Loading [MathJax]/jax/element/mml/optable/Arrows.js
Research article

Fixed point results of an implicit iterative scheme for fractal generations

  • Received: 06 May 2021 Accepted: 23 August 2021 Published: 15 September 2021
  • MSC : 37F45, 37F50, 47J25

  • In this paper, we derive the escape criteria for general complex polynomial f(x)=pi=0aixi with p2, where aiC for i=0,1,2,,p to generate the fractals. Moreover, we study the orbit of an implicit iteration (i.e., Jungck-Ishikawa iteration with s-convexity) and develop algorithms for Mandelbrot set and Multi-corn or Multi-edge set. Moreover, we draw some complex graphs and observe how the graph of Mandelbrot set and Multi-corn or Multi-edge set vary with the variation of ai's.

    Citation: Haixia Zhang, Muhammad Tanveer, Yi-Xia Li, Qingxiu Peng, Nehad Ali Shah. Fixed point results of an implicit iterative scheme for fractal generations[J]. AIMS Mathematics, 2021, 6(12): 13170-13186. doi: 10.3934/math.2021761

    Related Papers:

    [1] Zhong-Jie Han, Enrique Zuazua . Decay rates for elastic-thermoelastic star-shaped networks. Networks and Heterogeneous Media, 2017, 12(3): 461-488. doi: 10.3934/nhm.2017020
    [2] Zhong-Jie Han, Enrique Zuazua . Decay rates for 1d heat-wave planar networks. Networks and Heterogeneous Media, 2016, 11(4): 655-692. doi: 10.3934/nhm.2016013
    [3] Dongyi Liu, Genqi Xu . Input-output L2-well-posedness, regularity and Lyapunov stability of string equations on networks. Networks and Heterogeneous Media, 2022, 17(4): 519-545. doi: 10.3934/nhm.2022007
    [4] Yaru Xie, Genqi Xu . The exponential decay rate of generic tree of 1-d wave equations with boundary feedback controls. Networks and Heterogeneous Media, 2016, 11(3): 527-543. doi: 10.3934/nhm.2016008
    [5] Martin Gugat, Mario Sigalotti . Stars of vibrating strings: Switching boundary feedback stabilization. Networks and Heterogeneous Media, 2010, 5(2): 299-314. doi: 10.3934/nhm.2010.5.299
    [6] Serge Nicaise, Julie Valein . Stabilization of the wave equation on 1-d networks with a delay term in the nodal feedbacks. Networks and Heterogeneous Media, 2007, 2(3): 425-479. doi: 10.3934/nhm.2007.2.425
    [7] Kaïs Ammari, Mohamed Jellouli, Michel Mehrenberger . Feedback stabilization of a coupled string-beam system. Networks and Heterogeneous Media, 2009, 4(1): 19-34. doi: 10.3934/nhm.2009.4.19
    [8] Gen Qi Xu, Siu Pang Yung . Stability and Riesz basis property of a star-shaped network of Euler-Bernoulli beams with joint damping. Networks and Heterogeneous Media, 2008, 3(4): 723-747. doi: 10.3934/nhm.2008.3.723
    [9] Caihong Gu, Yanbin Tang . Global solution to the Cauchy problem of fractional drift diffusion system with power-law nonlinearity. Networks and Heterogeneous Media, 2023, 18(1): 109-139. doi: 10.3934/nhm.2023005
    [10] Abdelaziz Soufyane, Adel M. Al-Mahdi, Mohammad M. Al-Gharabli, Imad Kissami, Mostafa Zahri . Stability results of a swelling porous-elastic system with two nonlinear variable exponent damping. Networks and Heterogeneous Media, 2024, 19(1): 430-455. doi: 10.3934/nhm.2024019
  • In this paper, we derive the escape criteria for general complex polynomial f(x)=pi=0aixi with p2, where aiC for i=0,1,2,,p to generate the fractals. Moreover, we study the orbit of an implicit iteration (i.e., Jungck-Ishikawa iteration with s-convexity) and develop algorithms for Mandelbrot set and Multi-corn or Multi-edge set. Moreover, we draw some complex graphs and observe how the graph of Mandelbrot set and Multi-corn or Multi-edge set vary with the variation of ai's.



    In recent years the problems of control and stabilisation for thermoelastic systems have been studied intensively. We refer for instance to [22], [42] in the context of controllability, and to [14], [18], [23], [24], [35] for stabilization, among others.

    A closely related interesting issue is the asymptotic behaviour of a system consisting of two different materials joined together at the interface, one being purely elastic and the other thermoelastic (see Figure 1). In these systems, other than the intrinsic coupling effects of thermal and elastic components, typical in thermoelasticity, the thermoelastic and purely elastic rods are also coupled through the interface.

    Figure 1.  Transmission problem in 1-d elasticity-thermoelasticity.

    Exponential and polynomial decay properties for this kind of systems were proved by Rivera et al. in [12], [28], [31] and Messaoudi et al. in [30], using the energy multiplier method. Han and Xu in [15] got a sharp polynomial decay rate for a thermoelastic transmission problem with joint mass, based on a detailed spectral analysis and resolvent operator estimates. We also refer to [29] for the analysis of the large time behaviour of transmission problems of multi-dimensional thermoelasticity.

    In this work, we consider similar transmission problems in multi-connected networks. More precisely, we are interested in the large time behaviour of star-shaped networks constituted by coupled thermoelastic and purely elastic rods (see Figure 2).

    Figure 2.  Star-shaped thermoelastic-elastic network.

    The aim of this work is to give a complete analysis on the large time behaviour of these systems proving exponential, polynomial and slow decay rates. The optimality of these results is also discussed.

    In order to present the problems under consideration more precisely some notations are needed. We denote by ej,j=1,2,,n the segments occupying the intervals (0,j),j0, respectively. Denote by x=0 the common node of this star-shaped network. Assume that the edges j=1,2,...,N1(0<N1<N) in the network are constituted by thermoelastic rods, given by the following equations:

    {uk,tt(x,t)uk,xx(x,t)+αkθk,x(x,t)=0,x(0,k),k=1,2,...,N1,t>0,θk,t(x,t)θk,xx(x,t)+βkuk,tx(x,t)=0,x(0,k),k=1,2,...,N1,t>0, (1)

    and the other edges in the network are all purely elastic ones given by

    uj,tt(x,t)uj,xx(x,t)=0,x(0,j),j=N1+1,,N,t>0. (2)

    Here and in the sequel uj(x,t),j=1,2,,N denote the displacements of the rods at time t, and θk(x,t),k=1,2,,N1 the temperature difference with respect to a fixed reference temperature.

    Assume that the exterior nodes of the network are all clamped, and the displacements and temperatures are all continuous at the common node. There is no heat exchange between thermoelastic components and purely elastic ones and the balance of forces at the common node is fulfilled. Thus, the boundary and transmission conditions read as follows:

    {uj(j,t)=0,j=1,2,N,t>0,uj(0,t)=uk(0,t),j,k=1,2,3,,N,t>0,θk(k,t)=0,k=1,2,...,N1,t>0,θk(0,t)=θj(0,t),j,k=1,2,...,N1,t>0,Nj=1uj,x(0,t)=N1j=1αjθj(0,t),N1j=1αjβjθj,x(0,t)=0,t>0 (3)

    with initial conditions

    u|t=0=u(0):=(u(0)j)Nj=1,ut|t=0=u(1):=(u(1)j)Nj=1,θ|t=0=θ(0):=(θ(0)j)N1j=1. (4)

    Overall the thermoelastic-elastic network system under consideration reads as follows:

    {uj,tt(x,t)uj,xx(x,t)+αjθj,x(x,t)=0,x(0,j),j=1,2,,N1,t>0,θj,t(x,t)θj,xx(x,t)+βjuj,tx(x,t)=0,x(0,j),j=1,2,,N1,t>0,uj,tt(x,t)uj,xx(x,t)=0,x(0,j),j=N1+1,,,N,t>0,uj(j,t)=0,j=1,2,,N,t>0,uj(0,t)=uk(0,t),j,k=1,2,3,,N,t>0,θk(k,t)=0,k=1,2,,N1,t>0,θk(0,t)=θj(0,t),j,k=1,2,...N1,t>0,Nj=1uj,x(0,t)=N1j=1αjθj(0,t),N1j=1αjβjθj,x(0,t)=0,t>0,u(t=0)=u(0),ut(t=0)=u(1),θ(t=0)=θ(0). (5)

    Remark 1. The transmission problem in Figure 1 can be considered as a special case of the above network (N=2,N1=1).

    The natural energy of this system is as follows

    E(t)=12Nj=1j0[u2j,t+u2j,x]dx+12N1j=1αjβjj0θ2jdx,

    and a direct calculation yields the dissipation law

    E(t)=N1k=1αkβkk0θ2k,xdx0. (6)

    Hence, the energy of system (5) is decreasing. Moreover, from (6), it is easy to see that the dissipation mechanism only acts in the thermoelastic rods. This motivates the problem of whether or not the dissipation is strong enough to make the total energy of the network decay to zero, and with which rate.

    In this paper, the large time behaviour of system (5) is mainly discussed based on frequency domain analysis ([5], [6], [13], [19], [25], [26] and [34]). In [37], Shel showed the exponential stability of networks of thermoelastic and elastic materials for some special cases by similar methods. However, it was assumed that there was no heat exchange between the thermoelastic rods connected at the common nodes. In this paper, we allow for the heat exchange between thermoelastic rods at common nodes, which is a natural assumption.

    By estimating the resolvent operator along the imaginary axis and employing multiplier techniques, we get a necessary and sufficient condition for system (5) to decay uniformly exponentially, namely that there is no more than one purely elastic rod entering in the network. If this condition fails, the system lacks exponential decay and we further show that the decay rate of the networks can not be faster than t1. Moreover, for a very special case that there are two purely elastic rods involved in the network, the optimal polynomial decay result is obtained and for the general case that there are more purely elastic rods involved, a nearly optimal polynomial decay result is also derived under certain Diophantine approximation conditions, that refer to the lengths of the purely elastic rods involved in the network.

    To discuss the sharpness of slow decay rates it is useful to get explicit information on the spectrum of the system, and compare its real and imaginary parts (see [6] and [41]). However, spectra of PDE networks are often difficult to calculate. Thus, we prove the optimality by estimating the norm of the resolvent operator along the imaginary axis (see [1]). But resolvent estimates are hard to be achieved due to the thermoelastic coupling. Thus, we employ diagonalisation argument to deal with the resolvent problem. This allows building explicit approximations of solutions ensuring that the polynomial decay rates we get are nearly optimal.

    The rest of the paper is organised as follows. In section 2, the main result of this paper is given. Section 3 is devoted to show the well-posedness and strong asymptotic stability of the system (5). In section 4, we prove the exponential and nearly optimal slow decay rates, under different conditions. Section 5 is devoted to discuss some more general slow decay rates for system (5). Especially, for the special case NN1=2, the condition to achieve optimal decay of the network is obtained. In section 6, the numerical simulations of the dynamical behaviour of system (5) are presented.

    The results in this paper contribute to the understanding of the decay properties of wave and thermoelastic wave networks, a topic in which important issues are still to be understood.

    There has been an extensive literature on other closely related issues such as the large time behaviour and controllability properties of elastic networks with node and boundary feedback controls. We refer, among others (the present list of references is by no means complete), to Lagnese et al.[21] for the modelling and control of elastic networks; Ammari et al. [2], [3] and [4] and Nicaise et al. [32] for stabilisation problems on networks of wave and Euler-Bernoulli beams with star-shaped and tree-shaped configurations; Dáger and Zuazua [8], [9] and [10] for boundary controllability of wave networks; Xu et al. [16], [17] and [40] for the stabilisation and spectral properties of the wave networks.

    This section is devoted to state the main result of this paper.

    As we will see later, system (5) can be rewritten as an abstract Cauchy problem in an appropriate Hilbert space H:

    dU(t)dt=AU(t),   t>0;   U(0)=U0, (7)

    where U(t)=(u,ut,θ)T and U(0)=(u(0),u(1),θ(0))TH are given.

    The problem of whether the energy of solutions tends to zero as time goes to infinity or not has a simple answer:

    Theorem 2.1. Operator A generates a C0 semigroup of contractions on H. Moreover, the energy of the system (5) decays to zero as t if and only if one of the following two conditions is fulfilled,

    1). NN1=1;

    2). NN12 and i/jQ,i,j=N1+1,N1+2,,N,ij.

    We now obtain explicit decay rates for network (5).

    It is well known that if all the components in the network are thermoelastic, that is N=N1, the energy of the system decays exponentially to zero.

    In fact, in Propositions 1 and 2 in section 4, we obtain the following necessary and sufficient condition for the exponential decay.

    Theorem 2.2. The energy of system (5) decays to zero exponentially if and only if NN11, that is, if no more than one purely elastic undamped rod is involved in the network.

    Accordingly, when NN1>1, one can only expect slow decay rate for network (5). In order to address this issue we need the following definition from [36] and [11].

    Definition 2.3. ([36], [11]) Real numbers 1,2,,m are said to verify the conditions (S), if 1,2,,m are linearly independent over the field Q of rational numbers; and the ratios i/j are algebraic numbers for i,j=1,2,,m.

    The notation (S) for this condition was introduced in [11] to refer to the fundamental contribution by Schmidt [36], that defined this class of irrational numbers for the simultaneous approximation by rational ones. It should be noted that the condition (S) denotes a narrow class of irrationals, since the set of algebraic numbers is countable and has Lebesgue measure zero.

    By a detailed frequency domain analysis, we have the following explicit polynomial decay rate for system (5).

    Theorem 2.4. When NN1>1, The following estimation always holds.

    lim infttE(t)>0. (8)

    Thus, we can not expect a decay rate which is beyond first order polynomial. Furthermore, if N1+1,N1+2,,N(NN1>1) satisfy the conditions (S), then for any ϵ>0, there always exists a constant Cϵ>0 such that the energy of network (5) satisfies

    E(t)Cϵt11+ϵ(u(0),u(1),θ(0))2D(A),t0, (9)

    for all (u(0),u(1),θ(0))D(A). Thus, this decay rate is nearly sharp in the sense of (8).

    Remark 2. Using the method of proof of Theorem 2.4, we can obtain more general slow decay rates (polynomial, logarithmic or arbitrarily slow decay), which will be presented in section 5. Especially, for a very special case that there are two purely elastic rods involved in the network (NN1=2), we show that the network can achieve the optimal decay rate t1 if the mutual ratio of the lengths of the purely elastic rods belongs to the set of irrational numbers having a continuous fraction expansion [a0,a1,...,an,...] with bounded (an).

    Remark 3. The optimality result in (8) is well-known for wave-like equations with velocity damping in the case of one single string with damping on an internal point, which is equivalent to a simple star-like network constituted only by two strings ([20], [11]).

    Here we show the same lower bound on the decay rate for the more general system involving thermoelastic rods. However, for general cases that more than two purely elastic rods entering in the networks, it is still an open problem to find the condition to guarantee achieving a sharp polynomial decay rate.

    This section is devoted to show the well-posedness of network (5) by the semigroup theory and prove Theorem 2.1, with a necessary and sufficient condition for strong stability of this system.

    Let us first introduce an appropriate Hilbert space setting for the well-posedness of the system.

    For n1, define

    L2(Rn)={u|ujL2(0,j),j=1,2,,n},
    Vn:={ϕnj=1H1(0,j)|ϕj(0)=ϕk(0),ϕj(j)=0,k,j=1,2,,n}.

    Set the state space H as follows:

    H=VN×L2(RN)×L2(RN1),

    equipped with inner product:

    (W,˜W)H=Nj=1j0uj,x¯˜uj,xdx+Nj=1j0wj¯˜wjdx+N1k=1αkβkk0θk¯˜θkdx,

    for W=(u,w,θ),˜W=(˜u,˜w,˜θ)H.

    (H,H) is a Hilbert space.

    Then, define the system operator A in H as follows:

    A(uwθ)=(wuxxαIN×N1θxθxxβw1,x)=(0I0xx0αIN×N1x0βITN×N1xxx)[uwθ]

    where α=diag(α1,α2,,αN1),β=diag(β1,β2,,βN1), IN×N1=[IN1,0]T and IN1 is the N1unit matrix with domain

    D(A)={(u,w,θ)[VNNj=1H2(0,j)]×VN×N1j=1H2(0,j)|Nj=1uj,x(0)=N1k=1αjθj(0)θj(j)=0,j=1,2,,N1θj(0)=θk(0),j,k=1,2,,N1N1j=1αjβjθj,x(0)=0}.

    Thus, system (5) can be rewritten as the evolution equation (7) in H.

    It is easy to check that A is dissipative in H. Moreover, A is injective and surjective and hence 0ρ(A). Then, by Lummer-Phillips theorem (see [33]), A generates a C0 semigroup of contractions S(t) on H.

    Now, we focus on proving the strong stability property of the system in Theorem 2.1. The proof by contradiction is mainly used here.

    Proof of Theorem 2.1.

    Sufficiency. If the strong stability of system (5) does not hold, then by the Lyubich-Phóng strong stability theorem (see [27]), there exists at least one ˜λ=i˜σσ(A),˜σR,˜σ0 on the imaginary axis. Assume that

    ˜W=((uj)Nj=1,˜λ(uj)Nj=1,(θk)N1k=1)TD(A)

    is an eigenvector of A corresponding to such ˜λ. We get

    0=˜λ˜W2H=(A˜W,˜W)H=N1k=1αkβkk0θ2k,xdx,

    which yields θk=θk,x=0,k=1,2,,N1. Then by the boundary and transmission conditions in (5), uj(x),j=1,2,,N satisfy the following equations:

    {˜λ2uj(x)uj,xx(x)=0,x(0,j),j=1,2,,N,βj˜λuj,x(x)=0,x(0,j),j=1,2,,N1,uj(j)=0,j=1,2,,N,uj(0)=uk(0),j,k=1,2,3,,N,Nj=1uj,x(0)=0. (10)

    If NN1=1, it is easy to get (u,˜λu,θ)=0, which contradicts the fact that (u,˜λu,θ)=0 is an eigenvector.

    If NN12, we get by direct calculation,

    {uk=0,k=1,2,,N1,uj=cjsinh˜λx,j=N1+1,N1+2,,N,

    which satisfy

    cjsinh˜λj=0,j=N1+1,N1+2,,N,   and   Nj=N1+1cj=0.

    Since (u,˜λu,θ) is the eigenvector corresponding to ˜λ, then there are at least cj1,cj2 for some j1,j2N1+1 such that

    cj1,cj20.

    Hence, sinh˜λj1=sinh˜λj2=0. Thus, j1/j2Q, which is in contradiction with condition (2) in Theorem 2.1.

    Necessity. If there exist i0,j0 satisfying N1+1i0,j0N,i0/j0=p/q, with p,q nonzero integers, we get easily that ((ˆuj(x))Nj=1,˜λ(ˆuj(x))Nj=1,(0)N1k=1) is an eigenvector corresponding to ˜λ=iqπ/j0=ipπ/i0, in which

    ˆui0(x)=sin(pπx/i0),ˆuj0(x)=sin(q0πx/j0),
    ˆuj(x)=0,j=1,2,,N,ji0,j0.

    This contradicts the strong stability property of system (5). Therefore, i/jQ,i,j=N1+1,N1+2,,N,ij. The proof is complete.

    Remark 4. For the proof of "Sufficiency", we also can use the unique continuation property for wave networks in Dager and Zuazua [11] (See Corollary 5.28, p.135). Indeed, (10), in the absence of thermal components, corresponds to the eigenproblem associated with the pure wave system and, according to the results in [11], its unique solution is the trivial one, which contradicts that ((uj)Nj=1,˜λ(uj)Nj=1,(θk)N1k=1)T is an eigenvector. Then the desired result follows.

    This section is devoted to achieve explicit decay rates of the total energy of solutions of system (5). The exponential decay and slow decay rates are deduced under different assumptions of the various components of the network.

    In this subsection, we analyse the decay rate of network (5) when NN1=1, namely, there is only one purely elastic rod involved in the network. To do this, let us introduce the following lemma as described in [13], [19], [25] and [34].

    Lemma 4.1. Let (S(t))t0 be a C0 semigroup on a Hilbert space H generated by A. Then the semigroup is exponentially stable if and only if

    iRρ(A) (11)

    and

    (iσIA)1HC,σR. (12)

    We are then in conditions to prove the following proposition, which is one of the main statements in Theorem 2.1:

    Proposition 1. When NN1=1, the energy of system (5) decays exponentially to zero.

    Proof. It is sufficient to show that the conditions in Lemma 4.1 are fulfilled. It should be noted that although the idea of this proof is similar to the one in [37], some different multipliers are employed to get certain estimates so as to deal with the transmission conditions in the present paper.

    By the argument of the proof of Theorem 2.1, it is easy to see that, in the present case, there is no eigenvalue of A on the imaginary axis, that is, the condition (11) in Lemma 4.1 holds. Now we proceed to prove condition (12) and to deduce the exponential decay rate.

    If condition (12) is not fulfilled, then there exists a sequence of real numbers σn such that the corresponding resolvents Tn=(iσnIA)1 satisfy TnH,n. By Banach-Steinhaus theorem, there exists FH such that

    TnF=(iσnIA)1F=˜Ψn,in H,n.

    Thus,

    (iσnIA)˜Ψn˜ΨnH=F˜ΨnH0,in H,n.

    So there exists a sequence Φn=(Un,Vn,Θn)D(A), with ΦnH=1, where Un=(unj)Nj=1,Vn=(vnj)Nj=1,Θn=(θnj)N1j=1, and a sequence σnR with σn such that

    limn(iσnIA)ΦnH=0,

    namely,

    iσnunjvnj0,  inH1(0,j),j=1,2,,N, (13)
    iσnvnjunj,xx+αjθnj,x0,  inL2(0,j),j=1,2,,N1, (14)
    iσnθnjθnj,xx+βjvnj,x0,  inL2(0,j),j=1,2,,N1, (15)
    iσnvnNunN,xx0,  inL2(0,N). (16)

    Note that ((iσnA)Φn,Φn)H=N1j=1αjβjj0|θnj,x|2dx0. Thus,

    θnj,x0,inL2(0,j),j=1,2,,N1 (17)

    and hence by Poincaré inequality, we get θnj0, in L2(0,j),j=1,2,,N1. Moreover, by the Gagliardo-Nirenberg inequality (see [25]), we get θnjLd1θnj,x12θnj12+d2θnj0, which implies that

    θnj(0)0,j=1,2,,N1. (18)

    Removing θnj,x in (14) and substituting (13) into (14), we get

    iσnvnjunj,xx0,inL2(0,j),j=1,2,,N1. (19)

    and taking the inner product of (19) with xunj,x in L2(0,j), we have

    (σ2nunj,xunj,x)(unj,xx,xunj,x)0,j=1,2,,N1.

    Note that

    2(σ2nunj,xunj,x)=σ2nunj(j)¯junj(j)(σ2nunj,unj)=(σ2nunj,unj),
    2(unj,xx,xunj,x)=unj,x(j)¯junj,x(j)(unj,x,unj,x).

    Hence, unj,x(j),j=1,2,,N1 are bounded. Similarly, taking the inner product of (19) with (xj)unj,x in L2(0,j) yields

    (σ2nunj,(xj)unj,x)(unj,xx,(xj)unj,x)0,j=1,2,,N1.

    Hence,

    2(σ2nunj,(xj)unj,x)2(unj,xx,(xj)unj,x)=σ2nunj(0)¯junj(0)(σ2nunj,unj)unj,x(0)¯junj,x(0)+(unj,x,unj,x). (20)

    Thus, unj,x(0) and σnunj(0) are bounded.

    Dividing (15) by iσn, together with (13), we get θnj,xx/(iσn)+βjunj,x0,j=1,2,,N1. Hence, θnj,xx/iσn is bounded in L2(0,j). Then taking the L2product of the above with unj,x yields

    (θnj,xxiσn,unj,x)+(βjunj,x,unj,x)0,j=1,2,,N1. (21)

    Integrating the above by parts, we have

    θnj,x(j)iσn¯unj,x(j)+θnj,x(0)iσn¯unj,x(0)+(θnj,xiσn,unj,xx)+(βjunj,x,unj,x)0,j=1,2,,N1. (22)

    Note that dividing (14) by iσn, we obtain the boundedness of unj,xx/(iσn) in L2(0,j). Thus,

    (θnj,xiσn,unj,xx)0,j=1,2,,N1.

    By the Gagliardo-Nirenberg inequality,

    θnj,xLiσnd1θnj,xxiσn12θnj,x12+d2θnj,xiσn0, (23)

    we have θnj,x(j)/(iσn),θnj,x(0)/(iσn)0,j=1,2,,N1. By the boundedness of unj,x(j) and unj,x(0), together with (22), we have

    unj,x0,   inL2(0,j),   j=1,2,,N1. (24)

    Thus, unj0,inH1(0,j),j=1,2,,N1. Dividing (14) by iσn and taking the product of the obtained identity with vnj from 0 to j yields

    (vnj,vnj)(unj,xxiσn,vnj)0,j=1,2,,N1,

    and hence (vnj,vnj)(unj,xx/iσn,iσunj)0,j=1,2,,N1.

    Integrating by parts, we have

    (vnj,vnj)unj,x(0)unj(0)(unj,x,unj,x)0,j=1,2,,N1. (25)

    Note that unj(0)0,j=1,2,,N1, due to unj,unj,x0,inL2(0,j),j=1,2,,N1 and Gagliardo-Nirenberg inequality. Hence, due to the boundedness of unj,x(0),

    vnj0,inL2(0,j),j=1,2,,N1. (26)

    Then from (20), we have

    unj,x(0),σnunj(0)0,j=1,2,,N1. (27)

    On the other hand, on the segment (0,N), unN(0),σnunN(0)0 due to (18), (27) and the transmission condition at x=0.

    Taking the inner product of (16) with (xN)unN,x, we have

    (iσnunN,(xN)unN,x)(unN,xx,(xN)unN,x)0. (28)

    Note that

    2(iσnunN,(xN)unN,x)=iσnunN(0)¯NwnN(0)+(vnN,iσnunN)(vnN,vnN),
    2(unN,xx,(xN)unN,x)=unN,x(0)¯NunN,x(0)(unN,x,unN,x)(unN,x,unN,x).

    Thus,

    (iσnvnN,(xN)unN,x)(wnj,xx,(xN)unN,x)=(vnN,vnN)+(unN,x,unN,x)0. (29)

    Hence,

    unN,x,vnN0,inL2(0,N). (30)

    Thus, by (17), (24) and (26), we get Φn0,inH, which is in contradiction with Φn=1. The desired result follows.

    In this subsection, we shall show that if more than one purely elastic undamped rod is involved in the network, the exponential decay rate does not hold.

    Proposition 2. If NN1>1 the network (5) lacks the property of exponential decay.

    Proof. Note that from Lemma 4.1, it is sufficient to show that the norm of the resolvent operator of system (5) along the imaginary axis is necessarily unbounded when NN1>1.

    To do it we consider the resolvent problem

    (λIA)U=F,λ=iσ,σR, (31)

    where U=((uj)Nj=1,(vj)Nj=1,(θj)N1j=1),F=((fj)Nj=1,(gj)Nj=1,(ηj)N1j=1), which is equivalent to

    {λujvj=fj,j=1,2,,N,λvj(uj,xxαjθj,x)=gj,j=1,2,N1,λθj(θj,xxβjvj,x)=ηj,j=1,2,N1,λvjuj,xx=gj,j=N1+1,N1+2,,N. (32)

    Choose fj=0,j=1,2,,N, and gj=ηj=0,j=1,2,N1. Then we have

    {λujvj=0,j=1,2,,N,λvj(uj,xxαjθj,x)=0,j=1,2,N1,λθj(θj,xxβjvj,x)=0,j=1,2,N1,λvjuj,xx=gj,j=N1+1,N1+2,,N. (33)

    Using explicit representation formulas we get

    uj(x)=u(0)sinσjsinσ(jx)+sinσ(jx)σsinσjj0gj(js)sinσ(js)ds         1σjx0gj(js)sinσ(jxs)ds,j=N1+1,N1+2,,N. (34)

    In order to calculate uj,θj,j=1,2,,N1, let us consider the following coupled system.

    {λujvj=0,j=1,2,N1,λvj(uj,xxαjθj,x)=0,j=1,2,N1,λθj(θj,xxβjvj,x)=0,j=1,2,N1,

    that is,

    {λ2uj(uj,xxαjθj,x)=0,j=1,2,N1,λθj(θj,xxβjλuj,x)=0,j=1,2,N1. (35)

    We rewrite (35) in the vector form

    dYj(x)dx=AjYj(x),j=1,2,,N1,

    where Yj(x)=(uj,uj,xλ,θj,θj,xλ)T,Aj=[0λ00λ00αj000λ0βjλ10].

    Set dj,1=λ[λ+αjβj+1+(λ+αjβj+1)24λ]2,dj,2=λ[λ+αjβj+1(λ+αjβj+1)24λ]2. In order to diagonalize the matrix Aj, we employ the transformation Yj:=PjZj,j=1,2,,N1, where

    Pj=[11˜bj˜bjdj,1λdj,1λ˜bjdj,2λ˜bjdj,2λ˜aj˜aj11˜ajdj,1λ˜ajdj,1λdj,2λdj,2λ]

    and

    ˜aj=1αjdj,1(λ2+dj,1),˜bj=1βdj,2(1+dj,2λ).

    We then have dZj(x)dx=P1jAjPjZj=[dj,10000dj,10000dj,20000dj,2]Zj.

    In this way we obtain the following system

    Zj(x)=[edj,10000edj,10000edj,20000edj,2]Z(0),

    and

    Yj(x)=PjZj(x)=Pj[edj,10000edj,10000edj,20000edj,2]P1jYj(0). (36)

    By the boundary and transmission conditions in (5), together with (34), we get the following estimate (the technical details are given as in Appendix):

    σu(0)=Nj=N1+1Nkjk=N1+1sinσkcosσjj0gj(js)sinσ(js)dsNk=N1+1sinσkN1j=1(icosh(iσ+αjβj2)jsinh(iσ+αjβj2)j+O(1σ))Nj=N1+1Nkjk=N1+1sinσkcosσjNj=N1+1Nk=N1+1sinσkj0gj(js)cosσ(js)dsNk=N1+1sinσkN1j=1(icosh(iσ+αjβj2)jsinh(iσ+αjβj2)j+O(1σ))Nj=N1+1Nkjk=N1+1sinσkcosσj. (37)

    Choosing σnN1+1=nπ,n+, then sinσnN1+1=0. Since the irrational numbers always can be approximated by rational ones, we can find a subsequence σnk, such that

    sinσnkj0,j=N1+2,N1+3,,N.

    In fact, note that there is always a sequence of rational numbers pnmqnm,pnm,qnmZ+,n=1,2, such that |mN1+1pnmqnm|0,m=N1+2,N1+3,,N,n. Thus choose nk=Nm=N1+2qkm. Then we have

    sinσnkj=sin(σnkN1+1jN1+1)=sin(π(Nm=N1+2qkm)jN1+1)              sin(π(Nm=N1+2qkm)pkjqkj)=0,k,j=N1+2,N1+3,,N.

    Set gnkN1+1(s):=sinσnk(N1+1s),gnkj(s)=0,j=N1+2,N1+3,,N. Then we have

    N1+10gnkN1+1(js)sinσnk(js)dsN1+12,nk,
    N1+10gnkN1+1(js)cosσnk(js)ds0,nk.

    Thus, from (37), we get

    σnkunk(0)N1+12,nk.

    Hence, by (34), we have

    iσnkunkj(x),inL2(0,j),j=N1+2,N1+3,,N,

    due to the fact that sinσnkj0,j=N1+2,N1+3,,N. Note that vnkj=iσnkunkj(x). Therefore,

    vnkj,inL2(0,j),j=N1+2,N1+3,,N.

    Summarising the developments above we find a sequence (σn,Fn), where FnH and FnH is bounded, satisfying

    (iσnIA)1FnH,σn+,

    which implies the lack of exponential decay rate of system (5). The proof is complete.

    This subsection is devoted to get the lower bounds on the polynomial decay rate as stated in Theorem 2.4.

    We need the following result from [6] (see also [26]).

    Lemma 4.2. A C0 semigroup etA of contractions on a Hilbert space satisfies

    etAU0Ct1U0D(A),U0D(A),t

    for some constant C>0, if and only if the following conditions hold:

    1).{iβ|βR}ρ(A);

    2).lim sup|β|1β(iβA)1<.

    Proposition 3. The decay rate of the energy of system (5) can be, at most, polynomial of order t1.

    Proof. From Lemma 4.2, it is sufficient to show that there exists at least one sequence (σn,Fn) such that σn+,n and

    (iσnIA)1Fn>˜Cσ2n, (38)

    where FnH and FnH is bounded, ˜C being some positive constant.

    For simplicity, we consider the case NN1=2, that is, there are two purely elastic rods involved in the network. Based on the proof of Proposition 2, we choose the sequence (σnk,Fnk) satisfying (38) in the following three steps.

    Step 1). Choose σn=1N1+1+N1+2nπ. Thus we get sinσn(N1+1+N1+2)=0.

    Since

    sinσN1+1=sinσ(N1+1+N1+2N1+2)                =sinσ(N1+1+N1+2)cosσ(N1+2)cosσ(N1+1+N1+2)sinσ(N1+2),

    we have

    sinσnN1+1=±sinσnN1+2,cosσnN1+1=±cosσnN1+2.

    Step 2). Choose fj=0,j=1,2,,N, gj=ηj=0,j=1,2,N1 and gnN1+1(s)=sinσn(N1+1s),gnN1+2(s)=0 in (32).

    Substituting the above into (37), we get

    σnu(0)=±sinσnN1+2cosσnN1+1N1+10sinσnssinσn(N1+1s)ds(sinσnN1+2)2N1j=1(icosh(iσn+αjβj2)jsinh(iσn+αjβj2)j+O(1σn))±(sinσnN1+2)2N1+10sinσnscosσn(N1+1s)ds(sinσnN1+2)2N1j=1(icosh(iσn+αjβj2)jsinh(iσn+αjβj2)j+O(1σn)). (39)

    Note that

    N1+10sinσnssinσn(N1+1s)dsN1+1cosσnN1+12,n,
    N1+10sinσnscosσnk(js)dsN1+1sinσnN1+12,n,

    and

    N1j=1(icosh(iσn+αjβj2)jsinh(iσn+αjβj2)j+O(1σn))=O(1)R.

    Hence,

    |σnu(0)|=|N1+1cos2σnN1+22sinσnN1+2±N1+1sinσnN1+22|O(1). (40)

    Step 3). Note that we have chosen in Step 1) that σn=1N1+1+N1+2nπ and thus

    sinσnN1+2=sin(N1+2N1+1+N1+2(N1+1+N1+2)σn)=sin(N1+2N1+1+N1+2nπ).

    By Dirichlet theorem, we can always find an infinite subsequence nkZ such that

    |||N1+2N1+1+N1+2nk|||<1nk,

    and hence |sinσnkN1+2|=|sin(N1+2N1+1+N1+2nkπ)|<πnk.

    Thus, there exists a positive constant C

    |σnku(0)|>Cnk. (41)

    Note that from (33) and (34), we get vnkN1+2(x)=iσnkunkN1+2(x) and

    uN1+2(x)=u(0)sinσN1+2sinσ(N1+2x)+sinσ(N1+2x)σsinσN1+2N1+20gN1+2(N1+2s)sinσ(N1+2s)ds1σN1+2x0gN1+2(N1+2s)sinσ(N1+2xs)ds.

    Hence, by (41), there exists a positive constant ˆC satisfying

    vnkN1+2>ˆCn2k.

    Summarizing above, we have found a sequence (σnk,Fnk), where

    σnk=1N1+1+N1+2nkπ,Fnk=F=((fj)Nj=1,(gj)Nj=1,(ηj)N1j=1),

    in which fj=0,j=1,2,,N, gj=ηj=0,j=1,2,N1 and gnN1+1(s)=sinσn(N1+1s),gnN1+2(s)=0, such that

    (iσnkIA)1Fnk>˜Cσ2nk,σnk,

    where ˜C>0 is some constant. The proof is complete.

    Remark 5. For general case NN1>2, We can prove that the networks also satisfy (38) by modifying the above construction slightly. Indeed, choose gnN1+1(s)=sinσn(N1+1s),gnj(s)=0,Nj>N1+1 in step 2), and choose nkZ such that |||N1+2N1+1+N1+2nk|||<1nk, and sinσnkj in step 3). Thus, by the same discussion as above, the desired result follows.

    In subsection 4.2 we proved that if there are more than one purely elastic rods involved in the network (5), the system can not achieve the exponential decay. We have also shown that the decay rate can be, at most, polynomial of order t^{-1}.

    The mutual-irrationality of the radii of \ell_j,\; j=N_1+1,N_1+2,\cdots,N in Theorem 2.1 is a sufficient and necessary condition for strong stability. So, for the rest of this section, we always assume this condition to be fulfilled.

    Now, we prove effective and explicit polynomial decay results when N-N_1>1, and \ell_j,\; j=N_1+1,N_1+2,\cdots,N satisfy the conditions (S) in Definition 2.3.

    Proof of Theorem 2.4. Similarly to the proof of Theorem 2.2, we argue by contradiction, on the basis of Lemma 4.2.

    Let \Phi_n be a sequence such that \Phi^n=(U^n, V^n, \Theta^n)\in\mathcal{D}(\mathcal{A}), with \|\Phi^n\|_\mathcal{H}=1, where U^n=(u_j^n)_{j=1}^N,\; V^n=(v_j^n)_{j=1}^N,\; \Theta^n=(\theta_j^n)_{j=1}^{N_1}, and a sequence \sigma_n\in\mathbb{R} with \sigma_n\to \infty such that

    \lim\limits_{n\to\infty} \sigma_n^{2(1+\epsilon)}\|(i\sigma_n I-\mathcal{A})\Phi^n\|_{\mathcal{H}}=0,

    where \epsilon is given as in Theorem 2.4. Hence,

    \sigma_n^{2(1+\epsilon)}(i\sigma_n u_j^n-v_j^n)=f_j^n \to 0, \; in\; H^1(0, \ell_j),j=1,2,\cdots,N, \label{nnss1+} (42)
    \sigma_n^{2(1+\epsilon)}(i\sigma_n v_j^n- u^n_{j,xx}+\alpha_j\theta^n_{j,x})=g_j^n \to 0,\; in\; L^2(0, \ell_j), j=1,2,\cdots, N_1, \label{nnss2+} (43)
    \sigma_n^{2(1+\epsilon)}(i\sigma_n \theta_j^n-\theta^n_{j,xx}+\beta_j v^n_{j,x})=y_j^n \to 0,\; in\; L^2(0, \ell_j), j=1,2,\cdots,N_1,\label{nnss3+} (44)

    and

    \sigma_n^{2(1+\epsilon)}( i\sigma_n v_j^n-u^n_{j,xx})=g_j \to 0,\; in\; L^2(0, \ell_j), j=N_1\!+\!1,N_1\!+\!2,\!\cdots\!,N. \label{nnss4+} (45)

    Note that (\sigma_n^{2(1\!+\!\epsilon)}(i\sigma_n\!-\!\mathcal{A})\Phi^n, \Phi^n)_{\mathcal{H}}\!=\!-\sum_{j=1}^{N_1}\alpha_j\beta_j \int_0^{\ell_j}|\sigma_n^{(1\!+\!\epsilon)}\theta^n_{j,x}|^2dx\!\to\! 0.

    Hence,

    \sigma_n^{(1+\epsilon)}\theta^n_{j,x}\to 0, in\; L^2(0, \ell_j),\; j=1,2,\cdots, N_1. (46)

    Thus, by Poincaré inequality,

    \sigma_n^{(1+\epsilon)}\theta^n_{j}\to 0, in\; L^2(0, \ell_j), j=1,2,\cdots,N_1.

    Similar to the proof of Theorem 2.2, we can get that

    \sigma_n^{(1+\epsilon)}u_{j,x}^n(x),\; \sigma_n^{(1+\epsilon)} v_j^n(x),\; \sigma_n^{(1+\epsilon)} \theta_j^n(x)\to 0,\; in\; L^2(0, \ell_j),\; j=1,2,\cdots,N_1, (47)
    \sigma_n^{(1+\epsilon)} u_{j,x}^n(0),\; \sigma_n^{(1+\epsilon)} \sigma_n u_j^n(0),\; \sigma_n^{(1+\epsilon)}\theta_j^n(0)\to 0,\; j=1,2,\cdots,N_1. (48)

    Thus, by the transmission conditions at the common node, we get

    \sigma_n^{(1+\epsilon)} \sigma_n u_j^n(0)\to0,\; j=N_1+1,N_1+2,\cdots,N, \label{ss-nnss5+} (49)
    \sigma_n^{(1+\epsilon)}\sum\limits_{j=N_1+1}^N u_{j,x}^n(0)\to0.\label{ss-nnss6+} (50)

    Then we will prove that \Phi^n\to 0 in \mathcal{H}, which leads to a contradiction. To do this, we discuss it by the following two cases:

    Case 1). Assume that \sin \sigma_n\ell_j\neq0,\; j=N_1+1, N_1+2,\cdots,N.

    By (42), (45), we can easily get that u_{j,xx}^n+\sigma_n^2 u_j=-\frac{g_j^n+i\sigma_nf_j^n}{ \sigma_n^{2(1+\epsilon)}}.

    Thus a direct calculation yields

    u_j^n(x) = \frac{u^n(0)}{\sin\sigma_n\ell_j}\sin\sigma_n(\ell_j-x)\\ \ \ \ \ \ \ \ \ \ \ +\frac{\sin\sigma_n(\ell_j-x)}{\sigma_n\sin\sigma_n\ell_j}\int_0^{\ell_j}\frac{g_j^n(\ell_j-s)+i\sigma_nf_j^n(\ell_j-s)}{\sigma_n^{2(1+\epsilon)}}\sin(\sigma_n(\ell_j-s))ds\\ \ \ \ \ \ \ \ \ \ -\frac{1}{\sigma_n}\int_0^{\ell_j-x}\frac{g_j^n(\ell_j-s)+i\sigma_n f_j^n(\ell_j-s)}{\sigma_n^{2(1+\epsilon)}}\sin\sigma_n(\ell_j-x-s)ds,\\ \ \ \ \ \ \ \ \ \ j=N_1+1, N_1+2,\cdots,N.

    Hence,

    u_{j,x}^n(0) = -\frac{\sigma_nu^n(0)\cos\sigma_n\ell_j}{\sin\sigma_n\ell_j} \\ \ \ \ \ \ \ \ \ \ -\frac{\cos\sigma_n\ell_j}{\sin\sigma_n\ell_j}\int_0^{\ell_j}\frac{g_j^n(\ell_j-s)+i\sigma_nf_j^n(\ell_j-s)}{\sigma_n^{2(1+\epsilon)}}\sin(\sigma_n(\ell_j-s))ds \\ \ \ \ \ \ \ \ \ \ +\int_0^{\ell_j}\frac{g_j^n(\ell_j-s)+i\sigma_n f_j^n(\ell_j-s)}{\sigma_n^{2(1+\epsilon)}}\cos\sigma_n(\ell_j-s)ds, \\ \ \ \ \ \ \ \ \ \ j=N_1+1, N_1+2,\cdots,N. (51)

    It is easy to show that

    \int_0^{\ell_j}[{g_j^n(\ell_j-s)+i\sigma_n f_j^n(\ell_j-s)}]\sin\sigma_n(\ell_j-s)ds \to 0, \\ \int_0^{\ell_j}[{g_j^n(\ell_j-s)+i\sigma_n f_j^n(\ell_j-s)}]\cos\sigma_n(\ell_j-s)ds \to 0.\label{ssnn6} (52)

    Set

    \gamma_n=\max\limits_{j=N_1+1,N_1+2,\cdots,N}\prod\limits_{i=N_1+1,\; i\neq j}^N\limits|\sin(\sigma_n\ell_i)|,\; \forall n\geq 1. (53)

    When \ell_j,\; j= N_1+1,N_1+2,\cdots,N satisfy the conditions (S) as given in Definition 2.3, by Corollary A.10 in [11], we get

    \gamma_n\geq \frac{c_\epsilon}{\sigma_n^{1+\epsilon}},\; \forall \epsilon > 0. (54)

    For any given sequence \sigma_n, assume that there exists j_0^n(N_1+1 \leq j^{n}_0\leq N) such that

    \gamma_n=\prod\limits_{i=N_1+1,\; i\neq j^n_0}^N\limits|\sin(\sigma_n\ell_i)|.

    Hence, for any j\neq j^n_0,\; N_1+1\leq j\leq N, we have

    {|\sin\sigma_n\ell_{j}|}\geq \gamma_n,

    which implies that

    \frac{1}{|\sin\sigma_n\ell_{j}|}\leq \frac{1}{\gamma_n}\leq \frac{1}{c_\epsilon}\sigma_n^{1+\epsilon}.

    Therefore, by (49) and (52) together with (51), we get that

    u_{j,x}^n(0)\to 0,\; N_1+1\leq j\leq N,\; j\neq j^n_0, n\to \infty. (55)

    Then by the transmission condition (50), we obtain that

    u_{j,x}^n(0)\to 0, \; N_1+1\leq j\leq N, n\to\infty. (56)

    Taking the inner product of (45) with (x-\ell_j)u^n_{j,x},\; j=N_1+1,\; N_1+2,\cdots,N, respectively,

    (i\sigma_n v^n_j,(x-\ell_j)u^n_{j,x})-({u^n_{j,xx}},(x-\ell_j)u^n_{j,x})\to 0, j=N_1+1,N_1+2,\cdots,N. (57)

    We have

    2\Re(i\sigma_n v^n_j,(x-\ell_j)u^n_{j,x})= i\sigma_n v^n_j(0)\overline{\ell_ju^{n}_j(0)}+(v^n_j,i\sigma_n u^n_j)\to (v^n_j,v^n_j),
    2\Re({u^n_{j,xx}},(x-\ell_j)u^n_{j,x})=u^n_{j,x}(0)\overline{\ell_j u^n_{j,x}(0)}-(u^n_{j,x}, u^n_{j,x}) \to -(u^n_{j,x}, u^n_{j,x}).

    Hence,

    \Re(i\sigma_n v^n_j,(x-\ell_j)u^n_{j,x})-\Re({u^n_{j,xx}},(x-\ell_j)u^n_{j,x})=(v^n_j,v^n_j)+(u^n_{j,x}, u^n_{j,x})\to0. (58)

    Therefore,

    u^n_{j,x},\; v^n_j\to 0, in\;\; L^2(0, \ell_j),\; j=N_1+1,N_1+2,\cdots,N. (59)

    Thus, by (47) and (59), we have obtained

    \Phi^n=((u_j^n)_{j=1}^N, (v_j^n)_{j=1}^N, (\theta_j^n)_{j=1}^{N_1})\to 0, in\; \mathcal{H}, n\to\infty,

    which contradicts \|\Phi^n\|_\mathcal{H}=1.

    Now let us continue to consider the second step:

    Case 2). Assume that there exists j_0,\; N_1+1\leq j_{0}\leq N such that \sin \sigma_n\ell_{j_0}=0.

    Note that there exists at most one j_0^n,\; N_1+1\leq j_0\leq N satisfying \sin \sigma_n\ell_{j_0}=0 due to the irrationality of the mutual ratii between \ell_j,\; N_1+1\leq j\leq N. In this case,

    \gamma_n=\prod\limits_{i=N_1+1,\; i\neq j_0^n}^N\limits|\sin(\sigma_n\ell_i)|,

    where \gamma_n is given as in (53). Thus, by the same arguments as in Step\; 1), we can get

    u_{j,x}^n(0)\to 0,\; N_1+1\leq j\leq N,\; j\neq j_0^n, n\to \infty,

    which together with the transmission condition (50), implies

    u_{j,x}^n(0)\to 0,\; N_1+1\leq j\leq N, n\to \infty.

    Then by (57)-(59) in Step \; 1), we get

    u^n_{j,x},\; v^n_j\to 0, in\;\; L^2(0, \ell_j),\; j=N_1+1,N_1+2,\cdots,N.

    Hence, the same contradiction holds as in Step \; 1).

    Therefore, by Lemma 4.2, we get the polynomial decay rate of system (5), that is

    E(t)\leq C_\epsilon{t^{-\frac{1}{1+\epsilon}}}\|(u^{(0)}, u^{(1)},\theta^{(0)})\|_{\mathcal{D}(\mathcal{A})}^2,\; \forall t\geq 0.

    Note that in Proposition 3, we have proved that the polynomial decay order of the energy of system (5) is at most t^{-1}. Hence, t^{-\frac{1}{1+\epsilon}} is the nearly sharp. The proof of Theorem 2.4 is complete.

    In the last section, we have shown that, when N-N_1>1, namely, when the network (5) involves more than one purely elastic rod, the system can not achieve the exponential decay rate. Then we further derived the nearly optimal polynomial decay rate, if the conditions (S) in Definition 2.3 are fulfilled.

    In fact, the discussion of the proof of Theorem 2.4 (see subsection 4.3), shows that the slow decay rate of the system is determined by \gamma_n given as (53), which depends on the property of the lengths of the purely elastic rods entering in the network. Hence, other more general slow decay rates can be obtained when different conditions on \ell_j,\; j=N_1+1,N_1+2,\cdots,N are imposed.

    To do this, let us introduce the following definition on the irrational sets (see p. 209 in [11]), which is deduced from [7] (see Theorem Ⅰ, p. 120).

    Definition 5.1. (Theorem [7], [11]) 1. Set B_\epsilon: for all \epsilon>0 there exists a set B_\epsilon\subset\mathbb{R}, such that the Lebesgue measure of \mathbb{R}\setminus B_\epsilon vanishes, and a constant C_\epsilon>0 for which, if \xi\in B_\epsilon, then |||\xi m|||\geq \frac{C_\epsilon}{m^{1+\epsilon}} for all integer number m. All the algebraic irrational numbers belong to B_\epsilon.

    2. Set \mathcal{F}: the set of all real numbers \rho such that \rho\notin Q and so that its expansion as a continued fraction [0, a_1, a_2,\cdots, a_n, \cdots] is such that (a_n) is bounded. In particular \mathcal{F} is contained in the sets B_\epsilon for every \epsilon>0. (see p. 209 in [11] for more details). It contains all quadratic algebraic irrational numbers and also some transcendental numbers. This set \mathcal{F} has Lebesgue measure zero and is not denumerable.

    We have the following result:

    Corollary 1. For any (u^{(0)}, u^{(1)},\theta^{(0)})\in \mathcal{D}(\mathcal{A}), there always exists a constant C>0 such that the energy of network (5) satisfies

    E(t)\leq C_{\widetilde{s}}{t^{-\frac{1}{\widetilde{s}}}}\|(u^{(0)}, u^{(1)},\theta^{(0)})\|_{\mathcal{D}(\mathcal{A})}^2,\; \forall t\geq 0, (60)

    where \widetilde{s} is given as follows:

    \bullet if \frac{\ell_i}{\ell_j}\in B_\epsilon,\; i, j=N_1\!+\!1, N_1\!+\!2,\cdots,N, i\neq j, then \widetilde{s}\!=\!{N-N_1-1+\epsilon};

    \bullet if \frac{\ell_i}{\ell_j}\in \mathcal{F},\; i, j=N_1+1, N_1+2,\cdots,N,\; i\neq j, then \widetilde{s}=N-N_1-1.

    Proof. Since the proof is similar to the one of Theorem 2.4, we only give a sketch of it.

    If the decay rate is not fulfilled, there exists a sequence \Phi^n=(U^n, V^n, \Theta^n)\in\mathcal{D}(\mathcal{A}), with \|\Phi^n\|_\mathcal{H}=1, where U^n=(u_j^n)_{j=1}^N,\;V^n=(v_j^n)_{j=1}^N,\; \Theta^n=(\theta_j^n)_{j=1}^{N_1}, and a sequence \sigma_n\in\mathbb{R} with \sigma_n\to \infty such that

    \lim\limits_{n\to\infty} \sigma_n^{2\widetilde{s}}\|(i\sigma_n I-\mathcal{A})\Phi^n\|_{\mathcal{H}}=0,

    where \widetilde{s} is given as in Corollary 1.

    By Diophantine approximation (see [11]), different estimates can be gotten for \gamma_n, when the irrational numbers \ell_i/\ell_j belong to the sets B_\varepsilon and \mathcal{F} (as in as in Definition (5.1), respectively. By Corollary A. 10 in [11] (see also [39]), we get that

    (1) if \frac{\ell_i}{\ell_j}\in B_\epsilon,\; i, j=N_1+1, N_1+2,\cdots,N,\; i\neq j, then \gamma_n\geq c_\epsilon/\sigma_n^{N-N_1-1+\epsilon},\; n\geq 1,\; \epsilon>0;

    (2) if \frac{\ell_i}{\ell_j}\in \mathcal{F},\; i, j=N_1+1, N_1+2,\cdots,N,\; i\neq j, then \gamma_n\geq c/\sigma_n^{N-N_1-1},\; n\geq 1.

    Thus, proceeding as in the discussion of the proof of Theorem 2.4, finally we can get

    \Phi^n\to 0, in\; \mathcal{H}, n\to\infty.

    This contradicts \|\Phi^n\|_\mathcal{H}=1. Hence, the desired result follows.

    Remark 6. By Corollary 1, it is easy to see that when N-N_1=2, that is there are two purely elastic rods in the network, the system can achieve optimal polynomial decay rate t^{-1}, if \frac{\ell_{N_1+1}}{\ell_N}\in\mathcal{F}.

    More generally, by the similar proof as the one for Theorem 2.4, together with the so called M_{\log}-Theorem in [5], we obtain the general slow decay rate of system (5) as follows.

    Corollary 2. Set M_{\log}(s)=M(s)(\log(1+M(s))+\log(1+s)), where M(s): \mathbb{R}^+\rightarrow (0,\infty) is continuous and increasing.

    If

    \gamma_n\geq\frac{1}{\sqrt{M(\sigma_n)}},\; n > 0,

    where

    \gamma_n=\max\limits_{j=N_1+1,N_1+2,\cdots,N}\limits\prod\limits_{i=N_1+1,\; i\neq j}^N\limits|\sin(\sigma_n\ell_i)|,\; \forall \sigma_n\to\infty,

    then

    E(t)\leq C\frac{1}{(M_{\log}^{-1}(ct))^2}\|(u^{(0)}, u^{(1)},\theta^{(0)})\|_{\mathcal{D}(\mathcal{A})}^2,\; \forall t\geq 0, (61)

    where M_{log}^{-1}(\cdot) is the inverse function of M_{log}.

    Especially, if M(s)\leq ce^{as},\; a, c>0, then the network (5) achieves logarithmic decay rate.

    Proof. We can still use the proof of Theorem 2.4 to derive

    \limsup\limits_{|\sigma|\to\infty}\limits\frac{1}{(M(\sigma))}\|(i\sigma-\mathcal{A})^{-1}\| < \infty,

    which together with M_{\log}-Theorem in Batty and Duyckaerts [5] yields (61).

    Remark 7. From Corollary 2, we see that in order to obtain an explicit decay rate, it is very important to estimate the lower bound of \gamma_n. Some techniques such as Diophantine approximations can be used to estimate it. However, it is still open to get a sharp estimate for it.

    This section is devoted to present some numerical simulations on the dynamical behaviour of system (5) to support the results obtained above.

    The backward Euler method in time (time step: dt=0.01) and the Chebyshev spectral method in space (spatial grid size S=40) were employed, in a MatLab environment (see [38]).

    For simplicity, we assume that the star-shaped network consists of three edges, the lengths of which are given as

    \ell_1=1,\; \ell_2=2,\; \ell_3=1.

    The following cases are considered:

    Case A. N_1=3 (Three thermoelastic rods)

    In this case, all the three edges of the network are constituted by thermoelastic ones. First, choose the initial conditions as follows:

    \left\{ \begin{array}{l} u_1(x,0)=5\sin(\pi x), \; u_2(x,0)=5\sin(\frac{\pi x}{\ell_2}),\; u_3(x,0)=-5\sin(\pi x),\\ u_{1,t}(x,0)=9\sin(\pi x), \; u_{2,t}(x,0)=4\sin(\frac{\pi x}{\ell_2}),\; u_{3,t}(x,0)=-9\sin(\pi x),\\ \theta_1(x,0)=3\sin(\frac{1}{2}\pi x+\frac{\pi}{2}), \theta_2(x,0)=3\sin(\frac{\pi}{2\ell_2}x+\frac{\pi}{2}), \theta_3(x,0)=3\sin(\frac{1}{2}\pi x+\frac{\pi}{2}). \end{array} \right. (62)

    and the parameters in system (5):

    \alpha_1=1,\; \beta_1=1,\; \alpha_2=2,\; \beta_2=1,\;\alpha_1=3,\; \beta_3=1. (63)

    The dynamical behaviour of u_j,\; \theta_j,\; j=1,2,3 in the time interval [0,100] is given as in Figure A-1, 2, 3, 4, 5, 6. We observe the exponential decay rate of system (5).

    Figure A-1.  u_1(x,t).
    Figure A-2.  u_2(x,t).
    Figure A-3.  u_3(x,t).
    Figure A-4.  \theta_1(x,t).
    Figure A-5.  \theta_2(x,t).
    Figure A-6.  \theta_3(x,t).

    Case B. N_1=0 (Three purely elastic rods)

    In this case, the network consists of purely elastic rods, that is, no thermal-damping dissipates energy. The initial conditions are the same as in Case A. Solutions are plotted in Figure B-1, 2, 3, which confirms the conservative character of the system.

    Figure B-1.  u_1(x,t).
    Figure B-2.  u_2(x,t).
    Figure B-3.  u_3(x,t).

    Case C. N_1=2 (Two thermoelastic rods and one purely elastic one)

    In this case, the network is constituted by two thermoelastic rods and one purely elastic one. The initial conditions and parameters in system (5) are chosen as (62) and (63). We get the dynamical behaviour in Figure C-1, 2, 3, 4, 5.

    Figure C-1.  u_1(x,t).
    Figure C-2.  u_2(x,t).
    Figure C-3.  u_3(x,t).
    Figure C-4.  \theta_1(x,t).
    Figure C-5.  \theta_2(x,t).

    In these figures, we can see that the behaviour of each u_j,\; j=1,2,3 and \theta_j,\; j=1,2 are convergent to zero very fast. This shows numerically that the energy of system (5) decays to zero exponentially, which is consistent with the result on exponential decay rate in previous sections.

    Case D. N_1=1 (One thermoelastic rod and two purely elastic rods)

    In this case, the network consists of one thermoelastic rod and two purely elastic ones. The lengths of rods are still given as \ell_1=1,\; \ell_2=2,\; \ell_3=1. Simulations are plotted in Figure D-1, 2, 3, 4. We observe the lack of decay as predicted by the theory.

    Figure D-1.  u_1(x,t).
    Figure D-2.  u_2(x,t).
    Figure D-3.  u_3(x,t).
    Figure D-4.  \theta_1(x,t).

    Case E. N_1=1,\; \ell_2=\sqrt{2} (One thermoelastic rod and two purely elastic ones)

    In this case, the network is still constituted by one thermoelastic rod and two purely elastic ones. This time the length \ell_2=\sqrt{2}, which leads to \frac{\ell_2}{\ell_3}=\sqrt{2}\notin Q.

    In Figure E-1, 2, 3, 4, we observe a very slow decay rate. It implies that the energy of this system decays to zero but lacks exponential growth rate.

    Figure E-1.  u_1(x,t).
    Figure E-2.  u_2(x,t).
    Figure E-3.  u_3(x,t).
    Figure E-4.  \theta_1(x,t).

    Moreover, we presented Figure F-1, 2 to compare the decay rate of the energy for each case. Figure F-1 shows the dynamical behaviours of the logarithmic scale of the energy for Case A, B and C. Figure F-2 shows the dynamical behaviours of the energies for Case D and E. From these two figures, we can clearly see the behaviours of energies for each case respect time, which are consistent with our theoretical results obtained in this paper.

    Figure F-1.  Logarithmic scale of energy for Case A, B, C.
    Figure F-2.  Energy for Case D, E.

    This appendix is devoted to show how to get (37).

    Note that \theta_j(0)=\theta_k(0), u_j(0)=u_k(0) due to the continuity at the common node in system (5).

    Thus, by (36), for j=1,2,\cdots,N_1,

    u_j(x) = \frac{u(0)}{-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}}[-\sqrt{d_{j,2}}\cosh\sqrt{d_{j,1}}x+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}\cosh\sqrt{d_{j,2}}x] \\ \ \ \ \ \ \ \ \ \ +\frac{u_{j,x}(0)}{\sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}}[\sinh\sqrt{d_{j,1}}x-\widetilde{a}_j\widetilde{b}_j\sinh\sqrt{d_{j,2}}x] \\ \ \ \ \ \ \ \ \ \ +\frac{\theta(0)}{\sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}}[-\widetilde{b}_j\sqrt{d_{j,2}}\sinh\sqrt{d_{j,1}}x+\widetilde{b}_j\sqrt{d_{j,1}}\sinh\sqrt{d_{j,2}}x] \\ \ \ \ \ \ \ \ \ \ +\frac{\theta_{j,x}(0)}{-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}}[\widetilde{b}_j\cosh\sqrt{d_{j,1}}x-\widetilde{b}_j\cosh\sqrt{d_{j,2}}x],\\ \theta_j(x) = \frac{u(0)}{-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}}[-\sqrt{d_{j,2}}\widetilde{a}_j\sinh\sqrt{d_{j,1}}x+\widetilde{a}_j\sqrt{d_{j,1}}\sinh\sqrt{d_{j,2}}x] \\ \ \ \ \ \ \ \ \ \ +\frac{u_{j,x}(0)}{\sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}}[\widetilde{a}_j\cosh\sqrt{d_{j,1}}x-\widetilde{a}_j\cosh\sqrt{d_{j,2}}x] \\ \ \ \ \ \ \ \ \ \ +\frac{\theta(0)}{\sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}}[-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}\cosh\sqrt{d_{j,1}}x+\sqrt{d_{j,1}}\cosh\sqrt{d_{j,2}}x] \\ \ \ \ \ \ \ \ \ \ +\frac{\theta_{j,x}(0)}{-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}}[\widetilde{a}_j\widetilde{b}_j\sinh\sqrt{d_{j,1}}x-\sinh\sqrt{d_{j,2}}x].

    Thus, by the boundary condition u_j(\ell_j)=\theta_j(\ell_j)=0,\;j=1,2,\cdots,N_1, we get

    0 = \frac{u(0)}{-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}}[-\sqrt{d_{j,2}}\cosh\sqrt{d_{j,1}}\ell_j +\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}\cosh\sqrt{d_{j,2}}\ell_j] \\ \ \ \ \ \ \ \ \ \ +\frac{u_{j,x}(0)}{\sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}}[\sinh\sqrt{d_{j,1}}\ell_j-\widetilde{a}_j\widetilde{b}_j\sinh\sqrt{d_{j,2}}\ell_j] \\ \ \ \ \ \ \ \ \ \ +\frac{\theta(0)}{\sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}}[-\widetilde{b}_j\sqrt{d_{j,2}}\sinh\sqrt{d_{j,1}}\ell_j+\widetilde{b}_j\sqrt{d_{j,1}}\sinh\sqrt{d_{j,2}}\ell_j] \\ \ \ \ \ \ \ \ \ \ +\frac{\theta_{j,x}(0)}{-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}}[\widetilde{b}_j\cosh\sqrt{d_{j,1}}\ell_j-\widetilde{b}_j\cosh\sqrt{d_{j,2}}\ell_j],\\ 0 = \frac{u(0)}{-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}}[-\sqrt{d_{j,2}}\widetilde{a}_j\sinh\sqrt{d_{j,1}}\ell_j+\widetilde{a}_j\sqrt{d_{j,1}}\sinh\sqrt{d_{j,2}}\ell_j] \\ \ \ \ \ \ \ \ \ \ +\frac{u_{j,x}(0)}{\sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}}[\widetilde{a}_j\cosh\sqrt{d_{j,1}}\ell_j-\widetilde{a}_j\cosh\sqrt{d_{j,2}}\ell_j] \\ \ \ \ \ \ \ \ \ \ +\frac{\theta(0)}{\sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}}[-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}\cosh\sqrt{d_{j,1}}\ell_j+\sqrt{d_{j,1}}\cosh\sqrt{d_{j,2}}\ell_j] \\ \ \ \ \ \ \ \ \ \ +\frac{\theta_{j,x}(0)}{-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}}[\widetilde{a}_j\widetilde{b}_j\sinh\sqrt{d_{j,1}}\ell_j-\sinh\sqrt{d_{j,2}}\ell_j].

    Then a direct calculation yields, for j=1,2,\cdots,N_1,

    \left( \begin{array}{c} u_{j,x}(0)\\ \theta_{j,x}(0) \end{array}\right) =\left( \begin{array}{cc} C^j_{11}&C^j_{12}\\ C^j_{21}&C^j_{22} \end{array}\right)\left( \begin{array}{c} u(0)\\ \theta(0) \end{array}\right), (A.1)

    where

    C^j_{11} = \Big[{(\sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}})(-\widetilde{a}_j\widetilde{b}_j\sinh\sqrt{d_{j,1}}\ell_j\cosh\sqrt{d_{j,2}}\ell_j+\cosh\sqrt{d_{j,1}}\ell_j\sinh\sqrt{d_{j,2}}\ell_j)}\Big]\Big/D_j,\\ C^j_{12} = \Big[-\widetilde{a}_j\widetilde{b}_j^2\sqrt{d_{j,2}}\!-\!\widetilde{b}_j\sqrt{d_{j,1}}\!+\!\widetilde{b}_j\sqrt{d_{j,1}}(-\!\widetilde{a}_j\widetilde{b}_j\sinh\sqrt{d_{j,1}}\ell_j\sinh\sqrt{d_{j,2}}\ell_j\!+\!\cosh\sqrt{d_{j,1}}\ell_j\cosh\sqrt{d_{j,2}}\ell_j) \\ \ \ \ \ \ \ \ \ \ +\widetilde{b}_j\sqrt{d_{j,2}}(-\sinh\sqrt{d_{j,1}}\ell_j\sinh\sqrt{d_{j,2}}\ell_j+\widetilde{a}_j\widetilde{b}_j\cosh\sqrt{d_{j,1}}\ell_j\cosh\sqrt{d_{j,2}}\ell_j)\Big]\Big/D_j,\\ C^j_{21} = \Big[-\widetilde{a}_j^2\widetilde{b}_j\sqrt{d_{j,1}}-\widetilde{a}_j\sqrt{d_{j,2}} +\widetilde{a}_j\sqrt{d_{j,1}}(\widetilde{a}_j\widetilde{b}_j\cosh\sqrt{d_{j,1}}\ell_j\cosh\sqrt{d_{j,2}}\ell_j-\sinh\sqrt{d_{j,1}}\ell_j\sinh\sqrt{d_{j,2}}\ell_j) \\ \ \ \ \ \ \ \ \ \ +\widetilde{a}_j\sqrt{d_{j,2}}(\cosh\sqrt{d_{j,1}}\ell_j\cosh\sqrt{d_{j,2}}\ell_j-\widetilde{a}_j\widetilde{b}_j\sinh\sqrt{d_{j,1}}\ell_j\sinh\sqrt{d_{j,2}}\ell_j)\Big]\Big/D_j,\\ C^j_{22} = \Big[{(-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}})(\widetilde{a}_j\widetilde{b}_j\cosh\sqrt{d_{j,1}}\ell_j\sinh\sqrt{d_{j,2}}\ell_j -\sinh\sqrt{d_{j,1}}\ell_j\cosh\sqrt{d_{j,2}}\ell_j)}\Big]\Big/D_j,

    and

    D_j={-2\widetilde{a}_j\widetilde{b}_j-(1+\widetilde{a}_j^2\widetilde{b}_j^2)\sinh\sqrt{d_{j,1}}\ell_j\sinh\sqrt{d_{j,2}}\ell_j+2\widetilde{a}_j\widetilde{b}_j\cosh\sqrt{d_{j,1}}\ell_j\cosh\sqrt{d_{j,2}}\ell_j} .

    Then by the transmission condition \sum_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}\theta_{j,x}(0)=0, we get

    \sum\limits_{j=1}^{N_1}\frac{\alpha_j}{\beta_j}(C^j_{21} u(0)+C^j_{22}\theta(0))=0, (A.2)

    which implies that

    \theta(0)=- u(0)\frac{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{21}}{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{22}}.

    Hence by (A.1),

    u_{j,x}(0)=C^j_{11} u(0)\!+\!C_{12}^j\theta(0)=u(0)\left(C_{11}^j-C_{12}^j\frac{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{21}}{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{22}} \right),\; j=1,2,\cdots,N_1. (A.3)

    On the other hand, by (34),

    u_{j,x}(0) = -\frac{\sigma u(0)\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}-\frac{\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}\int_0^{\ell_{j}}g_{j}(\ell_{j}-s)\sin\sigma(\ell_{j}-s)ds \\ \ \ \ \ \ \ \ \ \ +\int_0^{\ell_{j}}g_j(\ell_{j}-s)\cos\sigma(\ell_{j}-s)ds,\; j=N_1+1,N_1+2,\cdots,N. (A.4)

    Then by the transmission condition \sum_{j=1}^N\limits u_{j,x}(0)=\sum_{j=1}^{N_1}\limits\alpha_j\theta_j(0), together with (A.3) and (A.4), we have

    \ \ \ \ \ \ \ \ \ u(0)\left( \sum\limits_{j=1}^{N_1} \Big(C_{11}^j-C_{12}^j\frac{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{21}}{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{22}}+\alpha_j \frac{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{21}}{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{22}}\Big) -\sum\limits_{j=N_1+1}^{N}\frac{\sigma\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}\right) \\ = \sum\limits_{j=N_1+1}^N\frac{\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}\!\int_0^{\ell_{j}}g_{j}(\ell_{j}\!-\!s)\!\sin\sigma(\ell_{j}\!-\!s)ds \\ \ \ \ \ \ \ \ \ \ -\sum\limits_{j=N_1+1}^N\int_0^{\ell_{j}}g_j(\ell_{j}\!-\!s)\cos\sigma(\ell_{j}\!-\!s)ds. (A.5)

    Hence,

    \sigma u(0)\!=\!\frac{\sum\limits_{j=N_1\!+\!1}^N\limits\!\Big(\!\frac{\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}\int_0^{\ell_{j}}g_{j}(\ell_{j}\!-\!s)\sin\sigma(\ell_{j}\!-\!s)ds\!-\!\int_0^{\ell_{j}}g_j(\ell_{j}\!-\!s)\!\cos\sigma(\ell_{j}\!-\!s)ds\!\Big)} {\frac{1}{\sigma}\sum\limits_{j=1}^{N_1}\limits \Big(C_{11}^j-C_{12}^j\frac{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{21}}{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{22}}+\alpha_j \frac{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{21}}{\sum\limits_{j=1}^{N_1}\limits\frac{\alpha_j}{\beta_j}C^j_{22}}\Big) -\sum\limits_{j=N_1+1}^{N}\limits\frac{\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}}. (A.6)

    As \sigma\to\infty, we have the following estimation:

    d_{j,1}=(i\sigma)^2+\alpha_j\beta_j(i\sigma)+\alpha_j\beta_j+O(\frac{1}{\sigma}), d_{j,2}=i\sigma-\alpha_j\beta_j+O(\frac{1}{\sigma}).

    Hence,

    \sqrt{d_{j,1}}=i\sigma+\frac{\alpha_j\beta_j}{2}+O(\frac{1}{\sigma}), \sqrt{d_{j,2}}=\sqrt{i\sigma}-\frac{\alpha_j\beta_j}{2\sqrt{i\sigma}}+O(\frac{1}{\sigma^{3/2}}),
    \widetilde{a}_j=\beta_j+O(\frac{1}{\sigma}), \widetilde{b}_j=-{\alpha_j} (i\sigma)^{-\frac{3}{2}}+O(\frac{1}{\sigma^{5/2}}).

    and

    -\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}}=-\sqrt{i\sigma}-\frac{\alpha_j\beta_j}{\sqrt{i\sigma}}+O(\frac{1}{\sigma^{3/2}}),
    \sqrt{d_{j,1}}-\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}=i\sigma+\frac{\alpha_j\beta_j}{2}+\frac{\alpha_j\beta_j}{\sqrt{i\sigma}}+O(\frac{1}{\sigma}).

    Substituting the above into (A.1), we get that

    C_{11}^j=\frac{(\!\sqrt{d_{j,1}}\!-\!\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,2}}\!)\cosh\sqrt{d_{j,1}}\ell_j}{(1+\widetilde{a}_j^2\widetilde{b}_j^2)\sinh\sqrt{d_{j,1}}\ell_j} \!+\!O(\!\frac{1}{\sigma^{3/2}}\!) \!=\!(i\sigma\!+\!\frac{\alpha_j\beta_j}{2}\!+\!\frac{\alpha_j\beta_j}{\sqrt{i\sigma}})\frac{\cosh\sqrt{d_{j,1}}\ell_j}{\sinh\sqrt{d_{j,1}}\ell_j}\!+\!O(\frac{1}{\sigma}),\\ (A.7)
    C_{12}^j = \frac{(\widetilde{b}_j\sqrt{d_{j,1}})\cosh\sqrt{d_{j,1}}\ell_j}{(1+\widetilde{a}_j^2\widetilde{b}_j^2)\sinh\sqrt{d_{j,1}}\ell_j} +O(\!\frac{1}{\sigma^{3/2}}\!)=-\frac{\alpha_j}{\sqrt{i\sigma}}\frac{\cosh\sqrt{d_{j,1}}\ell_j}{\sinh\sqrt{d_{j,1}}\ell_j}+O(\frac{1}{\sigma}), (A.8)
    C_{21}^j = -\frac{(\widetilde{a}_j\sqrt{d_{j,1}})}{(1+\widetilde{a}_j^2\widetilde{b}_j^2)} +O(\!\frac{1}{\sigma^{3/2}}\!)=-{i\beta_j\sigma}+\frac{\alpha_j\beta_j^2}{2}+O(\frac{1}{\sigma}), (A.9)
    C_{22}^j = -\frac{(-\sqrt{d_{j,2}}+\widetilde{a}_j\widetilde{b}_j\sqrt{d_{j,1}})}{(1+\widetilde{a}_j^2\widetilde{b}_j^2)} +O(\!\frac{1}{\sigma^{3/2}}\!) =\sqrt{i\sigma}+\frac{\alpha_j\beta_j}{\sqrt{i\sigma}}+O(\frac{1}{\sigma}). (A.10)

    Then by (A.6),

    \sigma u(0) = \frac{\sum\limits_{j=N_1+1}^N\limits\Big(\frac{\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}\int_0^{\ell_{j}}g_{j}(\ell_{j}-s)\sin\sigma(\ell_{j}-s)ds-\int_0^{\ell_{j}}g_j(\ell_{j}-s)\cos\sigma(\ell_{j}-s)ds\Big)} {\sum\limits_{j=1}^{N_1}\limits \Big(i\frac{\cosh(i\sigma+\frac{\alpha_j\beta_j}{2})\ell_j}{\sinh(i\sigma+\frac{\alpha_j\beta_j}{2})\ell_j}+O(\frac{1}{\sqrt{\sigma}})\Big) -\sum\limits_{j=N_1+1}^{N}\limits\frac{\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}} \\ = \frac{\sum\limits_{j=N_1+1}^N\limits\Big(\frac{\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}\int_0^{\ell_{j}}g_{j}(\ell_{j}-s)\sin\sigma(\ell_{j}-s)ds-\int_0^{\ell_{j}}g_j(\ell_{j}-s)\cos\sigma(\ell_{j}-s)ds\Big)} {\sum\limits_{j=1}^{N_1}\limits \Big(i\frac{\cosh(i\sigma+\frac{\alpha_j\beta_j}{2})\ell_j}{\sinh(i\sigma+\frac{\alpha_j\beta_j}{2})\ell_j}+O(\frac{1}{\sqrt{\sigma}})\Big) -\sum\limits_{j=N_1+1}^{N}\limits\frac{\cos\sigma\ell_{j}}{\sin\sigma\ell_{j}}} \\ = \frac{\sum\limits_{j=N_1+1}^N\limits\prod\limits_{\tiny\begin{array}{c}k\neq j\\ k=N_1+1\end{array}}^N\limits\sin\sigma\ell_k\cos\sigma\ell_j \int_0^{\ell_{j}}g_{j}(\ell_{j}-s)\sin\sigma(\ell_{j}-s)ds} {\prod\limits_{\tiny k=N_1+1}^N\limits\sin\sigma\ell_k\sum\limits_{j=1}^{N_1}\limits \Big(i\frac{\cosh(i\sigma+\frac{\alpha_j\beta_j}{2})\ell_j}{\sinh(i\sigma+\frac{\alpha_j\beta_j}{2})\ell_j}+O(\frac{1}{\sqrt{\sigma}})\Big) -\sum\limits_{j=N_1+1}^{N}\limits\prod\limits_{\tiny\begin{array}{c}k\neq j\\ k=N_1+1\end{array}}^N\limits\sin\sigma\ell_k\cos\sigma\ell_{j}} \\ \ \ \ \ \ \ \ \ \ -\!\frac{\sum\limits_{\tiny j=N_1+1}^N\limits\prod\limits_{\tiny k=N_1+1}^N\limits\sin\sigma\ell_k\int_0^{\ell_{j}}g_j(\ell_{j}-s)\cos\sigma(\ell_{j}-s)ds} {\prod\limits_{\tiny k=N_1+1}^N\limits\sin\sigma\ell_k\sum\limits_{j=1}^{N_1}\limits \Big(i\frac{\cosh(i\sigma+\frac{\alpha_j\beta_j}{2})\ell_j}{\sinh(i\sigma+\frac{\alpha_j\beta_j}{2})\ell_j}\!+\!O(\frac{1}{\sqrt{\sigma}})\Big) \!-\!\sum\limits_{\tiny j=N_1+1}^{N}\limits\!\prod\limits_{\tiny\begin{array}{c}k\neq j\\ k=N_1+1\end{array}}^N\limits\sin\sigma\ell_k\cos\sigma\ell_{j}}.

    Thus, (37) has been obtained.



    [1] B. B. Mandelbrot, The fractal geometry of nature, New York: W. H. Freeman, 1982.
    [2] Y. C. Kwun, M. Tanveer, W. Nazeer, K. Gdawiec, S. M. Kang, Mandelbrot and Julia sets via Jungck-CR iteration with s convexity, IEEE Access, 7 (2019), 12167–12176. doi: 10.1109/ACCESS.2019.2892013
    [3] A. Lakhtakia, V. V. Varadan, R. Messier, V. K. Varadan, On the symmetries of the Julia sets for the process z = z^p+ c, J. Phys. A-Math. Gen., 20 (1987), 3533.
    [4] P. Blanchard, R. L. Devaney, A. Garijo, E. D. Russell, A generalized version of the McMullen domain, Int. J. Bifurcat. Chaos, 18 (2008), 2309–2318. doi: 10.1142/S0218127408021725
    [5] T. Kim, Quaternion Julia set shape optimization, Comput. Graph. Forum, 34 (2015), 167–176.
    [6] V. Drakopoulos, N. Mimikou, T. Theoharis, An overview of parallel visualisation methods for Mandelbrot and Julia sets, Comput. Graph., 27 (2003), 635–646. doi: 10.1016/S0097-8493(03)00106-7
    [7] Y. Sun, L. Chen, R. Xu, R. Kong, An image encryption algorithm utilizing Julia sets and Hilbert curves, PloS one, 9 (2014), e84655. doi: 10.1371/journal.pone.0084655
    [8] M. Rani, V. Kumar, Superior julia set, Res. Math. Edu., 8 (2004), 261–277.
    [9] M. Rani, V. Kumar, Superior mandelbrot set, Res. Math. Edu., 8 (2004), 279–291.
    [10] M. Rani, R. Agarwal, Effect of stochastic noise on superior Julia sets, J. Math. Imaging Vis., 36 (2010), 63–68. doi: 10.1007/s10851-009-0171-0
    [11] M. Rani, R. Chugh, Julia sets and Mandelbrot sets in Noor orbit, Appl. Math. Comput., 228 (2014), 615–631.
    [12] S. M. Kang, A. Rafiq, A. Latif, A. A. Shahid, Y. C. Kwun, Tricorns and multicorns of-iteration scheme, J. Funct. Space., 2015 (2015), 417167.
    [13] K. Goyal, B. Prasad, Dynamics of iterative schemes for quadratic polynomial, AIP Conference Proceedings, 1897 (2017), 020031. doi: 10.1063/1.5008710
    [14] W. Nazeer, S. M. Kang, M. Tanveer, A. A. Shahid, Fixed point results in the generation of Julia and Mandelbrot sets, J. Inequal. Appl., 2015 (2015), 1–16. doi: 10.1186/1029-242X-2015-1
    [15] S. M. Kang, W. Nazeer, M. Tanveer, A. A. Shahid, New fixed point results for fractal generation in Jungck Noor orbit with-convexity, J. Funct. Space., 2015 (2015), 963016.
    [16] D. Li, M. Tanveer, W. Nazeer, X. Guo, Boundaries of filled julia sets in generalized jungck mann orbit, IEEE Access, 7 (2019), 76859–76867. doi: 10.1109/ACCESS.2019.2920026
    [17] Y. C. Kwun, M. Tanveer, W. Nazeer, M. Abbas, S. M. Kang, Fractal generation in modified Jungck-S orbit, IEEE Access, 7 (2019), 35060–35071. doi: 10.1109/ACCESS.2019.2904677
    [18] Y. C. Kwun, M. Tanveer, W. Nazeer, K. Gdawiec, S. M. Kang, Mandelbrot and Julia sets via Jungck-CR iteration with s-convexity, IEEE Access, 7 (2019), 12167–12176. doi: 10.1109/ACCESS.2019.2892013
    [19] M. Barnsley, Fractals everywhere, 2nd Edition, Academic Press, 1993.
    [20] R. L. Devaney, A first course in chaotic dynamical systems: Theory and experiment, CRC Press, 2018.
    [21] X. Liu, Z. Zhu, G. Wang, W. Zhu, Composed accelerated escape time algorithm to construct the general Mandelbrot sets, Fractals, 9 (2001), 149–154. doi: 10.1142/S0218348X01000580
    [22] N. Hussain, V. Kumar, M. A. Kutbi, On rate of convergence of Jungck-type iterative schemes, Abstr. Appl. Anal., 2013 (2013), 132626.
  • This article has been cited by:

    1. Kaïs Ammari, Farhat Shel, 2022, Chapter 4, 978-3-030-86350-0, 57, 10.1007/978-3-030-86351-7_4
    2. Kaïs Ammari, Farhat Shel, 2022, Chapter 2, 978-3-030-86350-0, 15, 10.1007/978-3-030-86351-7_2
    3. Victor Verbiţchi, Radu Cojocaru, Lia Nicoleta Boțilă, Cristian Ciucă, Ion Aurel Perianu, Considerations on the Ultimate Tensile Strength of Butt Welds of the EN AW 5754 Aluminium Alloy, Made by Friction Stir Welding (FSW), 2020, 1157, 1662-8985, 38, 10.4028/www.scientific.net/AMR.1157.38
    4. Vaibhav Mehandiratta, Mani Mehra, Günter Leugering, Existence and uniqueness results for a nonlinear Caputo fractional boundary value problem on a star graph, 2019, 477, 0022247X, 1243, 10.1016/j.jmaa.2019.05.011
    5. Vaibhav Mehandiratta, Mani Mehra, A difference scheme for the time-fractional diffusion equation on a metric star graph, 2020, 158, 01689274, 152, 10.1016/j.apnum.2020.07.022
    6. Victor Verbiţchi, Horia Florin Daşcău, Lia Nicoleta Boțilă, Radu Cojocaru, Experiments on Friction Stir Processing (FSP), with Tilted Parent Metal, 2021, 890, 1662-9795, 95, 10.4028/www.scientific.net/KEM.890.95
    7. Hai-E. Zhang, Gen-Qi Xu, Hao Chen, Min Li, Stability of a Variable Coefficient Star-Shaped Network with Distributed Delay, 2022, 35, 1009-6124, 2077, 10.1007/s11424-022-1157-x
    8. Mo Faheem, Arshad Khan, A collocation method for time‐fractional diffusion equation on a metric star graph with η \eta edges, 2023, 0170-4214, 10.1002/mma.9023
    9. Karim El Mufti, Rania Yahia, Polynomial stability in viscoelastic network of strings, 2022, 15, 1937-1632, 1421, 10.3934/dcdss.2022073
    10. Alaa Hayek, Serge Nicaise, Zaynab Salloum, Ali Wehbe, Exponential and polynomial stability results for networks of elastic and thermo-elastic rods, 2022, 15, 1937-1632, 1183, 10.3934/dcdss.2021142
    11. Mohammad Akil, Genni Fragnelli, Ibtissam Issa, The energy decay rate of a transmission system governed by the degenerate wave equation with drift and under heat conduction with the memory effect, 2024, 297, 0025-584X, 3766, 10.1002/mana.202300571
    12. Bienvenido Barraza Martínez, Jonathan González Ospino, Jairo Hernández Monzón, Analysis of the exponential stability of a beam-string-beam transmission problem with local damping on the string, 2024, 77, 09473580, 100988, 10.1016/j.ejcon.2024.100988
    13. Bienvenido Barraza Martinez, Jairo Hernandez Monzon, Gustavo Vergara Rolong, Exponential stability of a damped beam-string-beam transmission problem, 2022, 2022, 1072-6691, 30, 10.58997/ejde.2022.30
    14. Gonzalo Arias, Eduardo Cerpa, Swann Marx, Stability analysis of a linear system coupling wave and heat equations with different time scales, 2025, 543, 0022247X, 128923, 10.1016/j.jmaa.2024.128923
    15. R. Ch. Kulaev, A. A. Urtaeva, On the Existence of a Solution of a Boundary Value Problem on a Graph for a Nonlinear Equation of the Fourth Order, 2023, 59, 0012-2661, 1175, 10.1134/S0012266123090033
    16. R. Ch. Kulaev, A. A. Urtaeva, On the Existence of a Solution of a Boundary Value Problem on a Graph for a Nonlinear Equation of the Fourth Order, 2023, 59, 0374-0641, 1181, 10.31857/S0374064123090030
    17. Junping Nan, Weimin Hu, You-Hui Su, Yongzhen Yun, EXISTENCE AND STABILITY OF SOLUTIONS FOR A COUPLED HADAMARD TYPE SEQUENCE FRACTIONAL DIFFERENTIAL SYSTEM ON GLUCOSE GRAPHS, 2024, 14, 2156-907X, 911, 10.11948/20230202
    18. Vaibhav Mehandiratta, Mani Mehra, Space-time fractional parabolic equations on a metric star graph with spatial fractional derivative of Sturm-Liouville type: analysis and discretization, 2025, 1311-0454, 10.1007/s13540-025-00376-7
    19. Mohammad Akil, Genni Fragnelli, Sarah Ismail, Ahmet Özkan Özer, Exponential Stabilization of an Euler-Bernoulli Beam by a Heat Equation Involving Memory Effects, 2025, 0921-7134, 10.1177/09217134241310063
    20. Shengda Zeng, Moncef Aouadi, Decay with respect to boundary conditions in nonsimple thermoelasticity with microtemperatures*, 2025, 0149-5739, 1, 10.1080/01495739.2025.2466102
  • Reader Comments
  • © 2021 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(2475) PDF downloads(114) Cited by(9)

Figures and Tables

Figures(18)  /  Tables(6)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog