Event Reconstruction of the Hyper-Kamiokande Detector

Ben Carew - IMAPP - 29/09/25

Supervisors: Dr Vladimir Gligorov and Dr Mathieu Guigue

Hyper-Kamiokande

SuperK, the predecessor of HyperK
  • Next-generation neutrino detector
  • Water Cherenkov - collects radiation from super-luminal particles
  • 20,000 photomultiplier tubes with charge and time data
  • Need to reconstruct physical events from the PMT data

Hyper-Kamiokande

Motivation

  • Charge-Parity Violation explains matter-antimatter asymmetry in the universe
  • HyperK will have world-leading sensitivity to CPV in neutrinos
  • Fast reconstruction directly reduces systematic errors for this analysis

Motivation

  • Event reconstruction needs to be efficient and robust
    • Unexpected deviations from simulated data and backgrounds
    • Malfunctions and unpredictable detector behaviour
  • Faster reconstruction \(\rightarrow\) larger MC sample \(\rightarrow\) greater discovery potential
    • Focus on CPV analysis, but will benefit HyperK's broad physics goals
  • Crucial to optimise reconstruction before calibration in July 2027

Scattered Light Table

\[A(\textcolor{orange}{s}) = A(\textcolor{blue}{x_{\text{PMT}}, z_{\text{vtx}}, R_{\text{vtx}}, \varphi, \theta, \phi}) = \frac{\mathrm{d}\textcolor{green}{\mu}^{\text{sct}}}{\mathrm{d}\textcolor{green}{\mu}^{\text{iso,sct}}}\]
  • Accounts for predicted charge from light scattered in water and reflected
  • Table requires computationally taxing 6D interpolation at each point \(\textcolor{orange}{s}\) along the track - ~35% of total runtime
  • Optimisable by studying the physics behind the table's structure

Scattered Light Table

Gradient Threshold Optimisation

  • Calculate 6D gradient between each table value
  • Apply a threshold:
    • Above -> interpolate
    • Below -> nearest neighbour
  • Find highest threshold value before error becomes large

Gradient Threshold Optimisation

Gradient Threshold Results

Gradient Threshold Results

Parametrisation Optimisation

  • Regions of the scattering table can be well-fit by analytical functions
  • Use physically-motivated relationships between dimensions to reduce interpolation complexity

Parametrisation Results

Selective Interpolation Optimisation

  • Determine which dimensions must be interpolated for each bin
  • Select most efficient interpolation strategy for a given point
  • Generate bespoke interpolation function at runtime

25% Error Threshold

25% Error Threshold

5% Error Threshold

5% Error Threshold

Momentum Reconstruction

Final Results

10% speed-up with 2% additional error

Next Steps

Recently completed HyperK cavern
  • Optimise reconstruction with machine learning before calibration phase
  • Construction shifts, installation and qualification of timing system
  • Generate simulation samples and procedure for data validation
  • Phenomenological studies preparing for CPV data analysis

Summary

  • Developed three optimisation methods for the scattering table interpolation
  • Selective interpolation provides good efficiency-accuracy trade-off
  • Machine learning techniques may produce greater efficiency improvements
  • Even 10% faster reconstruction = 10% more calibration data = improved systematics for HyperK