dataset excluded from NNPDF4.0: CMS_1JET_7TEV_R07

[andrpie]

This is the initial implementation.

• the observable chosen is inclusive jet cross section with jet radius parameter R=0.7. There is another observable available, the same thing but with R=0.5. Should I implement both?
• for each bin, there is a nonperturbative correction factor that ensures a data-theory agreement. Should these also be implemented?

[enocera]

This is the initial implementation.

* the observable chosen is inclusive jet cross section with jet radius parameter R=0.7. There is another observable available, the same thing but with R=0.5. Should I implement both?

* for each bin, there is a nonperturbative correction factor that ensures a data-theory agreement. Should these also be implemented?

Dear @andrpie let me try to clarify some delicate points that we need to keep in mind with this data set.

1. We want to implement the most recent measurement arXiv:1406.0324, consistently with what we did in NNPDF4.0. The measurement is available for values of the jet radius of 0.5 and 0.7.
2. The NNLO ploughshare grids are unfortunately not available for this measurement. They are, however, for the previous measurement arXiv:1212.6660, albeit only for R=0.7 and for a slightly different binning.
   The two measurements are based on an analysis of the same rawdata (same data taking period, same luminosity). The kinematic binning of the two measurements overlaps, albeit not exactly: 1406.0324 has more bins at low pT (that we would cut anyways), whereas 1212.6660 has a couple more bins at large pT.

Because of 1) and 2), the idea is to implement only the most recent measurement with R=0.7 for NNPDF4.1. There are therefore three tasks.

• Implement the data set in commondata
• Take the relevant grids from plougshare and convert them to PineAPPL grids
• Make sure to take the bins that are the intersection of those in 1212.6660 and in 1406.0324 and match them to the points available from the grids.

As soon as you're done with the first two points, I can help with the third.

[achiefa]

Hi @enocera, thanks for your comment. I can also help if needed.

[andrpie]

Dear @enocera,

Thank you for the instructions. I have implemented the measurement in commondata as advised: it is the arXiv:1406.0324 R=0.7 double differential cross section. Then I also looked at the bins of the corresponding measurement in arXiv:1212.6660, and only left the ones that appeared in both (intersection). So point 1 can be considered done.

Regarding the grids, I have downloaded them and tried to perform the conversion to pineappl format. I ran into the following error:

thread 'main' (254931) panicked at pineappl_cli/src/import/fastnlo.rs:441:29: assertion `left == right` failed left: 1 right: 0

@Radonirinaunimi have you encountered this before?

@achiefa was able to reproduce this issue for applgrid format (for me, it happens with fastNLO). I suppose we can't proceed without fixing this problem, as the next step is to compare the bins in the grids to the ones in commondata.

[Radonirinaunimi]

Hi @andrpie, I just opened an issue in NNPDF/pineappl#383 and will look into it asap.

[andrpie]

Thank you @Radonirinaunimi for fixing the pineappl issue!

@enocera, I just converted the grids and removed the bins that were outside of the new measurement binning. Please have a look at data-theory comparison below (at the grid level, not FK table level, using NNPDF40_nnlo_as_01180).

data-theory plots

The experimental chi2 is 0.398. I suppose this is due to large uncertainties.

[Radonirinaunimi]

Thank you @Radonirinaunimi for fixing the pineappl issue!

Btw, I don't think NNPDF/pineappl#387 will be merged very soon so if you have grids that should be converted, I could do so in the meantime.

[andrpie]

Dear @enocera,

A way of proceeding is the following. You take the luminosity uncertainty (if I'm not wrong it's 2.5% for 7 TeV measurements), you construct a 100% correlated "luminosity" covariance matrix (i.e. the matrix that you get from the luminosity uncertainty only), then you subtract this "luminosity" covariance matrix from the total covariance matrix and you get a "difference" covariance matrix. You can then decompose this into Ndat artificial systematic uncertainties. You have then to separately list the luminosity uncertainty as MULT CORR.

I've implemented this with the luminosity uncertainty of 2.2% (cited in various 7TeV measurements). Unfortunately, the resulting "difference" matrix is not positive definite; all but one of its eigenvalues is negative, and that one negative eigenvalue equals -290. I've constructed the artificial uncertainties anyway (populating the "artificial" matrix with zeros in places corresponding to the negative eigenvalue), and the resulting chi2 and data-theory comparison are pretty much the same. Shall we stick to this implementation or is there a more clever way about it?

[andrpie]

@enocera the experimentalists say that the luminosity uncertainty is indeed included in the combined systematics. As per the HEPDATA entry:

[enocera]

Thanks for checking @andrpie . With that, my suggestion is to forget about the subtraction strategy and variant. Thanks.