{"id":"a5e22c64-9157-4b8b-8128-35d95d539b0e","arxiv_id":"2607.20421","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"For the Tanner-Hamming [7,4,3] LDPC model on a high-girth random regular graph, low-energy valleys separate canonical and microcanonical instability, yielding ensemble inequivalence in a non-random, unfrustrated spin glass.","lead":"Classical LDPC codes on random expander graphs are shown to harbor stable glassy valleys; a specific Tanner-Hamming code is found to have low-energy valleys that canonically destabilize before they microcanonically destabilize. The result combines tree-recursion analytics with flat-histogram Monte Carlo and matters for understanding spin-glass order without disorder and for LDPC-based error correction.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Microcanonical stability boundary relies on a finite-size-drifted lower bound; no extrapolation shows ε_micro persists above ε_canonical in thermodynamic limit.","rationale":"The reader's weakest assumption focuses on the 1D reaction-coordinate projection, arguing that a hidden escape path would make the reported ε_micro an artifact. My analysis shows that the projection actually yields a rigorous lower bound on the true barrier, so hidden structure cannot invalidate the stability certificate in the direction claimed; it would only strengthen it. The genuine threat to the central claim is the unextrapolated, significant finite-size drift of this lower bound. The reader did flag this drift as a symptom, but attributed it to the projection; I see it as the primary concern independent of the projection. Because the paper is heavily hedged ('consistent with hypothesis', 'suggestive evidence') and the SM already provides a partial escape-time check at N_v = 1298, the conditional verdict remains appropriate. No internal inconsistency or fatal error was found; the request for a finite-size scaling analysis is an addressable condition. Thus I recommend no change to the reader's CONDITIONAL verdict, while noting that the load-bearing concern is the finite-size extrapolation, not the projection per se.","tokens_in":34123,"tokens_out":10221,"duration_ms":131511,"concrete_test":"Extract the barrier-vanishing energy density ε_micro(N) for the ground-state valley (ε_min = 0) from SM Fig. S7a at the three available sizes N_v = 218, 1298, 7778 (girths 4, 5, 6). Fit ε_micro(N) to ε_micro(∞) + a N_v^{-b} (and also as a function of inverse girth) with proper uncertainty propagation. If the extrapolated ε_micro(∞) is not strictly greater than ε_canonical (≈ ε_ferro(x_mem) from the tree calculation) within error bars, the central claim of a persistent ensemble inequivalence fails. Additionally, for ε_min > 0, the absence of data beyond N_v = 1298 means no extrapolation is possible; a minimal extension would be to measure ΔS at N_v = 7778 for one nonzero ε_min to test whether the barrier-vanishing energy continues to drift downward.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim of a microcanonically stable yet canonically unstable regime for low-ε_min valleys depends on the entropy barrier ΔS(E) = S(E,M*(E)) − S(E,M_barrier(E)) computed from the one-dimensional reaction coordinate M(σ) (SM Sec. VI.C, Eq. S56/S64). The paper correctly labels ΔS a lower bound on the true barrier: because any escape path must cross every M slice, the transition-state entropy cannot exceed S(E,M) at the barrier M, so hidden orthogonal structure (SM Fig. S6) would only make the true barrier larger, not smaller. Thus the projection issue by itself is conservative. The load-bearing problem is the finite-size behavior of this lower bound. SM Fig. S7a shows that for ε_min = 0 the barrier-vanishing energy drifts significantly downward with N_v = 218, 1298, 7778, and for ε_min > 0 the data exist only at N_v = 1298 (SM Fig. S7b,c). No finite-size extrapolation is provided, and the paper itself calls the ε_micro curve a 'rough estimate' and 'conjecture'. If the drift continues to larger N, the extrapolated ε_micro could cross below ε_canonical, eliminating the proposed intermediate phase. The escape-time comparison quoted in SM Sec. VI.C is a useful check, but it is limited to ε_min = 0, N_v = 1298, and is described only as 'qualitative'. Therefore the thermodynamic-limit existence of the separation is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a classical LDPC model—the Tanner-Hamming [7,4,3] code on locally tree-like random regular graphs—as a statistical-mechanics model with glassy behavior but no quenched disorder or frustration. Using cavity-method recursion with a two-parameter (α,γ) family of boundary conditions, the authors derive a low-temperature expansion for the bulk free energy, energy, and configurational entropy, and construct high-energy valleys via fine-tuned node compositions. On closed graphs, adaptive multicanonical Monte Carlo is used to detect (i) the canonical instability (vanishing of the within-valley free-energy minimum) and (ii) a lower bound on the microcanonical entropic barrier using a one-dimensional overlap coordinate M(σ). The central claim is that for valleys with low minimum energy density ε_min, the canonical instability occurs at an energy below the microcanonical barrier-vanishing energy, giving an intermediate regime that is canonically unstable but microcanonically stable—an inequivalence of ensembles. The paper presents the phase diagram as a heuristic and labels the microcanonical boundary a 'rough estimate' and 'conjecture'.","tokens_in":34433,"tokens_out":6618,"duration_ms":55963,"significance":"If the central claim were established in the thermodynamic limit, the model would provide a concrete, non-random, unfrustrated example of microcanonical/canonical ensemble inequivalence in a finite-connectivity system with extensive barriers, complementing the rigorous spin-glass results of Ref. [35] and connecting to quantum LDPC constructions. The paper's strengths include a transparent analytic low-temperature expansion cross-checked against population dynamics and the exact α=γ=2 second moment; a clear and honest acknowledgment that the intermediate phase is a hypothesis; and careful numerical methodology with multiple system sizes (N_v = 218, 1298, 7778) and explicit algorithmic details. The tree-level calculations and the agreement between the codeword-polarized tree prediction for the canonical instability and the closed-graph data at N_v = 7778 are convincing.","major_comments":[{"comment":"The existence of the intermediate regime (canonically unstable, microcanonically stable) depends on ε_micro > ε_canonical in the thermodynamic limit. The quantity ε_micro is read off from the vanishing of the lower-bound barrier ΔS(E) in SM Eq. (S64). For ε_min=0, SM Fig. S7a shows a significant downward drift of the barrier-vanishing energy with increasing N_v (218 → 1298 → 7778); for ε_min>0, the data exist only at N_v=1298. No finite-size extrapolation is provided, and the paper itself labels ε_micro a 'rough estimate'. If the downward drift persists, extrapolated ε_micro could fall below ε_canonical, eliminating the proposed intermediate phase. The thermodynamic-limit existence of the ensemble inequivalence is therefore not established.","section":"Main text, 'Microcanonical ensemble'; SM Sec. VI.D, Fig. S7"},{"comment":"Even apart from finite-size effects, ε_micro as defined is the energy where a lower bound on the true entropic barrier vanishes, not necessarily the energy where the true barrier vanishes. The paper's own Fig. S6 shows that a one-dimensional projection can overestimate the entropy at the barrier coordinate, making the bound less tight; the paper notes this in SM Sec. VI.D. Consequently, the reported ε_micro may underestimate the true microcanonical instability. A positive separation at finite N_v therefore does not directly imply a positive separation in the thermodynamic limit; the manuscript should either supply a scaling analysis of ε_micro or clearly restrict the claim to finite systems.","section":"Main text, 'Microcanonical ensemble' and Fig. 3; SM Eq. (S64), Fig. S6"}],"minor_comments":[{"comment":"The abstract states that the numerical study 'reveals the inequivalence of the microcanonical and canonical ensemble', whereas the main text (Discussion and 'Dynamics on closed graphs') describes 'suggestive evidence' and a 'hypothesis'. Given the finite-size caveats, the abstract overstates the current support; recommend softening to 'provides evidence for' or 'suggests'.","section":"Abstract"},{"comment":"The thick light-blue curve is called a 'rough estimate' of the barrier-vanishing energy. Since this curve is load-bearing for the central claim, the main figure should display the finite-size data points (or at least the range of drift seen in SM Fig. S7) directly on the phase diagram, so the reader can assess the uncertainty.","section":"Fig. 3a"},{"comment":"The caption states that error bars are smaller than markers for N_v=218 and 1298, but does not comment on N_v=7778. Clarify whether error bars are shown for the largest size and how the drift is quantified.","section":"SM Sec. VI.D, Fig. S7 caption"},{"comment":"The subscript notation for ε_min is inconsistent between the main text (ε_min) and the SM (εmin, e.g., Sec. VI.A). Unify the notation throughout.","section":"Notation"}],"recommendation":"major_revision","confidential_remarks":"The paper is honest and technically careful, and the tree-level analysis is a solid contribution. The main obstacle is that the central numerical claim—the thermodynamic-limit separation of ε_micro and ε_canonical—is explicitly conjectural and rests on a finite-size-drifted lower bound. The authors can address this either by providing a finite-size scaling analysis (even suggestive) or by consistently framing the result as evidence on finite graphs. Given the 'reveals' wording in the abstract, I recommend major revision rather than rejection: the claim as stated exceeds what the data establish, but the underlying methodology and the qualitative picture are valuable and likely correct at low ε_min."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the canonical-instability half is in good shape: the two-parameter (α,γ) boundary-condition family is new, the low-temperature expansion is internally consistent and cross-checked (population dynamics, the exact γ=α=2 second moment), and the multicanonical data on N_v=7778 agree with the tree calculation for ε_canonical — that is a prediction-comparison, not a fit. Second, the marquee claim, a microcanonically stable but canonically unstable regime for low-ε_min valleys, is genuinely suggestive but not established in the thermodynamic limit.\n\nWhat is new: the (α,γ) BC family, the high-energy valley construction (type 0/1/2 nodes) giving bounds on the stable region, and the numerical evidence for the ε_canonical/ε_micro separation. Ref. [35] proved spin-glass order under code-expansion conditions; this paper maps the landscape for one code that does not meet those conditions. The paper is honest about that: it says code expansion and subextensive redundancies are unproven for Tanner-Hamming [7,4,3].\n\nSoft spots, in proportion. The microcanonical boundary ε_micro is estimated from ΔS(E) = S(E,M*) − S(E,M_barrier), a lower bound on the true entropic barrier. The worry that the one-dimensional reaction coordinate hides the true escape path is actually conservative — any escape must cross every M slice, so the projection makes the barrier estimate smaller, not larger. The load-bearing problem is finite-size drift: SM Fig. S7a shows the barrier-vanishing energy drifting downward for ε_min=0 as N_v goes 218→1298→7778, and for ε_min>0 the data exist only at N_v=1298, with no extrapolation. If the drift continues, ε_micro could cross below ε_canonical and the intermediate phase disappears. The paper itself calls ε_micro a 'rough estimate' and 'conjecture', so the language is honest, but the claim is not yet backed by a thermodynamic-limit argument. The escape-time check is qualitative and limited to ε_min=0, N_v=1298. No code or data are supplied, which matters for a numerics-heavy paper.\n\nWho this is for: people working on glassy landscapes without disorder or frustration, and the quantum LDPC community interested in passive error correction. A serious referee should engage — this is a carefully executed case study with a new toolset, and the ensemble-inequivalence observation is worth testing further. It deserves peer review rather than desk rejection, with the microcanonical section flagged for stronger finite-size analysis before publication.","headline":"A careful, honest case study whose canonical-instability result is solid; the marquee microcanonical-canonical gap is suggestive but rests on a finite-size-drifted lower bound with no extrapolation to back it.","tokens_in":34996,"tokens_out":3098,"would_cite":true,"duration_ms":26381,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper argues that in a disorder-free classical LDPC code on an expander graph, low-energy valleys of the energy landscape can be thermally (canonically) unstable while remaining protected by an extensive entropy barrier at fixed energy,","keywords":["low-density parity-check code","spin glass without disorder","energy landscape valleys","microcanonical versus canonical ensemble","entropy barrier","expander graph","cavity method / tree recursion","flat-histogram Monte Carlo"],"falsifier":"A calculation or simulation that exhibits a valley and an explicit escape path at an energy density below the predicted ε_micro while keeping the reaction coordinate M above its barrier value, or an escape-time measurement showing τ(E) does not grow exponentially with ΔS(E)/T, would disprove the claim that low-energy valleys are microcanonically stable.","tokens_in":33934,"feed_emoji":"🧊","tokens_out":2706,"duration_ms":26098,"temperature":0.7,"pith_summary":"The paper studies a classical low-density parity-check code, which is a spin-glass model without quenched disorder or frustration, and maps out the stability of its energy valleys. It claims that valleys with low minimum energy leave the valley at a canonical instability energy that is lower than the energy where the microcanonical entropy barrier vanishes. If true, this is a concrete demonstration of nonequivalence between canonical and microcanonical ensembles in a finite-connectivity glassy system with extensive barriers. The authors support this with tree recursion calculations and flat-histogram Monte Carlo on closed random regular graphs.","feed_headline":"Entropy barriers keep canonically unstable valleys frozen","feed_subtitle":"In a disorder-free Gaussian-like spin glass, valleys that escape when heated remain trapped at fixed energy, showing ensemble inequivalence.","key_machinery":"Valleys are parameterized by their minimum energy density ε_min. The canonical instability is detected as an inflection point in the within-valley free energy F_valley(E,T), extracted from an adaptive multicanonical Boltzmann factor x(E) whose local maximum gives ε_canonical. The microcanonical instability is probed through the entropy barrier ΔS(E) = S(E,M*(E)) − S(E,M_barrier(E)), where M(σ) is a single reaction coordinate measuring overlap with the valley bottom; ΔS is a lower bound on the true entropic barrier. Tree-based recursion with (α,γ) boundary conditions and a low-temperature expansion in the defect density supplement the closed-graph simulations.","core_discovery":"The central claim is that valleys of the Tanner-Hamming [7,4,3] cLDPC model on locally tree-like expander graphs exhibit a regime where the valley is canonically unstable yet microcanonically stable: heating the valley makes it escape at an energy density ε_canonical, but a state constrained to a fixed energy density below ε_micro cannot leave because of an extensive entropic barrier. The numerical evidence, stated in the main text, is 'consistent with the hypothesis that as ε_min increases, the entropy barrier at a given ε decreases, but ε_micro and ε_canonical remain separated for low ε_min.'","pith_inferences":["If this ensemble inequivalence is generic, it should appear in other sparse glassy models such as Gallager codes and diluted p-spin models, which share the same structure of extensive barriers and non-random interactions.","The single-reaction-coordinate estimate of the entropy barrier is a lower bound; a stronger test would be to compute escape times and compare their scaling with ΔS, or to search for escape paths orthogonal to M that would invalidate the reported ε_micro.","The finite-size drift in the barrier-vanishing energy (seen in the Supplemental Material) leaves open the possibility that the separation between ε_canonical and ε_micro shrinks or vanishes at larger system sizes, which is directly testable.","An analogous thermally-unstable-but-microcanonically-stable regime might exist in quantum LDPC codes, which are built from products of classical LDPC codes and exhibit topological quantum spin glass order."],"forward_implications":["If the central claim is correct, canonical and microcanonical phase boundaries differ for these valleys, so the equilibrium Gibbs measure at a given temperature would leave the valley while energy-constrained dynamics would remain trapped.","The canonical instability energy matches the codeword-polarized (memory) transition temperature computed on trees, suggesting that the passive error-correction threshold is set by the canonical instability, while microcanonical decoding might operate at higher energies.","The paper's phase diagram implies that high-energy valleys (large ε_min) lose the canonical instability and instead escape only via a vanishing entropy barrier, so there is a bounded region of 'intermediate' valleys.","Fine-tuned boundary conditions on trees produce a family of high-energy valleys with a downward cusp in energy at their transition, bounding the region of thermally stable valleys and confirming a rich landscape beyond codeword valleys."],"fun_headline_variants":["Entropy barriers freeze valleys that heat would melt","Glass valleys stay put at fixed energy, escape when heated","Microcanonical stability beats canonical in glassy code","Heating frees valleys, but entropy locks them at fixed energy"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The conclusion that a valley is microcanonically stable relies on estimating the entropy barrier from a single reaction coordinate M(σ); if the true escape path is hidden by this projection, the reported ε_micro is an artifact.","fun_headline_variants_meta":{"raw":{"variants":["Entropy barriers freeze valleys that heat would melt","Glass valleys stay put at fixed energy, escape when heated","Microcanonical stability beats canonical in glassy code","Heating frees valleys, but entropy locks them at fixed energy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000712,"raw_usage":{"total_tokens":2994,"prompt_tokens":651,"completion_tokens":2343,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":395,"completion_tokens_details":{"reasoning_tokens":2285}},"tokens_in":395,"tokens_out":2343,"duration_ms":14196,"temperature":1.0,"reasoning_tokens":2285,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T09:51:51.285331+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A calculation or simulation that exhibits a valley and an explicit escape path at an energy density below the predicted ε_micro while keeping the reaction coordinate M above its barrier value, or an escape-time measurement showing τ(E) does not grow exponentially with ΔS(E)/T, would disprove the claim that low-energy valleys are microcanonically stable.","supporting_citations":[],"review_version":1}