{"id":"3f395803-da54-4f1e-a119-3621d6a86be4","arxiv_id":"2508.05376","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Establishes Bernstein and Nikolskii inverse inequalities for kernel approximation spaces on bounded Lipschitz domains and compact Riemannian manifolds, with a mild extra kernel smoothness requirement.","lead":"This mathematics paper proves inverse inequalities for kernel-based approximation spaces on bounded domains and on curved manifolds. These are tools for controlling high-order or maximum norms by lower-order or average norms, which matter for error analysis of numerical methods.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'slightly more smoothness' condition is unquantified; if it requires k>d/2+1, the Nikolskii inequality may be a Sobolev-embedding consequence rather than a new inverse-inequality result.","rationale":"The reader's verdict is UNVERDICTED due to lack of the full text, and I cannot identify an internal error at the abstract level. I partially agree with the reader's weakest assumption about the smoothness condition, but my concern is more specific: the abstract leaves the 'slightly more smoothness' threshold unspecified, and the threshold determines whether the claimed Nikolskii inequality is a substantive new theorem or a near-immediate corollary of the Sobolev embedding. This does not change the verdict—without the full text the paper remains unverified—but it sharpens the question that the full text must answer. I therefore keep the verdict UNCHANGED and recommend a targeted check of the smoothness assumption.","tokens_in":639,"tokens_out":4415,"duration_ms":46144,"concrete_test":"Retrieve the full text from arXiv and inspect the main theorem on bounded Lipschitz domains (likely Theorem 3.1 or 4.1). Record the exact smoothness assumption on the kernel—specifically the Sobolev index k such that the native space is H^k(Ω). Then determine whether k > d/2 + 1 is required or whether k > d/2 + ε for arbitrarily small positive ε suffices. As a check, re-derive the Nikolskii inequality from the proof; if the proof invokes the standard embedding H^k(Ω) ↪ L∞(Ω) at threshold k > d/2, then the extra smoothness is not doing the same work.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is unverifiable from the abstract alone, but the one visible soft spot is the unquantified condition 'slightly more smoothness' on the kernel. If the proof requires the kernel's native space to be H^k with k > d/2 + 1, then the Nikolskii inequality (L∞ ≤ C L2) follows almost directly from the Sobolev embedding H^k ↪ L∞, and the 'all Sobolev orders on the lower bound' part may be an interpolation consequence rather than a new inverse-inequality mechanism. The paper would still be correct, but the advertised 'desired form' would be weaker than it appears. Alternatively, if k > d/2 + ε for arbitrarily small ε is sufficient, the proof must be using a sharp nonstandard argument. The abstract does not disclose the threshold, so the contribution's significance is genuinely uncertain. This is a scope/novelty concern, not an internal-error allegation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper announces inverse inequalities for kernel-based approximation spaces on bounded Lipschitz domains in R^d and on compact Riemannian manifolds. For domains, it claims to extend prior Bernstein inequalities to all Sobolev orders on the lower bound and to L_2 on the upper, and to derive Nikolskii inequalities bounding L_∞ by L_2. The authors explicitly state that these results may require slightly more kernel smoothness than the standard >d/2 assumption. For manifolds, the theory is developed for restricted kernels, i.e., positive definite kernels on the ambient Euclidean space restricted to the manifold, with corresponding counterpart inequalities. The manuscript provided for review consists solely of the abstract; no proofs or technical statements are available.","tokens_in":890,"tokens_out":2221,"duration_ms":23233,"significance":"If the claimed results are correct, they would fill a notable gap in kernel-based approximation theory: inverse inequalities are classical for polynomial spaces but have had limited extension to kernel trial spaces. The paper's explicit caveat about the required kernel smoothness is a sign of care and makes the claim falsifiable. The extension to Riemannian manifolds via restricted kernels is a plausible and potentially useful step, though its scope is narrower than the general kernel setting. However, because the full text is not available, the technical novelty and correctness cannot be assessed, and the advertised 'desired form' is clouded by the unquantified smoothness condition.","major_comments":[{"comment":"The condition 'slightly more smoothness' is never quantified. This is load-bearing: if the requirement is k > d/2 + 1 for the native space H^k, then the Nikolskii inequality L_∞ ≤ C L_2 is an immediate consequence of Sobolev embedding, and the Bernstein extension may be an interpolation artifact rather than a new inverse-inequality mechanism. The authors should state the precise smoothness threshold and explain how their proof differs from a direct Sobolev-embedding argument. As written, the principal novelty claim is unverifiable and potentially weaker than advertised.","section":"Abstract"},{"comment":"The claim of extending Bernstein inequalities 'to all orders on the lower bound and L_2 on the upper' is not supported by any technical statement in the provided text. Since the manuscript for review contains only the abstract, the central derivation cannot be checked. In a revised submission, the authors should state the precise theorem (the range of Sobolev orders, the exact norm on the upper bound, and the dependence of the constant on the kernel and domain) so that the scope of the extension is unambiguous.","section":"Abstract"},{"comment":"For compact Riemannian manifolds, the restriction to kernels induced by ambient Euclidean positive definite kernels is a significant limitation. The abstract does not indicate whether the results apply to intrinsically defined kernels (e.g., heat kernels on the manifold) or only to embeddings into Euclidean space. The authors should clarify whether this restriction is essential to the proof or merely a convenience, and discuss how limiting it is for applications. This is a scope concern, not an internal-error allegation.","section":"Abstract"}],"minor_comments":[{"comment":"The notation 'L_2' and 'L_∞' should be consistently typeset as L^2 and L^∞, or at least defined, to avoid confusion with sequence spaces.","section":"Abstract"},{"comment":"The phrase 'the regular >d/2 assumption' is informal; the standard assumption for kernel-based approximation is typically stated as the kernel being in H^s with s > d/2 (or equivalently s > d/2 + ℓ for derivatives). The authors should use the standard terminology to make the comparison exact.","section":"Abstract"}],"recommendation":"uncertain","confidential_remarks":"The manuscript provided is abstract-only, so a definitive judgment is impossible. The stress-test concern about the unquantified smoothness threshold is legitimate and should be raised with the authors. If the full paper is available, I would need to verify that the proof of the Nikolskii inequality does not reduce to Sobolev embedding and that the 'all orders' Bernstein claim is not a tautology. The scope restriction to restricted kernels on manifolds also deserves scrutiny. Given the lack of technical content, 'uncertain' is the only defensible recommendation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper looks like a real contribution to kernel approximation theory, but the abstract alone doesn't let me judge whether the advertised 'desired form' is a deep new mechanism or a repackaged Sobolev embedding. I'd want the full proof before citing.\n\nWhat's new: prior Bernstein inequalities for kernel spaces only covered a limited range of Sobolev orders; the authors claim to cover all orders on the lower bound and L2 on the upper, and add Nikolskii inequalities and a manifold version. That is a concrete, useful extension for people doing error analysis of meshless methods. The abstract is honest about the cost: they need a bit more kernel smoothness than the usual d/2, and on manifolds they restrict to kernels induced from ambient Euclidean space. Those are clear, stated limitations, and stating them up front is good practice.\n\nSoft spots: the phrase 'slightly more smoothness' is doing a lot of work. If the proof needs the native space to be H^k with k > d/2+1, then the Nikolskii inequality is basically the classical Sobolev embedding and the Bernstein extension may just be interpolation between known cases — correct, but not as big a deal as 'all orders' sounds. If instead they can get k > d/2 + epsilon for arbitrary epsilon, that would be genuinely new. The abstract does not say which. That is not a flaw in the result — it may actually be the strong version — it just means the significance is unverifiable from this piece. Same for the manifold part: restricting to restricted kernels is a real limitation, but it's explicit.\n\nThe stress-test worry about interpolation is plausible and worth checking, but it is not an internal-error allegation. I can't confirm or refute it from the abstract. The citation pattern can't be checked either.\n\nBottom line: this deserves a proper referee. If the proofs deliver what the abstract promises, it's a solid paper for a specialized journal. I would not cite it yet, but I'd read the full text if you have it.\n\nRecommendation: send to peer review, not desk reject.","headline":"A plausible, clearly described extension of inverse inequalities to all Sobolev orders and manifolds, but the abstract leaves the key smoothness threshold unquantified and I cannot weigh the novelty without seeing the proofs.","tokens_in":1293,"tokens_out":1671,"would_cite":false,"duration_ms":15817,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["41A17","41A25","65D15"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves Bernstein and Nikolskii inverse inequalities for kernel-based approximation spaces on bounded Lipschitz domains in $\\mathbb{R}^d$ and on compact Riemannian manifolds, extending prior results to all Sobolev orders and to $L","keywords":["inverse inequalities","Bernstein inequality","Nikolskii inequality","kernel-based approximation","reproducing kernel Hilbert space","Sobolev spaces","Lipschitz domains","Riemannian manifolds"],"falsifier":"Take a bounded Lipschitz domain and a positive definite kernel with smoothness exactly just above $d/2$, for example a Matérn kernel whose smoothness parameter is only slightly larger than $d/2$. For a fixed fill distance $h$, compute the quantity $\\sup_{v \\in \\mathcal{V}_h} \\|v\\|_{H^k}/\\|v\\|_{L_2}$ over the kernel approximation space $\\mathcal{V}_h$. If for some integer $k$ this quantity grows faster than the $h^{-k}$ rate predicted by the Bernstein inequality, the claimed extension to all Sobolev orders fails.","tokens_in":576,"feed_emoji":"📐","tokens_out":2610,"duration_ms":27676,"temperature":0.7,"pith_summary":"The paper establishes inverse inequalities for spaces spanned by translates of a positive definite kernel, in two settings: bounded Lipschitz domains in $\\mathbb{R}^d$ and compact Riemannian manifolds. In the domain setting, it extends earlier Bernstein inequalities, which held only for a limited range of Sobolev orders, to all orders on the lower bound and to $L_2$ on the upper bound. It also derives Nikolskii inequalities, which bound $L_\\infty$ norms by $L_2$ norms. The price is that the kernel must be slightly smoother than the usual $>d/2$ assumption. For manifolds, the results are proven for restricted kernels, meaning kernels obtained by restricting positive definite kernels from the ambient Euclidean space to the manifold.","feed_headline":"Kernel spaces gain full inverse inequalities","feed_subtitle":"Bernstein and Nikolskii bounds now hold on Lipschitz domains and compact manifolds, at the cost of extra kernel smoothness.","key_machinery":"The key machinery is the reproducing kernel Hilbert space structure of the approximation space, combined with norm equivalences across the Sobolev scale. The inverse inequalities relate derivatives of a kernel-based approximant to its $L_2$ norm through powers of the fill distance $h$, with exponents determined by the Sobolev orders involved. The proof requires the kernel to be slightly smoother than the minimal $>d/2$ smoothness usually required, so that the relevant interpolation and projection operators are stable.","core_discovery":"The central claim is that kernel-based trial spaces admit inverse inequalities of the same shape as polynomial spaces, provided the kernel has one degree of extra smoothness. On a bounded Lipschitz domain, the Bernstein inequality holds for all Sobolev orders on the lower side with an $L_2$ upper bound, and the Nikolskii inequality bounds the $L_\\infty$ norm by a constant times a power of the $L_2$ norm depending on the fill distance $h$. On compact Riemannian manifolds the same inequalities are proven for restricted kernels, those inherited from ambient Euclidean positive definite kernels. The paper presents these as extensions of prior Bernstein-type results to a much wider range of Sobole","pith_inferences":["The gap between the standard $>d/2$ smoothness assumption and the slightly stronger condition used here suggests that the minimal-regularity case may still admit inverse inequalities, though by different techniques; this is left open by the paper.","The restricted-kernel condition on manifolds may not be essential; testing intrinsic positive definite kernels defined directly on the manifold would clarify whether the ambient Euclidean restriction is a genuine limitation or an artifact of the proof.","The inequalities could be used to derive convergence rates for kernel interpolation in negative Sobolev norms, a direction the authors do not explicitly pursue.","Because the extension covers all Sobolev orders on Lipschitz domains, it may feed directly into error analyses for kernel-based solvers of higher-order PDEs, where inverse estimates with high-order norms are currently missing."],"forward_implications":["Kernel-based numerical methods gain inverse estimates usable in a priori error analysis, adaptive refinement, and multiscale approaches.","The Nikolskii inequality enables conversion between $L_2$ and $L_\\infty$ error measures in kernel approximation, supporting pointwise error bounds.","The manifold extension transfers these inverse estimates to kernel collocation and approximation problems posed on compact Riemannian manifolds.","The extra smoothness requirement makes explicit the trade-off between kernel regularity and the validity of a full Bernstein–Nikolskii theory."],"supporting_citations":[],"fun_headline_variants":["Kernel spaces match polynomial inverse inequalities","All Sobolev orders now covered by kernel inverse inequalities","Inverse inequalities for kernels: full orders, slight smoothness cost","Kernel approximation gains Bernstein and Nikolskii bounds"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The argument assumes the kernel has slightly more smoothness than the usual $>d/2$ threshold, and in the manifold case that the kernel is the restriction of a positive definite kernel on the ambient Euclidean space; if the kernel is only minimally smooth, or the manifold kernel is not of that restricted type, the stated inequalities may fail.","fun_headline_variants_meta":{"raw":{"variants":["Kernel spaces match polynomial inverse inequalities","All Sobolev orders now covered by kernel inverse inequalities","Inverse inequalities for kernels: full orders, slight smoothness cost","Kernel approximation gains Bernstein and Nikolskii bounds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000432,"raw_usage":{"total_tokens":2012,"prompt_tokens":685,"completion_tokens":1327,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":429,"completion_tokens_details":{"reasoning_tokens":1263}},"tokens_in":429,"tokens_out":1327,"duration_ms":13307,"temperature":1.0,"reasoning_tokens":1263,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T23:21:53.513768+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a bounded Lipschitz domain and a positive definite kernel with smoothness exactly just above $d/2$, for example a Matérn kernel whose smoothness parameter is only slightly larger than $d/2$. For a fixed fill distance $h$, compute the quantity $\\sup_{v \\in \\mathcal{V}_h} \\|v\\|_{H^k}/\\|v\\|_{L_2}$ over the kernel approximation space $\\mathcal{V}_h$. If for some integer $k$ this quantity grows faster than the $h^{-k}$ rate predicted by the Bernstein inequality, the claimed extension to all Sobolev orders fails.","supporting_citations":[],"review_version":1}