Switched Control of Complex-Valued Kuramoto Networks

arXiv:2604.07249 · eess.SY, cs.SY · Submitted 2026-04-08 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: We're looking at the paper: "Switched Control of Complex-Valued Kuramoto Networks".

Dev: Here's the gist of the paper. Synchronization in networks of coupled oscillators is classically studied via the Kuramoto model, whose intrinsic nonlinearity limits analytical tractability and complicates control design.

Rosa: Complex-valued extensions circumvent this by embedding phase dynamics into a higher-dimensional linear state space, where regulating complex-state moduli to a common value recovers Kuramoto phase behavior.

Dev: The main contributions of this work are: 1) A switched feedforward law that achieves exact, continuous phase matching with the real-valued Kuramoto dynamics for all t ≥ 0 under unit initial magnitudes.

Rosa: 2) A feedforward + SMC law that upgrades the asymptotic convergence of

15: to finite-time convergence, with an explicit and tunable reaching-time bound.

Paper summary: Rosa: We’re talking about "Switched Control of Complex-Valued Kuramoto Networks" here. Basically, they start by saying that the standard Kuramoto model is hard to control because it's nonlinear, but they solve this by taking those phase dynamics and embedding them into a higher dimensional linear state space >

Dev: That higher dimensional space lets you regulate the magnitude of these complex states to a common value which recovers the original Kuramoto phase behavior, which is key for synchronization >

Taro: The paper claims they propose two switched control designs: first, a switched feedforward law that achieves exact continuous phase correspondence for all time under unit initial magnitudes >

Rosa: Right. And then they have this feedforward plus sliding mode law that upgrades the asymptotic convergence from the state feedback method to finite time convergence, with a tunable reaching-time bound >

Dev: And they also introduce a third thing, a non autonomous complex valued MIMO sliding mode controller that enforces phase locking at a prescribed frequency in finite time, independent of natural frequencies and coupling strengths >

Taro: The paper also points out the limitations of the existing methods: state feedback only gives asymptotic convergence with constrained gain selection, and hybrid reset methods only match phases at those specific reset instants >

Rosa: So, they show that their switched feedforward law eliminates that drift-induced mismatch you get from those reset-based schemes because it keeps the phase perfectly matched for all time >

Dev: That’s a big difference. And the combined feedforward plus sliding mode law removes the need for spectral gain tuning entirely, which is another major win for real systems engineers >

Taro: And that final controller, enforcing synchronization at a prescribed frequency, is what really shows the advantage of using this complex valued formulation when dealing with networks that are actually messy >

Rosa: So to sum up, it’s about moving from methods that get asymptotic results or only work in discrete time to these new switched and sliding mode designs that offer exact matches or finite time convergence depending on what you need >

Dev: And the paper's overall goal is to provide a unified control-theoretic framework for this complex domain, showing how these different strategies build upon each other >

Conclusion: Rosa: So, thinking about "Switched Control of Complex-Valued Kuramoto Networks" and the authors Lorenzo Giordano, Josep M. Olm, and Mario di Bernardo. It sounds like they're providing a unified set of tools for managing these complex oscillator networks >

Dev: They’ve introduced three specific switched control strategies: exact continuous phase correspondence, finite time convergence without tuning, and frequency locking in finite time >

Taro: The implication is that for anyone working with heterogeneous networks, even the ones where simple models fail to synchronize anymore, they now have a robust way to achieve synchronization at a desired frequency >

Rosa: It means these tools are useful because they don't depend on tuning parameters as much, which simplifies the design process significantly for engineers trying to get these systems running reliably >

Dev: And this whole complex valued extension is a smart way to handle the nonlinearity of Kuramoto dynamics without getting bogged down in it analytically >

Taro: So, we’re looking at a set of methods that offer better performance than what was available before for synchronization problems in complex network settings >

Scuola Superiore Meridionale · Department of Mathematics & Institute of Industrial and Control Engineering, Universitat Politècnica de Catalunya · Department of Electrical Engineering and Information Technologies, Università degli Studi di Napoli Federico II

eess.SY, cs.SY

Submitted: 2026-04-08

Updated: 2026-11-16

Journal ref: IEEE Control Systems Letters, vol. 10, pp. 2029-2034, 2026

DOI: 10.1109/LCSYS.2026.3713251

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 78/100

The gist: Synchronization in networks of coupled oscillators is classically studied via the Kuramoto model, whose intrinsic nonlinearity limits analytical tractability and complicates control design.

Key concepts

Kuramoto Model
The classical model describes how coupled oscillators synchronize. It involves phases evolving based on their natural frequencies and the coupling strength between them. Synchronization is measured by how close these phases get to each other over time.
Complex-Valued Extension
To make the Kuramoto dynamics easier to control, the paper embeds phase information into a higher-dimensional linear system using complex numbers. This allows researchers to control the magnitude and argument of complex states separately, which simplifies achieving synchronization.
Sliding Mode Control (SMC)
SMC is a robust control technique used here to enforce synchronization at specific frequencies. It defines a switching manifold where the error dynamics are forced onto this surface, ensuring that the system reaches and stays on the desired state in finite time, regardless of initial conditions.

Terminology

Summary

Synchronization in networks of coupled oscillators is classically studied via the Kuramoto model, whose intrinsic nonlinearity limits analytical tractability and complicates control design. p. 0

Where the paper says it

1 Complex-Valued Kuramoto Networks: A Unified Control-Theoretic Framework Lorenzo Giordano, Josep M. Olm, and Mario di Bernardo Abstract—Synchronization in networks of coupled oscillators is classically studied via the Kuramoto model, whose intrinsic nonlinearity limits analytical tractability and complicates control design.

Complex-valued extensions circumvent this by embedding phase dynamics into a higher-dimensional linear state space, where regulating complex-state moduli to a common value recovers Kuramoto phase behavior. p. 0
Where the paper says it

Olm, and Mario di Bernardo Abstract—Synchronization in networks of coupled oscillators is classically studied via the Kuramoto model, whose intrinsic nonlinearity limits analytical tractability and complicates control design. Complex-valued extensions circumvent this by embedding phase dynamics into a higher-dimensional linear state space, where regulating complex-state moduli to a common value recovers Kuramoto phase behavior.

The main contributions of this work are: 1) A switched feedforward law that achieves exact, continuous phase matching with the real-valued Kuramoto dynamics for all t ≥ 0 under unit initial magnitudes. p. 0
Where the paper says it

Second, we seek to enforce phase locking at a prescribed frequency, regardless of natural frequencies and coupling strengths, achieving robust and controlled synchronization via complex-valued slidingmode control (SMC) [18]. Hence, the main contributions of this work are: 1) A switched feedforward law that achieves exact, continuous phase matching with the real-valued Kuramoto dynamics for all t ≥ 0 under unit initial magnitudes, overcoming the discrete-instant limitation of [9].

2) A feedforward + SMC law that upgrades the asymptotic convergence of [15] to finite-time convergence, with an explicit and tunable reaching-time bound. p. 0
Where the paper says it

Hence, the main contributions of this work are: 1) A switched feedforward law that achieves exact, continuous phase matching with the real-valued Kuramoto dynamics for all t ≥ 0 under unit initial magnitudes, overcoming the discrete-instant limitation of [9]. 2) A feedforward + SMC law that upgrades the asymptotic convergence of [15] to finite-time convergence, with an explicit and tunable reaching-time bound.

p. 0
Where the paper says it

Hence, the main contributions of this work are: 1) A switched feedforward law that achieves exact, continuous phase matching with the real-valued Kuramoto dynamics for all t ≥ 0 under unit initial magnitudes, overcoming the discrete-instant limitation of [9]. 2) A feedforward + SMC law that upgrades the asymptotic convergence of [15] to finite-time convergence, with an explicit and tunable reaching-time bound.

3) An extension of the complex-valued MIMO SMC framework of [19] to non-autonomous systems, applied to arXiv:2604. p. 0
Where the paper says it

No spectral gain tuning is required. 3) An extension of the complex-valued MIMO SMC framework of [19] to non-autonomous systems, applied to arXiv:2604.

07249v1 [eess. p. 0
Where the paper says it

3) An extension of the complex-valued MIMO SMC framework of [19] to non-autonomous systems, applied to arXiv:2604. 07249v1 [eess.

SY] 8 Apr 2026. p. 0
Where the paper says it

07249v1 [eess. SY] 8 Apr 2026

The Mathematical Foundation of the Model

The classical Kuramoto model describes the dynamics of coupled oscillators where phases evolve according to a nonlinear equation. By subtracting i cos(θj−θk) from the coupling term and setting x = e iθ, the nonlinear phase dynamics are embedded into a linear complex state space given by x˙ = (i diag(ω) + σA) x. p. 1

Where the paper says it

Complex-valued extension and control formulation Subtracting i cos(θj−θk) from the coupling in (1), performing straightforward algebraic manipulation and setting x = e iθ, yields the linear system [8]: x˙ = (i diag(ω) + σA) x. (2) Splitting (2) into magnitude and argument dynamics gives d dt |xk| = σ X N j=1 akj |xj | cosϕxj − ϕxk , (3a) d dt ϕxk = ωk + σ X N j=1 akj |xj | |xk| sinϕxj − ϕxk .

This linear system is further augmented by a control input u in (4), yielding x˙ = (i diag(ω) + σA) x + u. p. 1
Where the paper says it

This can be achieved by introducing a control input u ∈ C N in (2): x˙ = (i diag(ω) + σA) x + u. STATE-OF-THE-ART: A CONTROL PERSPECTIVE This section reviews the existing linear reformulations of the Kuramoto model from a control-theoretic standpoint, making explicit the shared objective that motivates the present work.

Limitations of Existing Control Strategies

Existing approaches to address the problem correspond, within a unified control framework, to statefeedback and hybrid reset-based strategies, each with performance constraints. State-feedback [15] ensures only asymptotic convergence and requires spectral gain tuning. p. 0

Where the paper says it

From a control viewpoint, [15] is a continuous state-feedback strategy, while [9] is a hybrid reset-based law. Both exhibit limitations: the former ensures only asymptotic convergence and requires spectral gain tuning, while the latter achieves exact matching only at reset instants, with magnitude drift accumulating between them.

Conversely, the hybrid reset method [9] achieves exact matching only at reset instants, with magnitude drift accumulating between them. p. 0
Where the paper says it

From a control viewpoint, [15] is a continuous state-feedback strategy, while [9] is a hybrid reset-based law. Both exhibit limitations: the former ensures only asymptotic convergence and requires spectral gain tuning, while the latter achieves exact matching only at reset instants, with magnitude drift accumulating between them.

Both exhibit fundamental performance limitations. p. 2
Where the paper says it

Unified perspective Although the approaches in [15] and [9] were developed independently, they share a common control objective: enforce a uniform modulus on the complex states of (4) so that their arguments reproduce the phase dynamics of (1). Both exhibit fundamental performance limitations: asymptotic convergence with constrained gain selection in the former, and discretetime matching with inter-reset magnitude drift in the latter.

Proposed Control Mechanisms

The proposed switched feedforward controller (Theorem 1) ensures exact phase correspondence for all time under unit initial magnitudes. The feedforward + SMC law (Theorem 2) upgrades the asymptotic convergence to finite-time convergence with a tunable reaching-time bound, requiring no spectral gain tuning. p. 0

Where the paper says it

2) A feedforward + SMC law that upgrades the asymptotic convergence of [15] to finite-time convergence, with an explicit and tunable reaching-time bound. No spectral gain tuning is required.

Additionally, the prescribed-frequency SMC strategy (Theorem 4) enforces phase locking at a desired frequency in finite time, independent of natural frequencies and coupling strengths. p. 5
Where the paper says it

The combined feedforward-SMC law (Theorem 2) relaxes the unit-magnitude initial condition and ensures finite-time convergence with an explicit, tunable reaching-time bound, improving upon the asymptotic result of [15] without spectral gain tuning. Finally, the prescribed-frequency SMC law (Theorem 4), based on a novel non-autonomous extension of complex-valued MIMO sliding-mode control, enforces phase locking at a desired frequency in finite time, independent of natural frequencies and coupling strengths.

Validation and Robustness

Numerical validation on an Erdös-Rényi network confirms the performance of the controllers. The switched feedforward controller achieves exact continuous-time phase correspondence with the real-valued model for all t ≥ 0. p. 0

Where the paper says it

Hence, the main contributions of this work are: 1) A switched feedforward law that achieves exact, continuous phase matching with the real-valued Kuramoto dynamics for all t ≥ 0 under unit initial magnitudes, overcoming the discrete-instant limitation of [9]. 2) A feedforward + SMC law that upgrades the asymptotic convergence of [15] to finite-time convergence, with an explicit and tunable reaching-time bound.

The complex SMC successfully enforces phase locking across the entire network at the prescribed frequency even when the uncontrolled real-valued model fails to synchronize. p. 4
Where the paper says it

Under these heterogeneous conditions, the real-valued Kuramoto model fails to synchronize, yet the SMC successfully enforces phase locking across the entire network at the prescribed frequency. This result highlights the fundamental advantage of the complex-valued control formulation: by operating on the linear embedding rather than the nonlinear phase equations, the controller can impose synchronization even in regimes where the uncontrolled realvalued model does not synchronize spontaneously.

This demonstrates the fundamental advantage of the complex-valued control formulation. p. 4
Where the paper says it

Under these heterogeneous conditions, the real-valued Kuramoto model fails to synchronize, yet the SMC successfully enforces phase locking across the entire network at the prescribed frequency. This result highlights the fundamental advantage of the complex-valued control formulation: by operating on the linear embedding rather than the nonlinear phase equations, the controller can impose synchronization even in regimes where the uncontrolled realvalued model does not synchronize spontaneously.

Conclusions and Future Directions

This work introduced a unified control-theoretic framework for complex-valued Kuramoto oscillator networks and proposed three switched control strategies that overcome key limitations of existing methods. The switched feedforward law eliminates the drift-induced mismatch inherent to reset-based schemes. p. 5

Where the paper says it

By interpreting the statefeedback approach of [15] and the hybrid reset method of [9] as mechanisms to enforce a common modulus, we identified a shared control objective and used it to guide improved designs. Within this framework, the switched feedforward law (Theorem 1) achieves exact continuous-time phase correspondence with classical Kuramoto dynamics for all t ≥ 0, eliminating the drift-induced mismatch inherent to reset-based schemes.

The combined feedforward-SMC law ensures finite-time convergence without spectral gain tuning. p. 5
Where the paper says it

Within this framework, the switched feedforward law (Theorem 1) achieves exact continuous-time phase correspondence with classical Kuramoto dynamics for all t ≥ 0, eliminating the drift-induced mismatch inherent to reset-based schemes. The combined feedforward-SMC law (Theorem 2) relaxes the unit-magnitude initial condition and ensures finite-time convergence with an explicit, tunable reaching-time bound, improving upon the asymptotic result of [15] without spectral gain tuning.

The prescribed-frequency SMC law enables synchronization of heterogeneous networks where the classical Kuramoto model fails. p. 5
Where the paper says it

Finally, the prescribed-frequency SMC law (Theorem 4), based on a novel non-autonomous extension of complex-valued MIMO sliding-mode control, enforces phase locking at a desired frequency in finite time, independent of natural frequencies and coupling strengths. This result is particularly significant, enabling synchronization of heterogeneous networks where the classical Kuramoto model fails.

Future work will explore output-feedback designs and applications to power systems and neuromorphic computing. p. 5
Where the paper says it

This result is particularly significant, enabling synchronization of heterogeneous networks where the classical Kuramoto model fails. Future work will explore output-feedback designs, extensions to directed and time-varying networks, tighter gain conditions exploiting sparse graph spectra, and applications to synchronization problems in power systems and neuromorphic computing.

--- Page 1 ---

Complex-Valued Kuramoto Networks: A Unified Control-Theoretic Framework

Lorenzo Giordano, Josep M. Olm, and Mario di Bernardo

Abstract—Synchronization in networks of coupled oscillators is classically studied via the Kuramoto model, whose intrinsic nonlinearity limits analytical tractability and complicates control design Complex-valued extensions circumvent this by embedding phase dynamics into a higher-dimensional linear state space, where regulating complex-state moduli to a common value recovers Kuramoto phase behavior. Existing approaches to address this problem correspond, within a unified control framework, to statefeedback and hybrid reset-based strategies, each with performance constraints. p. 0

Where the paper says it

Complex-valued extensions circumvent this by embedding phase dynamics into a higher-dimensional linear state space, where regulating complex-state moduli to a common value recovers Kuramoto phase behavior. Existing approaches to address this problem correspond, within a unified control framework, to statefeedback and hybrid reset-based strategies, each with performance constraints.

p. 0
Where the paper says it

Complex-valued extensions circumvent this by embedding phase dynamics into a higher-dimensional linear state space, where regulating complex-state moduli to a common value recovers Kuramoto phase behavior. Existing approaches to address this problem correspond, within a unified control framework, to statefeedback and hybrid reset-based strategies, each with performance constraints.

We propose two switched control designs that overcome these limitations: a switched feedforward law ensuring exact phase correspondence at all times, and a feedforward plus sliding-mode law achieving finite-time convergence without spectral gain tuning. p. 0
Where the paper says it

Existing approaches to address this problem correspond, within a unified control framework, to statefeedback and hybrid reset-based strategies, each with performance constraints. We propose two switched control designs that overcome these limitations: a switched feedforward law ensuring exact phase correspondence at all times, and a feedforward plus sliding-mode law achieving finite-time convergence without spectral gain tuning.

p. 0
Where the paper says it

Existing approaches to address this problem correspond, within a unified control framework, to statefeedback and hybrid reset-based strategies, each with performance constraints. We propose two switched control designs that overcome these limitations: a switched feedforward law ensuring exact phase correspondence at all times, and a feedforward plus sliding-mode law achieving finite-time convergence without spectral gain tuning.

Additionally, we present a non-autonomous complex-valued MIMO sliding-mode controller that enforces phase locking at a prescribed frequency in finite time, independent of natural frequencies and coupling strengths. p. 0
Where the paper says it

We propose two switched control designs that overcome these limitations: a switched feedforward law ensuring exact phase correspondence at all times, and a feedforward plus sliding-mode law achieving finite-time convergence without spectral gain tuning. Additionally, we present a non-autonomous complex-valued MIMO sliding-mode controller that enforces phase locking at a prescribed frequency in finite time, independent of natural frequencies and coupling strengths.

p. 0
Where the paper says it

We propose two switched control designs that overcome these limitations: a switched feedforward law ensuring exact phase correspondence at all times, and a feedforward plus sliding-mode law achieving finite-time convergence without spectral gain tuning. Additionally, we present a non-autonomous complex-valued MIMO sliding-mode controller that enforces phase locking at a prescribed frequency in finite time, independent of natural frequencies and coupling strengths.

Simulations confirm improved transient response, steady-state accuracy, and robustness, including synchronization of heterogeneous networks where the classical real-valued Kuramoto model fails. p. 0
Where the paper says it

Additionally, we present a non-autonomous complex-valued MIMO sliding-mode controller that enforces phase locking at a prescribed frequency in finite time, independent of natural frequencies and coupling strengths. Simulations confirm improved transient response, steady-state accuracy, and robustness, including synchronization of heterogeneous networks where the classical real-valued Kuramoto model fails.

Index Terms—Kuramoto oscillators, complex-valued systems, sliding-mode control, feedforward control, synchronization, switched systems, network control. p. 0

Where the paper says it

Simulations confirm improved transient response, steady-state accuracy, and robustness, including synchronization of heterogeneous networks where the classical real-valued Kuramoto model fails. Index Terms—Kuramoto oscillators, complex-valued systems, sliding-mode control, feedforward control, synchronization, switched systems, network control.

p. 0
Where the paper says it

Simulations confirm improved transient response, steady-state accuracy, and robustness, including synchronization of heterogeneous networks where the classical real-valued Kuramoto model fails. Index Terms—Kuramoto oscillators, complex-valued systems, sliding-mode control, feedforward control, synchronization, switched systems, network control.

--- Page 2 ---

II. NOTATION

For z = z R +izI ∈ C, z R = Re(z) and z I = Im(z) are the real and imaginary parts of z respectively. In turn, z∗ is the conjugate of z, z = √z∗z its magnitude, and ϕz its phase. p. 1

Where the paper says it

NOTATION For z = z R +izI ∈ C, z R = Re(z) and z I = Im(z) are the real and imaginary parts of z, respectively. In turn, z ∗ is the conjugate of z, |z| = √ z ∗z its magnitude, and ϕz its phase.

For w ∈ C n, Re(w) and Im(w) are vectors that collect real and imaginary parts of w respectively. p. 1
Where the paper says it

In turn, z ∗ is the conjugate of z, |z| = √ z ∗z its magnitude, and ϕz its phase. For w ∈ C n, Re(w) and Im(w) are vectors that collect real and imaginary parts of w, respectively, w ∗ is the conjugate transpose of w, and |w|, ϕw are vectors of magnitudes and phases, respectively.

Moreover, cos(ϕw) ∈ R n and e iϕw ∈ C n are componentwise defined functions. p. 1
Where the paper says it

For w ∈ C n, Re(w) and Im(w) are vectors that collect real and imaginary parts of w, respectively, w ∗ is the conjugate transpose of w, and |w|, ϕw are vectors of magnitudes and phases, respectively. Moreover, cos(ϕw) ∈ R n and e iϕw ∈ C n are componentwise defined functions.

The 1-norm is defined as∥w∥1 = Pn k=1 wk, the 2-norm as∥w∥2 = √w∗w, and the ∞-norm as∥w∥∞ = supkwk. p. 1
Where the paper says it

sign(w) := (sign (wk)) ∈ C n with sign (wk) = wk |wk| wk ̸= 0, 0 wk = 0. The 1-norm is defined as ∥w∥1 = Pn k=1 |wk|, the 2-norm as ∥w∥2 = √ w∗w, and the ∞-norm as ∥w∥∞ = supk{|wk|}.

Finally, 1n denotes the n-dimensional vector of ones. p. 1
Where the paper says it

The 1-norm is defined as ∥w∥1 = Pn k=1 |wk|, the 2-norm as ∥w∥2 = √ w∗w, and the ∞-norm as ∥w∥∞ = supk{|wk|}. Finally, 1n denotes the n-dimensional vector of ones.

--- Page 3 ---

III. THE KURAMOTO MODEL AND COMPLEX-VALUED EXTENSION

A. Classical model

Given a simple undirected graph G = (V, E) with N nodes and adjacency matrix A, the Kuramoto model [2] reads as ˙θk = ωk + σ X N j=1 akj sin(θj − θk), (1) where θk ∈ R are oscillator phases, ωk ∈ R natural frequencies, σ > 0 coupling strength, and akj ∈ [0, 1]. Synchronization is measured by the order parameter r(t) = 1/N Pk e iθk(t); r = 1 indicates perfect synchronization. p. 1

Where the paper says it

Classical model Given a simple undirected graph G = (V, E) with N nodes and adjacency matrix A, the Kuramoto model [2] reads as ˙θk = ωk + σ X N j=1 akj sin(θj − θk), (1) where θk ∈ R are oscillator phases, ωk ∈ R natural frequencies, σ > 0 coupling strength, and akj ∈ {0, 1}. Synchronization is measured by the order parameter r(t) = 1 N P k e iθk(t) ; |r| = 1 indicates perfect synchronization.

B. Complex-valued extension and control formulation

Subtracting i cos(θj−θk) from the coupling in (1), performing straightforward algebraic manipulation and setting x = e iθ, yields the linear system [8]: x˙ = (i diag(ω) + σA) x. Splitting (2) into magnitude and argument dynamics gives d/dt xk = σ X N j=1 akj xj cosϕxj − ϕxk, (3a) and d/dt ϕxk = ωk + σ X N j=1 akj xj / xk sinϕxj − ϕxk, (3b) for all k. p. 1

Where the paper says it

Complex-valued extension and control formulation Subtracting i cos(θj−θk) from the coupling in (1), performing straightforward algebraic manipulation and setting x = e iθ, yields the linear system [8]: x˙ = (i diag(ω) + σA) x. (2) Splitting (2) into magnitude and argument dynamics gives d dt |xk| = σ X N j=1 akj |xj | cosϕxj − ϕxk , (3a) d dt ϕxk = ωk + σ X N j=1 akj |xj | |xk| sinϕxj − ϕxk .

If xk → c for all k, then (3b) matches (1). p. 1
Where the paper says it

(3b) for all k. If |xk| → c for all k, then (3b) matches (1).

This can be achieved by introducing a control input u ∈ C N in (2): x˙ = (i diag(ω) + σA) x + u. p. 1
Where the paper says it

If |xk| → c for all k, then (3b) matches (1). This can be achieved by introducing a control input u ∈ C N in (2): x˙ = (i diag(ω) + σA) x + u.

--- Page 4 ---

IV. STATE-OF-THE ART: A CONTROL PERSPECTIVE

A. State-feedback control

In [15] the following linear system on C N is introduced: x˙ = (i diag(ω) − diag(µ) + σA) x, (5) where µ ∈ R N is chosen so that suitable spectral conditions are satisfied, enabling the arguments of (5) to reproduce the dynamics of the real-valued Kuramoto model (1), specifically requiring the matrix in (5) must have a single purely imaginary eigenvalue where all the entries of the associated eigenvector have equal modulus, while the remaining eigenvectors must have strictly negative real part.

B. Hybrid reset control

The method in [9] uses a time-window approach exploiting the linearity of (2): time is divided into intervals of length T, where at each reset the state x is projected onto the unit circle while preserving phases, and then evolves linearly. This construction matches the real-valued Kuramoto dynamics over finite, nontrivial time intervals. p. 1

Where the paper says it

From a hybrid-control viewpoint, this is described by the flow–jump system x˙ = (i diag(ω) + σA) x, (6a) τ˙ = 1, (6b) for (x, τ ) ∈ C = {(x, τ ) ∈ C N × R≥0 : 0 ≤ τ < T}, with jumps at τ = T given by x + = sign(x(t)), (7a) τ + = 0, (7b) for (x, τ ) ∈ D = {(x, τ ) ∈ C N × R≥0 : τ = T}. This construction matches the real-valued Kuramoto dynamics over finite, nontrivial time intervals, with accuracy improving for smaller T due to reduced inter-reset amplitude drif

Unified Perspective

Although the approaches in [15] and [9] were developed independently, they share a common control objective: enforce a uniform modulus on the complex states of (4) so that their arguments reproduce the phase dynamics of (1). Both exhibit fundamental performance limitations: asymptotic convergence with constrained gain selection in the former, and discretetime matching with inter-reset magnitude drift in the latter. p. 2

Where the paper says it

Unified perspective Although the approaches in [15] and [9] were developed independently, they share a common control objective: enforce a uniform modulus on the complex states of (4) so that their arguments reproduce the phase dynamics of (1). Both exhibit fundamental performance limitations: asymptotic convergence with constrained gain selection in the former, and discretetime matching with inter-reset magnitude drift in the latter.

Proposed Solutions

Motivated by this unified perspective, this work leverages the linear complex-valued formulation to address two control problems for the Kuramoto model. First, we aim to exactly reproduce the phase dynamics of the original real-valued system. p. 0

Where the paper says it

Motivated by this unified perspective, this work leverages the linear complex-valued formulation to address two control problems for the Kuramoto model. First, we aim to exactly reproduce the phase dynamics of the original real-valued system, improving upon [15] and [9].

Second, we seek to enforce phase locking at a prescribed frequency at a prescribed frequency regardless of natural frequencies and coupling strengths. p. 0
Where the paper says it

First, we aim to exactly reproduce the phase dynamics of the original real-valued system, improving upon [15] and [9]. Second, we seek to enforce phase locking at a prescribed frequency, regardless of natural frequencies and coupling strengths, achieving robust and controlled synchronization via complex-valued slidingmode control (SMC) [18].

Switched Control Designs

The main contributions of this work are: 1) A switched feedforward law that achieves exact, continuous phase matching with the real-valued Kuramoto dynamics for all t ≥ 0 under unit initial magnitudes. 2) A feedforward + SMC law that upgrades the asymptotic convergence of [15] to finite-time convergence, with an explicit and tunable reaching-time bound. 3) An extension of the complex-valued MIMO SMC framework of [19] to non-autonomous systems. p. 0

Where the paper says it

No spectral gain tuning is required. 3) An extension of the complex-valued MIMO SMC framework of [19] to non-autonomous systems, applied to arXiv:2604.

--- Page 5 ---

V. SWITCHED CONTROLLERS

A. Switched feedforward control

Theorem 1. Let u = (uke iϕuk) ∈ C N with uk = fk(x), (8a)ϕuk = ϕxk + π/2(1 + sign(fk(x))). p. 2

Where the paper says it

, N} we define the vector fields fk, gk : C N → R from (3): fk(x) = σ X N j=1 akj |xj | cosϕxj−ϕxk , gk(x) = ωk + σ X N j=1 akj |xj | |xk| sinϕxj−ϕxk . Let u = (|uk|e iϕuk ) ∈ C N with |uk| = |fk(x)|, (8a) ϕuk = ϕxk + π 2 (1 + sign(fk(x))).

If xk(0) = 1 for all k, then xk ≡ 1 for all t ≥ 0 and the argument dynamics of (4) coincide exactly with (1). p. 2
Where the paper says it

Let u = (|uk|e iϕuk ) ∈ C N with |uk| = |fk(x)|, (8a) ϕuk = ϕxk + π 2 (1 + sign(fk(x))). (8b) If |xk(0)| = 1 for all k, then |xk(t)| ≡ 1 for all t ≥ 0 and the argument dynamics of (4) coincide exactly with (1).

Theorem 2. Let u = u 1 + u 2, with u 1 defined by (8) and u 2k = αsign(xk − 1), ϕu 2k = ϕxk + π/2(1 + sign(xk − 1)). p. 2

Where the paper says it

Feedforward action with SMC To overcome the requirement of unit initial magnitudes, we augment the controller (8) with an SMC term that compensates for any drift of the state magnitudes away from unity. Let u = u 1 + u 2 , with u 1 defined by (8) and |u 2 k | = α|sign(|xk| − 1)|, (11a) ϕu 2 k = ϕxk + π 2 (1 + sign(|xk| − 1)), (11b) with α ∈ R +.

Then, for any initial condition, the arguments of (4) converge to (1) in finite time T ≤ √2α x(0) − 1N squared. p. 2
Where the paper says it

Let u = u 1 + u 2 , with u 1 defined by (8) and |u 2 k | = α|sign(|xk| − 1)|, (11a) ϕu 2 k = ϕxk + π 2 (1 + sign(|xk| − 1)), (11b) with α ∈ R +. Then, for any initial condition, the arguments of (4) converge to (1) in finite time T ≤ √ 2 α ∥|x(0)| − 1N ∥2.

--- Page 6 ---

VI. SYNCHRONIZATION VIA COMPLEX-VALUED SMC

Consider the complex-valued, non-autonomous system x˙ = F(x, t) + G(x, t)u. We define the complex switching manifold S = (x, t) ∈ Ω × R≥0: s(x, t) = 0. p. 3

Where the paper says it

Let the complex switching manifold be defined by S = {(x, t) ∈ Ω × R≥0 : s(x, t) = 0}, (15) where the complex switching function s:Ω×R → CM is also assumed holomorphic in x and continuous in t. Finally, we define the matrix-valued function D(x, t) = ∂s ∂xG(x, t).

Theorem 3. Suppose that Assumption 1 holds. p. 3
Where the paper says it

(17) Theorem 3. Suppose that Assumption 1 holds.

Then, the switched control action u = - diag(K) sign(s) (18) induces a sliding motion of (14) on Ω1 × R≥0 ∩ S. p. 3
Where the paper says it

Suppose that Assumption 1 holds. Then, the switched control action u = − diag(K) sign(s) (18) induces a sliding motion of (14) on Ω1 × R≥0 ∩ S.

Building on Theorem 3, we can now introduce a control strategy to actively enforce synchronization at a prescribed frequency ω¯ ∈ R>0. p. 3
Where the paper says it

Second, the reaching-time bound must absorb this additional term, which is accommodated through condition (16b) via the constant ϵ2. Building on Theorem 3, we can now introduce a control strategy to actively enforce synchronization at a prescribed frequency ω¯ ∈ R>0.

A sufficient condition for this is given in the following theorem. p. 3
Where the paper says it

, N} that |xk| = 1 and ϕxk = ¯ωt, indeed having all the oscillators phase locked at frequency ω¯. A sufficient condition for this is given in the following theorem.

--- Page 7 ---

VII. NUMERICAL VALIDATION

The results are shown in Figure 1. The phases of the complex-valued model achieve complete phase locking (panel a), and behave as the real-valued model, as evidenced by the synchronization order parameters (panel c). p. 4

Where the paper says it

The results are shown in Figure 1. The phases of the complex-valued model achieve complete phase locking (panel a), and behave as the real-valued model, as evidenced by the synchronization order parameters (panel c).

The robustness of the proposed strategy is assessed in panel (d), where the natural frequencies are drawn from a normal distribution N (2π, 1) and the coupling strength is reduced to σ = 0. p. 4
Where the paper says it

The synchronization order parameter in panel (c) further confirms fast finite-time convergence; for comparison, it also includes the order parameter of the corresponding real-valued Kuramoto model, which synchronizes under the same conditions but at its natural frequency 2π rad/s rather than the prescribed 4π rad/s. The robustness of the proposed strategy is assessed in panel (d), where the natural frequencies are drawn from a normal distribution N (2π, 1) and the coupling strength is reduced to σ = 0.

1. This result highlights the fundamental advantage of the complex-valued control formulation. p. 4
Where the paper says it

Under these heterogeneous conditions, the real-valued Kuramoto model fails to synchronize, yet the SMC successfully enforces phase locking across the entire network at the prescribed frequency. This result highlights the fundamental advantage of the complex-valued control formulation: by operating on the linear embedding rather than the nonlinear phase equations, the controller can impose synchronization even in regimes where the uncontrolled realvalued model does not synchronize spontaneously.

--- Page 8 ---

VIII. CONCLUSIONS

This work introduced a unified control-theoretic framework for complex-valued Kuramoto oscillator networks and proposed three switched control strategies that overcome key limitations of existing methods. The switched feedforward law achieves exact continuous-time phase correspondence with classical Kuramoto dynamics for all t ≥ 0. p. 5

Where the paper says it

By interpreting the statefeedback approach of [15] and the hybrid reset method of [9] as mechanisms to enforce a common modulus, we identified a shared control objective and used it to guide improved designs. Within this framework, the switched feedforward law (Theorem 1) achieves exact continuous-time phase correspondence with classical Kuramoto dynamics for all t ≥ 0, eliminating the drift-induced mismatch inherent to reset-based schemes.

The combined feedforward-SMC law ensures finite-time convergence with an explicit, tunable reaching-time bound. p. 5
Where the paper says it

Within this framework, the switched feedforward law (Theorem 1) achieves exact continuous-time phase correspondence with classical Kuramoto dynamics for all t ≥ 0, eliminating the drift-induced mismatch inherent to reset-based schemes. The combined feedforward-SMC law (Theorem 2) relaxes the unit-magnitude initial condition and ensures finite-time convergence with an explicit, tunable reaching-time bound, improving upon the asymptotic result of [15] without spectral gain tuning.

Finally, the prescribed-frequency SMC law enforces phase locking at a desired frequency in finite time, independent of natural frequencies and coupling strengths. p. 5
Where the paper says it

The combined feedforward-SMC law (Theorem 2) relaxes the unit-magnitude initial condition and ensures finite-time convergence with an explicit, tunable reaching-time bound, improving upon the asymptotic result of [15] without spectral gain tuning. Finally, the prescribed-frequency SMC law (Theorem 4), based on a novel non-autonomous extension of complex-valued MIMO sliding-mode control, enforces phase locking at a desired frequency in finite time, independent of natural frequencies and coupling strengths.

--- Page 1 ---

Complex-Valued Kuramoto Networks: A Unified Control-Theoretic Framework

Lorenzo Giordano, Josep M. Olm, and Mario di Bernardo

Abstract—Synchronization in networks of coupled oscillators is classically studied via the Kuramoto model, whose intrinsic nonlinearity limits analytical tractability and complicates control design Complex-valued extensions circumvent this by embedding phase dynamics into a higher-dimensional linear state space, where regulating complex-state moduli to a common value recovers Kuramoto phase behavior. Existing approaches to address this problem correspond, within a unified control framework, to statefeedback and hybrid reset-based strategies, each with performance constraints. We propose two switched control designs that overcome these limitations: a switched feedforward law ensuring exact phase correspondence at all times, and a feedforward plus sliding-mode law achieving finite-time convergence without spectral gain tuning. Additionally, we present a non-autonomous complex-valued MIMO sliding-mode controller that enforces phase locking at a prescribed frequency in finite time, independent of natural frequencies and coupling strengths.

Index Terms—Kuramoto oscillators, complex-valued systems, sliding-mode control, feedforward control, synchronization, switched systems, network control.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Switched feedforward control can achieve exact continuous-time phase correspondence with classical Kuramoto dynamics for all t >= 0 provided initial magnitudes are unity, overcoming the discrete-instant limitation of reset-based methods.

  2. Feedforward plus SMC law upgrades asymptotic convergence to finite-time convergence with an explicit and tunable reaching-time bound, requiring no spectral gain tuning.

  3. A non-autonomous complex valued MIMO sliding mode controller can enforce phase locking at a prescribed frequency in finite time, independent of natural frequencies and coupling strengths, even in heterogeneous networks where the classical real valued model fails to synchronize.

Abstract

Synchronization in networks of coupled oscillators is classically studied via the Kuramoto model, whose intrinsic nonlinearity limits analytical tractability and complicates control design. Complex-valued extensions circumvent this by embedding phase dynamics into a higher-dimensional linear state space, where regulating complex-state moduli to a common value reproduces the nominal Kuramoto phase dynamics on an invariant manifold. Existing approaches to address this problem correspond, within a control framework, to state-feedback and hybrid reset-based strategies, each with performance constraints. We propose two switched control designs that overcome these limitations: a switched feedforward law ensuring exact phase correspondence at all times, and a feedforward plus sliding-mode law achieving finite-time mapping onto the original Kuramoto dynamics without spectral gain tuning. Additionally, we present a non-autonomous complex-valued MIMO sliding-mode controller that enforces phase locking at a prescribed frequency in finite time, regardless of natural frequencies and coupling strengths, under appropriate control gain selection. Simulations confirm improved transient response, steady-state accuracy, and robustness, including synchronization of heterogeneous networks where the classical real-valued Kuramoto model fails.

Related papers