Lund Plane to Bloch (LP2B) Encoding for Object and Polarization Tagging with Quantum Jet Substructure

TL;DR

This paper introduces the LP2B encoding and a quantum neural network for jet substructure analysis, achieving competitive performance with fewer parameters and better robustness than classical methods.

Contribution

The work presents a novel quantum encoding and architecture for jet tagging that matches classical deep learning performance with significantly fewer parameters and improved robustness.

Findings

QTTN matches LundNet in polarization tagging performance.

QTTN requires three orders of magnitude fewer parameters.

QTTN outperforms classical methods in low-data regimes.

Abstract

The application of quantum algorithms to jet substructure analysis is of growing interest as NISQ hardware continues to mature in qubit count and gate depth. Jet substructure remains essential for addressing demanding and complementary challenges at the LHC and beyond, notably object classification and polarization tagging. However, existing quantum machine learning approaches typically rely on data representations that suffer from infrared and collinear unsafety, sensitivity to non-perturbative effects, or poor scalability. In this work, we introduce the Lund Plane to Bloch (LP2B) encoding, designed to map a theoretically clean and robust representation of jet kinematics directly into qubit states. Leveraging this encoding, we implement a Quantum Tree-Topology Network (QTTN) that natively embeds the hierarchical structure of the Lund tree. We evaluate the QTTN across multiple benchmarks and observe that it matches the performance of large classical deep learning architectures, such as LundNet, on polarization tagging, while maintaining competitive accuracy for W boson and top quark tagging. The architecture demonstrates enhanced sensitivity compared to standard 1P1Q encodings on both polarization and W tagging, and pushes the Pareto front when compared against MLP of similar size and BDTs. Remarkably, the QTTN requires three orders of magnitude fewer parameters than LundNet, demonstrating promises for low-latency FPGA implementations in trigger systems. Furthermore, the QTTN outperforms classical methods in the low-data regime, making it suitable for low-yield, data-driven analyses. We also find that the quantum model is less susceptible to overfitting generator-specific parton shower and hadronization models than classical deep learning approaches, pointing toward potentially smaller systematic uncertainties. We validate the QTTN on real quantum hardware using a 3-qubit SpinQ device.