ttH
- The K-factor should not go below 0.5 : it is true that this matches better the fact the total uncertainties rise only slowly with pTH.However this may not be fully reliable with the current MC so probably more realistic to not suppress the migration effects too much => Stick to K=0.5.
- Parton shower uncertainty should not be included in the STXS normalization uncertainty scheme : pTH is a very inclusive quantity, so there are no intrinsic resummation effects, and therefore the residual PS effects can be safely considered as covered by the STXS uncertainties. However there should be an acceptance uncertainty associated with the PS (on the BDT shapes, etc.) : in general the STXS uncertainties are not sufficient or meant to account for acceptance/shape uncertainties within a given bin, so other uncertainties should be included to cover this.
- The inclusive uncertainty used in the scheme is the average of the up/down variations (-5.8% +9.2%). Probably more realistic to use the larger value as estimate as this is typically due to the lower one being too low.
- The uncertainty on the first bin is artificially high due to the fact that the spectrum peaks in the second bin: so computing relative uncertainty from the second bin leads to an uncertainty size that's very large relatively to the first bin. Could be better to estimate the uncertainty from the first bin in this particular case (i.e. take the uncertainty always from the bin with the smaller yield, instead of always from the one on the left).
ttH Update
- The numbers were updated during the meeting to address points 3. and 4. Updated numbers show as expected a more sensible uncertainty in the first bin, partially compensated in other bins by the increase of the inclusive uncertainty.
- The K=0.5 case can still be seen as flattening the uncertainty curve a bit too much, by suppressing in particular the migrations at 120 GeV and 200 GeV. So taking K=0.7 or 0.75 as the baseline is probably the better choice.
EW qqH
- Variations in μR,F in PS seem to almost cover the differences between Pythia and Herwig, so this seems understood -- large effect, but makes sense given the computation order.
- The Delta_25 uncertainty, currently from FO NNLO, could perhaps be more sensible to get from HJets++ : the 25 GeV cut falls in the resummation region, so one should at least check the effect of shower uncertainties, and then decide based on this which value to use. At the moment the HJets value in the table on p.11 is essentially coming from stat fluctuations since shower uncertainties are not included, but could re-run with those included to get a reliable uncertainty value.
- The EW uncertainty scheme (separate NPs for the Sudakov and photon-induced parts) seems sensible but we should try to get the opinion of some EW-correction experts before concluding on this
- The tool can already be used as it is now, although the numbers could change slightly (e.g. from the outcome o point 2 above)
VH
- Would be interesting to see if some error bands can be computed around the pTj preductions to see if this covers the difference between PowHeg and Geneva. Not possible in PowHeg, but can be done with Geneva. However may not cover all the difference since the uncertainty should be smaller than the one from PowHeg .
- Current numbers don't use the full available statistics. Running with full stats could help get rid of the unphysical behavior in the >400 GeV bins (e.g. on p.10).
- Uncertainties are currently derived from the K=0.5 scheme, not subtraction in quadrature. For PowHeg this makes sense, since the uncertainty in the first bin is larger than the second, so cannot subtract; and it may not make sense to apply subtraction in higher bins if this isn't done for the first bin already. This feature could be stat fluctuations, but more likely a true effect given the large statistics in this bin. It would be interesting to check if this is specific to PowHeg : if the uncertainties in Geneva increase monotonically, one could consider using subtraction in quadrature (if statistics allow)
- One should check at what order gg->ZH is implemented in Geneva. If this at least matches PowHeg, one could update the uncertainty scheme to use these numbers instead of those from PowHeg, since the decrease in Njet bin uncertainties seems well understood.
There are minutes attached to this event.
Show them.