The LCFIVertex Package: vertex detector-based Reconstruction at the ILC

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.

💡 Research Summary

The paper presents the LCFIVertex software package, a comprehensive toolkit designed to exploit vertex‑detector information for high‑level event reconstruction at the International Linear Collider (ILC). The authors begin by motivating the need for precise vertex reconstruction in the ILC environment, where the detector’s ultra‑thin pixel layers and sub‑micron spatial resolution enable detailed tracking of heavy‑flavour decays. LCFIVertex builds on the established ZVTOP algorithm, offering two complementary vertex‑finding modules: ZVRES, which searches for common vertex candidates using impact‑parameter and covariance information, and ZVKIN, which incorporates track momentum vectors to locate secondary vertices from boosted B‑hadrons. Together they achieve high efficiency even in dense jet topologies.

Flavour tagging is performed through a multivariate analysis (MVA) that combines a rich set of variables derived from tracks (impact parameters, uncertainties, momentum) and reconstructed vertices (distance, invariant mass, number of sub‑vertices, surrounding energy flow). The package integrates with the TMVA framework, allowing users to train Boosted Decision Trees or neural networks on simulated samples and to evaluate performance via ROC curves and efficiency‑purity plots.

Validation proceeds in two stages. First, a fast parametric simulation (SimDet) provides an idealised environment with perfect tracking. In this setting LCFIVertex attains b‑tag efficiencies above 85 % at a light‑flavour mis‑identification rate of roughly 10 %, while c‑tag efficiencies exceed 70 % under similar conditions. The second stage employs a full GEANT4‑based detector simulation (LDC) together with realistic reconstruction code (LCIO, Marlin). Here material effects, non‑linear track distortions, and realistic detector geometry reduce vertex‑finding efficiency by about 5 %, yet b‑tag efficiency remains near 80 % and the overall flavour‑tagging performance stays robust.

A significant portion of the study investigates the impact of hadronic interactions (hard‑scatter secondary interactions) occurring near the interaction point. Such interactions generate spurious tracks that can be mistakenly clustered into fake vertices, degrading tagging performance. LCFIVertex includes a filtering option that uses energy‑loss information, particle‑identification clues, and geometric consistency to reject suspicious tracks. Without this filter, b‑tag efficiency drops by ~3 % and the light‑flavour mis‑tag rate rises by ~2 %; with the filter applied, the degradation is limited to less than 1 %.

The software architecture is deliberately modular. All inputs and outputs conform to the LCIO data format, and each functional block is implemented as a Marlin processor, facilitating seamless integration into existing ILC analysis chains. Users can customise parameters such as vertex‑search radius, minimum number of tracks per vertex, and MVA training datasets via simple configuration files, allowing optimisation for specific physics analyses (e.g., Higgs → bb̄, top‑pair studies, supersymmetric signatures).

In conclusion, LCFIVertex demonstrates that a well‑engineered vertex‑reconstruction and flavour‑tagging suite can fully exploit the ILC’s high‑precision vertex detector while remaining resilient to realistic detector effects. The package delivers high vertex‑finding efficiency, strong b‑ and c‑tag performance, and includes tools to mitigate the adverse effects of hadronic interactions. Its modular design and compatibility with the broader ILC software ecosystem make it a valuable asset for the upcoming physics program. Future developments outlined by the authors include the incorporation of deep‑learning‑based MVAs and real‑time vertex feedback loops, promising further gains in performance and adaptability.