Parallel multilevel methods for solving the Darcy--Forchheimer model based on a nearly semicoercive formulation

Jongho Park; S. Majid Hassanizadeh

REVIEW 3 major objections 6 minor 73 references

Parallel multilevel methods for solving the Darcy--Forchheimer model based on a nearly semicoercive formulation

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The Darcy–Forchheimer model of fast flow through porous media can be solved by a parallel multilevel optimization method whose convergence rate remains robust as the augmented Lagrangian parameter goes to zero.

desk verdict A credible, honest extension of the nearly semicoercive subspace-correction framework to Darcy–Forchheimer; the unproved kernel-decomposition assumption is the decisive thing to check. read the letter →

arxiv 2507.03192 v3 pith:3CLA2IFN submitted 2025-07-03 math.NA cs.NA

Jongho Park , S. Majid Hassanizadeh This is my paper · ORCID

classification math.NAcs.NA MSC 65N5565N2076S0590C25

keywords Darcy–ForchheimermodelmixedfiniteelementmethodsmultileveladditiveSchwarzmethodsubspacecorrectionnearlysemicoerciveconvexoptimizationaugmentedLagrangianH(div)spacefullapproximationscheme

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 sets out to give the Darcy–Forchheimer model—the standard nonlinear description of high-velocity flow through porous media—a numerical solver that is both parallel and backed by a global convergence proof. The route is to rewrite the mixed finite element discretization as a constrained convex minimization problem, then apply one iterate of the augmented Lagrangian method, which turns it into a nearly semicoercive convex program: the divergence constraint is penalized with weight $1/\epsilon$ while the nonlinear flow energy enters with weight $\epsilon$. The main claim is that the parallel multilevel additive Schwarz method built on this formulation converges with a rate that does not degrade as $\epsilon$ becomes small (Theorem 4.4), so the augmented Lagrangian parameter can be chosen small enough to make the outer iteration almost immediate without destroying the inner solver. If correct, this converts a coupled nonlinear saddle-point problem, for which monolithic multigrid solvers lack global convergence guarantees, into a hierarchy of small convex patch problems that can be solved in parallel. The numerical experiments support the claim by showing outer iteration counts that stay bounded as the mesh is refined and that remain stable as $\epsilon$ varies from $10^0$ down to $10^{-3}$.

What carries the argument

The load-bearing object is the nearly semicoercive energy $F^\epsilon(v;q_h)=\frac12\|\mathrm{div}\,v-g_h\|_{L^2}^2+\epsilon(F(v)-\int_\Omega q_h\,\mathrm{div}\,v\,dx)$ on the mixed finite element space $X_h\subset H(\mathrm{div};\Omega)\cap L^3(\Omega)^d$, meaning the energy is coercive only through the constraint residual $\|\mathrm{div}\,v-g_h\|$ and becomes increasingly ill-conditioned as $\epsilon\to0$. The method is a parallel subspace correction iteration built on the vertex-patch multilevel decomposition $X_h=\sum_{j=1}^J\sum_{k=1}^{n_j}V_{j,k}$, where each $V_{j,k}$ contains functions supported in the elements touching one interior vertex of mesh level $j$. From the current iterate, each patch solves a small convex minimization, and the patch updates are summed with a step size chosen by a backtracking line search. Convergence is carried by four conditions from the abstract nearly semicoercive theory—local smoothness, local uniform convexity, the kernel decomposition Assumption B.3, and a triangle-inequality property—together with the strengthened convexity estimate $\tau_0\gtrsim J^{-1}\simeq|\log h|^{-1}$. The kernel decomposition is the piece imported from the linear nearly singular theory: divergence-free parts of the space must split stably into the patches, and the paper invokes it for the nonlinear space $H(\mathrm{div})\cap L^3$ without a proof.

What would settle it

Run Algorithm 4.1 on the two benchmark problems for $\epsilon=10^{-3}$ on meshes $h=2^{-6},\dots,2^{-9}$ and count outer iterations to a fixed energy tolerance; a visible growth of iteration counts with $h$, or a sharp increase as $\epsilon$ decreases below $10^{-2}$, would contradict the claimed robustness. A more direct check targets Assumption B.3: take a divergence-free Raviart–Thomas field oscillating at the finest scale, and compute the infimum of $\sum_{j,k}\|w_{j,k}\|_X$ over representations $u=\sum_{j,k}w_{j,k}$ with $w_{j,k}\in V_{j,k}\cap\ker(\mathrm{div})$; if this stability constant grows like a power of the number of levels rather than like $|\log h|$, the kernel decomposition fails and Theorem 4.4 loses its foundation.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the discrete Darcy–Forchheimer problem reduces to the nearly semicoercive convex program (3.2) via the augmented Lagrangian method, and that this reduction is not just a reformulation but a practical algorithmic route. For the energy $F^\epsilon(v;q_h)=\frac12\int_\Omega(\mathrm{div}\,v-g_h)^2\,dx+\epsilon(F(v)-\int_\Omega q_h\,\mathrm{div}\,v\,dx)$, the paper proves that augmented Lagrangian iterates converge with rate $(\epsilon/(\mu+\epsilon))^{2n}$ in the symmetrized Bregman divergence, so smaller $\epsilon$ gives arbitrarily fast outer convergence. Theorem 4.4 then shows the parallel subspace correction Algorithm 4.1 preserves this: for $\tau\in(0,\tau_0]$ with $\tau_0\gtrsim J^{-1}$, large initial energy gaps are contracted by factor $1-\tau/2$ per iteration and small gaps decay at the sublinear rate $O(n^{-3})$, with the constants independent of $\epsilon$. This provides a global convergence guarantee of a type that previous monolithic multigrid approaches to this model lacked, and it is leveraged further by a backtracking line search and a full approximation scheme that reduce the practical cost without changing the convergence picture in the experiments.

Load-bearing premise

The convergence guarantee rests on an assumption, taken from the linear theory and not proved here, that every discrete divergence-free velocity field can be split stably into the small overlapping patch subspaces of the multilevel grid; if that stable splitting fails in $H(\mathrm{div})\cap L^3$, the $\epsilon$-robust rate in Theorem 4.4 collapses.

Editorial extensions

If this is right

Small $\epsilon$ can actually be used in practice: because convergence of the inner multilevel solver is $\epsilon$-robust, the augmented Lagrangian parameter can be set to $10^{-2}$ or smaller and still yield very few outer iterations, avoiding the usual penalty of nearly semicoercive subproblems.
The algorithm is parallel by construction: all patch minimizations across the multilevel hierarchy are independent within one sweep, and the only global operations per iteration are one energy evaluation per trial step and one global reduction, so the method maps naturally onto distributed-memory machines.
The backtracking variant removes the need to know the conservative step-size bound $\tau_0$ in advance, while the full approximation scheme evaluates the energy only at coarse-level interpolants, cutting local cost; numerical results show the two additions do not degrade the convergence curves.
The total computational work scales nearly linearly with the number of degrees of freedom in the tested range, because local Hessian assemblies grow by about a factor of four when $h$ is halved, while the idealized parallel patch-solve count grows only slowly.
The analysis leaves $h$-independence open; the theorem controls the $\epsilon$-dependence explicitly, but the constant may depend on mesh size through the stable decomposition, which the paper identifies as the main remaining question.

Reading between the lines

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

If the $\epsilon$-robust convergence extends to other monotone flux laws, the same template—augmented Lagrangian penalty on the continuity constraint plus multilevel convex subspace correction—would apply to nonlinear saddle-point problems such as generalized Forchheimer variants or Richards-type flow, where global convergence analyses are similarly scarce.
The unproved kernel decomposition for $H(\mathrm{div})\cap L^3$ could be tested numerically in isolation: pick a fine-scale divergence-free Raviart–Thomas field, compute the minimal sum of local patch energies needed to represent it, and check whether that stability constant grows with the number of levels; a growth that is not logarithmic would contradict Assumption B.3.
A sharp convergence theory for the FAS variant is a natural next step: the paper justifies Algorithm 4.3 only through existing inexact-local-solver theory, so an explicit rate bound for the practical variant would close the gap between what is proved and what is implemented.
Because the method reduces the PDE to unconstrained convex minimization with small local problems, it could be combined with inexact or randomized local solvers, provided the strengthened convexity and kernel-decomposition assumptions are preserved; the paper does not explore this.

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

Colleague,

My read: the contribution is narrow but legitimate. The paper extends the linear nearly-singular augmented Lagrangian analysis to the nonlinear Darcy–Forchheimer mixed method and fits the resulting penalized convex problem into the nearly semicoercive subspace correction framework. Theorem 3.1, giving arbitrarily fast ALM convergence as ε→0 in the nonlinear case, is proved cleanly through convex conjugacy. Theorem 4.4 is an honest application of the authors' own published framework [43], with the L3-norm verifications done in Lemma 4.3. They also state plainly that h-independence of the multilevel decomposition is not proved, which is more candid than many papers of this type.

The soft spots are real but not fatal. First, Assumption B.3—the kernel decomposition—is load-bearing for ε-robustness, and it is imported without proof from the linear Hilbert-space setting of [45]. Here the space is H(div)∩L3, and the patch decomposition carries an extra L3 norm that the linear theory never has to control. The paragraph after Theorem 4.4 essentially concedes this. This is the right thing to demand: either prove the decomposition for the discrete RT spaces or state it as an explicit assumption and mark the theorem conditional. The definition of F0 as (1/2)||div v − g_h||² also needs a clarification of which kernel is meant in the abstract theory, since the affine shift is not handled in Assumption B.3.

Second, the numerics do not test the full pipeline. The multilevel experiments in Section 5 solve (3.2) with q_h = 0, so they demonstrate convergence to the penalized subproblem, not to the original Darcy–Forchheimer solution. The ALM tests in Tables 1–2 are separate. A combined ALM+multilevel experiment measuring error against the exact u would make the practical claim credible.

Third, the parallel-cost claims are honestly labeled as implementation-independent proxies; there are no wall-clock timings. That is acceptable, but it means the "parallel multilevel" headline should be read carefully.

Overall: the mathematics is clear, the self-citations are legitimate, and the paper is honest about its open problems. It deserves a serious referee, not a desk reject. I would ask for a proof or explicit assumption for B.3, the affine-kernel clarification, and an end-to-end numerical experiment. If the kernel decomposition holds up, the central theorem stands.

Referee Report

3 major / 6 minor

Summary. The paper studies parallel multilevel solvers for the finite element discretization of the Darcy–Forchheimer model. It first proves (Theorem 3.1) that the augmented Lagrangian method applied to the constrained convex reformulation (2.7) converges at a rate that improves as the penalty parameter epsilon decreases, and it identifies the resulting subproblem (3.2) as nearly semicoercive. The core of the paper is a vertex-patch multilevel additive Schwarz method (Algorithm 4.1), together with a backtracking variant (Algorithm 4.2) and a full approximation scheme variant (Algorithm 4.3). Convergence of Algorithm 4.1 is stated in Theorem 4.4 as a consequence of the abstract nearly semicoercive subspace-correction theory summarized in Appendix B, with the required assumptions asserted to hold in Section 4.2. Numerical experiments on two benchmark problems report iteration counts, cost proxies, and idealized parallel communication counts as functions of epsilon, h, and the coarsening factor.

Significance. If the missing verification is supplied, the paper would give a global, epsilon-robust convergence guarantee for a parallel multilevel solver of a nonlinear mixed finite element problem of Darcy–Forchheimer type, complementing Newton-based solvers that only provide local convergence. The paper is transparent in several respects: it explicitly concedes that h-independence is not proved, that the FAS variant is not covered by Theorem 4.4, and that some constants may depend on the energy level set. The numerical section sensibly reports implementation-independent cost proxies rather than wall-clock timings from a nonexistent distributed implementation. The main risk is that the central epsilon-robustness claim rests on Assumption B.3, which is imported from a linear Hilbert-space theory without a proof for X = H(div) cap L^3, and on constants whose dependence on epsilon through the level set K0 is not controlled.

major comments (3)

[Section 4.2, Assumption B.3 and Theorem 4.4] The kernel-decomposition assumption is the load-bearing step for the epsilon-robust convergence claim, but it is asserted rather than proved. The text states that 'The kernel decomposition condition (Assumption B.3) is satisfied by the patch-based multilevel decomposition (4.1), as discussed in [45, Section 5.1].' That reference addresses linear nearly singular systems in Hilbert spaces, whereas the present space V = H(div; Omega) cap L^3(Omega)^d has an additional L^3 component in its norm, and the paragraph after Theorem 4.4 concedes exactly this difficulty. Because the strengthened convexity inequality (B.3) and the convergence rate in Theorem B.5 depend on representing divergence-free functions in X_h as sums of patch-supported divergence-free functions with norm control in X, a citation to the linear theory does not close the argument. Please provide a proof, or a precise theorem from [45] or another reference with norm bounds tracked, that Assumption B.3 holds for the Raviart-Thomas spaces (2.3) and the vertex-patch decomposition (4.1) in X. In addition, when g_h is nonzero the set ker F0 = {v : div v = g_h} is affine, not a linear subspace, so Assumption B.3 as stated is not even well-defined for Example 5.2; state the affine or Bregman-kernel version that is actually used.
[Section 4.2, Theorem 4.4 and equations (4.6)-(4.7)] The claimed robustness with respect to epsilon is not established by the stated result. The constants C1 and C2 in Theorem 4.4 depend on K0 defined in (4.7), and K0 depends on epsilon through F^epsilon. This dependence is not merely technical: F0 controls only div, so a solenoidal component w with div w = 0 and ||w||_{L2} approx ||w||_{L3} approx R can be added to any admissible v without changing F0, while F1 grows like R^3. Hence K0 can contain vectors with ||w||_X up to O(epsilon^{-1/3}). For such w, the ratio (d0 + d1)(w;v)/||w||^3_X behaves like R^{-3/2}, so mu_{1,K0} in (4.6) can tend to zero as epsilon -> 0 and the factor C2 in Theorem 4.4 degenerates. To support the abstract's claim of robustness with respect to epsilon, the paper must either prove that K0 is bounded in the X norm independently of epsilon or restate Theorem 4.4 and Theorem B.5 with explicit control of the epsilon-dependence of C1 and C2.
[Section 4.4, Algorithm 4.3] The recommended practical variant is not covered by Theorem 4.4. The paper states that its convergence 'relies on existing theory for inexact local solvers and FAS, such as [20],' but no inexactness condition from [20] or another reference is verified for the specific FAS local problems (4.12). Since all production experiments in Tables 4-7 use Algorithm 4.3, the theoretical support offered for the main practical solver is incomplete. Please either verify the relevant assumptions of the cited inexact-solver theory for the local problems (4.12), or explicitly mark Algorithm 4.3 as a numerically justified heuristic whose convergence analysis is deferred.

minor comments (6)

[Section 4.1] The symbol V_j is used both for the finite element space on level j and for the set of interior vertices of T_j ('Let V_j be the set of all interior vertices'). Please use a different symbol, for example mathcal{V}_j, to avoid ambiguity.
[Equation (4.6)] The ratios defining mu_{0,K} and mu_{1,K} are not bounded below by constants of order one on a general bounded K; for instance, mu_{0,K} = 1/(2 sup_{v, v+w in K} ||div w||_{L2}). The text should state that the infima are positive on each bounded K and should specify which dependencies on the diameter of K, h, and epsilon are intended.
[Theorem 3.1 proof] The proof asserts without detail that local smoothness of F implies local strong convexity of F* composed with div* via [31], after checking injectivity of div*. Please spell out how the local strong convexity constant is controlled and why it can be chosen independent of epsilon.
[Appendix B, Theorem B.5] The solution of the minimization problem (B.1) is denoted by u without being introduced in the statement of the theorem. Please add a sentence defining u.
[References] Reference [31] is cited with only a single page number, and reference [37] is an arXiv preprint; please update these if a published version is available.
[Section 5.1] The statement that iteration counts 'remain uniformly bounded as the mesh is refined' is a numerical observation, not a theoretical result; consider phrasing it as numerical evidence to avoid implying a proved h-independence.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the core derivations are independent applications of established frameworks, and the unproved kernel-decomposition hypothesis is a proof gap, not a circular step.

full rationale

The paper's central chain is: the mixed Darcy-Forchheimer discretization (2.4) is recast as the constrained convex problem (2.7); one step of the augmented Lagrangian method produces the nearly semicoercive problem (3.2); Algorithm 4.1 is the abstract parallel subspace correction method applied to (3.2); and Theorem 4.4 follows by invoking Theorem B.5, which is a summary of the published framework [43] after checking Assumptions B.1-B.4 in Section 4.2. No equation in the paper is used to define the quantity that the paper then claims to prove: no constant is fitted to convergence data, no 'prediction' re-reports a fitted parameter, and the augmented-Lagrangian reformulation does not presuppose convergence of the multilevel solver. The only fragile point is Assumption B.3: the paper asserts 'The kernel decomposition condition (Assumption B.3) is satisfied by the patch-based multilevel decomposition (4.1), as discussed in [45, Section 5.1]' and later concedes that controlling the L3 norm in X = H(div) ∩ L3 makes the h-independent decomposition 'highly nontrivial' and leaves it for future work. That is an unsupported imported hypothesis and a correctness risk, but it is not circular: the abstract theorem in [43] is published with stated assumptions that do not include the Darcy-Forchheimer result, and [45] is not authored by the present authors. Heavy reliance on [43] involves self-citation, but the cited result is independent support (peer-reviewed mathematics with explicit assumptions) and does not make the derivation equivalent to its inputs. The numerical benchmarks from [54] provide external validation of the algorithmic behavior. Overall, no circular reduction is exhibited in the paper's own equations.

Assumptions & free parameters 0 free parameters · 7 assumptions · 0 invented entities

No data-fitted constants appear in the paper. The solver parameters epsilon, tau, gamma, and the tolerances are user choices rather than fitted values. Every theorem states existence of constants without computing them from data. The central claims rest on the established abstract theory of [43] and on the cited kernel decomposition result of [45], both of which are assumed rather than derived.

assumptions (7)

standard math Fenchel-Rockafellar duality and the equivalence of the saddle-point problem (2.6) and the constrained problem (2.7) in Proposition 2.1.
Invoked in Section 2 to move from the mixed weak form to a constrained convex minimization; treated as established convex analysis.
standard math Local strong convexity of F* via [31] and inheritance of strong convexity by F* composed with the injective operator div*.
Used in the proof of Theorem 3.1 to obtain the augmented Lagrangian contraction factor; relies on the inf-sup condition of Raviart-Thomas spaces for a uniform lower bound.
standard math The abstract convergence theory of subspace correction for nearly semicoercive problems from [43], including Assumptions B.1-B.4, strengthened convexity (B.3), and Theorem B.5.
Section 4.2 is a verification of this framework; the framework itself is not reproved in this paper.
domain assumption Kernel decomposition Assumption B.3 for the patch-based decomposition (4.1), cited to [45, Section 5.1].
Critical for epsilon-robust convergence; asserted by citation for X = H(div) intersected with L^3 and not proved in the present nonlinear setting.
standard math Coloring argument giving tau0 at least order J^{-1}, approximately |log h|^{-1}, from [56].
Used to claim the strengthened convexity condition (B.3) for the overlapping vertex-based decomposition.
standard math Quasi-norm estimates and the triangle-inequality-like property from [46, Lemma 5.4] for the Bregman divergence d1.
Used to verify Assumption B.4.
domain assumption Existence and uniqueness of the discrete Raviart-Thomas solution (2.4) from [54, Theorem 3.5].
Underpins the well-posedness of the discrete problem being solved.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Parallel multilevel methods for solving the Darcy--Forchheimer model based on a nearly semicoercive formulation." pith.science (2026). https://pith.science/paper/3CLA2IFN

@misc{pith2026250703192,
  author       = {Pith},
  title        = {Pith review of: Parallel multilevel methods for solving the Darcy--Forchheimer model based on a nearly semicoercive formulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3CLA2IFN}},
  note         = {Machine review of arXiv:2507.03192}
}

abstract

High-velocity fluid flow through porous media is modeled by prescribing a nonlinear relationship between the flow rate and the pressure gradient, called the Darcy--Forchheimer equation. This paper is concerned with the analysis of parallel multilevel methods for solving the Darcy--Forchheimer model. We begin by reformulating the Darcy--Forchheimer model as a nearly semicoercive convex optimization problem via the augmented Lagrangian method. Building on this formulation, we develop a parallel multilevel method, also known as a multilevel additive Schwarz method, within the framework of subspace correction for nearly semicoercive convex problems, yielding a theoretically supported and computationally efficient solver for the Darcy--Forchheimer model. The convergence analysis establishes robustness with respect to the augmented Lagrangian parameter $\epsilon$. To further enhance convergence, we incorporate a backtracking line search and a full approximation scheme. Numerical results support the theoretical findings and demonstrate the effectiveness of the proposed approach.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 66 canonical work pages

[43]

Lee and J

Y.-J. Lee and J. Park, Parallel subspace correction methods for semicoercive and nearly semi- coercive convex optimization with applications to nonlinear PDEs , Math. Comp., (2025), https://doi.org/10.1090/mcom/4172

work page doi:10.1090/mcom/4172 2025
[45]

Y.-J. Lee, J. Wu, J. Xu, and L. Zikatanov , Robust subspace correction methods for nearly singular systems , Math. Models Methods Appl. Sci., 17 (2007), pp. 1937–1963

work page 2007
[20]

L. Chen, X. Hu, and S. Wise , Convergence analysis of the fast subspace descent method for convex optimization problems , Math. Comp., 89 (2020), pp. 2249–2282

work page 2020
[1]

Acharyya, A

S. Acharyya, A. Banerjee, and D. Boley , Bregman divergences and triangle inequality , in Proceedings of the 2013 SIAM International Conference on Data Mining (SDM), Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 2013, pp. 476–484

work page 2013
[2]

E. L. Allgower, K. B ¨ohmer, F. Potra, and W. Rheinboldt , A mesh-independence prin- ciple for operator equations and their discretizations , SIAM J. Numer. Anal., 23 (1986), pp. 160–169

work page 1986
[3]

A. C. Aristotelous, O. Karakashian, and S. M. Wise , A mixed discontinuous Galerkin, convex splitting scheme for a modified Cahn-Hilliard equation and an efficient nonlinear multigrid solver , Discrete Contin. Dyn. Syst. Ser. B, 18 (2013), pp. 2211–2238

work page 2013
[4]

D. N. Arnold, R. S. F alk, and R. Winther, Multigrid in H(div) and H(curl), Numer. Math., 85 (2000), pp. 197–217

work page 2000
[5]

Arrar ´as, F

A. Arrar ´as, F. J. Gaspar, L. Portero, and C. Rodrigo , Geometric multigrid methods for Darcy–Forchheimer flow in fractured porous media , Comput. Math. Appl., 78 (2019), pp. 3139–3151

work page 2019

Show all 73 references

[6]

H. H. Bauschke and P. L. Combettes , Convex Analysis and Monotone Operator Theory in Hilbert Spaces, Springer, New York, 2011

2011
[7]

Boffi, F

D. Boffi, F. Brezzi, and M. Fortin , Mixed Finite Element Methods and Applications , Springer, Heidelberg, 2013

2013
[8]

Boyd and L

S. Boyd and L. V andenberghe , Convex Optimization , Cambridge University Press, Cam- bridge, 2004

2004
[9]

J. H. Bramble, J. E. Pasciak, and J. Xu , Parallel multilevel preconditioners, Math. Comp., 55 (1990), pp. 1–22

1990
[10]

Brandt , Multi-level adaptive solutions to boundary-value problems , Math

A. Brandt , Multi-level adaptive solutions to boundary-value problems , Math. Comp., 31 (1977), pp. 333–390

1977
[11]

Brandt, Multigrid Techniques: 1984 Guide with Applications to Fluid Dynamics , Gesell- schaft f¨ ur Mathematik und Datenverarbeitung mbH, St

A. Brandt, Multigrid Techniques: 1984 Guide with Applications to Fluid Dynamics , Gesell- schaft f¨ ur Mathematik und Datenverarbeitung mbH, St. Augustin, 1984

1984
[12]

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

2008
[13]

Bueler and P

E. Bueler and P. E. F arrell, A full approximation scheme multilevel method for nonlinear variational inequalities, SIAM J. Sci. Comput., 46 (2024), pp. A2421–A2444

2024
[14]

X.-C. Cai, W. D. Gropp, D. E. Keyes, R. G. Melvin, and D. P. Young , Parallel Newton– Krylov–Schwarz algorithms for the transonic full potential equation, SIAM J. Sci. Comput., 19 (1998), pp. 246–265

1998
[15]

X.-C. Cai, W. D. Gropp, D. E. Keyes, and M. D. Tidriri, Newton–Krylov–Schwarz methods in CFD, in Numerical Methods for the Navier–Stokes Equations: Proceedings of the Inter- national Workshop Held at Heidelberg, October 25–28, 1993, Springer, 1994, pp. 17–30

1993
[16]

Cai and D

X.-C. Cai and D. E. Keyes , Nonlinearly preconditioned inexact Newton algorithms , SIAM J. Sci. Comput., 24 (2002), pp. 183–200

2002
[17]

X.-C. Cai, D. E. Keyes, and L. Marcinkowski, Non-linear additive Schwarz preconditioners and application in computational fluid dynamics , Internat. J. Numer. Methods Fluids, 40 (2002), pp. 1463–1470

2002
[18]

Cai and M

X.-C. Cai and M. Sarkis , A restricted additive Schwarz preconditioner for general sparse linear systems, SIAM J. Sci. Comput., 21 (1999), pp. 792–797

1999
[19]

L. Chen, R. Guo, and J. Wei , Transformed primal-dual methods with variable- preconditioners, arXiv preprint arXiv:2312.12355, (2023)

2023 arXiv
[21]

Z. Chen, Y. Ding, Z. Qin, Y. He, B. Liang, Y. Qing, Y. Xing, and B. Li , Effects of high- velocity flow on the temperature field near the wellbore: A review , in 2023 International Conference on Energy Engineering, Springer, 2023, pp. 887–912

2023
[22]

D. Cho, J. Xu, and L. Zikatanov , New estimates for the rate of convergence of the method of subspace corrections, Numer. Math. Theory Methods Appl., 1 (2008), pp. 44–56

2008
[23]

P. G. Ciarlet , The Finite Element Method for Elliptic Problems , Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 2002

2002
[24]

Dolean, M

V. Dolean, M. J. Gander, W. Kheriji, F. Kwok, and R. Masson , Nonlinear precondition- ing: How to use a nonlinear Schwarz method to precondition Newton ’s method , SIAM J. Sci. Comput., 38 (2016), pp. A3357–A3380

2016
[25]

Dryja, M

M. Dryja, M. V. Sarkis, and O. B. Widlund , Multilevel Schwarz methods for elliptic prob- lems with discontinuous coefficients in three dimensions, Numer. Math., 72 (1996), pp. 313– 348

1996
[26]

Efstathiou and M

E. Efstathiou and M. J. Gander , Why restricted additive Schwarz converges faster than additive Schwarz , BIT Numer. Math., 43 (2003), pp. 945–959. 30 JONGHO PARK AND S. MAJID HASSANIZADEH

2003
[27]

F. A. F airag and J. D. Audu , Two-level Galerkin mixed finite element method for Darcy– Forchheimer model in porous media , SIAM J. Numer. Anal., 58 (2020), pp. 234–253

2020
[28]

Frommer and D

A. Frommer and D. B. Szyld, An algebraic convergence theory for restricted additive Schwarz methods using weighted max norms , SIAM J. Numer. Anal., 39 (2001), pp. 463–479

2001
[29]

Gavaskar, N

A. Gavaskar, N. Gupta, B. Sass, R. Janosy, and J. Hicks , Design guidance for applica- tion of permeable reactive barriers for groundwater remediation, tech. report, BATTELLE Project Report, 2000

2000
[30]

Girault and M

V. Girault and M. F. Wheeler , Numerical discretization of a Darcy–Forchheimer model , Numer. Math., 110 (2008), pp. 161–198

2008
[31]

Goebel and R

R. Goebel and R. T. Rockafellar , Local strong convexity and local Lipschitz continuity of the gradient of convex functions , J. Convex Anal., 15 (2008), p. 263

2008
[32]

S. M. Hassanizadeh and W. G. Gray , High velocity flow in porous media , Transp. Porous Media, 2 (1987), pp. 521–531

1987
[33]

M. R. Hestenes, Multiplier and gradient methods , J. Optim. Theory Appl., 4 (1969), pp. 303– 320

1969
[34]

Hiptmair and A

R. Hiptmair and A. Toselli , Overlapping and multilevel Schwarz methods for vector valued elliptic problems in three dimensions , in Parallel Solution of Partial Differential Equations (Minneapolis, MN, 1997), vol. 120 of IMA Vol. Math. Appl., Springer, New York, 2000, pp. 181–208

1997
[35]

Horneber, C

T. Horneber, C. Rauh, and A. Delgado , Fluid dynamic characterisation of porous solids in catalytic fixed-bed reactors, Micropor. Mesomor. Mat., 154 (2012), pp. 170–174

2012
[36]

Huang, L

J. Huang, L. Chen, and H. Rui , Multigrid methods for a mixed finite element method of the Darcy–Forchheimer model, J. Sci. Comput., 74 (2018), pp. 396–411

2018
[37]

Jiang, J

B. Jiang, J. Park, and J. Xu , Connections between convex optimization algorithms and subspace correction methods, arXiv preprint arXiv:2505.09765, (2025)

2025 arXiv
[38]

Kim , Accelerated proximal point method for maximally monotone operators , Math

D. Kim , Accelerated proximal point method for maximally monotone operators , Math. Pro- gram., 190 (2021), pp. 57–87

2021
[39]

E. L. Kuniansky , Simulating groundwater flow in karst aquifers with distributed parameter models—Comparison of porous-equivalent media and hybrid flow approaches , tech. report, US Geological Survey, 2016

2016
[40]

Lee and E.-H

C.-O. Lee and E.-H. Park, A dual iterative substructuring method with a penalty term, Numer. Math., 112 (2009), pp. 89–113

2009
[41]

Lee and E.-H

C.-O. Lee and E.-H. Park , A dual iterative substructuring method with a small penalty pa- rameter, J. Korean Math. Soc., 54 (2017), pp. 461–477

2017
[42]

Lee and J

Y.-J. Lee and J. Park , On the linear convergence of additive Schwarz methods for the p- Laplacian, IMA J. Numer. Anal., 45 (2025), pp. 2655–2684

2025
[44]

Y.-J. Lee, J. Wu, and J. Chen , Robust multigrid method for the planar linear elasticity problems, Numer. Math., 113 (2009), pp. 473–496

2009
[46]

Liu and N

W. Liu and N. Yan , Quasi-norm local error estimators for p-Laplacian, SIAM J. Numer. Anal., 39 (2001), pp. 100–127

2001
[47]

L´opez, B

H. L´opez, B. Molina, and J. J. Salas, Comparison between different numerical discretizations for a Darcy–Forchheimer model , Electron. Trans. Numer. Anal., 34 (2009), pp. 187–203

2009
[48]

Martinet , R´ egularisation d’in´ equations variationnelles par approximations successives, Rev

B. Martinet , R´ egularisation d’in´ equations variationnelles par approximations successives, Rev. Fran¸ caise Informat. Recherche Op´ erationnelle, 4 (1970), pp. 154–158

1970
[49]

Medici and L

G. Medici and L. J. West, Groundwater flow velocities in karst aquifers; importance of spatial observation scale and hydraulic testing for contaminant transport prediction , Environ. Sci. Pollut. Res. Int., 28 (2021), pp. 43050–43063

2021
[50]

R. E. Megginson, An Introduction to Banach Space Theory, Springer-Verlag, New York, 1998

1998
[51]

Mustapha, L

H. Mustapha, L. de Langavant, and M. A. Giddins , Darcy and non-Darcy flows in frac- tured gas reservoirs , in SPE Reservoir Characterisation and Simulation Conference and Exhibition, SPE, 2015, p. D021S019R001

2015
[52]

Nesterov , Gradient methods for minimizing composite functions , Math

Y. Nesterov , Gradient methods for minimizing composite functions , Math. Program., 140 (2013), pp. 125–161

2013
[53]

Nielsen and R

F. Nielsen and R. Nock , Sided and symmetrized Bregman centroids , IEEE Trans. Inform. Theory, 55 (2009), pp. 2882–2904

2009
[54]

Pan and H

H. Pan and H. Rui, Mixed element method for two-dimensional Darcy–Forchheimer model , J. Sci. Comput., 52 (2012), pp. 563–587. MULTILEVEL METHODS FOR FORCHHEIMER 31

2012
[55]

Park, Mixed finite element methods for generalized Forchheimer flow in porous media , Numer

E.-J. Park, Mixed finite element methods for generalized Forchheimer flow in porous media , Numer. Methods Partial Differential Equations, 21 (2005), pp. 213–228

2005
[56]

Park, Additive Schwarz methods for convex optimization as gradient methods , SIAM J

J. Park, Additive Schwarz methods for convex optimization as gradient methods , SIAM J. Numer. Anal., 58 (2020), pp. 1495–1530

2020
[57]

Park, Additive Schwarz methods for convex optimization with backtracking , Comput

J. Park, Additive Schwarz methods for convex optimization with backtracking , Comput. Math. Appl., 113 (2022), pp. 332–344

2022
[58]

Park, Additive Schwarz methods for semilinear elliptic problems with convex energy func- tionals: Convergence rate independent of nonlinearity , SIAM J

J. Park, Additive Schwarz methods for semilinear elliptic problems with convex energy func- tionals: Convergence rate independent of nonlinearity , SIAM J. Sci. Comput., 46 (2024), pp. A1373–A1396

2024
[59]

M. J. D. Powell , A method for nonlinear constraints in minimization problems , in Opti- mization (Sympos., Univ. Keele, Keele, 1968), Academic Press, London-New York, 1969, pp. 283–298

1968
[60]

R. T. Rockafellar, Convex Analysis, Princeton University Press, Princeton, NJ, 1970

1970
[61]

R. T. Rockafellar, Augmented Lagrangians and applications of the proximal point algorithm in convex programming, Math. Oper. Res., 1 (1976), pp. 97–116

1976
[62]

R. T. Rockafellar, Monotone operators and the proximal point algorithm , SIAM J. Control Optim., 14 (1976), pp. 877–898

1976
[63]

J. J. Salas, H. L ´opez, and B. Molina , An analysis of a mixed finite element method for a Darcy–Forchheimer model, Math. Comput. Model., 57 (2013), pp. 2325–2338

2013
[64]

Sedghi-Asl, E

M. Sedghi-Asl, E. Morales-Casique, and S. M. Hassanizadeh , Dispersion in high-porosity porous medium, J. Porous Media, 26 (2023), pp. 1–12

2023
[65]

Setzer , Operator splittings, Bregman methods and frame shrinkage in image processing , Int

S. Setzer , Operator splittings, Bregman methods and frame shrinkage in image processing , Int. J. Comput. Vis., 92 (2011), pp. 265–280

2011
[66]

Tai and J

X.-C. Tai and J. Xu, Global and uniform convergence of subspace correction methods for some convex optimization problems , Math. Comp., 71 (2002), pp. 105–124

2002
[67]

S. M. Wise , Unconditionally stable finite difference, nonlinear multigrid simulation of the Cahn–Hilliard–Hele–Shaw system of equations , J. Sci. Comput., 44 (2010), pp. 38–68

2010
[68]

Xu and L

J. Xu and L. Zikatanov , The method of alternating projections and the method of subspace corrections in Hilbert space, J. Amer. Math. Soc., 15 (2002), pp. 573–597

2002
[69]

Xu and L

J. Xu and L. Zikatanov, Algebraic multigrid methods, Acta Numer., 26 (2017), pp. 591–721

2017
[70]

Y. Yao, G. Li, and P. Qin, Seepage features of high-velocity non-Darcy flow in highly produc- tive reservoirs, J. Nat. Gas Sci. Eng., 27 (2015), pp. 1732–1738

2015
[71]

Zhang, Multilevel Schwarz methods , Numer

X. Zhang, Multilevel Schwarz methods , Numer. Math., 63 (1992), pp. 521–539

1992
[72]

Zhang , Multilevel Schwarz methods for the biharmonic Dirichlet problem , SIAM J

X. Zhang , Multilevel Schwarz methods for the biharmonic Dirichlet problem , SIAM J. Sci. Comput., 15 (1994), pp. 621–644

1994
[73]

Zi ´o lkowska and D

I. Zi ´o lkowska and D. Zi´o lkowski, Fluid flow inside packed beds , Chem. Eng. Process., 23 (1988), pp. 137–164

1988

Pith tools

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

[1] [43]

Lee and J

Y.-J. Lee and J. Park, Parallel subspace correction methods for semicoercive and nearly semi- coercive convex optimization with applications to nonlinear PDEs , Math. Comp., (2025), https://doi.org/10.1090/mcom/4172

work page doi:10.1090/mcom/4172 2025

[2] [45]

Y.-J. Lee, J. Wu, J. Xu, and L. Zikatanov , Robust subspace correction methods for nearly singular systems , Math. Models Methods Appl. Sci., 17 (2007), pp. 1937–1963

work page 2007

[3] [20]

L. Chen, X. Hu, and S. Wise , Convergence analysis of the fast subspace descent method for convex optimization problems , Math. Comp., 89 (2020), pp. 2249–2282

work page 2020

[4] [1]

Acharyya, A

S. Acharyya, A. Banerjee, and D. Boley , Bregman divergences and triangle inequality , in Proceedings of the 2013 SIAM International Conference on Data Mining (SDM), Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 2013, pp. 476–484

work page 2013

[5] [2]

E. L. Allgower, K. B ¨ohmer, F. Potra, and W. Rheinboldt , A mesh-independence prin- ciple for operator equations and their discretizations , SIAM J. Numer. Anal., 23 (1986), pp. 160–169

work page 1986

[6] [3]

A. C. Aristotelous, O. Karakashian, and S. M. Wise , A mixed discontinuous Galerkin, convex splitting scheme for a modified Cahn-Hilliard equation and an efficient nonlinear multigrid solver , Discrete Contin. Dyn. Syst. Ser. B, 18 (2013), pp. 2211–2238

work page 2013

[7] [4]

D. N. Arnold, R. S. F alk, and R. Winther, Multigrid in H(div) and H(curl), Numer. Math., 85 (2000), pp. 197–217

work page 2000

[8] [5]

Arrar ´as, F

A. Arrar ´as, F. J. Gaspar, L. Portero, and C. Rodrigo , Geometric multigrid methods for Darcy–Forchheimer flow in fractured porous media , Comput. Math. Appl., 78 (2019), pp. 3139–3151

work page 2019