GPU-based optical simulation of the DARWIN detector

L. Althueser, B. Antunović, E. Aprile, D. Bajpai, L. Baudis, D. Baur, A. Baxter, L. Bellagamba, R. Biondi, Y. Biondi, A. Bismark, A. Brown, R. Budnik, A. Chauvin, A. P. Colijn, J. J. Cuenca-García, V. D’Andrea, P. Di Gangi, J. Dierle, S. Diglio, M Doerenkamp, K. Eitel, S. Farrell, A. D. Ferella, C. Ferrari, C. Findley, H. Fischer, M. Galloway, F. Girard, R. Glade-Beucke, L. Grandi, M. Guida, S. Hansmann-Menzemer, Florian Jörg, Lyndon Jones, P. Kavrigin, L. M. Krauss, B. von Krosigk, F. Kuger, H. Landsman, R. F. Lang, S. Li, S. Liang, M. Lindner, J. Loizeau, F. Lombardi, T. Marrodán Undagoitia, J. Masbou, E. Masson, J. A. M. Lopes, S. Milutinovic, C. M. B. Monteiro, M. Murra, K. Ni, U. Oberlack, I. Ostrovskiy, M. Pandurović, R. Peres, J. Qin, M. Rajado Silva, D. Ramírez García, P. Sánchez-Lucas, J.m.f. dos Santos, M. Schümann, M. Selvi, F. Semeria, H. Simgen, M. Steidl, P.-l. Tan, A. Terliuk, K. Thieme, Roberto Trotta, C. D. Tunnell, F. Tönnies, K. Valerius, S Vetter, G. Volta, W. Wang, C. Wittweg, Y. XingJournal of Instrumentation 2022 · DOI · arXiv

TL;DR

Researchers from the DARWIN collaboration have successfully ported the optical simulation of their next-generation dark matter detector to Chroma, a GPU-based photon tracking framework. This move delivers a massive 8x to 24x speedup over traditional CPU-based Geant4 simulations, enabling the team to rapidly iterate on novel detector designs, such as reflective electrode coatings and 4$\pi$ sensor coverage.

What Problem Does This Paper Address?

The DARWIN (Dark Matter WIMP Search with Liquid Xenon) experiment aims to build the "ultimate" liquid xenon (LXe) observatory, utilizing a massive 40-tonne dual-phase Time Projection Chamber (TPC). To maximize its sensitivity to rare events like dark matter interactions or neutrinoless double-beta decay, the detector must collect as much scintillation light as possible.

However, simulating light propagation in such a large, complex volume is a computational nightmare. In standard simulation toolkits like Geant4, photon tracking can consume over 95% of the total CPU time. This bottleneck makes it nearly impossible to quickly test how different geometries or materials affect Light Collection Efficiency (LCE). The authors tackle this by offloading the heavy lifting of ray-tracing to GPUs.

Methods

The authors utilized Chroma, an open-source framework that uses CUDA C to track photons on the GPU and Python for the user interface.

Key methodological highlights include:

  • CAD Integration: Unlike Geant4, which often requires tedious manual geometry coding, Chroma allows for the direct import of CAD files, significantly speeding up the modeling of complex components like electrode frames and PMT arrays.
  • Validation & Bug Fixes: The team performed rigorous "sanity checks" against Geant4, investigating refraction at the liquid-gas interface and Rayleigh scattering. Notably, they identified and fixed a discrepancy in Chroma’s original Rayleigh scattering algorithm to match theoretical expectations and Geant4’s performance.
  • Comparative Benchmarking: They tested the simulation speed across various hardware configurations, including NVIDIA GTX TITAN X and TITAN Xp GPUs against several Intel and AMD CPUs.

Datasets

As this is a simulation-focused physics paper rather than a machine learning study, there are no "training datasets." Instead, the "data" consists of:

  • Optical Property Tables: A comprehensive compilation of VUV (vacuum ultraviolet) properties for LXe, gaseous xenon, PTFE, and synthetic quartz, derived from existing literature.
  • Simulated Photon Batches: Evaluations were performed using batches of $10^5$ to $10^7$ isotropically generated photons within the TPC's active volume.

Key Insights and Ideas

  • The Electrode Sink: One of the most significant "aha" moments was quantifying exactly where photons go to die. In the baseline design, electrodes and their frames account for ~18% of all photon losses.
  • Reflective Coatings are a Game Changer: The study suggests that coating stainless steel electrodes with highly reflective Al/MgF2 (90% reflectivity in the VUV) could increase the total LCE by as much as 20% (relative).
  • The 4$\pi$ Obstruction: While adding SiPMs or PMTs to the barrel of the TPC (a 4$\pi$ design) seems intuitive, the authors found that the copper Field Shaping Rings (FSRs) act as a significant shield, blocking roughly half of the potential light gain unless the FSRs themselves are made reflective.
  • Single-Phase Advantage: Removing the liquid-gas interface (Single-phase TPC) not only simplifies the detector but also increases LCE by ~11% relative and narrows the photon arrival time distribution, which could improve event timing.

Key Results

  • Performance: Chroma demonstrated a speedup of 8x to 24x compared to Geant4. On a TITAN Xp GPU, the simulation was consistently over 12x faster than on high-end CPUs.
  • Baseline LCE: The baseline DARWIN design achieves an average LCE of approximately 43%.
  • Design Optimization:
    • Coating all electrodes boosted LCE to 51.8%.
  • A 4$\pi$ design with 2-inch PMTs and reflective FSRs reached an LCE of 46.8%.
  • Removing the bottom PMT array (to save cost/radioactivity) was deemed unfeasible, as it slashed LCE to a mere 18%.

Limitations and Future Work

  • Software Maturity: The authors note that Chroma is not as well-documented or supported as the industry-standard Geant4.
  • Physical Scope: Currently, the simulation only handles optical photons. Future work involves integrating this GPU engine into a full simulation chain that includes energy deposition, electron drift, and electronics noise.
  • Experimental Uncertainty: The accuracy of these simulations is limited by our knowledge of material properties in the VUV range. The authors emphasize the need for more precise experimental measurements of VUV reflectivity and absorption in liquid xenon environments to refine their models.

Summary updated

Abstract

Abstract Understanding propagation of scintillation light is critical for maximizing the discovery potential of next-generation liquid xenon detectors that use dual-phase time projection chamber technology. This work describes a detailed optical simulation of the DARWIN detector implemented using Chroma, a GPU-based photon tracking framework. To evaluate the framework and to explore ways of maximizing efficiency and minimizing the time of light collection, we simulate several variations of the conventional detector design. Results of these selected studies are presented. More generally, we conclude that the approach used in this work allows one to investigate alternative designs faster and in more detail than using conventional Geant4 optical simulations, making it an attractive tool to guide the development of the ultimate liquid xenon observatory.
BibTeX
@article{althueser2022gpu,
  author = {L. Althueser and B. Antunović and E. Aprile and D. Bajpai and L. Baudis and D. Baur and A. Baxter and L. Bellagamba and R. Biondi and Y. Biondi and A. Bismark and A. Brown and R. Budnik and A. Chauvin and A. P. Colijn and J. J. Cuenca-García and V. D’Andrea and P. Di Gangi and J. Dierle and S. Diglio and M Doerenkamp and K. Eitel and S. Farrell and A. D. Ferella and C. Ferrari and C. Findley and H. Fischer and M. Galloway and F. Girard and R. Glade-Beucke and L. Grandi and M. Guida and S. Hansmann-Menzemer and Florian Jörg and Lyndon Jones and P. Kavrigin and L. M. Krauss and B. von Krosigk and F. Kuger and H. Landsman and R. F. Lang and S. Li and S. Liang and M. Lindner and J. Loizeau and F. Lombardi and T. Marrodán Undagoitia and J. Masbou and E. Masson and J. A. M. Lopes and S. Milutinovic and C. M. B. Monteiro and M. Murra and K. Ni and U. Oberlack and I. Ostrovskiy and M. Pandurović and R. Peres and J. Qin and M. Rajado Silva and D. Ramírez García and P. Sánchez-Lucas and J.M.F. dos Santos and M. Schümann and M. Selvi and F. Semeria and H. Simgen and M. Steidl and P.-L. Tan and A. Terliuk and K. Thieme and Roberto Trotta and C. D. Tunnell and F. Tönnies and K. Valerius and S Vetter and G. Volta and W. Wang and C. Wittweg and Y. Xing},
  title = {{GPU-based optical simulation of the DARWIN detector}},
  journal = {Journal of Instrumentation},
  year = {2022},
  doi = {10.1088/1748-0221/17/07/p07018},
}
Figure