{"id":"14c00b00-7b37-4d70-bd5f-945166ebd7a2","arxiv_id":"2501.01718","paper_version":4,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For one-dimensional block band matrices with W > N^{1/2+c}, the paper proves the local semicircle law, eigenvector delocalization, quantum unique ergodicity, and GUE universality.","lead":"This paper proves that one-dimensional random band matrices with band width larger than the square root of the matrix size have delocalized eigenvectors and GUE eigenvalue statistics. The proof introduces a new loop hierarchy method that may extend to other random matrix models.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.6 rests on Section 7.2's flat-profile estimates (7.27)-(7.29), which are sketched rather than derived; if those estimates fail, universality collapses, so the current conditional verdict is appropriate.","rationale":"The reader's weakest-assumption analysis correctly identifies Section 7.2 as the decisive gap for the universality theorem. The rest of the paper, especially the primitive hierarchy, the tree representation of K, the sum-zero property, and the six-step proof of Theorem 2.21, is presented at a high level of technical detail and appears internally coherent. I do not see a concrete error in the main loop-hierarchy estimates, and the delocalization, local semicircle law, quantum diffusion, and QUE statements for the block model are likely unaffected by the Section 7.2 issue. The concern is concentrated on the short-time OU-flow estimates needed only for Theorem 2.6: those estimates are used to control L1 and L2 in Lemma 4.18 of [48], and without them the correlation-function comparison between H and GUE does not go through. Since Section 7.2 gives only a schematic derivation, with key flat-profile bounds asserted rather than proved, the honest verdict remains CONDITIONAL rather than ACCEPT. I would not move to REJECT, because the assertion is plausible and may well be fillable; the missing details are exactly the kind of technical verification that a conditional acceptance should request. The abstract overclaim about general one-dimensional band matrices is real but secondary, since the theorems themselves are stated for the block model; correcting the abstract would not change the mathematical verdict.","tokens_in":81767,"tokens_out":18002,"duration_ms":179460,"concrete_test":"Write out the flat-profile analogue of Lemma 5.10 with S^(B)_GUE = 1/L, and verify the omitted summation and Ward-identity steps in (7.39)-(7.44). Concretely, for n=1 at eta = N^{-1+2tau_U}, compute the martingale term in (7.38) by explicitly summing b,b' in Definition 5.4 with the flat profile: the claimed prefactor N^{-1}eta_u^{-2} sup|eL^(2)| must emerge from the W,L power count after two Ward summations; if the correct prefactor is instead (N/W^2)eta_u^{-2} or the bootstrap term |t-t1|^{1/2}(N^{-1}eta_u^{-2})^{1/2} sup|eL^(n)| in (7.45) does not become O(N^{-2tau_U}) for n=2, then (7.27)-(7.29) fail and Theorem 2.6 is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The delocalization results, Theorems 2.2-2.5, are supported by a detailed loop-hierarchy argument and appear internally consistent. The load-bearing uncertainty is in Theorem 2.6, whose proof needs the weak local law and QUE estimates (2.26)-(2.27) for the Ornstein-Uhlenbeck flow H_t at times t <= N^{-1+tau_U}. These are deferred to Section 7.2 and are not obtained by literally repeating the main proof: the replacement S^(B) -> S^(B)_GUE = 1/L destroys the block-profile length scale ell_t on which Lemma 5.10's E-term bounds and Lemma 7.3's kernel estimates depend. The flat-profile bounds (7.39)-(7.44) are asserted after 'similar' reasoning, but the summation factors from the constant S_GUE, the Ward-identity reductions, and the bootstrap inequality (7.45) require separate verification. In particular, if the martingale bound (7.43) or the (eL-eK) inequality (7.45) fails at eta = N^{-1+2tau_U}, then the L1,L2 bounds (2.29)-(2.30) and the comparison estimate (2.23) collapse, so the bulk universality theorem is not established. The abstract's omission of the block-structure restriction is a second, lesser issue; it affects the scope of the advertised theorem, not the internal correctness of the block-model proof.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies N×N Hermitian one-dimensional random band matrices in a block-band model with variance profile S = S^(B) ⊗ S_W and band width W > N^{1/2+c}. The main claims, stated in Theorems 2.2–2.6, are: local semicircle law down to scale N^{-1+ε}; simultaneous L^∞ delocalization of all bulk eigenvectors; quantum diffusion and generalized quantum unique ergodicity; and bulk GUE universality. The method introduces a stochastic flow with a time-dependent spectral parameter and analyzes a hierarchy of G-loop observables, approximating it by a 'primitive hierarchy' whose solutions K are constructed explicitly via tree representations. Theorems 2.3–2.5 are derived from loop estimates (Lemmas 2.18–2.20) through the six-step Theorem 2.21. Theorem 2.6 instead uses an Ornstein–Uhlenbeck comparison between the band matrix and GUE, relying on short-time local law and QUE estimates for the flow H_t that are deferred to Section 7.2.","tokens_in":82013,"tokens_out":9550,"duration_ms":91804,"significance":"If the proofs are correct, these results are a substantial advance: they establish the delocalization side of the conjectured W ≈ √N transition for the block-band model, with essentially optimal power W > N^{1/2+c}, and simultaneously give the local semicircle law, eigenvector delocalization, quantum diffusion, and bulk universality. The primitive-hierarchy/tree representation is an original and structurally explicit method: K is defined by an exact differential equation, solved in Lemma 3.4 by tree sums, and the sum-zero property of Lemma 3.10 is identified as the mechanism that converts the exponential instability of the hierarchy into polynomial bounds. There are no fitted parameters, and the main delocalization argument does not appear circular. The caveat is that the universality theorem is not on the same footing: its proof depends on a sketched flat-profile analysis in Section 7.2, and until that section is completed, Theorem 2.6 should be regarded as conditional.","major_comments":[{"comment":"The proof of Theorem 2.6 rests on the flat-profile estimates for eS^(B) = 1/L, but these estimates are asserted rather than derived. Replacing S^(B) by the constant matrix S_GUE^(B)=1/L changes Θ_t = (1 − ξS^(B))^{-1} from an exponentially decaying kernel of scale ℓ_t to a rank-one-type kernel with no such length scale, so the ℓ_t-summation factors used in Lemmas 5.10 and 7.3 and in the derivation of (5.34)–(5.36) are no longer available. The bounds (7.39)–(7.44) are introduced with phrases such as 'similar to' or 'one can prove' without the required kernel and Ward-identity estimates, and (7.45) is not closed: the sentence 'With a continuity argument and an induction on n ≥ 2, this inequality thus implies the estimate (7.27)' is a claim, not a proof. This is load-bearing: if (7.43) or (7.45) fails, then (2.29)–(2.30) and the comparison estimate (2.23) collapse, and Theorem 2.6 is not established.","section":"Section 7.2, Eqs. (7.27)–(7.29) and (7.39)–(7.45)"},{"comment":"The displayed estimate 'η_u ∼ 1 − u ≥ N^{1−2τU} ≫ N^{1−τU} ∼ |t0 − t1|' is false as written: in the stated range N^{-1+2τU} ≤ η ≤ N^{-1+c/3}, the quantity η_u is of order N^{-1+O(τU)}, not N^{1−2τU}. The subsequent continuity/bootstrap argument for (7.27) depends on the ordering η_u ≫ |t0 − t1|, so the exponents must be corrected (presumably to N^{-1+2τU} and N^{-1+τU}) and the continuity argument supplied. As it stands, the step from (7.45) to (7.27) is unverifiable.","section":"Section 7.2, paragraph after Eq. (7.36)"},{"comment":"The proof of Theorem 2.6 assumes the weak local law (2.26) and the QUE estimate (2.27) for all 0 ≤ t ≤ t_U and for the parameter range (2.22), but Section 7.2 proves these only for t = t_U and explicitly narrows to 'only the case η = N^{-1+4τU}', whereas (2.26) is stated at η = N^{-1+2τU}. The assertions that the other cases 'can be handled similarly' or 'should also hold' are not carried out. Since (2.23) needs the full range in (2.22), this is another unproved input of the universality theorem and must be addressed before Theorem 2.6 can be accepted.","section":"Section 2.3, Eqs. (2.26)–(2.27)"}],"minor_comments":[{"comment":"The abstract claims results for 'one-dimensional random band matrices', but the theorems are proved only for the block-band model S = S^(B) ⊗ S_W defined in Section 2.1. The abstract and introduction should state this block-structure restriction explicitly.","section":"Abstract and Section 2.1"},{"comment":"There is a repeated word in the first sentence of Section 2.1: 'a complex complex Hermitian random band matrix'.","section":"Section 2.1"},{"comment":"The name 'Sooster' in the discussion of [46, 45] should be 'von Soosten'.","section":"Section 1"},{"comment":"The displayed sum '\\sum_{x ∈ I_a} \\sum_{a ∈ A}' is notationally garbled; it should presumably be '\\sum_{a ∈ A} \\sum_{x ∈ I_a}'.","section":"Theorem 2.5, Eq. (2.13)"}],"recommendation":"major_revision","confidential_remarks":"The delocalization and local-law part (Theorems 2.2–2.5) appears structurally sound and is supported by a very detailed loop-hierarchy argument. My main concern is Theorem 2.6: the flat-profile estimates in Section 7.2 are the only support for universality, and they are asserted rather than proved. If the authors can supply the missing derivation, the paper would be a strong candidate for acceptance; otherwise, Theorem 2.6 should be removed or explicitly stated as conditional. The reported line-by-line checking was concentrated on the load-bearing Section 7.2, since that is where the manuscript's central remaining gap lies."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does what it says for delocalization, and that's a big deal. For a one-dimensional block band matrix with W > N^{1/2+eps}, it proves the local semicircle law at the optimal scale, simultaneous L^infty delocalization N^{-1/2+eps}, quantum diffusion, and probabilistic QUE. Theorems 2.2-2.5 are supported by a detailed, internally coherent argument; the loop hierarchy and primitive hierarchy are genuinely new, and the explicit tree representation with the sum-zero property is clever and well executed. This is the first rigorous proof at the conjectured threshold, and the delocalization side deserves full credit.\n\nThe soft spots are real but contained. The abstract overclaims: the results are for the block covariance S = S^(B) ⊗ S_W, not general 1D band matrices. That's a scope issue, not a proof flaw, but the abstract and intro should be corrected.\n\nThe load-bearing concern is Theorem 2.6, and the stress-test note lands. The universality proof needs the weak local law and QUE estimates (2.26)-(2.27) for the Ornstein-Uhlenbeck flow at short times. Section 7.2 asserts (7.27)-(7.29) follow from the main proof with 'very minor changes' after S^(B) → 1/L. This is not a literal repetition: the flat profile destroys the block-level scale ℓ_t that Lemma 5.10 and Lemma 7.3 depend on. The bounds (7.39)-(7.44), especially the martingale bound (7.43) and the bootstrap inequality (7.45), need separate verification. If those fail, universality collapses. But the gap is isolated to Theorem 2.6; Theorems 2.2-2.5 do not rest on Section 7.2, and the conditional verdict is right.\n\nMy recommendation is simple: send it to review. The delocalization result alone is a major within-field advance, and the method deserves serious referee time. Ask the referee to require a full derivation of (7.27)-(7.29) in the revision and a corrected abstract. The paper is honest about the omission—it says 'we claim' and defers—but a published universality theorem cannot rest on an unverified assertion.","headline":"A major within-field advance on delocalization for 1D band matrices at the conjectured W > N^{1/2} threshold, with a real but isolated gap in the universality theorem.","tokens_in":82577,"tokens_out":1912,"would_cite":true,"duration_ms":22106,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60B20","15B52","82B44"],"pacs":[],"model":"deepseek-v4-flash","headline":"One-dimensional random band matrices delocalize once band width exceeds $N^{1/2+\\varepsilon}$.","keywords":["random band matrices","delocalization","local semicircle law","quantum unique ergodicity","eigenvalue universality","GUE","loop hierarchy","sum-zero property"],"falsifier":"A numerical experiment on $W=N^{0.6}$ band matrices would settle the main claim: if any bulk eigenvalue correlation statistic at scale $1/N$ deviates from GUE, or any bulk eigenvector has a component larger than $N^{-1/2+0.01}$, the central theorems fail. A cheaper check targets the deferred estimates: for $t=N^{-1+\\tau}$ and $\\eta=N^{-1+2\\tau}$, test whether the Ornstein-Uhlenbeck-flow resolvent obeys the asserted local law and QUE bounds.","tokens_in":81535,"feed_emoji":"🧮","tokens_out":9059,"duration_ms":90388,"temperature":0.7,"pith_summary":"The paper proves that in one dimension a random band matrix with band width $W \\ge N^{1/2+c}$ for any fixed $c>0$ behaves like a Wigner matrix in the bulk. Specifically, the semicircle law holds at scales down to $N^{-1+\\varepsilon}$, every bulk eigenvector has maximum entry at most $N^{-1/2+\\varepsilon}$ with overwhelming probability, eigenvectors equidistribute over macroscopic blocks, and local eigenvalue statistics converge to GUE. The proof works by tracking traces of resolvent products through a stochastic flow and comparing them with exactly solvable deterministic counterparts. This gives a rigorous delocalization result for one-dimensional band matrices above the conjectured $W \\sim N^{1/2}$ threshold, without settling what happens at the threshold itself.","feed_headline":"Wide 1D random band matrices delocalize and match GUE","feed_subtitle":"Above band width $N^{1/2+\\varepsilon}$, bulk eigenvectors spread out and the local eigenvalue statistics are GUE.","key_machinery":"The workhorse is the family of G-loop observables $L_{t,\\sigma,a}=\\langle \\prod_i G_t(\\sigma_i)E_{a_i}\\rangle$, traces of alternating resolvents and block projections. These satisfy a loop hierarchy that is not closed. The paper replaces it by the primitive hierarchy, whose solutions $K_{t,\\sigma,a}$ have an explicit tree representation in terms of the block-level propagator $\\Theta^{(B)}_\\xi=(1-\\xi S^{(B)})^{-1}$. The crucial estimate is that $L-K$ is bounded by $(W\\ell_t\\eta_t)^{-n}$, with the sum-zero property of the self-energy controlling the delicate near-$t=1$ behavior of the flow.","core_discovery":"On its own terms, the paper's central claim is Theorems 2.2 through 2.6: for the block band matrix with $W\\ge N^{1/2+c}$, uniformly in the bulk $|E|<2-\\kappa$, the Green's function satisfies $\\max_{x,y}|(G(z)-m(z))_{xy}|\\prec (W\\ell\\eta)^{-1/2}$, the eigenvectors satisfy $\\max_k\\|\\psi_k\\|_\\infty^2 \\le N^{-1+\\tau}$ with overwhelming probability, the two-point resolvent product has the quantum diffusion profile $W^{-1}(|m|^2/(1-|m|^2 S^{(B)}))_{ab}$, and the $k$-point correlation functions converge to those of GUE. The unifying bound is the loop estimate $|L_{t,\\sigma,a}-K_{t,\\sigma,a}|\\prec(W\\ell_t\\eta_t)^{-n}$, where $K$ is the exact solution of a simplified hierarchy. The proof of universality depends on two short-time estimates for the Ornstein-Uhlenbeck flow that are stated in Section 2.3 and deferred to Section 7.2.","pith_inferences":["Because the proof controls all eigenvectors simultaneously with high probability, it automatically rules out any single bulk eigenvector concentrating on a small block; a quantitative version of these probability bounds would be a natural next step.","The band-width threshold $N^{1/2+c}$ leaves a gap of size $N^c$ to the conjectured transition at $W\\sim N^{1/2}$, and the method does not address that edge; testing whether the hierarchy degenerates as $c\\to 0$ could indicate where the transition actually begins.","The loop-hierarchy machinery as written relies on Gaussianity through the stochastic flow and integration by parts; replacing that comparison by a non-Gaussian argument would be the direct route to extending the conclusions to more general disorder models."],"forward_implications":["For every band width $W \\ge N^{1/2+c}$, all bulk eigenvectors are delocalized simultaneously with overwhelming probability, ruling out sparse bulk eigenvectors (Theorem 2.2).","The local semicircle law holds at scale $N^{-1+\\varepsilon}$, the finest scale on which a density statement can be expected for this model (Theorem 2.3).","Quantum diffusion and generalized quantum unique ergodicity hold: the resolvent has the block-level diffusion profile and local eigenvector mass equidistributes (Theorems 2.4 and 2.5).","The bulk $k$-point correlation functions converge to GUE for every fixed $k$, while individual eigenvalue fluctuations remain on the larger order $(WN)^{-1/2}$ scale (Theorem 2.6).","The authors state the loop-hierarchy method is expected to extend to other variance profiles, higher dimensions, and blocked random potentials."],"supporting_citations":[{"why":"Supplies the time-dependent flow approach and linearized T-observable dynamics that the loop hierarchy develops.","marker":"[16]"},{"why":"Provides the stochastic domination definitions and fluctuation averaging estimates used throughout the proof.","marker":"[19]"},{"why":"The dynamical approach to random matrices and the correlation-function comparison theorem used for bulk universality.","marker":"[21]"},{"why":"The resolvent decomposition and Gaussian integration-by-parts lemmas used to control individual Green's function entries.","marker":"[23]"},{"why":"The local semicircle law and fluctuation averaging for general band matrices that the present estimates extend.","marker":"[25]"},{"why":"The fixed-energy universality for Dyson Brownian motion used to compare the short-time flow with GUE.","marker":"[32]"},{"why":"The high-dimensional band matrix proof whose strategy derives universality from local law, delocalization, and QUE estimates.","marker":"[48]"},{"why":"Introduces the self-energy sum-zero property in high-dimensional band matrices, repurposed here for the primitive hierarchy.","marker":"[49]"}],"fun_headline_variants":["Wide 1D band matrices: eigenvectors spread, stats match GUE","For W>N^(1/2+ε): 1D band matrices delocalize, go GUE","Delocalization and GUE universality for wide 1D band matrices"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the local law and quantum unique ergodicity estimates continue to hold for the Ornstein-Uhlenbeck flow at the very short times used in the universality proof; the paper asserts these follow from the main argument with only minor changes, but it does not supply the detailed derivation.","fun_headline_variants_meta":{"raw":{"variants":["Wide 1D band matrices: eigenvectors spread, stats match GUE","For W>N^(1/2+ε): 1D band matrices delocalize, go GUE","Delocalization and GUE universality for wide 1D band matrices"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001949,"raw_usage":{"total_tokens":7636,"prompt_tokens":976,"completion_tokens":6660,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":592,"completion_tokens_details":{"reasoning_tokens":6597}},"tokens_in":592,"tokens_out":6660,"duration_ms":48418,"temperature":1.0,"reasoning_tokens":6597,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:21:45.016140+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A numerical experiment on $W=N^{0.6}$ band matrices would settle the main claim: if any bulk eigenvalue correlation statistic at scale $1/N$ deviates from GUE, or any bulk eigenvector has a component larger than $N^{-1/2+0.01}$, the central theorems fail. A cheaper check targets the deferred estimates: for $t=N^{-1+\\tau}$ and $\\eta=N^{-1+2\\tau}$, test whether the Ornstein-Uhlenbeck-flow resolvent obeys the asserted local law and QUE bounds.","supporting_citations":[{"cited_title":"Averaging Fluctuations in Resolvents of Random Band Matri- ces","cited_arxiv_id":null,"evidence_quote":"Provides the stochastic domination definitions and fluctuation averaging estimates used throughout the proof."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The dynamical approach to random matrices and the correlation-function comparison theorem used for bulk universality."},{"cited_title":"Rigidity of eigenvalues of generalized Wigner matrices","cited_arxiv_id":null,"evidence_quote":"The resolvent decomposition and Gaussian integration-by-parts lemmas used to control individual Green's function entries."},{"cited_title":"The local semicircle law for a general class of random matrices","cited_arxiv_id":null,"evidence_quote":"The local semicircle law and fluctuation averaging for general band matrices that the present estimates extend."}],"review_version":1}