[Rivet] [Fwd: Re: Combining generator runs]

[Andy Buckley]

For general interest, I'm forwarding below a snippet from an email 
conversation with Eric Feng about generating weighted events from 
Fortran Herwig & Pythia with a flatt(er) pT sampling cf. the Sherpa (and 
ThePEG, apparently) enhancement function. This could be used as a 
simpler replacement for our many kinematic subsample choppings in UE 
analyses for Rivet/gen validation and particularly generator tunings. 
Anyone have any experience of these methods?

Eric reports success with Pythia (the cross-section is fairly stable 
across multiple independent 5000 event runs), but for his jet events the 
cross-sections in the "equivalent" Fortran Herwig samples varied by 
orders of magnitude... doesn't sound like success to me. (I think he's 
asked you about this, Jon: anything to add?)

Andy


-------- Original Message --------
Subject: Re: Combining generator runs
Date: Sat, 26 Sep 2009 11:00:21 +0200
From: Eric Feng <Eric.Feng at cern.ch>
To: Andy Buckley <andy.buckley at ed.ac.uk>
References: <4AB8915F.2030306 at cern.ch> <4AB9059D.4060004 at durham.ac.uk> 
<4AB91142.1080309 at cern.ch> <4ABB8A96.2090004 at ed.ac.uk>

Hi Andy,

> Cool, I thought that couldn't be done in either of those generators. Out
> of interest, what's the setup for this? We could make good use of that
> in our Professor tunes, Rivet validation, etc. etc.!

In Pythia, to generate with a weighting function, one can do this with
PYEVWT.  For convenience I will just point to this example in the interface:
https://svnweb.cern.ch/trac/atlasoff/browser/Generators/Pythia_i/trunk/src/PythiaDummies/pyevwt.F

As discussed in the example, one can construct a weighting function,
e.g. that is ~proportional to the cross-section in ptHat, then set WTXS
= 1/CROSSSECTION.

Then one needs to specify MSTP(142)=1 to generate weighted events.  And
histograms should be filled with the weight PARI(10), and normalize by
PARI(2) which is the total cross-section divided by sum of weights.

For Herwig, one can just set PTPOW, which specifies the power for jet
sampling as \frac{1}{ p_T^{PTPOW} }.  The default power is 4, which I
set to 0 to get a flat distribution of events in pt:
http://webber.home.cern.ch/webber/hw65_manual.html#htoc89

Then I generate weighted events using NOWGT=.FALSE.:
http://webber.home.cern.ch/webber/hw65_manual.html#htoc97

Lastly, in each case I generate one unbinned sample by setting
CKIN(3)=8, CKIN(4)=4000 for Pythia, and similarly for
PTMIN and PTMAX for Herwig.

In the JetEtMiss group we're investigating use of these weights to
generate weighted samples that are flat in pt, rather than these ugly
piece-wise flat JX samples we have now.