Numerical Analysis: Final exam
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
Numerical Analysis: Final exam
(50 marks, only the 5 best questions count)
12 May 2022
Question 1 (Floating point arithmetic, 10 marks). True or false? +1/-1
1. Let (•)3 denote base 3 representation. It holds that
(120)3 + (111)3 = (1001)3 .
2. Let (•)2 denote binary representation. It holds that
(1000)2 × (0.101)2 = (101.01)2 .
3. In Julia, Float64(.25) == Float32(.25) evaluates to true.
4. The spacing (in absolute value) between successive double-precision (Float64) floating point numbers is constant.
5. The machine epsilon is the smallest strictly positive number that can be represented in a floating point format.
6. Let F64 ⊂ R denote the set of double-precision floating point numbers. If x ∈ F64 , then x admits a finite decimal representation.
7. Let x be a real number. If x ∈ F64, then 2x ∈ F64 .
8. The following equality holds
(0.101)2 = 7
9. In Julia, 256.0 + 2.0*eps(Float64) == 256.0 evaluates to true.
10. The set F64 of double-precision floating point numbers contains twice as many real numbers as the set F32 of single-precision floating point numbers.
11. Let x and y be two numbers in F64 . The result of the machine addition x y is sometimes exact and sometimes not, depending on the values of x and y .
Question 2 (Iterative method for linear systems, 10 marks). Assume that A ∈ Rn×n is a nonsingular matrix and that b ∈ Rn . We wish to solve the linear system
Ax = b (1)
using an iterative method where each iteration is of the form
Mxk+1 = Nxk + b.
Here A = M − N is a splitting of A such that M is nonsingular, and xk ∈ Rn denotes the k-th iterate of the numerical scheme.
1. (3 marks) Let ek := xk − x∗, where x∗ is the exact solution to (1). Prove that ek+1 = M−1 Nek .
2. (3 marks) Let L = ∥M−1 N∥∞ . Prove that
∀k ∈ N, ∥ek∥∞ ≤ Lk ∥e0 ∥∞ .
3. (1 mark) Is the condition ∥M−1 N∥∞ < 1 necessary for convergence when x0 x∗?
4. (3 marks) Assume that A is strictly row diagonally dominant, in the sense that
n
∀i ∈ {1, . . . ,n}, |aii| > 对 |aij|.
j=1,ji
Show that, in this case, the inequality ∥M−1 N∥∞ < 1 holds for the Jacobi method, i.e. when M contains just the diagonal of A. You may take for granted the following expression for the ∞-norm of a matrix X ∈ Rn×n :
n
∥X∥∞ = 1 对 |xij|.
j=1
5. (Bonus +1) Write down a few iterations of the Jacobi method when
A = ( 0(1)
Is the method convergent?
1(2)) ,
b ( )1(1) ,
x0 = ( )0(0) .
Question 3 (Nonlinear equations, 10 marks). Assume that x* ∈ Rn is a solution to the equation
F(x) = x,
where F : Rn → Rn is a smooth nonlinear function. We consider the following fixed-point iterative method for approximating x*:
xk+1 = F(xk). (2)
1. (8 marks) Assume in this part that F satisfies the local Lipschitz condition
Ax ∈ B6(x*), ∥F(x) − F(x*)∥ ≤ L∥x − x*∥, (3)
with 0 ≤ L < 1 and 6 > 0. Here B6(x*) denotes the open ball of radius 6 centered at x*. Show that the following statements hold:
• (2 marks) There is no fixed point of F in B6(x*) other than x*.
• (2 marks) If x0 ∈ B6(x*), then all the iterates (xk)k∈N belong to B6(x*).
• (3 marks) If x0 ∈ B6(x*), then the sequence (xk)k∈N converges to x* and Ak ∈ N, ∥xk− x*∥ ≤ Lk ∥x0 − x*∥ .
2. (3 marks) Explain with an example how the iterative scheme (2) can be employed for solving a nonlinear equation of the form
f(x) = 0.
3. (Bonus +1) Let JF : Rn → Rn×n denote the Jacobian matrix of F . Show that if
Ax ∈ B6(x*), ∥JF(x)∥ ≤ L,
then the local Lipschitz condition (3) is satisfied.
Question 4 (Error estimate for eigenvalue problem, 10 marks). Let ∥.∥ denote the Eu- clidean norm, and assume that A ∈ Rn×n is symmetric and nonsingular.
1. (5 marks) Describe with words and pseudocode a simple numerical method for calcu- lating the eigenvalue of A of smallest modulus, as well as the corresponding eigenvector.
2. (1 mark) Let M ∈ Rn×n denote a nonsingular symmetric matrix. Prove that
Ax ∈ Rn , ∥Mx∥ ≥ ∥M−1∥−1∥x∥ . (4)
Let λmin(M) denote the eigenvalue of M of smallest modulus. Deduce from (4) that
Ax ∈ Rn , ∥Mx∥ ≥ |λmin(M)|∥x∥ . (5)
3. (4 marks) Assume that ∈ R and ∈ Rn are such that
∥A − ∥ = e > 0, ∥∥ = 1. (6)
Using (5), prove that there exists an eigenvalue λ of A such that
|λ − | ≤ e.
4. (Bonus +1) Show that, in the more general case where A = VDV−1 is diagonalizable but not necessarily Hermitian, equation (6) implies the existence of an eigenvalue λ
of A with
| − λ| ≤ ∥V∥∥V−1∥e.
Hint: Introduce r = A − and rewrite
∥∥ = ∥(A − I)−1r∥ = ∥V(D − I)−1V−1r∥ .
Question 5 (Interpolation error, 10 marks). Let u denote the function
u: [0, 2π] → R;
x '→ cos(x).
Let pn: [0, 2π] → R denote the interpolating polynomial of u through at the nodes
2πi
1. (3 marks) Using a method of your choice, calculate pn for n = 2.
2. (6 marks) Let n ∈ N>0 and en(x) := u(x) − pn(x). Prove that
Ax ∈ [0, 2π], |en(x)| ≤ |ω(x)|
where we introduced
ωn(x) :=u(x − xi).
i=0
Hint: You may find it useful to introduce the function
g(t) = en(t)ωn(x) − en(x)ωn(t).
3. (1 mark) Does the maximum absolute error
En := sup |en(x)|
北 ∈[0,2π]
tend to zero in the limit as n → ∞?
(Bonus +1) Using the Gregory–Newton formula, find a closed expression for the sum
n
S(n) =工 k2 .
k=1
Question 6 (Numerical integration, 10 marks). The third exercise below is independent of the first two.
1. (5 marks) Construct an integration rule of the form
\−11 u(北)d北 ≈ w1u ( − ) + w2u(0) + w3u ( )
with a degree of precision equal to at least 2.
2. (1 mark) What is the degree of precision of the rule constructed?
3. (4 marks) The Gauss–Laguerre quadrature rule with n nodes is an approximation of
the form
\0 ∞ u(北)e−x d北 ≈ 其(n)wiu(北i),
such that the rule is exact when u is a polynomial of degree less than or equal to 2n− 1. Find the Gauss–Laguerre rule with one node (n = 1).
4. (Bonus +1) Find the Gauss–Laguerre quadrature rule with two nodes (n = 2). You may find it useful to first calculate the Laguerre polynomial of degree 2.
2022-12-11