{"id":"7d86c7a0-b50e-4828-92f6-f095d7161f44","arxiv_id":"1909.00381","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"In RCD(K,N) spaces, the reduced boundary of a set of finite perimeter has a unique Euclidean half-space tangent at almost every point and is rectifiable by bi-Lipschitz charts.","lead":"Boundaries of finite-perimeter sets inside very general curved spaces with lower Ricci bounds are shown to be almost everywhere flat-like and coverable by finitely many distorted coordinate patches. This closes a program begun in the authors' earlier work and gives new tools, including a Gauss-Green formula, that apply even to Ricci limit spaces.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Main theorems are structurally coherent; the least externally secure premise is the capacitary calculus of [22], used as a black box for the boundary tangent module.","rationale":"I read the main line of the paper as follows: Theorem 3.2 proves uniqueness of tangents by combining the existence of a Euclidean half-space tangent (Theorem 3.3) with harmonic δ-splitting maps and a weighted maximal-function propagation (Propositions 3.7–3.11); Theorem 4.1 then uses the boundary normal ν_E, the δ-orthogonality condition, and a quantitative isometry argument (Propositions 4.5 and 4.7) to produce bi-Lipschitz charts. I checked the scale-invariance of the bad-set estimates in Proposition 3.11, the use of H^{h2}-null sets to deduce Per-null sets via Lemma 1.10, and the alignment argument in Lemma 3.14; these steps are internally consistent. The proof of Theorem 1.12 that H^{hα}≪Cap for α<2 is also coherent. Thus I do not find a demonstrable flaw in the central rectifiability argument. The reader's formal weakest assumption was the codimension-one control of perimeter by capacity; I agree that this is foundational, but it is actually proved in the paper. The more precise soft spot is the external black-box reliance on [22] for the capacitary tangent module and the quasi-continuous representatives of vector fields. This does not change the conditional verdict: the paper should be accepted only after that dependence is checked against [22] and the sketch of Corollary 3.15 is expanded or explicitly deferred. Since the reader already made the verdict conditional for closely related reasons, I recommend no change.","tokens_in":47760,"tokens_out":25348,"duration_ms":232986,"concrete_test":"Independently verify that the standing assumptions of [22, Theorems 1.20, 2.14 and Proposition 2.8] are satisfied by every RCD(K,N) space as used in this paper, namely completeness, separability, local finiteness of m, density of continuous functions in H^{1,2}(X), and the existence of the quasi-continuous representative for every element of H^{1,2}_C(TX)∩L∞(TX). Then reproduce the trace identity (2.3)–(2.4) of Lemma 2.6 using only those stated results; if any step requires an extra hypothesis, such as global finiteness of m or a non-local Poincaré inequality, the construction of the boundary module and of the Gauss–Green formula collapses.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central theorems 3.2 and 4.1 rest on the construction of the boundary tangent module L^2_E(TX) (Theorem 2.1) and the Gauss–Green formula (Theorem 2.2). These are obtained by projecting the capacitary tangent module L^0_Cap(TX) of [22] onto the perimeter measure, using that Per(E,·) is absolutely continuous with respect to Cap. The absolute continuity itself (Lemma 1.10 plus Theorem 1.12) is proved in the paper and is not the fragile part. The fragile part is that the quoted results from [22] — in particular the existence of the capacitary tangent module (Theorem 1.35), the quasi-continuous representative for Sobolev vector fields (Theorem 2.14 of [22]), and the projection Lemma 1.36 — are used as a black box. If those results require hypotheses beyond those explicitly recalled here (e.g., σ-finiteness, local compactness, or a global density property that is not available when projecting onto the boundary measure), then the trace operator tr_E and the exterior normal ν_E of Theorem 2.2 may fail to exist, and the orthogonality propagation in Propositions 4.5 and 4.7 loses its foundation. Since [22] is described as very recent and shares an author with the present paper, this is the least independently verified load-bearing premise. No internal inconsistency in the main argument was found.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves uniqueness of tangents and rectifiability for the reduced boundary of sets of finite perimeter in RCD(K,N) metric measure spaces. Starting from the existence of Euclidean half-space tangents established in [2], the authors develop a Gauss–Green integration-by-parts formula using a new tangent module over the boundary, built by projecting the capacitary tangent module of [22] onto the perimeter measure. A weighted maximal-function argument with harmonic δ-splitting maps then propagates regularity to yield, for |DχE|-almost every point, a unique tangent equal to a Euclidean half-space (Theorem 3.2), the (|DχE|,k−1)-rectifiability of the strata F_k^E (Theorem 4.1), and a representation of the perimeter measure in terms of codimension-one Hausdorff measure (Corollary 3.15).","tokens_in":47927,"tokens_out":26452,"duration_ms":228461,"significance":"These are major results. Theorems 3.2 and 4.1 resolve open questions from [2] and provide the first De Giorgi-type rectifiability theorem for sets of finite perimeter in full RCD(K,N) generality, with the non-collapsed case stated in Corollary 4.2. The novelty is significant even for Ricci limit spaces. The paper is technically substantial: it constructs a boundary tangent module, proves a Gauss–Green formula, and adapts δ-splitting techniques to codimension-one objects. The proofs are detailed and, apart from the issues below, internally coherent. The result is a clear advance in the structure theory of RCD spaces, and the perimeter-measure representation in Corollary 3.15 is a strong additional payoff.","major_comments":[{"comment":"The application of Corollary 3.12 in the proof of the claim is not justified. Corollary 3.12 states that a δ-splitting map on B_{4r}(p) yields a good set G contained in B_{2r}(p) with the splitting property only for scales s<r. In this proof, the map u_i is defined on B_{5r_i}(x_i); taking r=5r_i/4, the corollary gives G_i⊂B_{2.5r_i}(x_i), not G_i⊂B_{5r_i}(x_i). The text instead asserts G_i⊂B_{5r_i}(x_i) with the estimate H^h_5(B_{5r_i}(x_i)\\G_i)≤ C_N√δ m(B_{5r_i}(x_i))/(5r_i), which does not follow. The annulus B_{5r_i}(x_i)\\B_{2.5r_i}(x_i) can carry H^h_5-measure comparable to Per(E,B_{r_i}(x_i)) per ball, and summing over the Vitali family gives a constant multiple of Per(E,B_2(p)) that is not small. Since (3.7) is the step needed to obtain Per(E,A_k\\G_η)=0 via Lemma 1.10, this is load-bearing. The proof should be repaired, for instance by choosing the initial covering balls so that the good set obtained from Corollary 3.12 actually controls the H^h_5-measure of the bad part of each covering ball.","section":"§2, Theorems 2.1–2.2"},{"comment":"The construction of the boundary tangent module L^2_E(TX) and the Gauss–Green formula depend on two black-box results from [22]: the existence of the capacitary tangent module (Theorem 1.35 in the present paper) and the quasi-continuous representative theorem for Sobolev vector fields ([22, Theorem 2.14]). These results are taken from a submitted preprint that shares an author with the present paper, and they are essential for the trace operator tr_E and the exterior normal ν_E. The manuscript should state the precise hypotheses under which these results hold and verify that an RCD(K,N) space as used here satisfies them; in particular the projection in Lemma 1.36 requires a finite Borel measure μ≪Cap, which is provided by Lemma 1.10 and Theorem 1.12, but the hypotheses for [22, Theorem 2.14] (σ-finiteness, local compactness, or other global properties) are not checked. Without this, the existence of ν_E and the subsequent orthogonality propagation in Propositions 4.5 and 4.7 lack a verified foundation.","section":"§3.2"}],"minor_comments":[{"comment":"The deduction of lim_{r→0} r^2 ⨏_{B_r(x)} |Hess φ|^2 dm = 0 for |Dχ_E|-a.e. x should explicitly apply Lemma 1.11 with α=1, not α=2. The inclusion {limsup r^2(f)_{x,r}>0} ⊂ {limsup r(f)_{x,r}>0} makes the argument correct after combining Lemma 1.11(α=1) with Lemma 1.10, but the text as written is ambiguous.","section":"§4.1"},{"comment":"In the displayed equation before (4.24), the convergence of (X_n,d_n,m_n,x_n) to (R^k, d_eucl, (1/ω_k)L^k, 0) follows from condition (i) with r=1/2 and |K|≤4; for clarity, it would help to state explicitly that the rescaling is the one given in condition (i), since the proof later uses both the normalized measure and the c_k-normalization of Definition 1.15.","section":"§4.2"}],"recommendation":"major_revision","confidential_remarks":"The main theorems are significant and the overall structure appears sound, but the proof of Theorem 3.2 contains a concrete scaling misapplication of Corollary 3.12 that affects the key smallness estimate (3.7). Also, the heavy reliance on the submitted preprint [22] should be made more explicit and self-contained. Given the importance of the results, I recommend major revision rather than rejection; these issues seem to be repairable within the manuscript's scope. I note that the paper's dependence on [2] and [22] includes results of the same authors, but the theorems are used as stated and no circularity is apparent."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is the paper that completes the De Giorgi program for sets of finite perimeter over RCD(K,N) spaces: uniqueness of Euclidean half-space tangents for the reduced boundary, rectifiability of the strata F^k_E, and a representation of the perimeter measure as a codimension-one Hausdorff measure. It genuinely removes the non-collapsed restriction from Ambrosio–Bruè–Semola [2], and the result is new for collapsed Ricci limits. The Gauss–Green formula (Theorem 2.2) with a boundary tangent module and a unit normal is substantial new technology, and the δ-splitting propagation with δ-orthogonality to the normal is well executed.\n\nThe main theorems are proved in detail. The compactness, splitting, and propagation arguments in Sections 3 and 4 are coherent; I found no internal inconsistency. The paper is also honest about its assumptions: it proves the needed absolute continuity Per(E,·) ≪ Cap itself (Lemma 1.10 and Theorem 1.12), so that particular premise is not hidden.\n\nSoft spots, in proportion. The boundary tangent module is built by projecting the capacitary tangent module of [22] onto the perimeter measure. Existence of that module, quasi-continuous representatives for Sobolev vector fields, and the projection lemma are quoted as black boxes. [22] is very recent and shares an author; this is a genuine dependency, though not a circular one, since [22] contains neither uniqueness nor rectifiability. A referee should ask the authors to state the precise hypotheses needed and verify that they are met when projecting onto a measure supported on a codimension-one set. Corollary 3.15 is left as a sketch, delegating a density argument to the non-collapsed case; that is a smaller issue, but it should be expanded for a journal version. The proof of Proposition 4.5 Step 3 refers to 'a slight modification of Proposition 3.9' to obtain good approximations; that is plausible but worth spelling out.\n\nThe authors themselves note the strategy would fail if perimeter were codimension at least two; that is an honest limitation, not a flaw. If the [22] black box holds up, the main theorems stand.\n\nThis deserves a serious referee. I would send it out with a request to expand Corollary 3.15 and to clarify the exact statements from [22] being used. Conditional accept, with the expectation that these points can be handled in revision.","headline":"This paper closes the De Giorgi program for sets of finite perimeter on RCD(K,N) spaces: uniqueness of Euclidean half-space tangents and rectifiability of the reduced boundary, removing the non-collapsed restriction and covering collapsed Ricci limits.","tokens_in":48550,"tokens_out":1663,"would_cite":true,"duration_ms":17740,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["26B30","26B20","53C23"],"pacs":[],"model":"deepseek-v4-flash","headline":"Reduced boundaries in RCD spaces are rectifiable: finite-perimeter sets have unique Euclidean half-space blow-ups and are covered by countably many bi-Lipschitz pieces.","keywords":["sets of finite perimeter","reduced boundary","rectifiability","RCD metric measure spaces","Gauss-Green formula","harmonic splitting maps","tangent cones","De Giorgi theorem"],"falsifier":"A concrete way to falsify the central claim is to find an RCD(K,N) space and a finite-perimeter set E with a point x in a stratum $F^k_E$ where the density ratio $r|D\\chi_E|(B_r(x))/m(B_r(x))$ does not converge to $\\omega_{k-1}/\\omega_k$ as $r\\to 0$, since Corollary 3.15 forces this limit at every point of the reduced boundary. Equally, exhibiting a point with two distinct half-space tangents of different dimensions on a set of positive perimeter would contradict Theorem 3.2.","tokens_in":47470,"feed_emoji":"📐","tokens_out":9153,"duration_ms":98875,"temperature":0.7,"pith_summary":"This paper proves that De Giorgi's theorem—the classical structure result for sets of finite perimeter—holds for RCD(K,N) metric measure spaces, the synthetic setting of spaces with Ricci curvature bounded below and dimension bounded above. The authors show that at almost every point of the reduced boundary of a set of finite perimeter there is one and only one tangent object: a Euclidean half-space of dimension k, for some k between 1 and the essential dimension. They then show that the stratum where this k occurs is rectifiable, meaning it is covered up to perimeter-negligible sets by countably many bi-Lipschitz images of $R^{{k-1}}$, with the perimeter measure represented by the codimension-one Hausdorff measure. These conclusions are new even for Ricci limit spaces and provide the missing boundary counterpart of the known structure theory for the ambient spaces.","feed_headline":"Reduced boundaries in RCD spaces are rectifiable","feed_subtitle":"Finite-perimeter sets have a unique half-space blow-up at almost every boundary point, completing De Giorgi's theorem.","key_machinery":"The load-bearing tool is a Gauss–Green integration-by-parts formula on RCD spaces (Theorem 2.2). Because the perimeter measure is absolutely continuous with respect to the 2-capacity, the paper can construct a capacitary tangent module over the boundary of E, define a trace of Sobolev vector fields there, and obtain a unit exterior normal $\\nu_E$ satisfying $\\int_E \\mathrm{div}(v)\\, dm = -\\int \\langle \\mathrm{tr}_E(v), \\nu_E\\rangle\\, d|D\\chi_E|$. This normal is then combined with harmonic $\\delta$-splitting maps—harmonic maps whose gradients form an almost orthonormal frame with small Hessian—to detect the Euclidean coordinate directions. A weighted maximal-function argument propagates the $\\delta$-splitting and its $\\delta$-orthogonality to the normal from one scale to all scales outside a set of small codimension-one Hausdorff content, which is what yields uniqueness of tangents and then bi-Lipschitz rectifiability.","core_discovery":"Let (X,d,m) be an RCD(K,N) space with essential dimension n and E a set of finite perimeter. The paper establishes that for $|D\\chi_E|$-almost every $x$ there is $k\\in\\{1,\\ldots,n\\}$ such that the tangent collection reduces to the single Euclidean half-space $(\\mathbb{R}^k, d_{\\mathrm{eucl}}, c_k \\mathcal{L}^k, 0_k, \\{x_k>0\\})$. Defining the reduced boundary stratum $F^k_E$ as the points with this unique tangent, the paper proves each $F^k_E$ is $(|D\\chi_E|, k-1)$-rectifiable; equivalently, up to $|D\\chi_E|$-negligible sets, $F^k_E$ is covered by countably many bi-Lipschitz images of subsets of $\\mathbb{R}^{k-1}$. A companion representation formula gives $|D\\chi_E| = \\sum_{k=1}^n \\frac{\\omega_{k-1}}{\\omega_k} \\mathcal{H}^{h}|_{F^k_E}$. In the non-collapsed case this reduces to the classical statement that the reduced boundary is $(N-1)$-rectifiable and its perimeter measure equals $\\mathcal{H}^{N-1}$ restricted to it.","pith_inferences":["[Inference] The same proof structure should work on any PI space where perimeter is controlled by codimension-one Hausdorff measure and harmonic functions satisfy L^2 Hessian estimates; RCD spaces are one natural setting, not the only one.","[Inference] The quantitative nature of the maximal-function argument suggests an $\\varepsilon$-regularity statement: if a finite-perimeter set is $\\delta$-close to a half-space at one scale and the splitting map is $\\delta$-orthogonal to the normal, then it is bi-Lipschitz to a hypersurface at all smaller scales with explicit constants—testable in concrete examples.","[Inference] Since the paper does not use constancy of dimension, a sharper version might show each stratum $F^k_E$ has a single dimension k and the sum in the representation formula collapses to one codimension-one measure; this is a natural next step.","[Inference] The authors' remark that the strategy fails for perimeter of codimension at least two suggests that extending these results to higher-codimension objects (e.g. minimal surfaces or clusters in RCD spaces) requires a new mechanism, not just a re-run of the same argument."],"forward_implications":["At almost every boundary point, the blow-up of a finite-perimeter set is unique: a Euclidean half-space of some dimension k, never a product with a nontrivial factor.","The reduced boundary stratum $F^k_E$ is countably covered by bi-Lipschitz images of subsets of $\\mathbb{R}^{k-1}$, so the classical De Giorgi rectifiability theorem holds in RCD spaces.","The perimeter measure is a weighted codimension-one Hausdorff measure on the reduced boundary, with density $\\omega_{k-1}/\\omega_k$ on $F^k_E$.","In non-collapsed RCD spaces, the reduced boundary is $(N-1)$-rectifiable and the perimeter measure equals $\\mathcal{H}^{N-1}$ restricted to it, matching the Euclidean statement.","Taking a product with a Euclidean line and the half-space $\\{t>0\\}$, rectifiability of the reduced boundary implies rectifiability of the ambient RCD space itself."],"supporting_citations":[{"why":"Establishes the starting point: every finite-perimeter set has a Euclidean half-space tangent at $|D\\chi_E|$-almost every point, and provides compactness and weak-convergence tools for perimeter measures.","marker":"[2]"},{"why":"Constructs the capacitary tangent module and quasi-continuous representatives for Sobolev vector fields, used to define traces and the tangent module over the boundary.","marker":"[22]"},{"why":"Proves $\\mathrm{Per}(E,\\cdot)\\ll \\mathcal{H}^{h^1}$ on PI spaces, the codimension-one control needed for the whole strategy.","marker":"[1]"},{"why":"Provides the structure-theory template: uniqueness of tangents through propagation of Euclidean splitting by maximal-function arguments in RCD spaces.","marker":"[42]"},{"why":"Introduces harmonic $\\delta$-splitting maps used as approximate coordinates in the blow-up analysis.","marker":"[15]"},{"why":"Supplies the weighted maximal-function propagation idea for $\\delta$-splitting and codimension-one regularity.","marker":"[20]"},{"why":"Gives stability and convergence results for Sobolev functions, Laplacians and Hodge Laplacians under pointed measured Gromov-Hausdorff convergence.","marker":"[8]"},{"why":"Develops the normed-module calculus (tangent modules, Hessian, divergence) on which the Gauss-Green formula is built.","marker":"[27]"}],"fun_headline_variants":["De Giorgi rectifiability extended to RCD spaces","Finite-perimeter sets have rectifiable reduced boundary","Unique tangents imply rectifiable reduced boundary","Gauss-Green formula enables boundary rectifiability","Rectifiability of reduced boundary proven in RCD"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the perimeter measure being a codimension-one object: $|D\\chi_E|$ is absolutely continuous with respect to the codimension-one Hausdorff measure, and hence with respect to the 2-capacity; if perimeter could concentrate in codimension two or higher, the trace, the normal, and the scale propagation would all break down.","fun_headline_variants_meta":{"raw":{"variants":["De Giorgi rectifiability extended to RCD spaces","Finite-perimeter sets have rectifiable reduced boundary","Unique tangents imply rectifiable reduced boundary","Gauss-Green formula enables boundary rectifiability","Rectifiability of reduced boundary proven in RCD"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000814,"raw_usage":{"total_tokens":3550,"prompt_tokens":908,"completion_tokens":2642,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":524,"completion_tokens_details":{"reasoning_tokens":2581}},"tokens_in":524,"tokens_out":2642,"duration_ms":17208,"temperature":1.0,"reasoning_tokens":2581,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:54:49.070016+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete way to falsify the central claim is to find an RCD(K,N) space and a finite-perimeter set E with a point x in a stratum $F^k_E$ where the density ratio $r|D\\chi_E|(B_r(x))/m(B_r(x))$ does not converge to $\\omega_{k-1}/\\omega_k$ as $r\\to 0$, since Corollary 3.15 forces this limit at every point of the reduced boundary. Equally, exhibiting a point with two distinct half-space tangents of different dimensions on a set of positive perimeter would contradict Theorem 3.2.","supporting_citations":[{"cited_title":"Quasi-continuous vector fields on RCD spaces","cited_arxiv_id":"1903.04302","evidence_quote":"Constructs the capacitary tangent module and quasi-continuous representatives for Sobolev vector fields, used to define traces and the tangent module over the boundary."},{"cited_title":"Ambrosio , Fine properties of sets of finite perimeter in doubling metri c measure spaces, Set-Valued Anal., 10 (2002), pp","cited_arxiv_id":null,"evidence_quote":"Proves $\\mathrm{Per}(E,\\cdot)\\ll \\mathcal{H}^{h^1}$ on PI spaces, the codimension-one control needed for the whole strategy."},{"cited_title":"Cheeger and T","cited_arxiv_id":null,"evidence_quote":"Introduces harmonic $\\delta$-splitting maps used as approximate coordinates in the blow-up analysis."},{"cited_title":"Cheeger and A","cited_arxiv_id":null,"evidence_quote":"Supplies the weighted maximal-function propagation idea for $\\delta$-splitting and codimension-one regularity."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives stability and convergence results for Sobolev functions, Laplacians and Hodge Laplacians under pointed measured Gromov-Hausdorff convergence."},{"cited_title":"Gigli , Nonsmooth differential geometry - an approach tailored for s paces with Ricci curvature bounded from below","cited_arxiv_id":null,"evidence_quote":"Develops the normed-module calculus (tangent modules, Hessian, divergence) on which the Gauss-Green formula is built."}],"review_version":1}