arXiv · 2611.17348
Topology-based fast neutron background suppression and sterile-neutrino sensitivity in a segmented reactor ${\overlineν}_{e}$ detector
Abstract
Reactor electron antineutrinos are commonly detected through the inverse beta decay interaction, with a prompt positron followed by delayed neutron capture. In shallow or above-ground segmented organic scintillator detectors, correlated fast neutrons are a dominant background because proton recoils can mimic the prompt IBD signal. We study topology-based machine learning using the Indian Scintillator Matrix for Reactor Anti-Neutrinos detector as a case study. A multilayer perceptron is trained on reconstructed topological observables from detailed Geant4 simulations. The $\mathrm{F_{1}}$ score and bootstrap resampling on an independent validation sample yield a fixed threshold of $P{\mathrm{IBD}}>0.40$. Applied without further optimization to an independent testing sample, the MLP retains $\mathrm{(83.4\pm 0.7)\%}$ of IBD events and rejects $\mathrm{(77.3\pm 0.7)\%}$ of fast neutron events. In a controlled spectral benchmark using an independent AmBe fast neutron spectrum, the selection reduces the net candidate-spectrum deformation by $\mathrm{84\%}$. The energy dependent signal acceptance and residual fast neutron component are subsequently incorporated into a binned likelihood, which recovers the injected IBD signal strength as $\mathrm{\hatα=0.993^{+0.053}_{-0.048}}$, with an empirical goodness-of-fit probability of $0.434$. The expected sterile-neutrino sensitivity for the ISMRAN detector is evaluated at baselines of 13 m and 18 m, with two years of exposure at each baseline. For the nominal pre-MLP $S/B=1:2$ benchmark and the background model, retaining the classifier score in a joint ($\mathrm{E_{prompt}\times P_{IBD}}$) likelihood improves the 95\% C.L. sensitivity to $\sin^{2}2θ_{14}$ by a logarithmically averaged 7.9\% over $0.1\leqΔm_{41}^{2}\leq1~\mathrm{eV}^{2}$ relative to an energy-only likelihood with the same fast neutron contamination.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Roni Dey. 2026-10-05. Topology-based fast neutron background suppression and sterile-neutrino sensitivity in a segmented reactor ${\overlineν}_{e}$ detector. https://searcharxiv.com/abs/2611.17348
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.