{"id":"70b2fe8b-b9d9-4fb9-8c8d-b0a548add64a","arxiv_id":"2512.03628","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For tridiagonal random matrices with independent entries of finite second moment, the empirical spectral distribution converges almost surely, and slowly varying weights produce a scale-mixture limit.","lead":"This paper proves that the eigenvalue distribution of certain random tridiagonal matrices converges to a known limit under only a finite second-moment condition, relaxing earlier all-moments assumptions. It also shows that slowly varying weights on the off-diagonal entries scale the limit law by an independent random factor.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Truncation step in Theorem 3.8 conflates truncating coefficients b_i with truncating the limiting variable X; the displayed limit is false as stated.","rationale":"The reader's weakest_assumption identifies exactly the truncation continuity step in Theorem 3.8. My analysis confirms the gap is real and, in fact, the stated limiting identification is literally false for fixed truncation levels because truncating b_i changes the second moment of the limiting spectral measure, whereas clipping X does not. The final claim that the limits coincide as C,M→∞ is plausible and can likely be recovered by a separate continuity argument, so the result is probably true but not proven as written. Secondary issues—Lemma 3.4 invokes Theorem 1.3 under conditions not verified for general σ, and Theorem 6.4 is stated without proof—reinforce the conditional verdict but are less central than the truncation step. Therefore the existing CONDITIONAL verdict remains appropriate; no adjustment is needed.","tokens_in":20014,"tokens_out":11186,"duration_ms":94175,"concrete_test":"Take σ≡1 (so T≡1), b_i i.i.d. N(0,1), and fix C=3. Compute the second moment of the limiting ESD for the truncated model X^{σ^M}_{N,C}: it equals E[b² 1_{|b|≤3}] ≈ 0.971. The paper's claimed limit L(X 1_{|X|≤3}) has second moment E[X²]=1 because the support of μ_X is contained in [-2,2]. If these second moments differ, the display in the proof of Theorem 3.8 is false as stated. To test repairability, verify the alternative route: show μ_{b^C} ⇒ μ_X as C→∞ by applying Corollary 2.12 to b_i = b_i 1_{|b_i|≤C} + b_i 1_{|b_i|>C} with residual L²-norm tending to 0, and then L(T_M Y_C) ⇒ L(TX) by independence and Slutsky. If this succeeds, the theorem stands after replacing the displayed step.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 3.8 asserts: 'By moment convergence, μ_{X^{σ^M}_{N,C}} → L(T 1_{|T|≤M}·X 1_{|X|≤C})'. This is the only bridge from the compact/truncated case to the general L² case, and it is not justified. The left side is the limit for a model with off-diagonal entries β_i = b_i 1_{|b_i|≤C}; its second moment is E[β_1²] = E[b² 1_{|b|≤C}] (times E[T² 1_{|T|≤M}] when T is included). The right side is the law of the clipped spectral variable T 1_{|T|≤M} · X 1_{|X|≤C}; its second moment is E[T² 1_{|T|≤M}] E[X² 1_{|X|≤C}]. These cannot be equal for fixed C. Truncating the coefficients is not the same operation as truncating the spectral variable. For example, take b_i ~ N(0,1), σ≡1, so T≡1 and the simple-model limit X has |X| ≤ 2 a.s. For C=3, the right side has second moment E[X²]=1, while the left side has second moment E[b² 1_{|b|≤3}] ≈ 0.971. Thus the displayed convergence fails at the level of second moments. What is missing is a continuity lemma: μ_{b^C} ⇒ μ_X as C→∞, and then L(T_M Y_C) ⇒ L(TX), where Y_C is the limit for the truncated coefficient model. This is plausible and likely repairable using the resolvent bound (3.9), but without it the truncation argument—and hence Theorem 3.8—is incomplete as written.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Stieltjes-transform approach to the empirical spectral distribution (ESD) of random tridiagonal matrices, complementing the moment method of [Pop09]. It first treats a \"simple\" model with i.i.d. off-diagonal entries in L^2 and zero diagonal, proving almost-sure convergence of the ESD to a law characterized by a fixed-point Stieltjes equation. It then considers a deformed model with an additional slowly varying factor sigma_{k,N} and a negligible diagonal, claiming that the limiting law is a scale mixture T·X, where X is the simple-model limit and T is independent with the empirical limit law of the sigma's. The paper also presents examples, a joint-distribution result for (Sigma_N, X_N), and a speculative algebraic interpretation via one-step paths.","tokens_in":20565,"tokens_out":16661,"duration_ms":150427,"significance":"If the main theorem were correct, this would be a substantive extension of [Pop09]: spectral convergence under only a finite second moment, analogous to the L^2 relaxation for Wigner matrices. The simple-model proof via Wasserstein contraction is coherent and genuinely analytic, and the compact-support deformation result (Theorem 3.1) is a clean use of the path expansion. The paper also gives explicit examples and a joint-convergence statement. However, the proof of the general theorem (Theorem 3.8) is not valid as written: the truncation bridge is missing, Lemma 3.5 is false, and Lemma 3.4 invokes an inapplicable result. These are load-bearing gaps, so the significance is conditional on a substantial repair.","major_comments":[{"comment":"The truncation step is the only bridge from the compact case to the L^2 case, and it is not justified. The paper asserts that, by moment convergence, μ_{X^{σ^M}_{N,C}} → L(T 1_{|T|≤M}·X 1_{|X|≤C}) as N→∞. This is false as stated because truncating the coefficient b_i is not the same operation as truncating the spectral variable X. For example, take b_i ~ N(0,1), σ≡1 (so the simple-model limit X is the arcsine law on [-2,2]). The truncated-entry model has limiting second moment 2E[b^2 1_{|b|≤C}], while L(X 1_{|X|≤C}) has second moment E[X^2 1_{|X|≤C}], and these are unequal for every finite C>2. What is needed is a continuity lemma: μ_{b^C} ⇒ μ_X as C→∞, and L(T_M Y_C) ⇒ L(TX), where Y_C is the limit for the truncated coefficient model. The paper neither states nor proves such a lemma. Moreover, after Eq. (3.9) the proof writes S_{μ_{X^σ}}(z) before the existence of the limit has been est","section":"§3.2, Theorem 3.8 proof"},{"comment":"Lemma 3.5 is false as stated. Eq. (3.6) claims that limsup_N (1/N)∑ σ_i^2 1_{|σ_i|>M} ≤ E[T^2 1_{|T|>M}] under empirical convergence and E[T^2]<∞. But weak convergence of probability measures does not control the empirical tail of the open set {|x|>M} in the claimed direction; Portmanteau gives liminf F_N({|x|>u}) ≥ P(|T|>u), not the reverse. The claimed limsup inequality can fail even with slow variation. Let T≡0 and define σ_{k,N}=0 for k≤N−N^{2/3}, and σ_{k,N}=(k−N+N^{2/3})N^{-1/3} for the last N^{2/3} indices. Then the empirical measure of the σ's converges to δ_0, E[T^2]=0, and sup_k |σ_{k,N}−σ_{k-1,N}| = N^{-1/3}→0, so all three hypotheses of Theorem 3.8 hold, yet (1/N)∑σ_k^2 ≈ (1/3)N^{1/3} → ∞. Thus Lemma 3.5 cannot supply the uniform integrability used in the bound (3.9). The main theorem in Section 1.3 includes an explicit L^2 uniform integrability condition; Theorem 3.8 omits i","section":"Lemma 3.5, Eqs. (3.6)–(3.7)"},{"comment":"Lemma 3.4 invokes Theorem 1.3 (Pop09) to assert that E(S_{\\tilde X_N^σ}(z)) converges for the zero-diagonal model, but Theorem 1.3 assumes all moments of b_n and a specific n^α scaling, namely conditions (2.7)–(2.8). It does not apply under the hypotheses of the present paper, where the b_i are only assumed to lie in L^2 and the σ_{k,N} are arbitrary. Consequently the lemma's claim that the diagonal does not affect the limiting distribution is not proved as written. The conclusion is plausible and can likely be derived from the Hoffman–Wielandt argument used in Corollary 2.12, but a self-contained proof under the actual hypotheses of Theorem 3.8 is required.","section":"Lemma 3.4"}],"minor_comments":[{"comment":"The proof establishes convergence of E(tr_N S_N(z)) on the half-plane ℑz > E(b^2) and then says \"by analyticity\" this holds on all of C^+. The analytic extension of the expectation is fine, but the almost-sure convergence is not explicitly extended from a half-plane to all z ∈ C^+. The standard Montel/diagonal argument should be spelled out.","section":"Theorem 2.11"},{"comment":"The displayed Theorem 1.3 references assumptions labelled (2.7)–(2.11), but those equation numbers have not been introduced at that point in the text; they belong to the quoted result from Pop09. Please renumber or clarify.","section":"Theorem 1.3 statement"},{"comment":"There are several typos and missing items: \"eivenvalues and eigenvectors\" in the introduction, \"reated\" for \"related\", \"reusult\" in §3.2, and Figure 1 is referenced but not visible in the text. Proposition 4.1's second proof also contains a typographical oddity in the display S(z)=1/(z−z+√(z^2−4)).","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The simple-model proof and the compact-case theorem are valuable, and the main idea is potentially repairable. However, the proof of Theorem 3.8 is not valid as written: Lemma 3.5 is false, Lemma 3.4 relies on an inapplicable result, and the truncation continuity step is missing. These are central, not cosmetic, so the manuscript needs substantial revision before it can be considered for publication. I recommend major revision rather than rejection because the gaps appear addressable within the paper's framework."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper does something genuinely useful: Theorem 2.11 replaces the all-moment assumptions in Popescu's tridiagonal theorem with finite second moment, using Stieltjes transforms, and the simple-model proof looks clean. Theorem 3.8, giving a scale-mixture limit under slow variation and L2, is new and plausible. The examples and the joint convergence result are also worthwhile.\n\nThe soft spot is the truncation step in the proof of Theorem 3.8. The authors truncate the coefficients b_i to b_i 1_{|b_i|≤C} and the weights σ to σ 1_{|σ|≤M}, then assert by moment convergence that the limit is L(T 1_{|T|≤M}·X 1_{|X|≤C}). That identification is false: the limit of the truncated model is L(T_M·X_C), where X_C is the limiting spectral variable for the simple model with coefficients truncated at C—not the clipped variable X 1_{|X|≤C}. The second moments differ (E[b² 1_{|b|≤C}] vs E[X² 1_{|X|≤C}]). A continuity lemma is missing: show μ_{b^C} ⇒ μ_b and then L(T_M X_C) ⇒ L(TX). This is likely repairable—the Hilbert–Schmidt bound (3.9) gives the right mechanism—but as written the general theorem is incomplete.\n\nOther issues: Lemma 3.4 cites Theorem 1.3 to justify convergence of the zero-diagonal model, but Theorem 1.3 requires the α-scaling and all moments; the proof should reference the compact-case Theorem 3.1 or an analogous moment argument. Theorem 6.4 is stated without proof; in an exploratory section that is acceptable if labeled as a conjecture, not a theorem. There are also minor sign typos in the Stieltjes transform formulas, though they appear cosmetic.\n\nAll that said, the core new results are probably true, and the simple-model theorem is a clean contribution. The paper deserves a serious referee: send it to peer review, but expect the referee to demand a fix for the truncation gap.","headline":"Useful L2 extension of Pop09 with a real but repairable truncation gap in the proof of Theorem 3.8.","tokens_in":20905,"tokens_out":9128,"would_cite":true,"duration_ms":72929,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60B20","60F15","15B52"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the eigenvalue distribution of a broad class of random tridiagonal matrices converges almost surely to a scale mixture T·X, under only a finite second moment on the off-diagonal entries.","keywords":["random tridiagonal matrices","empirical spectral distribution","Stieltjes transform","scale mixture","slow variation","almost sure convergence","Wasserstein distance","second moment conditions"],"falsifier":"Simulate the model with b_i having a distribution with finite second moment but heavy tail (e.g., Pareto with tail index 3) and with σ_{k,N} satisfying (SV), (EM), (SM); compare the empirical spectral distribution for large N with the law of T·X computed from the Stieltjes equation. If the histograms systematically deviate, the claimed truncation continuity fails.","tokens_in":19917,"feed_emoji":"📈","tokens_out":7240,"duration_ms":56631,"temperature":0.7,"pith_summary":"The paper establishes a convergence theorem for the eigenvalue distribution of random tridiagonal matrices whose off-diagonal entries are independent with finite second moment and whose diagonal entries are asymptotically negligible. The key advance over earlier moment-based results is that the proof works through the Stieltjes transform, so no higher moments are required. Under slow variation and empirical convergence of the scaling coefficients σ, the limiting spectrum is the law of a product T·X of an independent pair, where X is the limit for the unscaled simple model and T is the limit of the coefficients. The paper also proves joint convergence of the diagonal coefficient matrix and the tridiagonal matrix, and explores an algebraic structure analogous to free probability.","feed_headline":"Random tridiagonal spectra converge almost surely to T·X","feed_subtitle":"A Stieltjes-transform proof extends the moment method to entries with only finite variance.","key_machinery":"The proof rests on the Stieltjes transform of the resolvent and its recursion. For the simple model, the diagonal resolvent entries s_i satisfy s_{i+1} = 1/(z − b_i^2 s_i), and the paper shows these converge in Wasserstein distance to a random variable S characterized by S ≃ 1/(z − b^2 S). This random continued fraction, together with slow variation of σ, lets the authors express the trace of the resolvent as an average over independent copies, leading to the scale-mixture limit. A truncation argument is then used to pass from bounded entries to the full L^2 assumption.","core_discovery":"The central claim of the paper is Theorem 3.8: for a tridiagonal matrix X^σ_N with i.i.d. off-diagonal entries b_i of finite second moment, diagonal entries a_{k,N} with sup_k E|a_{k,N}|^2 → 0, and coefficients σ_{k,N} that are slowly varying (sup_k |σ_{k,N}−σ_{k+1,N}|→0) and whose empirical measure converges to a law μ_T with E[T^2]<∞, the empirical spectral distribution converges almost surely to the law of T·X, where X is independent of T and has the limiting spectral distribution of the simple model with entries b_i. The limit is characterized by a Stieltjes transform equation involving the distribution of b^2 and two independent copies of a random continued fraction.","pith_inferences":["A natural next step is to supply the missing continuity lemma; if it fails, the theorem may still hold but needs a different proof, and the counterexample would be a distribution with finite second moment whose truncated ESD converges to a different limit.","The scale-mixture factorization suggests that fluctuations of linear statistics should decouple into a term driven by the empirical measure of σ and a term driven by the base model; a CLT could be derived under additional moment conditions.","The algebraic structure based on the shift operator and colored paths may be a precursor to a 'tridiagonal free probability'; one testable question is whether the addition operation defined in Section 6 is associative when applied to non-independent models.","For band matrices with growing width, the same Stieltjes-transform recursion might work with paths of multiple steps, giving an analytic route to known results that currently rely on moment methods."],"forward_implications":["The empirical spectral distribution converges almost surely for any i.i.d. off-diagonal entries with finite second moment, even when higher moments are infinite.","The limiting law is a scale mixture: the moments of the limit factor into moments of the base law and moments of the coefficient law, with all odd moments zero.","Diagonal entries whose average squared magnitude goes to zero do not affect the limit.","The joint empirical distribution of the diagonal coefficient matrix and the tridiagonal matrix converges to that of independent (T,X), so all mixed moments factorize.","The main theorem includes a truncation argument that reduces general L^2 entries to bounded entries."],"fun_headline_variants":["Tridiagonal spectra: finite variance suffices for a.s. limit","Almost sure spectral limit for random tridiagonal matrices","Analytic proof relaxes moment assumptions for tridiagonal spectra","Tridiagonal eigenvalues: convergence under second moment only","Stieltjes transform yields a.s. limit for tridiagonal matrices"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proof of Theorem 3.8 passes from truncated off-diagonal entries to the full ones by asserting, without proof, that the limiting spectral measures for the truncated models converge to the limit for the untruncated model as the truncation levels grow; this continuity step is the load-bearing assumption.","fun_headline_variants_meta":{"raw":{"variants":["Tridiagonal spectra: finite variance suffices for a.s. limit","Almost sure spectral limit for random tridiagonal matrices","Analytic proof relaxes moment assumptions for tridiagonal spectra","Tridiagonal eigenvalues: convergence under second moment only","Stieltjes transform yields a.s. limit for tridiagonal matrices"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000595,"raw_usage":{"total_tokens":2568,"prompt_tokens":632,"completion_tokens":1936,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":376,"completion_tokens_details":{"reasoning_tokens":1862}},"tokens_in":376,"tokens_out":1936,"duration_ms":12392,"temperature":1.0,"reasoning_tokens":1862,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T18:44:07.542464+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the model with b_i having a distribution with finite second moment but heavy tail (e.g., Pareto with tail index 3) and with σ_{k,N} satisfying (SV), (EM), (SM); compare the empirical spectral distribution for large N with the law of T·X computed from the Stieltjes equation. If the histograms systematically deviate, the claimed truncation continuity fails.","supporting_citations":[],"review_version":1}