Position Reconstruction in the DEAP-3600 Dark Matter Search Experiment

TL;DR

The paper presents three complementary position-reconstruction algorithms for DEAP-3600: a hit-pattern maximum-likelihood method, a time-of-flight maximum-likelihood method, and a neural-network-based mapper using PMT charge patterns. Together with a detailed photon-counting and timing framework and a data-driven resolution assessment, these methods enable effective fiducialization and background rejection, particularly for neck-region shadowing. Validation against $^{39}$Ar, $^{40}$Ar, and $^{22}$Na calibration data shows good data–MC agreement and demonstrates the neck-region advantages of the neural-network approach. The work enhances the experiment’s WIMP sensitivity by improving event localization in the bulk LAr and under challenging surface geometries, with robust uncertainty characterization via a data-driven resolution technique.

Abstract

In the DEAP-3600 dark matter search experiment, precise reconstruction of the positions of scattering events in liquid argon is key for background rejection and defining a fiducial volume that enhances dark matter candidate events identification. This paper describes three distinct position reconstruction algorithms employed by DEAP-3600, leveraging the spatial and temporal information provided by photomultipliers surrounding a spherical liquid argon vessel. Two of these methods are maximum-likelihood algorithms: the first uses the spatial distribution of detected photoelectrons, while the second incorporates timing information from the detected scintillation light. Additionally, a machine learning approach based on the pattern of photoelectron counts across the photomultipliers is explored.

Figures (16)

• Figure 1: Schematic view of the DEAP-3600 detector (The water tank in which it is immersed is not shown). Variables used in the charge-based hit-pattern position reconstruction ($\vec{x}$, $\vec{r_i}$, $\theta_i$) are indicated.
• Figure 2: The light propagation model and the coordinate system for the time-of-flight reconstruction algorithm.
• Figure 3: The probability density function is shown as a two-dimensional histogram for the time-of-flight and spatial "cell" ID, with the color scale proportional to the probability density. The variables $r$ and $\theta$ in the definition of cell ID are given in Fig. \ref{['fig:TOF_schematic']}. Left: The histogram includes all cells. Right: The histogram is zoomed to show cells 8400 to 8700, representing three distance values $r$ = 840 mm, 850 mm, and 860 mm with $\sin\theta$ varying from 0.00 to 0.99 for each distance value.
• Figure 4: Distributions of reconstructed angular position variables, normalized reconstructed cubed radius ($(r/850\,\rm{mm})^3$) and reconstructed $z$, comparing $^{39}\rm{Ar}$ data (solid) and $^{39}\rm{Ar}$ simulation (dotted) and $^{40}\rm{Ar}$ nuclear recoil simulation (dashed), for the hit pattern algorithm (top row), the time-of-flight algorithm (middle row), and the neural network algorithm (bottom row). All the plots are normalized with their respective area.
• Figure 5: Estimates from the hit pattern and time-of-flight based algorithms of the contained mass of LAr within a radius of the reconstructed position. The estimate is based on the fraction of $^{39}$Ar decays observed in the 90--200 PE range reconstructing within a given radius. It is assumed that the true positions of the $^{39}$Ar nuclei are uniformly distributed throughout the LAr target. The analytical contained LAr mass calculated using the geometric volume and the density is also shown.