[Rivet] Validation meeting

Andy Buckley andy.buckley at durham.ac.uk
Thu Nov 27 19:20:13 GMT 2008


James Monk wrote:
> On 27 Nov 2008, at 17:00, Hendrik Hoeth wrote:
> 
>> I guess that's bad news if you want to run within the ATLAS framework?
>> Anyhow: With a single run without CKIN(3) cut you are still able to  
>> use
>> some of the plots.
> 
> The Atlas setup won't use any of the RivetGun params files, since the  
> generator is steered by Athena.  I could certainly run several jobs  
> for different ckin(3) and ckin(4) but then the problem becomes merging  
> them properly.  I think you need the dsigma/dPT of the leading jet in  
> order to know the contribution from each plot.  Might it be worth  
> adding that plot to the analysis in future for that reason?

You can merge without having to track the relative normalisation of each
run if you're plotting profile histograms binned in pT or Et. Anything
else requires more information, which is a driving force behind the YODA
design.

> If I remember correctly, the data is split up into a min bias sample  
> and a PT>20GeV sample anyway.  So I could take the min bias plots from  
> a low ckin(3) sample and the pt 20 plots from a ckin(3) around 15 or  
> so.

That's exactly what I meant by the two runs. We use three for this
analysis, and a lot more for the CDF 2004 and CDF 2008 UE jet analyses,
but where the histograms match one-to-one from run -> plot, then it's easy.

> Running in RTT I think we'll be limited to ~100k events (I think  
> there are ~10 cpus spread over all the RTT jobs over 24 hours).  Of  
> course we can have as many as we like from our own batch system.

For producing these plots, you'll need more --- as I said before we use
1M per run (== ME pT cut). So around 2M or more to validate this
analysis "properly". For RTT testing fewer results may be okay, but it
would be nicer to run more: 100k events won't cut it for validating
against the more recent UE data. Our runs took over 30 hours for the C++
generators: it needs to be considered for the future of RTT validation
that meaningful automated testing may not be possible on a daily basis.

Andy


More information about the Rivet mailing list