{"id":"47c98ccf-76df-4816-a080-a8a1ee36e7ff","arxiv_id":"1908.03384","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For m-th order gradients, Sobolev embeddings between rearrangement-invariant spaces are characterized by a one-dimensional Hardy inequality, with optimal target and domain spaces identified, including in Orlicz spaces.","lead":"This paper characterizes exactly when an m-th order gradient norm of a function on R^n controls its norm in a rearrangement-invariant target space, reducing the question to a one-dimensional Hardy inequality. It also identifies the smallest and largest spaces for which such inequalities hold, including for Orlicz spaces.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central reduction and the Orlicz-space optimality results rest on normability and optimality assertions imported from [15], [20], [29], and [30] rather than proved here; a gap in any of those would invalidate the main characterization.","rationale":"The paper presents a clean reduction of the Sobolev embedding inequality to a one-dimensional Hardy-type inequality, and the internal computations I traced (e.g., Proposition 4.1, the radial construction in Theorem 3.1, the use of interpolation for the operators in (4.9)) are consistent. However, the central claims are not fully proved in the text: the normability of sigma_m and tau_m, the equivalence in Theorem 3.6, and the optimal Orlicz-space results in Section 6 are all delegated to external sources, including a PhD thesis. The reader correctly identified this as the weakest assumption. Since I found no actual mathematical error, only a verification gap, the appropriate verdict remains CONDITIONAL rather than ACCEPT or REJECT. The proposed check targets the most load-bearing delegated step: without tau_m being a genuine rearrangement-invariant norm, the optimal-domain half of the characterization does not hold together.","tokens_in":27187,"tokens_out":14193,"duration_ms":143278,"concrete_test":"Independently reproduce the proof of normability of tau_m from [20, Theorem 4.1] in the model case Y = L^p(R^n) with p in (n/(n-m), inf), verifying the triangle inequality for the functional in (3.8) using only (3.7). If the triangle inequality cannot be shown without an extra condition, or if the proof requires an assumption not present in (3.7), then Theorem 3.5 and the reduction principle are unsupported. Conversely, a complete verification for this leading case would confirm that external reliance is not hiding a structural failure.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's completeness claim is conditioned on several external assertions that are not proved in the text. Theorem 3.1 defines sigma_m by (3.2) and asserts it is a rearrangement-invariant norm exactly under (3.1), citing [15, Theorem 5.4] and [20, Theorem 4.4]; the proof only notes that triangle inequality follows from (2.1). Theorem 3.5 asserts that tau_m defined by (3.8) is a rearrangement-invariant norm, calling the matter 'rather deep' and citing [20, Theorem 4.1]. Theorem 3.6 further relies on [20, Theorems 4.2 and 4.7], and Proposition 4.3's first inequality is asserted with details omitted. Section 6 is even more explicit: 'We omit proofs in this section' and the entire optimal Orlicz-space theory (Theorems 6.1, 6.4, 6.8) is imported from [29, Chapter 3] (a PhD thesis) and [30]. If, for example, tau_m failed the triangle inequality for some Y satisfying (3.7), then Y_m would not be a Banach space and Theorem 3.5's 'optimal domain' would be undefined; the proof of Theorem 3.3 goes through Theorem 3.1, so the equivalence (3.4)-(3.6) inherits the gap. This is not a detected internal contradiction, but the central result is not self-contained at the exact point where the main claims are anchored.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies embeddings of homogeneous Sobolev-type spaces into rearrangement-invariant spaces: it seeks necessary and sufficient conditions, in terms of one-dimensional inequalities, for the validity of ||u||_Y <= C ||∇^m u||_X on R^n, with m < n. The main contributions are Theorem 3.1, which gives an optimal target space X^m for a fixed domain space X under condition (3.1); Theorem 3.3, the reduction principle equating (3.4), (3.5), and (3.6); and Theorem 3.5, which gives an optimal domain space Y_m for a fixed target space Y under condition (3.7). Section 5 contains explicit examples for Lorentz--Zygmund and Orlicz spaces, and Section 6 treats optimal embeddings within the class of Orlicz spaces, including a reduction principle, Theorem 6.8. The proofs are largely built on earlier work of the author and collaborators, and several load-bearing statements are cited rather than proved.","tokens_in":27501,"tokens_out":5242,"duration_ms":50457,"significance":"If the main theorems are correct, this is a substantial contribution to the theory of optimal Sobolev embeddings: the reduction principle is elegant, the optimal spaces are given by explicit formulas in many classical cases, and the paper supplies concrete examples that sharpen Peetre's result and clarify limiting cases. The duality computation in Proposition 4.1 is clean, and the paper gives falsifiable, concrete predictions about optimal spaces. The significance is tempered by the fact that several central assertions, notably the normability of σ_m and τ_m and the Orlicz-space optimality theorems, are imported from other papers without proof, so a reader cannot fully verify the main claims from the present text alone.","major_comments":[{"comment":"The normability of the functionals σ_m and τ_m is load-bearing for the entire characterization. For σ_m, the proof says only that 'It can be proved that σ_m is a rearrangement-invariant norm if and only if (3.1) is satisfied (cf. [15, Theorem 5.4] and [20, Theorem 4.4])' and notes that the triangle inequality follows from (2.1). For τ_m, the proof says the normability is 'rather deep' and refers to [20, Theorem 4.1]. If τ_m fails to be a norm, then Y_m is not a rearrangement-invariant space and the optimality statement in Theorem 3.5 is not well defined; the proof of Theorem 3.3 uses Theorem 3.1, so the reduction principle inherits this gap. I request that the full norm proofs, or at least precise statements of the cited theorems with all hypotheses made explicit, be included in the paper.","section":"Section 3, Theorems 3.1 and 3.5"},{"comment":"The optimality half of Theorem 5.2 is asserted but not proved: the text says 'The optimality can be shown along the same lines of [12, Theorem 1.1, pp. 457] and we omit it here.' This is half of a stated theorem and is not a routine check. The omitted argument should be supplied, or the theorem should be marked as conditional on a detailed calculation available elsewhere.","section":"Section 5, Theorem 5.2"},{"comment":"The paper states 'We omit proofs in this section because they are lengthy and technical.' The entire Orlicz-space theory rests on this omission. Theorem 6.1 and Theorem 6.4 are described as applications of Theorem 3.3 together with results from [29, Chapter 3] and [30], but the transfer from the one-dimensional Orlicz results to the n-dimensional gradient inequality is not shown. Since the Boyd-index conditions and the non-existence claims are delicate, I cannot verify these theorems from the present text. At minimum, the relevant statements from [29] and [30] should be reproduced and the argument connecting them to Theorem 3.3 should be outlined.","section":"Section 6, Theorems 6.1, 6.4, and 6.8"},{"comment":"Two auxiliary results used in the main proofs are only partially proved. Proposition 4.3's first inequality is dismissed with 'For the sake of brevity, the details are omitted,' and Proposition 5.4's final case is treated with 'we can proceed similarly, omitting the proof here.' Proposition 4.3 is needed for the iteration principle used in the induction in Theorem 3.1, and Proposition 5.4 supports Theorem 5.3. These gaps, while less central than the issues above, should be closed for the paper to be self-contained.","section":"Section 4, Proposition 4.3 and Section 5, Proposition 5.4"}],"minor_comments":[{"comment":"The text contains several typographical spacing artifacts, such as 'SP ACES' in the title and 'con siderably' in the abstract; these should be corrected in the final version.","section":"Throughout"},{"comment":"The definition of the Lorentz spaces L^{p,q} lists the admissible parameter ranges, but it would help the reader to state explicitly that the functional ρ_{p,q} is only equivalent to a norm in those cases and that equality cases such as L^{1,1} and L^{∞,∞} are included.","section":"Section 2"},{"comment":"The description of the spaces Y1 and Y2 in Theorem 5.1 is terse; a sentence explaining how these spaces embed into the ambient function-space scale would improve readability.","section":"Section 5, Theorem 5.1"},{"comment":"The phrase 'an open set of Orlicz spaces' is informal; a precise statement of what is meant by the absence of an optimal Orlicz domain space is given in Theorem 6.4, but the remark could be tightened.","section":"Section 6, Remark 6.5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript relies heavily on the author's own prior results, particularly [20] and [29], for assertions that are central to the main characterization. I recommend that the editor verify that those cited results indeed establish the exact statements needed here, and that the author be asked to make the dependency explicit in the text. The paper is within the scope of the journal and the main ideas are promising, but the omitted proofs and conditional dependencies currently prevent a fully informed assessment."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on arXiv:1908.03384. The headline: this is a real advance. Mihula characterizes the m-th order gradient inequality for rearrangement-invariant spaces by reducing it to a one-dimensional Hardy-type condition, and then identifies the optimal target and domain spaces. The right-hand side uses only the top-order gradient, which distinguishes it from the full-gradient work of Alberico–Cianchi–Pick–Slavíková and Vybíral. The Orlicz section goes beyond Cianchi's m=1 target-side result by treating arbitrary m and both sides. That is a genuine addition.\n\nWhat it does well: Proposition 4.1 is a clean duality computation—the equivalence between the Hardy-type inequality and the condition on X' is immediate from the associate norm and Hardy–Littlewood. The iteration principle in Theorem 3.2 is a nice touch and the examples in Section 5 are concrete. No parameters are fitted and no result is built into its own premises. The optimal spaces are defined explicitly from X or Y.\n\nThe soft spots are about proof delegation, not about a detected error. Theorem 3.1 asserts sigma_m is a norm exactly under (3.1) and cites [15,20]. Theorem 3.5 calls the normability of tau_m 'rather deep' and defers to [20, Theorem 4.1]. Section 6 states 'We omit proofs in this section' and imports the Orlicz optimality theory from a PhD thesis [29] and [30]. Theorem 5.2 says optimality can be shown along the lines of [12] and omits it. If any of those external results has a gap, the completeness claim would shrink. That is a real concern for a paper whose selling point is completeness. It is not, by itself, a flaw in the mathematics; the external results are published (or in a thesis) and the author is open about the delegation. The self-citation to [20] is heavy, but the cited theorems are on point and published in JFA.\n\nMy conclusion: the central reduction argument holds up as far as I can see, and the omissions are packaging rather than a discovered contradiction. The stress-test note about propagation from [20] is accurate but it identifies a dependency, not an error. I would not desk-reject this. I would send it to a referee with a request that the final version either prove the normability and Orlicz optimality statements or state them in full as quoted theorems. With that, this is a paper worth citing in the function-space literature.\n\nWorth a serious referee and I'd bring it to our reading group.","headline":"A genuine characterization of m-th order gradient embeddings, with explicit optimal spaces; the main weakness is heavy delegation of normability and Orlicz optimality to external sources, not a detected error.","tokens_in":28019,"tokens_out":2718,"would_cite":true,"duration_ms":29703,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["26D10","46E30","46E35","47B38"],"pacs":[],"model":"deepseek-v4-flash","headline":"One inequality settles when Sobolev embeddings exist.","keywords":["rearrangement-invariant spaces","optimal target space","optimal domain space","reduction principle","weighted Hardy inequality","Orlicz spaces","homogeneous Sobolev spaces","fractional maximal operator"],"falsifier":"Exhibit a pair of rearrangement-invariant spaces $X,Y$ on $\\mathbb{R}^n$ with $m<n$ for which the one-dimensional inequality (3.5) fails while the gradient inequality (1.3) holds; Theorem 3.3 asserts this is impossible. A concrete check: for $n=2$, $m=1$, $X=L^2$, the theorem predicts no rearrangement-invariant target space exists, so finding any single nontrivial $Y$ satisfying $\\|u\\|_{Y}\\le C\\|\\nabla u\\|_{L^2}$ would refute it. Conversely, showing for some $X$ that condition (3.1) holds but the optimal target $X^m$ is not the smallest valid target would settle the optimality claim.","tokens_in":26982,"feed_emoji":"🎯","tokens_out":8339,"duration_ms":78739,"temperature":0.7,"pith_summary":"The paper characterizes exactly when the gradient inequality $\\|u\\|_{Y(\\mathbb{R}^n)}\\le C\\|\\nabla^m u\\|_{X(\\mathbb{R}^n)}$ can hold, for rearrangement-invariant spaces $X,Y$ and $m<n$, and it identifies the optimal (smallest) target space and the optimal (largest) domain space whenever one exists. The central device is a reduction principle: the $n$-dimensional inequality is equivalent to a one-dimensional weighted Hardy inequality, and to its dual form, so existence can be decided by a single Hardy-type test. When the test fails, no rearrangement-invariant space on the other side works at all. The same program is carried out inside the class of Orlicz spaces, where the optimal spaces are described by Young functions built from the data, with the caveat that an optimal Orlicz domain need not exist even when Orlicz domains do exist. Worked examples cover Lebesgue, Lorentz, Lorentz-Zygmund, and Orlicz spaces, including the classical fact that Peetre's Lorentz space $L^{p^*,p}$ is the smallest possible target for $L^p$.","feed_headline":"One inequality settles when Sobolev embeddings exist","feed_subtitle":"It reduces the n-dimensional question to a one-dimensional Hardy test and names optimal spaces on both sides.","key_machinery":"The load-bearing object is the reduction principle (Theorem 3.3), which transfers the $n$-dimensional gradient inequality into a one-dimensional Hardy-type inequality with kernel $s^{m/n-1}$. The optimal target norm is $\\sigma_m(f)=\\|t^{m/n}f^{**}(t)\\|_{X'(0,\\infty)}$, built from the maximal rearrangement $f^{**}$ and the associate norm of $X$; the optimal domain norm $\\tau_m$ is the supremum of the same Hardy expression over all one-dimensional functions equimeasurable with $f$. These operators make the problem tractable: checking the Hardy inequality on $(0,\\infty)$ decides the embedding, and the norms literally describe the smallest target and largest domain. In the Orlicz setting, the same principle is implemented through Young functions $A_m$ and $B_m$ defined by integrals of the data, giving the reduction principle of Theorem 6.8.","core_discovery":"On the paper's own terms, the discovery is a complete equivalence. Theorem 3.3 states that for $m<n$ and rearrangement-invariant spaces $X,Y$ over $\\mathbb{R}^n$, the embedding inequality (1.3) holds if and only if the one-dimensional inequality $\\bigl\\|\\int_t^\\infty f(s)s^{m/n-1}\\,ds\\bigr\\|_{Y(0,\\infty)}\\le C\\|f\\|_{X(0,\\infty)}$ holds for all nonnegative $f$, and also if and only if the dual inequality $\\|t^{m/n}g^{**}(t)\\|_{X'(0,\\infty)}\\le C\\|g\\|_{Y'(0,\\infty)}$ holds. For fixed $X$, the smallest target space $X^m$ is defined by the norm $\\sigma_m(f)=\\|t^{m/n}f^{**}(t)\\|_{X'(0,\\infty)}$; it exists exactly when condition (3.1) holds. For fixed $Y$, the largest domain space $Y_m$ is defined by a supremum over equimeasurable rearrangements of the same Hardy expression; it exists exactly when condition (3.7) on the fundamental function holds. The paper also proves an iteration principle for optimal targets and gives the analogous optimal Orlicz-space statements, including a four-way reduction principle for Orlicz spaces (Theorem 6.8).","pith_inferences":["One testable extension is to compute optimal target and domain spaces for weighted variants of the spaces, or for quasinormed spaces such as $L^{p,q}$ with $0<p<1$, where the current framework does not apply but the same reduction may be expected to hold after an appropriate renorming.","The reduction principle suggests that for numerical or computational checks of Sobolev inequalities, it is enough to test one-dimensional radial profiles, since validity for all functions is equivalent to validity of the Hardy operator acting on rearrangements.","The Orlicz 'no optimal domain' phenomenon is likely a general feature of any proper subclass of rearrangement-invariant spaces that excludes the universal optimal domain: one should expect open families of valid domains rather than a largest element whenever the universal optimum falls outside the subclass.","The equivalence with boundedness of the fractional maximal operator (Remark 3.4) connects the paper directly to harmonic analysis: optimal Sobolev target spaces are exactly optimal range spaces for the fractional maximal operator, so sharp constants or restricted weak-type estimates for $M_{m/n}$ could be read off from the examples."],"forward_implications":["If $X$ satisfies condition (3.1), the optimal target space $X^m$ exists and any valid target $Y$ must contain it; if (3.1) fails, no rearrangement-invariant target exists.","If $Y$ satisfies condition (3.7), the optimal domain space $Y_m$ exists and is the largest rearrangement-invariant space from which the gradient inequality maps into $Y$; otherwise no domain exists.","Optimal target spaces are stable under iteration: applying the construction twice with orders $k$ and $l$ yields the same space as applying it once with order $k+l$ (Theorem 3.2).","In the Orlicz class, the optimal target exists exactly when condition (5.3) holds, but the optimal domain can fail to exist even when Orlicz domains do exist; in that case the valid Orlicz domains form an open family with no largest member (Remark 6.5).","For Lorentz-Zygmund data, the paper's formulae give explicit optimal spaces, recovering the classical fact that $L^{p^*,p}$ is the smallest target for $L^p$ in the first-order case."],"supporting_citations":[{"why":"supplies the normability of the optimal-domain functional $\\tau_m$, the boundedness of the operator $T_{m/n}$, and the connection to the fractional maximal operator used in Remark 3.4.","marker":"[20]"},{"why":"provides the reduction principle for higher-order Sobolev embeddings and the normability criteria used in Theorem 3.1.","marker":"[15]"},{"why":"constructs the Orlicz-Lorentz target space $L(n/m,1,E_m)$ and the inequality (5.5) used in Theorem 5.2.","marker":"[12]"},{"why":"supplies the optimal Orlicz embedding theorems, from a PhD thesis, from which Theorems 6.1, 6.4, and 6.8 are deduced.","marker":"[29]"},{"why":"gives the optimality results for the fractional maximal operator in Orlicz spaces used in Section 6.","marker":"[30]"},{"why":"supplies the rearrangement-based gradient estimate used in the base step of the optimal-target proof.","marker":"[13]"},{"why":"provides the radial-function construction whose argument pattern is used to prove optimality of $X^m$.","marker":"[3]"},{"why":"supplies the associate-space descriptions of (generalized) Lorentz-Zygmund spaces used to compute the examples in Section 5.","marker":"[34]"},{"why":"provides the working theory of rearrangement-invariant spaces, including the Hardy-Littlewood inequality and the Luxemburg representation theorem.","marker":"[6]"}],"fun_headline_variants":["Hardy test pins down Sobolev embeddings","Sobolev embeddings fully characterized via Hardy","Exact Sobolev embedding spaces characterized","One-dimensional test settles Sobolev embeddings","Sobolev embeddings reduced to one Hardy test"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The characterization is complete only to the extent that the external boundedness and optimality results it invokes—normability of $\\sigma_m$ and $\\tau_m$, the fractional-maximal characterizations, and the Orlicz embedding theorems—are themselves correct; the paper does not reproduce those proofs, and Section 6 explicitly says 'We omit proofs in this section.'","fun_headline_variants_meta":{"raw":{"variants":["Hardy test pins down Sobolev embeddings","Sobolev embeddings fully characterized via Hardy","Exact Sobolev embedding spaces characterized","One-dimensional test settles Sobolev embeddings","Sobolev embeddings reduced to one Hardy test"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001411,"raw_usage":{"total_tokens":5690,"prompt_tokens":925,"completion_tokens":4765,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":541,"completion_tokens_details":{"reasoning_tokens":4695}},"tokens_in":541,"tokens_out":4765,"duration_ms":32688,"temperature":1.0,"reasoning_tokens":4695,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:14:33.685296+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit a pair of rearrangement-invariant spaces $X,Y$ on $\\mathbb{R}^n$ with $m<n$ for which the one-dimensional inequality (3.5) fails while the gradient inequality (1.3) holds; Theorem 3.3 asserts this is impossible. A concrete check: for $n=2$, $m=1$, $X=L^2$, the theorem predicts no rearrangement-invariant target space exists, so finding any single nontrivial $Y$ satisfying $\\|u\\|_{Y}\\le C\\|\\nabla u\\|_{L^2}$ would refute it. Conversely, showing for some $X$ that condition (3.1) holds but the optimal target $X^m$ is not the smallest valid target would settle the optimality claim.","supporting_citations":[{"cited_title":"Edmunds, Z","cited_arxiv_id":null,"evidence_quote":"supplies the normability of the optimal-domain functional $\\tau_m$, the boundedness of the operator $T_{m/n}$, and the connection to the fractional maximal operator used in Remark 3.4."},{"cited_title":"Cianchi, L","cited_arxiv_id":null,"evidence_quote":"provides the reduction principle for higher-order Sobolev embeddings and the normability criteria used in Theorem 3.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"constructs the Orlicz-Lorentz target space $L(n/m,1,E_m)$ and the inequality (5.5) used in Theorem 5.2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the optimal Orlicz embedding theorems, from a PhD thesis, from which Theorems 6.1, 6.4, and 6.8 are deduced."},{"cited_title":"Cianchi and L","cited_arxiv_id":null,"evidence_quote":"supplies the rearrangement-based gradient estimate used in the base step of the optimal-target proof."},{"cited_title":"Alberico, A","cited_arxiv_id":null,"evidence_quote":"provides the radial-function construction whose argument pattern is used to prove optimality of $X^m$."},{"cited_title":"Opic and L","cited_arxiv_id":null,"evidence_quote":"supplies the associate-space descriptions of (generalized) Lorentz-Zygmund spaces used to compute the examples in Section 5."},{"cited_title":"Bennett and R","cited_arxiv_id":null,"evidence_quote":"provides the working theory of rearrangement-invariant spaces, including the Hardy-Littlewood inequality and the Luxemburg representation theorem."}],"review_version":1}