[Rivet] Rivet with weighted events of indefinite sign

[Andy Buckley]

On 04/02/13 14:30, Paolo Nason wrote:
> Thanks Andy, but I was not asking about groups of correlated events, as
> they occour in fixed order NLO calculations, but rather about those that
> occour in NLO+MC programs.

Ah, I see: sorry for the misunderstanding.

> At the end I gave up trying to use agile, it seems that it does not
> propagate the event weight to the HepMC file, so I wrote a little
> interface from Pythia6 to HepMC, to be used with POWHEG, using the HepMC
> library to copy the event from the hepevt common block to the HepMC
> format.

Sounds good. I'm surprised that AGILe didn't work, since I added the
code to read in LHE files with HERWIG and PYTHIA together with Simone
Alioli, but must admit that it's some time since I tried it. AGILe is
anyway only a convenience interface for things like command-line
parameter setting: your program sounds perfect.

> I found a little LHEF to HepMC converter, written by you, and I used the
> same method found there to transfer the weights to the HepMC common
> block. I also found that you need to set the cross section, computed as
> the average weight at the given event.

Right -- the xsec handling in that program was also updated based on
exchanges with Simone, so if there are improvements that can be made
then please let me know.

> It seems to work, but it would help if I understood how does Rivet fill
> histograms. I guess it does the following:
> 
> sum of weights falling in a given bin
> ------------------------------------- X sigma_last
>        sum of all weights
> 
> where sigma_last is the cross section entered in the last event.
> 
> 
> That would be the correct thing for me;
> Is this what it does?

Essentially, yes. The error treatment is a bit special for weighted
events but I believe should be correct for negative weights (modulo the
currently missing correlated event group support that needs a bit more
work).

In Rivet the overall normalisation (and other post-event-loop
manipulations) are up to the analysis author, as different experimental
observables may normalise differently. But the "normal" way, which looks
something like normalize(_hist, crossSection()/picobarn) or scale(_hist,
crossSection()/picobarn/sumOfWeights()) in the Rivet analysis code,
should be equivalent to what you've written.

I will be at least partially out of contact in coming weeks due to
parental leave, so I've CC'd the main Rivet developer list (anyway the
better contact point) for further discussion. Do I understand correctly
that the problem was in AGILe (with which shower generator?) rather than
in Rivet, and that with your own POWHEG -> HepMC setup everything looks ok?

Andy



> On Mon, 2013-02-04 at 13:15 +0100, Andy Buckley wrote:
>> On 01/02/13 08:19, Paolo Nason wrote:
>>> Hi Andy,
>>> I am trying to learn how to use Rivet with POWHEG+PYTHIA6 (with Agile).
>>> I have lhe events with variable weights, that can also be negative, and
>>> I don't seem to be able to get a sensible result.
>>>
>>>
>>> Can rivet handle this? I could not find anything clear in the
>>> documentation ...
>>
>> Hi Paolo,
>>
>> It's been requested a lot, but we've not had the resources to implement
>> it: doing this "automatically" requires us to build in a closer
>> connection between the histograms and the event loop so that the
>> "triggering" on same/different event numbers can work automatically. So
>> no, it doesn't work at the moment: the central values should be fine,
>> but the errors will be wrong as they end up showing the huge spread
>> between the normal events and their negative counterterms.
>>
>> Once we get the next release of Rivet out (with new histogramming to
>> allow for proper run merging) we can work on finally solving this
>> problem (and handling of multiple weights per event). If we can get a
>> short term student from MCnet or elsewhere to work on it, it'll go
>> faster: at the moment the Rivet development manpower is < 5% of my time,
>> and similar for others (Hendrik Hoeth, Frank Siegert, David
>> Grellscheid), hence the slow progress of "core" development :(
>>
>> Andy
>>
> 


-- 
Dr Andy Buckley, Royal Society University Research Fellow
Particle Physics Expt Group, University of Edinburgh

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