
The rapid development and wide application of artificial intelligence is deeply affecting all aspects of human society. Combine artificial intelligence with the accounting industry, use computers to efficiently and automatically process accounting information, and let the accounting industry move towards the intelligent era. This can help people reduce the workload and speed up work efficiency. In recent years, with the rapid development of economy and technology, the use of financial instrument vouchers has exploded, but the processing requirements of financial instrument vouchers have become more and more efficient. Traditional accounting information processing methods, due to the staff's energy and ability, it is often difficult to quickly and accurately handle accounting information. This makes the processing of accounting information lack of timeliness, the degree of utilization of accounting information by enterprises is relatively low, and the demand for intelligent processing of accounting information is constantly pressing. In view of the above problems, this paper uses image processing technology to intelligently identify the content of accounting information to achieve automatic ticket input, improve work efficiency, reduce error rate and reduce labor costs. By simulating the actual 230 invoice images, the results show that the recognition accuracy rate is as high as 98.7%. The results show that the method is effective and has great application value, which is of great significance to the artificial intelligence of accounting information processing.
Citation: Juanjuan Tian, Li Li. Research on artificial intelligence of accounting information processing based on image processing[J]. Mathematical Biosciences and Engineering, 2022, 19(8): 8411-8425. doi: 10.3934/mbe.2022391
[1] | Saheb Pal, Nikhil Pal, Sudip Samanta, Joydev Chattopadhyay . Fear effect in prey and hunting cooperation among predators in a Leslie-Gower model. Mathematical Biosciences and Engineering, 2019, 16(5): 5146-5179. doi: 10.3934/mbe.2019258 |
[2] | Chunmei Zhang, Suli Liu, Jianhua Huang, Weiming Wang . Stability and Hopf bifurcation in an eco-epidemiological system with the cost of anti-predator behaviors. Mathematical Biosciences and Engineering, 2023, 20(5): 8146-8161. doi: 10.3934/mbe.2023354 |
[3] | Kawkab Al Amri, Qamar J. A Khan, David Greenhalgh . Combined impact of fear and Allee effect in predator-prey interaction models on their growth. Mathematical Biosciences and Engineering, 2024, 21(10): 7211-7252. doi: 10.3934/mbe.2024319 |
[4] | Sangeeta Kumari, Sidharth Menon, Abhirami K . Dynamical system of quokka population depicting Fennecaphobia by Vulpes vulpes. Mathematical Biosciences and Engineering, 2025, 22(6): 1342-1363. doi: 10.3934/mbe.2025050 |
[5] | Ranjit Kumar Upadhyay, Swati Mishra . Population dynamic consequences of fearful prey in a spatiotemporal predator-prey system. Mathematical Biosciences and Engineering, 2019, 16(1): 338-372. doi: 10.3934/mbe.2019017 |
[6] | Yuanfu Shao . Bifurcations of a delayed predator-prey system with fear, refuge for prey and additional food for predator. Mathematical Biosciences and Engineering, 2023, 20(4): 7429-7452. doi: 10.3934/mbe.2023322 |
[7] | Fang Liu, Yanfei Du . Spatiotemporal dynamics of a diffusive predator-prey model with delay and Allee effect in predator. Mathematical Biosciences and Engineering, 2023, 20(11): 19372-19400. doi: 10.3934/mbe.2023857 |
[8] | Hongqiuxue Wu, Zhong Li, Mengxin He . Dynamic analysis of a Leslie-Gower predator-prey model with the fear effect and nonlinear harvesting. Mathematical Biosciences and Engineering, 2023, 20(10): 18592-18629. doi: 10.3934/mbe.2023825 |
[9] | Jin Zhong, Yue Xia, Lijuan Chen, Fengde Chen . Dynamical analysis of a predator-prey system with fear-induced dispersal between patches. Mathematical Biosciences and Engineering, 2025, 22(5): 1159-1184. doi: 10.3934/mbe.2025042 |
[10] | Rongjie Yu, Hengguo Yu, Chuanjun Dai, Zengling Ma, Qi Wang, Min Zhao . Bifurcation analysis of Leslie-Gower predator-prey system with harvesting and fear effect. Mathematical Biosciences and Engineering, 2023, 20(10): 18267-18300. doi: 10.3934/mbe.2023812 |
The rapid development and wide application of artificial intelligence is deeply affecting all aspects of human society. Combine artificial intelligence with the accounting industry, use computers to efficiently and automatically process accounting information, and let the accounting industry move towards the intelligent era. This can help people reduce the workload and speed up work efficiency. In recent years, with the rapid development of economy and technology, the use of financial instrument vouchers has exploded, but the processing requirements of financial instrument vouchers have become more and more efficient. Traditional accounting information processing methods, due to the staff's energy and ability, it is often difficult to quickly and accurately handle accounting information. This makes the processing of accounting information lack of timeliness, the degree of utilization of accounting information by enterprises is relatively low, and the demand for intelligent processing of accounting information is constantly pressing. In view of the above problems, this paper uses image processing technology to intelligently identify the content of accounting information to achieve automatic ticket input, improve work efficiency, reduce error rate and reduce labor costs. By simulating the actual 230 invoice images, the results show that the recognition accuracy rate is as high as 98.7%. The results show that the method is effective and has great application value, which is of great significance to the artificial intelligence of accounting information processing.
Interactions between the predators and the preys are diverse and complex in ecology. The predators increase the mortality rate of preys by direct predation. In existing literatures, many predator-prey models only involve direct predation for predator-prey interactions [1,2,3,4]. However, besides the population loss caused by direct predation, the prey will modify their behavior, psychology and physiology in response to the predation risk. This is defined as the fear effect by Cannon [5]. Zanette et al. [6] investigated the variation of song sparrows offspring reproduction when the sounds and calls of predators were broadcasted to simulate predation risk. They discovered that the fear effect alone can cause a significant reduction of song sparrows offspring reproduction. Inspired by this experimental result, Wang et al. [7] incorporated the fear effect into the predator-prey model. In their theoretical analysis, linear and Holling type Ⅱ functional response are chosen respectively. According to their results, the structure of the equilibria will not be affected by the fear effects, but the stability of equilibria and Hopf bifurcation are slightly different from models with no fear effects.
Soon afterward, Wang and Zou [8] considered a model with the stage structure of prey (juvenile prey and adult prey) and a maturation time delay. Additionally, Wang and Zou [9] pointed out anti-predation behaviors will not only decrease offspring reproduction of prey but also increase the difficulty of the prey being caught. Based on this assumption, they derived an anti-predation strategic predator-prey model.
Recently, the aforementioned ordinary or delayed differential equation models were extended to reaction diffusion equation [10,11,12]. Wang et al. [11] used several functional responses to study the effect of the degree of prey sensitivity to predation risk on pattern formation. Following this work, Wang et al. [10] introduced spatial memory delay and pregnancy delay into the model. Their numerical simulation presented the effect of some biological important variables, including the level of fear effect, memory-related diffusion, time delay induced by spatial memory and pregnancy on pattern formation. Moreover, Dai and Sun [13] incorporated chemotaxis and fear effect into predator-prey model, and investigated the Turing-Hopf bifurcation by selecting delay and chemotaxis coefficient as two analysis parameters.
Denote by u1(x,t) and u2(x,t), the population of the prey and adult predator at location x and time t, respectively. We suppose juvenile predators are unable to prey on. By choosing simplest linear functional response in the model of Wang and Zou [7], the equation for prey population is given by
∂tu1−d1Δu1=u1(rg(K,u2)−μ1−mu1)−pu1u2, |
where d1 is a diffusion coefficient for prey, μ1 is the mortality rate for prey, m is the intraspecies competition coefficient, p is predation rate, r is reproduction rate for prey and g(K,u2) represents the cost of anti-predator defense induced by fear with K reflecting response level. We assume g(K,u2) satisfies the following conditions.
(H1) g(0,u2)=g(K,0)=1, limu2→∞g(K,u2)=limK→∞g(K,u2)=0, ∂g(K,u2)∂K<0 and ∂g(K,u2)∂u2<0.
Considering the maturation period of the predator, we set b(x,t,a1) be the density of the predator at age a1, location x and time t. Establish the following population model with spatial diffusion and age-structure
(∂a1+∂t)b(x,t,a1)=d(a1)Δb(x,t,a1)−μ(a1)b(x,t,a1),b(x,t,0)=cpu1(x,t)u2(x,t), | (1.1) |
where x∈Ω, a bounded spatial habitat with the smooth boundary ∂Ω, t,a1>0, c is the conversion rate of the prey to predators, τ>0 be the maturation period of predator and age-specific functions
d(a1)={d0, a1≤τ,d2, a1>τ, μ(a1)={γ, a1≤τ,μ2, a1>τ, |
represent the diffusion rate and mortality rate at age a1, respectively. We introduce the total population of the matured predator as u2=∫∞τb(x,t,a1)da1. Thus, (1.1) together with b(x,t,∞)=0 yields
∂tu2=d2Δu2+b(x,t,τ)−μ2u2. | (1.2) |
Let s1=t−a1 and w(x,t,s1)=b(x,t,t−s1). Along the characteristic line, solving (1.1) yields
∂tw(x,t,s1)={d0Δw(x,t,s1)−γw(x,t,s1), x∈Ω, 0≤t−s1≤τ,d2Δw(x,t,s1)−μ2w(x,t,s1), x∈Ω, t−s1>τ,w(x,s1,s1)=b(x,s1,0)=cpu1(x,s1)u2(x,s1), s1≥0; w(x,0,s1)=b(x,0,−s1), s1<0. |
Assume linear operator d0Δ−γ with Neumann boundary conditions yields the C0 semigroups T1(t). Therefore,
b(x,t,τ)=w(x,t,t−τ)={T1(τ)b(⋅,t−τ,0), t>τ,T1(t)b(⋅,0,τ−t), t≤τ. |
In particular, G(x,y,t) denotes the kernel function corresponding to T1(t). Thus
b(x,t,τ)=T1(τ)b(⋅,t−τ,0)=cp∫ΩG(x,y,τ)u1(y,t−τ)u2(y,t−τ)dy, t>τ. |
The above equation together with (1.2) yields a nonlocal diffusive predator-prey model with fear effect and maturation period of predators
∂u1∂t=d1Δu1+u1(x,t)[rg(K,u2(x,t))−μ1−mu1(x,t)]−pu1(x,t)u2(x,t),∂u2∂t=d2Δu2+cp∫ΩG(x,y,τ)u1(y,t−τ)u2(y,t−τ)dy−μ2u2(x,t). | (1.3) |
Since the spatial movement of mature predators is much bigger than that of juvenile predators, we assume that diffusion rate of juvenile predators d0 approaches zero. Hence, the kernel function becomes G=e−γτf(x−y) with a Dirac-delta function f. Thus, the equation of u2(x,t) in (1.3) becomes
∂u2∂t=d2Δu2+cpe−γτ∫Ωf(x−y)u1(y,t−τ)u2(y,t−τ)dy−μ2u2(x,t). |
It follows from the properties of Dirac-delta function that
∫Ωf(x−y)u1(y,t−τ)u2(y,t−τ)dy=limϵ→0∫Bϵ(x)f(x−y)u1(y,t−τ)u2(y,t−τ)dy=u1(x,t−τ)u2(x,t−τ)limϵ→0∫Bϵ(x)f(x−y)dy=u1(x,t−τ)u2(x,t−τ) |
where Bϵ(x) is the open ball of radius ϵ centered at x. Therefore, model (1.3) equipped with nonnegative initial conditions and Neumann boundary conditions is
∂u1(x,t)∂t=d1Δu1(x,t)+u1(x,t)[rg(K,u2(x,t))−μ1−mu1(x,t)]−pu1(x,t)u2(x,t), x∈Ω,t>0,∂u2(x,t)∂t=d2Δu2(x,t)+cpe−γτu1(x,t−τ)u2(x,t−τ)−μ2u2(x,t), x∈Ω,t>0,∂u1(x,t)∂ν=∂u2(x,t)∂ν=0,x∈∂Ω,t>0,u1(x,ϑ)=u10(x,ϑ)≥0,u2(x,ϑ)=u20(x,ϑ)≥0,x∈Ω, ϑ∈[−τ,0]. | (1.4) |
If r≤μ1, then as t→∞, we have (u1,u2)→(0,0) for x∈Ω, namely, two species will extinct. Throughout this paper, suppose r>μ1, which ensures that the prey and predator will persist.
This paper is organized as follows. We present results on well-posedness and uniform persistence of solutions and prove the global asymptotic stability of predator free equilibrium in Section 2. The nonexistence of nonhomogeneous steady state and steady state bifurcation are proven in Section 3. Hopf bifurcation analysis is carried out in Section 4. In Section 5, we conduct numerical exploration to illustrate some theoretical conclusions and further explore the dynamics of the nonlocal model numerically. We sum up our paper in Section 6.
Denote by C:=C([−τ,0],X2) the Banach space of continuous maps from [−τ,0] to X2 equipped with supremum norm, where X=L2(Ω) is the Hilbert space of integrable function with the usual inner product. C+ is the nonnegative cone of C. Let u1(x,t) and u2(x,t) be a pair of continuous function on Ω×[−τ,∞) and (u1t,u2t)∈C as (u1t(ϑ),u2t(ϑ))=(u1(⋅,t+ϑ),u2(⋅,t+ϑ)) for ϑ∈[−τ,0]. By using [14,Corollary 4], we can prove model (1.4) exists a unique solution. Note (1.4) is mixed quasi-monotone [15], together with comparison principle implies that the solution of (1.4) is nonnegative.
Lemma 2.1. For any initial condition (u10(x,ϑ),u20(x,ϑ))∈C+, model (1.4) possesses a unique solution (u1(x,t),u2(x,t)) on the maximal interval of existence [0,tmax). If tmax<∞, then lim supt→t−max(‖u1(⋅,t)‖+‖u2(⋅,t)‖)=∞. Moreover, u1 and u2 are nonnegative for all (x,t)∈¯Ω×[−τ,tmax).
We next prove u1 and u2 are bounded which implies that tmax=∞.
Theorem 2.2. For any initial condition φ=(u10(x,ϑ),u20(x,ϑ))∈C+, model (1.4) possesses a global solution (u1(x,t),u2(x,t)) which is unique and nonnegative for (x,t)∈¯Ω×[0,∞). If u10(x,ϑ)≥0(≢0),u20(x,ϑ)≥0(≢0), then this solution remains positive for all (x,t)∈¯Ω×(0,∞). Moreover, there exists a positive constant ξ independent of φ such that lim supt→∞u1≤ξ, lim supt→∞u2≤ξ for all x∈Ω.
Proof. Consider
∂w1∂t=d1Δw1+rw1−μ1w1−mw21, x∈Ω,t>0,∂w1∂ν=0, x∈∂Ω,t>0, w1(x,0)=supϑ∈[−τ,0]u10(x,ϑ), x∈Ω. | (2.1) |
Clearly, w1(x,t) of (2.1) is a upper solution to (1.4) due to ∂u1/∂t≤d1Δu1+ru1−μ1u1−mu21. By using Lemma 2.2 in [16], (r−μ1)/m of (2.1) is globally asymptotically stable in C(¯Ω,R+). This together with comparison theorem indicates
lim supt→∞u1≤limt→∞w1=r−μ1m uniformly for x∈¯Ω. | (2.2) |
Thus, there exists ˜ξ>0 which is not dependent on initial condition, such that ‖u1‖≤˜ξ for all t>0. T2(t) denotes the C0 semigroups yielded by d2Δ−μ2 with the Neumann boundary condition. Then from (1.4),
u2=T2(t)u20(⋅,0)+cpe−γτ∫t−τ−τT2(t−τ−a)u1(⋅,a)u2(⋅,a)da. |
Let −δ<0 be the principle eigenvalue of d2Δ−μ2 with the Neumann boundary condition. Then, ‖T2(t)‖≤e−δt. The above formula yields
‖u2(⋅,t)‖≤e−δt‖u20(⋅,0)‖+cpe−γτ˜ξ∫t−τ−τe−δ(t−τ−a)‖u2(⋅,a)‖da,≤˜B1+∫t0˜B2‖u2(⋅,a)‖da, |
by choosing constants ˜B1≥‖u2(⋅,0)‖+cpτ˜ξsupa∈[−τ,0]‖u2(⋅,a)‖ and ˜B2≥cp˜ξ. Using Gronwall's inequality yields ‖u2(⋅,t)‖≤˜B1e˜B2t for all 0≤t<tmax. Lemma 2.1 implies tmax=∞. So (u1,u2) is a global solution. Moreover, if u10(x,ϑ)≥0(≢0),u20(x,ϑ)≥0(≢0), then by [17,Theorem 4], this solution is positive for all t>0 and x∈¯Ω.
We next prove (u1,u2) is ultimately bounded by a constant which is not dependent on the initial condition. Due to (2.2), there exist t0>0 and ξ0>0 such that u1(x,t)≤ξ0 for any t>t0 and x∈¯Ω. Let z(x,t)=cu1(x,t−τ)+u2(x,t), μ=min{μ1,μ2} and I1=∫Ωz(x,t)dx. We integrate both sides of (1.4) and add up to obtain
I′1(t)≤∫Ω(c(r−μ1)u1(x,t−τ)−μ2u2(x,t))dx≤crξ0|Ω|−μI1(t), t≥t0+τ. |
Comparison principle implies
lim supt→∞‖u2‖1≤lim supt→∞I1≤crξ0|Ω|/μ. |
Especially, there exist t1>t0 and ξ1>0 such that ‖u2‖1≤ξ1 for all t≥t1.
Now, we define Vl(t)=∫Ω(u2(x,t))ldx with l≥1, and estimate the upper bound of V2(t). For t>t1, the second equation of (1.4) and Young's inequality yield
12V′2(t)≤−d2∫Ω|∇u2|2dx+cpξ0∫Ωu2(x,t−τ)u2(x,t)dx−μ2V2(t)≤−d2‖∇u2‖22+cpξ02V2(t−τ)+cpξ02V2(t). |
The Gagliardo-Nirenberg inequality states:
∀ ϵ>0,∃ ˆc>0, s.t. ‖P‖22≤ϵ‖∇P‖22+ˆcϵ−n/2‖P‖21, ∀ P∈W1,2(Ω). |
We obtain
V′2(t)≤C1+C2V2(t−τ)−(C2+C3)V2(t), |
where C1=ˆcϵ−n/2−12d2ξ21>0, C2=cpξ0>0 and C3=2(d2/ϵ−C2)>0 with small ϵ∈(0,d2/C2). Using comparison principle again yields lim supt→∞V2(t)≤C1/C3, which implies there exist t2>t1 and ξ2>0 such that V2≤ξ2 for t≥t2.
Let Ll=lim supt→∞Vl(t) with l≥1, we want to estimate L2l with the similar method of estimation for L2. Multiply the second equation in (1.4) by 2lv2l−1 and integrate on Ω. Young's inequality implies
V′2l(t)≤−2d2∫Ω|∇ul2|2dx+cpξ0V2l(t−τ)+(2l−1)cpξ0V2l(t). |
Then
V′2l(t)≤2d2ˆcϵ−n/2−1V2l(t)−2d2ϵV2l(t)+cpξ0(V2l(t−τ)+(2l−1)V2l(t)), |
via Gagliardo-Nirenberg inequality. Since Ll=lim supt→∞Vl(t), there exists tl>0 such that Vl≤1+Ll when t>tl. Hence,
V′2l(t)≤2d2ˆcϵ−(n/2+1)(1+Ll)2−2d2ϵV2l(t)+lC4(V2l(t)+V2l(t−τ)) |
with C4=2cpξ0. We choose ϵ−1=(2C4+1)l/(2d2) and C5=2d2ˆc[(2C4+1)/(2d2)]n/2+1. Then for t>tl, we obtain
V′2l(t)≤C5ln/2+1(1+Ll)2+lC4V2l(t−τ)−(lC4+l)V2l(t). |
By comparison principle, the above inequality yields L2l≤C5ln/2(1+Ll)2, with a constant C5 which is not dependent on l and initial conditions. Finally, prove L2s<∞ for all s∈N0. Let B=1+C5 and {bs}∞s=0 be an infinite sequence denoted by bs+1=B(1/2)(s+1)2sn((1/2)(s+2))bs with the first term b0=L1+1. Clearly, L2s≤(bs)2s and
lims→∞lnbs=lnb0+lnB+n2ln2. |
Therefore,
lim sups→∞(L2s)(1/2)s≤lims→∞bs=B(1+L1)2n/2≤B(1+ξ1)2n/2≤ξ:=max{B(ξ1+1)2n/2,(r−μ1)/m}. |
The above inequality leads to lim supt→∞u1≤ξ and lim supt→∞u2≤ξ for all x∈Ω.
In Theorem 2.2, we proved that the solution of model (1.4) is uniformly bounded for any nonnegative initial condition, this implies the boundedness of the population of two species. Clearly model (1.4) exists two constant steady states E0=(0,0) and E1=((r−μ1)/m,0), where E0 is a saddle. Define the basic reproduction ratio [18] by
R0=cpe−γτ(r−μ1)mμ2. | (2.3) |
Thus model (1.4) possesses exactly one positive constant steady state E2=(u∗1,u∗2) if and only if R0>1, which is equivalent to cp(r−μ1)>mμ2 and 0≤τ<τmax:=1γlncp(r−μ1)mμ2. Here,
u∗1=μ2eγτpc, u∗2 satisfies rg(K,u2)−pu2=μ1+mu∗1. |
The linearization of (1.4) at the positive constant steady state (˜u1,˜u2) gives
∂W/∂t=DΔW+L(Wt), | (2.4) |
where domain Y:={(u1,u2)T:u1,u2∈C2(Ω)∩C1(ˉΩ),(u1)ν=(u2)ν=0 on ∂Ω}, W=(u1(x,t),u2(x,t))T, D=diag(d1,d2) and a bounded linear operator L:C→X2 is
L(φ)=Mφ(0)+Mτφ(−τ), for φ∈C, |
with
M=(rg(K,˜u2)−μ1−2m˜u1−p˜u2˜u1(rg′u2(K,˜u2)−p)0−μ2), Mτ=(00cpe−γτ˜u2cpe−γτ˜u1). |
The characteristic equation of (2.4) gives
ρη−DΔη−L(eρ⋅η)=0, for some η∈Y∖{0}, |
or equivalently
det(ρI+σnD−M−e−ρτMτ)=0, for n∈N0. | (2.5) |
Here, σn is the eigenvalue of −Δ in Ω with Neumann boundary condition with respect to eigenfunction ψn for all n∈N0, and
0=σ0<σ1≤σ2≤⋯≤σn≤σn+1≤⋯ and limn→∞σn=∞. | (2.6) |
Theorem 2.3. (i) The trivial constant steady state E0=(0,0) is always unstable.
(ii) If R0>1, then E1=((r−μ1)/m,0) is unstable, and model (1.4) possesses a unique positive constant steady state E2=(u∗1,u∗2).
(iii) If R0≤1, then E1 is globally asymptotically stable in C+.
Proof. (ⅰ) Note, (2.5) at E0 takes form as (ρ+σnd1−r+μ1)(ρ+σnd2+μ2)=0 for all n∈N0. Then r−μ1>0 is a positive real eigenvalue, namely, E0 is always unstable.
(ⅱ) The characteristic equation at E1 gives
(ρ+σnd1+r−μ1)(ρ+σnd2+μ2−cpe−γτr−μ1me−ρτ)=0 for n∈N0. | (2.7) |
Note that one eigenvalue ρ1=−σnd1−r+μ1 remains negative. Hence, we only need to consider the root distribution of the following equation
ρ+σnd2+μ2−cpe−γτr−μ1me−ρτ=0 for n∈N0. | (2.8) |
According to [19,Lemma 2.1], we obtain that (2.8) exists an eigenvalue ρ>0 when R0>1, namely, E1 is unstable when R0>1.
(ⅲ) By using [19,Lemma 2.1] again, any eigenvalue ρ of (2.8) satisfies Re(ρ)<0 when R0<1, namely, E1 is locally asymptotically stable when R0<1. Now, consider R0=1. 0 is an eigenvalue of (2.7) for n=0 and all other eigenvalues satisfy Re(ρ)<0. To prove the stability of E1, we shall calculate the normal forms of (1.4) by the algorithm introduced in [20]. Set
Υ={ρ∈C,ρ is the eigenvalue of equation (2.7) and Reρ=0}. |
Obviously, Υ={0} when R0=1. System (1.4) satisfies the non-resonance condition relative to Υ. Denote ¯u1=(r−μ1)/m, and let w=(w1,w2)T=(¯u1−u1,u2)T and (1.4) can be written as
˙wt=A0wt+F0(wt) on C. |
Here linear operator A0 is given by (A0φ)(ϑ)=(φ(ϑ))′ when ϑ∈[−τ,0) and
(A0φ)(0)=(d1Δ00d2Δ)φ(0)+(−r+μ1(p−rg′u2(K,0))¯u10−μ2)φ(0)+(000μ2)φ(−τ), |
and the nonlinear operator F0 satisfies [F0(φ)](ϑ)=0 for −τ≤ϑ<0. By Taylor expansion, [F0(φ)](0) can be written as
[F0(φ)](0)=(mφ21(0)−r¯u1g″u2(K,0)φ22(0)/2+(rg′u2(K,0)−p)φ1(0)φ2(0)−cpe−γτφ1(−τ)φ2(−τ))+h.o.t. | (2.9) |
Define a bilinear form
⟨β,α⟩=∫Ω[α1(0)β1(0)+α2(0)β2(0)+μ2∫0−τβ2(ϑ+τ)α2(ϑ)dϑ]dx, β∈C([0,τ],X2), α∈C. |
Select α=(1,m/(p−rg′u2(K,0)) and β=(0,1)T to be the right and left eigenfunction of A0 relative to eigenvalue 0, respectively. Decompose wt as wt=hα+δ and ⟨β,δ⟩=0. Notice A0α=0 and ⟨β,A0δ⟩=0. Thus,
⟨β,˙wt⟩=⟨β,A0wt⟩+⟨β,F0(wt)⟩=⟨β,F0(wt)⟩. |
Moreover,
⟨β,˙wt⟩=˙h⟨β,α⟩+⟨β,˙δ⟩=˙h⟨β,α⟩. |
It follows from the above two equations that
˙hm(1+μ2τ)|Ω|p−rg′u2(K,0)=⟨β,F0(hα+δ)⟩=∫ΩβT[F0(hα+δ)](0)dx=∫Ω[F0(hα+δ)]2(0)dx. |
When the initial value is a small perturbation of E1, then δ=O(h2), together with Taylor expansion yields
[F0(hα+δ)]2(0)=−cpe−γτ(hα1(−τ)+δ1(−τ))(hα2(−τ)+δ2(−τ))=−cpe−γτmp−rg′u2(K,0)h2+O(h3). |
Therefore, we obtain the norm form of (1.4) as follows
˙h=−cpe−γτ1+μ2τh2+O(h3). | (2.10) |
Then for any positive initial value, the stability of zero solution of (2.10) implies E1 is locally asymptotically stable when R0=1.
Next, it suffices to show the global attractivity of E1 in C+ when R0≤1. Establish a Lyapunov functional V:C+→R as
V(ϕ1,ϕ2)=∫Ωϕ2(0)2dx+cpe−γτr−μ1m∫Ω∫0−τϕ2(θ)2dθdx for (ϕ1,ϕ2)∈C+. |
Along solutions of (1.4), taking derivative of V(ϕ1,ϕ2) with respect to t yields
dVdt≤−2d2∫Ω|∇u2|2dx+∫Ω2cpe−γτr−μ1mu2(x,t−τ)u2(x,t)−2μ2u22(x,t)+cpe−γτr−μ1m[u22(x,t)−u22(x,t−τ)]dx≤∫Ω2μ2(R0−1)u22(x,t)dx≤0 if R0≤1. |
Note {E1} is the maximal invariant subset of dV/dt=0, together with LaSalle-Lyapunov invariance principle [21,22] implies E1 is globally asymptotically stable if R0≤1.
In Theorem 2.3, E0 is always unstable which suggests that at leat one species will persist eventually. Moreover, if R0≤1, then E1 is globally asymptotically stable in C+, which implies that when the basic reproduction ratio is no more than one, the predator species will extinct and only the prey species can persist eventually. Next, we will prove the solution is uniformly persistent. Θt denotes the solution semiflow of (1.4) mapping C+ to C+; namely, Θtφ:=(u1(⋅,t+⋅),u2(⋅,t+⋅))∈C+. Set ζ+(φ)=∪t≥0{Θtφ} be the positive orbit and ϖ(φ) be the omega limit set of ζ+(φ). Denote
Z:={(φ1,φ2)∈C+:φ1≢0 and φ2≢0}, ∂Z:=C+∖Z={(φ1,φ2)∈C+:φ1≡0 or φ2≡0}, |
Γ∂ as the largest positively invariant set in ∂Z, and Ws((˜u1,˜u2)) as the stable manifold associated with (˜u1,˜u2). We next present persistence result of model (1.4).
Theorem 2.4. Suppose R0>1. Then there exists κ>0 such that lim inft→∞u1(x,t)≥κ and lim inft→∞u2(x,t)≥κ for any initial condition φ∈Z and x∈¯Ω.
Proof. Note Θt is compact, and Theorem 2.2 implies Θt is point dissipative. Then Θt possesses a nonempty global attractor in C+ [23]. Clearly, Γ∂={(φ1,φ2)∈C+:φ2≡0}, and ϖ(φ)={E0,E1} for all φ∈Γ∂. Define a generalized distance function ψ mapping C+ to R+ by
ψ(φ)=minx∈ˉΩ{φ1(x,0),φ2(x,0)}, ∀φ=(φ1,φ2)∈C+. |
Following from strong maximum principle [24], ψ(Θtφ)>0 for all φ∈Z. Due to ψ−1(0,∞)⊂Z, assumption (P) in [25,Section 3] holds. Then verify rest conditions in [25,Theorem 3].
First, prove Ws(E0)∩ψ−1(0,∞)=∅. Otherwise, there exists an initial condition φ∈C+ with ψ(φ)>0, such that (u1,u2)→E0 as t→∞. Thus, for any sufficiently small ε1>0 satisfying rg(K,ε1)−μ1>pε1, there exists t1>0 such that 0<u1,u2<ε1 for all x∈Ω and t>t1. Note that rg(K,0)−μ1>0 and ∂g(K,u2)/∂u2<0 ensure the existence of small ε1>0. Then the first equation in (1.4) and (H1) lead to
∂tu1>d1Δu1+u1[(rg(K,ϵ1)−μ1−pϵ1)−mu1],t>t1. |
Notice
∂tˆu1−d1Δˆu1=ˆu1[(rg(K,ϵ1)−pϵ1−μ1)−mˆu1], x∈Ω,t>t1,∂νˆu1=0, x∈∂Ω,t>t1, |
has a globally asymptotically stable positive steady state (rg(K,ϵ1)−μ1−pϵ1)/m due to [16,Lemma 2.2], together with comparison principle yields limt→∞u1≥limt→∞ˆu1>0. A contradiction is derived, so Ws(E0)∩ψ−1(0,∞)=∅.
Next check Ws(E1)∩ψ−1(0,∞)=∅. If not, there exists φ∈C+ with ψ(φ)>0 such that (u1,u2) converges to E1 as t→∞. According to (2.2), for any small ε2>0 satisfying cpe−γτ((r−μ1)/m−ϵ2)>μ2, there exists t2>0 such that u1>(r−μ1)/m−ε2 for all x∈¯Ω and t>t2−τ. Note that R0>1 ensures the existence of ε2>0. Thus, the second equation of (1.4) yields
∂tu2>d2Δu2+cpe−γτ(r−μ1m−ε2)u2(x,t−τ)−μ2u2(x,t), t>t2. |
In a similar manner, we derive limt→∞u2(x,t)>0 by above inequality, cpe−γτ((r−μ1)/m−ϵ2)>μ2 and comparison principle. A contradiction yields again. Hence, it follows from Theorem 3 in [25] that, for any φ∈C+, there exists κ>0 such that lim inft→∞ψ(Θtφ)≥κ uniformly for any x∈¯Ω.
Theorem 2.4 implies that when the basic reproduction ratio is bigger than one, both the predator species and prey species will persist eventually. We next investigate the stability of E2. The corresponding characteristic equation at E2 gives
ρ2+a1,nρ+a0,n+(b1,nρ+b0,n)e−ρτ=0, n∈N0, | (2.11) |
with
a1,n=σn(d1+d2)+μ2+mu∗1>0, a0,n=(σnd1+au∗1)(σnd2+μ2)>0,b1,n=−μ2<0, b0,n=−μ2(σnd1+mu∗1+u∗2(rg′u2(K,u∗2)−p)). |
Characteristic equation (2.11) with τ=0 is
ρ2+(a1,n+b1,n)ρ+a0,n+b0,n=0, n∈N0. | (2.12) |
We observe that a0,n+b0,n=σnd2(σnd1+mu∗1)−μ2u∗2(rg′u2(K,u∗2)−p)>0, and a1,n+b1,n=σn(d1+d2)+mu∗1>0 for all integer n≥0, which yields that any eigenvalue ρ of (2.12) satisfies Re(ρ)<0. Then, local asymptotic stability of E2 is derived when τ=0 which implies Turing instability can not happen for the non-delay system of (1.4). In addition, a0,n+b0,n>0 for any n∈N0 leads to that (2.11) can not have an eigenvalue 0 for any τ≥0. This suggests we look for the existence of simple ρ=±iδ (δ>0) for some τ>0. Substitute ρ=iδ into (2.11) and then
Gn(δ,τ)=δ4+(a21,n−2a0,n−b21,n)δ2+a20,n−b20,n=0, n∈N0, | (2.13) |
with
a21,n−2a0,n−b21,n=(σnd1+mu∗1)2+(σnd2)2+2μ2σnd2>0,a0,n+b0,n=σnd2(σnd1+mu∗1)−μ2u∗2(rg′u2(K,u∗2)−p)>0,a0,n−b0,n=σ2nd1d2+σn(2μ2d1+mu∗1d2)+μ2(2mu∗1+u∗2(rg′u2(K,u∗2)−p)). |
Thus, a0,n−b0,n≥0 for all n∈N0 is equivalent to
(A0): 2mu∗1≥u∗2(p−rg′u2(K,u∗2)). |
If (A0) holds, then (2.13) admits no positive roots, together with for τ=0, any eigenvalues ρ of (2.11) satisfies Re(ρ)<0, yields the next conclusion.
Theorem 2.5. Suppose R0>1. Then, E2 is locally asymptotically stable provided that (A0) holds.
Now, we consider positive nonhomogeneous steady states. The steady state (u1(x),u2(x)) of (1.4) satisfies the elliptic equation
−d1Δu1=rg(K,u2)u1−μ1u1−mu21−pu1u2, x∈Ω,−d2Δu2=cpe−γτu1u2−μ2u2, x∈Ω,∂νu1=∂νu2=0, x∈∂Ω. | (3.1) |
From Theorem 2.3, all the solutions converge to E1 when R0≤1 and the positive nonhomogeneous steady state may exist only if R0>1. Throughout this section, we assume that R0>1. In what follows, the positive lower and upper bounds independent of steady states for all positive solutions to (3.1) are derived.
Theorem 3.1. Assume that R0>1. Then any nonnegative steady state of (3.1) other than (0,0), and ((r−μ1)/m,0) should be positive. Moreover, there exist constants ¯B,B_>0 which depend on all parameters of (3.1) and Ω, such that B_≤u1(x),u2(x)≤¯B for any positive solution of (3.1) and x∈¯Ω.
Proof. We first show any nonnegative solution (u1,u2) other than E0 and E1, should be u1>0 and u2>0 for all x∈¯Ω. To see this, suppose u2(x0)=0 for some x0∈¯Ω, then u2(x)≡0 via strong maximum principle and
0≤d1∫Ω|∇(u1−r−μ1m)|2dx=∫Ω−mu1(x)(u1(x)−r−μ1m)2dx≤0. |
Thus the above inequality implies u1(x)≡0 or u1(x)≡(r−μ1)/m. Now, we assume u2>0 for all x∈¯Ω. Strong maximum principle yields u1>0 for all x∈¯Ω. Hence, u1>0 and u2>0 for all x∈¯Ω.
We now prove u1 and u2 have a upper bound which is a positive constant. Since −d1Δu1(x)≤(r−μ1−mu1(x))u1(x), we then obtain from Lemma 2.3 in [26] that u1(x)≤(r−μ1)/m for any x∈¯Ω.
By two equations in (3.1), we obtain
−Δ(d1cu1+d2u2)≤rc(r−μ1)/m−min{μ1d1,μ2d2}(d1cu1+d2u2). |
By using [26,Lemma 2.3] again, we conclude
u1(x),u2(x)≤¯B=rc(r−μ1)mmin{cμ1,μ2,μ1d2/d1,μ2d1c/d2}. |
Next, we only need to prove ‖u1(x)‖ and ‖u2(x)‖ have a positive lower bound which is not dependent on the solution. Otherwise, there exists a positive steady states sequence (u1,n(x),u2,n(x)) such that either limn→∞‖u1,n‖∞=0 or limn→∞‖u2,n‖∞=0. Integrating second equation of (3.1) gives
0=∫Ωu2,n(cpe−γτu1,n−μ2)dx. | (3.2) |
If ‖u1,n(x)‖∞→0 as n→∞, then cpe−γτu1,n(x)−μ2<−μ2/2 for sufficiently large n, which yields u2,n(cpe−γτu1,n−μ2)<0, a contradiction derived. Thus, ‖u2,n(x)‖∞→0 as n→∞ holds. We then assume that (u1,n,u2,n)→(u1,∞,0), as n→∞ where u1,∞≥0. Similarly, we obtain that either u1,∞≡0 or u1,∞≡(r−μ1)/m. Obviously, u1,∞≢0 based on the above argument, thus u1,∞≡(r−μ1)/m and limn→∞cpe−γτu1,n(x)−μ2=μ2(R0−1)>0. This again contradicts (3.2). Hence, we have shown ‖u1(x)‖∞ and ‖u2(x)‖∞ have a positive constant lower bound independent on the solution. Therefore, u1(x) and u2(x) have a uniform positive constant lower bound independent on the solution of (3.1) via Harnack's inequality [26,Lemma 2.2]. This ends the proof.
Theorem 3.2. Suppose R0>1. There exists a constant χ>0 depending on r,μ1,p,c,γ,τ,μ2,g and σ1, such that if min{d1,d2}>χ then model (1.4) admits no positive spatially nonhomogeneous steady states, where σ1 is defined in (2.6).
Proof. Denote the averages of the positive solution (u1,u2) of system (3.1) on Ω by
~u1=∫Ωu1(x)dx|Ω| and ~u2=∫Ωu2(x)dx|Ω|. |
By (2.2), we have u1(x)≤¯u1, where ¯u1=(r−μ1)/m, which implies ~u1≤¯u1. Multiplying the first equation by ce−γτ and adding two equations of (3.1) lead to
−(d1ce−γτΔu1+d2Δu2)=ce−γτ(rg(K,u2)u1−μ1u1−mu21)−μ2u2. |
The integration of both sides for the above equation yields
~u2=ce−γτμ2|Ω|∫Ω(rg(K,u2)u1−μ1u1−mu21)dx≤(r−μ1)¯u1ce−γτμ2:=Mv. |
It is readily seen that ∫Ω(u1−~u1)dx=∫Ω(u2−~u2)dx=0. Note u1 and u2 are bounded by two constants ¯u1>0 and ¯u2:=rc¯u1/min{μ1d2/d1,μ2}>0 by Theorem 3.1. Denote Mf=maxu2∈[0,¯u2]|g′u2(K,u2)| and we then obtain
d1∫Ω|∇(u1−~u1)|2dx=∫Ω(u1−~u1)(rg(K,u2)u1−μ1u1−mu21)dx−∫Ωpu1u2(u1−~u1)dx=∫Ω(u1−~u1)(rg(K,u2)u1−μ1u1−mu21−(rg(K,~u2)~u1−μ1~u1−m~u12))dx+∫Ωp(~u1~u2−u1u2)(u1−~u1)dx≤(r−μ1+(rMf+p)¯u12)∫Ω(u1−~u1)2dx+(rMf+p)¯u12∫Ω(u2−~u2)2dx, |
d2∫Ω|∇(u2−~u2)|2dx=∫Ω(cpe−γτu1u2−μ2u2)(u2−~u2)dx=cpe−γτ∫Ω(u1u2−~u1~u2)(u2−~u2)dx−∫Ωμ2(u2−~u2)2dx≤cpe−γτ∫ΩMv2(u1−~u1)2dx+cpe−γτ(¯u1+Mv2)∫Ω(u2−~u2)2dx. |
Set A1=(r−μ1)+((rMf+p)¯u1+cpe−γτMv)/2, and A2=((rMf+p)¯u1+cpe−γτ(2¯u1+Mv))/2. Then, the above inequalities and Poincarˊe inequality yield that
d1∫Ω|∇(u1−~u1)|2dx+d2∫Ω|∇(u2−~u2)|2dx≤χ∫Ω(|∇(u1−~u1)|2+|∇(u2−~u2)|2)dx, |
with a positive constant χ=max{A1/σ1,A2/σ1} depending on r,f,μ1,p,c,γ,τ,μ2 and σ1. Hence, if χ<min{d1,d2}, then ∇(u1−~u1)=∇(u2−~u2)=0, which implies (u1,u2) is a constant solution.
Select u∗1:=ν as the bifurcation parameter and study nonhomogeneous steady state bifurcating from E∗. Let u2,ν satisfy rg(K,u2)−pu2=μ1+mν, E2=(ν,u2,ν), and (ˆu1,ˆu2)=(u1−ν,u2−u2,ν). Drop ˆ⋅. System (3.1) becomes
H(ν,u1,u2)=(d1Δu1+(u1+ν)(rg(K,u2+u2,ν)−μ1−m(u1+ν)−p(u2+u2,ν))d2Δu2+cpe−γτ(u1+ν)(u2+u2,ν)−μ2(u2+u2,ν))=0, |
for (ν,u1,u2)∈R+×Y with Y={(u1,u2):u1,u2∈H2(Ω),(u1)ν=(u2)ν=0, on ∂Ω}. Calculating Frˊechet derivative of H gives
D(u1,u2)H(ν,0,0)=(d1Δ−mνν(rg′u2(K,u2,ν)−p)cpe−γτu2,νd2Δ). |
Then the characteristic equation follows
ρ2+Pi(ν)ρ+Qi(ν)=0 for i∈N0, | (3.3) |
where
Pi(ν)=mν+(d1+d2)σi, Qi(ν)=d1d2σ2i+d2mνσi−μ2u2,ν(rg′u2(K,u2,ν)−p). |
Obviously, Qi>0 and Pi>0 for all ν∈R+ and i∈N0. Therefore, (3.3) does not have a simple zero eigenvalue. According to [4], we obtain the nonexistence of steady state bifurcation bifurcating at E2.
Theorem 3.3. Model (1.4) admits no positive nonhomogeneous steady states bifurcating from E2.
Next, the stability switches at E2 and existence of periodic solutions of (1.4) bifurcating from E2 are studied. Suppose R0>1, namely, cp(r−μ1)>mμ2 and 0≤τ<τmax:=1γlncp(r−μ1)mμ2 to guarantee the existence of E2.
Recall the stability of E2 for τ=0 is proved and 0 is not the root of (2.11) for τ≥0. So, we only consider eigenvalues cross the imaginary axis to the right which corresponds to the stability changes of E2. Now, we shall consider the positive root of Gn(δ,τ). Clearly, there exists exactly one positive root of Gn(δ,τ)=0 if and only if a0,n<b0,n for n∈N0. More specifically,
(A1): 2mu∗1<u∗2(p−rg′u2(K,u∗2)) |
is the sufficient and necessary condition to ensure G0(δ,τ) has exactly one positive zero. For some integer n≥1, the assumption (A1) is a necessary condition to guarantee Gn(δ,τ) exists positive zeros. Set
Jn={τ:τ∈[0,τmax) satisfies a0,n(τ)<b0,n(τ)}, n∈N0. | (4.1) |
Implicit function theorem implies Gn(δ,τ) has a unique zero
δn(τ)=√([b21,n+2a0,n−a21,n+√(b21,n+2a0,n−a21,n)2−4(a20,n−b20,n)]/2) |
which is a C1 function for τ∈Jn. Hence, iδn(τ) is an eigenvalue of (2.11), and δn(τ) satisfies
sin(δn(τ)τ)=δn(−μ2δ2n+μ2a0,n+b0,na1,n)μ22δ2n+b20,n:=h1,n(τ),cos(δn(τ)τ)=b0,n(δ2n−a0,n)+a1,nμ2δ2nμ22δ2n+b20,n:=h2,n(τ), | (4.2) |
for n∈N0. Let
ϑn(τ)={arccosh2,n(τ), if δ2n<a0,n+b0,na1,n/μ2,2π−arccosh2,n(τ), if δ2n≥a0,n+b0,na1,n/μ2, |
which is the unique solution of sinϑn=h1,n and cosϑn=h2,n and satisfies ϑn(τ)∈(0,2π] for τ∈In.
According to [3,27], we arrive at the next properties.
Lemma 4.1. Suppose that R0>1 and (A1) holds.
(i) There exists a nonnegative integer M1 such that Jn≠∅ for 0≤n≤M1, with JM1⊂JM1−1⊂⋯⊂J1⊂J0, and Jn=∅ for n≥1+M1, where Jn is defined in (4.1).
(ii) Define
Skn(τ)=δn(τ)τ−ϑn(τ)−2kπ for integer 0≤n≤M1, k∈N0, and τ∈Jn. | (4.3) |
Then, S00(0)<0; for 0≤n≤M1 and k∈N0, we have limτ→ˆτ−nSkn(τ)=−(2k+1)π, where ˆτn=supJn; Sk+1n(τ)<Skn(τ) and Skn(τ)>Skn+1(τ).
(iii) For each integer n∈[0,N1] and some k∈N0, Skn(τ) has one positive zero ¯τn∈Jn if and only if (2.11) has a pair of eigenvalues ±iδn(¯τn). Moreover,
Sign(Reλ′(¯τn))=Sign((Skn)′(¯τn)). | (4.4) |
When (Skn)′(¯τn)<0, ±iδn(¯τn) cross the imaginary axis from right to left at τ=¯τn; when (Skn)′(¯τn)>0, ±iδn(¯τn) cross the imaginary axis from left to right.
If supτ∈I0S00≤0, then Skn<0 in Jn holds for any 0≤n≤M1 and k∈N0; or only S00 has a zero with even multiplicity in J0 and Skn<0 for any positive integers n and k. Therefore, E2 is locally asymptotically stable for τ∈[0,τmax). The following assumption ensures Hopf bifurcation may occur at E2.
(A2) supτ∈J0S00(τ)>0 and Skn(τ) has at most two zeros (counting multiplicity) for integer 0≤n≤M1 and k∈N0.
Note supτ∈JnS0n is strictly decreasing in n due to Lemma 4.1. It then follows from (A2), and Skn(0)<S00(0)<0, limτ→ˆτ−nSkn(τ)<0 for any integer 0≤n≤M1 and k∈N0, that we can find two positive integers
M={n∈[0,M1]: supS0n>0 and supS0n+1≤0}≥0, | (4.5) |
and
Kn={j≥1: supSj−1n>0 and supSjn≤0}≥1, for any integer 0≤n≤M. | (4.6) |
Then Skn(τ) admits two simple zeros τkn and τ2Kn−k−1n for k∈[0,Kn−1] and no zeros for k≥Kn. The above analysis, together with Lemma 4.1(ⅲ), yields the next result.
Lemma 4.2. Suppose R0>1 and (A1) and (A2) hold. Let Skn(τ), M and Kn be defined in (4.3), (4.5) and (4.6).
(i) For integer n∈[0,M], there are 2Kn simple zeros τjn (0≤j≤2Kn−1) of Sin(τ) (0≤i≤Kn−1), 0<τ0n<τ1n<τ2n<⋯<τ2Kn−1n<ˆτn, and dSin(τin)/dτ>0 and dSin(τ2Kn−i−1n)/dτ<0 for each 0≤i≤Kn−1.
(ii) If there exist exactly two bifurcation values τjn1=τin2:=τ∗ with n1≠n2 and (n1,j),(n2,i)∈[0,M]×[0,2Kn−1], then the double Hopf bifurcation occurs at E2 when τ=τ∗.
Collect all values τjn with 0≤n≤M and 0≤j≤2Kn−1 in a set. To ensure Hopf bifurcation occurs, remove values which appear more than once. The new set becomes
Σ={τ0,τ1,⋯,τ2L−1}, with τi<τj if i<j and 1≤L≤M∑n=0Kn. | (4.7) |
Lemma 4.1(ⅱ) implies S00(τ) exists two simple zeros τ0<τ2L−1. When τ=τi with 0≤i≤2L−1, the Hopf bifurcation occurs at E2. Moreover, E2 is locally asymptotically stable for τ∈[0,τ0)∪(τ2L−1,τmax) and unstable for τ∈(τ0,τ2L−1). Define
Σ0={τ∈Σ:Sj0(τ)=0 for integer 0≤j≤K0}. | (4.8) |
Theorem 4.3. Suppose R0>1. Let Jn, Skn(τ), Σ and Σ0 be defined in (4.1), (4.3), (4.7) and (4.8), respectively.
(i) E2 is locally asymptotically stable for all τ∈[0,τmax) provided that either J0=∅ or supτ∈J0S00(τ)≤0.
(ii) If (A1) and (A2) hold, then a Hopf bifurcation occurs at E2 when τ∈Σ. E2 is locally asymptotically stable for τ∈[0,τ0)∪(τ2L−1,τmax), and unstable for τ∈(τ0,τ2L−1). Further, for τ∈Σ∖Σ0, the bifurcating periodic solution is spatially nonhomogeneous; for τ∈Σ0, the bifurcating periodic solution is spatially homogeneous.
Next investigate the properties of bifurcating periodic solutions by global Hopf bifurcation theorem [28]. Set y(t)=(y1(t),y2(t))T=(u1(⋅,τt)−u∗1,u2(⋅,τt)−u∗2)T and write (1.4) as
y′(t)=Ay(t)+Z(yt,τ,Q), (t,τ,Q)∈R+×[0,τmax)×R+, yt∈C([−1,0],X2), | (4.9) |
where yt(ν)=y(t+ν) for ν∈[−1,0], A=diag(τd1Δ−τμ1,τd2Δ−τμ2) and
Z(yt)=τ((y1t(0)+u∗1)(rg(K,y2t(0)+u∗2)−m(y1t(0)+u∗1)−p(y2t(0)+u∗2))−μ1u∗1cpe−γτ(y1t(−1)+u∗1)(y2t(−1)+u∗2)−μ2u∗2). |
{Ψ(t)}t≥0 denotes the semigroup yielded by A in Ω, with Neumann boundary condition. Clearly, limt→∞Ψ(t)=0. The solution of (4.9) can be denoted by
y(t)=Ψ(t)y(0)+∫t0Ψ(t−σ)Z(yσ)dσ. | (4.10) |
If y(t) is a a−periodic solution of (4.9), then (4.10) yields
y(t)=∫t−∞Ψ(t−s)Z(yσ)dσ, | (4.11) |
since Ψ(t+na)y(0)→0 as n→∞. Hence we only need to consider (4.11). The integral operator of (4.11) is differentiable, completely continuous, and G-equivariant by [24,Chapter 6.5]. The condition min{d1,d2}>χ ensures (1.4) admits exactly one positive steady state E2. Using a similar argument as in [29,Section 4.2], (H1)–(H4) in [24,Chapter 6.5] hold and we shall study the periodic solution.
Lemma 4.4. Suppose R0>1, then all nonnegative periodic solutions of (4.9) satisfies κ≤u1(x,t),u2(x,t)≤ξ for all (x,t)∈¯Ω×R+, where ξ and κ are defined in Theorems 2.2 and 2.4, respectively.
Lemma 4.4 can be obtained by Theorems 2.2 and 2.4. We further assume
(A3): g(K,u2)−g(K,u∗2)u1−u∗1−mr≤0 for u1∈[κ,r−μ1m] and u2∈[κ,ξ]. |
This technical condition is used to exclude the τ−periodic solutions. Note assumption (A3) holds when K=0. This, together with that (g(K,u2)−g(K,u∗2))/(u1−u∗1) is continuous in K, implies there exists ε>0 such that (A3) holds for 0≤K<ε, that is, (A3) holds for the model (1.4) with weak fear effect.
Lemma 4.5. Assume that R0>1 and (A3) holds, then model (1.4) admits no nontrivial τ−periodic solution.
Proof. Otherwise, let (u1,u2) be the nontrivial τ−periodic solution, that is, (u1(x,t−τ),u2(x,t−τ))=(u1(x,t),u2(x,t)). Thus, we have
∂tu1=d1Δu1+u1(rg(K,u2)−μ1−mu1−pu2), x∈Ω,t>0,∂tu2=d2Δu2+cpe−γτu1u2−μ2u2, x∈Ω,t>0,∂νu1=∂νu2=0,x∈∂Ω,t>0,u1(x,ϑ)=u10(x,ϑ)≥0,u2(x,ϑ)=u20(x,ϑ)≥0,x∈Ω,ϑ∈[−τ,0]. | (4.12) |
Claim
(u1,u2)→E2 as t→∞. |
To see this, establish the Lyapunov functional L1:C(ˉΩ,R+×R+)→R,
L1(ϕ1,ϕ2)=∫Ω(ce−γτ(ϕ1−u∗1lnϕ1)+(ϕ2−u∗2lnϕ2))dx for (ϕ1,ϕ2)∈C(ˉΩ,R+×R+). |
Along the solution of system (4.12), the time derivative of L1(ϕ1,ϕ2) is
dL1dt=∫Ω[−d1μ2|∇u1|2pu21−d2u∗2|∇u2|2u22+r(u1−u∗1)2(g(K,u2)−g(K,u∗2)u1−u∗1−mr)]dx. |
The assumption (A3) ensures dL1/dt≤0 for all (u1,u2)∈C(ˉΩ,R+×R+). The maximal invariant subset of dL1/dt=0 is {E2}. Therefore, E2 attracts all positive solution of (4.12) by LaSalle-Lyapunov invariance principle [21,22] which excludes the nonnegative nontrivial τ−periodic solution.
To obtain the nonexistence of τ−periodic solution for model (1.4), we must use the condition (A3) which is very restrictive. However, in numerical simulations, Lemma 4.5 remains true even (A3) is violated. Thus, we conjecture the nonexistence of τ-periodic solution for (1.4).
In the beginning of this section, when τ=τi with 0≤i≤2L−1, ±iδn(τi) are a pair of eigenvalues of (2.11). Give the next standard notations:
(i) For 0≤i≤2L−1, (E2,τi,2π/(δn(τi)τi)) is an isolated singular point.
(ii) Γ=Cl{(y,τ,Q)∈X2×R+×R+: y is the nontrivial Q-periodic solution of (4.9)} is a closed subset of X2×R+×R+.
(iii) For 0≤i≤2L−1, Ci(E2,τi,Qi) is the connected component of (E2,τi,Qi) in Γ.
(iv) For integer 0≤k≤maxn∈[0,M]Kn−1, let ΣkH={τ∈Σ:Skn(τ)=0 for integer 0≤n≤M}.
We are ready to present a conclusion on the global Hopf branches by a similar manner in [30,Theorem 4.12].
Theorem 4.6. Assume R0>1, min{d1,d2}>χ and (A1)–(A3) hold. Then we have the following results.
(i) The global Hopf branch Ci(E2,τi,Qi) is bounded for τi∈ΣkH with k≥1 and i∈[0,2L−1].
(ii) For any τ∈(minkΣkH,maxkΣkH), model (1.4) possesses at least one periodic solution.
(iii) For τi1∈Σk1H, τi2∈Σk2H with i1,i2∈[0,2L−1] and k1,k2∈[0,maxn∈[0,M]Kn−1], we have Ci1(E2,τi1,Qi1)∩Ci2(E2,τi2,Qi2)=∅ if k1≠k2.
To verify obtained theoretical results, the numerical simulation is presented in this part. We choose the fear effect function as g(K,u2)=e−Ku2 and let
Ω=(0,2π),d1=d2=1,r=8,μ1=0.1,a=0.2,p=1,γ=0.3,μ2=0.2,c=0.1. |
Figure 1 shows the existence and stability of E0, E1 and E2, and Hopf bifurcation curve of model (1.4) in K−τ plane. Above the line τ=τmax, E2 does not exist, E1 is globally asymptotically stable and E0 is unstable; Below the line τ=τmax, there exist three constant steady states E0,E1 and E2. In the region which is bounded by τ=τmax and 2mu∗1+u∗2(rg′u2(K,u∗2)−p)=0, no Hopf bifurcation occurs and E2 is stable. In the region which is bounded by τ=τ0 and τ=τ1, there exist periodic solutions through Hopf bifurcation bifurcating at E2
Fix K=1.07, then by simple calculation, we have τmax≈9.944, τ0≈0.85, τ1≈3.55, J0=[0,5.25], J1=[0,3.05], Jn=∅ for n≥2, supτ∈J0S00>0 and supτ∈J1S01<0. Thus, all Hopf bifurcation values τ0 and τ1 are the all zeros of Sk0(τ) for integer k≥0. We summarize the dynamics of model (1.4) as follows.
(i) For τ∈[τmax,∞), we obtain E1 is globally asymptotically stable and E0 is unstable, see Figure 2(a).
(ii) For τ∈(0,τ0)∪(τ1,τmax), we obtain E2 is locally asymptotically stable, and two constant steady state E0 and E1 are unstable, as shown in Figure 2(b).
(iii) For τ∈(τ0,τ1), we obtain E0,E1 and E2 are unstable, a periodic solution bifurcates from E2, as shown in Figure 2(c). Further, a Hopf bifurcation occurs at E2 when τ=τ0,and τ1.
Next, we explore the global Hopf branches by choosing Ω=(0,4π) and
d1=1,d2=1,r=10,μ1=5,a=0.4,p=1,γ=0.05,μ2=4.75,c=2.5,K=0.4. | (5.1) |
As shown in Figure 3, we collect all zeros of Skn(τ) for nonnegative n,k in set Bi with i=0,1,2,3, namely,
B0={0.06,1.88,3.86,6.14,9.54,11.8,13.36,13.8,13.98,14.04},B1={0.08,1.96,4.04,6.56,12.36,13.01,13.25,13.35},B2={0.15,2.31,4.96,10.16,10.85,11.07}, B3={0.39,6.45}. |
From Theorem 4.3, E2 is locally asymptotically stable when τ∈(0,0.06)∪(14.04,τmax) and unstable when τ∈(0.06,14.04), at least one periodic solution emerges for τ∈(0.06,14.04). Moreover, a spatially homogeneous periodic solution bifurcates from τ∈B0, see Figure 4(a); a spatially nonhomogeneous periodic solution bifurcates from τ∈B1∪B2∪B3, see Figure 4(b).
In model (1.3), the kernel function takes form as G=e−γτf(x−y) by reasonable assumptions and our theoretical results are derived by choosing f(x−y) as Dirac-delta function. Next, we choose
f=e−2|x−y|2∫Ωe−2|x−y|2dy. | (5.2) |
Here, f is the truncated normal distribution. Clearly, ∫Ωf(x−y)dy=1. Let Ω=(0,4π) and the parameter values chosen according to (5.1). As shown in Figure 5, when τ=1.03, a stable nonhomogeneous periodic solution emerges; when τ=1.5, a homogeneous periodic solution emerges; when τ=16, the positive constant steady state is stable. Numerical simulation suggests nonlocal interaction can produce more complex dynamics.
We formulate an age-structured predator-prey model with fear effect. For R0≤1, the global asymptotic stability for predator-free constant steady state is proved via Lyapunov-LaSalle invariance principle. For R0>1, we prove the nonexistence of spatially nonhomogeneous steady states and exclude steady state bifurcation. Finally, we carry out Hopf bifurcation analyses and prove global Hopf branches are bounded.
Our theoretical results are obtained by choosing a special kernel function in model (1.3). However, in numerical results, we explore rich dynamics when the nonlocal interaction is incorporated into the delayed term. The theoretical results concerning the nonlocal model are left as an open problem.
H. Shu was partially supported by the National Natural Science Foundation of China (No. 11971285), the Fundamental Research Funds for the Central Universities (No. GK202201002), the Natural Science Basic Research Program of Shaanxi (No. 2023-JC-JQ-03), and the Youth Innovation Team of Shaanxi Universities. W. Xu was partially supported by a scholarship from the China Scholarship Council while visiting the University of New Brunswick. P. Jiang was partially supported by the National Natural Science Foundation of China (No. 72274119).
The authors declare no conflicts of interest in this paper.
[1] | J. Wang, Y. Su, Artificial intelligence and accounting model reform, Commun. Finance Account., 22 (2017), 41-43. Available from: https://d.wanfangdata.com.cn/periodical/cktx201722011. |
[2] | C. Zhang, Analysis of intelligent improvement of accounting information processing, Commun. Finance Account., 34 (2017), 102-105. Available from: https://d.wanfangdata.com.cn/periodical/cktx201734030. |
[3] | C. Alippi, F. Pessina, M. Roveri, An adaptive system for automatic invoice-documents classification, in IEEE International Conference on Image Processing 2005, 2 (2005), Ⅱ-526. https://doi.org/10.1109/ICIP.2005.1530108 |
[4] |
Y. Zhang, C. Zhang, S. Yu, J. Yang, Classification of bill image based on frame line detection, J. Nanjing Univ. Sci. Technol. (Nat. Sci.), 31 (2007), 409-413. https://doi.org/10.14177/j.cnki.32-1397n.2007.04.022 doi: 10.14177/j.cnki.32-1397n.2007.04.022
![]() |
[5] | F. Y. Bu, Q. G. Hu, Y. Wang, A valid frame line removal algorithm for financial document, Comput. Knowl. Technol., 12 (2016), 148-150. Available from: https://d.wanfangdata.com.cn/periodical/dnzsyjs-itrzyksb201623062. |
[6] | H. Jin, T. Xia, B. Wang, Research on adaptive character segmentation and extraction algorithm, J. Xi'an Univ. Technol., 32 (2016), 399-402+415. Available from: https://xuebao.xaut.edu.cn/__local/2/53/FA/00833B6F761060DF2461CB46334_A0C213DD_120738.pdf?e=.pdf. |
[7] | W. Sun, A study on XBRL based value chain accounting information processing, in 2017 6th International Conference on Industrial Technology and Management (ICITM), (2017), 171-175. https://doi.org/10.1109/ICITM.2017.7917916 |
[8] |
W. Cui, L. Ren, Y. Liu, Invoice number recognition algorithm based on numerical structure characteristics, J. Data Acquis. Process., 32 (2017), 119-125. https://doi.org/10.16337/j.1004-9037.2017.01.014 doi: 10.16337/j.1004-9037.2017.01.014
![]() |
[9] |
H. Ouyang, D. Fan, D. Li, Invoice seal recognition for multi-feature fusion decision making, Comput. Eng. Des., 39 (2018), 2842-2847. https://doi.org/10.16208/j.issn1000-7024.2018.09.026 doi: 10.16208/j.issn1000-7024.2018.09.026
![]() |
[10] | F. Jiang, H. Chen, L. J. Zhang, FCN-biLSTM based VAT invoice recognition and processing, in Edge Computing - EDGE 2018, (2018), 135-143. https://doi.org/10.1007/978-3-319-94340-4_11 |
[11] |
J. Shen, L. Han, Design process optimization and profit calculation module development simulation analysis of financial accounting information system based on particle swarm optimization (PSO), Inf. Syst. e-Bus. Manage., 18 (2020), 809-822. https://doi.org/10.1007/s10257-018-00398-0 doi: 10.1007/s10257-018-00398-0
![]() |
[12] |
Y. Xu, C. Lv, S. Li, P. Xin, S. Ma, H. Zou, et al., Review of development of visual neural computing, Comput. Eng. Appl., 53 (2017), 30-34. https://doi.org/10.3778/j.issn.1002-8331.1709-0474 doi: 10.3778/j.issn.1002-8331.1709-0474
![]() |
[13] |
Q. Wang, P. Lu, Research on application of artificial intelligence in computer network technology, Int. J. Pattern Recognit. Artif. Intell., 33 (2019), 1959015. https://doi.org/10.1142/S0218001419590158 doi: 10.1142/S0218001419590158
![]() |
[14] |
X. Liu, Y. Li, Q. Wang, Multi-view hierarchical bidirectional recurrent neural network for depth video sequence based action recognition, Int. J. Pattern Recognit. Artif. Intell., 32 (2018), 1850033. https://doi.org/10.1142/S0218001418500337 doi: 10.1142/S0218001418500337
![]() |
[15] |
Y. Hou, Q. Wang, Research and improvement of content-based image retrieval framework, Int. J. Pattern Recognit. Artif. Intell., 32 (2018), 1850043. https://doi.org/10.1142/S021800141850043X doi: 10.1142/S021800141850043X
![]() |
[16] |
Q. Peng, G. Ji, L. Xie, S. Zhang, Application of convolutional neural network in vehicle recognition, J. Front. Comput. Sci. Technol., 12 (2018), 282-291. https://doi.org/10.3778/j.issn.1673-9418.1704055 doi: 10.3778/j.issn.1673-9418.1704055
![]() |
[17] | X. Zhao, Z. Sun, M. Xia, Vehicle image recognition method based on local learning, J. Zhejiang Univ. Technol., 45 (2017), 439-444. http://xb.qks.zjut.edu.cn/CN/Y2017/V45/I4/439 |
[18] |
L. Shi, S. Qiang, License plate character recognition based on combined support vector machine. Comput. Eng. Des., 38 (2017), 1619-1623. https://doi.org/10.16208/j.issn1000-7024.2017.06.040 doi: 10.16208/j.issn1000-7024.2017.06.040
![]() |
[19] |
Q. Gao, Y. Liang, Q. Pan, Y. Chen, H. Zhang, The problem existed in face recognition using SVD and its solution, J. Image Graphics, 11 (2006), 1784-1791. https://doi.org/10.11834/jig.2006012312 doi: 10.11834/jig.2006012312
![]() |
[20] |
M. Adil, M. K. Khan, M. Jamjoom, A. Farouk, MHADBOR: AI-enabled administrative distance based opportunistic load balancing scheme for an agriculture internet of things network, IEEE Micro, 42 (2022), 41-50. https://doi.org/10.1109/MM.2021.3112264 doi: 10.1109/MM.2021.3112264
![]() |
[21] |
K. C. Rim, P. K. Kim, H. Ko, K. Bae, T. G. Kwon, Restoration of dimensions for ancient drawing recognition, Electronics, 10 (2021), 2269. https://doi.org/10.3390/electronics10182269 doi: 10.3390/electronics10182269
![]() |
[22] | A. Xing, X. Tao, R. Peng, Thinking about the accounting industry in the era of artificial intelligence, Account. Learn., 160 (2017), 112+114. Available from: http://www.ckxx.cbpt.cnki.net/WKG/WebPublication/paperDigest.aspx?paperID=9d9ee325-d998-420c-9e18-d24984315f65. |
[23] | T. Shi, The impact of the rise of artificial intelligence on the future accounting industry, Mod. Bus., 28 (2017), 122-123. https://doi.org/10.14097/j.cnki.5392/2017.28.053 |
[24] |
M. Ren, Research on the functional transformation of accounting personnel under artificial intelligence, Mod. Bus. Ind., 38 (2017), 68-69. https://doi.org/10.19311/j.cnki.1672-3198.2017.23.030 doi: 10.19311/j.cnki.1672-3198.2017.23.030
![]() |
[25] | R. Sarkar, S. Halder, S. Malakar, N. Das, S. Basu, M. Nasipuri, Text line extraction from handwritten document pages based on line contour estimation, in 2012 Third International Conference on Computing, Communication and Networking Technologies (ICCCNT'12), (2012), 1-8. https://doi.org/10.1109/ICCCNT.2012.6395873 |
[26] | W. Wang, An image skew correction method by using directional white run-length, Sci. Technol. Eng., 12 (2012), 5642-5644. Available from: http://www.stae.com.cn/jsygc/article/abstract/121573?st=article_issue. |
[27] |
J. S. Goldstein, I. S. Reed, L. L. Scharf, A multistage representation of the wiener filter based on orthogonal projections, IEEE Trans. Inf. Theory, 44 (1998), 2943-2959. https://doi.org/10.1109/18.737524 doi: 10.1109/18.737524
![]() |
[28] |
Y. Yu, J. Li, Denoising research based on wiener filter, J. Luoyang Norm. Univ., 36 (2017), 29-32. https://doi.org/10.16594/j.cnki.41-1302/g4.2017.02.008 doi: 10.16594/j.cnki.41-1302/g4.2017.02.008
![]() |
[29] |
S. Palanisamy, B. Thangaraju, O. I. Khalaf, Y. Alotaibi, S. Alghamdi, Design and synthesis of multi-mode bandpass filter for wireless applications, Electronics, 10 (2021), 2853. https://doi.org/10.3390/electronics10222853 doi: 10.3390/electronics10222853
![]() |
[30] |
H. Song, M. Brandt-Pearce, A 2-D discrete-time model of physical impairments in wavelength-division multiplexing systems, J. Lightwave Technol., 30 (2012), 713-726. https://doi.org/10.1109/JLT.2011.2180360 doi: 10.1109/JLT.2011.2180360
![]() |
[31] | L. Wang, Y. Chen, L. Liu, Research on text image layout segmentation algorithm based on projection contour analysis, Digital Technol. Appl., 3 (2017), 164-165. Available from: http://www.cqvip.com/QK/95792B/20173/671781608.html. |
[32] |
J. Schmidhuber, Deep learning in neural networks: An overview, Neural Networks, 61 (2015), 85-117. https://doi.org/10.1016/j.neunet.2014.09.003 doi: 10.1016/j.neunet.2014.09.003
![]() |
[33] | Y. Feng, S. Zeng, Y. Yang, Y. Zhou, B. Pan, Study on the optimization of CNN based on image identification, in 2018 17th International Symposium on Distributed Computing and Applications for Business Engineering and Science (DCABES), (2018), 123-126. https://doi.org/10.1109/DCABES.2018.00041 |
[34] |
A. Krizhevsky, I. Sutskever, G. E. Hinton, Imagenet classification with deep convolutional neural networks, Commun. ACM, 60 (2017), 84-90. https://doi.org/10.1145/3065386 doi: 10.1145/3065386
![]() |
1. | Xiaozhou Feng, Kunyu Li, Haixia Li, Bifurcation and stability analysis of a Leslie–Gower diffusion predator–prey model with prey refuge and Beddington–DeAngelis functional response, 2024, 0170-4214, 10.1002/mma.10470 | |
2. | Sangeeta Kumari, Sidharth Menon, Abhirami K, Dynamical system of quokka population depicting Fennecaphobia by Vulpes vulpes, 2025, 22, 1551-0018, 1342, 10.3934/mbe.2025050 |