Pith. sign in

REVIEW 1 major objections 4 minor 45 references

For composite functions with weakly smooth f, the high-order forward-backward envelope is differentiable near calm points.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review

2026-08-03 22:27 UTC pith:JKY3MHDD

load-bearing objection Solid analytic work on high-order forward-backward envelopes, but the weak-smoothness theorem has a genuine proof gap that needs fixing. the 1 major comments →

arxiv 2511.10421 v2 pith:JKY3MHDD submitted 2025-11-13 math.OC

On fundamental properties of high-order forward-backward envelope

Alireza Kabgani , Masoud Ahookhosh This is my paper
classification math.OC MSC 90C2665K0549J5290C3049M27
keywords nonconvex composite optimizationhigh-order forward-backward envelopehigh-order forward-backward splittingweakly smooth functionsprox-regularityp-calmnessHölder continuityscaled gradient method
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper is about a smoothing mechanism for nonconvex composite problems φ=f+g where f has a Hölder-continuous gradient rather than a Lipschitz one. The authors study the high-order forward-backward splitting map (HiFBS) and its associated envelope (HiFBE), a p-power generalization of the classical forward-backward envelope. Their central claim: if g is prox-regular and the point of interest is p-calm, then the HiFBS map is locally single-valued and Hölder continuous, and the HiFBE is continuously differentiable with a gradient that is Hölder continuous with an explicitly quantified exponent. Since the HiFBE shares the infimum with φ, this gives a smooth surrogate on which gradient methods can be run, even though the original problem is nonsmooth and nonconvex. The differentiability order is derived explicitly as η=(ν/2)min{µ,ν}.

Core claim

The core discovery is a two-step bridge: local single-valuedness and Hölder continuity of HiFBS imply differentiability of HiFBE. Under f∈C^{1,ν}_{Lν}, with g prox-regular at a p-calm point x̄ and −∇f(x̄)∈∂g(x̄), there exists γ̄>0 such that for γ∈(0,γ̄) the map T^p_γφ is single-valued, continuous and ν/2-Hölder on a neighborhood of x̄; consequently φ^p_γ is C^1 there, and when f∈C^{2,µ} its gradient is Hölder of order η=(ν/2)min{µ,ν}. The gradient formula is ∇φ^p_γ(x̄)=∇²f(x̄)(ȳ−x̄)+γ^{-1}‖x̄−ȳ‖^{p−2}(x̄−ȳ) with ȳ=T^p_γφ(x̄), so smoothness of the envelope is transported from regularity of the splitting map.

What carries the argument

The central object is the pair (T^p_γφ, φ^p_γ): the high-order forward-backward splitting map defined by minimizing f(x)+⟨∇f(x), y−x⟩+g(y)+(1/pγ)‖x−y‖^p, and the envelope φ^p_γ(x) defined as that minimum value. The mechanism that carries the argument is the identity connecting the envelope gradient to the residual x−y, together with p-calmness (a local sharpness inequality) and prox-regularity of g, which yield a coercive quadratic bound on the map. This converts the set-valued stability of the splitting map into single-valuedness, Hölder continuity, and ultimately C^1 regularity of the envelope.

Load-bearing premise

The headline differentiability result requires that the nonsmooth term g be prox-regular at the calm point and that the function satisfy a global p-calmness inequality φ(x)+M‖x−x̄‖^p > φ(x̄); if either of these fails, the local single-valuedness of HiFBS—and therefore the C^1 property of the envelope—is no longer guaranteed.

What would settle it

Construct a function with f∈C^{1,ν}_{Lν}, g prox-regular at a p-calm point, yet T^p_γφ has multiple selections arbitrarily close to the calm point, or compute ∇φ^p_γ and show it violates the claimed ν/2-Hölder bound on arbitrarily small neighborhoods. Concretely, test f(x)=|x|^{3/2} composite with a prox-regular but nonconvex g, and check numerically whether the residual map x−T^p_γφ(x) is pointwise unique as γ→0.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • The iterative high-order forward-backward algorithm is equivalent to a scaled gradient method on the HiFBE, so convergence guarantees for gradient methods can be transferred to the nonconvex composite setting.
  • For the classical p=2 case, the differentiability result recovers the known FBE gradient formula ∇φ_γ=Q_γ(x)R_γ(x) as a special case.
  • For f=0 the HiFBE reduces to the high-order Moreau envelope, and the gradient method on it has the form of a scaled gradient step, unifying the theory.
  • The Hölder exponent η=(ν/2)min{µ,ν} gives a concrete, quantitative smoothness certificate that can be checked or exploited in complexity analyses.
  • If g is prox-regular at x̄, the local minimizer of the HiFBS subproblem is unique near the calm point, so the envelope is well-defined as a single-valued function.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A testable extension: compute the HiFBE on weakly smooth functions with known ν and µ and verify numerically that the gradient's Hölder exponent matches η; mismatches would reveal constants or assumptions needing refinement.
  • A natural next step the authors leave implicit is whether prox-regularity can be relaxed to a weaker local growth condition, which would widen the class of nonsmooth g for which the envelope is smooth.
  • The p-calmness condition is a sharpness/growth requirement; in unconstrained minimization it holds at isolated minimizers, but checking it at non-minimizing points may be delicate—an extension to p-calm neighborhoods rather than points would strengthen the theory.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 4 minor

Summary. The paper develops a variational-analytic theory of the high-order forward-backward splitting mapping (HiFBS) and high-order forward-backward envelope (HiFBE) for composite objectives φ=f+g in which f is weakly smooth, i.e. ∇f is ν-Hölder, and g is proper lsc (possibly nonconvex). The main results are: boundedness and uniform boundedness of HiFBS, Hölder (and, for p=2, Lipschitz) continuity of HiFBE, explicit Fréchet/limiting subdifferential formulas, a necessary-and-sufficient differentiability characterization, and — under p-calmness and prox-regularity of g at a calm point — local single-valuedness/continuity of HiFBS and C^1 weak smoothness of HiFBE with explicit Hölder exponents. The bridge is the identity ∇φ^p_γ(x)=∇²f(x)(T^p_γφ(x)-x)+γ^{-1}∥x-T^p_γφ(x)∥^{p-2}(x-T^p_γφ(x)), which converts single-valuedness of HiFBS into differentiability of HiFBE. Theorems 30 and 31 are the headline results.

Significance. If the main results hold, the paper substantially extends the classical forward-backward envelope framework from Lipschitz-smooth f to weakly smooth nonconvex composite problems, providing a principled basis for gradient-type algorithms with high-order regularization. The paper is careful about hypotheses, contains explicit constants, and many auxiliary results (Proposition 18, Theorem 24, Theorem 28) are independently valuable. The proofs are largely detailed and rigorous, and the p-calmness/prox-regularity hypotheses are genuine extra assumptions rather than disguised restatements of the conclusions. However, the proof of the central weak-smoothness statement in Theorem 31(b) contains an invalid Hölder estimate; the theorem is likely repairable, but the argument as written does not establish the claimed result.

major comments (1)
  1. [§4, Theorem 31(b), Eq. (4.17)] The displayed estimate ||∇²f(x2)(y2−x2) − ∇²f(x1)(y1−x1)|| ≤ L_μ||x2−x1+y2−y1||^μ is not a consequence of f∈C^{2,μ}. For a quadratic f with constant Hessian A, L_μ=0 while the left side is ||A[(y2−y1)−(x2−x1)]||, which is generally nonzero. The natural decomposition (∇²f(x2)−∇²f(x1))(y2−x2) + ∇²f(x1)((y2−y1)−(x2−x1)) gives a valid local bound of order ||x2−x1||^μ + ||x2−x1||^{ν/2}, i.e. Hölder order min(μ, ν/2). Since the stated exponent η=(ν/2)min{μ,ν} is no larger than min(μ,ν/2), Theorem 31(b) is plausibly salvageable by replacing this step, but as written the proof of the central weak-smoothness claim is not valid.
minor comments (4)
  1. [§4, Theorem 31(b), final display] The last display writes ||∇φ^p_γ(x2)−∇φ^p_γ(x1)|| ≤ L_μ||x2−x1||^μ; the exponent and constant should be L_η||x2−x1||^η as stated before the display.
  2. [§3.1, Theorem 20] The symbol L_ν is reused for both the given Hölder constant of ∇f and the newly constructed Hölder constant of φ^p_γ in the proof. Use L̄_ν or another symbol for the envelope constant to avoid confusion.
  3. [§3.2 and §4, Assumption 29] The proofs of Theorem 24 and Theorem 30 set x̄=0 and φ(x̄)=0. This is presumably a translation and normalization, but it is not stated explicitly; please clarify that the results hold for a general p-calm point x̄.
  4. [Preliminaries, Facts 10, 14, 16] Several foundational facts used throughout the paper (well-definedness of HOME/HOPE, basic properties of HiFBE/HiFBS) are quoted from the authors' preprints [23,24,26]. Since these are not yet peer-reviewed, the presentation would be more self-contained if those results were stated with proofs or at least with precise pointers to the relevant statements.

Circularity Check

0 steps flagged

No significant circularity: the differentiability and weak-smoothness results rest on genuine p-calmness/prox-regularity assumptions and on previously established lemmas, not on the conclusions being assumed.

full rationale

The paper's central claims (Theorems 30 and 31) are not reductions of their inputs. p-calmness (Definition 22) and prox-regularity of g at x̄=0 (Assumption 29) are substantive extra hypotheses about local growth and a quadratic lower model for g; they do not by themselves assert single-valuedness of HiFBS or differentiability/Hölder continuity of HiFBE. The proof of Theorem 30 derives local single-valuedness and the ν/2-Hölder estimate from strong monotonicity (Lemma 1(a)) together with the prox-regularity inequality, while Theorem 31 derives the gradient formula from the explicit expression (4.5) obtained in Theorem 28, whose proof uses only the definition of T^p_γφ and the envelope. The use of the authors' earlier work for Facts 14 and 16 and Lemma 1 is not circular under the stated criterion: those results have explicit assumptions (p=1+ν, prox-boundedness, f in C^{1,ν}) that do not include the target differentiability conclusions, and they are used as lemmas rather than as an unverified uniqueness or ansatz-forcing citation. No fitted parameters are being renamed as predictions, and no known result is merely relabeled. Even if the Hölder estimate at Eq. (4.17) were questionable, that would be a proof-gap/correctness concern, not an instance of the derivation being equivalent to its inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 6 axioms · 0 invented entities

The results rely on a chain of standard variational-analysis facts plus the authors' earlier high-order framework; there are no data-fitted constants. The main structural price is the p-calmness and prox-regularity assumptions placed on φ/g.

axioms (6)
  • domain assumption f∈C^{1,ν}_{Lν}(R^n) with ν∈(0,1] (Assumption 12)
    Defines the weakly smooth class; p=1+ν and all Hölder bounds for HiFBE rest on this.
  • domain assumption g proper, lsc, high-order prox-bounded with threshold γ_{g,p}>0 (Assumption 13)
    Guarantees HiFBS nonempty/compact and HiFBE finite; used throughout via Fact 14/16.
  • domain assumption x̅=0 is p-calm for φ with φ(0)=0 and g prox-regular at 0 for −∇f(0) (Assumption 29)
    Load-bearing for uniform boundedness, single-valuedness, continuity, differentiability near calm points.
  • domain assumption f∈C^2(U), and in Theorem 31 f∈C^{2,µ}
    Subdifferential formulas and gradient formula require second-order smoothness of the smooth part.
  • standard math Lemma 1(a): p-Laplacian strong monotonicity on bounded sets, from [26, Lemma 2]
    Used to separate two HiFBS values in Theorems 28 and 30; accepted as prior result.
  • standard math Hölderian descent lemma (Fact 5) from [45, Lemma 1]
    Underlies the majorant property and all f-majorization estimates.

reviewed 2026-08-03 · how reviews work

0 comments
Cite this review

Pith. "Pith review of On fundamental properties of high-order forward-backward envelope." pith.science (2026). https://pith.science/paper/JKY3MHDD

@misc{pith2026251110421,
  author       = {Pith},
  title        = {Pith review of: On fundamental properties of high-order forward-backward envelope},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JKY3MHDD}},
  note         = {Machine review of arXiv:2511.10421}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

This paper studies the fundamental properties of the high-order forward-backward splitting mapping (HiFBS) and its associated high-order forward-backward envelope (HiFBE) through the lens of high-order regularization for nonconvex composite functions. Specifically, we (i) establish the boundedness and uniform boundedness of HiFBS, along with the H\"older and Lipschitz continuity of HiFBE; (ii) derive an explicit form for the subdifferentials of HiFBE; and (iii) investigate necessary and sufficient conditions for the differentiability and weak smoothness of HiFBE under suitable assumptions. By leveraging the prox-regularity of $g$ and the concept of $p$-calmness, we further demonstrate the local single-valuedness and continuity of HiFBS, which in turn guarantee the differentiability of HiFBE in neighborhoods of calm points. This paves the way for the development of gradient-based algorithms tailored to nonconvex composite optimization problems.

Figures

Figures reproduced from arXiv: 2511.10421 by Alireza Kabgani, Masoud Ahookhosh.

Figure 1
Figure 1. Figure 1: Graphs of φ, M0.2(x, y), M0.5(x, y), and M1(x, y) The next example illustrates how the choice of γ affects the shape of the HiFBE and validates Fact 14 (d). Example 15 Consider f, g : R → R defined by f(x) = 0.5|x| 3 2 and g(x) = |0.3 sin(5x)| + 0.2x 2 e −x 2 and let φ(x) = f(x) + g(x). Here, f exhibits a 1 2 -H¨older continuous gradient, while g is nonconvex and oscillatory, introducing multiple local min… view at source ↗
Figure 2
Figure 2. Figure 2: Graphs of φ and φ p γ for different values of γ in Example 15. value, such as γ = 1 ( [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 1 canonical work pages

  1. [1]

    Mathematical Methods of Operations Research 89(3):319–353

    Ahookhosh M (2019) Accelerated first-order methods for large-scale convex optimization: Nearly optimal complexity under strong convexity. Mathematical Methods of Operations Research 89(3):319–353

    2019

  2. [2]

    SIAM Journal on Optimization 31(1):653–685

    Ahookhosh M, Themelis A, Patrinos P (2021) A Bregman forward-backward linesearch algorithm for non- convex composite optimization: Superlinear convergence to nonisolated local minima. SIAM Journal on Optimization 31(1):653–685

    2021

  3. [3]

    arXiv:250520484 URLhttps://doi.org/10.48550/arXiv.2505

    Ahookhosh M, Iusem A, Kabgani A, Lara F (2025) Asymptotic convergence analysis of high-order proximal- point methods beyond sublinear rates. arXiv:250520484 URLhttps://doi.org/10.48550/arXiv.2505. 20484

    doi:10.48550/arxiv.2505 2025

  4. [4]

    Springer International Publishing

    Bauschke H, Combettes P (2017) Convex Analysis and Monotone Operator Theory in Hilbert Spaces, 2nd edn. Springer International Publishing

    2017

  5. [5]

    Society for Industrial and Applied Mathematics

    Beck A (2017) First-Order Methods in Optimization. Society for Industrial and Applied Mathematics

    2017

  6. [6]

    SIAM Journal on Optimization 22:557–580

    Beck A, Teboulle M (2012) Smoothing and first order methods: A unified framework. SIAM Journal on Optimization 22:557–580

    2012

  7. [7]

    In: Dolecki S (ed) Optimization: Proceedings of the Fifth French-German Conference held in Castel-Novel (Varetz), France, Oct

    Ben-Tal A, Teboulle M (1989) A smoothing technique for nondifferentiable optimization problems. In: Dolecki S (ed) Optimization: Proceedings of the Fifth French-German Conference held in Castel-Novel (Varetz), France, Oct. 3–8, 1988, Lecture Notes in Mathematics, No. 1405, Springer-Verlag, pp 1–11

    1989

  8. [8]

    Journal of Optimization Theory and Applications 185:17–33

    Berger GO, Absil PA, Jungers RM, Nesterov Y (2020) On the quality of first-order approximation of functions with H¨ older continuous gradient. Journal of Optimization Theory and Applications 185:17–33

    2020

  9. [9]

    Athena Scientific, Belmont, MA

    Bertsekas DP (1999) Nonlinear Programming, 2nd edn. Athena Scientific, Belmont, MA

    1999

  10. [10]

    Open Journal of Mathematical Optimization 4:1–17

    Bolte J, Glaudin L, Pauwels E, Serrurier M (2023) The backtrack H¨ older gradient method with application to min-max and min-min problems. Open Journal of Mathematical Optimization 4:1–17

    2023

  11. [11]

    SIAM Journal on Imaging Sciences 18(2):1314–1343

    Bonettini S, Franchini G, Pezzi D, Prato M (2025) Linesearch-enhanced forward–backward methods for inexact nonconvex scenarios. SIAM Journal on Imaging Sciences 18(2):1314–1343

    2025

  12. [12]

    Journal of Scientific Computing 85:33

    Bot ¸ RI, B¨ ohm A (2020) Variable smoothing for convex optimization problems using stochastic gradients. Journal of Scientific Computing 85:33

    2020

  13. [13]

    TOP 23:124–150

    Bot, RI, Hendrich C (2015) A variable smoothing algorithm for solving convex optimization problems. TOP 23:124–150

    2015

  14. [14]

    Inverse Problems 25:015005

    Bredies K (2009) A forward–backward splitting algorithm for the minimization of non-smooth convex func- tionals in Banach space. Inverse Problems 25:015005

    2009

  15. [15]

    SIAM Journal on Optimization 23:1457–1479

    Burke JV, Hoheisel T (2013) Epi-convergent smoothing with applications to convex composite functions. SIAM Journal on Optimization 23:1457–1479

    2013

  16. [16]

    Set-Valued and Variational Analysis 25:1–23

    Burke JV, Hoheisel T (2017) Epi-convergence properties of smoothing by infimal convolution. Set-Valued and Variational Analysis 25:1–23

    2017

  17. [17]

    Set-Valued and Variational Analysis 21:359–376

    Burke JV, Hoheisel T, Kanzow C (2013) Gradient consistency for integral-convolution smoothing functions. Set-Valued and Variational Analysis 21:359–376

    2013

  18. [18]

    Optimization Methods and Soft- ware 32:1273–1298

    Cartis C, Gould N, Toint P (2017) Worst-case evaluation complexity of regularization methods for smooth unconstrained optimization using H¨ older continuous gradients. Optimization Methods and Soft- ware 32:1273–1298

    2017

  19. [19]

    Mathematical Programming 134:71–99

    Chen X (2012) Smoothing methods for nonsmooth, nonconvex minimization. Mathematical Programming 134:71–99

    2012

  20. [20]

    Multiscale Mod- eling & Simulation 4(4):1168–1200

    Combettes PL, Wajs VR (2005) Signal recovery by proximal forward-backward splitting. Multiscale Mod- eling & Simulation 4(4):1168–1200

    2005

  21. [21]

    Mathematical Programming 178:503–558

    Drusvyatskiy D, Paquette C (2019) Efficiency of minimizing compositions of convex functions and smooth maps. Mathematical Programming 178:503–558

    2019

  22. [22]

    Applied Mathematics and Computation 464:128377

    Ghaderi S, Ahookhosh M, Arany A, Skupin A, Patrinos P, Moreau Y (2024) Smoothing unadjusted langevin algorithms for nonsmooth composite potential functions. Applied Mathematics and Computation 464:128377

    2024

  23. [23]

    arXiv URLhttps://arxiv.org/abs/2410.19928

    Kabgani A, Ahookhosh M (2024) ItsOPT: An inexact two-level smoothing framework for nonconvex opti- mization via high-order Moreau envelope. arXiv URLhttps://arxiv.org/abs/2410.19928

    Pith/arXiv arXiv 2024

  24. [24]

    arXiv URLhttps://doi.org/10.48550/arXiv

    Kabgani A, Ahookhosh M (2025) First-order majorization-minimization meets high-order majorant: Boosted inexact high-order forward-backward method. arXiv URLhttps://doi.org/10.48550/arXiv. 2510.22231

    work page doi:10.48550/arxiv.2510.22231 2025

  25. [25]

    arXiv URLhttps://doi.org/10.48550/arXiv.2501.02155 20 Alireza Kabgani, Masoud Ahookhosh

    Kabgani A, Ahookhosh M (2025) ItsDEAL: Inexact two-level smoothing descent algorithms for weakly convex optimization. arXiv URLhttps://doi.org/10.48550/arXiv.2501.02155 20 Alireza Kabgani, Masoud Ahookhosh

    Pith/arXiv arXiv doi:10.48550/arxiv.2501.02155 2025

  26. [26]

    arXiv URLhttps://doi.org/10.48550/arXiv.2503.04577

    Kabgani A, Ahookhosh M (2025) Moreau envelope and proximal-point methods under the lens of high-order regularization. arXiv URLhttps://doi.org/10.48550/arXiv.2503.04577

    doi:10.48550/arxiv.2503.04577 2025

  27. [27]

    Nonlinear Analysis: Theory, Methods & Applications 127:157–181

    Kecis I, Thibault L (2015) Moreau envelopes ofs-lower regular functions. Nonlinear Analysis: Theory, Methods & Applications 127:157–181

    2015

  28. [28]

    Society for Industrial and Applied Mathematics

    Lange K (2016) MM Optimization Algorithms. Society for Industrial and Applied Mathematics

    2016

  29. [29]

    Springer Cham

    Mordukhovich BS (2018) Variational Analysis and Applications. Springer Cham

    2018

  30. [30]

    Bulletin de la Soci´ et´ e Math´ ematique de France 93:273–299

    Moreau JJ (1965) Proximit´ e et dualit´ e dans un espace Hilbertien. Bulletin de la Soci´ et´ e Math´ ematique de France 93:273–299

    1965

  31. [31]

    Mathematical programming 103:127–152

    Nesterov Y (2005) Smooth minimization of non-smooth functions. Mathematical programming 103:127–152

    2005

  32. [32]

    Mathematical Program- ming 152:381–404

    Nesterov Y (2015) Universal gradient methods for convex optimization problems. Mathematical Program- ming 152:381–404

    2015

  33. [33]

    Mathematical Programming 197:1–26

    Nesterov Y (2023) Inexact accelerated high-order proximal-point methods. Mathematical Programming 197:1–26

    2023

  34. [34]

    Foundations and Trends ® in Optimization 1:127–239

    Parikh N, Boyd S (2014) Proximal algorithms. Foundations and Trends ® in Optimization 1:127–239

    2014

  35. [35]

    In: 52nd IEEE Conference on Decision and Control, IEEE, pp 2358–2363

    Patrinos P, Bemporad A (2013) Proximal Newton methods for convex composite optimization. In: 52nd IEEE Conference on Decision and Control, IEEE, pp 2358–2363

    2013

  36. [36]

    Transactions of the American Mathematical Society 348:1805–1838

    Poliquin RA, Rockafellar RT (1996) Prox-regular functions in variational analysis. Transactions of the American Mathematical Society 348:1805–1838

    1996

  37. [37]

    Springer Berlin, Heidelberg

    Rockafellar RT, Wets RJB (2009) Variational Analysis. Springer Berlin, Heidelberg

    2009

  38. [38]

    Journal of Optimiza- tion Theory and Applications 185:303–326

    Rodomanov A, Nesterov Y (2020) Smoothness parameter of power of Euclidean norm. Journal of Optimiza- tion Theory and Applications 185:303–326

    2020

  39. [39]

    Mathematical Programming 159(1-2):1370–164

    Shefi R, Teboulle M (2016) A dual method for minimizing a nonsmooth objective over one smooth inequality constraint. Mathematical Programming 159(1-2):1370–164

    2016

  40. [40]

    Computational Optimization and Applications 67:443–487

    Stella L, Themelis A, Patrinos P (2017) Forward–backward quasi-Newton methods for nonsmooth opti- mization problems. Computational Optimization and Applications 67:443–487

    2017

  41. [41]

    IEEE Transactions on Signal Processing 65(3):794–816

    Sun Y, Babu P, Palomar DP (2016) Majorization-minimization algorithms in signal processing, communi- cations, and machine learning. IEEE Transactions on Signal Processing 65(3):794–816

    2016

  42. [42]

    SIAM Journal on Optimization 30:149–181

    Themelis A, Patrinos P (2020) Douglas–Rachford splitting and ADMM for nonconvex optimization: Tight convergence results. SIAM Journal on Optimization 30:149–181

    2020

  43. [43]

    SIAM Journal on Optimization 28(3):2274–2303

    Themelis A, Stella L, Patrinos P (2018) Forward-backward envelope for the sum of two nonconvex functions: Further properties and nonmonotone linesearch algorithms. SIAM Journal on Optimization 28(3):2274–2303

    2018

  44. [44]

    In: Bauschke HH, Burachik RS, Luke DR (eds) Splitting Algorithms, Modern Operator Theory, and Applications, Springer Cham, pp 363–412

    Themelis A, Ahookhosh M, Patrinos P (2019) On the acceleration of forward-backward splitting via an inexact Newton method. In: Bauschke HH, Burachik RS, Luke DR (eds) Splitting Algorithms, Modern Operator Theory, and Applications, Springer Cham, pp 363–412

    2019

  45. [45]

    Optimization Letters 10:1361–1370

    Yashtini M (2016) On the global convergence rate of the gradient descent method for functions with H¨ older continuous gradients. Optimization Letters 10:1361–1370

    2016

This paper was first reviewed by deepseek-v4-flash on August 3, 2026.