Performance of an Optical TPC Geant4 Simulation with Opticks GPU-Accelerated Photon Propagation

TL;DR

This paper addresses the computational bottleneck of simulating optical photons in high-precision TPC detectors by evaluating GPU-accelerated photon propagation with Opticks against CPU Geant4 in the NEXT-CRAB-0 model. The authors implement a Geant4-Opticks hybrid workflow, leveraging a modular Geant4 framework, Garfield++ drift modeling, and COMSOL-derived electric-field maps, with Opticks handling photon propagation in batches to manage VRAM. They report substantial performance gains, with speedups ranging from $58.47±0.02$ to $181.39±0.28$ over CPU simulations and an average improvement of $2.09±0.36$ per RTX generation, while demonstrating that key observables such as S1/S2 photon counts, arrival times, and wavelengths remain in good agreement between the two approaches. The results validate GPU-accelerated photon propagation as a practical tool for rapid, flexible optical simulations in large-scale detectors, enabling more detailed modeling and faster Monte Carlo production. The work has implications for detector design and data analysis workflows in xenon-based TPC experiments and related photon-detection systems.

Abstract

We investigate the performance of Opticks, a NVIDIA OptiX API 7.5 GPU-accelerated photon propagation tool compared with a single-threaded Geant4 simulation. We compare the simulations using an improved model of the NEXT-CRAB-0 gaseous time projection chamber. Performance results suggest that Opticks improves simulation speeds by between 58.47+/-0.02 and 181.39+/-0.28 times relative to a CPU-only Geant4 simulation and these results vary between different types of GPU and CPU. A detailed comparison shows that the number of detected photons, along with their times and wavelengths, are in good agreement between Opticks and Geant4.

Figures (8)

• Figure 1: Left: the interior of the NEXT-CRAB-0 detector. Three needle sources are shown by the thin metal rods attached to the field cage with distances 4, 10, and 14 cm from the electroluminescence region for needles 1, 2, and 3, respectively. Center and Right: Images from the NEXT-CRAB-0 detector with Single (center) and Dual MCP II (right) at $6$ bar xenon with three Pb-210 alpha sources. The images contain the averaged intensities from alpha events. The color of the images is normalized to their mean.
• Figure 2: Rendering of the field cage with three needle sources from Fusion 360, and simulated electric fields and potentials from COMSOL. The gradient of electric potential is shown by the color bar while the direction of the field lines is displayed by black arrows.
• Figure 3: Geometry rendering of NEXT-CRAB-0 in Opticks and Geant4. The $z$ direction points along the drift axis (with $z=0$ closest to the electroluminescence region), while the $x$ and $y$ coordinates are perpendicular to this.
• Figure 4: This flowchart depicts a Geant4-Opticks hybrid simulation pipeline. Initially, the Geant4 geometry is translated via the G4CXOpticks interface. During an event, Geant4 tracks primary and secondary particles other than optical photons. When optical photons are produced (e.g., by S1 or S2), their gensteps are transferred via the U4 interface to Opticks in memory-managed batches. Opticks then performs GPU-accelerated simulations of these photons using its simulate function. The hits resulting from the Opticks simulation are collected via the SEvt interface component and saved. Finally, after Geant4 processes and saves all hits, it signals Opticks through G4CXOpticks to reset for the next event.
• Figure 5: Performance comparison between Geant4 and Opticks using different CPU (AMD/i7/i9 devices) and GPU (RTX devices) combinations. The ratios are given in Table \ref{['tab:percentages']}.