MATH3603 Numerical Methods 2017-2018
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
MATH3603
1. Consider the nonlinear equation f (x) = 0 where f (x) = cosx - x - x2 .
(a) Prove that f has a unique root α in the interval [0, 1].
(b) Estimate the number of iterations required to approximate α to an absolute
tolerance of 10_8 using the bisection method with initial interval [0, 1].
(c) Now consider the fixed point iteration x(k+1) = φ(x(k)) where
φ(x) = (cos x + 2x - x2 ).
Show that the iteration converges to α for all initial guesses x(0) e [0, 1]. Determine a lower bound on the number of iterations required to approximate α to an absolute tolerance of 10_8 using this method.
(d) Write down the Newton method for the solution of f (x) = 0 (give an explicit formula for the iteration). What can you say about the rate of convergence to the root α?
2. Let f : R → R be three times continuously differentiable, and let α e R satisfy f (α) = 0, f\ (α) = 0, f\ (x) 0 for x α, and f\\ (α) 0.
Consider the fixed point method defined by x(k+1) = φ(x(k)), where
f (x)
φ(x) = x - 2
(a) Write down a formula for φ\ (x), valid for x α, and use this to compute the value of φ\ (α) via Taylor approximations.
Apply a suitable result from fixed point theory to prove that, for a sufficiently good initial guess, the method converges to α at least linearly.
(b) Now, assuming convergence of x(k) to α, prove that the method actually con-
verges quadratically, with
(x(k+1) - α) f\\\ (α)
k →o (x(k) - α)2 6f\\ (α) .
Hint: use Taylor approximation to prove that
x(k+1) - α = (x(k) - α)2 + o((x(k) - α)2 ).
You may use without proof the fact that = 1 - e + O(e2 ) as e → 0 .
(c) Assuming the approximation
(x(k+1) - α) f\\\ (α)
(x(k) - α)2 6f\\ (α) ,
derive an estimate for the number of iterations required to approximate the root α = 0 of the function f (x) = x2 + x3 to within a given tolerance 0 < tol < 1, when the above method is applied with the initial guess x(0) = 1.
3. Consider the linear system Ax = b where
A = ╱ µ(1) 4(µ) 、
and µ is a real parameter.
(a) Prove that the basic stationary iterative method does not converge for A.
(b) For which values of µ is A symmetric positive definite? Calculate the optimal
choice of α giving the fastest convergence for the stationary Richardson method x(k+1) = (I - αA)x(k) + b,
along with the associated convergence constant 0 < C < 1 such that |x(k+1) - x|2 ≤ C|x(k) - x|2 .
(c) Without considering eigenvalues, give a sN巩O!au4 Ooup!4!ou on µ for the Ja- cobi method to be convergent. Now, by considering eigenvalues, determine a uaOassv儿h vup sN巩O!au4 condition on µ for the Jacobi method to be convergent.
(d) Prove that when |µ| < 2 the iterates for the Jacobi method satisfy the estimate |x(k+2) - x|2 ≤ C |x\(k) - x|2 ,
for some constant 0 < C\ < 1 that you should specify.
)No4a: TVa (k + 2) sNda儿sO儿!d4 Va儿a !s uo4 v 4hdo.(
4. Consider the method
un+1 = un + hf (tn , un ),
for the solution of the Cauchy problem
n = 1, 2, . . . ,
(*)
(a) Classify the method as either explicit or implicit.
(b) State the definition of truncation error for a general one-step method. Show
that the method (*) is of first order as h → 0. State clearly the smoothness assumptions under which your analysis is valid.
(c) Given δ > 0, let vn satisfy the perturbed system
Assume that
-Λ ≤ (t, y) ≤ -λ < 0, for some 0 < λ ≤ Λ .
Show that if h < 2/Λ is fixed then the method (*) is stable to perturbations, in the sense that vn - un is bounded as n → o.
(d) Write down a sufficient condition on h under which the method (*) is stable to perturbations for the problem
(You may use without proof the fact that (arctan(z)) = 1/(1 + z2 ) .)
5. Consider the second order boundary value problem
y(0) = y0 , y(1) = y1 ,
where q , r and f are continuous functions on [0, 1].
Given an integer N > 0, define the finite difference method
y(n) + qn + rnun = fn , n = 1, . . . , N - 1,
where h = 1/N , xn = nh, qn = q(xn ), rn = r(xn ), fn = f (xn ) and un is the approximation of y(xn ), for each n = 0, . . . , N .
(a) Write the method as a linear system of the form Au = f, specifying the system
matrix A and the vectors u and f .
(b) Without computing eigenvalues, show that if r(x) > 0 and h ≤ 2/ maxxo[0,1] |q(x)|
then the Jacobi and Gauss-Seidel methods both converge for this system.
(c) Prove that the truncation error Tn := LN (y(xn )) - fn satisfies |Tn | ≤ Ch2 ,
for some constant C > 0 that you should specify.
(You may assume that y(x) is four times continuously differentiable on [0, 1] .)
(d) Briefly explain the idea of the shooting method for solving the original boundary value problem.
2023-05-28