[Rivet] OPAL_2004_S6132243

[Andy Buckley]

Hi again Peter,

I've put in some debug printout and found that the first of several 
problem events happens on event 53360. In this event, the hemispheres 
projection returns a negative mass^2_L value (~10^-9 GeV, so it's at the 
numerical precision level), and so the sqrt of that introduces the NaNs. 
I'm not yet sure why the "normal" histograms tolerate being filled with 
a NaN but the moments profiles die, but it's certainly not good because 
4 vector calculations shouldn't be *able* to return negative mass2s! I'm 
looking into that now: will hopefully get it sorted this evening and be 
able to release the new version.

On a more physics-oriented note, I wonder if there is an undeclared 4 
jet (or at least 4 particle) cut on filling this distribution: it looks 
to me like there is an excess of entries in the zero bin of the main 
distribution. But I'm struggling to see how in a 4pi detector in the CoM 
frame it's even possible for a sphericity- or thrust-based hemispheres 
decomposition to ever have no particles in one of the hemispheres: the 
momentum has to be balanced somewhere.

Andy


On 11/11/11 07:16, Peter Skands wrote:
> Hi Andy,
>
> I was just wondering. ML is normally a 4-jet quantity, i.e., it is
> undefined (zero) for 3 particles or less. It may even be undefined for 4
> partons, since it needs at least 2 of the partons to be on the "light"
> hemisphere which is not always the case. I don't know exactly how you
> compute the moments, but is it possible that something bad happens if
> there is an event with too few particles in the light hemisphere? That
> could conceivably happen only very rarely.
>
> Peter
>
> On 11/10/11 11:22 PM, Anton Karneyeu wrote:
>> Hi Andy,
>> I made a recheck with 1.7.0beta0 and still can see the problem - see
>> attached
>> pdf. Did you tried on lxplus?
>>
>> The full output which I see and script is here:
>>
>> /afs/cern.ch/user/a/akorneev/public/rivet
>>
>> Maybe something is wrong with my environment but I see the same issue in
>> histograms coming from our BOINC cluster which are SLC5 i686 machines
>> with
>> minimal environment. Also, problem do not appear running small number
>> of events,
>> at least 100k events are necessary (probably it depends on random seed).
>>
>> Cheers,
>> Anton
>>
>>
>> On 10.11.2011 16:53, Andy Buckley wrote:
>>> Hi Anton,
>>>
>>> Sorry about the delay, but with the trunk version of Rivet I don't see
>>> any issue with these plots. I ran the PYTHIA default tune at 91 GeV and
>>> your z2.params which specifies 133 GeV, and both worked fine. No NaNs in
>>> any of the output files, and the plots look very reasonable (see
>>> attached).
>>>
>>> Do you still see this issue with the 1.7.0beta0 version? We do now also
>>> have a beta1 tarball available, but unless you found any problems in the
>>> beta0 testing I think we are ready to make our proper 1.7.0 release
>>> tomorrow.
>>>
>>> Cheers,
>>> Andy
>>>
>>>
>>> On 06/10/11 21:19, Anton Karneyeu wrote:
>>>> Hi guys,
>>>>
>>>> could you have a look on OPAL_2004_S6132243 analysis there is a problem
>>>> with HemiMassLMom (d25-x01-yYY) histogram. In some cases histogram bins
>>>> contents NaNs.
>>>>
>>>> Please, find in attachment the script which reproduce the problem on
>>>> lxplus:
>>>>
>>>> $ ssh lxplus
>>>> $ ./bug.sh
>>>>
>>>> Problematic histogram is
>>>>
>>>> S6132243/OPAL_2004_S6132243/d25-x01-y02.{dat,pdf,png}
>>>>
>>>>
>>>> Cheers,
>>>> Anton
>>>>
>>>>
>>>> _______________________________________________
>>>> Rivet mailing list
>>>> Rivet at projects.hepforge.org
>>>> http://www.hepforge.org/lists/listinfo/rivet
>>>
>>>
>


-- 
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.