[Rivet] our use of fastjet

[Andy Buckley]

Hendrik Hoeth wrote:
> Hi,
> 
> I've seen that in the transition to fastjet 2.4 we changed the overlap
> threshold in the CDF cone algorithms from 0.5 to 0.75. The svn log says
> "Using overlap_threshold for midpoint and jetclu, as advised by the
> FastJet authors.", so I guess the reasoning behind this change is the
> following comment in the fastjet code:
> 
>   /// NB: as of version 2.4, the default value for the
>   /// overlap_threshold threshold has been removed, to avoid
>   /// misleading people into using the value of 0.5 without thinking,
>   /// which is known to have adverse effects in high-noise
>   /// environments. A recommended value is 0.75.
> 
> But we shouldn't use the recommended value without thinking, in the same
> way nobody should use the old 0.5 without thinking. Most CDF analyses
> are using 0.5 and our aim is not to do a new analysis, but to reproduce
> whatever the experiments have done. So I think we should stick to the 0.5.

It's actually based on me having an email discussion with Gavin where he 
said that he believed CDF had actually used 0.75... I hadn't seen that 
comment, which makes it sound more like a FastJet opinion on how to make 
the algorithm behave best.

If you are sure that CDF actually used 0.5, then we should change it 
back in Rivet (it was briefly set to 0.5 in SVN, while I waited for 
Gavin's reply) and let the FastJetters know: I've already fed back that 
this "historical metadata" should go into the FastJet manual.

Andy