{"id":"f426d55b-0479-49ef-bbd2-b69d515f4b69","arxiv_id":"1908.04753","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a general class of 3D generalized Radon transforms, the resolution of reconstructed jump discontinuities from discrete data is an explicit edge profile determined by the interpolation kernel, with no non-local artifacts under generic conditions.","lead":"This paper derives an explicit formula for how edge sharpness changes when a 3D generalized Radon transform is inverted from discretely sampled data. The formula ties the reconstructed edge profile to the sampling step size and the interpolation kernel, and it shows distant edges do not create false structures.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The no-artifact part of Theorem 1 rests on GG2, a global tangency-curve condition that is neither shown to be generic nor exercised by the numerical experiment; if Γ_{x0} contains a lattice-aligned straight segment, the averaging argument in Theorem 3 can fail.","rationale":"The concern is not a defect in the proof under Assumption GG2; the argument is careful, and the constants in Sections 5–7 check out, including the factor in the δ(P0) computation and the π from polar integration in (5.18). The issue is that GG2 is a global, non-explicit hypothesis, and the paper's own introduction states that violation of the imposed conditions may cause artifacts and is left for future work. Thus the theorem is honestly conditional. I keep the reader's ACCEPT verdict because acceptance is of a conditional theorem, not of a fully quantified genericity statement. The proposed experiment would determine whether the no-artifact result is merely sufficient or also necessary in a practical sense; if artifacts appear in the GG2-violating example, it confirms the limitation but does not invalidate the theorem as stated.","tokens_in":26349,"tokens_out":23611,"duration_ms":250751,"concrete_test":"Run the Section 8 experiment with a second, remote boundary component engineered so that the tangency curve Γ_{x0} in data space contains a straight segment with rational direction d satisfying m·d=0 for some nonzero m∈Z^3 (say d=(1,1,0)). For ϵ=0.02, 0.01, and 0.005, compute fχϵ(xϵ)−fχ(x0) at a point x0 that sees only the remote component. If the difference does not tend to 0, while the same difference with a slightly perturbed GG2-valid surface does, then the no-artifact conclusion genuinely requires GG2. An analytic counterpart is to check directly that the phase in (7.13) is constant on that segment, so the oscillatory integral equals the segment length rather than o(1).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central formula (3.6) is obtained by adding the local edge response (Theorem 2, Sections 4–6) and the assertion that remote singularities produce no extra contribution (Theorem 3, Section 7). The remote half is the less secure part. Its proof reduces to Lemma 6, and the decisive decay statement (7.13) is exactly Condition GG2 of Definition 1. GG2 is not proved to be generic (no density argument is given), and the numerical experiment in Section 8 uses a single spherical object for which no remote tangency occurs, so Theorem 3 is never tested. If Γ_{x0} contains a segment on which m·\\dotΓ_{x0}≡0 for some nonzero integer vector m — for example a straight segment with rational direction — then for that Fourier mode the phase m·Y(s2)/ϵ in Lemma 6 is constant and the oscillatory integral has size O(1) instead of o(1). Consequently, the limit in (7.15) contains an extra term not present in the continuous reconstruction, so lim fχϵ(x0)=fχ(x0) can fail. Since Theorem 1 discards remote contributions using Theorem 3, the stated universality of the edge profile is conditional on a global geometric assumption whose failure is acknowledged in the introduction but neither excluded for typical surfaces nor checked in the reported experiment.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes the resolution of reconstruction from discrete data for a generalized Radon transform (GRT) in R^3. The reconstruction is obtained by applying the inversion formula (2.14) to interpolated data g_epsilon. The main result, Theorem 1 (eq. (3.6)), gives an explicit limit for f_chi^epsilon at points x0+epsilon x~ within O(epsilon) of a jump surface S: the edge response is f_chi(x0+) minus f0 times an integral of the classical Radon transform of the interpolation kernel. The proof splits into a local part (Theorem 2, Sections 3-6), where the leading singular contribution is computed via a Weyl-type averaging argument, and a remote part (Theorem 3, Section 7), which uses condition GG2 on the tangency curve to show that remote singularities do not contribute. A numerical experiment in Section 8 compares the predicted transition curve with a reconstruction for a ball and reports a good match.","tokens_in":26639,"tokens_out":5317,"duration_ms":49967,"significance":"If Theorem 1 is correct, it provides a parameter-free, explicit prediction for the edge profile in a broad class of tomographic problems, going well beyond the classical Radon transform cases treated in the author's prior work. The derivation is genuinely microlocal: the only ingredients are the defining function assumptions DF1-DF4, the local nondegeneracy LG1-LG2, and the global condition GG2, and no fitted parameters enter the final formula. The numerical experiment supports the local part of the theory for one spherical object. The main caveat is that the global no-artifact theorem rests on a delicate condition GG2 whose genericity is not established and whose failure mode is acknowledged but not analyzed.","major_comments":[{"comment":"The no-artifact part of Theorem 1 depends on Condition GG2, which is used in Section 7 to reduce Lemma 6 to the decay statement (7.13). The paper does not prove that GG2 is generic in any suitable Baire or measure-theoretic sense, despite the abstract and introduction referring to 'generic conditions on S'. If Gamma_{x0} contains a straight segment along which m·dotGamma_{x0} identically vanishes for some nonzero integer vector m, the oscillatory integral in Lemma 6 has size O(1) rather than o(1), and the limit in (7.15) contains an extra term, so the conclusion of Theorem 3 can fail. Because Theorem 1 discards remote contributions using Theorem 3, the universal edge profile is conditional on a global geometric assumption that is neither proved generic nor tested by the numerical experiment in Section 8, which uses a ball and only one tangency pair. Please add a proof of genericity of GG2 (or of a concrete subclass of surfaces that satisfies it) or, alternatively, state the main result as genuinely conditional and add a numerical test with a surface that produces remote tangencies.","section":"Definition 1, Section 7, Lemma 6"},{"comment":"The numerical validation reports a single experiment at epsilon = 0.01 and compares predicted and actual transition curves only visually. Since the central claim is an asymptotic limit as epsilon tends to zero, the experiment should show at least two or three values of epsilon with a quantitative error measure (e.g., sup-norm difference between the predicted profile (3.8) and the reconstruction) to demonstrate the convergence rate O(epsilon^{1/2}) indicated by the error terms in (4.25)-(5.6). Adding a case with a non-spherical surface or with remote tangencies would also exercise the part of the theorem that is currently the least secure.","section":"Section 8"},{"comment":"The derivation of the local edge response contains several estimates that are stated with only a brief justification, for example the claim that the O(epsilon^{1/2}) terms in (4.27) depend smoothly on tilde t and tilde alpha_perp and survive differentiation, and the passage from (5.10) to (5.11) via a Weyl-type argument. These are standard but nontrivial; please expand the justifications or give precise statements of the uniform bounds used, so that the reader can verify that the error terms cannot contribute to the limit in (3.8).","section":"Sections 4-5"}],"minor_comments":[{"comment":"Typo: 'descrete data' should read 'discrete data'.","section":"Abstract"},{"comment":"Typo: 'sometime' should read 'sometimes' in the sentence introducing the coordinates (3.1).","section":"Section 3.1"},{"comment":"The notation g_epsilon(y) is introduced with a sum over j in r+Z^3, but the index set is not restated when the sum is used later (e.g., in (4.24) and (7.2)); adding the index set explicitly at each occurrence would improve readability.","section":"Section 2, eq. (2.13)"},{"comment":"The cubic B-spline notation B_n is standard but not defined; please state that B_n is the cardinal B-spline of degree n supported on [0,n+1].","section":"Section 8, eq. (8.1)"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically strong and the local theory seems sound, but the global claim in Theorem 1 relies on Condition GG2, which is not shown to be generic and is not exercised numerically. This is the main reason for major revision rather than accept. The author may find it sufficient to (a) prove a genericity statement for GG2 under natural hypotheses on the surface and the GRT, or (b) rephrase the abstract and introduction so that the result is explicitly conditional on a geometric hypothesis that is verified in the experiment. The manuscript is otherwise appropriate for the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: this is a real result, not a repackaging. The explicit edge-response formula (3.6) for a general family of surfaces is new, and the proof is detailed and parameter-free. The main soft spot is the global no-artifact half: Theorem 3 depends on GG2, a tangency-curve condition that is neither shown to be generic nor tested numerically.\n\nWhat's new and good: The paper derives the leading singular behavior of the discrete reconstruction in an O(epsilon) neighborhood of a jump, for surfaces satisfying Bolker-type conditions. The connection to uniform distribution theory is apt and gives a clean handle on the lattice sums. The local analysis (Sections 4–6), culminating in Theorem 2, is carefully done, and the key asymptotic (3.6) follows from stated assumptions rather than being fit to data. The numerical experiment with spherical tangents is honest and matches the predicted transition curve.\n\nWhere it's soft: The remote-singularity part (Theorem 3, Section 7) is the weaker load-bearing wall. The decisive decay statement (7.13) is exactly GG2. GG2 requires that for every nonzero integer vector m, the set of parameters on Γ_x0 where |m·\\dotΓ_x0| is small has total length tending to zero. If Γ_x0 contains a straight segment whose direction is rational, then for some m we have m·\\dotΓ ≡ 0 on that segment, Lemma 6's oscillatory integral does not decay, and the limit in (7.15) can contain extra terms. The stress-test concern is real and it lands. The paper explicitly acknowledges that violation of GG2 may produce artifacts, and the numerical experiment uses a single sphere with no remote tangency, so Theorem 3 is never exercised. This doesn't sink the paper, but it means the universality of the edge profile in Theorem 1 is conditional on a geometric assumption whose genericity is not established. A density argument or an example satisfying GG2 would strengthen the claim considerably.\n\nWho it's for: anyone working on resolution analysis of tomographic reconstructions from discrete data, especially in ultrasound, photoacoustic, or Compton camera imaging. The local formula is the main useful takeaway; the global part is a useful warning about when artifacts may appear.\n\nRecommendation: deserves a serious referee. I would accept with minor revision: ask for a discussion or proof of how restrictive GG2 is, and ideally a numerical test with a configuration that actually produces remote tangencies. The core result is sound.","headline":"A serious, honest extension of Katsevich's CRT resolution program to generalized Radon transforms, with a clean local edge-response formula and a global no-artifact theorem that rests on a delicate, untested tangency-curve condition.","tokens_in":27109,"tokens_out":3034,"would_cite":true,"duration_ms":30049,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["44A12","65R10","35S30","65D05"],"pacs":[],"model":"deepseek-v4-flash","headline":"Discretely sampled generalized Radon transforms reconstruct 3D jump surfaces with a universal edge profile determined by the interpolation kernel and the sampling grid.","keywords":["generalized Radon transform","resolution analysis","edge response","interpolation kernel","discrete data","singularity reconstruction","uniform distribution","tomographic inversion"],"falsifier":"Take an object whose boundary is a cylinder with axis parallel to a data-grid direction, so that the tangency curve $\\Gamma_{x_0}$ contains a straight segment aligned with the lattice, reconstruct with the same interpolation kernel and inversion formula, and check whether the value at a point $x_0$ off the surface converges to $f_\\chi(x_0)$ or acquires an $O(1)$ artifact. A nonzero artifact would show that condition GG2 is doing real work; convergence would suggest the theorem holds under weaker conditions.","tokens_in":26174,"feed_emoji":"📐","tokens_out":9787,"duration_ms":90654,"temperature":0.7,"pith_summary":"The paper asks what happens when a three-dimensional image is reconstructed from generalized Radon data known only on a fine regular grid of step $O(\\epsilon)$. For any modality captured by the generalized Radon transform—spherical, elliptical, or conical integration surfaces—the author proves that near a generic point where the object has a jump, the reconstruction approaches a fixed edge profile whose shape depends only on the interpolation kernel and the local sampling geometry. The profile is the cumulative distribution of the Radon transform of that kernel along the data-space normal direction, and it is independent of the rest of the object surface. A second theorem shows that, under a delicate genericity condition on the tangency curve, distant parts of the surface produce no non-local artifacts in the limit. The result makes the resolution of discrete-data inversion quantitative for a broad class of integral transforms.","feed_headline":"3D scan jump edges follow a profile set by the grid","feed_subtitle":"Discrete Radon inversion smears every jump identically, with the shape fixed by the interpolation kernel.","key_machinery":"The load-bearing object is the identity $\\lim_{\\epsilon\\to 0} f_\\chi^\\epsilon(x_\\epsilon) = f_\\chi(x_0-) + f_0 \\int_{-\\infty}^{\\nu h} \\hat\\phi(\\beta_0,s)\\,ds$. The derivation tracks the singular part of the data near the tangency parameter $y_0$ using a known asymptotic for the classical Radon transform, expands the interpolated data through the chain rule of the inversion formula, and then applies a Weyl-type equidistribution argument over the lattice to replace the discrete sum by an integral over $[0,1]^3$. The exactness of $\\phi$ through degree 2 (condition IK1) is what lets all subleading terms drop out, and the non-degeneracy conditions DF1–DF4 (essentially the Bolker condition) make the local coordinates well defined. Condition LG2, which requires $\\lambda\\Phi'_y\\notin\\mathbb Z^3$, is what makes the nonzero Fourier modes cancel in the local limit.","core_discovery":"The paper establishes that if the data $g(y)=(\\mathcal R f)(y)$ are sampled on the lattice $r+\\epsilon\\mathbb Z^3$ and interpolated with a kernel $\\phi$ that is exact through degree 2, then for a generic tangency pair $(x_0,y_0)$ the reconstruction $f_\\chi^\\epsilon$ evaluated at $x_\\epsilon=x_0+\\epsilon\\tilde x$ obeys $\\lim_{\\epsilon\\to 0} f_\\chi^\\epsilon(x_\\epsilon) = f_\\chi(x_0+) - f_0 \\int_{\\nu h}^{\\infty} \\hat\\phi(\\beta_0,s)\\,ds$, or equivalently $f_\\chi(x_0-) + f_0\\int_{-\\infty}^{\\nu h}\\hat\\phi(\\beta_0,s)\\,ds$, where $h=\\tilde x\\cdot\\alpha_0$, $\\nu=|\\Phi'_x|/|\\Phi'_y|$, $\\beta_0=\\Phi'_y/|\\Phi'_y|$, and $\\hat\\phi$ is the classical Radon transform of the kernel. This is the limiting edge response: the jump is smeared into a ramp-like transition of width $O(\\epsilon)$ whose shape is universal. The proof has a local part, which computes the leading singular contribution by a Taylor expansion and an equidistribution step, and a global part, which shows under global genericity that remote singularities contribute nothing. A numerical experiment with spheres tangent to a plane reproduces the predicted transition curve.","pith_inferences":["The paper does not pursue calibration, but the universal profile suggests a protocol: measure the edge response once on a known phantom with a fixed sampling geometry, then use the inferred kernel Radon transform to deconvolve or sharpen edges in any GRT modality that shares that geometry.","If LG2 fails (that is, if $\\lambda\\Phi'_y$ lands on the integer lattice), the equidistribution step in (5.14) breaks down, so resonant sampling directions should produce extra oscillatory or aliased structure in the edge profile; this is a testable prediction beyond the paper's generic-position assumption.","The GG2 condition singles out ruled or cylindrical surfaces as the natural place to look for non-local artifacts, since their tangency curves can contain straight segments parallel to the grid; one could probe this by aligning a cylinder with the grid and comparing against a sphere with the same sampling.","The same uniform-distribution mechanism might extend to other Fourier integral operators beyond the adjoint-based inversion studied here, with the Weyl step being the part that needs the most adaptation."],"forward_implications":["Resolution near a jump is anisotropic: the transition scale along the normal is multiplied by $\\nu=|\\Phi'_x|/|\\Phi'_y|$, so the same object edge is reconstructed with different sharpness depending on local data geometry.","The interpolation kernel entirely determines the limiting edge profile; a kernel whose Radon transform is closer to a step function yields a sharper transition, and any kernel satisfying the exactness assumptions gives the same $O(\\epsilon)$ profile shape up to that Radon transform.","Object curvature and global surface shape do not affect the leading edge response at a generic point; they enter only through lower-order corrections.","Under global genericity, reconstruction away from the surface is artifact-free to leading order, so local inversion methods need not fear non-local contamination from distant jumps.","The formula gives a quantitative prediction that can be checked in any modality fitting the GRT assumptions, such as photoacoustic or ultrasound tomography with spherical integration surfaces."],"supporting_citations":[{"why":"defines the generalized Radon transform inversion and the FIO conditions that frame the whole analysis.","marker":"[2]"},{"why":"supplies the explicit piecewise-polynomial interpolation kernel used in the numerical experiment.","marker":"[3]"},{"why":"establishes the edge-response approach for the classical Radon transform that this paper extends to the GRT.","marker":"[10, 11]"},{"why":"provides the uniform-distribution theorem behind the Weyl-type step that turns lattice sums into integrals.","marker":"[13]"},{"why":"gives the asymptotic form of the transform near a tangency point that Lemma 2 generalizes to the GRT.","marker":"[16, 17]"},{"why":"supplies the Fourier-integral-operator calculus and Bolker-condition background for the reconstruction formula.","marker":"[20]"}],"fun_headline_variants":["3D scan jumps blur to a universal kernel shape","Grid spacing sets edge smear in 3D Radon inversion","Discrete Radon data yield kernel-defined edge response","In 3D, jump resolution is fixed by the grid step","Universal edge profile in 3D Radon reconstruction"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of the no-artifact theorem requires that the curve of data parameters where a measuring surface is tangent to the object surface never contains a straight segment parallel to a lattice direction; if it does, the proof that distant singularities produce no artifacts collapses.","fun_headline_variants_meta":{"raw":{"variants":["3D scan jumps blur to a universal kernel shape","Grid spacing sets edge smear in 3D Radon inversion","Discrete Radon data yield kernel-defined edge response","In 3D, jump resolution is fixed by the grid step","Universal edge profile in 3D Radon reconstruction"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0003,"raw_usage":{"total_tokens":1843,"prompt_tokens":1169,"completion_tokens":674,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":785,"completion_tokens_details":{"reasoning_tokens":593}},"tokens_in":785,"tokens_out":674,"duration_ms":7242,"temperature":1.0,"reasoning_tokens":593,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:33:16.094692+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take an object whose boundary is a cylinder with axis parallel to a data-grid direction, so that the tangency curve $\\Gamma_{x_0}$ contains a straight segment aligned with the lattice, reconstruct with the same interpolation kernel and inversion formula, and check whether the value at a point $x_0$ off the surface converges to $f_\\chi(x_0)$ or acquires an $O(1)$ artifact. A nonzero artifact would show that condition GG2 is doing real work; convergence would suggest the theorem holds under weaker conditions.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"defines the generalized Radon transform inversion and the FIO conditions that frame the whole analysis."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the explicit piecewise-polynomial interpolation kernel used in the numerical experiment."},{"cited_title":"Kuipers and H","cited_arxiv_id":null,"evidence_quote":"provides the uniform-distribution theorem behind the Weyl-type step that turns lattice sums into integrals."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the Fourier-integral-operator calculus and Bolker-condition background for the reconstruction formula."}],"review_version":1}