⚛️ quantum physics

Renormalization group and long-range conditional mutual information in hierarchical models

This paper investigates the relationship between the renormalization group (RG) and long-range conditional mutual information (CMI), demonstrating that while RG constraints typically ensure UV-finiteness, specific hierarchical models can exhibit divergent Markov lengths or polynomial CMI decay under RG flow despite admitting simple, stable RG descriptions.

Original authors: Yu-Hsueh Chen

Published 2026-09-03

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Yu-Hsueh Chen

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the study of matter, physicists often look for patterns that persist even as things change. When a material is heated or cooled, its atoms rearrange, yet certain fundamental behaviors remain the same. To understand these deep truths, scientists use a tool called the renormalization group. Imagine looking at a forest from a distance: you stop seeing individual leaves and start seeing the shape of the trees, then the shape of the groves. This tool works similarly, zooming out from the tiny details of atoms to reveal the large-scale behavior of the whole system. Usually, this process is straightforward: as you zoom out, the complex details fade away, leaving behind a simple, predictable picture. However, nature sometimes hides surprises. There are states of matter that look simple from a distance but hold onto complex, long-range connections that refuse to fade, even when the system is disturbed. Detecting these hidden connections is difficult because standard measurements often miss them entirely.

A researcher has now developed a new way to spot these elusive connections and has used it to map out two strange new types of matter. They focused on a specific measurement called conditional mutual information. To understand this, imagine three friends, Alice, Bob, and Charlie, sitting in a line. Alice and Charlie are far apart, with Bob sitting between them. If Alice and Charlie are talking to each other only through Bob, then knowing what Bob is doing explains everything about their connection. But if Alice and Charlie are sharing a secret that Bob cannot hear, that is a different kind of connection. The researcher used this idea to measure how much information two distant parts of a system share that is not explained by the middle part. They found that for many systems, this hidden connection disappears quickly as the middle part gets larger. But in the specific models they studied, this connection remained strong, defying the usual rules of how matter behaves.

The researcher began by proving a general rule about how these hidden connections behave. They showed that if a system can be simplified step-by-step without losing its essential long-range links, then the hidden connection between distant parts must be finite and well-behaved. This might sound technical, but it is a powerful guarantee: it means that if a system can be "renormalized" or simplified in a clean way, we can trust that its large-scale properties are stable and calculable. This rule acts as a filter, separating systems that are truly complex from those that only look complex because of short-range noise. With this rule in hand, the researcher turned their attention to two specific models of matter to see how they behaved under this new scrutiny.

The first model they examined was a variation of a classic system known as the hierarchical Ising model. In this setup, spins—tiny magnetic arrows—are arranged in a tree-like structure where groups of three are combined into a single unit, which is then combined with others, and so on. The researcher found something surprising: at any temperature above absolute zero, the distance over which the hidden connections fade away becomes infinite. In other words, the "buffer" needed to separate two parts of the system so they stop sharing secrets would have to be infinitely large. This suggests the system is deeply entangled over long distances. Yet, when they applied their renormalization tool to this system, it flowed smoothly into a completely simple, random state, as if all the complexity had vanished. This created a tension: the system looked infinitely complex in one way, but perfectly simple in another. The researcher resolved this by showing that while the system is connected to the simple state, the connection requires a specific type of circuit that is longer than what is typically allowed in standard definitions of "simple." It is a state that is technically connected to simplicity but refuses to be classified as simple under stricter rules.

The second model they studied was even more intriguing. It was built by combining two different rules for how spins interact: one that creates a strong, ordered pattern and another that enforces a global balance. This hybrid system behaved in a way that had never been seen before. In this model, the hidden connection between distant parts vanished when the middle section was small, meaning the system followed the local rules of physics perfectly. However, when the middle section was large, the hidden connection remained strong, violating the global rules. This means the system is locally well-behaved but globally mysterious. The researcher then introduced noise, or random errors, into the system. They found that at a specific critical level of noise, the hidden connection between two points did not disappear instantly. Instead, it faded away very slowly, following a power law, while the standard measure of connection between those same two points vanished completely. This is a rare phenomenon where the standard tools say "nothing is connected," but the more sensitive tool says "there is still a deep link."

The study concludes that these hierarchical models provide a clear window into how matter can maintain long-range order without being obvious. The first model shows that a system can have an infinite range of hidden connections yet still flow toward a trivial state under renormalization, provided the rules for simplification are flexible enough. The second model demonstrates that a system can be stable against noise while breaking global rules, maintaining a hidden link that standard measurements miss. These findings suggest that our current understanding of how matter organizes itself is incomplete. There are phases of matter that are stable and robust, yet they hide their true nature behind a veil of local simplicity. By using the renormalization group to probe these hidden layers, the researcher has shown that the universe of possible states is richer than previously thought, containing structures that are locally innocent but globally profound. This work does not just describe new materials; it offers a new way of thinking about what it means for a system to be simple or complex, revealing that the most interesting physics often lies in the connections we cannot see.

Technical Summary: Renormalization Group and Long-Range Conditional Mutual Information in Hierarchical Models

Problem Statement
The paper addresses the challenge of characterizing mixed quantum states that deviate from a local Gibbs description. While local observables often fail to detect such deviations, the Conditional Mutual Information (CMI), defined as $I(A:C|B) = I(A:BC) - I(A:B)$ , serves as a precise diagnostic for correlations not mediated by an intervening buffer region $B$ . A central question is how the Renormalization Group (RG) constrains the CMI, particularly regarding its ultraviolet (UV) finiteness. Specifically, the author investigates whether the existence of a locally reversible RG scheme guarantees that the CMI remains bounded in the scaling limit, and how this relates to the "Markov length" (the scale over which CMI decays). The work also explores the tension between states possessing divergent Markov lengths and those that flow to trivial fixed points under RG, as well as the stability of phases that satisfy local Markov conditions but violate global ones.

Methodology
The author employs a combination of information-theoretic proofs and exact calculations on hierarchical models:

Ideal RG Framework: They utilize the definition of an "ideal RG" scheme (following Ref. [43]), which consists of a sequence of finite-depth local channel circuits that are correlation-preserving and locally reversible. A channel $R$ is correlation-preserving if it maintains the mutual information between a coarse-grained region and its complement.
UV-Finiteness Proof: The author constructs a proof showing that if a mixed state admits an ideal RG with a fixed on-site Hilbert space dimension, the CMI between non-adjacent regions $A$ and $C$ (conditioned on $B$ ) is UV-finite. The proof involves three steps: (i) coarse-graining the bulk while preserving CMI via correlation-preserving gates; (ii) using Strong Subadditivity (SSA) to enlarge regions such that interface causal tubes are contained within single regions; and (iii) completing the RG within these tubes to bound the remaining CMI by the finite dimension of the coarse-grained sites.
Hierarchical Models: To test these theoretical constraints, the author analyzes two specific hierarchical models where RG transformations can be implemented exactly:
- Majority-Rule Hierarchical Ising Model: A variant of Dyson's hierarchical Ising model where spins are coarse-grained via a majority rule.
- SSB–SWSSB Tree Model: A tree-structured model constructed by concatenating rules generating the classical Spontaneous Symmetry Breaking (SSB) and Strong-to-Weak Symmetry Breaking (SWSSB) fixed points.

Key Contributions and Results

UV-Finiteness of CMI under Ideal RG:
The paper proves that for any mixed state admitting a locally reversible RG with a fixed finite on-site Hilbert space dimension at every scale, the CMI is UV-finite. This means $I(A:C|B)$ remains uniformly bounded in the scaling limit where the buffer size $N_B \to \infty$ while the ratios of region sizes are fixed. This result establishes a direct link between the existence of a specific type of RG flow and the finiteness of long-range conditional correlations.
Majority-Rule Hierarchical Ising Model:
- Divergent Markov Length vs. Trivial Flow: The author demonstrates that this model exhibits a divergent Markov length at every finite temperature $0 < T < \infty$ , meaning the CMI decays only algebraically with the buffer size.
- RG Flow: Despite the divergent Markov length, an ideal RG can be implemented exactly, driving the system to the trivial infinite-temperature product state.
- Phase Equivalence Nuance: The study reveals a distinction in phase definitions based on circuit range. The state is connected to the trivial product state by circuits with sublinear range (satisfying the criteria of Ref. [43]), but not by circuits with polylogarithmic range (the stricter criterion of Ref. [10]). This highlights that a divergent Markov length does not necessarily obstruct local reversibility in the broader sense, but it does prevent equivalence under stricter locality constraints.
SSB–SWSSB Tree Model:
- Local vs. Global Markov Properties: This model satisfies the local Markov condition (CMI vanishes for microscopic $A$ and $C$ ) but violates the global Markov condition (CMI remains non-zero, specifically $\ln 2$ , for macroscopic $A$ and $C$ separated by a subextensive buffer).
- Stability: The zero-noise fixed point is stable against weak local perturbations (noise), unlike the SWSSB phase which is destabilized by symmetry-breaking perturbations.
- Critical Behavior: At a critical noise strength $p_c \approx 0.2802$ , the model exhibits a unique critical point where the two-point Mutual Information (MI) vanishes identically, yet the two-point CMI decays only polynomially with system size ( $N^{-\alpha}$ with $\alpha \approx 1.5$ ). This contrasts with equilibrium criticality where MI typically decays algebraically.
- Global CMI Scaling: In the global tripartition (macroscopic $A, C$ ), the CMI is $\ln 2$ for $p < p_c$ and vanishes for $p > p_c$ .

Significance and Claims
The paper claims to resolve the relationship between RG and CMI by establishing that the existence of an ideal RG with fixed local dimension is a sufficient condition for CMI UV-finiteness. This provides a rigorous criterion for identifying genuine long-distance properties in mixed states.

The author emphasizes that their results challenge the intuition that a divergent Markov length implies a non-trivial phase or a lack of reversibility. The majority-rule model serves as a counterexample where long-range conditional correlations persist (divergent Markov length) even though the RG flow leads to a trivial state. Furthermore, the SSB–SWSSB tree model provides a concrete example of a stable phase that breaks the global Markov property while preserving the local one, offering a new perspective on the stability of mixed-state phases and the nature of nonequilibrium critical points.

The work suggests that CMI is a robust diagnostic for nonequilibrium phenomena, capable of detecting transitions that local observables and standard mutual information miss. The author concludes that their findings open avenues for studying learnability in regimes with infinite Markov lengths and polynomially decaying CMI, as well as exploring UV-finite quantities in systems with branching RG structures (like Fermi surfaces or fracton systems) where the on-site dimension grows.

Technical Summary: Renormalization Group and Long-Range Conditional Mutual Information in Hierarchical Models

More like this