Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Optimizing b-Jet Performance in the CMS High-Level Trigger with Run-3 Data

Uttiya Sarkar

TL;DR

The paper addresses the challenge of real-time b-jet tagging at the CMS HLT under Run-3 conditions, where tracker aging and high data rates limit older taggers. It adopts ParticleNet@HLT, a graph neural network that treats jets as sets of PF candidates and SVs, trained on simulated AK4 jets with $p_T>30$ GeV and $|5|<2.5$, to output per-jet probabilities for $b$, $c$, $uds$, gluon, and $ au$. The results show a $10$–$15\%$ gain in b-tagging efficiency at fixed mistag rates compared with DeepJet, with stable performance across Run-3 datasets and enabling dedicated $c$- and $ au$-jet triggers. In Run-3 the 2024 HLT menu achieved roughly a $6\%$ improvement in the $4j2b$ trigger efficiency for HH→4b events, at the cost of about a 50 Hz rate increase, illustrating practical trigger gains.

Abstract

The real-time identification and selection of b-jets play a crucial role in the CMS experiment, particularly in searches involving heavy-flavor jets. The High-Level Trigger (HLT) is designed to efficiently select events of interest while maintaining a manageable output rate of a few kilohertz. This report presents the commissioning and performance evaluation of b-jet triggers in the CMS HLT system using proton-proton collision data collected during Run-3 (2022-2024). Key aspects include algorithm optimization, efficiency studies, and comparisons with offline reconstruction. The results provide valuable insights into the current b-jet selection strategy and highlight potential refinements for future data-taking campaigns.

Optimizing b-Jet Performance in the CMS High-Level Trigger with Run-3 Data

TL;DR

The paper addresses the challenge of real-time b-jet tagging at the CMS HLT under Run-3 conditions, where tracker aging and high data rates limit older taggers. It adopts ParticleNet@HLT, a graph neural network that treats jets as sets of PF candidates and SVs, trained on simulated AK4 jets with GeV and , to output per-jet probabilities for , , , gluon, and . The results show a – gain in b-tagging efficiency at fixed mistag rates compared with DeepJet, with stable performance across Run-3 datasets and enabling dedicated - and -jet triggers. In Run-3 the 2024 HLT menu achieved roughly a improvement in the trigger efficiency for HH→4b events, at the cost of about a 50 Hz rate increase, illustrating practical trigger gains.

Abstract

The real-time identification and selection of b-jets play a crucial role in the CMS experiment, particularly in searches involving heavy-flavor jets. The High-Level Trigger (HLT) is designed to efficiently select events of interest while maintaining a manageable output rate of a few kilohertz. This report presents the commissioning and performance evaluation of b-jet triggers in the CMS HLT system using proton-proton collision data collected during Run-3 (2022-2024). Key aspects include algorithm optimization, efficiency studies, and comparisons with offline reconstruction. The results provide valuable insights into the current b-jet selection strategy and highlight potential refinements for future data-taking campaigns.
Paper Structure (7 sections, 2 equations, 4 figures)

This paper contains 7 sections, 2 equations, 4 figures.

Table of Contents

  1. Introduction
  2. The CMS High-Level Trigger
  3. ParticleNet@HLT for b-Jet Triggers
  4. b-tagging performance on data
  5. ParticleNet@HLT performance in 2024 data
  6. Online b-tagging performance in physics triggers
  7. Conclusions

Figures (4)

  • Figure 1: Illustration of jets initiated by (a) b-quark, (b) c-quark, and (c) light quark or gluon. Secondary vertices (SV) from heavy-flavor hadron decays are displaced from the primary vertex (PV), providing a key handle for heavy-flavor jet tagging at the trigger level.
  • Figure 2: Performance of the ParticleNet@HLT algorithm for AK4 b-jet tagging at the HLT. Shown is the b-jet tagging efficiency versus the misidentification probability for c and light-flavor jets in simulated $t\bar{t}$ events with $\mathrm{p_{T}} > 30$ GeV and $|\eta| < 2.5$. Results are compared to the Run--3 DeepJet and Run--2 DeepCSV taggers. ParticleNet demonstrates a $\sim$10--15% gain in efficiency at fixed mistag rates, representing the state-of-the-art for online AK4 b-tagging in CMS.
  • Figure 3: Per-jet ParticleNet@HLT efficiency vs. transformed offline ParticleNet BvsAll score for 2024 data. Plots from left to right show the loose (L), medium (M), and tight (T) online WPs. The top panel compares the efficiency across different data-taking eras, from RunC to RunI.I n the bottom panel, RunC is used as the reference, and the ratios of all other run eras are shown normalized to it.
  • Figure 4: Left: Trigger efficiency vs. mean offline ParticleNet score in the $e\mu$ control region for 2022 (red), 2023 (orange), and 2024 (blue). Right: $4j2b$ trigger efficiency on simulated $HH\to 4b$ events vs. $M_{HH}$ for 2022–2024.