Monte Carlo tuning and generator validation

TL;DR

This work addresses the challenge of tuning Monte Carlo event generators to describe a broad suite of collider observables at LHC energies. It introduces a systematic tuning approach (Professor) that parameterizes generator responses bin-by-bin with second-order polynomials and uses a pseudoinverse fit, coupled with Rivet/HepData-based data comparison. Applying this framework to Pythia 6.4, it yields new tunes based on LEP/SLD e+e- data and Tevatron underlying-event measurements, including a three-stage parameter-factoring strategy and separate tunes for virtuality- and pT-ordered showers; these tunes are comparable to Peter Skands' Perugia tunes. The work provides ready-to-use tuned configurations through PYTUNE and advocates collaboration with generator authors to ensure reliable predictions for LHC analyses.

Abstract

We present the Monte Carlo generator tuning strategy followed, and the tools developed, by the MCnet CEDAR project. We also present new tuning results for the Pythia 6.4 event generator which are based on event shape and hadronisation observables from e+e- experiments, and on underlying event and minimum bias data from the Tevatron. Our new tunes are compared to existing tunes and to Peter Skands' new "Perugia" tunes.

Figures (7)

• Figure 1: Some example distributions for $e^+e^-$ collisions using the virtuality-ordered shower. The solid line shows the new tune, the dashed line is the default. Even though the virtuality-ordered shower is well-tested and Pythia has been tuned several times, especially by the LEP collaborations, there is still room for improvement in the default settings. Note the different scale in the ratio plot of the rapidity distribution. The data in these plots has been published by Delphi Abreu:1996nadelphi-2002.
• Figure 2: Some example distributions for $e^+e^-$ collisions using the $p_\perp$-ordered shower. The solid line shows the new tune, the dashed line is the old recommendation for using the $p_\perp$-ordered shower (i. e. changing $\Lambda_\text{QCD}$ to 0.23), and the dashed-dotted line is produced by switching on the $p_\perp$-ordered shower leaving everything else at its default. The latter is the unfortunate choice made for the ATLAS-tune. The data has been published by Delphi Abreu:1996nadelphi-2002.
• Figure 3: The upper plots show the $Z$$p_\perp$ distribution as measured by CDF Affolder:1999jh compared to different tunes of the virtuality-ordered shower with the old MPI model (left) and the $p_\perp$-ordered shower with the interleaved MPI model (right). Except for tune A all tunes describe this observable, and also the fixed version of tune A, called AW, is basically identical to DW. The lower plots show the average track $p_\perp$ as function of the charged multiplicity in minimum bias events Aaltonen:2009ne. This observable is quite sensitive to colour reconnection. Only the recent tunes hit the data here (except for ATLAS).
• Figure 4: These plots show the average charged multiplicity in the toward and transverse regions as function of the leading jet $p_\perp$ in minimum bias events Affolder:2001xt. On the left side tunes of the virtuality-ordered shower with the old MPI model are shown, while on the right side the $p_\perp$-ordered shower with the interleaved MPI model is used. The old model is known to be a bit too "jetty" in the toward region, which can be seen in the first plot. Other than this, all tunes are very similar.
• Figure 5: These plots show the average track $p_\perp$ in the transverse region (top) and the $\sum p_\perp$ density in the transMIN region (bottom) in leading jet events cdf-leadingjet. The new model (on the right) seems do have a slight advantage over the virtuality-ordered shower with the old MPI model shown on the left, both in the turn-on hump and in overall activity.