Pith. sign in

REVIEW 3 major objections 4 minor 44 references

A primal-dual splitting algorithm for monotone inclusions with applications

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read One resolvent per component per iteration solves composite monotone inclusions, with weak convergence always and strong convergence under uniform monotonicity.

desk verdict A credible unification of two splitting lines, but the algorithm's resolvent notation is undefined and the fixed-point algebra doesn't match the printed iteration; worth refereeing, not accepting as-is. read the letter →

arxiv 2606.28010 v2 pith:MRO2L7HQ submitted 2026-06-26 math.OC

Changchi Huang , Jigen Peng , Liqian Qin , Yuchao Tang This is my paper · ORCID
classification math.OC MSC 47H0565K1590C25
keywords monotoneinclusionprimal-dualsplittingresolventcocoerciveoperatorweakconvergencestrongimagedeblurringdenoising
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 introduces Algorithm 1, a chain-structured primal-dual splitting method for finding a zero of the sum of n maximally monotone operators, n-1 cocoercive operators, and a composite term L*BL. At each iteration it performs one resolvent evaluation per monotone operator, one forward evaluation per cocoercive term, and one dual update for L. The central result is weak convergence of every iterate sequence to a solution of the primal and dual inclusions, and strong convergence of the primal components when one monotone operator is uniformly monotone. The scheme generalizes and unifies several existing splitting methods while keeping per-iteration cost low, and the authors demonstrate competitive performance on image deblurring and denoising. A reader should care because operator splitting is the standard tool for large-scale convex optimization, and this is a frugal, unifying construction.

What carries the argument

The load-bearing object is the fixed-point operator T = T_1 x T_2 defined by (10)-(11): its first component propagates a chain of resolvent evaluations x_i = J_{alpha A_i}(...) with forward cocoercive corrections, and its second component updates the dual variable via y = J^B_gamma((L(x_1 + x_n) - v)/gamma). The convergence proof works through a weighted inner product <.,.>^alpha_gamma on H^{n-1} x G; Lemma 3.2 shows T is lambda-averaged nonexpansive exactly when alpha < 2/beta and gamma < (1 - alpha beta / 2)/(alpha ||L||^2), which turns the iteration into a Krasnoselskii-Mann iteration. Lemma 3.1 is the chain identity that transfers fixed points of T to solutions of the primal-dual pair.

What would settle it

Implement Algorithm 1 for a problem with n = 2 and two cocoercive terms without merging them, using the stated resolvent definition J^B_gamma = (I + gamma B)^{-1} for a simple B with known resolvent (e.g., B = lambda I). If the iterates fail to converge to a solution of inclusion (1) or fail the inclusion (23) test, then both the m = n-1 reduction and the gamma-inverse resolvent convention are essential.

Watch

Extended reading notes

Core claim

Under the hypothesis m = n-1, Algorithm 1 defines a fixed-point iteration z^{k+1} = T(z^k) with T a lambda-averaged nonexpansive operator on H^{n-1} x G. Lemma 3.2 establishes the averaged nonexpansiveness inequality (18); Theorem 3.1(i)-(ii) then gives weak convergence of (z^k, v^k) to a point in Fix T and of (x^k, y^k) to (x-bar, L x-bar) with x-bar solving the primal inclusion (2), while Theorem 3.1(v) upgrades to strong convergence if some A_i is uniformly monotone. The proof relies on a resolvent-chain identity linking fixed points of T to points (x-bar, u-bar) satisfying -L*u-bar in sum_i A_i x-bar + sum_k C_k x-bar and u-bar in B(L x-bar).

Load-bearing premise

The convergence proof only covers the case m = n-1, so the algorithm's advertised generality for arbitrary numbers of cocoercive terms rests on merging or zero-padding those terms, and the dual resolvent step must be read as (I + gamma^{-1} B)^{-1} rather than as the gamma-resolvent defined in the preliminaries.

Editorial extensions

If this is right

  • For every instance with m = n-1, the sequence (z^k, v^k) converges weakly to a fixed point of T, and the reconstructed primal sequence (x^k, y^k) converges weakly to (x-bar, L x-bar) with x-bar solving inclusion (2); the dual variables converge to a solution of the dual inclusion (3).
  • Strong convergence of all x_i^k to the solution x-bar follows as soon as one of the maximally monotone operators A_i is uniformly monotone, covering the strongly monotone and uniformly convex data cases.
  • Because each monotone operator is used only through one resolvent and each cocoercive term through one forward evaluation per iteration, the method keeps the frugality of the best existing splitting schemes while adding the composite term L*BL.
  • Specializing B = 0, L = 0 recovers the ring-network forward-backward algorithm; setting C_k = 0 and alpha = 1 recovers a parameterized extension of a known primal-dual resolvent splitting method.
  • Applied to convex minimization with A_i = partial g_i, C_k = gradient f_k, and B_j = partial h_j, the algorithm becomes a chain-structured primal-dual proximal-gradient method for sums of nonsmooth, smooth convex, and linear-composite terms.

Reading between the lines

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

  • Editorial: The abstract advertises arbitrary m >= 1, but the convergence proof is literally for m = n-1; the transformations in Remark 3.1 for m != n-1 (merging cocoercive terms or padding with zero operators) alter the chain structure, so the advertised generality is a reduction rather than a directly proven fact.
  • Editorial: A reader implementing the algorithm should use the resolvent convention J^B_gamma(u) = (I + gamma^{-1} B)^{-1} u in the dual step; with the definition J_{gamma A} = (I + gamma A)^{-1} stated in Section 2, the key monotonicity inclusion (23) in Lemma 3.2 would not hold.
  • Editorial: The alpha/gamma step-size condition suggests a stability-speed trade-off: larger alpha accelerates the chain and improves conditioning, but forces smaller gamma when ||L|| is large; a quantitative convergence rate in terms of alpha, gamma, and the cocoercivity constants is a natural next question.
  • Editorial: Because the chain is sequential in the n primal operators, the method is not automatically parallel across components; a distributed or graph-structured variant preserving the single-resolvent budget would be a direct extension of this line.
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

3 major / 4 minor

Summary. The paper proposes a chain-structured primal-dual splitting algorithm for monotone inclusions of the form (1), combining resolvent steps for n maximally monotone operators, forward steps for n−1 cocoercive operators, and a dual update for the composite term L^*BL. It claims weak convergence of the iterates under standard monotonicity/cocoercivity assumptions and strong convergence under an additional uniform-monotonicity condition. A product-space extension handles finitely many composite terms L_j^*B_jL_j, and the method is specialized to convex minimization and tested on image deblurring and denoising problems.

Significance. The algorithmic template is natural and, if the proof is repaired, extends the ring-network forward-backward method [19] and the minimal-lifting primal-dual scheme [28] to include both cocoercive and composite terms while keeping one resolvent/forward evaluation per component. The proof strategy uses the standard Krasnosel'skii-Mann/Opial machinery, so the convergence framework is recognizable and checkable. The numerical experiments compare against PFDR [34] on several images and noise levels and report competitive restoration quality. However, the manuscript currently contains a fundamental inconsistency in the definition of the resolvent J^B_γ, and the main nonexpansiveness proof has multiple algebraic errors; these issues must be resolved before the convergence claims are usable.

major comments (3)
  1. [Section 2 and Algorithm 1, Eq. (7)] The symbol J^B_γ is never defined. With the Section 2 convention J_{γA}=(I+γA)^{-1}, the update y^k=J^B_γ((L(x_1^k+x_n^k)-v^k)/γ) gives L(x_1+x_n)-v-γy ∈ γ^2By, not the inclusion γL(x_1+x_n)-v-γy ∈ By used in Eq. (23) of Lemma 3.2. All subsequent fixed-point and nonexpansiveness results (Lemma 3.1, Lemma 3.2, Theorem 3.1) require J^B_γ=(I+γ^{-1}B)^{-1}. The same ambiguity affects Algorithm 2 and Theorem 3.3. Please define this resolvent explicitly, reconcile the notation with Section 2, and re-check the numerical implementation against the corrected convention.
  2. [Lemma 3.2, Eqs. (24)-(34)] The proof of the central averaged-nonexpansiveness estimate (18) contains several displayed algebraic errors. For example, Eq. (30) writes ||L(x_1-x_n)-L(x_1-x_n)||^2, which is identically zero, and the equality preceding it is wrong; the intended term is ||L(x_1-\bar{x}_1)-L(x_n-\bar{x}_n)||^2. Equations (32) and (34) also have misplaced factors and signs involving (1-λ)/λ. Since Lemma 3.2 is the load-bearing step for Theorem 3.1, this algebra must be rederived carefully and verified.
  3. [Section 3, Remark 3.1; Eq. (1)] The abstract and Section 1 advertise an algorithm for (1) with arbitrary m≥1 cocoercive terms, but all convergence theorems are for the case m=n−1. Remark 3.1 addresses m≠n−1 only by merging cocoercive terms or padding with zero operators. For m>n−1 this destroys the advertised fully-split per-component structure: the algorithm then treats the cocoercive part as a single operator. This is a gap between the claimed scope and the proven statement. Please state the scope more precisely in the abstract and introduction, or extend the proof to general m.
minor comments (4)
  1. [Proof of Theorem 3.2] The proof asserts ||L||^2 = Σ_j ||L_j||^2 for L=(L_1,...,L_m). This equality is false in general; only ||L||^2 ≤ Σ_j ||L_j||^2 holds. The step-size condition using Σ_j||L_j||^2 is a safe upper bound, so the convergence conclusion is not damaged, but the displayed equality should be corrected to an inequality.
  2. [Theorem 3.1(ii), Eq. (37)] In Eq. (37), the summation involving c^k_{i+1} is missing its upper limit, and the definition c_i^k=C_{i-1}(x_i^k)-C_{i-1}(x_{i-1}^k)) contains an unmatched parenthesis. These typographical errors should be fixed.
  3. [Section 4.1.1, p. 26] The text says that a larger λ 'strengthens the regularization constraint', but λ is a step size, not a regularization parameter. This wording is confusing and should be rephrased.
  4. [Throughout] The superscript notation J^B_γ conflicts with the convention J_{γA} established in Section 2. After fixing the definition, consider using a distinct notation such as J_{γ^{-1}B} or J_{\gamma^{-1}B} throughout to avoid ambiguity for readers and implementers.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: convergence proof uses standard operator-splitting estimates; self-citations are comparative only.

full rationale

The derivation chain is self-contained. Algorithms 1 and 2 are explicit splitting schemes, and Theorem 3.1 is proved from Lemma 3.1 (fixed-point/solution equivalence), Lemma 3.2 (averaged nonexpansiveness via monotonicity, cocoercivity, and the stated step-size ranges), and standard Opial/Krasnoselskii-Mann results from references [40] and [42]. No parameter is fitted to data and then reported as a prediction: the numerical experiments tune regularization parameters for image quality, which is standard benchmarking, not part of the convergence derivation. The only self-references ([34], [35]) appear in Table 1 and as comparison methods in Remark 3.3 and Section 4; they are not used to justify convergence or to forbid alternatives. The m = n-1 restriction is explicitly acknowledged in Remark 3.1 and handled by merging or padding cocoercive terms, so it is a scope limitation rather than a circular premise. The notational issue with J^B_gamma (Algorithm 1 vs the resolvent definition in Section 2) and the norm identity in Theorem 3.2 are correctness concerns, not circular reductions: they do not define an output in terms of an input. Thus no step of the claimed derivation reduces to its own input by construction.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The central theoretical claim rests on standard convex analysis results and on the m=n−1 structural restriction. The numerical section fits regularization parameters to the test data, which affects only the experimental comparison. No new physical or abstract entities are introduced; the auxiliary z and v variables are standard lifting variables.

free parameters (1)
  • µ1, µ2 (and nonconvexity parameter a) = Varies per image/kernel/noise; see Tables 2 and 4
    Selected to maximize PSNR on the test images in Sections 4.1 and 4.2. These affect the numerical comparison but not the theoretical convergence claim.
assumptions (5)
  • standard math Krasnoselski-Mann iteration for averaged nonexpansive operators converges weakly (Theorem 5.15 of [40]).
    Used in Theorem 3.1(i) to conclude weak convergence of (z^k,v^k) from the averagedness of T established in Lemma 3.2.
  • standard math The sum of a maximally monotone operator and a monotone Lipschitz operator is maximally monotone (Corollary 25.5 of [40]).
    Used in Theorem 3.1(ii) to claim R+S is maximally monotone and hence to take weak-strong limits in inclusion (37).
  • domain assumption Real Hilbert space setting; Ai maximally monotone, Ck is 1/βk-cocoercive, L bounded linear with adjoint L*.
    Problem statement (2) and Algorithm 1 assumptions.
  • ad hoc to paper The number of cocoercive terms equals n−1 (m=n−1).
    Remark 3.1: 'we introduce the assumption m = n−1 in (1).' This is needed for the chain-structured Algorithm 1; other cases are transformed by merging or padding.
  • ad hoc to paper J^B_γ in Algorithm 1 denotes the resolvent of B with parameter 1/γ (so that γL(x1+xn)−v−γy ∈ By holds), not J_{γB} as the Section 2 definition would suggest.
    The proof of Lemma 3.2 relies on inclusion (23) which is only consistent with the 1/γ interpretation. The convex specialization (46) confirms this via prox_{h_j/γ}.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A primal-dual splitting algorithm for monotone inclusions with applications." pith.science (2026). https://pith.science/paper/MRO2L7HQ

@misc{pith2026260628010,
  author       = {Pith},
  title        = {Pith review of: A primal-dual splitting algorithm for monotone inclusions with applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MRO2L7HQ}},
  note         = {Machine review of arXiv:2606.28010}
}
read the original abstract

In this paper, we study a broad class of structured monotone inclusion problems in real Hilbert spaces. We propose a novel primal-dual splitting algorithm for solving such inclusions, which accommodates multiple monotone operators and cocoercive terms, as well as a composite monotone operator involving the linear map. The algorithm combines forward evaluations for the cocoercive components with backward resolvent steps for the monotone operators and employs a dual update for the linear composition term. It generalizes and unifies several existing methods, while requiring only a single resolvent or operator evaluation per iteration. We prove weak convergence of the iterates under standard assumptions on monotonicity and cocoercivity. Furthermore, we establish strong convergence under a mild regularity condition, such as uniform monotonicity. Numerical experiments on image deblurring and denoising problems demonstrate the efficiency and flexibility of the proposed algorithm.

Figures

Figures reproduced from arXiv: 2606.28010 by the authors.

Figure 1
Figure 1. Test images (pixel intensity range:0-255). (a) Building, siz [PITH_FULL_IMAGE:figures/full_fig_p022_1.png] view at source ↗
Figure 2
Figure 2. Influence of α and γ on PSNR for different λ settings. 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 γ/γ max (α) 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100 k λ=0.1 α=0.1 α=0.5 α=1.0 α=1.5 α=1.9 (a) λ = 0.1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 γ/γ max (α) 700 800 900 1000 1100 1200 1300 1400 1500 1600 k λ=0.3 α=0.1 α=0.5 α=1.0 α=1.5 α=1.9 (b) λ = 0.3 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 γ/γ max (α) 500 600 700 800 900 100… view at source ↗
Figure 3
Figure 3. Influence of α and γ on number of iterations k for different λ settings. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Influence of α and γ on CPU time in seconds for different λ settings. Influence of γ. From the PSNR curves, it can be observed that varying γ/γmax(α) within the admissible range introduces only negligible changes in reconstruction quality. For all combinations of λ and…
Figure 5
Figure 5. Figure 5: Objective function values and PSNR versus the number of [PITH_FULL_IMAGE:figures/full_fig_p028_5.png]
Figure 6
Figure 6. Figure 6: Objective function values and PSNR versus the number of [PITH_FULL_IMAGE:figures/full_fig_p029_6.png]
Figure 7
Figure 7. Figure 7: Objective function values and PSNR versus the number of [PITH_FULL_IMAGE:figures/full_fig_p030_7.png]
Figure 8
Figure 8. Figure 8: Corrupted and restored results of the “Building” image. T [PITH_FULL_IMAGE:figures/full_fig_p031_8.png]
Figure 9
Figure 9. Figure 9: Corrupted and restored results of the “Goldhill” image. Th [PITH_FULL_IMAGE:figures/full_fig_p032_9.png]
Figure 10
Figure 10. Figure 10: Corrupted and restored results of the “Castle” image. [PITH_FULL_IMAGE:figures/full_fig_p033_10.png]
Figure 11
Figure 11. Figure 11: Objective function values and PSNR versus the number o [PITH_FULL_IMAGE:figures/full_fig_p036_11.png]
Figure 12
Figure 12. Figure 12: Corrupted and restored results for image denoising of t [PITH_FULL_IMAGE:figures/full_fig_p037_12.png]
Figure 13
Figure 13. Figure 13: Corrupted and restored results for image denoising of t [PITH_FULL_IMAGE:figures/full_fig_p038_13.png]
Figure 14
Figure 14. Figure 14: Corrupted and restored results for image denoising of t [PITH_FULL_IMAGE:figures/full_fig_p039_14.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 2 linked inside Pith

  1. [19]

    Aragon-Artacho, Y

    F.J. Aragon-Artacho, Y. Malitsky, M.K. Tam, and D. Torregros a-Belen. Dis- tributed forward-backward methods for ring networks. Comput. Optim. Appl. , 86:845–870, 2023. 41

    2023

  2. [28]

    Aragon-Artacho, R.I

    F.J. Aragon-Artacho, R.I. Bot, and D. Torregrosa-Belen. A p rimal-dual split- ting algorithm for composite monotone inclusions with minimal lifting. Numer. Algorithms, 93:103–130, 2023

    2023

  3. [34]

    Y.C. Tang, M. Wen, and T.Y. Zeng. Preconditioned three-opera tor splitting algorithm with applications to image restoration. J. Sci. Comput. , 92:106, 2022. 42

    2022

  4. [1]

    Condat, G

    L. Condat, G. Malinovsky, and P. Richtarik. Distributed proximal splitting algorithms with rates and acceleration. Front. Signal Process, 1(776825), 2022

    2022

  5. [2]

    Condat, D

    L. Condat, D. Kitahara, A. Contreras, and A. Hirabayashi. Pro ximal splitting algorithms for convex optimization: a tour of recent advances, wit h new twists. SIAM Review , 65(2):375–435, 2023

    2023

  6. [3]

    Combettes

    P.L. Combettes. The geometry of monotone operator splitting m ethods. Acta Numer., 33:487–632, 2024

    2024

  7. [4]

    Lions and B

    P.L. Lions and B. Mercier. Splitting algorithms for the sum of two no nlinear operators. SIAM J. Numer. Anal. , 16(6):964–979, 1979

    1979

  8. [5]

    Combettes and J.-C

    P.L. Combettes and J.-C. Pesquet. A Douglas-Rachford splitting approach to nonsmooth convex variational signal recovery. IEEE J. Sel. Top. Signal Process , 1(4):564–574, 2007

    2007

Show all 44 references
  1. [6]

    P. Tseng. A modified forward-backward splitting method for max imal monotone mappings. SIAM J. Control Optim. , 38(2):431–446, 2000. 40

    2000

  2. [7]

    Lorenz and T

    D.A. Lorenz and T. Pock. An inertial forward-backward algorith m for monotone inclusions. J. Math. Imaging Vis. , 51:311–325, 2015

    2015

  3. [8]

    Bot ¸, E.R

    R.I. Bot ¸, E.R. Csetnek, and C. Hendrich. Inertial Douglas-Rac hford splitting for monotone inclusion problems. Appl. Math. Comput. , 256:472–487, 2015

    2015

  4. [9]

    Attouch and A

    H. Attouch and A. Cabot. Convergence rates of inertial forwa rd-backward algorithms. SIAM J. Optim. , 28(1):849–874, 2018

    2018

  5. [10]

    Brice˜ no-Arias and P.L

    L.M. Brice˜ no-Arias and P.L. Combettes. A monotone+skew split ting splitting model for composite monotone inclusions in duality. SIAM J. Control Optim. , 21(4):1230–1250, 2011

    2011

  6. [11]

    P. L. Combettes and J.C. Pesquet. Primal-dual splitting algorith m for solv- ing inclusions with mixtures of composite, lipschitzian, and parallel-sum type monotone operators. Set-Valued Var. Anal., 20(2):307–330, 2012

    2012

  7. [12]

    L. Condat. A primal-dual splitting method for convex optimizatio n involving lipschitzian, proximable and linear composite terms. J. Optim. Theory Appl. , 158:460–479, 2013

    2013

  8. [13]

    B.C. V˜ u. A splitting algorithm for dual monotone inclusions involvin g cocoer- cive operators. Adv. Comput. Math. , 38:667–681, 2013

    2013

  9. [14]

    Valkonen

    T. Valkonen. A primal-dual hybrid gradient method for nonlinear operators with applications to MRI. Inverse Probl. , 30:055012, 2014

    2014

  10. [15]

    Bot ¸ and E.R

    R.I. Bot ¸ and E.R. Csetnek. An inertial forward-backward-fo rward primal-dual splitting algorithm for solving monotone inclusion problems. Numer. Algo- rithms, 71:519–540, 2016

    2016

  11. [16]

    Combettes and J

    P.L. Combettes and J. Eckstein. Asynchronous block-iterativ e primal-dual de- composition methods for monotone inclusions. Math. Program., 168:645–672, 2018

    2018

  12. [17]

    Gao, X.C

    Y. Gao, X.C. Pan, and C. Chen. An extended primal-dual algorith m framework for nonconvex problems:application to image reconstruction in spec tral CT. Inverse Probl. , 38:085011, 2022

    2022

  13. [18]

    V.C. Bang, D. Papadimitriou, and V.X. Nham. A primal-dual backwa rd re- flected forward splitting algorithm for strutured monotone inclusio ns. Acta Math. Vietnam. , 49(2):159–172, 2024

    2024

  14. [20]

    Davis and W.T

    D. Davis and W.T. Yin. A three-operator splitting scheme and its o ptimization applications. Set-Valued Var. Anal., 25(4):829–858, 2017

    2017

  15. [21]

    Malitsky and M.K

    Y. Malitsky and M.K. Tam. Resolvent splitting for sums of monoton e operators with minimal lifting. Math. Program., 201:231–262, 2023

    2023

  16. [22]

    Arag´ on-Artacho, R

    F.J. Arag´ on-Artacho, R. Campoy, and C. L´ opez-Pastor. F orward-backward algorithms devised by graphs. SIAM J. Optim. , 35(4):2423–2451, 2025

    2025

  17. [23]

    Bredies, E

    K. Bredies, E. Chenchene, and E. Naldi. Graph and distribued ex tensions of the Douglas-Rachford method. SIAM J. Optim. , 34(2):1569–1594, 2024

    2024

  18. [24]

    M. N. Dao, M. K. Tam, and T. D. Truong. A general approach to distributed operator splitting. arXiv:2504.14987, 2025

    2025 arXiv

  19. [25]

    Bredies, E

    K. Bredies, E. Chenchene, and D.A. Lorenz. Degenerate prec onditioned proxi- mal point algorithms. SIAM J. Optim. , 32(3):2376–2401, 2022

    2022

  20. [26]

    Raguet, J

    H. Raguet, J. Fadili, and G. Peyr´ e. A generalized forward-bac kward splitting. SIAM J. Imaging Sci. , 6(3):1199–1226, 2013

    2013

  21. [27]

    Akerman, E

    A. Akerman, E. Chenchene, P. Giselsson, and E. Naldi. Splitting t he forward- backward algorithm: a full characterization. arXiv:2504.10999v1, 2025

    2025 arXiv

  22. [29]

    S. Banert. A relaxed forward-backward splitting algorithm for inclusions of sums of monotone operators. Master’s thesis, 2012

    2012

  23. [30]

    Brice˜ no-Arias

    L.M. Brice˜ no-Arias. Foward-partial inverse forward splitting for solving mono- tone inclusions. J. Optim. Theory Appl. , 166:391–413, 2015

    2015

  24. [31]

    Raguet and L

    H. Raguet and L. Landrieu. Preconditioning of a generalized for ward-backward splitting and application to optimization on graphs. SIAM J. Imaging Sci. , 8(4):2706–2739, 2015

    2015

  25. [32]

    Brice˜ no-Arias

    L.M. Brice˜ no-Arias. Forward-Douglas-Rachford splitting and forward-partial inverse method for solving monotone inclusions. Optim., 64:1239–1261, 2015

    2015

  26. [33]

    H. Raguet. A note on the forward-Douglas-Rachford splitting for monotone inclusion and convex optimization. Optim. Lett., 13(4):717–740, 2019

    2019

  27. [35]

    Yang, Y.C

    Y.X. Yang, Y.C. Tang, M. Wen, and T.Y. Zeng. Preconditioned Dou glas- Rachford type primal-dual method for solving composite monotone inclusion problems with applications. Inverse Probl. Imaging , 15(4):787–825, 2021

    2021

  28. [36]

    Zong, Y.C

    C.X. Zong, Y.C. Tang, and G.F. Zhang. Solving monotone inclusions involving the sum of three maximally monotone operators and a cocoercive op erator with applications. Set-Valued Var. Anal., 31:16, 2023

    2023

  29. [37]

    Zhang and P.C

    H.M. Zhang and P.C. Duan. Inertial four-operator splitting algo rithm for solv- ing monotone inclusion problems. J. Appl. Numer. Optim. , 7(2):253–274, 2025

    2025

  30. [38]

    Tang, L.Q

    Y.C. Tang, L.Q. Qin, G. Gibali, and J.G. Peng. A new splitting algorithm for monotone inclusion problems involving three maximally monotone opera tors and a cocoercive operator. Optimization, 2025

    2025

  31. [39]

    Cao, Y.H

    Y. Cao, Y.H. Wang, H. Rehman, and Y. Shehu. Convergence ana lysis of a new forward-reflected-backward algorithm for four operator witho ut cocoercivity. J. Optimiz. Theory App. , 203(1):256–284, 2024

    2024

  32. [40]

    Bauschke and P.L

    H.H. Bauschke and P.L. Combettes. Convex Analysis and Monotone Operator Theory in Hilbert Spaces . Springer, London, second edition, 2017

    2017

  33. [41]

    Selesnick, A

    I. Selesnick, A. Lanza, S. Morigi, and F. Sgallari. Nonconvex tot al variation regularization for convex denoising of signals. J. Math. Imaging Vision , 62:825– 841, 2020

    2020

  34. [42]

    Matsushita

    S.-Y. Matsushita. On the convergence rate of the Krasnosels ki-Mann iteration. Bull. Aust. Math. Soc , 96(1):162–170, 2017

    2017

  35. [43]

    Wang, A.C

    Z. Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality a s- sessment: From error visibility to structural similarity. IEEE Trans. Image Process., 13(4):600–612, 2004

    2004

  36. [44]

    Micchelli, L

    C.A. Micchelli, L. Shen, and Y. Xu. Proximity algorithms for image mo dels: denoising. Inverse Probl. , 27:045009, 2011. 43

    2011

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.