implementation of CMS_2JET_8TEV_3D

[andrpie]

I've run into a few issues during the implementation of this dataset that I would like to ask for clarifications on.

• the triple differential cross section is presented in bins of: average of transverse momenta of the two leading jets, half rapidity separation y^* and the boost of the two leading jets y_b. Having looked at the allowed variables for the DIJET process, it seems that only ystar appears there (unless validphys understands pTavg as (pT+pT2)/2). How should I define the independent variables?
• the data is presented in 6 separate sets, one for each value of the pair (ystar, yboost). Initially I thought of aggregating it all together, but then I realised that for each set there is a statistical correlation matrix to be transformed into artificial uncertainties. This means that if all 6 tables are aggregated, there is going to be a total of 122 artificial statistical uncertainties per bin (and they will be uncorrelated across the tables). Should I stick to the aggregation, or is it more sensible to implement 6 separate observables?

Thank you for your help in advance @enocera!

[andrpie]

@enocera may I ask for advice on the above? It would be great to finish this implementation before the end of 2025.

[enocera]

Dear @andrpie thanks for your patience. Here are my suggestions.

1. I would say that, for the commondata implementation, the relevant variables to implement are the two rapidities (y* and y_b) and the average pT of the two jets. As of how validphys is going to interpret them, let me say two things. a - For the qualitative x-Q2 plot we need the two rapidities and the average pT, which work exactly as any rapidity and, say, the invariant mass of the dijet pair when you want to determine the LO values of x1, x2 and Q2. b - You can in principle alter the relevant functions in https://github.com/NNPDF/nnpdf/blob/master/nnpdf_data/nnpdf_data/process_options.py. To get a better idea of the implementation, you may want to have a look at the companion PR on CMS 3D dijets at 13 TeV CMS_2JET_13TEV #2361. My recommendation, however, is to make sure to have the kinematics.yaml done, I'll take care of the rest, if need be.
2. Concerning the implementation of the data sets, I suggest to collate all the rapidity bins into a single data set. In order to generate artificial systematic uncertainties, I suggest to construct first a block matrix in which blocks are represented by the covariance matrices for the individual rapidity bins and zeroes are put everywhere else. You can then make the outer decomposition of this matrix and attach the 122 artificial uncertainties to each data point, on top of the correlated systematic uncertainties. Alternatively, you could implement each rapidity bin as a separate observable, however you will need to correlate systematic uncertainties across different rapidity bins. I discourage this second option.