Simulating and reconstructing PLATON scintillator tracks

What this research found

The PLATON detector concept proposes replacing the fine internal segmentation of particle trackers with a single unsegmented scintillator block watched only on its surface, recovering three-dimensional position computationally from photon arrival times. K-Dense built a first-principles Monte Carlo of a 20 cm scintillator cube with 600 surface channels to test that claim, then reconstructed interaction vertices and short tracks from the simulated light. At a realistic operating point the reconstruction localized vertices to a median 1.31 mm and track direction to 7.0 degrees, and stayed near 1.5 mm once silicon-photomultiplier noise was modelled and filtered.

• At the nominal operating point of 100 ps timing jitter and 1000 detected photons, the reconstruction reached a median vertex resolution of 1.31 mm, a median track-direction error of 7.0 degrees, and a median track-length error of 2.37 mm using only 600 surface channels.
• Timing precision dominates every other factor. Doubling the per-photon jitter from 100 ps to 200 ps degraded the median vertex error from 1.31 mm to 3.63 mm and the median angular error from 7.0 to 22.4 degrees.
• Vertex resolution improves with photon statistics roughly as expected from counting noise: 1.92 mm at 500 detected photons, 1.31 mm at 1000, and 0.68 mm at 2000, a factor of about 2.8 against the factor of 2 a square-root scaling predicts.
• Dark counts break tracking but not vertexing. The outlier-robust vertex fit held near its noise-free value up to 1 MHz per channel, while unfiltered track-direction error jumped from 5.4 to roughly 60 degrees as soon as any dark counts appeared, and head-tail identification fell to 33%, 67%, and 17% correct at 100 kHz, 1 MHz, and 10 MHz.
• A two-stage spatial-temporal coincidence filter retained 98.5 to 99.2% of genuine photons while rejecting 100% of dark counts at 100 kHz, 95.9% at 1 MHz, and 91.2% at 10 MHz, restoring the vertex resolution to 1.47 mm at 1 MHz and holding it below 2.6 mm across four decades of dark rate.
• Because a monolith instruments only its boundary, channel count scales with surface area rather than volume, so the saving grows linearly with detector size. The report concludes that large-volume neutrino experiments, not PET scanners, are the most plausible first beneficiary.

How it was done

The simulated detector is a homogeneous 20 cm cube with refractive index 1.58, giving an in-medium light speed of 18.974 cm per nanosecond, tiled on all six faces with a 10 by 10 grid of pixels at 2 cm pitch for 600 channels and no interior readout. Photons were emitted isotropically from point vertices or uniformly along short straight tracks, propagated in straight lines to their exit face, and detected with Gaussian timing jitter and a binomial photodetection efficiency. Two estimators inverted the arrival pattern: a four-parameter time-of-flight trilateration solver using a robust Huber loss with analytic gradients for the vertex, and a seven-parameter maximum-likelihood solver for the track that marginalizes over each photon's unknown emission point. Detected light yield and timing jitter were swept across six cells with 8 vertex and 6 track trials each, and a second sweep stepped the per-channel dark count rate over four decades with crosstalk and afterpulsing held fixed, reconstructing each event both raw and coincidence-filtered.

Data sources

• DUNE far detector technical design report, Journal of Instrumentation 15:T08008 (2020) — comparison figures for liquid-argon time-projection chambers
• Blondel et al., SuperFGD prototype charged particle beam tests, Journal of Instrumentation 15:P12003 (2020) — segmented plastic tracker comparison
• Surti & Karp, Physica Medica 80:251 (2020) — time-of-flight PET performance comparison
• Piemonte & Gola, Nuclear Instruments and Methods A 926:2 (2019) — silicon photomultiplier noise parameters
• Nemallapudi et al., Journal of Instrumentation 11:P10016 (2016) — state-of-the-art single-photon time resolution