Unweighting multijet event generation using factorisation-aware neural networks

TL;DR

This paper introduces a neural network-based method to significantly accelerate the generation of unweighted multijet events in high-energy physics simulations, maintaining accuracy for complex LHC processes.

Contribution

It combines factorisation-aware neural network emulation with an unbiased unweighting algorithm to drastically reduce computational costs in multijet event generation.

Findings

Achieved a 16 to 350-fold reduction in computational cost for event generation.

Successfully emulated QCD multijet matrix elements including initial state and massive quarks.

Demonstrated unbiased event generation suitable for phenomenological and experimental analyses.

Abstract

In this article we combine a recently proposed method for factorisation-aware matrix element surrogates with an unbiased unweighting algorithm. We show that employing a sophisticated neural network emulation of QCD multijet matrix elements based on dipole factorisation can lead to a drastic acceleration of unweighted event generation. We train neural networks for a selection of partonic channels contributing at the tree-level to $Z+4,5$ jets and $t\bar{t}+3,4$ jets production at the LHC which necessitates a generalisation of the dipole emulation model to include initial state partons as well as massive final state quarks. We also present first steps towards the emulation of colour-sampled amplitudes. We incorporate these emulations as fast and accurate surrogates in a two-stage rejection sampling algorithm within the Sherpa Monte Carlo that yields unbiased unweighted events suitable for phenomenological analyses and post-processing in experimental workflows, e.g. as input to a time-consuming detector simulation. For the computational cost of unweighted events we achieve a reduction by factors between $16$ and $350$ for the considered channels.