betaFeedback

WorksMathematics

Exponential boundary stabilization of the KdV–KdV system with time-varying delay: the nonlinear case

Abstract

We consider a Boussinesq system of KdV–KdV type on a bounded interval, with a boundary feedback that includes a damping term with a time-varying delayed term. Although the exponential stability of the linearized system was established in [1] by a Lyapunov approach, global existence and stabilization for the nonlinear problem remained open. Here, we prove that, for sufficiently small initial data and under a suitable restriction on the interval length, the nonlinear system admits a unique global solution whose energy decays exponentially. In particular, we provide an explicit decay rate and a quantitative smallness condition on the initial energy. Our approach combines a Lyapunov argument with well-posedness estimates that are uniform under time translations of the delay, allowing the solutions to be continued globally.

Introduction

The boundary stabilization of the KdV–KdV system with a time-varying delay was studied in [1], where exponential stability was established for the linearized system and local well-posedness was obtained for the nonlinear problem. In the present note, we improve this analysis by addressing the global existence and exponential stability of the nonlinear solutions for sufficiently small initial data.

Related results on nonlinear boundary stabilization with delay were obtained for the Hirota–Satsuma system in [2], where exponential decay for small initial data is established through Lyapunov and observability arguments. A useful structural feature of that system is the cancellation of the nonlinear terms in the natural energy balance, which preserves the dissipativity of the linearized problem. For the original KdV–KdV system, a cubic term of indefinite sign remains in this balance, and the corresponding energy estimate does not directly provide the bound needed for global continuation. In the present note, this difficulty is overcome by retaining the spatial dissipation in the Lyapunov estimate and absorbing the nonlinear contribution for sufficiently small initial energy.

Setting

Let L>0L > 0. We study the Boussinesq system of KdV–KdV type

{ηt+ωx+ωxxx+(ηω)x=0,ωt+ηx+ηxxx+ωωx=0,η(t,0)=η(t,L)=0,ω(t,0)=ω(t,L)=0,ωx(t,L)=−αηx(t,L)+βηx(t−τ(t),L),ηx(θ,L)=h0(θ),θ∈(−τ(0),0),(η,ω)(0,x)=(η0,ω0)(x).(1)\left\{ \begin{aligned} \eta_t+\omega_x+\omega_{xxx}+( \eta\omega)_x&=0,\\ \omega_t+\eta_x+\eta_{xxx}+\omega\omega_x&=0,\\ \eta(t,0)&=\eta(t,L)=0,\\ \omega(t,0)&=\omega(t,L)=0,\\ \omega_x(t,L)&=-\alpha\eta_x(t,L)+\beta\eta_x(t-\tau(t),L),\\ \eta_x(\theta,L)&=h_0(\theta),\qquad\theta\in(-\tau(0),0),\\ (\eta,\omega)(0,x)&=(\eta_0,\omega_0)(x). \end{aligned} \right. \tag*{(1)}

for t>0t>0 and x∈(0,L)x\in(0,L), which corresponds to system (1.7) in [1]. Here, we recall that α,β∈R\alpha,\beta\in\mathbb{R} are the feedback gains, τ\tau is the time-dependent delay, and h0h_0 prescribes the delayed state. For the derivation of the KdV–KdV system, we refer to [3]; boundary stabilization without delay is studied in [4, 5]. Throughout this note, we assume that the delay satisfies

{τ∈Wloc2,∞([0,∞)),0<τ∗≤τ(t)≤M,τ˙(t)≤d,0≤d<1.(2)\left\{ \begin{aligned} \tau&\in W_{\mathrm{loc}}^{2,\infty}([0,\infty)),\qquad0<\tau_*\leq\tau(t)\leq M,\\ \dot{\tau}(t)&\leq d,\qquad0\leq d<1. \end{aligned} \right. \tag*{(2)}

for constants τ∗,M,d\tau_*,M,d, and the feedback gains satisfy

β≠0,(2α−∣β∣)(1−d)>∣β∣.(3)\beta\ne0,\qquad(2\alpha-|\beta|)(1-d)>|\beta|. \tag*{(3)}

Note that the second condition in (3) is equivalent to [1 (1.6)]. Moreover, the case β=0\beta=0 (no delay) is discussed in Remark 1.4.

Remark 1.1 (On the delay hypothesis). The lower bound τ(t)≥τ(0)\tau(t)\geq\tau(0) in [1 (1.5)] depends on the choice of the initial time. For the translated delay τs(t)=τ(s+t)\tau_s(t)=\tau(s+t), the corresponding condition would require τ(s+t)≥τ(s)\tau(s+t)\geq\tau(s) for t≥0t\geq0, which does not follow from the original assumption. For example, τ(t)=1+14(1−cos⁡t)\tau(t)=1+\frac{1}{4}(1-\cos t) satisfies τ(0)=1≤τ(t)≤32\tau(0)=1\le\tau(t)\le\frac{3}{2} and τ˙(t)≤14\dot{\tau}(t)\le\frac{1}{4}, whereas τπ(π)=1<32=τπ(0)\tau_{\pi}(\pi)=1<\frac{3}{2}=\tau_{\pi}(0).

In (2), we impose a fixed positive lower bound τ∗\tau_*, independent of the initial time. Every translated delay τs\tau_s then satisfies the same bounds with constants τ∗,M,d\tau_*,M,d, and retains the required local regularity. This formulation includes the delays considered in [1], with τ∗=τ(0)\tau_*=\tau(0), and allows the local existence estimates to be applied at each restart with unchanged constants; see Lemma 3.4.

Following [6], we introduce z(t,ρ)=ηx(t−τ(t)ρ,L)z(t,\rho)=\eta_x(t-\tau(t)\rho,L), ρ∈(0,1)\rho\in(0,1), which solves the transport equation

{τ(t)zt(t,ρ)+(1−τ˙(t)ρ)zρ(t,ρ)=0,z(t,0)=ηx(t,L),z(0,ρ)=Z0(ρ),\begin{cases} \tau(t)z_t(t,\rho)+(1-\dot{\tau}(t)\rho)z_\rho(t,\rho)=0,\\ z(t,0)=\eta_x(t,L),\qquad z(0,\rho)=Z_0(\rho), \end{cases}

where h0∈L2(−τ(0),0)h_0\in L^2(-\tau(0),0) is the initial delayed profile of ηx(⋅,L)\eta_x(\cdot,L), and

Z0(ρ)=h0(−τ(0)ρ),ρ∈(0,1).Z_0(\rho)=h_0(-\tau(0)\rho),\qquad\rho\in(0,1).

Let us define the initial state by U0:=(η0,ω0,Z0)U_0:=(\eta_0,\omega_0,Z_0). From the definition of the delayed profile Z0Z_0, we have

τ(0)∥Z0∥L2(0,1)2=∥h0∥L2(−τ(0),0)2.\tau(0)\lVert Z_0\rVert_{L^2(0,1)}^2=\lVert h_0\rVert_{L^2(-\tau(0),0)}^2.

We consider the state space H=X0×L2(0,1)H=X_0\times L^2(0,1), where X0=[L2(0,L)]2X_0=[L^2(0,L)]^2, and define the associated energy by

E(t)=12∫0L(η2+ω2)(t,x) dx+∣β∣2τ(t)∫01z2(t,ρ) dρ.(4)E(t)=\frac{1}{2}\int_0^L(\eta^2+\omega^2)(t,x)\,\mathrm{d}x+\frac{|\beta|}{2}\tau(t)\int_0^1z^2(t,\rho)\,\mathrm{d}\rho. \tag*{(4)}

Consequently, the initial energy can be expressed either in terms of Z0Z_0 or in terms of the prescribed delayed profile h0h_0,

E(0)=12∥(η0,ω0)∥X02+∣β∣τ(0)2∥Z0∥L2(0,1)2=12∥(η0,ω0)∥X02+∣β∣2∥h0∥L2(−τ(0),0)2.E(0)=\frac{1}{2}\lVert(\eta_0,\omega_0)\rVert_{X_0}^2+\frac{|\beta|\tau(0)}{2}\lVert Z_0\rVert_{L^2(0,1)}^2 =\frac{1}{2}\lVert(\eta_0,\omega_0)\rVert_{X_0}^2+\frac{|\beta|}{2}\lVert h_0\rVert_{L^2(-\tau(0),0)}^2.

When there is no risk of confusion, we write ∥⋅∥\lVert\cdot\rVert for the norm of L2(0,L)L^2(0,L) and employ the notation

∥U∥2=∥η∥2+∥ω∥2,∥∇U∥2=∥ηx∥2+∥ωx∥2.\lVert U\rVert^2=\lVert\eta\rVert^2+\lVert\omega\rVert^2,\qquad \lVert\nabla U\rVert^2=\lVert\eta_x\rVert^2+\lVert\omega_x\rVert^2.

Additionally, to have clear computations, let us introduce the boundary quantities

a(t):=ηx(t,L),b(t):=z(t,1)=ηx(t−τ(t),L),a(t):=\eta_x(t,L),\qquad b(t):=z(t,1)=\eta_x(t-\tau(t),L),
v(t)=(a(t),b(t))T.v(t)=(a(t),b(t))^T.

Let us consider

Φ=(−2α+∣β∣ββ∣β∣(d−1)),c0=−12λmax⁡(Φ).(5)\Phi=\begin{pmatrix} -2\alpha+|\beta| & \beta\\ \beta& |\beta|(d-1) \end{pmatrix},\qquad c_0=-\frac{1}{2}\lambda_{\max}(\Phi). \tag*{(5)}

Observe that (3) ensures that Φ\Phi is negative definite. Hence c0>0c_0>0 and is such that

12vTΦv≤−c0∣v∣2,v∈R2.\frac{1}{2}v^T\Phi v\le-c_0|v|^2,\qquad v\in\mathbb{R}^2.

Exponential decay of the energy of the linearized system was established in [1] for 0<L<3π0<L<\sqrt{3}\pi through a Lyapunov argument. Following that approach, we can consider the same Lyapunov functional

V=E+μ1V1+μ2V2,(6)V=E+\mu_1V_1+\mu_2V_2, \tag*{(6)}

where

V1(t)=∫0Lxηω dx,V2(t)=∣β∣2τ(t)∫01(1−ρ)z2(t,ρ) dρ.(7)\begin{aligned} V_1(t)&=\int_0^Lx\eta\omega\,\mathrm{d}x,\\ V_2(t)&=\frac{|\beta|}{2}\tau(t)\int_0^1(1-\rho)z^2(t,\rho)\,\mathrm{d}\rho. \tag*{(7)} \end{aligned}

For the nonlinear problem, local well-posedness was obtained in [1 Theorem 2.7], while global existence and stabilization remained open; see [1 Remark 2 and Section 4]. The main difficulty is to obtain an a priori L2L^2 estimate that accounts for the nonlinear contribution. Indeed, for regular solutions, there holds

ddt12∥U∥2=−αa2+βab−12∫0Lωxη2 dx.\frac{\mathrm{d}}{\mathrm{d}t}\frac{1}{2}\lVert U\rVert^2=-\alpha a^2+\beta ab-\frac{1}{2}\int_0^L\omega_x\eta^2\,\mathrm{d}x.

The nonlinear term has no prescribed sign, and controlling this contribution is the main step in extending the energy estimates to the nonlinear system.

For clarity, we collect in Section 2 the constants and estimates from [1] used in the present analysis, together with the adjustments needed. Throughout this note, we use the values specified in that section. Let us now introduce the admissibility condition for the parameters of the Lyapunov functional.

Definition 1.2. We say that a pair (μ1,μ2)(\mu_1,\mu_2) is admissible if μ1,μ2>0\mu_1,\mu_2>0, μ1L<1\mu_1L<1, and the matrix

Ψ=Φ+Lμ1(α2+1−αβ−αββ2)+∣β∣μ2(1000)(8)\Psi=\Phi+L\mu_1 \begin{pmatrix} \alpha^2+1 & -\alpha\beta\\ -\alpha\beta& \beta^2 \end{pmatrix} +|\beta|\mu_2 \begin{pmatrix} 1&0\\ 0&0 \end{pmatrix} \tag*{(8)}

is negative definite. In particular, an explicit family of admissible pairs is given in (C4) of Section 2.

Main results

Let us now state the main result. Fix 0<L<3π0<L<\sqrt{3}\pi and an admissible pair (μ1,μ2)(\mu_1,\mu_2) in the sense of Definition 1.2. We choose the decay parameter λ\lambda so that

0<λ<min⁡{μ1(3π2−L2)L2(1+μ1L),μ2(1−d)M(1+μ2)}.(9)0<\lambda<\min\left\{\frac{\mu_1(3\pi^2-L^2)}{L^2(1+\mu_1L)},\frac{\mu_2(1-d)}{M(1+\mu_2)}\right\}. \tag*{(9)}

and introduce the constants

{δ=μ1(3π2−L2)−λL2(1+μ1L)2π2,K=14(Lπ)1/2+μ1(Lπ)3/2,ζ=1+max⁡{μ1L,μ2}1−μ1L.(10)\begin{cases} \delta=\dfrac{\mu_1(3\pi^2-L^2)-\lambda L^2(1+\mu_1L)}{2\pi^2},\\[6pt] K=\dfrac{1}{4}\left(\dfrac{L}{\pi}\right)^{1/2}+\mu_1\left(\dfrac{L}{\pi}\right)^{3/2},\\[6pt] \zeta=\dfrac{1+\max\{\mu_1L,\mu_2\}}{1-\mu_1L}. \tag*{(10)} \end{cases}

The restriction (9) ensures that δ>0\delta>0. Then, for small enough initial data, we have the following result.

Theorem 1.3 (Global existence and exponential stability for small data). Assume (1.2), (1.3), and 0<L<3π0 < L < \sqrt{3}\pi, with the parameters chosen as above. If U0=(η0,ω0,Z0)∈HU_0=(\eta_0,\omega_0,Z_0)\in H satisfies

E(0)<r2≔δ22ζK2,(11)E(0)<r^2\coloneqq\frac{\delta^2}{2\zeta K^2}, \tag*{(11)}

then system (1.1) admits a unique global solution such that

(η,ω)∈C([0,T];X0)∩L2(0,T;H01(0,L)2)(\eta,\omega)\in C([0,T];X_0)\cap L^2(0,T;H_0^1(0,L)^2)

for every T>0T>0. Moreover, its energy satisfies E(t)≤ζE(0)e−λtE(t)\leq\zeta E(0)e^{-\lambda t}, for t≥0t\geq0. Finally, setting θ=K2ζE(0)/δ<1\theta=K\sqrt{2\zeta E(0)}/\delta<1, we have

sup⁡t≥0∥U(t)∥+(∫0∞∥∇U(t)∥2 dt)1/2≤2ζE(0)+(1+max⁡{μ1L,μ2})E(0)δ(1−θ).(12)\sup_{t\geq0}\|U(t)\|+\left(\int_0^\infty\|\nabla U(t)\|^2\,dt\right)^{1/2} \leq\sqrt{2\zeta E(0)}+\sqrt{\frac{(1+\max\{\mu_1L,\mu_2\})E(0)}{\delta(1-\theta)}}. \tag*{(12)}

Theorem 1.3 addresses, for sufficiently small initial data, the question of global existence raised in [1 Remark 2]. It also extends the linear stabilization result of [1 Theorem 1.1] to the nonlinear system, with the decay bound specified in (1.9). The proof combines a Lyapunov estimate with a continuation argument. A key point is that the estimates developed in § 3 depend on the delay only through τ∗\tau_*, MM, and dd, the latter through c0c_0. Consequently, the same bounds remain available when the initial time is shifted. This allows us to continue the solution in the energy space, without imposing additional regularity on the transport component; see also the question discussed in [1 Section 4.1(3)].

To illustrate the dependence of the estimates on the parameters, consider

L=5,d=12,α=1,β=12,M=3.L=5,\qquad d=\frac{1}{2},\qquad\alpha=1,\qquad\beta=\frac{1}{2},\qquad M=3.

Along the family of admissible pairs given in (C4), the choice μ1≈0.0478\mu_1\approx0.0478, μ2≈0.0446\mu_2\approx0.0446, and λ=λ∗/2≈0.00355\lambda=\lambda^*/2\approx0.00355 yields

δ≈5.58⋅10−3,K≈0.411,\delta\approx5.58\cdot10^{-3},\qquad K\approx0.411,
ζ≈1.63,r2≈5.65⋅10−5.\zeta\approx1.63,\qquad r^2\approx5.65\cdot10^{-5}.

Here, λ∗\lambda^* denotes the maximum of the decay bound along that family; see (C8). On the other hand, the admissible pair μ1=0.08\mu_1=0.08, μ2=0.15\mu_2=0.15, with λ=0.005\lambda=0.005, gives

δ≈9.81⋅10−3,K≈0.476,\delta\approx9.81\cdot10^{-3},\qquad K\approx0.476,
ζ≈2.33,r2≈9.1⋅10−5.\zeta\approx2.33,\qquad r^2\approx9.1\cdot10^{-5}.

These choices illustrate how both the decay estimate and the sufficient smallness threshold depend on the admissible pair. For a fixed pair, increasing λ\lambda decreases δ\delta and therefore reduces the radius rr. Figure 1 illustrates these bounds.

Figure 1. Illustration of the decay bounds in Theorem 1.3 for L=5L=5, d=12d=\frac{1}{2}, α=1\alpha=1, β=12\beta=\frac{1}{2}, and M=3M=3. (a) Along the parameter curve in (C4), the upper bound for the decay parameter is min⁡{f(μ1),g(μ1)}\min\{f(\mu_1),g(\mu_1)\}, where ff is given in (C1) and g(μ1)=μ2(1−d)/(M(1+μ2))g(\mu_1)=\mu_2(1-d)/(M(1+\mu_2)). The dashed line corresponds to the expression in [1 (1.10)], before the adjustment explained in (C1). (b) Bounds ζe−λt\zeta e^{-\lambda t} for E(t)/E(0)E(t)/E(0) associated with three parameter choices, each valid under its corresponding condition E(0)<r2E(0)<r^2. The comparison illustrates the dependence of the energy bound on the decay rate, the prefactor ζ\zeta, and the sufficient smallness threshold.

For initial data of arbitrary size, global existence for (1.1) remains open in the setting considered here. The smallness assumption enters through the control of the cubic term −12∫0Lωη2 dx-\frac{1}{2}\int_0^L\omega\eta^2\,dx, for which the present estimates do not yield a global energy bound for arbitrary data.

Remark 1.4 (The case β=0\beta=0). When β=0\beta=0, the feedback contains no delayed term. The state space reduces to X0X_0, and the energy is E=12∥U∥2E=\frac{1}{2}\|U\|^2. Let α>0\alpha>0 and choose

0<μ1<2αL(1+α2).0<\mu_1<\frac{2\alpha}{L(1+\alpha^2)}.

Then μ1L<1\mu_1L<1 and

−α+Lμ12(1+α2)<0.-\alpha+\frac{L\mu_1}{2}(1+\alpha^2)<0.

For 0<L<3π0<L<\sqrt{3}\pi, we take V=E+μ1V1V=E+\mu_1V_1 and choose

0<λ<μ1(3π2−L2)L2(1+μ1L).0<\lambda<\frac{\mu_1(3\pi^2-L^2)}{L^2(1+\mu_1L)}.

With δ\delta and KK defined as above and

ζ=1+μ1L1−μ1L,\zeta=\frac{1+\mu_1L}{1-\mu_1L},

the proof of Theorem 1.3 applies after omitting the terms associated with the delay. We therefore obtain the same global existence and exponential decay conclusions for sufficiently small initial data.

This article is organized as follows: In Section 2, we provide some clarifications on the previous work [1] so that the presentation of the current paper becomes clearer and more accurate. In § 3, we prove some estimates for the linearized system that will be useful for the local existence and continuation arguments for the nonlinear problem. § 4 is devoted to the estimates of the nonlinear term in the problem and then the existence of a unique maximal solution. § 5 is devoted to the proof of the main result Theorem 1.3. Finally, § 6 contains some numerical experiments that illustrate Theorem 1.3, and also concluding remarks.

Corrections and clarifications to [1]

Some constants and estimates in [1] contain typographical errors or inconsistencies that require clarification. Since these quantities play an important role in the analysis developed here, we collect below the necessary corrections, together with some refinements concerning the admissibility of the Lyapunov parameters and the optimization of the decay rate. Throughout this note, we systematically use the corrected expressions stated below.

(C1) Decay rate. The estimate of the term S1S_1 in [1 Section 3] yields the restriction

0<λ<μ1(3π2−L2)L2(1+μ1L).0<\lambda<\frac{\mu_1(3\pi^2-L^2)}{L^2(1+\mu_1L)}.

rather than the expression with denominator 1+μ11+\mu_1 stated in [1 Theorem 1.1, (1.10)]. In particular, for L>1L>1, the printed bound overestimates the decay rate justified by the proof. We observe that the function f(μ1)f(\mu_1) introduced in [1 Proposition 3.1] already contains the correct denominator.

(C2) The functional V1V_1. The calculations in [1 Section 3] are performed with

V1(t)=∫0Lxη(t,x)ω(t,x) dx.V_1(t)=\int_0^L x\eta(t,x)\omega(t,x)\,\mathrm{d}x.

without the factor 1/21/2 appearing in its original definition. Consequently, the estimate consistent with those calculations is

∣μ1V1(t)∣≤μ1L2∥U(t)∥2.|\mu_1V_1(t)|\leq\frac{\mu_1L}{2}\lVert U(t)\rVert^2.

(C3) Equivalence of the energy and the Lyapunov functional. Recalling that we are considering the Lyapunov V=E+μ1V1+μ2V2V=E+\mu_1V_1+\mu_2V_2 and V2≥0V_2\geq0, we get

(1−μ1L)E(t)≤V(t)≤(1+max⁡{μ1L,μ2})E(t).(13)(1-\mu_1L)E(t)\leq V(t) \leq(1+\max\{\mu_1L,\mu_2\})E(t). \tag*{(13)}

Thus, under the admissibility condition μ1L<1\mu_1L<1, the right constant to be considered is

ζ=1+max⁡{μ1L,μ2}1−μ1L.\zeta=\frac{1+\max\{\mu_1L,\mu_2\}}{1-\mu_1L}.

This improves the constant stated in [1 (1.9)], whose denominator is 1−max⁡{μ1L,μ2}1-\max\{\mu_1L,\mu_2\}. In particular, the latter requires the additional restriction μ2<1\mu_2<1 to remain positive, whereas the estimate above holds for every admissible pair. Such a restriction on μ2\mu_2 is not guaranteed by the parameter choice described in [1 Section 4.1].

(C4) Admissible parameters. The argument in [1 Section 3] requires the matrix Ψ\Psi, introduced in Definition 1.2, to be negative definite. An explicit family of admissible pairs is obtained by choosing

0<μ1<(2α−∣β∣)(1−d)−∣β∣L(1−d)(1+α2)0<\mu_1<\frac{(2\alpha-|\beta|)(1-d)-|\beta|}{L(1-d)(1+\alpha^2)}

and setting

μ2=(2α−∣β∣)(1−d)−∣β∣∣β∣(1−d)−L(1+α2)∣β∣μ1.\mu_2=\frac{(2\alpha-|\beta|)(1-d)-|\beta|}{|\beta|(1-d)}-\frac{L(1+\alpha^2)}{|\beta|}\mu_1.

Indeed, along this curve, a direct computation gives

Ψ11=−∣β∣1−d<0\Psi_{11}=-\frac{|\beta|}{1-d}<0

and

det⁡Ψ=Lβ2μ11−d[(1−d)(2α−Lα2μ1)−∣β∣].\det\Psi=\frac{L\beta^2\mu_1}{1-d}\left[(1-d)(2\alpha-L\alpha^2\mu_1)-|\beta|\right].

The gain condition (3), together with the upper bound on μ1\mu_1, ensures that the expression in brackets is strictly positive. Hence det⁡Ψ>0\det\Psi>0, and Ψ\Psi is negative definite throughout the prescribed interval. Moreover, the same choice guarantees μ2>0\mu_2>0 and

μ1L<2α1+α2≤1.\mu_1L<\frac{2\alpha}{1+\alpha^2}\leq1.

Therefore, every pair on this curve is admissible, without requiring the additional condition ∣α∣<1|\alpha|<1 imposed in [1 Section 4.1(5)]. We emphasize, however, that this construction provides only a particular family of admissible pairs and does not characterize the entire admissible set (see (C8) below).

(C5) Energy identity For the linearized system, the exact derivative of the energy involves ι˙(t)\dot{\iota}(t) rather than its upper bound dd. More precisely,

E′(t)=12ν(t)TΦν(t)−∣β∣2(d−τ˙(t))b(t)2≤12ν(t)TΦν(t)≤−c0(a(t)2+b(t)2).E'(t)=\frac{1}{2}\nu(t)^{T}\Phi\nu(t)-\frac{|\beta|}{2}\left(d-\dot{\tau}(t)\right)b(t)^2 \leq\frac{1}{2}\nu(t)^{T}\Phi\nu(t)\leq-c_0\left(a(t)^2+b(t)^2\right).

Thus, the identity stated in [1 Proposition 2.3] with the matrix Φ\Phi should be understood as an upper bound. Furthermore, the inequality

E′(t)+ηx2(t,L)+z2(t,1)≤0,E'(t)+\eta_x^2(t,L)+z^2(t,1)\leq0,

used in the proof of [1 Proposition 2.4], must retain the coercivity constant c0c_0, namely,

E′(t)+c0(ηx2(t,L)+z2(t,1))≤0.E'(t)+c_0\left(\eta_x^2(t,L)+z^2(t,1)\right)\leq0.

(C6) Weighted a priori estimates. As a consequence of the (C5) correction, the energy and boundary-trace estimates in [1 Proposition 2.4] must retain the time-dependent weights and the dissipation constant. In particular,

∥(η,ω)(t)∥X02+∣β∣τ(t)∥z(t)∥L2(0,1)2≤∥(η0,ω0)∥X02+∣β∣τ(0)∥Z0∥L2(0,1)2,\|(\eta,\omega)(t)\|_{X_0}^2+|\beta|\tau(t)\|z(t)\|_{L^2(0,1)}^2 \leq\|(\eta_0,\omega_0)\|_{X_0}^2+|\beta|\tau(0)\|Z_0\|_{L^2(0,1)}^2,

while integration of the corrected dissipation inequality gives

∫0T(a(t)2+b(t)2) dt≤E(0)c0=∥(η0,ω0)∥X02+∣β∣τ(0)∥Z0∥L2(0,1)22c0.\int_0^T\left(a(t)^2+b(t)^2\right)\,dt\leq\frac{E(0)}{c_0} =\frac{\|(\eta_0,\omega_0)\|_{X_0}^2+|\beta|\tau(0)\|Z_0\|_{L^2(0,1)}^2}{2c_0}.

(C7) Linear identity for V1V_1. The term

−∫0Lxηωωx dx-\int_0^L x\eta\omega\omega_x\,dx

appearing in [1 (3.5)] arises from the nonlinear equation and should not be present in the calculation for the linearized system.

(C8) On the optimality of λ\lambda. The optimization in [1 Proposition 3.1] concerns the decay bound restricted to the particular family of admissible pairs described in (C4), rather than the full admissible set. To illustrate this distinction, consider the parameters used in [1],

L=5,d=12,α=1,β=12,M=3.L=5,\qquad d=\frac{1}{2},\qquad\alpha=1,\qquad\beta=\frac{1}{2},\qquad M=3.

In this case, the curve in (C4) is given by

0<μ1<120,μ2=1−20μ1.0<\mu_1<\frac{1}{20},\qquad\mu_2=1-20\mu_1.

The maximum of the corrected decay bound along this curve is

λ∗≈0.007108903,\lambda^*\approx0.007108903,

corresponding to μ1≈0.047772311\mu_1\approx0.047772311, and μ2≈0.044553787\mu_2\approx0.044553787.

However, the pair μ1=0.08\mu_1=0.08, and μ2=0.15\mu_2=0.15 is also admissible, since

Ψ=(−0.6250.30.3−0.15),det⁡Ψ=0.00375>0.\Psi=\begin{pmatrix}-0.625&0.3\\0.3&-0.15\end{pmatrix},\qquad\det\Psi=0.00375>0.

For this choice, the corrected decay restriction is

0<λ<min⁡{0.010534430,0.021739130},0<\lambda<\min\{0.010534430,0.021739130\},

which allows, for instance, λ=0.01>λ∗\lambda=0.01>\lambda^*.

In fact, for these parameters,

Ψ=(−32+10μ1+12μ212−52μ112−52μ1−14+54μ1),\Psi=\begin{pmatrix} -\frac{3}{2}+10\mu_1+\frac{1}{2}\mu_2 & \frac{1}{2}-\frac{5}{2}\mu_1\\ \frac{1}{2}-\frac{5}{2}\mu_1 & -\frac{1}{4}+\frac{5}{4}\mu_1 \end{pmatrix},

and

det⁡Ψ=(5μ1−1)(10μ1+μ2−1)8.\det\Psi=\frac{(5\mu_1-1)(10\mu_1+\mu_2-1)}{8}.

Consequently, the full admissible set A\mathcal{A} consists of the pairs (μ1,μ2)(\mu_1,\mu_2) with

0<μ1<110,0<μ2<1−10μ1.0<\mu_1<\frac{1}{10},\qquad0<\mu_2<1-10\mu_1.

Since the decay bound is nondecreasing with respect to μ2\mu_2, its supremum over A\mathcal{A} is

Λadm=max⁡0≤s≤1/10min⁡{s(3π2−25)25(1+5s),1−10s6(2−10s)}≈0.011658099.\Lambda_{\mathrm{adm}}=\max_{0\leq s\leq1/10}\min\left\{\frac{s(3\pi^2-25)}{25(1+5s)},\frac{1-10s}{6(2-10s)}\right\} \approx0.011658099.

The first function is strictly increasing, whereas the second is strictly decreasing. Their unique intersection occurs at s∗≈0.092479062s_*\approx0.092479062, and determines the above supremum. Since the corresponding limiting parameter μ2=1−10s∗\mu_2=1-10s_* satisfies det⁡Ψ=0\det\Psi=0, the supremum is not attained by any admissible pair. Therefore, the optimization in [1 Proposition 3.1] yields the largest decay bound along the prescribed parameter curve, but not over the full admissible set. Moreover, this optimization concerns the decay estimates obtained by the Lyapunov method and does not establish an optimal exponential decay rate for the underlying evolution system.

With these corrections, the exponential-decay conclusion of [1 Theorem 1.1] remains valid for the linearized system, with the constants and admissibility conditions specified above. In the next section, we revisit the well-posedness estimates and establish bounds that are uniform under time translations of the delay, which will be essential for the analysis of the nonlinear problem.

Well-posedness with constants uniform under time translations

In this section, we establish the estimates for the linearized system that will be used in the local existence and continuation arguments for the nonlinear problem. First, let us set the time-dependent norm such that for each t≥0t \ge0, we equip HH with the norm

∥U∥t2=∥η∥2+∥ω∥2+∣β∣τ(t)∥z∥L2(0,1)2,\|U\|_{t}^{2}=\|\eta\|^{2}+\|\omega\|^{2}+|\beta|\tau(t)\|z\|_{L^{2}(0,1)}^{2},

for U=(η,ω,z)U=(\eta,\omega,z). Thus, along a solution UU, we have:

∥U(t)∥t2=2E(t).\|U(t)\|_{t}^{2}=2E(t).

Note that, by setting

c∗=min⁡{1,∣β∣τ∗},C∗=max⁡{1,∣β∣M},c_{*}=\min\{1,|\beta|\tau_{*}\},\qquad C_{*}=\max\{1,|\beta|M\},

we get

c∗∥U∥H2≤∥U∥t2≤C∗∥U∥H2c_{*}\|U\|_{H}^{2}\leq\|U\|_{t}^{2}\leq C_{*}\|U\|_{H}^{2}

for all U∈HU\in H and $t\geq 0.

For 0≤s<t10\leq s<t_{1}, let us set the solution spaces

Bs,t1=C([s,t1];X0)∩L2(s,t1;H01(0,L)),B_{s,t_{1}}=C([s,t_{1}];X_{0})\cap L^{2}(s,t_{1};H_{0}^{1}(0,L)),

endowed with the norm

∥(η,ω)∥Bs,t1=sup⁡t∈[s,t1]∥U(t)∥+(∫st1∥∇U(t)∥2 dt)1/2.\|(\eta,\omega)\|_{B_{s,t_{1}}}=\sup_{t\in[s,t_{1}]}\|U(t)\|+\left(\int_{s}^{t_{1}}\|\nabla U(t)\|^{2}\,dt\right)^{1/2}.

Similarly, we write BT=B0,TB_{T}=B_{0,T} and use the same notation for scalar functions, that is,

∥u∥BT=sup⁡t∈[0,T]∥u(t)∥+∥ux∥L2(0,T;L2(0,L)).\|u\|_{B_{T}}=\sup_{t\in[0,T]}\|u(t)\|+\|u_{x}\|_{L^{2}(0,T;L^{2}(0,L))}.

Recall that the linearized system is governed by the family {A(t)}\{A(t)\} defined by (see [1 (2.4)–(2.5)])

A(t)(η,ω,z):=(−ωx−ωxxx,−ηx−ηxxx,τ˙(t)ρ−1τ(t)zρ),A(t)(\eta,\omega,z):=\left(-\omega_{x}-\omega_{xxx},-\eta_{x}-\eta_{xxx},\frac{\dot{\tau}(t)\rho-1}{\tau(t)}z_{\rho}\right),

whose domain D(A(t))D(A(t)) consists of the triples

(η,ω,z)∈[H3(0,L)∩H01(0,L)]2×H1(0,1)(\eta,\omega,z)\in[H^{3}(0,L)\cap H_{0}^{1}(0,L)]^{2}\times H^{1}(0,1)

such that

ηx(0)=0,z(0)=ηx(L),\eta_{x}(0)=0,\qquad z(0)=\eta_{x}(L),
ωx(L)=−αηx(L)+βz(1).\omega_{x}(L)=-\alpha\eta_{x}(L)+\beta z(1).

Clearly, D(A(t))=D(A(0))D(A(t))=D(A(0)) is independent of tt, and Kato’s theory [7] applies under the conditions (1.2)–(1.3). Hence {A(t)}\{A(t)\} generates an evolution system {S(t,s)}0≤s≤t\{S(t,s)\}_{0\leq s\leq t} on HH; see [1 Theorem 2.2].

Now, aiming at the nonlinear problem, we set the mild solution for a system with a source. Let s≥0s\geq0, given Us∈HU_{s}\in H and f=(f1,f2)∈L1(s,t1;X0)f=(f_{1},f_{2})\in L^{1}(s,t_{1};X_{0}), we consider

{ηt+ωx+ωxxx=f1,ωt+ηx+ηxxx=f2,(14)\begin{cases} \eta_{t}+\omega_{x}+\omega_{xxx}=f_{1},\\ \omega_{t}+\eta_{x}+\eta_{xxx}=f_{2}, \tag*{(14)} \end{cases}

coupled with the transport equation for zz and the boundary conditions of (1.1), with initial state UsU_{s} at time ss. Its mild solution is then given by the Duhamel formula

U(t)=S(t,s)Us+∫stS(t,r)(f1(r),f2(r),0) dr,U(t)=S(t,s)U_{s}+\int_{s}^{t}S(t,r)(f_{1}(r),f_{2}(r),0)\,dr,

for t∈[s,t1]t\in[s,t_{1}].

Proposition 3.1 (Linear estimates). Assume (1.2)–(1.3). Let s≥0s\geq0, T>0T>0, Us∈HU_{s}\in H, and f∈L1(s,s+T;X0)f\in L^{1}(s,s+T;X_{0}), and set F:=∥f∥L1(s,s+T;X0)F:=\|f\|_{L^{1}(s,s+T;X_{0})}. Then the mild solution on [s,s+T][s,s+T] satisfies (η,ω)∈Bs,s+T(\eta,\omega)\in B_{s,s+T} and z∈C([s,s+T];L2(0,1))z\in C([s,s+T];L^{2}(0,1)), a,b∈L2(s,s+T)a,b\in L^{2}(s,s+T).

Moreover, we have

sup⁡t∈[s,s+T]∥U(t)∥t≤∥Us∥s+2F,\sup_{t\in[s,s+T]}\|U(t)\|_{t}\leq\|U_{s}\|_{s}+2F,
c0∫ss+T(a2+b2) dt≤∥Us∥s2+3F2,c_{0}\int_{s}^{s+T}(a^{2}+b^{2})\,dt\leq\|U_{s}\|_{s}^{2}+3F^{2},
∫ss+T∥∇U(t)∥2 dt≤Λ(T)(∥Us∥s+F)2,\int_{s}^{s+T}\|\nabla U(t)\|^{2}\,dt\leq\Lambda(T)(\|U_{s}\|_{s}+F)^{2},

where

Λ(T)=23[2T+6L+3L(1+2α2+2β2)2c0].\Lambda(T)=\frac{2}{3}\left[2T+6L+\frac{3L(1+2\alpha^{2}+2\beta^{2})}{2c_{0}}\right].

In particular, with CT=2+Λ(T)C_{T}=2+\sqrt{\Lambda(T)},

∥(η,ω)∥Bs,s+T≤CT(∥Us∥s+F).\|(\eta,\omega)\|_{B_{s,s+T}}\leq C_{T}(\|U_{s}\|_{s}+F).

The constant CTC_{T} is nondecreasing in TT, independent of ss, and uniform over the delays satisfying (1.2). In addition, for every t∈[s,s+T]t\in[s,s+T],

12∥U(t)∥t2−12∥Us∥s2=∫st[(−α+∣β∣2)a2+βab−∣β∣2(1−τ˙)b2] dr(15)\frac{1}{2}\|U(t)\|_{t}^{2}-\frac{1}{2}\|U_{s}\|_{s}^{2} =\int_{s}^{t}\left[\left(-\alpha+\frac{|\beta|}{2}\right)a^{2}+\beta ab-\frac{|\beta|}{2}(1-\dot{\tau})b^{2}\right]\,dr \tag*{(15)}
+∫st∫0L(ηf1+ωf2) dx dr,+\int_{s}^{t}\int_{0}^{L}(\eta f_{1}+\omega f_{2})\,dx\,dr,
V1(t)−V1(s)=∫st[12∥U∥2−32∥∇U∥2] drV_{1}(t)-V_{1}(s)=\int_{s}^{t}\left[\frac{1}{2}\|U\|^{2}-\frac{3}{2}\|\nabla U\|^{2}\right]\,dr
+L2∫st((αa−βb)2+a2) dr+\frac{L}{2}\int_{s}^{t}((\alpha a-\beta b)^{2}+a^{2})\,dr
+∫st∫0Lx(ωf1+ηf2) dx dr,(16)+\int_{s}^{t}\int_{0}^{L}x(\omega f_{1}+\eta f_{2})\,dx\,dr, \tag*{(16)}

and

V2(t)−V2(s)=∣β∣2∫st[a(r)2−∫01(1−τ˙(r)ρ)z(r,ρ)2 dρ] dr.(17)V_{2}(t)-V_{2}(s) =\frac{|\beta|}{2}\int_{s}^{t}\left[a(r)^{2}-\int_{0}^{1}(1-\dot{\tau}(r)\rho)z(r,\rho)^{2}\,d\rho\right]\,dr. \tag*{(17)}

Proof. Let us split the proof in several steps.

Step 1. Identities for regular data. Let Us∈D(A(0))U_{s}\in D(A(0)) and f∈C1([s,s+T];Cc∞(0,L)2)f\in C^{1}([s,s+T];C_{c}^{\infty}(0,L)^{2}). Then (f1,f2,0)(f_{1},f_{2},0) belongs to C1([s,s+T];D(A(0)))C^{1}([s,s+T];D(A(0))), and consequently the corresponding solution is classical; see [7, 8], that is,

U∈C([s,s+T];D(A(0)))∩C1([s,s+T];H).U\in C([s,s+T];D(A(0)))\cap C^{1}([s,s+T];H).

Multiplying the first equation by η\eta and the second by ω\omega, integrating over (0,L)(0,L), and using the boundary conditions we obtain

ddt12∥U∥2=−αa2+βab+∫0L(ηf1+ωf2) dx.(18)\frac{d}{dt}\frac{1}{2}\|U\|^2=-\alpha a^2+\beta ab+\int_0^L(\eta f_1+\omega f_2)\,dx . \tag*{(18)}

On the other hand, the transport equation yields

ddt(∣β∣2τ∫01z2 dρ)=∣β∣2[τ˙∫01z2 dρ−∫01(1−τ˙ρ)(z2)ρ dρ]=∣β∣2(a2−(1−τ˙)b2).(19)\begin{aligned} \frac{d}{dt}\left(\frac{|\beta|}{2}\tau\int_0^1z^2\,d\rho\right) \\ &=\frac{|\beta|}{2}\left[\dot{\tau}\int_0^1z^2\,d\rho-\int_0^1(1-\dot{\tau}\rho)(z^2)_\rho\,d\rho\right] \\ &=\frac{|\beta|}{2}\left(a^2-(1-\dot{\tau})b^2\right). \tag*{(19)} \end{aligned}

Adding (18) and (19) and integrating from ss to tt proves (15). For V2V_2, the same calculation with the weight 1−ρ1-\rho gives

V2′=∣β∣2[τ˙∫01(1−ρ)z2 dρ−∫01(1−ρ)(1−τ˙ρ)(z2)ρ dρ]=∣β∣2[a2−∫01(1−τ˙ρ)z2 dρ].\begin{aligned} V_2'=\frac{|\beta|}{2}\left[\dot{\tau}\int_0^1(1-\rho)z^2\,d\rho-\int_0^1(1-\rho)(1-\dot{\tau}\rho)(z^2)_\rho\,d\rho\right] \\ &=\frac{|\beta|}{2}\left[a^2-\int_0^1(1-\dot{\tau}\rho)z^2\,d\rho\right]. \end{aligned}

Integrating from ss to tt proves (17).

Next, multiplying the first equation by xωx\omega and the second by xηx\eta, integration by parts implies

∫0Lxηtω dx=12∥ω∥2−32∥ωx∥2+L2ωx(t,L)2+∫0Lxf1ω dx,\int_0^Lx\eta_t\omega\,dx=\frac{1}{2}\|\omega\|^2-\frac{3}{2}\|\omega_x\|^2+\frac{L}{2}\omega_x(t,L)^2+\int_0^Lxf_1\omega\,dx,
∫0Lxηωt dx=12∥η∥2−32∥ηx∥2+L2a2+∫0Lxηf2 dx.\int_0^Lx\eta\omega_t\,dx=\frac{1}{2}\|\eta\|^2-\frac{3}{2}\|\eta_x\|^2+\frac{L}{2}a^2+\int_0^Lx\eta f_2\,dx.

Adding the results and utilizing the boundary condition implies (16).

Step 2. Uniform estimates. By (C5) and (1.5), the boundary terms in (15) satisfy

(−α+∣β∣2)a2+βab−∣β∣2(1−τ˙)b2≤−c0(a2+b2).\left(-\alpha+\frac{|\beta|}{2}\right)a^2+\beta ab-\frac{|\beta|}{2}(1-\dot{\tau})b^2\leq-c_0(a^2+b^2).

Set Y:=sup⁡r∈[s,s+T]∥U(r)∥rY:=\sup_{r\in[s,s+T]}\|U(r)\|_r. Since ∥U(t)∥≤∥U(t)∥t\|U(t)\|\leq\|U(t)\|_t, the Cauchy–Schwarz inequality and (15) imply

12∥U(t)∥t2+c0∫st(a2+b2) dr≤12∥Us∥s2+YF,\frac{1}{2}\|U(t)\|_t^2+c_0\int_s^t(a^2+b^2)\,dr\leq\frac{1}{2}\|U_s\|_s^2+YF,

for every t∈[s,s+T]t\in[s,s+T]. Dropping the integral and taking the supremum in tt, we get Y2−2FY−∥Us∥s2≤0Y^2-2FY-\|U_s\|_s^2\leq0, and hence

Y≤F+F2+∥Us∥s2≤∥Us∥s+2F.Y\leq F+\sqrt{F^2+\|U_s\|_s^2}\leq\|U_s\|_s+2F.

Taking t=s+Tt=s+T in the same inequality yields

c0∫ss+T(a2+b2) dt≤12∥Us∥s2+YF≤12∥Us∥s2+(∥Us∥s+2F)F≤∥Us∥s2+3F2.\begin{aligned} c_0\int_s^{s+T}(a^2+b^2)\,dt\leq\frac{1}{2}\|U_s\|_s^2+YF \\ &\leq\frac{1}{2}\|U_s\|_s^2+(\|U_s\|_s+2F)F \leq\|U_s\|_s^2+3F^2. \end{aligned}

From (16) together with

∣V1(t)∣≤L2∥U(t)∥2,(αa−βb)2≤2α2a2+2β2b2,|V_1(t)|\leq\frac{L}{2}\|U(t)\|^2,\qquad(\alpha a-\beta b)^2\leq2\alpha^2a^2+2\beta^2b^2,
∣∫0Lx(ωf1+ηf2) dx∣≤L∥U∥∥f∥X0,\left|\int_0^Lx(\omega f_1+\eta f_2)\,dx\right|\leq L\|U\|\|f\|_{X_0},

we obtain

32∫ss+T∥∇U∥2 dt≤T2Y2+LY2+LYF+L2(1+2α2+2β2)∫ss+T(a2+b2) dt.\frac{3}{2}\int_s^{s+T}\|\nabla U\|^2\,dt\leq\frac{T}{2}Y^2+LY^2+LYF+\frac{L}{2}(1+2\alpha^2+2\beta^2)\int_s^{s+T}(a^2+b^2)\,dt.

Since Y≤2(∥Us∥s+F)Y\leq2(\|U_s\|_s+F) and

∥Us∥s2+3F2≤3(∥Us∥s+F)2,\|U_s\|_s^2+3F^2\leq3(\|U_s\|_s+F)^2,

we obtain

32∫ss+T∥∇U∥2 dt≤32Λ(T)(∥Us∥s+F)2.\frac{3}{2}\int_s^{s+T}\|\nabla U\|^2\,dt\leq\frac{3}{2}\Lambda(T)(\|U_s\|_s+F)^2.

This proves the estimate with Λ(T)\Lambda(T), and the bound in Bs,s+TB_{s,s+T} follows immediately. Notice that the estimates are uniform, i.e., none of these constants depends on ss or on the local derivatives of τ\tau beyond the fixed upper bound dd for τ˙\dot{\tau}.

Step 3. Passage to general data. Let Us∈HU_s\in H and f∈L1(s,s+T;X0)f\in L^1(s,s+T;X_0) be arbitrary. Since D(A(0))D(A(0)) is dense in HH and C1([s,s+T];C∞(0,L)2)C^1([s,s+T];C^\infty(0,L)^2) is dense in L1(s,s+T;X0)L^1(s,s+T;X_0), there exist Usn∈D(A(0))U_s^n\in D(A(0)) and fn∈C1([s,s+T];C∞(0,L)2)f^n\in C^1([s,s+T];C^\infty(0,L)^2) such that

Usn→Usin H,fn→fin L1(s,s+T;X0).U_s^n\to U_s\quad\text{in }H,\qquad f^n\to f\quad\text{in }L^1(s,s+T;X_0).

Let Un=(ηn,ωn,zn)U^n=(\eta^n,\omega^n,z^n) be the corresponding classical solutions, with traces an,bna^n,b^n. By linearity, Un−UmU^n-U^m is the classical solution with initial state Usn−UsmU_s^n-U_s^m and source fn−fmf^n-f^m. Hence the estimates of Step 2, applied to Un−UmU^n-U^m, together with the uniform equivalence of the norms, show that (Un)(U^n), (ηn,ωn)(\eta^n,\omega^n) and (an,bn)(a^n,b^n) are Cauchy sequences, so that

Un→Uin C([s,s+T];H),U^n\to U\quad\text{in }C([s,s+T];H),
(ηn,ωn)→(η,ω)in Bs,s+T,(\eta^n,\omega^n)\to(\eta,\omega)\quad\text{in }B_{s,s+T},
(an,bn)→(a,b)in L2(s,s+T;R2).(a^n,b^n)\to(a,b)\quad\text{in }L^2(s,s+T;\mathbb{R}^2).

We claim that UU is the mild solution with initial state UsU_s and source ff. Indeed, each UnU^n satisfies the Duhamel formula with data (Usn,fn)(U_s^n,f^n), and Step 2 with f=0f=0 gives

∥S(t,r)∥L(H)≤(C∗/c∗)1/2for s≤r≤t≤s+T;\|S(t,r)\|_{\mathcal{L}(H)}\leq(C_*/c_*)^{1/2}\quad\text{for }s\leq r\leq t\leq s+T;

thus both sides of the formula converge in C([s,s+T];H)C([s,s+T];H). Moreover, (η,ω)(t)∈H01(0,L)(\eta,\omega)(t)\in H^1_0(0,L) for almost every tt, since L2(s,s+T;H01(0,L))L^2(s,s+T;H^1_0(0,L)) is complete.

Finally, the identities (15)–(17) hold for each UnU^n, and every term in them passes to the limit: the terms that are quadratic in (a,b)(a,b) and in ∇U\nabla U by the L2L^2 convergences above; the terms involving zz because zn→zz^n\to z in C([s,s+T];L2(0,1))C([s,s+T];L^2(0,1)) and τ˙∈L∞(s,s+T)\dot{\tau}\in L^\infty(s,s+T); and the source terms because Un→UU^n\to U uniformly in time while fn→ff^n\to f in L1(s,s+T;X0)L^1(s,s+T;X_0). Therefore (15)–(17) hold for UU and every t∈[s,s+T]t\in[s,s+T], and the estimates of Step 2, which only use these identities, remain valid for UU. □

It is noteworthy that the preceding estimates also give a uniform bound in the fixed norm of HH.

Corollary 3.2 (Uniform estimate). Under the assumptions of Proposition 3.1, we have

sup⁡t∈[s,s+T]∥U(t)∥H+∥(a,b)∥L2(s,s+T;R2)+(∫ss+T(∥U(t)∥2+∥∇U(t)∥2) dt)1/2≤C~T(∥Us∥H+F).\sup_{t\in[s,s+T]}\|U(t)\|_H+\|(a,b)\|_{L^2(s,s+T;\mathbb{R}^2)} +\left(\int_s^{s+T}\left(\|U(t)\|^2+\|\nabla U(t)\|^2\right)\,dt\right)^{1/2} \le\widetilde C_T\left(\|U_s\|_H+F\right).

where

C~T=C∗c∗[2+(1+L2π2)Λ(T)+3c0].\widetilde C_T=\sqrt{\frac{C_*}{c_*}\left[2+\sqrt{\left(1+\frac{L^2}{\pi^2}\right)\Lambda(T)}+\sqrt{\frac{3}{c_0}}\right]}.

This constant depends only on TT, LL, α\alpha, β\beta, dd, τ∗\tau_*, MM and is independent of ss.

Proof. The equivalence of the norms gives

∥U(t)∥H≤c∗−1/2∥U(t)∥t,∥Us∥s≤C∗1/2∥Us∥H.\|U(t)\|_H\le c_*^{-1/2}\|U(t)\|_t,\qquad\|U_s\|_s\le C_*^{1/2}\|U_s\|_H.

Moreover, Poincaré’s inequality yields

∥U(t)∥2≤L2π2∥∇U(t)∥2,\|U(t)\|^2\le\frac{L^2}{\pi^2}\|\nabla U(t)\|^2,

for a.e. t∈(s,s+T)t\in(s,s+T). Applying Proposition 3.1, we therefore bound the left-hand side by

[2c∗+(1+L2π2)Λ(T)+3c0](∥Us∥s+F).\left[\frac{2}{\sqrt{c_*}}+\sqrt{\left(1+\frac{L^2}{\pi^2}\right)\Lambda(T)}+\sqrt{\frac{3}{c_0}}\right](\|U_s\|_s+F).

The conclusion follows from c∗≤1≤C∗c_*\le1\le C_* and

∥Us∥s+F≤C∗1/2(∥Us∥H+F).\|U_s\|_s+F\le C_*^{1/2}(\|U_s\|_H+F).

□

Remark 3.3 (Traces in the energy space). Notice that (η,ω)∈BT(\eta,\omega)\in B_T alone does not provide a classical trace for ηx(⋅,L)\eta_x(\cdot,L). Indeed, the traces aa and b=z(⋅,1)b=z(\cdot,1) in Proposition 3.1 are defined as L2(0,T)L^2(0,T) limits of regular boundary traces. The linear estimates ensure that these limits are independent of the approximation and agree with the classical traces for regular solutions.

Let a^\widehat a denote the extension of aa to (−τ(0),T)(-\tau(0),T) by the initial delayed profile h0(σ)=Z0(−σ/τ(0))h_0(\sigma)=Z_0(-\sigma/\tau(0)) on (−τ(0),0)(-\tau(0),0). Then

z(t,⋅)=a^(t−τ(t)⋅)in L2(0,1),z(t,\cdot)=\widehat a(t-\tau(t)\cdot)\quad\text{in }L^2(0,1),

for t∈[0,T]t\in[0,T], and b(t)=a^(t−τ(t))b(t)=\widehat a(t-\tau(t)) for almost every t∈(0,T)t\in(0,T). Indeed, τ(t)≥τ∗\tau(t)\ge\tau_* and 1−τ˙(t)≥1−d>01-\dot\tau(t)\ge1-d>0 give, by changes of variables,

sup⁡t∈[0,T]∫01∣a^(t−τ(t)ρ)∣2 dρ≤1τ∗∥a^∥L2(−τ(0),T)2,\sup_{t\in[0,T]}\int_0^1|\widehat a(t-\tau(t)\rho)|^2\,d\rho \le\frac{1}{\tau_*}\|\widehat a\|^2_{L^2(-\tau(0),T)},
∫0T∣a^(t−τ(t))∣2 dt≤11−d∥a^∥L2(−τ(0),T)2.\int_0^T|\widehat a(t-\tau(t))|^2\,dt \le\frac{1}{1-d}\|\widehat a\|^2_{L^2(-\tau(0),T)}.

Applied to differences of regular approximations, these bounds extend the characteristic formulas to all initial states in HH, uniformly in time in L2(0,1)L^2(0,1) for zz and in L2(0,T)L^2(0,T) for bb.

We conclude with the time-translation property that will allow us to restart the nonlinear problem from any state reached by its evolution.

Lemma 3.4 (Time translations). Let UU be a mild solution of the linear problem with source f∈L1(0,t1;X0)f\in L^1(0,t_1;X_0) on [0,t1][0,t_1], and let 0≤s<t10\le s<t_1. Then τs(t)=τ(s+t)\tau_s(t)=\tau(s+t) satisfies (1.2) with the same constants τ∗,M,d\tau_*,M,d. Moreover, t↦U(s+t)t\mapsto U(s+t) is the mild solution on [0,t1−s][0,t_1-s] corresponding to the delay τs\tau_s, the source f(s+⋅)f(s+\cdot), and the initial state U(s)∈HU(s)\in H.

Proof. The assumptions in (1.2) are preserved by time translation. Since τ˙≤d<1\dot\tau\le d<1, we have t−τ(t)≥s−τ(s)t-\tau(t)\ge s-\tau(s) for t≥st\ge s. By Remark 3.3, z(s,ρ)=a^(s−τ(s)ρ)z(s,\rho)=\widehat a(s-\tau(s)\rho) for a.e. ρ∈(0,1)\rho\in(0,1), so z(s)z(s) becomes the initial delayed profile to restart the problem at time ss.

The operator family associated with τs\tau_s is {A(s+t)}t≥0\{A(s+t)\}_{t\ge0}, with domain D(A(0))D(A(0)) and evolution operators S(s+t,s+r)S(s+t,s+r). Splitting the Duhamel integral at ss, we get

U(s+t)=S(s+t,s)U(s)+∫0tS(s+t,s+r)(f1(s+r),f2(s+r),0) dr,U(s+t)=S(s+t,s)U(s) +\int_0^t S(s+t,s+r)\left(f_1(s+r),f_2(s+r),0\right)\,dr,

for 0≤t≤t1−s0\le t\le t_1-s. This is the mild formulation of the problem with delay τs\tau_s, source f(s+⋅)f(s+\cdot), and initial state U(s)U(s). □

The nonlinear system

In order to consider the nonlinear terms, we first estimate the terms in the source space of Proposition 3.1.

Lemma 4.1. For every ω∈H01(0,L)\omega\in H^1_0(0,L),

∥ω∥L∞2≤∥ω∥∥ωx∥.(20)\|\omega\|_{L^\infty}^2\le\|\omega\|\|\omega_x\|. \tag*{(20)}

Moreover, for scalar u,v∈BTu,v\in B_T,

∫0T∥(uv)x(t)∥ dt≤2T1/4∥u∥BT∥v∥BT.(21)\int_0^T\|(uv)_x(t)\|\,dt\le2T^{1/4}\|u\|_{B_T}\|v\|_{B_T}. \tag*{(21)}

Consequently, the nonlinear operator

N(η,ω)=((ηω)x,ωωx)\mathcal{N}(\eta,\omega)=\bigl((\eta\omega)_x,\omega\omega_x\bigr)

satisfies, for (ηi,ωi)∈BT(\eta_i,\omega_i)\in B_T, i=1,2i=1,2,

∥N(η1,ω1)−N(η2,ω2)∥L1(0,T;X0)≤3T1/4(∥(η1,ω1)∥BT+∥(η2,ω2)∥BT)×∥(η1−η2,ω1−ω2)∥BT,\left\|\mathcal{N}(\eta_1,\omega_1)-\mathcal{N}(\eta_2,\omega_2)\right\|_{L^1(0,T;X_0)} \le3T^{1/4}\left(\left\|(\eta_1,\omega_1)\right\|_{B_T}+\left\|(\eta_2,\omega_2)\right\|_{B_T}\right) \times\left\|(\eta_1-\eta_2,\omega_1-\omega_2)\right\|_{B_T},

and

∥N(η,ω)∥L1(0,T;X0)≤3T1/4∥(η,ω)∥BT2.(22)\left\|\mathcal{N}(\eta,\omega)\right\|_{L^1(0,T;X_0)} \le3T^{1/4}\left\|(\eta,\omega)\right\|_{B_T}^{2}. \tag*{(22)}

The same holds on every interval [s,s+T][s,s+T].

Proof. Since w(0)=w(L)=0w(0)=w(L)=0,

w2(x)=∫0xw(y)w′(y) dy−∫xLw(y)w′(y) dy≤∥w∥∥wx∥.w^2(x)=\int_0^x w(y)w'(y)\,\mathrm{d}y-\int_x^L w(y)w'(y)\,\mathrm{d}y \le\|w\|\|w_x\|.

Thus, inequality (20) gives

∥(uv)x∥≤∥ux∥∥v∥1/2∥vx∥1/2+∥u∥1/2∥ux∥1/2∥vx∥.\|(uv)_x\|\le\|u_x\|\|v\|^{1/2}\|v_x\|^{1/2}+\|u\|^{1/2}\|u_x\|^{1/2}\|v_x\|.

Then, applying Hölder’s inequality after integrating in time,

∫0T∥ux(t)∥∥v(t)∥1/2∥vx(t)∥1/2 dt≤T1/4∥ux∥L2(0,T;L2)∥v∥C([0,T];L2)1/2∥vx∥L2(0,T;L2)1/2≤T1/4∥u∥BT∥v∥BT,\int_0^T\|u_x(t)\|\|v(t)\|^{1/2}\|v_x(t)\|^{1/2}\,\mathrm{d}t \le T^{1/4}\|u_x\|_{L^2(0,T;L^2)}\|v\|_{C([0,T];L^2)}^{1/2}\|v_x\|_{L^2(0,T;L^2)}^{1/2} \le T^{1/4}\|u\|_{B_T}\|v\|_{B_T},

where L2=L2(0,L)L^2=L^2(0,L). Exchanging uu and vv gives (21). On the other hand, for the nonlinear operator N\mathcal{N}, we can rewrite the terms as

η1ω1−η2ω2=η1(ω1−ω2)+(η1−η2)ω2,ω1ω1,x−ω2ω2,x=12((ω1+ω2)(ω1−ω2))x,\begin{aligned} \eta_1\omega_1-\eta_2\omega_2&=\eta_1(\omega_1-\omega_2)+(\eta_1-\eta_2)\omega_2,\\ \omega_1\omega_{1,x}-\omega_2\omega_{2,x}&=\frac{1}{2}\bigl((\omega_1+\omega_2)(\omega_1-\omega_2)\bigr)_x, \end{aligned}

and applying (21) to each term implies (22). Notice that time translation gives the same bounds on [s,s+T][s,s+T]. □

A solution of (1.1) on [0,T][0,T] is a state U=(η,ω,z)∈C([0,T];H)U=(\eta,\omega,z)\in C([0,T];H) with (η,ω)∈BT(\eta,\omega)\in B_T that satisfies Duhamel’s formula of the linear problem (3.1) with initial state U0U_0 and source −N(η,ω)∈L1(0,T;X0)-\mathcal{N}(\eta,\omega)\in L^1(0,T;X_0).

Theorem 4.2. Assume (1.2)–(1.3) and L>0L>0.

(a) Let T>0T>0. If 24CT2T1/4∥U0∥0≤124C_T^2T^{1/4}\|U_0\|_0\le1, then system (1.1) has a unique solution on [0,T][0,T], satisfying

∥(η,ω)∥BT≤2CT∥U0∥0.(23)\|(\eta,\omega)\|_{B_T}\le2C_T\|U_0\|_0. \tag*{(23)}

The data-to-solution map is Lipschitz on this set of initial states, with values in

BT×C([0,T];L2(0,1)).B_T\times C([0,T];L^2(0,1)).

(b) For every U0∈HU_0\in H, the conclusion of (a) holds on [0,T∗(∥U0∥0)][0,T_*(\|U_0\|_0)], where

T∗(∥U0∥0)={min⁡{1,(24CT2∥U0∥0)−4},if ∥U0∥0>0,1,if ∥U0∥0=0.T_*(\|U_0\|_0)= \begin{cases} \min\{1,(24C_T^2\|U_0\|_0)^{-4}\},&\text{if }\|U_0\|_0>0,\\ 1,&\text{if }\|U_0\|_0=0. \end{cases}

(c) There is a unique maximal solution on [0,Tmax⁡)[0,T_{\max}). If Tmax⁡<∞T_{\max}<\infty, then

sup⁡0≤t<Tmax⁡∥U(t)∥t=∞.\sup_{0\le t<T_{\max}}\|U(t)\|_t=\infty.

We highlight that the same conclusions hold for the problem started at any time s≥0s\ge0 from Us∈HU_s\in H, with ∥U0∥0\|U_0\|_0 replaced by ∥Us∥s\|U_s\|_s and the same constants.

Proof. (a) Let us consider R=2CT∥U0∥0R=2C_T\|U_0\|_0 be the radius of the closed ball

BR={(η,ω)∈BT:∥(η,ω)∥BT≤R}.\mathcal{B}_R=\left\{(\eta,\omega)\in B_T:\|(\eta,\omega)\|_{B_T}\le R\right\}.

For (η,ω)∈BR(\eta,\omega)\in\mathcal{B}_R, let Γ(η,ω)\Gamma(\eta,\omega) denote the first two components of the mild linear solution with initial state U0U_0 and source −N(η,ω)-\mathcal{N}(\eta,\omega). By Proposition 3.1, Lemma 4.1, and the assumption in (a),

∥Γ(η,ω)∥BT≤CT∥U0∥0+3CTT1/4∥(η,ω)∥BT2≤CT∥U0∥0+12CT3T1/4∥U0∥02≤2CT∥U0∥0,\begin{aligned} \|\Gamma(\eta,\omega)\|_{B_T} &\le C_T\|U_0\|_0+3C_TT^{1/4}\|(\eta,\omega)\|_{B_T}^{2}\\ &\le C_T\|U_0\|_0+12C_T^3T^{1/4}\|U_0\|_0^2\\ &\le2C_T\|U_0\|_0, \end{aligned}

and

∥Γ(η1,ω1)−Γ(η2,ω2)∥BT≤12CT3T1/4∥U0∥0∥(η1−η2,ω1−ω2)∥BT≤12∥(η1−η2,ω1−ω2)∥BT.\begin{aligned} \|\Gamma(\eta_1,\omega_1)-\Gamma(\eta_2,\omega_2)\|_{B_T} &\le12C_T^3T^{1/4}\|U_0\|_0 \|(\eta_1-\eta_2,\omega_1-\omega_2)\|_{B_T}\\ &\le\frac{1}{2}\|(\eta_1-\eta_2,\omega_1-\omega_2)\|_{B_T}. \end{aligned}

Thus Γ\Gamma is a contraction of the ball into itself, and Banach’s fixed point theorem yields a mild solution satisfying (23).

To prove uniqueness, let U=(η,ω,z)U=(\eta,\omega,z) and U~=(η~,ω~,z~)\widetilde{U}=(\widetilde{\eta},\widetilde{\omega},\widetilde{z}) be two mild solutions on [0,T][0,T] with the same initial state. For 0≤s<t≤T0\le s<t\le T with U(s)=U~(s)U(s)=\widetilde{U}(s), Proposition 3.1 and Lemmas 4.1 and 3.4 give

∥(η−η~,ω−ω~)∥Bs,t≤3CT(t−s)1/4(∥(η,ω)∥BT+∥(η~,ω~)∥BT)×∥(η−η~,ω−ω~)∥Bs,t.\|(\eta-\widetilde{\eta},\omega-\widetilde{\omega})\|_{B_{s,t}} \le3C_T(t-s)^{1/4}\left(\|(\eta,\omega)\|_{B_T}+\|(\widetilde{\eta},\widetilde{\omega})\|_{B_T}\right) \times\|(\eta-\widetilde{\eta},\omega-\widetilde{\omega})\|_{B_{s,t}}.

For t−st-s sufficiently small, one can claim that U=U~U=\widetilde{U} on [s,t][s,t]. Starting at s=0s=0 and iterating over finitely many such intervals proves uniqueness on [0,T][0,T].

Now, given two initial states satisfying (a), their difference can be estimated as follows

∥(η−η~,ω−ω~)∥BT≤CT∥U0−U~0∥0+12∥(η−η~,ω−ω~)∥BT,\|(\eta-\widetilde{\eta},\omega-\widetilde{\omega})\|_{B_T} \le C_T\|U_0-\widetilde{U}_0\|_0+\frac{1}{2}\|(\eta-\widetilde{\eta},\omega-\widetilde{\omega})\|_{B_T},

and hence Lipschitz property follows.

(b) Since CT≤C1C_T \le C_1 for T≤1T \le1, the stated choice of T∗(∥U0∥0)T_*(\|U_0\|_0) satisfies the condition in (a). For U0=0U_0=0, the solution is identically zero. The conclusions for a starting time ss follow from Lemma 3.4 and the uniformity of the linear estimates.

(c) Uniqueness allows the local solutions to be joined into a maximal solution. Suppose that Tmax⁡<∞T_{\max}<\infty and

sup⁡0≤t<Tmax⁡∥U(t)∥t<∞.\sup_{0\le t<T_{\max}}\|U(t)\|_t<\infty.

By (b), every restart from U(s)U(s), 0≤s<Tmax⁡0\le s<T_{\max}, has an existence time bounded below by

T∗(sup⁡0≤t<Tmax⁡∥U(t)∥t)>0.T_*\left(\sup_{0\le t<T_{\max}}\|U(t)\|_t\right)>0.

Choosing s<Tmax⁡s<T_{\max} sufficiently close to Tmax⁡T_{\max}, one extends the solution beyond Tmax⁡T_{\max}, a contradiction.

Proof of Theorem 1.3

Here, we are going to estimate the isolated identities that are involved in the Lyapunov estimates. Recall that the linear ones are already estimated.

Lemma 5.1. For η,ω∈H01(0,L)\eta,\omega\in H_0^1(0,L) and f=−N(η,ω)f=-\mathcal{N}(\eta,\omega),

∫0L(ηf1+ωf2) dx=−12∫0Lωxη2 dx,\int_0^L(\eta f_1+\omega f_2)\,dx=-\frac{1}{2}\int_0^L\omega_x\eta^2\,dx,
∫0Lx(ωf1+ηf2) dx=∫0Lηω2 dx,\int_0^Lx(\omega f_1+\eta f_2)\,dx=\int_0^L\eta\omega^2\,dx,

and

∣−12∫0Lωxη2 dx+μ1∫0Lηω2 dx∣≤K∥η∥∥∇U∥2.\left|-\frac{1}{2}\int_0^L\omega_x\eta^2\,dx+\mu_1\int_0^L\eta\omega^2\,dx\right|\le K\|\eta\|\|\nabla U\|^2.

Proof. Notice that all boundary terms vanish since η,ω∈H01(0,L)\eta,\omega\in H_0^1(0,L). Then, integrating by parts gives

∫0Lη(ηω)x dx=12∫0Lωxη2 dx,\int_0^L\eta(\eta\omega)_x\,dx=\frac{1}{2}\int_0^L\omega_x\eta^2\,dx,
∫0Lω2ωx dx=0,\int_0^L\omega^2\omega_x\,dx=0,
∫0Lxω(ηω)x dx=−∫0Lηω2 dx−∫0Lxηωωx dx.\int_0^Lx\omega(\eta\omega)_x\,dx=-\int_0^L\eta\omega^2\,dx-\int_0^Lx\eta\omega\omega_x\,dx.

Substituting f=−N(η,ω)f=-\mathcal{N}(\eta,\omega) proves the identities. By Lemma 4.1 and Poincaré’s inequality, we reach

12∣∫0Lωxη2 dx∣≤12∥ωx∥∥η∥L∞∥η∥≤14(Lπ)1/2∥η∥(∥ηx∥2+∥ωx∥2),\frac{1}{2}\left|\int_0^L\omega_x\eta^2\,dx\right| \le\frac{1}{2}\|\omega_x\|\|\eta\|_{L^\infty}\|\eta\| \le\frac{1}{4}\left(\frac{L}{\pi}\right)^{1/2}\|\eta\|(\|\eta_x\|^2+\|\omega_x\|^2),
∣∫0Lηω2 dx∣≤∥η∥∥ω∥L∞∥ω∥≤(Lπ)3/2∥η∥∥ωx∥2.\left|\int_0^L\eta\omega^2\,dx\right| \le\|\eta\|\|\omega\|_{L^\infty}\|\omega\| \le\left(\frac{L}{\pi}\right)^{3/2}\|\eta\|\|\omega_x\|^2.

The estimate follows from the definition of KK in (1.10).

Now, with the previous estimates in hand, we can prove our main Theorem. For simplicity, we are going to split the proof in three steps

Proof of Theorem 1.3. Let UU be the maximal solution given by Theorem 4.2.

Step 1. Lyapunov estimate. For every T<Tmax⁡T<T_{\max}, Lemma 4.1 gives −N(η,ω)∈L1(0,T;X0)-\mathcal{N}(\eta,\omega)\in L^1(0,T;X_0). Proposition 3.1 therefore yields V∈W1,1(0,T)V\in W^{1,1}(0,T). Combining (3.2)–(3.4) with Lemma 5.1, we obtain

V′+λV≤12(a  b)Ψ(ab)−∣β∣2(d−τ˙)b2+μ1+λ2∥U∥2+λμ1V1−3μ12∥∇U∥2+K∥η∥∥∇U∥2+∣β∣2∫01[−μ2(1−τ˙ρ)+λτ(1+μ2(1−ρ))]z2 dρ,V'+\lambda V\le\frac{1}{2}(a\ \ b)\Psi\begin{pmatrix}a\\b\end{pmatrix} -\frac{|\beta|}{2}(d-\dot{\tau})b^2+\frac{\mu_1+\lambda}{2}\|U\|^2+\lambda\mu_1V_1 -\frac{3\mu_1}{2}\|\nabla U\|^2+K\|\eta\|\|\nabla U\|^2 +\frac{|\beta|}{2}\int_0^1\left[-\mu_2(1-\dot{\tau}\rho)+\lambda\tau(1+\mu_2(1-\rho))\right]z^2\,d\rho,

for almost every t∈(0,Tmax⁡)t\in(0,T_{\max}). The first two terms are nonpositive by Definition 1.2 and (1.2). Now, recalling that

∣V1∣≤L2∥U∥2,∥U∥2≤L2π2∥∇U∥2,1−τ˙ρ≥1−d,|V_1|\le\frac{L}{2}\|U\|^2,\qquad\|U\|^2\le\frac{L^2}{\pi^2}\|\nabla U\|^2,\qquad1-\dot{\tau}\rho\ge1-d,

together with τ≤M\tau\le M, we can claim that

V′+λV≤[L22π2(μ1+λ(1+μ1L))−3μ12+K∥η∥]∥∇U∥2+∣β∣2[λM(1+μ2)−μ2(1−d)]∥z∥L2(0,1)2.V'+\lambda V\le\left[\frac{L^2}{2\pi^2}\bigl(\mu_1+\lambda(1+\mu_1L)\bigr)-\frac{3\mu_1}{2}+K\|\eta\|\right]\|\nabla U\|^2 +\frac{|\beta|}{2}\left[\lambda M(1+\mu_2)-\mu_2(1-d)\right]\|z\|_{L^2(0,1)}^2.

By (1.9) and the definition of δ\delta in (1.10), we conclude that

V′(t)+λV(t)≤(−δ+K∥η(t)∥)∥∇U(t)∥2(24)V'(t)+\lambda V(t)\le(-\delta+K\|\eta(t)\|)\|\nabla U(t)\|^2 \tag*{(24)}

for almost every t∈(0,Tmax⁡)t\in(0,T_{\max}).

Step 2. Energy decay and global existence. The smallness assumption (1.11) gives

∥η(0)∥≤2E(0)≤2ζE(0)<δK.\|\eta(0)\|\le\sqrt{2E(0)}\le\sqrt{2\zeta E(0)}<\frac{\delta}{K}.

As long as ∥η(t)∥<δ/K\|\eta(t)\|<\delta/K, integrating (5.1) and using (2.1) yields

E(t)≤V(0)1−μ1Le−λt≤ζE(0)e−λt.(25)E(t)\le\frac{V(0)}{1-\mu_1L}e^{-\lambda t}\le\zeta E(0)e^{-\lambda t}. \tag*{(25)}

If t∗<Tmax⁡t_*<T_{\max} were the first time at which ∥η(t∗)∥=δ/K\|\eta(t_*)\|=\delta/K, continuity and (5.2) implies

δK=∥η(t∗)∥≤2ζE(0)e−λt∗/2<δK,\frac{\delta}{K}=\|\eta(t_*)\|\le\sqrt{2\zeta E(0)}e^{-\lambda t_*/2}<\frac{\delta}{K},

a contradiction. Thus (5.2) holds on [0,Tmax⁡)[0,T_{\max}). In particular,

sup⁡0≤t<Tmax⁡∥U(t)∥t2=sup⁡0≤t<Tmax⁡E(t)≤2ζE(0),\sup_{0\le t<T_{\max}}\|U(t)\|_t^2=\sup_{0\le t<T_{\max}}E(t)\le2\zeta E(0),

so Theorem 4.2(c) gives Tmax⁡=∞T_{\max} = \infty.

Step 3. Uniform bound. Since K∥η(t)∥≤K2ζE(0)<δK\|\eta(t)\| \le K\sqrt{2\zeta E(0)} < \delta, integrating (5.1) gives

V(t)+λ∫0tV(r) dr+(δ−K2ζE(0))∫0t∥∇U(r)∥2 dr≤V(0).V(t) + \lambda\int_0^t V(r)\,dr + \left(\delta- K\sqrt{2\zeta E(0)}\right)\int_0^t \|\nabla U(r)\|^2\,dr \le V(0).

Using V≥0V \ge0 and (2.1), and letting t→∞t \to\infty, we obtain

∫0∞∥∇U(t)∥2 dt≤(1+max⁡{μ1L,μ2})E(0)δ−K2ζE(0).\int_0^\infty\|\nabla U(t)\|^2\,dt \le\frac{(1+\max\{\mu_1L,\mu_2\})E(0)}{\delta-K\sqrt{2\zeta E(0)}}.

Together with the bound

sup⁡t≥0∥U(t)∥≤2ζE(0),\sup_{t\ge0}\|U(t)\| \le\sqrt{2\zeta E(0)},

this proves (1.12). □

Remark 5.2 (A perturbative proof). Local exponential stability can also be deduced from the linear decay by a perturbation argument, in the spirit of [9, 10]. We sketch it. Fix T>0T>0 such that ζe−λT≤14\zeta e^{-\lambda T}\le\frac14, and let U0U_0 satisfy

24CT2T1/4∥U0∥0≤14.24C_T^2T^{1/4}\|U_0\|_0 \le\frac14.

By Theorem 4.2(a), the solution UU exists on [0,T][0,T] and ∥(η,ω)∥BT≤2CT∥U0∥0\|(\eta,\omega)\|_{B_T}\le2C_T\|U_0\|_0. Write U=Ulin+WU=U_{\mathrm{lin}}+W, where Ulin(t)=S(t,0)U0U_{\mathrm{lin}}(t)=S(t,0)U_0 solves the linearized system and WW is the mild solution of (3.1) with zero initial state and source −N(η,ω)-\mathcal{N}(\eta,\omega). Since ∥Ulin(T)∥T2\|U_{\mathrm{lin}}(T)\|_T^2 is twice the energy of UlinU_{\mathrm{lin}} at time TT, [1 Theorem 1.1], with the corrections of Section 2, gives

∥Ulin(T)∥T≤(ζe−λT)1/2∥U0∥0≤12∥U0∥0.\|U_{\mathrm{lin}}(T)\|_T \le\left(\zeta e^{-\lambda T}\right)^{1/2}\|U_0\|_0 \le\frac12\|U_0\|_0.

while Proposition 3.1 and (4.3) give

∥W(T)∥T≤6T1/4∥(η,ω)∥BT2≤24CT2T1/4∥U0∥02.\|W(T)\|_T \le6T^{1/4}\|(\eta,\omega)\|_{B_T}^2 \le24C_T^2T^{1/4}\|U_0\|_0^2.

Hence

∥U(T)∥T≤(12+24CT2T1/4∥U0∥0)∥U0∥0≤34∥U0∥0.\|U(T)\|_T \le\left(\frac12+24C_T^2T^{1/4}\|U_0\|_0\right)\|U_0\|_0 \le\frac34\|U_0\|_0.

In particular, U(T)U(T) satisfies the same smallness condition as U0U_0 and, by Lemma 3.4, the problem can be restarted at time TT with the same constants. Iterating, ∥U(nT)∥nT≤(3/4)n∥U0∥0\|U(nT)\|_{nT}\le(3/4)^n\|U_0\|_0 for every n≥1n\ge1, which yields the exponential decay of EE. This argument provides a decay rate depending on TT and a less explicit smallness condition, whereas the Lyapunov proof retains the linear rate λ\lambda and the explicit threshold (1.11).

Numerics and concluding remarks

Numerical illustration and verification

We illustrate Theorem 1.3 for system (1.1) with L=5L=5, α=1\alpha=1, β=1/2\beta=1/2 and τ(t)=2+sin⁡(t/2)\tau(t)=2+\sin(t/2), so that 1≤τ(t)≤31\le\tau(t)\le3 and τ˙(t)≤1/2\dot{\tau}(t)\le1/2. The admissible choice

(μ1,μ2)≃(0.04777231,0.04455379),(\mu_1,\mu_2)\simeq(0.04777231,0.04455379),

with λ=λ∗/2≃0.00355445\lambda=\lambda^*/2\simeq0.00355445, gives ζ≃1.62764\zeta\simeq1.62764 and

r2≃5.64783×10−5.r^2\simeq5.64783\times10^{-5}.

For zero initial history, we take

η0(x)=As2(1−s)2(1+2s),ω0(x)=As(1−s)2,\eta_0(x)=As^2(1-s)^2(1+2s),\qquad\omega_0(x)=As(1-s)^2,

with s=x/Ls=x/L, where A>0A>0 is chosen so that E(0)=κr2E(0)=\kappa r^2 for κ∈{0.1,0.5,0.9}\kappa\in\{0.1,0.5,0.9\}. These smooth data satisfy the boundary compatibility conditions and the strict smallness assumption E(0)<r2E(0)<r^2.

We use Legendre–Galerkin spaces of degrees NN and N−1N-1 for η\eta and ω\omega, respectively, imposing the boundary conditions and the feedback through the boundary term. Gauss quadrature integrates the polynomial terms exactly. Time integration uses second-order semi-implicit backward differentiation, with explicit extrapolation of the nonlinearity and a backward-Euler starting step. The delayed trace is linearly interpolated from the prescribed delayed profile and previously computed values. The primary runs use N=24N=24, Δt=0.005\Delta t=0.005 and T=400T=400. Writing aN(t)=ηN,x(t,L)a_N(t)=\eta_{N,x}(t,L), we compute

EN(t)=12∫0L(ηN2+ωN2) dx+∣β∣2∫t−τ(t)taN(s)2 ds.E_N(t)=\frac12\int_0^L(\eta_N^2+\omega_N^2)\,dx+\frac{|\beta|}{2}\int_{t-\tau(t)}^t a_N(s)^2\,ds.

Figure 2 compares the computed energies with ζEN(0)e−λt\zeta E_N(0)e^{-\lambda t}. The normalized trajectories almost coincide, and their maximum ratios to this envelope are below 0.615 on [0,400][0,400]. Additional experiments with the negative primary profile, a sign-changing profile and compatible nonzero history, at total energy approximately r2/2r^2/2, also remain below the envelope; their definitions are supplied in [11].

Figure 3 reports numerical resolution checks. A manufactured nonlinear solution with a nonzero delayed profile and time-varying delay yields temporal orders between 2.001 and 2.005. A manufactured nonlinear solution with a nonzero delayed profile and time-varying delay yields temporal orders between 2.001 and 2.005. For κ=0.5\kappa=0.5, the maximum energy difference on [0,100][0,100], normalized by EN(0)E_N(0), is 2.89×10−52.89\times10^{-5} when comparing Δt=0.005\Delta t=0.005 with N=32N=32, and 9.21×10−109.21\times10^{-10} when comparing N=24N=24 with 32 at Δt=0.0025\Delta t=0.0025. The boundary-feedback residual also decreases with NN. The larger-amplitude curves test numerical resolution outside the theorem’s range. These findings support the reported computations; they are neither rigorous error bounds, they are mainly designed to verify the findings without being a numerical test.

Figure 3. Numerical verification. Left: manufactured-solution L2L^2 errors at t=2t = 2 and a second-order reference line. Centre: energy differences on [0,100][0, 100] at N=32N = 32, relative to Δt=0.00125\Delta t = 0.00125. Right: normalized time-L2L^2 boundary-feedback residuals at Δt=0.0025\Delta t = 0.0025. The refinement curves correspond to κ=0.5\kappa= 0.5 and 3000030000; the latter is a test of numerical resolution outside the theorem’s small-data regime.

Final comments

  1. Theorem 1.3 establishes global existence and exponential decay for sufficiently small initial data, with

Figure 2. Small-data experiments with N=24N = 24, Δt=0.005\Delta t = 0.005 and τ(t)=2+sin⁡(t/2)\tau(t) = 2 + \sin(t/2). Left: normalized energies for κ=0.1,0.5,0.9\kappa= 0.1, 0.5, 0.9 and the theoretical envelope ζe−λt\zeta e^{-\lambda t}. Right: ratios to the envelope for four compatible initial states with EN(0)≃r2/2E_N(0) \simeq r^2/2, shown on [0,100][0, 100]. The horizontal dashed line marks equality with the envelope.

the same admissible range of decay rates as in the linear estimate. Global existence for arbitrary data remains open for the feedback considered here. The difficulty lies in the cubic contribution −12∫0Lαxη2 dx-\frac{1}{2}\int_0^L \alpha_x\eta^2\,\mathrm{d}x, which has no prescribed sign and prevents the present estimates from providing a global energy bound without a smallness assumption. This difficulty persists in the absence of delay.

  1. The decay rate and the smallness threshold depend on the admissible pair (μ1,μ2)(\mu_1,\mu_2). For a fixed pair, increasing λ\lambda decreases δ\delta and therefore reduces the threshold in (1.11). Maximizing the decay bound and maximizing the size of the guaranteed neighborhood of attraction are thus different optimization problems. As shown in (C8) of Section 2, optimization along the curve of (C4) need not give the best decay bound over the full admissible set. These Lyapunov estimates do not determine the optimal decay rate of the system or the full basin of attraction.

  1. The restriction L<3πL < \sqrt{3}\pi arises from Poincaré’s inequality in the Lyapunov estimate. The stability result cannot hold for every length: if L=2πnL = 2\pi n, with n≥1n \ge1, then (η,ω,z)=(c(1−cos⁡x),0,0)(\eta,\omega,z) = (c(1-\cos x),0,0), c∈R∖{0}c \in\mathbb{R} \setminus\{0\}, is a stationary solution of (1.1), with zero initial delayed profile, for every delay satisfying (1.2) ...

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

The Article Processing Charge (APC) for the publication of this research was funded by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior -Brasil (CAPES) (ROR identifier: 00x0ma614).

This work was completed while J. R. Muñoz was visiting R. de A. Capistrano–Filho and V. H. Gonzalez Martinez at UFPE. J. R. Muñoz acknowledges the support of ANID, Chile, through FONDECYT Postdoctorado Grant No. 3260141. R. de A. Capistrano–Filho was partially supported by CAPES/COFECUB grant number 88887.879175/2023-00, CNPq grant numbers 301744/2025-4 and 421573/2023-6, and PROPG (UFPE) via PROAP resources. V. H. Gonzalez Martinez was supported by CAPES/COFECUB grant 88887.879175/2023-00 and CNPq grants 421573/2023-6, 402652/2024-0 and 300997/2025-6.

CRediT authorship contribution statement

Roberto de A. Capistrano–Filho: Conceptualization, Formal analysis, Writing – original draft. Boumediène Chentouf: Formal analysis, Writing – review & editing. Victor H. Gonzalez Martinez: Formal analysis, Writing – review & editing. Juan Ricardo Muñoz: Formal analysis, Software, Validation, Writing – review & editing.

Data availability

The code used for the numerical illustrations in § 6 is available in [11].

References

  1. [1]R. de A. Capistrano–Filho, B. Chentouf, V. H. Gonzalez Martinez, J. R. Muñoz, On the boundary stabilization of the KdV–KdV system with time-dependent delay, Nonlinear Anal. Real World Appl. 79 (2024) 104122.
  2. [2]V. H. Gonzalez Martinez, J. R. Muñoz, Feedback boundary stabilization for the Hirota–Satsuma system with time-delay, arXiv preprint arXiv:2408.15705 (2024).arxiv.org/abs/2408.15705
  3. [3]J. L. Bona, M. Chen, J.-C. Saut, Boussinesq equations and other systems for small-amplitude long waves in nonlinear dispersive media. II. The nonlinear theory, Nonlinearity 17 (2004) 925–952.DOI
  4. [4]A. F. Pazoto, L. Rosier, Stabilization of a Boussinesq system of KdV–KdV type, Systems Control Lett. 57 (2008) 595–601.DOI
  5. [5]R. A. Capistrano–Filho, A. F. Pazoto, L. Rosier, Control of Boussinesq system of KdV–KdV type on a bounded interval, ESAIM Control Optim. Calc. Var. 25 (2019) 58.
  6. [6]S. Nicaise, J. Valein, E. Fridman, Stability of the heat and of the wave equations with boundary time-varying delays, Discrete Contin. Dyn. Syst. Ser. S 2 (2009) 559–581.
  7. [7]T. Kato, Linear evolution equations of “hyperbolic” type, J. Fac. Sci. Univ. Tokyo Sect. I 17 (1970) 241–258.
  8. [8]A. Pazy, Semigroups of Linear Operators and Applications to Partial Differential Equations, Springer, New York, 1983.DOI
  9. [9]L. Baudouin, E. Crépeau, J. Valein, Two approaches for the stabilization of nonlinear KdV equation with boundary time-delay feedback, IEEE Trans. Automat. Control 64 (2019) 1403–1414.arxiv.org/abs/1711.09696
  10. [10]H. Parada, C. Timimoun, J. Valein, Stability results for the KdV equation with time-varying delay, Systems Control Lett. 177 (2023) 105547.DOI
  11. [11]R. de A. Capistrano–Filho, B. Chentouf, V. H. Gonzalez Martinez, J. R. Muñoz, KdV–KdV Delay Numerics (version 1.0.0), Zenodo, 2026. https://doi.org/10.5281/zenodo.24283843.

Paper details

vlnc:W_k6usnckr
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents