{"id":"f92c6784-b41e-4c81-a6f2-69cf0725042d","arxiv_id":"2506.08891","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A distributional-derivative Fourier transform is constructed on variable exponent Lebesgue spaces over R, with an isometric Banach-space norm and norm-convergent inversion.","lead":"This paper defines a Fourier transform on variable exponent Lebesgue spaces by using the distributional derivative of an auxiliary Hölder continuous function, and equips the space of transforms with a norm making it isometric to L^{p(·)}(R). It also proves exchange and norm-inversion theorems, extending a constant-exponent construction to exponents that vary, including some that take the value infinity.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 6.4's norm-inversion claim is false for the LH1∞ class as stated: an exponent with p=∞ on a bounded set gives an L∞ component that convolution cannot regularize.","rationale":"The reader's duality objection to Theorem 6.2 is legitimate: Proposition 2.79 of [3] typically requires p+<∞ and cannot supply bg∈L^{q(·)} in the LH1∞ case. That gap is fixable, however, because for LH1∞ one has q(s)≥C^{-1}log(e+|s|) and the BV decay g(s)=o(|s|^{-1/p−}) gives bg(s)=O(1/|s|), so bg∈L^{q(·)} can be shown directly. The inversion theorem is a harder obstruction: the counterexample above is a genuine element of the stated class for which the claimed norm convergence fails. The norm of L^{p(·)} controls the L∞ norm on Ω∞, so the discontinuous f in the counterexample cannot be approximated in norm by its continuous mollifications. This is not a matter of missing justification; the statement is false. The core construction in Theorem 4.1 appears sound: the estimates for Ψ_f and the injectivity argument via the heat kernel are plausible, and for p+<∞ the inversion and exchange theorems are likely correct. But the abstract's claim of inversion in norm for the large class including LH1∞ exponents is false, so the current version should not be accepted as is; restricting Theorem 6.4 to p+<∞ would be a necessary revision.","tokens_in":19302,"tokens_out":26859,"duration_ms":301511,"concrete_test":"Set p(x)=∞ on [-1,1], p(x)=1+1/log(e+|x|) outside, f=χ_[0,1], and ψ_a the Gaussian mollifier. Compute ||f*ψ_a-f||_{L∞([-1,1])} for a=10^{-k}, k=1,...,6; if it stays above roughly 1/2, Theorem 6.4's LH1∞ case is refuted. Analytically, use (f*ψ_a)(x)=Φ(x/a) for x∈(0,a) and Φ(0+)=1/2, so the essential supremum over Ω∞ does not tend to 0.","verdict_should_be":"REJECT","load_bearing_attack":"The load-bearing flaw is in Theorem 6.4, not only in the duality step of Theorem 6.2. Taking LH1∞ in the theorem's sense (1/p log-Hölder at infinity with p∞=1), choose p(x)=∞ on [-1,1] and p(x)=1+1/log(e+|x|) for |x|>1. Then p∈LH1∞, p+=∞, and Ω∞=[-1,1] has positive measure. Let f=χ_[0,1] and ψ the Gaussian approximate identity; ψ satisfies the theorem's integrability hypotheses. Then f∈L^{p(·)}, and the conclusion would require ||f*ψ_a-f||_{p(·)}→0. But ||·||_{p(·)} dominates ||·||_{L∞(Ω∞)}. For x∈(0,a), (f*ψ_a)(x)=Φ(x/a), the standard normal CDF, while f(x)=1; hence ess sup_{Ω∞}|f*ψ_a-f| has liminf ≥1/2. Therefore f*ψ_a does not converge to f in L^{p(·)}. The proof attributes convergence to [3, Theorem 5.4], an approximate-identity/density result that normally requires p+<∞; it cannot cover this case. The theorem needs p+<∞ or an additional assumption excluding unbounded exponent sets; the abstract's inversion claim for the full LH1∞ class is false as stated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript defines, for f in L^{p(·)}(R) with p+<∞ or 1/p(·) log-Hölder continuous at infinity with limit 1 (the LH1∞ class), an auxiliary function Ψ_f(s)=∫(1−e^{-ist})/(it) f(t)dt, proves Hölder or Lipschitz estimates for Ψ_f, and then defines the Fourier transform of f as the distributional derivative Ψ'_f. Theorem 4.1 equips the spaces B^{p(·)}(R) and A^{p(·)}(R) of such transforms with the norm ||Ψ_f||=||f||_{p(·)} and proves that they are isometrically isomorphic to L^{p(·)}(R). Section 5 shows consistency with the tempered-distribution Fourier transform and derives standard operational rules. Section 6 introduces continuous-primitive integrals of the transform and states an exchange theorem and a norm-inversion theorem. The main difficulty is in Section 6: the exchange and inversion results are claimed for the full LH1∞ class, but their proofs require p+<∞ at a load-bearing point, and the inversion theorem is false as stated for exponents that take the value ∞ on a set of positive measure.","tokens_in":19625,"tokens_out":8424,"duration_ms":100062,"significance":"The construction in Theorem 4.1 is a genuine and clean extension of Talvila's Fourier transform to variable exponent Lebesgue spaces, with an explicit norm and a careful injectivity argument via the heat kernel. The estimates in Section 3 are explicit and the reduction of the isometry to injectivity is well motivated. If the scope were restricted to exponents with p+<∞, the exchange and inversion results would be plausible and would give a useful new framework. As written, however, the paper advertises the LH1∞ class in the abstract and in Theorems 6.2 and 6.4, and in that class the inversion claim is false and the exchange proof has a duality gap. The central isometry theorem is not the problem; the Section 6 claims need correction before the paper can be accepted.","major_comments":[{"comment":"Theorem 6.4 is false for the LH1∞ class as stated. Let p(x)=∞ on [-1,1] and p(x)=1+1/log(e+|x|) for |x|>1; this exponent satisfies 1/p(·) log-Hölder continuity at infinity with p∞=1, so it belongs to the LH1∞ class. Take f=χ_{[0,1]} and let ψ be a standard Gaussian. Then ψ satisfies all the kernel hypotheses of the theorem: ψ∈L^1(R), ∫ψ=1, ψ^ is absolutely continuous, and ∫|s||ψ^(s)|ds and ∫|s||ψ^'(s)|ds are finite. The conclusion would require ||f∗ψ_a−f||_{p(·)}→0 as a→0+. But for x∈(0,a), f∗ψ_a(x)=Φ(x/a), the standard normal CDF, while f(x)=1, so ||f∗ψ_a−f||_{p(·)} ≥ esssup_{[-1,1]}|f∗ψ_a−f| = sup_{u∈(0,1)}(1−Φ(u)) = 1/2 for every a>0. Hence the claimed norm convergence fails. The proof's appeal to [3, Theorem 5.4] is an approximate-identity/norm-density result that is not valid when p+<∞ is not assumed, and the LH1∞ class explicitly allows p=∞ on a set of positive measure. The theorem, and the corresponding claim in the abstract, must be restricted, for example to p+<∞ or to exponents with |Ω∞|=0.","section":"Theorem 6.4"},{"comment":"The proof of Theorem 6.2 concludes that bg∈L^{q(·)} by invoking [3, Proposition 2.79] after showing that f↦∫\\hat f g is a bounded linear functional on L^{p(·)}. That duality identification requires p+<∞. In the LH1∞ case with p=∞ on a set of positive measure, the dual of L^{p(·)} is strictly larger than L^{q(·)}, so the boundedness of the functional does not imply bg∈L^{q(·)}. The exchange theorem is therefore unproved for the full stated class. This is load-bearing because Theorem 6.4 relies on the exchange formula to identify I_a[f] with f∗ψ_a; the LH1∞ version of the inversion theorem cannot inherit a valid proof from Theorem 6.2.","section":"Theorem 6.2, final step"}],"minor_comments":[{"comment":"The notation ∫_∞^∞ \\hat f g is nonstandard and should be replaced by an explicitly defined improper integral over R, for example ∫_{−∞}^{∞} with the continuous-primitive convention explained in the preamble.","section":"Definition 6.1(2)"},{"comment":"In equation (20) the expressions 'dg_s(s)' and 'dg_i(s)' appear to be typographical errors; the intended notation is dg_1(s) and dg_2(s).","section":"Theorem 6.2 proof"},{"comment":"The claim that when restricted to constant exponents the results 'coincide precisely' with [12] should be qualified, since [12] treats 1≤p<∞ whereas the present paper also allows p=∞; the exact sense in which the constant-exponent case matches should be stated.","section":"Introduction"},{"comment":"In the first part of Example 5.5, the notation such as ∥\\hat f∥_{p±}^{∞} is ambiguous; writing the relevant sup norms explicitly would improve readability.","section":"Example 5.5"}],"recommendation":"major_revision","confidential_remarks":"The false LH1∞ inversion claim is prominent in the abstract, so this is not a purely local blemish; however, the core isometry theorem is sound and the Section 6 defects appear repairable by restricting the scope to p+<∞ or by excluding positive-measure sets where p=∞. I therefore recommend major revision rather than rejection, but the authors must either correct Theorems 6.2 and 6.4 for the full LH1∞ class or remove that class from the claimed results."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know about arXiv:2506.08891. First, the main isometry construction is genuinely new and largely correct: for p(·) with p+<∞ or 1/p(·) log-Hölder at infinity with p∞=1, the paper defines Ψ_f and spaces B^{p(·)} and A^{p(·)} as the distributional derivative, and proves these are isometrically isomorphic to L^{p(·)}(R). The proof of injectivity via the heat kernel is sound, and the growth estimates in Propositions 3.2–3.4 are real work; the Lambert-W constants are a nice touch. Second, the headline inversion theorem, Theorem 6.4, is false as stated for the LH1∞ class. A concrete counterexample: take p(x)=∞ on [-1,1] and p(x)=1+1/log(e+|x|) outside, f=χ_[0,1], and ψ a Gaussian approximate identity. The L^{p(·)} norm dominates the L∞ norm on Ω∞=[-1,1], and f*ψ_a cannot converge to f in that L∞ component because f is discontinuous at 0. The proof invokes [3, Theorem 5.4], which requires p+<∞. So the inversion claim for the full infinite-exponent class is not merely unproved; it is false. The same suspicion applies to Theorem 6.2, which uses a duality result that also requires p+<∞. The fix is straightforward: state these theorems with p+<∞, or add a condition like |Ω∞|=0, or else prove a weaker form of convergence that does not require L∞ approximation. The paper is worth a serious referee because the core isometry theorem is solid and useful; but the abstract overclaims, and the current version should not be accepted. I would want to see the authors address the infinite-exponent case before citing it.","headline":"Solid isometry construction for variable-exponent Fourier transforms, but the headline inversion theorem is false as stated for exponents that take the value infinity on a positive-measure set.","tokens_in":20124,"tokens_out":6092,"would_cite":false,"duration_ms":69344,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42A38","46E30","26A42","46B04"],"pacs":[],"model":"deepseek-v4-flash","headline":"A Fourier transform is defined for every function in a variable-exponent Lebesgue space, and the space of transforms is shown to be a Banach space isometrically isomorphic to the original space, with norm inversion and an exchange theorem.","keywords":["variable exponent Lebesgue space","Fourier transform","tempered distribution","continuous primitive integral","Banach space","log-Hölder continuity","distributional derivative","isometric isomorphism"],"falsifier":"Construct a specific $p(\\cdot)$ that equals $\\infty$ on a set of positive measure and satisfies the log-Hölder decay condition at infinity with $p_\\infty=1$. For a function $g$ of bounded variation with $g(\\pm\\infty)=0$ and $\\int |s|^{1/p_-}\\,d|g|<\\infty$, evaluate the functional $f \\mapsto \\int \\hat{f}\\,g$; if this functional is not given by integration against any $h \\in L^{q(\\cdot)}(\\mathbb{R})$, the exchange theorem's proof step fails and the theorem would need repair for that exponent.","tokens_in":19133,"feed_emoji":"","tokens_out":12760,"duration_ms":119149,"temperature":0.7,"pith_summary":"The paper extends the Fourier transform to variable-exponent Lebesgue spaces $L^{p(\\cdot)}(\\mathbb{R})$, where the exponent $p(\\cdot)$ can vary over $[1,\\infty]$ rather than being a fixed number. For each $f \\in L^{p(\\cdot)}(\\mathbb{R})$ it constructs the auxiliary function $\\Psi_f(s)=\\int_{\\mathbb{R}} \\frac{1-e^{-ist}}{it} f(t)\\,dt$ and defines the Fourier transform as the distributional derivative $\\hat{f}=\\Psi'_f$. The main theorem shows that, whenever $p_+<\\infty$ or $1/p(\\cdot)$ is log-Hölder continuous at infinity with $p_\\infty=1$, the spaces of these transforms are Banach spaces isometrically isomorphic to $L^{p(\\cdot)}(\\mathbb{R})$ with the inherited norm. The paper goes on to prove that the transform agrees with the classical Fourier transform on tempered distributions, that inversion holds in norm, and that an exchange formula extends Parseval-type identities. If the results are correct, every function in a variable-exponent Lebesgue space of this type has a genuine Fourier calculus, recovering the constant-exponent theory as the special case where $p(\\cdot)$ is constant.","feed_headline":"Every variable-exponent function gets a Fourier transform","feed_subtitle":"A norm-convergent inversion formula and an exchange theorem make the transform usable in analysis.","key_machinery":"The load-bearing object is the auxiliary function $\\Psi_f(s)=\\int_{\\mathbb{R}} \\frac{1-e^{-ist}}{it} f(t)\\,dt$, shown to be Hölder continuous with exponent $1/p_+$ when $p_+<\\infty$ and Lipschitz continuous when $1/p(\\cdot)$ is log-Hölder at infinity with limit $1$. The proof of these estimates uses the generalized Hölder inequality for variable-exponent spaces and bounds the $L^{q(\\cdot)}$ norm of the kernel $u_s(t)=(1-e^{-ist})/it$, producing constants that involve the integral $C_q$ of $|\\sin y / y|^q$ and the Lambert $W$ function in the unbounded-exponent case. With $\\Psi_f$ continuous, the Fourier transform is defined as the distributional derivative $\\Psi'_f$, which allows the continuous primitive integral and Stieltjes integration to be used for exchange and inversion results. The isometry from $L^{p(\\cdot)}$ to the transform space is what turns these estimates into a Banach space isomorphism.","core_discovery":"The central discovery is that a single family of Hölder continuous primitives $\\Psi_f$ carries the entire Fourier analysis of variable-exponent Lebesgue spaces. For exponents with finite essential supremum, or with $1/p(\\cdot)$ log-Hölder continuous at infinity and $p_\\infty=1$, the space $\\{\\Psi_f : f \\in L^{p(\\cdot)}(\\mathbb{R})\\}$ and its distributional derivative space $\\{\\Psi'_f : f \\in L^{p(\\cdot)}(\\mathbb{R})\\}$ are Banach spaces isometrically isomorphic to $L^{p(\\cdot)}(\\mathbb{R})$ under the norm $\\|\\Psi_f\\| = \\|f\\|_{p(\\cdot)}$. The transform defined this way coincides with the tempered-distribution Fourier transform, and the paper establishes an exchange formula $\\int_{\\mathbb{R}} \\hat{f}\\, g = \\int_{\\mathbb{R}} f\\, \\hat{g}$ for suitable $g$ and norm-convergent inversion with standard approximate identities. This gives a Fourier theory that works even when the exponent is unbounded, a regime where many tools of the constant-exponent theory fail.","pith_inferences":["The isometric isomorphism suggests the transform could be iterated, but the image space $A^{p(\\cdot)}$ is not shown to be closed under the transform, so a true Fourier inversion on the image is an open question rather than a corollary.","If the duality step in the exchange theorem fails for exponents with $p=\\infty$ on positive measure, the inversion theorem may still be salvageable by adding a separate restriction on the set where the exponent is infinite.","The explicit constants involving $C_q$ and the Lambert $W$ function could be optimized to give sharp operator norms for the maps $f \\mapsto \\Psi_f$ in simple cases such as $p(x)=1+\\kappa \\ln(e+|x|)$.","The pointwise-kernel construction suggests a route to variable-exponent Fourier analysis on $\\mathbb{R}^n$ by taking products of the one-dimensional kernel, though the distributional derivative step would need a higher-dimensional Hölder condition."],"forward_implications":["Every $f \\in L^{p(\\cdot)}(\\mathbb{R})$ with $p_+<\\infty$ or with $1/p(\\cdot)$ log-Hölder continuous at infinity and $p_\\infty=1$ has a generalized Fourier transform $\\hat{f}=\\Psi'_f$, and the map $f \\mapsto \\hat{f}$ is an isometric isomorphism onto a Banach space.","The generalized transform agrees with the classical Fourier transform on tempered distributions, so it inherits standard identities for translations, modulations, and differentiation.","For the Cesàro–Fejér, Abel–Poisson, and Gauss–Weierstrass kernels, the approximation $I_a[f]$ converges to $f$ in $L^{p(\\cdot)}$ norm as $a \\to 0^+$.","The exchange formula $\\int \\hat{f}\\, g = \\int f\\, \\hat{g}$ holds under mild assumptions on $g$, yielding a variable-exponent analog of Parseval's identity.","When $p(\\cdot)$ is constant, all results reduce to the known Fourier transform theory for $L^p$ spaces."],"supporting_citations":[{"why":"Supplies the generalized Hölder inequality, log-Hölder continuity theory, duality proposition, and convolution approximation used in the isometry and inversion proofs.","marker":"[3]"},{"why":"The constant-exponent construction that this paper extends; the claims recover its results when p(·) is constant.","marker":"[12]"},{"why":"Provides the continuous primitive integral that defines integration of the Fourier transform in Section 6.","marker":"[11]"},{"why":"Supplies the Jordan-decomposition fact for functions of bounded variation used in the exchange theorem proof.","marker":"[2]"},{"why":"Foundational reference for variable-exponent Lebesgue spaces, used for basic structural properties.","marker":"[9]"}],"fun_headline_variants":["Fourier transform defined for all variable-exponent functions","Isometric Fourier transform for variable exponent Lebesgue spaces","Exchange theorem and norm inversion for new Fourier transform","Fourier analysis extended to unbounded exponents","Hölder primitives provide a Fourier transform for every p(·)"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of the exchange formula assumes that every bounded linear functional on $L^{p(\\cdot)}(\\mathbb{R})$ is integration against a function of the conjugate variable-exponent space $L^{q(\\cdot)}(\\mathbb{R})$, a duality result whose standard form requires $p_+<\\infty$; the paper applies it to exponents that may take the value $\\infty$ on a set of positive measure, where the dual space is larger.","fun_headline_variants_meta":{"raw":{"variants":["Fourier transform defined for all variable-exponent functions","Isometric Fourier transform for variable exponent Lebesgue spaces","Exchange theorem and norm inversion for new Fourier transform","Fourier analysis extended to unbounded exponents","Hölder primitives provide a Fourier transform for every p(·)"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000908,"raw_usage":{"total_tokens":3851,"prompt_tokens":840,"completion_tokens":3011,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":456,"completion_tokens_details":{"reasoning_tokens":2931}},"tokens_in":456,"tokens_out":3011,"duration_ms":25776,"temperature":1.0,"reasoning_tokens":2931,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:03:20.963001+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a specific $p(\\cdot)$ that equals $\\infty$ on a set of positive measure and satisfies the log-Hölder decay condition at infinity with $p_\\infty=1$. For a function $g$ of bounded variation with $g(\\pm\\infty)=0$ and $\\int |s|^{1/p_-}\\,d|g|<\\infty$, evaluate the functional $f \\mapsto \\int \\hat{f}\\,g$; if this functional is not given by integration against any $h \\in L^{q(\\cdot)}(\\mathbb{R})$, the exchange theorem's proof step fails and the theorem would need repair for that exponent.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the generalized Hölder inequality, log-Hölder continuity theory, duality proposition, and convolution approximation used in the isometry and inversion proofs."},{"cited_title":"The Fourier Transform in Lebesgue spaces,Czech","cited_arxiv_id":null,"evidence_quote":"The constant-exponent construction that this paper extends; the claims recover its results when p(·) is constant."},{"cited_title":"The distributional Denjoy integral,Real Anal","cited_arxiv_id":null,"evidence_quote":"Provides the continuous primitive integral that defines integration of the Fourier transform in Section 6."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Jordan-decomposition fact for functions of bounded variation used in the exchange theorem proof."},{"cited_title":"and R´ akosn ´ ık, J","cited_arxiv_id":null,"evidence_quote":"Foundational reference for variable-exponent Lebesgue spaces, used for basic structural properties."}],"review_version":1}