Page 1 of 2 C3 Theory of Computational Dynamics MSc Examination, May 1994 2 hours. Full marks may be obtained by correct answers to FOUR questions. 1. Suppose xt∈Rn satisfies the differential equation dx t dt = F(xt) Write out the 2nd order Taylor expansion of x(t+h). Use this to derive the Euler method for integrating the equation and to estimate the error over one time step for the Runge-Kutta scheme given by xt+h ≈ xt + 1 2h (F(x )+ F (x +hF(x ))) t t t Compare these two methods with the exact solution for the one dimensional equation dxt dt = axt + b where a,b∈R are two constants. 2. A forced pendulum is allowed to repeatedly collide with a fixed wall at θ = a. If the collisions are perfectly elastic the motion may be described by θ˙˙ + k θ̇ + sin θ θ̇ = → cosωt − θ̇ for θ < a at θ = a Describe an algorithm to numerically compute trajectories of this system (hint: recall Hénon’s method for computing a Poincaré map).Where would you expect problems with your method to occur? 3. i) Answer either i) or ii) Describe and justify the Newton method for finding roots of a function g : R n → R n . For the case n = 1 show that generically the method converges quadratically. Explain how it can be used to find periodic orbits of a map f : Rn → R n (make sure that you indicate how the various quantities required to implement the Newton method are calculated in this case). ii) Sketch how you would find a periodic orbit of the Lorenz equations ẋ ẏ ż = = σ(y - x) rx - y - xz = xy - bz You need not give any derivations, but should give sufficient detail to allow a competent programmer to implement the algorithm from your description. Turn over Page 2 of 2 4. i) Answer either i) or ii) Describe a method of computing the LR decomposition of a matrix and briefly sketch how this can be used to invert the matrix. Suppose a matrix has the block structure A11 0 0 0 A12 0 A22 A23 0 A33 0 0 0 0 A34 A44 where each Aij is a 100×100 matrix. Indicate how you would numerically invert such a matrix. ii) Find an LR decomposition of the matrix 1 1 1 1 1 1 2 1 3 2 2 3 a b c d and compute its determinant. Determine for which choices of a, b, c, d and e the equation 1 1 1 1 1 1 2 1 3 2 2 3 a b c d x y z w = 1 1 1 e has a solution, and find this solution when it exists. 5. Suppose that the n×n matrix A has a real eigenvalue of maximum modulus of multiplicity 2, i.e. if the eigenvalues of A are λ1, …, λn, then λ1 = λ2 and |λi| < | λ1| for i = 3, …, n. How might one compute λ1 numerically? You may assume that A is diagonalizable. Now suppose that λ1 and λ2 are a complex conjugate pair, so that λ1 = re iθ and λ2 = re -iθ . What will happen when you try to use your method? Briefly suggest how you might modify your algorithm to compute λ1 and λ2 in this case. 6. Let dx t dt = F(xt,µ) xt∈Rn, µ∈R be a one parameter familiy of differential equations. Explain how you would find a parameter value µ0 where this family undergoes a saddle node of equilibrium points. End of paper