This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Phase synchronization between collective rhythms of fully locked oscillator groups

Yoji Kawamura ykawamura@jamstec.go.jp Department of Mathematical Science and Advanced Technology, Japan Agency for Marine-Earth Science and Technology, Yokohama 236-0001, Japan
(April 30, 2014)
Abstract

We study the phase synchronization between collective rhythms of fully locked oscillator groups. For weakly interacting groups of two oscillators with global sinusoidal coupling, we analytically derive the collective phase coupling function, which determines the dynamics of the collective phase difference between the groups. We demonstrate that the groups can exhibit anti-phase collective synchronization in spite of microscopic in-phase external coupling and vice versa.

Synchronization, Coupled oscillators, Phase reduction, Collective phase description, Fully locked states, Laplacian matrix
pacs:
05.45.Xt

A system of coupled oscillators can exhibit a rich variety of dynamical behaviors. When we investigate the dynamical properties of the system, we first analyze individual oscillators and the microscopic interactions between them. However, the structure of a coupled oscillator system is often hierarchical, so that the collective behaviors of the system cannot be fully clarified by simply analyzing each element of the system. For example, we found that two weakly interacting groups of coupled oscillators can exhibit anti-phase collective synchronization between the groups even though all microscopic interactions are in-phase coupling. This counter-intuitive phenomenon can occur even when the number of oscillators belonging to each group is only two, that is, when the total number of oscillators is only four. In this paper, we clarify the mechanism underlying this counter-intuitive phenomenon for two weakly interacting groups of two oscillators with global sinusoidal coupling.

I Introduction

A system of coupled oscillators provides abundant examples of dynamical behaviors including synchronization phenomena ref:winfree80 ; ref:kuramoto84 ; ref:pikovsky01 ; ref:strogatz03 ; ref:manrubia04 ; ref:osipov07 ; ref:mikhailov13 ; ref:hoppensteadt97 ; ref:izhikevich07 ; ref:ermentrout10 ; ref:schultheiss12 . Among them, collective synchronization emerging from coupled phase oscillators has been widely investigated not only for globally coupled systems but also for complex network systems ref:strogatz00 ; ref:acebron05 ; ref:boccaletti06 ; ref:arenas08 ; ref:dorogovtsev08 ; ref:barrat08 . Furthermore, the dynamical behaviors exhibited by interacting groups of globally coupled phase oscillators have been intensively investigated ref:okuda91 ; ref:montbrio04 ; ref:abrams08 ; ref:barreto08 ; ref:sheeba08 ; ref:sheeba09 ; ref:laing09 ; ref:skardal12 ; ref:anderson12 ; ref:laing12 . The appearance of the Ott-Antonsen ansatz ref:ott08 ; ref:ott09 ; ref:ott11 has considerably facilitated theoretical investigations on interacting groups of noiseless nonidentical phase oscillators with global sinusoidal coupling. In addition, interacting groups of globally coupled phase oscillators as well as a system of globally coupled phase oscillators have been experimentally realized using electrochemical oscillators ref:kiss02 ; ref:kiss07 , discrete chemical oscillators ref:taylor09 ; ref:tinsley12 , and mechanical oscillators ref:pantaleone02 ; ref:martens13 .

To study the phase synchronization between macroscopic rhythms, we recently formulated a theory for the collective phase description of macroscopic rhythms emerging from coupled phase oscillators for the following three representative cases: (A) phase coherent states in globally coupled noisy identical oscillators ref:kawamura08 ; ref:ykawamura13a ; ref:kawamura10a , (B) partially phase-locked states in globally coupled noiseless nonidentical oscillators ref:kawamura10b , and (C) fully phase-locked states in networks of coupled noiseless nonidentical oscillators ref:kori09 . The theory enables us to describe the dynamics of a macroscopic rhythm by a single degree of freedom called the collective phase. Accordingly, different mathematical treatments were required for the physical situation in each case. The keystone of the collective phase description method for each case is the following: (A) the nonlinear Fokker-Planck equation ref:kuramoto84 , (B) the Ott-Antonsen ansatz ref:ott08 ; ref:ott09 ; ref:ott11 , and (C) the Laplacian matrix ref:boccaletti06 ; ref:arenas08 ; ref:dorogovtsev08 ; ref:barrat08 . Here, we note that there exist several investigations ref:masuda09a ; ref:masuda09b ; ref:masuda10 ; ref:ko09 ; ref:toenjes09 ; ref:cross12 ; ref:cross13 related to case (C).

In Ref. ref:kawamura10a for case (A) and Ref. ref:kawamura10b for case (B), we investigated the phase synchronization between collective rhythms of globally coupled oscillator groups. In particular, the collective phase coupling function, which determines the dynamics of the collective phase difference between the groups, was systematically analyzed for sinusoidal coupling functions. As a result, for both cases, we found counter-intuitive phenomena in which the groups can exhibit anti-phase collective synchronization in spite of microscopic in-phase external coupling and vice versa.

In this paper, we study the phase synchronization between collective rhythms of coupled oscillator groups for case (C). We analytically derive the collective phase coupling function for two weakly interacting groups of two oscillators with global sinusoidal coupling (see Fig. 1). We thereby demonstrate counter-intuitive phenomena similar to those found in cases (A) and (B), that is, effective anti-phase (in-phase) collective synchronization with microscopic in-phase (anti-phase) external coupling. Therefore, this paper and Refs. ref:kawamura10a ; ref:kawamura10b are mutually complementary and together provide a deeper understanding of the collective phase synchronization phenomena.

This paper is organized as follows. In Sec. II, we review the collective phase description of fully locked states. In Sec. III, we analyze weakly interacting groups of globally coupled two phase oscillators. In Sec. IV, we perform further analytical calculations for the case of sinusoidal phase coupling. In Sec. V, we illustrate the collective phase coupling function for several representative cases. In Sec. VI, we demonstrate collective phase synchronization by direct numerical simulations. In App. A, we also consider interacting groups of weakly coupled Stuart-Landau oscillators. Concluding remarks are given in Sec. VII.

II Collective phase description of fully locked states

In this section, we review the collective phase description method for fully phase-locked states in networks of coupled noiseless nonidentical oscillators with an emphasis on the derivation of the collective phase coupling function. More details and other applications of this collective phase description method are given in Refs. ref:kori09 ; ref:masuda09a ; ref:masuda09b ; ref:masuda10 .

We consider weakly interacting groups of coupled noiseless nonidentical phase oscillators described by the following equation:

ϕ˙j(σ)(t)=ωj+k=1NΓjk(ϕj(σ)ϕk(σ))+ϵk=1NΓjkστ(ϕj(σ)ϕk(τ)),\dot{\phi}_{j}^{(\sigma)}(t)=\omega_{j}+\sum_{k=1}^{N}\Gamma_{jk}\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\sigma)}\right)+\epsilon\sum_{k=1}^{N}\Gamma_{jk}^{\sigma\tau}\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\tau)}\right), (1)

for j=1,,Nj=1,\cdots,N and (σ,τ)=(1,2),(2,1)(\sigma,\tau)=(1,2),(2,1), where ϕj(σ)(t)𝕊1\phi_{j}^{(\sigma)}(t)\in\mathbb{S}^{1} is the phase of the jj-th oscillator at time tt in the σ\sigma-th group consisting of NN oscillators and ωj\omega_{j} is the natural frequency of the jj-th phase oscillator. The second term on the right-hand side represents the microscopic internal coupling within the same group, while the third term represents the microscopic external coupling between the different groups. The characteristic intensity of the external coupling is given by ϵ0\epsilon\geq 0. When the external coupling is absent, i.e., ϵ=0\epsilon=0, Eq. (1) is assumed to have a stable fully phase-locked collective oscillation solution ref:izhikevich07 ; ref:ermentrout10 ; ref:ermentrout92 ; ref:mirollo05

ϕj(σ)(t)=Θ(σ)(t)+ψj,Θ˙(σ)(t)=Ω=ωj+k=1NΓjk(ψjψk),\phi_{j}^{(\sigma)}(t)=\Theta^{(\sigma)}(t)+\psi_{j},\qquad\dot{\Theta}^{(\sigma)}(t)=\Omega=\omega_{j}+\sum_{k=1}^{N}\Gamma_{jk}\left(\psi_{j}-\psi_{k}\right), (2)

where Θ(σ)(t)𝕊1\Theta^{(\sigma)}(t)\in\mathbb{S}^{1} is the collective phase at time tt for the σ\sigma-th group, Ω\Omega is the collective frequency, and the constants ψj\psi_{j} represent the relative phases of the individual oscillators for the fully phase-locked state.

When the external coupling is sufficiently weak, i.e., ϵ1\epsilon\ll 1, each group of oscillators obeying Eq. (1) is always in the near vicinity of the fully phase-locked solution (2). Therefore, we can approximately derive a collective phase equation in the following form ref:kori09 :

Θ˙(σ)(t)=Ω+ϵϝστ(Θ(σ)Θ(τ)),\dot{\Theta}^{(\sigma)}(t)=\Omega+\epsilon\digamma^{\sigma\tau}\left(\Theta^{(\sigma)}-\Theta^{(\tau)}\right), (3)

where the collective phase coupling function is given by

ϝστ(Θ(σ)Θ(τ))=j=1Nk=1NUjΓjkστ(Θ(σ)Θ(τ)+ψjψk).\digamma^{\sigma\tau}\left(\Theta^{(\sigma)}-\Theta^{(\tau)}\right)=\sum_{j=1}^{N}\sum_{k=1}^{N}U_{j}^{\ast}\Gamma_{jk}^{\sigma\tau}\left(\Theta^{(\sigma)}-\Theta^{(\tau)}+\psi_{j}-\psi_{k}\right). (4)

Here, UjU_{j}^{\ast} is the left zero eigenvector of the Jacobi matrix LjkL_{jk} at the fully phase-locked collective oscillation solution defined in Eq. (2). The Jacobi matrix LjkL_{jk} is given by

Ljk=δjkljΓjl(ψjψl)(1δjk)Γjk(ψjψk),L_{jk}=\delta_{jk}\sum_{l\neq j}\Gamma_{jl}^{\prime}\left(\psi_{j}-\psi_{l}\right)-\left(1-\delta_{jk}\right)\Gamma_{jk}^{\prime}\left(\psi_{j}-\psi_{k}\right), (5)

which is a Laplacian matrix ref:boccaletti06 ; ref:arenas08 ; ref:dorogovtsev08 ; ref:barrat08 . That is, the Jacobi matrix LjkL_{jk} possesses the following property for each jj: k=1NLjk=0\sum_{k=1}^{N}L_{jk}=0. In Eq. (5), we have used the Kronecker delta δjk\delta_{jk} and derivative notation Γjk(ϕ)=dΓjk(ϕ)/dϕ\Gamma_{jk}^{\prime}(\phi)=d\Gamma_{jk}(\phi)/d\phi. We also note that the Jacobi matrix LjkL_{jk} defined in Eq. (5) is generally asymmetric and weighted. Using the (j,j)(j,j)-cofactor of the Jacobi matrix and the summation over the index jj, i.e.,

Mj=detL^(j,j),M=j=1NMj,M_{j}=\det\hat{L}(j,j),\qquad M=\sum_{j=1}^{N}M_{j}, (6)

the left zero eigenvector UjU_{j}^{\ast} of the Jacobi matrix that takes the form of the Laplacian matrix can be generally written in the following form ref:kori09 ; ref:masuda09a ; ref:masuda09b ; ref:masuda10 :

Uj=MjM,j=1NUj=1.U_{j}^{\ast}=\frac{M_{j}}{M},\qquad\sum_{j=1}^{N}U_{j}^{\ast}=1. (7)

In Eq. (6), the matrix L^(j,j)\hat{L}(j,j) is the Jacobi matrix L^\hat{L} with the jj-th row and column removed, and the cofactor MjM_{j} is equal to the sum of the weights of all directed spanning trees rooted at the node jj according to the matrix tree theorem ref:biggs97 ; ref:agaev00 . Finally, we note that the collective phase Θ(σ)\Theta^{(\sigma)} can be written in the following form ref:kori09 :

Θ(σ)=j=1NUj(ϕj(σ)ψj),\Theta^{(\sigma)}=\sum_{j=1}^{N}U_{j}^{\ast}\left(\phi_{j}^{(\sigma)}-\psi_{j}\right), (8)

under the linear approximation of the isochron ref:winfree80 ; ref:kuramoto84 ; ref:pikovsky01 ; ref:izhikevich07 ; ref:ermentrout10 ; ref:schultheiss12 . In the following section, we analyze globally-coupled two-oscillator systems using this collective phase description method for fully locked states.

III Interacting groups of globally coupled two phase oscillators

We first consider weakly interacting groups of globally coupled phase oscillators. That is, the microscopic internal and external coupling functions are given by

Γjk(ϕj(σ)ϕk(σ))\displaystyle\Gamma_{jk}\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\sigma)}\right) =1NΓ(ϕj(σ)ϕk(σ)),\displaystyle=\frac{1}{N}\,\Gamma\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\sigma)}\right), (9)
Γjkστ(ϕj(σ)ϕk(τ))\displaystyle\Gamma_{jk}^{\sigma\tau}\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\tau)}\right) =1NΓστ(ϕj(σ)ϕk(τ)).\displaystyle=\frac{1}{N}\,\Gamma^{\sigma\tau}\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\tau)}\right). (10)

In this global coupling case, Eq. (1) is written in the following form:

ϕ˙j(σ)(t)=ωj+1Nk=1NΓ(ϕj(σ)ϕk(σ))+ϵNk=1NΓστ(ϕj(σ)ϕk(τ)).\dot{\phi}_{j}^{(\sigma)}(t)=\omega_{j}+\frac{1}{N}\sum_{k=1}^{N}\Gamma\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\sigma)}\right)+\frac{\epsilon}{N}\sum_{k=1}^{N}\Gamma^{\sigma\tau}\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\tau)}\right). (11)

We further focus on the case in which the number of oscillators within each group is two, i.e., N=2N=2; a schematic diagram of the case is shown in Fig. 1. In this case, the internal dynamics for each group, i.e., Eq. (11) with ϵ=0\epsilon=0, is described as follows:

ϕ˙1(t)\displaystyle\dot{\phi}_{1}(t) =ω1+12Γ(0)+12Γ(ϕ1ϕ2),\displaystyle=\omega_{1}+\frac{1}{2}\,\Gamma(0)+\frac{1}{2}\,\Gamma\left(\phi_{1}-\phi_{2}\right), (12)
ϕ˙2(t)\displaystyle\dot{\phi}_{2}(t) =ω2+12Γ(0)+12Γ(ϕ2ϕ1),\displaystyle=\omega_{2}+\frac{1}{2}\,\Gamma(0)+\frac{1}{2}\,\Gamma\left(\phi_{2}-\phi_{1}\right), (13)

where we dropped the group index σ\sigma for simplicity. From Eqs. (12) and (13), we obtain the following equation by subtraction:

ddtΔϕ(t)=Δω+12Γ(Δϕ)12Γ(Δϕ),\frac{d}{dt}\Delta\phi(t)=\Delta\omega+\frac{1}{2}\,\Gamma\left(\Delta\phi\right)-\frac{1}{2}\,\Gamma\left(-\Delta\phi\right), (14)

where the phase difference Δϕ(t)\Delta\phi(t) and frequency mismatch Δω\Delta\omega are defined as

Δϕ(t)=ϕ1(t)ϕ2(t),Δω=ω1ω2.\Delta\phi(t)=\phi_{1}(t)-\phi_{2}(t),\qquad\Delta\omega=\omega_{1}-\omega_{2}. (15)

Now, we assume that Eq. (14) has a fully phase-locked collective oscillation solution. The phase difference of the stable phase-locked solution, Δψ=ψ1ψ2\Delta\psi=\psi_{1}-\psi_{2}, is determined by the following equation:

Δω+12Γ(Δψ)12Γ(Δψ)=0.\Delta\omega+\frac{1}{2}\,\Gamma\left(\Delta\psi\right)-\frac{1}{2}\,\Gamma\left(-\Delta\psi\right)=0. (16)

Using the phase difference Δψ\Delta\psi obtained from Eq. (16), the collective frequency Ω\Omega is written in the following form:

Ω=ω1+12Γ(0)+12Γ(Δψ)=ω2+12Γ(0)+12Γ(Δψ).\Omega=\omega_{1}+\frac{1}{2}\,\Gamma(0)+\frac{1}{2}\,\Gamma\left(\Delta\psi\right)=\omega_{2}+\frac{1}{2}\,\Gamma(0)+\frac{1}{2}\,\Gamma\left(-\Delta\psi\right). (17)

For these globally-coupled two-oscillator systems, the Jacobi matrix L^\hat{L} defined in Eq. (5) is given by

L^=12(Γ(Δψ)Γ(Δψ)Γ(Δψ)Γ(Δψ)).\hat{L}=\frac{1}{2}\begin{pmatrix}\Gamma^{\prime}\left(\Delta\psi\right)&-\Gamma^{\prime}\left(\Delta\psi\right)\\ -\Gamma^{\prime}\left(-\Delta\psi\right)&\Gamma^{\prime}\left(-\Delta\psi\right)\\ \end{pmatrix}. (18)

Therefore, the cofactors of the Jacobi matrix are given by

M1=Γ(Δψ)2,M2=Γ(Δψ)2,M_{1}=\frac{\Gamma^{\prime}\left(-\Delta\psi\right)}{2},\qquad M_{2}=\frac{\Gamma^{\prime}\left(\Delta\psi\right)}{2}, (19)

and the sum of these cofactors is written as

M=M1+M2=Γ(Δψ)+Γ(Δψ)2.M=M_{1}+M_{2}=\frac{\Gamma^{\prime}\left(-\Delta\psi\right)+\Gamma^{\prime}\left(\Delta\psi\right)}{2}. (20)

As found from Eq. (7), using these cofactors and the sum, Eq. (19) and Eq. (20), the left zero eigenvector UjU_{j}^{\ast} is obtained as

U1=M1M=Γ(Δψ)Γ(Δψ)+Γ(Δψ),U2=M2M=Γ(Δψ)Γ(Δψ)+Γ(Δψ).U_{1}^{\ast}=\frac{M_{1}}{M}=\frac{\Gamma^{\prime}\left(-\Delta\psi\right)}{\Gamma^{\prime}\left(-\Delta\psi\right)+\Gamma^{\prime}\left(\Delta\psi\right)},\qquad U_{2}^{\ast}=\frac{M_{2}}{M}=\frac{\Gamma^{\prime}\left(\Delta\psi\right)}{\Gamma^{\prime}\left(-\Delta\psi\right)+\Gamma^{\prime}\left(\Delta\psi\right)}. (21)

Finally, we note that the Jacobi matrix L^\hat{L} possesses not only the zero eigenvalue but also the following non-zero eigenvalue:

λ=M=Γ(Δψ)+Γ(Δψ)2.\lambda=M=\frac{\Gamma^{\prime}\left(-\Delta\psi\right)+\Gamma^{\prime}\left(\Delta\psi\right)}{2}. (22)

When the external coupling intensity is sufficiently small compared to the absolute value of this non-zero eigenvalue, i.e., ϵ|λ|\epsilon\ll|\lambda|, the collective phase description is valid ref:kori09 .

IV Analytical formulas for the case of sinusoidal phase coupling

In this section, we consider the case of sinusoidal phase coupling functions for both microscopic internal and external couplings. First, the microscopic internal phase coupling function is given by

NΓjk(ϕ)=Γ(ϕ)=sin(ϕ+α),|α|<π2,N\Gamma_{jk}\left(\phi\right)=\Gamma\left(\phi\right)=-\sin\left(\phi+\alpha\right),\qquad\left|\alpha\right|<\frac{\pi}{2}, (23)

which is in-phase coupling (i.e., attractive). By substituting Eq. (23) into Eq. (16), the phase difference of the fully phase-locked state is obtained as

sin(Δψ)=η,ηΔωcosα,|η|<1,\sin(\Delta\psi)=\eta,\qquad\eta\equiv\frac{\Delta\omega}{\cos\alpha},\qquad\left|\eta\right|<1, (24)

which indicates that the fully phase-locked solution emerge from a saddle-node bifurcation and exists under the condition of |Δω|<cosα|\Delta\omega|<\cos\alpha. Owing to the in-phase coupling, i.e., Eq. (23), one solution of |Δψ|<π/2|\Delta\psi|<\pi/2 is stable, and the other solution of |Δψ|>π/2|\Delta\psi|>\pi/2 is unstable. Hereafter, the fully phase-locked solution indicates the stable one, |Δψ|<π/2|\Delta\psi|<\pi/2. Substituting Eqs. (23) and (24) into Eq. (17), we obtain the collective frequency Ω\Omega as

Ω=ω1η2cosα1+1η22sinα.\Omega=\omega_{1}-\frac{\eta}{2}\cos\alpha-\frac{1+\sqrt{1-\eta^{2}}}{2}\sin\alpha. (25)

Similarly, substituting Eqs. (23) and (24) into Eq. (19), we obtain the cofactors as follows:

M1=1η2cosαηsinα2,M2=1η2cosα+ηsinα2,M_{1}=\frac{-\sqrt{1-\eta^{2}}\,\cos\alpha-\eta\sin\alpha}{2},\qquad M_{2}=\frac{-\sqrt{1-\eta^{2}}\,\cos\alpha+\eta\sin\alpha}{2}, (26)

which yield M=1η2cosαM=-\sqrt{1-\eta^{2}}\,\cos\alpha. From Eqs. (21) and (26), the left zero eigenvector UjU_{j}^{\ast} is thus written as

U1=12(1+ηtanα1η2),U2=12(1ηtanα1η2).U_{1}^{\ast}=\frac{1}{2}\left(1+\frac{\eta\tan\alpha}{\sqrt{1-\eta^{2}}}\right),\qquad U_{2}^{\ast}=\frac{1}{2}\left(1-\frac{\eta\tan\alpha}{\sqrt{1-\eta^{2}}}\right). (27)

In addition, the non-zero eigenvalue λ\lambda defined in Eq. (22) is obtained as

λ=1η2cosα.\lambda=-\sqrt{1-\eta^{2}}\,\cos\alpha. (28)

Next, the microscopic external phase coupling function is given by

NΓjkστ(ϕ)=Γστ(ϕ)=sin(ϕ+β),N\Gamma_{jk}^{\sigma\tau}\left(\phi\right)=\Gamma^{\sigma\tau}\left(\phi\right)=-\sin\left(\phi+\beta\right), (29)

which can be either in-phase coupling (i.e., attractive) under the condition of |β|<π/2|\beta|<\pi/2 or anti-phase coupling (i.e., repulsive) under the condition of |β|>π/2|\beta|>\pi/2. By plugging Eqs. (24), (27), and (29) into Eq. (4), the collective phase coupling function takes the following form:

ϝστ(Θ)=ρsin(Θ+δ),\digamma^{\sigma\tau}\left(\Theta\right)=-\rho\sin\left(\Theta+\delta\right), (30)

where the complex number with modulus ρ\rho and argument δ\delta is given by

ρeiδ=\displaystyle\rho e^{i\delta}= 12[(1+1η2)cosβη21η2tanαsinβ]\displaystyle\,\frac{1}{2}\left[\left(1+\sqrt{1-\eta^{2}}\right)\cos\beta-\frac{\eta^{2}}{\sqrt{1-\eta^{2}}}\tan\alpha\sin\beta\right]
+i2[(1+1η2)sinβ+η21η2tanαcosβ].\displaystyle+\frac{i}{2}\left[\left(1+\sqrt{1-\eta^{2}}\right)\sin\beta+\frac{\eta^{2}}{\sqrt{1-\eta^{2}}}\tan\alpha\cos\beta\right]. (31)

This formula is the main result of the present paper. It determines the collective phase coupling function for two weakly interacting groups of two oscillators with global sinusoidal coupling. The coupling type can be found from the real part, i.e.,

ρcosδ=12[(1+1η2)cosβη21η2tanαsinβ],\rho\cos\delta=\frac{1}{2}\left[\left(1+\sqrt{1-\eta^{2}}\right)\cos\beta-\frac{\eta^{2}}{\sqrt{1-\eta^{2}}}\tan\alpha\sin\beta\right], (32)

where ρcosδ>0\rho\cos\delta>0 and ρcosδ<0\rho\cos\delta<0 indicate in-phase and anti-phase couplings, respectively. Finally, we note that Eq. (32) possesses origin symmetry in the α\alpha-β\beta plane.

V Type of the collective phase coupling function for representative cases

In this section, we study the type of the collective phase coupling function for the following five representative cases.

(i) The first case is η=0\eta=0, which indicates that two oscillators within each group are identical, i.e., Δω=0\Delta\omega=0. Substituting η=0\eta=0 into Eq. (31), we obtain the following result:

η=0,ρeiδ=eiβ.\eta=0,\qquad\rho e^{i\delta}=e^{i\beta}. (33)

That is, the collective phase coupling function is the same as the microscopic external phase coupling function, i.e., ϝστ(Θ)=Γστ(Θ)=sin(Θ+β)\digamma^{\sigma\tau}(\Theta)=\Gamma^{\sigma\tau}(\Theta)=-\sin(\Theta+\beta).

(ii) The second case is |η|1|\eta|\simeq 1, which indicates the proximity of the saddle-node bifurcation point, i.e., the onset of fully phase-locked collective oscillation. Substituting |η|1|\eta|\simeq 1 into Eq. (31), we obtain the following result:

|η|1,ρeiδ12[η21η2tanαsinβ]+i2[η21η2tanαcosβ].\left|\eta\right|\simeq 1,\qquad\rho e^{i\delta}\simeq-\frac{1}{2}\left[\frac{\eta^{2}}{\sqrt{1-\eta^{2}}}\tan\alpha\sin\beta\right]+\frac{i}{2}\left[\frac{\eta^{2}}{\sqrt{1-\eta^{2}}}\tan\alpha\cos\beta\right]. (34)

For the case of |η|1|\eta|\to 1 (excluding α=0\alpha=0), the amplitude of the collective phase coupling becomes infinity, i.e., ρ\rho\to\infty. Here, we note that this property for the fully phase-locked states is quite different from those for phase coherent states and partially phase-locked states ref:kawamura10a ; ref:kawamura10b . For the latter two states, the amplitude of the collective phase coupling is finite at the onset of collective oscillations. This difference in the properties results from the difference of bifurcations. The fully phase-locked states emerge from saddle-node bifurcations as mentioned above, whereas the phase coherent states and partially phase-locked states emerge from supercritical Hopf bifurcations ref:kawamura10a ; ref:kawamura10b .

(iii) The third case is α=0\alpha=0, which yields a microscopic antisymmetric internal coupling function. For this case, η=Δω\eta=\Delta\omega. Substituting α=0\alpha=0 into Eq. (31), we obtain the following result:

α=0,ρeiδ=1+1η22eiβ.\alpha=0,\qquad\rho e^{i\delta}=\frac{1+\sqrt{1-\eta^{2}}}{2}e^{i\beta}. (35)

That is, the phase shift δ\delta of the collective phase coupling function is the same as the phase shift β\beta of the microscopic external phase coupling function.

(iv) The fourth cases are special values of β\beta. Substituting β=0\beta=0, ±π\pm\pi, ±π/2\pm\pi/2 into Eq. (31), we obtain the following results:

β=0,ρeiδ\displaystyle\beta=0,\qquad\rho e^{i\delta} =+12[1+1η2]+i2[η21η2tanα],\displaystyle=+\frac{1}{2}\left[1+\sqrt{1-\eta^{2}}\right]+\frac{i}{2}\left[\frac{\eta^{2}}{\sqrt{1-\eta^{2}}}\tan\alpha\right], (36)
β=±π,ρeiδ\displaystyle\beta=\pm\pi,\qquad\rho e^{i\delta} =12[1+1η2]i2[η21η2tanα],\displaystyle=-\frac{1}{2}\left[1+\sqrt{1-\eta^{2}}\right]-\frac{i}{2}\left[\frac{\eta^{2}}{\sqrt{1-\eta^{2}}}\tan\alpha\right], (37)
β=±π2,ρeiδ\displaystyle\beta=\pm\frac{\pi}{2},\qquad\rho e^{i\delta} =12[η21η2tanα]±i2[1+1η2].\displaystyle=\mp\frac{1}{2}\left[\frac{\eta^{2}}{\sqrt{1-\eta^{2}}}\tan\alpha\right]\pm\frac{i}{2}\left[1+\sqrt{1-\eta^{2}}\right]. (38)

For microscopic antisymmetric external coupling functions, i.e., β=0\beta=0, ±π\pm\pi, the type of the collective phase coupling function coincides with that of the microscopic external coupling function. In contrast, for microscopic symmetric external coupling functions, i.e., β=±π/2\beta=\pm\pi/2, the type of the collective phase coupling function is determined by the sign of the microscopic internal coupling parameter α\alpha.

(v) The fifth case is β=α\beta=\alpha, which indicates that the microscopic external coupling has the same phase shift as the microscopic internal one. Substituting β=α\beta=\alpha into Eq. (31), we obtain the following result:

β=α,ρeiδ=12[1+1η21η2cosαη21η21cosα]+i2[1+1η21η2sinα].\beta=\alpha,\qquad\rho e^{i\delta}=\frac{1}{2}\left[\frac{1+\sqrt{1-\eta^{2}}}{\sqrt{1-\eta^{2}}}\cos\alpha-\frac{\eta^{2}}{\sqrt{1-\eta^{2}}}\frac{1}{\cos\alpha}\right]+\frac{i}{2}\left[\frac{1+\sqrt{1-\eta^{2}}}{\sqrt{1-\eta^{2}}}\sin\alpha\right]. (39)

From the condition of |α|<π/2|\alpha|<\pi/2, both microscopic internal and external coupling functions are in-phase coupling. However, the type of the collective phase coupling function is anti-phase coupling under the following condition:

cos2α<η21+1η2.\cos^{2}\alpha<\frac{\eta^{2}}{1+\sqrt{1-\eta^{2}}}. (40)

For the case of |η|1|\eta|\to 1, the above condition becomes cos2α<1\cos^{2}\alpha<1, which is satisfied for all α\alpha except for α=0\alpha=0.

VI Collective phase synchronization between two interacting groups

Now, we study counter-intuitive cases under the condition of η=3/4\eta=3/4. The type of the collective phase coupling function is shown in Fig. 2, where the solid curves are determined by Eq. (32), i.e., ρcosδ=0\rho\cos\delta=0. Here, we note that the type of the collective phase coupling function can be different from that of the microscopic external phase coupling function. Two sets of parameters, which were used in Fig. 3, are also shown in Fig. 2.

Two groups of two-oscillators exhibiting phase-locked states were separately prepared with their corresponding phases being nearly identical. Then, these states were used as the initial condition in Fig. 3(a). In spite of the microscopic in-phase external coupling, β=3π/8\beta=3\pi/8, the external phase difference |ϕ1(1)ϕ1(2)||\phi_{1}^{(1)}-\phi_{1}^{(2)}| approached π\pi after some time; this indicates anti-phase collective synchronization between the groups. In contrast, Fig. 3(b) shows in-phase collective synchronization between the groups in spite of the microscopic anti-phase external coupling, β=5π/8\beta=-5\pi/8.

Finally, it should be noted that we can also consider interacting groups of weakly coupled Stuart-Landau oscillators as mentioned in App. A.

VII Concluding remarks

In this paper, we considered the phase synchronization between collective rhythms of fully locked oscillator groups, clarified the relation between the collective phase coupling and microscopic external phase coupling functions, analytically determined the type of the collective phase coupling function for weakly interacting groups of two oscillators with global sinusoidal coupling, and demonstrated that the groups can exhibit anti-phase (in-phase) collective synchronization in spite of microscopic in-phase (anti-phase) external coupling. The theoretical predictions were successfully confirmed by direct numerical simulations of the phase oscillator model and Stuart-Landau oscillator model.

In Refs. ref:kawamura10a ; ref:kawamura10b , we investigated the phase synchronization between collective rhythms of globally coupled oscillator groups under two typical situations: phase coherent states in the noisy identical case ref:kawamura10a and partially phase-locked states in the noiseless nonidentical case ref:kawamura10b . In particular, we found the counter-intuitive phenomena similar to the results in this paper. That is, weakly interacting groups can exhibit anti-phase collective synchronization in spite of microscopic in-phase external coupling and vice versa. Here, we note that these three papers considered different physical situations and utilized different mathematical methods, but arrived at the similar counter-intuitive phenomena.

We also remark that fully phase-locked states emerge from a finite number of oscillators ref:izhikevich07 ; ref:ermentrout10 ; even two is possible as actually studied in this paper. In contrast, phase coherent states and partially phase-locked states emerge from a large population of oscillators ref:kuramoto84 ; the number of oscillators is infinite in theory. From this point of view, fully phase-locked states can be more easily realized in experiments such as electrochemical oscillators ref:kiss02 ; ref:kiss07 , discrete chemical oscillators ref:taylor09 ; ref:tinsley12 , and mechanical oscillators ref:pantaleone02 ; ref:martens13 . We hope that the counter-intuitive phenomena studied in this paper, i.e., effective anti-phase (in-phase) collective synchronization with microscopic in-phase (anti-phase) external coupling, will be experimentally confirmed in the near future and that the formula (31) will help in such experiments.

Acknowledgements.
The author is grateful to Yoshiki Kuramoto, Hiroya Nakao, Hiroshi Kori, and Kensuke Arai for fruitful discussions. This work was supported by JSPS KAKENHI Grant Number 25800222.

Appendix A Interacting groups of weakly coupled Stuart-Landau oscillators

In this appendix, we consider interacting groups of globally coupled Stuart-Landau oscillators described by the following equation:

W˙j(σ)(t)=(1+ibj)Wj(σ)(1+ic2)|Wj(σ)|2Wj(σ)+K(1+ic1)Nk=1NWk(σ)+ϵJ(1+ic3)Nk=1NWk(τ),\dot{W}_{j}^{(\sigma)}(t)=(1+ib_{j})W_{j}^{(\sigma)}-(1+ic_{2})\left|W_{j}^{(\sigma)}\right|^{2}W_{j}^{(\sigma)}+\frac{K(1+ic_{1})}{N}\sum_{k=1}^{N}W_{k}^{(\sigma)}+\frac{\epsilon J(1+ic_{3})}{N}\sum_{k=1}^{N}W_{k}^{(\tau)}, (41)

for j=1,,Nj=1,\cdots,N and (σ,τ)=(1,2),(2,1)(\sigma,\tau)=(1,2),(2,1), where Wj(σ)(t)1W_{j}^{(\sigma)}(t)\in\mathbb{C}^{1} is the complex amplitude of the jj-th limit-cycle oscillator at time tt in the σ\sigma-th group consisting of NN oscillators. The first and second terms on the right-hand side represent the intrinsic dynamics of each oscillator, the third term represents the microscopic internal coupling within the same group, and the fourth term represents the microscopic external coupling between the different groups. When the internal and external couplings are sufficiently weak compared to the absolute value of the amplitude Floquet exponent, we can approximately derive a phase equation in the following form ref:kuramoto84 :

ϕ˙j(σ)(t)=ωjPKNk=1Nsin(ϕj(σ)ϕk(σ)+α)ϵPJNk=1Nsin(ϕj(σ)ϕk(τ)+β),\dot{\phi}_{j}^{(\sigma)}(t)=\omega_{j}-\frac{P_{K}}{N}\sum_{k=1}^{N}\sin\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\sigma)}+\alpha\right)-\frac{\epsilon P_{J}}{N}\sum_{k=1}^{N}\sin\left(\phi_{j}^{(\sigma)}-\phi_{k}^{(\tau)}+\beta\right), (42)

where the parameters of phase oscillators are given by

ωj\displaystyle\omega_{j} =bjc2,\displaystyle=b_{j}-c_{2}, (43)
PKeiα\displaystyle P_{K}\,e^{i\alpha} =K(1+ic2)(1ic1),\displaystyle=K(1+ic_{2})(1-ic_{1}), (44)
PJeiβ\displaystyle P_{J}\,e^{i\beta} =J(1+ic2)(1ic3).\displaystyle=J(1+ic_{2})(1-ic_{3}). (45)

The phase of each Stuart-Landau oscillator is given by the following equation ref:winfree80 ; ref:kuramoto84 ; ref:pikovsky01 ; ref:izhikevich07 ; ref:ermentrout10 ; ref:schultheiss12 : ϕ=argWc2ln|W|\phi=\arg W-c_{2}\ln|W|. As in the main text, we focus on the case in which the number of oscillators within each group is two, i.e., N=2N=2. Using the following constants, r=0.01r=0.01 and a=3π/8a=3\pi/8, the parameters of the Stuart-Landau oscillators are fixed at K=J=rcos(a)K=J=r\cos(a), c1=c3=0c_{1}=c_{3}=0, c2=tan(a)c_{2}=\tan(a), b1=c2+3rcos(a)/4b_{1}=c_{2}+3r\cos(a)/4, and b2=c2b_{2}=c_{2}. Under these conditions, the parameters of the phase oscillators are obtained as PK=PJ=r=0.01P_{K}=P_{J}=r=0.01, α=β=a=3π/8\alpha=\beta=a=3\pi/8, ω1=3rcos(a)/4\omega_{1}=3r\cos(a)/4, and ω2=0\omega_{2}=0, which correspond to the parameters in Fig. 3(a). In particular, we note that η=(Δω)/(PKcosα)=(3rcos(a)/4)/(rcos(a))=3/4\eta=(\Delta\omega)/(P_{K}\cos\alpha)=(3r\cos(a)/4)/(r\cos(a))=3/4. The external coupling intensity is fixed at ϵ=0.001\epsilon=0.001. The direct numerical simulation result of Eq. (41) is shown in Fig. 4. Similarly to Fig. 3(a), Fig. 4 shows anti-phase collective synchronization between the groups in spite of the microscopic in-phase external coupling.

References

  • (1) A. T. Winfree, The Geometry of Biological Time (Springer, New York, 1980; Springer, Second Edition, New York, 2001).
  • (2) Y. Kuramoto, Chemical Oscillations, Waves, and Turbulence (Springer, New York, 1984; Dover, New York, 2003).
  • (3) A. Pikovsky, M. Rosenblum, and J. Kurths, Synchronization: A Universal Concept in Nonlinear Sciences (Cambridge University Press, Cambridge, 2001).
  • (4) S. H. Strogatz, Sync: How Order Emerges from Chaos in the Universe, Nature, and Daily Life (Hyperion Books, New York, 2003).
  • (5) S. C. Manrubia, A. S. Mikhailov, and D. H. Zanette, Emergence of Dynamical Order: Synchronization Phenomena in Complex Systems (World Scientific, Singapore, 2004).
  • (6) G. V. Osipov, J. Kurths, and C. Zhou, Synchronization in Oscillatory Networks (Springer, New York, 2007).
  • (7) A. S. Mikhailov and G. Ertl (Editors), Engineering of Chemical Complexity (World Scientific, Singapore, 2013).
  • (8) F. C. Hoppensteadt and E. M. Izhikevich, Weakly Connected Neural Networks (Springer, New York, 1997).
  • (9) E. M. Izhikevich, Dynamical Systems in Neuroscience: The Geometry of Excitability and Bursting (MIT Press, Cambridge, MA, 2007).
  • (10) G. B. Ermentrout and D. H. Terman, Mathematical Foundations of Neuroscience (Springer, New York, 2010).
  • (11) N. Schultheiss, R. Butera, and A. Prinz (Editors), Phase Response Curves in Neuroscience: Theory, Experiment, and Analysis (Springer, New York, 2012).
  • (12) S. H. Strogatz, Physica D 143, 1 (2000).
  • (13) J. A. Acebrón, L. L. Bonilla, C. J. Pérez Vicente, F. Ritort, and R. Spigler, Rev. Mod. Phys. 77, 137 (2005).
  • (14) S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D.-U. Hwang, Phys. Rep. 424, 175 (2006).
  • (15) A. Arenas, A. Díaz-Guilera, J. Kurths, Y. Moreno, and C. Zhou, Phys. Rep. 469, 93 (2008).
  • (16) S. N. Dorogovtsev, A. V. Goltsev, and J. F. F. Mendes, Rev. Mod. Phys. 80, 1275 (2008).
  • (17) A. Barrat, M. Barthelemy, and A. Vespignani, Dynamical Processes on Complex Networks (Cambridge University Press, Cambridge, 2008).
  • (18) K. Okuda and Y. Kuramoto, Prog. Theor. Phys. 86, 1159 (1991).
  • (19) E. Montbrió, J. Kurths, and B. Blasius, Phys. Rev. E 70, 056125 (2004).
  • (20) D. M. Abrams, R. E. Mirollo, S. H. Strogatz, and D. A. Wiley, Phys. Rev. Lett. 101, 084103 (2008).
  • (21) E. Barreto, B. Hunt, E. Ott, and P. So, Phys. Rev. E 77, 036107 (2008).
  • (22) J. H. Sheeba, V. K. Chandrasekar, A. Stefanovska, and P. V. E. McClintock, Phys. Rev. E 78, 025201(R) (2008).
  • (23) J. H. Sheeba, V. K. Chandrasekar, A. Stefanovska, and P. V. E. McClintock, Phys. Rev. E 79, 046210 (2009).
  • (24) C. R. Laing, Chaos 19, 013113 (2009).
  • (25) P. S. Skardal and J. G. Restrepo, Phys. Rev. E 85, 016208 (2012).
  • (26) D. Anderson, A. Tenzer, G. Barlev, M. Girvan, T. M. Antonsen, and E. Ott, Chaos 22, 013102 (2012).
  • (27) C. R. Laing, Chaos 22, 043104 (2012).
  • (28) E. Ott and T. M. Antonsen, Chaos 18, 037113 (2008).
  • (29) E. Ott and T. M. Antonsen, Chaos 19, 023117 (2009).
  • (30) E. Ott, B. R. Hunt, and T. M. Antonsen, Chaos 21, 025112 (2011).
  • (31) I. Z. Kiss, Y. Zhai, and J. L. Hudson, Science 296, 1676 (2002).
  • (32) I. Z. Kiss, C. G. Rusin, H. Kori, and J. L. Hudson, Science 316, 1886 (2007).
  • (33) A. F. Taylor, M. R. Tinsley, F. Wang, Z. Huang, and K. Showalter, Science 323, 614 (2009).
  • (34) M. R. Tinsley, S. Nkomo, and K. Showalter, Nature Physics 8, 662 (2012).
  • (35) J. Pantaleone, Am. J. Phys. 70, 992 (2002).
  • (36) E. A. Martens, S. Thutupalli, A. Fourriére, and O. Hallatschek, Proc. Natl. Acad. Sci. USA 110, 10563 (2013).
  • (37) Y. Kawamura, H. Nakao, K. Arai, H. Kori, and Y. Kuramoto, Phys. Rev. Lett. 101, 024101 (2008). [arXiv:0807.1285]
  • (38) Y. Kawamura, Physica D 270, 20 (2014). [arXiv:1312.7054]
  • (39) Y. Kawamura, H. Nakao, K. Arai, H. Kori, and Y. Kuramoto, Chaos 20, 043109 (2010). [arXiv:1007.4382]
  • (40) Y. Kawamura, H. Nakao, K. Arai, H. Kori, and Y. Kuramoto, Chaos 20, 043110 (2010). [arXiv:1007.5161]
  • (41) H. Kori, Y. Kawamura, H. Nakao, K. Arai, and Y. Kuramoto, Phys. Rev. E 80, 036207 (2009). [arXiv:0812.0118]
  • (42) N. Masuda, Y. Kawamura, and H. Kori, New J. Phys. 11, 113002 (2009). [arXiv:0907.0900]
  • (43) N. Masuda, Y. Kawamura, and H. Kori, Phys. Rev. E 80, 046114 (2009). [arXiv:0907.0901]
  • (44) N. Masuda, Y. Kawamura, and H. Kori, New J. Phys. 12, 093007 (2010). [arXiv:0911.5013]
  • (45) T.-W. Ko and G. B. Ermentrout, Phys. Rev. E 79, 016211 (2009).
  • (46) R. Tönjes and B. Blasius, Phys. Rev. E 80, 026202 (2009).
  • (47) M. C. Cross, Phys. Rev. E 85, 046214 (2012).
  • (48) J.-M. A. Allen and M. C. Cross, Phys. Rev. E 87, 052902 (2013).
  • (49) G. B. Ermentrout, SIAM J. Appl. Math. 52, 1665 (1992).
  • (50) R. E. Mirollo and S. H. Strogatz, Physica D 205, 249 (2005).
  • (51) N. Biggs, Bull. London Math. Soc. 29, 641 (1997).
  • (52) R. P. Agaev and P. Yu. Chebotarev, Autom. Remote Control 61, 1424 (2000).
Refer to caption
Figure 1: (Color online) Schematic diagram of two weakly interacting groups of two oscillators with global coupling. The microscopic internal and external couplings are represented by the solid and dotted arrows, respectively, whereas the self-coupling is not shown. The phase of the jj-th oscillator in the σ\sigma-th group is denoted by ϕj(σ)\phi_{j}^{(\sigma)}.
Refer to caption
Figure 2: (Color online) Effective type of phase coupling between collective rhythms of fully locked oscillator groups with α(π/2,π/2)\alpha\in(-\pi/2,\pi/2), β[π,π]\beta\in[-\pi,\pi], and η=3/4\eta=3/4. The solid curves are analytically determined by Eq. (32), i.e., ρcosδ=0\rho\cos\delta=0. The filled circle (\bullet) indicates α=β=3π/8\alpha=\beta=3\pi/8 corresponding to Fig. 3(a) and Fig. 4. The times sign (×\times) indicates α=3π/8\alpha=3\pi/8 and β=5π/8\beta=-5\pi/8 corresponding to Fig. 3(b).
Refer to caption
Figure 3: (Color online) Interacting groups of phase oscillators. Time evolution of the internal and external phase differences, i.e., |ϕ1(1)ϕ2(1)||\phi_{1}^{(1)}-\phi_{2}^{(1)}| and |ϕ1(1)ϕ1(2)||\phi_{1}^{(1)}-\phi_{1}^{(2)}|. The other internal and external phase differences are approximated as |ϕ1(2)ϕ2(2)||ϕ1(1)ϕ2(1)||\phi_{1}^{(2)}-\phi_{2}^{(2)}|\simeq|\phi_{1}^{(1)}-\phi_{2}^{(1)}| and |ϕ2(1)ϕ2(2)||ϕ1(1)ϕ1(2)||\phi_{2}^{(1)}-\phi_{2}^{(2)}|\simeq|\phi_{1}^{(1)}-\phi_{1}^{(2)}|, respectively. The collective phase difference is approximated as the external phase difference, i.e., |Θ(1)Θ(2)||ϕ1(1)ϕ1(2)||\Theta^{(1)}-\Theta^{(2)}|\simeq|\phi_{1}^{(1)}-\phi_{1}^{(2)}|. The parameters are α=3π/8\alpha=3\pi/8, ω1=3cos(α)/4\omega_{1}=3\cos(\alpha)/4, ω2=0\omega_{2}=0, and ϵ=0.001\epsilon=0.001. (a) Effective anti-phase collective synchronization with microscopic in-phase external coupling, β=3π/8\beta=3\pi/8. (b) Effective in-phase collective synchronization with microscopic anti-phase external coupling, β=5π/8\beta=-5\pi/8.
Refer to caption
Figure 4: (Color online) Interacting groups of weakly coupled Stuart-Landau (SL) oscillators. Effective anti-phase collective synchronization with microscopic in-phase external coupling. The parameters are K=J=rcos(a)K=J=r\cos(a), c1=c3=0c_{1}=c_{3}=0, c2=tan(a)c_{2}=\tan(a), b1=c2+3rcos(a)/4b_{1}=c_{2}+3r\cos(a)/4, b2=c2b_{2}=c_{2}, and ϵ=0.001\epsilon=0.001, where r=0.01r=0.01 and a=3π/8a=3\pi/8. (a) Time evolution of the internal and external phase differences, i.e., |ϕ1(1)ϕ2(1)||\phi_{1}^{(1)}-\phi_{2}^{(1)}| and |ϕ1(1)ϕ1(2)||\phi_{1}^{(1)}-\phi_{1}^{(2)}|. (b) Snapshot of the asymptotic state of individual oscillators, i.e., W1(1)W_{1}^{(1)}, W2(1)W_{2}^{(1)}, W1(2)W_{1}^{(2)}, and W2(2)W_{2}^{(2)}.