Open cschwan opened 2 years ago
It seems the problem is more complicated than I initially thought. For the CDF measurements (proton + anti-proton) the table that displayed at the end shows wrong results for PineAPPL, although the lines shown above should be correct.
Wait, I don't understand: in ŧable.py
we're only calling convolute_with_one
(never convolute_with_two
) and initial_state_1
and initial_state_2
are never accessed.
So, unless you're testing a locally upgraded version, I still believe the OP is right, and we should first try in that direction. Isn't it? (if the grid have not been conjugated in place of the PDF)
@AleCandido CDF measurements should be working (proton + anti-protons) should be working, but what I see is they don't. Could you try to reproduce that behaviour? Try the CDF_WMASS
branch and pick any distribution and reduce the statistics so that it runs faster.
I'll do, definitely, but according to me the problem is really in the convolution: https://github.com/NNPDF/runcards/blob/b4da9bb8ddee30ed56d5ae38d81fae9450b81286/runcardsrunner/table.py#L33-L35
As you see, pineappl_results
are only computed using convolute_with_one
, and the information about initial states is hard-coded in the 2212
. So there is really no chance to get a correct prediction (unless you transformed the grid to convolute with two initial protons).
I think I found the culprit: https://github.com/N3PDF/pineappl/commit/64a1b30c96f0b5e5ddac34aa7fa830652ca83940.
Commit bb3fb45fd640482a61f9f64b7773c601bb5dcea8 should fix the problem.
Anything other than protons yields a wrong convolution:
https://github.com/NNPDF/runcards/blob/master/runcardsrunner/table.py#L33-L35
This should read out the metadata
initial_state_1
andinitial_state_2
and callGrid::convolute_with_two
if the initial states are different.