{"id":"0037945d-367a-4daf-b1e1-0cd18ae6e475","arxiv_id":"1909.00708","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"low","formal_verification":"none","parameter_count":1,"one_line_summary":"Nonlocality is a generic outcome of homogenization and model reduction, and Bloch dispersion relations provide a practical route to construct nonlocal effective kernels.","lead":"This review connects homogenization, the study of effective equations for heterogeneous materials, with nonlocal models in which interactions act over a distance. It shows that model reduction and homogenization generically produce nonlocal operators, and that dispersion relations can guide the construction of nonlocal surrogate models.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 2.3's nonlocal surrogate is underdetermined: Eq. (2.19) admits many kernels, and the compact-support/fast-decay claim rests on an unproved smooth-extension step, so the advertised characterization of nonlocal interactions is conditional.","rationale":"The reader's weakest assumption correctly identifies the underdetermined kernel construction in Section 2.3 as the softest point. I agree with that identification and sharpen it: the smooth-extension argument is not merely a technical gap in a proof, it is the step that would turn a dispersion relation into a concrete nonlocal interaction kernel. Without it, the paper's statement that homogenization 'helps characterizing the nature and form of nonlocal interactions' is not fully supported. However, this concern does not overturn the paper's central claim. The nonlocal homogenized limit in Section 2.2 is a concrete local-PDE example with a non-polynomial symbol, and the model-reduction examples in Sections 3.1 and 3.4 show nonlocality in a rigorous, parameter-free way. The paper is a review that explicitly acknowledges the non-uniqueness in (2.19) and frames rigorous derivations as an open problem. The 'generic' phrasing is ambitious, but the underlying evidence is real and the limitations are disclosed rather than hidden. I therefore keep the reader's ACCEPT verdict unchanged, while recommending that the Section 2.3 construction be read as heuristic and conditional on a suitable kernel selection.","tokens_in":21191,"tokens_out":15615,"duration_ms":173703,"concrete_test":"Take the two-phase coefficient (2.2), compute \\(\\lambda_0(k)\\) on \\(Z\\) numerically, form two distinct \\(C^\\infty\\) compactly supported extensions of \\(\\lambda_0\\) to 1 outside \\(Z\\), and compute the two kernels \\(\\gamma\\) by inverse Fourier transform. If the kernels differ materially in sign, decay rate, or total variation, then (2.19) does not determine the nonlocal model and the Section 2.3 construction must be presented as a modeling choice, not a derived consequence.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim survives the Section 2.2 example, so this is a narrowing concern rather than a refutation. In Section 2.3, the nonlocal effective operator \\(\\bar L^\\epsilon\\) is defined by choosing \\(\\gamma\\) satisfying (2.19) on the Brillouin zone \\(Z\\). Equation (2.19) fixes only the Fourier symbol on \\(Z\\); infinitely many kernels satisfy it. The paper proposes to normalize \\(\\int\\gamma=1\\) and smoothly extend \\(\\lambda_0(k)\\) to 1 outside \\(Z\\), and says this 'naturally indicates' a kernel with compact support or fast decay. That inference is not proved: \\(\\lambda_0\\) is not shown to admit a \\(C^\\infty\\) extension on all of \\(\\mathbb R^d\\), and even with a smooth compactly supported extension, the inverse Fourier transform of \\(1-\\lambda_0\\) is entire and hence not compactly supported, only rapidly decaying at a rate set by the regularity of the extension. Moreover, different extensions yield different \\(\\gamma\\) with different sign and decay behavior, so the 'nature and form' of the nonlocal interaction is not characterized uniquely. The paper acknowledges the non-uniqueness, but the practical appeal of (2.20) as a canonical upscaled model depends on choosing a kernel with acceptable sign and decay, and that choice is not justified. Since Section 2.2 and the Schur-complement/Mori-Zwanzig discussions establish nonlocality independently, the main review claim stands, but the stronger 'characterization' claim is conditional.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript is a review article that develops the thesis that nonlocal operators arise generically from homogenization and model reduction of local differential equations. The authors use Peetre's theorem to observe that translation-invariant local operators must have polynomial Fourier symbols, so any effective operator with a non-polynomial symbol is nonlocal. They present a one-dimensional worked example in §2.2 in which homogenization in x of an operator with mixed derivatives yields a homogenized symbol that is genuinely non-polynomial in the Fourier variable k, thereby rigorously exhibiting a nonlocal homogenized limit. They also discuss a Bloch-wave-based construction of a nonlocal effective wave equation (§2.3), Tartar's memory-effect example (§2.1), Schur-complement/LOD equivalences (§3.1), MsFEM/HMM (§3.2), a coarse-grained lattice model (§3.3), and the Mori–Zwanzig formalism (§3.4), followed by open questions on asymptotically compatible schemes.","tokens_in":21499,"tokens_out":14949,"duration_ms":128028,"significance":"The paper's main mathematical contribution is the clean Peetre-theorem criterion and the explicit example in §2.2, which is complete and correct. The equivalence between the projection/Schur-complement homogenized system (3.1) and the LOD formulation (3.3) is also carefully derived and is a useful observation. The survey nature of the paper is a strength: it connects disparate literatures (analytical homogenization, numerical homogenization, nonlocal modeling, and molecular coarse graining) and gives specific, falsifiable expectations about when nonlocal operators should arise. The authors are honest about limitations, explicitly noting the non-uniqueness of kernels satisfying (2.19) and the heuristic nature of the smooth-extension step. If the review claims are accepted, the paper will be a valuable reference for researchers in multiscale modeling and nonlocal methods.","major_comments":[],"minor_comments":[{"comment":"In the paragraph following Eq. (2.19), the statement that 'The smoothness of 1 − λ0(k) for large k naturally indicates the possibility of a kernel γ with a compact support or fast decay' is imprecise for the construction described: if λ0 is smoothly extended to 1 outside Z, then 1 − λ0 has compact support, so its inverse Fourier transform is entire and can only be rapidly decaying, not compactly supported; please clarify that compactly supported kernels would require a different choice of solution to (2.19).","section":"2.3"},{"comment":"Section 3.3 attributes the main results on the coarse-grained lattice kernel (super-algebraic decay, sign properties, and the convergence of the rescaled kernel in (3.11)) to the unpublished preprint [26]; the transparency of the review would be improved by stating the provenance of these results and whether they are proven or numerically observed.","section":"3.3"},{"comment":"The statement that 'local operators are differential operators of finite order' is a consequence of the globally finite order version of Peetre's theorem cited as [10]; please remind the reader that the classical Peetre theorem yields locally finite sums, to avoid overstating the classical result.","section":"1.1"},{"comment":"In the expansion (2.8), the convergence of the geometric series relies on |a|/(1+k^2) < 1, which follows from |a| ≤ c0 < 1; it would be helpful to state this explicitly since the subsequent infinite series manipulations require uniform convergence in k.","section":"2.2"},{"comment":"The sentence about the Heterogeneous Multiscale Method being 'similar to the nonlocal version of the quasicontinuum method [66]' would benefit from a specific reference to a nonlocal quasicontinuum formulation, as the cited [66] is the classical local quasicontinuum method.","section":"3.2"},{"comment":"In Figure 3, the label 'Discrete Nonlocal' and the surrounding text use 'nonlocal' to denote both the continuum model and the discrete approximation; consider distinguishing 'nonlocal continuum model' from 'nonlocal discrete approximation' to avoid ambiguity.","section":"4.3"}],"recommendation":"minor_revision","confidential_remarks":"The paper is well within scope for the proceedings volume. The reliance on the unpublished preprint [26] for several concrete claims is a minor concern and should be flagged in the published version. The Section 2.3 heuristic about kernel support should be corrected for mathematical precision, but the central claims of the paper are not in doubt."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick read, before you spend time on it: this is a real review-plus-original-examples paper, and the central claim—homogenization and model reduction generically produce nonlocal effective operators, with classical local limits as special cases—holds up. It is not a breakthrough, but it deserves referee time. What is actually new is modest but real: Section 2.2 gives a one-dimensional PDE example where the homogenized symbol is non-polynomial in the transverse wave number, making the limit operator nonlocal; Section 2.3 builds a nonlocal surrogate wave equation from the first Bloch eigenvalue; Section 3.1 proves that the Schur-complement projection method and the basic LOD method are identical. The equivalence is the cleanest piece, though connecting two known methods is incremental. The survey parts are honest and anchored to standard references. No circularity. Heavy self-citation, including unpublished preprint [26] for Section 3.3, but the central claims do not depend on it. Soft spots, in proportion. Section 2.3 is the real one. Equation (2.19) fixes the kernel symbol only on the Brillouin zone; infinitely many kernels satisfy it. The claim that a smooth extension of lambda_0 'naturally indicates' compact support or fast decay is not proved, and different extensions give different kernels. So the advertised characterization of nonlocal interactions is conditional, not canonical. The paper acknowledges non-uniqueness, but the practical appeal of the surrogate in (2.20) is weaker than the text implies. This narrows the paper's strongest claim; it does not refute it, because Section 2.2 already establishes nonlocality independently. Section 3.3's reliance on an unpublished preprint for exponential decay of the coarse-grained lattice kernel is a minor credibility tax. Net: a solid, useful survey with a couple of original derivations, one overclaimed step, and no load-bearing flaw. I would bring it to reading group and cite it for the Section 2.2 example and the Schur/LOD equivalence. I would send it to peer review, expecting the referee to ask for a more careful statement in Section 2.3 about what is and is not proved.","headline":"A solid, honest survey with real but modest new examples; the generic nonlocality thesis holds, though the Section 2.3 characterization is conditional.","tokens_in":22049,"tokens_out":3754,"would_cite":true,"duration_ms":133611,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65-02","00A71","35B27","65N99","70-08","74Q99"],"pacs":[],"model":"deepseek-v4-flash","headline":"Nonlocal operators are the generic homogenization limit of local differential equations.","keywords":["homogenization","nonlocal operators","model reduction","Bloch waves","dispersion relation","memory effects","Schur complement","numerical homogenization"],"falsifier":"Take a one-dimensional periodic two-material composite, compute the first Bloch eigenvalue $\\lambda_0(k)$ on the Brillouin zone, extend it smoothly to $1$ outside, and numerically Fourier-invert $1-\\lambda_0(k)$ to obtain candidate kernels $\\gamma$. If every admissible kernel decays only algebraically, fails to be integrable, or changes sign in a way that violates the intended interaction structure, then the paper's claim that smoothness of the symbol 'naturally indicates' a compactly supported or fast-decaying kernel is refuted; equivalently, the nonlocal effective wave equation would lose its practical appeal.","tokens_in":20979,"feed_emoji":"📐","tokens_out":12662,"duration_ms":112143,"temperature":0.7,"pith_summary":"Homogenization and model reduction look like recipes for replacing fine-scale local equations by simpler local ones, but this paper argues the opposite is generic: the effective, coarse-grained laws that come out are usually nonlocal. The key diagnostic is a dispersion relation: any spatially homogeneous local operator must have a polynomial Fourier symbol, so any effective operator with a non-polynomial symbol cannot be reproduced by a differential equation. The paper demonstrates this with concrete linear examples, including a memory effect produced by oscillatory coefficients, a homogenized PDE whose symbol is non-polynomial, an epsilon-dependent nonlocal wave equation built from the first Bloch eigenvalue, Schur-complement coarse graining with dense but exponentially decaying couplings, and lattice reductions whose kernels decay faster than any polynomial. If the claim holds, homogenization becomes a constructive tool for deriving nonlocal interaction kernels, and nonlocal modeling becomes the natural language for effective equations.","feed_headline":"Homogenization generically yields nonlocal equations","feed_subtitle":"A dispersion-relation test separates rare local limits from the generic nonlocal ones.","key_machinery":"The load-bearing mechanism is the Peetre theorem combined with the Fourier symbol, or dispersion relation. Peetre's theorem says a local linear operator has finite-order differential form; for a translation-invariant operator this forces its Fourier symbol to be a polynomial in the wave number. Consequently, any effective operator whose symbol is not a polynomial, such as the first Bloch eigenvalue $\\lambda_0(k)$ of a periodic elliptic operator, is necessarily nonlocal. To turn this into a model, the paper sets up equation (2.19), $\\int \\gamma(s)(1-e^{2\\pi i k\\cdot s})\\,ds = \\lambda_0(k)$, and constructs the nonlocal interaction kernel $\\gamma$ from the dispersion relation, then rescales it as $\\gamma^\\epsilon(s)=\\epsilon^{-d-2}\\gamma(s/\\epsilon)$. In the numerical-homogenization examples, the analogous mechanism is the Schur complement: eliminating fine degrees of freedom turns a sparse local matrix into a dense, exponentially decaying coupling, i.e., a discrete nonlocal operator.","core_discovery":"On its own terms, the paper's central discovery is that nonlocality is not a pathology but the generic output of homogenization and model reduction of local equations, and that the classical local homogenized PDEs are special cases. The argument rests on a theorem of Peetre: a translation-invariant operator that is local must have a polynomial Fourier symbol, or dispersion relation; therefore any effective operator with a non-polynomial symbol is necessarily nonlocal. Working from this, the paper shows that the homogenized limit of a simple two-scale elliptic-PDE operator has a symbol that is not a polynomial unless the coefficient is essentially constant, that wave propagation in periodic media is better approximated for all times by the nonlocal operator whose symbol is the first Bloch eigenvalue than by the classical homogenized wave equation, and that projection, Schur-complement, and lattice-coarse-graining reductions produce discrete nonlocal kernels with exponential or super-algebraic decay. The same mechanism appears in time through Tartar's memory effect and the Mori-Zwanzig formalism, where reduced dynamics carry convolution memory terms that can sometimes be localized by adding auxiliary variables, but the nonlocal form is the intrinsic one.","pith_inferences":["The polynomial-symbol criterion suggests a practical diagnostic for model reduction: compute the effective dispersion relation numerically, for example from Bloch waves or transfer matrices, and check whether it is algebraic; generically it will not be, so a local surrogate should be treated as an approximation with a quantifiable bandwidth limit.","Because equation (2.19) admits multiple kernels, one could exploit the freedom to enforce desired kernel properties, such as positivity, compact support, or a chosen horizon, by adding null-space terms that vanish on the Brillouin zone; this may make dispersion-based kernel construction useful for designing peridynamic-type models from measured or computed data.","In high-contrast or strongly heterogeneous media, where exponential decay of Schur-complement kernels can fail, the decay rate of the effective kernel itself becomes an observable indicator of how nonlocal the problem really is; measuring that rate may inform whether localization is safe.","A natural testable extension is to random or quasiperiodic media: if the homogenized symbol remains non-polynomial almost surely, then the generic-nonlocality claim extends beyond the periodic examples treated here."],"forward_implications":["If the homogenized symbol is non-polynomial, matching it with a local PDE is at best a low-order approximation; long-time or high-frequency behavior requires the nonlocal surrogate.","The first Bloch eigenvalue gives a constructive, parameter-free way to build an $\\epsilon$-dependent nonlocal effective wave equation that tracks the true wave for all time with error $O(\\epsilon)$, unlike classical homogenization or truncated dispersive PDEs.","Numerical homogenization by Schur complement or corrector projection produces coarse-grid operators that are dense but have exponentially decaying kernels; localization of basis functions is justified precisely when this decay holds, and high-contrast media may break it.","Memory terms from coarse-graining dynamics are intrinsic; exponential kernels can be localized by introducing extra variables, giving a bridge from nonlocal models to extended-state local models.","The dispersion-relation test, asking whether the symbol is a polynomial, can serve as a general diagnostic for whether a proposed effective model is genuinely local or must be treated as nonlocal."],"supporting_citations":[{"why":"Peetre's theorem: a local linear operator has finite-order differential form; this yields the polynomial-symbol criterion that drives the nonlocality argument.","marker":"[61]"},{"why":"Schwartz kernel theorem: continuous linear maps have distribution kernels, and locality means the kernel is supported on the diagonal.","marker":"[42, Chapter 5]"},{"why":"Classical periodic homogenization: supplies the cell problem, the homogenized local equations (1.5) and (1.7), and the Bloch-wave expansion used throughout.","marker":"[6]"},{"why":"Tartar's oscillatory-coefficient example: weak convergence produces a nonlocal-in-time memory limit, the first clear homogenization-induced nonlocal effect.","marker":"[68]"},{"why":"Bloch-wave dispersive effective model: identifies the first-band dispersion relation as the object that determines the effective wave operator.","marker":"[62]"},{"why":"Wavelet-based numerical homogenization: shows the Schur complement of a sparse local matrix is dense with exponentially decaying coefficients.","marker":"[22]"},{"why":"Localized orthogonal decomposition: global corrector basis functions decay exponentially, linking projection homogenization to quasi-local discrete integral operators.","marker":"[56]"},{"why":"Coarse-grained lattice model: the reduced coarse-lattice interaction kernel has no compact support but decays faster than any algebraic power.","marker":"[26]"},{"why":"Mori-Zwanzig projection formalism: deriving reduced dynamics introduces convolution memory terms and the fluctuation-dissipation relation.","marker":"[58, 76]"}],"fun_headline_variants":["Homogenization: nonlocality is the rule, not the exception","Nonlocal effective equations: generic, not pathological","Homogenized models are typically nonlocal, local ones rare","Peetre's theorem: why homogenization yields nonlocal operators","The homogenized limit is almost never local"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that the first Bloch eigenvalue $\\lambda_0(k)$, after being smoothly extended to $1$ outside the Brillouin zone, Fourier-inverts to a nonlocal kernel that has compact support or fast decay; the paper states this is 'naturally indicated' but does not prove it, and the defining equation admits more than one kernel, so the practical utility of the nonlocal surrogate depends on this unproved regularity.","fun_headline_variants_meta":{"raw":{"variants":["Homogenization: nonlocality is the rule, not the exception","Nonlocal effective equations: generic, not pathological","Homogenized models are typically nonlocal, local ones rare","Peetre's theorem: why homogenization yields nonlocal operators","The homogenized limit is almost never local"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000279,"raw_usage":{"total_tokens":1602,"prompt_tokens":838,"completion_tokens":764,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":454,"completion_tokens_details":{"reasoning_tokens":682}},"tokens_in":454,"tokens_out":764,"duration_ms":6457,"temperature":1.0,"reasoning_tokens":682,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:38:50.928571+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a one-dimensional periodic two-material composite, compute the first Bloch eigenvalue $\\lambda_0(k)$ on the Brillouin zone, extend it smoothly to $1$ outside, and numerically Fourier-invert $1-\\lambda_0(k)$ to obtain candidate kernels $\\gamma$. If every admissible kernel decays only algebraically, fails to be integrable, or changes sign in a way that violates the intended interaction structure, then the paper's claim that smoothness of the symbol 'naturally indicates' a compactly supported or fast-decaying kernel is refuted; equivalently, the nonlocal effective wave equation would lose its practical appeal.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Peetre's theorem: a local linear operator has finite-order differential form; this yields the polynomial-symbol criterion that drives the nonlocality argument."},{"cited_title":"374, American Mathematical Soc., 2011","cited_arxiv_id":null,"evidence_quote":"Classical periodic homogenization: supplies the cell problem, the homogenized local equations (1.5) and (1.7), and the Bloch-wave expansion used throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Tartar's oscillatory-coefficient example: weak convergence produces a nonlocal-in-time memory limit, the first clear homogenization-induced nonlocal effect."},{"cited_title":"4, 984–1005","cited_arxiv_id":null,"evidence_quote":"Bloch-wave dispersive effective model: identifies the first-band dispersion relation as the object that determines the effective wave operator."},{"cited_title":"2, 540–559","cited_arxiv_id":null,"evidence_quote":"Wavelet-based numerical homogenization: shows the Schur complement of a sparse local matrix is dense with exponentially decaying coefficients."},{"cited_title":"290, 2583–2603","cited_arxiv_id":null,"evidence_quote":"Localized orthogonal decomposition: global corrector basis functions decay exponentially, linking projection homogenization to quasi-local discrete integral operators."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Coarse-grained lattice model: the reduced coarse-lattice interaction kernel has no compact support but decays faster than any algebraic power."}],"review_version":1}