Propagate montecarlo uncertainty from fastnlo grids

[scarlehoff]

In the fastnlo grid that we download from plougshare, at least for the dijet data, the MC uncertainties are already included, e.g.,

~$ fnlo-tk-yodaout applfast-cms-dijets-v2-fc-fnlo-arxiv-1705.02628-xsec000.tab.gz NNPDF40_nnlo_as_01180 ST NNLO

#=================================================================================
 # Relative statistical uncertainties (ST)
 #---------------------------------------------------------------------------------
 # bin      cross_section           lower_uncertainty       upper_uncertainty
 #---------------------------------------------------------------------------------
    1       1.17712104601E+03      -8.93967724580E-03       8.93967724580E-03
    2       5.88123036467E+02      -7.37521019135E-03       7.37521019135E-03
    3       3.06752769868E+02      -8.68429710285E-03       8.68429710285E-03
    4       1.65291105626E+02      -6.68794255972E-03       6.68794255972E-03
    5       8.94284730323E+01      -9.92508174274E-03       9.92508174274E-03
    6       4.92202483062E+01      -7.94142733975E-03       7.94142733975E-03
    7       2.92246985763E+01      -1.37335993342E-02       1.37335993342E-02
    8       1.61952732333E+01      -1.20941659544E-02       1.20941659544E-02
    9       9.50513815304E+00      -7.51627623106E-03       7.51627623106E-03
   10       5.59814170314E+00      -8.88502884738E-03       8.88502884738E-03
.
.
.

This information should be propagated to the grid. If not directly by pineappl (not sure whether it can be added, @Radonirinaunimi?) the pinefarm script should run the command above, parse the uncertainty and add it to the grid as metadata.

I think we will need to take it into account if we want to go towards the pt tail. For this grid for instance, already for the 10th bin the MC uncertainty on the NNLO coefficient is about 30%:

 ##################################################################################
 # fnlo-tk-yodaout: Evaluating uncertainties
 ##################################################################################
 #=================================================================================
 # Relative statistical uncertainties (ST)
 #---------------------------------------------------------------------------------
 # bin      cross_section           lower_uncertainty       upper_uncertainty
 #---------------------------------------------------------------------------------
    1       3.71518823259E+01      -2.82864000000E-01       2.82864000000E-01
    2       2.36089186045E+01      -1.83394100000E-01       1.83394100000E-01
    3       1.37816639030E+01      -1.93069790000E-01       1.93069790000E-01
    4       9.54858780265E+00      -1.15584010000E-01       1.15584010000E-01
    5       4.71713438802E+00      -1.88037690000E-01       1.88037690000E-01
    6       2.14050531455E+00      -1.82440820000E-01       1.82440820000E-01
    7       2.57801390436E+00      -1.55641770000E-01       1.55641770000E-01
    8       8.61484910806E-01      -2.27279070000E-01       2.27279070000E-01
    9       6.06435811476E-01      -1.17713530000E-01       1.17713530000E-01
   10       3.65014957690E-01      -1.36194550000E-01       1.36194550000E-01
   11       2.61725375041E-01      -8.91347580000E-02       8.91347580000E-02
   12       1.43491842804E-01      -8.24604230000E-02       8.24604230000E-02
   13       9.55372317654E-02      -7.67335580000E-02       7.67335580000E-02
   14       4.35619442853E-02      -1.42809410000E-01       1.42809410000E-01

cc @andrpie @enocera

[enocera]

This means that EWK K-factors, that at high pT are of the order of a few percent, are a complete waste of time...

[scarlehoff]

I haven't looked at the bin-kinematics correspondance, but basically yes:

$ fnlo-tk-yodaout applfast-cms-dijets-v2-fc-fnlo-arxiv-1705.02628-xsec004.tab.gz NNPDF40_nnlo_as_01180 ST NNLO

 ##################################################################################
 # fnlo-tk-yodaout: Evaluating uncertainties
 ##################################################################################
 #=================================================================================
 # Relative statistical uncertainties (ST)
 #---------------------------------------------------------------------------------
 # bin      cross_section           lower_uncertainty       upper_uncertainty
 #---------------------------------------------------------------------------------
    1       1.91960607302E+02      -1.61620047427E-02       1.61620047427E-02
    2       7.65669429259E+01      -1.64663168627E-02       1.64663168627E-02
    3       3.63202842585E+01      -1.81552180674E-02       1.81552180674E-02
    4       1.54034210732E+01      -2.29719974305E-02       2.29719974305E-02
    5       7.27211828009E+00      -1.81247885808E-02       1.81247885808E-02
    6       3.08760541660E+00      -1.57620098025E-02       1.57620098025E-02
    7       1.31197434727E+00      -1.89447104339E-02       1.89447104339E-02
    8       5.79699995027E-01      -1.98820861023E-02       1.98820861023E-02
    9       2.31995097409E-01      -2.55205409349E-02       2.55205409349E-02
   10       1.01824096210E-01      -1.95990161797E-02       1.95990161797E-02
   11       3.93070321213E-02      -2.60511435855E-02       2.60511435855E-02
   12       1.52022156172E-02      -2.71006097128E-02       2.71006097128E-02
   13       5.61651518170E-03      -2.70353005633E-02       2.70353005633E-02
   14       1.81006637967E-03      -3.43360140140E-02       3.43360140140E-02
   15       6.05505504673E-04      -3.65114350132E-02       3.65114350132E-02
   16       1.55975856902E-04      -3.83391337545E-02       3.83391337545E-02
   17       4.51362648531E-05      -4.39236857396E-02       4.39236857396E-02

Beyond the 8th the error on the total cross section is getting close to 20%(which makes sense, the error on the NNLO xs is about 200%)

    8       6.88182118526E-02      -1.67014160000E-01       1.67014160000E-01

But to be completely fair, the cross section is also 1e3 from the peak.