AI Meets Antimatter: Unveiling Antihydrogen Annihilations

TL;DR

This work tackles the challenge of precisely determining the vertical annihilation position $z$ in ALPHA-g to enable a $1\%$ precision test of antimatter gravity. It introduces PEAR, a modified PointNet ensemble that regresses the $z$-vertex directly from 3D spacepoints collected by the radial Time Projection Chamber, bypassing traditional track reconstruction. On Monte Carlo data, PEAR delivers substantially improved resolution while maintaining comparable $z$-bias to the conventional Helix Fit, achieving a Pearson correlation of about $0.9997$ and reducing the full width at half maximum from $14.12$ mm to $6.62$ mm. The results suggest that deep learning can enable more precise gravity measurements for antimatter and potentially transfer to other high-resolution vertex reconstruction tasks in particle detectors; the approach is validated on calibration data and future work includes extending to $(x,y)$ vertices and uncertainty estimation.

Abstract

The ALPHA-g experiment at CERN aims to perform the first-ever direct measurement of the effect of gravity on antimatter, determining its weight to within 1% precision. This measurement requires an accurate prediction of the vertical position of annihilations within the detector. In this work, we present a novel approach to annihilation position reconstruction using an ensemble of models based on the PointNet deep learning architecture. The newly developed model, PointNet Ensemble for Annihilation Reconstruction (PEAR) outperforms the standard approach to annihilation position reconstruction, providing more than twice the resolution while maintaining a similarly low bias. This work may also offer insights for similar efforts applying deep learning to experiments that require high resolution and low bias.

Figures (4)

• Figure 1: Conceptual schematics of the vertex reconstruction approach using our deep learning model (bottom), in contrast to the standard method that requires identification of particle tracks and fitting helix functions (top).
• Figure 2: Schematics of our modified PointNet architecture for the vertex reconstruction regression task, heavily based on Qi2017.
• Figure 3: Heat plot of predicted $z$-vertex with PEAR versus true $z$-vertex (left). Histogram of residuals for PEAR and Helix Fit with Gaussian fits (right).
• Figure 4: Box plot of residuals from PEAR and Helix Fit predictions for each 200 mm slice of the detector (top). For this study, the line within the box denotes the median, the box covers the interquartile range (IQR), and the whiskers extend to the furthest residuals within 1.5 times the IQR on either side of the distribution. Residual mean ($\mu_i$ used in ARA to probe $z$-basis) for each 200 mm slice of the detector (bottom). Note that the ends of the detector are greyed out, as generally no scientific measurements are made outside these bounds.