Pith. sign in

REVIEW 5 major objections 5 minor 31 references

Inverse source problems for the stochastic wave equations

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read The paper claims that the two source terms of a stochastic wave equation driven by a finite-jump Lévy process can be reconstructed from the final-time wave field, and that multi-frequency Tikhonov regularization mitigates the ill-posed…

desk verdict The Lévy half is a real, salvageable contribution, but the Helmholtz numerical experiments invert an operator that does not match the derived stochastic model, breaking the paper's central claim. read the letter →

arxiv 2507.01789 v2 pith:4HMAENQE submitted 2025-07-02 math.NA cs.NA

classification math.NAcs.NA MSC 35R3035R6060H1565J20
keywords stochasticwaveequationinversesourceproblemHelmholtzwithoutattenuationfinite-jumpLévyprocessill-posednessTikhonovregularizationmulti-frequencydatafusionfinal-time
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 tries to show that two ill-posed stochastic inverse source problems become numerically tractable when the right statistical features of the wave field are used. For the one-dimensional stochastic Helmholtz equation without attenuation, it argues that second moments alone give only the average source strength, but a feature function built from real and imaginary parts, combined with multi-frequency data fusion and Tikhonov regularization, can recover the strength point by point. For the wave equation driven by a finite-jump Lévy process, it claims that final-time data $u(x,T)$ determine the deterministic source $f$ through the mean of the modal coefficients and the random source $g$ through their covariance, up to the sign of $g$. If these claims hold, sources in settings with jump-type stochastic fluctuations, such as seismic or financial-volatility models, could be characterized from a single final observation.

What carries the argument

The machinery is mode-by-mode statistical inversion. In the Helmholtz half, the operative object is the integral operator $T_\nu$ with kernel $K_\nu(x,y)=\cos^2(2\nu|x-y|)$; the paper proves this operator is compact and self-adjoint, so a single frequency is ill-posed, while a dense set of frequencies $\{4\nu_k\}$ forces the null space to be trivial and makes the Tikhonov-regularized fit well-posed and $O(\sqrt{\delta})$-convergent. In the Lévy half, the Laplace eigenbasis diagonalizes the wave equation, so each mode $u_k(t)$ solves $u_k''+\lambda_k u_k=f_k h(t)+g_k\dot L_t$; the mean $\mathbb E[u_k(T)]$ and covariance $\mathrm{Cov}(u_k,u_l)$ become explicit linear expressions in $f_k$ and $g_kg_l$, inverted through (4.14)--(4.15). The covariance kernel $I_{kl}=\int_0^T \sin(k(T-\tau))\sin(l(T-\tau))/(kl)\,d\tau$ is shown to be nonzero for algebraic $T$, which is what carries the uniqueness claim for $g$.

What would settle it

Generate synthetic observations by solving the actual stochastic Helmholtz equation (2.1)--(2.2) with a known compactly supported $\mu$, then feed the second-moment data into the paper's multi-frequency inversion with the $\cos^2(2\nu|x-y|)$ kernel; if the recovered $\mu$ differs from the truth beyond the reported error levels, the numerical scheme is not inverting the physical forward map. For the Lévy equation, run the covariance-only reconstruction (4.15) on a sign-changing $g$ without phase correction; the paper's Remark 7 predicts it will return $|g|$ rather than $g$.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the obstruction to inverse source recovery in these stochastic settings is not the randomness itself but the choice of statistical data and the numerical treatment of the resulting operators. For the no-attenuation Helmholtz equation $\Delta u+\nu^2 u=f$ with $f=\sqrt{\mu}\,W'$, the variance $\mathbb E|u(x)|^2$ depends only on $\int_M \mu\,dy$, so the paper defines the feature functions $H^1(x,\nu,\mu)=\mathbb E|\mathrm{Re}(2\mathrm i\nu u)|^2=\int_M \cos^2(\nu|x-y|)\mu(y)\,dy$ and $H^2=\mathbb E|\mathrm{Re}(2\mathrm i\nu u)|^2-\mathbb E|\mathrm{Im}(2\mathrm i\nu u)|^2=\int_M \cos(2\nu|x-y|)\mu(y)\,dy$, then discretizes with kernel $\cos^2(2\nu|x-y|)$ and inverts with multi-frequency Tikhonov regularization. For the Lévy-driven equation $u_{tt}-\Delta u=f(x)h(t)+g(x)\dot L_t$, the paper establishes the stability estimate (4.13) for the mild solution and derives the reconstruction formulas (4.14) and (4.15): the modal means give $f_k$ when the drift is zero, and the modal covariances give $g_kg_l$ up to sign, with uniqueness of $f$ under monotonicity conditions and of $g$ for algebraic $T$.

Load-bearing premise

The Helmholtz reconstruction presupposes that the feature function with kernel $\cos^2(2\nu|x-y|)$ used in the discretization and Definition 2 is the forward map that generated the observations; the paper's own derivation from the wave equation yields kernels $\cos^2(\nu|x-y|)$ and $\cos(2\nu|x-y|)$ instead, so the numerical operator and the physical model may be inconsistent. The Lévy reconstruction additionally presupposes that the Lévy parameters $(b,\sigma,\lambda_p,\sigma_j^2)$ are known and that prior information fixes the sign of $g$, since formula (4.15) determines only the product $g_kg_l$ and hence $|g|$.

Editorial extensions

If this is right

  • Under the dense-frequency uniqueness theorem, the multi-frequency Helmholtz inverse problem has a unique regularized solution that converges to the true strength at rate $O(\sqrt{\delta})$ as the data noise $\delta\to0$.
  • The numerical experiments indicate that 3--4 frequency data points suffice for single-Fourier-mode strengths and about 10 for multi-mode strengths, compared with the 16 frequencies used in the attenuated case.
  • In the Lévy-driven equation, the modal means recover the deterministic source $f$ exactly when the drift $b=0$, and the modal covariances recover the random source $g$ up to sign, with relative $L^2$ errors of order $10^{-3}$ for simple modes and $10^{-2}$ for multi-mode reconstructions.
  • The stability estimate (4.13) bounds the expected space-time energy of the mild solution by the $L^2$ norms of $f$ and $g$, so the direct problem is mean-square well posed whenever the Gaussian-Poisson jump parameters are finite.

Reading between the lines

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

  • One could test whether the Helmholtz inversion still works with the physically derived kernels $\cos^2(\nu|x-y|)$ and $\cos(2\nu|x-y|)$ in place of $\cos^2(2\nu|x-y|)$; the paper's uniqueness and regularization theorems do not directly cover those operators.
  • Because the covariance data are insensitive to the sign of $g$, an immediate extension is to couple the final-time covariance with third-order moments or a known deterministic source term to break the sign symmetry; the paper notes the ambiguity but does not resolve it.
  • The same spectral-covariance scheme should extend to infinite-jump Lévy processes by truncating or renormalizing the jump measure; the paper lists this as future work, and the stability estimate suggests control is retained when $\lambda_p\sigma_j^2$ stays finite.
  • The dense-frequency condition in the Helmholtz half is stated for $\{4\nu_k\}$ dense in $[0,\infty)$; in higher dimensions the same idea would need a different oscillatory-integral argument, since the one-dimensional $|x-y|$ kernel structure does not transpose directly.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper treats two inverse source problems. For the one-dimensional stochastic Helmholtz equation without attenuation, with random source f = sqrt(mu) W', it proposes to reconstruct the strength mu from second-moment data of the wave field at several frequencies, using a feature function, multi-frequency data fusion, and Tikhonov regularization. For the stochastic wave equation driven by a finite-jump Lévy process, it proves a stability estimate for the mild solution and derives spectral formulas for recovering the deterministic source f and the spatial factor g from final-time data u(x,T). The paper contains theoretical statements on ill-posedness, uniqueness, stability, and convergence, together with numerical experiments for both problems.

Significance. If the central claims were correct, the paper would offer a genuinely new computational route to the no-attenuation Helmholtz inverse source problem and one of the first reconstruction schemes for a Lévy-driven stochastic wave equation. The direct stability estimate for the Lévy problem (Theorem 7) appears sound and is a useful contribution. However, the Helmholtz numerical framework as implemented does not invert the operator derived from the stochastic model, and the uniqueness statement for g in the Lévy section is overclaimed, as the paper itself partially acknowledges in Remark 7. In its current form the manuscript does not support its principal advertised conclusions.

major comments (5)
  1. [3.1.1, Definition 2, 3.1.3] Equations (3.11) and (3.12) correctly derive H1(x,ν,µ) = ∫ cos²(ν|x−y|)µ(y)dy and H2(x,ν,µ) = ∫ cos(2ν|x−y|)µ(y)dy. The next display, however, approximates H by Δy Σ cos²(2ν|x−y_j|)µ(y_j), and Definition 2 defines the forward operator Tν with kernel cos²(2ν|x−y|). Since cos²(2νr) is not an approximation of cos²(νr) or cos(2νr), there is no chain of reasoning connecting the physical data to the operator being inverted. Consequently, the numerical reconstructions in Section 3.1.3 and the theorems built on Definition 2 (Theorems 2–6) concern a different integral operator, and the paper's central claim that multi-frequency fusion mitigates the ill-posedness of the no-attenuation stochastic Helmholtz inverse problem is unsupported as written.
  2. [4.2.2, Theorem 8(2), Remark 7] Formula (4.15) determines the products g_k g_l, so for every solution g the function −g gives exactly the same covariance data. Theorem 8(2) therefore cannot be correct as stated: the source g is not uniquely determined by Cov(u_k(T), u_l(T)). Remark 7 concedes the sign ambiguity and indicates that the code applies a phase correction using prior information, but this limitation is not reflected in the abstract, in the statement of Theorem 8, or in the numerical claims. The final-time covariance data can at best determine g up to a global sign unless additional information is supplied.
  3. [Theorem 5] The theorem states the stability constant C(α) = 1/√α, but the continuous proof derives ∥µ_{α,1}−µ_{α,2}∥ ≤ (1/α)(Σ_k ∥T_{ν_k}∥)δ. The discrete proof first obtains (σ_max(A)/α)δ and then asserts the SVD bound δ/(2√α), so the rate 1/√α is not established by the arguments given. The statement and proof need to be aligned, since the quantitative stability estimate is one of the advertised theoretical results.
  4. [4.2.3] The numerical reconstruction of f is inconsistent with formula (4.14). Equation (4.14) divides by ∫_0^T h(τ) sin((T−τ)√λ_k)/√λ_k dτ, while the numerical section writes "the theoretical relationship" with ar U_k equal to f_k times Σ_j A_k(t_j)Δt and then defines f^rec_k = ar U_k. Unless that time integral equals 1, which is not stated, the computed f_k are not the modal coefficients of f. This needs to be corrected or the normalization made explicit.
  5. [4.2.2, proof of Theorem 8(2)] The claimed proof that I_{kl} ≠ 0 for every algebraic T is not valid. The Lindemann–Weierstrass/Vandermonde argument concerns a family of algebraically independent parameters T_j, not a fixed algebraic T, and it does not establish the desired non-degeneracy. Moreover, for T = π (an algebraic number), I_{kl} = 0 for all k ≠ l, so the statement "for any algebraic T, I_{kl} ≠ 0" is false. The numerical setting uses T = 1, so the non-degeneracy of the covariance kernel used in the reconstruction is not proved.
minor comments (5)
  1. [3.1.1] Several occurrences of "2viu" should read "2iνu"; the same typo appears in the sentences defining H1 and H2.
  2. [Definition 2 vs. 3.1.1] Definition 2 states that Tν maps L²([0,1]) into L²([0,1]), but the numerical design places observation points on I = [−1.2,−0.2] ∪ [1.2,2.2]. The theoretical framework should either use a data space L²(I) or explain how the kernel and observations are extended.
  3. [Theorem 2(2)] The null-space example µ(y) = cos(8νy) is not shown to satisfy Tνµ = 0 at ν = π/2; the displayed integral does not vanish by the argument given. Replace it with a correct construction, e.g., functions orthogonal to 1, cos(4νy), and sin(4νy).
  4. [Assumption 2] The text says the jump time interval follows a Poisson distribution; in a Poisson process the number of jumps in an interval is Poisson-distributed while the interarrival times are exponential. The wording should be corrected.
  5. [Figures 1–6] The figure captions are highly repetitive and do not identify which subplot corresponds to which true strength function or which noise level. Please expand the captions and label the axes.

Circularity Check

3 steps flagged · score 6.0 of 10

The Helmholtz numerical experiments invert the same cos²(2ν|x−y|) operator used to generate the synthetic data, while the Lévy g-reconstruction inserts the true sign via code-level phase correction; the claimed predictions are therefore partly self-consistency checks rather than tests against the stochastic models.

  1. self definitional [Section 3.1.1, equations (3.11)-(3.12), then Definition 2 and the Hobs generation sentence.]
    "In order to maximize the retention of the characteristics of the real part while eliminating the influence of the imaginary part on strength recovery, making the recovery process more stable, and taking into account the forms of the two coefficients, we use discretization and summation to approximate the integral H(x, ν, µ) ≈ ∆y Σ cos2(2ν | x − yj |)µ(yj). ... Definition 2 (Direct Problem). ... Kν(x, y) = cos2(2ν|x − y|). The observed data Hobs(x, ν, µ) in the above equation is generated using different frequency data ν and the known true strength distribution."

    The physical model gives H1(x,ν,µ) = ∫ cos²(ν|x−y|)µ(y)dy and H2(x,ν,µ) = ∫ cos(2ν|x−y|)µ(y)dy. The next display replaces these by the kernel cos²(2ν|x−y|) without a derivation; cos²(2νr) = (1+cos(4νr))/2 is not an approximation of cos²(νr) or cos(2νr). Definition 2 then makes the substituted kernel the forward operator Tν, and the reported Hobs is generated from the known true strength with the same framework. If Hobs is generated by the discretized cos²(2ν|x−y|) operator, the Tikhonov inversion is inverting the very operator that produced the data, so the numerical reconstructions and Theorems 2-6 certify a self-defined operator rather than the stochastic Helmholtz inverse source problem.

  2. fitted input called prediction [Section 4.2.2 formula (4.15), Remark 7, and the phase correction in the numerical implementation.]
    "It should be noted that since the covariance of the observed data is insensitive to the sign of g(x), the sign of its modal coefficient gk is ambiguous. Therefore, we perform phase correction in the code to obtain results that are more consistent with the actual situation, but this usually requires additional prior information. However, in practical applications, since the true sign is unknown, such direct correction is often infeasible, necessitating the reconstruction of g2 or the use of other methods to address the issue of sign uncertainty."

    Formula (4.15) determines only the products gkgl from the covariance data, so after fixing the known diagonal factors it identifies |gk| but not the sign of gk. The numerical code then applies a 'phase correction' that sets the signs equal to those of the true g used in the forward simulation. The reported reconstructed g curve therefore contains the target sign as an inserted input rather than as an output of the data. The paper itself concedes that the true sign is unknown in practice and that such direct correction is often infeasible, which means the g-reconstruction shown in the experiments is not a free prediction from final-time data.

1 more flagged steps
  1. other [Theorem 8(2) and Remark 7.]
    "(2) If T is an arbitrary algebraic number, then the source term g is uniquely determined by Cov(uk(T ), ul(T )); k, l∈ N. ... since the covariance of the observed data is insensitive to the sign of g(x), the sign of its modal coefficient gk is ambiguous."

    The theorem states uniqueness of g from the covariance data, but the covariance identifies only the products gkgl via (4.15). The proof shows Ikl ≠ 0, i.e., that the products are recoverable; it does not determine the signs of the individual gk. The remark immediately following the numerical design acknowledges this sign ambiguity. Thus Theorem 8(2) is stronger than its premises and, to be true, would need an unstated sign prior. The numerical section obtains a full signed g only by using the target signs through phase correction, so the uniqueness claim and the reported reconstruction both depend on input information that the stated data do not contain.

full rationale

The paper contains two genuine reductions-by-construction. First, in the Helmholtz section the authors correctly derive the model-based features (3.11) and (3.12), but they then replace the kernels cos²(ν|x−y|) and cos(2ν|x−y|) by cos²(2ν|x−y|) without proof, define the forward operator Tν with that new kernel, and generate the synthetic observations from the known true strength using this same framework. Consequently the reported multi-frequency Tikhonov reconstructions are self-consistency checks of an operator chosen by the authors, not tests of the stochastic Helmholtz inverse source problem; the ill-posedness and convergence theorems describe the chosen operator, not the physical model. Second, in the Lévy part formula (4.15) identifies only gkgl, so the sign of g is unidentifiable from the covariance; the code's phase correction inserts the true sign, and Remark 7 admits this. The numerical g-reconstructions therefore use the target sign as an input. The references to [20], [26], and [28] are not self-citations by the present authors, and no load-bearing self-citation chain is present. Overall, the central claim that the methods reconstruct f and g from final-time data is only partially supported: f follows directly from the mean data under b=0, while g is recovered only up to sign unless the true sign is supplied, and the Helmholtz numerical validation is circular with respect to the operator definition.

Assumptions & free parameters 6 free parameters · 7 assumptions · 1 invented entities

The central claims rest on several model assumptions (Gaussian-Poisson jump structure, known Lévy parameters, non-degenerate h) and on a feature function that is disconnected from the forward model. The numerical tests additionally rely on oracle sign information and on unspecified regularization and frequency choices.

free parameters (6)
  • Tikhonov regularization parameter α = not reported
    Appears in J(µ) and in the g-reconstruction objective; no selection rule or value is given in the text.
  • Set of frequencies {ν_k} = ν=1:3, 1:4, 2:11 in experiments
    The number and spacing of frequencies are chosen ad hoc; the claim that 2-4 frequencies suffice for simple sources is empirical.
  • Number of spectral modes K = not specified
    The spectral truncation for the Lévy reconstruction in Section 4.2.3 is not stated, but errors grow with mode count.
  • Lévy parameters (b, σ, λ_p, σ_j²) = assumed known
    Formula (4.15) requires these to be known to compute g; if unknown, the covariance data cannot be rescaled.
  • Sign of g_k (phase correction) = set from the true source in the code
    Remark 7 states sign ambiguity is resolved by a phase correction using additional prior information, and the numerical experiments use the true sign.
  • Noise variance σ_ε² = σ=0.001 and 0.005 in experiments
    The observation noise level is treated as known in the covariance model.
assumptions (7)
  • domain assumption Assumption 1: the random source f is a real-valued, centrally symmetric, locally isotropic Gaussian random field of order -s with covariance symbol μ(x)|ξ|^{-s}.
    Invoked in Section 2.1 to give the source representation f = sqrt(μ)(-Δ)^{-s/4}W' and to obtain Sobolev regularity.
  • domain assumption Assumption 2: jump amplitudes are i.i.d. N(0, σ_j²) and jump times follow a Poisson process with rate λ_p.
    Stated in Section 2.3 and used in the stability estimate and the covariance formula (4.15).
  • domain assumption The time-dependent source h is non-degenerate, with support of positive measure, and for the uniqueness result in Theorem 8(1) h is monotone.
    Used in Section 4.2 to ensure the denominator in (4.14) can be inverted.
  • domain assumption The drift coefficient b is zero for the reconstruction of f in (4.14).
    Imposed in Section 4.2.1; without b=0, the mean of the final-time data also involves the unknown g.
  • standard math Itô isometry for stochastic integrals.
    Used in the derivation of the variance of u(x) in (3.9) and in the covariance of the diffusion part.
  • standard math Campbell's theorem for compound Poisson processes.
    Used in the jump part of the stability estimate in Section 4.1 to convert a stochastic sum into an integral.
  • standard math Lindemann-Weierstrass theorem for algebraic numbers.
    Invoked in the proof of Theorem 8(2) to argue that certain exponential sums do not vanish for algebraic T.
invented entities (1)
  • Feature function H(x,ν,µ) with kernel cos²(2ν|x−y|)
    purpose: Serves as the forward operator for the numerical inverse problem in Definition 2 and the discretization in Section 3.1.1.
    This object is not equal to any second-order statistic of the stochastic Helmholtz solution derived in (3.11)-(3.12); its introduction is not justified by the physical model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Inverse source problems for the stochastic wave equations." pith.science (2026). https://pith.science/paper/4HMAENQE

@misc{pith2026250701789,
  author       = {Pith},
  title        = {Pith review of: Inverse source problems for the stochastic wave equations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4HMAENQE}},
  note         = {Machine review of arXiv:2507.01789}
}
abstract

To address the ill-posedness of the inverse source problem for the one-dimensional stochastic Helmholtz equations without attenuation, this study develops a novel computational framework designed to mitigate this inherent challenge at the numerical implementation level. For the stochastic wave equation driven by a finite-jump L\'evy process (assuming that its jump amplitude obeys a Gaussian distribution and the jump time interval obeys a Poisson distribution), this paper firstly establish the existence of a mild solution to its direct problem satisfying a particular stability estimate. Building upon these theoretical foundations, we further investigate the well-posedness of the inverse problem and develop a methodology to reconstruct the unknown source terms $f$ and $g$ using the data of the wave field at the final time point $u(x,T)$. This work not only provides rigorous theoretical analysis and effective numerical schemes for solving inverse source problems in these two specific classes of stochastic wave equations, but also offers new perspectives and methodological approaches for addressing a broader range of wave propagation inverse problems characterized by non-Gaussian stochastic properties. The proposed framework demonstrates significant relevance for characterizing physical phenomena influenced by jump-type stochastic perturbations, offering promising applications in diverse domains including but not limited to seismic wave propagation analysis and financial market volatility modeling.

Figures

Figures reproduced from arXiv: 2507.01789 by the authors.

Figure 1
Figure 1. Single Fourier mode i) and ii) correspond to (a) and (b) in [PITH_FULL_IMAGE:figures/full_fig_p022_1.png] view at source ↗
Figure 2
Figure 2. Multiple Fourier mode 4. Stochastic wave equations driven by finite-jump L´evy processes In this section, we analyze the direct problem of a stochastic wave equation driven by a finite￾jump L´evy process and obtain stability estimates for its mild solutions. Subsequently, we discuss the inverse problem and reconstruct the source terms f(x) and g(x) based on a spectral decompo￾sition method using global modes. 4.1. D… view at source ↗
Figure 3
Figure 3. Reconstructed effect with noise σ = 0.005 [PITH_FULL_IMAGE:figures/full_fig_p035_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Reconstructed effect with noise σ = 0.001 σ 0.001 0.005 f 0.0060 0.0088 g 0.0022 0.0069 [PITH_FULL_IMAGE:figures/full_fig_p035_4.png]
Figure 5
Figure 5. Figure 5: Reconstructed effect with noise σ = 0.005 [PITH_FULL_IMAGE:figures/full_fig_p036_5.png]
Figure 6
Figure 6. Figure 6: Reconstructed effect with noise σ = 0.001 σ 0.001 0.005 f 0.0033 0.0064 g 0.0370 0.0410 [PITH_FULL_IMAGE:figures/full_fig_p036_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

31 extracted references · 28 canonical work pages

  1. [1]

    Sobolev spaces, volume 140

    Robert A Adams and John JF Fournier. Sobolev spaces, volume 140. Elsevier, 2003

  2. [2]

    Application of inverse source concepts to photoacoustic tomography

    Mark A Anastasio, Jin Zhang, Dimple Modgil, and Patrick J La Rivi` ere. Application of inverse source concepts to photoacoustic tomography. Inverse Problems, 23(6):S21, 2007

  3. [3]

    L´ evy processes and stochastic calculus

    David Applebaum. L´ evy processes and stochastic calculus. Cambridge university press, 2009

  4. [4]

    Transcendental number theory

    Alan Baker. Transcendental number theory. Cambridge university press, 2022

  5. [5]

    An inverse source problem for maxwell’s equations in magnetoencephalography

    Gang Bao, Habib Ammari, and John L Fleming. An inverse source problem for maxwell’s equations in magnetoencephalography. SIAM Journal on Applied Mathematics , 62(4):1369– 1382, 2002

  6. [6]

    Inverse random source scattering problems in several dimensions

    Gang Bao, Chuchu Chen, and Peijun Li. Inverse random source scattering problems in several dimensions. SIAM/ASA Journal on Uncertainty Quantification , 4(1):1263–1287, 2016

  7. [7]

    Inverse random source scattering for elastic waves

    Gang Bao, Chuchu Chen, and Peijun Li. Inverse random source scattering for elastic waves. SIAM Journal on Numerical Analysis , 55(6):2616–2643, 2017

  8. [8]

    Stability for the inverse source problems in elastic and electromagnetic waves

    Gang Bao, Peijun Li, and Yue Zhao. Stability for the inverse source problems in elastic and electromagnetic waves. Journal de Math´ ematiques Pures et Appliqu´ ees, 134:122–178, 2020

Show all 31 references
  1. [9]

    A multi-frequency inverse source problem

    Gang Bao, Junshan Lin, and Faouzi Triki. A multi-frequency inverse source problem. Journal of Differential Equations , 249(12):3443–3465, 2010

  2. [10]

    An inverse source problem with multiple frequency data

    Gang Bao, Junshan Lin, and Faouzi Triki. An inverse source problem with multiple frequency data. Comptes Rendus. Math´ ematique, 349(15-16):855–859, 2011

  3. [11]

    Numerical solution of the inverse source problem for the helmholtz equation with multiple frequency data

    Gang Bao, Junshan Lin, Faouzi Triki, et al. Numerical solution of the inverse source problem for the helmholtz equation with multiple frequency data. Contemp. Math , 548:45–60, 2011. 37

  4. [12]

    Imaging and time reversal in random media

    Liliana Borcea, George Papanicolaou, Chrysoula Tsogka, and James Berryman. Imaging and time reversal in random media. Inverse Problems, 18(5):1247–1279, 2002

  5. [13]

    Increasing stability in the inverse source problem with many frequencies

    Jin Cheng, Victor Isakov, and Shuai Lu. Increasing stability in the inverse source problem with many frequencies. Journal of Differential Equations , 260(5):4786–4804, 2016

  6. [14]

    Nonuniqueness in inverse source and scattering problems

    A Devaney and G Sherman. Nonuniqueness in inverse source and scattering problems. IEEE Transactions on Antennas and Propagation , 30(5):1034–1037, 2003

  7. [15]

    The inverse problem for random sources

    AJ Devaney. The inverse problem for random sources. Journal of Mathematical Physics , 20(8):1687–1691, 1979

  8. [16]

    Inverse source and scattering problems in ultrasonics

    Anthony J Devaney. Inverse source and scattering problems in ultrasonics. IEEE transactions on sonics and ultrasonics , 30(6):355–363, 1983

  9. [17]

    Convex analysis and variational problems

    Ivar Ekeland and Roger Temam. Convex analysis and variational problems . SIAM, 1999

  10. [18]

    Regularization of inverse problems, volume 375

    Heinz Werner Engl, Martin Hanke, and Andreas Neubauer. Regularization of inverse problems, volume 375. Springer Science & Business Media, 1996

  11. [19]

    Increasing stability in acoustic and elastic inverse source problems

    Mozhgan Entekhabi and Victor Isakov. Increasing stability in acoustic and elastic inverse source problems. SIAM Journal on Mathematical Analysis , 52(5):5232–5256, 2020

  12. [20]

    An inverse source problem for the stochastic wave equation

    Xiaoli Feng, Meixia Zhao, Peijun Li, and Xu Wang. An inverse source problem for the stochastic wave equation. arXiv preprint arXiv:2101.04744 , 2021

  13. [21]

    Real analysis: modern techniques and their applications

    Gerald B Folland. Real analysis: modern techniques and their applications . John Wiley & Sons, 1999

  14. [22]

    A stochastic galerkin method for the direct and inverse

    Ning Guan, Dingyu Chen, Peijun Li, and Xinghui Zhong. A stochastic galerkin method for the direct and inverse. Communications in mathematical sciences , 22(2):563–581, 2024

  15. [23]

    Discrete inverse problems: insight and algorithms

    Per Christian Hansen. Discrete inverse problems: insight and algorithms . SIAM, 2010

  16. [24]

    Numerical solution of two point boundary value problems

    Herbert B Keller. Numerical solution of two point boundary value problems . SIAM, 1976

  17. [25]

    Linear integral equations, volume 82

    Rainer Kress. Linear integral equations, volume 82. Springer, 1999

  18. [26]

    Inverse scattering problem for a two dimensional random potential

    Matti Lassas, Lassi P¨ aiv¨ arinta, and Eero Saksman. Inverse scattering problem for a two dimensional random potential. Communications in mathematical physics , 279:669–703, 2008

  19. [27]

    Inverse source problems for the stochastic wave equations: far-field patterns

    Jianliang Li, Peijun Li, and Xu Wang. Inverse source problems for the stochastic wave equations: far-field patterns. SIAM Journal on Applied Mathematics , 82(4):1113–1134, 2022

  20. [28]

    An inverse random source problem for the one-dimensional helmholtz equation with attenuation

    Peijun Li and Xu Wang. An inverse random source problem for the one-dimensional helmholtz equation with attenuation. Inverse Problems, 37(1):015009, 2020. 38

  21. [29]

    Inverse random source scattering for the helmholtz equation with attenuation

    Peijun Li and Xu Wang. Inverse random source scattering for the helmholtz equation with attenuation. SIAM Journal on Applied Mathematics , 81(2):485–506, 2021

  22. [30]

    Stability on the inverse random source scattering problem for the one-dimensional helmholtz equation

    Peijun Li and Ganghua Yuan. Stability on the inverse random source scattering problem for the one-dimensional helmholtz equation. Journal of Mathematical Analysis and Applications , 450(2):872–887, 2017

  23. [31]

    Methods of modern mathematical physics: Functional anal- ysis, volume 1

    Michael Reed and Barry Simon. Methods of modern mathematical physics: Functional anal- ysis, volume 1. Gulf Professional Publishing, 1980. 39

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.