Clustering of Conditional Mutual Information via Quantum Belief-Propagation Channels

Kohtaro Kato 1, Tomotaka Kuwahara 2,3,4 · 2025

Read Clustering of Conditional Mutual Information via Quantum Belief-Propagation Channels by Kohtaro Kato 1, Tomotaka Kuwahara 2,3,4 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-02235.txt.

Introduction

Kohtaro Kato and Tomotaka Kuwahara attack a conjecture about how fast quantum correlations fade with distance, in a 2025 paper. Conditional mutual information measures the correlations a middle region mediates between two sides of a many-body system. In thermal Gibbs states it is expected to decay quickly, but a general proof has resisted the field.

The authors recast the problem through belief propagation channels, local maps that enact small changes to the Hamiltonian. If those channels are quasi-local, information decay follows, so the conjecture becomes a statement about channels. Under conditions like rapid mixing they build such channels, proving decay at all temperatures conditionally and outright at high temperatures. A fresh look at cluster expansions rounds out the picture.

Catalog record

CREATORKohtaro Kato 1, Tomotaka Kuwahara 2,3,4
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE147 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-02235

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Clustering of Conditional Mutual Information via Quantum Belief-Propagation Channels Kohtaro Kato 1, Tomotaka Kuwahara 2,3,4 Abstract Conditional mutual information (CMI) has recently attracted significant attention as a key quantity for characterizing quantum correlations in many-body systems. While it is conjectured that CMI decays rapidly in finite-temperature Gibbs states, a complete and general proof remains elusive. In this work, we introduce a new formulation of the problem based on the belief propagation (BP) channel , namely a completely positive trace-preserving (CPTP) map that realizes local perturbations of the Hamiltonian. Within this framework, we prove that establishing the quasi-locality of BP channels implies the decay of CMI, thereby reducing the original conjecture to a more tractable problem. We show that such quasi-local BP channels can be constructed under natural physical conditions, such as uniform rapid mixing or uniform clustering. Under these assumptions, we obtain conditional proofs of CMI decay valid at all temperatures. Moreover, because these assumptions are automatically satisfied at high temperatures, our results in that regime yield unconditional proofs of CMI decay. At the same time, in order to better understand the high-temperature behavior of Gibbs states, we revisit the cluster expansion method. Contrary to common intuition, we demonstrate that when multipartite correlations such as CMI are considered, the cluster expansion suffers from intrinsic divergence problems rooted in the Baker–Campbell–Hausdorff formula, revealing fundamental limitations of this traditional approach. I Introduction One of the central goals of quantum many-body theory is to uncover universal principles that apply regardless of the microscopic details of individual systems. In recent years, information-theoretic tools have emerged as powerful means to characterize such universal behavior, with quantities like mutual information and quantum entanglement playing critical roles in understanding correlations between subsystems. It is widely recognized that these bipartite information measures exhibit clustering—i.e., exponential decay with spatial separation—away from critical points [ 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 ] . Although there have been proposals for information-theoretic quantities that exhibit exponential decay even at low temperatures [ 12 , 13 , 14 ] , clustering is generally expected to break down in this regime due to the emergence of long-range correlations. In contrast, it is widely assumed, although often not explicitly stated, that at sufficiently high temperatures, all physically relevant correlations decay rapidly with distance, reflecting the underlying locality of thermal equilibrium states. This viewpoint was indeed mathematically formulated in terms of the cluster expansion technique [ 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 ] . Among the information-theoretic quantities that have attracted increasing attention in recent years, a particularly important one is the conditional mutual information (CMI). This quantity has emerged as a central tool in quantum information theory [ 23 , 24 , 25 ] , yet many fundamental aspects of its behavior remain poorly understood both in the low-temperature and high-temperature regimes. Given a tripartition of a quantum system into regions A , B , and C , and a quantum state \rho defined on the joint Hilbert space of ABC , the CMI is defined as I_{\rho}(A:C|B):=S_{\rho}(AB)+S_{\rho}(BC)-S_{\rho}(ABC)-S_{\rho}(B), (1) where S_{\rho}(X) denotes the von Neumann entropy of the reduced state on region X . As a genuinely tripartite quantity, CMI captures correlations beyond pairwise interactions and plays a central role in characterizing many-body correlations such as topological order [ 26 , 27 , 28 ] . Furthermore, through its connection to quantum Markovianity , CMI is deeply linked to the concept of quantum recoverability [ 29 , 30 , 31 , 32 , 33 , 34 ] . This connection has led to important applications, including the definition of quantum mixed phases [ 35 , 36 , 37 , 38 , 39 ] and the design of quantum Gibbs sampling algorithms [ 40 , 41 ] . A central open question in this context is the following: Conjecture 1 (CMI decay at arbitrary temperatures) . For general quantum Gibbs states \rho_{\beta} at any temperature \beta , the conditional mutual information I_{\rho_{\beta}}(A:C|B) with A\cup B\cup C=\Lambda decays rapidly (e.g., super-polynomially) with the distance between the arbitrary regions A and C , where \Lambda denotes the entire system. This is a quantum analogue of the Hammersley-Clifford theorem in the case of classical or commutative Hamiltonians [ 42 , 43 ] . A conditional version under (uniform) clustering has been discussed in the literature (see, e.g., Brandão–Kastoryano [ 41 ] ): Conditional version of Conjecture 1 . Under the assumption of the uniform clustering of correlations (or exponential decay of correlations), the quantum Gibbs state exhibits decay of the CMI. We defer the precise definition to Assumption 4 . At high temperatures, an even stronger form is conjectured: Conjecture 2 (CMI decay at high temperatures) . For general quantum Gibbs states at sufficiently high temperatures, the conditional mutual information I_{\rho_{\beta}}(A:C|B) with A\cup B\cup C\subseteq\Lambda decays rapidly with the distance between the regions A and C . These two conjectures capture a hierarchy of decay behavior. When A\cup B\cup C=\Lambda , the decay of CMI is expected to be a universal feature, independent of temperature. In contrast, at high temperatures, CMI is expected to exhibit exponential decay even when A\cup B\cup C is strictly contained in \Lambda , indicating a stronger form of spatial locality. However, this stronger decay property does not generally hold at low temperatures: explicit counterexamples (e.g., quantum topological order) are known where I_{\rho_{\beta}}(A:C|B) fails to decay when A\cup B\cup C\subset\Lambda in the low-temperature regime [ 44 , 45 ] . The formulation of conjectures concerning the decay of CMI is a relatively recent development. In 2016, a general proof for one-dimensional systems was provided by Kato and Brandão [ 46 ] , marking a significant first step in this direction. For Conjecture 1 , a partial resolution was later achieved in 2024 through the development of a systematic method to construct effective Hamiltonians on subsystems [ 47 ] . This allowed for proofs of the CMI decay at arbitrary temperatures as long as the regions A and C are small. Therefore, a key open problem is whether the regions A and C can be taken arbitrarily large, or, equivalently, whether the |A|,|C| dependence of the CMI decay is at most polynomial. Conjecture 2 , concerning the high-temperature regime, was initially believed to be resolved by the 2020 work of Kuwahara, Kato, and Brandão [ 48 ] , who introduced a technique known as the generalized cluster expansion. In this method, physical quantities of interest are expanded perturbatively in terms of Hamiltonian parameters, and the convergence of this expansion is then analyzed. This approach has proven effective in a variety of contexts [ 18 , 19 ] . However, when applying this technique to CMI, it was later pointed out that the treatment of the logarithm of reduced density matrices involves uncontrolled approximations, which undermines the convergence argument in the original proof. As a result, the applicability of the generalized cluster expansion to establishing CMI decay remains an open question [ 49 ] . In this work, we propose a new approach to establishing the decay of CMI for arbitrarily large subsystems

Read the complete text