Spectrally local geometric response at the onset of many-body quantum chaos

AI summary

Overview

Research area: Quantum physics — quantum chaos, quantum geometry, and random matrix theory in many-body spin chains. Technical level: Intermediate (readers should know basis states, eigenstates, and what a Hamiltonian is; the geometric concepts are built up from scratch in the paper). Scope: The paper introduces an energy-resolved "spectral density of geometric response" that tracks how sensitive each eigenstate is to a change of Hamiltonian parameter, and uses it to map how chaos nucleates and spreads across the spectrum in two spin-chain models and in two random-matrix constructions.

What This Paper Is About

Physicists usually study a parameter-dependent Hamiltonian in two separate ways: through its energy levels (spectral statistics) and through how its eigenstates deform when a parameter is tuned (quantum geometry). This paper fuses the two by assigning each eigenstate a geometric weight and plotting that weight against energy, producing a spectrum-resolved picture of eigenstate sensitivity. The goal is to understand not just whether a many-body system becomes chaotic, but where in the spectrum and in what order the transition happens.

Key Contributions

1. A new diagnostic, the spectral density of geometric response (SDGR). Defined as B_λ(E) = Σ_n δ(E − E_n) χ_λ^n, where χ_λ^n is the fidelity susceptibility of eigenstate |n⟩ with respect to parameter λ. It is the density of states reweighted by geometric sensitivity, and its integral recovers the normalized Hilbert–Schmidt norm of the adiabatic gauge potential (AGP), ‖A_λ‖².
2. Entropic characterization of spectral concentration. The normalized SDGR defines probabilities p_λ^n = χ_λ^n / Σ_m χ_λ^m, whose Shannon entropy S_λ, participation fraction Π_1^λ = D^{-1}e^{S_λ}, Rényi-2 fraction Π_2^λ, Rényi gap Δ_12^λ = log(Π_1^λ/Π_2^λ), and spectral center-of-mass offset η_λ = (⟨E⟩_λ − Ē)/Γ quantify how concentrated the geometric response is.
3. Demonstration of a three-to-four-regime crossover in physical models. Applied to a transverse-field Ising chain (ω = 1, g = 0.75, broken by longitudinal field h_x) and an XXZ chain (J = 1, Δ = 1.1, broken by a single-site σ_z defect of strength ε_d), the entropy reveals an integrable plateau, an entropic crossover at a scale Y*, a distinct entropic minimum at Y_min ~ e^{−N log 2} ~ D^{-1}, and a chaotic rise.
4. Random-matrix comparison isolating what is universal. A "Pauli-RMT" ensemble built from one- and two-body Pauli strings reproduces the intermediate entropic minimum (β_min ~ e^{−0.51N} for N = 8, 10, 12), whereas standard Gaussian ensembles do not, despite matching level-spacing statistics and AGP norms across constructions.

Main Findings

• Chaos nucleates locally rather than homogeneously. As integrability breaking increases past Y*, the entropy S_X/log D falls and the Rényi gap rises, meaning the eigenbasis deformation concentrates in selected spectral regions rather than spreading evenly. The authors describe this as chaos "nucleating" through rare, spectrally local rearrangements.
• Maximum localization occurs at an exponentially small scale. The entropic minimum sits at Y_min ~ e^{−N log 2} ~ D^{-1}, with the same leading scaling for both the TFIM and XXZ chains. At this point the participation fraction Π_1^X reaches its minimum and Δ_12^X its maximum.
• The AGP norm crossover is decomposed into two factors. The identity ‖A_X‖² = Π_1^X [χ_X^{(p)}]_typ shows the exponential growth of the norm is initially driven by a sharp rise in the typical susceptibility of a shrinking subset of eigenstates, not by a broadening of participation.
• The entropy crossover precedes conventional spectral diagnostics. In the Pauli-RMT model, S_λ/log D departs from its integrable plateau while the mean level-spacing ratio is still essentially Poissonian; the entropic minimum then occurs on the scale where level statistics begin to depart from Poisson.
• Integrable plateaus differ between the two models. For the TFIM the plateau saturates near S_ω/log D ≃ 1 for all sizes shown, with Π_1^ω showing the response is carried by most of the eigenbasis. For the XXZ chain the plateau is size-dependent, although S_J/log D → 1 with N while Π_1^J → 0.5, so the response is carried by a reduced fraction. The authors attribute the difference to interacting versus non-interacting integrability and to local versus global perturbations.
• Gaussian random-matrix ensembles lack the entropic minimum. No analogous minimum was found in standard Gaussian ensembles (details deferred to the supplemental material), even though level-spacing statistics and AGP norms agree across RMT constructions. The AGP norm, as a global measure, washes out the spectral redistribution that distinguishes the crossovers.
• An equivalent formulation via quantum Fisher information and AGP fluctuations. Since I_λ^n = 4χ_λ^n, the QFI spectral density is 4B_λ(E), and the factor cancels on normalization. The SDGR can also be written as B_λ(E) = Tr[δ(E − Ĥ)A_λ²], an energy-resolved density of AGP fluctuations, with total weight acting as a "geometric partition function" Z_λ^χ = D‖A_λ‖².
• The scaling exponent of Y is not pinned down.* The authors state that extracting a precise scaling exponent for the crossover is technically challenging and leave it open, while noting Y* is consistent with the AGP-norm crossover of earlier work.
• η_X provides a complementary witness. The spectral center-of-mass offset stays constant over the integrable plateau and becomes non-monotonic as the entropy falls, showing the concentration of response is accompanied by a redistribution of sensitivity across energy.

Methodology in Plain English

The authors take a Hamiltonian that depends on a tunable parameter and compute, for every eigenstate, how much that state shifts when the parameter is nudged — a quantity called fidelity susceptibility. Because near-degenerate levels can make this quantity diverge, they regularize it with an energy window μ = N D^{-1}, the same prescription as earlier AGP work, so that results can be compared directly against the known AGP-norm crossover.

They then histogram these susceptibilities against energy, giving a weighted density of states, and normalize it into a probability distribution. Standard information-theoretic measures of that distribution — entropy, participation fractions, a Rényi gap — tell them whether the geometric weight is spread across the whole spectrum or piled into a few states. They track these measures as the integrability-breaking perturbation is increased at fixed system size, and separately as system size grows at fixed perturbation.

Two spin chains are used because their integrability breaks in different ways: the TFIM's is broken globally by a longitudinal field, the XXZ chain's by a single local defect. Open boundary conditions are used, and calculations are restricted to a symmetry sector (the largest fixed-magnetization subspace for XXZ, the even spatial-reflection sector for TFIM), with D denoting the sector dimension. Finally, they repeat the analysis on a random-matrix model made of random one- and two-body Pauli strings, where integrability at β = 0 and chaos at increasing β can be dialed in, and compare against standard Gaussian ensembles.

Why This Matters

The work gives a microscopic explanation for a striking known fact: the AGP norm detects exponentially weaker integrability breaking than conventional spectral statistics, but as a single global number it cannot say why. The SDGR shows the exponential growth starts from a handful of anomalously sensitive eigenstates in narrow spectral windows, which then leak into neighbors as resonances proliferate. This reframes the route to chaos as a sequence with recognizable stages rather than a smooth interpolation, and it suggests that while the integrable and chaotic endpoints are universal (well captured by featureless random matrices), the path between them retains model-dependent structure.

Potential applications:

• Quantum sensing and metrology. The SDGR is built from fidelity susceptibility / quantum Fisher information, the standard resource for enhanced parameter estimation; energy-resolved structure could guide where to operate a sensor.
• Optimal control and state preparation. Knowing which eigenstates are hypersensitive to a parameter identifies where control protocols are fragile and where they can be exploited.
• Benchmarking quantum simulators. The entropy crossover appears before level statistics move away from Poisson, so it could serve as an early warning of unwanted chaotic behavior in a device.
• Characterizing complex many-body phases. The framework is proposed as generally applicable to criticality and non-Markovianity, not only to chaos.

Industry relevance: The paper is theoretical and reports no device experiments. Its practical pull is toward quantum-computing hardware characterization — cheap, early indicators of chaos onset are relevant to error accumulation and to validating that a simulator is operating in a controllable regime — and toward quantum sensing, where Fisher-information-based figures of merit are the working currency.

Future Directions

• Nailing down the crossover exponent. The authors explicitly leave open the precise scaling exponent of Y* with system size, which would sharpen the comparison to the AGP-norm crossover.
• Explaining why Gaussian ensembles fail. Understanding which coarse-grained physical features of Pauli-RMT (locality, one- and two-body structure) produce the entropic minimum, while featureless Gaussian ensembles do not, is posed as an open question about what is genuinely universal on the route to chaos.
• Finite-temperature and linear-response extensions. The supplemental material sketches a finite-temperature formulation and a connection to linear response theory, which invite follow-up work.
• Transfer to other phenomena and higher moments. The authors anticipate applying the joint spectral-geometric resolution to quantum phase transitions and non-Markovianity, and the supplement examines variance and skewness of the SDGR, suggesting higher moments as further diagnostics.

Target Audience

Researchers in quantum many-body physics, quantum chaos, and quantum information who already work with fidelity susceptibility, the adiabatic gauge potential, or level statistics — and who want a finer-grained tool than a single global norm. The paper is also accessible to graduate students entering the field, since it defines its geometric objects explicitly and works through two canonical spin models before moving to random matrix theory. Readers focused only on applications in quantum sensing or hardware will find relevant connections but no experimental data.

Authors’ abstract

We introduce the spectral density of geometric response, a measure of eigenstate sensitivity across an energy spectrum. Applied to many-body quantum systems, it reveals an exponentially sensitive integrability-to-chaos crossover, where eigenbasis deformations first accumulate in localized spectral regions before spreading throughout the spectrum. Physically motivated random-matrix ensembles reproduce this behaviour, whereas Gaussian ensembles do not, indicating universal features of the route to chaos that are absent from featureless random-matrix models.