[Rivet] [Rivet-svn] r2095 - in trunk: data/plotinfo src/Analyses

[Andy Buckley]

Frank Siegert wrote:
> Hendrik Hoeth, Wednesday 25 November 2009:
>> Thus spake Frank Siegert (frank.siegert at durham.ac.uk):
>>> Now I'm completely confused: I thought that's what we agreed not to
>>> do, and weren't you especially in favour of that decision?
>> In this analysis the data is indeed normalised to the cross-section, so
>> following the logic that we should normalise everything to
>>  cross-section where the data is given in terms of cross-section, this
>>  change in the .cc file is correct. That is different from trying to
>> cross-section-normalise event shape variables where the data is
>> normalised to 1, and then normalise the plot to 1 afterwards.
> 
> Ah, now I understand, what you are doing (not that I like it though).
> 
>> Having said that, I still think that's a absolutely stupid thing to
>> normalise this plot to cross-section. Nobody who looks at this plot is
>> interested in its normalisation. Everything people want to see is
>> whether the peak is at the right position and has the right shape. I
>> think it is plain wrong to blindly normalise everything to
>>  cross-section where the data is given in units of cross-section, but
>>  that's been the decision here, hasn't it? And in order not to mess our
>>  poor users too much, I added the two lines to the .plot file to repair
>>  at least the plotted result.
> 
> So count me as a user whose plots are probably going to break, because I 
> would have a line in my make-plots configuration file which sets the 
> correct kfactor for that process.
> If we are doing the change on which we agreed, I think we should do it 
> consistently. Otherwise it's very difficult to interpret plots in the end, 
> because you always have to wonder whether they've already had a kfactor 
> applied or not.

This is exactly my view. I want to remove the fitting to data from all 
Rivet plots, including this one, not because I think that this is a 
particularly good place to look for the Z production cross-section but 
because of consistency: we will just have to get used to passing all 
plots through whichever post-processing scaling we feel is appropriate 
for our *individual* purposes. Almost all Professor tuning will want to 
scale up by a fitted K-factor or to normalise to the reference data, but 
POWHEG validation might want to prove a point by comparing directly to 
the data without any scaling or fitting. It's not really Rivet's role to 
presume how it's going to be used, IMO.

Eike and Holger, what's the status of the normalisation scripts that you 
were writing? Are they ready for production use? Can you post a quick 
description and Frank, Hendrik, David, etc. can try them out and iterate 
the design/functionality/implementation before the 1.2.0 release.

Andy

-- 
Dr Andy Buckley
SUPA Advanced Research Fellow
Particle Physics Experiment Group, University of Edinburgh