Pith. sign in

REVIEW 3 major objections 4 minor 23 references

Resolution analysis of inverting the generalized Radon transform from discrete data in $\mathbb R^3$

T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Discretely sampled generalized Radon transforms reconstruct 3D jump surfaces with a universal edge profile determined by the interpolation kernel and the sampling grid.

desk verdict 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. read the letter →

arxiv 1908.04753 v2 pith:5XVL24FK submitted 2019-08-13 math.NA cs.NA

Alexander Katsevich This is my paper · ORCID
classification math.NAcs.NA MSC 44A1265R1035S3065D05
keywords generalizedRadontransformresolutionanalysisedgeresponseinterpolationkerneldiscretedatasingularityreconstructionuniformdistributiontomographicinversion
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

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.

What carries the argument

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.

What would settle it

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.

Watch

Extended reading notes

Core claim

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.

Load-bearing premise

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.

Editorial extensions

If this is right

  • 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.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • 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.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

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.

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 (3)
  1. [Definition 1, Section 7, Lemma 6] 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.
  2. [Section 8] 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.
  3. [Sections 4-5] 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).
minor comments (4)
  1. [Abstract] Typo: 'descrete data' should read 'discrete data'.
  2. [Section 3.1] Typo: 'sometime' should read 'sometimes' in the sentence introducing the coordinates (3.1).
  3. [Section 2, eq. (2.13)] 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.
  4. [Section 8, eq. (8.1)] 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].

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the edge response is a parameter-free asymptotic derivation from stated microlocal assumptions, and the numerical comparison is not a fit.

full rationale

The central claim, Theorem 1 (equation 3.6), is derived from a genuine asymptotic analysis rather than from the definition of the reconstruction or from fitted parameters. Lemma 2 obtains the local singular behavior of g(y) via the Morse lemma and an explicit constant G(z,0); Sections 4 and 5 compute the leading contribution f^(1)_chi_epsilon using the interpolation properties IK1–IK5 and a Weyl-type averaging argument; Section 6 shows the f^(2) term converges to the continuous value; Section 7 proves that remote singularities do not contribute under the explicitly stated global genericity conditions GG1 and GG2, using Lemma 6. No step reduces equation (3.6) to the input by construction: the interpolation kernel phi is specified in advance, not fitted to the reconstruction, and the predicted curve is compared with an independently computed GRT of a ball. The paper's self-citations to [10,11] supply the general approach and an explicit kernel formula (8.1), but the theorem does not depend on an unverified self-cited result. The limitation passage in Section 1, 'It is possible that violation of the imposed conditions leads to artifacts,' is an acknowledged robustness caveat about GG2, not a circularity; it affects generality but not the derivation's independence from its inputs.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No free parameters are fitted; the edge response formula depends only on the interpolation kernel and the GRT geometry. No new entities are introduced.

assumptions (5)
  • domain assumption DF1-DF4: Phi is non-degenerate, data complete (DF2), no conjugate points (DF3), and the mixed Hessian is non-degenerate (DF4, local Bolker condition).
    These conditions define the class of GRTs for which R is an FIO with local canonical graph and the inversion formula (2.14) is a parametrix. They are standard in GRT inversion (see Beylkin [2]).
  • domain assumption IK1-IK5: interpolation kernel phi is exact up to order 2, compactly supported, C^2 with piecewise continuous bounded third derivatives, and normalized.
    These properties make g_epsilon a good approximation to g and allow the Weyl-type argument in Section 5. The kernel's classical Radon transform enters the final edge profile.
  • domain assumption Generic conditions LG1, LG2, GG1, GG2 on the singularity surface S and the tangency curve.
    LG2 ensures the normal direction is not lattice-resonant, while GG2 controls the geometry of the tangency curve and is needed to rule out non-local artifacts in Section 7.
  • domain assumption f is a finite sum of smooth functions times characteristic functions of bounded domains with piecewise smooth boundaries, as in (2.6).
    Defines the class of piecewise smooth objects with jump discontinuities that the theorem analyzes.
  • standard math Standard microlocal analysis: R and R* are FIOs, and their composition is a pseudodifferential operator.
    Used to justify the inversion formula and the reduction of the reconstruction operator to a parametrix. This is standard theory from Hormander and Treves.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Resolution analysis of inverting the generalized Radon transform from discrete data in $\mathbb R^3$." pith.science (2026). https://pith.science/paper/5XVL24FK

@misc{pith2026190804753,
  author       = {Pith},
  title        = {Pith review of: Resolution analysis of inverting the generalized Radon transform from discrete data in $\mathbb R^3$},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5XVL24FK}},
  note         = {Machine review of arXiv:1908.04753}
}
abstract

A number of practically important imaging problems involve inverting the generalized Radon transform (GRT) $\mathcal R$ of a function $f$ in $\mathbb R^3$. On the other hand, not much is known about the spatial resolution of the reconstruction from discretized data. In this paper we study how accurately and with what resolution the singularities of $f$ are reconstructed. The GRT integrates over a fairly general family of surfaces $\mathcal S_y$ in $\mathbb R^3$. Here $y$ is the parameter in the data space, which runs over an open set $\mathcal V\subset\mathbb R^3$. Assume that the data $g(y)=(\mathcal R f)(y)$ are known on a regular grid $y_j$ with step-sizes $O(\epsilon)$ along each axis, and suppose $\mathcal S=\text{singsupp}(f)$ is a piecewise smooth surface. Let $f_\epsilon$ denote the result of reconstruction from the descrete data. We obtain explicitly the leading singular behavior of $f_\epsilon$ in an $O(\epsilon)$-neighborhood of a generic point $x_0\in\mathcal S$, where $f$ has a jump discontinuity. We also prove that under some generic conditions on $\mathcal S$ (which include, e.g. a restriction on the order of tangency of $\mathcal S_y$ and $\mathcal S$), the singularities of $f$ do not lead to non-local artifacts. For both computations, a connection with the uniform distribution theory turns out to be important. Finally, we present a numerical experiment, which demonstrates a good match between the theoretically predicted behavior and actual reconstruction.

Figures

Figures reproduced from arXiv: 1908.04753 by the authors.

Figure 1
Figure 1. Comparison of the predicted and actual transition curves for  = 0.01. References [1] F. Andersson, M. V. De Hoop, and H. Wendt. Multiscale Discrete Approximation of Fourier Integral Operators. Multiscale Modeling and Simulation, 10:111–145, 2012. [2] G. Beylkin. The inversion problem and applications of the generalized Radon transform. Comm. Pure and Appl. Math., 37:579–599, 1984. [3] T. Blu, P. Th´evenaz, and M. U… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

  1. [1]

    Andersson, M

    F. Andersson, M. V. De Hoop, and H. Wendt. Multiscale Discrete Approximation of Fourier Integral Operators. Multiscale Modeling and Simulation , 10:111–145, 2012

    work page 2012

  2. [2]

    G. Beylkin. The inversion problem and applications of the generalized Radon transform. Comm. Pure and Appl. Math. , 37:579–599, 1984

    work page 1984

  3. [3]

    T. Blu, P. Th´ evenaz, and M. Unser. Complete Parameterization of Piecewise-Polynomial Interpolation Kernels. IEEE Transactions on Image Processing , 12:1297–1309, 2003

    work page 2003

  4. [4]

    Candes, L

    E. Candes, L. Demanet, and L. Ying. Fast computation of Fourier integral operators. SIAM Journal on Scientific Computing , 29:2464–2493, 2007

    work page 2007

  5. [5]

    Candes, L

    E. Candes, L. Demanet, and L. Ying. A fast buttery algorithm for the computation of Fourier integral operators. SIAM Multiscale Modeling and Simulation , 7:1727–1750, 2009

    work page 2009

  6. [6]

    Cheney and B

    M. Cheney and B. Borden. Synthetic Aperture Radar Imaging. In O. Scherzer, editor, Hand- book of Mathematical Methods in Imaging , pages 763–799. Springer, New York, NY, 2015

    work page 2015

  7. [7]

    Grathwohl, P

    C. Grathwohl, P. Kunstmann, E. T. Quinto, and A. Rieder. Approximate inverse for the common offset acquisition geometry in 2D seismic imaging.Inverse Problems, 34, 2018. article id 014002

    work page 2018

  8. [8]

    Hormander

    L. Hormander. The Analysis of Linear Partial Differential Operators, Vol I . Springer Verlag, New York, 1983. 26 A KATSEVICH

    work page 1983

Show all 23 references
  1. [9]

    Katsevich

    A. Katsevich. An accurate approximate algorithm for motion compensation in two- dimensional tomography. Inverse Problems, 26, 2010. article ID 065007 (16 pp)

    2010

  2. [10]

    Katsevich

    A. Katsevich. A local approach to resolution analysis of image reconstruction in tomography. SIAM Journal on Applied Mathematics , 77:1706–1732, 2017

    2017

  3. [11]

    Katsevich

    A. Katsevich. Analysis of reconstruction from discrete radon transform data in r3 when the function has jump discontinuities. SIAM Journal on Applied Mathematics , 2019. to appear

    2019

  4. [12]

    Kuchment and L

    P. Kuchment and L. Kunyansky. Mathematics of Photoacoustic and Thermoacoustic To- mography. In O. Scherzer, editor, Handbook of Mathematical Methods in Imaging , pages 1117–1167. Springer, New York, NY, 2015

    2015

  5. [13]

    Kuipers and H

    L. Kuipers and H. Niederreiter. Uniform Distribution of Sequences. Dover Publications, Inc., Mineola, NY, 2006

    2006

  6. [14]

    Natterer

    F. Natterer. The Mathematics of Computerized Tomography . SIAM, Philadelphia, 2001

    2001

  7. [15]

    E. T. Quinto, A. Rieder, and Th. Schuster. Local inversion of the sonar transform regularized by the approximate inverse. Inverse Problems, 27, 2011. article id 035006

    2011

  8. [16]

    Ramm and A.I

    A.G. Ramm and A.I. Zaslavsky. Reconstructing singularities of a function given its Radon transform. Math. and Comput. Modelling , 18(1):109–138, 1993

    1993

  9. [17]

    Ramm and A.I

    A.G. Ramm and A.I. Zaslavsky. Singularities of the Radon transform. Bull. Amer. Math. Soc., 25:109–115, 1993

    1993

  10. [18]

    Stefanov

    P. Stefanov. Semiclassical sampling and discretization of certain linear inverse problems. arXiv:1811.01240, 2018

    2018 arXiv

  11. [19]

    Terzioglu, P

    F. Terzioglu, P. Kuchment, and L. Kunyansky. Compton camera imaging and the cone trans- form: a brief overview. Inverse Problems, 34, 2018. article id 054002

    2018

  12. [20]

    F. Treves. Introduction to Pseudodifferential and Fourier Integral Operators. Volume 2: Fourier Integral Operators. The University Series in Mathematics. Plenum, New York, 1980

    1980

  13. [21]

    Wang and M

    K. Wang and M. A. Anastasio. Photoacoustic and Thermoacoustic Tomography: Image For- mation Principles. In O. Scherzer, editor, Handbook of Mathematical Methods in Imaging , pages 1081–1116. Springer, New York, NY, 2015

    2015

  14. [22]

    H. Yang. Oscillatory Data Analysis and Fast Algorithms for Integral Operators . PhD thesis, Stanford University, 2015

    2015

  15. [23]

    M. Zworski. Semiclassical analysis, volume 138 of Graduate Studies in Mathematics . Amer- ican Mathematical Society, Providence, RI, 2012

    2012

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.