Orthogonal Catastrophe in Dynamical Quantum Kernels: Spectral Scrambling, Initial-State Ergodicity, and the Informative Time Window

Ergodicity, and the Informative Time WindowQuantum kernel methods promise advantage by mapping data into high-dimensional Hilbert spaces, yet current parameterized quantum circuit (PQC) architectures suffer from the orthogonal kernel catastrophe-the exponential concentration of kernel entries that destroys trainability. This crisis arises because deep PQCs rapidly scramble quantum states into indistinguishable Haar-like distributions. To overcome this, we introduce dynamical quantum kernels generated by continuous-time Hamiltonian evolution. We rigorously prove that although dynamical evolution postpones the instantaneous concentration characteristic of deep PQCs by generating short-time signals, the asymptotic long-time catastrophe persists via distinct physical mechanisms: the global fidelity kernel concentrates due to spectral scrambling (arising from non-resonance conditions typical of chaotic systems), whereas the projected kernel collapses solely from initial-state ergodicity, where Haar-random delocalization extinguishes local signals independent of the Hamiltonian's chaoticity. Crucially, this catastrophe is not immediate. Through rigorous short-time perturbation analysis of local Hamiltonians, we prove the existence of a robust informative time window. Utilizing linearly parameterized Hamiltonians H(θ) = H0 + θV , we derive an exact Ω(n)t 2 feature variance scaling and show that localized product states guarantee an Ω(1) discriminative signal per site, escaping the PQK catastrophe. Our work connects kernel trainability to the spectral and ergodic properties of Hamiltonian dynamics, dictating that practical advantage requires leveraging localized initial states and the dynamical grace period before global scrambling dominates.

Authors

Institutions

Publication Details

Journal
Quantum Science and Technology
Published
2026-11-18
DOI
https://doi.org/10.1088/2058-9565/aec3cf
Primary Topic
Quantum Computing Algorithms and Architecture
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Orthogonal Catastrophe in Dynamical Quantum Kernels: Spectral Scrambling, Initial-State Ergodicity, and the Informative Time Window

Youle Wang, Wenbin Yu, Lei Zhang, Hanmiao Cheng
2 links
Quantum Science and Technology
Quantum Computing Algorithms and Architecture
article

Orthogonal Catastrophe in Dynamical Quantum Kernels: Spectral Scrambling, Initial-State Ergodicity, and the Informative Time Window

Youle Wang, Wenbin Yu, Lei Zhang, Hanmiao Cheng
article en

Abstract

Ergodicity, and the Informative Time WindowQuantum kernel methods promise advantage by mapping data into high-dimensional Hilbert spaces, yet current parameterized quantum circuit (PQC) architectures suffer from the orthogonal kernel catastrophe-the exponential concentration of kernel entries that destroys trainability. This crisis arises because deep PQCs rapidly scramble quantum states into indistinguishable Haar-like distributions. To overcome this, we introduce dynamical quantum kernels generated by continuous-time Hamiltonian evolution. We rigorously prove that although dynamical evolution postpones the instantaneous concentration characteristic of deep PQCs by generating short-time signals, the asymptotic long-time catastrophe persists via distinct physical mechanisms: the global fidelity kernel concentrates due to spectral scrambling (arising from non-resonance conditions typical of chaotic systems), whereas the projected kernel collapses solely from initial-state ergodicity, where Haar-random delocalization extinguishes local signals independent of the Hamiltonian's chaoticity. Crucially, this catastrophe is not immediate. Through rigorous short-time perturbation analysis of local Hamiltonians, we prove the existence of a robust informative time window. Utilizing linearly parameterized Hamiltonians H(θ) = H0 + θV , we derive an exact Ω(n)t 2 feature variance scaling and show that localized product states guarantee an Ω(1) discriminative signal per site, escaping the PQK catastrophe. Our work connects kernel trainability to the spectral and ergodic properties of Hamiltonian dynamics, dictating that practical advantage requires leveraging localized initial states and the dynamical grace period before global scrambling dominates.

Quantum Science and Technology
Nanjing University of Information Science and Technology (CN), Instituto de Óptica "Daza de Valdés" (ES), Xi'an Institute of Optics and Precision Mechanics (CN)
Basic Research Program of Jiangsu Province
Quantum Computing Algorithms and Architecture
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon