Scrambling the spectral form factor: unitarity constraints and exact results

I. INTRODUCTION

• Unitary dynamics prevents indefinite late-time decay of correlations in finite systems, requiring time-averaged stabilization at a scale set by the entropy or Hilbert-space dimension.Source: §I · p. 2
• The spectral form factor, an analytically continued partition function, probes finite-spectrum effects and has been studied in SYK and holographic CFT models.Source: §I · p. 3
• The paper relates the spectral form factor to the survival probability of a thermofield double state and applies quantum-speed-limit bounds to its evolution.Source: §I · p. 3

THERMOFIELD DOUBLE STATE

• For the thermofield double, unitarity bounds on dynamics can be expressed using equilibrium thermodynamic properties through the partition function.Source: §I · p. 3, THERMOFIELD DOUBLE STATE · p. 4

I. INTRODUCTION

• The study analyzes short- and long-time survival-probability behavior, quantum-speed-limit decay bounds, and examples including AdS2, Calogero-Sutherland, and Gaussian unitary ensemble models.Source: §I · p. 4

A. Quantum Speed Limits on the evolution of a thermofield state

• Quantum speed limits bound the rate of survival-probability change using the energy dispersion of the initial state and the geometry of evolution in projective Hilbert space.Source: A. Quantum Speed Limits on the… · p. 5
• For a thermofield double, the relevant mean energy and fluctuations equal those of the canonical state and connect energy fluctuations to specific heat.Source: A. Quantum Speed Limits on the… · p. 6
• The generalized Bhattacharyya bound yields an exponential lower bound on survival probability for arbitrary initial states, specialized to the thermofield double.Source: A. Quantum Speed Limits on the… · p. 6
• The thermofield-double bound holds for S(β,t) ≥ 1/2 through the half-lifetime, when S(β,t_h) = 1/2.Source: A. Quantum Speed Limits on the… · p. 6
• The resulting connection between quantum dynamics and equilibrium properties is restricted to evolution times comparable to the short-time asymptotic regime.Source: A. Quantum Speed Limits on the… · p. 6

B. Short-time nonexponential decay

• The short-time survival probability is Gaussian through O(t^3), with energy fluctuations setting the evolution speed and no linear-in-time term.The quadratic behavior is a signature of unitary dynamics and depends on the existence of the Hamiltonian’s second moment.Source: B. Short-time nonexponential decay · pp. 6–7
• The short-time Gaussian law can deviate when energy fluctuations diverge or when the Hamiltonian lacks a finite second moment.Nonunitary dynamics generally introduces an O(t) correction, while strongly perturbed systems can extend Gaussian decay beyond the short-time regime.Source: B. Short-time nonexponential decay · p. 7

C. Long-time nonexponential decay

• Continuous and discrete spectra have qualitatively different long-time survival behavior.Discrete spectra support recurrence and a long-time average fixed by the thermal state’s purity, whereas continuous spectra exhibit decay dynamics without the same recurrence structure.Source: C. Long-time nonexponential decay · pp. 7–8
• In discrete spectra, the long-time survival average equals the purity of the thermal reduced state and therefore measures its delocalization over energy eigenstates.The purity is bounded between the pure-state and maximally mixed limits, with the latter associated with infinite temperature.Source: C. Long-time nonexponential decay · p. 8
• For continuous spectra bounded below, analytic constraints require long-time decay to be slower than exponential, with the low-energy spectral density determining the asymptotic form.If ρ(E) ∼ E^k near the ground-state energy, the survival probability follows a corresponding power law at long times.Source: C. Long-time nonexponential decay · pp. 9–10
• The memory term links long-time survival to rare unitary processes in which decay products reconstruct the initial state.Suppressing this term, for example through nonunitary dynamics, leads to exponential behavior.Source: C. Long-time nonexponential decay · pp. 9–10

IV. EXAMPLES

• The examples progress from the harmonic oscillator and integrable Calogero-Sutherland model to random matrices and holographic field theories.This sequence is used to illustrate survival dynamics across discrete and continuous spectra and increasing model complexity.Source: §IV · p. 10

A. Quantum harmonic oscillator and the xp model in AdS2

• The oscillator’s short-time asymptotics expose the role of energy fluctuations in setting the initial decay.The thermofield-double state can also be represented as a two-mode squeezed state.Source: A. Quantum harmonic oscillator and the… · p. 11
• The harmonic oscillator survival probability is periodic, with period 2π/ω, and oscillates at a single nominal frequency ω.This behavior follows from the harmonic spectrum and the resulting single-frequency time dependence.Source: A. Quantum harmonic oscillator and the… · pp. 10–12
• The xp model in AdS2 has exactly the same thermofield-double survival probability as the harmonic oscillator.The correspondence is specific to the thermofield-double state and connects the CFT1 realization to the oscillator result.Source: A. Quantum harmonic oscillator and the… · pp. 10–12

B. Rational Calogero-Sutherland model

• The thermofield-double survival probability is independent of the coupling constant because the coupling only renormalizes the zero-point energy.It is likewise independent of E0 and λ in the stated setting, unlike more complex quench scenarios.Source: B. Rational Calogero-Sutherland model · p. 13
• For N > 1, the survival probability acquires modulation at frequencies kω for k = 1, …, N and has period 2πN/ω.For N = 1, the model recovers the single-harmonic-oscillator behavior.Source: B. Rational Calogero-Sutherland model · p. 13
• The model’s spectrum is equivalent to that of N renormalized harmonic oscillators and applies to ideal bosons, fermions, and geons.The rational Calogero model is also related to fractional exclusion statistics and SU(N) super Yang-Mills spectra.Source: B. Rational Calogero-Sutherland model · p. 13
• At low temperature, the survival probability becomes approximately monotonic and follows a t^-3 power law, while higher temperatures retain oscillations.The oscillation amplitude diminishes as temperature decreases.Source: B. Rational Calogero-Sutherland model · p. 14
• At long times, the memory term accurately reproduces the exact survival probability, indicating decay followed by reconstruction of the thermofield-double state.The comparison is made for a GUE random-matrix ensemble at large N with λβ = 0.1.Source: B. Rational Calogero-Sutherland model · p. 14

1. Survival probability at large N

• The large-N GUE calculation uses the Wigner semicircle density on E ∈ [−λ, λ] and evaluates the survival amplitude through Bessel-function expressions.The ensemble is the Gaussian unitary ensemble with Dyson β = 2.Source: §1 · pp. 14–15
• The GUE survival probability initially decays parabolically, with the inverse energy fluctuations setting the Zeno time.The full decay is evaluated for Hamiltonians whose large-N density of states follows Wigner’s semicircle law.Source: §1 · pp. 14–15
• At intermediate times, the survival probability follows a t^-3 power law caused by the square-root behavior of the density of states near the low-energy edge.The relevant edge scaling is ρ(E) ∼ (E + λ)^1/2.Source: §1 · pp. 15–16
• At high temperatures, oscillations are superimposed on the power-law decay, while lowering the temperature reduces them until the decay becomes monotonic.The oscillatory behavior arises from the coherent superposition of many modes in the thermofield-double state.Source: §1 · p. 16
• At long times, the memory term dominates and describes reconstruction of the thermofield-double state from its decay products.The dynamics can pass through an intermediate time at which the evolving state is orthogonal to the initial state before reconstruction occurs.Source: §1 · p. 16

Exact finite-N calculation

• The exact GUE spectral form factor is obtained for arbitrary N by evaluating disconnected and connected contributions and combining them with the Z(2β) term.The calculation uses the analytically continued partition function, averaged eigenvalue densities, Hermite polynomials, and exact Fourier transforms.Source: §2 · pp. 17–18

Spectral form-factor behavior

• The finite-N curve shows an initial decay, approximate power-law revivals, a dip, a linear ramp, and a plateau of order 1/N.For N = 105, the power-law envelope scales as 1/t^3 before the ramp saturates.Source: §2 · pp. 18–19

Temperature dependence

• Increasing inverse temperature makes the initial decay more nearly linear, while the dip, ramp, and plateau remain relatively stable.The comparison uses N = 35 and β = 0, 0.05, 0.1, 0.2, 0.4.Source: §2 · p. 19

Decay and dip

• At short times the disconnected contribution produces Gaussian decay, while at later times a power-law or oscillatory envelope joins the connected contribution near a dip at t_d ∼ O(1).The initial Gaussian behavior is governed by the Laguerre polynomial and the energy-fluctuation-dependent Zeno time.Source: §2 · pp. 19–20

Ramp and plateau

• The connected contribution drives the late-time linear rise toward the purity plateau, while the spectral form factor approaches the purity for finite N and vanishes as N → ∞.The late-time approach follows from the Riemann–Lebesgue lemma; the plateau height also tends to zero in the infinite-N limit.Source: §2 · pp. 18–20

Normalization and N scaling

• The dip-time scaling depends on normalization: it is O(1) in the paper’s finite-N convention and acquires the large-N scaling after rescaling energies, time, and temperature.The rescaling keeps typical energies O(1) in the infinite-N analysis.Source: §2 · p. 20

• The holographic thermofield double is dual to an eternal black hole, while the spectral form factor diagnoses information loss and recovery.Source: §V · p. 21
• Holographic spectral-form-factor evolution has three phases: Gaussian decay, a power-law envelope, and an RMT-like linear ramp toward a plateau.Source: §V · p. 21
• Poincaré recurrences are not expected until times of order t ∼eeS, far beyond the ramp-and-plateau regime.Source: §V · p. 21
• The analytic RMT curves provide a surprisingly good description of the holographic spectral form factor.Source: §V · p. 21

• The holographic fidelity has a non-exponential initial decay, followed by algebraic late-time decay; in CFT2, modular images yield a 1/t3 power law [11].Source: §1 · p. 22
• Semiclassical analysis probes only a coarse-grained density of states, so it predicts decay to zero even though finite spectra forbid complete decay.Source: §1 · p. 22
• Assuming late-time RMT behavior gives an average linear ramp to a plateau, allowing estimates of the dip time and ramp-to-plateau crossover.Source: §1 · p. 22
• In the AdS3/CFT2 case, the dip time is exponentially large in central charge c, while the plateau joins the ramp at a parametrically later time.Source: §1 · p. 22
• The dip-time exponent may receive an O(1) modification from including matter contributions beyond the vacuum character.Source: §1 · p. 22
• Analogous behavior occurs in SYK, and the RMT spectral-form-factor calculation is proposed as an analytical model for holographic information loss and restoration.Source: §1 · p. 23

2. Bound on Fidelity Decay

• The holographic discussion focuses on compact-horizon black holes because they correspond to finite Hilbert spaces and discrete spectra.Source: §2 · p. 23
• The holographic fidelity bound is evaluated using energy fluctuations derived from the thermal ensemble and holographic partition function.Source: §2 · p. 23

VI. CONCLUSIONS AND DISCUSSION

• The survival amplitude maps the spectral-form-factor evolution to thermofield-double fidelity and its associated decay probability.Source: §VI · p. 24
• Quantum-speed-limit and unitarity bounds yield an early-time Gaussian decay bound and a sub-exponential bound for later fidelity decay.Source: §VI · p. 24
• For discrete spectra, the spectral form factor remains generally non-vanishing at very late times because it is governed by thermofield-double purity.Source: §VI · p. 24
• The exact all-time spectral form factor is obtained for generalized unitary ensembles using orthogonal polynomials.Source: §VI · p. 24
• The analytical random-matrix result matches earlier numerical studies [8] and resembles spectral form factors in holographic models including SYK [8] and the BTZ black hole [11].Source: §VI · p. 24
• The GUE provides an analytical toy model for studying information loss and retrieval in black holes.Source: §VI · p. 24

Appendix A: Bhattacharyya bound for arbitrary initial states

• The generalized Bhattacharyya bound applies to arbitrary initial states, including mixed states represented by density matrices.Survival probability is expressed through fidelity between the initial and evolved states, and Uhlmann’s bound leads to the Mandelstam-Tamm speed limit.Source: Appendix A · p. 25
• The half-lifetime satisfies S(th) = 1/2, and th can exceed π/(2∆E).Equation (A7) gives the generalized bound until the fidelity first reaches one-half.Source: Appendix A · p. 25

Appendix B: Complete decay for systems with continuous spectrum

• For continuous Hamiltonian spectra, the survival probability vanishes as t →∞ because it is the Fourier transform of an integrable energy distribution.The argument assumes a ground-state energy and support on [E0, ∞).Source: Appendix B · pp. 26–27

1. Integrals of Hermite polynomials

• The derivations require sufficiently rapid decay of the integrands so integration-by-parts boundary terms can be neglected.The appendix states this condition holds for the integrals considered.Source: §1 · p. 27
• The GUE appendix reduces products of Hermite-polynomial integrals to sums of single-polynomial integrals using differentiation identities and binomial expansions.The resulting master integrals are used to compute the spectral form factor, including disconnected contributions.Source: §1 · pp. 28–29

2. Spectral density

• The exact eigenvalue density yields the survival amplitude in terms of Laguerre polynomials, recovering the main-text expression in Eq. (63).The derivation first produces a sum over Laguerre polynomials and then reexpresses it as a single associated Laguerre polynomial.Source: §2 · p. 29

3. Two-level correlation

• The connected two-level correlation function is obtained by factorizing its defining integral and combining the second master integral with the relevant coefficients.This produces the expression quoted in the main text as Eq. (67).Source: §3 · pp. 29–30

4. Map to Wilson loop

• The GUE survival amplitude maps to the computation of a single Wilson loop in a Gaussian matrix model.The identification matches the Gaussian matrix-model weight after formally relating ν to the matrix-model parameter.Source: §4 · pp. 30–31
• The spectral form factor maps to a correlation function of two Wilson loops under the same formal gauge-coupling identification.The appendix derives the relation by rewriting the GUE measure over eigenvalues and inserting the corresponding Wilson-loop operators.Source: §4 · pp. 30–31