{"id":"40a8cff9-dae0-4e77-8198-bca5c1897c8f","arxiv_id":"1908.01911","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Local Hardy spaces on spaces of homogeneous type are characterized equivalently by maximal functions, atoms, and Littlewood–Paley functions, and their dual spaces are identified.","lead":"This paper builds a complete theory of local Hardy spaces, tools for studying oscillation and singular integrals, on very general rough metric measure spaces. It shows that five different definitions of these spaces coincide and identifies their dual spaces, completing a program that started in the 1970s.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 5 applies the discrete Calderón reproducing formula beyond its stated hypothesis at the lowest scale; without a low-frequency estimate, the Lusin-area characterization lacks proof.","rationale":"I read the paper as an extension of the authors' global Hardy-space theory to local Hardy spaces on spaces of homogeneous type. The main logical structure is plausible, and many estimates are written out in detail. The reader's weakest assumption — the existence of an exp-IA TI and the inhomogeneous Calderón reproducing formulae from [28] — is indeed load-bearing. My stress-test identifies a more specific gap within that assumption: Theorem 2.7 is stated for (˚Gη0(β,γ))′, but Section 5 applies it to f ∈ (Gη0(β,γ))′ for the l=0 term, whose test function E_0(x,·) has nonzero integral. If this application is not justified, the proof of Theorem 5.1, and hence the Lusin-area characterization of h^p(X), is incomplete. Other concerns, such as omitted proofs for Proposition 3.9 and parts of Section 7, support a conditional verdict but are less decisive because they are routine adaptations of cited arguments. I do not believe the paper should be rejected: the gap appears fixable by either extending the discrete reproducing formula to (Gη0(β,γ))′ or handling the low-frequency term separately. Since the reader already assigned CONDITIONAL, I keep that verdict; the concern reinforces the need for the authors to clarify the exact domain of Theorem 2.7 and to supply the missing low-frequency estimate.","tokens_in":568,"tokens_out":7960,"duration_ms":407681,"concrete_test":"Verify whether Theorem 2.7 in [28] is actually proved for all f ∈ (Gη0(β,γ))′ or only for f ∈ (˚Gη0(β,γ))′. If only the latter is available, independently derive the identity (2.6) when paired with the non-cancellating test function E_0(x,·), or produce a separate bound of the form |E_0 f(x)| ≤ C {M([S0(f)]^r)(x)}^{1/r} for some r<p. A concrete computational check on X = R with standard dyadic wavelets and f a non-zero-mean indicator of a unit ball would reveal whether the right-hand side of (2.6), truncated and tested against E_0(x,·), converges to E_0 f(x). If the identity or the low-frequency bound fails, Theorem 5.1's proof breaks for l=0.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central equivalences in Sections 3–5 rely on the inhomogeneous discrete reproducing formula of Theorem 2.7. As stated, Theorem 2.7 holds for f ∈ (˚Gη0(β,γ))′, i.e. for distributions acting on test functions with zero integral. However, the local Hardy space h^p(X) is defined as a subspace of (Gη0(β,γ))′, which admits distributions without zero moments. In the proof of Theorem 5.1, the formula is applied pointwise to E_l f(x)=⟨f,E_l(x,·)⟩. For l≥1, E_l(x,·) has zero integral, so the ˚G restriction is harmless. For l=0, however, E_0(x,·) has integral 1 and is not in ˚G. Thus Theorem 2.7 as stated cannot justify the expansion of E_0 f(x). This l=0 term is precisely the local low-frequency part; without a separate bound for ∥E_0 f∥ in terms of S0(f), the proof that h^p(X) equals the Lusin-area local Hardy space is incomplete. The same issue recurs in Proposition 5.5 and Theorem 5.7, where f ∈ (Gη0(β,γ))′ but Theorem 2.7 is invoked. This is not a demonstrated contradiction: the formula may extend to (Gη0(β,γ))′ because the first N terms are designed to absorb constants. But the text neither states nor proves that extension, and it does not handle l=0 separately. The reader's weakest assumption correctly identifies the external Calderón reproducing formulae as load-bearing; the present concern is a sharper, internal mismatch: the theorem's stated domain is too narrow for one of its central applications.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a real-variable theory of local Hardy spaces on spaces of homogeneous type, for p in the range (omega/(omega+eta), 1], with omega the upper dimension and eta the Holder regularity index of the underlying wavelet system. It introduces local Hardy spaces via grand, radial, and non-tangential maximal functions; proves their mutual equivalence; establishes atomic, Lusin-area, Littlewood-Paley g-function and g*-lambda-function characterizations; compares the local spaces with the Coifman--Weiss Hardy spaces, including the finite-measure case; and obtains finite atomic decompositions and dual spaces modeled on local Campanato and local Lipschitz spaces. The main tools are the inhomogeneous Calderon reproducing formulae from the authors' earlier work [28] and the global Hardy-space theory from [27].","tokens_in":73627,"tokens_out":5571,"duration_ms":57378,"significance":"If the claims are correct, the paper gives a substantial and coherent extension of Goldberg's local Hardy-space theory to general spaces of homogeneous type without reverse-doubling assumptions, and it answers the finite-measure version of the Coifman--Weiss question about radial maximal characterizations. The finite atomic characterization and the dual-space result for the full range p in (omega/(omega+eta),1) go beyond earlier work of Dafni and Yue, which treated only p=1. The authors are explicit that the results depend on the existence of exp-IATIs and on the inhomogeneous Calderon reproducing formulae established in [28]; those are published upstream results, not circular assumptions of this paper. A weakness is that several central proofs are deferred to 'similar' arguments in [27], while the local maximal functions are deliberately modified, so the local adaptation is not a purely formal corollary.","major_comments":[{"comment":"There is a domain mismatch in the use of the discrete reproducing formula. Theorem 2.7 is stated for f in (˚G_eta^0(beta,gamma))', the dual of test functions with zero integral. In the proof of Theorem 5.1, however, the formula is applied pointwise to write E_l f(x) = <f, E_l(x,.)> for f in (G_eta^0(beta,gamma))'. For l=0 the kernel E_0(x,.) has integral 1 by Definition 2.4(iii), so E_0(x,.) is not in ˚G_eta^0(beta,gamma), and the identity in (˚G_eta^0(beta,gamma))' cannot be evaluated at E_0(x,.). The same gap occurs in the proof of Theorem 5.7 when Q_l f(z) is expanded for l=0. The paper needs either a proof that Theorem 2.7 extends to (G_eta^0(beta,gamma))' with the first N terms absorbing constants, or a separate estimate for the low-frequency term E_0 f in terms of the Lusin-area function. Without this, the proof of the Lusin-area characterization is incomplete.","section":"Theorem 5.1 and Section 5"},{"comment":"Several load-bearing equivalences are stated without proof or with only a statement that the argument is 'similar' to [27]. In particular, Theorem 3.8 is obtained by combining propositions with the details omitted; Theorem 4.13 and Theorem 5.6 are stated with the details omitted; Proposition 7.1 is described as 'quite similar' to [27, Theorem 7.1] and only key points are given; and Theorem 7.4 explicitly proves only the case q=1, with the case q in (1,infty] deferred. Because the local radial and non-tangential maximal functions are not the same as in [27] (see Remark 3.2), these results are not formal restatements of the global theory. The manuscript should either include the local adaptations or state precisely which statements in [27] imply them, especially for the finite atomic characterization and the dual-space theorem.","section":"Theorems 3.8, 4.13, 5.6, 7.4; Proposition 7.1"},{"comment":"Proposition 5.5 relies on the reproducing identity f = sum_{k=0}^infty D_k D_k f in (G_eta^0(beta,gamma))', whose derivation is only described as 'a similar argument to that used in the proof of Theorem 2.6'. The operators D_k are asserted to be self-adjoint and orthogonal from [1], but the exp-IATI of Definition 2.4 is not itself orthogonal. If {D_k} are special orthogonal projections from [1], the proof needs to justify carefully that they form an exp-IATI and that the identity (5.16) holds for distributions in (G_eta^0(beta,gamma))', not only for ˚G. This identity is the bridge between the Lusin-area space and the molecular characterization, so the omission is not merely technical.","section":"Proposition 5.5, Eq. (5.16)"}],"minor_comments":[{"comment":"The abstract and title contain typographical artifacts such as 'S paces'; 'Holder' should be 'Holder' with the umlaut throughout.","section":"Abstract and title"},{"comment":"'Calder´ om' should be 'Calderon'; this typo occurs in the remark and possibly elsewhere.","section":"Remark 2.5"},{"comment":"In the first paragraph of the proof, the phrase 'which further implies that (5.25) holds true' appears to refer to equation (5.5), not (5.25), since (5.25) is the display in Theorem 5.7.","section":"Proposition 5.2"},{"comment":"After proving the cancellation of a-Pa, the text says this 'shows that a-Pa satisfies Definition 5.3(ii)'; the intended reference appears to be Definition 5.3(iii), the cancellation condition.","section":"Proposition 6.3"},{"comment":"The statement of the theorem mixes the cases q=1 and q in (1,infty]; the proof explicitly handles only q=1 and says the other case is similar. Since the dual-space theorem is a main application, at least a clear indication of the modifications for q>1 should be given.","section":"Theorem 7.4"}],"recommendation":"major_revision","confidential_remarks":"The paper is a natural extension of the authors' own prior work [27] and [28], and much of the verification burden is transferred to those papers. The editor may wish to ensure that the published versions of [27] and [28] contain the precise statements needed for the local theory, since the present manuscript does not reproduce them. The domain-mismatch issue with Theorem 2.7 in Section 5 is the most serious technical point and should be resolved before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth engaging with. It does the expected job: extends the local Hardy space h^p from the h^1 result of Dafni-Yue to the full range p in (omega/(omega+eta),1] on arbitrary spaces of homogeneous type, with no reverse doubling. The maximal-function, atomic, and finite-atomic characterizations are there, the dual space c_{1/p-1,q'} is identified, and the finite-measure equality with Coifman-Weiss H_CW^p answers a real open question. The debt to the authors' own earlier papers [27] and [28] is heavy, but those are published upstream results; leaning on them is legitimate, not circular. The main proof machinery is mostly checked. Proposition 3.4 and the atomic decomposition in Section 4 contain real estimates rather than hand-waving. What gives me pause is Section 5. Theorem 2.7, the discrete Calderon reproducing formula, is stated for distributions on the zero-integral test function space (G_eta^0(beta,gamma))'. The local Hardy spaces are subspaces of (G_eta^0(beta,gamma))', which includes test functions with nonzero integral. In Theorem 5.1 the authors expand E_l f(x)=<f,E_l(x,.)> using Theorem 2.7. For l >= 1 this is fine, since those kernels have zero integral. For l = 0, E_0(x,.) has integral 1 and is not in the dual space to which Theorem 2.7 applies. The same issue recurs in Theorem 5.7. The formula may well extend because the first N terms are designed to absorb constants, but the text neither states nor proves that extension, and it does not handle the l = 0 term separately. So as written, the proof that h^p equals the Lusin-area local Hardy space is incomplete, and the Littlewood-Paley characterizations rest on that missing step. The reader's stress-test note lands. There are also several places where central statements are deferred to 'similar' arguments in [27] - Theorems 3.8, 4.13, 5.6, Proposition 7.1(iii). That makes the paper hard to verify from the text alone, but it is a checkability problem, not evidence of error. The typos in the notation (including '1-exp-ITAI') are minor. Who is this for? Researchers working on Hardy spaces on metric measure spaces. If the Section 5 gap is patched, this becomes a standard reference for local Hardy spaces on spaces of homogeneous type. I would send it to a serious referee, with the advice to focus on the domain of Theorem 2.7 and the l = 0 term. It deserves revision, not rejection.","headline":"A serious extension of local Hardy space theory to general spaces of homogeneous type, but Section 5 has a real domain mismatch with the reproducing formula that needs fixing before the Littlewood-Paley characterizations are fully proven.","tokens_in":881,"tokens_out":1951,"would_cite":true,"duration_ms":43970,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42B30","42B35","42B20","30L99"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that on any space of homogeneous type, for p in the sharp range ω/(ω+η) < p ≤ 1, the local Hardy spaces defined by grand, radial, non-tangential, atomic, and Lusin-area Littlewood–Paley maximal functions all coincide…","keywords":["local Hardy space","space of homogeneous type","maximal function","atom","Littlewood–Paley function","dual space","inhomogeneous approximation of the identity","Calderón reproducing formula"],"falsifier":"Take a compact space of homogeneous type with quasi-metric constant $A_0>1$ and no reverse doubling, set $p=1$, and compute the local radial and grand maximal norms of the constant function $f\\equiv 1$ at points where coarse dyadic cubes meet; the paper predicts these norms are finite, comparable, and equal (up to constants independent of the cube partition) to the atomic norm of $f$, which is a bounded multiple of a local atom. If the ratio of the two norms depends on $A_0$ or on the particular dyadic cube system, the claimed equivalences fail.","tokens_in":73006,"feed_emoji":"📐","tokens_out":13077,"duration_ms":122580,"temperature":0.7,"pith_summary":"On any space of homogeneous type—a quasi-metric space whose balls satisfy a doubling condition—the paper proves that every natural real-variable definition of a local Hardy space yields the same space, for all exponents $p$ with $\\omega/(\\omega+\\eta) < p \\le 1$. The grand-maximal, radial-maximal, non-tangential-maximal, atomic, and Lusin-area Littlewood–Paley versions are shown to carry equivalent (quasi-)norms, and when the whole space has finite measure they also agree with the classical atomic Hardy space of Coifman and Weiss. The authors identify the duals of these spaces as local Campanato or Lipschitz spaces, extending the previously known $p=1$ case. The proof runs on inhomogeneous Calderón reproducing formulae with exponential decay, so no reverse-doubling condition is needed.","feed_headline":"Five local Hardy space definitions coincide on doubling spaces","feed_subtitle":"For p in the sharp range, atomic, maximal, and Littlewood–Paley versions match—no extra geometry needed.","key_machinery":"The engine is the inhomogeneous approximation of the identity with exponential decay (exp-IAI): a sequence of operators whose kernels have exponential off-diagonal decay, Hölder regularity of order $\\eta$, integral $1$ at the coarse scale, and zero integral at fine scales. From such a sequence, the inhomogeneous Calderón reproducing formulae (Theorems 2.6 and 2.7) express every distribution as a sum of coarse-scale averages and fine-scale kernel evaluations. The discrete version, built on a dyadic cube system, dictates the form of the local radial maximal function and the local Littlewood–Paley $g$-function, replacing function values on coarse cubes by local averages to compensate for the missing cancellation. The comparison lemmas (3.5 and 3.6) convert kernel differences at separated scales into a factor $\\delta^{|k-l|\\eta'}$, which turns the maximal-function equivalences into applications of the Hardy–Littlewood maximal theorem.","core_discovery":"The central claim is that the local Hardy space $h^p(X)$ is one object with five equivalent descriptions for $p \\in (\\omega/(\\omega+\\eta),1]$. As subspaces of the distribution space $(G^\\eta_0(\\beta,\\gamma))'$, the local grand maximal space $h^{*,\\,p}(X)$, the local radial maximal space $h^{+,\\,p}(X)$, each non-tangential maximal space $h^p_\\theta(X)$, each atomic space $h^{p,q}_{\\mathrm{at}}(X)$, and the Lusin-area space $h^p(X)$ have equivalent (quasi-)norms; Theorem 5.7 adds equivalences with the local Littlewood–Paley $g$-function and $g^*_\\lambda$-function. When $\\mu(X)<\\infty$, Proposition 6.5 identifies $h^p(X)$ with the Coifman–Weiss atomic Hardy space $H^p_{\\mathrm{CW}}(X)$. When $\\mu(X)=\\infty$, $H^p_{\\mathrm{CW}}(X)$ is a proper subspace, and every normalized test function with nonzero integral lies in $h^p(X)$ but not in $H^p_{\\mathrm{CW}}(X)$. The dual statement, Theorem 7.4, is that for $q\\in(1,\\infty]$ the dual of $h^{p,q}_{\\mathrm{at}}(X)$ is the local Campanato space $c_{1/p-1,q'}(X)$, with the local Lipschitz space $\\ell_{1/p-1}(X)$ in the case $q=1$.","pith_inferences":["My extension: the finite atomic characterization of Section 7 should make endpoint estimates for commutators and singular integrals on local Hardy spaces immediate, since norm control reduces to finite atomic decompositions.","My extension: replacing the $L^p$ target norm by a Musielak–Orlicz or variable-exponent norm in the same exponential-decay argument should produce local versions of those generalized Hardy spaces on spaces of homogeneous type.","My extension: in the finite-measure case, the equality $h^p=H^p_{\\mathrm{CW}}$ suggests that Hardy spaces associated with operators whose heat kernels have good decay should also coincide with these local spaces whenever the operator's Calderón reproducing structure matches the exp-IAI."],"forward_implications":["For every $p$ in the range, a distribution can be certified in whichever local metric is easiest: maximal, atomic, or Littlewood–Paley norms are interchangeable with constants independent of the function.","For $p>1$, the maximal-function local Hardy spaces coincide with $L^p(X)$, so the spaces behave like Hardy spaces only at small scales and like Lebesgue spaces at large scales.","In the finite-measure case, the local Hardy spaces coincide with the Coifman–Weiss Hardy spaces, so radial maximal characterizations of $H^1_{\\mathrm{CW}}(X)$ hold without any additional geometric condition.","In the infinite-measure case, the local Hardy space is strictly larger than the Coifman–Weiss space: nonzero-integral test functions belong to $h^p(X)$ but not to $H^p_{\\mathrm{CW}}(X)$.","The dual spaces are explicitly described by local Campanato norms $c_{1/p-1,q'}$ for $q>1$ and by the local Lipschitz norm $\\ell_{1/p-1}$ for $q=1$, giving a local counterpart of the classical $H^p$–Campanato duality."],"supporting_citations":[{"why":"Builds the orthonormal wavelet system and the exp-IAI with Hölder index $\\eta$, the kernel family used throughout the paper.","marker":"[1]"},{"why":"Defines the Coifman–Weiss atomic Hardy spaces and the atoms that Section 6 compares with the local spaces.","marker":"[11]"},{"why":"Establishes the p=1 local Hardy space and its bmo dual, the endpoint case the paper extends to p<1.","marker":"[12]"},{"why":"Supplies the complete real-variable theory of global Hardy spaces whose methods are adapted here.","marker":"[27]"},{"why":"Provides the inhomogeneous Calderón reproducing formulae with exponential decay that power the equivalences and characterizations.","marker":"[28]"},{"why":"Supplies the dyadic cube system used in the discrete reproducing formula and the local radial maximal function.","marker":"[29]"}],"fun_headline_variants":["Five local Hardy definitions unify on spaces of homogeneous type","No extra geometry: five local Hardy space definitions coincide","Sharp range found: local Hardy space has five equivalent definitions","For p in sharp range, all five local Hardy space definitions match"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that an inhomogeneous approximation of the identity with exponential decay (with Hölder exponent $\\eta$) exists on every space of homogeneous type, and that the associated inhomogeneous Calderón reproducing formulae hold without any reverse-doubling assumption; if this premise gives way, the maximal-function equivalences, the atomic decomposition, and the finite-measure identification with the Coifman–Weiss space would all fail.","fun_headline_variants_meta":{"raw":{"variants":["Five local Hardy definitions unify on spaces of homogeneous type","No extra geometry: five local Hardy space definitions coincide","Sharp range found: local Hardy space has five equivalent definitions","For p in sharp range, all five local Hardy space definitions match"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00135,"raw_usage":{"total_tokens":5578,"prompt_tokens":1138,"completion_tokens":4440,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":754,"completion_tokens_details":{"reasoning_tokens":4372}},"tokens_in":754,"tokens_out":4440,"duration_ms":36749,"temperature":1.0,"reasoning_tokens":4372,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:00:23.329632+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a compact space of homogeneous type with quasi-metric constant $A_0>1$ and no reverse doubling, set $p=1$, and compute the local radial and grand maximal norms of the constant function $f\\equiv 1$ at points where coarse dyadic cubes meet; the paper predicts these norms are finite, comparable, and equal (up to constants independent of the cube partition) to the atomic norm of $f$, which is a bounded multiple of a local atom. If the ratio of the two norms depends on $A_0$ or on the particular dyadic cube system, the claimed equivalences fail.","supporting_citations":[{"cited_title":"Auscher and T","cited_arxiv_id":null,"evidence_quote":"Builds the orthonormal wavelet system and the exp-IAI with Hölder index $\\eta$, the kernel family used throughout the paper."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the Coifman–Weiss atomic Hardy spaces and the atoms that Section 6 compares with the local spaces."},{"cited_title":"Dafni and H","cited_arxiv_id":null,"evidence_quote":"Establishes the p=1 local Hardy space and its bmo dual, the endpoint case the paper extends to p<1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the inhomogeneous Calderón reproducing formulae with exponential decay that power the equivalences and characterizations."},{"cited_title":"Hyt¨ onen and A","cited_arxiv_id":null,"evidence_quote":"Supplies the dyadic cube system used in the discrete reproducing formula and the local radial maximal function."}],"review_version":1}