felixhekhorn
commented
1 year ago Note that for
ATLAS_WP_7TEVthe two runs are digit by digit the same and good and this is re-computed, not convertedATLAS_WP_JET_8TEV_PT-atlas-atlas-wjets-arxiv-1711.03296-xsec002the thing went wrong as @andreab1997 said and it is convertedATLASWZRAP36PB-ATLAS-arXiv:1109.5141-Wminus_eta3both versions are fine, but not the same digit by digit and it is converted
so the conjecture by @cschwan seems correct there is something fishy with the APPL conversion; however, keep in mind the grids were (most likely) not re-converted, but they were converted long ago
cschwan
andreab1997
Computing FKs with
pineappl==0.6.a5and comparing their predictions with the predictions that can be obtained from the grids (i.e. usingpineko compare) results in some of the FK predictions to be off by up to a 100% factor. Considering all the hadronic FKs in NNPDF40 datasets, most of them are fine but there are a (not negligible) few which are off.This is the comparison I mentioned
comparison_424.txt
Just skipping the grid optimization in
pineko(i.e.grid.optimize()) the problem disappears.Here is the same comparison but with the new FKs comparison_424.txt
To simplify the debug process, here is the list of the mostly affected dataset:
which of course can be inferred by the comparisons I posted.
@cschwan @AleCandido @felixhekhorn
As far as I know at the moment only me, @niclaurenti and @giacomomagni are computing fktables. I tag you here so that you are aware and you can check your FKs as well.