FINAL Problem 1 To demonstrate L-stability concept we consider x0 = λ(x − cos(t)) − sin(t), λ = −106 , x(0) = x0 . The exact solution of this equation is given by x(t) = eλt (x0 − 1) + cos(t). Use backward Euler and trapezoidal rule (2nd order Adams-Moulton, xn+1 − xn = 0.5h(fn+1 + fn )) to solve numerically this equation with x0 = 1 and x0 = 1.5. In all cases use h = 0.1, N = 400. Compare the solutions with the exact one (by plotting). Note that both trapezoidal and backward Euler methods are A-stable, while only Backward Euler method is L-stable Problem 2 Consider the multi-step method (2nd order BDF method) 3xn+2 − 4xn+1 + xn = 2hfn+2 . (1) a) Find the region of absolute stability of the method (note that the method is A-stable). b) Show that the method is L-stable. c) Note that explicit methods can not be L-stable. To show this consider an “explicit version” of (1) 3xn+2 − 4xn+1 + xn = 2hfn+1 . Though for this explicit method q(z) = z satisfies root condition show that the method is not L-stable. d) (bonus) Why explicit multi-step methods can not be L-stable. Problem 3 Consider A-stability of one-step methods (e.g., Runge-Kutta methods). a) Write down in terms of ω (do not plot) A-stability for 2nd and 4th order Runge-Kutta methods. (note you only need to write down A-stability regions as W (ω) = 0). b) How can you adopt the idea of boundary locus method to plot A-stability region for 2nd order Runge-Kutta method? Plot the A-stability region. Problem 4 Linear multi-step methods can be considered as one step methods. Assume ak = 1, bk = 0 (without loss of generality) and x0 = λx. Show that the multi-step method ak xn + ...a0 xn−k = h(bk fn + ... + b0 fn−k ) (2) un = Aun−1 , (3) can be written as where un = (xn−k+1 , ..., xn )T with some matrix A. Find A. Show that A-stability of (3) is the same as the A-stability we have derived for (2). Note. If you do not want to work for arbitrary k you can assume k = 4. 1 Problem 5 Given xn + αxn−1 + (−1 − α)xn−2 = h(βfn + (2 + α − β)fn−1 ). a) Determine the relationship between α, β such that the method is consistent, convergent with second order accuracy. Find all possible values of α for which the method is (zero) stable. b) Find a value for α (or β) such that the method is absolutely stable. c) Is the method with calculated values of α and β L-stable? Problem 6 Consider a sequence of polynomials φn (x) (degree n) that is w-orthogonal to q(x) of lesser degree on (a, b), i.e., Z b w(x)φn (x)q(x)dx = 0. a a) Show that φn (x) can be written as φn (x) = 1 dn Un (x), w(x) dxn where Un (x) is a solution of dn+1 1 dn Un (x) = 0, dxn+1 w(x) dxn (4) and Un (x) and its derivatives up to n − 1 order vanishes at x = a and x = b. Note that the solution of (4) exists and unique up to a constant multiplier. b) Consider quadrature formula Z ∞ −x e f (x)dx ≈ 0 n X Ai f (xi ). (5) i=1 Show that if w(x) = e−x , then Un (x) = e−x xn satisfy (4). Using the latter write down formula for w-orthogonal polynomials φn (x) for n = 3. 2