @scarlehoff I have changed the H1 351pb-1 datasets to now use the tables given by kt (instead of anti-kt). I think with this, the numbers should match with the old implementation.

[enocera]

@enocera @scarlehoff I have changed the H1 351pb-1 datasets to now use the tables given by kt (instead of anti-kt). I think with this, the numbers should match with the old implementation.

The data.yaml should match, however, for a dataset the problem was also with the full breakdown of sys uncs. Let me take care of that.

[t7phy]

This PR is now fully ready with all the suggestions from above implemented.

[scarlehoff]

Thanks for this.

I guess you already tested this with the reader and checked that it worked after your changes?

[t7phy]

Thanks for this.

I guess you already tested this with the reader and checked that it worked after your changes?

@scarlehoff Thanks for reminding me of this, I was counting on from pdf4lhc times. I just tried to do a DTC and got this error: KeyError: 'Label HERAJET not found in PROCESS_DESCRIPTION_LABEL' , so ofc we need HERAJET and HERADIJET added to possible process types in the reader/validphys. This can be added in #1678 I guess.

However, to avoid delaying this PR merging, I remove the 'hera' part from my local copy metadatfiles in process types to be able to produce DTC, and it works well for me, here it is for reference: https://vp.nnpdf.science/92VrFzd4RI-HkEUf4F5zIg== so to sum up: yes, it's tested and working well, however we need the above mentioned fix in #1678

[scarlehoff]

Completely forgot about them!

This is a great candidate to start adding only the entries that are required (such as the q-x mapping)

@enocera @felixhekhorn how would you compute x for the DIS +1/2j? There's no simple x the moment you have a recoil at LO...

[felixhekhorn]

Completely forgot about them!

This is a great candidate to start adding only the entries that are required (such as the q-x mapping)

@enocera @felixhekhorn how would you compute x for the DIS +1/2j? There's no simple x the moment you have a recoil at LO...

if possible I'd like to change the behaviour on the kinPlot to a more physical one (in my opinion) ... @enocera, I guess there are reasons on why it was chosen in the old way it was chosen, so maybe I'm missing something ...

however, speaking as a theorist I would argue the following way:

1. for fully inclusive DIS, fully inclusive DY and SIA there is a 1:1 mapping at LO between measured variables and the variables in the coll. distribution, because they all rely basically in the same diagram (suitable rotated around)
2. for anything else there is no clear mapping between measured stuff and PDF(/FF) variables, but there is a convolution involved (or in more physical speak: there is a recoiling something), i.e. an integral with running PDF variables
3. what I would suggest for the x* mapping is to always map to the lower bound of the integral (*a name is just a name, but exactly for this mess we avoid in yadism this name)
   • this would leave 1. unaltered at higher order, since the lower integration bound is always the measured Bjorken-x
   • we would thus need to read the kinPlot not as a point, but rather as a range: the dataset is sensitive between the marker and x=1
   • in the specific case of DIS + 1/2j, I would need to look up the specific formulae ....
   • specifically I don't like this in the old format: there is no PDF at x=1 ...
   • (another option would be to mark the spot where the contribution to the observable is the biggest, but a) this is complicated to determine and b) is a PDF dependent statement)
4. for the Q2 mapping, I suggest to do the same thing: always use the lowest (central) scale
   • same as above: this would leave 1. unaltered, since there is only a static scale choice
   • same as above: we would need to read the kinPlot as a range: the dataset is sensitive between the marker and Q2=oo (which does not harm due to asymptotic freedom)
   • the choice of the lowest scale can e.g. be relevant in the case at hand as jet observables can and do have a dynamic scale

[scarlehoff]

I'm not sure the x coverage is 100% correct, maybe some of the approximations were too approximated (or I made a mistake coding the formula).

https://vp.nnpdf.science/vxS8mWGySsu-cehRFnhSOA==

Both the process label and the formula can be modified in this file:

nnpdf/validphys2/src/validphys/process_options.py

Line 233 in 7e6599a

HERAJET = _Process(

[t7phy]

The formula looks correct (from what ERN told us that day). Why do you say the x plot seems incorrect?

[scarlehoff]

I've merged and rebased this branch onto master: 625372d

[scarlehoff]

The formula looks correct (from what ERN told us that day). Why do you say the x plot seems incorrect?

Almost 10^-7 looks too small for me. Maybe there was a square missing.

[felixhekhorn]

The formula looks correct (from what ERN told us that day). Why do you say the x plot seems incorrect?

Almost 10^-7 looks too small for me. Maybe there was a square missing.

can you check against the grids?