[Rivet] OPAL_2004_S6132243

[Andy Buckley]

On 11/11/11 23:53, Peter Skands wrote:
> Hi Andy,
>
> Thanks for the speedy reaction. I'm afraid I have not much insight into
> how the original analysis was done. I could imagine that if the whole
> event hadronized into neutral energy, in the extreme case everything
> producing neutrons, the experimental systematics would be extremely hard
> to control. Correct me if I am wrong, but neutrons are basically
> invisible to you? Or do you see something in the EM/HAD calorimeters? Of
> course, it would be *extremely* rare for an entire Z decay to produce
> only two neutrons!

Yep, this entered my mind! Although I didn't then go off and check 
whether this was a charged-only or charged+neutral analysis. Re. neutron 
visibility, they should be seen in the HCAL -- they are pretty much the 
worst-modelled detector interaction, though, and using the Geant4 mode 
that does it a bit more accurately takes several times longer to run!

> But it could happen that no charged particles would
> be produced, and hence the event would consist of only neutral energy,
> say some K0L mesons (that live a long time) and maybe some photons from
> pi0 or eta0 decays. That might still be a situation not under extremely
> good control, and I really have no good feeling for how often it might
> happen, apart from the indication given by the charged-multiplicity
> distribution that says how often there are at least no charged tracks in
> the event. Still, they had pretty good EM calorimeters at LEP I guess,
> so should be able to detect an event reliably even if went to all pi0 ->
> photons, right? A small negative mass squared can be just the result of
> rounding errors. I agree that above the numerical precision, no
> spacelike quantities should be able to appear, but at or around
> numerical precision, rounding is always an issue, and the algorithm
> should be stable against small fluctuations induced by it.

Yep, I agree! In the end I couldn't track down what was going on -- the 
4-vector object was not returning a negative mass2 (good: I've spent 
enough time tweaking precision aspects of that code!), but the 
projection returning the stored result of that operation *was* negative. 
And the guilty event changed when I made a small change in that 
projection -- I suspect that there is some rarely-occurring memory issue 
going on.

Since it's important for us to get this release out now, and this is a 
rare effect -- although I don't like the knowledge that some technical 
aspect is not quite right -- I've put some isnan() protection into the 
OPAL 2004 analysis code to alleviate the problem of the plot completely 
screwing up. We'll return to fix it properly...

Cheers,
Andy

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

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