{"id":"3c69b45e-c9c9-4866-be87-0812845a8665","arxiv_id":"1908.07977","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Using periodic approximations and the first Bloch eigenvalue, the authors recover the homogenization limit for almost periodic elliptic operators and prove a power-law rate for the convergence of approximate homogenized tensors.","lead":"This paper extends the Bloch wave method for periodic materials to almost periodic media by approximating the coefficients with periodic ones on larger and larger boxes. It proves the homogenized equation is recovered and gives a convergence rate for the approximate effective coefficients, with numerical tests.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sign error in Eq. (5.21) contradicts Theorem 4.6(3) and invalidates the flux-identification step of Theorem 5.1 as written; the corrected sign would yield M(a)-M(a∇w) instead of the homogenized tensor (4.12).","rationale":"The reader's conditional verdict is reasonable, but the weakest assumption they flagged (§8.7) is not the most load-bearing defect. The sign error in §5.2.4 is an internal inconsistency in the proof of the paper's main qualitative theorem: the sign in (5.21) is opposite to what Theorem 4.6(3) gives, and restoring the correct sign flips the sign of the corrector contribution in (5.22), so the flux-limit identification would yield M(a)-M(a∇w) instead of (4.12). The theorem statement may still be true, and the error might be a typographical slip, but as written the proof of Theorem 5.1 does not establish the conclusion. The §8.7 estimate is also under-supported ('by a similar analysis to [45]'), but it affects only the quantitative rate. Both issues support the conditional verdict; the sign error should be an explicit condition for acceptance. The 1D periodic example in the concrete test provides a decisive check.","tokens_in":28071,"tokens_out":35655,"duration_ms":348039,"concrete_test":"Re-derive Eq. (5.21) from Theorem 4.6(3) by differentiating ∂η_s φ_R^1 - i φ_R^1 w_{R,s} = const and taking the mean against a_R,kl; if the mean is +i(2π)^{-d/2} M(a_R,kl ∂_{y_k}w_{R,s}), substitute this into (5.12)-(5.22) and check the sign of the resulting flux-limit against (4.12). As a decisive numerical check in one dimension with a(y)=2+sin y (taking R an integer multiple of 2π so a_R=a), compare the tensor implied by the printed proof, M(a)-M(a w'), with the true a* = 1/M(1/a); they differ by roughly 2.27 vs 1.73.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of the qualitative homogenization theorem has an internal sign inconsistency. Theorem 4.6(3) states that ∂η_s φ_R^1(y;0) - i φ_R^1(y;0) w_{R,s}(y) is constant. Since φ_R^1(·;0) is itself constant by Remark 4.1, differentiating in y_k gives ∂_{y_k}∂_{η_s}φ_R^1 = i φ_R^1 ∂_{y_k} w_{R,s}, and hence M(a_R,kl ∂_{y_k}∂_{η_s}φ_R^1) = + i (2π)^{-d/2} M(a_R,kl ∂_{y_k}w_{R,s}). Equation (5.21) asserts the same mean equals -i (2π)^{-d/2} M(a_R,kl ∂_{y_k}w_{R,s}). Using the correct sign in (5.12), the first term contributes -i φ0 ξ_s M(a_R,sl)(∂ψ0 u*)ˆ and the second contributes +i φ0 ξ_s M(a_R,kl ∂_{y_k}w_{R,s})(∂ψ0 u*)ˆ; the combination is -i φ0 ξ_s [M(a_R,sl)-M(a_R,kl ∂_{y_k}w_{R,s})](∂ψ0 u*)ˆ, not the '+' combination claimed in (5.22). Thus the Fourier-space limit (5.27) and the identification (5.33) do not follow as written; the proof would identify a tensor M(a)-M(a∇w), which contradicts (4.12) (in 1D with a=2+sin x this is about 2.27 instead of the correct 1.73). This is a separate, more basic defect than the §8.7 rate estimate.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Bloch-wave homogenization method for scalar elliptic operators with almost periodic coefficients. It replaces the almost periodic operator by periodic truncations on cubes of side length 2πR, studies the first Bloch eigenvalue of each truncation, and claims that the almost periodic effective tensor is the limit of half the Hessian of this eigenvalue at the origin (Theorem 5.1 and Eq. (4.17)). It further claims a quantitative rate for the approximation of the effective tensor under a power-law modulus of almost periodicity (Theorem 8.1). The last part of the paper contains numerical experiments for periodic and quasiperiodic examples. The central qualitative theorem is proved by passing to the limit in the first Bloch coefficient, and the quantitative theorem is proved by splitting the error into regularized, truncated, boundary, and periodic-cell terms.","tokens_in":28503,"tokens_out":24298,"duration_ms":227268,"significance":"If the main results were correct, the paper would give a spectral route to almost periodic homogenization and a quantitative justification for computing effective coefficients from Bloch eigenvalue data of periodic truncations. The paper has several strengths: it formulates a clear two-parameter limiting procedure, states a useful module-containment result for correctors (Lemma 6.3), and complements the analytic claims with numerical experiments. However, the manuscript as written does not establish its main theorem: the flux-identification step in the proof of Theorem 5.1 contains a sign inconsistency, and the rate estimate in Theorem 8.1 depends on an unproved decay estimate in Section 8.7. These are load-bearing issues, not presentation problems.","major_comments":[{"comment":"The identity (5.21) has the wrong sign. Theorem 4.6(3) states that ∂_{η_s}φ_R^1(y;0) − i φ_R^1(y;0) w_{R,s}(y) is constant in y, and by Remark 4.1 φ_R^1(·;0) is the constant (2π)^{-d/2}. Differentiating in y_k gives ∂_{y_k}∂_{η_s}φ_R^1(y;0) = i (2π)^{-d/2} ∂_{y_k} w_{R,s}(y). Hence the mean on the left of (5.21) equals + i(2π)^{-d/2} M(a_R,kl ∂_{y_k}w_{R,s}), not − i(2π)^{-d/2} M(a_R,kl ∂_{y_k}w_{R,s}). With the corrected sign, the second term in (5.12) contributes + i(2π)^{-d/2} ξ_s M(a_R,kl ∂_{y_k}w_{R,s}) times the common factor, so the combination in (5.22) becomes −i ξ_s [M(a_R,kl) − M(a_R,kl ∂_{y_k}w_{R,s})] instead of −i ξ_s a_R,∗_kl. Thus the Fourier-space limit (5.27) and the flux identification (5.33) do not follow as written; in the one-dimensional example a = 2 + sin x the resulting tensor would be approximately 2.27 rather than the correct 1.73. This sign inconsistency is a load-bearing defect in the proof of Theorem 5.1.","section":"§5.2.4, Eq. (5.21)"},{"comment":"The estimate ‖~w_{R,ξ}‖_{L2(Y1)} ≤ C_γ R^{−γ} for 0 < γ < τ/(τ+1) is introduced with the phrase 'by a similar analysis to [45]', but no theorem or lemma in [45] is stated that gives this exact estimate, and no derivation is provided. This estimate controls the term R^{4−2γ} T^{−2} in (8.35), so the power-law rate |A* − A^{R,*}| ≲ R^{−β} in Theorem 8.1 is not established. In addition, the passage from (8.31)–(8.32) to the limiting zero solution is not justified as written: (8.33) is a boundary-value problem with a forcing h, whereas (8.31) is a Y1-periodic cell problem with h = 0, and the claimed limit equation −∇·(A*(ξ+∇~w_∞)) = 0 has no Y1-periodic solution for general ξ ≠ 0. A precise proof or a precise reference with all hypotheses is required.","section":"§8.7, Theorem 8.9"},{"comment":"The displayed estimates for ‖v^R_k‖_{L2} and ‖z^R_s‖_{L2} are written with the norm ‖a_R − a‖_{L∞(ϵK)}; after the substitution y = x/ϵ, the quantity that appears is ‖a_R − a‖_{L∞(ϵ^{-1}K)}, and with either reading the displayed bound does not justify the claimed vanishing in the iterated limit ϵ→0 then R→∞. The conclusion that the terms involving v^R and z^R drop out of (5.26) is therefore unsupported. Since these terms are part of the proof of Theorem 5.1, a rigorous argument for their vanishing, or an explicit computation of their limits, is needed.","section":"§5.3, estimates before (5.27)"}],"minor_comments":[{"comment":"In the displayed line after (7.2), 'λ^{R,ϵ}_m(ξ) = ϵ^{-2}λ^{R,ϵ}_m' should read 'λ^{R,ϵ}_m(ξ) = ϵ^{-2}λ^R_m(ϵξ)'.","section":"§7, proof of Proposition 7.1"},{"comment":"The sentence 'As a consequence, ϵ^{-2}∂γ_R/∂y_k ∈ ...' appears inconsistent with the factor ϵ^{-1} in (5.15); the exponent should be checked and the convergence statement made precise.","section":"§5.2.4, Eqs. (5.15)–(5.17)"},{"comment":"There are minor notational inconsistencies, for example a^ϵ_kl(ϵ) in (2.4) should presumably be a_kl(x/ϵ), and the cell correctors in (4.12) are written sometimes as w_{R,l} and sometimes as w_{R,p}; harmonizing these notations would improve readability.","section":"§2.3 and §4.4"}],"recommendation":"reject","confidential_remarks":"The sign error in Section 5.2.4 is not a local typo: it lies at the center of the proof of the paper's main theorem, and the corrected sign leads to a different tensor. Combined with the unproved decay estimate in Section 8.7, the manuscript does not currently provide a valid proof of either advertised theorem. The topic is interesting and the overall strategy may be salvageable, but the revision would require reworking the central derivation rather than a small correction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing to know: the claimed rate theorem for restrict-and-periodize approximations is a plausible new result, but the proof of the qualitative homogenization theorem has a sign error that breaks the flux identification. The paper needs serious revision, not just small fixes.\n\nWhat is genuinely good: the strategy of assembling Bourgeat–Piatnitski, Shen, Shen–Zhuge, and Gloria–Otto estimates into a rate for almost periodic effective tensors is reasonable, and the rate theorem (8.1) is not verbatim in the literature. The module containment lemmas in Section 6 are a nice byproduct. The literature engagement is honest; the self-citations are to related work, not self-promotion.\n\nThe soft spots are real. In Section 5.2.4, equation (5.21) is sign-flipped. Theorem 4.6(3) gives ∂_η φ − i φ w = const; since φ is constant at η=0, ∂_y ∂_η φ = +i φ ∇w, so the mean in (5.21) should be +i(2π)^{-d/2} M(a∇w), not −i. With the correct sign, the flux limit in (5.22) becomes M(a) − M(a∇w) rather than the homogenized tensor; in 1D with a=2+sin x that is about 2.27 instead of the correct 1.73. This is a load-bearing internal contradiction, not a typo in isolation.\n\nThe Section 5.3 vanishing of v^R and z^R is also shaky: the argument x/ϵ lives in ϵ^{-1}K, not ϵK, and the iterated limit is not justified by the displayed bound. Section 8.7 is weaker still: the estimate ||w̃_{R,ξ}||_{L2(Y1)} ≤ C R^{-γ} is asserted by “a similar analysis to [45]” without a precise statement or proof, and Theorem 8.1 depends on it.\n\nNone of this is fatal to the underlying ideas; the rate theorem is likely recoverable with a real proof, and the qualitative theorem is already known from Kozlov's work. But as written, the paper's own equations contradict each other. I would send it to a serious referee only with the expectation of major revision. The audience is homogenization specialists, particularly people interested in quantitative effective tensor approximation for almost periodic media. I would not cite the rate theorem until the gaps are closed.","headline":"The rate theorem is a plausible new contribution, but the qualitative proof has a sign error that contradicts the paper's own Theorem 4.6(3); major revision needed before Theorem 5.1 is reliable.","tokens_in":28997,"tokens_out":23176,"would_cite":false,"duration_ms":186698,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["47A55","35J15","35B27","34C27"],"pacs":[],"model":"deepseek-v4-flash","headline":"Periodizing an almost periodic medium on growing cubes lets Bloch-wave spectral analysis recover its effective tensor, with a power-law error rate when the deviation from periodicity decays algebraically.","keywords":["Bloch eigenvalues","Almost periodic operators","Homogenization","Periodic approximation","Effective coefficients","Convergence rate","Besicovitch space","Quasiperiodic media"],"falsifier":"Solve the periodic cell problem for $A=4+\\cos(2\\pi x)+\\cos(2\\pi\\sqrt{2}\\,x)$ on $Y_R$ for $R$ from 10 to 400, form $A^{R,*}$ via (4.12), and regress $\\log|A^{R,*}-A^*|$ on $\\log R$ against a converged reference value; Theorem 8.1 predicts an eventual negative slope $-\\beta$ for every $R$ beyond some threshold, so an asymptotic slope of zero, or error curves that flatten, would falsify the rate claim for that coefficient. The qualitative theorem is instead checked by testing weak flux convergence on a quasiperiodic coefficient with a spectrally solvable cell problem.","tokens_in":27856,"feed_emoji":"📐","tokens_out":10461,"duration_ms":100724,"temperature":0.7,"pith_summary":"Bloch-wave homogenization is a spectral recipe for periodic media: the effective tensor is read off from the curvature of the lowest Bloch band. This paper claims the same recipe works for almost periodic media, provided one first truncates the medium to a large cube, periodizes it, runs Bloch analysis on that periodic approximation, and then lets the cube size tend to infinity. The limiting effective tensor is identified as the limit of half the Hessian of the first Bloch eigenvalue of these periodizations, and it matches the tensor obtained by the established abstract almost periodic homogenization theory. When the coefficients' deviation from periodicity decays like a power law, the paper further claims a power-law convergence rate for the approximated tensors, and its numerical experiments on periodic and quasiperiodic examples show errors that appear to decay polynomially.","feed_headline":"Bloch waves recover effective tensors of almost periodic media","feed_subtitle":"Truncating, periodizing, and reading the lowest Bloch band yields the homogenized tensor and a power-law error estimate.","key_machinery":"The machinery has two parts. First, the \"restrict and periodize\" approximation: take $f\\in AP(\\mathbb{R}^d)$, set $f^R=f$ on $Y_R$, and extend $2\\pi R$-periodically; the operator $A^R$ then has a genuine Bloch direct-integral decomposition with fiber operators on $L^2(Y_R)$. Second, the identity $\\frac{1}{2}\\frac{\\partial^2\\lambda^R_1}{\\partial\\eta_k\\partial\\eta_l}(0)=a^{R,*}_{kl}$, which expresses the homogenized tensor of the periodization as the curvature at zero of the lowest Bloch band; analyticity of the first Bloch eigenvalue and eigenvector near $\\eta=0$, a consequence of the first spectral gap, makes this Hessian meaningful. The paper combines these with the almost periodic cell problem in the Besicovitch space $B^2(\\mathbb{R}^d)$, whose solution $N_\\xi$ has frequencies contained in those of $A$, and with a four-term splitting of the error $|A^*-A^{R,*}|$ into regularized-corrector errors, a mean-ergodic truncation error, a Green's-function boundary term, and a periodic-corrector decay term.","core_discovery":"The central discovery is that an almost periodic operator, which has no genuine direct-integral Bloch decomposition, can nonetheless be homogenized from Bloch data of its periodic truncations. For each $R$, let $A^R$ be the $Y_R$-periodic function obtained by restricting $A$ to the cube $Y_R=[-\\pi R,\\pi R)^d$ and periodizing. The paper proves that after sending $\\epsilon\\to 0$ and then $R\\to\\infty$, solutions of $A^\\epsilon u^\\epsilon=f$ converge weakly in $H^1(\\Omega)$ to the solution of $A_hom u=f$, with flux convergence, where the homogenized tensor is $A^*_{kl}=\\lim_{R\\to\\infty}\\frac{1}{2}\\frac{\\partial^2\\lambda^R_1}{\\partial\\eta_k\\partial\\eta_l}(0)$, the limit of half the Hessian of the first Bloch eigenvalue $\\lambda^R_1$ of $A^R$. It then shows this tensor coincides with the abstract almost periodic homogenized tensor. Under the hypothesis that the modulus of almost periodicity $\\rho(A,L)$ decays as $L^{-\\tau}$, it proves $|A^*-A^{R,*}|\\lesssim R^{-\\beta}$ for some $\\beta\\in(0,1)$, and it provides numerical evidence for such rates.","pith_inferences":["A testable extension not pursued in the paper: for quasiperiodic coefficients with a finite frequency module, the modulus $\\rho(A,L)$ is controlled by the Diophantine quality of the frequency vector, so one could compute the expected exponent $\\beta$ explicitly and check it against finite-element error slopes.","The same \"periodize, read the lowest-band Hessian, let $R$ grow\" construction should transfer to systems and to other operators with a spectral-gap structure, since only the analyticity of the first band and the Hessian identity are used; the missing ingredient in each case is the Section 8.7 periodic-corrector decay estimate.","If the Section 8.7 estimate fails for some almost periodic coefficient with $\\rho(A,L)\\lesssim L^{-\\tau}$, the qualitative theorem would survive but the power-law rate would not: rate and qualitative convergence are logically independent in this proof, so numerical rate experiments are the right discriminator between the two claims.","The paper's log-log plots do not report fitted slopes, so a reader should treat the numerics as indicative rather than as a measurement of a specific $\\beta$."],"forward_implications":["For any coefficient satisfying (A1)-(A3), the almost periodic oscillations force only a constant macroscopic tensor: the weak limit of $u^\\epsilon$ solves $A_hom u=f$ and the oscillating fluxes converge weakly to $a^*\\nabla u^*$.","The homogenized tensor of an almost periodic medium can be computed by finite-cell periodic problems, with the cell side $R$ as the only numerical parameter; Theorem 8.1 guarantees power-law accuracy once $\\rho(A,L)\\lesssim L^{-\\tau}$.","Higher Bloch modes of the periodization are negligible: their contribution to the solution is bounded by $C R\\epsilon$, so the lowest Bloch band alone determines homogenization.","The tensor obtained by the Bloch route is not a new object: it coincides with the established almost periodic homogenized tensor, so the spectral construction is a valid alternative representation of $A^*$.","Approximations by Dirichlet and Neumann cell problems can be handled by the same error splitting, replacing the periodic corrector decay estimate with the corresponding boundary-value estimates."],"supporting_citations":[{"why":"Foundational Bloch-decomposition machinery: direct integral representation, Bloch coefficients, analyticity of the first Bloch band near zero, and the Hessian identity.","marker":"[24]"},{"why":"Introduces the restrict-and-periodize approximation of the coefficient and proves convergence of the approximate homogenized tensors to the almost periodic tensor.","marker":"[18]"},{"why":"Defines the abstract almost periodic cell problem in the Besicovitch space and identifies the homogenized coefficients used as the limit object.","marker":"[40]"},{"why":"Supplies the homogenization theorem for almost periodic operators, including the convergence-of-solutions result used in the periodic-corrector step.","marker":"[33]"},{"why":"Source of the regularized-corrector rate estimate and the decay estimate for periodic correctors that Section 8.7 invokes by analogy.","marker":"[45]"},{"why":"Gives the quantified mean-ergodic theorem for truncated averages over $Y_R$ used to bound the truncation error.","marker":"[46]"},{"why":"Provides the pointwise Green's function bounds for the screened elliptic operator used to control the boundary term.","marker":"[31]"},{"why":"Introduces the modulus of almost periodicity $\\rho(A,L)$ and its decay hypothesis $L^{-\\tau}$ on which the rate theorem rests.","marker":"[9]"}],"fun_headline_variants":["Bloch waves crack almost periodic homogenization","Periodic truncations drive Bloch homogenization","Almost periodic tensors from Bloch band Hessians","Bloch eigenvalues approximate effective coefficients","Truncate, periodize, homogenize with Bloch waves"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the periodic correction field on a cube of side $R$, rescaled to the unit cell, shrinks in $L^2$ norm at rate $R^{-\\gamma}$ with $\\gamma<\\tau/(\\tau+1)$; the proof invokes a cited result for this decay rather than deriving it, and the power-law rate $|A^*-A^{R,*}|\\lesssim R^{-\\beta}$ depends on that decay.","fun_headline_variants_meta":{"raw":{"variants":["Bloch waves crack almost periodic homogenization","Periodic truncations drive Bloch homogenization","Almost periodic tensors from Bloch band Hessians","Bloch eigenvalues approximate effective coefficients","Truncate, periodize, homogenize with Bloch waves"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000189,"raw_usage":{"total_tokens":1319,"prompt_tokens":913,"completion_tokens":406,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":529,"completion_tokens_details":{"reasoning_tokens":337}},"tokens_in":529,"tokens_out":406,"duration_ms":3914,"temperature":1.0,"reasoning_tokens":337,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:53:46.134208+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve the periodic cell problem for $A=4+\\cos(2\\pi x)+\\cos(2\\pi\\sqrt{2}\\,x)$ on $Y_R$ for $R$ from 10 to 400, form $A^{R,*}$ via (4.12), and regress $\\log|A^{R,*}-A^*|$ on $\\log R$ against a converged reference value; Theorem 8.1 predicts an eventual negative slope $-\\beta$ for every $R$ beyond some threshold, so an asymptotic slope of zero, or error curves that flatten, would falsify the rate claim for that coefficient. The qualitative theorem is instead checked by testing weak flux convergence on a quasiperiodic coefficient with a spectrally solvable cell problem.","supporting_citations":[{"cited_title":"and Vanninathan, M","cited_arxiv_id":null,"evidence_quote":"Foundational Bloch-decomposition machinery: direct integral representation, Bloch coefficients, analyticity of the first Bloch band near zero, and the Hessian identity."},{"cited_title":"and Piatnitski, A","cited_arxiv_id":null,"evidence_quote":"Introduces the restrict-and-periodize approximation of the coefficient and proves convergence of the approximate homogenized tensors to the almost periodic tensor."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the abstract almost periodic cell problem in the Besicovitch space and identifies the homogenized coefficients used as the limit object."},{"cited_title":"V., Kozlov, S","cited_arxiv_id":null,"evidence_quote":"Supplies the homogenization theorem for almost periodic operators, including the convergence-of-solutions result used in the periodic-corrector step."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the regularized-corrector rate estimate and the decay estimate for periodic correctors that Section 8.7 invokes by analogy."},{"cited_title":"and Zhuge, J","cited_arxiv_id":null,"evidence_quote":"Gives the quantified mean-ergodic theorem for truncated averages over $Y_R$ used to bound the truncation error."},{"cited_title":"and Otto, F","cited_arxiv_id":null,"evidence_quote":"Provides the pointwise Green's function bounds for the screened elliptic operator used to control the boundary term."},{"cited_title":"N., Cardaliaguet, P., and Souganidis, P","cited_arxiv_id":null,"evidence_quote":"Introduces the modulus of almost periodicity $\\rho(A,L)$ and its decay hypothesis $L^{-\\tau}$ on which the rate theorem rests."}],"review_version":1}