Pith. sign in

REVIEW 2 major objections 4 minor 44 references

Cluster-based Distributed Augmented Lagrangian Algorithm for a Class of Constrained Convex Optimization Problems

T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read This paper claims that a cluster-based distributed continuous-time algorithm, built on the augmented Lagrangian method, converges to KKT solutions of a constrained convex optimization problem without requiring a shared penalty parameter.

desk verdict Worth a serious referee: the cluster-based algorithm and convergence proofs are solid, but the penalty-weight bound in Theorem 3.3 is false as written for p>1. read the letter →

arxiv 1908.06634 v4 pith:X7URT345 submitted 2019-08-19 cs.MA math.OC

classification cs.MAmath.OC MSC 90C2590C46
keywords distributedconvexoptimizationaugmentedLagrangiancontinuous-timealgorithmprimal-dualdynamicsconstrainedepsilon-exactpenaltycluster-basednetworksresourceallocation
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 claims that a networked, clustered set of agents can solve a constrained convex optimization problem in a fully distributed way using a continuous-time algorithm built on the augmented Lagrangian idea. The algorithm assigns, for each affine equality constraint, a small consensus network among only the agents that the constraint couples, and each agent may choose its own local penalty parameter. Under the paper's assumptions, the primal and dual states converge asymptotically to a KKT point when local costs are convex, and exponentially when costs are strongly convex with Lipschitz gradients. To handle box constraints, the paper replaces them with a smooth epsilon-exact penalty function and proves an explicit finite lower bound on the penalty weight, so the iterates land in an epsilon-neighborhood of the global minimum. A reader would care because the cluster-based structure cuts communication and computation, and the explicit penalty bound removes the usual guesswork in penalty methods.

What carries the argument

The load-bearing mechanism is the pair of distributed dual dynamics (7a)-(7b) running over each constraint-specific subgraph G_k, together with the primal update (7c). The dual states v_l^k act as local copies of the Lagrange multiplier for constraint k, and the consensus term beta_k sum a_lj(v_l^k - v_j^k) makes the copies agree; summing over the cluster recovers the centralized augmented-Lagrangian dynamics, so the augmentation term rho_i ||...||^2 can be formed locally with each agent's own rho_i. The inequality constraints are smoothed by the piecewise-quadratic penalty p_epsilon(y), and Theorem 3.3's multiplier bound, built from the minimum singular value of invertible submatrices of W, sizes the penalty weight gamma.

What would settle it

Solve the one-variable problem min x subject to x=1 and 0<=x<=1, which has nonempty feasible set {1} but no Slater point. Its KKT multipliers for the equality and upper-bound constraints are unbounded, so the bound in Theorem 3.3 cannot be formed. Checking this instance against the paper's claims tests whether Slater's condition is truly implied by Assumption 3.1 and whether the epsilon-neighborhood guarantee survives without it.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is a continuous-time distributed primal-dual algorithm (Eq. 7) that solves problem (1) without a shared augmented-Lagrangian penalty parameter. For each equality constraint k, only the agents in a connected subgraph G_k maintain local copies of the dual state and reach agreement among themselves; the sum of their dual dynamics reproduces the central multiplier dynamics. The paper proves that when every G_k is connected and the auxiliary states satisfy sum_{l in V_k} y_l^k(0)=0, the trajectories converge asymptotically to a KKT solution of the equality-constrained problem for convex costs, and exponentially to the unique solution for strongly convex costs with Lipschitz gradients. For inequality constraints, the paper invokes an epsilon-exact penalty function and supplies an explicit bound on Lagrange multipliers (under LICQ) that makes the penalty weight finite and computable, yielding convergence to an epsilon-neighborhood of the optimal value.

Load-bearing premise

The central premise is that the feasible region contains a strictly feasible point (Slater's condition); the paper states this follows from its assumptions, but nonemptyness of the feasible set does not guarantee an interior point, and the KKT characterization plus the finite penalty-weight bound depend on that strict feasibility.

Editorial extensions

If this is right

  • Each agent chooses its own augmented-Lagrangian penalty rho_i > 0, so no network-wide coordination of penalty parameters is needed for convergence.
  • For a cluster with n_i subagents, the communication and computation load for dual dynamics is |T_i| <= p per agent instead of n_i * p, because only one dual copy is kept per agent.
  • The convergence holds over time-varying connected topologies when costs are strongly convex with Lipschitz gradients and rho_i > 0.
  • With the epsilon-exact penalty and gamma chosen from the multiplier bound, the solution is epsilon-feasible and the achieved cost is within epsilon gamma N of the global optimum.
  • The explicit multiplier bound is presented as a standalone result useful for sizing exact and epsilon-exact penalty weights in other penalty-based formulations.

Reading between the lines

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

  • The multiplier bound (20) is expressed only through W, LICQ, and gradients evaluated on the box, so in principle it can be computed locally and reused by other distributed primal-dual or penalty-based solvers beyond the specific algorithm (7).
  • The zero-sum initialization sum_{l in V_k} y_l^k(0)=0 is the only coupling across agents at startup; if it is deliberately violated, the equilibrium shifts to a solution of the same problem with b_k replaced by b_k + sum_{l in V_k} y_l^k(0), which could be used for tracking demand changes without reinitializing.
  • Because the exponential convergence proof uses a quadratic Lyapunov function and a uniform negative-definite bound, a sampled-data or event-triggered version with sufficiently frequent communication should inherit exponential convergence; the authors list event-triggered implementation as future work.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper proposes a continuous-time distributed augmented Lagrangian algorithm for a class of constrained convex optimization problems: agents are clusters of subagents, affine equality constraints couple the decision variables, and affine box constraints may bound components. The main algorithmic contribution is a cluster-based primal-dual dynamics in which only agents involved in a given equality constraint communicate over a subgraph, and each agent uses its own local penalty parameter. The paper claims asymptotic convergence for convex local costs (Theorem 3.1), exponential convergence for strongly convex costs with Lipschitz gradients (Theorem 3.2), and, for the inequality-constrained problem, convergence to an epsilon-neighborhood of the global optimum through an epsilon-exact penalty function with an explicit lower bound on the penalty weight (Proposition 3.1 and Theorem 3.3). Two numerical examples illustrate the method.

Significance. If the results were fully valid, the paper would make a useful contribution: the cluster-based communication structure is a natural way to exploit sparsity in the affine equality constraints, the local penalty parameters avoid network-wide coordination, and the Lyapunov-based convergence proofs for the equality-only problem appear coherent and detailed. The claimed explicit bound on Lagrange multipliers, which is used to select the penalty weight, would also be of independent interest for exact and epsilon-exact penalty methods. However, the multiplier bound in Theorem 3.3 is false as stated, and the paper's claim that Assumption 3.1 implies Slater's condition is incorrect. These issues directly affect the epsilon-exact penalty guarantees advertised in the abstract and in the contributions, so the central claim is not established in its current form. The errors appear localized and a corrected bound with proper norm constants may repair the argument, which is why the result is reported as a major revision rather than a rejection.

major comments (2)
  1. [Section 3.1, paragraph after Assumption 3.1] The statement 'Given Assumption 3.1, the Slater condition is satisfied' is false. Assumption 3.1 only requires the feasible set X_fe to be nonempty. For problem (1), which has affine equality constraints and box constraints, nonemptiness does not imply strict feasibility: for example, minimize x subject to x = 1 and 0 <= x <= 1 has a nonempty feasible set but no point satisfying all inequality constraints strictly. This matters because Lemma 3.2's KKT characterization, and hence the multiplier boundedness used to size gamma in Proposition 3.1 and Theorem 3.3, relies on Slater's condition or another constraint qualification. Please either add Slater's condition (or a relative-interior assumption) to Assumption 3.1, or prove the required KKT and multiplier-boundedness results under a different explicit constraint qualification.
  2. [Section 3.1, Theorem 3.3 proof, Eqs. (20)-(23)] Theorem 3.3's bound is not established and, as stated, is false. In Eq. (23) the proof bounds ||nu*||_infinity by (1/sigma_min(W_se))||J||_infinity. The smallest singular value sigma_min(W_se) controls the Euclidean norm of W_se^{-1}, not its infinity norm; the inequality ||W_se^{-1}||_infinity <= 1/sigma_min(W_se) is not valid in general. Separately, the step bounding |[w_i]_l^T nu*| by wbar ||nu*||_infinity omits the dimension factor p: for a p-dimensional column a = [w_i]_l, the correct bound is |a^T nu*| <= ||a||_1 ||nu*||_infinity <= p wbar ||nu*||_infinity. These are not cosmetic constants. A concrete counterexample in the theorem's scope is N=1, x in R^4, W = [[1,1,0,0],[1,0,1,0],[1,0,0,1]], equality constraint Wx = 1_3, costs f_1 = 0 and f_l = -x_l for l=2,3,4, and bounds 0 <= x_1 <= 1, 0 <= x_l <= 10 for l=2,3,4. The unique minimizer is x* = (0,1,1,1), LICQ holds, KKT gives nu* = (1,1,1), and the active lower-bound multiplier for x_1 is mu_1 = 3. Meanwhile the right-hand side of (20) evaluates to about 2.93, so the claimed bound (20) is violated. Consequently the penalty-weight rule from Proposition 3.1 is not guaranteed for p > 1. Please correct the bound by using a valid infinity-norm estimate for W_se^{-1} (e.g., ||W_se^{-1}||_infinity or a factor of sqrt(p) through the Euclidean norm) and by replacing the entrywise wbar with the maximum column 1-norm (or equivalently multiplying by p). As written, Theorem 3.3 is false.
minor comments (4)
  1. [Section 3.1, Proposition 3.1] The proposition says 'Let (x*, nu*, {mu*}, {bar mu*}) be any solution of the KKT equations (5)', but equations (5) are the KKT conditions of the equality-only problem (4) and contain no inequality multipliers. The reference should be to the KKT conditions (13) of problem (1).
  2. [Appendix A, proof of Theorem 3.2] In the last paragraph of the proof, for the case rho_i = 0 for some i, the text concludes 'converges exponentially fast to ...', which contradicts the theorem statement, which says the convergence is asymptotic for that case. Please clarify which statement is intended and align the proof accordingly.
  3. [Section 3.1, Table 1] The caption says 'The values of actual mu_bound and the bound in (17)', but the bound displayed in the table and discussed in the surrounding text is Eq. (18) (or Eq. (20) in the final numbering). Please correct the cross-reference.
  4. [Throughout] There are several typographical issues, including 'equlibia' and 'Lyapounv' in Appendix A, and the author name 'Jovanovic /acute.ts1' in references [22] and [29]. These should be cleaned up in revision.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the convergence analyses and penalty-weight bound are derived from standard KKT/Lyapunov/LaSalle arguments and from explicit multiplier inequalities, with self-citations only contextual.

full rationale

The paper's central claims do not reduce to their inputs by construction. Theorem 3.1 and Theorem 3.2 are proved in Appendix A by a Lyapunov function and LaSalle/semistability arguments, with external textbook results [31] and [45] supplying the invariant-set and exponential-stability criteria; the equilibrium set (9) is related to the KKT equations through the invariant sum (8a), but convergence to that set is an analyzed property of the dynamics, not an assumed one. The penalty analysis is likewise non-circular: Proposition 3.1 imports a standard epsilon-exact penalty equivalence from [33], and the admissible gamma range is then sized by the closed-form Lagrange-multiplier bound of Theorem 3.3 (and Proposition 3.2 for the p=1 case). These bounds are derived from the KKT equalities (21a)-(21f), with only the local gradient norms and submatrix singular values as data; they are not obtained by fitting gamma to the numerical examples, and the examples are illustrative rather than load-bearing. Self-citations appear only as a preliminary conference version [1] and as comparative prior results [21], [30]; none of the paper's convergence or penalty-size claims is justified by asserting an unproved result of the same authors. The reviewer-identified issues (the Slater claim following Assumption 3.1 and the missing factor p in the infinity-norm bound in the proof of Theorem 3.3) are mathematical-validity concerns rather than circularity: an incorrect bound or an unjustified regularity condition does not make the derivation equivalent to its inputs. Accordingly, no circular step can be quoted with a specific equation-to-equation reduction, and the appropriate finding is no significant circularity.

Assumptions & free parameters 3 free parameters · 7 assumptions · 0 invented entities

The central convergence theorems rest on standard convex analysis, graph connectivity, and regularity assumptions. The inequality-constrained extension additionally imports Slater's condition and LICQ. The numerical scheme introduces user-chosen gains rho_i, beta_k and penalty parameters gamma and epsilon. No new physical entities are postulated.

free parameters (3)
  • Penalty weight gamma = gamma = 200 in second numerical example
    Hand-chosen in simulations; the theory requires gamma at least (1+sqrt(N)) mu_bound, but the authors do not verify the bound against computed multipliers in the p=2 example. The epsilon-neighborhood guarantee depends on this choice.
  • Smoothing parameter epsilon = epsilon = 0.001 or 0.01 in examples
    User-selected tolerance controlling constraint violation; it appears in Proposition 3.1 and in the set X^epsilon_fe.
  • Algorithm gains beta_k and rho_i = not specified in the numerical examples
    Any beta_k>0 and rho_i>0 is covered by Theorems 3.1 and 3.2, but the reported simulations do not state the selected values, so exact figure reproduction is not possible.
assumptions (7)
  • domain assumption Local cost functions f_i^l are convex and differentiable with locally Lipschitz gradients (Assumption 3.1).
    This is the main regularity condition used for existence, uniqueness, and convergence of the continuous-time dynamics.
  • domain assumption Constraint matrix W is full row rank and the feasible set X_fe is nonempty (Assumption 3.1).
    Full row rank ensures the dual feasibility condition in KKT, and nonemptiness gives a well-posed problem.
  • domain assumption Each subgraph G_k is connected for every equality constraint k (Theorems 3.1 and 3.2).
    Connectivity of the constraint-specific communication graph is required for consensus on each dual variable.
  • domain assumption Initialization satisfies sum_{l in V_k} y_l^k(0)=0 for every k (Theorem 3.1).
    This keeps the sum of helper variables at zero, which is used to recover the equality constraint at the equilibrium.
  • ad hoc to paper Slater's condition holds for problem (1).
    The text asserts before Lemma 3.2 that Slater follows from Assumption 3.1, but Assumption 3.1 only states nonempty feasibility. This is a load-bearing premise for the KKT characterization and multiplier boundedness.
  • domain assumption LICQ holds at the minimizers of (1) for Theorem 3.3.
    Linear independence of active constraint gradients is assumed to guarantee uniqueness of multipliers and to build the p x p invertible submatrix used in the bound.
  • standard math KKT conditions and the Pinar-Zenios epsilon-exact penalty relation in Proposition 3.1 are accepted from [39] and [33].
    These are cited external results used without proof in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cluster-based Distributed Augmented Lagrangian Algorithm for a Class of Constrained Convex Optimization Problems." pith.science (2026). https://pith.science/paper/X7URT345

@misc{pith2026190806634,
  author       = {Pith},
  title        = {Pith review of: Cluster-based Distributed Augmented Lagrangian Algorithm for a Class of Constrained Convex Optimization Problems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X7URT345}},
  note         = {Machine review of arXiv:1908.06634}
}
abstract

We propose a distributed solution for a constrained convex optimization problem over a network of clustered agents each consisted of a set of subagents. The communication range of the clustered agents is such that they can form a connected undirected graph topology. The total cost in this optimization problem is the sum of the local convex costs of the subagents of each cluster. We seek a minimizer of this cost subject to a set of affine equality constraints, and a set of affine inequality constraints specifying the bounds on the decision variables if such bounds exist. We design our distributed algorithm in a cluster-based framework which results in a significant reduction in communication and computation costs. Our proposed distributed solution is a novel continuous-time algorithm that is linked to the augmented Lagrangian approach. It converges asymptotically when the local cost functions are convex and exponentially when they are strongly convex and have Lipschitz gradients. Moreover, we use an $\epsilon$-exact penalty function to address the inequality constraints and derive an explicit lower bound on the penalty function weight to guarantee convergence to $\epsilon$-neighborhood of the global minimum value of the cost. A numerical example demonstrates our results.

Figures

Figures reproduced from arXiv: 1908.06634 by the authors.

Figure 1
Figure 1. A group of clustered agents (generators) with undirected connected graph topology aim to solve x ? = arg minx∈R12 P6 i=1 f i (x i ), subject to [1 1]x 1 + x 2 + [0.5 0.5 0.5]x 3 + [1 1 1]x 4 = 450, [0.5 0.5 0.5]x 3 + [1 1] x 5 + x 6 = 700, and x i l ≤ x i l ≤ ¯x i l , i ∈ Z 6 1 , l ∈ Z n i 1 in a distributed manner. Here, f i (x i ) = Pn i l=1 f i l (x i l ), where f i l (x i l ) = α i l x i l 2 + β i l x i l + γ i … view at source ↗
Figure 2
Figure 2. shows the time history of x i l ’s generated by im￾plementing the distributed optimization algorithm (7) (using f i p (x i ) as defined in (12) in place of f i (x i ) in (7c)) in comparison to the solution obtained using MATLAB’s constraint optimization solver ‘fmincon’. As expected the decision variable x i of each agent i ∈ {1, . . . , 6} converges closely to its corresponding minimizer, using  = 0.001 [PITH_FUL… view at source ↗
Figure 3
Figure 3. Constraint violation error while solving the optimization problem described in [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Schematic representation of the events, sensors and relay nodes in the second example. lier this configuration leads to extra computational and communication efforts. Here, instead, we form 4 cyber￾layers Gk, k ∈ {1, 2, 3, 4}, where V1 = {1, 2}, V2 = {2, 3}, V3 = {3, 4…
Figure 6
Figure 6. Figure 6: Trajectories of algorithm (6) when it is used to solve optimization problem (B.1) with ρ = 0 and ρ = 1. Note that the optimization problem (B.1) has infi￾nite number of minimizers that correspond to the minimum cost of f ? = 0. One of these minimizers is (x 1? , x2? ) …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 43 canonical work pages

  1. [1]

    An augmented lagrangian distributed algorithm for an in-network optimal resource allocation problem,

    S. S. Kia, “An augmented lagrangian distributed algorithm for an in-network optimal resource allocation problem,” in American Control Conference, (WA, USA), 2017

  2. [2]

    A. J. Wood, F. Wollenberg, and G. B. Sheble, Power Generation, Operation and Control . New York: John Wiley, 3rd ed., 2013

  3. [4]

    Simultaneous routing and resource allocation via dual decomposition,

    L. Xiao, M. Johansson, and S. P. Boyd, “Simultaneous routing and resource allocation via dual decomposition,” IEEE Transactions on Communications , vol. 52, no. 7, pp. 1136–1144, 2004

  4. [5]

    Distributed algorithms for maximum lifetime routing in wireless sensor networks,

    R. Madan and S. Lall, “Distributed algorithms for maximum lifetime routing in wireless sensor networks,”IEEE Transactions on Wireless Communications , vol. 5, no. 8, pp. 2185–2193, 2006

  5. [6]

    Convergence analysis of saddle point problems in time varying wireless systems – control theoretical approach,

    J. Chen and V. K. N. Lau, “Convergence analysis of saddle point problems in time varying wireless systems – control theoretical approach,” IEEE Transactions on Signal Processing, vol. 60, no. 1, pp. 443–452, 2012. 10 Table 2 The global cost value and the inequality constraint evaluation at x⋆ p obtained by using ϵ-exact penalty function method x1 −x2 − 5 ...

  6. [7]

    Network resource allocation for users with multiple connections: fairness and stability,

    A. Ferragut and F. Paganini, “Network resource allocation for users with multiple connections: fairness and stability,” IEEE/ACM Transactions on Networking , vol. 22, no. 2, pp. 349–362, 2014

  7. [8]

    Dual coupled diffusion for distributed optimization with affine constraints,

    S. A. Alghunaim, K. Yuan, and A. H. Sayed, “Dual coupled diffusion for distributed optimization with affine constraints,” in IEEE Conf. on Decision and Control , (FL, USA), 2018

  8. [9]

    ADMM- based distributed model predictive control: Primal and dual approaches,

    R. Rostami, G. Costantini, and D. G¨ orges, “ADMM- based distributed model predictive control: Primal and dual approaches,” in IEEE Conf. on Decision and Control , (Melbourne, Australia), 2017

Show all 44 references
  1. [10]

    Distributed optimization and statistical learning via the alternating direction method of multipliers,

    S. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, “Distributed optimization and statistical learning via the alternating direction method of multipliers,” Foundations and Trends in Machine Learning , vol. 3, pp. 1–122, 2010

  2. [11]

    Dual averaging for distributed optimization: Convergence analysis and network scaling,

    J. Duchi, A. Agarwal, and M. Wainwright, “Dual averaging for distributed optimization: Convergence analysis and network scaling,” IEEE Transactions on Automatic Control , vol. 57, no. 3, pp. 592–606, 2012

  3. [12]

    A control perspective for centralized and distributed convex optimization,

    J. Wang and N. Elia, “A control perspective for centralized and distributed convex optimization,” in IEEE Conf. on Decision and Control , (FL, USA), 2011

  4. [13]

    Distributed convex optimization via continuous-time coordination algorithms with discrete-time communication,

    S. S. Kia, J. Cort´ es, and S. Mart´ ınez, “Distributed convex optimization via continuous-time coordination algorithms with discrete-time communication,” Automatica, vol. 55, pp. 254–264, 2014

  5. [14]

    Newton-raphson consensus for distributed convex optimization,

    D. Varagnolo, F. Zanella, A. Cenedese, G. Pillonetto, and L. Schenato, “Newton-raphson consensus for distributed convex optimization,” IEEE Transactions on Automatic Control, vol. 61, no. 4, pp. 994 – 1009, 2015

  6. [15]

    Convergence analysis of the incremental cost consensus algorithm under different communication network topologies in a smart grid,

    Z. Zhang and M. Chow, “Convergence analysis of the incremental cost consensus algorithm under different communication network topologies in a smart grid,” IEEE Transactions on Power Systems , vol. 27, no. 4, pp. 1761– 1768, 2012

  7. [16]

    Distributed robust economic dispatch in power systems: A consensus + innovations approach,

    S. Kar and G. Hug, “Distributed robust economic dispatch in power systems: A consensus + innovations approach,” in Power & Energy Society General Meeting , (San Diego, CA), pp. 1–8, July 2012

  8. [17]

    Decentralized optimal dispatch of distributed energy resources,

    A. D. Dominguez-Garcia, S. T. Cady, and C. N. Hadjicostis, “Decentralized optimal dispatch of distributed energy resources,” in IEEE Conf. on Decision and Control , (Hawaii, USA), pp. 3688–3693, Dec. 2012

  9. [18]

    Optimal scaling of a gradient method for distributed resource allocation,

    L. Xiao and S. Boyd, “Optimal scaling of a gradient method for distributed resource allocation,” Journal of optimization theory and applications , vol. 129, no. 3, pp. 469–488, 2006

  10. [19]

    A consensus-based distributed augmented lagrangian method,

    Y. Zhang and M. M. Zavlanos, “A consensus-based distributed augmented lagrangian method,” in IEEE Conf. on Decision and Control , (CA, USA), 2018

  11. [20]

    Initialization-free distributed algorithms for optimal resource allocation with feasibility constraints and its application to economic dispatch of power systems,

    P. Yi, Y. Hong, and F. Liu, “Initialization-free distributed algorithms for optimal resource allocation with feasibility constraints and its application to economic dispatch of power systems,” Automatica, vol. 74, pp. 259–269, 2016

  12. [21]

    Distributed optimal in-network resource allocation algorithm design via a control theoretic approach,

    S. S. Kia, “Distributed optimal in-network resource allocation algorithm design via a control theoretic approach,” Systems and Control Letters , vol. 107, pp. 49–57, 2017

  13. [22]

    A primal-dual Laplacian gradient flow dynamics for distributed resource allocation problems,

    D. Ding and M. Jovanovic /acute.ts1, “A primal-dual Laplacian gradient flow dynamics for distributed resource allocation problems,” in American Control Conference , (WI, USA), 2018

  14. [23]

    Initialization-free distributed coordination for economic dispatch under varying loads and generator commitment,

    A. Cherukuri and J. Cort´ es, “Initialization-free distributed coordination for economic dispatch under varying loads and generator commitment,” Automatica, vol. 74, pp. 183–193, 2016

  15. [24]

    Distributed generator coordination for initialization and anytime optimization in economic dispatch,

    A. Cherukuri and J. Cort´ es, “Distributed generator coordination for initialization and anytime optimization in economic dispatch,” IEEE Transactions on Control of Network Systems , vol. 2, no. 3, pp. 226–237, 2015

  16. [25]

    Bertsekas and J

    D. Bertsekas and J. Tsitsiklis, Parallel and Distributed Computation: Numerical Methods . 1997

  17. [26]

    Linear convergence rate of a class of distributed augmented lagrangian algorithms,

    D. Jakovetic, J. Moura, and J. Xavier, “Linear convergence rate of a class of distributed augmented lagrangian algorithms,” IEEE Transactions on Automatic Control , vol. 60, no. 4, pp. 922–936, 2015

  18. [27]

    Distributed augmentation- regularization for robust online convex optimization,

    M.Vaquero and J.Cortes, “Distributed augmentation- regularization for robust online convex optimization,” IF AC- PapersOnLine, vol. 51, no. 23, pp. 230–235, 2018

  19. [28]

    K. J. Arrow, L. Hurwicz, and H. Uzawa, Studies in linear and nonlinear programming. 1958

  20. [29]

    An exponentially convergent primal-dual algorithm for nonsmooth composite minimization,

    D. Ding, B. Hu, N. Dhingra, and M. Jovanovic /acute.ts1, “An exponentially convergent primal-dual algorithm for nonsmooth composite minimization,” in IEEE Conf. on Decision and Control , (FL, USA), 2018

  21. [30]

    Distributed optimal resource allocation over networked systems and use of an epsilon-exact penalty function,

    S. S. Kia, “Distributed optimal resource allocation over networked systems and use of an epsilon-exact penalty function,” in IF AC Symposium on Large Scale Complex Systems, (CA, USA), 2016

  22. [31]

    Haddad and V

    W. Haddad and V. Chellaboina, Nonlinear Dynamical Systems and Control . Princeton University Press, 2008

  23. [32]

    Nondifferentiable optimization via approximation,

    D. P. Bertsekas, “Nondifferentiable optimization via approximation,” Mathematical Programing Study , vol. 3, pp. 1–25, 1975

  24. [33]

    On smoothing exact penalty functions for convex constrained optimization,

    M. C ¸ . Pinar and S. A. Zenios, “On smoothing exact penalty functions for convex constrained optimization,” IEEE Transactions on Communications , vol. 4, no. 3, pp. 1136–1144, 1994

  25. [34]

    Optimal power flow of radial networks and its variations: A sequential convex optimization approach,

    W. Wei, J. Wang, N. Li, and S. Mei, “Optimal power flow of radial networks and its variations: A sequential convex optimization approach,” IEEE Transactions on Smart Grid , vol. 8, no. 6, pp. 2974–2987, 2017

  26. [35]

    Resilient distributed optimal generation dispatch for lossy ac microgrids,

    M. Zholbaryssov, D. Fooladivanda, and A. D.Dom´ ınguez- Garc´ ıa, “Resilient distributed optimal generation dispatch for lossy ac microgrids,” Systems and Control Letters , vol. 123, pp. 47–54, 2019. 11

  27. [36]

    Computable numerical bounds for lagrange multipliers of stationary points of non-convex differentiable non-linear programs,

    O. Mangasarian, “Computable numerical bounds for lagrange multipliers of stationary points of non-convex differentiable non-linear programs,” Operations Research Letters , vol. 4, no. 2, pp. 1757–1780, 1985

  28. [37]

    Towards computational complexity certification for constrained MPC based on Lagrange relaxation and the fast gradient method,

    S. Richter, M. Morari, and C. Jones, “Towards computational complexity certification for constrained MPC based on Lagrange relaxation and the fast gradient method,” in IEEE Conf. on Decision and Control , (Orlando, Florida, USA), pp. 5223 – 5229, 2011

  29. [38]

    Bullo, J

    F. Bullo, J. Cort´ es, and S. Mart´ ınez,Distributed Control of Robotic Networks . Applied Mathematics Series, Princeton University Press, 2009

  30. [39]

    Boyd and L

    S. Boyd and L. Vandenberghe, Convex Optimization . Cambridge University Press, 2004

  31. [40]

    Bertsekas, Nonlinear Programming

    D. Bertsekas, Nonlinear Programming. 1999

  32. [41]

    The fritz john necessary optimality conditions in the presence of equality and inequality constraints,

    O. L. Mangasarian and S. Fromovitz, “The fritz john necessary optimality conditions in the presence of equality and inequality constraints,” Operations Research Letters , vol. 17, pp. 37–47, 1967

  33. [42]

    On LICQ and the uniqueness of Lagrange multipliers,

    G. Wachsmuth, “On LICQ and the uniqueness of Lagrange multipliers,” Operations Research Letters , vol. 41, no. 1, pp. 78–80, 2013

  34. [43]

    Distributed algorithm via continuously differentiable exact penalty method for network optimization,

    P. Srivastava and J. Cortes, “Distributed algorithm via continuously differentiable exact penalty method for network optimization,” in IEEE Conf. on Decision and Control , (FL, USA), 2018

  35. [44]

    http://motor.ece.iit.edu/data/JEAS_IEEE118.doc

    2004. http://motor.ece.iit.edu/data/JEAS_IEEE118.doc

  36. [45]

    H. K. Khalil, Nonlinear Control. Prentice Hall, 2002. Appendix A PROOF. [Proof of Theorem 3.1] Let ( {xi⋆}N i=1,ν⋆) satisfy the KKT equation (5) and y⋆ k = [{[wl]kxl⋆− ¯bl k}l∈Vk]. For convenience in analysis, we apply the change of variables qk = [ r⊤ k R⊤ k ] (yk−y⋆ k), pk =...

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.