{"id":"ef3c86ae-4329-41a0-bed5-0bb4e9e49c69","arxiv_id":"2607.24188","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For even C² potentials and completely resonant phases, L(E)<2β(α) precludes eigenvalues, resolving the missing half of the Avila–Jitomirskaya transition for almost Mathieu.","lead":"The paper proves that quasiperiodic Schrödinger operators with even potentials have no eigenvalues at completely resonant phases whenever the Lyapunov exponent is less than twice the arithmetic resonance strength β(α). This finishes the Avila–Jitomirskaya conjecture for the almost Mathieu operator in that phase class, giving purely singular continuous spectrum in the remaining open window.","discovery_kind":"new_method","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"No significant objection identified. The argument is short, fully on the page, and every load-bearing matrix identity I re-derived (Lemma 3.1, Eq. (30), the coordinate limits to −1) checks out; the reader's flagged C² assumption is a hypothesis of the theorem, not a gap in its proof.","rationale":"I agree with the reader that the C²-regularity input to Lemma 2.2 is the argument's most exposed assumption: it is exactly where the h² gain comes from, and it is the point that would fail first under any weakening of hypotheses. But exposure is not the same as a gap. For the theorem as stated, C² is a standing hypothesis, the estimate (8)–(9) is correctly derived from the product formula with the q² factor accounted for, and the application (Corollary 1.2) uses v=2λcos, which is analytic. I independently re-verified the algebraic core of both cases: the reflection identities (18) and (47), the conjugations (23)–(24) and (48), the explicit B⁻ formulas (30) and (50), and the terminal computations showing a coordinate of (2B₀−B⁻)X₁ tends to −1 while the vector must tend to 0. The contradiction is forced by det B₀=1 alone once a,b (or c,d) → 0, which in turn follows from u∈ℓ². The parity splitting is licensed by eigenvalue simplicity. I found no circularity: the only external inputs are the standard uniform Lyapunov bound (Lemma 2.1) and, for the corollary, the known AMO Lyapunov exponent and absence of ac spectrum for |λ|>1. Given that this resolves the remaining half of a named conjecture with an elementary, fully self-contained proof, and my stress pass surfaced no correctness risk beyond the (explicitly stated) regularity hypothesis, the reader's ACCEPT with HIGH confidence stands. The concrete test proposed — high-precision verification of the exact block identity and the genuine h² (not h) scaling of the symmetric difference — is cheap and would convert the already-high confidence into near-certainty about the central mechanism.","tokens_in":9459,"tokens_out":7222,"duration_ms":199377,"concrete_test":"Numerically stress the quadratic-gain mechanism and the block identity for a concrete instance: take v(x)=2λcos(2πx) with λ=2, α=(√5−1)/2, θ=0, q=F_n a large Fibonacci number, h=qα−p. Compute A_q(0), A_q(±h) in high-precision arithmetic. Check (i) T⁻R(h)T⁰=R(0) holds to machine precision (validates Lemma 3.1), and (ii) the ratio ‖A_q(h)+A_q(−h)−2A_q(0)‖/(h²‖A_q(0)‖) stays bounded (≈O(q²)) as n grows, rather than the ratio with h¹ in the denominator — i.e., confirm the first-order term cancels. If (ii) failed and the second difference were O(h·e^{Lq}), the doubled threshold mechanism would collapse back to L<β.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I attempted to find a genuine soft spot and could not. The proof has three load-bearing pieces, and each survives direct scrutiny. (1) Lemma 2.2's quadratic estimate: the double sum in A_q'' has ~q² terms each bounded by C³‖v'‖²e^{(L+3ε)q} via the uniform bound of Lemma 2.1, giving (8); the second difference (9) then follows from Taylor with sup‖A_q''‖, and the q² polynomial factor dropped in (32) is legitimately absorbed into ε since L−2γ+ε<0. (2) The block identities: in Case I, T⁻R(s+h)T⁰=R(s) (Lemma 3.1) follows from applying the transfer relation to the reflected solution φ̃(n)=φ(−n); I verified the phase arithmetic (s−qα=s−h mod 1 is the starting phase of the block from −q to 0) and the determinant signs (−1 both sides). B⁻=DB₀⁻¹D_h and the explicit form (30) follow from D_h²=I; the key computation (2B₀−B⁻)X₁=(2a²−1+η_h ac, 2bd+η_h a²)ᵀ verifies against (28)–(30), and the −1 limit uses a→0 plus |η_h|‖B₀‖→0 (both justified: η_h=O(h²) needs only v'(s)=0 and C², i.e. (11) and (25)). Case II is cleaner (identity (47) is exact, no η term) and I verified PΦ(q)=(φ(q−1),φ(q))ᵀ and B⁻=DB₀⁻¹D=[[d,c],[b,a]]. (3) The parity reduction uses simplicity of eigenvalues (Lemma 2.3), which is standard for 1D discrete Schrödinger. The reader's weakest assumption — C² regularity feeding Lemma 2.2 — is real as a *sharpness* question (for v∈C^{1+ε} the second difference degrades to O(h^{1+ε}) and the threshold would drop to (1+ε)β), but the theorem as stated hypothesizes C² and the AMO application is analytic, so this does not touch the central claim. I also checked the reduction of 2θ∈αZ+Z to the four canonical phases, the reflection centers at s∈{0,1/2}, and the choice of the resonance sequence from the limsup definition of β(α); all are correct. The one place the argument could conceivably hide an error is the claim that the first-order errors genuinely cancel in the *conjugated* quantity Γ_q rather than only in the raw T-blocks — but since S_s is x-independent, Γ_q=S_s⁻¹(A_q(s+h","agreement_with_reader":"agree"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The paper studies one-frequency quasiperiodic Schrödinger operators H = Δ + v(θ+nα) on ℓ²(ℤ) with v ∈ C²(𝕋,ℝ) even. The main result, Theorem 1.1, states that for completely resonant phases (2θ ∈ αℤ+ℤ), an energy E with Lyapunov exponent L(E) < 2β(α) cannot be an eigenvalue — doubling the classical Gordon threshold β(α). The mechanism is new: instead of a one-sided comparison of transfer blocks (which yields one factor of ‖qα‖), the author forms the symmetric second difference A_q(x+h)+A_q(x−h)−2A_q(x), gaining ‖qα‖², and applies the resulting block identity at the midpoint U(q) rather than at U(0). After reducing to four canonical phases, the resonant block relation T⁻R(s+h)T⁰ = R(s) (Case I) or T⁻PT⁰ = P (Case II) is conjugated so that parity eigenfunctions map to standard basis vectors, and a contradiction follows from the exact formula B⁻ = DB₀⁻¹D_h (resp. DB₀⁻¹D) together with ad−bc = 1. Corollary 1.2 gives purely singular continuous spectrum for the almost Mathieu operator at completely resonant phases when 1 < |λ| < e^{2β(α)}, settling the absence-of-eigenvalues half of the Avila–Jitomirskaya conjecture on the sharp phase transition.","tokens_in":9951,"tokens_out":7693,"duration_ms":219458,"significance":"If correct — and I believe it is — this completes, together with the author's earlier localization result [Liu25] for |λ| > e^{2β(α)}, the Avila–Jitomirskaya conjecture on the sharp arithmetic spectral transition at completely resonant phases of the almost Mathieu operator: purely singular continuous for 1 < |λ| < e^{2β(α)}, Anderson localized for |λ| > e^{2β(α)}. This is a well-known open problem and the threshold e^{2β(α)} is the expected sharp one, so the result is of clear interest to the community. Equally valuable is the method: it is the first argument to combine repetition (Gordon) and reflection (palindromic) symmetries so as to extract a quadratic resonance factor, and it does so in a short, elementary, and fully self-contained proof. Every load-bearing identity is on the page and verifiable by hand — no black boxes beyond standard facts. The method should apply beyond the almost Mathieu case (the theorem is already stated for general even C² sampling functions). This is a strong paper for the journal.","major_comments":[],"minor_comments":[{"comment":"§1.2: the statement \"For every energy in the spectrum of the almost Mathieu operator, its Lyapunov exponent is L(E) = max{log|λ|,0}\" is a theorem (Avila–Jitomirskaya; also Bourgain–Jitomirskaya for a.e. α) and should be cited at the point of use — [AJ09] is already in the reference list, so only an in-text citation is needed.","section":"§1.2"},{"comment":"§1.2: the deduction of pure singular continuity uses that the supercritical (|λ|>1) AMO has no absolutely continuous spectrum. Please add a citation (e.g., the Last–Simon inequality together with positivity of L on the spectrum, or the global theory of Avila–Jitomirskaya).","section":"§1.2"},{"comment":"Lemma 2.1 (uniform upper bound sup_x ‖A_n(x)‖ ≤ C_ε e^{(L+ε)n}) is quoted without a reference; a citation to Furman's theorem on uniform bounds for SL(2,ℝ) cocycles (or an equivalent standard source) should be added.","section":"§2, Lemma 2.1"},{"comment":"In (32) and (33) the polynomial factor q² from Lemma 2.2 is silently absorbed into the exponential: one should say explicitly that ε is first chosen so that L − 2γ + 2ε < 0 (possible since γ > L/2), after which q²e^{(L−2γ+ε)q} ≤ e^{(L−2γ+2ε)q} → 0. As written, the displayed inequality Cq²h²e^{(L+ε)q} ≤ Ce^{(L−2γ+ε)q} reuses the same ε on both sides.","section":"§3.1, Eqs. (32)–(33)"},{"comment":"Notation: the abstract defines β(α) with ‖kα‖_{R/Z} while the body defines and uses ‖x‖_𝕋; please unify. Also, a sentence noting that the empty interval 1<|λ|<e^{2β(α)} when β(α)=0 makes Corollary 1.2 vacuous (and that β=∞ is allowed) would orient the reader.","section":"Notation"},{"comment":"Typos/presentation: §3.1 \"We refer readers to the remark in Section 4.3 at the of this paper\" (missing word); §4.2 \"As in the Case I\" (drop \"the\"); §1 sentence ending \"...for further discussion. [Jit23].\" has doubled punctuation; §4.3 remark ends with a stray comma after \"estimate B⁻ + B⁺ − 2B₀,\".","section":"Various"},{"comment":"The reference [Liu] (\"A new proof of the sharp Gordon's lemma\", Pure Appl. Funct. Anal., to appear) should be updated with full bibliographic data if it has appeared by the time of revision.","section":"References"},{"comment":"Optional: a brief remark on sharpness would be valuable. For the AMO the 2β threshold is sharp by [Liu25]; for general even C² potentials, Theorem 1.1 gives the natural barrier of the method (the quadratic gain saturates at two resonance factors), and whether the bound is optimal in that generality is an interesting open question worth one sentence.","section":"§1.1 or §4.3"}],"recommendation":"accept","confidential_remarks":"I read the proof in full and independently re-derived the load-bearing algebra: the second-difference estimate in Lemma 2.2 (the q² factor from the double sum in A_q'' is legitimately absorbed into ε), the block identity T⁻R(s+h)T⁰ = R(s) of Lemma 3.1 (including the phase arithmetic s−qα = s−h mod 1 and the determinant signs), the conjugations (23)–(24) and (48), the exact formulas (30) and (50) for B⁻, and the coordinate limits to −1 in both parity subcases of Case I and both σ = ±1 subcases of Case II. Everything checks; the argument is genuinely self-contained modulo standard facts (uniform Lyapunov bound, simplicity of eigenvalues, the AMO Lyapunov formula). The C² hypothesis is a hypothesis of the theorem, not a gap: for v merely C^{1+τ} the second difference degrades and the argument would only give (1+τ)β, but that is a sharpness question outside the stated result. Note that the same author proved the complementary localization half [Liu25], so acceptance of this paper completes the Avila–Jitomirskaya conjecture in print; the editor may wish to keep that context in mind. No concerns about overlap or citation practice. The manuscript is short and clean; the minor items (mostly missing citations for three quoted background facts) can be handled at proof stage or in a light revision."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The paper closes the remaining absence-of-eigenvalues half of the Avila–Jitomirskaya conjecture for completely resonant phases of the almost Mathieu operator: when 2θ ∈ αℤ + ℤ and 1 < |λ| < e^{2β(α)}, the spectrum is purely singular continuous. The localization side was already in Liu25; this supplies the matching upper bound.\n\nWhat is new is the simultaneous use of approximate repetition and palindromic symmetry. Instead of the usual one-sided Gordon comparison (which only sees a single factor of ||qα|| and stops at L < β), the author works with the centered second difference ||A_q(x+h) + A_q(x−h) − 2A_q(x)||. First-order errors cancel, the quadratic |h|² appears, and the threshold doubles to L < 2β. The comparison is applied after one transfer block (at site q rather than at 0), so the two symmetries interact with the dynamics. After reducing to four canonical phases and conjugating so that even/odd (or σ = ±1) eigenfunctions become standard basis vectors, the contradictions are short matrix multiplications whose coordinates tend to −1.\n\nThe argument is fully written, elementary, and self-contained once standard facts (Lyapunov exponent, simplicity of eigenvalues, definition of β) are granted. I re-checked the load-bearing identities (Lemma 3.1, the explicit form of B_−, the coordinate limits) and they hold; the stress-test note is right that there is no hidden gap. The C² hypothesis is used exactly where it must be (second-derivative bound on the cocycle) and is stated up front; for analytic AMO it is automatic. The only minor soft spot is that the method as written does not immediately give the same threshold for merely C^{1+ε} potentials, but that is outside the theorem’s claim.\n\nThis is for people who work on quasiperiodic spectral theory or arithmetic transitions. It deserves a serious referee and should be engaged with; I would cite the doubled-threshold statement and the AMO corollary.","headline":"Clean elementary proof that doubles the Gordon threshold at completely resonant phases and finishes the open half of the Avila–Jitomirskaya conjecture for AMO.","tokens_in":11292,"tokens_out":541,"would_cite":true,"duration_ms":8517,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["47B36","37A30","81Q10"],"pacs":[],"model":"grok-4.5","headline":"For even quasiperiodic potentials at completely resonant phases, no energy with Lyapunov exponent below twice the frequency-resonance strength can be an eigenvalue.","keywords":["quasiperiodic Schrödinger operators","Gordon lemma","palindromic potentials","Lyapunov exponent","almost Mathieu operator","singular continuous spectrum","completely resonant phases","arithmetic transitions"],"falsifier":"Exhibit a nonzero ℓ² eigenfunction for an even C² potential at a completely resonant phase with L(E) < 2β(α), or show that the second-difference estimate already fails for some C^{1,α} even potential while an eigenvalue still appears below 2β(α).","tokens_in":10810,"feed_emoji":"🔁","tokens_out":960,"duration_ms":26440,"temperature":0.7,"pith_summary":"This paper proves that one-frequency quasiperiodic Schrödinger operators with an even C² sampling function have no square-summable eigenfunctions at completely resonant phases whenever the Lyapunov exponent is strictly less than twice the arithmetic resonance strength β(α). The usual Gordon and palindrome arguments each only reach the single factor β(α); the new method compares three neighboring transfer blocks symmetrically so that first-order errors cancel and a quadratic factor ∥qα∥² appears, doubling the threshold. Applied to the almost Mathieu operator, the result shows that the spectrum is purely singular continuous throughout the window 1 < |λ| < e^{2β(α)} at those phases. That closes the remaining absence-of-eigenvalues half of the sharp spectral-transition picture for completely resonant phases.","feed_headline":"Doubled Gordon bound bars eigenvalues for resonant phases","feed_subtitle":"Even quasiperiodic operators have no point spectrum below twice the frequency-resonance strength","key_machinery":"A symmetric second-difference comparison of transfer-matrix blocks: ∥A_q(x+h) + A_q(x−h) − 2A_q(x)∥ is controlled by |h|² times a sub-exponential factor. First-order errors cancel, producing the doubled arithmetic threshold 2β(α); the comparison is applied at the intermediate site after one block so that repetition and palindromic symmetry act together.","core_discovery":"If v is even and C², α is irrational, and the phase satisfies 2θ ∈ αℤ + ℤ, then any energy E with Lyapunov exponent L(E) < 2β(α) admits no nonzero ℓ²(ℤ) solution of H_{v,α,θ}u = Eu. For the almost Mathieu operator this yields purely singular continuous spectrum whenever 1 < |λ| < e^{2β(α)} at the same phases.","pith_inferences":["If the C² hypothesis can be relaxed to C^{1,α} while keeping a usable second-difference bound, the same doubled threshold would extend to a larger class of sampling functions.","The intermediate-site comparison may adapt to other reflection-symmetric quasiperiodic models (e.g., extended Harper or CMV cocycles) where both repetition and palindrome structure are present.","Quantitative decay rates of the would-be eigenfunction along the resonant scales q could yield effective estimates on the spectral measure even when L(E) is only slightly below 2β(α)."],"forward_implications":["At completely resonant phases the almost Mathieu operator has empty point spectrum throughout 1 < |λ| < e^{2β(α)}.","Combined with existing localization above e^{2β(α)}, the spectral type is completely classified for those phases: singular continuous below the threshold and pure point above it.","Any even C² quasiperiodic potential inherits the same doubled absence-of-eigenvalues threshold L(E) < 2β(α) at completely resonant phases.","The method supplies a template for combining repetition and reflection symmetries in other one-dimensional cocycle settings."],"fun_headline_variants":["Doubled Gordon bound bars eigenvalues at resonant phases","Even potentials: no eigenvalues if L(E)<2β(α)","Palindromic Gordon method doubles eigenvalue threshold","Resonant phases eigenvalue-free below twice β(α)","Almost Mathieu singular continuous for |λ|<e^{2β(α)}"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The sampling function must be twice continuously differentiable; that smoothness is what turns the first-order cancellation into a usable quadratic bound on the transfer-matrix difference.","fun_headline_variants_meta":{"raw":{"variants":["Doubled Gordon bound bars eigenvalues at resonant phases","Even potentials: no eigenvalues if L(E)<2β(α)","Palindromic Gordon method doubles eigenvalue threshold","Resonant phases eigenvalue-free below twice β(α)","Almost Mathieu singular continuous for |λ|<e^{2β(α)}"]},"model":"grok-4.5","effort":"low","cost_usd":0.00421,"raw_usage":{"total_tokens":1332,"prompt_tokens":891,"num_sources_used":0,"completion_tokens":71,"cost_in_usd_ticks":42104000,"prompt_tokens_details":{"text_tokens":891,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":370,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":891,"tokens_out":71,"duration_ms":7456,"temperature":1.0,"reasoning_tokens":370,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T21:27:18.317040+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Exhibit a nonzero ℓ² eigenfunction for an even C² potential at a completely resonant phase with L(E) < 2β(α), or show that the second-difference estimate already fails for some C^{1,α} even potential while an eigenvalue still appears below 2β(α).","supporting_citations":[],"review_version":1}