Heat Conduction in a Rod with a Non-uniform Thermal Di¤usivity J. L. Taylor1 Abstract Let L be the length of a rod and u (x; t) be its temperature for (x; t) 2 [0; L] [0; 1). We assume that the initial and boundary temperatures of the rod are f (x) and 0 respectively. This heat conduction problem is formulated as the following …rst initial-boundary value problem: ut = 1 u(x; 0) = f (x) for 0 uxx for 0 < x < L; t > 0; x L; u(0; t) = 0 = u(L; t) for t > 0: where is a positive function and f is a nonnegative function. In this paper, we study an approximated solution to the above problem. 1. Introduction The heat equation can be used to describe the transfer of heat in a given material. The transfer of heat from one material to another creates problems in the machines’functions. This is the motivation to study heat conduction in materials. Suppose that a rod is placed along the x-axis with x = 0 at the left end of the rod and x = L at the right end. The heat conduction problem of the rod is described by (cf. Nagle et al. [3, p. 578]): c @u (x; t) @ = @t @x K0 @u (x; t) @x + Q (x; t) where c, , K0 , Q, and u are corresponding to speci…c heat, mass density, thermal conductivity, heat source, and temperature respectively. In the following discussion, we assume that there is no heat source, and K0 is a positive constant. Then, the above expression becomes ut = 1 uxx where = c =K0 and 1= is called the thermal di¤usivity. We assume that is a twice-di¤erentiable positive function of x. Furthermore, the rod satis…es the homogeneous boundary condition. That is, the temperature of the rod at each end is 0. Also, the rod has an initial temperature 1 This paper is the research work of the independent study in Spring 2006 semester under the supervision of Prof. W. Y. Chan in the Department of Mathematics at Southeast Missouri State University. Email Address: taylorseries1580@yahoo.com 1 f (x) which is a di¤erentiable function. Then, the following …rst initial-boundary value problem is formulated: ut = 1 u(x; 0) = f (x) for 0 uxx for 0 < x < L; t > 0; x L; u(0; t) = 0 = u(L; t) for t > 0: (1) (2) To solve the problem (1)-(2), the method of separation of variables is used. We assume that u(x; t) = (x)h(t). Substitute this expression into (1); it leads to the following two ordinary di¤erential equations: dh + h = 0; dt (3) d2 + = 0; (0) = 0 = (L); (4) dx2 where is a constant. Equation (4) is called Sturm-Liouville eigenvalue problem. Let n and n be the corresponding eigenvalues and eigenfunctions of (4) respectively. f n g forms an orthogonal set with the weight function (x). That is, Z L (x) n (x) m (x) dx = 0 for n 6= m: 0 To each n , the solution of (3) is hn (t) = e n t . From the result of Haberman [2, p. 142], the solution of the problem (1)-(2) is given by the following in…nite series: 1 X an n (x) e n t (5) u (x; t) = n=1 where an = RL 0 (x) f (x) n (x) dx : RL (x) 2n (x) dx 0 (6) However, the analytic solution to n (x) in (4) may not be obtained for any given (x). In Section 2, we study the approximated solution to n (x). In Section 3, we give an example to show the approximated solution of the problem (1)-(2) with given functions (x) and f (x). 2. Sturm-Liouville eigenvalue problem The general form of Sturm-Liouville eigenvalue problem (cf. Gustafson [1, p.175]) is given by d d p +q + r =0 dx dx where p, q, and r are functions of x, and q = 0, and r = , then it gives d2 + dx2 is the eigenvalue. If we let p = K0 , = 0: 2 (7) The Liouville-Green approximation is used to determine the approximated value of (x) when is large. The following theorem is from Olver [4, pp. 190-191]. The book only provide the outline of the proof. We give the detail of the proof. Theorem 1. If 1, the approximation for (x) is given by (x) A~ 1 4 Rx e 0 ( )1=2 ds 1 4 ~ +B e Rx 0 ( )1=2 ds ~ are arbitrary constants. where A~ and B Proof. Let g (x) = (x), then (7) becomes d2 = g (x) : dx2 Let (x) = and Rx 0 g 1=2 (s) ds and (x) = ( 0 (x)) 1=2 (8) W ( ). Then, d = g 1=2 dx dW 0 d = ( ) dx dx W 0 ( ) 2 1 2 3 2 00 : Also, W d h 0 23 00 i ( ) 2 dx 3 0 25 00 2 W ( ) ( ) + ( 0) 2 2 dW d h 0 23 00 i ( ) d dx d2 W 0 21 dW 0 23 00 d2 = ( ) ( ) dx2 dx2 dx dW 0 23 00 d2 W 0 21 ( ) ( ) = 2 dx dx 1 d dW d d = ( 0) 2 d d dx dx 3W 0 25 00 2 W 0 23 ( ) ( ) ( ) + 4 2 000 3 : 3 2 000 By 1 d d2 = g = g 1=2 , dx dx2 2 1=2 0 g , and 1 d3 = g dx3 4 3=2 1 2 (g 0 ) + g 2 1 1 d dW 1 dW 3W d2 1 1 0 = g2 g4 g 4 g 2 g + g 2 dx d d d 2 4 1 3 0 2 1 1 00 W 3 g 4 g 2 (g ) + g 2 g 2 4 2 1 1 dg d2 W 3 dW 1 3 dg dW = g4 + g 4 g 4 2 d d 2d 2 dx d W 3 9 0 2 1 9 0 2 1 5 00 g 4 (g ) g 4 (g ) + g 4 g 2 8 4 2 d2 W 3 1 1 dW dg 1 3 dg dW d g4 + g 4 g 4 = d 2 2 d d 2 d d dx 5 9 0 2 1 5 00 W g 4 (g ) + g 4 g 2 8 2 2 d W 3 W 5 9 0 2 1 5 00 = g4 g 4 (g ) + g 4 g : 2 d 2 8 2 5 4 1=2 00 1 g 4 g , we have 1 2 (g 0 ) Then (8) becomes, d2 W 3 g4 d 2 W 2 5 g 8 9 4 1 2 (g 0 ) + g 2 5 4 3 g 00 = g 4 W: It is equivalent to d2 W =W d 2 Let ' = g 3=4 d2 g dx2 1=4 5 g 16 3 . By g = 1 2 (g 0 ) + g 4 2 00 g +1 : (9) , we have 5 3 0 2 1 g (g ) + g 16 4 2 00 5 ( 0) = : 16 3 4 2 '= 2 00 g Therefore, (9) becomes d2 W = (1 + ') W: d 2 Now, if 1, then ' is negligible. The approximated equation of (10) is d2 W d 2 W: Therefore, the approximated solution to W is W Ae + Be 4 (10) where A and B are arbitrary constants. By 1=2 ( 0 (x)) W , we obtain (x) By g = 1 4 Ag Rx e 0 g 1=2 ds (x) = 1 4 + Bg Rx 0 e Rx e Rx 0 g 1=2 (s) ds and g 1=2 ds (x) = : , we get (x) where A~ = A ( ) A~ 1=4 1 4 Rx e 0 )1=2 ds ( ~ =B( and B 1 4 ~ +B 1=4 ) 0 ( )1=2 ds ; . The proof is completed. Adjustment can be made to the Liouville-Green approximation. This adjustment is from Haberman [2, p. 184], and it deals with the envelope function, ~ 1=4 . The idea is to set the envelope function as an in…nite namely A~ 1=4 or B series. Theorem 2. If 1, there exists an in…nite series E(x) = 1 X n 2 En (x) (11) n=0 where En satis…es En (x) = i 2 1 4 Z 1=2 1 4 Rx 1=2 1=4 0 with E0 = such that E(x)e i (7). Rx Proof. Let ! = i 1=2 0 1=2 (s)ds, then d! =i dx and If d2 ! =i dx2 (x) En00 1 2 ds 1 2 1 2 2 (x) dx is an approximation solution to ; 0 1 2 1 (12) : (13) E (x) e! , then d dx E 0 e! + Ee! d! ; dx d! + Ee! dx d! dx and d2 dx2 E 00 e! + 2E 0 e! 2 + Ee! d2 ! : dx2 Thus, (7) is approximated by " d! E 00 e! + 2E 0 e! + E e! dx d! dx 5 2 # 2 d ! + e! 2 + dx Ee! 0: By (12), we have E 00 e! + 2i 1 2 1 2 E 0 e! + Ee! + d2 ! + dx2 0: By (13), the above expression becomes 1 2 00 E + 2i Now, use E(x) = 1 P n 2 1 2 1 2 i 0 E + 0 1 2 2 E 0: En (x), we get n=0 (E000 + 1 +2i 2 1 +i 2 0: 2 1 2 E100 + 1 2 (E 0 + 0 0 (E 1 0+ 2 n 1 2 E 00 E200 + ::: + n 1 0 1 0 2E + E2 + ::: + 1 1 1 2E + E2 + ::: + 1 1 n 2 En00 + :::) n 1 n 0 2 E 2 E 0 + :::) n n 1+ n 1 n 2 E 2 E + n 1 n + :::) + 1 9 > > > = > > > ; (14) Again, using the fact that 1; we can say that higher powers of will give the most important terms in our approximation. Equate the coe¢ cient of 1=2 to 0, we have 2 1 2 0 E00 + 1 E0 = 0: 2 2 Integrate the above expression with respect to x, then 1=4 E0 = In (14), equate the coe¢ cient of E000 + 2i 0 1 2 to 0, it gives E10 + n Similarly, equate the coe¢ cient of En00 1 + 2i : 1 2 1 2 i 0 2 1 2 E1 = 0: to 0, we obtain En0 + i 0 2 1 2 En = 0: The above expression is equivalent to 2i 1 2 En0 + i 0 1 2 2 En0 + 0 4 En00 En = En = Multiplying both side by the integrating factor d En dx 1=4 = 6 i 2 1 iEn00 1 1=4 1=2 1 2 2 1 4 En00 : , we have 1: Integrating both side with respect to x, Z 1 i 4 En = 2 1 4 Similarly, the above proof is true for ! = completed. En00 1 2 i 1 dx: Rx 1 2 0 (t)dt. Thus, the proof is Note that E (x) is precisely the Liouville-Green approximation if the …rst term in the in…nite series in (11) is used. We remark that a better approximation of (x) can be obtained if more terms in the in…nite series of (11) is used. From the result of Theorem 1, (x) A~ 1=4 i e 1=2 Rx 1=2 0 ds ~ +B 1=4 e i 1=2 Rx 0 1=2 ds : We use sine and cosine functions instead of exponential function, it leads to Z x Z x 1 1 1 1 1 1 4 2 2 4 2 2 ds) cos( ds) + C2 sin( (x) C1 0 0 where C1 and C2 are arbitrary constants. Substitute x = 0 in the above expression and by the boundary condition in (4), (0) = 0, it gives 1 4 C1 =0 which implies C1 = 0. Therefore, (x) C2 1 4 1 2 sin( Z x 1 2 ds): 0 Substitute x = L in the above expression and by another boundary condition in (4), (L) = 0, we have 1 4 C2 sin( Z 1 2 L 1 2 ds) = 0: 0 As C2 6= 0, otherwise (x) 1 2 Z 0, we obtain L 1 2 ds = n for n = 1; 2; 3:::: (15) 0 This expression is equivalent to n = RL n 0 Therefore, n (x) C2n 1 4 1 2 sin( ds 1 2 n !2 Z 0 7 : x (16) 1 2 ds): We choose C2n such that f Z ng forms an orthonormal set, that is, L 0 if n 6= m : 1 if n = m dx = n m 0 Thus, if n = m, then 2 (C2n ) Z L 1 2 Z 1 2 sin2 n 0 Let u = 1=2 R x n 0 1=2 x 1 2 ds dx 1: 0 ds, then du = Z n 2 (C2n ) 1=2 1=2 dx. n 2 n 2 n 1 2 By (15), we have sin2 udu 1: 0 It implies 1 2 (C2n ) Rn sin udu 0 = n 2 1 2 : n From (16), we obtain 2 (C2n ) RL 0 2 1 2 : ds This gives our approximation to the solution of (7) n (x) RL 2 1 2 0 ds ! 12 1 4 sin n Rx 0 RL 0 1 2 ds 1 2 ds ! : (17) 3. An Example In this section, we let L = 1. We want to look at the behavior of the approximated solution to the problem (1)-(2) for some (x) and f (x). Let (x) = 1 + 50x and f (x) = x (1 x). From (4), d2 + (1 + 50x) dx2 = 0; (0) = 0 = (1): Using (16), we have n We note that f 1 n gn=1 75n 513=2 1 = 2 : is an increasing sequence. That is, 1 < 2 < ::: < 8 n < :::: (18) From (17), 1=2 150 n (x) 513=2 (1 + 50x) 1 Let n = 10, then 10 = 513=2 1 sin 75n 513=2 1 Z x 1 (1 + 50s) 2 ds : (19) 0 2 and 1=2 150 10 (x) 750 513=2 1 1 4 (1 + 50x) 1 4 750 513=2 sin 1 Z x 1 (1 + 50s) 2 ds : 0 The graph of this approximated eigenfunction is: y 0.5 0.25 0 0 0.25 0.5 0.75 1 x -0.25 -0.5 In the above graph, there are exactly 9 roots in the interval (0; 1). This matches to the result of the following theorem (cf. Haberman [2, p. 135]). Theorem 3. The eigenfunction n (x) has exactly n 1 roots in the interval 0 < x < L. Substitute (18) and (19) into (5), we get the approximated solution to the problem (1)-(2) u (x; t) 1 X an 150 513=2 1 1=2 (1 + 50x) 1 4 75n 513=2 1 sin n=1 Z x 1 (1 + 50s) 2 ds e 75n 513=2 2 1 0 Rewrite the above expression, it yields u (x; t) 1 X n=1 a ~n (1 + 50x) 1 4 sin 75n 513=2 1 9 Z 0 x 1 (1 + 50s) 2 ds e 75n 513=2 2 1 t t : where a ~n = an When t = 0, 1 X 1=2 150 513=2 1 . To solve for a ~n , we use the initial condition of (2). 1 4 a ~n (1 + 50x) n=1 sin 75n 513=2 1 Z x 1 (1 + 50s) 2 ds u (x; 0) 0 = f (x) = x (1 x) : From (6), a ~n is given by R1 1 (1 + 50x) x (1 x) (1 + 50x) 4 sin 0 a ~n h R1 1 (1 + 50x) (1 + 50x) 4 sin 5175n 3=2 0 75n 513=2 1 Simplify the above expression, we have R1 0 a ~n 3 (1 + 50x) 4 x (1 R1 0 1 2 x) sin 2 (1 + 50x) sin Rx 1 (1 + 50s) 2 ds dx : i2 Rx 1 2 dx (1 + 50s) ds 1 0 0 [(1+50x)3=2 1]n 513=2 1 [(1+50x)3=2 1]n 513=2 1 dx : dx Since the analytic solution of the right-hand side of the above expression cannot be determined, we use numerical method to compute the approximated solution. Adaptive Gaussian Quadrature is used to compute the value of the numerator and denominator, this numerical integration method is built in Maple R 2 version 9.03. We use the …nite sum Z x 100 2 X 75n 1 1 75n t u ~ (x; t) = a ~n (1 + 50x) 4 sin (1 + 50s) 2 ds e 513=2 1 3=2 51 1 0 n=1 to approximate u (x; t). The table below shows that the values of f (x) = x (1 x) and u ~ (x; 0) are close to each other at x = 0; 0:1; 0:2; 0:3; :::; 1: x 0 0:1 0:2 0:3 0:4 0:5 0:6 0:7 0:8 0:9 1 2 Maple R f (x) 0 0:09 0:16 0:21 0:24 0:25 0:24 0:21 0:16 0:09 0 u ~(x; 0) 0 0:08997841586 0:1599906913 0:2100016569 0:2399981029 0:2499991097 0:2400001991 0:2100001255 0:1599993234 0:08999867687 9:377258464 10 12 is a registered trademark of Waterloo Maple Inc., Waterloo, Ontario, Canada 10 The graphs of u ~ (x; t) when t = 0, 0:5, and 2 are: t=0 t = 0:5 t=2 The graphs above show that u ~ (x; t) skews to the right-hand side when t is increasing. For each x 2 [0; 1], u ~ (x; t) is a non-increasing function of t. If (x) = 1 and f (x) = x (1 x), the exact solution to problem (1)-(2) is 1 Z 1 X 2 u (x; t) = 2 x (1 x) sin (n x) dx sin (n x) e (n ) t (20) n=1 0 (cf. Haberman [2, pp. 42-43]). We use the …nite sum 100 Z 1 X u ^ (x; t) = 2 x (1 x) sin (n x) dx sin (n x) e n=1 (n )2 t 0 to approximate the solution in (20). Use Maple R to graph u ^ (x; t) for t = 0; 0:5; and 2. The graphs are: t=0 t = 0:5 11 t=2 The graphs above show that u ^ (x; t) is symmetric with respect to the line x = 0:5 for all t. For each x 2 [0; 1], u ^ (x; t) is a non-increasing function of t. Compare the graphs of u ~ (x; t) and u ^ (x; t), the rate of decrease of u ^ (x; t) is faster than u ~ (x; t). References [1] K. Gustafson, Introduction to Partial Di¤ erential Equations and Hilbert Space Methods, Mineola, NY, Dover, 1999, p. 175. [2] R. Haberman, Elementary Applied Partial Di¤ erential Equations, Englewood Cli¤s, NJ, Prentice-Hall, 1987, pp. 42-43, 135, 142, and 184. [3] R. Nagle, E. Sa¤, and A. Snider, Fundamentals of Di¤ erential Equations and Boundary Value Problems, New York, NY, Addison Wesley, 2004, p. 578. [4] F. Olver, Asymptotics and Special Functions, New York, NY, Academic Press, 1974, pp. 190-191. 12