Optimal design of higher-order interactions for mutual synchronization among three collectively oscillating networks
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Nonlinear Sciences. The day in nonlinear science: chaos, patterns, networks and adaptation.
Kai: We're looking at the paper: "Optimal design of higher-order interactions for mutual synchronization among three collectively oscillating networks".
Mira: Here's the gist of the paper. Optimal design of higher-order interactions for mutual synchronization among three collectively oscillating networks investigates how to optimize higher-order internetwork coupling for mutual synchronization among three collectively oscillating networks The study derives optimal coupling tensors that maximize either the linear stability or the rotational property for a fixed overall strength of the internetwork coupling.
Kai: Model Setup and Phase Reduction The authors introduce a general model of three collectively oscillating networks, labeling them by X ∈ A, B, C, where network X consists of N X dynamical elements.
Mira: The full dynamical system is described by equations (1)–(3), involving intrinsic dynamics fX i and interaction terms gX ij and hX ijk.
Kai: Assuming weak internetwork coupling ϵ (0 < ϵ ≪ 1) and sufficiently close collective frequencies, the dynamics are reduced to a two-dimensional system of phase-difference equations by applying collective phase reduction and averaging.
Title and authors: Mira: Okay, so the title itself points out the core idea: optimizing higher-order interactions for mutual synchronization among three collectively oscillating networks. It sets up the problem right there—it’s not just two networks anymore, it’s three and beyond, and we need to get those couplings right.
Kai: Right. And what they're doing is using phase reduction theory to take the full, complex dynamics of these coupled systems and boil them down to a much simpler two-dimensional system of phase differences.
Lev: That reduction step is crucial because it lets them characterize things like linear stability and rotational properties using a coupling tensor shared by all three networks. That’s where the actual math gets applied to the physics.
Mira: Precisely. They then derive optimal coupling tensors that maximize either the linear stability or the rotational property for a fixed overall strength of that internetwork coupling. It’s about finding the best way to connect them up when you can only control how strong those connections are globally.
Kai: So, instead of just picking random connections, they find these specific patterns—these optimal tensors—that lead to better results than what you'd get by just being random with the same total coupling strength.
The paper's summary: Lev: The summary boils down to applying phase reduction theory to three networks and then using optimization techniques based on that framework to find coupling tensors that either maximize the linear stability or the rotational property.
Mira: That’s right. They use the reduced system of phase differences, which we call phi one and phi two and they show that both linear stability and rotational properties are determined by a higher-order coupling tensor shared across all three networks <ref:2611.02589#pg1>.
Kai: And they then mathematically derive these optimal tensors for the coupling vector, subject to a norm constraint on the overall strength of the internetwork coupling. It’s an optimization problem in itself.
Lev: The results show that for maximizing linear stability, they find a specific coupling vector c and it scales like kappa times some vector w.
Mira: And for maximizing rotational property, under a certain condition where four tau is greater than or equal to sigma squared, they find another optimal coupling vector c which is proportional to the maximum value of some quantity called nu max.
Kai: The numerical example they use involves three networks, each with four FitzHugh–Nagumo elements coupled in a tree structure. They show that these two optimized couplings give faster convergence to the phase-locked state and stronger rotational motion around it compared to a random coupling tensor of the same overall strength.
The paper's improvements: Lev: The suggested improvements focus on extending this work beyond just finding one optimal coupling for stability or rotation. They suggest ways to treat linear stability and rotational property as a single objective function instead of optimizing them separately.
Mira: That’s interesting because it means you could design a coupling tensor that balances both convergence speed and rotational dynamics at the same time, instead of picking one or the other based on what you need more right then.
Kai: And they also point out something practical: they suggest using L1 norm constraints to identify sparse internetwork coupling matrices. This is important because in real systems, the number of hyperedges can grow very fast, and sparsity helps us find useful connections.
Lev: Sparsity would be huge for me on hardware implementation; if we can identify which three-body interactions actually matter most, we don't waste resources trying to model every possible connection.
Mira: And they also mention that the framework can be extended to include amplitude dynamics when you need stronger internetwork coupling. That lets us move beyond just phase differences and consider the actual state variables of the elements too.
Conclusion: Kai: So, to wrap up on this paper, the optimal design of higher-order interactions for mutual synchronization among three collectively oscillating networks gives us explicit coupling tensors that beat random couplings in terms of both convergence speed and rotational motion.
Lev: For real hardware, this means we have a blueprint for designing those three-body connections that can make synchronization happen faster and with more dynamic behavior than just guessing or using random configurations.
Mira: They show how to use phase reduction theory to systematically find these optimal tensors by maximizing either the linear decay rate or the rotational property for a fixed overall coupling strength, which is a key way to design better systems.
Kai: It’s about moving from guesswork to finding specific, mathematically derived connection strengths that lead directly to better synchronization behavior in multi-network setups.
Lev: I think this framework is valuable because it gives us the tools to actually test these optimal designs on physical systems and see if they hold up under real-world noise and imperfections.
Mira: It’s a solid piece of work that shows how higher-order interactions matter significantly when you move beyond simple pairwise coupling between two networks.
Kai: That’s it for this paper on the optimal design of higher-order interactions for mutual synchronization among three collectively oscillating networks.
Sotaro Kuroiwa, *, Yuzuru Kato, *
Department of Systems and Control Engineering, Institute of Science Tokyo · Department of Complex and Intelligent Systems, Future University Hakodate
nlin.AO
Submitted: 2026-11-02
Updated: 2026-11-02
Comments: 35 pages, 6 figures
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 78/100
The gist: Optimal design of higher-order interactions for mutual synchronization among three collectively oscillating networks investigates how to optimize higher-order internetwork coupling for mutual…
Key concepts
- Collective Phase Reduction
- This technique simplifies the complex dynamics of many interacting elements into a simpler system describing only the collective phases of the networks. It is used because the networks oscillate collectively and their frequencies are close, reducing high-dimensional dynamics to a manageable two-dimensional phase-difference system.
- Higher-Order Coupling Tensor
- This tensor describes how three networks interact with each other directly (three-body coupling). The paper derives optimal forms of this tensor that maximize desired synchronization properties like linear stability or rotational motion, given a fixed total strength of the coupling.
- Linear Stability and Rotational Property
- These are key measures used to determine how stable a synchronized state is. Linear stability relates to how quickly small perturbations decay (convergence speed), while the rotational property describes the motion of phase differences around the synchronized state.
Terminology
Summary
Optimal design of higher-order interactions for mutual synchronization among three collectively oscillating networks investigates how to optimize higher-order internetwork coupling for mutual synchronization among three collectively oscillating networks The study derives optimal coupling tensors that maximize either the linear stability or the rotational property for a fixed overall strength of the internetwork coupling. p. 0
Model Setup and Phase Reduction
The authors introduce a general model of three collectively oscillating networks, labeling them by X ∈ A, B, C, where network X consists of N X dynamical elements. The full dynamical system is described by equations (1)–(3), involving intrinsic dynamics fX i and interaction terms gX ij and hX ijk. p. 4 Assuming weak internetwork coupling ϵ (0 < ϵ ≪ 1) and sufficiently close collective frequencies, the dynamics are reduced to a two-dimensional system of phase-difference equations by applying collective phase reduction and averaging. p. 4 This leads to coupled equations for the slow phases ΦX, which evolve at a rate of O(ϵ) because the common fast rotation ωt has been removed. p. 5
Linear Stability Analysis
The synchronization dynamics are reduced to a two-dimensional system of phase-difference equations where linear stability and rotational property are determined by a higher-order coupling tensor shared by the three networks. The eigenvalues of the Jacobian matrix J are given by λ± = −σ2 ± √σ2 − 4τ2, where σ = −trJ and τ = det J. p. 7 The fixed point is linearly stable if and only if Λ > 0, where Λ gives the exponential decay rate of small perturbations from the fixed point. p. 8
Optimization Results
The optimization problems are formulated for the coupling vector c subject to a norm constraint∥c∥2 = κ. The optimal coupling vector maximizing the linear decay rate Λ is given by c⋆ Λ = κ w / w squared, yielding a maximum rate of Λmax(ϕ∗ 1, ϕ∗ 2) = κ2 w squared. p. 9 For the rotational component R under the condition 4τ ≥ σ2, the optimal coupling vector is c⋆ R = κ vmax, attaining Rmax(ϕ∗ 1, ϕ∗ 2) = κ2 √νmax. p. 9
Numerical Example and Comparison
The numerical example considers three networks of four FitzHugh–Nagumo elements coupled in a tree structure. The two optimized couplings yield faster convergence to the phase-locked state and stronger rotational motion of the phase differences around it compared to a random coupling tensor with the same overall strength. p. 2 For identical networks, the optimal coupling yields Λ = 4. p. 66220 and R = 0 at (0, 0), while a random cyclically invariant coupling tensor gives Λ = 1. p. 00342 and R = 0 at (0, 0). For non-identical networks, the optimal coupling yields R = 35. p. 37 times larger than that of the random coupling. p. 0
Conclusion and Future Directions
The framework allows for extensions such as treating linear stability and rotational property as a single objective or designing pairwise and three-body couplings together. The study also suggests using L1 norm constraints to identify sparse internetwork coupling matrices, which is practically important because the number of hyperedges grows rapidly. p. 0 Furthermore, the framework can be extended by incorporating amplitude dynamics to treat stronger internetwork coupling and by using data-driven methods to estimate necessary quantities from time-series data. p. 1 The optimal design of higher-order interactions is an important problem at the intersection of network science and control theory. p. 0
Acknowledgments
The authors acknowledge JSPS KAKENHI JP22K14274 and JST PRESTO JPMJPR24K3 for financial support. The references listed are [1] to [35] on page 18. The reference list is provided on page 26. p. 5 The appendix A is on page 27. p. 8 The appendix B is on page 31. p. 8 The appendix C is on page 32. p. 8 The appendix D starts on page 34 and ends on page 35. p. 8
--- Page 1 ---
Optimal design of higher-order interactions for mutual synchronization among three collectively oscillating networks Sotaro Kuroiwa1, ∗ and Yuzuru Kato2, † Department of Systems and Control Engineering, Institute of Science Tokyo, Tokyo 152-8552, Japan Department of Complex and Intelligent Systems, Future University Hakodate, Hokkaido 041-8655, Japan (Dated: November 3, 2026) Abstract Many networked dynamical systems consist of interacting subnetworks, each of which can exhibit collective oscillations. Mutual synchronization among such collectively oscillating networks is ubiquitously observed, and its efficient control is an important engineering problem. p. 0 While control of synchronization has been extensively studied for pairwise interactions, many systems also involve higher-order interactions among three or more subnetworks that cannot be decomposed into independent pairwise couplings. p. 0 In this study, we investigate the optimization of higher-order internetwork coupling for mutual synchronization among three collectively oscillating networks. p. 0 Using phase reduction theory for collectively oscillating networks, we reduce the full network dynamics to a two-dimensional system of phase-difference equations, in which the linear stability of the phase-locked state and the rotational property of the phase differences around it are characterized by a higher-order coupling tensor shared by the three networks. p. 0 We then derive the optimal coupling tensors that maximize either the linear stability or the rotational property for a fixed overall strength of the internetwork coupling. p. 0 As a numerical example, we consider three collectively oscillating networks, each consisting of four FitzHugh–Nagumo elements coupled in a tree structure, and demonstrate that, compared with a random coupling tensor of the same overall strength, the two types of optimal coupling yield faster convergence to the phase-locked state and stronger rotational motion of the phase differences around it. p. 0 p. 0 arXiv:2611. p. 002589v1 [nlin. p. 0AO] 2 Nov 2026
--- Page 3 ---
I. INTRODUCTION Many natural and engineered systems can be described as networks of interacting dynamical elements, such as power grids [1], neuronal networks [2], gene-regulatory networks [3], ecological communities [4], and socio-technical systems [5]. p. 1 In such systems, the collective dynamics strongly depends on the topology of interactions, including small-world [6] and scale-free [7] structures. p. 1 Synchronization is one of the most fundamental collective phenomena on networks and is observed in a wide range of physical, biological, and engineering systems [8–10]. p. 1 Understanding and controlling synchronization is therefore one of the central problems in the analysis and design of networked dynamical systems. p. 1 A standard theoretical framework for analyzing the synchronization of weakly coupled nonlinear oscillators is phase reduction theory [8, 11, 12]. p. 1 When each oscillator possesses a stable limit-cycle solution, its multidimensional dynamics can be reduced to a one-dimensional phase equation, and the effect of weak coupling is described by a phase coupling function that depends only on the phase differences between the oscillators. p. 1 The phase reduction framework has also enabled systematic optimization of synchronization. p. 1 For example, optimal waveforms of periodic forcing for entrainment [13, 14] and optimal coupling functions for mutual synchronization [15, 16] have been derived under constraints on the intensity of the forcing or coupling. p. 1 More recently, phase reduction theory has been extended from individual oscillators to networks of coupled dynamical elements exhibiting collective oscillations, allowing the dynamics of an entire network to be described by a single phase variable, called the collective phase. p. 1 Within this framework, the internetwork coupling between a pair of collectively oscillating networks has been optimized for mutual synchronization. p. 1 Conventional models of network dynamics typically assume pairwise interactions, in which each link connects only two nodes. p. 1 However, many real-world systems involve higher-order interactions among three or more units that cannot be decomposed into independent pairwise couplings. p. 1 Such interactions arise, for example, in scientific collaborations, ecological communities, brain networks, and social contagion [20]. p. 1 Hypergraphs and simplicial complexes have therefore been introduced as natural mathematical frameworks for describing such higher-order networks. p. 1 Higher-order interactions can qualitatively alter collective dynamics, giving rise to phenomena such as multistability [26] and abrupt synchronization transitions [27, 28], and modifying the stability of synchronized states [29, 30]. p. 2
--- Page 3 ---
In this study, we consider three collectively oscillating networks interacting through weak higher-order internetwork coupling. By extending the phase reduction framework for collective oscillations [18], we reduce the full network dynamics to a two-dimensional system of phase-difference equations. p. 2 In the reduced system, both the linear stability of the phaselocked state and the rotational property of the phase differences around it are determined by a higher-order coupling tensor shared by the three networks. p. 2 Following optimization approaches developed in Ref. p. 2 [19], we formulate optimization problems for this coupling tensor for a fixed overall strength of the internetwork coupling and analytically derive the optimal tensors that maximize either the linear stability or the rotational property. p. 2 As a numerical example, we consider three collectively oscillating networks, each consisting of four FitzHugh–Nagumo elements coupled in a tree structure, and demonstrate that, compared with a random coupling tensor of the same overall strength, the two types of optimal coupling yield faster convergence to the phase-locked state and stronger rotational motion of the phase differences around it.
--- Page 4 ---
FIG. 1. Schematic of three collectively oscillating networks with higher-order internetwork coupling Each colored triangle represents a higherorder internetwork coupling among one node from each of the three networks, and the whole set of such couplings is specified by a coupling tensor. p. 3
--- Page 5 ---
In this section, we introduce a general model of three collectively oscillating networks, which can be non-identical in their internal structures and in the dynamics of their individual elements. We label the networks by X ∈ A, B, C and assume that network X consists of N X dynamical elements. p. 3 p. 3 The state of the ith element in network X is denoted by xX i ∈ R mX i, where mX i is the dimension of its state space. p. 3 p. 3 The full dynamical system is given by equations (1) to (3), which include intrinsic dynamics fX i, pairwise interactions gX ij, and three-body internetwork coupling hX ijk. p. 4 p. 4
--- Page 6 ---
A. Three collectively oscillating networks We first introduce a general model of the three collectively oscillating networks, which can be non-identical in their internal structures and in the dynamics of their individual elements. p. 3 We label the networks by X ∈ A, B, C and assume that network X consists of N X dynamical elements. The state of the ith element in network X is denoted by xX i ∈ R mX i, where mX i is the dimension of its state space. The full dynamical system is given by equations (1) to (3), which include intrinsic dynamics fX i, pairwise interactions gX ij, and three-body internetwork coupling hX ijk.
--- Page 7 ---
B. Phase reduction and averaging Assuming that the internetwork coupling is weak and that the collective frequencies of the three networks are sufficiently close to each other, we apply collective phase reduction and averaging to derive coupled equations for the collective phases of the three networks. p. 4 This leads to a two-dimensional system of phase-difference equations. p. 0 The effective collective phase coupling function ΓX(ϕXY, ϕXZ) is derived by summing contributions from all combinations of elements. p. 6
--- Page 8 ---
C. Phase-difference equations and linear stability analysis Among these phase differences ϕXY = θY − θX introduced in equation (10), only two are independent, taken as ϕ1 = ϕAB = θB − θA and ϕ2 = ϕAC = θC − θA. p. 6 Subtracting the equations yields the phase-difference equations (18) and (19). p. 6 The Jacobian matrix J of (F1, F2) with respect to (ϕ1, ϕ2), evaluated at the fixed point is given by J = (−ΓB 1 − ΓB 2 − ΓA 1 ΓB 1 − ΓA 2 ΓC 2 − ΓA 1 −ΓC 1 −ΓC 2. p. 7 The eigenvalues are λ± = -σ2 ± √σ2 - 4τ2, where σ = trJ and τ = det J. p. 7
--- Page 9 ---
III. OPTIMAL DESIGN OF HIGHER-ORDER COUPLINGS In this section, we optimize the strengths of the higher-order (three-body) internetwork couplings among the three networks on the basis of the quantities obtained in the linear stability analysis in the previous section. p. 8 The coupling tensor C is expressed as a product of a scalar coefficient and a prescribed coupling function. p. 8
--- Page 10 ---
In what follows, for notational simplicity, we introduce a vector representation of the coupling tensor c = vec(C) ∈ R L with L = N AN BNC. Denoting by l(i, j, k) the position of Cijk in c. p. 8 We define the gradient vectors consisting of the partial derivatives of V A ijk and V B jki and V C kij in equation (27) as (aA1) l(i,j,k) = ∂1V A ijk(ϕ∗1, ϕ∗2), (aA2) l(i,j,k) = ∂2V A ijk(ϕ∗1, ϕ∗ 2), (aB1) l(i,jki) = ∂1V B jki(ϕ∗2 - ϕ∗ 1, −ϕ∗ 1), (aB2) l(i,jki) = ∂2V B jki(ϕ∗2 - ϕstar-phi star-phi star-phi. p. 9 The derivatives ΓXm defined in equations (20)–(24) are expressed as ΓXm = (aXm) T c, m ∈ 1, 2. p. 9
--- Page 11 ---
Defining w = X A 1 + aA 1 + aA 2 and z = X A 1 - aA 2. We can express σ in equation (24) as the linear form σ = wT c. p. 9 Similarly, with sym(K) = (K + KT)/2 we define the symmetric matrix M as M = sym(aA 1(aB 1) T + aB 1(aC 1) T + aC 1(aA 1) T + aA 2(aB 2) T + aB 2(aC 2) T + aC2(aA2) T. p. 9 Thus, the determinant τ in equation (24) is expressed as τ = c TMc. p. 9 Substituting equations (31) and (33) into equation (26), we express the linear decay rate and the rotational component in terms of the coupling vector as Λ = 1/2 wT c, R = 1/2 pmax(4τ − σ2, 0). p. 9
--- Page 12 ---
Thus, Λ is a linear function and R2 is a quadratic function of c in the regime cTPc ≥ 0. To fix the overall magnitude of the coupling tensor, we impose the norm constraint∥c∥2 = κ. p. 9
--- Page 13 ---
A. Optimization of linear stability
The problem of maximizing the linear decay rate Λ in equation (34) under the norm constraint (35) is formulated as max c Λ(c) = 1/2 wT c subject to c2 = κ. The optimal coupling vector is given by c⋆ Λ = κ w / w squared, and the maximum linear decay rate is Λmax(ϕ∗ 1, ϕ∗ 2) = κ2 w squared. p. 9
B. Optimization of the rotational component
Under the condition 4τ ≥ σ2, we have R = √cTPc/2, so maximizing R is equivalent to maximizing the quadratic form cTPc. The optimal coupling vector is c⋆ R = κ vmax, attaining Rmax(ϕ∗ 1, ϕ∗ 2) = κ2 √νmax. p. 8
C. Random Coupling
To evaluate the optimization results, random coupling vectors are sampled from a standard normal distribution and normalized to satisfy the norm constraint. The resulting fixed point is retained only if it is a stable spiral, i. p. 8e., Λ > 0 and 4τ > σ2. p. 8
D. Numerical Calculations
The authors use an alternating iteration to compute the optimal coupling vector and corresponding fixed point by evaluating gradient vectors at each iteration.
--- Page 14 ---
FIG. 2. Collective oscillation of a tree network of four FitzHugh–Nagumo elements (a) Network topology (b) Collective limit cycle in the (u, v) plane (c) Time evolution of the v components over one period (d) The v components of the collective phase sensitivity functions Qi(θ) defined in equation (6), obtained numerically by the adjoint method. p. 5
--- Page 15 ---
The three collectively oscillating networks are coupled through the higher-order internetwork coupling as illustrated in figure 1.
--- Page 16 ---
Figure 2 shows the tree network and its collective oscillation (a) Network topology (b) Collective limit cycle in the (u, v) plane (c) Time evolution of the v components over one period (d) The v components of the collective phase sensitivity functions Qi(θ) defined in equation (6), obtained numerically by the adjoint method.
--- Page 17 ---
The optimal coupling gives a …
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
-
Optimization of higher-order internetwork coupling for mutual synchronization among three networks can be used to design specific coupling tensors that maximize either linear stability or rotational property for a fixed overall strength of the internetwork coupling. This allows AI systems to achieve faster convergence to the phase-locked state or stronger rotational motion around it compared to random couplings.
-
The derived optimal coupling vectors, such as c star lambda, can be used in alternating iteration procedures with root-finding methods to determine the self-consistent fixed point and coupling tensor for non-identical networks. This enables the AI system to find specific synchronization states that are not achievable through simple random sampling.
-
The framework can be extended by treating linear stability and rotational property as a single objective, such as maximizing a weighted sum of lambda and R, allowing for synchronized designs that balance both convergence speed and rotational dynamics simultaneously.
Related papers
- Scale-Dependent Collective Adaptation in Self-Amending LLM Societies: A Cross-Family Study of Emergent Governance
- Paths to synchronization in the Kuramoto model with inertia
- Oscillations in two-stage contagion on networks: contagion structure, turnover and connectivity
- Phase Locking in Oscillator Networks with Uncertain and State-Dependent Couplings
- Cooperation Dynamics with Environmental Feedback: Steady States and Transitions
- When higher-order interactions enhance synchronization: the case of the Kuramoto model