[Rivet] CMS dressed top question

[Andy Buckley]

On 12/11/14 09:45, Lars Sonnenschein wrote:
> Hello Andy,

Hi Lars,

Please contact the Rivet list rather than me personally. I've CC'd it
(again ;-) )

> we are still struggling with the first top analysis getting into Rivet.
> In the ttbar dilepton channel we get two oppositely charged leptons and
> two neutrinos we are touching in Rivet to reconstruct the two W bosons.
> 
> Are we allowed to make use of the information if it is a neutrino or an
> anti-neutrino we are dealing with?

It's up to you, but I'd advise against it unless CMS has some
extraordinary subdetector that I'm not aware of which can do detailed
neutrino particle ID ;-)  The experimental analysis needs to resolve the
ambiguity, therefore it's best that the equivalent MC analysis does so, too.

And of course it's actually a worse problem than you say -- you can only
see one missing energy vector, not two separate (anti)neutrino
components within it, so you can't treat the two tops independently.

> This would remove one ambiguity.
> The temporary version of the Rivet code is right now making a choice:
> 
>   |(lepton1+neutrino1).M()-W.M()|
>   + |(lepton2+neutrion2).M()-W.M()|
>   < |(lepton1+neutrino2).M()-W.M()|
>   + |(lepton2+neutrion1).M()-W.M()|
> 
> for the pair which matches better the W boson mass without exploiting
> neutrino/anti-neutrino information.

Again, how can you experimentally determine the momentum vectors of
multiple neutrinos in a single event? If you think you can do that
safely, then being able to tell whether they are nu's or anti-nu's is
not a big leap. The important thing to think about is whether you can
safely use this sort of experimentally unavailable truth information at
all -- naively the answer is no.

The (only?) situation in which it *is* appropriate is where the data has
been unfolded, using MC models and adding significant modelling
systematics, to an MC analysis which used this full information. Then
you would be writing a Rivet analysis using identical logic to the MC
truth analysis that went into the data analysis. That is fine, but doing
the data analysis that way (i.e. relying on unfolding to perfectly deal
with an intrinsic experimental analysis ambiguity) is rather
unsatisfactory and I suspect is not the case.

Andy

-- 
Dr Andy Buckley, Royal Society University Research Fellow
Particle Physics Expt Group, University of Glasgow / PH Dept, CERN