Predicting the single-site and multi-site event discrimination power of dual-phase time projection chambers

TL;DR

This work quantifies the fundamental limits of single-site and multi-site discrimination in dual-phase xenon TPCs using Fisher Information and the Cramér-Rao bound, and validates that practical methods such as maximum likelihood reconstruction and convolutional neural networks can approach these limits. By building a generic light-based TPC model and predicting photon hit patterns with a Geant4 optical simulation, the study isolates how sensor size, photon budget, and geometry shape SS/MS separability. The results show that MS discrimination remains significantly more challenging than SS discrimination due to photon-pattern degeneracy, but that FI-guided predictions align closely with reconstruction performance for SS and MS2 scenarios. These insights offer a practical framework for optimizing detector design—sensor pixellation, light-sensor spacing, and optical readout strategy—to enhance rare-event searches like neutrinoless double beta decay and Migdal interactions in liquid xenon TPCs.

Abstract

Dual-phase xenon time projection chambers (TPCs) are widely used in searches for rare dark matter and neutrino interactions, in part because of their excellent position reconstruction capability in 3D. Despite their millimeter-scale resolution along the charge drift axis, xenon TPCs face challenges in resolving single-site (SS) and multi-site (MS) interactions in the transverse plane. In this paper, we build a generic TPC model with an idealized light-based signal readout, and use Fisher Information (FI) to study its theoretical capability of differentiating SS and MS events. We also demonstrate via simulation that this limit can be approached with conventional reconstruction algorithms like maximum likelihood estimation, and with a convolutional neural network classifier. The implications of this study on future TPC experiments will be discussed.

Figures (18)

• Figure 1: Left - an illustration of the generic TPC model used in our studies; Right - a cross-section view of the TPC model detailing the photosensor array configuration; the asterisks represent two light emission sites separated by a distance $d$ and $r_1$ and $r_2$ represent the radial distances between the light sources and the PMT under investigation.
• Figure 2: Left - the simulated number of photons detected by 2" PMTs, as a function of the radial distance between the PMT center and the light source, for every 10,000 emitted photons; different colors represent different maximum photon acceptance angles (black-90 degree, red-85 degree, blue-80 degree); Right - mean photon detection probability as a function of radial distance, along with their best fit functions, for individual sensors of different sizes (from top to bottom: 3", 2", 1", 15 mm and 6 mm).
• Figure 3: Predicted $x$ position reconstruction accuracy limits for SS events in our TPC model using FI and CRLB. Left - Position resolution limits estimated for 2" PMTs, as a function of the detected photon number, at event positions A (solid), B (dashed) and C (dotted); Right- position resolution limits estimated for 3", 2", 1", 15 mm, and 6 mm circular photosensors (from top to bottom) for events at position A.
• Figure 4: Predicted $d$ reconstruction accuracy with 500 total detected photons for MS2 events in our TPC detector model (2" PMTs) using FI and CRLB. Left -$\sigma(d)$ ($d$=1") as a function of azimuth angle of second vertex relative to the first, at event positions A (solid), B (dashed) and C (dotted); Right- same as left figure but with $d$=2". The total photon budget is evenly split between the two vertices.
• Figure 5: Predicted $\sigma(d)$ as a function of vertex separation ($d$), with 500 total detected photons for MS2 events with 2" (left) and 6 mm (right) photosensor sizes. The lines indicate different photon budget splitting between the two vertices: 1-1 splitting (solid line), 1-2 splitting (dotted), 1-4 splitting (dashed), 1-9 splitting (dot dashed). The red line represents $\sigma(d)=d$. The diverging trend near $d=0$ reflects an instability of the FI matrix and is unphysical.