Optimized Schwarz methods for heterogeneous heat transfer problems

Liu-Di Lu; Martin J. Gander; Tingting Wu

REVIEW 3 major objections 5 minor 22 references

Optimized Schwarz methods for heterogeneous heat transfer problems

T0 review · 3 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 heterogeneous heat problems, the best local transmission conditions must scale with both diffusion coefficients.

desk verdict Solid scaling insight and clean analysis for Versions I and II, but the two-parameter Version III overclaims: the equioscillation system has no solution for some admissible parameters, so the theorem needs repair. read the letter →

arxiv 2505.22103 v1 pith:7KRMZ7RZ submitted 2025-05-28 math.NA cs.NA

Martin J. Gander , Liu-Di Lu , Tingting Wu This is my paper

classification math.NAcs.NA MSC 65M5565M1235K0565Y05

keywords optimizedSchwarzmethodswaveformrelaxationheterogeneousheatequationtransmissionconditionsconvergencefactormin-maxproblemequioscillationdomaindecomposition

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 studies optimized Schwarz waveform-relaxation iterations for the heat equation when the diffusion coefficient jumps between subdomains, as in layered thermal protection materials. It derives the iteration's convergence factor and shows that the optimal transmission operators are nonlocal in time, so it replaces them with three local (constant-in-time) approximations. The central claim is that the approximation must be scaled with the diffusion coefficient of the opposite subdomain: the version that uses $\sigma_1=\sqrt{\nu_2}p$ and $\sigma_2=\sqrt{\nu_1}q$, with parameters from a three-point equioscillation condition (equal worst-case values at the two band endpoints and at the interior maximum), stays fast and robust even when the coefficient ratio is $10^4$, while the version scaled with only one coefficient degrades badly. A sympathetic reader should care because this gives analytically computed, implementable interface conditions for a practically important class of heterogeneous heat problems.

What carries the argument

The load-bearing object is the Laplace-domain convergence factor (Equation (8)) for the two-subdomain iteration, together with the min-max problem (P) that minimizes its worst case over the frequency interval $[\tilde{\omega}_1,\tilde{\omega}_2]$. The three versions differ in how the real transmission parameters scale with the diffusion coefficients: Version I uses $\sigma_1=\sigma_2=\sqrt{\nu_2}p$; Version II uses $\sigma_1=\sqrt{\nu_2}q$ and $\sigma_2=\sqrt{\nu_1}q$; Version III uses $\sigma_1=\sqrt{\nu_2}p$ and $\sigma_2=\sqrt{\nu_1}q$. The argument proceeds by restricting the parameter ranges (Lemmas 3.1 and 3.4), locating the local maxima of $\rho$ in $\tilde{\omega}$ (Lemmas 3.2 and 3.5), and applying equioscillation, meaning the local maxima of the convergence factor are equalized at the minimizer, with the crucial nuance that for Version I the equioscillation system is not always the true minimizer (Theorem 3.3), whereas for Version III the three-point equioscillation system (Theorem 3.5) gives the unique minimizer.

What would settle it

For a concrete two-material problem with a very large ratio, e.g., $\nu_1/\nu_2=10^6$, take $T=5$, and march the Schwarz iteration with Version III parameters from Theorem 3.5 for several time steps (say $\Delta t=10^{-2},10^{-3},10^{-4}$). If a direct parameter sweep over $p,q$ finds a pair with a smaller measured worst-case convergence factor, or if the iteration needs more iterations than some non-optimized fixed parameter as $\Delta t$ shrinks, the claimed optimality over the frequency band fails. More directly, compute $\max_{s\in i\mathbb{R}} \rho(s,p^*,q^*)$ numerically and compare with the equioscillation value; if the maximum is attained at a frequency outside $[\tilde{\omega}_1,\tilde{\omega}_2]$ or exceeds the predicted value, the band assumption is violated.

Watch

Extended reading notes

Core claim

On the paper's own terms: after Laplace transforming the error equations for two half-line subdomains, the convergence factor is $\rho(\tilde{\omega},\sigma_1,\sigma_2)$, a product of two rational factors. The paper solves, for each of three local parameter scalings, the min-max problem of minimizing the worst-case convergence factor over the frequency band $\tilde{\omega}\in[\sqrt{\pi/(4T)},\sqrt{\pi/(2\Delta t)}]$. Version I (one-coefficient scaling) requires a case split in $\mu=\sqrt{\nu_1/\nu_2}$ and, for large $\mu$, equioscillation of the two endpoint values is not always the minimizer; sometimes an interior maximum $R_c$ dominates, so the minimizer is non-unique. Version II (symmetric scaling) yields the single parameter $q^*=\sqrt{2\tilde{\omega}_1\tilde{\omega}_2}$. Version III (two parameters, each scaled with the opposite coefficient) yields the unique pair $(p^*,q^*)$ solving $\rho(\tilde{\omega}_1,p^*,q^*)=\rho(\tilde{\omega}_2,p^*,q^*)=\rho(\sqrt{\tilde{\omega}_1\tilde{\omega}_2},p^*,q^*)$, with a closed-form root analysis and the asymptotic $p^*\approx 2\mu/(\mu-1)\tilde{\omega}_1$. Numerically, Version III converges in 6 iterations for $\nu_1/\nu_2=10^4$, where Version I needs 169.

Load-bearing premise

The whole optimization hinges on the assumption that the worst convergence mode lies on the imaginary axis of the Laplace variable and inside the frequency band set by the total time $T$ and the time step $\Delta t$; if the true worst mode has a nonzero real part or falls outside that band, the 'optimized' parameters may not be optimal.

Editorial extensions

If this is right

For large coefficient ratios ($\nu_1/\nu_2$ from $10$ to $10^4$), Version III needs far fewer iterations than Version I, and the gap grows with the ratio.
Version III's convergence is nearly insensitive to the time step $\Delta t$ for large ratios, while Versions I and II deteriorate as $\Delta t$ decreases.
The transmission parameters have closed-form or asymptotic formulas, so the method is directly implementable without solving the min-max problem numerically.
The analytical two-subdomain results carry over to multiple asymmetric subdomains, as demonstrated in the three-layer thermal-protection simulation.
When the diffusion discontinuity is small, the three versions perform similarly, so the scaling matters mainly for strongly heterogeneous media.

Reading between the lines

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

Editorial inference: because the frequency band is set by $T$ and $\Delta t$, the optimal parameters automatically adapt to the time discretization; a natural extension is to derive band parameters from a rigorous error estimate instead of the cited heuristic, which would likely produce comparable or better parameters for extreme time steps.
Editorial inference: the same min-max machinery could be applied to frequency-dependent local conditions (for example, low-order rational approximations of $\sqrt{s}$), which would shrink the convergence factor further while remaining local in time.
Editorial inference: a testable prediction is that for fixed coefficient ratio and time interval there is a crossover $\Delta t$ below which Version III's advantage over Version II becomes significant; the crossover should follow from the asymptotic $p^*\approx 2\mu/(\mu-1)\tilde{\omega}_1$ and could be checked in a table of iteration counts.
Editorial inference: the Version I equioscillation caveat suggests that for two-parameter versions, interior maxima could also dominate in some parameter regimes; a complete characterization would require proving that the root of equation (22) is the only real root in $I_p$ for all $\mu$, $\tilde{\omega}_1$, and $\tilde{\omega}_2$, which the paper verifies only numerically.

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.

Desk Editor's Note

read the letter

Worth reading, with a caveat: the scaling insight is real, but the flagship Theorem 3.5 overclaims. The paper derives local Robin parameters for optimized Schwarz for heterogeneous heat equation, and the key message—both diffusion coefficients should enter the transmission parameters—is well supported by analysis and experiments. Versions I and II have clean proofs and explicit formulas. Version III, the two-parameter version, is the centerpiece, and the numerical experiments show it is indeed the best, especially for large coefficient ratios.

The soft spot is the proof of Theorem 3.5. The equioscillation system (22) is claimed to have a unique root p* in I_p for every μ>1. That is not true as stated. For μ=2, ω̃1=1, ω̃2=4, the ratio R(p) on the left of (22) is >1 over the whole interval I_p ≈ [1.236, 2.828], with the only root below the interval. So the system has no admissible solution. The proof leans on Lemma 3.5, whose sign analysis of (21) is justified by numerical observation and Figure 4, and on Remark 3.1, which essentially asserts the root existence numerically. This is a load-bearing gap: the 'unique minimizer' claim for Version III is not established for the full parameter range. The paper's own asymptotic formula p* ≈ 2μ/(μ−1) ω̃1 is for the small-Δt regime, and the experiments all lie in that regime, so the practical tables may survive. But the theorem needs a condition on ω̃2/ω̃1 (or a different argument) before the Version III parameters can be called analytically optimal.

Other soft spots are minor: the η=0 reduction and the frequency band [sqrt(π/(4T)), sqrt(π/(2Δt))] are heuristic, taken from [9], and no code or data are shipped. These are common in the subfield and not disqualifying.

Bottom line: the paper deserves a serious referee. The scaling principle and the clean treatment of Versions I and II are useful, and the flaw in Theorem 3.5 is fixable. A referee should push for a corrected version that either restricts the parameter range or proves the equioscillation system has a solution under stated conditions.

Referee Report

3 major / 5 minor

Summary. The paper analyzes nonoverlapping optimized Schwarz waveform relaxation methods for the heterogeneous heat equation with a discontinuous diffusion coefficient. After a Laplace transform in time, the authors derive the convergence factor (4) and observe that the optimal transmission operators are nonlocal in time. They then propose three families of local (constant-in-time) Robin parameters, each scaled differently with respect to the two diffusion coefficients. For each version they formulate a min-max problem over a heuristic frequency band and derive, or claim to derive, the optimized parameters. Version I uses one parameter scaled by one diffusion coefficient, Version II uses one parameter scaled by both coefficients, and Version III uses two independent parameters, one per subdomain, each scaled by the opposite coefficient. Numerical experiments in one space dimension, including a three-subdomain thermal-protection-system example, support the main practical claim that scaling the transmission conditions appropriately with respect to both coefficients yields faster and more robust convergence, especially for large coefficient ratios.

Significance. If the analytical results are correct, the paper provides useful practical guidance for choosing Robin transmission parameters in heterogeneous heat transfer simulations. The derivation of the Laplace-domain convergence factor is clean, and the analyses of Versions I and II are detailed and lead to explicit formulas. The numerical study is systematic and supports the conclusion that Version III is the most robust local choice, with the strength that the paper ships reproducible, clearly described experiments. The main weakness is that the centerpiece analytical statement for Version III, Theorem 3.5, is not rigorously established and appears to be false for part of the stated parameter range; this undermines the claim that the Version III parameters are analytically optimal. Because the numerical experiments use the small-time-step regime in which the asymptotic formula is plausible, the practical conclusions may survive, but the theorem and its proof need substantial revision.

major comments (3)

[§3.3, Theorem 3.5 and Eq. (22)] The theorem claims that for every μ>1 the unique minimizer pair of (P3) is obtained from the equioscillation system, and the proof asserts that Eq. (22) has a unique real root p* in I_p=[ω1(√(μ^2+1)−(μ−1)), √(2ω1ω2)]. This existence claim is supported only by Remark 3.1 and Figure 5, which is not a proof. More seriously, the claim is false as stated: for μ=2, ω1=1, ω2=4, the interval I_p is approximately [1.236, 2.828], and for R(p)=A(p,1)A(p,4)/A(p,2)^2 with A(p,ω)=((p−ω)^2+ω^2)/((p+μω)^2+μ^2ω^2), the values are R(1.236)≈1.036, R(1.5)≈1.314, and R(2.828)≈2.72, so R(p)>1 throughout I_p; the only root of R(p)=1 lies near p≈1.14, below I_p. Thus the equioscillation system has no admissible solution for these parameters. The theorem needs an explicit condition on ω2/ω1, or a different argument, before the Version III parameters can be called the unique minimizers, and the claimed 'unique minimizer' wording must be corrected.
[§3.3, Lemma 3.5] The characterization of the local maxima of the convergence factor for Version III is not proven rigorously. The proof states that 'in practice for common choices of ωj we numerically find that the convergence factor behaves as in Figure 4' and then uses this numerical observation to conclude that Eq. (21) must be negative. The contradiction argument only shows that the assumed shape of Figure 4 is incompatible with the alternative sign pattern; it does not establish the shape from the definition of ρ. Since Lemma 3.5 is the basis for the three-point equioscillation used in Theorem 3.5, this is a load-bearing gap. A rigorous proof should identify, for the full admissible parameter region, the sign of the second polynomial in (20) at ω^2=pq/2 and the behavior of the maxima, without relying on a single numerical figure.
[§3, problem (P) and preceding paragraph] The min-max analysis is carried out with the Laplace variable restricted to the imaginary axis (η=0) and over the heuristic frequency band ω̃∈[√(π/(4T)), √(π/(2Δt))], citing [9] for the constant-coefficient case. This reduction is not proved for the heterogeneous problem, and the worst-case convergence mode for the time-domain iteration could in principle have a nonzero real part or lie outside this band. If that happens, the optimized parameters are not true minimizers of the convergence factor. The paper should state this limitation explicitly and provide a numerical check, for example by evaluating ρ(s,σ1,σ2) on a rectangle in the s-plane for representative μ and (σ1,σ2), to show that the maximum is attained on the imaginary axis and within the chosen band.

minor comments (5)

[§3.2, paragraph before Theorem 3.4] The sentence 'we can find a unique optimized transmission parameter p' should refer to q, since the free parameter in Version II is denoted q throughout this subsection.
[§3.3, first paragraph] There is a typo: 'funnd a unique optimized transmission parameter' should be 'found a unique optimized transmission parameter'.
[§4.1, Table 1] The text refers to 'Table 4.1' when presenting the iteration counts; the table is numbered Table 1 in the captions. The cross-reference should be corrected.
[§3.3, Theorem 3.5 proof] The statement that 'there exist closed forms for the roots of this polynomial' and that three simple solutions are 0 and ±i√(2μω1ω2) is not accompanied by the polynomial itself; writing out the polynomial or its factorization would make the claim checkable.
[§3.3, Remark 3.1] Remark 3.1 says that 'we can show numerically the graph of (22) in Figure 5' for one set of (ω1,ω2,μ) and that the behavior is similar for all numerical experiments. A remark of this kind cannot replace the missing existence and uniqueness proof in Theorem 3.5, and the phrase 'for all our numerical experiments' should not be stated as a mathematical conclusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: analytic optimization is self-contained, and numerical experiments independently validate the derived parameters.

full rationale

The paper's central derivation chain is not circular. The convergence factor (8) is derived from the Laplace-transformed error equations with the chosen Robin transmission conditions, and each optimized parameter is obtained by solving an explicit min-max problem over a stated frequency interval, not by fitting iteration counts or by assuming the numerical results. Version I and Version II parameters follow from closed-form equioscillation arguments (Theorems 3.2 and 3.4). Version III parameters are characterized as the solution of the equioscillation system in Theorem 3.5, with the proof reducing to a polynomial root condition (22); while the proof relies on a numerically observed shape of the convergence factor in Lemma 3.5 and Remark 3.1, this is an analytical gap or correctness concern, not circularity, because the numerical observation is not used as the definition of the optimized parameters and the experiments are not used to fabricate the formulas. The heuristic frequency range [ω̃1, ω̃2] and the η=0 reduction are imported from prior literature including [9], which is partly self-citation, but this is a standard modeling choice rather than a uniqueness theorem or a fitted input, and it does not presuppose the paper's main conclusion. The numerical experiments in Section 4 serve as independent validation of the analytically chosen parameters. No step was found in which an input is defined in terms of an output, a fitted parameter is renamed as a prediction, or a load-bearing conclusion reduces to an unverified self-citation.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The paper introduces no fitting parameters: the transmission parameters p and q are derived from min-max problems, not tuned to iteration data. The only hand-chosen inputs are the frequency range endpoints, which are standard heuristics from the waveform relaxation literature, and the reduction to the imaginary axis. The Version III analysis further assumes a specific shape of the convergence factor based on numerical observation. No new physical or mathematical entities are postulated.

free parameters (2)

ω̃1 = sqrt(π/(4T)) = sqrt(π/(4T))
Chosen as the lower bound of the frequency range for the min-max problem; heuristic from [9], directly enters the optimized parameter formulas.
ω̃2 = sqrt(π/(2Δt)) = sqrt(π/(2Δt))
Chosen as the upper frequency bound based on the time step; heuristic from [9], directly enters the optimized parameter formulas.

assumptions (5)

domain assumption The error analysis uses solutions decaying to zero at spatial infinity on half-line subdomains.
Invoked in Section 2.1 to write the general solutions of the Laplace-transformed error equations as decaying exponentials.
domain assumption The Laplace variable s is restricted to the imaginary axis (η=0) with a heuristic frequency range ω ∈ [π/(2T), π/Δt].
Stated in Section 3 before problem (P); this is standard practice for waveform relaxation analysis, but the reduction is not rigorously proven for the heterogeneous problem.
domain assumption The error has no constant-in-time component, so the frequency ω=0 is excluded.
Justified in Section 3 by the fact that the error vanishes at t=0, so a constant function cannot appear in the error.
ad hoc to paper For Version III, the convergence factor is assumed to have an interior local maximum at sqrt(pq/2) with the shape of Figure 4 for the frequency range of interest.
The proof of Lemma 3.5 argues by contradiction and relies on a numerical observation that the second polynomial factor is negative; the rigorous range of validity is not established.
ad hoc to paper For small time steps, the asymptotic approximation p* ≈ 2μ/(μ-1)ω̃1 is used for Version III.
Stated in Theorem 3.5 without a rigorous error bound; it is used in practice to approximate the root of the equioscillation equation (22).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Optimized Schwarz methods for heterogeneous heat transfer problems." pith.science (2026). https://pith.science/paper/7KRMZ7RZ

@misc{pith2026250522103,
  author       = {Pith},
  title        = {Pith review of: Optimized Schwarz methods for heterogeneous heat transfer problems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7KRMZ7RZ}},
  note         = {Machine review of arXiv:2505.22103}
}

read the original abstract

We present here nonoverlapping optimized Schwarz methods applied to heat transfer problems with heterogeneous diffusion coefficients. After a Laplace transform in time, we derive the error equation and obtain the convergence factor. The optimal transmission operators are nonlocal, and thus inconvenient to use in practice. We introduce three versions of local approximations for the transmission parameter, and provide a detailed analysis at the continuous level in each case to identify the best local transmission conditions. Numerical experiments are presented to illustrate the performance of each local transmission condition. As shown in our analysis, local transmission conditions, which are scaled appropriately with respect to the heterogeneous diffusion coefficients, are more efficient and robust especially when the discontinuity of the diffusion coefficient is large.

Figures

Figures reproduced from arXiv: 2505.22103 by the authors.

**Figure 2.** 2D illustration of the decomposition. 2 Model problem To model the heat transfer between different materials as shown in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Illustration of the convergence factor ρ as a function of ωe with different values of the parameter p. Left: p ∈ Ic. Right: p ∈ Ir. (i) if kr > h2(µ), then one value of the parameter p minimizing the convergence factor is p ∗ = p 2µωe1ωe2 ∈ Ic. This optimized parameter p ∗ is unique when ρ(ωe1, p∗ ) ≥ Rc. Otherwise, the minimum of the convergence factor is also attained for any p chosen in a closed interval around… view at source ↗

Figures from the paper (7 more)

**Figure 4.** Figure 4: Illustration of the convergence factor with respect to [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]

**Figure 5.** Figure 5: Illustration of the left and rights part in ( [PITH_FULL_IMAGE:figures/full_fig_p021_5.png]

**Figure 6.** Figure 6: Convergence behavior of the three local transmission conditions [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]

**Figure 7.** Figure 7: Comparison of the convergence factor ρ with respect to the frequency ω˜ for all three versions. Left: ν1 ν2 = 10. Right: ν1 ν2 = 102 [PITH_FULL_IMAGE:figures/full_fig_p023_7.png]

**Figure 8.** Figure 8: Convergence behavior of the three local transmission conditions [PITH_FULL_IMAGE:figures/full_fig_p024_8.png]

**Figure 9.** Figure 9: Convergence behavior of the three local transmission conditions [PITH_FULL_IMAGE:figures/full_fig_p025_9.png]

**Figure 10.** Figure 10: Solution of the heat distribution within a thermal protection [PITH_FULL_IMAGE:figures/full_fig_p026_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

[9]

Gander and Thibault Lunet.Time Parallel Time Integration

M.J. Gander and Thibault Lunet.Time Parallel Time Integration. So- ciety for Industrial and Applied Mathematics, Philadelphia, PA, 2024

work page 2024
[1]

Anderson.Hypersonic and high temperature gas dynamics

J.D. Anderson.Hypersonic and high temperature gas dynamics. New York, McGraw-Hill, 1989

work page 1989
[2]

Bennequin, M.J

D. Bennequin, M.J. Gander, L. Gouarin, and L. Halpern. Optimized Schwarz waveform relaxation for advection reaction diffusion equations in two dimensions.Numerische Mathematik, 134:513–567, 2016

work page 2016
[3]

Birken, M.J

P. Birken, M.J. Gander, and N. Kotarsky. Continuous analysis of wave- form relaxation for heterogeneous heat equations. Submitted, 2024. 27

work page 2024
[4]

Divo, A.J

E. Divo, A.J. Kassab, and F. Rodriguez. Parallel domain decomposition approach for large-scale three-dimensional boundary-element models in linear and nonlinear heat conduction.Numerical Heat Transfer, Part B, 44:417–437, 2003

work page 2003
[5]

Dubois and M.J

O. Dubois and M.J. Gander. Optimized Schwarz methods for a diffusion problem with discontinuous coefficient.Numerical Algorithms, 69:109– 144, 2015

work page 2015
[6]

W.Z. Feng, K. Yang, M. Cui, and X.W. Gao. Analytically-integrated radial integration bem for solving three-dimensional transient heat con- duction problems.International Communications in Heat and Mass Transfer, 79:21–30, 2016

work page 2016
[7]

M.J. Gander. Schwarz methods over the course of time.Electronic Transactions on Numerical Analysis, 31:228–255, 2008

work page 2008

Show all 22 references

[8]

Gander, F

M.J. Gander, F. Kwok, and B.C. Mandal. Dirichlet–Neumann and Neumann–Neumann waveform relaxation algorithms for parabolic prob- lems.Electronic Transactions on Numerical Analysis, 45:424–456, 2016

2016
[10]

Gander and T

M.J. Gander and T. Vanzan. Heterogeneous optimized Schwarz methods for second order elliptic PDEs.SIAM Journal on Scientific Computing, 41(4):A2329–A2354, 2019

2019
[11]

Gerardo-Giorda and F

L. Gerardo-Giorda and F. Nataf. Optimized Schwarz methods for un- symmetric layered problems with strongly discontinuous and anisotropic coefficients.Journal of Numerical Mathematics, 13(4):265–294, 2005

2005
[12]

Girault and D

M. Girault and D. Petit. Identification methods in nonlinear heat con- duction. part I: model reduction.International Journal of Heat and Mass Transfer, 48:105–118, 2005

2005
[13]

Girault and D

M. Girault and D. Petit. Identification methods in nonlinear heat con- duction. part II: inverse problem using a reduced model.International Journal of Heat and Mass Transfer, 48:119–133, 2005

2005
[14]

Goldfeld, L.F

P. Goldfeld, L.F. Pavarino, and O.B. Widlund. Balancing Neumann– Neumann preconditioners for mixed approximations of heterogeneous problems in linear elasticity.Numerische Mathematik, 95:283–324, 2003. 28

2003
[15]

Kumar and S.P

S. Kumar and S.P. Mahulikar. Selection of materials and design of multilayer lightweight passive thermal protection system.Journal of Thermal Science and Engineering Applications, 8:021003, 2016

2016
[16]

Kumar and S.P

S. Kumar and S.P. Mahulikar. Design of thermal protection system for reusable hypersonic vehicle using inverse approach.Journal of Spacecraft and Rockets, 54(2):436–446, 2017

2017
[17]

Lemarié, L

F. Lemarié, L. Debreu, and E. Blayo. Toward an optimized global- in-time Schwarz algorithm for diffusion equations with discontinuous and spatially variable coefficients. Part 1: the constant coefficients case. Electronic Transactions on Numerical Analysis, 40:148–169, 2013

2013
[18]

Lemarié, L

F. Lemarié, L. Debreu, and E. Blayo. Toward an optimized global- in-time Schwarz algorithm for diffusion equations with discontinuous and spatially variable coefficients. Part 2: the variable coefficients case. Electronic Transactions on Numerical Analysis, 40:170–186, 2013

2013
[19]

Maday and F

Y. Maday and F. Magoulès. Optimized Schwarz methods without over- lap for highly heterogeneous media.Computer Methods in Applied Me- chanics and Engineering, 196(8):1541–1553, 2007. Domain Decomposi- tion Methods: recent advances and new challenges in engineering

2007
[20]

H.A. Schwarz. Über einen Grenzübergang durch alternierendes Ver- fahren.Vierteljahrsschrift der Naturforschenden Gesellschaft in Zürich, 15:272–286, 1870
[21]

Uyanna and H

O. Uyanna and H. Najafi. Thermal protection systems for space vehi- cles: A review on technology development, current challenges and future prospects.Acta Astronautica, 176:341–356, 2020

2020
[22]

Zhang, J

B. Zhang, J. Mei, M. Cui, X.-W. Gao, and Y. Zhang. A general approach forsolvingthree-dimensionaltransientnonlinearinverseheatconduction problems in irregular complex structures.International Journal of Heat and Mass Transfer, 140:909–917, 2019. 29

2019

Pith tools

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

[1] [9]

Gander and Thibault Lunet.Time Parallel Time Integration

M.J. Gander and Thibault Lunet.Time Parallel Time Integration. So- ciety for Industrial and Applied Mathematics, Philadelphia, PA, 2024

work page 2024

[2] [1]

Anderson.Hypersonic and high temperature gas dynamics

J.D. Anderson.Hypersonic and high temperature gas dynamics. New York, McGraw-Hill, 1989

work page 1989

[3] [2]

Bennequin, M.J

D. Bennequin, M.J. Gander, L. Gouarin, and L. Halpern. Optimized Schwarz waveform relaxation for advection reaction diffusion equations in two dimensions.Numerische Mathematik, 134:513–567, 2016

work page 2016

[4] [3]

Birken, M.J

P. Birken, M.J. Gander, and N. Kotarsky. Continuous analysis of wave- form relaxation for heterogeneous heat equations. Submitted, 2024. 27

work page 2024

[5] [4]

Divo, A.J

E. Divo, A.J. Kassab, and F. Rodriguez. Parallel domain decomposition approach for large-scale three-dimensional boundary-element models in linear and nonlinear heat conduction.Numerical Heat Transfer, Part B, 44:417–437, 2003

work page 2003

[6] [5]

Dubois and M.J

O. Dubois and M.J. Gander. Optimized Schwarz methods for a diffusion problem with discontinuous coefficient.Numerical Algorithms, 69:109– 144, 2015

work page 2015

[7] [6]

W.Z. Feng, K. Yang, M. Cui, and X.W. Gao. Analytically-integrated radial integration bem for solving three-dimensional transient heat con- duction problems.International Communications in Heat and Mass Transfer, 79:21–30, 2016

work page 2016

[8] [7]

M.J. Gander. Schwarz methods over the course of time.Electronic Transactions on Numerical Analysis, 31:228–255, 2008

work page 2008