A Bayesian tune of the Herwig Monte Carlo event generator

TL;DR

This paper introduces a Bayesian-based tuning method for the Herwig Monte Carlo event generator, enabling statistically rigorous parameter optimization and uncertainty propagation using Markov Chain Monte Carlo techniques.

Contribution

It presents a novel Bayesian tuning procedure for Herwig, incorporating experimental correlations and parameter-to-observable uncertainty propagation, improving the robustness of generator calibration.

Findings

Successful application to Herwig7 with LEP data

Quantified effects of experimental correlations

Enhanced uncertainty propagation from parameters to observables

Abstract

The optimisation (tuning) of the free parameters of Monte Carlo event generators by comparing their predictions with data is important since the simulations are used to calculate experimental efficiency and acceptance corrections, or provide predictions for signatures of hypothetical new processes in experiments. We present a tuning procedure that is based on Bayesian reasoning and that allows for a proper statistical interpretation of the results. The parameter space is fully explored using Markov Chain Monte Carlo. We apply the tuning procedure to the Herwig7 event generator with both the cluster and the string hadronization models and a large set of measurements from hadronic Z-boson decays produced at LEP in $e^{+}e^{-}$ collisions. Furthermore, we introduce a coherent propagation of uncertainties from the realm of parameters to the realm of observables and we show the effects of including experimental correlations of the measurements. To allow comparison with the approaches of other groups, we repeat the tuning considering weights for individual measurements.