DISCRETE AND CONTINUOUS DYNAMICAL SYSTEMS SUPPLEMENT 2007 Website: www.AIMSciences.org pp. 373–381 EXISTENCE OF POSITIVE SOLUTIONS FOR SECOND ORDER DIFFERENTIAL EQUATIONS ARISING FROM CHEMICAL REACTOR THEORY Wenying Feng Computer Science and Mathematics, Trent University Peterborough, ON Canada K9J 7B8 Guang Zhang School of Science, Tianjin University of Commerce Tianjin 300134, P. R. China Yikang Chai Department of Mathematics, Qingdao Technological University No. 11 Fushun Road, Qingdao 266033, P. R. China Dedicated to Prof. J.R.L. Webb on the occasion of his 60th birthday Abstract. In this paper, we study second-order differential equations that represent the steady state model in an adiabatic tubular chemical reactor. Theoretical results on existence and range of positive solutions are proved by applying a fixed point theorem. At the mean time, numerical solutions are obtained by computer programming. Results from mathematical analysis are compared with the numerical solutions. 1. Introduction. As mathematical models for an adiabatic tubular chemical reactor, the following boundary value problem of second order differential equations has been extensively studied (see [2]-[7]): x00 − λx0 + λµ (β − x) ex = 0, t ∈ (0, 1) , 0 0 x (0) = λx (0) , x (1) = 0. (1) (2) BVP (1)-(2) describes the steady state of the reactor that processes an irreversible exothermic chemical reaction (see [5], [6]). The three parameters, λ > 0, µ > 0 and β ≥ 0, represent the Peclet number, the Damkohler number and the dimensionless adiabatic temperature rise, respectively. A positive solution x to the system represents the steady state temperature of the reaction. Most of the results on problem (1)-(2) are obtained by numerical methods ([4], [5], [7]). For example, Madbouly, McGhee and Roach applied the Adomian’s method for Hammerstein integral equations and the shooting method [5]. In a recent paper [7], Saadatmandi, Razzaghi and Dehghan proposed the application of the Sinc-Galerkin 2000 Mathematics Subject Classification. Primary: 34B15; Secondary 34B18. Key words and phrases. Steady state model, positive solution, fixed point theorem, numerical simulation. The first author is supported by a grant from NSERC (Natural Sciences and Engineering Research Council of Canada). 373 374 WENYING FENG, GUANG ZHANG AND YIKANG CHAI method for solving (1)-(2). Existence of multiple solutions for parameters in some ranges are also discussed (see [2], [3]). The following existence and uniqueness theorem is given in [5]: Theorem 1. For any M > 0, the problem (1)-(2) has a unique solution in B (M ) = {x ∈ C [0, 1] : ||x||∞ ≤ M } provided ¾ ½ M 1 µ < min , . (3) kKk (β + M ) eM kKk (|β − 1| + M ) eM Here, ||x||∞ = supt∈[0,1] |x(t)|. Theorem 1 is proved by the contraction mapping principle [4]. It is noticed that lim µ = 0. M →∞ Therefore, Theorem 1 specifies that to have a solution in a bigger range (M is bigger), the value of µ should be smaller. This is not clearly suitable. In this paper, we apply the following well known fixed point theorem [1] to obtain results on the existence of positive solutions for (1)-(2). Lemma 1. [1] Let E be a Banach space. Assume Ω1 and Ω2 are two bounded open¡ sets in ¢ E such that 0 ∈ Ω1 and Ω1 ⊂ Ω2 , and P is a cone in E. Let T : P ∩ Ω2 \Ω1 → P is completely continuous and one of the two conditions is satisfied: 1. kT xk ≤ kxk for x ∈ P ∩ ∂Ω1 and kT xk ≥ kxk for x ∈ P ∩ ∂Ω2 ; 2. kT xk ≥ kxk for x ∈ P ∩ ∂Ω1 and kT xk ≤ kxk for x ∈ P ∩ ∂Ω2 . ¡ ¢ Then T has at least one fixed point in P ∩ Ω2 \Ω1 . The proofs of the theorems are based on the construction of various open sets. Our results not only provide sufficient conditions on existence of a positive solution, but also specify the possible ranges, thus the upper and lower bounds for the solutions. Using Matlab programming, we are able to obtain some numerical solutions with various input parameters. Comparisons among the theoretical results and the numerical solutions provide more information on the positive solutions. For example, it proves that the solution is not unique for some specified parameters. In Section 2, we present the theorems on existence of a positive solution. Solutions from computer simulation and comparisons are given in Section 3. 2. Existence of positive solutions. The existence of solutions for (1)-(2) is equivalent to the existence of solutions to the Hammerstein integral equation [5] Z 1 x (t) = µ G (t, s) (β − x (s)) ex(s) ds, 0 ≤ t ≤ 1, (4) 0 where the Green’s function G (t, s) is defined by ½ λ(t−s) e , 0≤t≤s≤1 G (t, s) = 1, 0≤s≤t≤1 and satisfies the condition e−λ ≤ G (t, s) ≤ 1. When β = 0, equation (4) has a unique zero solution, see [3]. Thus, we will assume that β > 0. Our first theorem gives a sufficient condition for a positive solution that has uniform upper bound. We also prove that the maximum of the solution is greater than a constant. In applications, this implies that the highest temperature during the reaction is higher than the constant. EXISTENCE OF POSITIVE SOLUTIONS FOR APPLIED ODES 375 Theorem 2. Let λ, µ and β be positive numbers with µeβ ≤ 1. Boundary value problem (1)-(2) has at least one positive solution x such that max x(t) > t∈[0,1] µβ and x(t) ≤ β for t ∈ [0, 1]. µ + eλ Proof: Since µeβ ≤ 1, we have µ < 1 < eλ and µβe−λ < β. Let E = {u : u (t) ∈ C [0, 1]} with the norm ||u||∞ = maxt∈[0,1] |u (t) | for any u ∈ E. Assume that P = {u ∈ E : u (t) ≥ 0} and Ω = {u ∈ E : ||u||∞ < β} . Clearly, P is a cone of E. Define an operator T : P → E by Z 1 (T x) (t) = µ G (t, s) (β − x (s)) ex(s) ds, 0 ≤ t ≤ 1, (5) 0 which is completely continuous on Ω ∩ P . Let r = µβ µ+eλ and define the open set Ω1 = {x ∈ E : ||x||∞ < r} . For x ∈ E and x ∈ ∂Ω1 ∩ P , Z (T x)(t) = 1 µ G (t, s) (β − x(s)) ex(s) ds 0 ≥ ≥ = Z µ eλ 1 (β − x(s)) ex(s) ds 0 ¶ Z µ µ 1 µβ ds β− eλ 0 µ + eλ µβ = ||x||∞ . µ + eλ Thus, ||T x||∞ ≥ ||x||∞ . On the other hand, let Ω2 = Ω ⊃ Ω1 . Then, for x ∈ ∂Ω2 ∩ P , we have Z 1 (T x)(t) = µ G (t, s) (β − x (s)) ex(s) ds 0 Z ≤ < 1 µ (β − x(s))ex(s) ds 0 β µβe ≤ β = ||x||∞ . Hence, ¡ ||T x|| ¢ ∞ ≤ ||x||∞ . Lemma 1 ensures that BVP (1)-(2) has a positive solution x ∈ Ω2 \Ω1 , which implies max x(t) > t∈[0,1] The proof is complete. µβ and x(t) ≤ β for t ∈ [0, 1]. µ + eλ 376 WENYING FENG, GUANG ZHANG AND YIKANG CHAI In the following, we denote y (t) = 1 − x(t) β . The integral equation (4) reduces to Z 1 y (t) = 1 − µeβ G (t, s) y (s) e−βy(s) ds. (6) 0 Existence of a solution for (1)-(2) is equivalent to the following BVP: Z 1 G(t, s)y(s)e−βy(s) ds, y(t) = 1 − µeβ (7) 0 y 0 (0) = λ(y(0) − 1), y 0 (1) = 0. (8) Our next result gives a sufficient condition on existence of a positive solution with a minimum value less than a constant. In applications, it implies that at one point, temperature of the reaction is lower than the constant. Theorem 3. Let λ, µ and β be positive numbers with µeβ ≤ 1. If r is a positive 1 number that satisfies r ≤ 1+µe β . Then BVP problem (1)-(2) has at least one positive solution x such that min x(t) < β(1 − r) and x(t) ≤ β for t ∈ [0, 1]. t∈[0,1] In particular, there exists a positive solution x(t) such that β min x(t) < and x(t) ≤ β for t ∈ [0, 1]. 2 t∈[0,1] Proof. We prove that there exists a fixed point for the operator T1 : E → E defined by Z 1 (T1 y)(t) = 1 − µeβ G(t, s)y(s)e−βy(s) ds. (9) 0 Let Ω3 = {y ∈ E : ||y||∞ < 1}. Note that βy < eβy and e−λ ≤ G(t, s) ≤ 1 for t, s ∈ [0, 1], thus µeβ ≤ 1 ensures that Z 1 0 ≤ 1 − µeβ G(t, s)y(s)e−βy(s) ds ≤ 1 for y ≥ 0 and ||y||∞ ≤ 1. 0 So, T1 : P ∩ Ω3 → P and ||T1 y||∞ ≤ ||y||∞ for y ∈ P ∩ ∂Ω3 . Let Ω4 be the open set defined by Ω4 = {y ∈ E : ||y||∞ < r}. Then for y ∈ P ∩ ∂Ω4 , we have Z 1 Z 1 µeβ G(t, s)y(s)e−βy(s) ds ≤ µeβ y(s)e−βy(s) ds ≤ µeβ r. Since r ≤ 0 1 , 1+µeβ 0 (T1 y)(t) ≥ 1 − µeβ r ≥ r = ||y||∞ . By Lemma 1, T1 has a fixed point y1 ∈ Ω3 \Ω4 , which is a positive solution of (7)-(8). Let x1 (t) = β(1 − y1 (t)), t ∈ [0, 1]; then x1 is a positive solution of (1)-(2). Obviously, x1 (t) ≤ β for t ∈ [0, 1]. Also, maxt∈[0,1] y1 (t) > r implies that minx∈[0,1] x1 (t) < β(1 − r) . 1 In particular, let r = 12 ≤ 1+µe β . We obtain that there exists a positive solution that satisfies β and x(t) ≤ β for t ∈ [0, 1]. min x(t) < 2 t∈[0,1] EXISTENCE OF POSITIVE SOLUTIONS FOR APPLIED ODES 377 Theorem 4. Assume that µ < βe1−β . For any λ > 0, the BVP (1)-(2) has at least one positive solution x(t) such that min x(t) < µeβ−1 and x(t) ≤ β for t ∈ [0, 1]. t∈[0,1] Proof. Similar to the proof of Theorem 3, we prove that there exists a fixed point for the integral operator T1 . Define the open set Ω5 as the following: ¾ ½ µeβ−1 . Ω5 = y ∈ E : ||y||∞ < 1 − β Then Ω5 ⊂ Ω3 . Consider the function f (x) = xe−βx , we have f 0 (x) = e−βx (1 − βx). The unique solution of f 0 (x) = 0, x ∈ (0, +∞) is x = 1/β. Therefore, 1 sup xe−βx = . βe x∈R For any y ∈ P ∩ ∂Ω5 , we have (T1 y)(t) = Z 1 − µeβ 1 G(t, s)y(s)e−βy(s) ds 0 β−1 Z 1 µe G(t, s)ds β 0 µeβ−1 ≥ 1− = ||y||∞ . β Thus, ||T1 y||∞ ≥ ||y||∞ for y ∈ P ∩ ∂Ω5 . Again, by Lemma 1, T1 has a fixed point in Ω3 \Ω5 , which is a positive solution for problem (7)-(8). The rest of the proof follows the proof of Theorem 3. ≥ 1− The following corollary can be easily verified by Theorem 4. Corollary 1. (a) Assume that µ < β ≤ 1. Then for any λ > 0, the BVP (1)-(2) has a positive solution x such that min x(t) < µeβ−1 and x(t) ≤ β for t ∈ [0, 1]. t∈[0,1] (b) In particular, let β = 1 and µ < 1. Then for any λ > 0, there exists a positive solution of (1)-(2) that satisfies min x(t) < µ and max x(t) ≤ 1 for t ∈ [0, 1]. t∈[0,1] t∈[0,1] Remark 1. In Theorem 1, let M = β, condition (3) is equivalent to the following: 1 , if 0 < β ≤ 32 , 2kKk β µe < 1 1 < , if β > 32 . (2β − 1)kKk 2kKk 1 Therefore, when (1) kKk > 12 or (2) kKk ≤ 12 and β > max{ 23 , 2kKk + 12 }, the condition of Theorem 2 is weaker than that of Theorem 1. It is easy to find examples 378 WENYING FENG, GUANG ZHANG AND YIKANG CHAI that satisfies Theorem 2 but Theorem 1 can not be applied. In addition, Theorem 2 gives the lower bound of kxk for a solution of problem (1)-(2). As mentioned earlier, this implies the highest temperature during the reaction is higher than a constant. Remark 2. Note that βe1−β − e−β = (βe − 1)e−β implies βe1−β > e−β for βe > 1, and that βe1−β < e−β for 0 < βe < 1. Thus, Theorem 4 is different from Theorems 2 and 3. 3. Comparison of numerical solutions. In this section, we solve for numerical solutions for given parameters by applying computer simulations. Let f (x) = λµ (β − x) ex . By a simple difference calculation, we can obtain the following system of algebraic equations: ½ (2 + λh) xk+1 − 4xk + (2 + λh) xk−1 + 2h2 f (xk ) = 0, k = 1, 2, ..., n, (10) x1 x0 = 1+hλ , xn+1 = xn , where the step length h = 1/ (n + 1). The simulation is done by using Matlab. The program used for the procedure is given below (here, b stands for β): function f=fff(x,lam,mu,b) f=zeros(size(x)); n=length(x); h=1/(n-1); f(1)=x(2)-(1+h*lam)*x(1); x1=x(1:n-2); x2=x(2:n-1); x3=x(3:n); f(2:n-1)=(2+lam*h).*x1-4.*x2+(2-lam*h).*x3 +2*h*h*lam*mu.*(b-x2).*exp(x2); f(n)=x(n)-x(n-1); Given the values of λ, µ and β, the program enables us to obtain sample solutions for (1)-(2). Some interesting facts are observed by comparison between the numerical solutions and the theoretical results obtained in Section 2. First, choose λ = 10, µ = 0.55 and β = 0.57. The numerical positive solution obtained is shown in Figure 1. The solution satisfies x(t) > 0.289 > 0.5β for t ∈ [0, 1]. Such a solution satisfies Theorem 2 but does not satisfies Theorem 3. Therefore, Theorem 3 actually ensures that the equation has at least two positive solutions. On the other hand, with the inputs λ = 0.05, µ = 0.5, β = 0.6, BVP (1)-(2) has the numerical solution shown in Figure 2. Such a solution satisfies both Theorems 2 and 3 In fact, β µβ = 0.3 > x(t) > 0.23 > = 0.1974, for all t ∈ [0, 1]. 2 µ + e−λ Second, let λ = 5, µ = 0.05, and β = 0.53 (which satisfy all conditions of Theorem 2), two numerical positive solutions are obtained as shown in Figures 3 and 4 respectively. EXISTENCE OF POSITIVE SOLUTIONS FOR APPLIED ODES 379 Figure 1. Numerical solution when λ = 10, µ = 0.55, β = 0.57 Figure 2. Numerical solution when λ = 0.05, µ = 0.5, β = 0.6 Figure 3. First solution when λ = 5, µ = 0.05, β = 0.53 It is noticed that the solution in Figure 3 does satisfy the bound conditions of Theorem 2. The solution in Figure 4, however, is not consistent with any of the 380 WENYING FENG, GUANG ZHANG AND YIKANG CHAI Figure 4. Second solution when λ = 5, µ = 0.05, β = 0.53 theorems of Section 2 because of ||x||∞ > β. This raises the open problem that does BVP (1)-(2) have at least one positive solution with kxk∞ > β? Third, when λ = 5, µ = 0.7, β = 0.8, it can be verified that all conditions of Theorem 4 hold but the conditions of Theorem 2 and 3 are not satisfied. In this case, the numerical solution can also be obtained. Please see Figure 5. Figure 5. Numerical solution when λ = 5, µ = 0.7, β = 0.8 At the last, we give an example of numerical solution with the parameter λ < 0. In fact, speaking from the mathematical side, our theorems are valid when λ < 0. As such an example, Figure 6 shows a solution with inputs λ = −3, µ = 0.05 and β = 0.6. EXISTENCE OF POSITIVE SOLUTIONS FOR APPLIED ODES 381 Figure 6. Numerical solution when λ = −3, µ = 0.05, β = 0.6 Acknowledgements. The authors thank both referees for valuable comments, especially the correction of Theorem 2 and the suggestion of Remark 1. REFERENCES [1] D. J. Guo and V. Lakshmikantham, “Nonlinear Problems in Abstract Cones,” Academic Press, 1988. [2] R. Heinemann and A. B. Poore, Multiplicity stability and oscillatory dynamics of the tubular reactor, Chemical Engineering Science, 36 (1981), 1411–1419. [3] R. Heinemann and A. B. Poore, The effect of activation energy on tubular reactor multiplicity, Chemical Engineering Science, 37 (1982), 128–131. [4] N. Madbouly, “Solutions of Hammerstein Integral Equations Arising from Chemical Reactor Theory,” Ph.D. Thesis, University of Strathclyde, 1996. [5] N. M. Madbouly, D. F. McGhee and G. F. Roach, Adomian’s method for Hammerstein integral equations arising from chemical reactor theory, Appl. Math. Comput., 117 (2001), 241–249. [6] A. B. Poore, A tubular chemical reactor model, in “A Collection of Nonlinear Model Problems Contributed to the Proceedings of the AMS-SIAM,” (1989), 28–31. [7] A. Saadatmandi, M. Razzaghi and M. Dehghan, Sinc-Galerkin solution for nonlinear twopoint boundary value problems with applications to chemical reactor theory, Math. Comput. Modelling, 42 (2005), 1237–1244. Received September 2006; revised March 2007. E-mail address: wfeng@trentu.ca