{"id":"79dc2a08-5e1d-43bc-a579-1787c803763b","arxiv_id":"2609.02141","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Existence of an ideal (locally reversible) renormalization group forces the conditional mutual information to be UV-finite, and two hierarchical models demonstrate that long-range CMI and simple RG flow can coexist.","lead":"This paper proves that the conditional mutual information (CMI) is bounded in large systems whenever a quantum or classical state can be coarse-grained by a locally reversible renormalization group with fixed local Hilbert space dimension. It then constructs two explicit hierarchical models that develop long-range CMI yet flow to simple states under RG, showing that such RG constraints can coexist with divergent Markov lengths and global Markov violations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"No significant objection identified; the proof appears sound under the stated assumptions.","rationale":"The reader's verdict is ACCEPT with HIGH confidence and identifies the weakest assumption correctly (existence of ideal RG). My stress-test finds no new load-bearing concern. The proof is logically consistent given its definitions, and the paper transparently discusses the conditions and models. The two examples satisfy the theorem's condition (majority-rule model) or are treated separately (SSB-SWSSB tree). No internal inconsistency or hidden circularity is present. Therefore the reader's judgment stands.","tokens_in":28006,"tokens_out":6997,"duration_ms":59743,"concrete_test":"Verify the preservation of I(A:BC) under a single gate acting within A, using the recovery argument on the marginal ABC. For a simple one-dimensional binary MERA circuit (b=2), implement a single isometric gate on two qubits in region A, with the recovery channel being its adjoit on the same two qubits. Compute the original and transformed I(A:BC) for a random input state; confirm equality to floating-point precision. This tests the crucial step that gates in the partial circuit preserve the CMI.","verdict_should_be":"UNCHANGED","load_bearing_attack":"After careful examination of the central claim and its proof (Section II A, Eqs.(5)-(10), I do not identify a load-bearing flaw. The argument that an ideal RG with fixed finite on-site Hilbert space dimension implies UV-finite CMI is logically structured: (i) coarse-grain the bulk via the partial circuit, (ii) enlarge regions using SSA, (iii) complete the interface tubes to O(1) sites, yielding a bound independent of system size. The subtle step concerning whether a correlation-preserving gate acting inside A preserves I(A:BC) and I(A:B) can be closed by noting that the gate on X ⊆ A is part of a reversible channel on the marginal ABC (since the recovery acts only on X and commutes with tracing out D), hence preserves all mutual informations on ABC. Lemma 1 in Appendix A correctly argues that retained gates remain correlation-preserving on the partial-circuit input. The assumption of existence of an ideal RG is transparently stated and is satisfied by the constructed models. The paper honestly notes where the condition fails (e.g., SSB-SWSSB tree does not have fixed finite dimension). The reader's weakest_assumption correctly identifies the central condition; no further hidden assumption undermines the result.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"This paper studies the relationship between the renormalization group (RG) and the conditional mutual information (CMI). It proves that the CMI between nonadjacent regions is UV-finite whenever the state admits an ideal (locally reversible) RG with a fixed finite on-site Hilbert space dimension. It then analyzes two hierarchical models that have long-range CMI: a majority-rule hierarchical Ising model with a divergent Markov length at every nonzero temperature, and an SSB–SWSSB tree that violates the global Markov condition while satisfying the local one. Both models are solved exactly using the RG, and the second model exhibits a phase transition under noise at which the CMI decays polynomially while the mutual information vanishes identically.","tokens_in":28267,"tokens_out":3327,"duration_ms":28856,"significance":"The paper establishes a rigorous connection between the existence of an ideal RG and the UV-finiteness of CMI, identifying CMI as a genuine long-distance property that is invariant under RG in the scaling limit. The hierarchical models provide clean, exactly solvable examples that illustrate subtle distinctions between local and global Markov properties, between different RG schemes, and between the behaviors of CMI and mutual information. These results are relevant to the classification of mixed-state phases, the stability of quantum memories, and the learnability of quantum states.","major_comments":[],"minor_comments":[{"comment":"The decomposition of A', B', C', D' into bulk and in/out factors is clear in the text but could benefit from a brief remark that the in/out factors contain the interface degrees of freedom left over from all RG scales.","section":"§II A, Eq. (6)"},{"comment":"The base of the logarithm in entropy and mutual information is not explicitly stated. It appears to be natural log (nats) from the form of the formulas (e.g., Eq. (10) mentions log d*). Please state the convention explicitly at first use.","section":"General"},{"comment":"Fig. 2 is a bit dense; consider adding a color legend for the tubes and the coarse-grained regions.","section":"Fig. 2"},{"comment":"In the proof of Lemma 1, the argument that the recovery channel D acts outside X and therefore commutes with the undoing of removed gates is correct, but the chain of reasoning could be made more explicit: write σ as the state after applying the recovery channels of the removed gates, and then note that D still recovers R because the recovery channels act on the complement of X.","section":"Appendix A"},{"comment":"The inequality I(A:η|B) ≤ H(η) follows from H(η|B) ≤ H(η) and the fact that conditioned entropy is nonnegative. This is fine, but note that η is a classical bit, so H(η) ≤ log 2. The bound is therefore at most log 2.","section":"§IV D, Eq. (37)"},{"comment":"The paper uses 'UV-finite' to mean uniformly bounded in the scaling limit; this is slightly nonstandard (usually UV-finite means no divergence in the continuum limit). Consider clarifying the definition early on (it is defined in the text, but a brief note would help).","section":"§II A"}],"recommendation":"accept","confidential_remarks":"This is a well-written paper that provides a clear theoretical result and two instructive exactly solvable examples. The proof appears rigorous, and the model calculations are exact. I recommend acceptance without hesitation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"This paper does two things, both well. First, it proves that conditional mutual information is UV-finite whenever the state admits an ideal RG (locally reversible, fixed finite on-site Hilbert space dimension). The proof is clear: use the bounded causal tube, strong subadditivity, and a dimension count. This is a genuine new result—it takes the idea of correlation-preserving RG from Sang et al. 2024 and turns it into a crisp bound. Second, it constructs two hierarchical models that satisfy (or nearly satisfy) the ideal-RG condition and show what happens when the Markov length diverges or when local vs. global Markov properties decouple. The models are exactly solvable and the calculations are explicit. The paper is transparent about where the assumption holds and where it does not (e.g., the SSB–SWSSB tree briefly violates the finite-dimension requirement, and the author flags this).\n\nSoft spots: The central assumption—existence of an ideal RG—is strong. The paper acknowledges this, and the models are built to satisfy it, so the theorem applies to them by construction. But this means the theorem’s scope is limited to states with that very special structure. The paper does not offer a way to check the condition for a generic state, which is left as future work. That is honest but limits immediate impact. Also, the UV-finiteness bound is independent of system size but could depend on blocking factor, dimension, etc.; that’s fine, but the bound is not tight enough to be used as a diagnostic unless the RG is also known. The model analyses are exact but rely on the hierarchical architecture; whether the lessons generalize to translation-invariant local Hamiltonians is unclear, and the paper does not overclaim this.\n\nWho the paper is for: Researchers working on information-theoretic diagnostics for mixed-state phases, RG for open systems, and exactly solvable models that test the boundaries of Markov length. The writing is clear and the figures are helpful.\n\nPeer review recommendation: This paper should be sent to a good journal (e.g., PRB or SciPost Physics). The proof is sound, the models are instructive, and the limitations are stated. It is a genuine contribution that advances the conversation about CMI and RG.","headline":"Well-crafted paper: clean theorem linking ideal RG to bounded CMI, plus exactly solvable models that test the limits. Honest about assumptions. Deserves a serious referee.","tokens_in":28836,"tokens_out":904,"would_cite":false,"duration_ms":9879,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":{"msc":["81P45","82B28","82B20"],"pacs":["03.67.-a","64.60.ae","05.70.Jk"],"model":"deepseek/deepseek-v4-flash","headline":"The paper proves that if a mixed quantum state admits an ideal renormalization group—an exactly reversible coarse-graining by local channels with fixed on-site Hilbert space dimension—then its conditional mutual information is bounded…","keywords":["conditional mutual information","renormalization group","mixed quantum states","Markov property","hierarchical models","strong-to-weak symmetry breaking","Markov length","many-body phases"],"falsifier":"Simulate the SSB–SWSSB tree at the claimed critical noise $p_c \\backsimeq 0.2802$ for sizes well beyond $6^5$, for instance up to $6^8$ or $6^{10}$, by population dynamics or direct sampling: the claim is a clean power-law decay of the local CMI, $N^{-\\ts{alpha}}$ with $\\ts{alpha} \\backsimeq 1.5$, and identically vanishing two-point mutual information at every noise strength; a log-log plot that bends away from a straight line, or a nonzero two-point MI at any $p$, would rule out the critical description. For the main theorem, the falsifier would be a state that provably admits an ideal RG with fixed on-site dimension but whose CMI in the annular tripartition grows without bound with the logarithmic size parameter, directly contradicting the explicit bound $2n_* \\ts{log} d_*$.","tokens_in":29705,"feed_emoji":"🌳","tokens_out":10470,"duration_ms":85167,"temperature":0.7,"pith_summary":"The paper establishes a constraint that the renormalization group places on the conditional mutual information (CMI), the information that two distant regions share beyond what the intervening region already explains. Its main theorem: if a mixed quantum state admits an ideal RG—a coarse-graining by short-range channels that can be exactly undone, on that state, by local channels, with the on-site Hilbert space dimension kept fixed at every scale—then $I(A:C|B)$ is UV-finite, bounded by a constant independent of system size as the regions are scaled up at fixed shape. This matters because CMI is the standard diagnostic for how far a state departs from a local Gibbs (thermal) description, and UV-finiteness is exactly what is needed for CMI to serve as a genuine long-distance, RG-invariant quantity rather than a lattice artifact. The paper then constructs two exactly solvable hierarchical models showing that long-range conditional correlations can coexist with an ideal RG: one has a divergent Markov length at every finite temperature yet flows to the infinite-temperature product state, and one violates the global Markov property while respecting the local one, resists weak noise, and at its critical noise strength shows power-law CMI decay with identically vanishing two-point mutual information.","feed_headline":"Reversible renormalization caps a state's hidden correlations","feed_subtitle":"Exact coarse-graining keeps conditional mutual information bounded, even when the Markov length diverges.","key_machinery":"The central object is the ideal (locally reversible) RG scheme: a sequence of coarse-graining maps $\rho^{(r)} \to \rho^{(r+1)}$, each a finite-depth circuit of local channels that are correlation-preserving, $I_{\rho'}(X':ar{X}) = I_{\rho}(X:ar{X})$, a condition equivalent to each coarse-graining step being exactly undone by a fine-graining channel acting only on the coarse-grained region. The proof's workhorse is the bounded causal-tube property: because each RG layer is local at its own scale and shrinks linear size by a factor $b>1$, the causal closure of a fixed-width boundary collar is a multiscale tube of uniformly bounded transverse width containing $O(1)$ sites per horizontal cut. Strong subadditivity enters twice—enlarging regions to absorb the tubes can only increase the CMI, and the final bound is $2\ts{min}\ts{ln}\ts{dim} A''$, $\ts{log}\ts{dim} C''\\} \ts{le} 2n_* \ts{log} d_*$. For the tree model the machinery is the posterior probability $q(S) = P(\ts{sigma}=+1|S)$ of the hidden parent spin, which the paper shows is exactly the variable that preserves correlations and obeys an exact recursion $E_{\ts{ideal}}(q_1,\ts{ldots},q_6)$; since $q$ is continuous, the fixed-dimension hypothesis of the theorem fails there, and UV-finiteness is instead proven directly by a strong-subadditivity argument giving $I(A:C|B) \ts{le} \ts{log} 2$.","core_discovery":"On the paper's own terms, the core result is a theorem plus two model constructions. The theorem: for fixed-shape annular tripartitions $A|B|C$ in $d$ dimensions, if a state $\rho$ admits an ideal RG—a finite-depth circuit of local correlation-preserving channels at each scale, each step exactly undoable by a channel acting only on the coarse-grained region, with fixed finite on-site dimension $d_*$—then the CMI is bounded by $2 n_* \\log d_*$ with $n_*=O(1)$ sites, uniformly in the scaling limit $N_A/N_B \to x_A$, $N_C/N_B \to x_C$, $N_B \to \\infty$. The proof removes the RG gates in the interface causal tubes, coarse-grains the bulk without changing the CMI, enlarges regions so each tube lies inside a single region (which can only increase the CMI by strong subadditivity), then completes the tubes and bounds the remainder by the dimension of the surviving $O(1)$ coarse sites. The first model, a majority-rule variant of Dyson's hierarchical Ising model, implements an exactly locally reversible RG at every scale with $T \to 3^{1/\nu} T$, flowing to the infinite-temperature product state, yet its CMI decays only algebraically in the buffer size, so the Markov length diverges at every finite $T$; sublinear-range circuits connect it two-way to the product state, while polylogarithmic-range circuits cannot. The second model, a tree concatenating the classical SSB repetition rule with the SWSSB parity rule (the classical counterpart of the Bell tree), has vanishing local CMI for $N_A/N_B < 2$ but global CMI equal to $\text{ln} 2$ for any subextensive buffer; it carries a strong Z$_2$ symmetry yet is stable against noise, in contrast to SWSSB, and at the critical noise strength $p_c acksimeq 0.2802$ the local CMI decays as $N^{-\ts{alpha}}$ with $\ts{alpha} acksimeq 1.5$ while the two-point mutual information vanishes identically.","pith_inferences":["The paper's learnability discussion points toward a testable prediction: a state whose CMI decays polynomially but whose RG flows to a trivial fixed point (the majority-rule model) should be easier to learn from measurements than a genuinely nonequilibrium critical point with the same Markov-length behavior; whether a single estimator separates the two is left open.","Read as a resource theory, the theorem says any RG that must branch into multiple copies—Fermi surfaces with branching MERA, or fracton systems—escapes the bound precisely because the effective on-site dimension grows; quantifying a branching number per RG step would extend the UV-finiteness criterion to those states.","A natural extension is to discretize the continuous posterior variable $q(S)$ of the ideal tree RG onto a finite alphabet; if a finite-alphabet approximation remains approximately correlation-preserving, the theorem's fixed-dimension hypothesis would apply directly to the noisy tree, and the approximation error could be compared with the critical noise strength $p_c$.","The sharp plateau structure of the global CMI ($\\log 2$ for $p < p_c$, 0 for $p > p_c$) acts as an order parameter for a transition invisible to local observables and to two-point correlations, suggesting that CMI measurements on subextensive buffers could serve as a practical detector of the phase boundary."],"forward_implications":["CMI becomes a genuine long-distance property: given an ideal RG it is invariant when all three regions are rescaled together, so the scaling limit is described by a scaling function $f(N_A/N_B, N_C/N_B, \\{t_i N_B^{1/(\\nu_i d)}\\})$ without finite-size corrections.","A divergent Markov length does not block locally reversible RG flow: the majority-rule hierarchy connects to the infinite-temperature product state in both directions by sublinear-range circuits, yet is not phase-equivalent to it under the polylogarithmic-range circuit definition, so the two phase classifications genuinely differ.","Stability of mixed-state phases is tied to the global, not local, Markov property: the SSB–SWSSB tree breaks the global Markov condition while satisfying the local one and is stable against noise, unlike SWSSB, and strong subadditivity then forces the global CMI to be exactly $\\log 2$ below $p_c$ and exactly 0 above it.","The tree's critical point is a one-dimensional example where the two-point mutual information vanishes identically while the CMI decays only polynomially, a pattern not present at equilibrium criticality, where mutual information decays algebraically and CMI is short-ranged."],"supporting_citations":[{"why":"Supplies the definition of the correlation-preserving, locally reversible 'ideal RG' scheme that is the central hypothesis of the main theorem.","marker":"[43]"},{"why":"Provides the polylogarithmic-range circuit definition of mixed-state phases and the local Markov-length stability criterion that the majority-rule example is tested against.","marker":"[10]"},{"why":"Shows the CMI scaling function is an RG monotone in nonequilibrium systems under a scaling assumption; the UV-finiteness theorem here supplies the missing justification for that assumption.","marker":"[36]"},{"why":"Dyson's hierarchical Ising model is the template for the majority-rule hierarchy whose RG can be implemented exactly.","marker":"[49]"},{"why":"The Bell-tree models whose classical counterpart the SSB–SWSSB tree is, contributing the repeated-encoding construction used for the second example.","marker":"[52, 53]"},{"why":"Establishes the instability of SWSSB states under strong-symmetry-breaking perturbations, the contrast against which the tree's stability is established.","marker":"[48]"},{"why":"Proves that a finite Markov length suffices for efficient learning of a quantum state, marking the regime (polynomial CMI decay with divergent Markov length) that the paper's models inhabit and its outlook targets.","marker":"[82]"}],"fun_headline_variants":["Reversible RG bounds hidden correlations in quantum states","Exact coarse-graining caps conditional mutual information","Long-range correlations persist under reversible renormalization","Renormalization tames nonlocal correlations in hierarchical models","CMI stays bounded when a reversible RG exists"],"cache_read_input_tokens":0,"weakest_assumption_plain":"Everything rests on assuming the state admits an ideal RG at all: a coarse-graining by local channels that can be exactly reversed, on that state, by local channels acting only on the coarse-grained region, with the on-site Hilbert space dimension never growing; the paper constructs its two models precisely to satisfy this condition and does not prove existence for generic states.","fun_headline_variants_meta":{"raw":{"variants":["Reversible RG bounds hidden correlations in quantum states","Exact coarse-graining caps conditional mutual information","Long-range correlations persist under reversible renormalization","Renormalization tames nonlocal correlations in hierarchical models","CMI stays bounded when a reversible RG exists"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000267,"raw_usage":{"cost":0.0002667,"is_byok":false,"cost_details":{"upstream_inference_cost":0.0002667,"upstream_inference_prompt_cost":0.00019166,"upstream_inference_completions_cost":7.504e-05},"total_tokens":1637,"prompt_tokens":1369,"completion_tokens":268,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":0,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":209}},"tokens_in":1369,"tokens_out":268,"cost_source":"openrouter_usage","duration_ms":3752,"temperature":1.0,"reasoning_tokens":209,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-09-08T22:32:12.410735+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"falsifier":"Simulate the SSB–SWSSB tree at the claimed critical noise $p_c \\backsimeq 0.2802$ for sizes well beyond $6^5$, for instance up to $6^8$ or $6^{10}$, by population dynamics or direct sampling: the claim is a clean power-law decay of the local CMI, $N^{-\\ts{alpha}}$ with $\\ts{alpha} \\backsimeq 1.5$, and identically vanishing two-point mutual information at every noise strength; a log-log plot that bends away from a straight line, or a nonzero two-point MI at any $p$, would rule out the critical description. For the main theorem, the falsifier would be a state that provably admits an ideal RG with fixed on-site dimension but whose CMI in the annular tripartition grows without bound with the logarithmic size parameter, directly contradicting the explicit bound $2n_* \\ts{log} d_*$.","supporting_citations":[{"cited_title":"Mixed-state topological order and the errorfield double formulation of decoherence-induced transitions","cited_arxiv_id":"2301.05687","evidence_quote":"Supplies the definition of the correlation-preserving, locally reversible 'ideal RG' scheme that is the central hypothesis of the main theorem."},{"cited_title":"Information Critical Phases under Decoherence","cited_arxiv_id":"2512.22121","evidence_quote":"Shows the CMI scaling function is an RG monotone in nonequilibrium systems under a scaling assumption; the UV-finiteness theorem here supplies the missing justification for that assumption."},{"cited_title":"Patil and Andreas W","cited_arxiv_id":null,"evidence_quote":"Dyson's hierarchical Ising model is the template for the majority-rule hierarchy whose RG can be implemented exactly."},{"cited_title":"Casini and M","cited_arxiv_id":null,"evidence_quote":"Establishes the instability of SWSSB states under strong-symmetry-breaking perturbations, the contrast against which the tree's stability is established."},{"cited_title":"Fannes, A continuity property of the entropy density for spin lattice systems, Communications in Mathemati- cal Physics31, 291 (1973)","cited_arxiv_id":null,"evidence_quote":"Proves that a finite Markov length suffices for efficient learning of a quantum state, marking the regime (polynomial CMI decay with divergent Markov length) that the paper's models inhabit and its outlook targets."}],"review_version":1}