{"id":"99f29f50-5bc6-447e-83f6-751b073ce2cf","arxiv_id":"2411.08829","paper_version":2,"verdict":"REJECT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The claimed anisotropic Hölder embedding is not established because the key lemma's box argument is wrong and variable-exponent Hölder inequalities are misapplied.","lead":"This mathematics paper defines a new space for measuring direction-dependent smoothness and claims that anisotropic Sobolev functions live in it. The central proof contains a suspected error in the key lemma and leaves out a required smoothness condition.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 2.3's final exponent estimate is invalid: with variable p_i, the step replacing s^{1/beta_i^- - sigma/p_i(x)} by s can fail (e.g. p=(3,100) vs (100,3) in N=2), so Theorem 1.1 is not proved.","rationale":"The paper introduces a plausible framework, and the idea of anisotropic variable-exponent Hölder spaces is not inherently wrong. However, the only proof of the main embedding is Lemma 2.3, and its decisive estimate is invalid under the stated variable-exponent assumptions. The failure is not merely a missing technical detail: the exponent in the final chain can be strictly below 1 even for a simple two-point configuration, so the claimed inequality |u(x)-u(y)| ≤ C s cannot be derived from the argument. The reader correctly identified serious problems, but their stated reason about box containment is mistaken; the stronger concern is the algebra after the Hölder step, which collapses precisely when exponents vary between points. Because the central theorem is unsupported by the manuscript's proof, the rejection verdict stands. A revision would need either a corrected proof that confronts the variable-exponent norm and the exponent inequality, or a reformulation of the theorem with weaker exponents that the available estimates actually deliver.","tokens_in":11394,"tokens_out":13721,"duration_ms":112796,"concrete_test":"Recompute Lemma 2.3 for N=2 with p_1(x)=3, p_2(x)=100, p_1(y)=100, p_2(y)=3, extended continuously to a rectangle, with x and y close enough that s=Σ|x_i-y_i|^{β_i(x,y)} < 1. Calculate β_i^-, σ, and e_i = 1/β_i^- - σ/p_i(x); the check should confirm e_1 ≈ 0.672 < 1, so the step s^{e_i} ≤ s fails. Then attempt to repair the proof by using pair-dependent β_i(x,y)=min(β_i(x),β_i(y)) in place of global minima; recompute the algebra and verify that the same numerical pair still gives e_1 < 1 for the choice β_i(x,y)=min(β_i(x),β_i(y)). If no choice of β_i^- makes e_i ≥ 1 for all pairs, Lemma 2.3 as stated is false.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The main theorem rests entirely on Lemma 2.3. In its Case 2 (0<s<1), the proof builds a box of side s^{1/beta_i^-}, where beta_i^- is the global minimum of beta_i, and arrives at |u(x)-u(y)| <= C ||u|| Σ_i s^{1/beta_i^- - sigma/p_i(x)}, with sigma = Σ_j 1/beta_j^-. It then replaces s^{1/beta_i^- - sigma/p_i(x)} by s. This requires the exponent to be at least 1. Under the intended formula beta_i(x) = (1-Σ_j 1/p_j(x)) / (1-Σ_j 1/p_j(x) + N/p_i(x)), set T_i(x) = p_i(x)(1-Σ_j 1/p_j(x)); then 1/beta_i(x) = 1 + N/T_i(x). With beta_i^- = inf beta_i, the exponent equals 1 + N/(inf T_i) - [N + N Σ_j 1/(inf T_j)]/p_i(x), which is not generally >= 1. Example in R^2: take p(x) = (3,100) and p(y) = (100,3). Then beta_1(x) = beta_2(y) ≈ 0.496 and beta_1(y) = beta_2(x) ≈ 0.970, so beta_1^- = beta_2^- ≈ 0.496, sigma ≈ 4.03, and for i=1 at x the exponent is 2.015 - 4.03/3 ≈ 0.672 < 1. The inferred factor s is therefore not obtained; the proof yields only s^{0.672}, which is not dominated by s as s -> 0. Since Lemma 2.3 is the only mechanism producing the Hölder estimate, Theorem 1.1 is not established. The reader's box-containment objection is not the right flaw: beta_i^- ≤ beta_i(x,y) implies s^{1/beta_i^-} ≥ |x_i-y_i|, so the translated box can contain both points; the exponent collapse above is decisive. Independently, the use of ||∂_i u||_{L^{p_i(x)}(Q)} after a constant-exponent Hölder inequality requires justification and is not generally valid for variable-exponent norms.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper defines an anisotropic variable-exponent H\\\"older space C^{0,\\vec{\\beta}(x)} on rectangular domains and claims a continuous embedding W^{1,\\vec{p}(x)} \\hookrightarrow C^{0,\\vec{\\beta}(x)} with explicit exponents \\beta_i given by (2.2), under the assumption that each p_i is continuous, bounded away from 1 and infinity, and satisfies p_i(x) > N. The proof combines Fan's embedding theorem with a Morrey-type estimate (Lemma 2.3) intended to control |u(x)-u(y)| by \\sum_i |x_i-y_i|^{\\beta_i(x,y)}. Corollaries address the isotropic reduction and monotonicity of the exponents; Section 4 gives informal examples on cusped domains, and Section 5 presents qualitative applications to heat conduction and porous-media flow.","tokens_in":11942,"tokens_out":11756,"duration_ms":106122,"significance":"If valid, the main theorem would be a useful generalization of Fan's isotropic variable-exponent embedding and of R\\'akosn\\'ik's constant-exponent anisotropic results, providing explicit, direction-dependent H\\\"older exponents. The definition of the anisotropic H\\\"older space through the pointwise minimum \\beta_i(x,y)=\\min\\{\\beta_i(x),\\beta_i(y)\\} is natural, and the paper contains no fitted parameters or circular normalizations; the claimed embedding is a genuine, falsifiable statement. However, the central estimate in Lemma 2.3 is false as written: the decisive exponent comparison fails on a concrete two-point example, and the proof also applies H\\\"older's inequality as if the exponent were constant. Because Lemma 2.3 is the only mechanism producing the pointwise H\\\"older bound, the main theorem is not established by the arguments given.","major_comments":[{"comment":"The step replacing s^{1/\\beta_i^- - \\sigma/p_i(x)} by s, i.e. asserting 1/\\beta_i^- - \\sigma/p_i(x) \\ge 1, is false in general. Take N=2, p_1(x)=3, p_2(x)=100, p_1(y)=100, p_2(y)=3. Using (2.2) with the denominator interpreted as N/p_i(x) (as Theorem 1.1 requires), one gets \\beta_1(x)=\\beta_2(y)\\approx 0.496 and \\beta_1(y)=\\beta_2(x)\\approx 0.970, so \\beta_1^-=\\beta_2^-\\approx 0.496 and \\sigma=\\sum_i 1/\\beta_i^-\\approx 4.03. For i=1 at x, the exponent is 1/0.496 - 4.03/3 \\approx 0.67 < 1, so the estimated factor is s^{0.67}, which is not dominated by s as s\\to 0. Since Lemma 2.3 is the only mechanism giving the pointwise H\\\"older estimate, Theorem 1.1 is not proved. The translated-box containment is not the difficulty: \\beta_i^-\\le \\beta_i(x,y) gives s^{1/\\beta_i^-}\\ge |x_i-y_i|, so a translated box of side s^{1/\\beta_i^-} can indeed contain both points.","section":"Lemma 2.3, Eq. (2.3) and following inequalities"},{"comment":"H\\\"older's inequality is applied as \\int_{\\Omega(s,t)} |\\partial_i u|\\,dz \\le \\|\\partial_i u\\|_{L^{p_i(x)}(Q)}\\,(\\operatorname{meas}\\Omega(s,t))^{1-1/p_i(x)}, which treats p_i(x) as a constant exponent. In variable-exponent Lebesgue spaces the dual norm of the characteristic function is not generally (meas)^{1-1/p_i(x)}; the correct estimate involves the norm in L^{p_i'(\\cdot)} and depends on the log-H\\\"older constant or on global bounds in a more subtle way. This step is therefore unjustified even if the exponent comparison in the previous comment were repaired.","section":"Lemma 2.3, after Eq. (2.3)"},{"comment":"Lemma 2.1 is stated for \\vec p\\in(C_+(\\Omega))^N only, but the introduction describes Fan's Theorem 2.5 as requiring the log-H\\\"older condition \\vec p\\in(C_{\\log}^+(\\Omega))^N. The proof of Theorem 1.1 invokes Lemma 2.1 at (3.1) and again in Case 1 of Lemma 2.3 to obtain continuity and the s\\ge 1 estimate. Under the stated hypotheses, that embedding into C(\\Omega) is not justified by the cited theorem. The authors should either add the log-H\\\"older condition to the hypotheses or provide a proof of Lemma 2.1 as stated.","section":"Lemma 2.1 and Theorem 1.1"}],"minor_comments":[{"comment":"The denominator in (2.2) reads 1-\\sum_j 1/p_j(x)+N/p_1(x) for every i, which would make all \\beta_i identical and contradict the anisotropic nature of the claimed embedding; it should presumably be N/p_i(x).","section":"Equation (2.2)"},{"comment":"The examples in Section 4 are informal and do not constitute rigorous counterexamples to any stated theorem: for instance, the function u(x,y)=\\sqrt{x} on the cusp domain is not shown to belong to W^{1,4}(\\Omega), nor is it shown to fail every H\\\"older condition.","section":"Section 4"},{"comment":"Corollary 3.2 is not a precise mathematical statement: saying that \\beta_i(x) approaches 1 as p_i(x)\\to\\infty does not formulate a comparative embedding statement, and the proof does not establish monotonicity as a theorem about the embedding.","section":"Corollary 3.2"},{"comment":"There are numerous typographical and notational inconsistencies (e.g., C^+ vs. C_{\\log}^+, L^{p_i(x)}(Q) vs. L^{p_i(\\cdot)}(Q), and missing spaces in displayed formulas); a careful revision is needed if the paper is resubmitted.","section":"References and text"}],"recommendation":"reject","confidential_remarks":"The paper appears to be an extended version with additional application and counterexample sections, but these do not compensate for the central gap. The failure of Lemma 2.3 is demonstrated by an explicit example, and the variable-exponent H\\\"older step is also invalid; these are load-bearing errors in the proof of the main theorem. The result might be salvageable with a substantially different proof and additional hypotheses (likely log-H\\\"older continuity), but the current manuscript does not support its central claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You can skip this one: the main theorem is unproved, and the proof fails at the central lemma. The idea—defining a Hölder space with a direction-dependent variable exponent and embedding anisotropic Sobolev spaces into it—is reasonable, and I don't see that exact definition or the explicit β_i formula in the cited literature. But Lemma 2.3, on which Theorem 1.1 rests entirely, has at least three independent breakdowns.\n\nFirst, the box used in Case 2 has side length s^{1/β_i^-}, where β_i^- is the global infimum of β_i. For 0<s<1, a smaller exponent gives a smaller side, so s^{1/β_i^-} ≤ s^{1/β_i(x,y)}. A component distance |x_i-y_i| can be as large as s^{1/β_i(x,y)}, so the box need not contain both x and y. The containment claim is false.\n\nSecond, even granting containment, the exponent collapse is invalid. The proof replaces s^{1/β_i^- - σ/p_i(x)} by s, which requires the exponent to be at least 1. In R^2, take p(x)=(3,100) and p(y)=(100,3). Then β_1(x)=β_2(y)≈0.496 and β_1(y)=β_2(x)≈0.970, so β_i^-≈0.496 and σ≈4.03; at x with i=1 the exponent is about 0.67, not ≥1. The factor s is not obtained.\n\nThird, the use of Hölder's inequality with a fixed exponent p_i(x) inside a variable-exponent norm is not justified; the characteristic function of Ω(s,t) has a variable-exponent norm that does not reduce to a simple power of the measure.\n\nThere are smaller issues: Fan's Lemma 2.1 is quoted under less regularity than Fan actually requires (the log-Hölder condition is stated in the introduction but absent from the lemma), and the applications in Section 5 are decorative—they evaluate Lipschitz functions of sine and exp, which says nothing about the embedding.\n\nWhat is good: the anisotropic variable-exponent Hölder space is a natural object, and the explicit formula for β_i is a plausible ansatz for the correct exponent. If the proof can be repaired, the result would be a modest but useful contribution to the variable-exponent community. As it stands, the central claim is unsupported.\n\nI would not send this to peer review in its current form. The flaws are too central and the surrounding text too padded. If the authors fix Lemma 2.3 and the assumptions, it might deserve a referee's time; for now, desk reject.","headline":"Lemma 2.3 is invalid in three independent places, so the main embedding theorem is unproved; the anisotropic variable-exponent Hölder space idea is worth keeping, but this paper needs a full rewrite before it merits refereeing.","tokens_in":12404,"tokens_out":6446,"would_cite":false,"duration_ms":96782,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["46E35","46E15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that on rectangular domains, the anisotropic variable-exponent Sobolev space $W^{1,\\vec p(x)}$ embeds continuously into an anisotropic Hölder space $C^{0,\\vec\\beta(x)}$, with each directional exponent $\\beta_i$ given by…","keywords":["anisotropic variable exponent Sobolev spaces","Sobolev embeddings","continuous embedding","anisotropic Hölder-continuous functions","critical Sobolev exponent","Morrey-type estimates","variable exponent"],"falsifier":"Take $N=2$, $Q=(0,1)^2$, exponents $p_1(x)=4+0.5\\sin(2\\pi x_1)$ and $p_2(x)=4+0.5\\cos(2\\pi x_2)$, so $p_m>N$ and $\\beta_1$ varies with $x$. Choose $x=(0.25,0.5)$ and $y=(0.35,0.5)$, so only the first coordinate differs. Compute $s=|x_1-y_1|^{\\beta_1(x,y)}$ and compare $s^{1/\\beta_1^-}$ with $|x_1-y_1|$: since $\\beta_1(x,y)>\\beta_1^-$, the box side is smaller than the coordinate gap, so $y$ lies outside the asserted box $\\Omega(s)$. Directly evaluating the Lemma 2.3 inequality for a smooth function along this pair determines whether the claimed constant $c$ can exist and therefore whether the proof's main estimate holds.","tokens_in":11244,"feed_emoji":"📐","tokens_out":8969,"duration_ms":69329,"temperature":0.7,"pith_summary":"The paper aims to show that functions with one distributional derivative in each direction, measured in different variable Lebesgue spaces, possess direction-dependent fractional regularity on rectangular domains. Concretely, it proves a continuous embedding from the anisotropic Sobolev space $W^{1,\\vec p(x)}$ into the anisotropic Hölder space $C^{0,\\vec\\beta(x)}$, where the Hölder exponent in direction $i$ is computed pointwise from the Sobolev exponents by $\\beta_i(x) = (1 - \\sum_j 1/p_j(x)) / (1 - \\sum_j 1/p_j(x) + N/p_i(x))$. This would extend the classical isotropic Morrey embedding and Fan's variable-exponent results to a setting where smoothness can differ along different axes, which matters for models of composite materials, turbulent flows, and image restoration. The proof works through a local oscillation estimate on cubes and a covering argument.","feed_headline":"Anisotropic Sobolev spaces embed in directional Hölder spaces","feed_subtitle":"On rectangular domains, each axis gets its own Hölder exponent fixed by the Sobolev exponents.","key_machinery":"The load-bearing object is the anisotropic variable-exponent Hölder space $C^{0,\\vec\\beta(x)}$ with norm (2.1): a function is measured by its sup norm plus the supremum over pairs $x,y$ of $|u(x)-u(y)| / \\sum_i |x_i-y_i|^{\\beta_i(x,y)}$. The proof of the embedding rests on Lemma 2.3, a local Morrey-type estimate on a unit cube: $|u(x)-u(y)| \\le c\\|u\\| \\sum_i |x_i-y_i|^{\\beta_i(x,y)}$. To prove it, the authors introduce a small translated box $\\Omega(s)$ with side lengths $s^{1/\\beta_i^-}$, where $s = \\sum |x_i-y_i|^{\\beta_i(x,y)}$, use the fundamental theorem of calculus along segments, Fubini, and Hölder's inequality to control the average oscillation, and then patch cubes across the rectangular domain.","core_discovery":"The central claim is Theorem 1.1: for a rectangular domain $\\Omega \\subset \\mathbb{R}^N$ and continuous exponents $p_i$ with $N < p_m(x) \\le p_M(x) < \\infty$, the space $W^{1,\\vec p(x)}(\\Omega)$ is continuously embedded in $C^{0,\\vec\\beta(x)}(\\Omega)$, where $\\beta_i(x)$ is given by the formula above. The embedding is witnessed by the pointwise bound $|u(x)-u(y)| \\le C\\|u\\|_{W^{1,\\vec p(x)}} \\sum_i |x_i-y_i|^{\\beta_i(x,y)}$ with $\\beta_i(x,y)=\\min(\\beta_i(x),\\beta_i(y))$. Two corollaries are drawn: the isotropic case $p_i=p$ reduces $\\beta$ to $1-N/p(x)$, and larger exponents $p_i$ yield larger Hölder exponents, approaching Lipschitz regularity in the limit.","pith_inferences":["The pointwise estimate implies a scale-invariant bound: after rescaling a rectangular domain, the embedding constant depends only on the anisotropy of the exponents and not on the domain diameter, so the same argument applies to cubes of arbitrary side length with adjusted constants.","One could test the sharpness of the exponent formula numerically: approximate functions such as $|x_i|^{\\alpha}$ on a rectangle and compare the largest achievable directional Hölder exponent with the predicted $\\beta_i$; agreement would support the formula, while disagreement would suggest a different critical exponent.","The counterexamples in Section 4 suggest that rectangularity is not merely a technical convenience: if a similar embedding holds on cuspidal domains, it would need a modified norm or boundary-dependent exponents, and characterizing that is a natural next step.","The anisotropic Hölder norm could serve as a regularity diagnostic in applied fields: the ratio $\\beta_i/\\beta_j$ measures how much smoother a signal or velocity field is along one axis than another, which is exactly the directional information image-restoration and turbulence models need."],"forward_implications":["Every element of $W^{1,\\vec p(x)}(\\Omega)$ on a rectangle has a continuous representative whose oscillation along coordinate direction $i$ is controlled by the local exponent $\\beta_i(x,y)$, giving quantitative anisotropic regularity rather than a single global modulus.","When the Sobolev exponents are all equal to $p(x)$, the embedding reduces to the known isotropic one with Hölder exponent $1 - N/p(x)$, so the theorem contains the classical scalar case as a special case.","Raising the Sobolev exponent in one direction raises the Hölder exponent in that direction, so the result interpolates between mere continuity and Lipschitz continuity as $p_i$ increases.","The embedding supplies a Banach-space framework for studying anisotropic variable-exponent PDEs on rectangular domains: solutions lying in $W^{1,\\vec p(x)}$ automatically have the directional Hölder regularity needed in compactness arguments."],"supporting_citations":[{"why":"Supplies the continuous embedding $W^{1,\\vec p(x)} \\hookrightarrow C(\\Omega)$ used as Lemma 2.1 and as the base case $s \\ge 1$.","marker":"[9]"},{"why":"Source of the Morrey-type oscillation estimate that the local box argument adapts.","marker":"[13]"},{"why":"Continues Morrey's method, cited as inspiration for the cube covering proof.","marker":"[14]"},{"why":"Provides earlier anisotropic Sobolev embedding results and techniques that the proof extends.","marker":"[17]"},{"why":"Companion of [17], also cited for the anisotropic embedding methods.","marker":"[18]"},{"why":"Documents the failure of $W^{1,N}$ to embed into $L^\\infty$, the boundary case that motivates the assumption $N < p_m(x)$.","marker":"[12]"}],"fun_headline_variants":["Anisotropic Sobolev spaces embed with per-axis Hölder exponents","Variable-exponent Sobolev spaces gain Hölder continuity per direction","New embedding links anisotropic Sobolev to Hölder spaces on rectangles","Rectangular domains: Sobolev embedding yields directional Hölder regularity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof's key local estimate rests on the assumption that a small anisotropic distance forces the two points into a box whose side lengths are set by the smallest Hölder exponent, and that Hölder's inequality can be applied with the exponent frozen at one point; if either premise fails for genuinely variable exponents, the estimate and hence the embedding proof collapse.","fun_headline_variants_meta":{"raw":{"variants":["Anisotropic Sobolev spaces embed with per-axis Hölder exponents","Variable-exponent Sobolev spaces gain Hölder continuity per direction","New embedding links anisotropic Sobolev to Hölder spaces on rectangles","Rectangular domains: Sobolev embedding yields directional Hölder regularity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000523,"raw_usage":{"total_tokens":2466,"prompt_tokens":818,"completion_tokens":1648,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":434,"completion_tokens_details":{"reasoning_tokens":1572}},"tokens_in":434,"tokens_out":1648,"duration_ms":13556,"temperature":1.0,"reasoning_tokens":1572,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T21:18:50.711576+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $N=2$, $Q=(0,1)^2$, exponents $p_1(x)=4+0.5\\sin(2\\pi x_1)$ and $p_2(x)=4+0.5\\cos(2\\pi x_2)$, so $p_m>N$ and $\\beta_1$ varies with $x$. Choose $x=(0.25,0.5)$ and $y=(0.35,0.5)$, so only the first coordinate differs. Compute $s=|x_1-y_1|^{\\beta_1(x,y)}$ and compare $s^{1/\\beta_1^-}$ with $|x_1-y_1|$: since $\\beta_1(x,y)>\\beta_1^-$, the box side is smaller than the coordinate gap, so $y$ lies outside the asserted box $\\Omega(s)$. Directly evaluating the Lemma 2.3 inequality for a smooth function along this pair determines whether the claimed constant $c$ can exist and therefore whether the proof's main estimate holds.","supporting_citations":[{"cited_title":"Fan; Anisotropic variable exponent Sobolev spaces and ⃗p(·)-Laplacian equations , Complex V ar","cited_arxiv_id":null,"evidence_quote":"Supplies the continuous embedding $W^{1,\\vec p(x)} \\hookrightarrow C(\\Omega)$ used as Lemma 2.1 and as the base case $s \\ge 1$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the Morrey-type oscillation estimate that the local box argument adapts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Continues Morrey's method, cited as inspiration for the cube covering proof."},{"cited_title":"R´ akosn´ ık;Some remarks to anisotropic Sobolev Spaces I , Beitr¨ age Anal","cited_arxiv_id":null,"evidence_quote":"Provides earlier anisotropic Sobolev embedding results and techniques that the proof extends."},{"cited_title":"R´ akosn´ ık;Some remarks to anisotropic Sobolev spaces II , Beitr¨ age Anal","cited_arxiv_id":null,"evidence_quote":"Companion of [17], also cited for the anisotropic embedding methods."},{"cited_title":"Haˇ skovec, C","cited_arxiv_id":null,"evidence_quote":"Documents the failure of $W^{1,N}$ to embed into $L^\\infty$, the boundary case that motivates the assumption $N < p_m(x)$."}],"review_version":1}