[Rivet] Question on normalisation of histograms containing weighted events

[Andy Buckley]

On 17/07/12 18:58, Holger Schulz wrote:
> On 17/07/12 19:47, Andy Buckley wrote:
>> A very quick reply...
>>
>> This plot is marked as thrust, but the shape suggests to me that it's
>> actually 1-T, i.e. the perfect back-to-back dijet config is at 0 rather
>> than 1: is that correct?
> No, it's (transverse) thrust, calculated for all charged particles
> in Z->ee events (minus the leptons from the Z-decay).

Oops! See, I told you it was a very reply ;) So actually the "glued"
distribution is not transverse-circular enough?

>> If so, then it looks like your glued distribution is more spherical than
>> the true inclusive distribution, which would imply that too much weight
>> is going on high #MPI, or too little on #low MPI. So, some questions:
>>
>> * have you included a #MPI = 0 "bin"?
> Yes.

Damn, that would have been a neat answer :P

>> * I think it's worth double checking that everything is consistent in
>> the normalisation, since Rivet now includes the overflow and underflow
>> bins in calls to normalize()
> Will do.
> 
> Any feeling if the normalistion kind of destroys information
> from the weighted events? The #MPI histo is also filled using the
> weights, so weight information should be recoverable, right?
>> * double-check that there's no bin-width silliness going on!

The normalisation does destroy information, but the correct bin heights
should be recoverable by scaling by the recorded sum(w) for each #MPI.
What probably won't be correct without some careful handling (or the new
YODA-based Rivet trunk!) is the errors.

It's possible that some weight is being lost because the normalisation
and rescaling treat overflows differently... if there *are* overflows.

>> Finally, just a comment that I don't think #MPI is not really a physical
>> variable, but more an artefact of the modelling with multiple
>> perturbative scatters. Is the analysis attempting to *measure*<#MPI>,
>> or is that just an intermediate thing that you're e.g. reweighting?
> The analysis is going to measure event shapes, as a bonus we
> would like to calculate (the model-dependent) sigma_eff
> which is possible by means of a template fit and the knowledge
> of the model-dependent 2->2 MPI cross-section up to our process-scale,
> m(Z).

Sounds good! Just checking ;) I'll stop here, since further discussion
would be ATLAS-centric/confidential!

Andy


>> On 17/07/12 18:33, Holger Schulz wrote:
>>> Hi,
>>>
>>> I am working on a template fit for some event shape
>>> distributions. In each event, I ask for the number of MPI scatters
>>> and thus get the distributions differential in #MPI.
>>> Those distributions are not normalised in finalize().
>>>
>>> I also obtain a histogram with the occurences of
>>> the number of MPI-scatters, which I normalise to one.
>>>
>>> Further I fill a histogram that contains the inclusive
>>> distribution, which I normalise to 1.
>>>
>>> When I try to "glue" the differential distributions
>>> together, using the relative #MPI contributions
>>> and afterwards scale the glued distribution to 1,
>>> I do not get the same shape as is present in the
>>> inclusive histo (see attachment).
>>>
>>> The event generator I use is Sherpa, generating
>>> weighted events and this is rivet still using the AIDA
>>> histogramming.
>>>
>>> Does anyone have a feeling if the weighted events
>>> and my normalisations do not get along well?
>>>
>>> Cheers,
>>> Holger
>>>
>>>
>>> _______________________________________________
>>> Rivet mailing list
>>> Rivet at projects.hepforge.org
>>> http://www.hepforge.org/lists/listinfo/rivet
>>>
>>
> 
> 


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



The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.