arXiv:2610.20724
Long-lived local coherences at high temperature from frozen domain walls
Sarang Gopalakrishnan, Romain Vasseur
Abstract
Local coherences between distinct charge sectors--measured by the single-particle Green's function or the transverse spin autocorrelation function--do not overlap with hydrodynamic modes. At finite temperature, therefore, they decay faster than any power law in generic chaotic systems. We present a mechanism by which these coherences can persist anomalously long, for integrable and chaotic spin chains (including the XXZ and Fredkin spin chains) that obey a quantum group symmetry. We show--using a Mazur bound derived from the quantum-group symmetry, together with locality and a "non-undershoot" assumption--that the transverse spin autocorrelation function $C(t) \equiv \langle σ^-_x(t) σ^+_x(0)\rangle$ can decay at most algebraically: $C(t) \gtrsim 1/t^2$. This bound is not tight: in the large-anisotropy limit of the XXZ spin chain, at infinite temperature, we explicitly compute $C(t) \sim \sqrt{3}/(18 \sqrt{t})$. We also compute this decay rate in the XXZ spin chain for general easy-axis anisotropy, and estimate it to scale as $C(t) \sim t^{-2/z}$ (for transport exponent $z$) in the chaotic models.
AI Summary
Why can local coherences between different conserved-charge sectors decay algebraically at high temperature, despite having no overlap with ordinary hydrodynamic modes? This paper studies transverse spin autocorrelations $C(t)=\langle\sigma_x^-(t)\sigma_x^+\rangle$ in easy-axis XXZ, Fredkin, and chaotic spin chains with $U_q(sl_2)$ quantum-group symmetry. In a height representation, the symmetry protects spin flips associated with deep minima until transport exposes them to deeper minima. Polar decomposition of the quantum-group raising operator produces a conserved partial isometry with a finite-size Mazur bound $\overline C\gtrsim4/(\pi^2L^2)$. Combining this with Lieb–Robinson locality and an assumed “non-undershoot” property gives a bound on the moving average, $t^{-1}\int_t^{2t}C(\tau)d\tau\gtrsim1/(4\pi^2v^2t^2)$. This is not a pointwise bound. In the folded XXZ limit, an explicit calculation gives $C(t)\sim\sqrt3/(18\sqrt t)$, supported by simulations up to 1,024 sites. For finite easy-axis anisotropy, a kinetic argument using heavy Bethe strings predicts $C(t)\sim[ D(\Delta)t]^{-1/2}[\ln q/\ln(t/t_0)]^2$. For chaotic chains, a survival-and-return argument predicts $C(t)\sim t^{-2/z}$, where $z$ is the spin-transport exponent. Away from half filling, chaotic chains are expected to recover stretched-exponential decay, whereas integrable chains retain algebraic decay. The finite-time bound depends on the numerically supported but unproved non-undershoot assumption. The finite-anisotropy and chaotic exponents are heuristic: accessible XXZ times do not isolate the logarithmic correction, and chaotic-chain numerics have not reached asymptotic power laws.
Central claim may not hold: The abstract states a lower bound on C(t) itself, whereas the conditional locality–Mazur argument establishes only a lower bound on its moving average and consequently large values at arbitrarily late times. The abstract describes the general easy-axis XXZ decay rate as computed, although its derivation depends on a heuristic coherence lifetime and cutoff whose predicted logarithmic correction is not resolved numerically.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 5 pages + supplemental material