
This paper investigates the global asymptotic stability problem for a class of Takagi-Sugeno fuzzy Clifford-valued delayed neural networks with impulsive effects and leakage delays using the system decomposition method. By applying Takagi-Sugeno fuzzy theory, we first consider a general form of Takagi-Sugeno fuzzy Clifford-valued delayed neural networks. Then, we decompose the considered n-dimensional Clifford-valued systems into 2mn-dimensional real-valued systems in order to avoid the inconvenience caused by the non-commutativity of the multiplication of Clifford numbers. By using Lyapunov-Krasovskii functionals and integral inequalities, we derive new sufficient criteria to guarantee the global asymptotic stability for the considered neural networks. Further, the results of this paper are presented in terms of real-valued linear matrix inequalities, which can be directly solved using the MATLAB LMI toolbox. Finally, a numerical example is provided with their simulations to demonstrate the validity of the theoretical analysis.
Citation: Abdulaziz M. Alanazi, R. Sriraman, R. Gurusamy, S. Athithan, P. Vignesh, Zaid Bassfar, Adel R. Alharbi, Amer Aljaedi. System decomposition method-based global stability criteria for T-S fuzzy Clifford-valued delayed neural networks with impulses and leakage term[J]. AIMS Mathematics, 2023, 8(7): 15166-15188. doi: 10.3934/math.2023774
[1] | Ekram E. Ali, Miguel Vivas-Cortez, Rabha M. El-Ashwah . New results about fuzzy γ-convex functions connected with the q-analogue multiplier-Noor integral operator. AIMS Mathematics, 2024, 9(3): 5451-5465. doi: 10.3934/math.2024263 |
[2] | Shahid Khan, Saqib Hussain, Maslina Darus . Inclusion relations of q-Bessel functions associated with generalized conic domain. AIMS Mathematics, 2021, 6(4): 3624-3640. doi: 10.3934/math.2021216 |
[3] | Saqib Hussain, Shahid Khan, Muhammad Asad Zaighum, Maslina Darus . Certain subclass of analytic functions related with conic domains and associated with Salagean q-differential operator. AIMS Mathematics, 2017, 2(4): 622-634. doi: 10.3934/Math.2017.4.622 |
[4] | Xiaoli Zhang, Shahid Khan, Saqib Hussain, Huo Tang, Zahid Shareef . New subclass of q-starlike functions associated with generalized conic domain. AIMS Mathematics, 2020, 5(5): 4830-4848. doi: 10.3934/math.2020308 |
[5] | Jianhua Gong, Muhammad Ghaffar Khan, Hala Alaqad, Bilal Khan . Sharp inequalities for q-starlike functions associated with differential subordination and q-calculus. AIMS Mathematics, 2024, 9(10): 28421-28446. doi: 10.3934/math.20241379 |
[6] | Shujaat Ali Shah, Ekram Elsayed Ali, Adriana Cătaș, Abeer M. Albalahi . On fuzzy differential subordination associated with q-difference operator. AIMS Mathematics, 2023, 8(3): 6642-6650. doi: 10.3934/math.2023336 |
[7] | Ibtisam Aldawish, Mohamed Aouf, Basem Frasin, Tariq Al-Hawary . New subclass of analytic functions defined by q-analogue of p-valent Noor integral operator. AIMS Mathematics, 2021, 6(10): 10466-10484. doi: 10.3934/math.2021607 |
[8] | Huo Tang, Kadhavoor Ragavan Karthikeyan, Gangadharan Murugusundaramoorthy . Certain subclass of analytic functions with respect to symmetric points associated with conic region. AIMS Mathematics, 2021, 6(11): 12863-12877. doi: 10.3934/math.2021742 |
[9] | F. Müge Sakar, Arzu Akgül . Based on a family of bi-univalent functions introduced through the Faber polynomial expansions and Noor integral operator. AIMS Mathematics, 2022, 7(4): 5146-5155. doi: 10.3934/math.2022287 |
[10] | Ebrahim Amini, Mojtaba Fardi, Shrideh Al-Omari, Rania Saadeh . Certain differential subordination results for univalent functions associated with q-Salagean operators. AIMS Mathematics, 2023, 8(7): 15892-15906. doi: 10.3934/math.2023811 |
This paper investigates the global asymptotic stability problem for a class of Takagi-Sugeno fuzzy Clifford-valued delayed neural networks with impulsive effects and leakage delays using the system decomposition method. By applying Takagi-Sugeno fuzzy theory, we first consider a general form of Takagi-Sugeno fuzzy Clifford-valued delayed neural networks. Then, we decompose the considered n-dimensional Clifford-valued systems into 2mn-dimensional real-valued systems in order to avoid the inconvenience caused by the non-commutativity of the multiplication of Clifford numbers. By using Lyapunov-Krasovskii functionals and integral inequalities, we derive new sufficient criteria to guarantee the global asymptotic stability for the considered neural networks. Further, the results of this paper are presented in terms of real-valued linear matrix inequalities, which can be directly solved using the MATLAB LMI toolbox. Finally, a numerical example is provided with their simulations to demonstrate the validity of the theoretical analysis.
From an ecological and financial perspective, toxicants have emerged as a major threat to terrestrial and aquatic environments. With increasing demand, industries are cranking out a flood of toxic chemicals. Toxic chemicals and substances, such as cadmium, arsenic, copper, lead, etc., are often dumped into lakes, rivers and oceans, where they can have a devastating effect on aquatic life [1]. Toxic oil, metals and synthetic organic chemicals are common contaminants of river, lake, and sea water [2]. A significant loss of biodiversity occurs in ecosystems where toxic contaminants are present [3]. Fish, birds and mammals that eat contaminated marine life can be exposed to the toxins themselves. Therefore, many species have become extinct, and many more are on the verge of extinction, due to the unchecked release of toxic substances into the environment. There are numerous species in the ocean that produce toxins, and these toxins, if released into the environment, can have serious consequences for the development of other organisms. For example, the grazing pressure of zooplankton can be greatly reduced by phytoplankton that are naturally toxic. Therefore, research into the effects of toxic substances on ecological communities is becoming increasingly significant from both an environmental and a conservationist point of view.
The mathematical modeling of the impact of toxicants on a population was a newly established area in the early 1980s [4,5,6]. In order to effectively estimate the qualitative impact of toxic substances on species, mathematical models are a great tool to use. Das et al. [7] investigated a predator-prey fishery model with harvesting and the effects of toxicants which are released by some other species. Chakraborty and Das [8] studied a two-zooplankton one-phytoplankton system in the presence of toxicity. Ang and Safuan [9] discussed an intraguild fishery model in which predators are thought to become infected through their prey, while fish are thought to be infected directly by an anthropogenic toxicant in the environment. Juneja and Agnihotri [10] addressed the issue of two competing fish species, each of which releases a harmful chemical for the other.
It is noted that in biological populations, delay plays an important role. In the last few years, theoretical and mathematical ecologists have paid a lot of attention to research on differential equations with time delays [11,12,13,14,15]. Even so, studying the effects of time delay on the dynamics of a system can be very complicated. For example, it can cause the system to lose its stability and lead to periodic solutions and chaotic behavior. Pal and Mahapatra [16] studied the combined effect of a toxicant and delay on the dynamical behaviors of a delayed two-species competitive system with imprecise biological parameters. Pal et al. [17] took into account two fish species that are in competition with one another, each of which releases a toxin that is poisonous to the other and each of which obeys the law of logistic growth. Meanwhile, in reality, species are spatially heterogeneous, so individuals seek out low population densities where they have a better chance of survival. As a result, the reaction-diffusion predator-prey model with toxic effects has been considered by some researchers. Zhang and Zhao [18] investigated a diffusive predator-prey model with toxins, and their research results show that toxic substances have a great impact on the dynamics of the model. Zhu et al. [19] investigated a delayed diffusive predator-prey model affected by a toxic substance. However, we find that the research results about the delayed diffusive predator-prey model with toxic substances are rare.
Motivated by these pioneer works, we hypothesize that prey produce toxins for predators, and that this process is not instantaneous but rather follows a discrete time lag that can be thought of as the species' maturation period. Toxic substances released by prey into the environment have a half-life of τ, which we introduce here. We consider a diffusive predator-prey model as follows
{∂u∂t=d1Δu+ru(1−uK)−αuva+u2,(x,t)∈Ω×(0,+∞),∂v∂t=d2Δv+βuva+u2−dv−γu(t−τ)v2,(x,t)∈Ω×(0,+∞),u(x,t)=u0(x,t),v(x,t)=v0(x,t),x∈¯Ω,t∈[−τ,0],∂u(t,x)∂n=∂v(t,x)∂n=0,t>0,x∈∂Ω, | (1.1) |
where u(x,t), v(x,t) denote the density of the prey and the predator, respectively. r is the birth rate of prey. ua+u2 is the Holling type-Ⅳ function. α is the maximum predator per capita consumption rate of u due to v. β is the conversion of the biomass constant. The parameter d is the death rate of predator. γv2 is the functional response of the u population to the density of the v population.
To explore the dynamics of system (1.1), we first do the non-dimensionalization described below
ˉt=rt,ˉu=uK,ˉv=αvK2r,ˉa=aK2,ˉβ=βKr,ˉd=dr,ˉγ=γK3α,ˉd1=d1r,ˉd2=d2r,ˉτ=rτ. |
Thus, system (1.1) is simplified (by removing the bars) to be
{∂u∂t=d1Δu+u(1−u)−uva+u2,(x,t)∈Ω×(0,+∞),∂v∂t=d2Δv+βuva+u2−dv−γu(t−τ)v2,(x,t)∈Ω×(0,+∞),u(x,t)=u0(x,t),v(x,t)=v0(x,t),x∈¯Ω,t∈[−τ,0],∂u(t,x)∂n=∂v(t,x)∂n=0,t>0,x∈∂Ω. | (1.2) |
In this study, we will study the dynamics of system (1.2), such as the existence of the solutions, local/global stability of the equilibria, and Hopf bifurcation induced by delay. In addition, we will also discuss the existence and non-existence of non-constant positive solutions of the following elliptic system
{−d1Δu=u(1−u)−uva+u2,x∈Ω,−d2Δv=βuva+u2−dv−γuv2,x∈Ω,∂u∂n=∂v∂n=0,x∈∂Ω. | (1.3) |
The structure of this paper is as follows. In Section 2, the existence of solutions and the persistence of system (1.2) are studied. In Section 3, the stability of the equilibria, Turing bifurcation, and Hopf bifurcation induced by delay are discussed. In Section 4, the global stability of the equilibria are investigated by using the Lyapunov functional method. In Section 5, the non-existence and existence of the non-constant steady state are studied. In Section 6, theoretical results are verified through numerical simulations. Finally, a brief conclusion is given in Section 7.
Theorem 2.1. Assume that u0(x,t)≥0, v0(x,t)≥0, and u0(x,t)≢0, v0(x,t)≢0. There is a unique solution (u(x,t),v(x,t))>(0,0), (t>0,x∈ˉΩ) of system (1.2), and
limsupmaxt→+∞x∈ˉΩu(x,t)≤1,∫Ωv(x,t)dx≤eK(m+r)m|Ω|. | (2.1) |
Additionally,
||u(⋅,t)||C(ˉΩ)≤K1,||v(⋅,t)||C(ˉΩ)≤K2, | (2.2) |
where ‖, \Vert v(x, t)\Vert_{C(\overline{\Omega})} = \mathop{\max}\limits_{x\in\overline{\Omega}} v(x, t) , K_1 = \max\{1, \max\limits_{\bar{\Omega}, t\in[-\tau, 0]}u_0(x, t)\} and K_2 depends upon \beta , a , d , \gamma , u_0(x, t) , v_0(x, t) and \Omega .
Proof. We consider the following auxiliary system
\begin{eqnarray} \left\{\begin{array}{lcr} \dfrac{\partial u}{\partial t} = d_1 \Delta u+u(1-u)-\dfrac{uv}{a+u^2}, \\ \dfrac{\partial v}{\partial t} = d_2\Delta v+v(\dfrac{\beta u}{a}-d), \\ \dfrac{\partial u}{\partial n} = \dfrac{\partial v}{\partial n} = 0, \\ u(x, 0) = u_0(x), v(x, 0) = v_0(x). \end{array}\right. \end{eqnarray} | (2.3) |
Obviously, (\underline{u}(x, t), \underline{v}(x, t)) = (0, 0) and (\bar{u}(x, t), \bar{v}(x, t)) = (\tilde{u}(t), \tilde{v}(t)) are the lower and upper solutions of system (2.3), respectively, where (\tilde{u}(t), \tilde{v}(t)) is the unique solution of
\begin{eqnarray} \left\{\begin{array}{lcr} \dfrac{d u}{d t} = u(1-u)-\dfrac{uv}{a+u^2}, \\ \dfrac{d v}{d t} = v(\dfrac{\beta u}{a}-d), \\ u(0) = \bar{u}_0, v(0) = \bar{v}_0, \\ \end{array}\right. \end{eqnarray} | (2.4) |
where \bar{u}_0 = \max\limits_{\overline{\Omega}}u_0(x) , \bar{v}_0 = \max \limits_{\overline{\Omega}}v_0(x) . As a result, according to Theorem 8.3.3 in [20], we obtain that system (1.2) has a unique globally defined solution which satisfies 0\leq u(x, t) \leq \tilde{u}(t), 0\leq v(x, t)\leq \tilde{v}(t). By the strong maximum principle, we have that u(x, t), v(x, t) > 0 (t > 0, x\in \bar{\Omega}) .
Evidently, from the first equation of system (2.4) we have that \lim \limits_{t\to+\infty}{u(t)}\leq1 , which implies \mathop {\lim \sup \max }\limits_{t \to + \infty \quad x \in \bar{\Omega}} u(x, t) \leq 1 . Therefore, ||u(\cdot, t) ||_{C(\bar{\Omega})}\leq K_1 for all t\geq 0 .
For v(x, t) , we let U(t) = \int_\Omega u(x, t)dx and V(t) = \int_\Omega v(x, t)dx ; then,
\begin{equation} \begin{split} & \frac{dU}{dt} = \int_\Omega u_tdx = d_1\int_\Omega\Delta udx+\int_\Omega[u(1-u)-\frac{ uv}{a+u^2}]dx = \int_\Omega[u(1-u)-\frac{ uv}{a+u^2}]dx, \end{split} \end{equation} | (2.5) |
\begin{equation} \frac{dV}{dt} = \int_\Omega v_tdx = d_2\int_\Omega\Delta vdx+\int_\Omega(v(\frac{\beta u}{a+u^2}- d-\gamma u v))dx = \int_\Omega v(\frac{\beta u}{a+u^2}- d-\gamma u(t-\tau) v)dx. \end{equation} | (2.6) |
Multiplying Eq (2.5) by \beta , and then addin it to Eq (2.6), we have
\begin{equation*} \begin{split} (\beta U+V)_t& = -d V+\beta \int_\Omega (u(1-u)dx -\gamma\int_\Omega u(t-\tau)v^2dx \leq -d(\beta U+V)+(1+d\beta)U. \end{split} \end{equation*} |
||u(\cdot, t) ||_{C(\bar{\Omega})}\leq K_1 , so we have that U(t)\leq K_1 |\Omega| . Thus,
\begin{equation} (\beta U+V)_t \leq -d(\beta U+V)+M_2, t > 0, \end{equation} | (2.7) |
where M_2 = (1+d\beta)|\Omega| . Integrating the inequality (2.7), we obtain
\begin{equation} \int_\Omega v(x, t)dx = V(t) < \beta U(t)+V(t)\leq (\beta U(0)+V(0))e^{-d t}+\frac{M_2}{d}(1-e^{-d t}). \end{equation} | (2.8) |
This means that ||v(\cdot, t) ||_{L^1(\Omega)}\leq \beta ||u_0(\cdot) ||_{L^1(\Omega)}+||v_0(\cdot) ||_{L^1(\Omega)}+\frac{M_2}{d}. According to Theorem 3.1 [21], there is an M_3 such that ||v(\cdot, t) ||_{L^\infty(\Omega)}\leq M_3 . Therefore, there exists a K_2 such that ||v(\cdot, t) ||_{C(\overline{\Omega})}\leq K_2 .
Theorem 2.2. If 1-\frac{K_2}{a} > 0 and \frac{\beta (1-\frac{K_2}{a})}{\gamma K_2(a+K_1)}-\frac{d}{\gamma K_2} > 0 , then system (1.2) has the persistence property.
Proof. From system (1.2), we obtain
\begin{equation} \begin{split} \frac{\partial u}{\partial t}& = d_1 \Delta u+ u(1-u-\frac{v}{a+u^2}) \geq d_1 \Delta u+u(1-u-\frac{K_2}{a}). \end{split} \end{equation} | (2.9) |
For small enough \varepsilon > 0 , it holds that 1-\frac{K_2}{a}- \varepsilon > 0 . Therefore, there is a t_1 such that
\begin{equation} u(x, t)\geq 1-\frac{K_2}{a}- \varepsilon: = \underline{c}_1, t > t_1. \end{equation} | (2.10) |
The second equation of system (1.2) is then solved using the upper and lower bounds of u , yielding
\begin{equation} \begin{split} \frac{\partial v}{\partial t}& = d_2\Delta v + \frac{\beta uv}{a+u^2}-dv-\gamma u(t-\tau) v^2, \geq d_2\Delta v + v(\frac{\beta \underline{c}_1}{a+K_1^2}-d-\gamma K_2v), \\ \end{split} \end{equation} | (2.11) |
for t > t_1+\tau . Then there exists t_2 > t_1+\tau such that for any t > t_2 ,
\begin{equation} v(x, t)\geq \frac{\beta \underline{c}_1 }{\gamma K_2(a+K_1)}-\frac{d }{\gamma K_2}: = \underline{c}_2. \end{equation} | (2.12) |
From (2.10) and (2.12), we can easily obtain that
\begin{equation} \mathop {\lim \inf \max }\limits_{t \to + \infty\quad x \in \bar{\Omega}} u(., t) \geq \underline{c}_1, \qquad \mathop {\lim \inf \max }\limits_{t \to + \infty\quad x \in \bar{\Omega}} v(., t) \geq \underline{c}_2. \end{equation} | (2.13) |
They are evidence that the system (1.2) is persistent. Regardless of the diffusion coefficients, this indicates that, from a biological point of view, a predator and its prey will always coexist within the habitable domain at any given time and in any given location.
Obviously, model (1.2) has a trivial equilibrium E_0 = (0, 0) and a predator-free equilibrium E_1 = (1, 0) , and the interior equilibrium must simultaneously meet the two non-trivial prey and predator nullcline conditions below:
\begin{align} & \Phi(u, v) = 1-u-\frac{v}{a+u^2} = 0, \end{align} | (3.1) |
\begin{align} & \Psi(u, v) = \frac{\beta u}{a+u^2}-d-\gamma u v = 0. \end{align} | (3.2) |
From (3.2), we obtain that v = -{\left(u^2 +a\right)}\, {\left(u-1\right)} , and substituting this into (3.1), we have
\begin{equation} \gamma \, u^6 +{\left(-\gamma \right)}\, u^5 +{\left(2\, a\, \gamma \right)}\, u^4 +{\left(-2\, a\, \gamma \right)}\, u^3 +{\left(a^2 \, \gamma -d\right)}\, u^2 +{\left(\beta -a^2 \, \gamma \right)}\, u-a\, d = 0. \end{equation} | (3.3) |
Obviously, Eq (3.3) has at least a positive root. Therefore, system (1.2) has at least an interior equilibrium E^* = (u^*, v^*) . To illustrate this, isoclines (3.1) and (3.2) are shown in Figure 1. Figure 1(a) shows that isoclines (3.1) and (3.2) intersect uniquely in the interior of the positive quadrant, i.e., system (1.2) has a unique interior equilibrium, and Figure 1(b) shows that isoclines (3.1) and (3.2) intersect two times in the interior of the positive quadrant, i.e., the system has two equilibria.
In order to study the stability of the equilibria. First, we define the real-valued Sobolev space
\begin{equation*} X = \left\{ (u, v)^T: u, v\in L^2(\Omega) \right\}, < u, v > = < u_1, v_1 > _{L^2}+ < u_2, v_2 > _{L^2}. \end{equation*} |
Then (X, < \cdot, \cdot > ) is a Hilbert space. In C([-\tau, 0], X) , system (1.2) can be thought of as a functional differential equation in abstract form.
Let U(t) = (u(\cdot, t), v(\cdot, t))^{T}. Thus, we linearize system (1.2) around a constant solution E = (u, v) ; we get
\begin{equation} \dot{U} = D\Delta U(t)+L(U_{t}), \end{equation} | (3.4) |
where D = {\rm diag}(d_1, d_2) ,
\begin{equation*} dom(d\Delta) = \left\{ (u, v)^T: u, v \in W^{2, 2}(\Omega), \frac{\partial u}{\partial n} = \frac{\partial v}{\partial n} = 0\right\}, \end{equation*} |
and U_t = {\bf col}(u(x, t), v(x, t))\in \mathcal{C}_\tau and L: C([-\tau, 0] \rightarrow X is given by
\begin{equation} L(\varphi) = \left(\begin{array}{ccc} a_{11} & -a_{12} \\ a_{21} & - a_{22}\\ \end{array}\right)\varphi(0) +\left(\begin{array}{ccc} 0 & 0 \\ -b & 0\\ \end{array}\right)\varphi(-1) \end{equation} | (3.5) |
with \varphi = (\varphi_1, \varphi_2)^T\in C([-\tau, 0] and
\begin{equation*} \begin{split} & a_{11} = \frac{2u^2v}{(a+u^2)^2}-u, a_{12} = \frac{u}{u^2+a}, \\ & a_{21} = \frac{ \beta v(a-u^2)}{(a+u^2)^2}, a_{22} = \gamma uv, b = \gamma v^2. \end{split} \end{equation*} |
Therefore, the characteristic equation of system (3.4) is
\begin{equation} \lambda y-D\Delta y-L(e^{\lambda\cdot}y) = 0, y\in dom(\Delta), y\neq 0. \end{equation} | (3.6) |
We know that with the corresponding eigenfunctions \psi_n(x) the problem
\begin{eqnarray*} \left\{\begin{array}{lcr} - \Delta \psi = \lambda \psi, x\in \Omega, \\ \frac{\partial \psi }{\partial n} = 0, x\in \partial \Omega, \\ \end{array}\right. \end{eqnarray*} |
has eigenvalues 0 = \mu_0 < \mu_1\leq \mu_2\leq \cdots\leq \mu_n\leq\mu_{n+1}\leq \cdots . Substituting
\begin{equation*} y = \sum\limits_{n = 0}^{\infty}\psi_n(x) \left(\begin{array}{lll} y_{1n}\\ y_{2n}\\ \end{array}\right) \end{equation*} |
into Eq (3.4), we have
\begin{equation*} \left(\begin{array}{lll} a_{11}-d_1\mu_n & -a_{12} \\ a_{21}-b e^{-\lambda\tau}& -a_{22}-d_2\mu_n\\ \end{array}\right)\left(\begin{array}{lll} y_{1n}\\ y_{2n}\\ \end{array}\right) = \lambda \left(\begin{array}{lll} y_{1n}\\ y_{2n}\\ \end{array}\right), \; \; n\in \{0, 1, 2, \cdots\}: = N_0. \end{equation*} |
Hence, Eq (3.6) equals
\begin{equation} \begin{split} W_n(\lambda, \beta, \tau) = \lambda^2 +A_n\lambda +B_n +Ce^{-\lambda \tau } = 0, \end{split} \end{equation} | (3.7) |
where
\begin{equation} \begin{split} & A_n = (d_1+d_2) \mu_n+a_{22}-a_{11}, \\ & B_n = d_1d_2\mu_n^2 +(d_1 a_{22}- d_2 a_{11})\mu_n+ (a_{12}a_{21}-a_{11}a_{22}), \\ & C = -a_{12}b. \end{split} \end{equation} | (3.8) |
For the extinct equilibrium E_0 = (0, 0) , J_{(0, 0)} = \left(\begin{array}{cc} 1 & 0 \\ 0 & -d \end{array}\right) . When n = 0 , the eigenvalue \lambda = 1 > 0 , therefore, E_0 is a saddle point, which is always unstable.
For the predator free equilibrium E_1 = (1, 0) , J_{(1, 0)} = \left(\begin{array}{cc} -1 & -\frac{1}{a+1} \\ 0 & \frac{\beta}{a+1}-d \end{array}\right) . Therefore, if \frac{\beta}{a+1}-d < 0 , then E_1 is stable.
In what follows, we first discuss the stability of the interior equilibrium of system (1.2) with \tau = 0 . When \tau = 0 , the characteristic equation (3.7) becomes
\begin{equation} \lambda^2+A_n\lambda+B_n+C = 0. \end{equation} | (3.9) |
Theorem 3.1. Assume that the conditions
\begin{align} &a_{22}-a_{11} > 0, \end{align} | (3.10) |
\begin{align} &a_{12}a_{21}-a_{11}a_{22}-a_{12}b > 0 \end{align} | (3.11) |
hold; then, the following results are true:
(i) If d_1a_{22}-d_2a_{11} > 0 , then the interior equilibrium is locally asymptotically stable.
(ii) If d_1a_{22}-d_2a_{11} < 0 and \Delta_1 < 0 , then the interior equilibrium is locally asymptotically stable, where
\begin{equation*} \Delta_1 = (d_1a_{22}+d_2a_{11})^2-4d_1d_2(a_{12}a_{21}-a_{12}b). \end{equation*} |
(iii) If d_1a_{22}-d_2a_{11} < 0 and \Delta_1 > 0 , but there is no n\in N_0 such that \mu_n\in (\mu^-, \mu^+) , then the interior equilibrium is asymptotically stable, where
\begin{equation*} \mu^\pm = \frac{d_2a_{11}-d_1a_{22}\pm\sqrt{\Delta_1}}{2d_1d_2}. \end{equation*} |
Proof. Obviously, if a_{22}-a_{11} > 0 , then A_n\leq A_0 > 0 for n\in N_0 . a_{12}a_{21}-a_{11}a_{22}-a_{12}b > 0 holds, so B_0+C = hb_3(b_2b_3-b_1) > 0 .
(ⅰ) If d_1a_{22}-d_2a_{11} > 0 , then B_n+C\geq B_0+C , which implies that all roots of the characteristic equation (3.9) have negative real parts. Therefore, the interior equilibrium is locally asymptotically stable.
(ⅱ) d_1a_{22}-d_2a_{11} < 0 and \Delta_1 < 0 hold, which implies that the equation
\begin{equation*} h(y) = d_1d_2y^2 +(d_1 a_{22}- d_2 a_{11})y+ (a_{12}a_{21}-a_{11}a_{22})-a_{12}b > 0 \end{equation*} |
for any y\geq0 . That is B_n+C > 0 for any \mu_n . Similar to the discussion in (i), we have that the interior equilibrium is locally asymptotically stable.
(ⅲ) There is no n\in N_0 such that \mu_n\in (\mu^-, \mu^+) . So, B_n+C > 0 for any \mu_n . Consequently, we have the results.
Theorem 3.2. Suppose that the conditions (3.10) and (3.11) hold, and assume further that
\begin{align} \begin{split} & \frac{d_2}{d_1} > \frac{2a_2b_1-a_1b_2+\sqrt{(2a_2b_1-a_1b_2)^2-a_1^2b_2^2}}{a_1^2} \end{split} \end{align} | (3.12) |
hold, then the Turing bifurcation occurs.
Proof. The condition (3.10) is satisfied, so E^* of the ODE model corresponding to model (1.2) is locally asymptotically stable.
The condition (3.10) holds, so A_n < 0 . Therefore, when E^* of model (1.2) is unstable i.e., Eq (3.9) has at least one positive real root i.e., B_n+C = 0 has two true roots, one positive and one negative. Note that
\begin{equation} B_n+C = d_1d_2\mu_n^2 +(d_1 a_{22}- d_2 a_{11})\mu_n+(a_{12}a_{21}-a_{11}a_{22})-a_{12}b.\\ \end{equation} | (3.13) |
It is easy to see that B_n+C reaches its minimal value at \mu_n = \mu_{\min} = \frac{d_2a_{11}-d_1a_{22}}{2d_1d_2} with d_1 a_{22}- d_2 a_{11} < 0 .
It implies that
\begin{equation} a_{11}^2\frac{d_2^2}{d_1^2}-2( 2a_{12}b-2a_{12}a_{21}-a_{11}a_{22})\frac{d_2}{d_1}+a_{22}^2 > 0. \end{equation} | (3.14) |
Hence, D_n is negative when Eq (3.14) is met, and it applies for \mu close to \mu_{\min} . By Eq (3.14), we obtain
\begin{equation} \frac{d_2}{d_1} > \frac{2a_{12}b-2a_{12}a_{21}-a_{11}a_{22}+\sqrt{(a_{12}b-a_{12}a_{21})^2-a_{11}a_{22}(a_{12}b-a_{12}a_{21})}}{a_{11}^2}, \end{equation} | (3.15) |
which completes the proof.
Let \lambda = \pm i\omega\ (\omega > 0) be a pure imaginary pair of eigenvalues of Eq (3.7). Thus, \omega satisfies
\begin{equation} -\omega^2-iA_n\omega +B_n+C(cos(\omega\tau)-i\sin(\omega \tau) = 0. \end{equation} | (3.16) |
Therefore, we have
\begin{equation} \left\{\begin{array}{lcr} -\omega ^2 +B_n = -C \cos(\omega \tau), \\ A_n\omega = C \sin(\omega \tau). \end{array}\right. \end{equation} | (3.17) |
From the above equation, we obtain
\begin{equation} \omega ^4+(A_n^2-2B_n)\omega^2+B_n^2-C^2 = 0. \end{equation} | (3.18) |
From Eq (3.8), we know that if the condition (3.11) holds, then B_n-C > B_n+C > 0 . Therefore, B_n^2-C^2 > 0 . We discuss the existence of roots of Eq (3.18) in two cases.
Case Ⅰ. Suppose that
\begin{equation} A_n^2-2B_n < 0, \Delta_n = (A_n^2-2B_n)^2-4(B_n^2-C^2) > 0. \end{equation} | (3.19) |
Thus, Eq (3.18) has two positive real roots \omega^\pm(n) = \sqrt{\frac{-(A_n^2-2B_n)\pm\sqrt{\Delta_n}}{2}} . Substituting \omega^\pm(n) into Eq (3.17), we have
\begin{equation} \tau_{j}^\pm(n) = \frac{1}{\omega^\pm(n)}\left[ \arccos \left( \frac{\omega^{\pm^2}(n)-B_{n}}{C}\right) +2j\pi \right], j\in\mathbb{N}.\\ \end{equation} | (3.20) |
Case Ⅱ. Suppose that either of the following two conditions are met
\begin{equation} (i)\quad \Delta_n < 0; (ii)\quad \Delta_n\geq0, A_n^2-2B_n\geq 0. \end{equation} | (3.21) |
So, Eq (3.18) does not have a positive real root.
We summarize the above discussions, and we have the following theorem.
Theorem 3.3. Assume that the conditions (3.10) and (3.11) hold and the following is true:
(i) When
A_n^2-2B_n < 0, \Delta_n = (A_n^2-2B_n)^2-4(B_n^2-C^2) > 0 , |
Eq (3.18) has two positive real roots \omega^\pm(n) ; then, Eq (3.7) has a pair of pure imaginary roots \pm i \omega^\pm(n) , when \tau = \tau_{j}^\pm(n) .
(ii) If either \Delta_n < 0 or \Delta_n\geq0, P_{1n}\geq 0 ; then, Eq (3.7) has no pure imaginary roots.
Lemma 3.1. Let \lambda(\tau) = \xi(\tau)\pm i\eta(\tau) be the root of Eq (3.7) satisfying \alpha(\tau_{j}^\pm(n)) = 0, \xi(\tau_{j}^\pm(n)) = \omega^\pm(n)) . Then the following transversality conditions are satisfied:
\begin{equation} \begin{split} & \left. \frac{d( {\rm{Re}}\lambda(\tau ))}{{d\tau }}\right|_{\tau = \tau_{j}^\pm(n) } = \pm \frac{\sqrt{\Delta_n}}{A_{3n}^2}.\\ \end{split} \end{equation} | (3.22) |
Proof. When both sides of Eq (3.7) are differentiated with regard to \lambda , the result is
2\lambda+A_{n}-C(\tau+\lambda\frac{d\tau_1}{d\lambda})e^{-\lambda\tau} = 0. |
Therefore,
\begin{equation*} \begin{split} &\left[\frac{d\lambda}{d\tau} \right]^{-1} = \frac{(2\lambda+C)e^{\lambda \tau}}{\lambda C} -\frac{\tau}{\lambda}. \end{split} \end{equation*} |
Combining with (3.17), we get
\begin{equation*} \begin{split} & {\rm{Re}}\left(\left[\frac{{\rm{d}}\lambda}{{\rm{d}}\tau} \right]^{-1}\right)_{\lambda = \pm {\rm{i}} \omega^\pm({\rm{n}})} = \left[ \frac{2\omega^2+{\rm{A}}_{{\rm{n}}}^2-2{\rm{B}}_{{\rm{n}}}}{{\rm{C}}^2} \right]_{\lambda = \pm {\rm{i}} \omega^\pm({\rm{n}})} = \pm \frac{\sqrt{\Delta_n}}{{\rm{C}}^2}. \end{split} \end{equation*} |
Denote
\Gamma = \left\{n\in\mathbb{N}_0| A_n^2-2B_n < 0, \Delta_n = (A_n^2-2B_n)^2-4(B_n^2-C^2) > 0 \right\}. |
For a given n\in \Gamma , it is seen that \tau_{j}^+(n) grows with respect to j . As a result, we can deduce that \tau_{0}^+(n) = \mathop {\min }\limits_{j\in\mathbb{N}_0}\tau_{j}^+(n) for some fixed n . We define
\begin{equation} \tau^* = \mathop{\min}\limits_{n\in\Gamma}\{\tau_{0}^+(n)\}. \end{equation} | (3.23) |
Theorem 3.4. Assume that the conditions (3.10) and (3.11) hold; we have the following results
(i) If either of the following two conditions are met
\begin{equation} (a)\quad \Delta_n < 0;\; \; (b)\quad \Delta_n\geq0, A_n^2-2B_n\geq 0, \end{equation} | (3.24) |
then the interior equilibrium E^* is locally asymptotically stable for \tau\geq 0 ;
(ii) If A_n^2-2B_n > 0 and \Delta_n > 0 , then the interior equilibrium E^* is locally asymptotically stable when \tau \in[0, \tau^*) and system (1.2) experiences a Hopf bifurcation at E^* . In addition, before it becomes unstable, the interior equilibrium E* will go through k changes from stable to unstable and back to stable.
Theorem 4.1. Suppose that \frac{\beta K_1}{a} < d and E_1 = (1, 0) is globally asymptotically stable.
Proof. Define a Lyapunov functional as follows
\begin{equation} V(t) = \beta\int_{\Omega} \int_{1}^{u} \frac{u-1}{u}d\xi dx+\int_\Omega vdx. \end{equation} | (4.1) |
Differentiating V(t) with respect to t , we have
\begin{align*} \frac{{\rm d}V}{{\rm d}t} = &\beta \int_{\Omega} \frac{u-1}{u}\left(d_1\Delta u+u(1-u)-\frac{uv}{a+u^2}\right) +\int_{\Omega}\left(d_2\Delta v+\frac{\beta uv}{a+u^2}-dv-\gamma u(t-\tau) v^2\right)dx\\ = & \beta d_1\int_{\Omega} \frac{u-1}{u}\Delta udx +\beta \int_{\Omega} (u-1)\left((1-u)-\frac{v}{a+u^2}\right)+d_2\int_\Omega \Delta vdx\\ & +\int_\Omega v\left(\frac{\beta u}{a+u^2}-d-\gamma u(t-\tau)v\right)dx\\ = & -\beta \int_{\Omega} \left( u-1\right)^2dx+\int_\Omega \left (\frac{\beta u}{a+u^2}-d-\gamma u(t-\tau)v\right)dx-\beta\int_{\Omega}\left( \frac{\nabla u}{u}\right)^2dx \\ \leq & -\beta \int_{\Omega} \left( u-1\right)^2dx+\int_\Omega \left (\frac{\beta K_1}{a}-d \right)vdx-\beta\int_{\Omega}\left( \frac{\nabla u}{u}\right)^2dx. \end{align*} |
\frac{\beta}{a} < d , so \frac{{\rm d}V}{{\rm d}t}\leq 0 , and \frac{{\rm d}V}{{\rm d}t} = 0 if and only if (u, v) = (1, 0) . We conclude that E_1 = (1, 0) is globally asymptotically stable.
Theorem 4.2. Assume that the conditions
\begin{equation} \begin{split} & \frac{a\beta}{a+u^{*2}}-\frac{a\beta v^*(u^*+K_1)}{a(a+u^{*2})^2}-\frac{\beta u^*K_1}{2a(a+u^{*2})} > 0, \\ & \gamma \underline{c}_1-\frac{\gamma v^*}{2}-\frac{\beta u^* k_1}{2a(a+u^{*2})} > 0, \frac{\gamma v^*}{2}-1 > 0\\ \end{split} \end{equation} | (4.2) |
hold. Then the unique interior equilibrium E^* = (u^*, v^*) is globally asymptotically stable.
Proof. Define a Lyapunov function as follows
\begin{equation*} \begin{split} & W(t) = \frac{a\beta}{a+u^{*^2}}I_1(t)+ I_2(t)+I_3(t), \end{split} \end{equation*} |
where
\begin{equation*} I_1(t) = \int_{\Omega}\int_{u^*}^{u} \frac{\xi-u^*}{\xi}{\rm d}\xi {\rm d}x, \; I_2(t) = \int_{\Omega}\int_{v^*}^{v} \frac{\xi-v^*}{\xi} {\rm d}\xi {\rm d}x, \; I_3(t) = \int_{\Omega}\int_{t-\tau}^{t} (u(\xi)-u^*) ^2{\rm d}\xi {\rm d}x. \end{equation*} |
Thus, we have
\begin{align*} \frac{{\rm d}I_1(t)}{dt}& = d_1\int_{\Omega}\frac{u-u^*}{u}\Delta udx + \int_{\Omega}(u-u^*)\left( 1-u-\frac{v}{a+u^2}d\right)dx \\ & = -d_1u^*\int_{\Omega} \left(\frac{\nabla u}{u}\right)^2dx+\int_{\Omega}(u-u^*)\left( -(u-u^*)-\left( \frac{v}{a+u^2}-\frac{v^*}{a+u^{*2}}\right) \right)dx\\ & = -d_1u^*\int_{\Omega} \left(\frac{\nabla u}{u}\right)^2dx- \int_{\Omega}(u-u^*)^2dx-\int_{\Omega}(u-u^*) \left(\frac{a(v-v^*)+u^{*2}v-u^2v^*}{(a+u^2)(a+u^{*2})}\right)dx\\ & = -d_1u^*\int_{\Omega} \left(\frac{\nabla u}{u}\right)^2dx- \int_{\Omega}\left(1+\frac{v^*(u^*+u)}{(a+u^2)(a+u^{*2})}\right)(u-u^*)^2dx-\int_{\Omega} \frac{1}{a+u^2}(u-u^*)(v-v^*){\rm d}x, \\&\\ \frac{{\rm d}I_2(t)}{dt}& = d_2\int_{\Omega}\frac{v-v^*}{v}\Delta udx + \int_{\Omega}(v-v^*)\left( \frac{\beta u}{a+u^2}-d-\gamma u(t-\tau)v\right)dx \\ & = -d_2v^*\int_{\Omega} \left(\frac{\nabla v}{v}\right)^2dx+\int_{\Omega}(v-v^*) \frac{\beta u(a+u^{*2})-\beta u^*(a+u^2)}{(a+u^2)(a+u^{*2})}dx-\gamma\int_{\Omega}(v-v^*)(u(t-\tau)v-u^*v^*){\rm d}x \\ & = -d_2u^*\int_{\Omega} \left(\frac{\nabla v}{v}\right)^2dx+ \int_{\Omega} (a\beta -\beta u^*u) \frac{(u-u^*)(v-v^*)}{(a+u^2)(a+u^{*2})} dx -\gamma \int_\Omega u(t-\tau)(v-v^*)^2\\ &\quad-\gamma v^*\int_\Omega (v-v^*)(u(t-\tau)-u^*){\rm d}x \\ &\leq -d_2u^*\int_{\Omega} \left(\frac{\nabla v}{v}\right)^2dx+ \int_{\Omega} (a\beta -\beta u^*u) \frac{(u-u^*)(v-v^*)}{(a+u^2)(a+u^{*2})} dx -\gamma \int_\Omega u(t-\tau)(v-v^*)^2\\ &\quad+\frac{\gamma v^*}{2}\int_\Omega (v-v^*)^2{\rm d}x +\int_\Omega(u(t-\tau)-u^*)^2{\rm d}x, \\ & \\ \frac{{\rm d}I_3(t)}{dt} & = \int_{\Omega} (u-u^*)^2{\rm d}x -\int_{\Omega} (u(t-\tau)-u^*) ^2{\rm d}x. \end{align*} |
Note that
\begin{equation*} \begin{split} & \frac{{\rm d}W}{{\rm d}t} = \frac{{\rm d}I_1(t)}{dt} +\frac{{\rm d}I_2(t)}{dt}+\frac{{\rm d}I_3(t)}{dt}. \end{split} \end{equation*} |
Therefore, we obtain
\begin{equation*} \begin{split} \frac{{\rm d}W(t)}{{\rm d}t}&\leq -\frac{a\beta u^*d_1}{a+u^{*2}}\int_{\Omega} \left(\frac{\nabla u}{u}\right)^2dx-d_2v^*\int_{\Omega} \left(\frac{\nabla v}{v}\right)^2dx\\ &-\int_\Omega \left(\frac{a\beta}{a+u^{*2}}-\frac{a\beta v^*(u^*+u)}{(a+u^2)(a+u^{*2})^2}-\frac{\beta u^*u}{2(a+u^2)(a+u^{*2})}\right)(u-u^*)^2dx\\ &-\int_{\Omega} \left(\gamma u(t-\tau)-\frac{\gamma v^*}{2}-\frac{\beta u^* u}{2(a+u^2)(a+u^{*2})}\right)(v-v^*)^2dx-(\frac{\gamma v^*}{2}-1)\int_\Omega (u(t-\tau)-u^*)^2dx.\\ \end{split} \end{equation*} |
From Eq (4.2), we know that \frac{{\rm d}W}{{\rm d}t}\leq 0 , and \frac{{\rm d}W}{{\rm d}t} = 0 if and only if (u, v) = (u^*, v^*) . Therefore, the interior equilibrium E^* is globally asymptotically stable.
In this section, we investigate the existence and non-existence of non-constant positive solutions of system (1.3). First, we will give a priori upper and lower bounds for the positive solutions of system (1.3).
Lemma 5.1 (Harnack inequality). [22] Assume that c(x)\in C(\bar{\Omega}) and \phi \in C^2(\Omega)\cap C^1(\bar{\Omega}) is a positive solution to
\begin{equation*} \Delta \phi+c(x)\phi, x\in \Omega, \frac{\partial \phi}{\partial n} = 0, x\in \partial \Omega. \end{equation*} |
Then there exists a positive constant C_* = C_*(\Vert c\Vert_\infty) such that
\begin{equation*} \max\limits_{\bar{\Omega}}\phi \leq C_* \min\limits_{\bar{\Omega}} \phi. \end{equation*} |
Theorem 5.1 (Upper bound). Any positive solution (u, v) of system (1.3) satisfies
\begin{equation*} 0 < \mathop { \max }\limits_{\overline{\Omega}}u(x)\leq 1, 0 < \mathop { \max }\limits_{\overline{\Omega}}v(x)\leq \frac{\beta(d_2+dd_1)}{dd_2}. \end{equation*} |
Proof. By the strong maximum principle, we know that if there is a x_0\in \bar{\Omega} such that v(x_0) = 0 , then v(x)\equiv 0 and u satisfies
\begin{eqnarray} \left\{\begin{array}{lcr} -d_1\Delta u = u(1-u), &x\in \Omega, \\ \dfrac{\partial u}{\partial n} = \dfrac{\partial v}{\partial n} = 0, &x\in \partial \Omega. \end{array}\right. \end{eqnarray} | (5.1) |
From the well known result, u\equiv 0 or u\equiv 1 . Hence, if (u, v) is not (0, 0) or (1, 0) , then u(x) > 0 and v(x) > 0 for x\in \bar{\Omega} .
From the maximum principle, we easily obtained that u(x)\leq 1 in \Omega . Multiplying the first equation of Eq (1.3) by \beta and adding it to the second equation of Eq (1.3), we have
\begin{equation} \begin{split} -(\beta d_1\Delta u+d_2\Delta v)& = \beta u(1-u)+\frac{d\beta d_1u}{d_2}-\frac{d}{d_2}(\beta d_1u+d_2v)-\gamma uv^2 \\ & \leq \beta+\frac{d\beta d_1}{d_2}-\frac{d}{d_2}(\beta d_1u+d_2v). \end{split} \end{equation} | (5.2) |
Then from the maximum principle,
\begin{equation*} \beta d_1u+d_2v\leq \frac{d_2\beta+d\beta d_1}{d}, \end{equation*} |
which leads to
\begin{equation*} v\leq \frac{\beta(d_2+dd_1)}{dd_2}. \end{equation*} |
Theorem 5.2 (Lower bound). There exists a positive constant \underline{C} depending on d_1 , d_2 , a , \beta , d and \gamma , such that any positive solution (u(x), v(x)) of Eq (1.3) satisfies
\begin{equation} u(x), v(x)\geq \underline{C}, x\in \bar{\Omega}. \end{equation} | (5.3) |
Proof.
\begin{equation} \begin{split} & u(x), v(x)\leq \overline{C}: = \max\left\{1, \frac{\beta(d_2+dd_1)}{dd_2}\right\}. \\ \end{split} \end{equation} | (5.4) |
Let
\begin{equation} \begin{split} & c_1(x): = 1-u(x)-\frac{v(x)}{a+u^2(x)}\, \, \text{and}\, \, c_2(x): = \frac{\beta u(x)}{a+u^2(x)} -d-\gamma u(x) v(x). \end{split} \end{equation} | (5.5) |
Then
\begin{equation*} \vert c_1(x) \vert 2+ \frac{\overline{C}}{a}, \vert c_2(x)\vert d+ \frac{\overline{C}}{a} +\gamma \overline{C}^2. \end{equation*} |
A positive constant C can be derived from Lemma 5.1 in such a way that
\begin{equation*} \begin{split} & \sup\limits_{\bar{\Omega}} u(x) \leq C \inf\limits_{\bar{\Omega}} u(x), \sup\limits_{\bar{\Omega}} v(x) \leq C \inf\limits_{\bar{\Omega}} v(x). \end{split} \end{equation*} |
Therefore, it must now show that there is some c > 0 such that
\begin{equation} \mathop {\sup}\limits_{ \bar{\Omega}}u(x)\geq c, \quad \mathop {\sup}\limits_{ \bar{\Omega}}v(x)\geq c. \end{equation} | (5.6) |
On the other hand, suppose that the outcome is incorrect. Then there exists a series of affirmative solutions (u_n(x), v_n(x)) such that
\begin{equation} \mathop {\sup}\limits_{ \overline{\Omega}}u_n\rightarrow0\;\text{or}\;\mathop {\sup}\limits_{ \overline{\Omega}}v_n\rightarrow0 \quad as \; n\rightarrow +\infty. \end{equation} | (5.7) |
By standard elliptic regularity, we obtain that there exists a subsequence of \{(u_n, v_n)\} , which is again denoted by \{(u_n, v_n)\} , such that \{(u_n, v_n)\}\rightarrow (u_\infty, v_\infty) in C^2(\bar{\Omega}) as n\rightarrow +\infty . Noting that u_\infty\leq 1 , since Eq (5.7) holds, u_\infty\equiv0 or v_\infty\equiv0 . So, we have the following:
(ⅰ) u_\infty\equiv0 , v_\infty\not\equiv0 ; or u_\infty\equiv0 , v_\infty\equiv0 ;
(ⅱ) u_\infty\not\equiv0 , v_\infty\equiv0 .
Moreover, we get the following two integral equations:
\begin{eqnarray} \left\{\begin{array}{lcr} \int_\Omega u_n(1-u_n)-\dfrac{ u_nv_n}{a+u_n^2})dx = 0, \\ \int_\Omega v_n(-d+\dfrac{\beta u_n}{a+u_n^2}-\gamma uv^{2})dx = 0. \end{array}\right. \end{eqnarray} | (5.8) |
(ⅰ) Case: u_\infty\equiv0 , so we have
\begin{equation*} -d+\frac{\beta u_n}{a+u_n^2}-\gamma uv^{2}\rightarrow -d < 0, n\rightarrow \infty, \end{equation*} |
and v_n > 0 . But, we integrate the equation of v_n ; then, we have
\begin{equation} \int_\Omega v_n(-\gamma+\frac{\beta u_n}{b+u_n})dx = 0. \end{equation} | (5.9) |
This is a contradiction.
(ⅱ) u_\infty\not\equiv0 and v_\infty\equiv0 , so u_\infty satisfies Eq (5.1). So, u_\infty\equiv 1 , and for a large n , we have
\begin{equation*} -d+\frac{\beta u_n}{a+u_n^2}-\gamma uv^{2} \rightarrow -d+\frac{\beta }{a+1} > \varepsilon > 0. \end{equation*} |
Thus, the second equation of Eq (5.9) does not hold, which is a contradiction. So, \sup\limits_{\bar{\Omega}} u(x) > 0 , \sup\limits_{\bar{\Omega}} v(x) > 0 , and consequently Eq (5.3) holds.
Theorem 5.3. There is a large constant d^* for which there is no non-constant positive solution to the problem (1.3) if d_1, d_2\geq d^* .
Proof. Suppose that (u(x), v(x)) is a non-constant positive solution of system (1.3). Denote
\overline{u} = |\Omega|^{-1}\int_\Omega u(x)dx\geq0, \; \; \overline{v} = |\Omega|^{-1}\int_\Omega v(x)dx\geq0. |
Then
\begin{equation} \int_\Omega (u-\overline{u})dx = 0, \int_\Omega (v-\overline{v})dx = 0. \end{equation} | (5.10) |
Multiplying the first equation in Eq (1.3) by u-\bar{u} and applying system (1.3), we get
\begin{equation} \begin{split} \int_\Omega d_1|\nabla (u-\bar{u})|^2dx& = \int_\Omega (u-\overline{u})u\left(1-u-\frac{uv}{a+u^2} \right)dx\\ & = -\int_\Omega (u-\bar{u})^2dx -\int_\Omega (u-\bar{u})\frac{uv}{a+u^2}{\rm d}x\\ & = -\int_\Omega (u-\bar{u})^2dx -\int_\Omega \frac{v(u-\bar{u})^2}{a+u^2}dx-\int_\Omega \frac{v\bar{u}(u-\overline{u})}{a+u^2}dx\\ & = -\int_\Omega (u-\bar{u})^2dx -\int_\Omega \frac{v(u-\bar{u})^2}{a+u^2}dx-\int_\Omega (u-\overline{u}) \left(\frac{v\overline{u}}{a+u^2}-\frac{\bar{v}\bar{u}}{a+\bar{u}^2}\right) dx\\ & = -\int_\Omega (u-\bar{u})^2dx -\int_\Omega \frac{v(u-\bar{u})^2}{a+u^2}dx-\int_\Omega \frac{\bar{u}}{a+u^2} (u-\bar{u})(v-\bar{v})dx\\ &\quad+\int_\Omega \frac{\bar{u}\bar{v}(\bar{u}+u)}{(a+u^2)(a+\bar{u}^2)} (u-\bar{u})^2dx\\ &\leq \left( \frac{\bar{C}}{2a}+\frac{2\bar{C}^3}{a^2} -\frac{\underline{C}}{a+\bar{C}^2}-1 \right) \int_{\Omega} (u-\bar{u})^2dx +\frac{\bar{C}}{2a}\int_{\Omega} (v-\bar{v})^2dx. \end{split} \end{equation} | (5.11) |
Multiplying the second equation in system (1.3) by v-\bar{v} and applying Theorem 5.1, we get
\begin{split} d_2\int_\Omega |\nabla(v-\overline{v})|^2dx& = \delta \int_\Omega(v-\overline{v})v\left(\frac{\beta u}{a+u^2}-d-\gamma u v\right)dx\\ & = -d\int_\Omega (v-\overline{v})^2dx + \int_\Omega (v-\overline{v})\frac{\beta u v}{a+u^2}dx-\int_\Omega \gamma u v^2(v-\bar{v})dx\\ & = -d\int_\Omega (v-\overline{v})^2dx+\beta \int_\Omega \frac{a\bar{v}(u-\bar{u})(v-\bar{v})}{(a+u^2)(a+\bar{u}^2)}dx+\beta\int_\Omega \bar{v}(v-\bar{v}) \frac{u\bar{u}^2-\bar{u}u^2}{(a+u^2)(a+\bar{u}^2)}dx\\ &\quad-\gamma \int_\Omega v(v-\bar{v})(u(v-\bar{v})+u\bar{v})dx\\ &\leq \frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2} \int_\Omega (u-\bar{u})^2dx+\left(\frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2}-d\right)\int_\Omega (v-\bar{v})^2dx-\gamma \int_\Omega uv(v-\bar{v})^2dx\\ &\quad -\gamma \int \bar{v}v(u-\bar{u})(v-\bar{v})dx-\gamma \int_\Omega \bar{v}\bar{u}(v-\bar{v})^2dx\\ &\leq \left( \frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2} +\frac{\gamma \bar{C}^2}{2}\right) \int_\Omega (u-\bar{u})^2dx+\left(\frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2}-d+\frac{\beta \bar{C}}{a}-\frac{3\gamma \underline{C}^2}{2}\right)\int_\Omega (v-\bar{v})^2dx. \end{split} | (5.12) |
From Eqs (5.11) and (5.12), we get
\begin{equation} \begin{split} &\int_\Omega d_1|\nabla (u-\bar{u})|^2dx+d_2\int_\Omega |\nabla(v-\overline{v})|^2dx\\ &\leq \left( \frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2} +\frac{\gamma \bar{C}^2}{2} +\frac{\bar{C}}{2a}+\frac{2\bar{C}^3}{a^2} -\frac{\underline{C}}{a+\bar{C}^2}-1 \right)\int_\Omega(u-\bar{u})^2dx\\ &\quad+\left(\frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2}-d+\frac{\beta \bar{C}}{a}-\frac{3\gamma \underline{C}^2}{2}+\frac{\bar{C}}{2a}\right)\int_\Omega (v-\bar{v})^2dx. \end{split} \end{equation} | (5.13) |
By the Poincar\acute{e} inequality, we have
\begin{equation} \begin{split} &\int_\Omega d_1|\nabla (N-\bar{N})|^2dx+d_2\int_\Omega |\nabla(u-\overline{u})|^2dx\\ &\leq \frac{1}{\mu_1}\Big(A\int_\Omega(u-\bar{u})^2dx+B\int_\Omega(v-\bar{v})^2dx, \end{split} \end{equation} | (5.14) |
where
\begin{align*} A& = \frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2} +\frac{\gamma \bar{C}^2}{2} +\frac{\bar{C}}{2a}+\frac{2\bar{C}^3}{a^2} -\frac{\underline{C}}{a+\bar{C}^2}-1, \\ B& = \frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2}-d+\frac{\beta \bar{C}}{a}-\frac{3\gamma \underline{C}^2}{2}+\frac{\bar{C}}{2a}. \end{align*} |
These equations mean that if
\begin{equation*} \begin{split} & \min\{ d_1, d_2 \} > \frac{1}{\mu_1}\max\{ A, B\} \end{split} \end{equation*} |
then
\begin{equation*} \nabla (u-\bar{u}) = \nabla (v-\bar{v}) = 0, \end{equation*} |
and (u, v) must be a constant solution.
For simplicity, denote
\begin{equation*} \begin{split} & b_1 = a_{21}-b, { {\bf{F}}_{\bf{u}}({\bf{u}}^*)} = \left(\begin{array}{lcr} a_{11} & -a_{12} \\ b_{1} & -a_{22}\\ \end{array}\right). \end{split} \end{equation*} |
By the maximum principle and the standard elliptic regularity, the embedding theorems, and the assumption that \partial \Omega \in C^{2+\alpha} , we obtain that (u, v)\in C^2\times C^2 for the elliptic system (1.3). Therefore, there is a positive constant M_1 , such that \|\nabla u\|_{C^1}\leq M_1 and \|\nabla v\|_{C^1}\leq M_1 . So, there exists a sufficiently large positive constant M such that -d_1\Delta u-u(1-u-\frac{v}{a+u^2})+Mu and -d_2\Delta v-\frac{\beta uv}{a+u^2}+dv+\gamma u v^2+Mv are monotonically increasing functions with respect to u and v .
Define \mathcal{A}:[C^1(\overline{\Omega})]^2\rightarrow [C^1(\overline{\Omega})]^2 by
\begin{equation*} \mathcal{A}({\bf u})\triangleq \left(\begin{array}{lcr} (M-d_1\Delta)^{-1}[f_1(u, v)+Mu]\\ (M-d_2\Delta)^{-1}[f_2(u, v)+Mv] \end{array}\right), \end{equation*} |
where f_1(u, v) = u(1-u-\frac{v}{a+u^2}) and f_2(u, v) = \frac{\beta uv}{a+u^2}-dv-\gamma u v^2 .
It is worth noting that solving system (1.3) equates to finding positive solutions to the equation {(M{\bf I}-{\bf \mathcal{A})u} = 0} . We investigate the eigenvalue of the following problem using index theory.
\begin{equation} -{(M{\bf I}-{\bf \mathcal{A}_u(u^*))}}\Psi = \lambda \Psi, \Psi\neq 0, \end{equation} | (5.15) |
where \Psi = (\psi_1, \psi_2)^T and {\bf u^*} = (u^*, v^*) .
The following lemma is used to calculate the index of ( M{\bf I}-\mathcal{A}, {\bf u^*} ).
Lemma 5.2. [23] Assume that (M{\bf I}-\mathcal{A}_u(u^*))\neq 0 . Then {\rm index}(M{\bf I}-\mathcal{A}, {\bf u^*}) = (-1)^\sigma, \sigma = \sum_{\lambda > 0}m_{\lambda}, where m_\lambda is the multiplicity of \lambda .
By direct calculation, Eq (5.15) can be written as
\begin{equation} \left\{\begin{array}{lcr} -(\lambda+M)d_1\Delta\psi_1+(\lambda-a_{11})\psi_1+a_{12}\psi_2 = 0, &x\in \Omega, \\ -(\lambda+M)d_2\Delta\psi_2-b_1\psi_1+(\lambda+a_{22})\psi_2 = 0, & x\in \Omega, \\ \frac{\partial \psi_1}{\partial n} = \frac{\partial \psi_2}{\partial n} = 0, & x \in \partial \Omega. \end{array}\right. \end{equation} | (5.16) |
Notice that Eq (5.16) has a non-trivial solution if and only if Q_n(\lambda; d_1, d_2) = 0 for some \lambda\geq0 and n\geq0 , where
\begin{equation*} Q_n(\lambda;d_1, d_2)\triangleq {\rm{det}} \left(\begin{array}{lcr} \lambda+\frac{M d_1\mu_n-a_{11}}{d_1\mu_n+1} &\frac{a_{12}}{d_1\mu_n+1} \\ -\frac{b_{1}}{d_2\mu_n+1} & \lambda+\frac{Md_2\mu_n+a_{22}}{d_2\mu_n+1} \\ \end{array}\right). \end{equation*} |
Then, \lambda is an eigenvalue of Eqs (5.15) and (5.16) if and only if \lambda is a positive root of the characteristic equation Q_n(\lambda; d_1, d_2) = 0 for n\geq0 .
Lemma 5.3. (i) When n = 0 , Q_0(\lambda, d_1, d_2) = 0 may have no positive root, or exactly one positive root with a multiplicity of two, or two positive roots with a multiplicity of one.
(ii) If d_1 > \frac{a_{11} }{M\mu_1}: = \hat{d}_1 , then Q_n(\lambda, d_1, d_2) = 0 has no positive root for n\geq1 .
Proof. (ⅰ) It is easily obtained that Q_0(\lambda, d_1, d_2) = \lambda^2- {\rm trace}({ {\bf{F}}_{\bf{u}}({\bf{u}}^*)})\lambda + {\rm{det}}({ {\bf{F}}_{\bf{u}}({\bf{u}}^*)}). Obviously, the result holds.
(ⅱ) It is clear that for n\geq1 ,
\begin{align*} Q_n(\lambda, d_1, d_2) = & \lambda^2+\left(\frac{Md_2\mu_n+a_{12} }{d_2\mu_n+1}+\frac{Md_1\mu_n-a_{11} }{d_1\mu_n+1}\right)\lambda \\ &+ \frac{Md_2\mu_n+a_{12} }{d_2\mu_n+1}\frac{Md_1\mu_n-a_{11} }{d_1\mu_n+1} +\frac{b_1a_{12} }{(d_2\mu_n+1)(d_1\mu_n+1)}. \end{align*} |
Since a_{12} > 0 , b_{1} < 0 and a_{22} > 0 , then the polynomial Q_n(\lambda, d_1, d_2) > 0 . So, if d_2 is big enough, the desired result is reached.
Lemma 5.4. Suppose that
\begin{equation} \begin{split} & \frac{\beta v^*(a-u^{*2})}{(a+u^{*2})^2}-\gamma v^{*2}-\frac{2\gamma u^{*2} v^{*2}}{a+u^{*2}} < 0 \\ \end{split} \end{equation} | (5.17) |
holds; then, the following is true:
(i) The quadratic equation M^2d_1d_2\mu_n^2+(a_{22}Md_1-a_{11}Md_2)\mu_n+b_{1}a_{12}-a_{11}a_{22} = 0 has two roots. One is positive, say \mu_1^* , and the other is negative.
(ii) For some n_1^*\geq1 , suppose that \mu_1^*\in(\mu_{n_1^*}, \mu_{n_1^*+1}) . Then, there is a \hat{d}_1: = \hat{d}_1(\Gamma, d_1, d_2) such that the characteristic equation Q_n(\lambda, d_1, d_2) = 0 has a unique positive root for 1\leq n\leq n_1^* and has no positive root for n_1^*+1\leq n provided that d_1\geq \hat{d}_1 .
Proof. (ⅰ) The condition (5.17) holds; then, b_{1}a_{12}-a_{11}a_{22} < 0 , which implies that the result is true.
(ⅱ) Obviously, according to the definitions of n_1^* , Q_n(\lambda, d_1, d_2) = 0 has a unique root with a multiplicity of one for 1\leq b\leq n_1^* , and it has no positive root for n\geq n_1^*+1 if d_1\geq \hat{d}_1 .
Theorem 5.4. Assume that \mu_1^*\in(\mu_{n_1^*}, \mu_{n_1^*+1}) for some n_1^*\geq1 , and that \sum_{k = 1}^{n^*}n_k is even. Thus, there is a \hat{d}_1 such that for any d_1 > \hat{d}_1 , system (1.3) has at least one non-constant positive solution.
Proof. Assume, on the other hand, that the assertion is false for some d_1 > \hat{d}_1 .
\begin{equation*} \mathcal{A}_t({\bf u})\triangleq \left(\begin{array}{lcr} (MI-(d_1^*+t(d_1-d_1^*))\Delta)^{-1}[f_1(u, v)+Mu]\\ (MI-(d_2^*+t(d_2-d_2^*))\Delta)^{-1}[f_2(u, v)+Mv] \end{array}\right), \end{equation*} |
where d_1^* and d_2^* are constants that are positive and will be found out later.
Consider the problem
\begin{equation} \mathcal{A}_t({\bf u}) = {\bf u} \quad in \quad \Omega, \quad \frac{\partial{\bf u}}{\partial n} = 0 \quad on \quad \partial \Omega. \end{equation} | (5.18) |
Its positive solution is contained in
\Lambda: = \Big\{{\bf u}\in[C^1(\bar{\Omega})]^2:\underline{C}\leq u, v < \overline{C}\Big\}. |
By the homotopy invariance of the Leray-Schauder degree,
\begin{equation} deg(I-\mathcal{A}_0, \Lambda, 0) = deg(I-\mathcal{A}_1, \Lambda, 0). \end{equation} | (5.19) |
Notice that {\bf u} is a non-constant positive solution of Eq (1.3) if and only if it is such a solution of (5.18) for t = 1 . And for any t\in[0, 1] , {\bf u}^* is a constant solution of Eq (5.18).
Since we assumed that there are no non-constant positive solutions of system (1.3), \mathcal{A}_t({\bf u}) = {\bf u} has only the constant solution u^* in \Lambda . By Lemmas 5.3 and 5.4, we can obtain
\begin{equation*} l_{\lambda_k} = \left\{\begin{array}{lcr} 0 \; {\rm{or}} \; 2, {\rm{if}}\; n = 0, \\ 1, \qquad {\rm if}\; 1\leq n\leq n_1^*, \\ 0, \qquad {\rm if}\; n\geq n_1^*+1 . \end{array}\right. \end{equation*} |
Thus, \sigma = \sum_{k = 1}^{k_1^*}n_k+ 1\:(\:{\rm{or}}\: 3\:)\: = \:{\rm{an}}\: {\rm{odd}}\: {\rm{number}}. So that
\begin{array}{c} deg(I-\mathcal{A}_1, \Lambda, 0) = {\rm {index}}(\mathcal{A}_1, {\bf u^*}) = -1. \\ A = \frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2} +\frac{\gamma \bar{C}^2}{2} +\frac{\bar{C}}{2a}+\frac{2\bar{C}^3}{a^2} -\frac{\underline{C}}{a+\bar{C}^2}-1, \\ B = \frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2}-d+\frac{\beta \bar{C}}{a}-\frac{3\gamma \underline{C}^2}{2}+\frac{\bar{C}}{2a}. \end{array} | (5.20) |
Let us take
\begin{equation*} \begin{split} & d_1^* = \frac{1}{\mu_1}\left(\frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2} +\frac{\gamma \bar{C}^2}{2} +\frac{\bar{C}}{2a}+\frac{2\bar{C}^3}{a^2} -\frac{\underline{C}}{a+\bar{C}^2}-1\right), \\ & d_2^* = \max \left\{\frac{1}{\mu_1}\left(\frac{\beta \bar{C}(a+\bar{C}^2)}{2a^2}-d+\frac{\beta \bar{C}}{a}-\frac{3\gamma \underline{C}^2}{2}+\frac{\bar{C}}{2a}\right) , \hat{d}_2\right\}+1, \\ \end{split} \end{equation*} |
where \hat{d}_1 and \hat{d}_2 are defined in Lemma 5.3. By Theorem 5.3, \mathcal{A}_0({\bf u}) = {\bf u} has only the positive constant solution u^* . In addition, by investigating the existence of positive roots \lambda_k of Q_k(\lambda, D_0, D_I, D_P) = 0 , we get
\begin{equation} {\rm deg}(I-\mathcal{A}_0, \Lambda, 0) = {\rm {index}}(\mathcal{A}_0, {\bf u^*}) = 1, \end{equation} | (5.21) |
since Lemma 5.3 gives \sigma = l_{\lambda_0} = 0\; {\rm or}\; 2 . Therefore, Eqs (5.19) and (5.20) contradict Eq (5.21). This completes the proof.
In this section, we use numerical simulations of a few different scenarios to illustrate our theoretical findings.
We choose the parameters a = 1 , \beta = 0.96 , d = 0.01 , \gamma = 0.3 , d_1 = 0.2 , d_2 = 2 and \Omega = (0, \pi) . By a direct calculation, we get that system (1.2) has a unique equilibrium E^* = (0.0151, 0.9852) , and Eq (3.18) has two positive roots: \omega_1 = 0.1336 and \omega_2 = 0.1018 . According to Eq (3.20), we obtain the critical values of \tau
\begin{equation*} \tau _1^j = 28.1499, 75.1690,122.1881,169.2072, \cdots, \end{equation*} |
and
\begin{equation*} \tau _2^j = 57.2235,118.9519,180.6803,242.4087, \cdots. \end{equation*} |
In addition, from Eq (3.1), we obtain
\begin{equation*} \left. {d\frac{{{\mathop{\rm Re}\nolimits} \lambda (\tau )}}{{d\tau }}} \right|_{\tau = \tau _1^{(j)}, \lambda = i\omega_1 } > 0, \left. {d\frac{{{\mathop{\rm Re}\nolimits} \lambda (\tau )}}{{d\tau }}} \right|_{\tau = \tau _2^{(j)}, \lambda = i\omega_2} < 0. \end{equation*} |
When \tau = \tau _1^j , a pair of eigenvalues crosses the imaginary axis from left to right. Figure 2 shows the delay time histories for different locations. When \tau \in [0, \tau_1^1)\bigcup(\tau_2^1, \tau_1^2)\bigcup(\tau_2^2, \tau_1^3) , the equilibrium of system (1.2) is asymptotically stable, but it becomes unstable when \tau \in (\tau_1^1, \tau_2^1)\bigcup(\tau_1^2, \tau_2^2)\bigcup (\tau_1^3, +\infty) . In other words, the delay \tau causes the system (1.2) to exhibit the phenomenon of multiple switching events, in which the state of system (1.2) alternates between stable and unstable and vice versa, and the equilibrium E^* is ultimately unstable.
If we keep the other parameters unchanged, only changing the value of \gamma to 0.1 , we find that the positive equilibrium E^* is locally stable for all \tau\geq0 (see Figure 3).
We apply the parameters a = 0.3 , \beta = 0.3 , d = 0.2 , \gamma = 0.3 , d_1 = 0.002 , d_2 = 4 and \Omega = (0, 60) By some calculations, we obtain that system (1.2) has a unique positive equilibrium E^* = (0.2824, 0.2725) . According to Theorem 3.1, by perturbing the initial value at the equilibrium E^* , we find that Turing bifurcation occurs (see Figure 4). Figure 4 shows that system (1.2) has a stable non-constant steady state. But, if we increase the value of \tau to 12 , we find that the stable non-constant steady state disappears, and that the system has a period solution (see Figure 5). However, if we further increase the value of \tau to 81 , we find that the system has chaotic behavior (see Figure 6); the bifurcation diagrams of system (1.2) are shown in Figure 7.
The focus of this paper is on analyzing the effects of the toxins on a delayed diffusive predator-prey model. Overall, the paper provides a thorough analysis of the dynamic behavior of the system, considering various steady states and their stability. The incorporation of a delay in the model allows for an exploration of the effects of time lags in the predator-prey interaction, which adds realism to the study. The findings are interesting and reasonable.
Our system's dynamics were investigated in detail at and near all feasible steady states. We demonstrated that the system is persistent under specific conditions, where both the prey and the predator can survive. Conditions for the Turing bifurcation and global stability of the system at all equilibria are derived and presented. With respect to the delay \tau , we found that the system displays a Hopf bifurcation near its interior equilibrium. Non-constant steady states were also discussed, along with the conditions under which they do and do not exist.
The theoretical findings are then illustrated by means of some numerical simulations. These results demonstrate that the system (1.2) displays spatial patterns and that a delay can lead to Hopf bifurcation and chaos. The findings might be useful for future qualitative research into a similar natural system.
The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.
This work was supported by the National Social Science Fund Youth Project of China (Grant No. 21CJY040).
The authors declare that they have no conflicts of interest.
[1] |
J. J. Hopfield, Neural networks and physical systems with emergent collective computational abilities, Proc. Natl. Acad. Sci. U.S.A, 79 (1982), 2554–2558. https://doi.org/10.1073/pnas.79.8.2554 doi: 10.1073/pnas.79.8.2554
![]() |
[2] |
C. M. Marcus, R. M. Westervelt, Stability of analog neural networks with delay, Phys. Rev. A, 39 (1989), 347–359. https://doi.org/10.1103/PhysRevA.39.347 doi: 10.1103/PhysRevA.39.347
![]() |
[3] |
J. Cao, Global exponential stability of Hopfield neural networks, Int. J. Syst. Sci., 32 (2001), 233–236. https://doi.org/10.1080/002077201750053119 doi: 10.1080/002077201750053119
![]() |
[4] |
J. Cao, D. W. C. Ho, A general framework for global asymptotic stability analysis of delayed neural networks based on LMI approach, Chaos Soliton. Fract., 24 (2005), 1317–1329. https://doi.org/10.1016/j.chaos.2004.09.063 doi: 10.1016/j.chaos.2004.09.063
![]() |
[5] |
Z. Zhang, J. Cao, D. Zhou, Novel LMI-based condition on global asymptotic stability for a class of Cohen-Grossberg BAM networks with extended activation functions, IEEE Trans. Neural Netw. Learn. Syst., 25 (2014), 1161–1172. https://doi.org/10.1109/TNNLS.2013.2289855 doi: 10.1109/TNNLS.2013.2289855
![]() |
[6] |
Q. Song, Exponential stability of recurrent neural networks with both time-varying delays and general activation functions via LMI approach, Neurocomputing, 71 (2008), 2823–2830. https://doi.org/10.1016/j.neucom.2007.08.024 doi: 10.1016/j.neucom.2007.08.024
![]() |
[7] |
S. Arik, Stability analysis of delayed neural networks, IEEE Trans. Circuits Syst. I: Fundam. Theory Appl., 47 (2000), 1089–1092. https://doi.org/10.1109/81.855465 doi: 10.1109/81.855465
![]() |
[8] |
A. Hirose, Nature of complex number and complex-valued neural networks, Front. Electr. Electron. Eng. China, 6 (2011), 171–180. https://doi.org/10.1007/s11460-011-0125-3 doi: 10.1007/s11460-011-0125-3
![]() |
[9] |
C. Aouiti, M. Bessifi, X. Li, Finite-time and fixed-time synchronization of complex-valued recurrent neural networks with discontinuous activations and time-varying delays, Circuits Syst. Signal Process, 39 (2020), 5406–5428. https://doi.org/10.1007/s00034-020-01428-4 doi: 10.1007/s00034-020-01428-4
![]() |
[10] |
Z. Zhang, X. Liu, D. Zhou, C. Lin, J. Chen, H. Wang, Finite-time stabilizability and instabilizability for complex-valued memristive neural networks with time delays, IEEE Trans. Syst. Man. Cybern. Syst., 48 (2018), 2371–2382. https://doi.org/10.1109/TSMC.2017.2754508 doi: 10.1109/TSMC.2017.2754508
![]() |
[11] |
R. Samidurai, R. Sriraman, S. Zhu, Leakage delay-dependent stability analysis for complex-valued neural networks with discrete and distributed time-varying delays, Neurocomputing, 338 (2019), 262–273. https://doi.org/10.1016/j.neucom.2019.02.027 doi: 10.1016/j.neucom.2019.02.027
![]() |
[12] |
K. Subramanian, P. Muthukumar, Global asymptotic stability of complex-valued neural networks with additive time-varying delays, Cogn. Neurodyn., 11 (2017), 293–306. https://doi.org/10.1007/s11571-017-9429-1 doi: 10.1007/s11571-017-9429-1
![]() |
[13] | Q. Song, Z. Zhao, Y. Liu, Stability analysis of complex-valued neural networks with probabilistic time-varying delays, Neurocomputing, 159 (2015), 96–104. https://doi.org/10.1016/j.neucom.2015.02.015 |
[14] |
X. Liu, J. Zheng. Convergence rates of solutions in a predator-prey system withindirect pursuit-evasion interaction in domains of arbitrary dimension, Discrete Contin. Dyn. Syst. B, 28 (2023), 2269–2293. https://doi.org/10.3934/dcdsb.2022168 doi: 10.3934/dcdsb.2022168
![]() |
[15] | M. Winkler, How far do chemotaxis-driven forces influence regularity in the Navier-Stokes system? Trans. Amer. Math. Soc., 369 (2017), 3067–3125. https://doi.org/10.1090/tran/6733 |
[16] |
M. Winkler, Global mass-preserving solutions in a two-dimensional chemotaxis Stokes system with rotation flux components, J. Evol. Eqns., 18 (2018), 1267–1289. https://doi.org/10.1007/s00028-018-0440-8 doi: 10.1007/s00028-018-0440-8
![]() |
[17] |
J. Zheng, An optimal result for global existence and boundedness in a three-dimensional Keller-Segel-Stokes system with nonlinear diffusion, J. Differ. Equ., 267 (2019), 2385–2415. https://doi.org/10.1016/j.jde.2019.03.013 doi: 10.1016/j.jde.2019.03.013
![]() |
[18] |
J. Zheng, A new result for the global existence (and boundedness) and regularity of a three-dimensional Keller-Segel-Navier-Stokes system modeling coral fertilization, J. Differ. Equ., 272 (2021), 164–202. https://doi.org/10.1016/j.jde.2020.09.029 doi: 10.1016/j.jde.2020.09.029
![]() |
[19] |
J. Zheng, Boundedness of solutions to a quasilinear parabolic–elliptic Keller-Segel system with logistic source, J. Differ. Equ., 259 (2015), 120–140. https://doi.org/10.1016/j.jde.2015.02.003 doi: 10.1016/j.jde.2015.02.003
![]() |
[20] |
J. Zheng, Eventual smoothness and stabilization in a three-dimensional Keller-Segel-Navier-Stokes system with rotational flux, Calc. Var. Partial Differ. Equ., 61 (2022), 52. https://doi.org/10.1007/s00526-021-02164-6 doi: 10.1007/s00526-021-02164-6
![]() |
[21] |
Y. Ke, J. Zheng, An optimal result for global existence in a three-dimensional Keller-Segel Navier-Stokes system involving tensor-valued sensitivity with saturation, Calc. Var. Partial Differ. Equ., 58 (2019), 109. https://doi.org/10.1007/s00526-019-1568-2 doi: 10.1007/s00526-019-1568-2
![]() |
[22] |
T. Isokawa, H. Nishimura, N. Kamiura, N. Matsui, Associative memory in quaternionic Hopfield neural network, Int. J. Neural Syst., 18 (2008), 135–145. https://doi.org/10.1142/S0129065708001440 doi: 10.1142/S0129065708001440
![]() |
[23] |
T. Parcollet, M. Morchid, G. Linares, A survey of quaternion neural networks, Artif. Intell. Rev., 53 (2020), 2957–2982. https://doi.org/10.1007/s10462-019-09752-1 doi: 10.1007/s10462-019-09752-1
![]() |
[24] |
Y. Li, X. Meng, Almost automorphic solutions for quaternion-valued Hopfield neural networks with mixed time-varying delays and leakage delays, J. Syst. Sci. Complex., 33 (2020), 100–121. https://doi.org/10.1007/s11424-019-8051-1 doi: 10.1007/s11424-019-8051-1
![]() |
[25] |
Z. Tu, Y. Zhao, N. Ding, Y. Feng, W. Zhang, Stability analysis of quaternion-valued neural networks with both discrete and distributed delays, Appl. Math. Comput., 343 (2019), 342–353. https://doi.org/10.1016/j.amc.2018.09.049 doi: 10.1016/j.amc.2018.09.049
![]() |
[26] |
H. Shu, Q. Song, Y. Liu, Z. Zhao, F. E. Alsaadi, Global stability of quaternion-valued neural networks with non-differentiable time-varying delays, Neurocomputing, 247 (2017), 202–212. https://doi.org/10.1016/j.neucom.2017.03.052 doi: 10.1016/j.neucom.2017.03.052
![]() |
[27] |
M. Tan, Y. Liu, D. Xu, Multistability analysis of delayed quaternion-valued neural networks with nonmonotonic piecewise nonlinear activation functions, Appl. Math. Comput., 341 (2019), 229–255. https://doi.org/10.1016/j.amc.2018.08.033 doi: 10.1016/j.amc.2018.08.033
![]() |
[28] | Y. Kuroe, Models of Clifford recurrent neural networks and their dynamics, Proc. 2011 Int. Joint Conf. Neural Netw., San Jose, CA, USA, (2011), 1035–1041. https://doi.org/10.1109/IJCNN.2011.6033336 |
[29] |
S. Buchholz, G. Sommer, On Clifford neurons and Clifford multi-layer perceptrons, Neural Netw., 21 (2008), 925–935. https://doi.org/10.1016/j.neunet.2008.03.004 doi: 10.1016/j.neunet.2008.03.004
![]() |
[30] | J. Zhu, J. Sun, Global exponential stability of Clifford-valued recurrent neural networks, Neurocomputing, 173 (2016), 685–689. https://doi.org/10.1016/j.neucom.2015.08.016 |
[31] |
Y. Liu, P. Xu, J. Lu, J. Liang, Global stability of Clifford-valued recurrent neural networks with time delays, Nonlinear Dyn., 84 (2016), 767–777. https://doi.org/10.1007/s11071-015-2526-y doi: 10.1007/s11071-015-2526-y
![]() |
[32] |
S. Shen, Y. Li, S^p-Almost periodic solutions of Clifford-valued fuzzy cellular neural networks with time-varying delays, Neural Process. Lett., 51 (2020), 1749–1769. https://doi.org/10.1007/s11063-019-10176-9 doi: 10.1007/s11063-019-10176-9
![]() |
[33] |
Y. Li, J. Xiang, B. Li, Globally asymptotic almost automorphic synchronization of Clifford-valued recurrent neural netwirks with delays, IEEE Access, 7 (2019), 54946–54957. https://doi.org/10.1109/ACCESS.2019.2912838 doi: 10.1109/ACCESS.2019.2912838
![]() |
[34] |
B. Li, Y. Li, Existence and global exponential stability of pseudo almost periodic solution for Clifford-valued neutral high-order Hopfield neural networks with leakage delays, IEEE Access, 7 (2019), 150213–150225. https://doi.org/10.1109/ACCESS.2019.2947647 doi: 10.1109/ACCESS.2019.2947647
![]() |
[35] |
G. Rajchakit, R. Sriraman, C. P. Lim, B. Unyong, Existence, uniqueness and global stability of Clifford-valued neutral-type neural networks with time delays, Math. Comput. Simulat., 201 (2021), 508–527. https://doi.org/10.1016/j.matcom.2021.02.023 doi: 10.1016/j.matcom.2021.02.023
![]() |
[36] |
C. Aouiti, F. Dridi, Weighted pseudo almost automorphic solutions for neutral type fuzzy cellular neural networks with mixed delays and D operator in Clifford algebra, Int. J. Syst. Sci., 51 (2020), 1759–1781. https://doi.org/10.1080/00207721.2020.1777345 doi: 10.1080/00207721.2020.1777345
![]() |
[37] |
G. Rajchakit, R. Sriraman, P. Vignesh, C. P. Lim, Impulsive effects on Cliffordvalued neural networks with time-varying delays: An asymptotic stability analysis, Appl. Math. Comput., 407 (2021), 126309. https://doi.org/10.1016/j.amc.2021.126309 doi: 10.1016/j.amc.2021.126309
![]() |
[38] |
Y. Li, J. Xiang, Existence and global exponential stability of anti-periodic solution for Clifford-valued inertial Cohen-Grossberg neural networks with delays, Neurocomputing, 332 (2019), 259–269. https://doi.org/10.1016/j.neucom.2018.12.064 doi: 10.1016/j.neucom.2018.12.064
![]() |
[39] |
T. Takagi, M. Sugeno, Fuzzy identification of systems and its applications to modeling and control, IEEE Trans. Syst. Man Cybern., 15 (1985), 116–132. https://doi.org/10.1109/TSMC.1985.6313399 doi: 10.1109/TSMC.1985.6313399
![]() |
[40] |
L. Wang, H. K. Lam, New stability criterion for continuous-time Takagi-Sugeno fuzzy systems with time-varying delay, IEEE Trans. Cybern., 49 (2019), 1551–1556. https://doi.org/10.1109/TCYB.2018.2801795 doi: 10.1109/TCYB.2018.2801795
![]() |
[41] | Y. Y. Hou, T. L. Liao, J. J. Yan, Stability analysis of Takagi-Sugeno fuzzy cellular neural networks with time-varying delays, 37 (2007), 720–726. https://doi.org/10.1109/TSMCB.2006.889628 |
[42] |
C. K. Ahn, Delay-dependent state estimation for T-S fuzzy delayed Hopfield neural networks, Nonlinear Dyn., 61 (2010), 483–489. https://doi.org/10.1007/s11071-010-9664-z doi: 10.1007/s11071-010-9664-z
![]() |
[43] |
J. Jian, P. Wan, Global exponential convergence of fuzzy complex-valued neural networks with time-varying delays and impulsive effects, Fuzzy Sets Syst., 338 (2018), 23–39. https://doi.org/10.1016/j.fss.2017.12.001 doi: 10.1016/j.fss.2017.12.001
![]() |
[44] |
R. Sriraman, N. Asha, Global stability of Clifford-valued Takagi-Sugeno fuzzy neural networks with time-varying delays and impulses, Kybernetika, 58 (2022), 498–521. https://doi.org/10.14736/kyb-2022-4-0498 doi: 10.14736/kyb-2022-4-0498
![]() |
[45] |
Z. Zhang, Z. Quan, Global exponential stability via inequality technique for inertial BAM neural networks with time delays, Neurocomputing, 151 (2015), 1316–1326. https://doi.org/10.1016/j.neucom.2014.10.072 doi: 10.1016/j.neucom.2014.10.072
![]() |
[46] |
Z. Zhang, J. Cao, Finite-time synchronization for fuzzy inertial neural networks by maximum value approach, IEEE Trans. Fuzzy Syst., 30 (2022), 1436–1446. https://doi.org/10.1109/TFUZZ.2021.3059953 doi: 10.1109/TFUZZ.2021.3059953
![]() |
[47] |
Z. Zhang, J. Cao, Novel Finite-time synchronization criteria for inertial neural networks with time delays via integral inequality method, IEEE Trans. Neural Netw. Learn. Syst., 30 (2019), 1476–1485. https://doi.org/10.1109/TNNLS.2018.2868800 doi: 10.1109/TNNLS.2018.2868800
![]() |
[48] |
K. Gopalsamy, Stability of artificial neural networks with impulses, Appl. Math. Comput., 154 (2004), 783–813. https://doi.org/10.1016/S0096-3003(03)00750-1 doi: 10.1016/S0096-3003(03)00750-1
![]() |
[49] |
R. Raja, Q. Zhu, S. Senthilraj, R. Samidurai, Improved stability analysis of uncertain neutral type neural networks with leakage delays and impulsive effects, Appl. Math. Comput., 266 (2015), 1050–1069. https://doi.org/10.1016/j.amc.2015.06.030 doi: 10.1016/j.amc.2015.06.030
![]() |
[50] |
R. Samidurai, R. Manivannan, Robust passivity analysis for stochastic impulsive neural networks with leakage and additive time-varying delay components, Appl. Math. Comput., 268 (2015), 743–762. https://doi.org/10.1016/j.amc.2015.06.116 doi: 10.1016/j.amc.2015.06.116
![]() |
[51] |
J. Chen, C. Li, X. Yang, Asymptotic stability of delayed fractional-order fuzzy neural networks with impulse effects, J. Franklin Inst., 355 (2018), 7595–7608. https://doi.org/10.1016/j.jfranklin.2018.07.039 doi: 10.1016/j.jfranklin.2018.07.039
![]() |
[52] |
Y. Tan, S. Tang, J. Yang. Z. Liu, Robust stability analysis of impulsive complex-valued neural networks with time delays and parameter uncertainties, J. Inequal. Appl., 2017 (2017), 215. https://doi.org/10.1186/s13660-017-1490-0 doi: 10.1186/s13660-017-1490-0
![]() |
[53] |
H. Xu, C. Zhang, L. Jiang, J. Smith, Stability analysis of linear systems with two additive timevarying delays via delay-product-type Lyapunov functional, Appl. Math. Model., 45 (2017), 955–964. https://doi.org/10.1016/j.apm.2017.01.032 doi: 10.1016/j.apm.2017.01.032
![]() |
[54] |
P. G. Park, J. W. Ko, C. Jeong, Reciprocally convex approach to stability of systems with time-varying delays, Automatica, 47 (2011), 235–238. https://doi.org/10.1016/j.automatica.2010.10.014 doi: 10.1016/j.automatica.2010.10.014
![]() |