arXiv:2609.38007
Improved estimate of local Markovianity for quantum Gibbs states
Tai Hsuan Yang
Abstract
We establish improved locally Markovian properties of quantum Gibbs states at every fixed positive temperature. For finite-range interactions of bounded interaction degree, the conditional mutual information (CMI) decays exponentially with separation, with a prefactor exponential in the boundary size of only one of the two regions and independent of the size of the other region and the total system. This improves the exponential volume dependence of the local Markov bounds established by Chen and Rouzé \cite{Chen2025}. For exponentially and stretched-exponentially decaying interactions, we obtain stretched-exponential CMI decay with a boundary-dependent prefactor. For power-law interactions, we prove inverse-logarithmic or algebraic decay with polynomial region-size dependence. Our proof is static, in contrast to the dynamical approach of Chen and Rouzé. We compare the Gibbs state with a product state obtained by removing boundary interactions, reducing the CMI bound to a relative-entropy loss that we control using locality estimates.
AI Summary
The paper asks how conditional mutual information (CMI) across a shielding region decays in quantum Gibbs states at arbitrary fixed positive temperature, without assuming clustering or rapid mixing. It considers full partitions $\Lambda=A\sqcup B\sqcup C$, with separation $r=d(A,C)$. For finite-range interactions with bounded term-overlap degree, it proves $I(A:C\mid B)\le C_\beta\exp(C_\beta g_{A\mid A^c}-c_\beta r)$, where $g_{A\mid A^c}$ is the summed interaction strength across the cut. On lattices this yields an exponential boundary-size prefactor, replacing previous exponential region-volume dependence, with no dependence on $|C|$ or total volume. Either region can be used by symmetry. For aggregate interaction decay $e^{-\mu r^\theta}$, unrestricted support sizes, and $D\ge2$, the bound is $C_\beta\exp[C_\beta(1+|\partial_{\mathrm e}A|)^{\theta/D}-c_\beta(1+r)^{\theta/(D+1-\theta)}]$. For aggregate power-law decay $(1+r)^{-\zeta}$, unrestricted supports are also allowed. The bounds have prefactor $(1+|A|)^{b_\zeta+1}$, where $b_\zeta=1+2\zeta/D$: distance decay is $[\log(e+r)]^{-b_\zeta}$ for $D<\zeta<2D$, with a squared iterated-logarithm correction at $\zeta=2D$, and $(1+r)^{-\nu_\zeta}$ for $\zeta>2D$, where $\nu_\zeta=b_\zeta(\zeta-2D)/(\zeta-D+b_\zeta)$. The static proof removes cut interactions to obtain a product reference state, bounds CMI by relative-entropy loss under partial trace, and controls that loss through a variational resolvent formula, Lieb–Robinson locality, and modular spectral estimates. Long-range proofs additionally use local truncation and Gibbs interpolation. The finite-range prefactor remains exponential, not polynomial. Arbitrary marginals with unobserved exterior are not covered. Algebraic decay throughout $D<\zeta\le2D$, sharper exponents, and stronger boundary dependence remain open.
The results presented in this paper were found by GPT-6 Astra Pro. The author verifies and organizes the proof and takes full responsibility for the correctness of the paper.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 35 pages, no figures, improved exposition