[Paper Review] The LCFIVertex Package: vertex detector-based Reconstruction at the ILC

S. Hillert|arXiv (Cornell University)|Aug 20, 2007

High-Energy Particle Collisions Research5 references3 citations

TL;DR

The LCFIVertex package is a C++ software framework for high-precision vertex-based reconstruction at the International Linear Collider (ILC), implementing the ZVTOP topological vertex finder (including ZVKIN and ZVRES variants) and a flexible neural network-based flavour tag with quark charge determination. It achieves improved tagging performance over prior FORTRAN implementations, especially at high energies, due to better detector resolution and suppression of hadronic interaction effects using MC information.

ABSTRACT

The contribution gives an overview of the LCFIVertex package, providing software tools for high-level event reconstruction at the International Linear Collider using vertex-detector information. The package was validated using a fast Monte Carlo simulation. Performance obtained with a more realistic GEANT4-based detector simulation and realistic tracking code is presented. The influence of hadronic interactions on flavour tagging is discussed.

Motivation & Objective

• To develop a modern, flexible, and high-performance software framework for vertex detector-based event reconstruction at the ILC.
• To enable precise flavour tagging and topological vertex reconstruction using high-resolution vertex detector data.
• To validate the new C++ implementation against legacy FORTRAN code and assess performance under realistic detector conditions.
• To quantify the impact of hadronic interactions in detector material on flavour tagging performance.
• To provide a community-ready tool with extensible neural network architecture and shared trained models via a dedicated repository.

Proposed method

• The package implements the ZVTOP topological vertex finder, including both ZVRES (probability tube-based) and ZVKIN (ghost track-based) algorithms for improved vertex resolution in short-lived B-hadron decays.
• Flavour tagging is performed using a neural network approach, with the Hawkings algorithm as the default, and supports customizable input variables, architecture, and training methods.
• The software uses the LCIO data format and integrates with MarlinReco for event reconstruction, with planned integration into the JAS3-based US software framework org.lcsim.
• A track cheater mode using MC truth is employed during initial validation to assign hits to tracks, later replaced with pattern recognition for realistic performance evaluation.
• Detector response is simulated using MOKKA with GEANT4, and tracking is performed via the LDCTracking package with Gaussian hit smearing and particle flow reconstruction.
• Performance is evaluated using PYTHIA-generated events at √s = 500 GeV and Z-resonance, with comparisons to previous BRAHMS and FORTRAN-based results.

Experimental results

Research questions

• RQ1 How does the performance of the new C++-based LCFIVertex package compare to the legacy FORTRAN implementation in terms of flavour tagging purity and efficiency?
• RQ2 To what extent do hadronic interactions in the beam pipe and vertex detector layers degrade b-jet tagging performance in realistic detector simulations?
• RQ3 Can the ZVKIN variant of the ZVTOP algorithm effectively reconstruct short-lived B-hadron decays and enable future charge determination for neutral hadrons?
• RQ4 How do improvements in vertex resolution (e.g., 2 μm vs. 3.5 μm) and suppression of K-short and Lambda decays affect flavour tagging separation power?
• RQ5 What is the impact of using MC truth-based track suppression on tagging performance, and how does it vary with center-of-mass energy?

Key findings

• The LCFIVertex package achieves excellent agreement with the legacy FORTRAN code in tagging purity and efficiency for b- and c-jets at the Z-resonance, confirming code correctness and validation.
• At √s = 500 GeV, the new package shows improved tagging performance compared to previous results, primarily due to a higher assumed vertex resolution of 2 μm instead of 3.5 μm.
• The suppression of tracks from K-short and Lambda decays, photon conversions, and hadronic interactions enhances flavour tagging separation power compared to earlier studies.
• Including tracks from hadronic interactions in the beam pipe and detector material significantly degrades b-jet tagging performance at 500 GeV, but has negligible impact at the Z-resonance.
• The track cheater mode yields similar tagging performance to realistic pattern recognition, indicating robustness to reconstruction method in initial evaluations.
• The current suppression of hadronic interaction effects relies on MC information and is specific to the LDC01Sc detector model, highlighting a need for future physics-motivated suppression in real-time reconstruction.