A numerical scheme for a SIS epidemic model with a delay is constructed by applying a nonstandard finite difference (NSFD) method. The dynamics of the obtained discrete system is investigated. First we show that the discrete system has equilibria which are exactly the same as those of continuous model. By studying the distribution of the roots of the characteristics equations related to the linearized system, we can provide the stable regions in the appropriate parameter plane. It is shown that the conditions for those equilibria to be asymptotically stable are consistent with the continuous model for any size of numerical time-step. Furthermore, we also establish the existence of Neimark-Sacker bifurcation (also called Hopf bifurcation for map) which is controlled by the time delay. The analytical results are confirmed by some numerical simulations.
In this paper we reconsider a SIS epidemic model with maturation delay developed in [
Here is a birth rate function. The size of mature population at time is divided into susceptible and infective classes, so that ., are the death rates of immature and mature population, respectively. The delay is considered as the maturation time. The parameter, and are respectively the constant contact rate, the disease induced death rate constant and the recovery constant rate, respectively.
In this paper we consider a special class of Equation (1) by assuming that the death rate in each stage prior to the adult stage and the disease induced death are negligible, i.e.. Using the Rick function as the birth rate function, Equation (1) becomes
Note that Equation (2.c) is decoupled and is called the Nicholson’s blowflies equation which proposed by [
In [
It has also been shown in [
For practical purposes, we need to do numerical simulations and therefore we have to transform the continuous model (2) into a discrete system. We expect that the dynamical properties of the discrete system are in accordance with its continuous counterpart (2). Kunnawuttipreechachan [
To overcome the dependence of stability condition on the time-step size, we will apply a nonstandard finite difference (NSFD) scheme. This method, which is developed by Mickens [9,10], has been applied to various problems; see e.g. [11-16], in which the numerical solutions preserve dynamical properties of the continuous model. We will show that the discrete SIS epidemic model with a delay obtained by the NSFD method maintains the stability properties of equilibria irrespective of the size of numerical time-step. Besides the stability conditions, the existence of bifurcation of the discrete model will also be investigated.
Using the fact that, we will only consider the last two equations of system (2). Using a transformation and with, Equation (2.b) and (2.c) can be written as
To discretize system (4) we first consider the step-size of the form where k is a positive integer. Then, applying the forward difference scheme for the derivative and a nonlocal approximation for the right hand sides of system (4) yields a system of difference equations
where and are numerical approximation of and , respectively. The numerical scheme (5) is a nonstandard because it uses a nonlocal approximation; see Mickens [9,10]. Here we consider initial conditions
and,(6)
where, for. It is easy to show that the implicit scheme (5) can be arranged to get its explicit version, i.e.
Direct calculations show that the discrete system (7) has a unique equilibrium: if
and. This equilibrium is called the disease free equilibrium (DFE). On the other hand, if and then, in addition to the DFE, there also exists an endemic equilibrium (EE): where
and.
We observe that those equilibria are exactly the same as those of the continuous system (4); see e.g. [
It is well known that the stability of equilibrium of a dynamical system depends on the distribution of the zeros of its associated characteristics equation. In this section, the distribution of the roots of the characteristics equation will be analyzed using the following results of Zhang, Zu and Zheng [
Theorem 1 (Zhang, Zu and Zheng [
Suppose that is a bounded, closed and connected set;
is continuous in and is a parameter. Then as varies, the sum of the order of the zeros of out of the unit circle:
can change only if a zero appears on or crosses the unit circle.
First we perform a linearization of system (7) about the DFE by taking and. The linearized system around the DFE is given by
By introducing new variables
we can rewrite system (8) in the form
where and the constant matrix A is defined by
with, and . The characteristic equation of matrix A is
where.
Lemma 1. If and then there exists a such that all roots of Equation (10) have modulus less than one for.
Proof. The trivial root of Equation (10) is
.
It is clear that if then for all. Other roots of Equation (10) are determined by
when, Equation (11) becomes.
Hence Equation (11), at, has a root of multiplicity k and a simple root. Consider the root such that. This root depends continuously on. Based on Equation (11) we have
whenever. Consequently all roots of Equation (10) lie in for all sufficiently small, and the existence of the maximal follows.
Lemma 2. If then Equation (11) has no root with modulus one for all.
Proof. Assume with be a root of Equation (11) when. Substituting this root to Equation (11) yields
.
Separating the real and imaginary parts, we have
and therefore we get
If then, which yields a contradiction. This completes the proof.
If, then the roots of Equation (11) with modulus one satisfy
Lemma 3. If then the roots of Equation (11) satisfy
where and satisfy Equation (15).
Proof. From Equations (11) and (15) we have
where
and
with
It is clear that if then and therefore
.
Based on Theorem 1 and Lemmas 1-3, we have the following results on stability and bifurcation of system (5) at the DFE.
Theorem 2.
1) If and then the DFE is asymptotically stable for all.
2) If and then there exists an infinite sequence of time delay parameter such that the DFE is asymptotically stable when and unstable when. Map (7) has a Neimark-Sacker bifurcation at the DFE when, where satisfy Equation (15).
Proof.
1) Let and then it is known from Lemmas 1 and 2 that the characteristic Equation (10) has no root with modulus one for all. Applying Theorem 1, all roots of Equation (10) have modulus less than one for all. Thus, conclusion (a) follows.
2) Let and. It is known from Lemma 1 that if then one of roots of characteristics Equation (10) has modulus less than one for all. Other roots of Equation (10) satisfy Equation (11). From Lemmas 1 and 3 we know that all roots of Equation (11) have modulus less than one when, and Equation (11) has at least a couple of roots with modulus greater than one when. Hence conclusion (b) follows.
The linearization of system (7) around the EE is performed by substituting and into system (7) to get
where
,
,
,
and
.
Using the same transformation as in the DFE, i.e., system (16) can be written as
where. The constant matrix B in Equation (17) is
.
It is easy to show that the characteristic equation of matrix B is
where Clearly this characteristic equation has a trivial root
and other roots are determined by. Direct calculations show that if then for all. Since and, is exactly the same as in Equation (11) and therefore the distribution of its roots is exactly the same as described by Lemmas 1-3. Hence we directly have the following results.
Lemma 4. If and then there exists a such that all roots of Equation (18) have modulus less than one for.
Lemma 5. If and then Equation (18) has no root with modulus one for all.
Lemma 6. If and then the modulus of trivial root of Equation (18), i.e., is less than one for all and other roots satisfy
where and satisfy Equation (15).
Theorem 3.
1) If and then the EE is asymptotically stable for all.
2) If and then there exists an infinite sequence of time delay parameter such that the EE is asymptotically stable when and unstable when. System (7) has a Neimark-Sacker bifurcation at the EE when, where satisfy Equation (15).
Theorems 2.1 and 3.1 give the sufficient conditions for DFE and EE to be asymptotically stable, respectively. Unlike the discrete system obtained by the Euler method where the stability conditions for equilibria depend on the size of time-step h, the proposed discrete system (7) has stability conditions which are consistent with the stability conditions for equilibria of continuous system (4) for any size of time step h; see [
To confirm our previous theoretical analysis, in this section we present some numerical simulations using nonstandard finite difference scheme (7) with time – step h = 0.1 (or k = 10). For the simulations we adopt parameters used by [
For simulations of the DFE scenario we use a constant contact rate and vary the value of p. Using this constant contact rate we have that. First we choose and therefore the conditions for the DFE to be stable are satisfied, i.e..
For the EE scenario we use such that R0 = 2.3810 > 1. Figures 3(a) and (b) show the numerical solutions for (i.e), using and, respectively. These results indicate that if then the EE (0.5524, 0.9523) is asymptotically stable irrespective of.
Next we take the same parameters as used in
. This prediction is confirmed by our numerical solutions; see
Finally we compare the numerical solutions obtained by Euler method with those obtained by our NSFD method. We found that Euler method with a relatively big numerical time-step (h) may produce unrealistic negative and unbounded solutions. However, using the same or even bigger numerical time-step, NSFD method always gives positive and bounded solution. Depending on the parameters used in the simulations, the solution will be periodic or convergent to a correct equilibrium point. For example, in
number of infectives tends to negative infinity (notice that the graph is shown in logarithmic scaled). Using the same value of h the solution of NSFD method oscillates about the EE; see
In this paper we have introduced a discrete delayed SIS epidemic model obtained by a nonstandard finite difference method. Unlike the Euler method, the proposed scheme reproduces exactly the same equilibria as well as their stability conditions as those of continuous model, i.e. if then the delay does not affect the stability of the equilibria. However, in the case of, the stability of equilibria changes when the delay passes a critical value. Here the discrete SIS model with a delay has periodic solutions when the stability is lost. In other words, a Neimark-Sacker bifurcation occurs. It is also shown that our numerical simulations have confirmed our analytical findings.
This research is supported by Direktorat Penelitian dan Pengabdian kepada Masyarakat, Direktorat Jenderal Pendidikan Tinggi Indonesia (Penelitian Unggulan Perguruan Tinggi—Fundamental) via DIPA Brawijaya University No. 0636/023-04.2.16/15/2012. The author would like to thank the anonymous reviewers for their valuable comments and suggestions to improve this article.