Pith. sign in

REVIEW 2 major objections 5 minor 2 cited by

Nonlinear boundary waves can be turned into an X-ray image

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 16:41 UTC pith:CWHD525W

load-bearing objection A genuine first 2D implementation of the Radon-based nonlinear wave reconstruction, with clean optimal spectral differentiation estimates; thin on quantitative validation and with a real but overstated admissible-window gap. the 2 major comments →

arxiv 2509.11951 v2 pith:CWHD525W submitted 2025-09-15 math.NA cs.NAmath.AP

X-ray imaging from nonlinear waves: numerical reconstruction of a cubic nonlinearity

Suvi Anttila , Markus Harju , Teemu Tyni This is my paper
classification math.NA cs.NAmath.AP MSC 35R3065M3244A1265D2565M06
keywords nonlinear wave equationDirichlet-to-Neumann maphigher order linearizationRadon transformspectral regularizationnumerical differentiationinverse boundary value problemfiltered back-projection
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper asks whether an unknown potential q in a nonlinear wave equation ∂_t^2 u − Δu + q u^3 = 0 can be recovered from boundary measurements alone. The answer it argues for is yes: the third derivative of the Dirichlet-to-Neumann map at zero boundary amplitude equals, up to the constant 3!π, the Radon transform of q along a chosen line in space–time, so a full sinogram can be built from boundary data and q follows by standard filtered back-projection. The paper provides the first numerical implementation of this higher-order linearization idea in two space dimensions, replacing the unstable finite-difference third derivative with a Gaussian-filtered spectral differentiator. In synthetic tests with 2% noise added to the boundary data, the regularized reconstruction locates the shape, size, and position of both time-independent and time-dependent potentials, while the unregularized version is overwhelmed by noise.

Core claim

The central claim is the identity 3!π R(q)(t0, θ0, η0) = lim_{τ→∞} ∫_Σ f0 ∂^3_{ε1}|_{ε1=0} Λ(ε1 f1) dS, where Λ is the Dirichlet-to-Neumann map of the cubic nonlinear wave equation, f1 is a plane-wave boundary value, f0 an auxiliary plane wave chosen so that the product of the three linearized waves and the auxiliary wave concentrates on the line x·θ0 = η0 at time t0. The paper realizes this identity numerically: it computes synthetic boundary data, forms g(ε1) = (1/3!π) ∫_Σ f0 Λ(ε1 f1) dS, extends it oddly in ε1, applies a Gaussian-filtered spectral differentiator D_R^{(3)} to obtain the third derivative at ε1 = 0, and feeds the resulting sinogram to filtered back-projection. It also proves

What carries the argument

The carrying identity is Eq. (13): differentiating the nonlinear wave equation three times with respect to the amplitude ε1 of a single plane wave produces a product of linear waves that converges, as the focusing parameter τ→∞, to a delta distribution on one Radon line; the boundary integral of the auxiliary wave against the third derivative of the DN-map then isolates R(q) at that line. Numerically, the key mechanism is regularized spectral differentiation D_R^{(p)} = F^{-1}[(iξ)^p m_R(ξ) F·], where the admissible filter m_R (truncation or Gaussian) obeys |ξ|^p |m_R(ξ)| ≤ C1 R^p and |1−m_R(ξ)| ≤ C2 min(1, (|ξ|/R)^{s−p}). With cutoff R = (E/δ)^{1/s}, this stabilizes the otherwise ill-posed

Load-bearing premise

The reconstruction identity is proven only for parameter tuples inside the causal window W_Radon defined by t1>2r and t2<T−2r; the numerical examples sweep t0 = T/2+η0 with η0∈[−0.4,0.4], which for T=3 and Ω=[−0.5,0.5]² extends beyond the admissible band (1.414,1.586), so the outer rows of the sinogram rest on an unstated assumption that the formula also holds outside that window.

What would settle it

Run the reconstruction on a synthetic potential with a high-accuracy forward solver, and evaluate Eq. (13) numerically at a tuple outside the admissible window, e.g. (t0,θ0,η0)=(1.1,0,0.4), comparing to the true Radon value at that line. If the boundary-derivative integral does not approach 3!π R(q) as τ grows, the outer rows of the implemented sinogram are artifacts; additionally, reconstructing q with η0 restricted to |η0| < T/2 − 2r (≈0.086) should then yield a noticeably cleaner back-projected image.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • For a time-independent potential, one forward plane wave per angle θ0 fixes the whole column of the sinogram: varying the auxiliary wave's time shift η0 sweeps the distance parameter, cutting the number of required full boundary measurements from a 3D grid to a 1D sweep per angle.
  • The stability estimate applies to any smooth function, not just the wave-equation data: regularized spectral differentiation with an admissible filter is minimax-optimal among all linear operators, so the same interpolation rate E^{p/s}δ^{1−p/s} is the best one can hope for in this setting.
  • Limited-angle data work as in conventional tomography: restricting the incident plane-wave directions produces a limited-angle Radon transform whose stably recoverable singularities are exactly those predicted by micro-local analysis, enabling imaging near boundaries or over short observation times.
  • The method handles time-dependent potentials q(x,t): because the reconstruction is pointwise in the Radon domain, a potential that moves or changes amplitude in time is recovered slice-by-slice at each chosen reconstruction time.
  • The pipeline is not specific to p=3 in principle: taking a p-th derivative of the DN-map extracts the p-th power nonlinearity, so the same boundary-data-to-sinogram procedure generalizes to any integer power p≥2.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A strict reading of the admissible window W_Radon (t1 > 2r, t2 < T−2r) would exclude most of the published sweep: with T=3 and Ω=[−0.5,0.5]², r≈0.707, the band is (1.414,1.586), while the implementation uses t0 = T/2 + η0 for η0∈[−0.4,0.4], i.e. t0∈[1.1,1.9]. If the causal window is genuinely required, the outer sinogram rows are unsupported by the theory and could be contaminating the back-projec
  • The sinogram-based formulation means that any future improvement in standard CT—sparse-angle solvers, iterative denoising, learned reconstruction—can be transplanted directly onto the output of this method, because the boundary-data-to-sinogram step is decoupled from the image-reconstruction step.
  • The identity (13) is stated for the continuous DN-map; the numerical experiments replace it with a finite-difference surrogate. A cleaner test of the core identity would use an independent, high-accuracy forward solver (e.g., spectral) and compare sinogram rows directly, isolating the discretization error of the forward solver from the ill-posedness of the derivative step.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper studies the inverse boundary value problem for the semilinear wave equation ∂_t^2 u − Δu + q u^3 = 0 in a bounded 2D domain, where the goal is to recover the potential q from the Dirichlet-to-Neumann map. The method is the higher-order linearization approach of [29]: using appropriately chosen real-valued focusing waves, the third derivative of the DN-map is shown, via the integral identity (6), to approximate the spatial Radon transform of q; filtered backprojection then gives q. The paper implements this numerically with a finite-difference forward solver, a regularized spectral differentiation step for the required third derivative, and a pointwise reconstruction method for comparison. It also proves Hölder-type stability estimates for regularized spectral differentiation (Theorems 1–2) and gives numerical experiments for smooth, discontinuous, high-frequency, limited-angle, and time-dependent potentials.

Significance. If the numerical claims hold, this is a useful practical step: it is, to my knowledge, the first two-dimensional numerical implementation of the [29] reconstruction, and it demonstrates that nonlinear wave interactions can produce Radon-type data for a potential using real-valued waves. The spectral differentiation analysis is clean and the minimax lower bound in Theorem 2 is a nice addition with independent interest. The experiments are honest about limitations such as blurring for high-frequency potentials. However, the numerical validation is almost entirely visual—no error tables, convergence studies in τ, ε, N_ε, R, or noise-level dependence are reported—and, more importantly, the reconstruction sweep in Section 3.3 uses parameter tuples outside the admissible window W_Radon for which the reconstruction formula is stated. These two issues leave the central feasibility claim not fully established as written.

major comments (2)
  1. [§3.3 and §4, Eqs. (9)–(10), (13), (27)] The reconstruction identity (13) is stated for tuples (t0,θ0,η0) in W_Radon = [t1,t2]×S^1×[−r,r], with t1>2r and t2<T−2r. For the numerical setup Ω=[−0.5,0.5]^2 and T=3, r=√0.5≈0.707, so the admissible time band is roughly (1.414,1.586). The implementation, however, sets t0=T/2+η0 and sweeps η0∈[−0.4,0.4], giving t0∈[1.1,1.9]. Only a small central band of rows (|η0|≲0.086) lies in W_Radon. For many of the outer rows the intended plane waves H1 or H2 are not guaranteed to satisfy the initial/final conditions required in (4)–(5), so v1 and v0 need not coincide with H1 and H2 inside Ω, and Lemma 1 does not apply. The paper gives no justification for these rows. This directly affects Figure 4 and all reconstructions that use the full sinogram. Please either restrict reconstruction to W_Radon, or prove that the specific choice t0=T/2+η0, together with a suitable choice of the cutoff φ_h in (1
  2. [§4, Figures 4–10] The central claim that the method is feasible is supported only by visual comparison of images. No quantitative error measure is reported for the reconstructed sinogram R(q)_rec or for the final q_rec, and no study of the dependence on the numerical parameters (τ, ε, N_ε, R, h, noise level σ) is presented. Since the paper proposes a numerical reconstruction method and claims robustness at 2% noise, the authors should provide at least relative L2-type errors for the examples and a short parameter-sensitivity or convergence table. Without such data, it is difficult to judge whether the good visual results are stable or fortuitous for the chosen parameter values.
minor comments (5)
  1. [§3.2, paragraph after Example 1] There is a typo: “different choise” should be “different choice.”
  2. [Eq. (24)] The bracket notation in the definition of the Fourier multiplier [D_R^{(p)} g(ξ) is malformed; it should read \widehat{D_R^{(p)} g}(ξ) or similar.
  3. [§3.1, Eqs. (15)–(17)] The convergence study in Figure 1 concerns the interior solution, while the inverse problem uses the boundary normal derivatives approximated by (16)–(17), which are second-order one-sided stencils. A short remark or test on the convergence of the resulting DN-map would strengthen the synthetic-data section.
  4. [§3.2, Example 2] In the piecewise definition of g, the inequalities use “<” and “≤” inconsistently at the breakpoints; this is harmless but should be cleaned up.
  5. [§2.2.1, Eq. (11)] The support width h of the cutoff φ_h is never specified or discussed. Since the vanishing of H1 at t=0 and H2 at t=T depends on h relative to the geometry, the authors should state the chosen h and explain its role in the numerical experiments.

Circularity Check

0 steps flagged

No significant circularity: reconstruction identities are imported from independent analytical prior work, and the new numerical contributions are self-contained.

full rationale

The central reconstruction formula (13) is imported from [29] and used as a black box; [29] is a peer-reviewed analytical paper with parameter-free theorems and stated assumptions that do not include the numerical reconstruction, so this is legitimate independent support rather than a self-citation chain. The paper's own contributions—the spectral regularization stability estimates in Section 3.2 and the finite-difference forward solver—are derived from first principles and do not presuppose the reconstructed potential. The numerical experiments compare reconstructed sinograms against independently computed true Radon transforms, so the prediction is not fitted to the data by construction. The only notable concern is that the implementation in Section 3.3 sweeps t0 = T/2 + η0 outside the admissible window W_Radon defined by Eqs. (9)–(10); however, this is a correctness/validation issue (the guarded reconstruction identity may not apply on those rows), not a circularity, because the identity is not defined in terms of its output. No step in the derivation reduces to an input by construction, and no load-bearing argument depends on unverified self-citation.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

The reconstruction pipeline rests on the analytical framework of [29] and [31] (overlapping authorship: Tyni is a co-author of this paper and of the cited theory papers), which supplies differentiability of the DN-map, the integral identity (6), and the delta-approximation Lemmas 1-2. The numerical method adds hand-chosen parameters (tau=700, epsilon=1.5, N_eps=16, R=10) and several discretization choices (boundary stencils (16)-(17), cutoff width h in (11)) whose fidelity is not independently verified. No new physical entities, forces, or dimensions are introduced.

free parameters (6)
  • tau (wave focusing width) = 700 (all examples)
    Width of the Gaussian delta-concentration in Eq. (11)-(12); chosen experimentally (Section 4). Lemma 1 requires tau->infty, so finite tau is a resolution-compromise parameter.
  • epsilon (linearization amplitude) = 1.5 (Radon), 0.1 (pointwise)
    Range of the small parameter in Eq. (25); chosen experimentally. The theory requires small enough epsilon for the small-solution branch of (2).
  • N_eps (number of DN-map samples) = 16
    Number of boundary-value evaluations used to form g~(eps1) before spectral differentiation; chosen experimentally.
  • R (spectral cutoff for differentiation) = 10 (Gaussian filter)
    Regularization parameter in Eq. (24); chosen experimentally, acknowledged in Section 3.2. Theorem 1's optimal choice R=(E/delta)^(1/s) requires a known noise level delta, which is unavailable in the inverse experiments.
  • h (support width of cutoff phi_h) = unspecified
    Support parameter of phi_h in Eq. (11), entering every plane-wave boundary datum (12); never given numerically in the implementation, a reproducibility gap.
  • noise level sigma = 2% of mean |Lambda| (~44 dB SNR)
    Simulated additive Gaussian noise on the DN-map; a single noise level with no sweep, so noise-dependence of the reconstruction is not demonstrated.
axioms (5)
  • domain assumption The DN-map Lambda_q for (2) is well-defined, smooth in epsilon, and the integral identity (6) plus reconstruction formulas (13)-(14) hold.
    Imported from [29, Prop 4 and Lemma 8] and [31, Lemma 20]; invoked in Section 2.1: 'For justification of the differentiability of the DN-map Lambda_q, see [29].'
  • domain assumption Plane-wave products approximate line/point delta distributions with error O(tau^{-1/2}) as in Lemma 1 and Lemma 2.
    Section 2.2.1, Eqs. (11)-(13); Lemma 1 requires G compactly supported and Lipschitz, but Examples 3-5 use characteristic functions, which are not Lipschitz.
  • ad hoc to paper The finite-difference discretization (15) with boundary normal-derivative stencils (16)-(17) produces a synthetic DN-map sufficiently close to the continuous one.
    Section 3.1; only interior convergence to a reference solution is checked (Figure 1); the boundary stencil fidelity is not tested, and the stencils approximate the derivative one grid cell inside the boundary.
  • domain assumption The zero-initial-condition linear solutions (4) equal the plane waves H1, H2, H3 inside the reconstruction window.
    Section 2.2, Eq. (12); requires wavefronts to enter before the reconstruction time and the cutoff h small enough; the chosen sweep t0 = T/2 + eta0 in [1.1,1.9] extends outside the admissible window (2r, T-2r) ~ (1.414,1.586).
  • domain assumption Small-solution existence and uniqueness for (2) on [0,T] for boundary data eps1 f1 with small norm.
    Needed for the oddness of Lambda_q(eps1 f1) used in the odd-extension step of Section 3.3 and for the higher-order linearization; from [29].

pith-pipeline@v1.3.0-alltime-deepseek · 18257 in / 29796 out tokens · 333404 ms · 2026-08-04T16:41:58.335392+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of X-ray imaging from nonlinear waves: numerical reconstruction of a cubic nonlinearity." pith.science (2026). https://pith.science/paper/CWHD525W

@misc{pith2026250911951,
  author       = {Pith},
  title        = {Pith review of: X-ray imaging from nonlinear waves: numerical reconstruction of a cubic nonlinearity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CWHD525W}},
  note         = {Machine review of arXiv:2509.11951}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We study an inverse boundary value problem for the nonlinear wave equation in $2 + 1$ dimensions. The objective is to recover an unknown potential $q(x, t)$ from the associated Dirichlet-to-Neumann map using real-valued waves. We propose a direct numerical reconstruction method for the Radon transform of $q$, which can then be inverted using standard X-ray tomography techniques to determine $q$. Our implementation introduces a spectral regularization procedure to stabilize the numerical differentiation step required in the reconstruction, improving robustness with respect to noise in the boundary data. We give rigorous justification and optimal stability estimates for the regularized spectral differentiation of noisy measurements, which may be of independent interest. Numerical experiments demonstrate the feasibility of recovering potentials from boundary measurements of nonlinear waves and illustrate the advantages of the Radon-based reconstruction.

Figures

Figures reproduced from arXiv: 2509.11951 by Markus Harju, Suvi Anttila, Teemu Tyni.

Figure 1
Figure 1. Figure 1: Comparison of the convergence of two finite difference approximation solutions to the nonlinear wave equation as a function of the total number of nodes in the discretization. We see that the finite difference approximation given by (15) (blue solid line) achieves a given accuracy with fewer discretization nodes compared to the classical second-order solution (red dot-dashed line). To evaluate the DN-map Λ… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of regularized differentiation via high-frequency truncation (red dashed) and Gaussian filtering (blue dot-dashed). The exact derivatives are shown in solid black. Left: Noise level σ = 10−3 . Truncation cutoffs: R = 64 (1st derivative) and 40 (2nd derivative). Gaussian parameters: R = q 105 5 (1st derivative) and q 105 8 (2nd derivative). RMS-errors: etrunc = 0.016, eGauss = 0.013 (1st derivati… view at source ↗
Figure 3
Figure 3. Figure 3: Left: The causal set where waves can be sent and measured lies between the two blue cones within the cylinder Ω × [0, T] (gray). This is the optimal set, which can be reached from the lateral boundary Σ by light rays and from where light can emanate to the boundary. Right: Illustration of how the Radon transform of q is obtained when q is time-independent. The space–time domain Ω × [0, T] is shown as the g… view at source ↗
Figure 4
Figure 4. Figure 4: Sinograms of Example 2. Left: True R(q) computed with Matlab’s radon. Middle: reconstructed R(q) rec by finite differences. Right: reconstructed R(q) rec via spectral differentiation. The finite difference approximation of D(3)ge(0) amplifies measurement error. A better approximation to the true sinogram is obtained via regularized differentiation as D (3) R ge(0). Here R = 10 [PITH_FULL_IMAGE:figures/ful… view at source ↗
Figure 5
Figure 5. Figure 5: Example 1. 1st: True unknown potential. 2nd: Pointwise reconstruction using finite differences. 3rd: Radon reconstruction using finite differences. 4th: Radon reconstruction using regularized spectral differentiation. In the unregularized cases, in the pointwise approach noise domi￾nates the reconstruction compared to the back-projection, where the location and rough shape of the potential can be detected.… view at source ↗
Figure 6
Figure 6. Figure 6: Example 2, smooth bump functions supported on a disc and two ellipses (tilted by ±22.5 ◦ from the vertical axis). 1st: True unknown potential. 2nd: Pointwise reconstruction using finite differences. 3rd: Radon reconstruction using finite differences. 4th: Radon reconstruction using regularized spectral differentiation. Location, shape, and size of the potential functions are captured accurately, but finer … view at source ↗
Figure 7
Figure 7. Figure 7: Example 4. 1st: True unknown potential. 2nd: Pointwise reconstruction using finite differences. 3rd: Radon reconstruction using finite differences. 4th: Radon reconstruction using regularized spectral differentiation. The jump discontinuity of the characteristic function is blurred in the reconstruction due to finite τ > 0 being used in the wave v1, see Lemma 1. Nevertheless, the overall geometry of the po… view at source ↗
Figure 8
Figure 8. Figure 8: Example 5 with k = 2, 3, 4, 5. Top row: True unknown potentials. Bottom row: Reconstructions via regularized differentiation and filtered back-projection. The method reliably recovers the coarse, low-frequency structures with correct contrast and location. As the frequency k in the potential q increases, however, the reconstructions become progressively blurred and their amplitudes damped, with fine-scale … view at source ↗
Figure 9
Figure 9. Figure 9: Example 3, limited angle tomography. 1st: True potential. 2nd: Reconstruction from Radon data over Θ′ = {0 ◦ , · · · , 134◦}. 3rd: Θ′ = {0 ◦ , · · · , 89◦}. 4th: Θ′ = {0 ◦ , · · · , 44◦}. All reconstructions are obtained via regularized differentiation and filtered backprojection of the Radon transform. The locations of stable singularities (see Remark 2) are highlighted with red curves. Increasing blurrin… view at source ↗
Figure 10
Figure 10. Figure 10: Example 6, time-dependent potential. Top row: Cross-sections of the true potential at times t = 1.1, 1.5, 1.9. Bottom row: Reconstructions via regularized differentiation and filtered back-projection. The potential is a smooth bump function, whose amplitude increases and position shifts linearly over time. Acknowledgements This work was supported by the Research Council of Finland (Flagship of Advanced Ma… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Gauge symmetry and uniqueness in inverse problems for the JMGT equation

    math.AP 2026-04 unverdicted novelty 6.0

    The nonlinear coefficient β in the JMGT equation is uniquely determined from boundary measurements, while α, q, and F are recovered up to gauge symmetry.

  2. Gauge symmetry and uniqueness in inverse problems for the JMGT equation

    math.AP 2026-04 unverdicted novelty 6.0

    Nonlinear acoustic coefficient β is uniquely determined from the all-boundary measurement map for the JMGT equation; linear coefficients α, q and source F are recovered up to gauge symmetry via linearization and geome...

Reference graph

Works this paper leans on

47 extracted references · 1 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Acosta, G

    S. Acosta, G. Uhlmann, and J. Zhai. Nonlinear Ultrasound Imaging Modeled by a Westervelt Equation.SIAM J. Appl. Math., 82(2):408–426, 2022. 22

    2022

  2. [2]

    Balehowsky, A

    T. Balehowsky, A. Kujanp¨ a¨ a, M. Lassas, and T. Liimatainen. An Inverse Problem for the Relativistic Boltzmann Equation.Commun. Math. Phys., 396(3):983–1049, 2022

    2022

  3. [3]

    Brezis and P

    H. Brezis and P. Mironescu. Gagliardo–Nirenberg inequalities and non- inequalities: the full story.Ann. l. H. Poincar´ e C – AN, 35(5):1355–1376, 2018

    2018

  4. [4]

    X. Chen, M. Lassas, L. Oksanen, and G. P. Paternain. Detection of Hermi- tian connections in wave equations with cubic non-linearity.J. Eur. Math. Soc., 24(7):2191–2232, 2021

    2021

  5. [5]

    X. Chen, M. Lassas, L. Oksanen, and G. P. Paternain. Inverse Problem for the Yang–Mills Equations.Commun. Math. Phys., 384(2):1187–1225, 2021

    2021

  6. [6]

    J. Cullum. Numerical differentiation and regularization.SIAM J. Numer. Anal., 8(2):254–265, 1971

    1971

  7. [7]

    C. I. Cˆ arstea, G. Nakamura, and M. Vashisth. Reconstruction for the coefficients of a quasilinear elliptic partial differential equation.Appl. Math. Lett., 98:121–127, 2019

    2019

  8. [8]

    de Hoop, G

    M. de Hoop, G. Uhlmann, and Y. Wang. Nonlinear interaction of waves in elas- todynamics and an inverse problem.Math. Ann., 376(1):765–795, 2020

    2020

  9. [9]

    M. d. de Hoop, G. Uhlmann, and Y. Wang. Nonlinear responses from the inter- action of two progressing waves at an interface.Ann. Inst. Henri Poincare (C), 36(2):347–363, 2019

    2019

  10. [10]

    Eld´ en, F

    L. Eld´ en, F. Berntsson, and T. Reginska. Wavelet and Fourier Methods for Solving the Sideways Heat Equation.SIAM J. Sci. Comput., 21(6):2187–2205, 2000

    2000

  11. [11]

    Feizmohammadi and L

    A. Feizmohammadi and L. Oksanen. An inverse problem for a semi-linear elliptic equation in Riemannian geometries.J. Differ. Equ., 269(6):4683–4719, 2020

    2020

  12. [12]

    Feizmohammadi and L

    A. Feizmohammadi and L. Oksanen. Recovery of zeroth order coefficients in non- linear wave equations.J. Inst. Math. Jussieu, 21(2):367–393, 2022

    2022

  13. [13]

    Griesmaier, M

    R. Griesmaier, M. Kn¨ oller, and R. Mandel. Inverse medium scattering for a non- linear Helmholtz equation.J. Math. Anal. Appl., 515(1):126356, 2022

    2022

  14. [14]

    P. C. Hansen.Discrete inverse problems: insight and algorithms. SIAM, 2010

    2010

  15. [15]

    Helgason.The Radon Transform, volume 5 ofProgress in Mathematics

    S. Helgason.The Radon Transform, volume 5 ofProgress in Mathematics. Birkh¨ auser, Boston, MA, 2nd edition, 1999

    1999

  16. [16]

    Hintz, G

    P. Hintz, G. Uhlmann, and J. Zhai. The Dirichlet-to-Neumann map for a semi- linear wave equation on Lorentzian manifolds.Commun. Partial Differ. Equ., 47(12):2363–2400, 2022. 23

    2022

  17. [17]

    Hintz, G

    P. Hintz, G. Uhlmann, and J. Zhai. An Inverse Boundary Value Problem for a Semilinear Wave Equation on Lorentzian Manifolds.Int. Math. Res. Not., 2022(17):13181–13211, 2022

    2022

  18. [18]

    A. J. Jerri.The Gibbs Phenomenon in Fourier Analysis, Splines and Wavelet Approximations. Mathematics and Its Applications. Springer New York, NY, 1998

    1998

  19. [19]

    Kaltenbacher and W

    B. Kaltenbacher and W. Rundell. On the identification of the nonlinearity param- eter in the Westervelt equation from boundary measurements.Inv. Probl. Imag., 15(5):865–891, 2021

    2021

  20. [20]

    Knowles and R

    I. Knowles and R. J. Renka. Methods for numerical differentiation of noisy data. Electron. J. Differ. Equ, 21:235–246, 2014

    2014

  21. [21]

    Knowles and R

    I. Knowles and R. Wallace. A variational method for numerical differentiation. Numer. Math., 70:91–110, 1995

    1995

  22. [22]

    Krupchyk and G

    K. Krupchyk and G. Uhlmann. A remark on partial data inverse problems for semilinear elliptic equations.Proc. Am. Math. Soc., 148(2):681–685, 2020

    2020

  23. [23]

    Krupchyk and G

    K. Krupchyk and G. Uhlmann. Partial data inverse problems for semilinear elliptic equations with gradient nonlinearities.Math. Res. Lett., 27(6):1801–1824, 2021

    2021

  24. [24]

    Kurylev, M

    Y. Kurylev, M. Lassas, L. Oksanen, and G. Uhlmann. Inverse problem for Einstein-scalar field equations.Duke Math. J., 171(16):3215–3282, 2022

    2022

  25. [25]

    Kurylev, M

    Y. Kurylev, M. Lassas, and G. Uhlmann. Inverse problems for Lorentzian mani- folds and non-linear hyperbolic equations.Invent. Math., 212(3):781–857, 2018

    2018

  26. [26]

    R.-Y. Lai, G. Uhlmann, and Y. Yang. Reconstruction of the Collision Kernel in the Nonlinear Boltzmann Equation.SIAM J. Math. Anal., 53(1):1049–1069, 2021

    2021

  27. [27]

    Lassas, T

    M. Lassas, T. Liimatainen, Y.-H. Lin, and M. Salo. Partial data inverse prob- lems and simultaneous recovery of boundary and coefficients for semilinear elliptic equations.Rev. Mat. Iberoam., 37(4):1553–1580, 2020

    2020

  28. [28]

    Lassas, T

    M. Lassas, T. Liimatainen, Y.-H. Lin, and M. Salo. Inverse problems for elliptic equations with power type nonlinearities.J. Math. Pures Appl., 145:44–82, 2021

    2021

  29. [29]

    Lassas, T

    M. Lassas, T. Liimatainen, L. Potenciano-Machado, and T. Tyni. Uniqueness, reconstruction and stability for an inverse problem of a semi-linear wave equation. J. Diff. Eq., 337:395–435, 2022

    2022

  30. [30]

    Lassas, T

    M. Lassas, T. Liimatainen, L. Potenciano-Machado, and T. Tyni. An inverse problem for a semi-linear wave equation: a numerical study.Inv. Probl. Imag., 18(1):62–85, 2024. 24

    2024

  31. [31]

    Lassas, T

    M. Lassas, T. Liimatainen, L. Potenciano-Machado, and T. Tyni. Stability and Lorentzian geometry for an inverse problem of a semilinear wave equation.Anal. PDE., 18(5):1065–1118, 2025

    2025

  32. [32]

    Lassas, G

    M. Lassas, G. Uhlmann, and Y. Wang. Determination of vacuum space-times from the Einstein-Maxwell equations.arXiv:1703.10704, 2017

    Pith/arXiv arXiv 2017

  33. [33]

    Lassas, G

    M. Lassas, G. Uhlmann, and Y. Wang. Inverse Problems for Semilinear Wave Equations on Lorentzian Manifolds.Commun. Math. Phys., 360(2):555–609, 2018

    2018

  34. [34]

    Liimatainen and Y.-H

    T. Liimatainen and Y.-H. Lin. Uniqueness results for inverse source problems for semilinear elliptic equations.Inverse Probl., 40(4):045030, 2024

    2024

  35. [35]

    A. R. Mitchell and D. F. Griffiths.The finite difference method in partial differ- ential equations. A Wiley-Interscience publication. Wiley, Chichester, 1980

    1980

  36. [36]

    J. L. Mueller and S. Siltanen.Linear and nonlinear inverse problems with practical applications. SIAM, 2012

    2012

  37. [37]

    Natterer.The Mathematics of Computerized Tomography

    F. Natterer.The Mathematics of Computerized Tomography. Vieweg+Teubner Verlag, Wiesbaden, 1986

    1986

  38. [38]

    Oksanen, M

    L. Oksanen, M. Salo, P. Stefanov, and G. Uhlmann. Inverse problems for real principal type operators.Am. J. Math, 146(1):161–240, 2024

    2024

  39. [39]

    Qian, C.-L

    Z. Qian, C.-L. Fu, X.-T. Xiong, and T. Wei. Fourier truncation method for high order numerical derivatives.Comput. Appl. Math, 181(2):940–948, 2006

    2006

  40. [40]

    E. T. Quinto. Singularities of the X-ray transform and limited data tomography inR 2 andR 3.SIAM J. Math. Anal., 24(5):1215–1225, 1993

    1993

  41. [41]

    Ramm and A

    A. Ramm and A. Smirnova. Stable numerical differentiation: when is it possible? J. Korean SIAM, 7:47–61, 2003

    2003

  42. [42]

    J. B. Roerdink and M. A. Westenberg. Data-parallel tomographic reconstruction: a comparison of filtered backprojection and direct fourier reconstruction.Parallel Comput., 24(14):2129–2142, 1998

    1998

  43. [43]

    E. M. Stein and G. Weiss.Introduction to Fourier analysis on Euclidean spaces, volume 1. Princeton university press, 1971

    1971

  44. [44]

    S´ a Barreto and P

    A. S´ a Barreto and P. Stefanov. Recovery of a Cubic Non-linearity in the Wave Equation in the Weakly Non-linear Regime.Commun. Math. Phys., 392(1):25–53, 2022

    2022

  45. [45]

    Uhlmann and Y

    G. Uhlmann and Y. Wang. Determination of Space-Time Structures from Gravi- tational Perturbations.Commun. Pure Appl., 73(6):1315–1367, 2020

    2020

  46. [46]

    Wang and T

    Y. Wang and T. Zhou. Inverse problems for quadratic derivative nonlinear wave equations.Commun. Partial Differ. Equ., 44(11):1140–1158, 2019. 25

    2019

  47. [47]

    F. Yang, C. L. Fu, and X. X. Li. The Generalized Tikhonov Regularization Method for High Order Numerical Derivatives.Comput. Model. Eng. Sci., 100(1):19–29, 2014. E-mail addresses: Suvi Anttila: suvi.m.anttila@oulu.fi (Corresponding author) Markus Harju: markus.harju@oulu.fi Teemu Tyni: teemu.tyni@oulu.fi 26

    2014