A model for mosquito-borne epidemic outbreaks with information-dependent protective behaviour Authors: Simone De Reggi, Andrea Pugliese, Mattia Sensi, Cinzia Soresina Published: 2026 ABSTRACT We investigate a model for a mosquito-borne epidemic in which human hosts may adopt protective behaviour against vector bites in response to information on both past and current disease prevalence. Assuming that mosquitoes can also feed on non-competent hosts (i.e. hosts that do not contribute to disease transmission), we first revisit existing results and show that behaviour-driven protection may either decrease or increase the basic reproduction number, depending on the interaction between behavioural response, host composition, and transmission parameters. Assuming that opinion dynamics evolves on a much faster time scale than disease transmission, we then apply Geometric Singular Perturbation Theory to effectively reduce the original two-group model to a model for a homogeneous host population. The reduced system enables a detailed investigation of the impact of information-induced behavioural changes on the transient dynamics of the epidemic, including scenarios in which protective measures lead to outbreaks with low attack rates. Our analysis shows that behavioural responses may either facilitate epidemic control or prolong disease persistence, potentially generating recurrent damped epidemic waves. Numerical simulations are provided to illustrate and support the analytical findings. Introduction In recent years, there has been growing concern over the rise in mosquito-borne diseases in both endemic and non-endemic areas (European Centre for Disease Prevention and Control 2025; World Health Organization 2025). Malaria, dengue, chikungunya, Japanese encephalitis, West Nile virus, yellow fever, and Zika are among the most prominent examples of mosquito-borne infections. These diseases are transmitted to humans through the bites of infected mosquitoes carrying pathogens (viruses or other parasites), and represent major threats to global public health (World Health Organization 2025). Factors such as climate change, globalisation, and urbanisation have strongly favoured the spread of many of these infections, such as dengue and chikungunya, to non-tropical areas, creating numerous new habitats suitable for mosquito life (Chala and Hamde 2021). Mathematical models play a crucial role in understanding the spread of mosquito-borne infections and in assessing the potential impact of control strategies. The classical Ross–Macdonald framework provides the foundation for much of the mathematical modelling of such diseases; see, for instance, the recent review by Pugliese and De Reggi (2026). Originally proposed by Ronald Ross between 1908 and 1911 to investigate the effect of control measures on malaria transmission (Ross 1911), the model was later extended by George Macdonald to incorporate, among other features, incubation periods, i.e. time delays between infection and infectiousness (Macdonald 1957) (the role of delays had already been considered by Sharpe and Lotka (1923); however, they neglected mosquito mortality during the incubation period. Moreover, Macdonald acknowledged Armitage (1953) for mathematical ideas related to delays; see Pugliese and De Reggi (2026), Smith et al. (2012)). In particular, the mathematical modelling of dengue, which strongly motivates the present work, poses significant epidemiological and methodological challenges; see the recent review Aguiar et al. (2022). A key feature distinguishing vector-borne infections from directly transmitted ones is that epidemic containment can be achieved through interventions targeting the vector population (Ogunlade et al. 2023). This insight was already recognised by Ross in his pioneering work on malaria, where he developed arguments that anticipated the modern concept of the Basic Reproduction Number (BRN), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0; see Bacaër (2011), Heesterbeek (2002), Pugliese and De Reggi (2026). However, effective control of mosquito-borne epidemics generally requires combining vector-targeted interventions with preventive measures adopted by humans, such as the use of repellents or bed nets. Theoretical investigations of the effectiveness of such strategies require mathematical models incorporating behavioural components. As clearly illustrated by the COVID-19 pandemic, human behaviour can substantially influence epidemic spread, since individuals may dynamically adjust their actions in response to available information. For example, individuals may decide whether to be vaccinated or to adopt self-protective measures, either in accordance with public health guidelines or on their own initiative. The behavioural epidemiology of infectious diseases addresses these aspects by incorporating behavioural dynamics into mathematical models of disease transmission (Manfredi and d’Onofrio 2013). Its origins can be traced back to the seminal work of Capasso and Serio (1978), who first proposed an SIR (Susceptible–Infectious–Removed) model in which the transmission rate was assumed to decrease as a function of the current prevalence, that is, the total number of infected individuals (d’Onofrio and Manfredi 2009). Since then, numerous approaches have been developed to incorporate behavioural feedback into epidemic models. We refer to the monograph by Manfredi and d’Onofrio (2013) and to the recent reviews by Bedson et al. (2021), Funk et al. (2010), Wang et al. (2016) for comprehensive overviews of this field. Although a substantial body of literature now addresses the interplay between the spread of directly transmitted infections and human behaviour, comparatively less attention has been devoted to information-induced behavioural effects in vector-borne diseases. In the following paragraphs, we briefly review selected contributions in this area. For directly transmitted infections, the adoption of preventive behaviour by a subset of individuals generally reduces contact rates across the population, thereby producing an overall beneficial effect. Mathematically, this is typically modelled by assuming that transmission rates decrease as functions of the level of awareness within the population (Manfredi and d’Onofrio 2013). In contrast, for vector-borne infections, modelling partial protective behaviour among hosts is far from straightforward, as predictions regarding its effectiveness critically depend on the assumptions governing vector–host ecological interactions (McCallum et al. 2001; Thongsripong et al. 2021). In particular, different assumptions on how mosquito biting rates depend on host availability may lead to substantially different predictions concerning the probability of epidemic invasion and the effectiveness of control strategies (Demers et al. 2018); see also Wonham et al. (2006). As an example, for vector-borne infections, the adoption of self-protective measures by a subset of individuals may divert mosquito bites towards unprotected hosts (Killeen and Smith 2007; Moore et al. 2007), thereby increasing the likelihood of mosquito–host–mosquito transmission cycles (Demers et al. 2018; Killeen and Smith 2007; Miller et al. 2016). More generally, partially adopted protective behaviour may enhance host heterogeneity, as individuals using repellents or bed nets become less attractive to mosquitoes than those who remain unprotected (Killeen and Smith 2007; Miller et al. 2016). Dye and Hasibeder (1986) showed that host heterogeneity in vector-borne epidemics may, perhaps counter-intuitively, increase the value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0, thereby enhancing the invasion potential of the disease. At the same time, they demonstrated that the final epidemic size may decrease under such heterogeneity (Hasibeder and Dye 1988). This phenomenon was revisited by Miller and Huppert (2013), who considered a simple vector–host model with two host classes and investigated the effects of host diversity, transmission competence, and vector preference on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0. In a subsequent work, Miller et al. (2016) analysed a one-vector–one-host model in which the host population is statically divided into two subgroups: individuals adopting protective measures and individuals who do not. They found that, depending on parameter values, the use of protective measures by only a fraction of the host population may increase \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0, thereby exacerbating the risk of epidemic invasion. On the other hand, if protective measures such as bed nets or repellents significantly increase the time mosquitoes spend attempting to bite protected individuals, then a beneficial effect may also extend to the unprotected subgroup. Recently, Demers et al. (2018) investigated, primarily through numerical simulations and analysis of the basic reproduction number, the potential effects of behavioural changes during the early phase of a mosquito-borne outbreak, including limiting cases of very small or very large behavioural switching rates. The authors showed numerically that incorporating behavioural changes can lead to a lower \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 compared with a model with static behaviour. Building on this framework, subsequent studies (Brito Da Cruz and Rodrigues 2021; Roosa and Fefferman 2022) have proposed models of increasing complexity to explore more elaborate scenarios involving behavioural adaptation. However, most of these contributions rely predominantly on numerical simulations and provide limited analytical insight into the impact of (possibly information-induced) protective behaviour on epidemic spread, beyond considerations based on the basic reproduction number. The idea of information-dependent behavioural switching in vector–host models has also been explored in Misra et al. (2013) and, more recently, in Hu and Nie (2023), where it is assumed that only the susceptible host subpopulation is divided into protected (with perfect protection) and unprotected individuals, with behavioural changes driven by exponentially waning information on infection prevalence. Incorporating host demography and thus focusing on an endemic setting, the authors showed that an endemic equilibrium exists whenever \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0>1$$\end{document}R0>1 and may become unstable via a Hopf bifurcation under suitable conditions, leading to the emergence of periodic solutions. The possibility of oscillatory, or even more complex, dynamics in epidemic models with information-induced behavioural responses has been widely discussed in the literature; see, for instance, Andò et al. (2026), d’Onofrio and Manfredi (2009), d’Onofrio and Manfredi (2022), Manfredi and d’Onofrio (2013), Poletti et al. (2009), Zhang et al. (2023). Motivated by the recent rise of autochthonous dengue cases in Europe (European Centre for Disease Prevention and Control 2025), we propose a model for a mosquito-borne outbreak in which host individuals may adopt protective behaviour in response to the information available on the current state of the epidemic. Protective behaviour may consist, e.g. in using repellents, wearing long-sleeved clothes, using screens on windows and doors, or using air conditioning (we neglect possible deadly effects on the vector population deriving from the use of such measures); see Centers for Disease Control and Prevention (2026). In contrast to Misra et al. (2013), Hu and Nie (2023), we focus on an outbreak scenario and therefore neglect host demography, waning immunity, and seasonal effects in the vector population. In this outbreak scenario, the epidemic will certainly go extinct; see, e.g. Propositions 6 and 10 below. Hence, the focus of our analysis is on characterising the transient dynamics before extinction. Following Miller et al. (2016), we assume that the host population is divided into protected and unprotected subclasses, with individuals dynamically switching behaviour as in Demers et al. (2018) at rates depending on epidemic information. To model this mechanism, we adopt the information index approach introduced by d’Onofrio, Manfredi, and collaborators, whereby the delayed influence of past infections on current perception is represented through a memory kernel, which may be either distributed or concentrated (Andò et al. 2020). This information may depend on prevalence (d’Onofrio and Manfredi 2009; d’Onofrio et al. 2007) or on incidence, namely the number of new cases per unit time (d’Onofrio and Manfredi 2022). Consistently, with the structure of the resulting system of Ordinary Differential Equations (ODEs), we assume that the human population is closed and constant in size, with no births or deaths. Furthermore, we allow vectors to feed on a non-human host population that does not contribute to transmission (Esteva and Vargas 1998; Nishiura 2006). For instance, field studies in Italy found that Aedes albopictus (main vector of dengue and chikungunya in Europe) feeds predominantly on humans in urban settings (79–96% of blood meals), whereas in rural environments humans account for only 23–55%, with horses and bovines representing common alternative hosts (Valerio et al. 2010). In the literature, it is usual to assume that epidemiological dynamics evolves either much more slowly or much more rapidly than behavioural changes in the host population; see, for example, Della Marca and Onofrio (2021), Della Marca et al. (2024), Manfredi and d’Onofrio (2013), Poletti et al. (2009). This is often motivated by the increasing role that social media play in accelerating opinion shifts and decision dynamics in different contexts, such as the prevention of infectious diseases (via, e.g. non-pharmaceutical interventions, such as social distancing, or vaccination campaigns). Indeed, several studies (e.g. Bolton et al. (2024), Urmi et al. (2025)) show that, during the COVID-19 pandemic, the number of contacts and other behaviours changed quickly in response to changes in the law or in the observed mortality. When such models are formulated as systems of ODEs, a powerful framework for analysing the qualitative behaviour under time-scale separation is the so-called Geometric Singular Perturbation Theory (GSPT), originating from the seminal work of Neil Fenichel (Fenichel 1979). GSPT has been widely applied to the study of natural systems characterised by interacting mechanisms evolving on distinct time scales; see, in particular, Hek (2010), Kuehn (2015) for biologically motivated examples. Under the assumption of separated time scales, perturbed ODE systems can be analysed in suitable singular limits that describe the fast and slow dynamics separately, and the resulting reduced problems provide insight into the behaviour of the full system. This approach has been employed in several epidemic models (Jardón-Kojakhmetov et al. 2021a, b; Kaklamanos et al. 2024), especially in settings where information (as well as misinformation and opinions) spreads faster than the disease itself (Bulai et al. 2024; Della Marca et al. 2024; Schecter 2021). In this work, we exploit GSPT to understand both the transient and the asymptotic behaviour of the system. In Sect. 4, we rigorously prove that the model in which individuals switch between protected and unprotected behaviours can be approximated, as the switching rates become very large, by model (38), in which only one type of hosts is present, but infection rate depends on the information index. In Sect. 5, we use another asymptotic analysis on model (38) that allows us to explain a slow decay in the infectious population, which is uncommon for compartmental models without behaviour dynamics, and which we observe in our base behavioural model (22), as well. It is worth noting that behavioural and epidemiological processes are not the only mechanisms operating on different time scales. In mosquito-borne diseases, demographic processes in the vector population typically occur on the same time scale as epidemic transmission and therefore cannot be neglected, whereas demographic changes in the human population, as well as waning disease-induced immunity, often evolve much more slowly and may be disregarded in early outbreak scenarios. The paper is organised as follows. In Sect. 2, we introduce the baseline vector-borne model without information and analyse its main properties and asymptotic behaviour. Section 3 incorporates information-induced behavioural changes through an information index based on a memory kernel; specialising to Erlang-distributed kernels, we derive the corresponding reproduction number. In Sect. 4, we further assume a separation of time scales and apply GSPT to investigate the transient dynamics of the system and present numerical simulations illustrating the analytical results in epidemiologically relevant scenarios. In Sect. 5, we examine the model towards the end of a first outbreak, thus assuming a low attack rate. We investigate, for selected memory kernels, the existence and stability of equilibria of the reduced system; these correspond to a slow decline of incidence in the original model. Finally, Sect. 6 summarises the main findings and outlines possible extensions of the model. Vector-host model with static protective behaviour In this section, we introduce the baseline model (a slight extension of the Ross–Macdonald framework incorporating non-human hosts). This model will later be extended to account for a division of the human population into individuals who adopt protective measures against mosquito bites and those who do not. At this stage, behaviour is assumed to be fixed. In the following section, we will allow individuals, irrespective of their epidemiological status, to switch dynamically between behavioural states. We remark that, while our model is motivated by recent outbreaks of vector-borne diseases such as dengue or chikungunya, the analysis carried out in this work does not aim at closely reproducing the dynamics of complex illnesses such as malaria or dengue. Rather, our goal is to introduce a base model which provides a useful starting point and a remarkable trade-off between modelling complexity and analytical tractability. Let H denote the total human population size. We assume H is constant over time (a closed population with no births or deaths). We also assume that the vector can take blood meals from a non-human host population whose individuals do not participate in transmission, i.e. a bite on a non-human host is “wasted” from the standpoint of infection spread. We will refer to L as the “effective” population size of “non-competent” hosts; thus, L/(H+L)L/(H+L) is the probability that a mosquito bite falls on a “non-competent” host, and H/(H+L)H/(H+L) is the probability that it falls on a human host. As for the epidemiological dynamics, we assume an SIR–SI framework. Humans follow an SIR model without demography, i.e. we assume no births or deaths, and permanent immunity after recovery. Mosquitoes follow an SI model with demography since their average lifespan is orders of magnitude shorter than that of humans. For simplicity, we neglect the incubation periods in both mosquitoes and hosts, during which individuals are infected but not yet infectious. It is well-known that including the extrinsic incubation period (the one in mosquitoes) reduces the value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0, while having only minor effects on the qualitative dynamics of the system (Macdonald 1957; Pugliese and De Reggi 2026). Thus, in the absence of protective behaviour, the dynamics is described by the following system of ODEs (Esteva and Vargas 1998; Nishiura 2006):1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} S_H' & = -bp_{H\leftarrow M}I_M \frac{S_H}{H+L},\\ I_H' & = bp_{H\leftarrow M}I_M \frac{S_{H} }{H+L} - \gamma I_{H} ,\\ R_H'& =\gamma I_H,\\ S_M' & = \Lambda -bp_{M\leftarrow H}S_M \frac{I_{H} }{H+L}-\mu S_M,\\ I_M' & = bp_{M\leftarrow H}S_M \frac{I_{H}}{H+L} -\mu I_M, \end{array} \right. \end{aligned}$$\end{document}SH′=-bpH←MIMSHH+L,IH′=bpH←MIMSHH+L-γIH,RH′=γIH,SM′=Λ-bpM←HSMIHH+L-μSM,IM′=bpM←HSMIHH+L-μIM,where we denote with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H(t),\, I_H(t),\, R_H(t)$$\end{document}SH(t),IH(t),RH(t) the number of human individuals who are susceptible, infectious, and removed, respectively, at time t≥ 0t≥0, and with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_M(t),\, I_M(t)$$\end{document}SM(t),IM(t) the number of mosquitoes who are susceptible and infectious, respectively, at time t≥ 0t≥0. The parameter b is the constant per-capita mosquito biting rate, assumed to be independent of the mosquito’s infection status. The quantities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{H\leftarrow M},\, p_{M \leftarrow H}\in [0, 1]$$\end{document}pH←M,pM←H∈[0,1] are the probabilities of transmission per bite from mosquitoes to humans and from humans to mosquitoes, respectively. The parameter γ >0γ>0 is the per-capita human recovery rate (so that the average infectious period is 1/γ 1/γ). The parameters Λ Λ and μ μ are the mosquito recruitment rate and per-capita death rate, respectively (hence the average mosquito lifespan is 1/μ 1/μ). For the mosquito population dynamics, we assume for simplicity that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda = \mu M$$\end{document}Λ=μM. This implies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M'=S_M'+I_M'=0$$\end{document}M′=SM′+IM′=0, so that the total mosquito population M remains constant. Similarly, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H'=S'_H+I'_H+R'_H=0$$\end{document}H′=SH′+IH′+RH′=0. Hence, the equation for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH can be omitted since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H(t)\equiv H-S_H(t)-I_H(t)$$\end{document}RH(t)≡H-SH(t)-IH(t), for all t≥ 0t≥0. Remark 1 Recall that L represents non-human targets for mosquito bites. An alternative interpretation of the constant L is a “penalty term” in the expression for the mean time mosquitoes spend searching for human hosts; for instance, it may reflect the spatial separation among humans. Then, by defining \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\zeta :=L^{-1}$$\end{document}ζ:=L-1 and\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hat{b}:=\frac{b H}{H+L}=\frac{b\zeta H}{1+\zeta H}, \end{aligned}$$\end{document}b^:=bHH+L=bζH1+ζH,we can interpret \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{b}$$\end{document}b^ as the effective mosquito biting rate, determined by a Holling-type II functional response, as in Demers et al. (2018), Miller et al. (2016), Yakob (2016). The force of infection acting on humans can then be written as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda _{H\leftarrow M}:=\hat{b}\ {p_{H\leftarrow M}} \frac{I_M}{H}. \end{aligned}$$\end{document}λH←M:=b^pH←MIMH.In summary, L can be viewed as accounting for the effect on the average mosquito searching time of both the presence of non-competent hosts and the spatial distribution (or mutual distance) of humans. Protective behaviour To incorporate protective behaviour into model (1), following recent works in the literature (Demers et al. 2018; Miller et al. 2016; Roosa and Fefferman 2022), we assume that the host population H can be partitioned as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H=H_{P}+H_{NP}$$\end{document}H=HP+HNP. Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{P}=pH$$\end{document}HP=pH, with p∈ [0,1]p∈[0,1], represents the subpopulation of protected individuals, i.e. those adopting protective behaviour, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{NP}=(1-p)H$$\end{document}HNP=(1-p)H represents those who do not (the subpopulation of non-protected individuals). To model the efficacy of protective behaviour, we assume that protected individuals are relatively less exposed to mosquito bites compared to non-protected individuals. Let q∈ [0, 1]q∈[0,1] denote the probability of protection failure (see, for instance, Roosa and Fefferman (2022)). The probability that a mosquito bites a protected individual in the group \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{P}$$\end{document}HP is2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} P_b^{H_P}(p,q)=\frac{q H_{P}}{q H_{P}+H_{NP}+L}=\frac{pqH}{c(p, q)H+L}, \end{aligned}$$\end{document}PbP(p,q)=qHPqHP+HNP+L=pqHc(p,q)H+L,where3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} c(p,q):=1-p(1-q). \end{aligned}$$\end{document}c(p,q):=1-p(1-q).Note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_b^{H_P}(p,q)< P_b^{H_P}(p,1)$$\end{document}PbP(p,q)P_b^{H_{NP}}(p,1),\quad \text {for}\quad p\in (0,1),\ q\in [0,1), \end{aligned}$$\end{document}PbNP(p,q):=(1-p)Hc(p,q)H+L>PbNP(p,1),forp∈(0,1),q∈[0,1),and\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{L}{c(p,q)H+L}> \frac{L}{H+L},\quad \text {for}\quad p\in (0,1),\ q\in [0,1), \end{aligned}$$\end{document}Lc(p,q)H+L>LH+L,forp∈(0,1),q∈[0,1),respectively. Then, taking human protective behaviour into account, the model (1) becomes4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} S_{P}' & = -\beta _{H\leftarrow M}I_M \frac{q S_{P} }{c(p,q)H+L},\\ I_{P}' & = \beta _{H\leftarrow M}I_M \frac{q S_{P} }{c(p,q)H+L} - \gamma I_{P} ,\\ S_{NP}' & = -\beta _{H\leftarrow M}I_M \frac{S_{NP} }{c(p,q)H+L},\\ I_{NP}' & = \beta _{H\leftarrow M}I_M \frac{S_{NP} }{c(p,q)H+L} - \gamma I_{NP},\\ I_M' & = \beta _{M\leftarrow H}(M-I_M) \frac{q I_{P} +I_{NP} }{c(p,q)H+L} -\mu I_M, \end{array} \right. \end{aligned}$$\end{document}SP′=-βH←MIMqSPc(p,q)H+L,IP′=βH←MIMqSPc(p,q)H+L-γIP,SNP′=-βH←MIMSNPc(p,q)H+L,INP′=βH←MIMSNPc(p,q)H+L-γINP,IM′=βM←H(M-IM)qIP+INPc(p,q)H+L-μIM,where now \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_P, I_P$$\end{document}SP,IP denote the susceptible and infectious individuals who adopt protective behaviour, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{NP}, I_{NP}$$\end{document}SNP,INP denote the susceptible and infectious individuals who do not. Moreover,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \beta _{H\leftarrow M}:=b p_{H\leftarrow M}, \qquad \beta _{M\leftarrow H}:=b p_{M\leftarrow H} \end{aligned}$$\end{document}βH←M:=bpH←M,βM←H:=bpM←Hare the transmission rates from mosquitoes to humans and from humans to mosquitoes, respectively. Remark 2 In system (4), we assume that protective behaviour does not substantially alter the mosquito handling time; that is, the time required for a mosquito to attempt or complete a bite is assumed to be the same for protected and non-protected individuals. Consequently, protection affects only the probability of successful biting (through the factor q), and not the biting rate itself. Remark 3 Consider the case L=0L=0. We note that neither of the limits\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lim _{(p, q)\rightarrow (1^-, 0^+)}P_b^{H_P}(p, q) \;\;\hbox { and } \;\;\lim _{(p, q)\rightarrow (1^-, 0^+)}P_b^{H_{NP}}(p, q) \end{aligned}$$\end{document}lim(p,q)→(1-,0+)PbP(p,q)andlim(p,q)→(1-,0+)PbNP(p,q)exists. This “inconsistency” in the model is expected, as we are assuming that the number of available hosts is zero while the mosquito biting rate remains fixed; that is, mosquitoes are assumed to bite a fixed number of hosts per unit time regardless of host availability. Clearly, this situation cannot occur when L>0L>0. In the following, we argue in terms of the fractions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_P/H$$\end{document}SP/H, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_P/H$$\end{document}IP/H, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{NP}/H$$\end{document}SNP/H, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_{NP}/H$$\end{document}INP/H, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_M/M$$\end{document}IM/M, which, for later convenience, will still be denoted by the same variables that have been used so far. The resulting model reads5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} S_{P}' & = -\beta _{H\leftarrow M}\rho I_M \frac{q S_{P} }{c(p,q)+l},\\ I_{P}' & = \beta _{H\leftarrow M}\rho I_M \frac{q S_{P} }{c(p,q)+l} - \gamma I_{P} ,\\ S_{NP}' & = -\beta _{H\leftarrow M}\rho I_M \frac{S_{NP} }{c(p,q)+l},\\ I_{NP}' & = \beta _{H\leftarrow M}\rho I_M \frac{S_{NP} }{c(p,q)+l} - \gamma I_{NP},\\ I_M' & = \beta _{M\leftarrow H}(1-I_M) \frac{q I_{P} +I_{NP} }{c(p,q)+l} -\mu I_M, \end{array} \right. \end{aligned}$$\end{document}SP′=-βH←MρIMqSPc(p,q)+l,IP′=βH←MρIMqSPc(p,q)+l-γIP,SNP′=-βH←MρIMSNPc(p,q)+l,INP′=βH←MρIMSNPc(p,q)+l-γINP,IM′=βM←H(1-IM)qIP+INPc(p,q)+l-μIM,where now6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} l:=\frac{L}{H},\quad \text {and}\quad \rho :=\frac{M}{H}. \end{aligned}$$\end{document}l:=LH,andρ:=MH.For the reader’s convenience, we report the model parameters and their interpretation in Table 1, while c(p, q) is given by (3). In Fig. 1, we represent the dynamics of model (5).Table 1Parameters appearing in model (5) and their interpretationsParameterInterpretationlRatio between non-competent hosts and human population sizesρ ρRatio between mosquito and human population sizespFraction of protected humansqProbability of protection failure\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta _{H\leftarrow M}$$\end{document}βH←MTransmission rate from mosquitoes to humans\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta _{M\leftarrow H}$$\end{document}βM←HTransmission rate from humans to mosquitoes1/γ 1/γAverage human infectious period1/μ 1/μAverage mosquito life span Table 1Parameters appearing in model (5) and their interpretationsParameterInterpretationlRatio between non-competent hosts and human population sizesρ ρRatio between mosquito and human population sizespFraction of protected humansqProbability of protection failure\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta _{H\leftarrow M}$$\end{document}βH←MTransmission rate from mosquitoes to humans\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta _{M\leftarrow H}$$\end{document}βM←HTransmission rate from humans to mosquitoes1/γ 1/γAverage human infectious period1/μ 1/μAverage mosquito life span Parameters appearing in model (5) and their interpretations Fig. 1Flow chart for system (5). Straight lines: compartmental movements within each population; dashed lines: infections between populations (mosquitoes infecting humans and vice versa) Flow chart for system (5). Straight lines: compartmental movements within each population; dashed lines: infections between populations (mosquitoes infecting humans and vice versa) As we are dealing with fractions, the biologically meaningful state space for the solutions of model (5) is the set\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Omega :=\{(&S_P, I_P, S_{NP}, I_{NP}, I_M) \in \mathbb {R}^5_{\ge 0}\ :\\&0\le S_P, I_P, S_{NP}, I_{NP}, I_M;\ {S_P+I_P\le p;\ S_{NP}+I_{NP}\le 1-p;}\ I_M\le 1\}, \end{aligned}$$\end{document}Ω:={(SP,IP,SNP,INP,IM)∈R≥05:0≤SP,IP,SNP,INP,IM;SP+IP≤p;SNP+INP≤1-p;IM≤1},for which the following classical result holds. Proposition 1 For every initial condition \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\textbf{X}}(0):=(S_P(0), I_P(0), S_{NP}(0), I_{NP}(0), I_M(0)) \in \Omega $$\end{document}X(0):=(SP(0),IP(0),SNP(0),INP(0),IM(0))∈Ω, the system (5) admits a unique solution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\textbf{X}}(t):=(S_P(t), I_P(t), S_{NP}(t), I_{NP}(t), I_M(t)) \in \Omega $$\end{document}X(t):=(SP(t),IP(t),SNP(t),INP(t),IM(t))∈Ω that is globally defined in the future. Furthermore, the following inequalities hold for every t≥ 0t≥0:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} S_P(t)+I_P(t)\le S_P(0)+I_P(0),\quad \text {and}\quad S_{NP}(t)+I_{NP}(t)\le S_{NP}(0)+I_{NP}(0). \end{aligned}$$\end{document}SP(t)+IP(t)≤SP(0)+IP(0),andSNP(t)+INP(t)≤SNP(0)+INP(0). Proof Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\textbf{X}}(0)\in \Omega $$\end{document}X(0)∈Ω. The local existence and uniqueness of the solution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\textbf{X}}(t)$$\end{document}X(t) to the Cauchy problem associated with (5) follow from the Cauchy–Lipschitz theorem. Now, for each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y\in \{ S_P, I_P, S_{NP}, I_{NP}, I_M\}$$\end{document}Y∈{SP,IP,SNP,INP,IM}, one has \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y'|_{Y=0}\ge 0$$\end{document}Y′|Y=0≥0, which implies that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\textbf{X}}(t)\in \mathbb {R}^5_{\ge 0}$$\end{document}X(t)∈R≥05 for all t≥ 0t≥0 for which the solution is defined. Furthermore, the inequalities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(S_j+I_j)'=-\gamma I_j\le 0,\quad j\in \{P,\ NP\}$$\end{document}(Sj+Ij)′=-γIj≤0,j∈{P,NP}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I'_M|_{I_M\ge 1} \le -\mu I_M\le 0$$\end{document}IM′|IM≥1≤-μIM≤0 hold. Hence, using standard arguments, one obtains that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\textbf{X}}(t)\in \Omega $$\end{document}X(t)∈Ω is globally defined in the future, and the inequalities in (7) hold. □ □ Finally, for later use, we introduce the removed fractions8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {R_P:=p-S_P-I_P,\quad \text {and}\quad R_{NP}:=(1-p) -S_{NP}-I_{NP},} \end{aligned}$$\end{document}RP:=p-SP-IP,andRNP:=(1-p)-SNP-INP,as well as the sums9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {S_H:=S_P+S_{NP},\quad I_H:=I_P+I_{NP},\quad \text {and}\quad R_H:=R_P+R_{NP}.} \end{aligned}$$\end{document}SH:=SP+SNP,IH:=IP+INP,andRH:=RP+RNP. Control reproduction number In epidemiological models, the BRN \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 is the expected number of secondary cases produced by a typical infected individual in a completely susceptible population during its entire infectious period (Diekmann et al. 1990). When intervention or control strategies are implemented in an otherwise fully susceptible population, the relevant threshold quantity is referred to as the Control Reproduction Number (CRN), denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c$$\end{document}Rc. This quantity shares the same threshold property as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 (Pellis et al. 2022). In this section, we compute the CRN \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c:=R_c(p,q)$$\end{document}Rc:=Rc(p,q) for the model with protective behaviour (5), following the general framework introduced in Diekmann et al. (1990). In particular, we apply the Next Generation Matrix (NGM) method described in Diekmann et al. (2010), van den Driessche and Watmough (2002). To this end, we first observe that system (5) admits the Disease-Free Equilibrium (DFE)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (S_{P}^*,\ I_{P}^*,\ S_{NP}^*,\ I_{NP}^*,\ I_M^*):=(p,\ 0,\ 1-p,\ 0,\ 0). \end{aligned}$$\end{document}(SP∗,IP∗,SNP∗,INP∗,IM∗):=(p,0,1-p,0,0).We then focus on the early stage of the epidemic. In particular, we assume that the total human susceptible fraction \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH in (9) satisfies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{H}\approx 1$$\end{document}SH≈1 and that the mosquito population is almost entirely susceptible, i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_M=1-I_M\approx 1$$\end{document}SM=1-IM≈1. We therefore consider the linearised system for the infected compartments around the DFE10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} I_{P}' & = \beta _{H\leftarrow M}\rho I_M\frac{q p}{c(p,q)+l} - \gamma I_{P} ,\\ I_{NP}' & = \beta _{H\leftarrow M} \rho I_M\frac{1-p}{c(p,q)+l} - \gamma I_{NP} ,\\ I_M' & = \beta _{M\leftarrow H}\frac{q I_{P} +I_{NP} }{c(p,q)+l} -\mu I_M, \end{array} \right. \end{aligned}$$\end{document}IP′=βH←MρIMqpc(p,q)+l-γIP,INP′=βH←MρIM1-pc(p,q)+l-γINP,IM′=βM←HqIP+INPc(p,q)+l-μIM,where ρ ρ and l are defined as in (6). Following Diekmann et al. (2010), van den Driessche and Watmough (2002), we define a matrix accounting for infection11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} B(p, q):=\begin{pmatrix} 0 & 0 & \dfrac{\rho qp\beta _{H\leftarrow M }}{c(p,q)+l}\\ 0 & 0 & \dfrac{\rho (1- p)\beta _{H\leftarrow M}}{c(p,q)+l}\\ \dfrac{q\beta _{M\leftarrow H}}{c(p,q)+l} & \dfrac{\beta _{M\leftarrow H}}{c(p,q)+l} & 0 \end{pmatrix},\quad p,q\in [0,1], \end{aligned}$$\end{document}B(p,q):=00ρqpβH←Mc(p,q)+l00ρ(1-p)βH←Mc(p,q)+lqβM←Hc(p,q)+lβM←Hc(p,q)+l0,p,q∈[0,1],and a matrix accounting for transition \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Sigma :={{\,\textrm{diag}\,}}(\gamma , \gamma , \mu )$$\end{document}Σ:=diag(γ,γ,μ), so that the Jacobian matrix of system (10) can be easily written as B(p,q)-Σ B(p,q)-Σ. Then, according to Diekmann et al. (1990), the CRN \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{c}(p, q)$$\end{document}Rc(p,q) is obtained as the spectral radius of the NGM \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K(p, q):=B(p, q)\Sigma ^{-1}$$\end{document}K(p,q):=B(p,q)Σ-1, and explicitly reads12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} R_c=R_0\; \frac{1+l}{c(p,q)+l}\;\sqrt{1-p(1-q^2)}\,, \end{aligned}$$\end{document}Rc=R01+lc(p,q)+l1-p(1-q2),where13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} R_{0}:=\frac{1}{1+l}\; \sqrt{\frac{\beta _{H\leftarrow M}\beta _{M\leftarrow H}}{\gamma \mu } \rho } \end{aligned}$$\end{document}R0:=11+lβH←MβM←Hγμρis the BRN (namely, the reproduction number in the absence of protective behaviour). Note that, if p=0p=0 (i.e. no individuals adopt protective behaviour) or q=1q=1 (i.e. protection is completely ineffective), then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c=R_0$$\end{document}Rc=R0. Moreover, it is interesting to observe that if p=1p=1 and l=0l=0, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c=R_0$$\end{document}Rc=R0. This is a direct consequence of the Ross–Macdonald-like assumptions underlying the model. Indeed, in this case, all individuals adopt protective behaviour. However, the mosquito biting rate is assumed to be fixed, meaning that mosquitoes feed on a given number of hosts per unit time regardless of host behaviour. Since l=0l=0, mosquitoes have no alternative hosts and therefore must feed on humans. Consequently, whenever protection fails, all mosquito bites are effectively concentrated on those individuals for whom protection is unsuccessful. As a result, the overall transmission potential remains unchanged. It is therefore natural to ask whether the presence of awareness in the human population reduces the reproduction number. In particular, we seek conditions ensuring that14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} R_c(p, q)< R_{0}. \end{aligned}$$\end{document}Rc(p,q)F(p,q),\qquad \text {with}\qquad F(p,q):=\frac{(1-p)(1-q)}{\sqrt{1-p(1-q^2)}+q}\,. \end{aligned}$$\end{document}l>F(p,q),withF(p,q):=(1-p)(1-q)1-p(1-q2)+q. Proof From (12), the inequality \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c(p, q)< R_{0}$$\end{document}Rc(p,q)0\iff (1-q)^2p(1-p)>0$$\end{document}1-p(1-q2)-[1-p(1-q)]>0⟺(1-q)2p(1-p)>0, which is obvious if q∈ [0, 1)q∈[0,1) and p∈ (0, 1)p∈(0,1). Hence\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sqrt{1-p+q^2 p}-(1-p+q p)> 0,\quad p\in (0,1),\quad q\in [0, 1). \end{aligned}$$\end{document}1-p+q2p-(1-p+qp)>0,p∈(0,1),q∈[0,1).As for the second term of inequality (16), it is clear that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt{1-p+q^2 p} = \sqrt{1-p(1-q^2)}<1$$\end{document}1-p+q2p=1-p(1-q2)<1, for p∈ (0, 1]p∈(0,1] and q∈ [0, 1).q∈[0,1). Hence17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} l\left[ \sqrt{1-p+q^2 p}-1\right] < 0,\qquad p\in (0, 1],\quad q\in [0, 1). \end{aligned}$$\end{document}l1-p+q2p-1<0,p∈(0,1],q∈[0,1).This shows that the presence of alternative hosts l always reduces \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c$$\end{document}Rc when protective behaviour is present. Using (17), inequality (16) can be rewritten as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$l[1-\sqrt{1-p+q^2 p}]> \sqrt{1-p+q^2 p}-(1-p+q p)$$\end{document}l[1-1-p+q2p]>1-p+q2p-(1-p+qp) which is equivalent to18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} l>\frac{\sqrt{1-p+q^2 p}-(1-p+q p)}{1-\sqrt{1-p+q^2 p}}. \end{aligned}$$\end{document}l>1-p+q2p-(1-p+qp)1-1-p+q2p.Finally, after algebraic manipulations, the right-hand side of (18) can be rewritten, for p,q∈ (0,1)p,q∈(0,1), as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{\sqrt{1-p+q^2 p}-(1-p+q p)}{1-\sqrt{1-p+q^2 p}}&=\frac{(1-p)(1-q)}{\sqrt{1-p(1-q^2)}+q}\,. \end{aligned}$$\end{document}1-p+q2p-(1-p+qp)1-1-p+q2p=(1-p)(1-q)1-p(1-q2)+q.□ □ Corollary 3 Let q∈ [0, 1)q∈[0,1) and l>0l>0. Then there exists \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{p} < 1$$\end{document}p¯<1 such that (14) holds if and only if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{p}< p < 1$$\end{document}p¯ 0Q′(1)>0 for l > 0l>0, which implies that the other root \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{p}$$\end{document}p¯ satisfies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{p}<1$$\end{document}p¯<1, and therefore \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Ap^2+Bp+C<0$$\end{document}Ap2+Bp+C<0 for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\bar{p},1)$$\end{document}(p¯,1). Finally, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{p} \le 0$$\end{document}p¯≤0 if and only if Q(0) = C ≤ 0Q(0)=C≤0, which holds precisely when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$l \ge \bar{l}$$\end{document}l≥l¯. □ □ Fig. 2The behaviour of the function F(p, q) defined in (15) The behaviour of the function F(p, q) defined in (15) The behaviour of the function F(p, q) is illustrated in Fig. 2. Note that it depends only on p and q, and is therefore independent of all other model parameters. Essentially, Proposition 2 shows that if there are not enough “other available hosts”, or, under the other interpretation of L (recall Remark 1), if human hosts are not sufficiently distanced from each other, then protective behaviour adopted by only part of the population may actually lead to an increase in the reproduction number \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0. From Proposition 2, we also immediately obtain the following corollary. Corollary 4 Let l=0l=0. Then, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c(p, q)\ge R_{0},\ \forall p\in [0,1),\ q\in (0,1].$$\end{document}Rc(p,q)≥R0,∀p∈[0,1),q∈(0,1]. In particular, defining \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r(p, q):=R_c(p, q)/R_0$$\end{document}r(p,q):=Rc(p,q)/R0, we have r(1, q)=1r(1,q)=1, r(p, q)>1r(p,q)>1, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\partial _q r(p,q) < 0$$\end{document}∂qr(p,q)<0 for all q ∈ [0,1)q∈[0,1) and p∈ (0,1)p∈(0,1). Thus, either in the absence of other hosts or under conditions of high population density, static protective behaviour in the human population always increases the reproduction number \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 when it is only partially adopted (p<1p<1) or partially effective (q>0q>0). This is consistent with the results of Dye and Hasibeder (1986); Hasibeder and Dye (1988) for vector-host models with heterogeneous host populations. In particular, Dye and Hasibeder (1986) show that, in the absence of non-competent hosts, static heterogeneity in the host population always leads to an increased value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0. Hasibeder and Dye (1988) further discuss the effect of such heterogeneity on the final epidemic size. See also the more recent work by Bolzoni et al. (2015), where, building on the results of Dye and Hasibeder (1986), Hasibeder and Dye (1988), the authors investigate the role of heterogeneity on the invasion probability of a vector-borne disease in multi-host models. Finally, it is worth observing from (12) that the presence of other hosts, i.e. l>0l>0, always decreases the value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 regardless of the level of awareness or intervention, and even in their absence (i.e. when q=1q=1 or p=0p=0). This phenomenon is commonly known (especially in the context of tick-borne infections) as the dilution effect (Rosà and Pugliese 2007). Finally, considering \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c(p, q)$$\end{document}Rc(p,q) as a function of q, it is straightforward to derive the following result. Proposition 5 Let l > 0l>0. Then, for every p∈ (0,1]p∈(0,1], the function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c(p, q)$$\end{document}Rc(p,q), viewed as a function of q, attains a minimum value, smaller than \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{0}$$\end{document}R0, at q= (1-p)/(1-p+l).q=(1-p)/(1-p+l). Proposition 5 shows that when protective behaviour is static, the largest reduction of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 is achieved at an intermediate level of protection effectiveness, 1-q1-q. Figure 3 illustrates the behaviour of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c$$\end{document}Rc as a function of q for different values of p and l.Fig. 3Behaviour of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c$$\end{document}Rc for model (5) as a function of q for different values of p and l, with ρ =2ρ=2 and epidemiological parameters as in Table 2 Behaviour of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c$$\end{document}Rc for model (5) as a function of q for different values of p and l, with ρ =2ρ=2 and epidemiological parameters as in Table 2 Asymptotic dynamics and final size In this section, we study the long-term dynamics of the solutions of system (5) and we numerically investigate the effect of static protective behaviour in the host population on the final epidemic size (Brauer 2017, 2019; Diekmann et al. 2013). We begin by observing that, since the model does not include human demography or loss of immunity, the only non-trivial equilibria of (5) are those satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_P=I_{NP}=I_M=0$$\end{document}IP=INP=IM=0. Consequently, after an outbreak, the disease is expected to die out in the long run. This result is established in the following proposition. Proposition 6 For system (5), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lim _{t\rightarrow +\infty } I_{P}(t) = \lim _{t\rightarrow +\infty } I_{NP}(t) =\lim _{t\rightarrow +\infty } I_M(t)=0.$$\end{document}limt→+∞IP(t)=limt→+∞INP(t)=limt→+∞IM(t)=0. Proof For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{P},\, S_{NP},\, I_{P},\, I_{NP}\, \in {\mathbb {R}}_{\ge 0}$$\end{document}SP,SNP,IP,INP∈R≥0, the inequalities in (7) hold. Hence, both \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_P:=S_{P}+I_{P}$$\end{document}XP:=SP+IP and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_{NP}:=S_{NP}+I_{NP}$$\end{document}XNP:=SNP+INP are non-increasing functions of t, thus they admit limits \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_{P}^ \infty ,\, X_{NP}^\infty \ge 0$$\end{document}XP∞,XNP∞≥0, respectively, for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\rightarrow +\infty $$\end{document}t→+∞. Then, we observe that\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} -\infty < X_P^\infty -X_P(0) =-\gamma \int _0^{+\infty } I_{P}(t)\,\!\textrm{d}t, \end{aligned}$$\end{document}-∞R_0$$\end{document}Rc≈2.24>R0. Observe that, although protective behaviour leads to an increased value of the reproduction number, the final epidemic size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R^\infty _H$$\end{document}RH∞ is smaller than for the model without protective behaviour. This indicates that personal protection against mosquito bites can help reduce the total number of infections in the long run. Yet, the bottom row of Fig. 4 shows that, while protective behaviour decreases the proportion of individuals infected in the protected subpopulation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_P^\infty $$\end{document}R^P∞), the proportion infected in the non-protected subpopulation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_{NP}^\infty $$\end{document}R^NP∞) is larger than in the model without protection. A similar phenomenon was also observed in Hasibeder and Dye (1988) in the context of a vector-borne epidemic model with heterogeneity in mosquito preferences. In contrast, in Fig. 5, where we take p=0.6p=0.6 but l=1l=1, protective behaviour not only effectively reduces the value of the reproduction number \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c\approx 1.820a(x),w(x)>0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a(0) \in \mathcal {O}(1)$$\end{document}a(0)∈O(1), a'(x)>0a′(x)>0 and w'(x)≤ 0w′(x)≤0 for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x\in [0, +\infty )$$\end{document}x∈[0,+∞). Assumption 1 models a scenario in which public health campaigns and increasing information about the current epidemic have a beneficial effect on the host population, enhancing its awareness of the disease and, consequently, its willingness to adopt self-protective measures. Such assumptions are reasonable in the case of a first large outbreak of an infectious disease for which the host population has no prior experience with past epidemics, as in the case of vector-borne diseases in temperate European countries such as Italy.Fig. 6Flow chart for model (20) Flow chart for model (20) Figure 6 provides a schematic representation of the flow of (20). Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p(t) :=H_P(t)/H$$\end{document}p(t):=HP(t)/H denote the fraction of individuals in the population adopting protective behaviour at time t (so that the fraction of those not adopting it is 1 - p1-p at all times). Using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(H_P + H_{NP})' \equiv 0$$\end{document}(HP+HNP)′≡0, it follows that (20) can be conveniently rewritten as21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p'=a(J)(1-p)-w(J)p =a(J)-[a(J)+w(J)]p. \end{aligned}$$\end{document}p′=a(J)(1-p)-w(J)p=a(J)-[a(J)+w(J)]p.Then, taking into account (21), we modify system (5) as follows:22\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} S_{P}' & = -\beta _{H\leftarrow M} \rho I_M \frac{q S_{P} }{c(p,q)+l} + a(J)S_{NP} - w(J) S_{P} ,\\ S_{NP}' & = -\beta _{H\leftarrow M} \rho I_M \frac{S_{NP} }{c(p,q)+l}-a(J)S_{NP} + w(J) S_{P} ,\\ I_{P}' & = \beta _{H\leftarrow M} \rho I_M \frac{q S_{P} }{c(p,q)+l} - \gamma I_{P} + a(J)I_{NP} - w(J) I_{P} ,\\ I_{NP}' & = \beta _{H\leftarrow M} \rho I_M \frac{S_{NP} }{c(p,q)+l} - \gamma I_{NP} - a(J)I_{NP} + w(J) I_{P} ,\\ p' & = a(J)-\left[ a(J)+w(J)\right] p,\\ I_M' & = \beta _{M\leftarrow H}(1-I_M) \frac{q I_{P} +I_{NP} }{c(p,q)+l} -\mu I_M ,\\ J(t) & =\displaystyle \int _{-\infty }^t (I_{P}(\theta ) + I_{NP}(\theta ))K(t-\theta )\,\!\textrm{d}\theta . \end{array} \right. \end{aligned}$$\end{document}SP′=-βH←MρIMqSPc(p,q)+l+a(J)SNP-w(J)SP,SNP′=-βH←MρIMSNPc(p,q)+l-a(J)SNP+w(J)SP,IP′=βH←MρIMqSPc(p,q)+l-γIP+a(J)INP-w(J)IP,INP′=βH←MρIMSNPc(p,q)+l-γINP-a(J)INP+w(J)IP,p′=a(J)-a(J)+w(J)p,IM′=βM←H(1-IM)qIP+INPc(p,q)+l-μIM,J(t)=∫-∞t(IP(θ)+INP(θ))K(t-θ)dθ.where c(p,q)=1-p(1-q)c(p,q)=1-p(1-q), recalling (3). Remark 4 A model similar to (22) was proposed in Roosa and Fefferman (2022), where the authors considered a much more complex system (including, among other features, human demography, stages of the mosquito life cycle, and additional interventions targeting the mosquito population) and allowed only susceptible individuals to change their behaviour with rates dependent on the current prevalence (i.e. in our notation, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J:=I_H$$\end{document}J:=IH). In contrast, protective behaviour in the infected and removed classes was assumed to be static. The model proposed here extends that of Roosa and Fefferman (2022) by allowing all human individuals to adjust their behaviour according to possibly delayed information on the prevalence. In the following, for the sake of simplicity, we assume, as is common in the literature (d’Onofrio and Manfredi 2022; d’Onofrio et al. 2007), that the memory kernel is given by23\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K(\theta )=\frac{k^{n}\theta ^{n-1}e^{-k \theta }}{(n-1)!},\qquad \theta \in \mathbb {R}_{\ge 0},\quad n\in \mathbb {N},\quad k\in (0,+\infty ), \end{aligned}$$\end{document}K(θ)=knθn-1e-kθ(n-1)!,θ∈R≥0,n∈N,k∈(0,+∞),that is, K is the density function of an Erlang distribution with shape parameter n, mean φ = n/kφ=n/k (where k is the rate parameter), and standard deviation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma = \sqrt{n}/k$$\end{document}σ=n/k. This “simplifying” assumption on K allows us to use the linear chain trick (Macdonald 1978; MacDonald 1989) to reduce the equation for J to a system of ODEs (Diekmann et al. 2018). In particular, for n=1n=1, one can rewrite the equation for J as24\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} J' =k\left( I_{P} + I_{NP} \right) -k J, \end{aligned}$$\end{document}J′=kIP+INP-kJ,while for n≥ 1n≥1, it can be reduced to the following system of ODEs:25\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} Z_1' & =k\left( I_{P} + I_{NP} \right) -k Z_1, \\ Z_i' & =kZ_{i-1} -kZ_i ,\qquad \qquad \qquad i=2,\dots , n, \end{array} \right. \end{aligned}$$\end{document}Z1′=kIP+INP-kZ1,Zi′=kZi-1-kZi,i=2,⋯,n,with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_n =J$$\end{document}Zn=J. In the following, for the sake of generality, we avoid using (24) and refer only to (25), with the convention that, for n=1n=1, system (25) implicitly coincides with (24) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_1=Z_{n-1}=J$$\end{document}Z1=Zn-1=J. Moreover, we refer to φ φ as the “delay”, as it qualitatively represents the lag between the current time and the peak of the memory kernel. Note that, as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k\rightarrow +\infty $$\end{document}k→+∞ and n is bounded, one has \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varphi , \sigma \rightarrow 0^+$$\end{document}φ,σ→0+; that is, K converges in distribution to a Dirac delta concentrated at 0. In this case, the information becomes instantaneous. On the other hand, if both n and k increase in such a way that φ φ remains constant, then the Erlang distribution becomes increasingly concentrated around its mean φ φ. In particular, if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n,\, k\rightarrow +\infty $$\end{document}n,k→+∞ with n/k=φ n/k=φ, then the n-Erlang distribution approaches a discrete delay at a linear rate (Andò et al. 2020), i.e. it converges to a Dirac delta concentrated at φ φ. In this case, the model reduces to one with memory concentrated at t-φ t-φ. Since we are dealing with fractions, the biologically meaningful state space for the solutions of model (22) turns out to be the set\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\begin{matrix} \Omega :=\{ (& S_P, S_{NP}, I_P, I_{NP}, p, I_M, Z_1, \dots , Z_{n-1}, J) \in \mathbb {R}^{6+n}_{\ge 0}\ :\\ & 0\le S_P, S_{NP}, I_P, I_{NP}, p, I_M, Z_i, J\le 1,\quad i=1,\dots , n-1,\\ & {S_P+I_P\le p,\ S_{NP}+I_{NP}\le 1-p}\}. \end{matrix}} \end{aligned}$$\end{document}Ω:={(SP,SNP,IP,INP,p,IM,Z1,⋯,Zn-1,J)∈R≥06+n:0≤SP,SNP,IP,INP,p,IM,Zi,J≤1,i=1,⋯,n-1,SP+IP≤p,SNP+INP≤1-p}.The following result holds. Proposition 7 For every initial condition\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}X(0):=(S_P(0), S_{NP}(0), I_P(0), I_{NP}(0), p(0), I_M(0), Z_1(0),\dots , Z_{n-1}(0), J(0)) \in \Omega ,\end{aligned}$$\end{document}X(0):=(SP(0),SNP(0),IP(0),INP(0),p(0),IM(0),Z1(0),⋯,Zn-1(0),J(0))∈Ω,system (22) admits a unique solution\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X(t):=(S_P(t), S_{NP}(t), I_P(t), I_{NP}(t), p(t), I_M(t), Z_1(t),\dots , Z_{n-1}(t), J(t)) \in \Omega ,$$\end{document}X(t):=(SP(t),SNP(t),IP(t),INP(t),p(t),IM(t),Z1(t),⋯,Zn-1(t),J(t))∈Ω,which is globally defined in the future. Furthermore, the inequality \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H(t)+I_H(t)\le S_H(0)+I_H(0)$$\end{document}SH(t)+IH(t)≤SH(0)+IH(0) holds for every t≥ 0t≥0. Proof Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X(0)\in \Omega $$\end{document}X(0)∈Ω. The local existence and uniqueness of the solution X(t) of the Cauchy problem relevant to (22) follows from the Cauchy–Lipschitz theorem. Now, for each\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y\in \{ S_P, S_{NP}, I_P, I_{NP}, p, I_M, Z_1, \dots , Z_{n-1}, J\},$$\end{document}Y∈{SP,SNP,IP,INP,p,IM,Z1,⋯,Zn-1,J},one has \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y'|_{Y=0}\ge 0$$\end{document}Y′|Y=0≥0, which implies that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X(t)\in \mathbb {R}^{6+n}_{\ge 0}$$\end{document}X(t)∈R≥06+n for all t≥ 0t≥0 for which the solution is defined. Furthermore, the inequalities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(S_H+I_H)'=-\gamma I_H\le 0$$\end{document}(SH+IH)′=-γIH≤0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p'|_{p\ge 1}\le -\omega (J) p\le 0$$\end{document}p′|p≥1≤-ω(J)p≤0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I'_M|_{I_M\ge 1} \le -\mu I_M\le 0$$\end{document}IM′|IM≥1≤-μIM≤0 imply that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_P, S_{NP}, I_P, I_{NP}, p, I_M\le 1$$\end{document}SP,SNP,IP,INP,p,IM≤1. From this, one also gets \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_P'|_{R_P=0}\ge 0$$\end{document}RP′|RP=0≥0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{NP}'|_{R_{NP}=0}\ge 0$$\end{document}RNP′|RNP=0≥0 for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_P$$\end{document}RP and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{NP}$$\end{document}RNP defined as in (8), which give \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_P+I_P\le p$$\end{document}SP+IP≤p and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{NP}+I_{NP}\le 1-p$$\end{document}SNP+INP≤1-p, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_i'|_{Z_i\ge 1}\le 0$$\end{document}Zi′|Zi≥1≤0 for i=1,… , ni=1,⋯,n (recall \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_n=J$$\end{document}Zn=J), which gives \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z_i\le 1$$\end{document}Zi≤1. Hence, standard arguments give that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X(t)\in \Omega $$\end{document}X(t)∈Ω is globally defined in the future and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H(t)+I_H(t)\le S_H(0)+I_H(0)$$\end{document}SH(t)+IH(t)≤SH(0)+IH(0) for all t≥ 0t≥0. □ □ Control reproduction number As for the model with static behaviour, in this section we compute the CRN \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c$$\end{document}R^c for model (22). For any \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_0:=a(0)>0$$\end{document}a0:=a(0)>0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0:=w(0)>0$$\end{document}w0:=w(0)>0, the model (22) admits the DFE26\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left( S_{P}^*, S_{NP}^*, I_{P}^*, I_{NP}^*, p^*, I_M^*, J^*\right) :=\left( p_0, 1-p_0, 0, 0, p_0, 0, 0\right) , \end{aligned}$$\end{document}SP∗,SNP∗,IP∗,INP∗,p∗,IM∗,J∗:=p0,1-p0,0,0,p0,0,0,where27\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p_0:=\frac{a_0}{a_0+w_0}. \end{aligned}$$\end{document}p0:=a0a0+w0.Linearising the equations for the infected individuals around the DFE (26), we obtain\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} I_{P}' & = \beta _{H\leftarrow M}\rho I_M \frac{p_0q}{c(p_0,q)+l} - \gamma I_{P} + a_0I_{NP} - w_0 I_{P} ,\\ I_{NP}' & = \beta _{H\leftarrow M} \rho I_M \frac{1-p_0}{c(p_0,q)+l} - \gamma I_{NP} - a_0I_{NP} + w_0 I_{P} ,\\ I_M' & = \beta _{M\leftarrow H} \frac{q I_{P} +I_{NP} }{c(p_0,q)+l} -\mu I_M. \end{array} \right. \end{aligned}$$\end{document}IP′=βH←MρIMp0qc(p0,q)+l-γIP+a0INP-w0IP,INP′=βH←MρIM1-p0c(p0,q)+l-γINP-a0INP+w0IP,IM′=βM←HqIP+INPc(p0,q)+l-μIM.Then, considering an infection matrix B defined as in (11) and the transition matrix28\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hat{\Sigma }:=\begin{pmatrix} \gamma + w_0 & -a_0 & 0\\ -w_0 & \gamma + a_0 & 0\\ 0 & 0 & \mu \end{pmatrix}, \end{aligned}$$\end{document}Σ^:=γ+w0-a00-w0γ+a0000μ,the CRN \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c=\hat{R}_c(p_0, q)$$\end{document}R^c=R^c(p0,q) is obtained as the spectral radius of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{K}:=B\hat{\Sigma }^{-1}$$\end{document}K^:=BΣ^-1, and, for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_0$$\end{document}p0 as in (27), it reads29\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hat{R}_c(p_0, q) = R_0\, \frac{1+l}{c(p_0,q) + l}\, \sqrt{ \Delta }, \end{aligned}$$\end{document}R^c(p0,q)=R01+lc(p0,q)+lΔ,where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \Delta = \frac{qp_0\left( a_0 q+\gamma q+w_0\right) +(1-p_0)\left( a_0q+w_0+\gamma \right) }{a_0+w_0+\gamma }. $$\end{document}Δ=qp0a0q+γq+w0+(1-p0)a0q+w0+γa0+w0+γ.Note that Δ Δ is non-negative for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_0,\, w_0\in (0,+\infty )$$\end{document}a0,w0∈(0,+∞); hence, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c(p_0, q)\ge 0$$\end{document}R^c(p0,q)≥0 is well-defined. In particular, the following result holds. Proposition 8 Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_c$$\end{document}Rc be defined as in (12) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_0,\, w_0\in (0, +\infty )$$\end{document}a0,w0∈(0,+∞). Then30\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hat{R}_c(p_0, q)2.12\approx R_0$$\end{document}Rc≈2.24>2.12≈R0. Let us take \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1,\ 1,\ 10,\ 100$$\end{document}w0=0.1,1,10,100 and let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_0=w_0p/(1-p)$$\end{document}a0=w0p/(1-p), so that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_0:=p$$\end{document}p0:=p. Then (29) gives \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c\approx 2.00$$\end{document}R^c≈2.00 for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c\approx 1.83$$\end{document}R^c≈1.83 for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c\approx 1.79$$\end{document}R^c≈1.79 for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=10,\ 100$$\end{document}w0=10,100, while the limit in (32) gives \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c\approx 1.79$$\end{document}R^c≈1.79. Note that for all these choices of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0$$\end{document}w0, one has \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c< R_00w(x)>0 for x≥ 0x≥0, we obtain\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} h(p(x), q)=\frac{w(x)+qa(x)}{[w(x)+qa(x)]+l[w(x)+a(x)]}=\frac{1}{1+l\frac{w(x)+a(x)}{w(x)+qa(x)}}\,. \end{aligned}$$\end{document}h(p(x),q)=w(x)+qa(x)[w(x)+qa(x)]+l[w(x)+a(x)]=11+lw(x)+a(x)w(x)+qa(x).Assumption 1 also ensures that\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p'(x)=\frac{a'(x)w(x)-a(x)w'(x)}{[a(x)+w(x)]^2}>0,\qquad x\in [0, +\infty ), \end{aligned}$$\end{document}p′(x)=a′(x)w(x)-a(x)w′(x)[a(x)+w(x)]2>0,x∈[0,+∞),thus40\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \partial _p h(p(x), q)\cdot p'(x)=-\frac{l(1-q)p'(x)}{[1-(1-q)p(x)+l]^2}<0,\qquad x\in [0,+\infty ). \end{aligned}$$\end{document}∂ph(p(x),q)·p′(x)=-l(1-q)p′(x)[1-(1-q)p(x)+l]2<0,x∈[0,+∞). Remark 5 We have h(p(x), q)>0h(p(x),q)>0 for q∈ [0, 1)q∈[0,1) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x\in [0, +\infty )$$\end{document}x∈[0,+∞). Thus, equation (40) ensures that h(p(x), q) admits limit for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x\rightarrow +\infty $$\end{document}x→+∞, and the largest reduction for the transmission rates \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta _{H\leftarrow M},\,\beta _{M\leftarrow H}$$\end{document}βH←M,βM←H is obtained at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{h}:=\lim _{x\rightarrow +\infty } h(p(x), q)$$\end{document}h¯:=limx→+∞h(p(x),q). Lastly, we establish the following result on the asymptotic behaviour of the solutions of system (38), which follows directly by arguing as in the proof of Proposition 6 and using Proposition 7. Proposition 10 For system (38), one has \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lim _{\tau \rightarrow +\infty } I_H(\tau ) =\lim _{\tau \rightarrow +\infty } I_M(\tau ) =0$$\end{document}limτ→+∞IH(τ)=limτ→+∞IM(τ)=0. Consequently, as expected, the dynamics of (38) always ends with epidemic extinction, independently of the hosts’ protective behaviour. However, we are interested in the transient behaviour of the solutions of system (38), which is far from trivial due to the absence of non-trivial equilibria (i.e. with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H\ne 0$$\end{document}IH≠0 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_M\ne 0$$\end{document}IM≠0). Numerical simulations In this section, we present numerical experiments for the proposed model with behavioural changes, comparing the outputs of model (22) with those of the reduced one-group model (38). The purpose of the following simulations is to assess which dynamical features of the two-group model are reproduced by the reduced one-group model as the behavioural switching rates increase. In particular, we compare epidemic peak size and timing, the overall infected fraction, and convergence to the critical manifold. For the simulations, we use epidemiological parameters appropriate for a dengue outbreak, listed in Table 2. In addition, we consider l=0.25l=0.25 and l=1l=1, and we vary ρ ρ to preserve the value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0\approx 2.12$$\end{document}R0≈2.12, recalling (13). More precisely, we take ρ =2ρ=2 for l=0.25l=0.25 and ρ = 5.12ρ=5.12 for l=1l=1. We assume that, at the beginning of the outbreak, the human population is at the opinion equilibrium, i.e. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p(0)=p_0$$\end{document}p(0)=p0, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_0$$\end{document}p0 defined as in (27). As for the rates a and w, we consider the following forms:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} a(x)=a_0+\alpha x,\quad a_0,\ \alpha> 0,\quad \text {and}\quad w(x)\equiv w_0,\quad w_0>0. \end{aligned}$$\end{document}a(x)=a0+αx,a0,α>0,andw(x)≡w0,w0>0.We then vary \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0>0$$\end{document}w0>0, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_0=w_0$$\end{document}a0=w0, so that, according to (27), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_0=0.5$$\end{document}p0=0.5, i.e. at the beginning of the epidemic, 50%50% of the host population adopts personal protective measures against mosquito bites, possibly an optimistic assumption. We set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha = \chi \, w_0$$\end{document}α=χw0, where χ >0χ>0 accounts for both the effect of PHS campaigns and the human reaction to information. Finally, we vary q∈ [0, 1)q∈[0,1), the shape parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n\in \mathbb {N}$$\end{document}n∈N, and the delay φ φ of the Erlang distribution in (23). Since we observe the outbreak from its very beginning, we consider the initial conditions\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}\begin{gathered} S_P(0)=p_0,\quad S_{NP}(0)=1-S_P(0),\quad I_P(0)=I_{NP}(0)=0,\quad I_M(0)=10^{-4}. \end{gathered}\end{aligned}$$\end{document}SP(0)=p0,SNP(0)=1-SP(0),IP(0)=INP(0)=0,IM(0)=10-4.The numerical results are obtained using the MATLAB built-in ODE solvers ode45 and ode23s with standard tolerances (the latter is employed when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0\ge 10$$\end{document}w0≥10). Remark 6 The term \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0$$\end{document}w0 in α α is used to normalise the human response to information with respect to the magnitude of the reaction term \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0$$\end{document}w0. To clarify the choice of α α, let us consider the case \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J=I_H$$\end{document}J=IH, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH is defined as in (9). In this case, taking \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104 yields, for the one-group model (36), that a(x)=2w(x)a(x)=2w(x) exactly when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x=J=I_H=10^{-4}$$\end{document}x=J=IH=10-4, since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_0=w_0$$\end{document}a0=w0. Hence, for instance, if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H=10^5$$\end{document}H=105, then a(x)=2w(x)a(x)=2w(x) when the total number of infected humans at t≥ 0t≥0 is equal to 10. Moreover, for these choices of a and w, it is straightforward to verify that\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} h(p(x), q)=\frac{1}{2}h(0, q)\iff x =\frac{a_0q(1+2l)+l(w_0-a_0)+w_0}{\alpha [l-q(1+2l)]},\quad l\ne q(1+2l). \end{aligned}$$\end{document}h(p(x),q)=12h(0,q)⟺x=a0q(1+2l)+l(w0-a0)+w0α[l-q(1+2l)],l≠q(1+2l).In particular, in the case of perfect protection, q=0q=0, if l=0.25l=0.25, the “contact rates” between mosquitoes and humans are halved exactly when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H=4\cdot 10^{-4}$$\end{document}IH=4·10-4 (i.e. when the total number of infected individuals is 40), while for q=0.1q=0.1 (90%90% of efficacy), one has \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H=1.2\cdot 10^{-3}$$\end{document}IH=1.2·10-3. Fig. 7Comparison between (22) (Two-group model) and (38) (One-group model) with model parameters and initial conditions as in Sect. 4.1 with l=0.25l=0.25, ρ =2ρ=2, q=0q=0 (perfect protection), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, n=1n=1, φ =20φ=20 days, and from bottom to top, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1, 1, 10, 100$$\end{document}w0=0.1,1,10,100 (which give \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c\approx 2.17,\ 1.85,\ 1.78,\ 1.77,$$\end{document}R^c≈2.17,1.85,1.78,1.77, respectively). Left: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (defined as in (9)) as a function of time for model (22) and (38). Right: proportions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_P/S_H$$\end{document}SP/SH, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_P/I_H$$\end{document}IP/IH, p and a(J)/[a(J)+w(J)]a(J)/[a(J)+w(J)] as functions of time. Note that in the plots for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=10, 100$$\end{document}w0=10,100, the line corresponding to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_P/S_H$$\end{document}SP/SH is perfectly overlapped with those relevant to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_P/I_H, p$$\end{document}IP/IH,p and a(J)/[a(J)+w(J)]a(J)/[a(J)+w(J)], which indicates a fast convergence to the manifold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {C}_0$$\end{document}C0 in (37) Comparison between (22) (Two-group model) and (38) (One-group model) with model parameters and initial conditions as in Sect. 4.1 with l=0.25l=0.25, ρ =2ρ=2, q=0q=0 (perfect protection), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, n=1n=1, φ =20φ=20 days, and from bottom to top, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1, 1, 10, 100$$\end{document}w0=0.1,1,10,100 (which give \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c\approx 2.17,\ 1.85,\ 1.78,\ 1.77,$$\end{document}R^c≈2.17,1.85,1.78,1.77, respectively). Left: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (defined as in (9)) as a function of time for model (22) and (38). Right: proportions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_P/S_H$$\end{document}SP/SH, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_P/I_H$$\end{document}IP/IH, p and a(J)/[a(J)+w(J)]a(J)/[a(J)+w(J)] as functions of time. Note that in the plots for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=10, 100$$\end{document}w0=10,100, the line corresponding to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_P/S_H$$\end{document}SP/SH is perfectly overlapped with those relevant to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_P/I_H, p$$\end{document}IP/IH,p and a(J)/[a(J)+w(J)]a(J)/[a(J)+w(J)], which indicates a fast convergence to the manifold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {C}_0$$\end{document}C0 in (37) We start by comparing the outputs of (22) with those of (38) for different orders of magnitude of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0$$\end{document}w0. In Fig. 7, we take l=0.25l=0.25 and ρ =2ρ=2, and we observe that, when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=10$$\end{document}w0=10 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text {days}^{-1}$$\end{document}days-1 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=100$$\end{document}w0=100 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text {days}^{-1}$$\end{document}days-1 (i.e. it is from two to three orders of magnitude larger than γ γ), the results for the two models are (almost) indistinguishable. In particular, note that in the plots for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=10, 100$$\end{document}w0=10,100 in the right column, the line corresponding to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_P/S_H$$\end{document}SP/SH perfectly overlaps with those corresponding to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_P/I_H, p$$\end{document}IP/IH,p and a(J)/[a(J)+w(J)]a(J)/[a(J)+w(J)], as the solution of (2) almost immediately converges to the critical manifold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {C}_0$$\end{document}C0 in (37). Interestingly, even for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, the simulation of system (38) provides a good approximation of model (22), although the convergence to the manifold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {C}_0$$\end{document}C0 is much slower in this case, leading to significantly different timing and magnitudes of the (first) epidemic peak. Finally, we can observe that for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1 (i.e. of the same order of magnitude as γ γ), the dynamics of the two models differ substantially, as the solution is still far from converging to the manifold (37). This is also the only case, among those considered here, for which \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_c\approx 2.17>2.12 \approx R_0 $$\end{document}R^c≈2.17>2.12≈R0; see the caption of Fig. 7. Hence, in this case, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0$$\end{document}w0 is not large enough to prevent protective behaviour from increasing the probability of outbreak. Motivated by the results shown in Fig. 7, in the following, we consider the outputs of model (22) with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, and compare them with those of (38). In particular, we take \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, φ =30φ=30 days, and vary the shape parameter n of the Erlang distribution in (23), considering n=1,~2,~5n=1,2,5.Fig. 8Plots for (22) (red for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, yellow for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1) and (38) (blue, one-group model) with model parameters and initial conditions as in Sect. 4.1, l=0.25l=0.25, ρ =2ρ=2, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104 for, from top to bottom, q=0, 0.1,\ 0.2q=0,0.1,0.2, with n=1n=1 and φ =30φ=30 days in (23). Left and centre: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH as a function of time. Right: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH as a function of time (colour figure online)Fig. 9Plots for (22) (red for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, yellow for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1) and (38) (blue, one-group model) with model parameters and initial conditions as in Sect. 4.1, l=0.25l=0.25, ρ =2ρ=2, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104 for, from top to bottom, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q=0,\ 0.1,\ 0.2$$\end{document}q=0,0.1,0.2, with n=2n=2 and φ =30φ=30 days in (23). Left and centre: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH as a function of time. Right: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH as a function of time (colour figure online)Fig. 10Plots for (22) (red for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, yellow for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1) and (38) (blue, one-group model) with model parameters and initial conditions as in Sect. 4.1, l=0.25l=0.25, ρ =2ρ=2, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104 for, from top to bottom, q=0, 0.1,\ 0.2q=0,0.1,0.2, with n=5n=5 and φ =30φ=30 days in (23). Left: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH as a function of time. Right: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH as a function of time (colour figure online) Plots for (22) (red for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, yellow for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1) and (38) (blue, one-group model) with model parameters and initial conditions as in Sect. 4.1, l=0.25l=0.25, ρ =2ρ=2, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104 for, from top to bottom, q=0, 0.1,\ 0.2q=0,0.1,0.2, with n=1n=1 and φ =30φ=30 days in (23). Left and centre: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH as a function of time. Right: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH as a function of time (colour figure online) Plots for (22) (red for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, yellow for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1) and (38) (blue, one-group model) with model parameters and initial conditions as in Sect. 4.1, l=0.25l=0.25, ρ =2ρ=2, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104 for, from top to bottom, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q=0,\ 0.1,\ 0.2$$\end{document}q=0,0.1,0.2, with n=2n=2 and φ =30φ=30 days in (23). Left and centre: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH as a function of time. Right: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH as a function of time (colour figure online) Plots for (22) (red for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, yellow for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1) and (38) (blue, one-group model) with model parameters and initial conditions as in Sect. 4.1, l=0.25l=0.25, ρ =2ρ=2, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104 for, from top to bottom, q=0, 0.1,\ 0.2q=0,0.1,0.2, with n=5n=5 and φ =30φ=30 days in (23). Left: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH as a function of time. Right: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH as a function of time (colour figure online) In Figs. 8, 9, and 10, we plot \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (left and centre) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH (right), defined as in (9), as functions of time for n=1, 2, 5n=1,2,5, respectively, and for three different values of q, namely q=0q=0 (upper row), q=0.1q=0.1 (middle row), and q=0.2q=0.2 (lower row). We observe that for q=0q=0 and q=0.1q=0.1, information-induced behavioural changes effectively prevent a large portion of infections for all choices of n (note the different scales on the vertical axes for the case q=0q=0 and q=0.1q=0.1). Interestingly, in the first and second rows, after the first epidemic peak, the solutions for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH appear to enter a quasi-stationary state, although the model does not admit any non-trivial equilibria (i.e. with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H, I_M\ne 0$$\end{document}IH,IM≠0). Notably, this phenomenon is entirely due to the behavioural adaptation in response to the epidemic and does not correspond to convergence of the solutions of (22) and (38) to true equilibria of their respective systems. Indeed, observe that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH increases in all cases, indicating that no stationary situation has been reached by the solution, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH will inevitably go extinct in the long run, as shown in Proposition 10. Furthermore, when the memory distribution is more concentrated around its mean (and hence further in the past), as in Figs. 9 and 10, we observe the possible emergence of multiple epidemic waves. Notably, this occurs precisely in the cases where behavioural changes effectively prevent a large proportion of infections, namely for q=0q=0 and q=0.1q=0.1. On the other hand, for q=0.2q=0.2, although the adoption of protective measures still contributes to containing the epidemic, a larger outbreak is observed compared to the cases q=0q=0 and q=0.1q=0.1, and no oscillatory behaviour or quasi-stationary regimes arise.Fig. 11Plots for (22) (red for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, yellow for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1) and (38) (blue, one-group model) with model parameters and initial conditions as in Sect. 4.1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, q=0.5q=0.5, n=1n=1, and φ =30φ=30 days in (23). Left: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH as a function of time. Right: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH as a function of time. The dashed lines represent the same simulations for the model without protective behaviour (q=1q=1). Upper row: l=0.25l=0.25 and ρ =2ρ=2. Lower row: l=1l=1 and ρ = 5.12ρ=5.12 (colour figure online) Plots for (22) (red for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=1$$\end{document}w0=1, yellow for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_0=0.1$$\end{document}w0=0.1) and (38) (blue, one-group model) with model parameters and initial conditions as in Sect. 4.1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, q=0.5q=0.5, n=1n=1, and φ =30φ=30 days in (23). Left: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH as a function of time. Right: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_H$$\end{document}RH as a function of time. The dashed lines represent the same simulations for the model without protective behaviour (q=1q=1). Upper row: l=0.25l=0.25 and ρ =2ρ=2. Lower row: l=1l=1 and ρ = 5.12ρ=5.12 (colour figure online) Finally, in Fig. 11 we compare the outputs of models (22) and (38) for l=0.25l=0.25 and l=1l=1 (hence with ρ =2ρ=2 ad ρ =5.12ρ=5.12, respectively) and q=0.5q=0.5 (i.e. a 50% probability of protection failure). In this case, we observe, in both the upper and lower rows, that the rates at which individuals change their behaviour in response to information do not significantly affect the overall dynamics. Nevertheless, both scenarios show that the model with protective behaviour leads to a smaller final size compared to the model without protection (q=1q=1). In particular, the model with l=1l=1 predicts a smaller final size than the model with l=0.25l=0.25. Behaviour-induced dynamics in the low attack rate regime In this section, motivated by the results of Sect. 4, we investigate the dynamics of model (38) in a scenario where interventions and behavioural changes have successfully prevented a large proportion of infections before herd immunity is achieved and seasonality becomes dominant. In this setting, the susceptible populations are still sufficiently large, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H\ll S_H$$\end{document}IH≪SH and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_M\ll S_M$$\end{document}IM≪SM, and only a small amount of information J is left in the system (for instance, after the first epidemic peak has occurred) (Zhang et al. 2023). To model this low attack rate scenario (see d’Onofrio et al. (2021)), we assume that there exists \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{t}\ge 0$$\end{document}t¯≥0 such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H(t),\, I_M(t),\, J(t)\in \mathcal {O}(\varepsilon )$$\end{document}IH(t),IM(t),J(t)∈O(ε), for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0<\varepsilon \ll 1$$\end{document}0<ε≪1 for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\ge \bar{t}$$\end{document}t≥t¯, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(t)\in \mathcal {O}(1)$$\end{document}S(t)∈O(1), and the functions a,\ wa,w satisfy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a(x)=\bar{a}(\varepsilon ^{-1} x)$$\end{document}a(x)=a¯(ε-1x) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w(x)=\bar{w}(\varepsilon ^{-1} x)$$\end{document}w(x)=w¯(ε-1x). Note that the latter condition, together with Assumption 1, implies that information associated with a very small infected fraction may still induce a substantial protective response (recall Assumption 1 and note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x \in \mathcal {O}(\varepsilon )$$\end{document}x∈O(ε) might still have a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathcal {O}(1)$$\end{document}O(1) impact on the dynamics). We then introduce rescaled variables \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{I}_H = \varepsilon ^{-1} I_H $$\end{document}I¯H=ε-1IH, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{I}_M=\varepsilon ^{-1} I_M$$\end{document}I¯M=ε-1IM and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{J} =\varepsilon ^{-1} J$$\end{document}J¯=ε-1J. With a slight abuse of notation, we retain the same symbols to avoid introducing additional variables or parameters (for instance, α α will denote \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha /\varepsilon $$\end{document}α/ε). The resulting model reads\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} \dot{S}_H & = -\beta _{H\leftarrow M}\rho \varepsilon I_M h(p(J), q)S_H ,\\ \varepsilon \dot{I}_H & = \beta _{H\leftarrow M} \rho \varepsilon I_M h(p(J), q)S_H - \gamma \varepsilon I_H ,\\ \varepsilon \dot{I}_M & = \beta _{M\leftarrow H}(1- \varepsilon I_M) h(p(J), q) \varepsilon I_H -\mu \varepsilon I_M,\\ \varepsilon \dot{Z}_1 & =k \varepsilon I_H -k \varepsilon Z_1, \\ \varepsilon \dot{Z}_i & = k \varepsilon Z_{i-1} - k \varepsilon Z_i, \qquad i=2,\dots , n-1,\\ \varepsilon \dot{J} & =k\varepsilon Z_{n-1} -k \varepsilon J, \end{array} \right. \end{aligned}$$\end{document}S˙H=-βH←MρεIMh(p(J),q)SH,εI˙H=βH←MρεIMh(p(J),q)SH-γεIH,εI˙M=βM←H(1-εIM)h(p(J),q)εIH-μεIM,εZ˙1=kεIH-kεZ1,εZ˙i=kεZi-1-kεZi,i=2,⋯,n-1,εJ˙=kεZn-1-kεJ,Then, by simplifying ɛ ε in every equation except the first one and taking the limit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon \rightarrow 0^+$$\end{document}ε→0+, we obtain the corresponding layer system41\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} \dot{I}_H & = \beta _{H\leftarrow M}\rho I_M h(p(J), q) S_H - \gamma I_H ,\\ \dot{I}_M & = \beta _{M\leftarrow H}h(p(J),q) I_H -\mu I_M,\\ \dot{Z}_1 & =k I_H -k Z_1 \\ \dot{Z}_i & = k Z_{i-1} - k Z_i , \qquad i=2,\dots , n-1,\\ \dot{J} & =kZ_{n-1} -k J, \end{array} \right. \end{aligned}$$\end{document}I˙H=βH←MρIMh(p(J),q)SH-γIH,I˙M=βM←Hh(p(J),q)IH-μIM,Z˙1=kIH-kZ1Z˙i=kZi-1-kZi,i=2,⋯,n-1,J˙=kZn-1-kJ,with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{S}_H=0$$\end{document}S˙H=0. Observe that the resulting systems resemble the model studied in Zhang et al. (2023), and subsequently extended in Andò et al. (2026), for outbreaks of a directly transmitted infection under a low-attack-rate assumption. In the following section, we will investigate the existence and stability of the equilibria of (41) to gain insights into the transient dynamics of models (22) and (38). Equilibria In this section, we investigate stationary solutions of system (41). Convergence to these equilibria corresponds to a rapid “collapsing” of the fast–slow system onto its critical manifold, in whose neighbourhood the slow dynamics play a central role. Characterising the existence of such equilibria and the associated convergence allows us to describe the transient behaviour of the system, even though the asymptotic dynamics inevitably leads to disease extinction, as shown in Proposition 10. We observe that the equilibria of system (41) are all points \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(I_H^*, I_M^*, J^*)\in \mathbb {R}^3_{\ge 0}$$\end{document}(IH∗,IM∗,J∗)∈R≥03 that satisfy the system of non-linear equations42\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} I_H^*& =\frac{\beta _{H\leftarrow M}}{\gamma }\rho S_H h(p(J^*), q)I_M^*,\\ I_M^* & = \frac{\beta _{M\leftarrow H}}{\mu } h(p(J^*),q) I_H^*,\\ Z_i^*& =J^* =I_H^*,\qquad i=1,\dots , n-1. \end{array} \right. \end{aligned}$$\end{document}IH∗=βH←MγρSHh(p(J∗),q)IM∗,IM∗=βM←Hμh(p(J∗),q)IH∗,Zi∗=J∗=IH∗,i=1,⋯,n-1.Substituting the second and third equations of (42) into the one for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H^*$$\end{document}IH∗, we obtain the non-linear equation43\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} I_H^* \left( \hat{R}_e^2 h(p(I_H^*),q)^2 -1\right) =0, \end{aligned}$$\end{document}IH∗R^e2h(p(IH∗),q)2-1=0,where44\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hat{R}_e:=\sqrt{\frac{\beta _{H\leftarrow M}\beta _{M\leftarrow H}}{\gamma \mu }\rho S_H} \end{aligned}$$\end{document}R^e:=βH←MβM←HγμρSHis the Effective Reproduction Number (ERN, i.e. the reproduction number when the population is not fully susceptible or partially immune (Pellis et al. 2022)) in the absence of protective behaviour (p=0p=0 or q=1q=1) and non-competent hosts (l=0l=0). From (43), it follows that (41) admits the DFE \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H^*=I_M^*=J^*=0$$\end{document}IH∗=IM∗=J∗=0. All other equilibria are determined by the solutions of the non-linear equation45\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hat{R}_e^2h(p(I_H^*),q)^2=1\,. \end{aligned}$$\end{document}R^e2h(p(IH∗),q)2=1.Note that, since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e>0$$\end{document}R^e>0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(\cdot , \cdot )\ge 0$$\end{document}h(·,·)≥0, the analysis reduces to studying the equation46\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} h(p(I_H^*),q)=\frac{1}{\hat{R}_e}\,. \end{aligned}$$\end{document}h(p(IH∗),q)=1R^e.We have the following result: Theorem 11 Let Assumption 1 hold and q∈ [0, 1)q∈[0,1). There exists a positive equilibrium of (41) if and only if47\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} 1+l<\hat{R}_e<1+\frac{l}{q}, \end{aligned}$$\end{document}1+l1}$$\end{document}R^eh(p(0),q)>1, independently of the choice of K in (19) and (23). Then, we consider the stability of the EsE, showing that it depends on the specific form of the memory kernel. In particular, for selected choices of K, we show that the EsE may either be LAS or lose its stability via Hopf, hence ensuring the possible emergence of self-sustained oscillations even in an outbreak scenario. Let52\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} f(x):=h(p(x), q),\qquad x\in [0,+\infty ), \end{aligned}$$\end{document}f(x):=h(p(x),q),x∈[0,+∞),so that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f'(x)= \partial _p h(p(x), q)\cdot p'(x)$$\end{document}f′(x)=∂ph(p(x),q)·p′(x). The linearisation of (41) around an equilibrium \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\bar{I}_H, \bar{I}_M, \bar{J})$$\end{document}(I¯H,I¯M,J¯) reads53\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \left\{ \begin{array}{rl} \dot{I}_H & = \beta _{H\leftarrow M}\rho S_Hf(\bar{J})I_M+\beta _{H\leftarrow M}\rho \bar{I}_M S_Hf'(\bar{J})J - \gamma I_H ,\\ \dot{I}_M & = \beta _{M\leftarrow H} f(\bar{J})I_H+\beta _{M\leftarrow H} f'(\bar{J}) \bar{I}_H J-\mu I_M,\\ \dot{Z}_1 & =k I_H -k Z_1 \\ \dot{Z}_i & = k Z_{i-1} - k Z_i ,\qquad \qquad \qquad i=2,\dots , n-1,\\ \dot{J} & =kZ_{n-1} -k J. \end{array} \right. \end{aligned}$$\end{document}I˙H=βH←MρSHf(J¯)IM+βH←MρI¯MSHf′(J¯)J-γIH,I˙M=βM←Hf(J¯)IH+βM←Hf′(J¯)I¯HJ-μIM,Z˙1=kIH-kZ1Z˙i=kZi-1-kZi,i=2,⋯,n-1,J˙=kZn-1-kJ.Then, for K as in (23), the Jacobian associated with system (53) is the following:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\textbf{J}=\begin{pmatrix} -\gamma & \beta _{H\leftarrow M}\rho S_Hf(\bar{J}) & 0 & 0 & \dots & \beta _{H\leftarrow M}\rho \bar{I}_M S_Hf'(\bar{J})\\ \beta _{M\leftarrow H} f(\bar{J}) & -\mu & 0 & 0 & \dots & \beta _{M\leftarrow H} f'(\bar{J}) \bar{I}_H\\ k & 0 & -k & 0 & \dots & 0\\ 0 & 0 & k & -k & \dots & 0 \\ \dots & \dots & \dots & \dots & \dots & \dots \\ 0 & 0 & 0 & \dots & k & -k \end{pmatrix}.} $$\end{document}J=-γβH←MρSHf(J¯)00⋯βH←MρI¯MSHf′(J¯)βM←Hf(J¯)-μ00⋯βM←Hf′(J¯)I¯Hk0-k0⋯000k-k⋯0⋯⋯⋯⋯⋯⋯000⋯k-k.By proceeding as in Sect. B, one obtains the following characteristic equation for the eigenvalues of the Jacobian of (53):54\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\lambda ^2+\lambda (\gamma +\mu )+\gamma \mu \left[ 1 -\hat{R}_e^2f^2(\bar{J})\right] - \hat{K}(\lambda )f(\bar{J})f'(\bar{J})\gamma \hat{R}_e^2(\lambda + 2\mu ) \bar{I}_H=0,\qquad \nonumber \\&\quad \Re (\lambda )>-k, \end{aligned}$$\end{document}λ2+λ(γ+μ)+γμ1-R^e2f2(J¯)-K^(λ)f(J¯)f′(J¯)γR^e2(λ+2μ)I¯H=0,ℜ(λ)>-k,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{K}$$\end{document}K^ denotes the Laplace transform of K, i.e.55\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hat{K}(\lambda ):=\displaystyle \int _{0}^{+\infty }{\mathop {\textrm{e}}\limits }^{-\lambda \theta } K(\theta )\,\!\textrm{d}\theta =\left( \frac{k}{\lambda +k}\right) ^n,\qquad \Re (\lambda )>-k. \end{aligned}$$\end{document}K^(λ):=∫0+∞e-λθK(θ)dθ=kλ+kn,ℜ(λ)>-k.In the following sections, we specialise (54) for the case of the DFE and the EsE. Remark 9 The derivation of (54) is provided in Sect. B, where we apply the relevant theory for delay equations (Diekmann et al. 1995) with, possibly, infinite delay (Diekmann and Gyllenberg 2012). As a result, equation (54) holds for any memory kernel K, not necessarily restricted to Erlang-distributed. Stability of the DFE Let us consider the DFE \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\bar{I}_H, \bar{I}_M, \bar{J})=(0, 0, 0)$$\end{document}(I¯H,I¯M,J¯)=(0,0,0). Then, the characteristic equation in (54) becomes56\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda ^2+\lambda (\gamma +\mu )+\gamma \mu \left[ 1 -\hat{R}_e^2f^2(0)\right] =0,\qquad \Re (\lambda )>-k. \end{aligned}$$\end{document}λ2+λ(γ+μ)+γμ1-R^e2f2(0)=0,ℜ(λ)>-k.Recalling equation (52), we have the following result. Theorem 13 The DFE is LAS when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_eh(p(0), q)<1$$\end{document}R^eh(p(0),q)<1 and unstable when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_eh(p(0), q)>1$$\end{document}R^eh(p(0),q)>1, independently of K (i.e. not restricted to the one in (23)). In particular, in the latter case, the characteristic equation (56) has exactly two real roots, one negative and one positive. Proof Being \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma ,\ \mu >0$$\end{document}γ,μ>0, Descartes’ rule of signs ensures that the DFE is stable when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1-\hat{R}_eh(p(0), q)>0$$\end{document}1-R^eh(p(0),q)>0, while it is unstable when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1-\hat{R}_eh(p(0), q)<0$$\end{document}1-R^eh(p(0),q)<0. □ □ Comparing this condition with (51), we see that, when the DFE is stable, no EsE exists. Conversely, when the DFE is unstable, one additional condition is still needed to guarantee the existence of the EsE. Stability of the EsE Let us consider the EsE defined as in (50). Using (45) and (55), the characteristic equation (54) reads57\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda ^2+\lambda (\gamma +\mu )+\left( \frac{k}{\lambda +k}\right) ^n\gamma \delta \left( \lambda + 2\mu \right) =0, \qquad \Re (\lambda )>-k, \end{aligned}$$\end{document}λ2+λ(γ+μ)+kλ+knγδλ+2μ=0,ℜ(λ)>-k,for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta :=-f'(I_H^*)\hat{R}_e I_H^*>0$$\end{document}δ:=-f′(IH∗)R^eIH∗>0. Note that, in this case, the stability of the equilibrium depends on the particular choice of n and k. In particular, from (57) we are led to study the roots of58\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda ^2(\lambda +k)^n+\lambda (\gamma +\mu )(\lambda +k)^n+ \lambda (k^n\gamma \delta )+2 k^n\gamma \delta \mu =0. \end{aligned}$$\end{document}λ2(λ+k)n+λ(γ+μ)(λ+k)n+λ(knγδ)+2knγδμ=0.It is worth observing that, if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k\rightarrow +\infty $$\end{document}k→+∞ and n is fixed, one has \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varphi :=n/k\rightarrow 0^+$$\end{document}φ:=n/k→0+ (that is, the information is instantaneous) and the characteristic equation reduces to59\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda ^2+\lambda \left( \gamma +\mu + \gamma \delta \right) + 2\mu \gamma \delta =0. \end{aligned}$$\end{document}λ2+λγ+μ+γδ+2μγδ=0.This case corresponds to the case in which K is a Dirac delta concentrated at 0, so that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{K}\equiv 1$$\end{document}K^≡1. On the other hand, if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n,\ k\rightarrow +\infty $$\end{document}n,k→+∞ in such a way that φ :=n/kφ:=n/k is constant, then the kernel concentrates at t-φ t-φ, i.e. the memory becomes discrete with delay φ φ. In this case, the characteristic equation takes the form\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda ^2+\lambda (\gamma +\mu )+ \gamma \delta \lambda {\mathop {\textrm{e}}\limits }^{-\lambda \varphi }+2 \gamma \mu \delta e^{-\lambda \varphi }. \end{aligned}$$\end{document}λ2+λ(γ+μ)+γδλe-λφ+2γμδe-λφ.We consider the following cases: instantaneous information, exponentially fading memory, and Erlang-2 distributed memory. Instantaneous information : the characteristic equation reads as in equation (59). Then, from (40), the Routh-Hurwitz criterion implies that the EsE is LAS whenever it exists. Exponentially fading memory : we assume that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K(\theta )=k{\mathop {\textrm{e}}\limits }^{-k\theta }$$\end{document}K(θ)=ke-kθ, corresponding to n=1n=1 in (23). Then, equation (58) reduces to60\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda ^2(\lambda +k)+\lambda (\gamma +\mu )(\lambda +k)+ k\gamma \delta \left( \lambda + 2\mu \right) =0. \end{aligned}$$\end{document}λ2(λ+k)+λ(γ+μ)(λ+k)+kγδλ+2μ=0.Applying the Routh–Hurwitz criterion for third-order polynomials (see Sect. C for details), we obtain the following result. Proposition 14 Let the assumptions of Theorem 11 hold, and let n=1n=1 in (23). Then the EsE is LAS if and only if61\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} k>-\frac{\gamma \delta (\gamma -\mu )+(\gamma +\mu )^2}{\gamma +\mu +\gamma \delta }. \end{aligned}$$\end{document}k>-γδ(γ-μ)+(γ+μ)2γ+μ+γδ.If equality holds, the EsE undergoes a Hopf bifurcation, and the characteristic equation (60) has one negative real root and a pair of purely imaginary conjugate roots. If the opposite inequality holds, then the EsE is unstable, and the characteristic equation (60) has one negative real root and a pair of complex conjugate roots with positive real parts. Since condition (40) holds and k>0k>0, condition (61) is automatically satisfied whenever \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma +k\ge \mu $$\end{document}γ+k≥μ. In particular, this is always true if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma \ge \mu $$\end{document}γ≥μ. This assumption is biologically reasonable, as the average human infectious period 1/γ 1/γ is typically shorter than the average mosquito lifespan 1/μ 1/μ. On the other hand, if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma <\mu $$\end{document}γ<μ, then condition (61) provides a (possibly positive) lower bound for k. This yields the following corollary. Corollary 15 Let the assumptions of Theorem 11 hold, and let n=1n=1 in (23). If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma \ge \mu $$\end{document}γ≥μ, then the EsE is LAS for all k>0k>0. Erlang-2 distributed memory : we assume that K follows an Erlang distribution of order 2, i.e. we take n=2n=2 in (23). Thus, the equation (58) becomes62\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda ^2(\lambda +k)^2+\lambda (\gamma +\mu )(\lambda +k)^2+ k^2\gamma \delta \left( \lambda + 2\mu \right) =0. \end{aligned}$$\end{document}λ2(λ+k)2+λ(γ+μ)(λ+k)2+k2γδλ+2μ=0.Then, applying the Routh-Hurwitz criterion to a fourth-order polynomial (see Sect. C.2 for the details), we obtain the following result. Proposition 16 Let the assumptions of Theorem 11 hold, and let n=2n=2 in (23). Then, there exist \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_+, \,k_-\ge 0$$\end{document}k+,k-≥0 with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_+>k_-$$\end{document}k+>k- such that the EsE is LAS for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k>k_+$$\end{document}k>k+, undergoes a Hopf bifurcation at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=k_+$$\end{document}k=k+, and is unstable for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k\in (k_-, k_+)$$\end{document}k∈(k-,k+). In particular, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_+$$\end{document}k+ is the largest positive real root of\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p(k)=\tilde{A} k^3+\tilde{B}k^2+\tilde{C} k+\tilde{D}, \end{aligned}$$\end{document}p(k)=A~k3+B~k2+C~k+D~,where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \tilde{A}&:=2(\gamma +\mu +\gamma \delta )>0,\\ \tilde{B}&:=4(\gamma +\mu )^2+\gamma \delta (3\gamma -5\mu -\gamma \delta ),\\ \tilde{C}&:=2(\gamma +\mu )[(\gamma +\mu )^2+\gamma \delta (\gamma -3\mu )],\\ \tilde{D}&:=2\gamma \mu \delta (\gamma +\mu )^2<0, \end{aligned}$$\end{document}A~:=2(γ+μ+γδ)>0,B~:=4(γ+μ)2+γδ(3γ-5μ-γδ),C~:=2(γ+μ)[(γ+μ)2+γδ(γ-3μ)],D~:=2γμδ(γ+μ)2<0,while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_-$$\end{document}k- is either 0 or a positive root of p(k)=0p(k)=0 such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p'(k_-)<0$$\end{document}p′(k-)<0. Remark 10 Note that, once the epidemiological parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta _{H\leftarrow M},\, \beta _{M\leftarrow H},\, \rho ,\, \gamma ,$$\end{document}βH←M,βM←H,ρ,γ, and μ μ are fixed, the thresholds \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_-$$\end{document}k- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_+$$\end{document}k+ are fully determined by δ δ, which itself depends on l, q, a, and w. In this section, we showed that a sufficiently large delay φ φ can destabilise the EsE via a Hopf bifurcation. This suggests that self-sustained epidemic oscillations may arise even in the absence of additional mechanisms such as demography or waning immunity. Here, we restricted the analysis to Erlang-distributed memory kernels with shape parameter n=1n=1 or n=2n=2; see (23). As noted above, the analysis may be extended to more general kernels, although this would likely result in considerably more involved computations (Andò et al. 2026; Zhang et al. 2023). Numerical results Fig. 12Time evolution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (blue) for model (38) (one-group model) and EsE (green) as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH. Model parameters and initial conditions are as in Sect. 4.1, with l=0.25l=0.25 and ρ =2ρ=2. The upper row corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, the lower row to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^3$$\end{document}χ=103; the left panels correspond to q=0q=0 and the right panels to q=0.1q=0.1. We take n=1n=1 and φ =30φ=30 days in (23) (colour figure online) Fig. 13Time evolution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (blue) for model (38) (one-group model) and EsE (green) as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH. Model parameters and initial conditions are as in Sect. 4.1, with l=0.25l=0.25 and ρ =2ρ=2. The upper row corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, the lower row to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^3$$\end{document}χ=103; the left panels correspond to q=0q=0 and the right panels to q=0.1q=0.1. We take n=2n=2 and φ =30φ=30 days in (23) (colour figure online) Fig. 14Time evolution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (blue) for model (38) (one-group model) and EsE (green) as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH. Model parameters and initial conditions are as in Sect. 4.1, with l=0.25l=0.25 and ρ =2ρ=2. The upper row corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, the lower row to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^3$$\end{document}χ=103; the left panels correspond to q=0q=0 and the right panels to q=0.1q=0.1. We take n=5n=5 and φ =30φ=30 days in (23) (colour figure online) Fig. 15Time evolution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (solid blue) for model (38) (one-group model), EsE (solid green) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e$$\end{document}R^e (dashed red) as functions of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH. Model parameters and initial conditions are as in Sect. 4.1, with l=0.25l=0.25 and ρ =2ρ=2 (left), l=1l=1 and ρ =5.12ρ=5.12 (right). We take \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, q=0.2, with n=1n=1 and φ =30φ=30 days in (23). For each panel, the left vertical axis (blue) corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH and EsE, while the right vertical axis (red) corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e$$\end{document}R^e. The upper and lower solid black horizontal lines represent the thresholds 1+l/q1+l/q and 1+l1+l in (47), respectively. Note that, in the left panel, the EsE only exists once \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e$$\end{document}R^e crosses the threshold value identified in Theorem 11 (colour figure online) Time evolution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (blue) for model (38) (one-group model) and EsE (green) as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH. Model parameters and initial conditions are as in Sect. 4.1, with l=0.25l=0.25 and ρ =2ρ=2. The upper row corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, the lower row to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^3$$\end{document}χ=103; the left panels correspond to q=0q=0 and the right panels to q=0.1q=0.1. We take n=1n=1 and φ =30φ=30 days in (23) (colour figure online) Time evolution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (blue) for model (38) (one-group model) and EsE (green) as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH. Model parameters and initial conditions are as in Sect. 4.1, with l=0.25l=0.25 and ρ =2ρ=2. The upper row corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, the lower row to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^3$$\end{document}χ=103; the left panels correspond to q=0q=0 and the right panels to q=0.1q=0.1. We take n=2n=2 and φ =30φ=30 days in (23) (colour figure online) Time evolution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (blue) for model (38) (one-group model) and EsE (green) as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH. Model parameters and initial conditions are as in Sect. 4.1, with l=0.25l=0.25 and ρ =2ρ=2. The upper row corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, the lower row to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^3$$\end{document}χ=103; the left panels correspond to q=0q=0 and the right panels to q=0.1q=0.1. We take n=5n=5 and φ =30φ=30 days in (23) (colour figure online) Time evolution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (solid blue) for model (38) (one-group model), EsE (solid green) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e$$\end{document}R^e (dashed red) as functions of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH. Model parameters and initial conditions are as in Sect. 4.1, with l=0.25l=0.25 and ρ =2ρ=2 (left), l=1l=1 and ρ =5.12ρ=5.12 (right). We take \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104, q=0.2, with n=1n=1 and φ =30φ=30 days in (23). For each panel, the left vertical axis (blue) corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH and EsE, while the right vertical axis (red) corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e$$\end{document}R^e. The upper and lower solid black horizontal lines represent the thresholds 1+l/q1+l/q and 1+l1+l in (47), respectively. Note that, in the left panel, the EsE only exists once \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e$$\end{document}R^e crosses the threshold value identified in Theorem 11 (colour figure online) In this section, we numerically assess the extent to which the further reduction leading from model (38) to model (41) preserves the transient dynamical features identified in Sect. 5. We use the same parameter values as in Sect. 4.1, with φ =30φ=30 days. In Figs. 12, 13 and 14, we show the time evolution of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (blue) and the EsE as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH (green, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH varying in time, see (50)) for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=1,\,2,\,5$$\end{document}n=1,2,5, respectively. We fix l=0.25l=0.25 and ρ =2ρ=2, and consider q=0q=0 (left panels) and q=0.1q=0.1 (right panels). For each case, we compare the model outputs for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104 (upper row) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^3$$\end{document}χ=103 (lower row). In all cases, the solution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH of model (38) approaches the EsE predicted by model (41), which can be shown to be LAS for these parameter values whenever it exists. In particular, for q=0q=0, increasing n leads to more pronounced oscillations around the EsE before the solution ultimately collapses onto this manifold, for both \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^3$$\end{document}χ=103. These oscillatory behaviours are instead much less pronounced (and nearly absent) for the case q=0.1q=0.1. As observed in the numerical simulations presented in Sect. 4.1, in all experiments shown in Figs. 12, 13 and 14, the choice q=0.2q=0.2 with l=0.25l=0.25 and ρ =2ρ=2 led to larger outbreaks compared to the other cases (except for q=0.5q=0.5), and no “quasi steady state” was observed. To explain this, in Fig. 15 we plot \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH (blue) and the EsE (green) as functions of time (with EsE depending on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH), together with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e$$\end{document}R^e as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_H$$\end{document}SH (dashed line; see (44)), for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =10^4$$\end{document}χ=104. We consider both l=0.25l=0.25, ρ =2ρ=2 (left panel) and l=1l=1, ρ =5.12ρ=5.12 (right panel). In the former case, at the beginning of the outbreak, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e$$\end{document}R^e exceeds the upper threshold 1+l/q1+l/q in (47), which is a necessary condition for the existence of the EsE (in the figure, the solid black horizontal lines represent the thresholds 1+l1+l and 1+l/q1+l/q). As a consequence, a larger outbreak occurs, reaching its peak when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e=1+l/q$$\end{document}R^e=1+l/q, after which the epidemic declines and eventually dies out. In contrast, for l=1l=1 and ρ =5.12ρ=5.12, the condition \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{R}_e<1+l/q$$\end{document}R^e<1+l/q holds, ensuring the existence of the EsE for model (41). In this case, the equilibrium is LAS, and consequently, the solution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_H$$\end{document}IH converges to it. Conclusions and outlook In this paper, we investigated the impact of protective behaviour on an SIR-SI host–vector compartmental model. We focused on single-outbreak scenarios (thus neglecting host demography and waning immunity) so that the systems under study admit only the DFE as a non-trivial equilibrium. Nevertheless, we were able to characterise the transient dynamics under various modelling assumptions. In Sect. 2, we first analysed a model in which individuals adopt a fixed behaviour (either protecting or not protecting themselves from mosquito bites). Such models have been extensively studied in the literature (Dye and Hasibeder 1986; Hasibeder and Dye 1988; Miller and Huppert 2013), where it has been shown that imperfect or partial protection strategies may increase the basic reproduction number \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0, thereby enhancing the risk of an outbreak. In contrast, our analysis (Proposition 2) shows that imperfect protection in a fraction of the population may decrease \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 whenever a function F(p, q), depending on the fraction p of protected individuals and on the protection leakage q, is less than the parameter l, representing the fraction of mosquito bites on non-human hosts. To the best of our knowledge, this result is new, although consistent with previous findings. For instance, in Miller et al. (2016), threshold conditions on p were derived to determine whether protective behaviour increases or decreases \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0. In particular, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0$$\end{document}R0 increases when protection diverts mosquito bites towards non-protected individuals (Killeen and Smith 2007; Moore et al. 2007), whereas it may decrease if protective measures force mosquitoes to spend additional time attempting to bite protected hosts, indirectly benefiting non-protected individuals. This interpretation is consistent with viewing l as a measure of the spatial separation between hosts (see Remark 1). We then considered a model in which behavioural changes depend on information about the epidemic. In this setting, we proved (Proposition 8) that the reproduction number is smaller when individuals switch between protected and unprotected behaviour than in the case of fixed behaviour, for the same values of p and q. While this effect had been observed numerically in Demers et al. (2018), a rigorous proof of (30) was, to our knowledge, not previously available. Biologically, this result shows that the concentration of mosquito bites on non-protected individuals (see Dye and Hasibeder (1986), Miller et al. (2016), Miller and Huppert (2013)) is mitigated when individuals switch between protected and unprotected behaviour, even when such changes occur at low rates (Demers et al. 2018). When it is assumed that the rate of behavioural changes is much faster than epidemic dynamics (a common assumption in this class of models; see Bulai et al. (2024), Della Marca et al. (2024), Poletti et al. (2009)), it is possible to separate the time scales of the two processes by applying methods from geometric singular perturbation theory. This leads to the epidemic model (38), which includes the information index while treating the host population as homogeneous. Note that Buonomo (2014) proposed an SIR–SI model similar to (38) in an endemic setting for a malaria-like infection, accounting for human behavioural responses to public health campaigns promoting the use of bed nets. In that work, the mosquito biting rate (interpreted as the human–mosquito contact rate) is assumed to depend on past prevalence through an information index, as well as on an effort function describing the actions of the PHS in the health-promotion campaign. However, the analysis in Buonomo (2014) mainly focuses on numerical results and optimal control aspects for specific choices of the information kernel, with less emphasis on a detailed analytical study of the model dynamics. In Sect. 5 we show that the inclusion of information-induced protective behaviour may lead to prolonged epidemic waves when the adoption of protective measures effectively prevents a large portion of infections before herd immunity and seasonality take over. Indeed, a heuristic asymptotic analysis of model (38) shows that, after the initial peak, solutions rapidly approach a slow manifold along which the infected population decays very slowly. Numerical simulations confirm this behaviour, showing that an outbreak may last for a very long time, even in the absence of host demography and recruitment of new susceptibles. Moreover, numerical simulations indicate that convergence to this slow manifold persists even when the assumption of a clear separation of time scales is relaxed. This suggests that prolonged outbreaks are an intrinsic feature of information-dependent behavioural models. We stress, however, that the model assumes constant parameters; in realistic settings, seasonal effects (e.g., temperature-driven mosquito dynamics) may significantly shorten the duration of an outbreak. Note, moreover, that in the numerical simulations we assumed that the switching functions a(J) and w(J) are linear, while non-linear functions might be more realistic. Based on our analytical and numerical results, several extensions of the model can be considered. A natural first step would be to include demography and/or loss of immunity in the human population, which would allow for the existence of an Endemic Equilibrium under standard assumptions on the parameters. Moreover, it is important to stress that lifelong immunity upon recovery is not universal for vector-borne diseases: an individual may contract malaria multiple times during their lifetime; there exist several strains of dengue, so that reinfection by a heterologous dengue serotype can give rise to antibody-dependent enhancement, possibly inducing complex dynamics (Katzelnick et al. 2017). Including loss of immunity and/or multiple viral strains to our model would increase its realism, however with a non-negligible toll on its analytical tractability. Another possible extension is to incorporate seasonality, for instance by accounting for seasonal fluctuations in the mosquito population (e.g., due to temperature and humidity); see, for instance, Rocha et al. (2016). However, in this setting, even the computation of the BRN and CRN becomes more involved, as they are defined as the spectral radius of an infinite-dimensional next-generation operator rather than of a next-generation matrix (NGM) (Bacaër and Guernaoui 2006; Inaba 2019), and typically require numerical approximations; see, e.g., Bacaër (2007), Breda et al. (2026). Additionally, it would be interesting to investigate models evolving on three time scales: a fast one associated with the spread of information, an intermediate one corresponding to disease dynamics, and a slow one related to human demography and/or loss of immunity. Finally, we note that individual behaviour depends on personal opinions about the infection, which may in turn be influenced by information on the epidemic as well as by social interactions. Several authors have recently investigated the interplay between opinion dynamics and epidemic spread through mathematical models (Albi et al. 2025; Chang et al. 2025; Tyson et al. 2020). Without explicitly introducing a variable describing opinion, an extension of the behavioural model (20) might be obtained by including imitation-dynamics terms of the form \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta _P(J)H_{NP}p$$\end{document}θP(J)HNPp and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta _{NP}(J)H_{P}(1-p)$$\end{document}θNP(J)HP(1-p), which describe a contagion of ideas among host individuals (Bauch 1573; Manfredi and d’Onofrio 2013; Wang et al. 2016). Such contagion may be driven by individuals’ knowledge of past epidemics (Della Marca et al. 2024), or by payoff considerations (Poletti et al. 2009). Accordingly, the equation for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_P$$\end{document}HP (and correspondingly for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{NP}$$\end{document}HNP) takes the form\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} H_P'=\left[ a(J) + \theta _P(J)p\right] H_{NP} - \left[ w(J)+\theta _{NP}(J)(1-p)\right] H_{P}. \end{aligned}$$\end{document}HP′=a(J)+θP(J)pHNP-w(J)+θNP(J)(1-p)HP.In the context of vector-borne epidemics, similar approaches have been adopted in Asfaw et al. (2018) and, more recently, in Laxmi and Bhattacharyya (2022) to model the effect of bed-net usage in malaria-like diseases. The latter, in particular, focuses on both the use and misuse of Insecticide-Treated Nets (ITNs). Both works deal with endemic scenarios: the former is based on imitation game dynamics with information-free behavioural changes, while the latter derives a related model in which behavioural changes are driven by payoff considerations linked to infection risk, which may depend on information about current prevalence, mosquito density, and seasonal effects (e.g., periodic replacement of bed nets). While Asfaw et al. (2018) provide analytical results on the model dynamics, Laxmi and Bhattacharyya (2022) mainly focus on reproduction numbers (and related considerations on optimal ITN usage) and numerical simulations, while also showing that information-induced imitation dynamics may generate recurrent epidemic waves. Interestingly, Laxmi and Bhattacharyya (2022) also discuss applications to malaria control in several African countries. In the present work, we did not consider such mechanisms. Nevertheless, the approach developed here could be extended to models incorporating evolutionary game-like dynamics, although this would inevitably lead to more involved computations (see, e.g., Della Marca et al. (2024)). This will be the subject of future work, together with the inclusion of additional factors relevant to the dynamics and control of mosquito-borne epidemics, and applications of the proposed behavioural models to data-informed real-world scenarios.