Loading [MathJax]/jax/output/SVG/jax.js
Research article

On Kuramoto-Sakaguchi-type Fokker-Planck equation with delay

  • Received: 01 July 2023 Revised: 16 November 2023 Accepted: 30 November 2023 Published: 08 December 2023
  • Recently, the Kuramoto model with transmission delay has been attracting increasing attention, accompanied by the increase in its practical applications. In this paper, we studied the Kuramoto-Sakaguchi-type Fokker-Planck equation of the above model proposed by Lee et al., in 2009. We proved the global-in-time solvability of the equation under some conditions on the initial data and distribution of delay.

    Citation: Hirotada Honda. On Kuramoto-Sakaguchi-type Fokker-Planck equation with delay[J]. Networks and Heterogeneous Media, 2024, 19(1): 1-23. doi: 10.3934/nhm.2024001

    Related Papers:

    [1] Hirotada Honda . Global-in-time solution and stability of Kuramoto-Sakaguchi equation under non-local Coupling. Networks and Heterogeneous Media, 2017, 12(1): 25-57. doi: 10.3934/nhm.2017002
    [2] Karoline Disser, Matthias Liero . On gradient structures for Markov chains and the passage to Wasserstein gradient flows. Networks and Heterogeneous Media, 2015, 10(2): 233-253. doi: 10.3934/nhm.2015.10.233
    [3] Michael Herty, Lorenzo Pareschi, Sonja Steffensen . Mean--field control and Riccati equations. Networks and Heterogeneous Media, 2015, 10(3): 699-715. doi: 10.3934/nhm.2015.10.699
    [4] Seung-Yeal Ha, Hansol Park, Yinglong Zhang . Nonlinear stability of stationary solutions to the Kuramoto-Sakaguchi equation with frustration. Networks and Heterogeneous Media, 2020, 15(3): 427-461. doi: 10.3934/nhm.2020026
    [5] Ioannis Markou . Hydrodynamic limit for a Fokker-Planck equation with coefficients in Sobolev spaces. Networks and Heterogeneous Media, 2017, 12(4): 683-705. doi: 10.3934/nhm.2017028
    [6] Diandian Huang, Xin Huang, Tingting Qin, Yongtao Zhou . A transformed $ L1 $ Legendre-Galerkin spectral method for time fractional Fokker-Planck equations. Networks and Heterogeneous Media, 2023, 18(2): 799-812. doi: 10.3934/nhm.2023034
    [7] Seung-Yeal Ha, Jeongho Kim, Jinyeong Park, Xiongtao Zhang . Uniform stability and mean-field limit for the augmented Kuramoto model. Networks and Heterogeneous Media, 2018, 13(2): 297-322. doi: 10.3934/nhm.2018013
    [8] Luis Almeida, Federica Bubba, Benoît Perthame, Camille Pouchol . Energy and implicit discretization of the Fokker-Planck and Keller-Segel type equations. Networks and Heterogeneous Media, 2019, 14(1): 23-41. doi: 10.3934/nhm.2019002
    [9] Yves Achdou, Victor Perez . Iterative strategies for solving linearized discrete mean field games systems. Networks and Heterogeneous Media, 2012, 7(2): 197-217. doi: 10.3934/nhm.2012.7.197
    [10] Seung-Yeal Ha, Yongduck Kim, Zhuchun Li . Asymptotic synchronous behavior of Kuramoto type models with frustrations. Networks and Heterogeneous Media, 2014, 9(1): 33-64. doi: 10.3934/nhm.2014.9.33
  • Recently, the Kuramoto model with transmission delay has been attracting increasing attention, accompanied by the increase in its practical applications. In this paper, we studied the Kuramoto-Sakaguchi-type Fokker-Planck equation of the above model proposed by Lee et al., in 2009. We proved the global-in-time solvability of the equation under some conditions on the initial data and distribution of delay.



    The study of synchronization phenomena has a rather long history. One of the most sophisticated formulations is the Kuramoto model, which describes the temporal behavior of oscillator phases. A range of models related to the Kuramoto model, which offers vast applications, have been considered and discussed to date.

    An important factor in practical applications is the consideration of transmission delay, as pointed out in past arguments (see, for instance, [23]). For example, when discussing the neural networks of the human brain, the effect of the transmission delay of the synaptic propagation cannot be disregarded.

    A general method among the models that takes into account the effect of transmission delay is that proposed by Lee et al. [17] because it approaches delay as a random variable that is subjected to a certain probability density. The discussion in this work [17] begins with a system of ordinary differential equations with a delay imposed on each pair of oscillators.

    ddtθi(t)=ωi+KNNj=1sin[θj(tτij)θi(t)](i=1,2,,N), (1.1)

    where θi(t) is the phase of oscillator i, ωi is the natural frequency of i, K is the coupling strength between oscillators, and N is the total number of oscillators. τij, which denotes the transmission delay on the link between oscillators i and j, satisfies τii=0(i=1,2,,N).

    As in the original Kuramoto model, Lee et al. [17] also introduced a measure of phase synchronization known as the order parameter:

    N1Nj=1sin[θj(tτij)θi(t)]=Im[rieiθi(t)], (1.2)
    ri(t)=N1Nj=1eiθj(tτij). (1.3)

    Hereafter, we use the notation for imaginary units i=1. Using Eqs (1.2) and (1.3), we can rewrite Eq (1.1) as a simpler system of ordinary differential equations, then by adding further white noise, we consider

    ddtθi(t)=ωi+K2i(ri(t)eiθi(t)ˉri(t)eiθi(t))+ξi(t), (1.4)

    where we use the fact that Im[z]=zˉz2i for zC. Here, ˉz denotes the complex conjugate of z. We also used the notation {ξi(t)}Ni=1, which are the independent Wiener processes satisfying ξi(t)=0 and ξj(t)ξk(τ)=2εδjkδ(tτ), in which δ() denotes the Dirac measure. Then, as a continuum limit of the infinite oscillator population N+, the Fokker-Planck equation of (1.4) as a limit of ε0 [6], which describes the temporal behavior of the probability density of the oscillators, can be obtained:

    ft+θ{[ω+k2i(eiθreiθˉr)]f}=0, (1.5)

    where

    r(t)=0ξ(tτ)h(τ)dτ,ξ(t)=2π0f(θ,t;ω)eiθdθdω. (1.6)

    Here, the unknown function f=f(θ,t;ω) is the distribution function of the single oscillator and h(τ) is that of the delay.

    In this study, we rigorously examine the well-posedness of models (1.5) and (1.6) under suitable initial and boundary conditions, and discuss the following:

    (ⅰ) We propose a Fokker-Planck equation [19] corresponding to the model developed by Lee et al. with additional noise.

    (ⅱ) We demonstrate the existence and uniqueness of a global-in-time solution in a suitable function space for the aforementioned model.

    Some difficulties are encountered owing to the delay τij in Eq (1.2) and its density h() in Eq (1.6). We demonstrate that, when using a method developed in the study of fluid mechanics, we can obtain an a-prioriestimate of the solution and its global existence. The novel contributions of this paper are summarized as follows.

    (ⅰ) We consider the uniqueness and existence of a global-in-time solution to the parabolic equation that corresponds to the formulation of Lee et al. [17]. To the best of our knowledge, this is the first study that rigorously discusses their formulation.

    (ⅱ) Unlike other models, our formulation includes the random delay, which causes some difficulties. We overcome this issue with the aid of the approach used in the mathematical analyses of fluid mechanics.

    The remainder of this paper is organized as follows. In the following section, we discuss the central objective of our analysis. The problem formulation is outlined in Section 2. Section 3 provides an overview of existing mathematical arguments. Section 4 introduces the notation used throughout this study. The main results of this study are presented in Section 5, followed by proofs in Sections 6 and 7. In Section 8, we discuss other properties of the solution. Section 9 concludes the paper.

    In this section, we discuss the problems identified in this study. For simplicity, hereafter, we use notations Ω=(0,2π), ΩT=Ω×(0,T) with an arbitrary T(0,+], and R+=(0,+). By substituting Eqs (1.5) and (1.6) into Eq (1.4) and imposing the periodic boundary conditions and initial condition, we obtain

    {ft+ωfθ+Kθ(F[f,f])=0inΩ×R,f(j,0)(0,t;ω)=f(j,0)(2π,t;ω)(j=0,1)onR+×R,f(θ,t;ω)=f0(θ;ω)onΩ×(,0]×R. (2.1)

    where

    F[f1,f2]f1(θ,t;ω)0h(τ)dτRdωf2(θ,tτ;ω)sin(θθ)dθ.

    Note that in Eq (2.1), the initial condition is provided on the interval (,0] with respect to t owing to the presence of a delay. In addition, in Eq (2.1), the initial condition is replaced by that over the half-infinite interval (,0] owing to the existence of a delay. Instead of Eq (2.1), we first consider the parabolic regularization:

    {ft+ωfθε2fθ2+Kθ(F[f,f])=0inΩ×R,f(j,0)(0,t;ω)=f(j,0)(2π,t;ω)(j=0,1)onR+×R,f(θ,0;ω)=f0(θ;ω)onΩ×(,0]×R. (2.2)

    Note that ˉfg(ω)/2π is a trivial stationary solution to Eq (2.2), where g(ω) is the probability density of the natural frequency ω. Moreover, based on appropriate assumptions for f0 and h, the following properties of f are derived (see, for instance, Lemma 2.1 in [11], Lemmas 1.1 and 1.2 in [24], and Lemma 8.1 of this paper):

    f(θ,t;ω)0.

    However, in the case of the typical Kuramoto–Sakaguchi equation [10], the unknown phase probability density satisfies 2π0f(θ,t;ω)=1, whereas for Eq (2.2), f satisfies 2π0f(θ,t;ω)dθ=const, and Rdω2π0f(θ,t;ω)dθ=1. Therefore, we find a suitable decay rate must be imposed with respect to ω on the initial data f0, which is conserved at all times.

    Similar to case of the original Kuramoto–Sakaguchi equation [9], we introduce the transformation of coordinates as

    θ=θ+ωt,t=t,

    and

    f(θ,t;ω)=f(θ+ωt,t;ω)=ˆf(θ,t;ω),

    then, ˆf satisfies

    {ˆftε2ˆfθ2+Kθ(F[ˆf,ˆf])=0,inΩ×R,ˆf(j,0)(0,t;ω)=ˆf(j,0)(2π,t;ω)(j=0,1)onR+×R,ˆf(θ,0;ω)=f0(θ;ω)onΩ×(,0]×R. (2.3)

    Next, we introduce a weight function with δ>1/2:

    w(δ)(ω)={1+|ω|1/2+δ(|ω|>1),2(|ω|1).

    After multiplying w(δ)(ω) and Eq (2.3), we introduce ˜f(θ,t;ω)w(δ)(ω)ˆf(θ,t;ω). This satisfies

    {˜ftε2˜fθ2+Kθ(F[˜f,˜fw(δ)])=0inΩ×R,˜f(j,0)(0,t;ω)=˜f(j,0)(2π,t;ω)(j=0,1)onR+×R,˜f(θ,0;ω)=˜f0(θ;ω)w(δ)(ω)f0(θ;ω)onΩ×(,0]×R. (2.4)

    In this study, we discuss the solvability of Eq (2.4).

    Mathematical arguments concerning the solvability of the Kuramoto–Sakaguchi equation, which corresponds to the original Kuramoto model, were first presented by Lavrentiev et al. [15,16]. In their earlier work [15], they constructed a classical global-in-time solution wherein the support of g(ω) is compact.

    They later removed this restriction [16] by applying a priori estimates derived from the energy method. They also studied the regularity of an unknown function with respect to ω.

    In terms of stability, pioneering work was conducted by Strogatz and Mirollo [21] who focused on the linear stability of the trivial stationary solution ˉϱ=1/2π. By investigating the spectrum of the linearized operator, they verified the existence of a critical coupling strength, over which the coherent state became stable.

    Remarkably, Ha and Xiao [7] discussed the nonlinear stability of ˉϱ and the convergence of the solution as D tends to zero. Their estimate was obtained in the space L with respect to θ. They also verified the instability of ˉϱ when the support from g(ω) is sufficiently narrow [8].

    In the case of vanishing diffusion (2.1), Chiba [4] argued for the nonlinear stability of a trivial stationary solution under the assumption of unbounded support for g(ω).

    Several contributions relating to the Kuramoto model with delay were made prior to the work by Lee et al. [17]. To the best of our knowledge, one of the earliest works in this direction was presented by Niebur et al. [18]. They investigated the behavior of a system with two oscillators and found that the collective frequency decreases with the increase in the delay.

    Subsequently, Kim et al. [12] discussed a Kuramoto model with a constant delay and pinning forces. Based on thorough numerical computations, they clarified the existence of a multi-stable state that depends on the delay and the coefficient of the pinning forces (in fact, the coexistence of coherent and incoherent states had also been reported in oscillator networks under nonlocal coupling [13]). In addition, they verified that the system may exhibit a hysteresis of the collective frequency when the coefficient of the pinning forces increases.

    Their motivation was the fact that the behavior of a neural network in the human brain is also significantly affected by the presence of delay [3]. Thereafter, Yeung and Strogatz [23] formulated a Kuramoto model with a constant delay τ, and analyzed the stability of coherent and incoherent states. They demonstrated that the incoherence becomes stable within certain regions of K and τ. They also showed that coherence and incoherence coexist in certain Kτ regions.

    This work was continued by Choi et al. [5], who derived the Fokker–Planck equation and its solution that corresponds to the coherent state. For each phase ϕi of each oscillator, they began with the transform ψi=ϕiΩt with a constant Ω. Subsequently, they derived the approximate solution using the asymptotic expansion of Ω under the assumption ΩK1, where is the magnitude of the order parameter. The restriction in their model was the convexity on the distribution g(ω) of natural frequencies at ω=0; that is, g(0)<0.

    As in the original Kuramoto model, the shape of the natural frequency density drastically affects the results, as reported by Strogatz and Mirillo [21]. Moreover, motivated by the study of the human brain, several recent works have investigated the pattern of the behavior of the oscillators under the presence of heterogeneous transmission delay and network structure [2,24].

    However, none of these studies considered models with diffusion. In addition, to the best of our knowledge, no mathematical arguments have been made regarding the solvability of the Kuramoto model with delays. In this regard, the novelty of this paper is characterized as follows. First, we consider the Kuramoto–Sakaguchi type Fokker-Planck equation with transmission delay. Second, by using the approaches developed in mathematical fluid analysis, we establish the global-in-time solvability of the equation. It is noteworthy that under the presence of a transmission delay, the standard method used in our previous works based on the energy method cannot be applied in the present case.

    The functional spaces used throughout this study are described in this subsection. Let T>0 and G be an open set in R. Hereafter, L2(G) represents a set of square-integrable functions defined on G, equipped with the norm

    uG|u(x)|2dx.

    The inner product is defined as follows:

    (u1,u2)Gu1(x)¯u2(x)dx,

    where ˉz denotes the complex conjugate of zC. We denote the L2-norm of the function f(θ,t;ω) with respect to θ as fL2(Ω). We use C(G) and Ck(G)(kN{+}) to denote the spaces of real continuous and k-times continuously differentiable functions on Ω, respectively. The notation C0(G) refers to the set of C(G) functions with compact support in G.

    For a Banach space E with norm E, we denote the space of E-valued measurable functions u(t) in the interval (a,b) as Lp(a,b;E), the norm of which is defined by

    uLp(a,b;E){(bau(t)pEdt)1/p(1p<),esssupatbu(t)Ep=.

    Similarly, we denote the space of continuous functions as C(a,b;E) (and Ck(a,b;E)), (resp. k continuously differentiable functions) from (a,b) to E. Wl2(G) denotes a space of functions u(θ),θG equipped with the norm u2Wl2(G)=|α|<lαuθα2L2(G)+u2˙Wl2(G), where

    {u2˙Wl2(G)=|α|=lαuθα2L2(G)=|α|=lG|αuθα|2dθif l is an integer,u2˙Wl2(G)=|α|=[l]GG|αu(x)θααu(y)θα|2|xy|n+2{l}dxdyif l is a non-integer,l=[l]+{l},0<{l}<1.

    Subsequently, we introduce anisotropic Sobolev–Slobodetskiĭ spaces [22] Wl,l22(GT1)Wl,02(GT1)W0,l22(GT1) (GT1G×(0,T1)), the norms of which are defined as

    u2Wl,l22(GT1)=T10u(,t)2Wl2(G)dt+Gu(x,)2Wl22(0,T1)dxu2Wl,02(GT1)+u2W0,l22(GT1).

    The set of functions with vanishing initial data Wr,r22(GT) is defined as [14]:

    Wr,r22(GT)={fWr,r22(GT)|kftk|t=0=0(k=0,1,2,,[r2])}.

    The norms of these spaces are denoted by Wr,r22(GT), etc. We define the function spaces for m>0 as

    VmT{u(θ,t;ω)Wm,m/22(ΩT)||||u|||(m,m2);TsupωRu(ω)2Wm,m22(ΩT)<+},VmT{u(θ,t;ω)Wm,m/22(ΩT)||||u|||(m,m2);T<+},Vm{u(θ;ω)Wm2(Ω)||||u|||(m)supωRu(ω)2Wm2(Ω)<+}.

    For brevity, we use the notation |||u|||2=supωRu(ω)2L2(Ω), |||u|||(0,0);T=|||u|||T with an arbitrary T(0,+]. Finally, we introduce the following notations:

    L(1)1(Ω){u(;ω)L1(Ω)|u0,Ωu(θ;ω)dθ=1ωR },L(1)1(R){u(ω)L1(R)Ru(ω)dω=1},L(1)1(T){u(,t;ω)L1(Ω)|u0,2π0u(θ,t;ω)dθc41/(1+|ω|2+δ)t(0,T),ωR },

    where T>0 denotes an arbitrary number.

    Hereafter, c with suffixes represents constants in the approximation of certain quantities. We also denote c(t) using suffixes depending on t. Furthermore, we use the notation u(j,k)(θ)j(t)ku(j,k=0,1,2,) for the function u=u(θ,t). We also use the notation

    F(1)[f1,f2]f1(θ,t;ω)0h(τ)dτRdωf2(θ,tτ;ω)cos(θθ)dθ.

    The following subsections present our main results. For the proof, we discuss the estimate of f in Eq (2.4) for the local-in-time (Theorem 5.1) and global-in-time (Theorem 5.2) solvabilities. We estimate ˜f=fg(ω)2π. First, we state the existence and uniqueness of the local-in-time solution to problem (2.4).

    Theorem 5.1. Let us assume that l(1/2,1), T>0, and the following hold true:

    (ⅰ) The natural frequency ω follows a distribution with probability density function g(ω), which satisfies gL(1)1(R)L(R).

    (ⅱ) f0V2+lL(1)1(Ω), and 2π0f0(θ;ω)dθc51/(1+|ω|2+δ) with a certain constant c51>0 and δ(0,1).

    (ⅲ) hL2(R+)L(R+).

    Then, there exists a certain T(0,T] and solution ˜f(θ,t;ω) to Eq (2.4) (and consequently, Eq (2.3)) on (0,T) such that

    ˜fV3+l(T).

    Based on Theorem 5.1, we obtain the global-in-time solvability of Eq (2.4).

    Theorem 5.2. In addition to the assumptions in Theorem 5.1, we assume that

    (ⅰ) |||˜f0ˉf|||2+lδ1 with sufficiently small δ1>0;

    (ⅱ) h(0)=h(0)=0, where h(t) is the first derivative of h;

    Then, there exists a solution f to Eq (2.4)(and consequently, Eq (2.3)) that satisfies ektf(θ,t;ω)V3+l for a certain k>0.

    In this section, we examine the solvability of Equation (2.4) (and consequently, (2.3)). We first discuss the local-in-time solvability of Eq (2.2); that is, the proof of Theorem 5.1: We construct the successive approximation of Eq (2.4), {˜f(m)}m=0, which is defined for m=0 by ˜f(0)=˜f0 and m1 using

    {˜f(m)tε2˜f(m)θ2=Kθ(F[˜f(m1),˜f(m1)w(δ)])inΩ×R,˜f(j,0)(m)(0,t;ω)=˜f(j,0)(m)(2π,t;ω)(j=0,1)onR+×R,˜f(m)(θ,0;ω)=˜f0(θ;ω)onΩ×(,0]×R. (6.1)

    For T>0, we introduce the following notation for m1:

    E(m)(T;ω)˜f(m)(;ω)W3+l,3+l22(ΩT),K(m)(T)|||˜f(m)|||2(3+l,3+l2);T.

    We also define

    E(0)(ω)˜f0(;ω)W2+l2(Ω),K(0)supωR˜f0(;ω)2W2+l2(Ω)=|||˜f0|||2(2+l).

    Owing to the elementary results for the initial and boundary value problems of the heat equation [14], we observe

    E(m)(T;ω)c61{θ(F[˜f(m1),˜f(m1)w(δ)])W1+l,1+l22(ΩT)}+E0(ω)(ωR). (6.2)

    At this point, we state the following:

    Lemma 6.1. Let T>0 and assume that l(1/2,1) and α(0,1). then for uW2+l,2+l22(ΩT), we obtain the following estimates.

    0h(τ)dτRdωw(δ)(ω)Ωu(θ,tτ;ω)sin(θθ)dθW2+l,2+l22(ΩT)c62ϕ(T)hW1+α2(0,T)(|||u|||T+|||u(0,1)|||T), (6.3)

    where ϕ(T) is a polynomial of its argument with a degree not exceeding (3l)/2, and c62>0 is a constant that is independent of T.

    Proof. For simplicity, we introduce the following notation:

    S(θ,t)0h(τ)dτRdωw(δ)(ω)Ωu(ϕ,tτ;ω)sin(ϕθ)dϕ.

    We first estimate S(,t)2L2(Ω) for each t. By an elementary variable change, it can be observed that

    |0h(τ)dτRdωw(δ)(ω)Ωu(ϕ,tτ;ω)sin(ϕθ)dϕ|2=|t0h(tt)dτRdωw(δ)(ω)Ωu(ϕ,t;ω)sin(ϕθ)dϕ|2(t0|h(tt)|2dt){t0|Rdωw(δ)(ω)Ωu(ϕ,t;ω)sin(ϕθ)dϕ|2dt}J0(t). (6.4)

    We apply the Cauchy–Schwartz inequality results in

    |Rdωw(δ)(ω)Ωu(ϕ,t;ω)sin(ϕθ)dϕ|2c63(Rdωw(δ)(ω))|||u(t)|||2(0)c64|||u(t)|||2(0). (6.5)

    Thus, Eqs (6.4) and (6.5) yield

    J0(t)c65h2L2(0,T)|||u|||2Tt(0,T].

    Next, we estimate S(2,0)Wl,02(ΩT). It is worth noting that S(2,0)=S holds. Thus, we obtain

    S(2,0)(θ1,t)S(2,0)(θ2,t)=t0h(tt)dtRdωw(δ)(ω)Ωu(ϕ,t;ω){sin(ϕθ1)sin(ϕθ2)}dϕ. (6.6)

    By virtue of the mean value theorem, we obtain

    sin(θθ1)sin(θθ2)=10ddσsin(σ(ϕθ1)+(1σ)(ϕθ2))dσ=(θ2θ1)10cos(σ(ϕθ1)+(1σ)(ϕθ2))dσ.

    From these, we obtain

    |S(2,0)(θ1,t)S(2,0)(θ2,t)|2c66|θ1θ2|2h2L2(0,T)|||u|||2T. (6.7)

    The approximation (6.7) and some elementary calculations yield

    S(2,0)2Wl,02(ΩT)c67Th2L2(0,T)|||u|||2T. (6.8)

    Next, we estimate S(0,1)W0,l22(ΩT). By noting that

    S(0,1)=0h(τ)dτRdωΩf(0,1)(ϕ,t;ω)sin(ϕθ)dϕ,

    we have

    S(0,1)(θ,t1;ω)S(0,1)(θ,t2;ω)=t10{h(t1t)h(t2t)}dtRdωw(δ)(ω)Ωu(0,1)(ϕ,t;ω)sin(ϕθ)dϕ+t1t2h(t2t)RdωΩu(0,1)(ϕ,t;ω)sin(ϕθ)dϕJ1+J2.

    Therefore, for each θΩ, we obtain

    T0dt1T0|S(0,1)(θ,t1)S(0,1)(θ,t2)|2|t1t2|1+ldt2c68(T0dt1T0|J1|2|t1t2|1+ldt2+T0dt1T0|J2|2|t1t2|1+ldt2). (6.9)

    To estimate the first term, we again apply the mean value theorem to obtain

    t10{h(t1t)h(t2t)}dtRdωw(δ)(ω)Ωu(0,1)(ϕ,t;ω)sin(ϕθ)dϕ=(t2t1)t10(10h((t1t)+σ(t2t))dσ)dt×Rdωw(δ)(ω)Ωu(0,1)(ϕ,t;ω)sin(ϕθ)dϕ

    Thus, we have

    T0dt1T0|J1|2|t1t2|1+ldt2c69T0dt1T0|t1t2|2|t1t2|1+l(t10|10h((t1t)+σ(t2t))dσ|2dt)×(t10|Rdωw(δ)(ω)Ωu(0,1)(ϕ,t;ω)sin(ϕθ)dϕ|2dt)dt2c610h2L2(0,T)T0dt1T0|t1t2|1l×(t10|Rdωw(δ)(ω)Ωu(0,1)(ϕ,t;ω)sin(ϕθ)dϕ|2dt)dt2. (6.10)

    To estimate the rightmost side of Eq (6.10) by virtue of the Cauchy–Schwartz inequality, we obtain

    |Rdωw(δ)(ω)Ωu(0,1)(ϕ,t;ω)sin(ϕθ)dϕ|2c611(Rdωw(δ)(ω))(supωRu(0,1)(,t;ω)2L2(Ω))c612|||u(0,1)(t)|||2.

    Together with Eq (6.10), this yields

    T0dt1T0|J1|2|t1t2|1+ldt2c610c612h2L2(0,T)|||u(0,1)|||2TT0dt1T0|t1t2|1ldt2=2c610c612T3l(2l)(3l)h2L2(0,T)|||u(0,1)|||2T. (6.11)

    J2 is approximated in a simpler manner using the fact that

    t2t1|h(t2t)|2dt|t1t2|2h2L(R+)c613T2h2W1+α2(R+).

    Here, the Sobolev embedding theorem is applied. From these considerations, we obtain:

    S(0,1)2W0,l22(ΩT)c614(T3lh2L2(R)+T2h2W1+α2(R))|||u(0,1)|||2T. (6.12)

    This completes the proof of Eq (6.3).

    We also prepare the following lemmas (see, for instance, [20]).

    Lemma 6.2. Letting l(1/2,1), the following holds:

    (ⅰ) For g1, g2W1+l2(Ω) in general,

    g1g2W1+l2(Ω)c615g1W1+l2(Ω)g2W1+l2(Ω)

    holds with some constant c615>0.

    (ⅱ) For g1W1+l2(Ω), g2Wl2(Ω) in general,

    g1g2Wl2(Ω)c616g1W1+l2(Ω)g2Wl2(Ω)

    holds with some constant c616>0.

    Lemma 6.3. Let l(1/2,1), m2 and mk, then for fWm+l,m+l22(Ω) and gWk+l,k+l22(Ω) in general, fgWk+l,k+l22(Ω) and

    fgWk+l,k+l22(Ω)c617fWm+l,m+l22(Ω)gWk+l,k+l22(Ω),

    with some constant c617>0.

    We estimate the first term on the right side of inequality (6.2). Considering that

    θ(F[˜f(m1),˜f(m1)w(δ)])W1+l,1+l22(ΩT)F[˜f(1,0)(m1),˜f(m1)w(δ)]W1+l,1+l22(ΩT)+F(1)[˜f(m1),˜f(m1)w(δ)]W1+l,1+l22(ΩT),

    we only estimate the first term. Using Lemma 6.3, we estimate

    F[˜f(1,0)(m1),˜f(m1)w(δ)]W1+l,1+l22(ΩT)˜f(1,0)(m1)(ω)W1+l,1+l22(ΩT)h(τ)dτRdωw(δ)(ω)Ω˜f(m1)(θ,tτ;ω)sin(θθ)dθW2+l,2+l22(ΩT), (6.13)

    then, by applying Lemma 6.1, we obtain

    F[˜f(1,0)(m1),˜f(m1)w(δ)](ω)W1+l,1+l22(ΩT)c618ϕ(T)˜f(m1)(ω)W1+l,1+l22(ΩT)˜f(m1)(ω)(2+l,2+l2);T. (6.14)

    We can then estimate the supremum of the function with respect to ω, as follows:

    supωRF[˜f(1,0)(m1),˜f(m1)w(δ)](ω)2W1+l,1+l22(ΩT)c618ϕ(T)|||˜f(m1)|||2(1+l,1+l2);T|||˜f(m1)|||2(2+l,2+l2);T.

    Thus, we integrate the square of Eq (6.2) with respect to t and take the supremum with respect to ω to obtain the following:

    K(m)(T)c619{ϕ(T)|||˜f(m1)|||2(2+l,2+l2);T|||f(m1)|||2(2+l,2+l2);T+K0}c619(η+CηT){K(m1)(T)+(K(m1)(T))2+K0}, (6.15)

    where η>0 is a constant that is determined later and Cη is a constant that is decreasingly dependent on η.

    We now inductively demonstrate that {Km(T)}m0 is bounded in V3+l. Consider a constant M69 that satisfies

    K0<M620.

    We also assume that, for a certain m1,

    K(m1)(T)<M620

    holds. By virtue of Eq (6.15), we initially consider η so that

    c613η(2M620+M2620)<M620.

    Next, let T621 be sufficiently small so that

    c619CηT621(2M620+M2620)<M620c619η(2M620+M2620).

    Subsequently, based on Eq (6.15), we obtain K(m)(T621)<M619. By induction, {K(m)(T621)}m is a bounded sequence in V3+lT621.

    Next, we demonstrate that ˜f(m) converges to some element in V3+lT622 with a certain T622(0,T621]. To demonstrate this, we define

    ˜˜f(m)=˜f(m)˜f(m1)

    for m1 and subtract Eq (6.1) with m replaced by m+1 from itself. This satisfies

    {˜˜f(m)tε2˜˜f(m)θ2=K[θ(F[˜f(m),˜f(m)w(δ)])θ(F[˜f(m1),˜f(m1)w(δ)])],˜˜f(j,0)(m)(0,t;ω)=˜˜f(j,0)(m)(2π,t;ω)(j=0,1),˜˜f(m)(θ,0;ω)=0. (6.16)

    Note that

    θ(F[˜f(m),˜f(m)w(δ)])θ(F[˜f(m1),˜f(m1)w(δ)])=F[˜˜f(1,0)(m),˜f(m)w(δ)]+F[˜f(1,0)(m1),˜˜f(m)w(δ)]+F(1)[˜˜f(m),˜f(m)w(δ)]+F(1)[˜f(m1),˜˜f(m)w(δ)]4j=1Lj(θ,t;ω).

    Using multiplicative and interpolation inequalities, we estimate the first term on the righthand side. By introducing

    S(m)(θ,t)t0h(τ)dτRdωΩ˜f(m)(θ,tτ;ω)sin(θθ)dθ,

    we can write

    F[˜˜f(1,0)(m),˜f(m)w(δ)]=˜˜f(1,0)(m)(θ,t;ω)S(m)(θ,t).

    Using Lemma 6.3 again, we can observe that

    L1(;ω)2W1+l,1+l22(ΩT)˜˜f(m)(;ω)2W1+l,1+l22(ΩT)S(m)2W2+l,2+l22(ΩT)c623ϕ(T)˜˜f(m)(;ω)2W1+l,1+l22(ΩT)|||˜f(m)|||2(2+l,2+l2);Tc623ϕ(T)(η+CηT)2˜˜f(m)(;ω)2W2+l,2+l22(ΩT)|||˜f(m)|||2(2+l,2+l2);T,

    where ϕ() is a polynomial of its argument. Thus, we obtain

    supωRL1(;ω)2W1+l,1+l22(ΩT)c623ϕ(T)(η+CηT)2|||˜˜f(m)|||2(2+l,2+l2);T|||˜f(m)|||2(2+l,2+l2);T.

    The other terms Lj(;ω)(j=2,3,4) are estimated similarly, and we obtain

    supωR|Lj(ω)|2c623(η+CηT)2ϕ(T)(|||˜f(m1)|||2(2+l,2+l2);T+|||˜f(m)|||2(2+l,2+l2);T)|||˜˜f(m)|||2(3+l,3+l2);T. (6.17)

    We introduce the following notation:

    ˜K(m)(T)supωR˜f(m)(ω)W3+l,3+l22(ΩT)=|||˜f(m)|||(3+l,3+l2);T.

    Then, we obtain

    ˜K(m)(t)c623(η+Cηt)ϕ(t){1+mj=m1K(j)(t)}˜K(m)(t).

    Let η be sufficiently small so that the following holds:

    c617ηϕ(T622)(1+2M620)<1,

    then, for this η, we take a sufficiently small t=T624(0,T622] so that

    c623Cηϕ(T622){1+2M619}<1c623ηϕ(T621)

    holds. We can now observe that

    ˜K(m+1)(t)<r˜K(m)(t),

    with r=c623(η+CηT624)ϕ(T624)(1+2M620)(0,1). We can conclude that {˜f(m)}m=1 forms a Cauchy sequence in V3+lT624, and the limitations

    f=limm+˜f(m)

    exist in the same function space. This is the desired solution. As the uniqueness of this solution can be verified in a similar manner, we omit the detailed discussion. This completes the proof of Theorem 5.1.

    We now prove Theorem 5.2. We examine the temporal behavior of the solution around a trivial stationary solution ˉf=g(ω)/2π. For this purpose, we subtract ˉf from Eq (2.4) to obtain the problem for ˘f˜fˉf, which satisfies

    {˘ftε2˘fθ2+Kθ(F[˘f+ˉf,˘fw(δ)])=0inΩ×R,˘f(j,0)(0,t;ω)=˘f(j,0)(2π,t;ω)(j=0,1)onR+×R,˘f(θ,t;ω)=˘f0(θ;ω)˜f0ˉfonΩ×(,0]×R. (7.1)

    We begin with the linear problem that corresponds to Eq (7.1):

    {˘˘ftε2˘˘fθ2+Kθ(F[ˉf,˘˘fw(δ)])=GinΩ×R,˘˘f(j,0)(0,t;ω)=˘˘f(j,0)(2π,t;ω)(j=0,1)onR+×R,˘˘f(θ,t;ω)=˘˘f0(θ,t;ω)onΩ×(,0]×R. (7.2)

    We introduce the new variable v(θ,t;ω)ekt˘˘f(θ,t;ω), where ˘˘f is the solution of Eq (7.2), which solves

    {vt+kvε2vθ2+Kˉfθ(0h(τ)ekτdτRdωΩv(θ,tτ;ω)w(δ)(ω)sin(θθ)dθ)=ektGinΩ×R,v(j,0)(0,t;ω)=v(j,0)(2π,t;ω)(j=0,1)onR+×R,v(θ,t;ω)=˘f0onΩ×(,0]×R. (7.3)

    We now obtain an a-priori estimate of the solution under a small initial data size. We first approximate the linear problem and then discuss the nonlinear model.

    First, we consider the linearized problem of Eq (7.3) with zero initial data.

    {vt+kvε2vθ2+Kˉfθ(0h(τ)ekτdτRdωΩv(θ,tτ;ω)w(δ)(ω)sin(θθ)dθ)=ektGinΩ×R,v(j,0)(0,t;ω)=v(j,0)(2π,t;ω)(j=0,1)onR+×R,v(θ,t;ω)=0onΩ×(,0]×R. (7.4)

    Lemma 7.1. Let T>0 be an arbitrary number, and G(θ,t;ω)V(1+l) be periodic with respect to θΩ. In addition, suppose that the assumptions on h(t) in Theorem 5.1 and Theorem 5.2 holds, then there exists a solution to Eq (7.4) for (0,+) that satisfies the estimate in the form of

    v(3+l,3+l2);Tc71ektG(1+l,1+l2);T. (7.5)

    Proof. First, we consider the problem with vanishing initial data.

    {vt+kvε2vθ2+Kˉfθ(0h(τ)ekτdτRdωΩv(θ,tτ;ω)w(δ)(ω)sin(θθ)dθ)=ektGinΩ×R,v(j,0)(0,t;ω)=v(j,0)(2π,t;ω)(j=0,1)onR+×R,v(θ,t;ω)=0onΩ×(,0]×R. (7.6)

    We now expand v and G using the Fourier series with respect to θ as follows:

    v(θ,t;ω)=+n=˜an(t;ω)einθ,G(θ,t;ω)=+n=bn(t;ω)einθ,

    then, Eq (7.6) becomes

    n˜anteinθ+n(k+εn2)˜aneinθKg(ω)2[+0h(τ)ekτdτR1w(δ)(ω){˜a1(tτ;ω)eiθ+˜a1(tτ;ω)eiθ}dω]=nbnekt+iθ.

    A comparison of the coefficients of eiθ yields:

    ˜a1t+(k+ε)˜a1Kg(ω)2+0h(τ)ekτdτR˜a1(tτ;ω)w(δ)(ω)dω=b1ekt, (7.7)
    ˜ant+(k+εn2)˜an=bnekt(n±2,±3,). (7.8)

    Next, we apply a transform similar to that of Beale [1] to Eqs (7.7) and (7.8).

    F[f]ˆf(ξ)=Reiξtf(t)dt.

    Note that the following equality holds:

    F[R˜a1(ts)g(s)ds]=Reiξt{R˜a1(ts)g(s)ds}dt=R(Reiξ(ts)˜a1(ts)dt)eiξsg(s)ds,

    then from the assumption that ˜an|t=0=0, we have

    (iξ+k+ε)ˆa±1(ξ;ω)Kg(ω)2Rˆa±1(ξ;ω)ˆh(ξik)w(δ)(ω)dω=ˆb±1(ξik;ω), (7.9)
    (iξ+k+εn2)ˆan(τ;ω)=ˆbn(ξik;ω)(n±1). (7.10)

    Now, we introduce a region D(+){zC|Re(z)>0} and λD(+). For Eq (7.9), by letting ξ=iλ, we obtain

    (λ+k+ε)ˆa1(iλ;ω)Kg(ω)2ˆh(i(λ+k))ˆa1(iλ;ω)w(δ)(ω)dω=ˆb1(i(λ+k);ω).

    Multiplying this by ¯(ˆa1(iλ;ω)) results in

    (λ+k+ε)|ˆa1(iλ;ω)|2Kg(ω)2ˆh(i(λ+k))¯ˆa1(iλ;ω)ˆa1(iλ;ω)w(δ)(ω)dω=¯ˆa1(iλ;ω)ˆb1(i(λ+k);ω).

    From this, we obtain

    (λ+k+ε)|ˆa1(iλ;ω)|212[|ˆa1(iλ;ω)|2+|Kg(ω)2ˆh(i(λ+k))(Rˆa1(iλ;ω)w(δ)(ω)dω)|2],

    then by taking the supremum with respect to ω, we obtain

    (λ+k+ε)supω|ˆa1(iλ;ω)|212supω|ˆa1(iλ;ω)|2+K24supω|g(ω)||ˆh(i(λ+k))|supω|ˆa1(iλ;ω)|2+supω|ˆa1(iλ;ω)|supω|ˆb1(i(λ+k);ω)|.

    By estimating K24supω|g(ω)||ˆh(i(λ+k))|˜c0 and letting k>0 be sufficiently large, we have

    (λ+k+ε˜c012)supω|ˆa1(iλ;ω)|2supω|ˆa1(iλ;ω)|supω|ˆb1(i(λ+k);ω)|,

    which yields

    supω|ˆa1(iλ;ω)|supω|ˆb1(i(λ+k);ω)|(λ+k+ε˜c0). (7.11)

    This implies that

    (1+|λ|3+l2)supω|ˆa1(iλ;ω)|(1+|λ|1+l2)supω|ˆb1(i(λ+k);ω)|. (7.12)

    For n±1, we have

    ˆan(ξ;ω)=ˆbn(ξ;ω)iξ+k+εn2.

    By setting ξ=iλ to λ=σ0+iσ1 and integrating with respect to σ1, we obtain:

    R|ˆan(iλ;ω)|2dσ1=R|ˆbn(iλ;ω)|2|λ+k+εn2|2dσ1.

    As |λ+k+εn2|k+εn2,

    |λ+k+εn2|1+n2c72

    holds with c72 independent of n. Therefore, for an arbitrary αR,

    R(1+n2)α|ˆan(iλ;ω)|2dσ1=R(1+n2)α|ˆbn(iλ;ω)|2|λ+k+εn2|2dσ1=R(1+n2)α|ˆbn(i(ηk);ω)|2|η+εn2|2dη1,

    where η=λ+k. From these, we eventually obtain

    |ˆan(iλ;ω)||ˆbn(iλ;ω)|λ+k+εn2(n±1).

    Next, by multiplying (7.11) by |λ|3+l, we obtain

    |λ|3+lsupωR|ˆa1(iλ;ω)|2|λ|3+lsupωˆb1(i(λ+k);ω)|λ+k+ε˜c0|2|λ|1+lsupωR|ˆb1(i(λ+k);ω)|2.

    Thus, we have

    (1+n3+l+|λ|3+l2)supω|ˆan(iλ;ω)|(1+n1+l+|λ|1+l2)supω|ˆbn(i(λ+k);ω)|λD(+)(n=±1,±2,). (7.13)

    Let us denote λ=σ0+iσ1 with σjR(j=0,1), then ˆb1(i(λ+k);ω)=Reλtb1(t;ω)dt is the Fourier transform of eσ0tb1(t;ω) with respect to σ1. We obtain

    ˆb1(i(λ+k);ω)=Re(λ+k)tb1(t;ω)dt=Rekte(σ0+iσ1)tb1(t;ω)dt=F[e(σ0+k)tb1(t;ω)].

    Thus, by virtue of the Plancherel theorem, we obtain

    R|ˆb1(i(λ+k);ω)|2dσ1=R|e(σ0+k)tb1(t;ω)|2dt. (7.14)

    Note that the righthand side of Eq (7.12) is finite for all σ0R. By letting σ00 there, the right side of Eq (7.12) tends to

    R|ektb1(t;ω)|2dt=ektb12L2(0,+).

    Likewise, we obtain

    R|ˆa1(iλ;ω)|2dσ1=R|eσ0ta1(t;ω)|2dtR|a1(t;ω)|2dt(σ00).

    Thus, if we take a sufficiently large k, we have

    |||v|||(m,m2);|||ektG|||(m,m2);.

    The uniqueness of the solution is guaranteed by construction; from Eqs (7.9) and (7.10), it is obvious that the solution {an} is uniquely determined. Thus, the proof is complete.

    Now, we consider problem (7.1). We use the following notations:

    L˘fk˘f+ε2˘fθ2+Kˉfθ(0h(s)eksdsRdωΩ˘f(θ,ts;ω)w(δ)(ω)sin(θθ)dθ),F[˘f]Kektθ[˘f(θ,t;ω)0h(τ)eksdsRdωw(δ)(ω)Ω˘f(θ,ts;ω)sin(θθ)dθ],

    then, the problem is expressed as

    A˘f=F[˘f], (7.15)

    where A is a linear operator defined on V3+l, which associates ˘f with ˘ftL˘f under the periodic boundary conditions. From Lemma 6.3, we obtain

    |||F[˘f]|||(1+l,1+l2);c73|||˘f|||(3+l,3+l2); (7.16)

    for v in a bounded set of V3+l. Similarly, it is apparent that for ˘fjV3+l(j=1,2),

    |||F[˘f1]F[˘f2]|||(1+l,1+l2);c74|||˘f1˘f2|||(3+l,3+l2);2j=1|||˘fj|||(3+l,3+l2);. (7.17)

    To solve Eq (7.15) iteratively, we first determine ˘f(0)V3+l, which satisfies Eq (7.3) at t=0 and

    |||˘f(0)|||(3+l,3+l2);c75˘f0W2+l2(Ω).

    This is obtained by tracing the classical argument ([14], Theorem Ⅳ. 4.3). Next, we rewrite the problem for the new variable ˘f(1)˘f˘f(0):

    A˘f(1)=F[˘f(0)+˘f(1)]A˘f(0). (7.18)

    If ˘f(1)V3+l, using our method of constructing ˘f(0), A˘f(0)=F[˘f(0)]=F[˘f(0)+˘f(1)] at t=0. Thus, the right-hand side of Equation (7.18) belongs to V1+l. Let A0 be a solution operator of Lemma 7.1 for a linear problem with zero initial data. Then, by virtue of Lemma 7.1, if

    ˘f(1)=A10[F[˘f(0)+˘f(1)]A˘f(0)], (7.19)

    the above assumption is satisfied because Lemma 7.1 indicates that A10 is a bounded operator from V1+l to V3+l. To demonstrate the solvability of Eq (7.19), we define a map

    M[˘f(1)]A10[F[˘f(0)+˘f(1)]A˘f(0)],

    and show that it has a fixed point assuming that

    |||˘f0|||(1+l)δ0

    with sufficiently small δ0>0. We obtain

    |||˘f(0)|||(3+l,3+l2);c75δ0

    and |||A˘f(0)|||(1+l,1+l2);c76δ0. Then, using Eq (7.16), we obtain

    |||F[˘f(0)+˘f(1)]|||(1+l,1+l2);c77(δ20+|||˘f(1)|||2(3+l,3+l2);). (7.20)

    Combining this result with the boundedness of A10 yields

    |||M[˘f(1)]|||(1+l,1+l2);c78(|||˘f(1)|||2(3+l,3+l2);+δ20+δ0). (7.21)

    Thus, if we consider ¯B{˘f(1)V3+l|˘f(1)V3+l2c78δ0}, M maps ¯B to itself provided that δ0 is sufficiently small, satisfying (4c278+1)δ01. Similarly, we obtain the following from Eq (7.17):

    |||M[v(1)]M[v(2)]|||(3+l,3+l2);c79δ0|||v(1)v(2)|||(3+l,3+l2);. (7.22)

    Thus, if we consider δ0<1/c79, M is a contraction map of ¯B and has a unique fixed point. This establishes the existence of a solution vV3+l.

    Because our method is based on the assumptions of Theorems 5.1 and 5.2, this approach will be available to use as long as those assumptions are satisfied.

    In this section, we prove other properties of the solution obtained thus far. We first prove that the solution satisfies the requirements of a probability density.

    Lemma 8.1. The solution f(θ,t;ω) of Eq (2.2) satisfies the following:

    (ⅰ) RdωΩf(θ,t;ω)dθ=1tR+.

    (ⅱ) f(θ,t;ω)0(θ,t,ω)Ω×R+×R.

    Proof. Note that it is sufficient to consider Eq (2.3). The first statement can be easily proven by considering

    ddtRdωΩˆf(θ,t;ω)dθ=εRdωΩ2ˆfθ2dθKRdωΩθ(F[ˆf,ˆf])dθ. (8.1)

    Clearly, the righthand side of Eq (8.1) vanishes owing to the periodicity of ˆf with respect to θ.

    Next, we prove the following lemma.

    Lemma 8.2. Suppose that f0(θ;ω)0(θ,ω)Ω×R. Then, the solution f(θ,t;ω) to Eq (2.2) satisfies f(θ,t;ω)0(θ,t,ω)Ω×R+×R.

    Proof. Again, if is sufficient to consider Eq (2.3). This time, we employ Stampaccia's truncation method. By introducing

    f+(|f|+f)20,f(|f|f)20,

    which are the positive and negative parts of ˆf at each point, we can decompose ˆf into ˆf=ˆf+ˆf. Then, if we multiply (2.3) by ˆf, and integrate with respect to θ over Ω, we obtain

    12ddtˆf(t;ω)2L2(Ω)+εˆfθ(t;ω)2L2(Ω)=K|Ωfθ(F(ˆf,ˆf))dθ|.

    By virtue of Lemmas 8.1 and 8.2, we obtain

    Ωfθ(F(ˆf,ˆf))dθ=Ωfθ[(f+f){0h(τ)dτRdωΩf(θ,tτ;ω)sin(θθ)dθ}]dθ=Ωffθ{0h(τ)dτRdωΩf(θ,tτ;ω)sin(θθ)dθ}dθ.

    Indeed, if we introduce notations Ω(t,ω){θΩ|f(θ,t;ω)0}, and g(θ,t)0h(τ)dτdωΩf(θ,tτ;ω)sin(θθ)dθ, we have

    Ωf(θ,t;ω)θ(f+(θ,t;ω)g(θ,t))dθ=Ω(t,ω)f(θ,t;ω)θ(f+(θ,t;ω)g(θ,t))dθ
    12ddtˆf(t;ω)2L2(Ω)+εˆfθ(t;ω)2L2(Ω)Kˆf(t;ω)2L2(Ω).

    As ˆf(θ,0)=0θΩ, we obtain ˆf0 by virtue of the Gronwall's inequality.

    We have proven the existence and uniqueness of a global-in-time solution to a parabolic-regularized Fokker–Planck equation that corresponds to the Kuramoto model with delay proposed by Lee et al. [17].

    This argument theoretically demonstrates the validity of the model. However, our approach has some limitations. First, we proved the existence of a global-in-time solution under small initial data. We will relax this restriction in our future work.

    Second, we will discuss the existence and structure of the invariant set or inertial manifold of the proposed model. This will be interesting because past arguments have implied the existence of multiple stable states under the presence of delay. Finally, we have considered only the parabolic-regularized problem. In our future work, we will consider the vanishing diffusion limit of the diffusion coefficient, which corresponds to the original model proposed by Lee et al. [17].

    The authors declare that they have not used Artificial Intelligence (AI) tools in the creation of this article.

    We thank the anonymous reviewers whose comments and suggestions have helped to improve and clarify this manuscript. This work was supported by Toyo University Top Priority Research Program.

    The authors declare there is no conflict



    [1] J. T. Beale, Large-time regularity of viscous surface waves, Arch. Ration. Mech. Anal., 84 (1983), 307–352. https://doi.org/10.1007/BF00250586 doi: 10.1007/BF00250586
    [2] R. C. Budzinski, T. T. Nguyen, G. B. Benigno, J. Doàn, Ján Mináč, T. J. Sejnowski, et al., Analytical prediction of specific spatiotemporal patterns in nonlinear oscillator networks with distance-dependent time delay, Phys. Rev. Lett., 5 (2023), 013159. https://doi.org/10.1103/PhysRevResearch.5.013159 doi: 10.1103/PhysRevResearch.5.013159
    [3] F. Chapeau-Blondeau, G. Chauvet, Stable, oscillatory, and chaotic regimes in the dynamics of small neural networks with delay, Neural Netw., 5 (1992), 735–743. https://doi.org/10.1016/S0893-6080(05)80134-0 doi: 10.1016/S0893-6080(05)80134-0
    [4] H. Chiba, A proof of the Kuramoto conjecture for a bifurcation structure of the infinite dimensional Kuramoto model, Ergod. Theory Dyn. Syst., 35 (2015), 762–834. https://doi.org/10.1017/etds.2013.68 doi: 10.1017/etds.2013.68
    [5] M. Y. Choi, H. J. Kim, D. Kim, H. Hong, Synchronization in a system of globally coupled oscillators with time delay, Phys. Rev. E, 61 (2000), 371–381. https://doi.org/10.1103/PhysRevE.61.371 doi: 10.1103/PhysRevE.61.371
    [6] J. D. Crawford, Amplitude expansions for instabilities in populations of Globally-Coupled oscillators, J. Stat. Phys., 74 (1994), 1047–1082. https://doi.org/10.1007/BF02188217 doi: 10.1007/BF02188217
    [7] S. Y. Ha, Q. Xiao, Remarks on the nonlinear stability of the Kuramoto-Sakaguchi equation, J. Diff. Eq., 259 (2015), 2430–2457. https://doi.org/10.1016/j.jde.2015.03.038 doi: 10.1016/j.jde.2015.03.038
    [8] S. Y. Ha, Q. Xiao, Nonlinear instability of the incoherent state for the Kuramoto-Sakaguchi-Fokker-Plank equation, J. Stat. Phys., 160 (2015), 477–496. https://doi.org/10.1007/s10955-015-1270-5 doi: 10.1007/s10955-015-1270-5
    [9] H. Honda, A. Tani, Mathematical analysis of synchronization from the perspective of network science, Mathematical Analysis of Continuum Mechanics and Industrial Applications (Proceedings of the International Conference CoMFoS15), Singapore: Springer, 2017.
    [10] H. Honda, Global-in-time solution and stability of Kuramoto-Sakaguchi equation under non-local coupling, Netwo. Heterog. Media, 12 (2017), 25–57. https://doi.org/10.3934/nhm.2017002 doi: 10.3934/nhm.2017002
    [11] Y. Kawamura, From the Kuramoto-Sakaguchi model to the Kuramoto-Sivashinsky equation, Phys. Rev. E, 89 (2014), 010901.
    [12] S. Kim, S. H. Park, C. S. Ryu, Multistability in coupled oscillator systems with time delay, Phys. Rev. Lett., 79 (1997), 2911–2914. https://doi.org/10.1103/PhysRevLett.79.2911 doi: 10.1103/PhysRevLett.79.2911
    [13] Y. Kuramoto, D. Battogtokh, Coexistence of coherence and incoherence in nonlocally coupled phase oscillators, Nonlinear Phenom. Complex Syst., 5 (2002), 380–385.
    [14] O. A. Ladyženskaja, V. A. Solonnikov, N. N. Ural'ceva, Linear and Quasi-linear Equations of Parabolic Type, Rhode Island: American Mathematical Society, 1968.
    [15] M. Lavrentiev, R. S. Spigler, Existence and uniqueness of solutions to the Kuramoto-Sakaguchi nonlinear parabolic integrodifferential equation, Differ. Integr. Equ., 13 (2000), 649–667.
    [16] M. Lavrentiev, R. S. Spigler, A. Tani, Existence, uniqueness, and regularity for the Kuramoto–Sakaguchi equation with unboundedly supported frequency distribution, Differ. Integral Equ., 27 (2014), 879–892.
    [17] W. S. Lee, E. Ott, T. M. Antonsen, Large Coupled Oscillator systems with heterogeneous interaction delays, Phys. Rev. Lett., 103 (2009), 044101.
    [18] E. Niebur, H. G. Schuster, D. M. Kammen, Collective frequencies and metastability in networks of limit-cycle oscillators with time delay, Phys. Rev. Lett., 67 (1991), 2753–2756. https://doi.org/10.1103/PhysRevLett.67.2753 doi: 10.1103/PhysRevLett.67.2753
    [19] H. Risken, The Fokker-Planck Equation, Berlin: Springer, 1989.
    [20] N. Tanaka, A. Tani, Surface waves for a compressible viscous fluid, J. Math. Fluid Mech., 5 (2003), 303–363. https://doi.org/10.1007/s00021-003-0078-2 doi: 10.1007/s00021-003-0078-2
    [21] S. H. Strogatz, E. Mirollo, Stability of incoherent in a population of coupled oscillators, J. Stat. Phys., 63 (1991), 613–635. https://doi.org/10.1007/BF01029202 doi: 10.1007/BF01029202
    [22] J. Wloka, Partielle Differentialgleichungen: sobolevraume und Randwertaufgaben, Stuttgart: Teubner, 1982.
    [23] M. K. S. Yeung, S. H. Strogatz, Time delay in the Kuramoto model of coupled oscillators, Phys. Rev. Lett., 82 (1999), 648–651. https://doi.org/10.1103/PhysRevLett.82.648 doi: 10.1103/PhysRevLett.82.648
    [24] T. Zhu, Synchronization of the generalized Kuramoto model with time delay and frustration, Netw. Heterog. Media, 18 (2023), 1772–1798. https://doi.org/10.3934/nhm.2023077 doi: 10.3934/nhm.2023077
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1232) PDF downloads(329) Cited by(0)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog