Stochastic threshold dynamics of an HIV model with latent cell activation under mean-reverting environmental noise Authors: Zhongwei Cao, Yuchen Jiang, Zhenfeng Shi, Daqing Jiang, Su Wei, Yaxin Zhou Published: 2027 ABSTRACT Motivated by the impact of environmental variability on within-host HIV dynamics, we develop a stochastic HIV model incorporating latent cell activation. Infection and activation rates are modeled as logarithmic mean-reverting Ornstein-Uhlenbeck processes, ensuring positive parameters and biologically realistic fluctuations. We prove the existence and uniqueness of a global positive solution and derive stochastic thresholds characterizing persistence and extinction. If the persistence threshold exceeds one, the model admits a stationary distribution, indicating sustained viral presence; conversely, sufficiently strong environmental noise can drive latent cells, actively infected cells, and free virus to extinction almost surely, even when the deterministic model predicts persistence. Analysis of the associated Fokker-Planck equation provides explicit expressions for the covariance structure near quasi-equilibrium, quantifying stochastic fluctuations. Numerical simulations support the theoretical findings and illustrate how environmental noise reshapes HIV dynamics. Our results highlight the crucial role of stochastic variability in modulating viral persistence and latent reservoir dynamics. Introduction Human immunodeficiency virus (HIV) is a retrovirus that targets the human immune system and has constituted a major global public health challenge since its discovery. Globally, HIV has resulted in approximately 40.4 million deaths and infected about 85.6 million individuals (UNAIDS, 2023). After entering the host, HIV primarily infects CD4+ T lymphocytes. Through reverse transcription, viral RNA is converted into DNA and subsequently integrated into the host genome. The infected cells then exploit the host cellular machinery to produce new virions, ultimately leading to cell destruction (Deeks et al., 2015). As CD4+ T cells are progressively depleted, HIV-infected individuals become increasingly susceptible to opportunistic infections. Consequently, many HIV-related complications, including mortality, are primarily attributable to secondary microbial infections rather than the direct cytopathic effect of the virus itself (Nowak & McMichael, 1995). Mathematical modeling has become an indispensable tool for elucidating the within-host dynamics of infectious diseases. By constructing appropriate mathematical frameworks, it is possible to quantitatively describe viral replication, immune responses, and treatment effects, thereby providing theoretical insights into disease progression and control strategies. Since the early development of HIV research, numerous mathematical models have been proposed to investigate viral dynamics within the human body (Merrill, 1989, 2018, pp. 371–385, pp. 59–75; Tan & Wu, 1998). Classical HIV models typically consist of three compartments: uninfected target cells, infected cells, and free virus particles. For example, Elaiw (Elaiw, 2010) studied the global dynamics of an HIV model incorporating an exposed class and a nonlinear incidence rate. By constructing suitable Lyapunov functions, the global asymptotic stability of the disease equilibria was rigorously established. Despite their success in capturing the fundamental mechanisms of HIV infection, classical models overlook an important biological feature: not all infected cells actively produce virus particles. Experimental studies indicate that a proportion of CD4+ T cells enters a long-lived dormant state following infection, during which viral replication is effectively suppressed (BAGASRA & POMERANTZ, 1993). These cells, commonly referred to as latent infected cells (Pace et al., 2011), form a persistent viral reservoir characterized by slow turnover. Although latent infected cells do not continuously produce free virions, they retain the capacity to rapidly release virus upon activation, particularly after ∗∗the∗∗ interruption of antiretroviral therapy. Moreover, because latent reservoirs evade immune surveillance and therapeutic elimination, complete viral eradication within the host's lifetime remains unattainable. In recent years, increasing attention has been devoted to the role of latent infected cells in HIV dynamics. A variety of mathematical models incorporating latency and activation mechanisms have been developed (Pankavich, 2016; Wang et al., 2017, 2021). For instance, Wang (Wang et al., 2015) proposed an HIV model with latent infected cells and a saturated incidence rate, while Rong (Rong & Perelson, 2009) further investigated antigen-induced activation of latent cells by extending earlier models (Jones & Perelson, 2007a; Kim & Perelson, 2006). These studies demonstrate that latent infection plays a crucial role in viral persistence and treatment outcomes. Recent developments in mathematical modeling of HIV dynamics have increasingly focused on linking within-host processes to population-level outcomes and intervention strategies. For instance, Wei et al. proposed a compartmental community transmission model incorporating HIV detection and monitoring, demonstrating that improved detection and monitoring rates can raise the stability threshold and help control infection spread in a community setting (Wei et al., 2026). Similarly, stochastic transmission models with protection awareness and fluctuating contact behaviors have been analyzed to understand how awareness and therapy can influence long-term epidemic outcomes (Zhai et al., 2023). In addition, models that explicitly incorporate HIV testing and contact tracing mechanisms have shown that early detection at the population level can significantly reduce hidden transmission and influence overall epidemic trajectories (Ouyang et al., 2025). These studies illustrate the importance of integrating detection mechanisms, stochastic effects, and behavioral factors into HIV models, suggesting promising avenues for extending our current within-host framework to multiscale, population-level settings where detection, monitoring, and intervention strategies play a central role. From a mathematical standpoint, incorporating a latent compartment introduces multiple time scales into the system, reflecting the fast dynamics of viral replication and the slow evolution of the latent reservoir. Furthermore, activation of latent infected cells acts as an endogenous source of actively infected cells, even when viral loads are substantially suppressed. Consequently, the qualitative analysis of such models becomes significantly more intricate, and analytical techniques developed for standard three-dimensional HIV models are not directly applicable. In particular, the presence of latency may modify the basic reproduction number, influence the global stability of equilibria, and generate more complex long-term dynamical behaviors. Motivated by the above biological and mathematical considerations, we formulate and analyze a deterministic HIV model that explicitly incorporates latent infected cells. The model describes the interactions among uninfected target cells, latent infected cells, actively infected cells, and free virus particles, thereby providing a tractable framework for investigating the impact of latency on within-host HIV dynamics. The model is given by the following system of ordinary differential equations:(1.1)dT(t)dt=λ−dTT(t)−(1−ϵ)k¯V(t)T(t),dL(t)dt=αL(1−ϵ)k¯V(t)T(t)−(dL+a¯)L(t),dT∗(t)dt=(1−αL)(1−ϵ)k¯V(t)T(t)−δT∗(t)+a¯L(t),dV(t)dt=NδT∗(t)−cV(t).Here, T(t) denotes the population of uninfected target CD4+ T cells, L(t) represents latent infected CD4+ T cells, T∗(t) denotes actively infected CD4+ T cells, and V(t) represents free HIV virus particles. All parameters are assumed to be positive and their biological interpretations and numerical values are summarized in Table 1.Table 1Parameters and values used in the model.Table 1ParameterValueDescriptionReferenceλ104 ml−1day−1Recruitment rate of target cellsCallaway and Perelson (2002)dT0.01 day−1Death rate of target cellsMohri et al. (1998)k¯2.4 × 10−8 ml day−1Infection ratePerelson et al. (1993)ɛ0.85Overall drug efficacysee textαL0.001Fraction of infections leading to latencyJones and Perelson (2007b)dL0.001 day−1Death rate of latent infected cellsCallaway and Perelson (2002)a¯0.1 day−1Activation rate of latent infected cellssee textδ1 day−1Death rate of actively infected cellsMarkowitz et al. (2003)N2000Burst sizeHockett et al. (1999)c23 day−1Clearance rate of free virusRamratnam et al. (1999) Table 1Parameters and values used in the model.Table 1ParameterValueDescriptionReferenceλ104 ml−1day−1Recruitment rate of target cellsCallaway and Perelson (2002)dT0.01 day−1Death rate of target cellsMohri et al. (1998)k¯2.4 × 10−8 ml day−1Infection ratePerelson et al. (1993)ɛ0.85Overall drug efficacysee textαL0.001Fraction of infections leading to latencyJones and Perelson (2007b)dL0.001 day−1Death rate of latent infected cellsCallaway and Perelson (2002)a¯0.1 day−1Activation rate of latent infected cellssee textδ1 day−1Death rate of actively infected cellsMarkowitz et al. (2003)N2000Burst sizeHockett et al. (1999)c23 day−1Clearance rate of free virusRamratnam et al. (1999) Parameters and values used in the model. Using the next-generation matrix method, the basic reproduction number of model (1.1) is given by(1.2)R0=αL(1−ϵ)k¯λa¯NdT(dL+a¯)c+(1−αL)(1−ϵ)k¯λNdTc. The following results describe the threshold dynamics of the model:•If R0 < 1, model (1.1) admits a disease-free equilibrium P0=λdT,0,0,0, which is locally asymptotically stable.•If R0 > 1, the disease-free equilibrium P0 is unstable and model (1.1) admits a unique endemic equilibriumP∗=(Tˆ,Lˆ,Tˆ∗,Vˆ),where(1.3)Tˆ=λdT+(1−ϵ)k¯Vˆ,Tˆ∗=cNδVˆ,Lˆ=αL(1−ϵ)k¯λVˆdT+(1−ϵ)k¯Vˆ(dL+a¯),Vˆ=Nλ(dL+a¯)(1−αL)+a¯αLc(dL+a¯)−dT(1−ϵ)k¯. •If R0 < 1, model (1.1) admits a disease-free equilibrium P0=λdT,0,0,0, which is locally asymptotically stable. If R0 < 1, model (1.1) admits a disease-free equilibrium P0=λdT,0,0,0, which is locally asymptotically stable. •If R0 > 1, the disease-free equilibrium P0 is unstable and model (1.1) admits a unique endemic equilibrium If R0 > 1, the disease-free equilibrium P0 is unstable and model (1.1) admits a unique endemic equilibrium Moreover, the endemic equilibrium P∗ is globally asymptotically stable. The detailed proof is provided in Appendix A. Although the deterministic model captures the essential mechanisms of within-host HIV dynamics, real biological systems are inevitably subject to environmental fluctuations arising from variability in immune responses, treatment adherence, and other external factors. Incorporating stochastic perturbations is therefore necessary for a more realistic and mathematically robust description of disease evolution. Stochastic epidemic models have been extensively studied in recent years (Cheng et al., 2019; Wang et al., 2019). For example, Gao (Gao et al., 2019) showed that environmental noise may significantly alter disease transmission patterns, while Cai (Cai et al., 2018) demonstrated that mean-reverting stochastic perturbations can suppress outbreaks. Qi (Qi & Meng, 2021) further observed that sufficiently strong noise may drive HIV to extinction. These results highlight that stochasticity can fundamentally change the long-term behavior predicted by deterministic models. In the literature, environmental noise is typically introduced either through linear white-noise perturbations (Mao et al., 2002; Meng et al., 2018) or via mean-reverting Ornstein–Uhlenbeck (OU) processes (Shi & Jiang, 2023; Zhou et al., 2022). Compared with white noise, OU processes incorporate mean reversion and temporal correlation, providing a more realistic description of regulated biological parameters while maintaining analytical tractability (Shi & Jiang, 2023). From a mathematical perspective, the use of logarithmic OU processes offers two additional advantages. First, it guarantees parameter positivity without imposing artificial constraints. Second, it transforms multiplicative environmental fluctuations into additive noise in logarithmic coordinates, which facilitates Lyapunov analysis and the derivation of stationary distributions. Although environmental variability may affect all parameters, we restrict stochastic perturbations to two key quantities that dominate HIV dynamics: the infection rate k(t) and the activation rate of latent infected cells a(t). The infection rate directly determines viral transmission efficiency, while the activation rate governs the release of virus from the latent reservoir, a central obstacle in HIV eradication strategies. Introducing stochasticity into these parameters allows us to quantify how environmental fluctuations influence both viral persistence and extinction. To preserve positivity and incorporate mean-reverting fluctuations, we assume that k(t) and a(t) satisfy the logarithmic Ornstein–Uhlenbeck processes:(1.4)dlogk(t)=λ1logk¯−logk(t)dt+σ1dB1(t),dloga(t)=λ2loga¯−loga(t)dt+σ2dB2(t),where B1(t) and B2(t) are independent standard Brownian motions, and λi > 0, σi > 0 (i = 1, 2) denote the mean-reversion rates and noise intensities, respectively. Under this formulation, the resulting system becomes a nonlinear stochastic differential equation with correlated multiplicative effects and multiple time scales. The presence of stochastic activation of latent cells introduces additional analytical challenges, particularly in establishing global positivity, identifying stochastic threshold parameters, and characterizing invariant probability measures. These features distinguish the present model from classical three-dimensional stochastic HIV systems and motivate the theoretical analysis developed in the subsequent sections. Accordingly, the stochastic counterpart of model (1.1) is formulated as follows:(1.5)dT(t)=λ−dTT(t)−(1−ϵ)k(t)V(t)T(t)dt,dL(t)=αL(1−ϵ)k(t)V(t)T(t)−(dL+a(t))L(t)dt,dT∗(t)=(1−αL)(1−ϵ)k(t)V(t)T(t)−δT∗(t)+a(t)L(t)dt,dV(t)=NδT∗(t)−cV(t)dt,dlogk(t)=λ1logk¯−logk(t)dt+σ1dB1(t),dloga(t)=λ2loga¯−loga(t)dt+σ2dB2(t). Compared with existing stochastic HIV models, the principal contributions of this work are summarized as follows:•We go beyond conventional linear white-noise perturbations by modeling environmental fluctuations with logarithmic OU processes. This formulation ensures parameter positivity, incorporates mean reversion and temporal correlation, and provides biologically realistic variability while maintaining analytical tractability.•Using appropriate Lyapunov functions and stochastic analysis, we define two threshold quantities, R0s and R0e, characterizing persistence and extinction, respectively. Specifically, R0s>1 implies the existence of at least one stationary distribution, indicating long-term stochastic persistence, while R0e<1 leads to almost-sure exponential extinction. In the deterministic limit (σ1 = σ2 = 0), these thresholds reduce to R0, demonstrating consistency with the classical basic reproduction number.•Through analysis of the associated five-dimensional Fokker-Planck equation, we derive explicit expressions for the covariance matrix near the quasi-equilibrium point. This provides quantitative insight into the magnitude and correlation structure of stochastic fluctuations.•Extensive numerical simulations verify the theoretical results and illustrate the impact of stochastic perturbations on HIV dynamics. The methodology developed is general and can be extended to other stochastic epidemic models with latent structures. •We go beyond conventional linear white-noise perturbations by modeling environmental fluctuations with logarithmic OU processes. This formulation ensures parameter positivity, incorporates mean reversion and temporal correlation, and provides biologically realistic variability while maintaining analytical tractability. We go beyond conventional linear white-noise perturbations by modeling environmental fluctuations with logarithmic OU processes. This formulation ensures parameter positivity, incorporates mean reversion and temporal correlation, and provides biologically realistic variability while maintaining analytical tractability. •Using appropriate Lyapunov functions and stochastic analysis, we define two threshold quantities, R0s and R0e, characterizing persistence and extinction, respectively. Specifically, R0s>1 implies the existence of at least one stationary distribution, indicating long-term stochastic persistence, while R0e<1 leads to almost-sure exponential extinction. In the deterministic limit (σ1 = σ2 = 0), these thresholds reduce to R0, demonstrating consistency with the classical basic reproduction number. Using appropriate Lyapunov functions and stochastic analysis, we define two threshold quantities, R0s and R0e, characterizing persistence and extinction, respectively. Specifically, R0s>1 implies the existence of at least one stationary distribution, indicating long-term stochastic persistence, while R0e<1 leads to almost-sure exponential extinction. In the deterministic limit (σ1 = σ2 = 0), these thresholds reduce to R0, demonstrating consistency with the classical basic reproduction number. •Through analysis of the associated five-dimensional Fokker-Planck equation, we derive explicit expressions for the covariance matrix near the quasi-equilibrium point. This provides quantitative insight into the magnitude and correlation structure of stochastic fluctuations. Through analysis of the associated five-dimensional Fokker-Planck equation, we derive explicit expressions for the covariance matrix near the quasi-equilibrium point. This provides quantitative insight into the magnitude and correlation structure of stochastic fluctuations. •Extensive numerical simulations verify the theoretical results and illustrate the impact of stochastic perturbations on HIV dynamics. The methodology developed is general and can be extended to other stochastic epidemic models with latent structures. Extensive numerical simulations verify the theoretical results and illustrate the impact of stochastic perturbations on HIV dynamics. The methodology developed is general and can be extended to other stochastic epidemic models with latent structures. The remainder of this paper is organized as follows. Section 2 presents preliminary definitions and auxiliary lemmas. Section 3 establishes the existence and uniqueness of a global positive solution and derives an invariant region for the stochastic model. Section 4 provides sufficient conditions for the existence of a stationary distribution via Lyapunov function techniques. Section 5 derives explicit expressions for the probability density function in a neighborhood of the quasi-equilibrium point. Section 6 establishes sufficient conditions for the exponential extinction of latent infected cells, actively infected cells, and free virus particles. Section 7 presents numerical simulations to illustrate the theoretical findings. Finally, concluding remarks and future research directions are discussed. Preliminaries Mathematical notation Throughout this paper, unless otherwise stated, Ω,F,{Ft}t≥0,P denotes a complete filtered probability space satisfying the usual conditions, namely, the filtration {Ft}t≥0 is right-continuous and F0 contains all P-null sets. The stochastic processes Bi(t) (i = 1, 2) are defined as independent standard Brownian motions on this probability space. For a positive integer d, we defineR+d=x=(x1,…,xd)T∈Rd:xi>0,i=1,…,d. For any matrix A, A−1 and AT denote its inverse (if it exists) and transpose, respectively. We denote by A(k) the submatrix formed by the first k rows and first k columns of A. The n × d zero matrix is denoted by On,d. For real numbers a1, …, an, we use the standard notations(2.1)a1∨a2∨⋯∨an≔max1≤j≤n{aj},a1∧a2∧⋯∧an≔min1≤j≤n{aj}. For a stochastic process X(t), we say that X(t) is positive ifP{X(t)∈R+d,∀t≥0}=1.Unless otherwise specified, all stochastic differential equations in this paper are interpreted in the Itô sense. Necessary lemmas Consider an l-dimensional nonlinear stochastic differential equation:dϒ(t)=f1(ϒ(t))dt+f2(ϒ(t))dB0(t),where B0(t) is a Brownian motion defined on the complete probability space Ω,F,Ftt≥0,P. If f1 and f2 are Borel-measurable functions, according to the literature (Dieu, 2018; Du et al., 2016; Meyn & Tweedie, 1993), Lemma 2.1 illustrates the existence of a stationary distribution of the model (1.5).Lemma 2.1If there exists a d-dimensional bounded closed domain D with a regular boundary, for any initial value, iflim inft→+∞1t∫0P(τ,ϒ(0),D)dτ>0 a.s., where P(τ,ϒ(0),⋅) is the transition probability of ϒ(t). Then model (1.5) admits a solution with the Feller property and at least one invariant probability measure g∗(⋅) on Rd, i.e., a stationary distribution.Lemma 2.2( (Ma et al., 2015)). Let A∈Rn×n be a real matrix and letφA(s)=det(sI−A)=ansn+an−1sn−1+⋯+a1s+a0,an>0,be the characteristic polynomial of A. Then A is a Hurwitz matrix if and only if all the leading principal minorsΔ1,Δ2,…,Δnof the associated Hurwitz matrix are positive, that is,Δk>0,k=1,2,…,n.Here, the Hurwitz matrix is defined asH=an−1an−3an−5⋯0anan−2an−4⋯00an−1an−3⋯00anan−2⋯0⋮⋮⋮⋱⋮000⋯a0,and Δk denotes the determinant of the k × k leading principal submatrix of H.Lemma 2.3( (Zhou et al., 2023)). For equation H02+C1ΣA+ΣAC1T=0, where H0=diag(1,0,0,0,0), and C1 is defined as follows:C1=−e1−e2−e3−e4−e510000010000010000010.If ΣA is a five-dimensional Routh-Hurwitz matrix, i.e.ΔH1=e1>0,ΔH2=e1e31e2>0,ΔH3=e1e3e51e2e40e1e3>0,ΔH4=e1e3e501e2e400e1e3e501e2e4>0,ΔH5=e1e3e5001e2e4000e1e3e5001e2e4000e1e3e5=e5ΔH4>0,then ΣA is a positive definite matrix, and its expression is as follows:ΣA=e2e3e4−e2e5−e4e1e4−e52ΔH40−e3e4−e2e52ΔH40e1e4−e52ΔH40e3e4−e2e52ΔH40−e1e4−e52ΔH40−e3e4−e2e52ΔH40e1e4−e52ΔH40−ΔH22ΔH40−e1e4−e52ΔH40ΔH22ΔH40e1e4−e52ΔH40−ΔH22ΔH40ΔH32e5ΔH4.Lemma 2.4( (Zhou et al., 2021)). For equation H02+C2ΣB+ΣBC2T=0, where H0=diag(1,0,0,0,0), and C2 is defined as follows:C2=−e1−e2−e3−e4−e51000001000001000000Δ4.If e1 > 0, e3 > 0, e4 > 0 and e1e2e3−e32−e12e4>0, then the matrix ΣB is semi-positive definite which followsΣB=e2e3−e1e42e1e2e3−e32−e12e40−e32e1e2e3−e32−e12e4000e32e1e2e3−e32−e12e40−e12e1e2e3−e32−e12e40−e32e1e2e3−e32−e22e40e12e1e2e3−e32−e12e4000−e12e1e2e3−e32−e12e40e1e2−e32e4e1e2e3−e32−e12e4000000.Lemma 2.5( (Zhou et al., 2020)). For equation H12+C3ΣC+ΣCC3T=0, where H1=diag(1,0,0), and C3 is defined as follows.C3=−l¯1−l¯2−l¯3100010.ΣC is a positive definite matrix ifl¯1>0,l¯2>0,l¯3>0 and l¯1l¯2−l¯3>0,which followsΣC=l¯22l¯1l¯2−l¯30−12l¯1l¯2−l¯3012l¯1l¯2−l¯30−12l¯1l¯2−l¯30l¯12l¯3l¯1l¯2−l¯3.Lemma 2.6( (Shi & Jiang, 2024)). For a stochastic equation(2.2)dX(t)=−rX(t)+ρdB(t),where r and ρ are positive constants and B(t) is a standard Brownian motion. Then.(i)(2.3)limt→∞1t∫0tex(s)−1ds≤1+eρ2r−2eρ24r12.(ii)(2.4)limt→∞1t∫0tenx(s)ds=en2ρ24r. If there exists a d-dimensional bounded closed domain D with a regular boundary, for any initial value, iflim inft→+∞1t∫0P(τ,ϒ(0),D)dτ>0 a.s., where P(τ,ϒ(0),⋅) is the transition probability of ϒ(t). Then model (1.5) admits a solution with the Feller property and at least one invariant probability measure g∗(⋅) on Rd, i.e., a stationary distribution. ( (Ma et al., 2015)). Let A∈Rn×n be a real matrix and letφA(s)=det(sI−A)=ansn+an−1sn−1+⋯+a1s+a0,an>0,be the characteristic polynomial of A. Then A is a Hurwitz matrix if and only if all the leading principal minorsΔ1,Δ2,…,Δnof the associated Hurwitz matrix are positive, that is,Δk>0,k=1,2,…,n.Here, the Hurwitz matrix is defined asH=an−1an−3an−5⋯0anan−2an−4⋯00an−1an−3⋯00anan−2⋯0⋮⋮⋮⋱⋮000⋯a0,and Δk denotes the determinant of the k × k leading principal submatrix of H. ( (Zhou et al., 2023)). For equation H02+C1ΣA+ΣAC1T=0, where H0=diag(1,0,0,0,0), and C1 is defined as follows:C1=−e1−e2−e3−e4−e510000010000010000010.If ΣA is a five-dimensional Routh-Hurwitz matrix, i.e.ΔH1=e1>0,ΔH2=e1e31e2>0,ΔH3=e1e3e51e2e40e1e3>0,ΔH4=e1e3e501e2e400e1e3e501e2e4>0,ΔH5=e1e3e5001e2e4000e1e3e5001e2e4000e1e3e5=e5ΔH4>0,then ΣA is a positive definite matrix, and its expression is as follows:ΣA=e2e3e4−e2e5−e4e1e4−e52ΔH40−e3e4−e2e52ΔH40e1e4−e52ΔH40e3e4−e2e52ΔH40−e1e4−e52ΔH40−e3e4−e2e52ΔH40e1e4−e52ΔH40−ΔH22ΔH40−e1e4−e52ΔH40ΔH22ΔH40e1e4−e52ΔH40−ΔH22ΔH40ΔH32e5ΔH4. ( (Zhou et al., 2021)). For equation H02+C2ΣB+ΣBC2T=0, where H0=diag(1,0,0,0,0), and C2 is defined as follows:C2=−e1−e2−e3−e4−e51000001000001000000Δ4.If e1 > 0, e3 > 0, e4 > 0 and e1e2e3−e32−e12e4>0, then the matrix ΣB is semi-positive definite which followsΣB=e2e3−e1e42e1e2e3−e32−e12e40−e32e1e2e3−e32−e12e4000e32e1e2e3−e32−e12e40−e12e1e2e3−e32−e12e40−e32e1e2e3−e32−e22e40e12e1e2e3−e32−e12e4000−e12e1e2e3−e32−e12e40e1e2−e32e4e1e2e3−e32−e12e4000000. ( (Zhou et al., 2020)). For equation H12+C3ΣC+ΣCC3T=0, where H1=diag(1,0,0), and C3 is defined as follows.C3=−l¯1−l¯2−l¯3100010.ΣC is a positive definite matrix ifl¯1>0,l¯2>0,l¯3>0 and l¯1l¯2−l¯3>0,which followsΣC=l¯22l¯1l¯2−l¯30−12l¯1l¯2−l¯3012l¯1l¯2−l¯30−12l¯1l¯2−l¯30l¯12l¯3l¯1l¯2−l¯3. ( (Shi & Jiang, 2024)). For a stochastic equation(2.2)dX(t)=−rX(t)+ρdB(t),where r and ρ are positive constants and B(t) is a standard Brownian motion. Then.(i)(2.3)limt→∞1t∫0tex(s)−1ds≤1+eρ2r−2eρ24r12.(ii)(2.4)limt→∞1t∫0tenx(s)ds=en2ρ24r. (i)(2.3)limt→∞1t∫0tex(s)−1ds≤1+eρ2r−2eρ24r12. (i)(2.3)limt→∞1t∫0tex(s)−1ds≤1+eρ2r−2eρ24r12. (ii)(2.4)limt→∞1t∫0tenx(s)ds=en2ρ24r. (ii)(2.4)limt→∞1t∫0tenx(s)ds=en2ρ24r. Existence and uniqueness of a global solution Depending on the biological meaning represented by T, L, V, T∗, k and a, they should all be non-negative. Therefore, we need to give the existence and uniqueness of the global positive solution.Theorem 3.1For any initial value T(0),L(0),T∗(0),V(0),k(0),a(0)T∈R+6, there exists a unique solution for model (1.5) on t ≥ 0 and the solution T(t),L(t),T∗(t),V(t),k(t),a(t)T will remain in R+6 with probability one.Proof. Since all the parameters of the stochastic model (5) are locally Lipschitz continuous, then model (5) has a unique local solution T(t),L(t),T∗(t),V(t),k(t),a(t)T on 0,τe, where τe is an explosion time. If we prove that τe = ∞ a.s., then the stochastic solution T(t),L(t),T∗(t),V(t),k(t),a(t) is global. Choose a sufficiently large constant n0 ≥ 1, which guarantees that T(0), L(0), T∗(0), V(0), k(0) and a(0) all lie in the interval 1n0,n0. For ∀n ≥ n0, define an explosion time:(3.1)τn=inft∈0,τe:minT(t),L(t),T∗(t),V(t),k(t),a(t)≤1n or maxT(t),L(t),T∗(t),V(t),k(t),a(t)≥n.From the expression of τn, it can be observed that τn increases as n increases and define τ∞ = limn→∞τn. In addition, since τ∞ ≤ τe, if we prove that τ∞ = ∞, then τe = ∞. Next, we prove τ∞ = ∞ with the proof by contradiction. Assuming that τ∞ < ∞, there is a positive integer T and χ ∈ (0, 1) such thatPτ∞≤T>χ.Therefore there is a positive integer n1 ≥ n0, which holdsPτn≤T≥χ,∀n≥n1Define a non-negative C2-functionPT,L,T∗,V,k,a=T−1−lnT+L−1−lnL+T∗−1−lnT∗+1N(V−1−lnV)+k−1−lnk+a−1−lna.Applying Itoˆ’s formula, we get(3.2)LP=λ−dTT−dLL−δT∗−λT+dT+(1−ϵ)kV−αL(1−ϵ)kVTL+dL+a−(1−αL)(1−ϵ)kVTT∗+δ−aLT∗−cVN−δT∗V+cN+kλ1lnk¯−λ1lnk+12σ12−λ1lnk¯−lnk+aλ2lna¯−λ2lna+12σ22−λ2lna¯−lna.In addition, for any t ≤ τn,dT=λ−dTT−(1−ϵ)kVTdt≤λ−dTTdt,anddT+L+T∗=λ−dTT−dLL−δT∗dt≤λ−min{dT,dL,δ}(T+L+T∗)dt,which implies that(3.3)T(t)≤T(0), if T(0)≥λdT,λdT, if T(0)<λdT≤K1,and(3.4)T(t)+L(t)+T∗(t)≤T(0)+L(0)+T∗(0), if T(0)+L(0)+T∗(0)≥λmin{dT,dL,δ},λmin{dT,dL,δ}, if T(0)+L(0)+T∗(0)<λmin{dT,dL,δ}≤K2,whereK1=maxT(0),λdT,K2=maxT(0)+L(0)+T∗(0),λmin{dT,dL,δ}.On the other hand,dV=NδT∗−cVdt≤NδK2−cVdt,one obtains that(3.5)V(t)≤V(0), if V(0)≥NδK2c,NδK2c, if V(0)χ.Therefore there is a positive integer n1 ≥ n0, which holdsPτn≤T≥χ,∀n≥n1Define a non-negative C2-functionPT,L,T∗,V,k,a=T−1−lnT+L−1−lnL+T∗−1−lnT∗+1N(V−1−lnV)+k−1−lnk+a−1−lna.Applying Itoˆ’s formula, we get(3.2)LP=λ−dTT−dLL−δT∗−λT+dT+(1−ϵ)kV−αL(1−ϵ)kVTL+dL+a−(1−αL)(1−ϵ)kVTT∗+δ−aLT∗−cVN−δT∗V+cN+kλ1lnk¯−λ1lnk+12σ12−λ1lnk¯−lnk+aλ2lna¯−λ2lna+12σ22−λ2lna¯−lna.In addition, for any t ≤ τn,dT=λ−dTT−(1−ϵ)kVTdt≤λ−dTTdt,anddT+L+T∗=λ−dTT−dLL−δT∗dt≤λ−min{dT,dL,δ}(T+L+T∗)dt,which implies that(3.3)T(t)≤T(0), if T(0)≥λdT,λdT, if T(0)<λdT≤K1,and(3.4)T(t)+L(t)+T∗(t)≤T(0)+L(0)+T∗(0), if T(0)+L(0)+T∗(0)≥λmin{dT,dL,δ},λmin{dT,dL,δ}, if T(0)+L(0)+T∗(0)<λmin{dT,dL,δ}≤K2,whereK1=maxT(0),λdT,K2=maxT(0)+L(0)+T∗(0),λmin{dT,dL,δ}.On the other hand,dV=NδT∗−cVdt≤NδK2−cVdt,one obtains that(3.5)V(t)≤V(0), if V(0)≥NδK2c,NδK2c, if V(0)1, then the model (1.5) has at least one stationary distribution g∗(⋅) on Γ.Proof. The proof will be divided into the following three steps:1)Constructing a non-negative C2-function WT,L,T∗,V,k,a.2)Constructing a compact set D such that for any T,L,T∗,V,k,a∈Γ∖D, LW≤−1.3)Demonstrating the existence of a stationary distribution.Step 1. (Constructing a non-negative C2-function).LetT¯=λdT,T¯∗=1,V¯=Nδc,L¯=αL(1−ϵ)k¯T¯dL+a¯eσ224λ2,T~=TT¯,T~∗=T∗T¯∗,V~=VV¯,k~=kk¯,a~=aa¯.Applying Itoˆ’s formula to −ln T, −ln L, −ln T∗ and −ln V respectively and combining inequality u − 1 − ln u > 0, it is calculated thatL(−lnT)=−λT+dT+(1−ϵ)kV==−λT¯−λT¯1T~−1+dT+(1−ϵ)kVdTlnT~+(1−ϵ)kV,L(−lnL)=−αL(1−ϵ)kVTL+dL+a=≤−αL(1−ϵ)k¯V¯T¯L¯−αL(1−ϵ)k¯V¯T¯L¯k~V~T~L~−1+dL+a−dL+a¯eσ224λ2(lnk~+lnV~+lnT~−lnL~)+a−a¯eσ224λ2,L(−lnV)=c−NδT∗V=c−NδT¯∗V¯−NδT¯∗V¯T~∗V~−1≤c(lnV~−lnT~∗),L−lnT∗=−1−αL(1−ϵ)kVTT∗+δ−aLT∗=≤≤=−1−αL(1−ϵ)k¯V¯T¯T¯∗−1−αL(1−ϵ)k¯V¯T¯T¯∗k~V~T~T~∗−1+δ−a¯L¯T¯∗−a¯L¯T¯∗a~L~T~∗−1−1−αL(1−ϵ)k¯V¯T¯T¯∗−1−αL(1−ϵ)k¯V¯T¯T¯∗lnk~V~T~T~∗+δ−a¯L¯T¯∗−a¯L¯T¯∗lna~L~T¯∗−δR0s−1−R02(lnk~+lnV~+lnT~−lnT~∗)−R01(lna~+lnL~−lnT~∗)−δR0s−1−R02lnT~−R01lnL~+R0δlnT~∗−R02lnV~−R02lnk~−R01lna~.Thus, define a non-negative C2-function W0 which takes the form:W0=−lnT∗−ς1lnT−ς2lnL−ς3lnV−ς4lnk−ς5lna,where ς1, ς2 and ς3 are positive constants and their values will be given later. Applying Itoˆ’s formula to W0, we derive thatLW0⩽−δR0s−1−R02lnT~−R01lnL~+R0sδlnT~∗−R02lnV~−R02lnk~−R01lna~+ς1dTlnT~+ς1(1−ϵ)kV+ς2a−a¯eσ224λ2−ς2dL+a¯eσ224λ2lnk~+λ1ς4lnk~+λ2ς5lna~−ς2dL+a¯eσ224λ2lnV~−ς2dL+a¯eσ224λ2lnT~+ς2dL+a¯eσ224λ2lnL~+ς3clnV~−ς3clnT~∗=−δR0s−1+R02−ς1dT+ς2dL+a¯eσ224λ2lnT~+−R01+ς2dL+a¯eσ224λ2lnL~+R0sδ−ς3clnT~∗+−R02−ς2dL+a¯eσ224λ2+ς3clnV~−R02+ς2dL+a¯eσ224λ2−λ1ς4lnk~−R01−λ2ς5lna~+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV.Chooseς1=R0sδdT,ς2=R01dL+a¯eσ224λ2,ς3=R0sδc,ς4=δR0sλ1,ς5=R01λ2.Then(4.1)LW0≤−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV.According to the Young inequality(4.2)k≤ℏk2+14ℏ,where ℏ is a positive constant which is small enough. Define a C2-function W1,W1=W0−ς11−ϵ4ℏcV.Applying Itoˆ’s formula to W1,(4.3)LW1=−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV+ς11−ϵ4ℏcNδT∗−cV≤≤=−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)ℏk2+14ℏV+ς11−ϵ4ℏcNδT∗−cV−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)ℏk2V+ς11−ϵNδ4ℏcT∗−δR0s−1+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k¯2eσ12λ1+ς11−ϵNδ4ℏcT∗+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2.According to the randomness of ℏ, choose(4.4)ℏ=δR0s−1cmin{dT,dL,δ}ς1(1−ϵ)ℏNδλk¯2eσ12λ1.Therefore,(4.5)LW1≤−δ2R0s−1+ς11−ϵNδ4ℏcT∗+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2.Next, defineW2=−lnT−lnL−lnV+k−1−lnk+a−1−lna.Then we obtain(4.6)LW2=−λT+dT+(1−ϵ)kV−αL(1−ϵ)kVTL+dL+a−NδT∗V+c+kλ1lnk¯−λ1lnk+12σ12−λ1(lnk¯−lnk)+aλ2lna¯−λ2lna+12σ22−λ2(lna¯−lna).Synthesize the above analysis, define a non-negative C2-function WT,L,T∗,V,k,a which holds the following forms:W=MW1+W2,where M is a positive constant holds(4.7)−Mδ2R0s−1+dT+dL+c+supk∈R+f1(k)+supa∈R+f2(a)≤−2.Combining (4.5) and (4.6),(4.8)LW≤−δM2R0s−1+Mς11−ϵNδ4ℏcT∗+Mς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+Mς2a−a¯eσ224λ2−λT+dT−αL(1−ϵ)kVTL+dL−NδT∗V+c+f1(k)+f2(a)=FT,L,T∗,V,k,a+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2,whereFT,L,T∗,V,k,a=−Mδ2R0s−1+Mς1(1−ϵ)Nδ4ℏcT∗+dT+dL+c−λT−αL(1−ϵ)kVTL−NδT∗V+f1(k)+f2(a).Step 2. (Construction of a compact set).We need to construct a compact set D such that for any T,L,T∗,V,k,a∈Γ∖D≔Dc, FT,L,T∗,V,k,a≤−1. Therefore defineD=T,L,T∗,V,k,a∈Γ∣ϵ≤k≤1ϵ,ϵ≤a≤1ϵ,ϵ≤T,ϵ4≤L,ϵ≤T∗,ϵ2≤V,ϵ is a sufficiently small constant, which satisfies(4.9)−Mδ2R0s−1+dT+dL+c+12λ1lnϵ+supk∈R+g1(k)+supa∈R+f2(a)≤−1,(4.10)−Mδ2R0s−1+dT+dL+c+λ1lnϵ2ϵ+supk∈R+g2(k)+supa∈R+f2(a)≤−1,(4.11)−Mδ2R0s−1+dT+dL+c+12λ2lnϵ+supa∈R+g3(a)+supk∈R+f1(k)≤−1,(4.12)−Mδ2R0s−1+dT+dL+c+λ2lnϵ2ϵ+supa∈R+g4(a)+supk∈R+f1(k)≤−1,(4.13)Mς1(1−ϵ)Nδ4ℏcϵ≤1,(4.14)Mς1(1−ϵ)Nδ4ℏcK2−min{λ,Nδ,αL(1−ϵ)}ϵ≤1,whereg1(k)=f1(k)−12λ1lnk,g2(k)=f1(k)+12λ1klnk,g3(a)=f2(a)−12λ2lna,g4(a)=f2(a)+12λ2alna,then, divide the compact set Dc into eight subsets i.e.Dc=⋃i=18Dic,whereD1c=T,L,T∗,V,k,a∈Γ∣01ϵ,D3c=T,L,T∗,V,k,a∈Γ∣01ϵ,D5c=T,L,T∗,V,k,a∈Γ∣00.Based on the transformation between probability and schematic function, (4.19) can be converted into the following formlimt→+∞inf1t∫0tPτ,T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ),Ddτ≥1G+1>0. a.s. Therefore, according to the Lemma 2.1, the model (1.5) has at least one stationary distribution g∗(⋅) on Γ. This completes the proof. If R0s>1, then the model (1.5) has at least one stationary distribution g∗(⋅) on Γ. Proof. The proof will be divided into the following three steps:1)Constructing a non-negative C2-function WT,L,T∗,V,k,a.2)Constructing a compact set D such that for any T,L,T∗,V,k,a∈Γ∖D, LW≤−1.3)Demonstrating the existence of a stationary distribution.Step 1. (Constructing a non-negative C2-function).LetT¯=λdT,T¯∗=1,V¯=Nδc,L¯=αL(1−ϵ)k¯T¯dL+a¯eσ224λ2,T~=TT¯,T~∗=T∗T¯∗,V~=VV¯,k~=kk¯,a~=aa¯.Applying Itoˆ’s formula to −ln T, −ln L, −ln T∗ and −ln V respectively and combining inequality u − 1 − ln u > 0, it is calculated thatL(−lnT)=−λT+dT+(1−ϵ)kV==−λT¯−λT¯1T~−1+dT+(1−ϵ)kVdTlnT~+(1−ϵ)kV,L(−lnL)=−αL(1−ϵ)kVTL+dL+a=≤−αL(1−ϵ)k¯V¯T¯L¯−αL(1−ϵ)k¯V¯T¯L¯k~V~T~L~−1+dL+a−dL+a¯eσ224λ2(lnk~+lnV~+lnT~−lnL~)+a−a¯eσ224λ2,L(−lnV)=c−NδT∗V=c−NδT¯∗V¯−NδT¯∗V¯T~∗V~−1≤c(lnV~−lnT~∗),L−lnT∗=−1−αL(1−ϵ)kVTT∗+δ−aLT∗=≤≤=−1−αL(1−ϵ)k¯V¯T¯T¯∗−1−αL(1−ϵ)k¯V¯T¯T¯∗k~V~T~T~∗−1+δ−a¯L¯T¯∗−a¯L¯T¯∗a~L~T~∗−1−1−αL(1−ϵ)k¯V¯T¯T¯∗−1−αL(1−ϵ)k¯V¯T¯T¯∗lnk~V~T~T~∗+δ−a¯L¯T¯∗−a¯L¯T¯∗lna~L~T¯∗−δR0s−1−R02(lnk~+lnV~+lnT~−lnT~∗)−R01(lna~+lnL~−lnT~∗)−δR0s−1−R02lnT~−R01lnL~+R0δlnT~∗−R02lnV~−R02lnk~−R01lna~.Thus, define a non-negative C2-function W0 which takes the form:W0=−lnT∗−ς1lnT−ς2lnL−ς3lnV−ς4lnk−ς5lna,where ς1, ς2 and ς3 are positive constants and their values will be given later. Applying Itoˆ’s formula to W0, we derive thatLW0⩽−δR0s−1−R02lnT~−R01lnL~+R0sδlnT~∗−R02lnV~−R02lnk~−R01lna~+ς1dTlnT~+ς1(1−ϵ)kV+ς2a−a¯eσ224λ2−ς2dL+a¯eσ224λ2lnk~+λ1ς4lnk~+λ2ς5lna~−ς2dL+a¯eσ224λ2lnV~−ς2dL+a¯eσ224λ2lnT~+ς2dL+a¯eσ224λ2lnL~+ς3clnV~−ς3clnT~∗=−δR0s−1+R02−ς1dT+ς2dL+a¯eσ224λ2lnT~+−R01+ς2dL+a¯eσ224λ2lnL~+R0sδ−ς3clnT~∗+−R02−ς2dL+a¯eσ224λ2+ς3clnV~−R02+ς2dL+a¯eσ224λ2−λ1ς4lnk~−R01−λ2ς5lna~+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV.Chooseς1=R0sδdT,ς2=R01dL+a¯eσ224λ2,ς3=R0sδc,ς4=δR0sλ1,ς5=R01λ2.Then(4.1)LW0≤−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV.According to the Young inequality(4.2)k≤ℏk2+14ℏ,where ℏ is a positive constant which is small enough. Define a C2-function W1,W1=W0−ς11−ϵ4ℏcV.Applying Itoˆ’s formula to W1,(4.3)LW1=−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV+ς11−ϵ4ℏcNδT∗−cV≤≤=−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)ℏk2+14ℏV+ς11−ϵ4ℏcNδT∗−cV−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)ℏk2V+ς11−ϵNδ4ℏcT∗−δR0s−1+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k¯2eσ12λ1+ς11−ϵNδ4ℏcT∗+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2.According to the randomness of ℏ, choose(4.4)ℏ=δR0s−1cmin{dT,dL,δ}ς1(1−ϵ)ℏNδλk¯2eσ12λ1.Therefore,(4.5)LW1≤−δ2R0s−1+ς11−ϵNδ4ℏcT∗+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2.Next, defineW2=−lnT−lnL−lnV+k−1−lnk+a−1−lna.Then we obtain(4.6)LW2=−λT+dT+(1−ϵ)kV−αL(1−ϵ)kVTL+dL+a−NδT∗V+c+kλ1lnk¯−λ1lnk+12σ12−λ1(lnk¯−lnk)+aλ2lna¯−λ2lna+12σ22−λ2(lna¯−lna).Synthesize the above analysis, define a non-negative C2-function WT,L,T∗,V,k,a which holds the following forms:W=MW1+W2,where M is a positive constant holds(4.7)−Mδ2R0s−1+dT+dL+c+supk∈R+f1(k)+supa∈R+f2(a)≤−2.Combining (4.5) and (4.6),(4.8)LW≤−δM2R0s−1+Mς11−ϵNδ4ℏcT∗+Mς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+Mς2a−a¯eσ224λ2−λT+dT−αL(1−ϵ)kVTL+dL−NδT∗V+c+f1(k)+f2(a)=FT,L,T∗,V,k,a+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2,whereFT,L,T∗,V,k,a=−Mδ2R0s−1+Mς1(1−ϵ)Nδ4ℏcT∗+dT+dL+c−λT−αL(1−ϵ)kVTL−NδT∗V+f1(k)+f2(a).Step 2. (Construction of a compact set).We need to construct a compact set D such that for any T,L,T∗,V,k,a∈Γ∖D≔Dc, FT,L,T∗,V,k,a≤−1. Therefore defineD=T,L,T∗,V,k,a∈Γ∣ϵ≤k≤1ϵ,ϵ≤a≤1ϵ,ϵ≤T,ϵ4≤L,ϵ≤T∗,ϵ2≤V,ϵ is a sufficiently small constant, which satisfies(4.9)−Mδ2R0s−1+dT+dL+c+12λ1lnϵ+supk∈R+g1(k)+supa∈R+f2(a)≤−1,(4.10)−Mδ2R0s−1+dT+dL+c+λ1lnϵ2ϵ+supk∈R+g2(k)+supa∈R+f2(a)≤−1,(4.11)−Mδ2R0s−1+dT+dL+c+12λ2lnϵ+supa∈R+g3(a)+supk∈R+f1(k)≤−1,(4.12)−Mδ2R0s−1+dT+dL+c+λ2lnϵ2ϵ+supa∈R+g4(a)+supk∈R+f1(k)≤−1,(4.13)Mς1(1−ϵ)Nδ4ℏcϵ≤1,(4.14)Mς1(1−ϵ)Nδ4ℏcK2−min{λ,Nδ,αL(1−ϵ)}ϵ≤1,whereg1(k)=f1(k)−12λ1lnk,g2(k)=f1(k)+12λ1klnk,g3(a)=f2(a)−12λ2lna,g4(a)=f2(a)+12λ2alna,then, divide the compact set Dc into eight subsets i.e.Dc=⋃i=18Dic,whereD1c=T,L,T∗,V,k,a∈Γ∣01ϵ,D3c=T,L,T∗,V,k,a∈Γ∣01ϵ,D5c=T,L,T∗,V,k,a∈Γ∣00.Based on the transformation between probability and schematic function, (4.19) can be converted into the following formlimt→+∞inf1t∫0tPτ,T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ),Ddτ≥1G+1>0. a.s. Therefore, according to the Lemma 2.1, the model (1.5) has at least one stationary distribution g∗(⋅) on Γ. This completes the proof. 1)Constructing a non-negative C2-function WT,L,T∗,V,k,a. Constructing a non-negative C2-function WT,L,T∗,V,k,a. 2)Constructing a compact set D such that for any T,L,T∗,V,k,a∈Γ∖D, LW≤−1. Constructing a compact set D such that for any T,L,T∗,V,k,a∈Γ∖D, LW≤−1. 3)Demonstrating the existence of a stationary distribution. Demonstrating the existence of a stationary distribution. Step 1. (Constructing a non-negative C2-function).LetT¯=λdT,T¯∗=1,V¯=Nδc,L¯=αL(1−ϵ)k¯T¯dL+a¯eσ224λ2,T~=TT¯,T~∗=T∗T¯∗,V~=VV¯,k~=kk¯,a~=aa¯.Applying Itoˆ’s formula to −ln T, −ln L, −ln T∗ and −ln V respectively and combining inequality u − 1 − ln u > 0, it is calculated thatL(−lnT)=−λT+dT+(1−ϵ)kV==−λT¯−λT¯1T~−1+dT+(1−ϵ)kVdTlnT~+(1−ϵ)kV,L(−lnL)=−αL(1−ϵ)kVTL+dL+a=≤−αL(1−ϵ)k¯V¯T¯L¯−αL(1−ϵ)k¯V¯T¯L¯k~V~T~L~−1+dL+a−dL+a¯eσ224λ2(lnk~+lnV~+lnT~−lnL~)+a−a¯eσ224λ2,L(−lnV)=c−NδT∗V=c−NδT¯∗V¯−NδT¯∗V¯T~∗V~−1≤c(lnV~−lnT~∗),L−lnT∗=−1−αL(1−ϵ)kVTT∗+δ−aLT∗=≤≤=−1−αL(1−ϵ)k¯V¯T¯T¯∗−1−αL(1−ϵ)k¯V¯T¯T¯∗k~V~T~T~∗−1+δ−a¯L¯T¯∗−a¯L¯T¯∗a~L~T~∗−1−1−αL(1−ϵ)k¯V¯T¯T¯∗−1−αL(1−ϵ)k¯V¯T¯T¯∗lnk~V~T~T~∗+δ−a¯L¯T¯∗−a¯L¯T¯∗lna~L~T¯∗−δR0s−1−R02(lnk~+lnV~+lnT~−lnT~∗)−R01(lna~+lnL~−lnT~∗)−δR0s−1−R02lnT~−R01lnL~+R0δlnT~∗−R02lnV~−R02lnk~−R01lna~.Thus, define a non-negative C2-function W0 which takes the form:W0=−lnT∗−ς1lnT−ς2lnL−ς3lnV−ς4lnk−ς5lna,where ς1, ς2 and ς3 are positive constants and their values will be given later. Applying Itoˆ’s formula to W0, we derive thatLW0⩽−δR0s−1−R02lnT~−R01lnL~+R0sδlnT~∗−R02lnV~−R02lnk~−R01lna~+ς1dTlnT~+ς1(1−ϵ)kV+ς2a−a¯eσ224λ2−ς2dL+a¯eσ224λ2lnk~+λ1ς4lnk~+λ2ς5lna~−ς2dL+a¯eσ224λ2lnV~−ς2dL+a¯eσ224λ2lnT~+ς2dL+a¯eσ224λ2lnL~+ς3clnV~−ς3clnT~∗=−δR0s−1+R02−ς1dT+ς2dL+a¯eσ224λ2lnT~+−R01+ς2dL+a¯eσ224λ2lnL~+R0sδ−ς3clnT~∗+−R02−ς2dL+a¯eσ224λ2+ς3clnV~−R02+ς2dL+a¯eσ224λ2−λ1ς4lnk~−R01−λ2ς5lna~+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV.Chooseς1=R0sδdT,ς2=R01dL+a¯eσ224λ2,ς3=R0sδc,ς4=δR0sλ1,ς5=R01λ2.Then(4.1)LW0≤−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV.According to the Young inequality(4.2)k≤ℏk2+14ℏ,where ℏ is a positive constant which is small enough. Define a C2-function W1,W1=W0−ς11−ϵ4ℏcV.Applying Itoˆ’s formula to W1,(4.3)LW1=−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV+ς11−ϵ4ℏcNδT∗−cV≤≤=−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)ℏk2+14ℏV+ς11−ϵ4ℏcNδT∗−cV−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)ℏk2V+ς11−ϵNδ4ℏcT∗−δR0s−1+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k¯2eσ12λ1+ς11−ϵNδ4ℏcT∗+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2.According to the randomness of ℏ, choose(4.4)ℏ=δR0s−1cmin{dT,dL,δ}ς1(1−ϵ)ℏNδλk¯2eσ12λ1.Therefore,(4.5)LW1≤−δ2R0s−1+ς11−ϵNδ4ℏcT∗+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2.Next, defineW2=−lnT−lnL−lnV+k−1−lnk+a−1−lna.Then we obtain(4.6)LW2=−λT+dT+(1−ϵ)kV−αL(1−ϵ)kVTL+dL+a−NδT∗V+c+kλ1lnk¯−λ1lnk+12σ12−λ1(lnk¯−lnk)+aλ2lna¯−λ2lna+12σ22−λ2(lna¯−lna).Synthesize the above analysis, define a non-negative C2-function WT,L,T∗,V,k,a which holds the following forms:W=MW1+W2,where M is a positive constant holds(4.7)−Mδ2R0s−1+dT+dL+c+supk∈R+f1(k)+supa∈R+f2(a)≤−2.Combining (4.5) and (4.6),(4.8)LW≤−δM2R0s−1+Mς11−ϵNδ4ℏcT∗+Mς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+Mς2a−a¯eσ224λ2−λT+dT−αL(1−ϵ)kVTL+dL−NδT∗V+c+f1(k)+f2(a)=FT,L,T∗,V,k,a+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2,whereFT,L,T∗,V,k,a=−Mδ2R0s−1+Mς1(1−ϵ)Nδ4ℏcT∗+dT+dL+c−λT−αL(1−ϵ)kVTL−NδT∗V+f1(k)+f2(a). Step 1. (Constructing a non-negative C2-function). LetT¯=λdT,T¯∗=1,V¯=Nδc,L¯=αL(1−ϵ)k¯T¯dL+a¯eσ224λ2,T~=TT¯,T~∗=T∗T¯∗,V~=VV¯,k~=kk¯,a~=aa¯.Applying Itoˆ’s formula to −ln T, −ln L, −ln T∗ and −ln V respectively and combining inequality u − 1 − ln u > 0, it is calculated thatL(−lnT)=−λT+dT+(1−ϵ)kV==−λT¯−λT¯1T~−1+dT+(1−ϵ)kVdTlnT~+(1−ϵ)kV,L(−lnL)=−αL(1−ϵ)kVTL+dL+a=≤−αL(1−ϵ)k¯V¯T¯L¯−αL(1−ϵ)k¯V¯T¯L¯k~V~T~L~−1+dL+a−dL+a¯eσ224λ2(lnk~+lnV~+lnT~−lnL~)+a−a¯eσ224λ2,L(−lnV)=c−NδT∗V=c−NδT¯∗V¯−NδT¯∗V¯T~∗V~−1≤c(lnV~−lnT~∗),L−lnT∗=−1−αL(1−ϵ)kVTT∗+δ−aLT∗=≤≤=−1−αL(1−ϵ)k¯V¯T¯T¯∗−1−αL(1−ϵ)k¯V¯T¯T¯∗k~V~T~T~∗−1+δ−a¯L¯T¯∗−a¯L¯T¯∗a~L~T~∗−1−1−αL(1−ϵ)k¯V¯T¯T¯∗−1−αL(1−ϵ)k¯V¯T¯T¯∗lnk~V~T~T~∗+δ−a¯L¯T¯∗−a¯L¯T¯∗lna~L~T¯∗−δR0s−1−R02(lnk~+lnV~+lnT~−lnT~∗)−R01(lna~+lnL~−lnT~∗)−δR0s−1−R02lnT~−R01lnL~+R0δlnT~∗−R02lnV~−R02lnk~−R01lna~.Thus, define a non-negative C2-function W0 which takes the form:W0=−lnT∗−ς1lnT−ς2lnL−ς3lnV−ς4lnk−ς5lna,where ς1, ς2 and ς3 are positive constants and their values will be given later. Applying Itoˆ’s formula to W0, we derive thatLW0⩽−δR0s−1−R02lnT~−R01lnL~+R0sδlnT~∗−R02lnV~−R02lnk~−R01lna~+ς1dTlnT~+ς1(1−ϵ)kV+ς2a−a¯eσ224λ2−ς2dL+a¯eσ224λ2lnk~+λ1ς4lnk~+λ2ς5lna~−ς2dL+a¯eσ224λ2lnV~−ς2dL+a¯eσ224λ2lnT~+ς2dL+a¯eσ224λ2lnL~+ς3clnV~−ς3clnT~∗=−δR0s−1+R02−ς1dT+ς2dL+a¯eσ224λ2lnT~+−R01+ς2dL+a¯eσ224λ2lnL~+R0sδ−ς3clnT~∗+−R02−ς2dL+a¯eσ224λ2+ς3clnV~−R02+ς2dL+a¯eσ224λ2−λ1ς4lnk~−R01−λ2ς5lna~+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV.Chooseς1=R0sδdT,ς2=R01dL+a¯eσ224λ2,ς3=R0sδc,ς4=δR0sλ1,ς5=R01λ2.Then(4.1)LW0≤−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV.According to the Young inequality(4.2)k≤ℏk2+14ℏ,where ℏ is a positive constant which is small enough. Define a C2-function W1,W1=W0−ς11−ϵ4ℏcV.Applying Itoˆ’s formula to W1,(4.3)LW1=−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)kV+ς11−ϵ4ℏcNδT∗−cV≤≤=−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)ℏk2+14ℏV+ς11−ϵ4ℏcNδT∗−cV−δR0s−1+ς2a−a¯eσ224λ2+ς1(1−ϵ)ℏk2V+ς11−ϵNδ4ℏcT∗−δR0s−1+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k¯2eσ12λ1+ς11−ϵNδ4ℏcT∗+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2.According to the randomness of ℏ, choose(4.4)ℏ=δR0s−1cmin{dT,dL,δ}ς1(1−ϵ)ℏNδλk¯2eσ12λ1.Therefore,(4.5)LW1≤−δ2R0s−1+ς11−ϵNδ4ℏcT∗+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2.Next, defineW2=−lnT−lnL−lnV+k−1−lnk+a−1−lna.Then we obtain(4.6)LW2=−λT+dT+(1−ϵ)kV−αL(1−ϵ)kVTL+dL+a−NδT∗V+c+kλ1lnk¯−λ1lnk+12σ12−λ1(lnk¯−lnk)+aλ2lna¯−λ2lna+12σ22−λ2(lna¯−lna).Synthesize the above analysis, define a non-negative C2-function WT,L,T∗,V,k,a which holds the following forms:W=MW1+W2,where M is a positive constant holds(4.7)−Mδ2R0s−1+dT+dL+c+supk∈R+f1(k)+supa∈R+f2(a)≤−2.Combining (4.5) and (4.6),(4.8)LW≤−δM2R0s−1+Mς11−ϵNδ4ℏcT∗+Mς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+Mς2a−a¯eσ224λ2−λT+dT−αL(1−ϵ)kVTL+dL−NδT∗V+c+f1(k)+f2(a)=FT,L,T∗,V,k,a+ς1(1−ϵ)ℏNδλcmin{dT,dL,δ}k2−k¯2eσ12λ1+ς2a−a¯eσ224λ2,whereFT,L,T∗,V,k,a=−Mδ2R0s−1+Mς1(1−ϵ)Nδ4ℏcT∗+dT+dL+c−λT−αL(1−ϵ)kVTL−NδT∗V+f1(k)+f2(a). Step 2. (Construction of a compact set).We need to construct a compact set D such that for any T,L,T∗,V,k,a∈Γ∖D≔Dc, FT,L,T∗,V,k,a≤−1. Therefore defineD=T,L,T∗,V,k,a∈Γ∣ϵ≤k≤1ϵ,ϵ≤a≤1ϵ,ϵ≤T,ϵ4≤L,ϵ≤T∗,ϵ2≤V,ϵ is a sufficiently small constant, which satisfies(4.9)−Mδ2R0s−1+dT+dL+c+12λ1lnϵ+supk∈R+g1(k)+supa∈R+f2(a)≤−1,(4.10)−Mδ2R0s−1+dT+dL+c+λ1lnϵ2ϵ+supk∈R+g2(k)+supa∈R+f2(a)≤−1,(4.11)−Mδ2R0s−1+dT+dL+c+12λ2lnϵ+supa∈R+g3(a)+supk∈R+f1(k)≤−1,(4.12)−Mδ2R0s−1+dT+dL+c+λ2lnϵ2ϵ+supa∈R+g4(a)+supk∈R+f1(k)≤−1,(4.13)Mς1(1−ϵ)Nδ4ℏcϵ≤1,(4.14)Mς1(1−ϵ)Nδ4ℏcK2−min{λ,Nδ,αL(1−ϵ)}ϵ≤1,whereg1(k)=f1(k)−12λ1lnk,g2(k)=f1(k)+12λ1klnk,g3(a)=f2(a)−12λ2lna,g4(a)=f2(a)+12λ2alna,then, divide the compact set Dc into eight subsets i.e.Dc=⋃i=18Dic,whereD1c=T,L,T∗,V,k,a∈Γ∣01ϵ,D3c=T,L,T∗,V,k,a∈Γ∣01ϵ,D5c=T,L,T∗,V,k,a∈Γ∣01ϵ,D3c=T,L,T∗,V,k,a∈Γ∣01ϵ,D5c=T,L,T∗,V,k,a∈Γ∣00.Based on the transformation between probability and schematic function, (4.19) can be converted into the following formlimt→+∞inf1t∫0tPτ,T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ),Ddτ≥1G+1>0. a.s. Therefore, according to the Lemma 2.1, the model (1.5) has at least one stationary distribution g∗(⋅) on Γ. This completes the proof. Step 3. (Existence of a stationary distribution). Taking the time average for both sides of equation (4.8),(4.16)0≤EWT(t),L(t),T∗(t),V(t),k(t),a(t)t==EWT(0),L(0),T∗(0),V(0),k(0),a(0)t+1t∫0tELWT(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)dτ=EWT(0),L(0),T∗(0),V(0),k(0),a(0)t+1t∫0tEFT(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)dτ+Mς1K3(1−ϵ)ℏE1t∫0tk2(τ)dτ−k¯2eσ12λ1+ς2ME1t∫0ta(τ)dτ−a¯eσ224λ2.According to lemma 2.6, let t → +∞, then take the inferior limit on both sides of (4.16), one derives that(4.17)0≤limt→+∞infEWT(0),L(0),T∗(0),V(0),k(0),a(0)t+limt→+∞inf1t∫0tEFT(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)dτ+Mς1K3(1−ϵ)ℏElimt→+∞1t∫0tk2(τ)dτ−k¯2eσ12λ1+ς2MElimt→+∞1t∫0ta(τ)dτ−a¯eσ224λ2=limt→+∞inf1t∫0tEFT(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)dτ.According to equation (4.15), it follows that(4.18)0≤limt→+∞inf1t∫0τEFT(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)dτ=≤limt→+∞inf1t∫0tEFT(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)1T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)∈Ddτ+limt→+∞inf1t∫0tEFT(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)1T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)∈Γ∖Ddτ≤Glimt→+∞inf1t∫0t1T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)∈Ddτ−limt→+∞inf1t∫0t1T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)∈Γ∖Ddτ≤−1+G+1limt→+∞inf1t∫0t1T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)∈Ddτ.Combining (4.17) and (4.18), it can be derived that(4.19)limt→+∞inf1t∫0t1T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ)∈Ddτ≥1G+1>0.Based on the transformation between probability and schematic function, (4.19) can be converted into the following formlimt→+∞inf1t∫0tPτ,T(τ),L(τ),T∗(τ),V(τ),k(τ),a(τ),Ddτ≥1G+1>0. a.s. Therefore, according to the Lemma 2.1, the model (1.5) has at least one stationary distribution g∗(⋅) on Γ. This completes the proof. Probability density function In this section, we mainly focus on the exact expressions of the probability density function near the quasi-equilibrium point. Firstly, transform the model (1.5). Main equivalent transformations of model (1.5) Assuming that the quasi-equilibrium P¨=T¨,L¨,T¨∗,V¨,k¨,a¨. Substituting it into the model (1.5), we haveλ−dTT¨−(1−ϵ)k¯V¨T¨=0,αL(1−ϵ)k¯V¨T¨−(dL+a)L¨=0,(1−αL)(1−ϵ)k¯V¨T¨−δT¨∗+aL¨=0,NδT¨∗−cV¨=0,−λ1lnk¨−lnk¯=0,−λ2lna¨−lna¯=0. By calculation, if R0 > 1, then the solution is unique andT¨=Tˆ,L¨=Lˆ,T∗¨=Tˆ∗,V¨=Vˆ,k¨=k¯,a¨=a¯.Letting X(t)=x1,x2,x3,x4,x5,x6T=T−Tˆ,L−Lˆ,T∗−Tˆ∗,V−Vˆ,lnk−lnk¯,lna−lna¯, model (1.5) can be converted to the form below:dx1=−a11x1−a14x4−a15x5dt,dx2=a21x1−a22x2+a24x4+a25x5−a26x6dt,dx3=a31x1+a32x2−a33x3+a34x4+a35x5+a36x6dt,dx4=a43x3−a44x4dt,dx5=−λ1x5dt+σ1dB1(t),dx6=−λ2x6dt+σ2dB2(t),wherea11=dT+(1−ϵ)k¯Vˆ,a14=(1−ϵ)k¯Tˆ,a15=(1−ϵ)k¯VˆTˆ,a21=αL(1−ϵ)k¯Vˆ,a22=dL+a¯,a24=αL(1−ϵ)k¯Tˆ,a25=αL(1−ϵ)k¯VˆTˆ,a26=a¯Lˆ,a31=(1−αL)(1−ϵ)k¯Vˆ,a32=a¯,a33=δ,a34=(1−αL)(1−ϵ)k¯Tˆ,a35=(1−αL)(1−ϵ)k¯VˆTˆ,a36=a¯Lˆ,a43=Nδ,a44=c.Accordingly,dX(t)=−a1100−a14−a150a21−a220a24a25−a26a31a32−a33a34a35a3600a43−a44000000−λ1000000−λ2X(t)dt+0000000000000000000000000000σ12000000σ22dB(t)=AX(t)dt+H2dB(t),where H = diag(0, 0, 0, 0, σ1, σ2), B(t) = (0, 0, 0, 0, B1(t), B2(t)). Density function of model (1.5) Theorem 5.1If R0s>1, then the stationary distribution g∗(⋅) has a normal density functionΨ(T,L,T∗,V,logk,loga)=(2π)−3|Σ|−12exp−12T−Tˆ,L−Lˆ,T∗−Tˆ∗,V−Vˆ,logk−logk¯,loga−loga¯Σ−1T−Tˆ,L−Lˆ,T∗−Tˆ∗,V−Vˆ,logk−logk¯,loga−loga¯T,where Σ is a positive definite matrix and its specific expressions can be divided into the following cases:•Δ1 = 0, Δ2 ≠ 0–If Δ6 ≠ 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ6 = 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,•Δ1 ≠ 0–If Δ3 ≠ 0, Δ6 ≠ 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ3 ≠ 0, Δ6 = 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,–If Δ3 = 0, Δ6 ≠ 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–Δ3 = 0, Δ6 = 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,•Δ1 = 0, Δ2 = 0–If Δ6 ≠ 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ6 = 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,where Δ1, Δ2, Δ3, Δ6, Σ11(5), Σ12(5), Σ13(5), Σ14(3), Σ21(5) and Σ22(5) will be given in (5.3), (5.6), (5.15), (5.5), (5.8), (5.9) and (5.11) respectively. In addition, ρi, Mi, Ijj and J11 will be listed in the following proof, ∀i = 1, …, 6; j = 1, 2, 3.Proof. The characteristic polynomial of A(4) isφA(4)(ξ)=ξ4+ϱ1ξ3+ϱ2ξ2+ϱ3ξ+ϱ4=0,whereϱ1=a11+a22+a33+a44,ϱ2=a22a33+a22a44+a33a44−a34a43+a11a22+a33+a44,ϱ3=a22a33a44−a24a32a43−a22a34a43+a11a22a33+a22a44+a33a44−a34a43+a14a31a43,ϱ4=a11a22a33a44−a24a32a43−a22a34a43+a14a43a22a31+a14a43a21a32.Through simple calculation, we derive thatϱ1ϱ2ϱ3−ϱ32−ϱ12ϱ4>0.Therefore A(4) is a four-dimensional Hurwitz matrix.φA(ξ)=ξ4+ϱ1ξ3+ϱ2ξ2+ϱ3ξ+ϱ4(ξ+λ1)(ξ+λ2).Therefore, all eigenvalues of A also have a negative real part, A is a six-dimensional Hurwitz matrix. According to the (Liu & Jiang, 2024), the value of Σ is determined by the algebraic equation that follows(5.1)H2+AΣ+ΣAT=0.According to Superposition theorem, (5.1) can be split into the following equations:H12+AΣ1+Σ1AT=0,H22+AΣ2+Σ2AT=0.Next, we will obtain the expression of Σ in two steps.Step 1. Consider the equation(5.2)H12+AΣ1+Σ1AT=0.Let A1=I4I3I2I1AI4I3I2I1−1, whereI1=000010100000010000001000000100000001,I2=1000000100000a25a1510000a35a150100000010000001,I3=100000010000000010001000000100000001,I4=10000001000000100000−b1b310000−b2b3010000001,whereb1=a21−a11αL+a22αL=αLdL+a¯−dT,b2=a31−a111−αL−a32αL+a331−αL=1−αLδ−dT−a¯αL,b3=−a431−αL.Through simple calculation, one derives thatA1=−λ100000−a15−a11−a140000b3−a44+a43b2b30a43000Δ1−a22−a43b1b3−a2600Δ2a32−a33−a43b2b3a3600000λ2,where(5.3)Δ1=a44b1b3−a22b1b3−a43b1b2b32,Δ2=a44b2b3+a32b1b3−a33b2b3−a43b22b32.Based on the uncertainty of the values of Δ1 and Δ2, we discuss from the following three cases:Case 1. If Δ1 = 0, Δ2 ≠ 0, let A2=I5A1I5−1, whereI5=100000010000001000000010000100000001.It is calculated thatA2=−λ100000−a15−a11−a140000b3−a44+a43b2b3a430000Δ2−a33−a43b2b3a320000−a43b1b3−a22000000−λ2.For convenience, let I11 = I5I4I3I2I1, equation (37) can be converted into the following form:(5.4)I11H12I11T+I11AI11−1I11Σ1I11T+I11Σ1I11TI11AI11−1T=0,and we can calculate thatI11Σ1I11T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA2=B1O5,1O5,1T−λ2,B1=−λ10000−a15−a11−a14000b3−a44+a43b2b30a4300Δ2a33−a43b2b3a32000−a43b1b3−a22.Next, define C1=M1B1M1−1, where the standardized transformation matrix M1 isM1=v1v2−v3v4−v50−m1m2−m3m400−a43b1b3Δ2a43b1b3a22+a33+a43b2b3−a43b1b3a32−a222000−a43b1b3−a2200001,wherev1=m1a15,v2=m1a11+m2b3,v3=m2a44−a43b2b3+m3Δ2−m1a14,v4=m2a43+m3a33+a43b2b3−m4a43b1b3,v5=m3a32+m4a22,m1=a43b1Δ2,m2=a43b1b3(a22+a33+a44),m4=a32a43b1b3a22+a33+a43b2b3+a22a43b1b3a32−a222,m3=a432b1b3Δ2+a33+a43b2b3a43b1b3a22+a33+a43b2b3−a43b1b3a43b1b3a32−a222.Then by simple calculationC1=−a1−a2−a3−a4−a510000010000010000010,wherea1=ϱ1+λ1,a2=ϱ2+λ1ϱ1,a3=ϱ3+λ1ϱ2,a4=ϱ4+λ1ϱ3,a5=λ1ϱ4.Therefore equation (5.4) can be converted into the following form:diagσ12,0,0,0,0+M1−1C1M1Σ1(5)+Σ1(5)M1TC1TM1−1T=0.That isdiag1,0,0,0,0+C1Σ11(5)+Σ11(5)C1T=0,whereΣ11(5)=ρ1−2M1Σ1(5)M1Tρ1=a15|m1|σ1.Since the eigenvalues of A have all negative real-parts, then C1 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ11(5) is a positive-definite matrix, and its specific form is as follows(5.5)Σ11(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ12M1−1Σ11(5)M1−1T is also a positive matrix.Case 2. If Δ1 ≠ 0, let A3=I6A1I6−1, whereI6=100000010000001000000100000−Δ2Δ100000001,It is calculated thatA3=−λ100000−a15−a11−a140000b3−a44+a43b2b3a43Δ2Δ1a43000Δ1−a22−a43b1b3Δ2Δ1−a43b1b3−a26000Δ3Δ4a36+a26Δ2Δ100000−λ2,where(5.6)Δ3=a32+a22Δ2Δ1−a33Δ2Δ1−a43b2b3Δ2Δ1+a43b1b3Δ22Δ12,Δ4=−a33−a43b2b3+a43b1b3Δ2Δ1.Based on the uncertainty of the values of Δ3 and Δ4, we discuss from the following two cases:Case 2.1. If Δ1 ≠ 0, Δ3 ≠ 0. For convenience, let I22 = I6I4I3I2I1, equation (37) can be converted into the following form:(5.7)I22H12I22T+I22AI22−1I22Σ1I22T+I22Σ1I22TI22AI22−1T=0,and we can calculate thatI22Σ1I22T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA3=B1AO5,1T−λ2,B2=−λ10000−a15−a11−a14000b3−a44+a43b2b3a43Δ2Δ1a4300Δ1−a22−a43b1b3Δ2Δ1−a43b1b3000Δ3Δ4,whereA=0,0,0,−a26,a36+a26Δ2Δ1T.Next, define C2=M2B2M2−1, where the standardized transformation matrix M2 isM2=−v6−v7v8−v9−v100m5−m6m7m800Δ1Δ3−Δ3a22+a43b1Δ2b3Δ1−Δ4Δ42−Δ3a43b1b3000Δ3Δ400001,wherev6=a15m5,v7=m5a11+m6b3,v8=m6a44−a43b2b3+m7Δ1−m5a14,v9=m6a43Δ2Δ1+m7a22+a43b1Δ2b3Δ1−m8Δ3,v10=m6a43+m7a43b1b3−m8Δ4,m5=Δ1Δ3b3,m6=Δ1Δ3a44−a43b2b3+a22+a43b1Δ2b3Δ1−Δ4,m7=Δ2Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a22+a43b1Δ2b3Δ1+Δ3Δ42−Δ3a43b1b3,m8=Δ1Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a43b1b3+Δ4Δ42−Δ3a43b1b3.Then by simple calculation,C2=−a1−a2−a3−a4−a510000010000010000010.Therefore equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M2−1C2M2Σ1(5)+Σ1(5)M2TC2TM2−1T=0.That isdiag1,0,0,0,0+C2Σ12(5)+Σ12(5)C2T=0,whereΣ12(5)=ρ2−2M2Σ1(5)M2Tρ2=a15|m5|σ1.Since the eigenvalues of A have all negative real-parts, then C2 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ12(5) is a positive-definite matrix, and its specific form is as follows(5.8)Σ12(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ22M2−1Σ12(5)M2−1T is also a positive matrix.Case 2.2. If Δ1 ≠ 0, Δ3 = 0, let C3=M3B3M3−1, where the standardized transformation matrix M3 isM3=−v11−v12v13−v14−v150m9−m10m11m1200Δ1−a22+a43b1Δ2b3Δ1−a43b1b30001000001,wherev11=a15m9,v12=m9a11+m10b3,v13=m10a44−a43b2b3+m11Δ1−m9a14,v14=m10a43Δ2Δ1+m11a22+a43b1Δ2b3Δ1,v15=m10a43+m11a43b1b3−m12Δ4,m9=Δ1b3,m10=Δ1a44−a43b2b3+a22+a43b1Δ2b3Δ1,m11=Δ2a43+a22+a43b1Δ2b3Δ12,m12=Δ1a43+a43b1b3a22+a43b1Δ2b3Δ1−a43b1b3Δ4.Then by simple calculation,C3=−τ1−τ2−τ3−τ4−τ51000001000001000000Δ4,whereτ1=q1+Δ4,τ2=q2+q1Δ4+Δ42,τ3=q3+q2Δ4+q1Δ42+Δ43,τ4=q4+q3Δ4+q2Δ42+q1Δ43+Δ44.Therefore, equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M3−1C3M3Σ1(5)+Σ1(5)M3TC3TM3−1T=0,That is,diag1,0,0,0,0+C3Σ13(5)+Σ13(5)C3T=0,where Σ13(5)=ρ3−2M3Σ1(5)M3T, ρ3=a15m9σ1. According to the Lemma 2.4, Σ13(5) is a semi-positive definite matrix and we compute(5.9)Σ13(5)=τ2τ3−τ1τ42τ0−τ32τ000τ32τ0−τ12τ0−τ32τ0τ12τ000−τ12τ0τ1τ2−τ32τ4τ000000,whereτ=τ1τ2τ3−τ32−τ12τ4.Thus, Σ1(5) is a semi-positive definite matrix.Case 3. If Δ1 = 0, Δ2 = 0. For convenience, let I33 = I4I3I2I1, equation (5.2) can be converted into the following form:(5.10)I33H12I33T+I33AI33−1I33Σ1I33T+I33Σ1I33TI33AI33−1T=0,and we can calculate thatI33Σ1I33T=Σ1(3)O3,3O3,3TO3,3,where Σ1(3) is a three-dimensional symmetric matrix. LetA1=B4BO3,3TC,B4=−λ100−a15−a11−a140b3−a44+a43b2b3,whereB=0000000a430,C=−a22−a43b1b3−a26a32−a33−a43b2b3a3600−λ2.Next, define C4=M4B4M4−1, where the standardized transformation matrix M4 isM4=−a15b3−b3a11+a44−a43b2b3a44−a43b2b32−b3a140b3−a44−a43b2b3001.Then, by simple calculation,C4=−l¯1−l¯2−l¯3100010,wherel¯1=a1−z1−z2,l¯2=a2−(z1+z2)a1+(z1+z2)2−z1z2,l¯3=a3−(z1+z2)a2+(z1+z2)2a1−(z1+z2)3+2z1z2(z1+z2)−a1z1z2,z1=a33+a43b2b3,z2=a22.Therefore, equation (5.10) can be converted into the following form:diagσ12,0,0+M4−1C4M4Σ1(3)+Σ1(3)M4TC4TM4−1T=0.That is,diag1,0,0+C4Σ14(3)+Σ14(3)C4T=0,whereΣ14(3)=ρ4−2M4Σ1(3)M4T,ρ4=a15b3σ1.According to Lemma 2.5, Σ14(3) is a positive-definite matrix and we compute(5.11)Σ14(3)=l¯22l¯1l¯2−l¯30−12l¯1l¯2−l¯3012l¯1l¯2−l¯30−12l¯1l¯2−l¯30l¯12l¯3l¯1l¯2−l¯3.Thus, Σ1(5) is a semi-positive definite matrix.Taking the above several cases together, Σ1 is at least a semi-positive definite matrix. Define a positive constant ν which is the smallest eigenvalue of Σ1(3) i.e.Σ1(3)⩾ν100010001.Therefore, by analysis,Σ1=I33−1Σ1(3)O3,3O3,3TO3,3I33−1T⩾ν1−a25a15−a35a15000−a25a15a252a152+b12b32a25a35a152+b1b2b32b1b300−a35a15a25a35a152+b1b2b32a352a152+b22b32b2b3000b1b3b2b3100000010000000.By simple computation,(5.12)XΣ1XT≥νx1−a25a15x2−a35a15x32+b1b3x2+b2b3x3+x42+x52.Step 2. Consider the equation(5.13)H22+AΣ2+Σ2AT=0.Let A4=(J7J6J5J4J3J2J1)A(J7J6J5J4J3J2J1)−1, whereJ1=000001100000010000001000000100000010,J2=100000010000001000001000000010000001,J3=100000001000010000000100000010000001,J4=100000010000000010000100001000000001,J5=10000001000000100000δ−dLa43100000010000001,J6=100000010000001000000010000100000001,J7=100000010000001000000100000Δ5a1410000001,where(5.14)Δ5=(1−ϵ)k¯Tˆ−a44(δ−dL)a43+a33(δ−dL)a43−(δ−dL)2a432.Through simple calculation, one derives thatA4=−λ200000−a26−a22a24a210a250−a43−c−δ+dL−a43Δ5a14a43000−a14−a110−a15000Δ6−dL(1−ϵ)k¯VˆTˆ−a15Δ5a1400000−λ1,where(5.15)Δ6=(1−ϵ)k¯Vˆ−a11Δ5a14+dLΔ5a14.Based on the uncertainty of the value of Δ6, we discuss from the following two cases:Case 1. If Δ6 ≠ 0, for convenience, let J11 = J7J6J5J4J3J2J1, equation (5.13) can be transformed into the following form:(5.16)J11H22J11T+J11AJ11−1J11Σ2J11T+J11Σ2J11TJ11AJ11−1T=0,and we can calculate thatJ11Σ2J11T=Σ2(5)O5,1O5,1T0,where Σ2(5) is a five-dimensional symmetric matrix. LetA4=B5DO5,1T−λ1,B5=A4(5)=−λ20000−a26−a22a24a2100−a43−c−δ+dL−a43Δ5a14a4300−a14−a110000Δ6−dL,whereD=0,a25,0,−a15,(1−ϵ)k¯VˆTˆ−a15Δ5a14T.Next, define C5=M5B5M5−1, where the standardized transformation matrix M5 isM5=−v16−v17−v18−v19−v200m13m14m15−m1600−a14Δ6−Δ6a11+dLdL2000Δ6−dL00001,wherev16=a26m13,v17=m13a22+a43m14,v18=m14(c+δ−dL)+m15a14−m13a24,v19=m14a43Δ5a14+m15a11+m16Δ6−m13a21,v20=m14a43+m16dL,m13=a14Δ6a43,m14=a14Δ6(a11+c+δ),m15=a43Δ5Δ6+a11Δ6(a11+dL)+dL2Δ6,m16=a14Δ6a43+dL3.Then by simple calculation,C5=−q1−q2−q3−q4−q510000010000010000010,whereq1=ϱ1+λ2,q2=ϱ2+λ2ϱ1,q3=ϱ3+λ2ϱ2,q4=ϱ4+λ2ϱ3,q5=λ2ϱ4.Therefore equation (5.16) can be converted into the following form:diagσ22,0,0,0,0+M5−1C5M5Σ2(5)+Σ2(5)M5TC5TM5−1T=0.That isdiag1,0,0,0,0+C5Σ21(5)+Σ21(5)C5T=0,whereΣ21(5)=ρ5−2M5Σ2(5)M5Tρ5=a26|m13|σ2.Since the eigenvalues of A have all negative real-parts, then C5 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ21(5) is a positive-definite matrix, and its specific form is as follows(5.17)Σ21(5)=q2q3q4−q2q5−q4q1q4−q52D40−q3q4−q2q52D40q1q4−q52D40q3q4−q2q52D40−q1q4−q52D40−q3q4−q2q52D40q1q4−q52D40−D22D40−q1q4−q52D40D22D40q1q4−q52D40−D22D40D32q5D4.Therefore, Σ2(5)=ρ52M5−1Σ21(5)M5−1T is also a positive matrix.Case 2. If Δ6 = 0, define C6=M6B5M6−1, where the standardized transformation matrix M6 isM6=−v21−v22−v23−v24v250m17m18m19−m2000−a14−a1100001000001,wherev21=m17a26,v22=m17a22+m18a43,v23=m18(c+δ−dL)+m19a14−m17a24,v24=m18a43Δ5a14+m19a11−m17a21,v25=m18a43+m20dL,m17=a14a43,m18=a14(c+δ+a11−dL),m19=a43Δ5+a112,m20=a14a43.Then by simple calculation,C6=−w1−w2−w3−w4−w51000001000001000000−dL,wherew1=q1−dL,w2=q2−q1dL+dL2,w3=q3−q2dL+q1dL2−dL3,w4=q4−q3dL+q2dL2−q1dL3+dL4.Therefore, equation (5.16) can be converted into the following form:diagσ22,0,0,0,0+M6−1C6M6Σ2(5)+Σ2(5)M6TC6TM6−1T=0,That is,diag1,0,0,0,0+C6Σ22(5)+Σ22(5)C6T=0.where Σ22(5)=ρ6−2M6Σ2(5)M6T, ρ6=a26m9σ2. According to the Lemma 2.4, Σ22(5) is a semi-positive definite matrix and we compute(5.18)Σ22(5)=w2w3−w1w42w0−w32w000w32w0−w12w0−w32w0w12w000−w12w0w1w2−w32w4w000000,wherew=w1w2w3−w32−w12w4.Thus, Σ2(5) is a semi-positive definite matrix.Taking the above several cases together, Σ2 is at least a semi-positive definite matrix. Define a positive constant ϑ which is the smallest eigenvalue of Σ2(5) i.e.Σ2(5)⩾ϑ1000001000001000001000000.Therefore, by analysis,Σ2=J11−1Σ2(5)O5,1O5,1T0J11−1T⩾ϑ10−Δ5a1400001−1000−Δ5a14−1δ−dL2a432+Δ52a142+1dL−δa430000dL−δa43100000000000001.By simple computation,(5.19)XΣ2XT≥ϑx1−Δ5a14x32+x2−x32+δ−dLa43x3−x42+x62.Combining equations (5.12) and (5.19), we can obtain(5.20)XΣXT⩾κx1−a25a15x2−a35a15x32+b1b3x2+b2b3x3+x42+x52+x1−Δ5a14x32+x2−x32+δ−dLa43x3−x42+x62,where κ = ν ∧ϑ. Only if X=(0,0,0,0,0,0)T, XΣXT=0. Then Σ is a positive-definite matrix. To sum up, we obtain the exact expression of the probability density function around the quasi-equilibrium P∗. This completes the proof. If R0s>1, then the stationary distribution g∗(⋅) has a normal density functionΨ(T,L,T∗,V,logk,loga)=(2π)−3|Σ|−12exp−12T−Tˆ,L−Lˆ,T∗−Tˆ∗,V−Vˆ,logk−logk¯,loga−loga¯Σ−1T−Tˆ,L−Lˆ,T∗−Tˆ∗,V−Vˆ,logk−logk¯,loga−loga¯T,where Σ is a positive definite matrix and its specific expressions can be divided into the following cases:•Δ1 = 0, Δ2 ≠ 0–If Δ6 ≠ 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ6 = 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,•Δ1 ≠ 0–If Δ3 ≠ 0, Δ6 ≠ 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ3 ≠ 0, Δ6 = 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,–If Δ3 = 0, Δ6 ≠ 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–Δ3 = 0, Δ6 = 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,•Δ1 = 0, Δ2 = 0–If Δ6 ≠ 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ6 = 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,where Δ1, Δ2, Δ3, Δ6, Σ11(5), Σ12(5), Σ13(5), Σ14(3), Σ21(5) and Σ22(5) will be given in (5.3), (5.6), (5.15), (5.5), (5.8), (5.9) and (5.11) respectively. In addition, ρi, Mi, Ijj and J11 will be listed in the following proof, ∀i = 1, …, 6; j = 1, 2, 3. •Δ1 = 0, Δ2 ≠ 0–If Δ6 ≠ 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ6 = 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, Δ1 = 0, Δ2 ≠ 0–If Δ6 ≠ 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ6 = 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, –If Δ6 ≠ 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T, If Δ6 ≠ 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T, –If Δ6 = 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, If Δ6 = 0,Σ=I11−1ρ12M1−1Σ11(5)M1−1TO5,1O5,1T0I11−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, •Δ1 ≠ 0–If Δ3 ≠ 0, Δ6 ≠ 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ3 ≠ 0, Δ6 = 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,–If Δ3 = 0, Δ6 ≠ 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–Δ3 = 0, Δ6 = 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, Δ1 ≠ 0–If Δ3 ≠ 0, Δ6 ≠ 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ3 ≠ 0, Δ6 = 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,–If Δ3 = 0, Δ6 ≠ 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–Δ3 = 0, Δ6 = 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, –If Δ3 ≠ 0, Δ6 ≠ 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T, If Δ3 ≠ 0, Δ6 ≠ 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T, –If Δ3 ≠ 0, Δ6 = 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, If Δ3 ≠ 0, Δ6 = 0,Σ=I22−1ρ22M2−1Σ12(5)M2−1TO5,1O5,1T0I22−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, –If Δ3 = 0, Δ6 ≠ 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T, If Δ3 = 0, Δ6 ≠ 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T, –Δ3 = 0, Δ6 = 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, Δ3 = 0, Δ6 = 0,Σ=I33−1ρ32M3−1Σ13(5)M3−1TO5,1O5,1T0I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T, •Δ1 = 0, Δ2 = 0–If Δ6 ≠ 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ6 = 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,where Δ1, Δ2, Δ3, Δ6, Σ11(5), Σ12(5), Σ13(5), Σ14(3), Σ21(5) and Σ22(5) will be given in (5.3), (5.6), (5.15), (5.5), (5.8), (5.9) and (5.11) respectively. In addition, ρi, Mi, Ijj and J11 will be listed in the following proof, ∀i = 1, …, 6; j = 1, 2, 3. Δ1 = 0, Δ2 = 0–If Δ6 ≠ 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T,–If Δ6 = 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,where Δ1, Δ2, Δ3, Δ6, Σ11(5), Σ12(5), Σ13(5), Σ14(3), Σ21(5) and Σ22(5) will be given in (5.3), (5.6), (5.15), (5.5), (5.8), (5.9) and (5.11) respectively. In addition, ρi, Mi, Ijj and J11 will be listed in the following proof, ∀i = 1, …, 6; j = 1, 2, 3. –If Δ6 ≠ 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T, If Δ6 ≠ 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ52M5−1Σ21(5)M5−1TO5,1O5,1T0J11−1T, –If Δ6 = 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,where Δ1, Δ2, Δ3, Δ6, Σ11(5), Σ12(5), Σ13(5), Σ14(3), Σ21(5) and Σ22(5) will be given in (5.3), (5.6), (5.15), (5.5), (5.8), (5.9) and (5.11) respectively. In addition, ρi, Mi, Ijj and J11 will be listed in the following proof, ∀i = 1, …, 6; j = 1, 2, 3. If Δ6 = 0,Σ=I33−1ρ42M4−1Σ14(3)M4−1TO3,3O3,3TO3,3I33−1T+J11−1ρ62M6−1Σ22(5)M6−1TO5,1O5,1T0J11−1T,where Δ1, Δ2, Δ3, Δ6, Σ11(5), Σ12(5), Σ13(5), Σ14(3), Σ21(5) and Σ22(5) will be given in (5.3), (5.6), (5.15), (5.5), (5.8), (5.9) and (5.11) respectively. In addition, ρi, Mi, Ijj and J11 will be listed in the following proof, ∀i = 1, …, 6; j = 1, 2, 3. Proof. The characteristic polynomial of A(4) isφA(4)(ξ)=ξ4+ϱ1ξ3+ϱ2ξ2+ϱ3ξ+ϱ4=0,whereϱ1=a11+a22+a33+a44,ϱ2=a22a33+a22a44+a33a44−a34a43+a11a22+a33+a44,ϱ3=a22a33a44−a24a32a43−a22a34a43+a11a22a33+a22a44+a33a44−a34a43+a14a31a43,ϱ4=a11a22a33a44−a24a32a43−a22a34a43+a14a43a22a31+a14a43a21a32.Through simple calculation, we derive thatϱ1ϱ2ϱ3−ϱ32−ϱ12ϱ4>0.Therefore A(4) is a four-dimensional Hurwitz matrix.φA(ξ)=ξ4+ϱ1ξ3+ϱ2ξ2+ϱ3ξ+ϱ4(ξ+λ1)(ξ+λ2).Therefore, all eigenvalues of A also have a negative real part, A is a six-dimensional Hurwitz matrix. According to the (Liu & Jiang, 2024), the value of Σ is determined by the algebraic equation that follows(5.1)H2+AΣ+ΣAT=0.According to Superposition theorem, (5.1) can be split into the following equations:H12+AΣ1+Σ1AT=0,H22+AΣ2+Σ2AT=0.Next, we will obtain the expression of Σ in two steps.Step 1. Consider the equation(5.2)H12+AΣ1+Σ1AT=0.Let A1=I4I3I2I1AI4I3I2I1−1, whereI1=000010100000010000001000000100000001,I2=1000000100000a25a1510000a35a150100000010000001,I3=100000010000000010001000000100000001,I4=10000001000000100000−b1b310000−b2b3010000001,whereb1=a21−a11αL+a22αL=αLdL+a¯−dT,b2=a31−a111−αL−a32αL+a331−αL=1−αLδ−dT−a¯αL,b3=−a431−αL.Through simple calculation, one derives thatA1=−λ100000−a15−a11−a140000b3−a44+a43b2b30a43000Δ1−a22−a43b1b3−a2600Δ2a32−a33−a43b2b3a3600000λ2,where(5.3)Δ1=a44b1b3−a22b1b3−a43b1b2b32,Δ2=a44b2b3+a32b1b3−a33b2b3−a43b22b32.Based on the uncertainty of the values of Δ1 and Δ2, we discuss from the following three cases:Case 1. If Δ1 = 0, Δ2 ≠ 0, let A2=I5A1I5−1, whereI5=100000010000001000000010000100000001.It is calculated thatA2=−λ100000−a15−a11−a140000b3−a44+a43b2b3a430000Δ2−a33−a43b2b3a320000−a43b1b3−a22000000−λ2.For convenience, let I11 = I5I4I3I2I1, equation (37) can be converted into the following form:(5.4)I11H12I11T+I11AI11−1I11Σ1I11T+I11Σ1I11TI11AI11−1T=0,and we can calculate thatI11Σ1I11T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA2=B1O5,1O5,1T−λ2,B1=−λ10000−a15−a11−a14000b3−a44+a43b2b30a4300Δ2a33−a43b2b3a32000−a43b1b3−a22.Next, define C1=M1B1M1−1, where the standardized transformation matrix M1 isM1=v1v2−v3v4−v50−m1m2−m3m400−a43b1b3Δ2a43b1b3a22+a33+a43b2b3−a43b1b3a32−a222000−a43b1b3−a2200001,wherev1=m1a15,v2=m1a11+m2b3,v3=m2a44−a43b2b3+m3Δ2−m1a14,v4=m2a43+m3a33+a43b2b3−m4a43b1b3,v5=m3a32+m4a22,m1=a43b1Δ2,m2=a43b1b3(a22+a33+a44),m4=a32a43b1b3a22+a33+a43b2b3+a22a43b1b3a32−a222,m3=a432b1b3Δ2+a33+a43b2b3a43b1b3a22+a33+a43b2b3−a43b1b3a43b1b3a32−a222.Then by simple calculationC1=−a1−a2−a3−a4−a510000010000010000010,wherea1=ϱ1+λ1,a2=ϱ2+λ1ϱ1,a3=ϱ3+λ1ϱ2,a4=ϱ4+λ1ϱ3,a5=λ1ϱ4.Therefore equation (5.4) can be converted into the following form:diagσ12,0,0,0,0+M1−1C1M1Σ1(5)+Σ1(5)M1TC1TM1−1T=0.That isdiag1,0,0,0,0+C1Σ11(5)+Σ11(5)C1T=0,whereΣ11(5)=ρ1−2M1Σ1(5)M1Tρ1=a15|m1|σ1.Since the eigenvalues of A have all negative real-parts, then C1 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ11(5) is a positive-definite matrix, and its specific form is as follows(5.5)Σ11(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ12M1−1Σ11(5)M1−1T is also a positive matrix.Case 2. If Δ1 ≠ 0, let A3=I6A1I6−1, whereI6=100000010000001000000100000−Δ2Δ100000001,It is calculated thatA3=−λ100000−a15−a11−a140000b3−a44+a43b2b3a43Δ2Δ1a43000Δ1−a22−a43b1b3Δ2Δ1−a43b1b3−a26000Δ3Δ4a36+a26Δ2Δ100000−λ2,where(5.6)Δ3=a32+a22Δ2Δ1−a33Δ2Δ1−a43b2b3Δ2Δ1+a43b1b3Δ22Δ12,Δ4=−a33−a43b2b3+a43b1b3Δ2Δ1.Based on the uncertainty of the values of Δ3 and Δ4, we discuss from the following two cases:Case 2.1. If Δ1 ≠ 0, Δ3 ≠ 0. For convenience, let I22 = I6I4I3I2I1, equation (37) can be converted into the following form:(5.7)I22H12I22T+I22AI22−1I22Σ1I22T+I22Σ1I22TI22AI22−1T=0,and we can calculate thatI22Σ1I22T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA3=B1AO5,1T−λ2,B2=−λ10000−a15−a11−a14000b3−a44+a43b2b3a43Δ2Δ1a4300Δ1−a22−a43b1b3Δ2Δ1−a43b1b3000Δ3Δ4,whereA=0,0,0,−a26,a36+a26Δ2Δ1T.Next, define C2=M2B2M2−1, where the standardized transformation matrix M2 isM2=−v6−v7v8−v9−v100m5−m6m7m800Δ1Δ3−Δ3a22+a43b1Δ2b3Δ1−Δ4Δ42−Δ3a43b1b3000Δ3Δ400001,wherev6=a15m5,v7=m5a11+m6b3,v8=m6a44−a43b2b3+m7Δ1−m5a14,v9=m6a43Δ2Δ1+m7a22+a43b1Δ2b3Δ1−m8Δ3,v10=m6a43+m7a43b1b3−m8Δ4,m5=Δ1Δ3b3,m6=Δ1Δ3a44−a43b2b3+a22+a43b1Δ2b3Δ1−Δ4,m7=Δ2Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a22+a43b1Δ2b3Δ1+Δ3Δ42−Δ3a43b1b3,m8=Δ1Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a43b1b3+Δ4Δ42−Δ3a43b1b3.Then by simple calculation,C2=−a1−a2−a3−a4−a510000010000010000010.Therefore equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M2−1C2M2Σ1(5)+Σ1(5)M2TC2TM2−1T=0.That isdiag1,0,0,0,0+C2Σ12(5)+Σ12(5)C2T=0,whereΣ12(5)=ρ2−2M2Σ1(5)M2Tρ2=a15|m5|σ1.Since the eigenvalues of A have all negative real-parts, then C2 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ12(5) is a positive-definite matrix, and its specific form is as follows(5.8)Σ12(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ22M2−1Σ12(5)M2−1T is also a positive matrix.Case 2.2. If Δ1 ≠ 0, Δ3 = 0, let C3=M3B3M3−1, where the standardized transformation matrix M3 isM3=−v11−v12v13−v14−v150m9−m10m11m1200Δ1−a22+a43b1Δ2b3Δ1−a43b1b30001000001,wherev11=a15m9,v12=m9a11+m10b3,v13=m10a44−a43b2b3+m11Δ1−m9a14,v14=m10a43Δ2Δ1+m11a22+a43b1Δ2b3Δ1,v15=m10a43+m11a43b1b3−m12Δ4,m9=Δ1b3,m10=Δ1a44−a43b2b3+a22+a43b1Δ2b3Δ1,m11=Δ2a43+a22+a43b1Δ2b3Δ12,m12=Δ1a43+a43b1b3a22+a43b1Δ2b3Δ1−a43b1b3Δ4.Then by simple calculation,C3=−τ1−τ2−τ3−τ4−τ51000001000001000000Δ4,whereτ1=q1+Δ4,τ2=q2+q1Δ4+Δ42,τ3=q3+q2Δ4+q1Δ42+Δ43,τ4=q4+q3Δ4+q2Δ42+q1Δ43+Δ44.Therefore, equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M3−1C3M3Σ1(5)+Σ1(5)M3TC3TM3−1T=0,That is,diag1,0,0,0,0+C3Σ13(5)+Σ13(5)C3T=0,where Σ13(5)=ρ3−2M3Σ1(5)M3T, ρ3=a15m9σ1. According to the Lemma 2.4, Σ13(5) is a semi-positive definite matrix and we compute(5.9)Σ13(5)=τ2τ3−τ1τ42τ0−τ32τ000τ32τ0−τ12τ0−τ32τ0τ12τ000−τ12τ0τ1τ2−τ32τ4τ000000,whereτ=τ1τ2τ3−τ32−τ12τ4.Thus, Σ1(5) is a semi-positive definite matrix.Case 3. If Δ1 = 0, Δ2 = 0. For convenience, let I33 = I4I3I2I1, equation (5.2) can be converted into the following form:(5.10)I33H12I33T+I33AI33−1I33Σ1I33T+I33Σ1I33TI33AI33−1T=0,and we can calculate thatI33Σ1I33T=Σ1(3)O3,3O3,3TO3,3,where Σ1(3) is a three-dimensional symmetric matrix. LetA1=B4BO3,3TC,B4=−λ100−a15−a11−a140b3−a44+a43b2b3,whereB=0000000a430,C=−a22−a43b1b3−a26a32−a33−a43b2b3a3600−λ2.Next, define C4=M4B4M4−1, where the standardized transformation matrix M4 isM4=−a15b3−b3a11+a44−a43b2b3a44−a43b2b32−b3a140b3−a44−a43b2b3001.Then, by simple calculation,C4=−l¯1−l¯2−l¯3100010,wherel¯1=a1−z1−z2,l¯2=a2−(z1+z2)a1+(z1+z2)2−z1z2,l¯3=a3−(z1+z2)a2+(z1+z2)2a1−(z1+z2)3+2z1z2(z1+z2)−a1z1z2,z1=a33+a43b2b3,z2=a22.Therefore, equation (5.10) can be converted into the following form:diagσ12,0,0+M4−1C4M4Σ1(3)+Σ1(3)M4TC4TM4−1T=0.That is,diag1,0,0+C4Σ14(3)+Σ14(3)C4T=0,whereΣ14(3)=ρ4−2M4Σ1(3)M4T,ρ4=a15b3σ1.According to Lemma 2.5, Σ14(3) is a positive-definite matrix and we compute(5.11)Σ14(3)=l¯22l¯1l¯2−l¯30−12l¯1l¯2−l¯3012l¯1l¯2−l¯30−12l¯1l¯2−l¯30l¯12l¯3l¯1l¯2−l¯3.Thus, Σ1(5) is a semi-positive definite matrix.Taking the above several cases together, Σ1 is at least a semi-positive definite matrix. Define a positive constant ν which is the smallest eigenvalue of Σ1(3) i.e.Σ1(3)⩾ν100010001.Therefore, by analysis,Σ1=I33−1Σ1(3)O3,3O3,3TO3,3I33−1T⩾ν1−a25a15−a35a15000−a25a15a252a152+b12b32a25a35a152+b1b2b32b1b300−a35a15a25a35a152+b1b2b32a352a152+b22b32b2b3000b1b3b2b3100000010000000.By simple computation,(5.12)XΣ1XT≥νx1−a25a15x2−a35a15x32+b1b3x2+b2b3x3+x42+x52.Step 2. Consider the equation(5.13)H22+AΣ2+Σ2AT=0.Let A4=(J7J6J5J4J3J2J1)A(J7J6J5J4J3J2J1)−1, whereJ1=000001100000010000001000000100000010,J2=100000010000001000001000000010000001,J3=100000001000010000000100000010000001,J4=100000010000000010000100001000000001,J5=10000001000000100000δ−dLa43100000010000001,J6=100000010000001000000010000100000001,J7=100000010000001000000100000Δ5a1410000001,where(5.14)Δ5=(1−ϵ)k¯Tˆ−a44(δ−dL)a43+a33(δ−dL)a43−(δ−dL)2a432.Through simple calculation, one derives thatA4=−λ200000−a26−a22a24a210a250−a43−c−δ+dL−a43Δ5a14a43000−a14−a110−a15000Δ6−dL(1−ϵ)k¯VˆTˆ−a15Δ5a1400000−λ1,where(5.15)Δ6=(1−ϵ)k¯Vˆ−a11Δ5a14+dLΔ5a14.Based on the uncertainty of the value of Δ6, we discuss from the following two cases:Case 1. If Δ6 ≠ 0, for convenience, let J11 = J7J6J5J4J3J2J1, equation (5.13) can be transformed into the following form:(5.16)J11H22J11T+J11AJ11−1J11Σ2J11T+J11Σ2J11TJ11AJ11−1T=0,and we can calculate thatJ11Σ2J11T=Σ2(5)O5,1O5,1T0,where Σ2(5) is a five-dimensional symmetric matrix. LetA4=B5DO5,1T−λ1,B5=A4(5)=−λ20000−a26−a22a24a2100−a43−c−δ+dL−a43Δ5a14a4300−a14−a110000Δ6−dL,whereD=0,a25,0,−a15,(1−ϵ)k¯VˆTˆ−a15Δ5a14T.Next, define C5=M5B5M5−1, where the standardized transformation matrix M5 isM5=−v16−v17−v18−v19−v200m13m14m15−m1600−a14Δ6−Δ6a11+dLdL2000Δ6−dL00001,wherev16=a26m13,v17=m13a22+a43m14,v18=m14(c+δ−dL)+m15a14−m13a24,v19=m14a43Δ5a14+m15a11+m16Δ6−m13a21,v20=m14a43+m16dL,m13=a14Δ6a43,m14=a14Δ6(a11+c+δ),m15=a43Δ5Δ6+a11Δ6(a11+dL)+dL2Δ6,m16=a14Δ6a43+dL3.Then by simple calculation,C5=−q1−q2−q3−q4−q510000010000010000010,whereq1=ϱ1+λ2,q2=ϱ2+λ2ϱ1,q3=ϱ3+λ2ϱ2,q4=ϱ4+λ2ϱ3,q5=λ2ϱ4.Therefore equation (5.16) can be converted into the following form:diagσ22,0,0,0,0+M5−1C5M5Σ2(5)+Σ2(5)M5TC5TM5−1T=0.That isdiag1,0,0,0,0+C5Σ21(5)+Σ21(5)C5T=0,whereΣ21(5)=ρ5−2M5Σ2(5)M5Tρ5=a26|m13|σ2.Since the eigenvalues of A have all negative real-parts, then C5 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ21(5) is a positive-definite matrix, and its specific form is as follows(5.17)Σ21(5)=q2q3q4−q2q5−q4q1q4−q52D40−q3q4−q2q52D40q1q4−q52D40q3q4−q2q52D40−q1q4−q52D40−q3q4−q2q52D40q1q4−q52D40−D22D40−q1q4−q52D40D22D40q1q4−q52D40−D22D40D32q5D4.Therefore, Σ2(5)=ρ52M5−1Σ21(5)M5−1T is also a positive matrix.Case 2. If Δ6 = 0, define C6=M6B5M6−1, where the standardized transformation matrix M6 isM6=−v21−v22−v23−v24v250m17m18m19−m2000−a14−a1100001000001,wherev21=m17a26,v22=m17a22+m18a43,v23=m18(c+δ−dL)+m19a14−m17a24,v24=m18a43Δ5a14+m19a11−m17a21,v25=m18a43+m20dL,m17=a14a43,m18=a14(c+δ+a11−dL),m19=a43Δ5+a112,m20=a14a43.Then by simple calculation,C6=−w1−w2−w3−w4−w51000001000001000000−dL,wherew1=q1−dL,w2=q2−q1dL+dL2,w3=q3−q2dL+q1dL2−dL3,w4=q4−q3dL+q2dL2−q1dL3+dL4.Therefore, equation (5.16) can be converted into the following form:diagσ22,0,0,0,0+M6−1C6M6Σ2(5)+Σ2(5)M6TC6TM6−1T=0,That is,diag1,0,0,0,0+C6Σ22(5)+Σ22(5)C6T=0.where Σ22(5)=ρ6−2M6Σ2(5)M6T, ρ6=a26m9σ2. According to the Lemma 2.4, Σ22(5) is a semi-positive definite matrix and we compute(5.18)Σ22(5)=w2w3−w1w42w0−w32w000w32w0−w12w0−w32w0w12w000−w12w0w1w2−w32w4w000000,wherew=w1w2w3−w32−w12w4.Thus, Σ2(5) is a semi-positive definite matrix. Step 1. Consider the equation Step 1. Consider the equation Case 1. If Δ1 = 0, Δ2 ≠ 0, let A2=I5A1I5−1, whereI5=100000010000001000000010000100000001.It is calculated thatA2=−λ100000−a15−a11−a140000b3−a44+a43b2b3a430000Δ2−a33−a43b2b3a320000−a43b1b3−a22000000−λ2.For convenience, let I11 = I5I4I3I2I1, equation (37) can be converted into the following form:(5.4)I11H12I11T+I11AI11−1I11Σ1I11T+I11Σ1I11TI11AI11−1T=0,and we can calculate thatI11Σ1I11T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA2=B1O5,1O5,1T−λ2,B1=−λ10000−a15−a11−a14000b3−a44+a43b2b30a4300Δ2a33−a43b2b3a32000−a43b1b3−a22.Next, define C1=M1B1M1−1, where the standardized transformation matrix M1 isM1=v1v2−v3v4−v50−m1m2−m3m400−a43b1b3Δ2a43b1b3a22+a33+a43b2b3−a43b1b3a32−a222000−a43b1b3−a2200001,wherev1=m1a15,v2=m1a11+m2b3,v3=m2a44−a43b2b3+m3Δ2−m1a14,v4=m2a43+m3a33+a43b2b3−m4a43b1b3,v5=m3a32+m4a22,m1=a43b1Δ2,m2=a43b1b3(a22+a33+a44),m4=a32a43b1b3a22+a33+a43b2b3+a22a43b1b3a32−a222,m3=a432b1b3Δ2+a33+a43b2b3a43b1b3a22+a33+a43b2b3−a43b1b3a43b1b3a32−a222.Then by simple calculationC1=−a1−a2−a3−a4−a510000010000010000010,wherea1=ϱ1+λ1,a2=ϱ2+λ1ϱ1,a3=ϱ3+λ1ϱ2,a4=ϱ4+λ1ϱ3,a5=λ1ϱ4.Therefore equation (5.4) can be converted into the following form:diagσ12,0,0,0,0+M1−1C1M1Σ1(5)+Σ1(5)M1TC1TM1−1T=0.That isdiag1,0,0,0,0+C1Σ11(5)+Σ11(5)C1T=0,whereΣ11(5)=ρ1−2M1Σ1(5)M1Tρ1=a15|m1|σ1.Since the eigenvalues of A have all negative real-parts, then C1 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ11(5) is a positive-definite matrix, and its specific form is as follows(5.5)Σ11(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ12M1−1Σ11(5)M1−1T is also a positive matrix. Case 1. If Δ1 = 0, Δ2 ≠ 0, let A2=I5A1I5−1, whereI5=100000010000001000000010000100000001.It is calculated thatA2=−λ100000−a15−a11−a140000b3−a44+a43b2b3a430000Δ2−a33−a43b2b3a320000−a43b1b3−a22000000−λ2.For convenience, let I11 = I5I4I3I2I1, equation (37) can be converted into the following form:(5.4)I11H12I11T+I11AI11−1I11Σ1I11T+I11Σ1I11TI11AI11−1T=0,and we can calculate thatI11Σ1I11T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA2=B1O5,1O5,1T−λ2,B1=−λ10000−a15−a11−a14000b3−a44+a43b2b30a4300Δ2a33−a43b2b3a32000−a43b1b3−a22.Next, define C1=M1B1M1−1, where the standardized transformation matrix M1 isM1=v1v2−v3v4−v50−m1m2−m3m400−a43b1b3Δ2a43b1b3a22+a33+a43b2b3−a43b1b3a32−a222000−a43b1b3−a2200001,wherev1=m1a15,v2=m1a11+m2b3,v3=m2a44−a43b2b3+m3Δ2−m1a14,v4=m2a43+m3a33+a43b2b3−m4a43b1b3,v5=m3a32+m4a22,m1=a43b1Δ2,m2=a43b1b3(a22+a33+a44),m4=a32a43b1b3a22+a33+a43b2b3+a22a43b1b3a32−a222,m3=a432b1b3Δ2+a33+a43b2b3a43b1b3a22+a33+a43b2b3−a43b1b3a43b1b3a32−a222.Then by simple calculationC1=−a1−a2−a3−a4−a510000010000010000010,wherea1=ϱ1+λ1,a2=ϱ2+λ1ϱ1,a3=ϱ3+λ1ϱ2,a4=ϱ4+λ1ϱ3,a5=λ1ϱ4.Therefore equation (5.4) can be converted into the following form:diagσ12,0,0,0,0+M1−1C1M1Σ1(5)+Σ1(5)M1TC1TM1−1T=0.That isdiag1,0,0,0,0+C1Σ11(5)+Σ11(5)C1T=0,whereΣ11(5)=ρ1−2M1Σ1(5)M1Tρ1=a15|m1|σ1.Since the eigenvalues of A have all negative real-parts, then C1 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ11(5) is a positive-definite matrix, and its specific form is as follows(5.5)Σ11(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ12M1−1Σ11(5)M1−1T is also a positive matrix. Case 2. If Δ1 ≠ 0, let A3=I6A1I6−1, whereI6=100000010000001000000100000−Δ2Δ100000001,It is calculated thatA3=−λ100000−a15−a11−a140000b3−a44+a43b2b3a43Δ2Δ1a43000Δ1−a22−a43b1b3Δ2Δ1−a43b1b3−a26000Δ3Δ4a36+a26Δ2Δ100000−λ2,where(5.6)Δ3=a32+a22Δ2Δ1−a33Δ2Δ1−a43b2b3Δ2Δ1+a43b1b3Δ22Δ12,Δ4=−a33−a43b2b3+a43b1b3Δ2Δ1.Based on the uncertainty of the values of Δ3 and Δ4, we discuss from the following two cases:Case 2.1. If Δ1 ≠ 0, Δ3 ≠ 0. For convenience, let I22 = I6I4I3I2I1, equation (37) can be converted into the following form:(5.7)I22H12I22T+I22AI22−1I22Σ1I22T+I22Σ1I22TI22AI22−1T=0,and we can calculate thatI22Σ1I22T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA3=B1AO5,1T−λ2,B2=−λ10000−a15−a11−a14000b3−a44+a43b2b3a43Δ2Δ1a4300Δ1−a22−a43b1b3Δ2Δ1−a43b1b3000Δ3Δ4,whereA=0,0,0,−a26,a36+a26Δ2Δ1T.Next, define C2=M2B2M2−1, where the standardized transformation matrix M2 isM2=−v6−v7v8−v9−v100m5−m6m7m800Δ1Δ3−Δ3a22+a43b1Δ2b3Δ1−Δ4Δ42−Δ3a43b1b3000Δ3Δ400001,wherev6=a15m5,v7=m5a11+m6b3,v8=m6a44−a43b2b3+m7Δ1−m5a14,v9=m6a43Δ2Δ1+m7a22+a43b1Δ2b3Δ1−m8Δ3,v10=m6a43+m7a43b1b3−m8Δ4,m5=Δ1Δ3b3,m6=Δ1Δ3a44−a43b2b3+a22+a43b1Δ2b3Δ1−Δ4,m7=Δ2Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a22+a43b1Δ2b3Δ1+Δ3Δ42−Δ3a43b1b3,m8=Δ1Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a43b1b3+Δ4Δ42−Δ3a43b1b3.Then by simple calculation,C2=−a1−a2−a3−a4−a510000010000010000010.Therefore equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M2−1C2M2Σ1(5)+Σ1(5)M2TC2TM2−1T=0.That isdiag1,0,0,0,0+C2Σ12(5)+Σ12(5)C2T=0,whereΣ12(5)=ρ2−2M2Σ1(5)M2Tρ2=a15|m5|σ1.Since the eigenvalues of A have all negative real-parts, then C2 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ12(5) is a positive-definite matrix, and its specific form is as follows(5.8)Σ12(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ22M2−1Σ12(5)M2−1T is also a positive matrix.Case 2.2. If Δ1 ≠ 0, Δ3 = 0, let C3=M3B3M3−1, where the standardized transformation matrix M3 isM3=−v11−v12v13−v14−v150m9−m10m11m1200Δ1−a22+a43b1Δ2b3Δ1−a43b1b30001000001,wherev11=a15m9,v12=m9a11+m10b3,v13=m10a44−a43b2b3+m11Δ1−m9a14,v14=m10a43Δ2Δ1+m11a22+a43b1Δ2b3Δ1,v15=m10a43+m11a43b1b3−m12Δ4,m9=Δ1b3,m10=Δ1a44−a43b2b3+a22+a43b1Δ2b3Δ1,m11=Δ2a43+a22+a43b1Δ2b3Δ12,m12=Δ1a43+a43b1b3a22+a43b1Δ2b3Δ1−a43b1b3Δ4.Then by simple calculation,C3=−τ1−τ2−τ3−τ4−τ51000001000001000000Δ4,whereτ1=q1+Δ4,τ2=q2+q1Δ4+Δ42,τ3=q3+q2Δ4+q1Δ42+Δ43,τ4=q4+q3Δ4+q2Δ42+q1Δ43+Δ44.Therefore, equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M3−1C3M3Σ1(5)+Σ1(5)M3TC3TM3−1T=0,That is,diag1,0,0,0,0+C3Σ13(5)+Σ13(5)C3T=0,where Σ13(5)=ρ3−2M3Σ1(5)M3T, ρ3=a15m9σ1. According to the Lemma 2.4, Σ13(5) is a semi-positive definite matrix and we compute(5.9)Σ13(5)=τ2τ3−τ1τ42τ0−τ32τ000τ32τ0−τ12τ0−τ32τ0τ12τ000−τ12τ0τ1τ2−τ32τ4τ000000,whereτ=τ1τ2τ3−τ32−τ12τ4.Thus, Σ1(5) is a semi-positive definite matrix. Case 2. If Δ1 ≠ 0, let A3=I6A1I6−1, whereI6=100000010000001000000100000−Δ2Δ100000001,It is calculated thatA3=−λ100000−a15−a11−a140000b3−a44+a43b2b3a43Δ2Δ1a43000Δ1−a22−a43b1b3Δ2Δ1−a43b1b3−a26000Δ3Δ4a36+a26Δ2Δ100000−λ2,where(5.6)Δ3=a32+a22Δ2Δ1−a33Δ2Δ1−a43b2b3Δ2Δ1+a43b1b3Δ22Δ12,Δ4=−a33−a43b2b3+a43b1b3Δ2Δ1.Based on the uncertainty of the values of Δ3 and Δ4, we discuss from the following two cases:Case 2.1. If Δ1 ≠ 0, Δ3 ≠ 0. For convenience, let I22 = I6I4I3I2I1, equation (37) can be converted into the following form:(5.7)I22H12I22T+I22AI22−1I22Σ1I22T+I22Σ1I22TI22AI22−1T=0,and we can calculate thatI22Σ1I22T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA3=B1AO5,1T−λ2,B2=−λ10000−a15−a11−a14000b3−a44+a43b2b3a43Δ2Δ1a4300Δ1−a22−a43b1b3Δ2Δ1−a43b1b3000Δ3Δ4,whereA=0,0,0,−a26,a36+a26Δ2Δ1T.Next, define C2=M2B2M2−1, where the standardized transformation matrix M2 isM2=−v6−v7v8−v9−v100m5−m6m7m800Δ1Δ3−Δ3a22+a43b1Δ2b3Δ1−Δ4Δ42−Δ3a43b1b3000Δ3Δ400001,wherev6=a15m5,v7=m5a11+m6b3,v8=m6a44−a43b2b3+m7Δ1−m5a14,v9=m6a43Δ2Δ1+m7a22+a43b1Δ2b3Δ1−m8Δ3,v10=m6a43+m7a43b1b3−m8Δ4,m5=Δ1Δ3b3,m6=Δ1Δ3a44−a43b2b3+a22+a43b1Δ2b3Δ1−Δ4,m7=Δ2Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a22+a43b1Δ2b3Δ1+Δ3Δ42−Δ3a43b1b3,m8=Δ1Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a43b1b3+Δ4Δ42−Δ3a43b1b3.Then by simple calculation,C2=−a1−a2−a3−a4−a510000010000010000010.Therefore equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M2−1C2M2Σ1(5)+Σ1(5)M2TC2TM2−1T=0.That isdiag1,0,0,0,0+C2Σ12(5)+Σ12(5)C2T=0,whereΣ12(5)=ρ2−2M2Σ1(5)M2Tρ2=a15|m5|σ1.Since the eigenvalues of A have all negative real-parts, then C2 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ12(5) is a positive-definite matrix, and its specific form is as follows(5.8)Σ12(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ22M2−1Σ12(5)M2−1T is also a positive matrix.Case 2.2. If Δ1 ≠ 0, Δ3 = 0, let C3=M3B3M3−1, where the standardized transformation matrix M3 isM3=−v11−v12v13−v14−v150m9−m10m11m1200Δ1−a22+a43b1Δ2b3Δ1−a43b1b30001000001,wherev11=a15m9,v12=m9a11+m10b3,v13=m10a44−a43b2b3+m11Δ1−m9a14,v14=m10a43Δ2Δ1+m11a22+a43b1Δ2b3Δ1,v15=m10a43+m11a43b1b3−m12Δ4,m9=Δ1b3,m10=Δ1a44−a43b2b3+a22+a43b1Δ2b3Δ1,m11=Δ2a43+a22+a43b1Δ2b3Δ12,m12=Δ1a43+a43b1b3a22+a43b1Δ2b3Δ1−a43b1b3Δ4.Then by simple calculation,C3=−τ1−τ2−τ3−τ4−τ51000001000001000000Δ4,whereτ1=q1+Δ4,τ2=q2+q1Δ4+Δ42,τ3=q3+q2Δ4+q1Δ42+Δ43,τ4=q4+q3Δ4+q2Δ42+q1Δ43+Δ44.Therefore, equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M3−1C3M3Σ1(5)+Σ1(5)M3TC3TM3−1T=0,That is,diag1,0,0,0,0+C3Σ13(5)+Σ13(5)C3T=0,where Σ13(5)=ρ3−2M3Σ1(5)M3T, ρ3=a15m9σ1. According to the Lemma 2.4, Σ13(5) is a semi-positive definite matrix and we compute(5.9)Σ13(5)=τ2τ3−τ1τ42τ0−τ32τ000τ32τ0−τ12τ0−τ32τ0τ12τ000−τ12τ0τ1τ2−τ32τ4τ000000,whereτ=τ1τ2τ3−τ32−τ12τ4.Thus, Σ1(5) is a semi-positive definite matrix. Case 2.1. If Δ1 ≠ 0, Δ3 ≠ 0. For convenience, let I22 = I6I4I3I2I1, equation (37) can be converted into the following form:(5.7)I22H12I22T+I22AI22−1I22Σ1I22T+I22Σ1I22TI22AI22−1T=0,and we can calculate thatI22Σ1I22T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA3=B1AO5,1T−λ2,B2=−λ10000−a15−a11−a14000b3−a44+a43b2b3a43Δ2Δ1a4300Δ1−a22−a43b1b3Δ2Δ1−a43b1b3000Δ3Δ4,whereA=0,0,0,−a26,a36+a26Δ2Δ1T.Next, define C2=M2B2M2−1, where the standardized transformation matrix M2 isM2=−v6−v7v8−v9−v100m5−m6m7m800Δ1Δ3−Δ3a22+a43b1Δ2b3Δ1−Δ4Δ42−Δ3a43b1b3000Δ3Δ400001,wherev6=a15m5,v7=m5a11+m6b3,v8=m6a44−a43b2b3+m7Δ1−m5a14,v9=m6a43Δ2Δ1+m7a22+a43b1Δ2b3Δ1−m8Δ3,v10=m6a43+m7a43b1b3−m8Δ4,m5=Δ1Δ3b3,m6=Δ1Δ3a44−a43b2b3+a22+a43b1Δ2b3Δ1−Δ4,m7=Δ2Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a22+a43b1Δ2b3Δ1+Δ3Δ42−Δ3a43b1b3,m8=Δ1Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a43b1b3+Δ4Δ42−Δ3a43b1b3.Then by simple calculation,C2=−a1−a2−a3−a4−a510000010000010000010.Therefore equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M2−1C2M2Σ1(5)+Σ1(5)M2TC2TM2−1T=0.That isdiag1,0,0,0,0+C2Σ12(5)+Σ12(5)C2T=0,whereΣ12(5)=ρ2−2M2Σ1(5)M2Tρ2=a15|m5|σ1.Since the eigenvalues of A have all negative real-parts, then C2 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ12(5) is a positive-definite matrix, and its specific form is as follows(5.8)Σ12(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ22M2−1Σ12(5)M2−1T is also a positive matrix. Case 2.1. If Δ1 ≠ 0, Δ3 ≠ 0. For convenience, let I22 = I6I4I3I2I1, equation (37) can be converted into the following form:(5.7)I22H12I22T+I22AI22−1I22Σ1I22T+I22Σ1I22TI22AI22−1T=0,and we can calculate thatI22Σ1I22T=Σ1(5)O5,1O5,1T0,where Σ1(5) is a five-dimensional symmetric matrix. LetA3=B1AO5,1T−λ2,B2=−λ10000−a15−a11−a14000b3−a44+a43b2b3a43Δ2Δ1a4300Δ1−a22−a43b1b3Δ2Δ1−a43b1b3000Δ3Δ4,whereA=0,0,0,−a26,a36+a26Δ2Δ1T.Next, define C2=M2B2M2−1, where the standardized transformation matrix M2 isM2=−v6−v7v8−v9−v100m5−m6m7m800Δ1Δ3−Δ3a22+a43b1Δ2b3Δ1−Δ4Δ42−Δ3a43b1b3000Δ3Δ400001,wherev6=a15m5,v7=m5a11+m6b3,v8=m6a44−a43b2b3+m7Δ1−m5a14,v9=m6a43Δ2Δ1+m7a22+a43b1Δ2b3Δ1−m8Δ3,v10=m6a43+m7a43b1b3−m8Δ4,m5=Δ1Δ3b3,m6=Δ1Δ3a44−a43b2b3+a22+a43b1Δ2b3Δ1−Δ4,m7=Δ2Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a22+a43b1Δ2b3Δ1+Δ3Δ42−Δ3a43b1b3,m8=Δ1Δ3a43+Δ3a22+a43b1Δ2b3Δ1−Δ4a43b1b3+Δ4Δ42−Δ3a43b1b3.Then by simple calculation,C2=−a1−a2−a3−a4−a510000010000010000010.Therefore equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M2−1C2M2Σ1(5)+Σ1(5)M2TC2TM2−1T=0.That isdiag1,0,0,0,0+C2Σ12(5)+Σ12(5)C2T=0,whereΣ12(5)=ρ2−2M2Σ1(5)M2Tρ2=a15|m5|σ1.Since the eigenvalues of A have all negative real-parts, then C2 is a five-dimensional Hurwitz matrix. According to the Lemma 2.3, Σ12(5) is a positive-definite matrix, and its specific form is as follows(5.8)Σ12(5)=a2a3a4−a2a5−a4a1a4−a52H40−a3a4−a2a52H40a4a4−a52H40a3a4−a2a52H40−a1a4−a52H40−a3a4−a2a52H40a1a4−a52H40−H22H40−a1a4−a52H40H22H40a1a4−a52H40−H22H40H32a5H4.Therefore, Σ1(5)=ρ22M2−1Σ12(5)M2−1T is also a positive matrix. Case 2.2. If Δ1 ≠ 0, Δ3 = 0, let C3=M3B3M3−1, where the standardized transformation matrix M3 isM3=−v11−v12v13−v14−v150m9−m10m11m1200Δ1−a22+a43b1Δ2b3Δ1−a43b1b30001000001,wherev11=a15m9,v12=m9a11+m10b3,v13=m10a44−a43b2b3+m11Δ1−m9a14,v14=m10a43Δ2Δ1+m11a22+a43b1Δ2b3Δ1,v15=m10a43+m11a43b1b3−m12Δ4,m9=Δ1b3,m10=Δ1a44−a43b2b3+a22+a43b1Δ2b3Δ1,m11=Δ2a43+a22+a43b1Δ2b3Δ12,m12=Δ1a43+a43b1b3a22+a43b1Δ2b3Δ1−a43b1b3Δ4.Then by simple calculation,C3=−τ1−τ2−τ3−τ4−τ51000001000001000000Δ4,whereτ1=q1+Δ4,τ2=q2+q1Δ4+Δ42,τ3=q3+q2Δ4+q1Δ42+Δ43,τ4=q4+q3Δ4+q2Δ42+q1Δ43+Δ44.Therefore, equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M3−1C3M3Σ1(5)+Σ1(5)M3TC3TM3−1T=0,That is,diag1,0,0,0,0+C3Σ13(5)+Σ13(5)C3T=0,where Σ13(5)=ρ3−2M3Σ1(5)M3T, ρ3=a15m9σ1. According to the Lemma 2.4, Σ13(5) is a semi-positive definite matrix and we compute(5.9)Σ13(5)=τ2τ3−τ1τ42τ0−τ32τ000τ32τ0−τ12τ0−τ32τ0τ12τ000−τ12τ0τ1τ2−τ32τ4τ000000,whereτ=τ1τ2τ3−τ32−τ12τ4.Thus, Σ1(5) is a semi-positive definite matrix. Case 2.2. If Δ1 ≠ 0, Δ3 = 0, let C3=M3B3M3−1, where the standardized transformation matrix M3 isM3=−v11−v12v13−v14−v150m9−m10m11m1200Δ1−a22+a43b1Δ2b3Δ1−a43b1b30001000001,wherev11=a15m9,v12=m9a11+m10b3,v13=m10a44−a43b2b3+m11Δ1−m9a14,v14=m10a43Δ2Δ1+m11a22+a43b1Δ2b3Δ1,v15=m10a43+m11a43b1b3−m12Δ4,m9=Δ1b3,m10=Δ1a44−a43b2b3+a22+a43b1Δ2b3Δ1,m11=Δ2a43+a22+a43b1Δ2b3Δ12,m12=Δ1a43+a43b1b3a22+a43b1Δ2b3Δ1−a43b1b3Δ4.Then by simple calculation,C3=−τ1−τ2−τ3−τ4−τ51000001000001000000Δ4,whereτ1=q1+Δ4,τ2=q2+q1Δ4+Δ42,τ3=q3+q2Δ4+q1Δ42+Δ43,τ4=q4+q3Δ4+q2Δ42+q1Δ43+Δ44.Therefore, equation (5.7) can be converted into the following form:diagσ12,0,0,0,0+M3−1C3M3Σ1(5)+Σ1(5)M3TC3TM3−1T=0,That is,diag1,0,0,0,0+C3Σ13(5)+Σ13(5)C3T=0,where Σ13(5)=ρ3−2M3Σ1(5)M3T, ρ3=a15m9σ1. According to the Lemma 2.4, Σ13(5) is a semi-positive definite matrix and we compute(5.9)Σ13(5)=τ2τ3−τ1τ42τ0−τ32τ000τ32τ0−τ12τ0−τ32τ0τ12τ000−τ12τ0τ1τ2−τ32τ4τ000000,whereτ=τ1τ2τ3−τ32−τ12τ4.Thus, Σ1(5) is a semi-positive definite matrix. Case 3. If Δ1 = 0, Δ2 = 0. For convenience, let I33 = I4I3I2I1, equation (5.2) can be converted into the following form:(5.10)I33H12I33T+I33AI33−1I33Σ1I33T+I33Σ1I33TI33AI33−1T=0,and we can calculate thatI33Σ1I33T=Σ1(3)O3,3O3,3TO3,3,where Σ1(3) is a three-dimensional symmetric matrix. LetA1=B4BO3,3TC,B4=−λ100−a15−a11−a140b3−a44+a43b2b3,whereB=0000000a430,C=−a22−a43b1b3−a26a32−a33−a43b2b3a3600−λ2.Next, define C4=M4B4M4−1, where the standardized transformation matrix M4 isM4=−a15b3−b3a11+a44−a43b2b3a44−a43b2b32−b3a140b3−a44−a43b2b3001.Then, by simple calculation,C4=−l¯1−l¯2−l¯3100010,wherel¯1=a1−z1−z2,l¯2=a2−(z1+z2)a1+(z1+z2)2−z1z2,l¯3=a3−(z1+z2)a2+(z1+z2)2a1−(z1+z2)3+2z1z2(z1+z2)−a1z1z2,z1=a33+a43b2b3,z2=a22.Therefore, equation (5.10) can be converted into the following form:diagσ12,0,0+M4−1C4M4Σ1(3)+Σ1(3)M4TC4TM4−1T=0.That is,diag1,0,0+C4Σ14(3)+Σ14(3)C4T=0,whereΣ14(3)=ρ4−2M4Σ1(3)M4T,ρ4=a15b3σ1.According to Lemma 2.5, Σ14(3) is a positive-definite matrix and we compute(5.11)Σ14(3)=l¯22l¯1l¯2−l¯30−12l¯1l¯2−l¯3012l¯1l¯2−l¯30−12l¯1l¯2−l¯30l¯12l¯3l¯1l¯2−l¯3.Thus, Σ1(5) is a semi-positive definite matrix.Taking the above several cases together, Σ1 is at least a semi-positive definite matrix. Define a positive constant ν which is the smallest eigenvalue of Σ1(3) i.e.Σ1(3)⩾ν100010001.Therefore, by analysis,Σ1=I33−1Σ1(3)O3,3O3,3TO3,3I33−1T⩾ν1−a25a15−a35a15000−a25a15a252a152+b12b32a25a35a152+b1b2b32b1b300−a35a15a25a35a152+b1b2b32a352a152+b22b32b2b3000b1b3b2b3100000010000000.By simple computation,(5.12)XΣ1XT≥νx1−a25a15x2−a35a15x32+b1b3x2+b2b3x3+x42+x52. Case 3. If Δ1 = 0, Δ2 = 0. For convenience, let I33 = I4I3I2I1, equation (5.2) can be converted into the following form:(5.10)I33H12I33T+I33AI33−1I33Σ1I33T+I33Σ1I33TI33AI33−1T=0,and we can calculate thatI33Σ1I33T=Σ1(3)O3,3O3,3TO3,3,where Σ1(3) is a three-dimensional symmetric matrix. LetA1=B4BO3,3TC,B4=−λ100−a15−a11−a140b3−a44+a43b2b3,whereB=0000000a430,C=−a22−a43b1b3−a26a32−a33−a43b2b3a3600−λ2.Next, define C4=M4B4M4−1, where the standardized transformation matrix M4 isM4=−a15b3−b3a11+a44−a43b2b3a44−a43b2b32−b3a140b3−a44−a43b2b3001.Then, by simple calculation,C4=−l¯1−l¯2−l¯3100010,wherel¯1=a1−z1−z2,l¯2=a2−(z1+z2)a1+(z1+z2)2−z1z2,l¯3=a3−(z1+z2)a2+(z1+z2)2a1−(z1+z2)3+2z1z2(z1+z2)−a1z1z2,z1=a33+a43b2b3,z2=a22.Therefore, equation (5.10) can be converted into the following form:diagσ12,0,0+M4−1C4M4Σ1(3)+Σ1(3)M4TC4TM4−1T=0.That is,diag1,0,0+C4Σ14(3)+Σ14(3)C4T=0,whereΣ14(3)=ρ4−2M4Σ1(3)M4T,ρ4=a15b3σ1.According to Lemma 2.5, Σ14(3) is a positive-definite matrix and we compute(5.11)Σ14(3)=l¯22l¯1l¯2−l¯30−12l¯1l¯2−l¯3012l¯1l¯2−l¯30−12l¯1l¯2−l¯30l¯12l¯3l¯1l¯2−l¯3.Thus, Σ1(5) is a semi-positive definite matrix. Taking the above several cases together, Σ1 is at least a semi-positive definite matrix. Define a positive constant ν which is the smallest eigenvalue of Σ1(3) i.e.Σ1(3)⩾ν100010001.Therefore, by analysis,Σ1=I33−1Σ1(3)O3,3O3,3TO3,3I33−1T⩾ν1−a25a15−a35a15000−a25a15a252a152+b12b32a25a35a152+b1b2b32b1b300−a35a15a25a35a152+b1b2b32a352a152+b22b32b2b3000b1b3b2b3100000010000000.By simple computation,(5.12)XΣ1XT≥νx1−a25a15x2−a35a15x32+b1b3x2+b2b3x3+x42+x52. Step 2. Consider the equation Step 2. Consider the equation Case 1. If Δ6 ≠ 0, for convenience, let J11 = J7J6J5J4J3J2J1, equation (5.13) can be transformed into the following form: Case 1. If Δ6 ≠ 0, for convenience, let J11 = J7J6J5J4J3J2J1, equation (5.13) can be transformed into the following form: Case 2. If Δ6 = 0, define C6=M6B5M6−1, where the standardized transformation matrix M6 is Case 2. If Δ6 = 0, define C6=M6B5M6−1, where the standardized transformation matrix M6 is Taking the above several cases together, Σ2 is at least a semi-positive definite matrix. Define a positive constant ϑ which is the smallest eigenvalue of Σ2(5) i.e.Σ2(5)⩾ϑ1000001000001000001000000.Therefore, by analysis,Σ2=J11−1Σ2(5)O5,1O5,1T0J11−1T⩾ϑ10−Δ5a1400001−1000−Δ5a14−1δ−dL2a432+Δ52a142+1dL−δa430000dL−δa43100000000000001.By simple computation,(5.19)XΣ2XT≥ϑx1−Δ5a14x32+x2−x32+δ−dLa43x3−x42+x62.Combining equations (5.12) and (5.19), we can obtain(5.20)XΣXT⩾κx1−a25a15x2−a35a15x32+b1b3x2+b2b3x3+x42+x52+x1−Δ5a14x32+x2−x32+δ−dLa43x3−x42+x62,where κ = ν ∧ϑ. Only if X=(0,0,0,0,0,0)T, XΣXT=0. Then Σ is a positive-definite matrix. To sum up, we obtain the exact expression of the probability density function around the quasi-equilibrium P∗. This completes the proof. Extinction Conditions for disease extinction play an important role in studying the transmission and control of disease. This section mainly discusses the sufficient conditions for disease extinction in model (1.5). For simplicity, defineR0e=R0+1A1dLdL+a¯+14αL1−R0a¯eσ22λ2−2eσ224λ2+1+1A1a¯αL(1−ϵ)λdTdL+a¯+1−dL(1−ϵ)λdTk¯eσ12λ1−2eσ124λ1+1,whereA1=maxc2N,δ4,dL+a¯1−dL+a¯αL2N.Theorem 6.1If R0 < 1 and R0e<1, latent infected cells, actively infected cells and free virus will be exponentially extinct.Proof. Define a non-negative C2-function UL,T∗,V, which owns the following form:U=T∗+a¯dL+a¯L+1NV.By simple calculation, the derivative of U isdUL,T∗,Vdt≤a¯αL(1−ϵ)k¯VλdT+a¯+(1−αL)(1−ϵ)k¯λVdT−cVN+dLdL+a¯|a−a¯|L+a¯αL(1−ϵ)λdT(dL+a¯)+(1−αL)(1−ϵ)λdT|k−k¯|V=cN(R0−1)V+a¯αL(1−ϵ)λdT(dL+a¯)+(1−αL)(1−ϵ)λdT|k−k¯|V+dLdL+a¯|a−a¯|L.On this basis, defineQ=U+14αL1−R01−αLL−αLT∗+12NR0−1V.Then, applying Itoˆ’s formula to lnQ, one obtains that(6.1)dlnQdt=1Qc2NR0−1V+δ4R0−1T∗+12NR0−1dL+a¯1−αL+a¯αLL+dLdL+a¯+14αL1−R0|a−a¯|LQ+a¯L(1−ϵ)λdTdL+a¯+1−αL(1−ϵ)λdT|k−k¯|VQ.Combining the previous formula (2.3), it is calculated thatlimt→∞1t∫0t|k(s)−k¯|ds≤k¯eσ12λ1−2eσ124λ1+1,limt→∞1t∫0t|a(s)−a¯|ds≤a¯eσ22λ2−2eσ224λ2+1.Therefore, taking the time average for both sides of the (6.1),(6.2)lnQ(t)−lnQ(0)t⩽maxc2N,δ4,dL+a¯1−αL+a¯αL2Nmin3+R04,a¯dL+a¯+1−R01−dL4dL,1+R02NR0−1+1a¯dL+a¯+1−R01−αL4αLdLdL+a¯+14αL1−R01t∫0t∣a(s)−a¯∣ds.+a¯αL(1−ϵ)λdT(dL+a)+(1−αL)(1−ϵ)λdT2N(1+R0)1t∫0t∣k(s)−k¯∣ds.For convenience, let(6.3)A1=maxc2N,δ4,dL+a¯1−dL+a¯αL2N,and(6.4)A2=min3+R04,a¯dL+a¯+1−R0(1−αL)4αL,1+R02N.Taking t tends to ∞ on both sides of (6.2), we obtainlimt→∞lnQ(t)−lnQ(0)t≤A1A2R0−1+1A2dLdL+a¯+14αL1−R0a¯eσ22λ2−2eσ224λ2+1+1A2a¯αL(1−ϵ)λdTdL+a¯+(1−αL)(1−ϵ)λdTk¯eσ12λ1−2eσ124λ1+1≤A1A2R0e−1,whereR0e=R0+1A1dLdL+a¯+14αL1−R0a¯eσ22λ2−2eσ224λ2+1+1A1a¯αL(1−ϵ)λdTdL+a¯+1−dL(1−ϵ)λdTk¯eσ12λ1−2eσ124λ1+1.It is indicated that if R0<1 and R0e<1, Q will decay at an exponential rate, which means thatlimt→∞L(t)=limt→∞T∗(t)=limt→∞V(t)=0.This completes the proof. If R0 < 1 and R0e<1, latent infected cells, actively infected cells and free virus will be exponentially extinct. Proof. Define a non-negative C2-function UL,T∗,V, which owns the following form:U=T∗+a¯dL+a¯L+1NV.By simple calculation, the derivative of U isdUL,T∗,Vdt≤a¯αL(1−ϵ)k¯VλdT+a¯+(1−αL)(1−ϵ)k¯λVdT−cVN+dLdL+a¯|a−a¯|L+a¯αL(1−ϵ)λdT(dL+a¯)+(1−αL)(1−ϵ)λdT|k−k¯|V=cN(R0−1)V+a¯αL(1−ϵ)λdT(dL+a¯)+(1−αL)(1−ϵ)λdT|k−k¯|V+dLdL+a¯|a−a¯|L.On this basis, defineQ=U+14αL1−R01−αLL−αLT∗+12NR0−1V.Then, applying Itoˆ’s formula to lnQ, one obtains that(6.1)dlnQdt=1Qc2NR0−1V+δ4R0−1T∗+12NR0−1dL+a¯1−αL+a¯αLL+dLdL+a¯+14αL1−R0|a−a¯|LQ+a¯L(1−ϵ)λdTdL+a¯+1−αL(1−ϵ)λdT|k−k¯|VQ.Combining the previous formula (2.3), it is calculated thatlimt→∞1t∫0t|k(s)−k¯|ds≤k¯eσ12λ1−2eσ124λ1+1,limt→∞1t∫0t|a(s)−a¯|ds≤a¯eσ22λ2−2eσ224λ2+1.Therefore, taking the time average for both sides of the (6.1),(6.2)lnQ(t)−lnQ(0)t⩽maxc2N,δ4,dL+a¯1−αL+a¯αL2Nmin3+R04,a¯dL+a¯+1−R01−dL4dL,1+R02NR0−1+1a¯dL+a¯+1−R01−αL4αLdLdL+a¯+14αL1−R01t∫0t∣a(s)−a¯∣ds.+a¯αL(1−ϵ)λdT(dL+a)+(1−αL)(1−ϵ)λdT2N(1+R0)1t∫0t∣k(s)−k¯∣ds.For convenience, let(6.3)A1=maxc2N,δ4,dL+a¯1−dL+a¯αL2N,and(6.4)A2=min3+R04,a¯dL+a¯+1−R0(1−αL)4αL,1+R02N.Taking t tends to ∞ on both sides of (6.2), we obtainlimt→∞lnQ(t)−lnQ(0)t≤A1A2R0−1+1A2dLdL+a¯+14αL1−R0a¯eσ22λ2−2eσ224λ2+1+1A2a¯αL(1−ϵ)λdTdL+a¯+(1−αL)(1−ϵ)λdTk¯eσ12λ1−2eσ124λ1+1≤A1A2R0e−1,whereR0e=R0+1A1dLdL+a¯+14αL1−R0a¯eσ22λ2−2eσ224λ2+1+1A1a¯αL(1−ϵ)λdTdL+a¯+1−dL(1−ϵ)λdTk¯eσ12λ1−2eσ124λ1+1.It is indicated that if R0<1 and R0e<1, Q will decay at an exponential rate, which means thatlimt→∞L(t)=limt→∞T∗(t)=limt→∞V(t)=0.This completes the proof. Numerical simulations Numerical simulation is an essential method for revealing the transmission patterns of diseases and predicting their developmental trends. In this section, we use numerical simulations to confirm the results of our analysis. According to Milstein higher-order numerical method, the corresponding discretization equation for the system is as follows:Tj+1=Tj+λ−dTTj−(1−ϵ)k¯ex1jVjTjΔt,Lj+1=Lj+αL(1−ϵ)k¯ex1jVjTj−(dL+a¯ex2j)LjΔt,T∗j+1=T∗j+(1−αL)(1−ϵ)k¯ex1jVjTj−δT∗j+a¯ex2jLjΔt,Vj+1=Vj+NδT∗j−cVjΔt,x1j+1=x1j−λ1x1jΔt+σ1Δtξ1,j,x2j+1=x2j−λ2x2jΔt+σ2Δtξ2,j,where Δt is the time increment and ξ1,j, ξ2,j are two independent random variables that follow a Gaussian distribution N(0, 1). (Tj,Lj,(T∗)j,Vj,kj,aj) represents the value of the j-th iteration of the system. Parameter values used in numerical simulations are defined in Table 1. Choose the initial values as T(0),L(0),T∗(0),V(0),k(0),a(0)=1.3,2,0.7,0.4,0,0.Example 7.1(Stationary distribution) To illustrate the existence of a stationary distribution, let ɛ = 2 × 10−6 and N = 8000, chooseσi=0.003,λi=160,i=1,2,and the values of other parameters are as shown in Table 1. Then we calculateR0s=αL(1−ϵ)k¯λa¯NdTdL+a¯eσ224λ2c+1−αL(1−ϵ)k¯λNdTc=1.6695>1.In this case, model (1.5) admits at least one stationary distribution g∗(⋅) on Γ, which indicates that the disease will persist in the long term. Fig. 1 effectively validates this statement.Fig. 1Numerical simulations of model (1.5) under stochastic perturbations (σ1, σ2) = (0.003, 0.003). Left column: Time evolution of uninfected target cells T, latent infected cells L, actively infected cells T∗, and free virus particles V. Right column: Corresponding frequency histograms and estimated marginal probability density functions, illustrating the stationary distribution and stochastic variability of each compartment.Example 7.2(Probability density function) According to the Example 1, there is at least one stationary distribution g∗(⋅) on Γ and in this example, we focus on the specific expression of the probability density function near the quasi-equilibrium point. The values of each parameter are the same as those in Example 7.1.R0=αL(1−ϵ)k¯λa¯NdT(dL+a¯)c+1−αL(1−ϵ)k¯λNdTc=1.7132>1.According to Theorem 5.1, the solution T(t),L(t),T∗(t),V(t),k(t),a(t) follows a normal density function(7.1)Ψ(T,L,T∗,V,logk,loga)∼N6P∗,Σ,with P∗=(Tˆ,Lˆ,Tˆ∗,Vˆ,lnk¯,lna¯)=59898,38.5595,4010,1394800,ln2.4×10−8,ln0.1 andΣ=1.7859×108−4.8173×104−2.3502×106−8.2594×108−151.3353−0.0245−4.8173×10435.95363.8864×1031.3545×1060.0334−0.0086−2.3502×1063.8864×1035.8458×1052.0333×1083.96888.1782×10−4−8.2594×1081.3545×1062.0333×1087.0723×10101.3794×1030.2843−151.33530.03343.96881.3794×1032.7000×10−40−0.0245−0.00868.1782×10−40.284302.7000×10−4.The marginal probability density functions of T(t), L(t), T∗(t) and V(t) are given as follows:∂Ψ(T,L,T∗,V,logk,loga)∂T=2.2338×10−9e−1.5677×10−17(T−59898)2,∂Ψ(T,L,T∗,V,logk,loga)∂L=0.0111e−3.8680×10−4(L−38.5598)2,∂Ψ(T,L,T∗,V,logk,loga)∂T∗=6.8244×10−7e−1.4631×10−12(T∗−4010)2,∂Ψ(T,L,T∗,V,logk,loga)∂V=5.6409×10−12e−9.9965×10−23(V−1394800)2.We plot the marginal density function curves (the red line in Fig. 2) and the fitting curves of frequency histograms (the blue line in Fig. 2) of T(t), L(t), T∗(t) and V(t) with 40,000,000 iteration points. Upon observation, it is clearly that the two curves roughly overlap, confirming the validity of Theorem 5.1.Fig. 2Marginal probability density functions and frequency histograms for T(t), L(t), T∗(t), and V(t) in model (1.5), computed over 40,000,000 iteration points. The curves illustrate the stationary distributions and stochastic variability of each compartment under the simulated stochastic dynamics.Example 7.3(Extinction) In order to obtain sufficient conditions for the extinction of the disease, chooseσi=0.3,λi=14,i=1,2,ϵ=0.8,and the values of other parameters are as shown in Table 1. Then we calculateR0=αL(1−ϵ)k¯λa¯NdT(dL+a¯)c+(1−αL)(1−ϵ)k¯λNdTc=0.6261<1,andR0e=R0+1A1dLdL+a¯+14αL1−R0a¯eσ22λ2−2eσ224λ2+1+1A1a¯αL(1−ϵ)λdTdL+a¯+1−dL(1−ϵ)λdTk¯eσ12λ1−2eσ124λ1+1=0.7482<1.According to Theorem 6.1, latent infected cells L(t), actively infected cells T∗(t) and free viruses V(t) will go extinction in a long term. Fig. 3 clearly illustrates this conclusion, showing that the number of L(t), T∗(t) and V(t) decreases rapidly to near zero.Fig. 3Time evolution of model (1.5) under stochastic perturbations (σ1, σ2) = (0.3, 0.3) with R0e<1. The plots show the dynamics of uninfected target cells T, latent infected cells L, actively infected cells T∗, and free virus particles V, highlighting the stochastic extinction behavior predicted by the threshold R0e.Example 7.4(The impact of the white noise intensity σi and recovery rate λi)As widely recognized, the intensity of white noise and the recovery rate in the OU process play crucial roles in simulating the spread of viruses. The intensity of white noise k affects the strength of random perturbations in the model, while the recovery rate a determines the speed of mean reversion. Employing a controlled variable approach, we conduct simulations on the stochastic solutions of model (1.5) under various scenarios with white noise intensities of σ1=0.01,σ2=0.01, σ1=0.03,σ2=0.01 and σ1=0.01,σ2=0.03 as depicted in Fig. 4. It can be observed that by increasing the magnitudes of σ1 and σ2, each component exhibits varying degrees of fluctuation. Under otherwise identical conditions, the fluctuations induced by changes in σ1 are notably greater than those caused by changes in σ2. In other words, σ1 has a larger destabilizing effect on each component compared to σ2.Fig. 4Simulations of the solution T(t),L(t),T∗(t),V(t) of model (1.5) under different noise intensities: (σ1, σ2) = (0.01, 0.01), (0.03, 0.01), and (0.01, 0.03). All other parameters are as in Example 7.1. The plots illustrate how variations in noise intensities influence the stochastic dynamics of each compartment.Furthermore, employing the same methodology, we depict the mean first passage time (MFPT) from the initial state to the persistent state under different recovery rates, as illustrated in Fig. 5. It can be observed that the larger the recovery rate λi, the longer the mean first passage time from the initial state to the persistent state.Fig. 5Mean first passage time (MFPT) for the transition from the initial state to the persistent state in model (1.5). All other parameter values are as in Example 7.1. The results illustrate the expected time required for the system to reach the persistent state under stochastic fluctuations.Example 7.5(The impact of the infection rate k and the activation rate a)Studying the impact of infection rates on HIV transmission is crucial for controlling the spread of the disease and formulating public health policies. The infection rate k is a key parameter in determining the threshold for HIV virus epidemic and extinction. Fig. 6 illustrates the changing trends of three thresholds R0, R0s and R0e under different values of k. It can be observed that.•If k∈1.436×10−8,1.43686184×10−8, latent infected cells L(t), effective infected cells T∗(t) and free virus V(t) will exponentially decay.•If k∈1.43755532×10−8,1.44×10−8, latent infected cells L(t), effective infected cells T∗(t) and free virus V(t) will persist in a long term.•If k∈1.43755529×10−8,1.44×10−8, the stochastic solution T(t),L(t),T∗(t),V(t),k(t),a(t) follows a normal density function.Fig. 6Trends of the basic reproduction numbers R0, R0s, and R0e as functions of the infection rate k within the range [1.436 × 10−8, 1.44 × 10−8]. All other parameters are the same as in Example 7.1. The figure illustrates the sensitivity of stochastic and deterministic thresholds to variations in k.In addition, a is also a crucial parameter influencing the thresholds for disease prevalence and extinction. Therefore, we further investigate the joint impact of k and a on R0s and R0e. Fig. 7, Fig. 8 respectively illustrate the values of R0s and R0e under different k,a.Fig. 7Trends of the stochastic basic reproduction number R0s as a function of (k, a) within the ranges [1 × 10−8, 2 × 10−8] × [0.05, 0.35]. All other parameters are the same as in Example 7.1. The figure illustrates how variations in both the infection rate k and latent-cell activation rate a affect the stochastic threshold for persistence.Fig. 8Trends of the extinction-related stochastic reproduction number R0e as a function of (k, a) within the ranges [1 × 10−8, 2 × 10−8] × [0.05, 0.35]. All other parameters are the same as in Example 7.1. The figure illustrates how variations in the infection rate k and latent-cell activation rate a influence the stochastic threshold for viral extinction. (Stationary distribution) To illustrate the existence of a stationary distribution, let ɛ = 2 × 10−6 and N = 8000, chooseσi=0.003,λi=160,i=1,2,and the values of other parameters are as shown in Table 1. Then we calculateR0s=αL(1−ϵ)k¯λa¯NdTdL+a¯eσ224λ2c+1−αL(1−ϵ)k¯λNdTc=1.6695>1.In this case, model (1.5) admits at least one stationary distribution g∗(⋅) on Γ, which indicates that the disease will persist in the long term. Fig. 1 effectively validates this statement.Fig. 1Numerical simulations of model (1.5) under stochastic perturbations (σ1, σ2) = (0.003, 0.003). Left column: Time evolution of uninfected target cells T, latent infected cells L, actively infected cells T∗, and free virus particles V. Right column: Corresponding frequency histograms and estimated marginal probability density functions, illustrating the stationary distribution and stochastic variability of each compartment. Numerical simulations of model (1.5) under stochastic perturbations (σ1, σ2) = (0.003, 0.003). Left column: Time evolution of uninfected target cells T, latent infected cells L, actively infected cells T∗, and free virus particles V. Right column: Corresponding frequency histograms and estimated marginal probability density functions, illustrating the stationary distribution and stochastic variability of each compartment. (Probability density function) According to the Example 1, there is at least one stationary distribution g∗(⋅) on Γ and in this example, we focus on the specific expression of the probability density function near the quasi-equilibrium point. The values of each parameter are the same as those in Example 7.1.R0=αL(1−ϵ)k¯λa¯NdT(dL+a¯)c+1−αL(1−ϵ)k¯λNdTc=1.7132>1.According to Theorem 5.1, the solution T(t),L(t),T∗(t),V(t),k(t),a(t) follows a normal density function(7.1)Ψ(T,L,T∗,V,logk,loga)∼N6P∗,Σ,with P∗=(Tˆ,Lˆ,Tˆ∗,Vˆ,lnk¯,lna¯)=59898,38.5595,4010,1394800,ln2.4×10−8,ln0.1 andΣ=1.7859×108−4.8173×104−2.3502×106−8.2594×108−151.3353−0.0245−4.8173×10435.95363.8864×1031.3545×1060.0334−0.0086−2.3502×1063.8864×1035.8458×1052.0333×1083.96888.1782×10−4−8.2594×1081.3545×1062.0333×1087.0723×10101.3794×1030.2843−151.33530.03343.96881.3794×1032.7000×10−40−0.0245−0.00868.1782×10−40.284302.7000×10−4.The marginal probability density functions of T(t), L(t), T∗(t) and V(t) are given as follows:∂Ψ(T,L,T∗,V,logk,loga)∂T=2.2338×10−9e−1.5677×10−17(T−59898)2,∂Ψ(T,L,T∗,V,logk,loga)∂L=0.0111e−3.8680×10−4(L−38.5598)2,∂Ψ(T,L,T∗,V,logk,loga)∂T∗=6.8244×10−7e−1.4631×10−12(T∗−4010)2,∂Ψ(T,L,T∗,V,logk,loga)∂V=5.6409×10−12e−9.9965×10−23(V−1394800)2. We plot the marginal density function curves (the red line in Fig. 2) and the fitting curves of frequency histograms (the blue line in Fig. 2) of T(t), L(t), T∗(t) and V(t) with 40,000,000 iteration points. Upon observation, it is clearly that the two curves roughly overlap, confirming the validity of Theorem 5.1.Fig. 2Marginal probability density functions and frequency histograms for T(t), L(t), T∗(t), and V(t) in model (1.5), computed over 40,000,000 iteration points. The curves illustrate the stationary distributions and stochastic variability of each compartment under the simulated stochastic dynamics. Marginal probability density functions and frequency histograms for T(t), L(t), T∗(t), and V(t) in model (1.5), computed over 40,000,000 iteration points. The curves illustrate the stationary distributions and stochastic variability of each compartment under the simulated stochastic dynamics. (Extinction) In order to obtain sufficient conditions for the extinction of the disease, chooseσi=0.3,λi=14,i=1,2,ϵ=0.8,and the values of other parameters are as shown in Table 1. Then we calculateR0=αL(1−ϵ)k¯λa¯NdT(dL+a¯)c+(1−αL)(1−ϵ)k¯λNdTc=0.6261<1,andR0e=R0+1A1dLdL+a¯+14αL1−R0a¯eσ22λ2−2eσ224λ2+1+1A1a¯αL(1−ϵ)λdTdL+a¯+1−dL(1−ϵ)λdTk¯eσ12λ1−2eσ124λ1+1=0.7482<1.According to Theorem 6.1, latent infected cells L(t), actively infected cells T∗(t) and free viruses V(t) will go extinction in a long term. Fig. 3 clearly illustrates this conclusion, showing that the number of L(t), T∗(t) and V(t) decreases rapidly to near zero.Fig. 3Time evolution of model (1.5) under stochastic perturbations (σ1, σ2) = (0.3, 0.3) with R0e<1. The plots show the dynamics of uninfected target cells T, latent infected cells L, actively infected cells T∗, and free virus particles V, highlighting the stochastic extinction behavior predicted by the threshold R0e. Time evolution of model (1.5) under stochastic perturbations (σ1, σ2) = (0.3, 0.3) with R0e<1. The plots show the dynamics of uninfected target cells T, latent infected cells L, actively infected cells T∗, and free virus particles V, highlighting the stochastic extinction behavior predicted by the threshold R0e. (The impact of the white noise intensity σi and recovery rate λi) As widely recognized, the intensity of white noise and the recovery rate in the OU process play crucial roles in simulating the spread of viruses. The intensity of white noise k affects the strength of random perturbations in the model, while the recovery rate a determines the speed of mean reversion. Employing a controlled variable approach, we conduct simulations on the stochastic solutions of model (1.5) under various scenarios with white noise intensities of σ1=0.01,σ2=0.01, σ1=0.03,σ2=0.01 and σ1=0.01,σ2=0.03 as depicted in Fig. 4. It can be observed that by increasing the magnitudes of σ1 and σ2, each component exhibits varying degrees of fluctuation. Under otherwise identical conditions, the fluctuations induced by changes in σ1 are notably greater than those caused by changes in σ2. In other words, σ1 has a larger destabilizing effect on each component compared to σ2.Fig. 4Simulations of the solution T(t),L(t),T∗(t),V(t) of model (1.5) under different noise intensities: (σ1, σ2) = (0.01, 0.01), (0.03, 0.01), and (0.01, 0.03). All other parameters are as in Example 7.1. The plots illustrate how variations in noise intensities influence the stochastic dynamics of each compartment. Simulations of the solution T(t),L(t),T∗(t),V(t) of model (1.5) under different noise intensities: (σ1, σ2) = (0.01, 0.01), (0.03, 0.01), and (0.01, 0.03). All other parameters are as in Example 7.1. The plots illustrate how variations in noise intensities influence the stochastic dynamics of each compartment. Furthermore, employing the same methodology, we depict the mean first passage time (MFPT) from the initial state to the persistent state under different recovery rates, as illustrated in Fig. 5. It can be observed that the larger the recovery rate λi, the longer the mean first passage time from the initial state to the persistent state.Fig. 5Mean first passage time (MFPT) for the transition from the initial state to the persistent state in model (1.5). All other parameter values are as in Example 7.1. The results illustrate the expected time required for the system to reach the persistent state under stochastic fluctuations. Mean first passage time (MFPT) for the transition from the initial state to the persistent state in model (1.5). All other parameter values are as in Example 7.1. The results illustrate the expected time required for the system to reach the persistent state under stochastic fluctuations. (The impact of the infection rate k and the activation rate a) Studying the impact of infection rates on HIV transmission is crucial for controlling the spread of the disease and formulating public health policies. The infection rate k is a key parameter in determining the threshold for HIV virus epidemic and extinction. Fig. 6 illustrates the changing trends of three thresholds R0, R0s and R0e under different values of k. It can be observed that.•If k∈1.436×10−8,1.43686184×10−8, latent infected cells L(t), effective infected cells T∗(t) and free virus V(t) will exponentially decay.•If k∈1.43755532×10−8,1.44×10−8, latent infected cells L(t), effective infected cells T∗(t) and free virus V(t) will persist in a long term.•If k∈1.43755529×10−8,1.44×10−8, the stochastic solution T(t),L(t),T∗(t),V(t),k(t),a(t) follows a normal density function.Fig. 6Trends of the basic reproduction numbers R0, R0s, and R0e as functions of the infection rate k within the range [1.436 × 10−8, 1.44 × 10−8]. All other parameters are the same as in Example 7.1. The figure illustrates the sensitivity of stochastic and deterministic thresholds to variations in k. •If k∈1.436×10−8,1.43686184×10−8, latent infected cells L(t), effective infected cells T∗(t) and free virus V(t) will exponentially decay. If k∈1.436×10−8,1.43686184×10−8, latent infected cells L(t), effective infected cells T∗(t) and free virus V(t) will exponentially decay. •If k∈1.43755532×10−8,1.44×10−8, latent infected cells L(t), effective infected cells T∗(t) and free virus V(t) will persist in a long term. If k∈1.43755532×10−8,1.44×10−8, latent infected cells L(t), effective infected cells T∗(t) and free virus V(t) will persist in a long term. •If k∈1.43755529×10−8,1.44×10−8, the stochastic solution T(t),L(t),T∗(t),V(t),k(t),a(t) follows a normal density function. If k∈1.43755529×10−8,1.44×10−8, the stochastic solution T(t),L(t),T∗(t),V(t),k(t),a(t) follows a normal density function. Trends of the basic reproduction numbers R0, R0s, and R0e as functions of the infection rate k within the range [1.436 × 10−8, 1.44 × 10−8]. All other parameters are the same as in Example 7.1. The figure illustrates the sensitivity of stochastic and deterministic thresholds to variations in k. In addition, a is also a crucial parameter influencing the thresholds for disease prevalence and extinction. Therefore, we further investigate the joint impact of k and a on R0s and R0e. Fig. 7, Fig. 8 respectively illustrate the values of R0s and R0e under different k,a.Fig. 7Trends of the stochastic basic reproduction number R0s as a function of (k, a) within the ranges [1 × 10−8, 2 × 10−8] × [0.05, 0.35]. All other parameters are the same as in Example 7.1. The figure illustrates how variations in both the infection rate k and latent-cell activation rate a affect the stochastic threshold for persistence.Fig. 8Trends of the extinction-related stochastic reproduction number R0e as a function of (k, a) within the ranges [1 × 10−8, 2 × 10−8] × [0.05, 0.35]. All other parameters are the same as in Example 7.1. The figure illustrates how variations in the infection rate k and latent-cell activation rate a influence the stochastic threshold for viral extinction. Trends of the stochastic basic reproduction number R0s as a function of (k, a) within the ranges [1 × 10−8, 2 × 10−8] × [0.05, 0.35]. All other parameters are the same as in Example 7.1. The figure illustrates how variations in both the infection rate k and latent-cell activation rate a affect the stochastic threshold for persistence. Trends of the extinction-related stochastic reproduction number R0e as a function of (k, a) within the ranges [1 × 10−8, 2 × 10−8] × [0.05, 0.35]. All other parameters are the same as in Example 7.1. The figure illustrates how variations in the infection rate k and latent-cell activation rate a influence the stochastic threshold for viral extinction. Conclusion In this study, we analyzed a stochastic within-host HIV model incorporating latent cell activation, with infection and activation rates subject to environmental fluctuations modeled by logarithmic mean-reverting Ornstein-Uhlenbeck processes. This formulation ensures parameter positivity while capturing correlated, temporally structured stochastic variability.•If the stochastic persistence thresholdR0s=αL(1−ϵ)k¯λa¯NdTdL+a¯eσ224λ2c+(1−αL)(1−ϵ)k¯λNdTc>1,the system admits at least one stationary distribution supported on the invariant set Γ, indicating stochastic persistence of all state variables in distribution.•If the stochastic extinction threshold R0e<1, then latent infected cells L(t), actively infected cells T∗(t), and free virus V(t) decay exponentially almost surely, demonstrating that sufficiently strong stochastic perturbations can induce extinction even when the deterministic model predicts persistence.•When the deterministic reproduction numberR0=αL(1−ϵ)kλa¯NdT(dL+a¯)c+(1−αL)(1−ϵ)kλNdTc>1,the solution in a neighborhood of the quasi-equilibrium point follows an approximate multivariate normal distribution, with explicit covariance derived from the associated Fokker-Planck equation, quantifying stochastic fluctuations near equilibrium see (Fig. 9).Fig. 9Three-dimensional trajectories of latent infected cells L, actively infected cells T∗, and free virus particles V under two threshold regimes. Left panel: joint dynamics when R0s>1, illustrating stochastic persistence of infection. Right panel: joint dynamics when R0e<1, illustrating stochastic extinction of infection. •If the stochastic persistence threshold If the stochastic persistence threshold •If the stochastic extinction threshold R0e<1, then latent infected cells L(t), actively infected cells T∗(t), and free virus V(t) decay exponentially almost surely, demonstrating that sufficiently strong stochastic perturbations can induce extinction even when the deterministic model predicts persistence. If the stochastic extinction threshold R0e<1, then latent infected cells L(t), actively infected cells T∗(t), and free virus V(t) decay exponentially almost surely, demonstrating that sufficiently strong stochastic perturbations can induce extinction even when the deterministic model predicts persistence. •When the deterministic reproduction number When the deterministic reproduction number Three-dimensional trajectories of latent infected cells L, actively infected cells T∗, and free virus particles V under two threshold regimes. Left panel: joint dynamics when R0s>1, illustrating stochastic persistence of infection. Right panel: joint dynamics when R0e<1, illustrating stochastic extinction of infection. These results show that environmental variability fundamentally reshapes within-host HIV dynamics. Noise acting on key parameters can suppress viral persistence or induce extinction, emphasizing the importance of incorporating stochastic effects in models with latent reservoirs. Mathematically, the study extends deterministic threshold theory to a stochastic framework with latent infection, multiplicative mean-reverting noise, and multi-scale dynamics. The distinct stochastic thresholds R0s and R0e highlight how environmental noise modifies the effective reproduction mechanism. Future work may incorporate stochasticity in additional biological parameters (Kang et al., 2026; Ma & Wang, 2026), explore non-Gaussian or state-dependent noise, and develop a unified framework linking deterministic and stochastic reproduction numbers. CRediT authorship contribution statement Zhongwei Cao: Writing – original draft, Funding acquisition, Formal analysis, Conceptualization. Yuchen Jiang: Writing – original draft, Software, Investigation. Zhenfeng Shi: Writing – review & editing, Supervision, Formal analysis. Daqing Jiang: Writing – review & editing, Methodology. Su Wei: Writing – original draft, Resources. Yaxin Zhou: Formal analysis, Writing – review & editing. Data availability statements The manuscript has no associated data. Declaration of competing interest The authors declare no conflict of interest.