Applied Mathematics and Computation 134 (2003) 51–67 www.elsevier.com/locate/amc
Innovation diffusion model in patch environment Wang Wendi
a,b,*,1
, P. Fergola c, C. Tenneriello
c
a
b
Department of Mathematics, Xi an Jiaotong University, Xi an 710049, China Department of Mathematics, Southwest Normal University, Chongqing 400715, China c Dipartimento di Matematica e Applicazioni, ‘‘R. Caccioppoli’’, Universit a di Napoli ‘‘Federico II’’, Italy
Abstract A mathematical model is proposed to describe the dynamics of users of one product in two different patches. Advertisement force, contact rate between users and non-users, population dispersal rate and returning rate from user class to non-user class are chosen as key parameters. For the product with a long life-span, it is assumed that the returning rate is proportional to the number of users. It is shown that this model has a unique positive equilibrium which is globally stable. For the product with a short life-span, time delays are introduced to represent the duration of the product in two different markets and the stability of a positive equilibrium is analyzed in one reasonable case. Periodic advertisements are also incorporated and the existence and uniqueness of positive periodic solutions are investigated. Ó 2002 Elsevier Science Inc. All rights reserved. Keywords: Innovation; Patch; Stability; Periodic; Delay
1. Introduction A problem of interest in market behavior is that of forecasting how a social system reacts to the offer of two different products. This problem can be
*
Corresponding author. E-mail address:
[email protected] (W. Wendi). 1 The author was supported by Doctorate Foundation of Xian Jiaotong University.
0096-3003/02/$ - see front matter Ó 2002 Elsevier Science Inc. All rights reserved. PII: S 0 0 9 6 - 3 0 0 3 ( 0 1 ) 0 0 2 6 8 - 5
52
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
formulated as an innovation diffusion problem according to the modeling ideas developed in many recent papers as, for instance, in [1–7]. By supposing that the channels of communication are represented by external sources (mass media like radio, TV, newspaper, poster, internet, etc.) and interpersonal contacts (word-of-mouth), we assume that the overall population at time t can be divided into two different classes, non-users of innovation and users of innovation. Furthermore, in order to take into account that individuals can give up the use of an innovation, we allow the possibility for users to leave their class and pass to non-user class after a given time period which we can interpret as the safe period of innovation. Therefore, we introduce delay terms in the model representing this process. Since it is important to consider the sales of products in different regions, we also include space structure into our model by incorporating patch environment. As a beginning, we will only consider two patches in this paper. When two patches are isolated, we suppose that the dynamics of both classes are governed by dN1 ¼ c1 N1 k1 A1 N1 d1 N1 þ m1 A1 þ b1 ; dt dA1 ¼ c1 N1 þ k1 A1 N1 d1 A1 m1 A1 ; dt
ð1:1Þ
where N1 is the number of non-users in the first patch, A1 is the number of users in the first patch, b1 is the birth rate of the population in the first patch, d1 is the death rate of the population in the first patch, c1 represents the intensity of advertisement for promoting product-users in the first patch, and k1 is the valid contact rate of users of the product with non-users in the first patch. Similarly, we suppose that the dynamics of non-user-class and user-class in the second patch are governed by dN2 ¼ c2 N2 k2 A2 N2 d2 N2 þ m2 A2 þ b2 ; dt dA2 ¼ c2 N2 þ k2 A2 N2 d2 A2 m2 A2 ; dt
ð1:2Þ
where N2 is the number of non-users in the second patch, A2 is the number of users in the second patch, b2 is the birth rate of the population in the second patch, d2 is the death rate of the population in the second patch, c2 represents the intensity of advertisement for promoting product-users in the second patch, and k2 is the valid contact rate of users of the product with non-users in the second patch. We now suppose that, when the two patches are connected, h1 is the probability that an individual migrates from the first patch to the second patch, h2 the probability that an individual migrates from the second patch to the first patch. Further, we assume that k2 is the fraction of new immigrants from the
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
53
second patch who remain in user-class and that k1 is the fraction of new immigrants from the first patch who remain in user-class. Under these assumptions, our model is: dN1 ¼ c1 N1 k1 A1 N1 d1 N1 þ m1 A1 þ b1 þ h2 ðN2 þ ð1 k2 ÞA2 Þ h1 N1 ; dt dA1 ¼ c1 N1 þ k1 A1 N1 d1 A1 m1 A1 h1 A1 þ k2 h2 A2 ; dt dN2 ¼ c2 N2 k2 A2 N2 d2 N2 þ m2 A2 þ b2 þ h1 ðN1 þ ð1 k1 ÞA1 Þ h2 N2 ; dt dA2 ¼ c2 N2 þ k2 A2 N2 d2 A2 m2 A2 h2 N2 þ k1 h1 A1 : dt ð1:3Þ We will suppose that all the parameters are positive constants except that ki are non-negative constants. If we set Ti ¼ Ni þ Ai , (1.3) implies that T10 ¼ b1 ðd1 þ h1 ÞT1 þ h2 T2 ; T20 ¼ b2 ðd2 þ h2 ÞT2 þ h1 T1 :
ð1:4Þ
Let us define C1 ¼
d2 b 1 þ h2 b 1 þ h2 b 2 ; d1 d2 þ d1 h2 þ h1 d2
C2 ¼
h1 b 1 þ d1 b 2 þ h1 b 2 : d1 d2 þ d1 h2 þ h1 d2
It is easy to see that any positive solution ðT1 ðtÞ; T2 ðtÞÞ of (1.4) satisfies Ti ðtÞ ! Ci
as t ! 1:
Since we are interested in the asymptotic behavior of (1.3), we may regard N1 þ A1 as C1 and N2 þ A2 as C2 . Thus, we have dA1 ¼ ðc1 þ k1 A1 ÞðC1 A1 Þ ðd1 þ m1 þ h1 ÞA1 þ k2 h2 A2 ; dt dA2 ¼ ðc2 þ k2 A2 ÞðC2 A2 Þ ðd2 þ m2 þ h2 ÞA2 þ k1 h1 A1 : dt For simplicity, we denote ðdi þ mi þ hi Þ by ai and denote ki hi by di , i ¼ 1; 2. Hence, we obtain dA1 ¼ ðc1 þ k1 A1 ÞðC1 A1 Þ a1 A1 þ d2 A2 ; dt dA2 ¼ ðc2 þ k2 A2 ÞðC2 A2 Þ a2 A2 þ d1 A1 : dt
ð1:5Þ
54
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
The organization of this paper is as follows. In Section 2, we show that (1.5) has a positive equilibrium which is globally stable. In Section 3, we introduce time delays into the model which represent the duration of the product in two patches and study its stability. In Section 4, we will suppose that the advertisements are periodic in time and will establish sufficient conditions that ensure the existence and uniqueness of positive periodic solutions.
2. Stability of the model Let us begin from d1 ¼ d2 ¼ 0. In this case, (1.5) has a unique positive equilibrium ðA10 ; A20 Þ, where qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2a þ k C þ 4a21 þ k21 C12 1 1 1 1 ; A10 ¼ 2 k1 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2a þ k C þ 4a22 þ k22 C22 2 2 2 1 A20 ¼ ; 2 k2 and (1.5) can be rewritten as dA1 a 1 A1 ¼ ðc1 þ k1 A1 Þ C1 A1 dt c 1 þ k 1 A1 a1 c1 ðA1 A10 Þ ¼ ðc1 þ k1 A1 Þ ðA1 A10 Þ ; ðc1 þ k1 A1 Þðc1 þ k1 A10 Þ dA2 a 2 A2 ¼ ðc2 þ k2 A2 Þ C2 A2 dt c 2 þ k 2 A2 a2 c2 ðA2 A20 Þ ¼ ðc2 þ k2 A2 Þ ðA2 A20 Þ : ðc2 þ k2 A2 Þðc2 þ k2 A20 Þ
ð2:1Þ
Since each equation of this system is independent, it is sufficient to consider the stability of each equation separately. Set Z A1 A1 ðsÞ A10 ds: V1 ¼ c A10 1 þ k1 A1 ðsÞ Calculating the derivative of V1 along the positive solutions of (2.1), we obtain a1 c 1 V_1 6 1 þ ðA1 A10 Þ2 : ðc1 þ k1 A1 Þðc1 þ k1 A10 Þ It follows that A10 is globally stable. Similarly, we see that A20 is globally stable. Thus, ðA10 ; A20 Þ is globally stable.
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
Suppose d1 ¼ 0 and d2 > 0. Then (1.5) is reduced to dA1 a 1 A1 ¼ ðc1 þ k1 A1 Þ C1 A1 ; dt c 1 þ k 1 A1 dA2 a 2 A2 ¼ ðc2 þ k2 A2 Þ C2 A2 þ d2 A1 : dt c 2 þ k 2 A2
55
ð2:2Þ
Since A10 is globally stable for the first equation of the system, the limiting equation of the system is dA2 a 2 A2 ¼ ðc2 þ k2 A2 Þ C2 A2 ð2:3Þ þ d2 A10 : dt c 2 þ k 2 A2 Suppose A21 is the positive solution of the following equation: a2 A2 ðc2 þ k2 A2 Þ C2 A2 þ d2 A10 ¼ 0; c 2 þ k 2 A2 where A2 is the variable of the equation. Since it can be written as dA2 ða2 c2 þ d2 A10 k2 ÞðA2 A21 Þ ¼ ðc2 þ k2 A2 Þ ðA2 A21 Þ : ðc2 þ k2 A2 Þðc2 þ k2 A21 Þ dt Define V2 ¼
Z
A2
A21
A2 ðsÞ A21 ds: c2 þ k2 A2 ðsÞ
Then V_2 ¼ 1 þ
ða2 c2 þ d2 A10 k2 Þ 2 ðA2 A21 Þ : ðc2 þ k2 A2 Þðc2 þ k2 A21 Þ
It follows from [11] that the positive equilibrium ðA10 ; A21 Þ of (2.2) is globally stable. Suppose d1 > 0 and d2 > 0. We will show that (1.5) has a unique positive equilibrium. By setting the right-hand side of (1.5) to 0, we obtain A2 ¼ ½ðc1 þ k1 A1 ÞðC1 A1 Þ a1 A1 =d2 ; A1 ¼ ½ðc2 þ k2 A2 ÞðC2 A2 Þ a2 A2 =d1 : These two curves are parabolic. Set f1 ¼ ½ðc1 þ k1 A1 ÞðC1 A1 Þ a1 A1 =d2 ; f2 ¼ ½ðc2 þ k2 A2 ÞðC2 A2 Þ a2 A2 =d1 : We have f10 ðA1 Þ ¼
k1 C1 þ 2k1 A1 þ c1 þ a1 ; d2
ð2:4Þ
56
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
f20 ðA2 Þ ¼
k2 C2 þ 2k2 A2 þ c2 þ a2 ; d1
f100 ðA1 Þ ¼ 2
k1 ; d2
f200 ðA2 Þ ¼ 2
k2 : d1
By the second equation of (2.4), we obtain 1 k2 C 2 a2 c 2 A2 ¼ 2k2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi þ k22 C22 2k2 C2 a2 þ 2k2 C2 c2 þ a22 þ 2a2 c2 þ c22 þ 4k2 A1 d1 : Denote the right-hand side of this equation by g2 ðA1 Þ. It is easy to see that 3 g2 ðA1 Þ > 0, g20 ðA1 Þ > 0 for A1 P 0. Then, since g200 ðA1 Þ ¼ f200 ðA2 Þ=ðf20 ðA2 ÞÞ < 0, g2 ðA1 Þ is concave. Notice that f1 ð0Þ < 0, g2 ð0Þ > 0, limA1 !þ1 ½f1 ðA1 Þ g2 ðA1 Þ ¼ þ1 and that f1 is convex. The parabolic curves have one and only one intersection point in the interior of first quadrant. Consequently, (1.5) has one and only one positive equilibrium ðA1 ; A2 Þ. Let us check the vector field of (1.5) at ðA; AÞ. When A is positive and sufficiently small, ðc1 þ k1 AÞðC1 AÞ a1 A þ d2 A ¼ ðc1 þ k1 AÞ½C1 A þ ðd2 a1 ÞA=ðc1 þ k1 AÞ > 0; ðc2 þ k2 AÞðC2 AÞ a2 A þ d1 A ¼ ðc2 þ k2 AÞ½C2 A þ ðd1 a2 ÞA=ðc2 þ k2 AÞ > 0: When A is sufficiently large, ðc1 þ k1 AÞðC1 AÞ a1 A þ d2 A ¼ ðc1 þ k1 AÞ½C1 A þ ðd2 a1 ÞA=ðc1 þ k1 AÞ < 0; ðc2 þ k2 AÞðC2 AÞ a2 A þ d1 A ¼ ðc2 þ k2 AÞ½C2 A þ ðd1 a2 ÞA=ðc2 þ k2 AÞ < 0: Notice that (1.5) is cooperative. It follows that (1.5) is permanent. Since the positive equilibrium of (1.5) is unique, it follows from [12] that it is globally attractive. At this time, we can state the main result of this section. Theorem 2.1. System (1.5) has one positive equilibrium which is globally attractive.
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
57
3. The model with time delays The purpose of this section is to modify (1.3) to incorporate duration of the product. The assumption that returning rate is proportional to the number of users seems reasonable when the length of life-span of the product is large. Otherwise, the effect of the length will be significant. Because of this, we introduce time delays due to the life-span of the product into (1.3) and obtain: dN1 ¼ c1 N1 ðtÞ k1 A1 ðtÞN1 ðtÞ d1 N1 ðtÞ þ ed1 s1 ðc1 N1 ðt s1 Þ dt þ k1 A1 ðt s1 ÞN1 ðt s1 ÞÞ þ b1 þ h2 ðN2 þ ð1 k2 ÞA2 Þ h1 N1 ; dA1 ¼ c1 N1 ðtÞ þ k1 A1 ðtÞN1 ðtÞ d1 A1 ðtÞ ed1 s1 ðc1 N1 ðt s1 Þ dt þ k1 A1 ðt s1 ÞN1 ðt s1 ÞÞ h1 A1 ðtÞ þ k2 h2 A2 ðtÞ; dN2 ¼ c2 N2 ðtÞ k2 A2 ðtÞN2 ðtÞ d2 N2 ðtÞ þ ed2 s2 ðc2 N2 ðt s2 Þ dt þ k2 A2 ðt s2 ÞN2 ðt s2 ÞÞ þ b2 þ h1 ðN1 þ ð1 k1 ÞA1 Þ h2 N2 ; dA2 ¼ c2 N2 ðtÞ þ k2 A2 ðtÞN2 ðtÞ d2 A2 ðtÞ ed2 s2 ðc2 N2 ðt s2 Þ dt þ k2 A2 ðt s2 ÞN2 ðt s2 ÞÞ h2 N2 ðtÞ þ k1 h1 A1 : ð3:1Þ Here s1 represents the life-span of the product in the first patch and s2 represents the life-span of the product in the second patch. By the same arguments as those in Section 1, we only need to consider the asymptotic behavior of the following system: dA1 ¼ ðc1 þ k1 A1 ðtÞÞðC1 A1 ðtÞÞ ðd1 þ h1 ÞA1 ðtÞ dt ed1 s1 ðc1 þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ þ k2 h2 A2 ðtÞ; dA2 ¼ ðc2 þ k2 A2 ðtÞÞðC2 A2 ðtÞÞ ðd2 þ h2 ÞA2 ðtÞ dt
ð3:2Þ
ed2 s2 ðc2 þ k2 A2 ðt s2 ÞÞðC2 A2 ðt s2 ÞÞ þ k1 h1 A1 ðtÞ: First, we analyze the local stability of (3.2). In order to make mathematics tractable, we consider the case where k1 ¼ k2 ¼ 0, which means that all the new immigrants are the non-users of the product. In this case, (3.2) is reduced to
58
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
dA1 ¼ ðc1 þ k1 A1 ðtÞÞðC1 A1 ðtÞÞ p1 A1 ðtÞ dt g1 ðc1 þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ; dA2 ¼ ðc2 þ k2 A2 ðtÞÞðC2 A2 ðtÞÞ p2 A2 ðtÞ dt g2 ðc2 þ k2 A2 ðt s2 ÞÞðC2 A2 ðt s2 ÞÞ;
ð3:3Þ
where pi ¼ di þ hi and gi ¼ edi si . It is easy to see that (3.3) has one and only one positive equilibrium ðA1 ; A2 Þ, where 1 A1 ¼ ð1 g1 Þc1 þ ð1 g1 Þk1 C1 p1 þ ð1 g1 Þ2 c21 2ð1 g1 Þk1 þ 2ð1 g1 Þ2 k1 c1 C1 þ 2ð1 g1 Þc1 p1 þ ð1 g1 Þ2 k21 C12 1=2 2 2ð1 g1 Þk1 C1 p1 þ p1 ; A2 ¼
1 ð1 g2 Þc2 þ ð1 g2 Þk2 C2 p2 þ ð1 g2 Þ2 c22 2ð1 g2 Þk2 2
2
þ 2ð1 g2 Þ k2 c2 C2 þ 2ð1 g2 Þc2 p2 þ ð1 g2 Þ k22 C22 1=2 2ð1 g2 Þk2 C2 p2 þ p22 : Linearizing (3.3) at ðA1 ; A2 Þ, we obtain dB1 ¼ q1 B1 ðtÞ g1 ðq1 þ p1 ÞB1 ðt s1 Þ; dt dB2 ¼ q2 B2 ðtÞ g2 ðq2 þ p2 ÞB2 ðt s2 Þ; dt
ð3:4Þ
where qi ¼ ki Ci ci pi 2ki Ai : Note that the two equations of (3.4) are independent. It is sufficient to concentrate on one equation. Let us consider the first equation. The characteristic equation of it is k ¼ q1 g1 ðq1 þ p1 Þ eks : The location of the roots of this equation is indicated by the Hayes theorem (see [9, p. 444]). For convenience, we state this theorem here. Theorem 3.1 (Hayers). If A; B 2 R; then all roots of A ez þ B z ez ¼ 0
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
59
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi have negative real parts if and only if A < 1 and A < B < a2 þ A2 , where a is the root of a ¼ A tan a, such that 0 < a < p. (If A ¼ 0, then a ¼ p=2). By this theorem, we obtain: Lemma 3.2. Assume: (i) s1 ðk1 C1 c1 p1 2k1 A1 Þ < 1. (ii) ð1 g1 Þðk1 C1 c1 2k1 A1 Þ < p1 . qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 (iii) s1 g1 ðk1 C1 c1 2k1 A1 Þ < a21 þ ½s1 ðk1 C1 c1 p1 2k1 A1 Þ , where a1 is the root of a1 ¼ s1 ðk1 C1 c1 p1 2k1 A1 Þ tan a1 , such that 0 < a1 < p. (If k1 C1 c1 p1 2k1 A1 ¼ 0, then a1 ¼ p=2.) Then A1 is asymptotically stable. Similarly, we have: Lemma 3.3. Assume: (i) s2 ðk2 C2 c2 p2 2k2 A2 Þ < 1. (ii) ð1 g2 Þðk2 C2 c2 2k2 A2 Þ < p2 . qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (iii) s2 g2 ðk2 C2 c2 2k2 A2 Þ < a22 þ ½s2 ðk2 C2 c2 p2 2k2 A2 Þ 2 , where a2 is the root of a2 ¼ s2 ðk2 C2 c2 p2 2k2 A2 Þ tan a2 , such that 0 < a2 < p. (If k2 C2 c2 p2 2k2 A2 ¼ 0, then a2 ¼ p=2.) Then A2 is asymptotically stable. By Lemmas 3.2 and 3.3, we have: Theorem 3.4. Assume: (i) si ðki Ci ci pi 2ki Ai Þ < 1; i ¼ 1; 2. (ii) ð1 gi Þðki Ci ci 2ki Ai Þ < pi ; i ¼ 1; 2. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (iii) si gi ðki Ci ci 2ki Ai Þ < a2i þ ½si ðki Ci ci pi 2ki Ai Þ 2 , where a is the root of ai ¼ si ðki Ci ci pi 2ki Ai Þ tan ai , such that 0 < ai < p. (If ki Ci ci pi 2ki Ai ¼ 0, then ai ¼ p=2.) Then the equilibrium ðA1 ; A2 Þ of (3.3) is asymptotically stable. Example. Suppose k2 ¼ 1, C2 ¼ 10, c2 ¼ 1, p2 ¼ 6 and s2 ¼ 1. By numerical calculations, we obtain A2 ¼ 4:356378697, a2 ¼ 2:434890483 and g2 ¼ 0:1353352832. Then we have k2 C2 c2 2k2 A2 p2 1=s2 ¼ 3:851792217; ð1 g2 Þðk2 C c2 2kA2 Þ p2 ¼ 4:009953137;
60
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi s2 g2 k2 C2 c2 2k2 A2 a22 þ ðs2 ðk2 C2 c2 2k2 A2 p2 ÞÞ2 ¼ 2:591693812: Thus, A2 is asymptotically stable. If k1 ¼ 3, c1 ¼ 0:5, C1 ¼ 5, p1 ¼ 4, d1 ¼ 1 and h1 ¼ 3, then ð1 g1 Þ k1 C c1 2kA1 p1 ¼ ð1 expðs1 ÞÞ 1 0 1=2 3 21 þ 29 expðs1 Þ 561 1458 expðs1 Þ þ 961 expðs1 Þ2 C B 4C B A: @14:5 6 þ 6 expðs1 Þ
By means of maple, we see that it is negative. Thus, we only need to consider conditions (i) and (iii) of Lemma 3.2. Note that k1 C1 c1 2k1 A1 p1 1=s1 ¼ 10:5
3½21 þ 29 expðs1 Þ ð561 1458 expðs1 Þ þ 961 expð2s1 ÞÞ1=2 1=s1
: 6 þ 6 expðs1 Þ
By numerical calculations, we see that it is negative if 0 < s1 < 0:1005 or 0:388 < s1 and is positive if 0:1005 < s1 < 0:388. Furthermore, condition (iii) is valid if 0:4427159803 < s or 0 < s1 < 0:08347740397. Thus, the positive equilibrium is asymptotically stable if s1 > 0:4427159803 or 0 < s1 < 0:08347740397 and is unstable if 0:08347740397 < s1 < 0:4427159803. This example shows that stability switch can occur in model (3.3). Let us turn to the global stability of the positive equilibrium. Since the first and second equations of (3.3) have the same form and are independent, we will concentrate on the first one. Let us rewrite the first equation as dA1 ¼ ðc1 þ k1 A1 ðtÞÞ C1 A1 ðtÞ dt p1 A1 ðtÞ þ g1 ðc1 þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ c1 þ k1 A1 ðtÞ c1 A1 ðtÞ A1 ¼ ðc1 þ k1 A1 ðtÞÞ A1 ðtÞ A1 p1 ðc1 þ k1 A1 ðtÞÞðc1 þ k1 A1 Þ h g1 c1 A1 ðt s1 Þ A1 þ k1 C1 A1 ðt s1 Þ A1 2 k1 A1 ðt s1 Þ A1 2k1 A1 ðt s1 Þ A1 A1 i. C1 k1 A1 ðtÞ A1 þ k1 A1 A1 ðtÞ A1 ðc1 þ k1 A1 ðtÞÞ : ð3:5Þ
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
61
Let us define Liapunov function V1 by Z A1 A1 ðsÞ A1 ds: V1 ¼ A1 c1 þ k1 A1 s Calculating the derivative of V1 along the positive solutions of (3.5), we have 2 c1 A1 ðtÞ A1 2 _ V1 ¼ A1 ðtÞ A1 p1 ðc1 þ k1 A1 ðtÞÞðc1 þ k1 A1 Þ h g1 A1 ðtÞ A1 c1 A1 ðt s1 Þ A1 þ k1 C1 A1 ðt s1 Þ A1 2 k1 A1 ðt s1 Þ A1 2k1 A1 ðt s1 Þ A1 A1 C1 k1 A1 ðtÞ A1 i. þ k1 A1 A1 ðtÞ A1 ð3:6Þ ðc1 þ k1 A1 ðtÞÞ: By (3.3), it is easy to see that 0 < A1 ðtÞ < C1 if t is large. It follows that for all large t, we have 2 V_1 6 A1 ðtÞ A1
2 g1 C1 k1 A1 ðtÞ A1 c1 ðc þ k1 C1 Þ h 1 g1 A1 ðtÞ A1 c1 A1 ðt s1 Þ A1 þ k1 C1 A1 ðt s1 Þ A1 2 i. k1 A1 ðt s1 Þ A1 2k1 A1 ðt s1 Þ A1 A1 ðc1 þ k1 A1 ðtÞÞ:
p 1 c1
2
A1 ðtÞ A1
2
þ
ð3:7Þ By Cauchy inequality, we have ! 2 g1 C1 k1 1þ A1 ðtÞ A1 2 c1 ðc1 þ k1 C1 Þ h 2 g1 c1 þ k1 C1 þ k1 þ 2k1 A1 A1 ðtÞ A1 þ 2ðc1 þ k1 A1 ðtÞÞ 2 4 i þ c1 þ k1 C1 þ 2k1 A1 A1 ðt s1 Þ A1 þ k1 A1 ðt s1 Þ A1 :
V_1 6
p1 c1
ð3:8Þ Notice that 0 < A1 ðtÞ < C1 for large t and 0 < A1 < C1 . We have ! 2 p1 c1 g1 C1 k1 _V1 6 1 þ A1 ðtÞ A1 2 c1 ðc1 þ k1 C1 Þ 2 g1 h þ ðc1 þ k1 C1 þ k1 þ 2k1 C1 Þ A1 ðtÞ A1 2c1 2 i þ c1 þ k1 C1 þ 2k1 C1 þ k1 C12 A1 ðt s1 Þ A1 :
ð3:9Þ
62
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
Let us define Liapunov functional V2 by Z t 2 g A1 ðsÞ A1 ds: V2 ¼ 1 c1 þ k1 C1 þ 2k1 C1 þ k1 C12 2c1 ts1 It follow from (3.9) that for all large t, we have, " p1 c1 g C 1 k1 g1 0 ðV1 þ V2 Þ 6 1 þ 1 ðc þ k1 C1 þ k1 2 c1 2c1 1 ðc1 þ k1 C1 Þ # 2 g1 2 þ 2k1 C1 Þ c1 þ k1 C1 þ 2k1 C1 þ k1 C1 A1 ðtÞ A1 : 2c1 ð3:10Þ By [8] we see that A1 is asymptotically stable if p 1 c1 g C1 k1 g1 1 ðc þ k1 C1 þ k1 þ 2k1 C1 Þ 2 c1 2c1 1 ðc1 þ k1 C1 Þ g 1 c1 þ k1 C1 þ 2k1 C1 þ k1 C12 > 0: 2c1
b,1þ
ð3:11Þ
If V ¼ V1 þ V2 , we have Z t 2 A1 ðsÞ A1 ds 6 V ð0Þ < 1: V ðtÞ þ b 0
It is easy to see that A1 ðtÞ is uniformly continuous on ½0; 1Þ. It follows from Barbalat lemma [10] that limt!1 ðA1 ðtÞ A1 Þ2 ¼ 0. Consequently, A1 is globally stable. We are now in a stage to state one main result of this section. Theorem 3.5. Assume gi C i gi ðc þ ki Ci þ ki þ 2ki Ci Þ ci 2ci i ðci þ ki Ci Þ g i ci þ ki Ci þ 2ki Ci þ ki Ci2 > 0; i ¼ 1; 2: 2ci
1þ
p i ci
2
ð3:12Þ
Then the positive equilibrium ðA1 ; A2 Þ of (3.3) is globally stable.
4. Periodic model In this section, we suppose that the advertisement intensity is periodic in time. This seems reasonable in reality. By introducing periodic advertisements into (3.3), we obtain
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
dA1 ¼ ðc1 ðtÞ þ k1 A1 ðtÞÞðC1 A1 ðtÞÞ p1 A1 ðtÞ g1 ðc1 ðt s1 Þ dt þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ; dA2 ¼ ðc2 ðtÞ þ k2 A2 ðtÞÞðC2 A2 ðtÞÞ p2 A2 ðtÞ dt g2 ðc2 þ k2 A2 ðt s2 ÞÞðC2 A2 ðt s2 ÞÞ;
63
ð4:1Þ
where ci ðtÞ, i ¼ 1; 2, are non-negative, continuous and periodic functions with common period T and all the other parameters have the same meaning as the last two sections. Theorem 4.1. Suppose ci ðtÞ > gi ðci ðt s1 Þ þ ki Ci Þ;
i ¼ 1; 2; t 2 ½0; T :
ð4:2Þ
Then there exists at least a positive periodic solution in (4.1). Proof. Since the two equations of this model are independent and have the same structure, we will concentrate on the first one. By (4.2), we can choose > 0 small enough such that c1 ðtÞ >
p1 g C1 ðc1 ðt s1 Þ þ k1 C1 Þ 1 ; C1 C1
t 2 ½0; T :
Define M ¼ maxfjðc1 ðtÞ þ k1 AÞðC1 AÞ p1 A ðc1 ðt s1 Þ þ k1 BÞðC1 BÞj : 0 6 A 6 C1 ; 0 6 B 6 C1 ; t 2 ½0; T g: Then set D ¼ f/ : 6 /ðhÞ 6 C1 ; h 2 ½s1 ; 0 ; j/ðh1 Þ /ðh2 Þj 6 Mjh1 h2 j; h1 ; h2 2 ½s; 0 g: Let Aðt; 0; /Þ be the solution of the first equation of (4.1) through ð0; /Þ. Suppose that P is the Poincare map of the first equation of (4.1). We show that D is positively invariant. Let us consider a positive solution Aðt; 0; /Þ of the first equation of (4.1) with / 2 D. If t is the first time such that A1 ðt; 0; /Þ ¼ C1 , we have dA1 ðt; 0; /Þ ¼ p1 C1 g1 ðc1 ðt s1 Þ þ k1 A1 ðt s1 ; 0; /ÞÞ dt ðC1 A1 ðt s1 ; 0; /ÞÞ < 0:
ð4:3Þ
64
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
If t is the first time such that A1 ðt; 0; /Þ ¼ , we have dA1 ðt; 0; /Þ p1 ¼ ðC1 Þ c1 ðtÞ þ k1 dt C1 ðc1 ðt s1 Þ þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ g1 C1 p1 ðc1 ðt s1 Þ þ k1 C1 ÞC1 g1 > 0: > ðC1 Þ c1 ðtÞ þ k1 C1 C1 ð4:4Þ It follows from (4.3) and (4.4) that D is positively invariant. By Schauder fixed point theorem, we conclude that the Poincare map has a fixed point in D, which shows the first equation of (4.1) has a positive periodic solution. The existence of positive periodic solution in the second equation of (4.2) follows in the same way. The proof is complete. Let us consider the global stability of positive periodic solution of (4.1). Suppose that A1 ðtÞ is a positive periodic solution of the first equation of (4.1). Set k ¼ maxt2½0;T c1 ðtÞ. Theorem 4.2. Let (4.2) be satisfied. Then A1 ðtÞ is globally stable if " # 2 c1 ðt s1 ÞC1 þ k1 C12 8k1 C12 þ 2c1 C1 p1 k1 g1 > 0; a, g1 Þ2 Þ2 C1 ðC1 A ðC1 A > 0 satisfies where A Þ p1 A ÞðC1 A ¼ 0: ðk þ k1 A Proof. The first equation of (4.1) can be rewritten as dA1 ðtÞ p1 A1 ðtÞ ¼ ðC1 A1 ðtÞÞ c1 ðtÞ þ k1 A1 ðtÞ dt C1 A1 ðtÞ ðc ðt s1 Þ þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ g1 1 : C1 A1 ðtÞ
ð4:5Þ
Obviously, dA1 ðtÞ p1 A1 ðtÞ ¼ C1 A1 ðtÞ c1 ðtÞ þ k1 A1 ðtÞ dt C1 A1 ðtÞ c1 ðt s1 Þ þ k1 A1 ðt s1 Þ C1 A1 ðt s1 Þ g1 : C1 A1 ðtÞ
ð4:6Þ
Let A1 ðtÞ be a positive solution of the first equation of (4.1). Then define V1 ðtÞ ¼ lnðC1 A1 ðtÞÞ ln C1 A1 ðtÞ :
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
65
It is not hard to obtain p1 C1 A1 ðtÞ A ðtÞ þ k1 A1 ðtÞ A ðtÞ 1 1 ðC1 A1 ðtÞÞðC1 A1 ðtÞÞ ðc ðt s1 Þ þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ þ g1 1 C1 A1 ðtÞ c1 ðt s1 Þ þ k1 A1 ðt s1 Þ C1 A1 ðt s1 Þ : C1 A1 ðtÞ
Dþ V1 6
Note that ðc1 ðt s1 Þ þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ C1 A1 ðtÞ c1 ðt s1 Þ þ k1 A1 ðt s1 Þ C1 A1 ðt s1 Þ C1 A1 ðtÞ h ¼ k1 C1 A1 ðtÞ A1 ðt s1 Þ A1 ðt s1 Þ 2k1 C1 A1 ðt s1 Þ A1 ðt s1 Þ A1 ðt s1 Þ þ 2k1 A1 ðtÞA1 ðt s1 ÞðA1 ðt s1 Þ A1 ðt s1 ÞÞ 2 þ k1 A1 ðt s1 ÞC1 A1 ðtÞ A1 ðtÞ k1 A1 ðt s1 Þ A1 ðtÞ A1 ðtÞ c1 ðt s1 ÞC1 A1 ðt s1 Þ A1 ðt s1 Þ þ c1 ðt s1 ÞA1 ðtÞ A1 ðt s1 Þ A1 ðt s1 Þ þ k1 C12 A1 ðt s1 Þ A1 ðt s1 Þ k1 C1 A1 ðt s1 Þ 2 2 A1 ðt s1 Þ þ k1 A1 ðtÞ A1 ðt s1 Þ A1 ðt s1 Þ i þ c1 ðt s1 ÞC1 A1 ðtÞ A1 ðtÞ c1 ðt s1 ÞA1 ðt s1 Þ A1 ðtÞ A1 ðtÞ . ½ðC1 A1 ðtÞÞ C1 A1 ðtÞ : Since 0 < A1 ðtÞ < C1 , 0 < A1 ðtÞ < C1 and c1 ðtÞ 6 k for all large t, we have ðc1 ðt s1 Þ þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ C1 A1 ðtÞ c1 ðt s1 Þ þ k1 A1 ðt s1 Þ C1 A1 ðt s1 Þ C1 A1 ðtÞ 2 2 6 2 kC1 þ k1 C1 A1 ðtÞ A1 ðtÞ þ 8k1 C1 þ 2kC1 A1 ðt s1 Þ A1 ðt s1 Þ ðC1 A1 ðtÞÞ C1 A1 ðtÞ : In order to estimate eventual lower bound of the denominator, let us consider the auxiliary equation dA1 ¼ ðk þ k1 A1 ÞðC1 A1 Þ p1 A1 : dt
ð4:7Þ
66
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
is a positive equilibrium of this equation and is globally It is easy to see that A þ and A ðtÞ 6 stable. Then by comparison principle, we see that A1 ðtÞ 6 A 1 A þ for all large t, where is a very small positive constant. For simplicity in and A ðtÞ 6 A for all large t. This will not innotation, we suppose A1 ðtÞ 6 A 1 fluence our final results. As a consequence, ðc1 ðt s1 Þ þ k1 A1 ðt s1 ÞÞðC1 A1 ðt s1 ÞÞ C1 A1 ðtÞ c1 ðt s1 Þ þ k1 A1 ðt s1 Þ C1 A1 ðt s1 Þ C1 A1 ðtÞ 6 2 kC1 þ k1 C12 A1 ðtÞ A1 ðtÞ þ 8k1 C12 þ 2kC1 A1 ðt s1 Þ Þ2 : A ðt s1 Þ =ðC1 A 1
Hence, we have "
# 2 kC1 þ k1 C12 p1 D V1 6 k1 g1 A1 ðtÞ A1 ðtÞ 2 Þ C1 ðC1 A 8k1 C12 þ 2kC1 þ g1 A1 ðt s1 Þ A1 ðt s1 Þ: 2 Þ ðC1 A þ
Set
8k1 C12 þ 2kC1 V2 ¼ g1 Þ2 ðC1 A
Z
t
A1 ðsÞ A ðsÞ ds: 1
ts
Then we have "
2 kC1 þ k1 C12 p1 D ðV1 þ V2 Þ 6 k1 g1 Þ2 C1 ðC1 A # 8k1 C12 þ 2kC1 g1 A1 ðtÞ A1 ðtÞ: 2 Þ ðC1 A þ
ð4:8Þ
It follows from [8] that A1 ðtÞ is asymptotically stable. If V ¼ V1 þ V2 , an integration of (4.8) leads to Z t A1 ðsÞ A ðsÞ ds 6 V ð0Þ < 1: V ðtÞ þ a 1 0
It is easy to see that A1 ðtÞ and A1 ðtÞ are uniformly continuous on ½0; 1Þ. It follows from Barbalat lemma [10] that limt!1 jA1 ðtÞ A1 ðtÞj ¼ 0. Hence, the positive periodic solution is globally stable. The proof is complete.
W. Wendi et al. / Appl. Math. Comput. 134 (2003) 51–67
67
References [1] V. Mahajan, E. Muller, R.A. Kerin, Introduction strategy for a new products with positive and negative word-of-mouth, Management Sci. 30 (1984) 1389–1404. [2] V. Mahajan, R. Peterson, Models for Innovation Diffusion, Sage, Beverly Hills, CA, 1985. [3] V. Mahajan, Y. Wind, Innovation Diffusion Models of New Product Acceptance, Bellinger, Cambridge, 1986. [4] R.W. Mizerski, An attribution explanation of the disproportionate influence of unfavorable information, J. Consumer. Res. 9 (1982) 301–310. [5] P. Fergola, C. Tenneriello, Z. Ma, F. Petrillo, Delayed innovation diffusion processes with positive and negative word-of-mouth, Int. J. Diff. Equa. Appl. 1 (2000) 131–147. [6] P. Fergola, C. Tenneriello, Z. Ma, F. Petrillo, An innovation diffusion model with time delay: positive and negative word of mouth, to appear. [7] P. Fergola, F. Petrillo, Some new results on the stability of a delayed innovation diffusion model, to appear. [8] J.K. Hale, S.M.V. Lunel, Introduction to Functional Differential Equations, Springer, New York, 1993. [9] R. Bellman, K.L. Cooke, Differential–Difference Equations, Academic Press, New York, 1963. [10] I. Barbalat, Systems d’equations differentielles d’oscillations non-lineaires, Rev. Roumaine. Math. Pures Appl. 4 (1959) 267–270. [11] J.P. Lasalle, The stability of dynamical systems, Hamilton Press, Berlin, 1976. [12] H.L. Smith, Monotone Dynamical Systems, Mathematical Surverys and Monographs, vol. 41, American Mathematical Society, Providence, RI, 1995.