C3 THEORY OF COMPUTATIONAL DYNAMICS EXAMPLE SHEET 2 Stars indicate level of difficulty. 1. Obtain an explicit error estimates for the modified Euler method (Q17, Ex. Sheet 1) and the 2nd order Runge-Kutta scheme given in lectures: i.e. compute constants K1 and K2 such that 1 ≤ | xt+h - xt - hF(xt + 2hF(xt)) | K1h3 1 ≤ | xt+h - xt - 2h[F(xt) + F(xt + hF(xt))] | 2. K2h3 Consider the numerical scheme obtained by taking two steps of the 2 nd order Runge-Kutta 1 scheme, each with step size 2h. Write down an expression for xt+h - xt, and hence compare the accuracy of this method to a 4t h order Runge-Kutta of step size h. How many function evaluations per time step of size h does each of these schemes require. 1 1 3***. The 4th order Runge-Kutta scheme is given by A = F(xt), B = F(xt + 2hA), C = F(xt + 2hB), D = F(xt + hC). Verify that (assuming I haven’t made any mistakes!) 1 1 1 3 48 1 1 1 B = F + 2hFF' + 8h2F 2 F'' + C = F + 2hFF' + 8h2F 2 F'' + 4h2F(F')2 + D = F + hFF' + 2h2F 2 F'' + 2h2F(F')2 + 8h3F 2 F'F'' + 6h3F 3 F''' + 1 h F 3 F''' + O(h4) 1 3 3 16 h F 2 F'F'' + 5 1 3 48 h F 3 F''' + O(h4) 1 1 3 4 h F(F')3 + O(h4) 1 Hence, using Q15. Ex. Sheet 1, verify that xt+h = xt + 6h(A + 2B + 2C +D) + O(h5). I suggest you do B by hand, and use Mathematica for the remaining expressions. 4. C3 May 1994, Q1. 5. C3 May 1995, Q1. 6. Suppose x(t)∈Rn satisfies the linear differential equation dx(t ) dt = A.x(t) where A is an n×n matrix. Suppose thatx(0) is an eigenvector of A, i.e. A.x(0) = λx(0) for some λ ∈R. Show that x(t) = e λtx(0), and hence deduce that the eigenspaces of A are invariant under the flow generated by this equation. Show that the Euler, modified Euler (Q17, Ex. Sheet 1) and 2nd and 4t h order Runge-Kutta schemes all preserve the eigenspaces. For x(t) an eigenvector, compare the exact solution of the equation with that obtained from these schemes. C3 Exercise Sheet 2 7. 2 This question investigates the convergence of the secant method. Let x* be the root of F : R → R and suppose that {xi} is a sequence generated by the secant scheme, so that xi+1 = x i−1 − x i F (x i−1 ) − F (x i ) xi - F(xi) a) Let εi = xi - x* and show that εi+1 εi - F(xi) = x i−1 − x i F (x i−1 ) − F (x i ) b) Hence show that εi+1 F (x i ) − F (x *) x i−1 − x i εi 1 − xi − x * F (x i−1 ) − F (x i ) = c) Thus deduce that εi+1 εi εi-1 = d) Show that F (x i ) − F (x *) F (x i−1 ) − F (x i ) − x −x* x i−1 − x i x i−1 − x i i x * − x i−1 F (x i−1 ) − F (x i ) F (x i−1 ) − F (x i ) = F'(x*) + O(εi) + O(εi-1 ) and x i−1 − x i F (x i ) − F (x *) F (x i−1 ) − F (x i ) − x −x* x i−1 − x i i x * − x i−1 = 1 2 F''(x*) + O(εi) + O(εi-1 ) e) Conclude that if εi and εi-1 are sufficiently small, then there exists a constant K > 0 such that |εi+1 | K |εi| |εi-1 | ≤ f) Let ei = K |εi|. Show that e i+1 ≤ e i ei-1 . Using induction, or otherwise, show ≤ ei e 0p e 1p i-1 i where pi is the ith Fibonacci number, given by pi+1 = pi + pi-1 , with p0 = 0, p1 = 1. g) Let e = max {e 0,e 1}. Show that e i ≤ e p i+1 h) Write down a 2×2 matrix such that pi p i+1 p A i−1 pi = By finding the eigenvectors and eigenvalues, or otherwise, show that pi ~ γi-1 , where γ is the golden mean γ = 1 2 (√5+1) j) Hence deduce that if e is sufficiently small, there exists a constant c > 0 such that ei ≤ c eγ i