Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Nonconvex Optimization Framework for Group-Sparse Feedback Linear-Quadratic Optimal Control: Penalty Approach

T0 review · 3 major / 4 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 This paper claims that group-sparse feedback LQ control can be solved directly by a penalty-based PALM algorithm with guaranteed convergence to a KKT point.

desk verdict The PALM machinery is real, but the control reformulation has a load-bearing gap: (16) restricts rather than relaxes, and the penalty is not shown exact. read the letter →

arxiv 2507.18114 v3 pith:UN7RHR3C submitted 2025-07-24 math.OC cs.LG

Lechen Feng , Xun Li , Yuan-Hua Ni This is my paper · ORCID
classification math.OCcs.LG MSC 90C2693B5249M37
keywords nonconvexoptimizationgroupl0normsparsefeedbackcontroldistributedLQPALMpenaltymethodlinear-quadraticregulatorconvergenceanalysis
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to solve the infinite-horizon linear-quadratic control problem with group-sparse feedback directly, as a finite-dimensional nonconvex optimization problem carrying a group $\ell_0$-norm penalty, instead of passing through $\ell_1$ or semidefinite relaxations or confining the gain to block-diagonal form. It claims that the sparse-feedback LQ problem and the distributed LQ problem with a fixed communication topology are two faces of one optimization model, obtained by replacing the constraint $K = W_2^\top W_1^{-1}$ with a linear constraint on $W_1$ and a group-$\ell_0$ term on $W_2^\top$. On the resulting penalized problem it runs a proximal alternating linearized minimization (PALM) algorithm and proves that, when the parameters satisfy explicit inequalities, the iterates are bounded, have finite length, and converge to a KKT point, with linear or sublinear rates depending on a Łojasiewicz exponent. If correct, this gives a controller-design method that returns group-sparse stabilizing gains together with a convergence certificate, without a convex surrogate or block-diagonal-only structure.

What carries the argument

The load-bearing construction is a Lyapunov parameterization: a set $\mathcal{C}$ of semidefinite matrices defined by $W \succeq 0$ and $-V_2(F_i W + W F_i^\top + Q)V_2^\top \succeq 0$, with $K = W_2^\top W_1^{-1}$, which guarantees closed-loop stability and gives $\langle R, W\rangle$ as an upper bound on the $H_2$ cost. The sparsity inheritance identity — block-diagonal $W_1$ forces the pattern of $W_2^\top$ onto $K$ — turns the nonconvex manifold constraint into the linear constraint $V_1 W V_2^\top - P = 0$, and the group-$\ell_0$ term lands on $P$. Around this, the paper builds a regularized augmented Lagrangian $\Psi$ that is semi-algebraic (hence satisfies the Kurdyka-Łojasiewicz inequality), and this is what makes the PALM convergence and rate theorems apply without coercivity.

What would settle it

Take a small group-sparse LQ instance, solve the exact mixed-integer reformulation (47) to global optimality, and run Algorithm 3 with parameters satisfying (43). If the PALM limit has strictly larger LQ cost than the MISDP optimum, or converges to a KKT point of (24) whose $W_1$ cannot be made block-diagonal, then the relaxation is lossy and the paper's central claim fails on that instance.

Watch

Extended reading notes

Core claim

The central claim is that the relaxation from (15) to (16) is benign and algorithmically useful: when $W_1$ is restricted to $\mathrm{Sparse}(\mathrm{blockdiag}(I_{n_1},\ldots,I_{n_t}))$, the sparsity pattern of $W_2^\top$ is inherited by the feedback gain $K = W_2^\top W_1^{-1}$, so the group-$\ell_0$ objective can be moved from $K$ to $W_2^\top$ at the cost of a linear constraint. The constrained problem is then written as $\min f(\vec W) + g(\vec P)$ subject to $A\vec W + B\vec P = 0$, and the linear constraint is penalized as $(\rho/2)\|A\vec W + B\vec P\|^2$. The paper shows that PALM applied to this penalty function has fully solvable subproblems — a pBCD scheme over a dual quadratic semidefinite problem for the $z$-update and a closed-form group-$\ell_0$ proximal step for the $P$-update — and that the whole iteration converges to a KKT point of the penalized problem with finite length and known rates, despite $g$ and the penalty term not being coercive.

Load-bearing premise

The load-bearing premise is that every optimal group-sparse stabilizing feedback gain $K$ can be represented as $K = W_2^\top W_1^{-1}$ with a block-diagonal $W_1$ satisfying the defining LMIs of $\mathcal{C}$; the paper assumes this without proving it, so if it fails on some instance, the problem being solved is a strict subset of the intended sparse LQ problem.

Editorial extensions

If this is right

  • Any run of Algorithm 3 with parameters satisfying the selection criterion (43) produces a bounded sequence of finite length that converges to a KKT point of the penalized problem (24).
  • The convergence is linear when the Łojasiewicz exponent is at most $1/2$ and sublinear otherwise; for the present semi-algebraic structure the paper argues the exponent is $1/2$ under suitable conditions, so linear convergence is expected.
  • The same update scheme solves the cardinality-constrained variant where at most $S$ communication links are allowed, because $\delta_{C_S}$ is semi-algebraic and the proximal step still has a closed form.
  • The DFT-LQ problem with a fixed communication topology, and the mixed problem with prescribed forbidden entries plus group sparsity, reduce to the same standard form (19) with extra linear constraints, so the same algorithm and convergence analysis apply.
  • In the numerical examples, PALM reaches a group-sparse gain with substantially lower LQ cost than the group-$\ell_1$ relaxation at the same sparsity level (1.428 versus 5.855), which is the failure mode of convex surrogates the paper targets.

Reading between the lines

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

  • The unproved step is the representability assumption that an optimal group-sparse stabilizing $K$ admits a block-diagonal $W_1$ in $\mathcal{C}$; if that fails on some instance, the solved problem is a strict subset and the KKT point need not correspond to the true sparse LQ optimum. A small comparison against the exact MISDP formulation (47) would settle how often this gap bites.
  • Because the convergence target is a KKT point of the penalized problem rather than of the constrained problem, the practical guarantee depends on $\rho$ being large enough; the paper's own ADMM experiments show oscillation and feasibility trade-offs with $\beta$, suggesting the penalty route may need a tuning rule or an exact-penalty variant.
  • The same proof architecture — semi-algebraic regularized augmented Lagrangian plus PALM without coercivity — is portable to other structured control problems with $\ell_0$ or cardinality terms, such as actuator and sensor selection with stability constraints, where coercivity also typically fails.
  • The framework's allowance of overlapping observation sets is a genuine generalization of the disjoint-block distributed designs in earlier LMI work; one consequence is that arbitrary communication patterns, not just chordal or block-diagonal ones, become algorithmically accessible.
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 nonconvex optimization framework for designing group-sparse state feedback gains for infinite-horizon LQ problems, covering both distributed (DFT-LQ) and sparse (SF-LQ) variants. It parameterizes stabilizing gains through an LMI set C, introduces a group-ℓ0 penalty, converts the problem into a linearly constrained composite form (19), and then proposes a quadratic penalty relaxation (24) solved by a PALM algorithm with closed-form subproblem solvers. Convergence to a KKT point of (24) and convergence rates are claimed under the parameter selection criterion (43), relying on the KL framework of [8,9]. Numerical experiments on small systems illustrate the behavior of both the analyzed PALM scheme and an ADMM scheme whose analysis is deferred to a companion paper.

Significance. If the formulation and convergence analysis were correct, the paper would offer a direct nonconvex alternative to ℓ1 relaxations and to block-diagonal-only designs, with provable convergence to critical points and with efficient subproblem solvers. The paper is transparent about its dependence on the machinery of [9] and it gives an explicit parameter selection criterion. However, the modeling chain from the original sparse LQ problem to the solved penalized problem contains a restrictive structural assumption that is presented as a relaxation, and the penalty equivalence and boundedness arguments are not established. Because these issues affect the central claims of the paper, the contribution in its current form does not meet the standard for publication.

major comments (3)
  1. [Section 2.1, Eq. (16)] The transition from (15) to (16) is described as a 'relaxation', but it imposes the additional constraint W1 ∈ Sparse(blockdiag(In1,...,Int)), i.e., W1 is block-diagonal. Adding a constraint shrinks the feasible set; the only implication proved is that sparse W2^T with block-diagonal W1 yields sparse K = W2^T W1^{-1}. The converse — that every optimal group-sparse stabilizing K in (15) admits such a block-diagonal Lyapunov certificate — is not proved and is known to be restrictive in decentralized control. Hence the optimal value of (16) can be strictly larger than that of (15), and a KKT point of (24) may be suboptimal or infeasible for the intended SF-LQ problem (4)/(12). The abstract's claim of solving the problem 'without restrictive structural assumptions' is therefore not supported by the formulation.
  2. [Section 3, problem (24)] The assertion 'As ρ→∞, optimization problem (19) becomes equivalent to optimization problem (24)' is stated without proof. No exact-penalty result is established: no constraint qualification, multiplier boundedness, or error bound is given. Theorem 4 concerns convergence to a KKT point of (24) for a fixed ρ satisfying (43), and the KKT conditions listed in Theorem 4 do not imply AfW* + B eP* = 0. Consequently, even a fully successful run of Algorithm 3 is not shown to produce a feasible or near-feasible point of the constrained problem (19). The empirical feasibility observed in Figures 1b and 1h is not a substitute for a theoretical guarantee.
  3. [Appendix, proof of Theorem 2] The proof defines B* = sup_n {‖fWn+1−fWn‖, ‖ePn+1−ePn‖, ‖zn+1−zn‖, ‖un+1−un‖} and states 'based on Theorem 2.8 of [9], it holds that B* < ∞'. This boundedness of successive differences is itself a nontrivial conclusion in [9] obtained under coercivity or related assumptions; the paper explicitly notes (Section 4, text after (40)) that those assumptions fail for g and H. No verification that the hypotheses of Theorem 2.8 hold in the present non-coercive setting is provided. Since boundedness of the iterates is a load-bearing input for Theorem 4, the convergence proof is incomplete.
minor comments (4)
  1. [Eq. (16)] The expression ‖W2^T‖0 does not specify the block structure; the group ℓ0-norm is only defined for partitioned matrices as ‖·‖s,t;0, so the notation should be ‖W2^T‖s,t;0 or the partition should be stated explicitly.
  2. [Section 6, Example 1] The feasibility check is written as '‖AfWk − B ePk‖→0', but the constraint in (19) is AfW + B eP = 0; as written, the displayed residual has the wrong sign and should be ‖AfWk + B ePk‖ (or the appropriate residual for the particular definition of B).
  3. [Section 3] The phrase 'As ρ→∞, optimization problem (19) becomes equivalent to optimization problem (24)' should be replaced by a formal lemma with explicit hypotheses; as written it is an assertion that is not used in a precise way in the subsequent convergence theorems.
  4. [Section 5 and Section 6] The numerical experiments of Section 6 include results for the ADMM algorithm (46), whose convergence analysis is deferred to a companion paper [18]; the paper should clearly indicate which figures correspond to the PALM algorithm analyzed here and which correspond to ADMM, since the current presentation mixes both without a clear separation.

Circularity Check

0 steps flagged · score 2.0 of 10

Convergence analysis is self-contained; self-citations are minor and not load-bearing.

full rationale

The paper's core convergence results (Theorems 4-5) are not circular: the sufficient-decrease bound, boundedness, KL/semi-algebraic property, finite length, and linear/sublinear rates are established in-paper or reduced to the external PALM machinery [8,9] with the missing hypotheses (boundedness, KL) proved inside the paper. No parameter is fitted to a target result and no fitted quantity is relabeled as a prediction. The self-citations are not load-bearing: [18] defers the ADMM follow-up, [19] supplies a comparison algorithm, and [20] provides vectorization-matrix details and a subproblem-solver methodology that is also rederived in Lemmas 1-2. The main weakness is the transition (15)-(16): the paper calls it a 'relaxation' but it adds the constraint W1∈Sparse(blockdiag(...)), which is a restriction, and the sparsity inheritance is a sufficient condition built into the new problem rather than an equivalence with the original SF-LQ problem. This is a correctness/completeness gap, not a definitional circularity; from (16) onward the equivalences and convergence chain are exact.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The ledger shows two user-chosen weights (γ and ρ) plus hand-tuned algorithm parameters, four domain assumptions imported from prior control theory, and two unproven reformulation assumptions that are load-bearing for the central claim. No new physical entities are introduced.

free parameters (3)
  • γ (sparsity regularization weight) = not reported for PALM examples; γ=50 and 200 for the ℓ1 baseline
    Balances LQ cost against the group-ℓ0 penalty; user-selected and central to the objective.
  • ρ (penalty weight for the affine constraint) = not reported for numerical examples
    Required for the PALM parameter conditions (43) but its value is omitted, which also affects feasibility and convergence behavior.
  • PALM step-size parameters (μ, β, τ, σ) = Example 1: σ=1/50, β=6618, μ=1309, τ=10454
    Hand-chosen to satisfy criterion (43); they depend on ρ and on problem data through κ1, κ2, κ3.
assumptions (6)
  • domain assumption Assumption 1: C^T D=0, D^T D≻0, C^T C≻0, B1B1^T≻0, (A,B2) stabilizable, and (A,C) has no unobservable modes on the imaginary axis.
    Standard hypotheses imported from [24,28,31] for the LQ cost and for Theorem 1; stated in Section 2.1.
  • domain assumption Assumption 2: F is convex-bounded, F = Σ ξ_i F_i with ξ_i ≥ 0 and Σ ξ_i = 1.
    Defines the uncertain domain with M vertices; for a known system, M=1; stated in Section 2.1.
  • domain assumption W1≻0 for all W in C (implicit assumption from [31]).
    Needed for K = W2^T W1^{-1} to be defined and for Theorem 1 to give a stabilizing controller; the paper says it generally holds in practice (Remark 2) but does not prove it.
  • ad hoc to paper Problem (26) is strictly feasible, i.e., Slater's condition holds.
    Required for strong duality and for the pBCD dual solver; the paper acknowledges the general conditions are open and only quotes partial sufficient conditions from [50] (Section 3.1, Assumption 3).
  • ad hoc to paper Optimal solutions of (15) can be restricted to W1 in Sparse(blockdiag(...)) without changing the optimal value.
    The paper imposes this constraint in (16) and calls it a relaxation, but gives no proof that all relevant sparse K are representable with block-diagonal W1 satisfying the LMIs (Section 2.1).
  • ad hoc to paper As ρ→∞, the penalized problem (24) becomes equivalent to the constrained problem (19).
    Stated as motivation, but no exact penalty theorem is proved; convergence to a KKT point of (24) does not guarantee feasibility or optimality for (19) (Sections 3 and 4).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Nonconvex Optimization Framework for Group-Sparse Feedback Linear-Quadratic Optimal Control: Penalty Approach." pith.science (2026). https://pith.science/paper/UN7RHR3C

@misc{pith2026250718114,
  author       = {Pith},
  title        = {Pith review of: Nonconvex Optimization Framework for Group-Sparse Feedback Linear-Quadratic Optimal Control: Penalty Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UN7RHR3C}},
  note         = {Machine review of arXiv:2507.18114}
}
abstract

This paper develops a unified nonconvex optimization framework for the design of group-sparse feedback controllers in infinite-horizon linear-quadratic (LQ) problems. We address two prominent extensions of the classical LQ problem: the distributed LQ problem with fixed communication topology (DFT-LQ) and the sparse feedback LQ problem (SF-LQ), both of which are motivated by the need for scalable and structure-aware control in large-scale systems. Unlike existing approaches that rely on convex relaxations or are limited to block-diagonal structures, we directly formulate the controller synthesis as a finite-dimensional nonconvex optimization problem with group $\ell_0$-norm regularization, capturing general sparsity patterns. We establish a connection between DFT-LQ and SF-LQ problems, showing that both can be addressed within our unified framework. Furthermore, we propose a penalty-based proximal alternating linearized minimization (PALM) algorithm and provide a rigorous convergence analysis under mild assumptions, overcoming the lack of coercivity in the objective function. The proposed method admits efficient solvers for all subproblems and guarantees global convergence to critical points. Our results fill a key gap in the literature by enabling the direct design of group-sparse feedback gains with theoretical guarantees, without resorting to convex surrogates or restrictive structural assumptions.

Figures

Figures reproduced from arXiv: 2507.18114 by the authors.

Figure 1
Figure 1. Figures of Section 6 26 [PITH_FULL_IMAGE:figures/full_fig_p026_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Douglas-Rachford Splitting for Group-Sparse Feedback Linear-Quadratic Control

    math.OC 2025-07 reject novelty 5.0 of 10

    Direct ADMM/DR splitting is proposed for group-sparse LQ control, with convergence claimed under an unverified smoothness condition on the epi-composed objective.

Reference graph

Works this paper leans on

50 extracted references · 41 canonical work pages · cited by 1 Pith paper

  1. [9]

    A proximal minimization algorithm for structured nonconvex and nonsmooth problems.SIAM Journal on Optimization, 29(2):1300– 1328, 2019

    Radu Ioan Bot, Erno Robert Csetnek, and Dang-Khoa Nguyen. A proximal minimization algorithm for structured nonconvex and nonsmooth problems.SIAM Journal on Optimization, 29(2):1300– 1328, 2019

    work page 2019

  2. [1]

    Closed-loop feedback sparsification under parametric uncertainties

    Reza Arastoo, MirSaleh Bahavarnia, Mayuresh V Kothare, and Nader Motee. Closed-loop feedback sparsification under parametric uncertainties. In 2016 IEEE 55th Conference on Decision and Control (CDC), pages 123–128. IEEE, 2016

    work page 2016

  3. [2]

    Sparse linear-quadratic feedback design using affine approximation.arXiv preprint arXiv:1507.08592, 2015

    MirSaleh Bahavarnia. Sparse linear-quadratic feedback design using affine approximation.arXiv preprint arXiv:1507.08592, 2015

    arXiv 2015

  4. [3]

    Convergence for nonconvex admm, with applications to ct imaging

    Rina Foygel Barber and Emil Y Sidky. Convergence for nonconvex admm, with applications to ct imaging. Journal of Machine Learning Research, 25(38):1–46, 2024

    work page 2024

  5. [4]

    Optimization problems involving group sparsity terms.Mathematical Programming, 178:39–67, 2019

    Amir Beck and Nadav Hallak. Optimization problems involving group sparsity terms.Mathematical Programming, 178:39–67, 2019

    work page 2019

  6. [5]

    KL property of exponent $1/2$ of $\ell_{2,0}$-norm and DC regularized factorizations for low-rank matrix recovery

    Shujun Bi, Ting Tao, and Shaohua Pan. Kl property of exponent1/2 of ℓ2,0-norm and dc regularized factorizations for low-rank matrix recovery.arXiv preprint arXiv:1908.09078, 2019

    work page Pith review arXiv 1908

  7. [6]

    The łojasiewicz inequality for nonsmooth subana- lytic functions with applications to subgradient dynamical systems.SIAM Journal on Optimization, 17(4):1205–1223, 2007

    Jérôme Bolte, Aris Daniilidis, and Adrian Lewis. The łojasiewicz inequality for nonsmooth subana- lytic functions with applications to subgradient dynamical systems.SIAM Journal on Optimization, 17(4):1205–1223, 2007

    work page 2007

  8. [7]

    Clarke subgradients of stratifiable functions

    Jérôme Bolte, Aris Daniilidis, Adrian Lewis, and Masahiro Shiota. Clarke subgradients of stratifiable functions. SIAM Journal on Optimization, 18(2):556–572, 2007

    work page 2007

Show all 50 references
  1. [8]

    Proximal alternating linearized minimization for nonconvex and nonsmooth problems.Mathematical Programming, 146(1):459–494, 2014

    Jérôme Bolte, Shoham Sabach, and Marc Teboulle. Proximal alternating linearized minimization for nonconvex and nonsmooth problems.Mathematical Programming, 146(1):459–494, 2014

    2014

  2. [10]

    The proximal alternating direction method of multipliers in the nonconvex setting: convergence analysis and rates

    Radu Ioan Boţ and Dang-Khoa Nguyen. The proximal alternating direction method of multipliers in the nonconvex setting: convergence analysis and rates. Mathematics of Operations Research, 45(2):682–712, 2020

    2020

  3. [11]

    Cambridge university press, 2004

    Stephen Boyd and Lieven Vandenberghe.Convex optimization. Cambridge university press, 2004

    2004

  4. [12]

    Outer approximation with conic certificates for mixed-integer convex problems.Mathematical Programming Computation, 12(2):249–293, 2020

    Chris Coey, Miles Lubin, and Juan Pablo Vielma. Outer approximation with conic certificates for mixed-integer convex problems.Mathematical Programming Computation, 12(2):249–293, 2020

    2020

  5. [13]

    An introduction to semialgebraic geometry, 2000

    Michel Coste. An introduction to semialgebraic geometry, 2000

    2000

  6. [14]

    The design of controllers for the multivariable robust servomech- anism problem using parameter optimization methods.IEEE Transactions on Automatic Control, 26(1):93–110, 1981

    Edward Davison and Ian Ferguson. The design of controllers for the multivariable robust servomech- anism problem using parameter optimization methods.IEEE Transactions on Automatic Control, 26(1):93–110, 1981

    1981

  7. [15]

    Sparse approximations with interior point methods.Siam review, 64(4):954–988, 2022

    Valentina De Simone, Daniela di Serafino, Jacek Gondzio, Spyridon Pougkakiotis, and Marco Viola. Sparse approximations with interior point methods.Siam review, 64(4):954–988, 2022

    2022

  8. [16]

    Solving infinite-dimensional optimization problems by polynomial approximation

    Olivier Devolder, François Glineur, and Yurii Nesterov. Solving infinite-dimensional optimization problems by polynomial approximation. InRecent Advances in Optimization and its Applications in Engineering: The 14th Belgian-French-German Conference on Optimization, pages 31–40...

    2010

  9. [17]

    Convex relaxation for optimal distributed control problems

    Ghazal Fazelnia, Ramtin Madani, Abdulrahman Kalbat, and Javad Lavaei. Convex relaxation for optimal distributed control problems. IEEE Transactions on Automatic Control, 62(1):206–221, 2016

    2016

  10. [18]

    Nonconvex optimization framework for group-sparse feedback linear-quadratic optimal control: Non-penalty approach.arXiv preprint arXiv:2507.19895, 2025

    Lechen Feng, Xun Li, and Yuan-Hua Ni. Nonconvex optimization framework for group-sparse feedback linear-quadratic optimal control: Non-penalty approach.arXiv preprint arXiv:2507.19895, 2025. 28

    2025 arXiv

  11. [19]

    Optimization-based design of communication topology for decentral- ized linear-quadratic problem

    Lechen Feng and Yuan-Hua Ni. Optimization-based design of communication topology for decentral- ized linear-quadratic problem. In2024 43rd Chinese Control Conference (CCC), pages 5865–5870. IEEE, 2024

    2024

  12. [20]

    Two-timescale optimization framework for decen- tralized linear-quadratic optimal control.arXiv preprint arXiv:2406.11168, 2024

    Lechen Feng, Yuan-Hua Ni, and Xuebo Zhang. Two-timescale optimization framework for decen- tralized linear-quadratic optimal control.arXiv preprint arXiv:2406.11168, 2024

    2024 arXiv

  13. [21]

    Sparsity invari- ance for convex design of distributed controllers.IEEE Transactions on Control of Network Systems, 7(4):1836–1847, 2020

    Luca Furieri, Yang Zheng, Antonis Papachristodoulou, and Maryam Kamgarpour. Sparsity invari- ance for convex design of distributed controllers.IEEE Transactions on Control of Network Systems, 7(4):1836–1847, 2020

    2020

  14. [22]

    A gauss–seidel type inertial proximal alternating linearized minimization for a class of nonconvex optimization problems

    Xue Gao, Xingju Cai, and Deren Han. A gauss–seidel type inertial proximal alternating linearized minimization for a class of nonconvex optimization problems. Journal of Global Optimization, 76:863–887, 2020

    2020

  15. [23]

    Courier Corporation, 2000

    Izrail Moiseevitch Gelfand, Richard A Silverman, et al.Calculus of variations. Courier Corporation, 2000

    2000

  16. [24]

    Decentralized control through parameter space optimization.Automatica, 30(10):1565–1578, 1994

    José Claudio Geromel, Jacques Bernussou, and Pedro Luis Dias Peres. Decentralized control through parameter space optimization.Automatica, 30(10):1565–1578, 1994

    1994

  17. [25]

    Iteration complexity analysis of block coordinate descent methods.Mathematical Programming, 163:85–114, 2017

    Mingyi Hong, Xiangfeng Wang, Meisam Razaviyayn, and Zhi-Quan Luo. Iteration complexity analysis of block coordinate descent methods.Mathematical Programming, 163:85–114, 2017

    2017

  18. [26]

    Prentice hall, 1996

    IS Khalil, JC Doyle, and K Glover.Robust and optimal control, volume 2. Prentice hall, 1996

    1996

  19. [27]

    Global convergence of splitting methods for nonconvex composite optimization

    Guoyin Li and Ting Kei Pong. Global convergence of splitting methods for nonconvex composite optimization. SIAM Journal on Optimization, 25(4):2434–2460, 2015

    2015

  20. [28]

    Yingying Li, Yujie Tang, Runyu Zhang, and Na Li. Distributed reinforcement learning for decentral- ized linear quadratic control: A derivative-free policy optimization approach.IEEE Transactions on Automatic Control, 67(12):6429–6444, 2021

    2021

  21. [29]

    Design of optimal sparse feedback gains via the alternating direction method of multipliers.IEEE Transactions on Automatic Control, 58(9):2426– 2431, 2013

    Fu Lin, Makan Fardad, and Mihailo R Jovanović. Design of optimal sparse feedback gains via the alternating direction method of multipliers.IEEE Transactions on Automatic Control, 58(9):2426– 2431, 2013

    2013

  22. [30]

    A practical and optimal first-order method for large-scale convex quadratic programming

    Haihao Lu and Jinwen Yang. A practical and optimal first-order method for large-scale convex quadratic programming. arXiv preprint arXiv:2311.07710, 2023

    2023 arXiv

  23. [31]

    Optimal decen- tralized control for uncertain systems by symmetric gauss–seidel semi-proximal alm.IEEE Trans- actions on Automatic Control, 66(11):5554–5560, 2021

    Jun Ma, Zilong Cheng, Xiaoxue Zhang, Masayoshi Tomizuka, and Tong Heng Lee. Optimal decen- tralized control for uncertain systems by symmetric gauss–seidel semi-proximal alm.IEEE Trans- actions on Automatic Control, 66(11):5554–5560, 2021

    2021

  24. [32]

    Regularization for design.IEEE Transactions on Au- tomatic Control, 61(12):3991–4006, 2016

    Nikolai Matni and Venkat Chandrasekaran. Regularization for design.IEEE Transactions on Au- tomatic Control, 61(12):3991–4006, 2016

    2016

  25. [33]

    Sparsity-promoting optimal control of cyber–physical systems over shared communication networks.Automatica, 122:109217, 2020

    Nandini Negi and Aranya Chakrabortty. Sparsity-promoting optimal control of cyber–physical systems over shared communication networks.Automatica, 122:109217, 2020

    2020

  26. [34]

    Kl property of exponent1/2 of quadratic functions under nonnegative zero-norm constraints and applications.arXiv preprint arXiv:1904.03784, 2019

    Shaohua Pan, Yuqia Wu, and Shujun Bi. Kl property of exponent1/2 of quadratic functions under nonnegative zero-norm constraints and applications.arXiv preprint arXiv:1904.03784, 2019

    1904 arXiv

  27. [35]

    Structured policy iteration for linear quadratic regulator

    Youngsuk Park, Ryan Rossi, Zheng Wen, Gang Wu, and Handong Zhao. Structured policy iteration for linear quadratic regulator. InInternational Conference on Machine Learning, pages 7521–7531. PMLR, 2020

    2020

  28. [36]

    Inertial proximal alternating linearized minimization (ipalm) for nonconvex and nonsmooth problems.SIAM journal on imaging sciences, 9(4):1756–1787, 2016

    Thomas Pock and Shoham Sabach. Inertial proximal alternating linearized minimization (ipalm) for nonconvex and nonsmooth problems.SIAM journal on imaging sciences, 9(4):1756–1787, 2016

    2016

  29. [37]

    A unified convergence analysis of block successive minimization methods for nonsmooth optimization

    Meisam Razaviyayn, Mingyi Hong, and Zhi-Quan Luo. A unified convergence analysis of block successive minimization methods for nonsmooth optimization. SIAM Journal on Optimization, 23(2):1126–1153, 2013. 29

    2013

  30. [38]

    Springer Science & Business Media, 2009

    R Tyrrell Rockafellar and Roger J-B Wets.Variational analysis, volume 317. Springer Science & Business Media, 2009

    2009

  31. [39]

    A characterization of convex problems in decentralized control

    Michael Rotkowitz and Sanjay Lall. A characterization of convex problems in decentralized control. IEEE transactions on Automatic Control, 50(12):1984–1996, 2005

    1984

  32. [40]

    Robust control and linear parameter varying approaches: application to vehicle dynamics, volume 437

    Olivier Sename, Peter Gaspar, and József Bokor. Robust control and linear parameter varying approaches: application to vehicle dynamics, volume 437. Springer, 2013

    2013

  33. [41]

    Linear systems: optimal and robust control

    Alok Sinha. Linear systems: optimal and robust control. CRC press, 2007

    2007

  34. [42]

    Learning optimal controllers by policy gradient: Global optimality via convex parameterization

    Yue Sun and Maryam Fazel. Learning optimal controllers by policy gradient: Global optimality via convex parameterization. In 2021 60th IEEE Conference on Decision and Control (CDC), pages 4576–4581. IEEE, 2021

    2021

  35. [43]

    Structured output feedback control for linear quadratic regulator using policy gradient method.IEEE Transactions on Automatic Control, 69(1):363–370, 2023

    Shokichi Takakura and Kazuhiro Sato. Structured output feedback control for linear quadratic regulator using policy gradient method.IEEE Transactions on Automatic Control, 69(1):363–370, 2023

    2023

  36. [44]

    Policy optimization over submanifolds for linearly constrained feedback synthesis.arXiv preprint arXiv:2201.11157, 2022

    Shahriar Talebi and Mehran Mesbahi. Policy optimization over submanifolds for linearly constrained feedback synthesis.arXiv preprint arXiv:2201.11157, 2022

    2022 arXiv

  37. [45]

    Douglas–rachford splitting and admm for nonconvex optimization: Tight convergence results.SIAM Journal on Optimization, 30(1):149–181, 2020

    Andreas Themelis and Panagiotis Patrinos. Douglas–rachford splitting and admm for nonconvex optimization: Tight convergence results.SIAM Journal on Optimization, 30(1):149–181, 2020

    2020

  38. [46]

    On the complexity of decentralized decision making and detec- tion problems

    John Tsitsiklis and Michael Athans. On the complexity of decentralized decision making and detec- tion problems. IEEE Transactions on Automatic Control, 30(5):440–446, 1985

    1985

  39. [47]

    Kurdyka–łojasiewicz property of zero-norm composite functions

    Yuqia Wu, Shaohua Pan, and Shujun Bi. Kurdyka–łojasiewicz property of zero-norm composite functions. Journal of Optimization Theory and Applications, 188:94–112, 2021

    2021

  40. [48]

    An accelerated proximal alternating direc- tion method of multipliers for optimal decentralized control of uncertain systems.arXiv preprint arXiv:2304.11037, 2023

    Bo Yang, Xinyuan Zhao, Xudong Li, and Defeng Sun. An accelerated proximal alternating direc- tion method of multipliers for optimal decentralized control of uncertain systems.arXiv preprint arXiv:2304.11037, 2023

    2023 arXiv

  41. [49]

    Deducing kurdyka-lojasiewicz exponent via infprojection

    Peiran Yu, Guoyin Li, and Ting Kei Pong. Deducing kurdyka-lojasiewicz exponent via infprojection. arXiv preprint arXiv:1902.03635, 2019

    1902 arXiv

  42. [50]

    Distributed design for decentralized control using chordal decomposition and admm.IEEE Transactions on Control of Network Systems, 7(2):614–626, 2019

    Yang Zheng, Maryam Kamgarpour, Aivar Sootla, and Antonis Papachristodoulou. Distributed design for decentralized control using chordal decomposition and admm.IEEE Transactions on Control of Network Systems, 7(2):614–626, 2019. 30

    2019

Pith tools

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