jet reconstruction performance in single top events

[Fabian Bach]

In a single top s channel sample with partonic final state evbb (50k events, shower/hadronization from Pythia with Perugia2011 tune), I did a matching of the truth and detector level jets in order to examine the reconstruction performance. Following the procedure described in the Delphes '09 paper, the matching condition was to find a reconstructed object inside a cone of R<0.2 around one of the input objects, so that the kinematic reconstruction could be compared. In the ROOT file attached, I looked at absolute angular distance ("deltaR"), difference of angular components ("eta_diff"/"phi_diff") as well as transverse momentum and energy differences normalized to the input value ("Pt_diff"/"E_diff"). I matched the partonic MC b momenta to truth level jets (suffix "_jt") and detector level jets ("_jr"), in each case reconstructed with Delphes' ATLAS default setting anti-kt with R=0.7, cf. detector chart attached. I also matched truth level to detector level jets as was done in the paper ("_jtr"), and would roughly expect a Gaussian around 0 in each kinematic parameter, but I observe an asymmetry in the Pt/E reconstruction tending towards lower values, and also a mean angular distance deltaR peaking not at 0 but at a small non-zero distance. Maybe you could comment on that? I varied the Pythia tune, jet algos (anti-kt/kt) and jet parameter R (0.4, 0.6, 0.7), always with similar results. (For completeness, the ROOT file also contains the matching results for the neutrino momentum reconstructed from the charged lepton and the W mass shell "_nr" and for the missing transverse energy parameters vs. input neutrino momentum "met_abs_diff"/"met_phi_diff".)

[Michele Selvaggi]

Hello,

thanks for the interesting plots. I am not surprised by what you see. Part of the gen level components are lost at the reco level (typically neutrinos). We do not apply any jet energy scale in Delphes (yet) so it is absolutely normal to observe that bias towards low values.
In the next Delphes release we will include an Energy Scale module that allows to correct for this effect as a function of pt,eta. Stay tuned for this.

Also, the DR distribution has a small bias but this looks very much reasonable. The peak you see in DR (and not in Deta and DPhi) is simply the bias due to the requirement that DR is positive (think about a Chi2 distribution). If you increase the binning in the "DR_jt" distribution, I expect you would see the same peak, but simply at a smaller value. If you look at the relative Deta and Dphi they are centered in 0.

Does this answer your questions?

Cheers,

Michele

[Fabian Bach]

Hi Michele,

thanks for the quick answer. I just wanted to make sure that Delphes behaves as intended by the developers in this case, since from reading the validation part of the paper I would have expected a symmetric Pt/E smearing.

Cheers,
Fabian