Spectral Clustering for Jet Reconstruction

TL;DR

This paper introduces a spectral clustering method for jet reconstruction in particle physics, which directly uses kinematic data and offers consistent parameter settings across different processes, potentially improving analysis robustness.

Contribution

The paper presents a novel spectral clustering approach for jet reconstruction that is IR-safe and does not require parameter tuning for different physics processes.

Findings

Spectral clustering achieves comparable or better jet reconstruction performance than anti-$k_T$.

The method maintains consistent parameters across various processes.

Spectral clustering is IR-safe and effective in complex event topologies.

Abstract

We present a new approach to jet definition alternative to clustering methods, such as the anti-$k_T$ scheme, that exploit kinematic data directly. Instead the new method uses kinematic information to represent the particles in a multidimensional space, as in spectral clustering. After confirming its Infra-Red (IR) safety, we compare its performance in analysing $gg \rightarrow H_{125~\rm GeV} \rightarrow H_{40~\rm GeV}H_{40~\rm GeV} \rightarrow b\bar{b}b\bar{b}$, $gg \rightarrow H_{500~\rm GeV} \rightarrow H_{125~\rm GeV}H_{125~\rm GeV} \rightarrow b\bar{b}b\bar{b}$ and $gg, q\bar{q} \rightarrow t\bar{t} \rightarrow b\bar{b}W^+W^- \rightarrow b\bar{b}jjl\nu l$ events from Monte Carlo (MC) samples, specifically, in reconstructing the relevant final states, to that of the anti-$k_T$ algorithm. Finally, we show that the results for spectral clustering are obtained without any change in the parameter settings of the algorithm, unlike the anti-$k_T$ case, which requires the cone size to be adjusted to the physics process under study.