Pith. sign in

REVIEW 2 major objections 5 minor 32 references

Numerical Approximation and Analysis of the Inverse Robin Problem Using the Kohn-Vogelius Method

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read For piecewise constant Robin coefficients, the FEM-discretized Kohn-Vogelius functional comes with provable, near noise-optimal reconstruction rates.

desk verdict Elliptic part is a genuine contribution and the parabolic part has a repair-worthy gap in the proof of the advertised convergence rate. read the letter →

arxiv 2506.07370 v1 pith:OAODLIXJ submitted 2025-06-09 math.NA cs.NA

Erik Burman , Siyu Cen , Bangti Jin , Zhi Zhou This is my paper · ORCID
classification math.NAcs.NA MSC 65N2165N3035R3065M60
keywords inverseRobinproblemKohn-Vogeliusvariationalmethodfiniteelementconditionalstabilitypiecewiseconstantcoefficienterrorestimategradedmeshparabolic
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

This paper aims to prove that a Galerkin finite-element discretization of the Kohn-Vogelius variational method recovers a piecewise constant Robin coefficient on the inaccessible part of the boundary, with reconstruction error controlled explicitly by the noise level $\delta$, the mesh size $h$, and (in the parabolic case) the time step $\tau$. The target guarantee matches the conditional stability of the continuous inverse problem: for a known partition, choosing $h\sim\delta^{2/3}$ and $\alpha\sim\delta^{-4/3}$ gives an $L^\infty(\Gamma_i)$ error of order $O(\delta^{1-\epsilon})$, and for an unknown partition the same bound holds with a H\"older exponent. The parabolic analysis adds new large-time stability estimates that justify recovering a time-independent coefficient from terminal boundary data. A sympathetic reader would care because this turns a widely used heuristic functional into a provably convergent numerical algorithm with explicit parameter choices.

What carries the argument

The engine is the regularized Kohn-Vogelius functional $J_\alpha(q)=\|\nabla u_N-\nabla u_D\|^2_{L^2(\Omega)}+\|q^{1/2}(u_N-u_D)\|^2_{L^2(\Gamma_i)}+\alpha\|u_N-u_D\|^2_{L^2(\Omega)}$, where $u_N$ solves the Neumann problem and $u_D$ solves the mixed problem with Dirichlet data on $\Gamma'_a$; the extra $\alpha$-weighted $L^2(\Omega)$ term supplies the $H^1(\Omega)$ control needed to estimate the boundary trace of $u_N-u_D$. The Galerkin discretization uses meshes graded near the singular vertices of the mixed boundary value problem, restoring optimal approximation rates for the Ritz projection and the Lagrange interpolant. The argument then runs a fixed chain: Lemma 2.2 bounds the finite-element and noise perturbation of the direct solves, Lemma 2.3 bounds $J_{\alpha,h}(q^*)$ by $c(\eta+\alpha\eta^2)$, and the Lipschitz stability estimate [28] (known partition) or the H\"older stability estimate [14] (unknown partition) converts the resulting boundary-data difference into the $L^\infty(\Gamma_i)$ coefficient error. In the parabolic case the same chain is run with backward Euler in time and spectral decay estimates replacing elliptic regularity.

What would settle it

Run the discrete Kohn-Vogelius algorithm on the class-$A$ example $q^\dagger(x_2)=1+\chi_{[0.2,1]}(x_2)$ of Example 4.1(i), rescaling $h\sim\delta^{2/3}$ and $\alpha\sim\delta^{-4/3}$ as $\delta$ decreases; if the measured $L^\infty(\Gamma_i)$ error does not decay like $O(\delta^{1-\epsilon})$ or stalls above that rate, the theorem's bound fails. A complementary check uses a coefficient with one segment of length below $c_0$ (outside class $B$): the proof's H\"older stability estimate then does not apply, and observing non-convergence as $\delta\to0$ would confirm that the structural assumption is necessary rather than technical.

Watch

Extended reading notes

Core claim

The central claim is Theorem 2.1: for the elliptic inverse Robin problem, if $q^\dagger$ belongs to the admissible class $A$ (piecewise constant on an a priori known partition), then every global minimizer $q^*$ of the discrete Kohn-Vogelius functional satisfies $\|q^*-q^\dagger\|_{L^\infty(\Gamma_i)} \le c\big((\alpha^{-1/4}+\alpha^{-1/2})(\eta^{1/2}+\alpha^{1/2}\eta)+\eta^{3/4}\big)$ with $\eta=h^{2(1-\epsilon)}+\delta^{\frac43(1-\epsilon)}$, and the choice $h\sim\delta^{2/3}$, $\alpha\sim\delta^{-4/3}$ yields $O(\delta^{1-\epsilon})$. For class $B$ (unknown partition) the same bound holds with an additional H\"older exponent $\kappa\theta<1$ coming from the conditional stability estimate. The parabolic analogue, Theorem 3.2, adds a time step $\tau$ to the bound and rests on new large-time stability estimates, Theorem 3.1, for the parabolic inverse Robin problem. The proof controls the data-fitting part of the functional at $q^\dagger$ through finite-element approximation and noise bounds, uses the minimizer property to transfer this control to $q^*$, converts smallness of the functional into smallness of $u(q^*)-u(q^\dagger)$ on the measured boundary, and finally applies Lipschitz or H\"older conditional stability to pass from data to coefficient.

Load-bearing premise

The load-bearing premise is that the true Robin coefficient lies in the structured piecewise-constant classes $A$ or $B$—either on a known partition, or on an unknown partition with a fixed lower bound on segment length and an upper bound on the number of pieces—because the Lipschitz and H\"older stability estimates imported into the proof are only known for such coefficients.

Editorial extensions

If this is right

  • For known-partition piecewise constant coefficients, the elliptic reconstruction converges at the near-optimal rate $O(\delta^{1-\epsilon})$ with $h\sim\delta^{2/3}$ and $\alpha\sim\delta^{-4/3}$.
  • For unknown partitions the method still converges, but only at a H\"older rate whose exponent is implicit; the parameter choices remain valid but the observed rate must be measured numerically.
  • In the parabolic case, the new large-time stability estimates justify recovering a time-independent Robin coefficient from terminal boundary data for sufficiently large $T$, with the full discrete scheme converging as $\tau\sim\delta^{4/3}$, $h\sim\delta^{2/3}$, $\alpha\sim\delta^{-4/3}$.
  • The discrete optimization problem always has a global minimizer, since the admissible coefficient sets are finite-dimensional, so the error bounds apply to a well-defined object without any extra discretization of the coefficient.
  • The error bound separates the roles of data noise, finite-element discretization, and time stepping, so a user can balance these error sources deliberately rather than treating the method as a black box.

Reading between the lines

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

  • If the continuous stability estimates were extended to smooth or space-time dependent Robin coefficients, the same discretization analysis would immediately supply convergence rates for those classes; the missing ingredient is the stability estimate, not the numerical analysis.
  • The scaling $h\sim\delta^{2/3}$ is coarser than the $h\sim\delta$ one might guess from the noise level; this reflects the regularity of the direct problem and suggests a practical mesh-selection rule that could be validated on other variational inverse problems.
  • The very slow observed rate $O(\delta^{0.16})$ in the parabolic unknown-partition example indicates that the H\"older exponent $\kappa\theta$ is quite small in practice, so users should regard the parabolic case as qualitative convergence unless the partition is known.
  • A natural testable extension is to reconstruct $q$ together with the flux $g$ or the initial data $u_0$ in the same Kohn-Vogelius framework; the main obstacle is the absence of a comparable conditional stability estimate for the enlarged parameter set.
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

2 major / 5 minor

Summary. The paper studies the numerical recovery of a piecewise constant Robin coefficient in the elliptic problem (1.1) and the parabolic problem (3.1) from boundary Cauchy data on a subset of the accessible boundary. The authors minimize a Kohn-Vogelius functional, (2.8) and (3.3), augmented by an L2(Ω) penalty with parameter α, after discretizing the state equations by piecewise linear finite elements on graded meshes; the parabolic scheme is fully discrete with backward Euler in time. The main theoretical results, Theorem 2.1 and Theorem 3.2, bound the L∞(Γ_i) error of a discrete global minimizer in terms of the noise level δ, mesh size h, time step τ, and penalty α, for coefficients in the known-partition class A and the unknown-partition class B. The proofs combine a priori bounds on the discrete functionals with the conditional stability estimates of Sincich and of Hu-Yamamoto, together with two new large-time stability estimates for the parabolic problem in Theorem 3.1. Numerical experiments in Section 4 illustrate convergence rates for elliptic and parabolic cases in both partition classes.

Significance. If the stated results were fully established, this would be one of the first quantitative convergence-rate guarantees for a FEM-discretized Kohn-Vogelius reconstruction of piecewise constant Robin coefficients, and the parabolic stability estimates in Theorem 3.1 would be a useful contribution in their own right. The paper is careful to separate discretization error from the conditional stability of the continuous inverse problem, and the graded-mesh construction is well motivated by the singularity decomposition in Proposition 2.1. The numerical studies cover both known and unknown partition classes and report rates consistent with the theoretical predictions. However, the proof of the parabolic convergence theorem currently rests on a regularity assumption that is neither stated nor generally satisfiable, and the elliptic analysis contains a smaller but real noise-regularity gap; until these are repaired, the advertised rates are not established as written.

major comments (2)
  1. [Section 2.3, Lemma 2.2, Eq. (2.14)] In bounding ∥G∥_{H^{1+ε}(Ω)} ≤ c ∥z^δ_h − I^∂_h f∥_{H^{1/2+ε}(Γ'_a)}, the proof uses the noise assumption (2.12) at s = 1/2 + ε, but Assumption 2.1 and (2.12) only cover −1/2 ≤ s ≤ 1/2. The estimate ∥z^δ_h − f∥_{H^{1/2+ε}(Γ'_a)} ≤ c δ^{2/3(1−ε)} is therefore not a consequence of the stated assumptions. This gap propagates into Lemma 2.2, the a priori bound in Lemma 2.3, and the elliptic error estimate in Theorem 2.1. A repair is likely possible by combining (2.12) at s = 1/2 with an inverse inequality and the scaling h ∼ δ^{2/3}, but the manuscript should either carry out that argument or state the additional noise regularity needed, rather than silently extending (2.12).
  2. [Assumption 3.1(i)] The compatibility condition in Assumption 3.1(i) is written as '∂_n u0 + q u0 = 0 on Γ_i' without specifying which coefficient q is meant. If q denotes the true coefficient q†, then the assumption does not apply to the state u(T; q∗) that is used in Theorem 3.2, because the minimizer q∗ need not coincide with q† on Γ_i. If the condition is intended to hold for every admissible q, that is a substantial additional restriction that is never stated or verified. This ambiguity is directly connected to the gap in the proof of Lemma 3.2 for q = q∗; the assumption should be reformulated precisely and the consequences for the error analysis traced explicitly.
minor comments (5)
  1. [Throughout] There are several typographical errors: 'a prioriestimates' in the abstract and Introduction, 'Cauchy-Schwartz' in the proof of Lemma 2.2, 'wth' in Section 3.2, and 'Lipschitz domain' in Lemma 2.1 should read 'Lipschitz domain'.
  2. [Remark 2.1] The construction z^δ_h = P_{h,Γ'_a} z^δ verifies condition (2.12) only under the scaling h ∼ δ^{2/3}; the remark should state this scaling explicitly before it is used in the derivation of Theorem 2.1's parameter choice.
  3. [Theorem 3.1 and Section 3] The symbol f is used both for the exact data u(q†)|_{Γ'_a} in Assumption 2.1/3.1 and for the pair f_i = u_i|_{Γ'_a×{T}} in Theorem 3.1; please disambiguate these notations.
  4. [Theorem 3.1(ii)] The dependence of κ is written as κ(c_g, c_q, c_q, c_0, Ω); the second c_q should presumably be ar c_q, the upper bound on the Robin coefficient.
  5. [Section 4.1, Tables 1 and 2] The 'rate' column reports values such as O(δ^{1.03}); since these are empirical rates over a finite set of δ values, the tables should specify that the rates are least-squares fits over the displayed noise levels.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the convergence rates are derived from external stability estimates plus independent FEM error analysis.

full rationale

The paper's derivation is not circular. Theorem 2.1 and Theorem 3.2 bound the reconstruction error by combining (i) conditional stability estimates imported from Sincich [28] and Hu-Yamamoto [14] for the continuous inverse problem, (ii) finite-element and time-discretization error estimates for the direct problems (Lemmas 2.2, 3.2, 3.3), and (iii) an a priori bound on the Kohn-Vogelius functional evaluated at any global minimizer (Lemmas 2.3 and 3.4). Each ingredient is proven from the stated data assumptions (Assumptions 2.1 and 3.1) rather than from the target error bound. The stability estimates are external and independent of the discrete scheme. The algorithmic parameters h, tau, and alpha are chosen a priori as functions of the noise level delta (Remarks 2.2 and 3.1), and no fitted quantity is renamed as a prediction. Self-citations ([18], [19], [20], [21], [22]) appear only for algorithmic context or for a standard contour-integral representation of the backward-Euler solution in Appendix A; they are not the source of the convergence claims. The one notable issue is a non-circular proof gap in the parabolic section: Lemma 3.2 is proved using the norm of A u0 in L2, which requires u0 in D(A), while Assumption 3.1 only gives u0 in H^{2-epsilon}, and Theorem 3.2 applies Lemma 3.2 with q = q* whose compatibility condition is not established. This is a correctness risk for the parabolic rate, but it does not make the derivation equivalent to its inputs and is not a circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

There are no fitted physical constants or invented mechanisms. The only chosen parameters are algorithmic (α, h, τ, grading exponent) and are explicit in the error bounds; the stability and regularity theorems are imported from prior literature, which is standard practice in numerical analysis and not circular.

free parameters (4)
  • penalty parameter α = α ~ δ^{-4/3}
    Chosen by hand to balance the α∥u_N,h − u_δ_D,h∥^2 term against the energy mismatch; all error bounds keep α explicit (Remarks 2.2 and 3.1).
  • mesh size h = h ~ δ^{2/3}
    Selected in experiments and rate discussion to equilibrate finite element error and noise; enters η = h^{2(1−ε)} + δ^{4/3(1−ε)}.
  • time step τ = τ ~ δ^{4/3}
    Backward Euler step chosen to match noise scaling in the parabolic case; appears in Theorem 3.2 and Remark 3.1.
  • mesh grading exponent r = r ∈ (0, min(1, π/(2ω_j)))
    Mesh grading parameter in (2.5) that controls the FEM interpolation error near mixed boundary vertices; the optimal rates require r in this range.
assumptions (5)
  • standard math Regularity decomposition of mixed boundary value solutions (Proposition 2.1)
    Says solutions to (1.1) and (2.1) are H^2 outside a finite set of singular functions; used in Lemmas 2.2, 2.3, 3.3 and Appendix A; taken from Grisvard and Mghazli.
  • domain assumption Sincich Lipschitz stability for q ∈ A and Hu-Yamamoto Hölder stability for q ∈ B
    Black-box stability estimates [28, Thm 2.4] and [14, Thm 4.1] convert data misfit into Robin coefficient error; the paper's error bounds inherit their Lipschitz and Hölder exponents.
  • domain assumption Noise model in (2.12) and (3.2)
    Assumes noisy data satisfies ∥zδ_h − f∥_{H^s(Γ'_a)} ≤ cδ^{1−2s/3} for s ∈ [−1/2, 1/2]; this is what turns δ into the explicit noise terms in η.
  • domain assumption Admissible classes A and B for q: piecewise constant with known or unknown partition, box bounds, minimum segment length c0, bounded N
    The stability theorems and the compactness argument for existence of a discrete minimizer are stated only for these classes (Equations (2.2) and (2.3)).
  • domain assumption Parabolic data assumptions: u0 ∈ H^{2−ε}(Ω) with compatibility, g ∈ H^{1/2}(Γ_a), f ∈ H^2(Γ'_a)
    Assumption 3.1 is needed for the regularity of solutions and for Lemma 3.2's error estimates for the backward Euler FEM.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Numerical Approximation and Analysis of the Inverse Robin Problem Using the Kohn-Vogelius Method." pith.science (2026). https://pith.science/paper/OAODLIXJ

@misc{pith2026250607370,
  author       = {Pith},
  title        = {Pith review of: Numerical Approximation and Analysis of the Inverse Robin Problem Using the Kohn-Vogelius Method},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OAODLIXJ}},
  note         = {Machine review of arXiv:2506.07370}
}
read the original abstract

In this work, we numerically investigate the inverse Robin problem of recovering a piecewise constant Robin coefficient in an elliptic or parabolic problem from the Cauchy data on a part of the boundary, a problem that commonly arises in applications such as non-destructive corrosion detection. We employ a Kohn-Vogelius type variational functional for the regularized reconstruction, and discretize the resulting optimization problem using the Galerkin finite element method on a graded mesh. We establish rigorous error estimates on the recovered Robin coefficient in terms of the mesh size, temporal step size and noise level. This is achieved by combining the approximation error of the direct problem, a priori estimates on the functional, and suitable conditional stability estimates of the continuous inverse problem. We present several numerical experiments to illustrate the approach and to complement the theoretical findings.

Figures

Figures reproduced from arXiv: 2506.07370 by the authors.

Figure 1
Figure 1. The reconstruction of Examples 4.1 and 4.2 with δ = 1% noise. Top row: reconstructions for Example 4.1(i)–(iii). Bottom row: reconstructions for Example 4.2(i)–(ii). The next example is about the parabolic case. Example 4.2 (Parabolic case). Consider the time independent boundary illumination g(x1, x2) = 1 − x 2 1 . To obtain an initial data u0 satisfying Assumption 3.1, let u0 solve the following elliptic problem w… view at source ↗
Figure 2
Figure 2. The optimization dynamics and reconstructions for Examples [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 28 canonical work pages

  1. [1]

    S. C. Brenner and L. R. Scott.The Mathematical Theory of Finite Element Methods. Springer, New York, third edition, 2008

    2008

  2. [2]

    Buttazzo and R

    G. Buttazzo and R. V . Kohn. Reinforcement by a thin layer with oscillating thickness.Appl. Math. Optim., 16(3):247–261, 1987

    work page 1987

  3. [3]

    K. Cao, D. Lesnic, and J. Liu. Simultaneous reconstruction of space-dependent heat transfer coeffi- cients and initial temperature.J. Comput. Appl. Math., 375:112800, 18, 2020

    work page 2020

  4. [4]

    Chaabane, C

    S. Chaabane, C. Elhechmi, and M. Jaoua. A stable recovery method for the Robin inverse problem. Math. Comput. Simulation, 66(4-5):367–383, 2004

    work page 2004

  5. [5]

    Chaabane, I

    S. Chaabane, I. Feki, and N. Mars. Numerical reconstruction of a piecewise constant Robin parameter in the two- or three-dimensional case.Inverse Problems, 28(6):065016, 19, 2012

    work page 2012

  6. [6]

    Chaabane and M

    S. Chaabane and M. Jaoua. Identification of Robin coefficients by the means of boundary measure- ments.Inverse Problems, 15(6):1425–1438, 1999

    work page 1999

  7. [7]

    D.-H. Chen, D. Jiang, I. Yousept, and J. Zou. Variational source conditions for inverse Robin and flux problems by partial measurements.Inverse Probl. Imaging, 16(2):283–304, 2022

    work page 2022

  8. [8]

    H. W. Engl, M. Hanke, and A. Neubauer.Regularization of Inverse Problems. Kluwer, Dordrecht, 1996

    work page 1996

Show all 32 references
  1. [9]

    Fang and E

    W. Fang and E. Cumberbatch. Inverse problems for metal oxide semiconductor field-effect transistor contact resistivity.SIAM J. Appl. Math., 52(3):699–709, 1992

    1992

  2. [10]

    Fujita and T

    H. Fujita and T. Suzuki. Evolution problems. InFinite Element Methods (Part 1), volume 2 ofHand- book of Numerical Analysis, pages 789–928. Elsevier, 1991

    1991

  3. [11]

    Grisvard.Elliptic Problems in Nonsmooth Domains

    P. Grisvard.Elliptic Problems in Nonsmooth Domains. Pitman, Boston, MA, 1985

    1985

  4. [12]

    D. N. H `ao, P. X. Thanh, and D. Lesnic. Determination of the heat transfer coefficients in transient heat conduction.Inverse Problems, 29(9):095020, 21, 2013

    2013

  5. [13]

    B. Harrach. Uniqueness, stability and global convergence for a discrete inverse elliptic Robin trans- mission problem.Numer. Math., 147(1):29–70, 2021

    2021

  6. [14]

    Hu and M

    G. Hu and M. Yamamoto. H ¨older stability estimate of Robin coefficient in corrosion detection with a single boundary measurement.Inverse Problems, 31(11):115009, 20, 2015

    2015

  7. [15]

    G. Inglese. An inverse problem in corrosion detection.Inverse Problems, 13(4):977–994, 1997

    1997

  8. [16]

    Ito and B

    K. Ito and B. Jin.Inverse Problems: Tikhonov Theory and Algorithms. World Scientific, Hackensack, NJ, 2015

    2015

  9. [17]

    Jaoua, S

    M. Jaoua, S. Chaabane, C. Elhechmi, J. Leblond, M. Mahjoub, and J. R. Partington. On some robust algorithms for the Robin inverse problem.ARIMA Rev. Afr. Rech. Inform. Math. Appl., 9:287–307, 2008

    2008

  10. [18]

    B. Jin. Conjugate gradient method for the Robin inverse problem associated with the Laplace equation. Internat. J. Numer. Methods Engrg., 71(4):433–453, 2007. 24

    2007

  11. [19]

    Jin and X

    B. Jin and X. Lu. Numerical identification of a Robin coefficient in parabolic problems.Math. Comp., 81(279):1369–1398, 2012

    2012

  12. [20]

    Jin and Z

    B. Jin and Z. Zhou.Numerical Treatment and Analysis of Time-Fractional Evolution Equations, vol- ume 214 ofApplied Mathematical Sciences. Springer, Cham, 2023

    2023

  13. [21]

    Jin and J

    B. Jin and J. Zou. Numerical estimation of piecewise constant Robin coefficient.SIAM J. Control Optim., 48(3):1977–2002, 2009

    1977

  14. [22]

    Jin and J

    B. Jin and J. Zou. Numerical estimation of the Robin coefficient in a stationary diffusion equation. IMA J. Numer. Anal., 30(3):677–701, 2010

    2010

  15. [23]

    R. V . Kohn and M. V ogelius. Relaxation of a variational method for impedance computed tomography. Comm. Pure Appl. Math., 40(6):745–777, 1987

    1987

  16. [24]

    Lions and E

    J.-L. Lions and E. Magenes.Non-homogeneous Boundary Value Problems and Applications. Vol. I. Springer-Verlag, New York-Heidelberg, 1972

    1972

  17. [25]

    Z. Mghazli. Regularity of an elliptic problem with mixed Dirichlet-Robin boundary conditions in a polygonal domain.Calcolo, 29(3-4):241–267, 1992

    1992

  18. [26]

    S. G. Pyatkov and V . A. Baranchuk. Determination of the heat transfer coefficient in mathematical models of heat and mass transfer.Mat. Zametki, 113(1):90–108, 2023

    2023

  19. [27]

    A. K. Rasmussen, F. Seizilles, M. Girolami, and I. Kazlauskaite. The Bayesian approach to inverse Robin problems.ASA/SIAM J. Uncertain. Quantif., 12(3):1050–1084, 2024

    2024

  20. [28]

    E. Sincich. Lipschitz stability for the inverse Robin problem.Inverse Problems, 23(3):1311–1326, 2007

    2007

  21. [29]

    Slodi ˇcka, D

    M. Slodi ˇcka, D. Lesnic, and T. T. M. Onyango. Determination of a time-dependent heat transfer coefficient in a nonlinear inverse heat conduction problem.Inverse Probl. Sci. Eng., 18(1):65–81, 2010

    2010

  22. [30]

    Thom´ee.Galerkin Finite Element Methods for Parabolic Problems

    V . Thom´ee.Galerkin Finite Element Methods for Parabolic Problems. Springer-Verlag, Berlin, second edition, 2006

    2006

  23. [31]

    Xu and J

    Y . Xu and J. Zou. Analysis of an adaptive finite element method for recovering the Robin coefficient. SIAM J. Control Optim., 53(2):622–644, 2015

    2015

  24. [32]

    L. Zhuo, D. Lesnic, and S. Meng. Reconstruction of the heat transfer coefficient at the interface of a bi-material.Inverse Probl. Sci. Eng., 28(3):374–401, 2020. 25

    2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.