Global Convergence of Control-Based Lagrangian Flows for Non-Convex Optimization

Daniele Astolfi; Diego Regruto; Francesco Ripa; Simone Pirrera; Sophie M. Fosson; Vito Cerone

arxiv: 2605.22486 · v1 · pith:YOAPPH57new · submitted 2026-05-21 · 🧮 math.OC · cs.SY· eess.SY

Global Convergence of Control-Based Lagrangian Flows for Non-Convex Optimization

Simone Pirrera , Francesco Ripa , Daniele Astolfi , Sophie M. Fosson , Vito Cerone , Diego Regruto This is my paper

Pith reviewed 2026-05-22 04:23 UTC · model grok-4.3

classification 🧮 math.OC cs.SYeess.SY

keywords global convergenceLagrangian dynamicsnon-convex optimizationequality constraintsexponential convergencecontrol-based methodsconstraint manifoldcontinuous-time flows

0 comments

The pith

Control-based Lagrangian flows achieve global exponential convergence for non-convex equality-constrained problems that are convex on the manifold.

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

The paper examines continuous-time dynamical systems for equality-constrained optimization that arise from proportional-integral and feedback linearization controllers. It establishes global exponential convergence to the optimum for non-convex objectives, provided the problem satisfies a suitable convexity property once restricted to the surface defined by the equality constraints. This result relies on the geometry induced by the constraints rather than on strong convexity of the objective or bounds on the variables. Readers interested in optimization would care because the approach removes common restrictive assumptions and still delivers a strong convergence guarantee for a wider class of problems.

Core claim

For equality-constrained optimization problems that satisfy a suitable convexity property when restricted to the constraint manifold, the continuous-time dynamics induced by proportional-integral and feedback linearization controllers converge globally and exponentially to the optimal solution.

What carries the argument

Control-based Lagrangian flows induced by proportional-integral and feedback linearization controllers, which exploit the geometric structure of the constraint manifold.

If this is right

The convergence guarantee holds globally without requiring strong convexity of the objective function over the whole space.
The same flows remain valid for non-convex problems as long as the restriction to the constraint surface is convex.
No boundedness assumptions on the variables or gradients are needed for the exponential rate.
The geometric structure of the equality constraints supplies the necessary stability properties.

Where Pith is reading between the lines

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

The continuous-time analysis suggests that discrete-time implementations of the same controllers could inherit similar guarantees under suitable discretization.
The manifold-convexity condition may apply to problems in robotics or physics where equality constraints define the feasible set naturally.
If the manifold convexity holds only locally, local exponential convergence might still be provable by restricting the initial conditions.

Load-bearing premise

The optimization problems satisfy a suitable convexity property when restricted to the constraint manifold.

What would settle it

A concrete equality-constrained problem that meets the manifold convexity condition but whose trajectories under the proposed flows fail to converge exponentially to the optimum.

Figures

Figures reproduced from arXiv: 2605.22486 by Daniele Astolfi, Diego Regruto, Francesco Ripa, Simone Pirrera, Sophie M. Fosson, Vito Cerone.

read the original abstract

This paper studies the flows of continuous-time dynamics for equality-constrained optimization based on control-theoretic Lagrangian methods. In particular, we consider dynamics induced by proportional-integral and feedback linearization controllers, which have been recently proposed as alternatives to primal-dual gradient methods. Unlike existing convergence results, which rely on strong convexity of the objective function or boundedness assumptions, we exploit the geometric structure induced by the constraints. Specifically, we show global exponential convergence for non-convex problems that satisfy a suitable convexity property when restricted to the constraint manifold.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper claims global exponential convergence for PI and feedback-linearization Lagrangian flows on non-convex objectives that are convex when restricted to the equality-constraint manifold, without strong convexity or boundedness.

read the letter

The key takeaway is that this work shows global exponential convergence for those two control-based Lagrangian flows on equality-constrained problems that are non-convex in the ambient space but satisfy a convexity property on the manifold. They avoid the usual strong-convexity or boundedness requirements by leaning on the geometry induced by the constraints instead. That is the main advance over prior results in this line of control-theoretic optimization methods. The paper does a clean job of setting up the dynamics for the proportional-integral and feedback-linearization controllers and then deriving the convergence claim from a Lyapunov function that combines primal error with the integral state. The geometric restriction lets them get a negative-definite derivative linear in the distance to the optimum, which is a reasonable way to relax the standard assumptions. Credit to them for targeting a broader class of problems while staying within the continuous-time flow framework. The soft spot is that the manifold convexity property is only sketched in the abstract, and the stress-test concern about possible implicit coercivity or compactness on non-compact manifolds is worth checking in the full derivation. If the Lyapunov analysis really closes without any hidden lower bound that prevents escape along the manifold, the result stands; if it leans on sublevel-set compactness that is not stated, then the claim is narrower than advertised. The citation pattern looks standard for this subfield and does not raise red flags. This paper is mainly for people working on continuous-time methods for constrained optimization or on control-based approaches to non-convex problems. A reader who already knows the recent Lagrangian-flow literature will get the most out of it. It deserves a serious referee to verify the exact hypotheses and the step where the exponential rate is obtained.

Referee Report

1 major / 2 minor

Summary. The manuscript develops continuous-time control-based Lagrangian flows for equality-constrained optimization, focusing on proportional-integral (PI) and feedback-linearization controllers. It claims to establish global exponential convergence for objectives that are non-convex in ambient Euclidean space but satisfy a suitable convexity property when restricted to the equality-constraint manifold, by exploiting the induced geometric structure rather than strong convexity of the objective or boundedness assumptions on the manifold.

Significance. If the central convergence result is rigorously established under the stated manifold-restricted convexity assumption, the work would meaningfully extend the theory of continuous-time optimization dynamics. It offers an alternative to primal-dual gradient flows by using control-theoretic constructions and manifold geometry to obtain exponential rates for a wider class of non-convex constrained problems, potentially informing the design of new optimization algorithms.

major comments (1)

[Theorem 1 (or the main convergence theorem in §4)] The skeptic's concern about implicit boundedness or coercivity on non-compact manifolds is a load-bearing issue for the global claim. The derivation of the exponential decay rate via the quadratic Lyapunov candidate (primal error plus integral control state) must be shown to produce a negative definite term linear in the distance to the optimum solely from the manifold convexity property, without any tacit compactness of sublevel sets or escape-preventing lower bounds. If this verification is absent or relies on unlisted geometric assumptions, the result becomes conditional in a way that contradicts the stated avoidance of boundedness conditions.

minor comments (2)

[Introduction / §2] The precise definition of the 'suitable convexity property' on the constraint manifold should be stated explicitly in the introduction or preliminaries, with a clear comparison to standard notions such as geodesic convexity.
[§3] Notation for the projection onto the tangent space of the manifold and the definition of the feedback-linearization controller could be made more self-contained to aid readers unfamiliar with the control-theoretic Lagrangian framework.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their thorough review and valuable comments on our work. The main concern regarding the global convergence claim is addressed point-by-point below.

read point-by-point responses

Referee: [Theorem 1 (or the main convergence theorem in §4)] The skeptic's concern about implicit boundedness or coercivity on non-compact manifolds is a load-bearing issue for the global claim. The derivation of the exponential decay rate via the quadratic Lyapunov candidate (primal error plus integral control state) must be shown to produce a negative definite term linear in the distance to the optimum solely from the manifold convexity property, without any tacit compactness of sublevel sets or escape-preventing lower bounds. If this verification is absent or relies on unlisted geometric assumptions, the result becomes conditional in a way that contradicts the stated avoidance of boundedness conditions.

Authors: We thank the referee for highlighting this important aspect of the proof. In Section 4, the proof of Theorem 1 proceeds by constructing the quadratic Lyapunov function V = 1/2 ||x - x^*||^2 + 1/2 ||z||^2, where x lies on the constraint manifold M and z is the integral state of the PI controller. The manifold-restricted convexity property (Definition 2) ensures that the inner product of the (projected) gradient with (x - x^*) satisfies <grad_M f(x), x - x^*> >= mu ||x - x^*||^2 for all x in M. Differentiating V along the closed-loop trajectories (Equations 12-14) yields dot V = (x - x^*)^T dot x + z^T dot z. The PI and feedback-linearization designs ensure that dot x aligns with the negative projected gradient while maintaining tangency to M, so that after substitution and cancellation of constraint terms, dot V <= -mu ||x - x^*||^2 - ||z||^2 <= -lambda V for lambda = min(mu, 1) > 0. This bound is obtained pointwise using only the manifold convexity assumption and holds globally on M. The exponential decay of V then implies both global convergence and boundedness of trajectories as a direct consequence (V(t) <= V(0) exp(-lambda t)), without any a priori compactness or coercivity assumptions. We agree that an explicit remark would improve clarity and will add one sentence after the proof of Theorem 1 emphasizing that boundedness follows from the Lyapunov analysis. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation relies on manifold geometry and control Lyapunov analysis

full rationale

The paper establishes global exponential convergence by constructing a Lyapunov function from the primal error and integral control state, then showing its derivative is negative definite using the convexity property restricted to the equality-constraint manifold. This step uses the induced Riemannian structure and feedback-linearization or PI controller design; it does not reduce to a fitted parameter, self-definition, or self-citation chain. The hypotheses explicitly avoid strong convexity and boundedness, and the exponential decay rate follows directly from the manifold-restricted convexity without importing uniqueness theorems or ansatzes from prior self-work. The derivation is therefore self-contained and independent of its target result.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption of a suitable convexity property restricted to the constraint manifold together with the geometric structure of the equality constraints; no free parameters or new entities are introduced in the abstract.

axioms (1)

domain assumption The optimization problem satisfies a suitable convexity property when restricted to the constraint manifold.
This replaces the strong convexity assumption used in prior work and is the key premise enabling the geometric argument for global exponential convergence.

pith-pipeline@v0.9.0 · 5631 in / 1259 out tokens · 63397 ms · 2026-05-22T04:23:22.061664+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

[1]

Supply chain design and optimization: Challenges and opportunities,

D. J. Garcia and F. You, “Supply chain design and optimization: Challenges and opportunities,”Com- put. Chem. Eng., vol. 81, pp. 153–170, 2015

work page 2015
[2]

Constrained optimization for machine learning: algorithms and applications,

J. Gallego-Posada, “Constrained optimization for machine learning: algorithms and applications,” 2024

work page 2024
[3]

A constrained optimization approach to 6 nonlinear system identification through simulation error minimization,

V. Cerone, S. M. Fosson, S. Pirrera, and D. Re- gruto, “A constrained optimization approach to 6 nonlinear system identification through simulation error minimization,”arXiv, no. 2509.01461, 2025

work page arXiv 2025
[4]

Semiglobal exponential stability of the discrete-time Arrow- Hurwicz-Uzawa primal-dual algorithm for con- strained optimization,

M. Bin, I. Notarnicola, and T. Parisini, “Semiglobal exponential stability of the discrete-time Arrow- Hurwicz-Uzawa primal-dual algorithm for con- strained optimization,”Math. Program., vol. 208, no. 1, pp. 629–660, 2024

work page 2024
[5]

An intercon- nected systems approach to convergence analysis of discrete-time primal-dual algorithms,

S. Kelly and J. W. Simpson-Porco, “An intercon- nected systems approach to convergence analysis of discrete-time primal-dual algorithms,” inAm. Con- trol Conf. (ACC). IEEE, 2024, pp. 662–668

work page 2024
[6]

A differential equation for modeling nesterov’s accelerated gradi- ent method: Theory and insights,

W. Su, S. Boyd, and E. J. Candes, “A differential equation for modeling nesterov’s accelerated gradi- ent method: Theory and insights,”J. Mach. Learn. Res., vol. 17, no. 153, pp. 1–43, 2016

work page 2016
[7]

Stability of primal-dual gradient dynamics and applications to network op- timization,

D. Feijer and F. Paganini, “Stability of primal-dual gradient dynamics and applications to network op- timization,”Automatica, vol. 46, no. 12, pp. 1974– 1981, 2010

work page 1974
[8]

Asymp- totic convergence of constrained primal–dual dy- namics,

A. Cherukuri, E. Mallada, and J. Cort´ es, “Asymp- totic convergence of constrained primal–dual dy- namics,”Syst. Control Lett., vol. 87, pp. 10–15, 2016

work page 2016
[9]

On the exponential stability of primal-dual gradient dynamics,

G. Qu and N. Li, “On the exponential stability of primal-dual gradient dynamics,”IEEE Control Syst. Lett., vol. 3, no. 1, pp. 43–48, 2019

work page 2019
[10]

Analysis and design of optimization algorithms via integral quadratic constraints,

L. Lessard, B. Recht, and A. Packard, “Analysis and design of optimization algorithms via integral quadratic constraints,”SIAM J. Optim., vol. 26, no. 1, pp. 57–95, 2016

work page 2016
[11]

A tutorial on convex design of optimization algorithms by in- tegral quadratic constraints,

C. W. Scherer and C. Ebenbauer, “A tutorial on convex design of optimization algorithms by in- tegral quadratic constraints,”Annu. Rev. Control Robot. Auton. Syst., vol. 9, 2025

work page 2025
[12]

A new framework for constrained optimiza- tion via feedback control of Lagrange multipliers,

V. Cerone, S. M. Fosson, S. Pirrera, and D. Re- gruto, “A new framework for constrained optimiza- tion via feedback control of Lagrange multipliers,” IEEE Trans. Autom. Control, vol. 70, no. 11, pp. 7141–7156, 2025

work page 2025
[13]

A feedback control approach to convex op- timization with inequality constraints,

——, “A feedback control approach to convex op- timization with inequality constraints,” inIEEE Conf. Decis. Control (CDC), 2024, pp. 2538–2543

work page 2024
[14]

Control-Barrier- Function-Based Design of Gradient Flows for Con- strained Nonlinear Programming,

A. Allibhoy and J. Cort´ es, “Control-Barrier- Function-Based Design of Gradient Flows for Con- strained Nonlinear Programming,”IEEE Trans. Autom. Control, vol. 69, no. 6, pp. 3499–3514, 2024

work page 2024
[15]

On the dif- ferentiability of projected trajectories and the ro- bust convergence of non-convex anti-windup gradi- ent flows,

A. Hauswirth, F. D¨ orfler, and A. Teel, “On the dif- ferentiability of projected trajectories and the ro- bust convergence of non-convex anti-windup gradi- ent flows,”IEEE Control Syst. Lett., vol. 4, no. 3, pp. 620–625, 2020

work page 2020
[16]

A control theoretical approach to online con- strained optimization,

U. Casti, N. Bastianello, R. Carli, and S. Zampieri, “A control theoretical approach to online con- strained optimization,”Automatica, vol. 176, p. 112107, 2025

work page 2025
[17]

Repetitive control tools for an original approach to convex optimiza- tion problems under affine periodic constraints,

D. Astolfi and C. M. Verrelli, “Repetitive control tools for an original approach to convex optimiza- tion problems under affine periodic constraints,” Syst. Control Lett., vol. 201, p. 106095, 2025

work page 2025
[18]

Zhang, A

R. Zhang, A. Raghunathan, J. Shamma, and N. Li, “Constrained Optimization From a Con- trol Perspective via Feedback Linearization,” 2026, arXiv:2503.12665

work page arXiv 2026
[19]

Anytime solvers for variational inequalities: The (recursive) safe mono- tone flows,

A. Allibhoy and J. Cort´ es, “Anytime solvers for variational inequalities: The (recursive) safe mono- tone flows,”Automatica, vol. 177, p. 112287, 2025

work page 2025
[20]

First-order methods for geodesically convex optimization,

H. Zhang and S. Sra, “First-order methods for geodesically convex optimization,” inConference on Learning Theory (COLT). PMLR, 2016, pp. 1617–1638

work page 2016
[21]

Boumal,An Introduction to Optimization on Smooth Manifolds

N. Boumal,An Introduction to Optimization on Smooth Manifolds. Cambridge University Press, 2023

work page 2023
[22]

Multiplier and gradient methods,

M. R. Hestenes, “Multiplier and gradient methods,” J. Optim. Theory Appl., vol. 4, no. 5, pp. 303–320, 1969

work page 1969
[23]

D. P. Bertsekas,Constrained Optimization and La- grange Multiplier Methods. Belmont, MA: Athena Scientific, 1996

work page 1996
[24]

Nocedal and S

J. Nocedal and S. J. Wright,Numerical optimiza- tion. Springer, 2006

work page 2006
[25]

A dynamical systems approach to constrained minimization,

J. Schropp and I. Singer, “A dynamical systems approach to constrained minimization,”Numer. Funct. Anal. Optim., vol. 21, pp. 537–551, 2000

work page 2000
[26]

D. G. Luenberger and Y. Ye,Linear and Nonlinear Programming, 4th ed. Springer Publishing Com- pany, Inc., 2015

work page 2015
[27]

Bertsekas,Nonlinear Programming, 2nd ed

D. Bertsekas,Nonlinear Programming, 2nd ed. Belmont, Massachusetts: Athena Scientific, 1999

work page 1999
[28]

Time-varying convex optimization: A contraction and equilibrium tracking approach,

A. Davydov, V. Centorrino, A. Gokhale, G. Russo, and F. Bullo, “Time-varying convex optimization: A contraction and equilibrium tracking approach,” IEEE Trans. Autom. Control, vol. 70, no. 11, pp. 7446–7460, 2025

work page 2025
[29]

On gradient descent ascent for nonconvex-concave minimax problems,

T. Lin, C. Jin, and M. Jordan, “On gradient descent ascent for nonconvex-concave minimax problems,” inInt. Conf. Mach. Learn. (ICML). PMLR, 2020, pp. 6083–6093

work page 2020
[30]

Isidori,Nonlinear Control Systems

A. Isidori,Nonlinear Control Systems. Springer London, 1995

work page 1995
[31]

Bullo,Contraction theory for dynamical systems

F. Bullo,Contraction theory for dynamical systems. Francesco Bullo, 2022

work page 2022
[32]

E. D. Sontag,Mathematical control theory: deter- ministic finite dimensional systems. Springer Sci- ence & Business Media, 2013, vol. 6

work page 2013
[33]

Kato and T

T. Kato and T. Kat˚ ao,Perturbation theory for lin- ear operators. Springer, 1966, vol. 132. 7 A Bound on the parameterk p In this section, we provide an explicit bound on the pa- rameterk p in terms of the problem’s data. Lemma 2(Explicit form of Lemma 1).Let m >0be such that∥∇ ⊤ x h(x)∇sh(x)∥ ≤ mfor allx∈R n. Under Assumptions 1, 4, and 6, the bound in(...

work page 1966

[1] [1]

Supply chain design and optimization: Challenges and opportunities,

D. J. Garcia and F. You, “Supply chain design and optimization: Challenges and opportunities,”Com- put. Chem. Eng., vol. 81, pp. 153–170, 2015

work page 2015

[2] [2]

Constrained optimization for machine learning: algorithms and applications,

J. Gallego-Posada, “Constrained optimization for machine learning: algorithms and applications,” 2024

work page 2024

[3] [3]

A constrained optimization approach to 6 nonlinear system identification through simulation error minimization,

V. Cerone, S. M. Fosson, S. Pirrera, and D. Re- gruto, “A constrained optimization approach to 6 nonlinear system identification through simulation error minimization,”arXiv, no. 2509.01461, 2025

work page arXiv 2025

[4] [4]

Semiglobal exponential stability of the discrete-time Arrow- Hurwicz-Uzawa primal-dual algorithm for con- strained optimization,

M. Bin, I. Notarnicola, and T. Parisini, “Semiglobal exponential stability of the discrete-time Arrow- Hurwicz-Uzawa primal-dual algorithm for con- strained optimization,”Math. Program., vol. 208, no. 1, pp. 629–660, 2024

work page 2024

[5] [5]

An intercon- nected systems approach to convergence analysis of discrete-time primal-dual algorithms,

S. Kelly and J. W. Simpson-Porco, “An intercon- nected systems approach to convergence analysis of discrete-time primal-dual algorithms,” inAm. Con- trol Conf. (ACC). IEEE, 2024, pp. 662–668

work page 2024

[6] [6]

A differential equation for modeling nesterov’s accelerated gradi- ent method: Theory and insights,

W. Su, S. Boyd, and E. J. Candes, “A differential equation for modeling nesterov’s accelerated gradi- ent method: Theory and insights,”J. Mach. Learn. Res., vol. 17, no. 153, pp. 1–43, 2016

work page 2016

[7] [7]

Stability of primal-dual gradient dynamics and applications to network op- timization,

D. Feijer and F. Paganini, “Stability of primal-dual gradient dynamics and applications to network op- timization,”Automatica, vol. 46, no. 12, pp. 1974– 1981, 2010

work page 1974

[8] [8]

Asymp- totic convergence of constrained primal–dual dy- namics,

A. Cherukuri, E. Mallada, and J. Cort´ es, “Asymp- totic convergence of constrained primal–dual dy- namics,”Syst. Control Lett., vol. 87, pp. 10–15, 2016

work page 2016

[9] [9]

On the exponential stability of primal-dual gradient dynamics,

G. Qu and N. Li, “On the exponential stability of primal-dual gradient dynamics,”IEEE Control Syst. Lett., vol. 3, no. 1, pp. 43–48, 2019

work page 2019

[10] [10]

Analysis and design of optimization algorithms via integral quadratic constraints,

L. Lessard, B. Recht, and A. Packard, “Analysis and design of optimization algorithms via integral quadratic constraints,”SIAM J. Optim., vol. 26, no. 1, pp. 57–95, 2016

work page 2016

[11] [11]

A tutorial on convex design of optimization algorithms by in- tegral quadratic constraints,

C. W. Scherer and C. Ebenbauer, “A tutorial on convex design of optimization algorithms by in- tegral quadratic constraints,”Annu. Rev. Control Robot. Auton. Syst., vol. 9, 2025

work page 2025

[12] [12]

A new framework for constrained optimiza- tion via feedback control of Lagrange multipliers,

V. Cerone, S. M. Fosson, S. Pirrera, and D. Re- gruto, “A new framework for constrained optimiza- tion via feedback control of Lagrange multipliers,” IEEE Trans. Autom. Control, vol. 70, no. 11, pp. 7141–7156, 2025

work page 2025

[13] [13]

A feedback control approach to convex op- timization with inequality constraints,

——, “A feedback control approach to convex op- timization with inequality constraints,” inIEEE Conf. Decis. Control (CDC), 2024, pp. 2538–2543

work page 2024

[14] [14]

Control-Barrier- Function-Based Design of Gradient Flows for Con- strained Nonlinear Programming,

A. Allibhoy and J. Cort´ es, “Control-Barrier- Function-Based Design of Gradient Flows for Con- strained Nonlinear Programming,”IEEE Trans. Autom. Control, vol. 69, no. 6, pp. 3499–3514, 2024

work page 2024

[15] [15]

On the dif- ferentiability of projected trajectories and the ro- bust convergence of non-convex anti-windup gradi- ent flows,

A. Hauswirth, F. D¨ orfler, and A. Teel, “On the dif- ferentiability of projected trajectories and the ro- bust convergence of non-convex anti-windup gradi- ent flows,”IEEE Control Syst. Lett., vol. 4, no. 3, pp. 620–625, 2020

work page 2020

[16] [16]

A control theoretical approach to online con- strained optimization,

U. Casti, N. Bastianello, R. Carli, and S. Zampieri, “A control theoretical approach to online con- strained optimization,”Automatica, vol. 176, p. 112107, 2025

work page 2025

[17] [17]

Repetitive control tools for an original approach to convex optimiza- tion problems under affine periodic constraints,

D. Astolfi and C. M. Verrelli, “Repetitive control tools for an original approach to convex optimiza- tion problems under affine periodic constraints,” Syst. Control Lett., vol. 201, p. 106095, 2025

work page 2025

[18] [18]

Zhang, A

R. Zhang, A. Raghunathan, J. Shamma, and N. Li, “Constrained Optimization From a Con- trol Perspective via Feedback Linearization,” 2026, arXiv:2503.12665

work page arXiv 2026

[19] [19]

Anytime solvers for variational inequalities: The (recursive) safe mono- tone flows,

A. Allibhoy and J. Cort´ es, “Anytime solvers for variational inequalities: The (recursive) safe mono- tone flows,”Automatica, vol. 177, p. 112287, 2025

work page 2025

[20] [20]

First-order methods for geodesically convex optimization,

H. Zhang and S. Sra, “First-order methods for geodesically convex optimization,” inConference on Learning Theory (COLT). PMLR, 2016, pp. 1617–1638

work page 2016

[21] [21]

Boumal,An Introduction to Optimization on Smooth Manifolds

N. Boumal,An Introduction to Optimization on Smooth Manifolds. Cambridge University Press, 2023

work page 2023

[22] [22]

Multiplier and gradient methods,

M. R. Hestenes, “Multiplier and gradient methods,” J. Optim. Theory Appl., vol. 4, no. 5, pp. 303–320, 1969

work page 1969

[23] [23]

D. P. Bertsekas,Constrained Optimization and La- grange Multiplier Methods. Belmont, MA: Athena Scientific, 1996

work page 1996

[24] [24]

Nocedal and S

J. Nocedal and S. J. Wright,Numerical optimiza- tion. Springer, 2006

work page 2006

[25] [25]

A dynamical systems approach to constrained minimization,

J. Schropp and I. Singer, “A dynamical systems approach to constrained minimization,”Numer. Funct. Anal. Optim., vol. 21, pp. 537–551, 2000

work page 2000

[26] [26]

D. G. Luenberger and Y. Ye,Linear and Nonlinear Programming, 4th ed. Springer Publishing Com- pany, Inc., 2015

work page 2015

[27] [27]

Bertsekas,Nonlinear Programming, 2nd ed

D. Bertsekas,Nonlinear Programming, 2nd ed. Belmont, Massachusetts: Athena Scientific, 1999

work page 1999

[28] [28]

Time-varying convex optimization: A contraction and equilibrium tracking approach,

A. Davydov, V. Centorrino, A. Gokhale, G. Russo, and F. Bullo, “Time-varying convex optimization: A contraction and equilibrium tracking approach,” IEEE Trans. Autom. Control, vol. 70, no. 11, pp. 7446–7460, 2025

work page 2025

[29] [29]

On gradient descent ascent for nonconvex-concave minimax problems,

T. Lin, C. Jin, and M. Jordan, “On gradient descent ascent for nonconvex-concave minimax problems,” inInt. Conf. Mach. Learn. (ICML). PMLR, 2020, pp. 6083–6093

work page 2020

[30] [30]

Isidori,Nonlinear Control Systems

A. Isidori,Nonlinear Control Systems. Springer London, 1995

work page 1995

[31] [31]

Bullo,Contraction theory for dynamical systems

F. Bullo,Contraction theory for dynamical systems. Francesco Bullo, 2022

work page 2022

[32] [32]

E. D. Sontag,Mathematical control theory: deter- ministic finite dimensional systems. Springer Sci- ence & Business Media, 2013, vol. 6

work page 2013

[33] [33]

Kato and T

T. Kato and T. Kat˚ ao,Perturbation theory for lin- ear operators. Springer, 1966, vol. 132. 7 A Bound on the parameterk p In this section, we provide an explicit bound on the pa- rameterk p in terms of the problem’s data. Lemma 2(Explicit form of Lemma 1).Let m >0be such that∥∇ ⊤ x h(x)∇sh(x)∥ ≤ mfor allx∈R n. Under Assumptions 1, 4, and 6, the bound in(...

work page 1966