#### Abstract

A family of discrete nonautonomous SIRVS models with general incidence is obtained from a continuous family of models by applying Mickens nonstandard discretization method. Conditions for the permanence and extinction of the disease and the stability of disease-free solutions are determined. Concerning extinction and persistence, the consistency of those discrete models with the corresponding continuous model is discussed: if the time step is sufficiently small, when we have extinction (permanence) for the continuous model, we also have extinction (permanence) for the corresponding discrete model. Some numerical simulations are carried out to compare the different possible discretizations of our continuous model using real data.

#### 1. Introduction

Most of the epidemiological models in the literature are continuous models. In spite of this, recently, there has been a growing interest in discrete-time models [1–7]. In this work, we will use Mickens nonstandard difference (NSFD) scheme to achieve a discretization of a family of continuous epidemiological models with vaccination and general incidence function considered in [8]. We have multiple objectives: firstly, we want to obtain conditions for extinction and permanence of the disease for the discrete family; next, having a continuous and a corresponding discrete family of models, we wish to discuss the problem of consistency of the discrete models with the corresponding continuous ones; finally, we intend to present some simulation results.

The dynamical consistency of a numerical scheme with the associated continuous system is not a precise definition. By the expression “dynamical consistency,” it is meant that the numerical solutions replicate some of the properties of the continuous systems solutions.

We will consider the dynamical consistency regarding the permanence and extinction of the disease: whenever there is extinction (permanence) of the disease for the continuous-time model, the same holds for the discrete-time one. Several papers [9–15] discuss the dynamical consistency with respect to some particular properties of discrete epidemiological models obtained from continuous models by some NSFD scheme [16]. We note that while the papers cited above consider autonomous models, in the present work we discuss dynamical consistency for a nonautonomous model. To the best of our knowledge, this is the first work where the consistency of a discretized epidemiological model with the original continuous model is discussed in the nonautonomous context.

We remark that, even in the very particular case of autonomous models with mass action incidence and assuming that there is no disease related death, it follows from results in [17] that, when , the continuous autonomous model has one or two endemic equilibrium points coexisting with the disease-free equilibrium. The existence and stability of the equilibriums are governed by and three additional thresholds: , , and . These thresholds determine the qualitative behaviour of the system. This very particular situation shows that, even in the autonomous context, the qualitative behaviour can be difficult to determine.

One of the motivations for our work was that the difficulties increase considerably when we deal with a general nonautonomous situation. Thus, we decided to discuss an aspect of consistency that, nevertheless, is very important from the point of view of biomathematics: can we obtain consistency between continuous and discrete-time models from the point of view of persistence and extinction of the disease?

Regarding our simulation results, we considered two different types of computational experiments. Our first set of simulation results are designed to compare different possible discretizations of our continuous models. After this discussion, we apply our model to a real situation, considering data from the incidence of measles in France in the period 2012-2016. To the possible extent, this data is used to estimate our model parameters and the computational results obtained are compared with real data.

The law of mass action states that the rate of change in the disease incidence is directly proportional to the product of the number of susceptible and infective individuals and was the paradigm in the classic models in epidemiology. This is why classical models usually consider a bilinear incidence rate , where and denote, respectively, the number of susceptible and infective individuals, to model the disease transmission. In spite of this, it is sometimes important to consider other forms of incidence functions. Another usual assumption is the time independence of the parameters model parameters: in fact, the majority of the epidemiological models in the literature are given by a system of autonomous differential or difference equations. Nevertheless, the assumption that the parameters are independent of time is not very realistic in many situations and it is useful to consider nonautonomous models that, for instance, allow the discussion of environmental and demographic effects that change with time [18, 19]. In this work the family of models considered is nonautonomous and the incidence rates are taken from a large class of functions.

Our model generalizes one obtained by Mickens nonstandard finite difference method from the continuous model [8] (see Section 2). In [20], a discrete nonautonomous epidemic model with vaccination and mass action incidence was obtained by Mickens method. We emphasize that, in the particular mass action case, our model is not exactly similar to the model in [20], although Mickens rules were considered in both. We briefly compare computationally these two slightly different models in Section 5.

The model we will consider is the following: , where the classes , , , and correspond, respectively, to susceptible, infective, recovered, and vaccinated individuals and the parameter functions have the following meanings: denotes the inflow of newborns in the susceptible class; the function is the incidence (into the infective class) from the susceptible individuals; the function is the incidence (into the infective class) from the vaccinated individuals; are the natural deaths; represents the susceptibles vaccination; represents the immunity loss and consequence influx in the susceptible class; are the deaths occurring in the infective class; is the recovery. We will assume that , , , and are bounded and nonnegative sequences and that there are positive constants , , , , and such that (H1)the functions and are nonnegative and differentiable in and the functions and are nondecreasing and Lipschitz, with Lipschitz constants and ,(H2)we have for all ,(H3),(H4) and ,(H5)functions and are nonincreasing.

In this work, we prove, when our conditions prescribe extinction (permanence) for the continuous model we also have extinction (permanence) for the corresponding discrete model as long as the time step is smaller than some constant (that depends on some model parameters and on the threshold condition). We also consider a family of examples of the periodic system of period such that the continuous and the discrete-time system with time step is not consistent, highlighting the importance of knowing that for time steps smaller than some explicit value we have consistency.

#### 2. Discretization of the Continuous Model

We start with a nonautonomous SIRVS model that is slightly less general than the one considered in [8] and generalizes the one in [21]. Namely, we consider the model:

We assume that the functions , , , and belong to the class and are nonnegative and bounded. We also require that(C1)the functions and are nonnegative, nondecreasing, differentiable, and Lipschitz with Lipschitz constants and , respectively;(C2);(C3)there is such that .

In order to obtain threshold conditions for model (2), it was considered in [8] the following auxiliary system:And for each solution of (3) with positive initial conditions, it was shown that the numbers are independent of the particular solution.

Using the above numbers, the following results are contained in results obtained in [8].

Theorem 1 (Theorem 1 of [8]). *Assume that conditions (C1), (C2), and (C3) hold. Then, if there is a constant such that , then the infectives are permanent in system (2).*

Theorem 2 (Theorem 2 of [8]). *Assume that conditions (C1), (C2), and (C3) hold. Then if there is a constant such that , then the infectives go to extinction in system (2).*

In the literature, several models were discretized using Mickens NSFD schemes [22–35]. Next, we will apply Micken’s nonstandard method to obtain a discrete version of system (2).

Let be a positive continuous function such thatGiven , we let , with , and identify with After deciding a nonlocal representation for the incidence function and that terms that do not correspond to an interaction will be considered in the time, the first equation in (2) becomesWriting , , , , , , , and , we have Proceeding similarly for the other equations, we obtain the following discrete model: . We will consider a model that contains this one to obtain some of our results. Namely, based on model (9), in Sections 3 and 4 we will study model (1) that has a more general form for the incidence function.

Now, we need to make some definitions. We say that(i)the infectives are* permanent* if for any solution of (1) with initial conditions there are constants such that(ii)the infectives go to* extinction* if for any solution of (1) with initial conditions we have .

Similar definitions can be made for the other compartments. For instance, if there exist constants such that for any solution of (1) with initial conditions we have we say that the susceptibles are permanent.

#### 3. Permanence and Extinction in the Discrete Model

In this section, we will extend the results obtained for the model with the usual mass action incidence in [20] to our generalized family of models. Namely, suitable thresholds are defined and conditions for persistence and extinction of the disease are obtained. As a corollary of our results, we consider the periodic case where we have a unique number that establishes the boundary between the regions of permanence and extinction. Although the proofs of our results are inspired in [20], some difficulties must be dealt with. In particular, it was necessary to understand the right conditions to impose to the incidence functions in order to overcome the technical difficulties.

To lighten the reading, the proofs of our results are presented in the appendix.

##### 3.1. Auxiliary Results

Consider the auxiliary system,Note that the auxiliary system describes the behaviour of the system in the absence of infection. If , , , , , , , and are constant sequences, then the linear system (12) becomes autonomous and corresponds to the linearization of the equations for and in the classical (autonomous) SIRVS model.

In order to proceed we need to recall some notions. A solution of some system of difference equations is said to be* attractive* if for all and all there is and such that if is a solution with then , for all . Additionally, if some solution is attractive and we can take to be only dependent on , we say that it is* uniformly attractive*.

The following theorem furnishes some simple properties of system (12).

Lemma 3 (lemma 2.2 of [20]). *Assume that conditions (H3) and (H4) hold. Then *(i)*all solutions of system (12) with initial condition and are nonnegative for all ;*(ii)*each fixed solution of (12) is bounded and globally uniformly attractive for all ;*(iii)*if is a solution of (12) and is a solution of the system* *with then there is a constant , only depending on , satisfying *(iv)*there exist constants such that, for each solution of (12), we have* (v)*when system (12) is –periodic, it has a unique positive –periodic solution which is globally uniformly attractive.*

We have the following lemma.

Lemma 4. *Assume that condition (H5) holds. Then we have the following:*(i)*all solutions of (1) with nonnegative initial conditions are nonnegative for all ;*(ii)*all solutions of (1) with positive initial conditions are positive for all ;*(iii)*there is a constant such that, if is a solution of (1) with nonnegative initial conditions, then*

*Proof. * the appendix.

For each and each particular solution of (12) with and we define the numberswhere denotes the partial derivative of with respect to the -th variable. Contrarily to what one could expect, the next lemma shows that the numbers above do not depend on the particular solution of (12) with and .

Lemma 5. *Assume that (H1), (H3), and (H4) hold. If and are two solutions of (12) with and , , then*

*Proof. *See Appendix.

By Lemma 5 we can drop the dependence of the particular solution and simply write and instead of an , respectively.

##### 3.2. Extinction and Permanence

We have the following result about the extinction of the disease.

Theorem 6 (extinction of the disease). *Assume that conditions (H1) to (H5) hold. Then*(a)*if there is a constant such that , then the infectives go to extinction;*(b)*any solution , where is a particular solution of system (12), is globally uniformly attractive.*

*Proof. *

We have the following result about the permanence of the disease.

Theorem 7 (permanence of the disease). *Assume that conditions (H1) to (H5) hold. If there is a constant such that then the infectives are permanent in system (1).*

*Proof. *ee the Appendix.

We consider now the particular periodic case: assume that all parameters of system (1) are periodic with period . By (v) in Lemma 3, there is an -periodic disease-free solution of (12), . Thus, in the periodic setting, (17) and (18) become both equal toTherefore, we obtain the corollary.

Corollary 8 (periodic case). *Assume that all coefficients are -periodic in (1) and that conditions (H1) to (H5) hold. Then *(a)*if then the infectives go to extinction;*(b)*the disease-free solution , where is an disease-free -periodic solution of (12), is globally attractive;*(c)*if , then the infectives are permanent.*

*Proof. * Appendix.

#### 4. Consistency

In this section, under the additional assumption that the parameter functions , , , and are constant, we will get a result stating that when our integral conditions prescribe extinction (persistence) for the continuous-time model, then the discrete-time conditions prescribe extinction (persistence) for the corresponding discrete-time models, as long as the time step is less than some constant. Throughout this section, we assume that the parameter functions , , , and are constant functions and will be the function used in the discretization of the derivative.

We consider the continuous-time model (2) and, for a given time step , the corresponding discrete-time model, that is, the discrete-time model with parameters , , , , , , , and .

For a given time step , the expressions and in (17) and (18) become, in our context, where is the solution of the (in our context autonomous) system (12).

We have the following result.

Theorem 9. *For system (2), assume that , , , and for all and that the functions , , , and are differentiable, nonnegative, and bounded and have bounded derivative. Assume also that conditions (C1) to (C3) hold and let where Then,*(a)*if , then for all ;*(b)*if , then for all for all .*

*Proof. *Observe that , and , , where are, respectively, solutions of system (12) and system (3). Thus,By contradiction, assume thatand that there is a sequence such that as andfor all . By (27), we conclude that, for each , there are sequences and such that as , as andBy (28), we havewhereand, multiplying both sides by , we get We also haveNoting that, by (5), we have and that a convergent sequence is bounded; by (34) there is such thatSimilarly, we haveUsing (5) again, we get There is such thatFinally, we have According to (5), we obtain There is such thatThus,for all . Since the right hand side of (43) is independent of , we conclude thatas , uniformly in .

On the other hand, we note that the function given byis Riemann-integrable on any bounded interval .

We have that is a Riemann sum of with respect to the partition of size of the interval . Note that and as , uniformly in .

Since is with bounded derivative, for any we have where . We conclude thatthus and therefore By (53) we conclude that, given , there is such that, for all ,Finally, recalling that , by assumption, by the arbitrariness of and the fact that as , we obtain for sufficiently large , which is a contradiction. We obtain (a).

A similar argument allows us to prove (b). In fact, assuming by contradiction thatand that there is a sequence such that as and it is possible to conclude thatwhere , , and are given, respectively, by (30), (31), and (42) and still satisfy (36), (39), and (42). Consequently, given , there is such that, for all ,Recalling that , by assumption and since is arbitrary, we obtain for sufficiently large , which is a contradiction. We obtain (b) and the theorem follows.

Next, for each , we give an example of a periodic system of period such that the continuous and the discrete-time system with time step are not consistent; namely, we will have persistence for the continuous-time model and extinction for the discrete-time model with time step .

*Example 10. *Let . Consider in system (2) that , that, with the exception of and , all parameters are constant, that , and that We obtain a periodic system of period .

In this context, , , and , , where are, respectively, solutions of system (12) and system (3). It is now possible to compute the number . In fact, noting that , we get We can also compute . Namely, we have If we let be sufficiently small so that , or in other words, and be sufficiently large so that , we obtain So we conclude that we do not have consistency for time step .

Let and consider the continuous model with the following parameters , , , , , , and . In Figure 1, we plot function (or similarly ) and the component of the solution of system (2) given by the solver of Mathematica® (that we take to represent the solution of the continuous-time model) and the solution of the discrete-time model (9) with time step . As can be seen, the infectives are persistent in the continuous-time model but go to extinction in the discrete-time model. We have inconsistency in this case.

Note that, changing and slightly, we can construct an example of a periodic system with period where the infectives in the continuous-time model go to extinction but, in the discrete-time model with time step , the infectives are persistent.

Furthermore, we emphasize that this lack of consistency is not a result of the discretization method used but simply a result of the fact that the time steps lead to a situation where the points where the functions and are evaluated (in order to obtain the discrete-time parameters) correspond to minimums of and .

#### 5. Simulation

Our objective in this section is twofold. On the one hand, we want to consider different incidence functions , corresponding to different discretizations of our continuous model, and compare the several discrete models obtained. We do this in the first subsection. On the other hand, we want to use our model to describe a real situation. We do this in the second subsection where we consider data from the incidence of measles in France in the period 2012-2016.

##### 5.1. Simulation with Several NSFD Schemes

In this subsection we do some simulation to illustrate our results. To begin, we compare our model (1) with mass action incidence ( and ) with Zhang’s model [20]. We use the following set of parameters: , , , , , and Setting we obtain and thus we conclude that we have extinction for the continuous model. Taking time steps equal to , , and , we get , , and and we conclude that we have extinction for time steps and . For these parameters, we have consistency in the sense of Theorem 9 as long as the time step is less than . Clearly, there is numerical evidence that there is consistency even for higher time steps. Figure 2 illustrates this situation.

Changing to we obtain and thus we conclude that we have persistence. Taking time steps equal to , , and , we get , , and and we conclude that we have persistence for all these time steps. Figure 3 illustrates this situation. Figures 2 and 3 suggest that numerically our model is slightly better than Zhang’s model, at least for large time steps.

Next, we compare our model with the discretized model obtained by Euler method and the output of the Mathematica solver ODE (that uses a Runge-Kutta method). Considering , we get extinction for the continuous-time model, as we already saw. Taking time steps equal to , , and , we can see in Figure 4 that for all methods considered and all time steps we have extinction, although the behaviour of our model shadows better the behaviour given by Mathematica’s solver, at least for these time steps.

Changing to we already saw that we get persistence for the continuous model. Figure 5 illustrates this situation.

Next, we change our incidence function and consider , maintaining the set of parameters. Letting we have extinction for the continuous model and letting we have persistence for the continuous-time model. Note that the thresholds , , , and are similar to the mass action case. Figures 6 and 7 illustrate this situation.

Doing corresponding simulations and comparisons for our model with instead of , we can draw the same conclusions regarding extinction/persistence, relation to Zhang’s model and the model obtained by Euler method.

##### 5.2. Simulation with Real Data

In this subsection, we present some simulation regarding measles. This disease is endemic in some countries such as France. In that country, with the measles outbreak in 2011, a vaccination policy that lowered the number of reported cases was introduced. We will focus on measles in France, between 2012 and 2016. For a study concerning the period before 2012, see [36]. For our parameters estimation, we gathered information from several websites. We considered standard incidence functions and , where is the total population. Inspired in the time series for the infectives (https://ecdc.europa.eu), we considered and given by