REVIEW 3 major objections 5 minor 43 references
A level-set structural approach for multi-physics joint inversion using full-waveform and gravity data
T0 review · 3 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash
Pith's one-line read Gravity data can be made to rescue the deep structures that full-waveform inversion misses, provided the density contrast of the target body is known and velocity and density share one level-set interface.
desk verdict A genuinely new algorithmic coupling of level-set FWI and gravity with a sensible weighting strategy, but the synthetic evidence is weaker than the prose suggests because the tests never leave the generative class of the model. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the shared level-set interface: one scalar field φ whose zero level set is the boundary of the anomalous body, appearing in both the velocity parameterization c(r)=c1(r)H(φ(r))+c2(r)(1−H(φ(r))) and the density parameterization ρ(r)=f(r)H(φ(r)), with H a smoothed Heaviside and δτ its derivative. It enforces structural similarity directly rather than through gradient penalties. The second mechanism is the balanced-decaying weight ω(n)=ω1(n)·ω2(n), set from the ratio of waveform and gravity gradient magnitudes in φ so that the two data misfits act comparably on the interface, and decaying exponentially so gravity dominates the early iterations and full-waveform data domina
What would settle it
Build a synthetic model with two separate anomalous bodies: one with a velocity anomaly and zero density contrast, the other with a density contrast and zero velocity anomaly. Run the proposed joint inversion with the density-contrast value of the second body imposed. If the shared level-set coupling is correct, the velocity reconstruction should ignore the density-only body; if the velocity model is pulled toward the density-only interface, the central claim fails. A second decisive experiment is to image a U-shaped or disconnected anomaly, where the uniqueness theorem's geometric conditions
Extended reading notes
Core claim
The paper claims that in the joint inversion of seismic and gravity data, the density function should be written as ρ(r)=f(r)H(φ(r)) — a known density-contrast value f times the indicator of a region D — and the velocity as c(r)=c1(r)H(φ(r))+c2(r)(1−H(φ(r))). Because the same level-set function φ carries both parameterizations, the boundary of the salt body or other anomaly is the single object being inverted. The authors further claim that with the weight ω(n)=ω1(n)ω2(n), where ω1 scales the gravity gradient to the waveform gradient and ω2 decays exponentially, the gravity data first establish the large-scale deep structure and the waveform data then refine it. This reverses the usual one-w
Load-bearing premise
The whole scheme hinges on two priors: the density contrast f(r) is known ahead of time, and the wave-velocity anomaly and the density anomaly are exactly the same body with one common boundary; if either fails in the field—say a velocity-only sediment boundary or an unknown salt density—the gravity data can pull the reconstructed interface to a false location.
Editorial extensions
If this is right
- Deep and laterally extended structures that full-waveform inversion alone cannot resolve become recoverable from surface data, because gravity with a known density contrast supplies large-scale information through the shared interface.
- The recovered velocity model remains nearly correct even when gravity noise degrades the density reconstruction, so the velocity and density outputs complement each other.
- The method requires no explicit velocity-density formula; structural coincidence replaces petrophysical composition, which is useful where rock-physics relations are unknown.
- The interface-penalty and reinitialization steps keep the level-set evolution stable, so the joint inversion can be run from a very general initial guess.
Reading between the lines
- Inference: The real test of the shared-interface assumption is geology where velocity and density boundaries do not coincide; in such settings a single level-set function would either flatten the velocity image or create a spurious interface, and a multi-level-set extension would be needed.
- Inference: Because the density contrast f is imposed a priori, the method does not estimate the magnitude of the density anomaly; treating f as unknown with its own regularization would make the method applicable where salt or basement densities are not well constrained.
- Inference: The balanced-decaying weight is essentially a continuation schedule on the data terms; the same schedule could be ported to other multi-physics inversions, with the early dataset chosen as the one providing the fastest large-scale constraint.
- Inference: Since the uniqueness theorem's geometric conditions are relaxed in the algorithm, a U-shaped or disconnected anomaly is the natural stress test for whether the gravity information remains reliable outside the proven regime.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a level-set structural method for the joint inversion of full-waveform and gravity data. The velocity and density fields are parameterized by a single level-set function and a known density-contrast function f(r), with the interior velocity c1(r) often fixed a priori. A balanced and exponentially decaying weight is introduced to let gravity dominate early iterations and full-waveform data dominate later. The method is tested on four synthetic examples, including a 2D SEG/EAGE salt model, and compared with isolated FWI and gravity inversion. The central claim is that the joint inversion integrates the strengths of both datasets and produces superior reconstructions, especially for deep or extended salt structures.
Significance. If the claims are robust, the paper offers a practically attractive way to impose structural coupling between velocity and density: instead of penalty terms, the shared interface is represented explicitly, and prior information such as f and c1 is inserted directly. The use of an existing open-source FWI engine (Deepwave) and the clear statement of Algorithm 1 are strengths. The main significance, however, is conditional: the four synthetic tests are generated inside the model class defined by Eq. (6), with f and c1 prescribed, so the numerical evidence does not yet establish robustness to the central structural assumption. The theoretical justification via Theorem 2.1 is also invoked in a regime the theorem does not cover.
major comments (3)
- [Section 2.2] Theorem 2.1 is the paper's principal justification for extracting 'unambiguous information' from gravity data, but the algorithm uses only the vertical component g_z (Eq. (5)) rather than the full modulus |∇U|, and it explicitly drops the geometric conditions (star-shaped, convex, or convex-in-one-direction) required by the theorem. No proof, counterexample analysis, or numerical test is given to show that the uniqueness benefit survives these relaxations. In addition, Example 3 uses f(r)=(1.8-z)*0.2, which is not constant and may violate condition (3) of Theorem 2.1 depending on the chosen direction. The theoretical claim that gravity contributes meaningful information therefore needs either an extended theorem, a precise statement of the regularity regime in which the method operates, or substantially softened wording.
- [Sections 5.2.1-5.2.4] All synthetic true models satisfy the shared-interface parameterization (6) exactly, with the same f(r) and c1(r) used as known priors in the inversion. The waveform and gravity data are generated with the same numerical solvers and the same spatial mesh employed in the inversion (an inverse-crime configuration). The claim in Section 5.2.3 that the joint inversion 'produces superior solutions that closely match the true model' is therefore a demonstration within the exact generative class, not a test of the method's robustness. No experiment considers a velocity-only or density-only boundary, a misspecified f, an unknown c1, or a different forward discretization. The conclusions in Section 6 should be reframed as proof-of-concept, or additional out-of-class experiments should be added.
- [Equation (27)] The gravity gradient term in Eq. (27) is f(r) δτ(φ) times the gravity residual. If the structural assumption fails or f is misspecified, this term still moves the shared interface φ, and through Eq. (6) it changes the velocity reconstruction via (c1-c2)δτ(φ). Thus a density anomaly with no velocity counterpart, or an incorrect density contrast, will not merely fail to help; it can actively create a spurious velocity interface. The manuscript does not analyze this mechanism or provide experiments that bound the resulting bias. Given that the paper's central novelty is the shared-interface coupling, this is a load-bearing robustness question.
minor comments (5)
- [Tables 2, 3, 4, 5] The notation α_{v2} appears to mean α_{c2}, and λ_{c1} is absent although the optimization problem in Eq. (16) includes λ_{c1} E_{c1}. Please clarify whether c1 is always frozen and, if so, state this consistently in the text and tables.
- [Section 3.2.3] The terms '2-1 norm TV regularization' and '1-1 norm TV regularization' are nonstandard as written. Please define them explicitly, e.g., as the l2 norm of the gradient magnitude or the sum of absolute partial derivatives, so that the distinction is clear.
- [Section 3.2.1] There is a typo: 'repeatedlly' should be 'repeatedly'.
- [Section 5.2.4] The SSIM comparison is only reported for the velocity model; an SSIM or similar metric for the density reconstruction would help support the claim of complementary information in the noisy case, where the density result is visibly degraded (Figure 16(f)).
- [Section 6] The limitations paragraph is candid about slow convergence, hyperparameter selection, and Deepwave scalability. It would be useful to also explicitly note that the shared-interface prior and the prescription of f and c1 are untested when violated, since that is the main limitation of the numerical evidence.
Circularity Check
No significant circularity: the joint inversion uses explicit priors and independent forward models; the cited uniqueness theorem is external (Isakov) and the adaptive weight is a scale-balancing heuristic.
full rationale
The paper's derivation chain does not reduce to its own inputs. The level-set coupling (Eq. 6) is an explicit structural prior: 'we assume that the wave velocity c(r) and the anomalous density ρ(r) have the same interface structure' (Section 3.1). The two data misfits (Eq. 11) come from genuinely different physical forward models: the acoustic wave equation (Eq. 1) and the gravity integral (Eq. 5). The shared level-set function φ is estimated by minimizing both, so the recovered interface is not defined by a fitted parameter renamed as a prediction. The balanced-decaying weight (Eqs. 30–33) is computed from current gradient magnitudes, not from the final answer; it is a scale-balancing heuristic and does not encode the recovered model. Theorem 2.1 is cited from Isakov's book [16] (external), with [26] and [5] as additional pointers; the uniqueness theorem is an external mathematical result, and the paper explicitly relaxes its geometric requirements ('we will not enforce these geometric constraints on D', Section 2.2), so it is not used to force the conclusion. The numerical tests construct true models inside the assumed form of Eq. (6) and impose the same density-contrast f and interior velocity c1 as priors (Algorithm 1 'freeze the density-contrast value f based on prior information'). This limits external validity and robustness to misspecified structure, as the skeptic notes, but it is not circular: the paper declares those values as priors rather than claiming to predict them. The acknowledged limitations in Section 6 (slow convergence, many hyper-parameters, Deepwave scalability) are flagged by the authors and are unrelated to circularity. No load-bearing argument reduces to a self-citation or to an equation's own definition.
Assumptions & free parameters
free parameters (6)
- Density contrast f(r) =
2.0 g/cm^3 (Examples 1-2); (1.8 - z)*0.2 g/cm^3 (Examples 3-4)
- Velocity value inside the anomaly c1(r) =
4.0 km/s (Ex1), 4.348 km/s (Ex2), 4.482 km/s (Ex3-4)
- Gravity lead-in weight omega0 =
5 (all examples)
- Decay exponent lambda =
ln(50)/n_max, n_max = 2e4
- Regularization weights lambda_phi, lambda_c1, lambda_c2 =
e.g., lambda_phi = 5e-5 or 2e-6, lambda_c2 = 1e-4 (Tables 2-5)
- Smoothing width tau and optimizer step sizes =
tau = grid size; epsilon and alpha_phi set per example
assumptions (4)
- domain assumption The acoustic wave equation (1) with PML and the gravity integral (5) with vertical gz are adequate forward models for the subsurface.
- domain assumption Theorem 2.1 (well-posedness of inverse gravimetry) is valid, and its guarantees transfer to the relaxed algorithm even though D is not enforced to be star-shaped, convex, or convex in one direction.
- domain assumption Velocity c and density rho share the same interface and can be represented by a single level-set function (Eq. 6).
- domain assumption Using only the vertical component gz, with known f(r), provides enough information to guide the inversion when combined with waveform data.
Cite this review
Pith. "Pith review of A level-set structural approach for multi-physics joint inversion using full-waveform and gravity data." pith.science (2026). https://pith.science/paper/DCQWWC2V
@misc{pith2026250906689,
author = {Pith},
title = {Pith review of: A level-set structural approach for multi-physics joint inversion using full-waveform and gravity data},
year = {2026},
howpublished = {\url{https://pith.science/paper/DCQWWC2V}},
note = {Machine review of arXiv:2509.06689}
}
read the original abstract
This paper presents a level-set based structural approach for the joint inversion of full-waveform and gravity data. The joint inversion aims to integrate the strengths of full-waveform inversion for high resolution imaging and gravity inversion for detecting density contrasts over extensive regions. Although common studies typically only observe full-waveform inversion assisting gravity inversion, we propose three key points that enable gravity data to complement full-waveform data in the joint inversion. (i) Based on the well-posedness theorem, we consider a volume mass distribution where the density-contrast value is imposed as a priori information, ensuring that the gravity data provide meaningful information. (ii) We utilize a level-set formulation to characterize the shared interface of wave velocity and density functions, connecting multi-physics datasets via the structural similarity of their inversion parameters. (iii) We develop a balanced and decaying weight to regulate the influence of multi-physics datasets during joint inversion. This weight comprises a balanced part that accounts for the differing scales of full-waveform and gravity data, and a decaying part designed to effectively utilize the features and advantages of each dataset.
Figures
Figures from the paper (14 more)
Reference graph
Works this paper leans on
-
[1]
Aria Abubakar, G Gao, Tarek M Habashy, and J Liu. Joint inversion approaches for geophysical electromagnetic and elastic full-waveform data.Inverse problems, 28(5):055016, 2012
work page 2012
-
[2]
Afnimar, Kazuki Koketsu, and Koichi Nakagawa. Joint inversion of refraction and gravity data for the three-dimensional topography of a sediment–basement interface.Geophysical Journal International, 151(1):243–254, 2002
work page 2002
-
[3]
F. Aminzadeh, J. Brac, and T. Kunz.3-Dsalt and overthrust models, SEG/EAGE 3-D Modeling Series No. 1. Society of Exploration Geophysicists, 1997
work page 1997
-
[4]
F. Birch. The velocity of compressional waves in rocks to 10 kilobars.Journal of Geophysical Research, 66:2199–2224, 1961
work page 1961
-
[5]
Yihang Chen and Wenbin Li. Learning on the correctness class for domain inverse problems of gravimetry.Machine Learning: Science and Technology, 5(3):035072, 2024
work page 2024
-
[6]
Benjamin Crestel, Georg Stadler, and Omar Ghattas. A comparative study of structural similarity and regularization for joint inverse problems governed by pdes.Inverse Problems, 35(2):024003, 2018
work page 2018
-
[7]
Paolo Dell’Aversana, Giancarlo Bernasconi, Fabio Chiappa, et al. A global integration platform for optimizing cooperative modeling and simultaneous joint inversion of multi-domain geophysical data. Aims Geosciences, 2(1):1–31, 2016
work page 2016
-
[8]
Dorn and D
O. Dorn and D. Lesselier. Level set methods for inverse scattering.Inverse Problems, 22:R67–R131, 2006
2006
Show all 43 references
-
[9]
Total variation regularization strategies in full-waveform inversion.SIAM Journal on Imaging Sciences, 11(1):376–406, 2018
Ernie Esser, Lluis Guasch, Tristan van Leeuwen, Aleksandr Y Aravkin, and Felix J Herrmann. Total variation regularization strategies in full-waveform inversion.SIAM Journal on Imaging Sciences, 11(1):376–406, 2018
2018
-
[10]
Joint two-dimensional dc resistivity and seismic travel time inversion with cross-gradients constraints.Journal of Geophysical Research: Solid Earth, 109(B3), 2004
Luis A Gallardo and Max A Meju. Joint two-dimensional dc resistivity and seismic travel time inversion with cross-gradients constraints.Journal of Geophysical Research: Solid Earth, 109(B3), 2004
2004
-
[11]
L. A. Gallardo-Delgado, M. A. P´ erez-Flores, and E. G´ omez-Trevi˜ no. A versatile algorithm for joint 3D inversion of gravity and magnetic data.Geophysics, 68:949–959, 2003
2003
-
[12]
G. H. F. Gardner, L. W. Gardner, and A. R. Gregory. Formation velocity and density-the diagnostic basics for stratigraphic traps.Geophysics, 39:770–780, 1974
1974
-
[13]
The split Bregman method for L1-regularized problems.SIAM journal on imaging sciences, 2(2):323–343, 2009
Tom Goldstein and Stanley Osher. The split Bregman method for L1-regularized problems.SIAM journal on imaging sciences, 2(2):323–343, 2009
2009
-
[14]
Joint inversion: a structural approach.Inverse problems, 13(1):63, 1997
E Haber and D Oldenburg. Joint inversion: a structural approach.Inverse problems, 13(1):63, 1997
1997
-
[15]
A joint inversion algorithm to process geoelectric and surface wave seismic data
A Hering, R Misiek, A Gyulai, T Ormos, M Dobr´ oka, and L Dresen. A joint inversion algorithm to process geoelectric and surface wave seismic data. part i: basic ideas.Geophysical Prospecting, 43:135–156, 1995
1995
-
[16]
American Mathematical Society, Providence, Rhode Island, Multi-physics joint inversion using full-waveform and gravity data33 1990
Victor Isakov.Inverse source problems. American Mathematical Society, Providence, Rhode Island, Multi-physics joint inversion using full-waveform and gravity data33 1990
1990
-
[17]
A fast local level set method for inverse gravimetry
Victor Isakov, Shingyu Leung, and Jianliang Qian. A fast local level set method for inverse gravimetry. Communications in Computational Physics, 10:1044–1070, 2011
2011
-
[18]
3-d joint inversion of seismic waveform and airborne gravity gradiometry data
Wenbin Jiang. 3-d joint inversion of seismic waveform and airborne gravity gradiometry data. Geophysical Journal International, 223(2):746–764, 2020
2020
-
[19]
Notes on perfectly matched layers (PMLs).arXiv preprint arXiv:2108.05348, 2021
Steven G Johnson. Notes on perfectly matched layers (PMLs).arXiv preprint arXiv:2108.05348, 2021
2021 arXiv
-
[20]
A convex formulation for binary tomography.IEEE Transactions on Computational Imaging, 6:1–11, 2019
Ajinkya Kadu and Tristan van Leeuwen. A convex formulation for binary tomography.IEEE Transactions on Computational Imaging, 6:1–11, 2019
2019
-
[21]
Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014
Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014
2014 arXiv
-
[22]
Efficient 1.5 d full waveform inversion in the laplace-fourier domain.Inverse Problems, 39(7):075012, 2023
Apostolos Kontakis, Diego Rovetta, Daniele Colombo, and Ernesto Sandoval-Curiel. Efficient 1.5 d full waveform inversion in the laplace-fourier domain.Inverse Problems, 39(7):075012, 2023
2023
-
[23]
Joint inversion of seismic traveltimes and gravity data on unstructured grids with application to mineral exploration.Geophysics, 77(1):K1–K15, 2012
P G Leli` evre, C G Farquharson, and C A Hurich. Joint inversion of seismic traveltimes and gravity data on unstructured grids with application to mineral exploration.Geophysics, 77(1):K1–K15, 2012
2012
-
[24]
A level-set adjoint-state method for crosswell transmission-reflection traveltime tomography.Geophysical Journal International, 199(1):348–367, 2014
Wenbin Li, Shingyu Leung, and Jianliang Qian. A level-set adjoint-state method for crosswell transmission-reflection traveltime tomography.Geophysical Journal International, 199(1):348–367, 2014
2014
-
[25]
A level set method for imaging salt structures using gravity data.Geophysics, 81(2):G35–G51, 2016
Wenbin Li, Wangtao Lu, and Jianliang Qian. A level set method for imaging salt structures using gravity data.Geophysics, 81(2):G35–G51, 2016
2016
-
[26]
Simultaneously recovering both domain and varying density in inverse gravimetry by efficient level-set methods.Inverse Problems and Imaging, 15(3):387–413, 2021
Wenbin Li and Jianliang Qian. Simultaneously recovering both domain and varying density in inverse gravimetry by efficient level-set methods.Inverse Problems and Imaging, 15(3):387–413, 2021
2021
-
[27]
Li and D
Y. Li and D. Oldenburg. 3D inversion of gravity data.Geophysics, 63:109–119, 1998
1998
-
[28]
Litman, D
A. Litman, D. Lesselier, and F. Santosa. Reconstruction of a two-dimensional binary obstacle by controlled evolution of a level-set.Inverse Problems, 14:685–706, 1998
1998
-
[29]
On the limited memory BFGS method for large scale optimization
Dong C Liu and Jorge Nocedal. On the limited memory BFGS method for large scale optimization. Mathematical programming, 45(1):503–528, 1989
1989
-
[30]
An optimal transport approach for seismic tomography: Application to 3d full waveform inversion.Inverse Problems, 32(11):115008, 2016
Ludovic M´ etivier, Romain Brossier, Quentin Merigot, Edouard Oudet, and Jean Virieux. An optimal transport approach for seismic tomography: Application to 3d full waveform inversion.Inverse Problems, 32(11):115008, 2016
2016
-
[31]
Moorkamp, B
M. Moorkamp, B. Heincke, M. Jegen, A. W. Roberts, and R. W. Hobbs. A framework for 3-D joint inversion of MT, gravity and seismic refraction data.Geophysical Journal International, 184:477–493, 2011
2011
-
[32]
Nielsen and B
L. Nielsen and B. H. Jacobsen. Integrated gravity and wide-angle seismic inversion for two-dimensional crustal modelling.Geophysical Journal International, 140:222–232, 2000
2000
-
[33]
S. J. Osher and J. A. Sethian. Fronts propagating with curvature dependent speed: algorithms based on Hamilton-Jacobi formulations.J. Comput. Phys., 79:12–49, 1988. Multi-physics joint inversion using full-waveform and gravity data34
1988
-
[34]
Springer Science & Business Media, 2006
Stanley Osher and Ronald Fedkiw.Level set methods and dynamic implicit surfaces, volume 153. Springer Science & Business Media, 2006
2006
-
[35]
R. E. Plessix. A review of the adjoint-state method for computing the gradient of a functional with geophysical applications.Geophysical Journal International, 167:495–503, 2006
2006
-
[36]
Deepwave
Alan Richardson. Deepwave. Software: https://doi.org/10.5281/zenodo.8381177, September 2023
2023 doi
-
[37]
Inversion modeling of multiple geophysical data sets for geothermal exploration: application to roosevelt hot springs area
J M Savino, W L Rodi, and J F Masso. Inversion modeling of multiple geophysical data sets for geothermal exploration: application to roosevelt hot springs area. final report. Technical report, S-Cubed, La Jolla, CA (USA), 1982
1982
-
[38]
Cooperative full waveform and gravimetric inversion.J
Raul U Silva, Jonas D De Basabe, Mrinal K Sen, Mario Gonzalez-Escobar, Enrique Gomez-Trevino, and Selene Solorza-Calderon. Cooperative full waveform and gravimetric inversion.J. Seismic Exploration, 29(6):549–573, 2020
2020
-
[39]
Sussman, P
M. Sussman, P. Smereka, and S. J. Osher. A level set approach for computing solutions to incompressible two-phase flows.Journal of computational physics, 114:146–159, 1994
1994
-
[40]
Mitigating local minima in full-waveform inversion by expanding the search space.Geophysical Journal International, 195(1):661–667, 2013
Tristan Van Leeuwen and Felix J Herrmann. Mitigating local minima in full-waveform inversion by expanding the search space.Geophysical Journal International, 195(1):661–667, 2013
2013
-
[41]
An overview of full-waveform inversion in exploration geophysics
Jean Virieux and St´ ephane Operto. An overview of full-waveform inversion in exploration geophysics. Geophysics, 74(6):WCC1–WCC26, 2009
2009
-
[42]
Fwigan: Full-waveform inversion via a physics-informed generative adversarial network.Journal of Geophysical Research: Solid Earth, 128(4):e2022JB025493, 2023
Fangshu Yang and Jianwei Ma. Fwigan: Full-waveform inversion via a physics-informed generative adversarial network.Journal of Geophysical Research: Solid Earth, 128(4):e2022JB025493, 2023
2023
-
[43]
Total variation regularization for seismic waveform inversion using an adaptive primal dual hybrid gradient method.Inverse Problems, 34(4):045006, 2018
Peng Yong, Wenyuan Liao, Jianping Huang, and Zhenchun Li. Total variation regularization for seismic waveform inversion using an adaptive primal dual hybrid gradient method.Inverse Problems, 34(4):045006, 2018
2018
Reviewed August 4, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.