Recent advances and trends in pattern recognition and data analysis for RICH detectors

Abstract

Ring Imaging Cherenkov (RICH) detectors are a key component of particle identification systems in many particle, nuclear and astroparticle physics experiments. Their ultimate performance depends not only on detector design and hardware implementation, but also crucially on the quality of pattern recognition and data analysis algorithms used to reconstruct Cherenkov ring images and to perform particle identification. In recent years, significant advances have been made both in traditional reconstruction approaches, such as likelihood-based methods and Hough-transform techniques, and in the application of modern machine learning tools. This contribution reviews the current state of RICH reconstruction algorithms, highlights representative use cases from operating experiments, and discusses emerging trends including global particle identification strategies and generative machine learning approaches for fast simulation and reconstruction.

Figures (10)

• Figure 1: Illustration of the Cherenkov ring emission and Cherenkov angle definition in the typical geometrical configuration of proximity focusing RICH detector. The yellow dots illustrate detected Cherenkov photons.
• Figure 2: Comparison of the expected number of detected photons for high-energy muons (left) with the corresponding measured number in data (right) in Belle II aerogel RICH, depending on the particle impact position on the aerogel radiator plane. $N^\mu$ is evaluated using toy simulation on track-by-track basis within the likelihood evaluation belle2_arich1.
• Figure 3: Comparison of Cherenkov ring image for high-energy muons in measured data (accumulated over $\sim 10$k tracks) and the PDF used for muon hypothesis likelihood evaluation. Both pictures are shown in the Cherenkov space. Several detailed effects of ring image mainly originating from photon reflections within the photon detectors can be seen on both pictures belle2_arich2.
• Figure 4: Left: Simulated event containing $B^0\to \pi^+\pi^-$ decay in LHCb RICH-1. Right: Expected number of photons in each pixel for the region of the left picture for a set of hypotheses that maximizes the global event PID likelihood lhcb_global.
• Figure 5: Global PID performance for kaon/pion separation at Belle for kaons and pions from $D^* \to (D\to K\pi)\pi$ decays. In both, simulated and measured data, notable improvement is observed when sub-detector PID likelihoods are combined by the neural network belle2_ml.