421a it may still use less computer time when

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: a) with the trapezoidalrule variant of the weighted average scheme (4.4.20b). Assuming that Dirichlet data is prescribed at x = 0 and 1, let's calculate the Jacobian of this system. Thus, di erentiating Fj with respect to Uin+1 . 8 f i=j m t n+1 m;1 @Fj = < ;+ 2mx2 (Uti (U ) +1 )m;1 iif i = j ; 1 j + 1 : n 1 (4.4.22a) x2 i @Uin+1 : 0 otherwise In matrix form FUn+1 (Vk ) = 3 2 2(U1n+1 )m;1 ;(U2n+1 )m;1 7 t 6 ;(U1n+1 )m;1 2(U2n+1 )m;1 ;(U3n+1 )m;1 7 6 : I + m x2 6 7 ... 5 4 +1 +1 ;(UJn;2 )m;1 2(UJn;1 )m;1 (Un+1 =Vk ) (4.4.22b) 44 Parabolic PDEs Thus, to solve this problem using Newton's method, we would use (4.4.21a) with F(Vk ) given by (4.4.20b - 4.4.20d) and the Jacobian given by (4.4.22b). The Jacobian is tridiagonal thus, one tridiagonal solution is required per iterative step. Richtmyer and Morton 17], p. 201) solve an initial-boundary value problem for (4.4.20a) with m = 5 when the exact solution is a \running wave" u(x t) = (v(vt ; x + x0 )) where v and x0 are prescribed constants. The function is implicitly determined as the solution u of 5 (u ; u )4 + 20 u (u ; u )3 + 15u2(u + u )2 + 20u3(u ; u )+ 0 0 0 0 0 0 4 30 5u4 ln(u ; u0) = v(vt ; x + x0) 0 where u0 is another prescribed constant. Computationally, Richtmyer and Morton 17] solve an initial-boundary value problem with data that is consistent with the exact running wave solution. Thus, the initial and boundary conditions for (4.4.20b) are Uj0 = (v(x0 ; xj )) U0n = (v(vtn + x0)) j = 0 1 ::: J n UJ = (v(vtn ; 1 + x0 )) n > 0: Richtmyer and Morton 17] solve a problem with = 0:4, v t= x = 0:75, and 5u4 t= x2 = 0:005. They only perform one Newton iteration per time step thus, they 0 take V1 = Un+1 . (This not good practice. The convergence of Newton's method should be checked before proceeding to the next time step.) Their results for the numerical and exact solutions as functions of x (j ) are shown for several times t (time steps n) in Figure 4.4.2. The weighted average scheme (4.4.20b) is expected to have a stability restriction with = 0:4. Thus, there is no logical reason for this choice relative to, e.g., the CrankNicolson scheme with = 0:5 Richtmyer and Morton 17] choose = 0:4 to show that a constant-coe cient stability analysis su ces for this example since its solution is smooth. To do this, they write (4.4.20a) ut = (mum;1 ux)x: 4.4. Variable-Coe cient and Nonlinear Problems 45 Figure 4.4.2: Richtmyer and Morton's 17] solution of Example 4.4.6. The solid curves are the exact solution and the dots show the numerical solution obtained by the weighted average scheme (4.4.20b) with m = 5 and = 0:4. In this form, it has the appearance of a heat conduction equation ut = ( ux)x with the di usion coe cient = mum;1: Based on the constant coe cient analysis of Section 4.1, we would expect this weighted average scheme to be absolutely stable when (4.1.20) is satis ed, i.e., when t 1 2 2(1 ; 2 ) : x With m = 5, we have = 5u4 thus, the stability condition is 1: 5u4 t 2 x 2(1 ; 2 ) Using the indicated parameter values 1 u4 0:005 u 2(1 ; 2(0:4)) = 2:5 0 or u 4:73: u 0 46 Parabolic PDEs The results of Figure 4.4.2 indicate that oscillations and a loss of absolute stability set in shortly after the solution at x = 0 exceeds the above value. Problems 1. Use (4.4.2b - 4.4.2d) with (4.4.7) to show that the forward time-centered space scheme (4.4.2a) is consistent. 2. Consider the initial-boundary value problem for Burgers' equation ut + uux = uxx u(x 0) = (x) 0<x<1 u(0 t) = gL(t) t>0 u(1 t) = gR(t): where is a positive constant. Burgers proposed this equation to model certain aspects of turbulent ows. It is also used as a proving ground for numerical methods because when is small the solution exhibits a shock like structure that is di cult to calculate. 2.1. Consider the two possibilities for discretizing the nonlinear term: n n n U n Uj +1 ; Uj ;1 (uux)j j 2x and (u2)x]n (Ujn+1)2 ; (Ujn;1)2 j : 2 4x Both formulas are O( x2) approximations of uux. Rather than decide which one is best, let's use the linear combination Q(Ujn) (uux) 2 x where (uux)n = j (Ujn+1)2 ; (Ujn;1)2 + (1 ; p)Ujn (Ujn+1 ; Ujn;1) 2 with p selected on 0 1]. Suppose that we discretize Burgers' equation using the Crank-Nicolson scheme Q(Ujn ) = p Ujn+1 ; Ujn Q((Ujn+1 + Ujn )=2) = t+ 2x 2 ((U n+1 + U n )=2) j j 2 x 4.4. Variable-Coe cient and Nonlinear Problems 47 where, as usual, 2 denotes the second-centered di erence. Develop a procedure for solving the Crank-Nicolson system using the tridiagonal algorithm and Newton's iteration. Give formulas for evaluating any Jacobians. 2.2. Write a program to solve the Crank-Nicolson system developed in Part 1. Choose the initial and Dirichlet boundary data so that the exact solution is u(x t) = 1 ; tanh x2; t : Solve the problem with = 0:01 for t 1 using p = 0, 2/3, and 1. Select x = 0:05, 0.02, and 0.01 and compute with t= x = 0:5 and 1. Plot the solutions as functions of x at t = 0, 0.5, and 1. Tabulate or plot the error in the discrete L2 norm at t = 1 as a functi...
View Full Document

Ask a homework question - tutors are online