Event-Level Classification of Boosted Multi-Tau Signatures in scNMSSM Di-Higgs Production
Abstract
In the semi-constrained NMSSM at large lambda and low tan(beta), di-Higgs production can proceed through h1h1 -> 4a1 -> 8tau when the lightest pseudoscalar lies below the b-bbar threshold. The tau pairs from each a1 decay are collimated and merge within a single jet. Using the benchmark point m_a1 = 8.2 GeV identified in earlier work as a representative signal topology, we examine which observables separate such events from Standard Model processes producing similar final states. The exotic branching fraction of this point exceeds present experimental limits, and all results concern the shape of the distributions, not the signal rate. The substructure observables normally used for two-prong tagging prove ineffective here: the jet mass and the N-subjettiness ratio tau_21 have separation powers of 0.02 and 0.09 against boosted Z -> tau tau. Both are ditau jets, and the neutrinos from tau decays displace the reconstructed masses to a common value near 18 GeV despite parent masses differing by an order of magnitude. Discrimination comes instead from the event as a whole. A boosted decision tree trained on single fat jets reaches an AUC of 0.759 +- 0.002, while an event-level classifier reaches 0.941 +- 0.001, driven by the multiplicity of boosted objects and the masses of the subleading ones. Top-quark pair production reproduces the multi-object structure, and against ttbar -> bbbar tau tau the AUC falls to 0.855 +- 0.008. For a cross-section-weighted combination of the two simulated backgrounds the classifier reaches 0.908 +- 0.002, with 71% signal efficiency at 10% background efficiency. These are classification metrics from fast simulation, not discovery significances.
Related
- Multi-Channel Di-Higgs Production in the scNMSSM: From Exotic 8b to Boosted 8tau Signatures at the HL-LHC
- Cosmic ray boosted dark matter with momentum dependent interactions can explain the LZ 248 keV event
- Boosted dark particles and the LZ nuclear recoil event
- A closer look at the LZ 248 keV event through the lens of cosmic-ray boosted dark matter
- High-Energy Nuclear Recoils from Boosted Dark Matter for the LZ 248-keV Event: Beyond the Halo-Dependent High-Velocity Tail
BibTeX
@misc{telba2026eventlevel,
title = {{Event-Level Classification of Boosted Multi-Tau Signatures in scNMSSM Di-Higgs Production}},
author = {Marwa Telba},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2609.29620},
eprint = {2609.29620},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.29620},
}
RIS
TY - UNPB
TI - Event-Level Classification of Boosted Multi-Tau Signatures in scNMSSM Di-Higgs Production
AU - Marwa Telba
PY - 2026
DA - 2026/10/06
PB - arXiv
DO - 10.48550/arxiv.2609.29620
UR - https://arxiv.org/abs/2609.29620
AB - In the semi-constrained NMSSM at large lambda and low tan(beta), di-Higgs production can proceed through h1h1 -> 4a1 -> 8tau when the lightest pseudoscalar lies below the b-bbar threshold. The tau pairs from each a1 decay are collimated and merge within a single jet. Using the benchmark point m_a1 = 8.2 GeV identified in earlier work as a representative signal topology, we examine which observables separate such events from Standard Model processes producing similar final states. The exotic branching fraction of this point exceeds present experimental limits, and all results concern the shape of the distributions, not the signal rate. The substructure observables normally used for two-prong tagging prove ineffective here: the jet mass and the N-subjettiness ratio tau_21 have separation powers of 0.02 and 0.09 against boosted Z -> tau tau. Both are ditau jets, and the neutrinos from tau decays displace the reconstructed masses to a common value near 18 GeV despite parent masses differing by an order of magnitude. Discrimination comes instead from the event as a whole. A boosted decision tree trained on single fat jets reaches an AUC of 0.759 +- 0.002, while an event-level classifier reaches 0.941 +- 0.001, driven by the multiplicity of boosted objects and the masses of the subleading ones. Top-quark pair production reproduces the multi-object structure, and against ttbar -> bbbar tau tau the AUC falls to 0.855 +- 0.008. For a cross-section-weighted combination of the two simulated backgrounds the classifier reaches 0.908 +- 0.002, with 71% signal efficiency at 10% background efficiency. These are classification metrics from fast simulation, not discovery significances.
ER -