{"id":"d6638454-758e-4263-a68e-a41096879036","arxiv_id":"2507.10507","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"In the zero-temperature Edwards-Anderson model, the spectral measure of the two-point spin correlation assigns exponentially small mass to all edge sets whose size is at most (1+ε) times the distance, so the spectral sample is strictly larger than a shortest path with high probability.","lead":"This paper proves new bounds on the Fourier spectrum of spin-glass ground states in the Edwards-Anderson model, showing that the disorder correlations are spread over many more edges than the minimal path. The result is an early rigorous step toward confirming numerical predictions of fractal behavior in lattice spin glasses.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3.1's energy comparison omits exterior edges incident to the boundary; a single large exterior coupling can pin a boundary spin, so Theorems 2.2–2.3 are not established as written.","rationale":"The reader's weakest assumption is exactly the load-bearing concern: Lemma 3.1 assumes a local barrier forces boundary-spin alignment irrespective of exterior couplings, but equation (3.5) silently omits edges from boundary vertices to the exterior. I reviewed the full text carefully. The Fourier setup and the review of Chatterjee's Proposition 2.1 are standard and not the issue. The entirely new engine is the barrier argument: Lemma 3.1 is used in Lemma 3.2 to freeze the effect of the conditioned edge e* and obtain the zero conditional expectation (3.7). Without Lemma 3.1, the proof of Theorem 2.2 collapses, and the same barrier step is reused in Lemma 3.5 part 2 for Theorem 2.3. The flaw is internal to the proof: (3.5) is not a valid lower bound for all configurations satisfying the stated Barrier(e) event, because it ignores unbounded exterior couplings. This is not a disagreement with consensus; it is a missing term in a finite energy comparison. The paper's idea may be repairable by enlarging the barrier event to constrain all edges incident to the boundary path, at the cost of rechecking the constants, but as written the central claims are not established. Therefore I keep the reader's REJECT verdict; I would not move it to ACCEPT or CONDITIONAL without a corrected Lemma 3.1.","tokens_in":11610,"tokens_out":9344,"duration_ms":113929,"concrete_test":"Test Lemma 3.1 by exact enumeration on a small graph. Take a 5×5 box around the central edge e, set the boundary path edges (w_i,w_{i+1}) for i=1..9 to J=+100, set all red edges shown in Fig. 3.2 to J=+1, and add one exterior edge from w_1 to a vertex z outside the box with J=−1000; pin z by adding a very large coupling between z and a fixed anchor so that σ_z is forced. Enumerate all spin assignments to find the ground state and check whether σ_{w_1}σ_{w_2}J_{(w_1,w_2)} > 0. If not, Lemma 3.1 is refuted. If the intended Barrier(e) includes such exterior edges in Low, repeat with that edge as Low; if inequality (3.5) becomes non-positive, the constant 160 in the current proof is wrong.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is Lemma 3.1. Its proof constructs a competitor bσ agreeing with σ off the ten boundary vertices w_1,...,w_10 and claims the energy gain (3.5): H(bσ)−H(σ) ≥ 2 min_i |J(w_i,w_{i+1})| − 2 Σ_{e∈Low}|J_e| ≥ 200−40 = 160. This comparison counts only the blue path edges and the Low edges drawn inside the box. It omits every edge from a boundary vertex w_i to the exterior, since Barrier(e) does not constrain those couplings. But bσ flips w_i, so each such edge changes the energy by ±2|J_e| and these terms are unbounded. Concretely, if one boundary vertex has both blue-path neighbors aligned by the proposed rule and an exterior neighbor z is pinned so that σ_z = −σ_{w_i} with |J(w_i,z)| ≥ 1000, the ground state will violate the blue constraint. Thus Lemma 3.1 is false as stated. The final appeal to the domain Markov property does not help: the boundary spins are not pinned by the local event and can be influenced by arbitrarily large exterior couplings. Lemma 3.2 relies on Lemma 3.1 to flip e* without changing σ_u σ_v; without Lemma 3.1 the conditional-expectation bound (3.7) is unsupported, and Theorem 2.2, Lemma 3.5 part 2, and Theorem 2.3 do not follow.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the zero-temperature Edwards-Anderson model on the n×n grid with i.i.d. standard Gaussian couplings. It defines a spectral probability measure μ for the two-point ground-state correlation σ_u σ_v via the Hermite expansion, and uses Chatterjee's Proposition 2.1 that the spectral sample must connect u and v. The main results are exponential upper bounds: Theorem 2.2 states μ(L) ≤ exp(−cn) for the straight horizontal line L connecting u=(1,n/2) and v=(n,n/2), and Theorem 2.3 states that the total spectral mass of sets of size at most (1+ε)n is also exp(−cn). The proof introduces local “barrier” events of high and low couplings and claims in Lemma 3.1 that a barrier forces alignment of the boundary spins of a small box, independently of all couplings outside. Lemma 3.2 then uses this to implement a cut-set sign-flipping symmetry and conclude that the conditional expectation of σ_u σ_v is zero on the good event, yielding the exponential decay. Theorem 2.3 is obtained by combining the barrier argument with a combinatorial classification of small connecting sets, with a union bound over the classes.","tokens_in":11766,"tokens_out":4483,"duration_ms":53883,"significance":"The intended contribution is a new percolation-type estimate on the size of the spectral sample in a lattice spin glass, a direction where rigorous results are scarce and where numerical work predicts fractal dimension strictly above one. The paper is transparent about its modest quantitative gain and explicitly builds on an external result of Chatterjee rather than on fitted parameters or circular reasoning; I see no circularity in the use of Proposition 2.1. If the barrier mechanism were valid, it would be a clean and potentially transferable idea. However, the central mechanism is not established: the proof of Lemma 3.1 omits uncontrolled exterior edges, and the main theorems depend directly on that lemma. The exposition is otherwise clear, and the combinatorial part of Theorem 2.3 is reasonable, but the probabilistic core of the argument needs substantial repair.","major_comments":[{"comment":"The energy comparison in (3.5) is incomplete. The configuration σ̂ flips the ten boundary vertices w_1,...,w_10, so every edge from a boundary vertex to the exterior of the box changes its energy by ±2|J_e|. These exterior couplings are not constrained by the event Barrier(e), and they can be arbitrarily large. A single exterior edge with |J_e| ≥ 1000 incident to a boundary vertex can dominate the claimed lower bound of 200 − 40 = 160 and pin that boundary spin, so the lemma's conclusion that σ_{w_i}σ_{w_{i+1}}J_{(w_i,w_{i+1})} > 0 for all i, irrespective of all couplings outside the box, is false as stated. The domain Markov property invoked at the end of the proof does not help because the boundary spins are not conditioned; the claim 'fixing σ_{w_i} for 1 ≤ i ≤ 10' assumes the very alignment that the lemma is supposed to prove. Since this lemma is the load-bearing step for Lemma 3.2 and hence for Theorems 2.2 and 2.3, the main results are not established by the present proof.","section":"Section 3, Lemma 3.1 and equation (3.5)"},{"comment":"Lemma 3.2 asserts E[σ_u σ_v | J_L, |J|, Good] = 0 by arguing that on Barrier(e*) the value of σ_u σ_v does not depend on the sign of J_{e*}, so the cut-set flipping symmetry applies. This step relies entirely on Lemma 3.1. Given the gap in Lemma 3.1, the sign of σ_u σ_v may depend on J_{e*} through large exterior couplings incident to the boundary of the barrier box, and the conditioned law of σ_u σ_v need not be uniform on {±1}. Consequently the conditional-expectation bound (3.3), and with it the exponential bound μ(L) ≤ exp(−cn), does not follow.","section":"Section 3, Lemma 3.2 and equation (3.7)"},{"comment":"The proof of part 2 of Lemma 3.5 repeats the barrier argument on straight segments of the sets in each class G and again concludes that the conditional expectation is zero on the good event. No new mechanism is introduced to control exterior edges incident to the boundary vertices of the barrier boxes. Thus the same omission that invalidates Lemma 3.1 propagates to the proof of Theorem 2.3, and the claimed exponential bound on the total spectral mass of small sets is not proved as written.","section":"Section 3, Lemma 3.5 part 2 (proof of Theorem 2.3)"}],"minor_comments":[{"comment":"The vertices w_1,...,w_10 are not labeled in Figure 3.2, which makes it difficult to verify the traversal order, the set of Low edges, and the top-right corner mentioned in the lemma statement.","section":"Section 3, Figure 3.2"},{"comment":"The sentence 'for all the vertices w including v, incident on edges e∈L, it must be must be the case, provided that σ_u is pinned to be 1, that σ_w = 1' contains a duplicated 'must be' and is grammatically unclear; the intended set of vertices should be specified precisely.","section":"Section 3, Remark 3.3"},{"comment":"The statement 'The theorem is an immediate consequence of the following lemma by a union bound' would benefit from spelling out the union bound over the collection C, since the balance between the entropic factor exp(n·ε·(log(1/ε)+c₂)) and the spectral bound exp(−cn) is the crux of the argument and is only implicit in the text.","section":"Section 3, proof of Theorem 2.3"},{"comment":"The paper advertises that the arguments are not planar and hold in any dimension, but the barrier construction is drawn and described only in two dimensions; the spanning-tree generalization in Remark 3.4 is only a sketch and would need a precise statement to be verifiable.","section":"Section 1 and Remark 3.4"}],"recommendation":"reject","confidential_remarks":"The gap in Lemma 3.1 is fundamental: the energy comparison omits exterior edges incident to the barrier-box boundary, and the lemma is false as stated. Both main theorems depend on this lemma, so the central claims are not proved. I do not see a repair within the present manuscript's scope; the barrier event would need to constrain all edges incident to the boundary vertices, which would change its probability and likely require a different construction. The combinatorial part of Theorem 2.3 may be salvageable, but the probabilistic mechanism needs substantive revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: the paper has a genuinely new idea — a local barrier configuration that would decouple a conditioned coupling from the relative spin — and the intended application to lower bounds on spectral sample size is a sensible next step after Chatterjee. But the main lemma has a proof gap that the current text does not fix, and until that is repaired the theorems are not established.\n\nWhat is new and good: the barrier construction itself, and the way the authors use it to get an exponential bound on µ(L) and then a union-bound upgrade to all sets of size at most (1+ε)n. The counting part of Lemma 3.5 is careful and the estimates look right, conditional on having the barrier lemma. The paper is also honest: it states clearly that the improvement over Chatterjee's qualitative result is modest, and it discusses the limitations of the method in Remarks 2.4 and 2.5. Citation practice is fine; the main external input is Chatterjee's Proposition 2.1, which is clearly cited.\n\nThe soft spot is Lemma 3.1. The energy comparison in (3.5) only counts the blue path edges and the Low edges inside the box. It omits every edge from the boundary vertices w_i to the exterior. The barrier event does not constrain those exterior couplings, so they can be arbitrarily large. When the competitor flips a boundary vertex, each such exterior edge contributes ±2|J_e| to the energy difference, and a single large exterior coupling can wipe out the claimed 160 gain. The lemma as stated is therefore not proven, and it may well be false as a deterministic statement. Since Lemma 3.2 uses Lemma 3.1 to flip e* without changing σ_uσ_v, both Theorem 2.2 and Theorem 2.3 rest on this step. The fix is plausible — add the exterior edges incident to the boundary to the Low set, or otherwise bound them — but that is not a cosmetic change; it alters the barrier event and the probability estimate.\n\nI don't think the paper is salvageable in its present form, but the idea deserves a careful referee. The structural parts — the philosophy of using barriers to get quantitative connectivity control, and the combinatorial reduction — are worth engaging with. I'd send it to review rather than desk-reject, but the referee should focus on whether Lemma 3.1 can be repaired.\n\nFor your purposes: would I cite it now? No, because the main results are not established. Would I bring it to reading group? Maybe; it is a good discussion piece about where noise-sensitivity arguments for spin glasses stand, and the gap is instructive.","headline":"Promising barrier idea for quantitative spectral bounds in the EA model, but Lemma 3.1 has an unaddressed proof gap that takes down Theorems 2.2 and 2.3 as written.","tokens_in":12415,"tokens_out":3890,"would_cite":false,"duration_ms":41762,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","82B44"],"pacs":[],"model":"deepseek-v4-flash","headline":"In the Edwards-Anderson spin glass, the Fourier spectrum is forced off the shortest path by a new barrier argument.","keywords":["Edwards-Anderson model","spin glass","zero temperature","Fourier spectrum","spectral sample","noise sensitivity","disorder chaos","percolation barriers"],"falsifier":"Run the barrier configuration around the middle edge e of the line in an n by n grid, set all exterior couplings adjacent to the box boundary to +1000 with outside spins free, and compare ground-state sigma_u sigma_v for J_e=+1 and J_e=-1; if the product changes, Lemma 3.1's pinning claim is false.","tokens_in":11237,"feed_emoji":"🧊","tokens_out":5630,"duration_ms":64170,"temperature":0.7,"pith_summary":"The paper studies the zero-temperature Edwards-Anderson spin glass on an n by n grid and asks how large the spectral sample of the two-point spin correlation must be. It proves that for two horizontally opposite vertices, the spectral measure assigns exponentially small mass to the straight path between them, and more generally to every edge set of size at most (1+epsilon)n. The proof introduces a percolation-type barrier: a local pattern of very strong and very weak couplings around an edge that forces the ground-state spin product to be insensitive to the sign of that edge. If correct, this is evidence that the Fourier spectrum of the EA ground state is genuinely larger than the shortest path, consistent with numerics suggesting a fractal dimension above one.","feed_headline":"Spectral mass on short edge sets is exponentially tiny in a spin glass","feed_subtitle":"A barrier construction forces the spin-glass ground state's Fourier spectrum to exceed (1+epsilon) times the distance.","key_machinery":"The carrying object is the barrier configuration around an edge e: a box around e whose boundary cycle consists of nine 'blue' edges with |J| at least 100 and whose interior 'red' edges have |J| at most 1. Lemma 3.1 asserts that in any ground state, the spin product along the boundary path must satisfy sigma_{w_i} sigma_{w_{i+1}} J_{(w_i,w_{i+1})} > 0 for every boundary edge, so the boundary spins are pinned relative to one corner and the relative spin sigma_u sigma_v is independent of the sign of J_e. Lemma 3.2 then flips the signs of all other edges in a vertical cutset through a barrier; because barriers occur independently with constant probability along the line, an exponentially small failure probability plus cutset symmetry forces the conditional expectation given the line couplings to vanish.","core_discovery":"The paper's central claim is Theorem 2.3: for u=(1,n/2) and v=(n,n/2) in the n by n grid, there exist epsilon>0 and c>0, independent of n, such that the total spectral mass of all edge sets S connecting u and v with |S| at most (1+epsilon)n is at most exp(-cn). Interpreted probabilistically, a sample from the spectral measure has size strictly larger than (1+epsilon) times the L1 distance with probability exponentially close to 1. The weaker Theorem 2.2 gives the same exponential decay for the single straight-line set L. The argument builds on Chatterjee's observation that any Fourier block whose support does not connect the two vertices has zero coefficient; the new work is a quantitative, geometrically local mechanism that kills the conditional expectation of the relative spin on a large conditioning set.","pith_inferences":["The method should transfer to positive temperature if the barrier energy comparison is augmented by an entropy penalty for flipping interior spins; that would turn Fourier-spectrum lower bounds into bounds on the Ornstein-Uhlenbeck decorrelation time.","Because the barrier event is defined only through absolute values, the same lower bound should hold for any continuous symmetric coupling distribution; the Gaussian assumption is used for the Hermite framework, not for the barrier mechanism.","One could probe the conjectured fractal dimension by computing the expected size of the spectral sample; the paper's bounds imply E|S| is at least (1+epsilon)n minus an exponentially small term, but not a power-law exponent."],"forward_implications":["The straight line L carries exponentially small spectral mass, so the spectral sample is not concentrated on the unique geodesic between u and v.","All small sets of size at most (1+epsilon)n together carry exponentially small mass, so with probability exponentially close to 1 the spectral sample has size at least (1+epsilon)n.","The same statements hold in any dimension, with columns replaced by slabs of width W.","A matching lower bound up to logarithmic factors shows the exponential decay in Theorem 2.2 is essentially sharp.","The barrier-plus-entropy method may be reused for other conditioning sets, since the barrier events are independent across disjoint straight segments."],"supporting_citations":[{"why":"Establishes Proposition 2.1, the zero-support condition that every nonzero Fourier block must connect the two vertices, the baseline this paper improves.","marker":"[Cha23]"},{"why":"Provides the numerical evidence for a fractal dimension of the spectral sample strictly above one that motivates Theorems 2.2 and 2.3.","marker":"[BM87]"},{"why":"Introduces the spectral-sample framework for noise-sensitive percolation events that the paper adapts to the Edwards-Anderson model.","marker":"[BKS99]"},{"why":"Gives the detailed spectral analysis of critical percolation whose fractal questions this paper takes up.","marker":"[GPS10]"},{"why":"Introduces the Edwards-Anderson lattice spin-glass Hamiltonian under study.","marker":"[EA75]"},{"why":"Supplies the Ornstein-Uhlenbeck eigenfunction fact that turns the covariance of relative spins into the Fourier weight expression used in the paper.","marker":"[Cha14]"}],"fun_headline_variants":["Spin glass: short spectral paths carry exponentially tiny mass","Edwards-Anderson: spectral mass on short edge sets decays exponentially","Barrier construction kills spectral mass on short edge sets","Spin glass spectrum: short sets are exponentially negligible"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on Lemma 3.1's assumption that a local box of prescribed high and low couplings pins the boundary spins of the box no matter what the couplings outside the box do, even though those exterior couplings are not part of the energy comparison in (3.5).","fun_headline_variants_meta":{"raw":{"variants":["Spin glass: short spectral paths carry exponentially tiny mass","Edwards-Anderson: spectral mass on short edge sets decays exponentially","Barrier construction kills spectral mass on short edge sets","Spin glass spectrum: short sets are exponentially negligible"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000892,"raw_usage":{"total_tokens":3818,"prompt_tokens":888,"completion_tokens":2930,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":504,"completion_tokens_details":{"reasoning_tokens":2865}},"tokens_in":504,"tokens_out":2930,"duration_ms":27580,"temperature":1.0,"reasoning_tokens":2865,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T17:32:19.544605+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the barrier configuration around the middle edge e of the line in an n by n grid, set all exterior couplings adjacent to the box boundary to +1000 with outside spins free, and compare ground-state sigma_u sigma_v for J_e=+1 and J_e=-1; if the product changes, Lemma 3.1's pinning claim is false.","supporting_citations":[],"review_version":1}