[Rivet] Event subsample analysis

[Frank Siegert]

Hi Riveters,

I have been playing around with an idea which I'd like to get some input 
for. I would like to be able to run the analysis separately for different 
subsamples of the events (and of course one total for all events). Use 
case: To better understand a ME+PS merging sample one typically wants to 
see the sub contributions from events where the matrix element generated N 
jets, which then should add up to the total in histograms.
To test without too much steering work I have used a direct Sherpa-Rivet 
interface which initialises multiple Rivet analysis handlers (kudos! to 
the clean static-less structure that allows me to do this). Each of them 
then gets a different event sample to analyze, in this example according 
to ME jet multiplicity.

For "normal" histograms, i.e. the ones which are not scaled using an 
internal event counter, there is only one small addition needed, to make 
this work. AnalysisHandler needs to know the total sum of weights (instead 
of using the one that it has counted from the events passed to it). This 
will make things like
  scale(hist, crossSection()/sumOfWeights());
just work.

For histograms which are scaled using an internal event counter, e.g.
  scale(hist, 1.0/_sum_of_weights_Z_inclusive)
or even normalised to a fixed number,
  normalize(hist, 1.0)
the result will look different, the subhistograms don't add up to the 
total anymore. I don't see any way around this and would argue, that in 
such cases the only conclusion one can/should draw from the subsample 
histos is about their shape.

Are there any comments or objections to this feature? For the time being, 
I would only like to add a setSumOfWeights() method to the 
AnalysisHandler, so that I can use this feature from my external 
interface. Do we want to make it available through bin/rivet as well, can 
somebody think of a use case where this would be useful? I guess if the 
generator is able to produce separate event samples for the different jet 
multiplicities (Alpgen?) this might be usable. But I myself would only 
want to use it through the comfort of my direct interface, so I don't 
care.

Frank