Practice Exercises for the Final Exam, Math 104B

1. True (T) or False (F). Suppose A is an n n positive denite matrix: a) ( ) The search directions for the conjugate gradient (CG) method are always the residuals. b) ( ) The CG method will converge to the exact solution of Ax = b in at most n2 iterations. c) ( ) Two vectors u and v are said to be conjugate with respect to A if and only if uT Av > 0. d) ( ) The most expensive part in a CG iteration is computing the product of A and a vector. e) ( ) For A sparse, the CG method generally beats Jacobi, GS, and S.O.R. 2. The concentration c of a radioactive material decays according to the law c(t) = beat where t represents time in seconds, a = 0.1 sec1 , and b is the initial concentration. a)Using the Least Squares method and the data table (Table 1) below nd b. b) Find the error in the least squares approximation. ti (sec) 1 2 3 4 Ci 0.91 0.80 0.76 0.65

Table 1: 3. Given a collection of data points {(xi , yi )}m i=1 nd the best least squares approx2 3 imation of the form y = ax + bx . 4. (a) Given a collection of data points {(xi , yi )}m i=1 nd the best least squares 2 approximation of the form y = ax + bx . (b) Use this approximation to t the data in Table 2. (c) Find the error in the least squares approximation. 5. a) Obtain the rst 4 Legendre polynomials. b) Find the least squares polynomial approximations of degrees 1, 2, and 3 for the function f (x) = ex on [1, 1]. c) What is the polynomial least squares approximation of degree 4 for f (x) = x3 on [1, 1]? Explain

xi 1 2 3 4

yi 3.1 9.8 21.2 36.1

Table 2: 6. Find the general continuous least squares trigonometric polynomial Sn (x) for square pulse: -1 if x 0, f (x) = 1 if 0 < x . 7. a) Determine the interpolating trigonometric polynomial S4 (x) of degree 4 for f (x) = esin(x) on [, ]. b) Find maxi |S4 (xi ) f (xi )|. 8. Circle the right answer(s) (no partial credit). The FFT is based on: a) the interpolation accuracy of the least squares polynomials. b) grouping even and odd terms. c) the identity (ei2/N )N = 1. d) none of the above. 9. Given a periodic array [y0 , y1 , . . . , yN 1 ], where the values correspond to the equally spaced points xj = j 2/N for j = 0, 1, . . . , N 1, the Discrete Fourier Transform (DFT) can be dened as
N 1

y k =
j =0

yj e2jk/N ,

for k = 0, 1, . . . , N 1

(1)

(a) Prove that the corresponding Inverse Discrete Fourier Transform (or IDFT) is given by 1 yj = N
N 1

y k e2jk/N ,
k=0

for j = 0, 1, . . . , N 1.

(2)

(b) What is the cost of evaluating directly the DFT or the IDFT? What is the cost using the Fast Fourier Transform (FFT)? (c) Illustrate the main ideas behind the FFT to compute the DFT for N = 4 in 7 operations instead of the 16 operations of the direct approach.

3 (the monic Chebyshev polynomial of degree 3) to con10. (a) Use the zeros of T struct and interpolating polynomial of degree 2 for f (x) = ex on [1, 1]. (b) Find a bound for the maximum error in the approximation of f by this interpolating polynomial. 11. Show that any polynomial P (x) of degree less or equal to n can be written on [1, 1] as a linear combination of Chebyshev polynomials:
n

P (x) =
k=0

ak Tk (x)

x [1, 1],

(3)

for some constants a0 , a1 , . . . , an . 12. Prove that 2

1 1

[T (x)]2 n dx = 1 1 x2

(4)

13. Use Gerschgorin Theorem to determine bounds for the eigenvalues and the spectral radius of the following matrix: 4 0 1 3 0 4 2 1 . A= 1 2 2 0 3 1 0 4 14. Find the rst 3 iterations obtained by the Power Method applied to the matrix 4 2 1 A = 0 3 2 , 1 1 4 using x(0) = [1 2 1]T . 15. Determine a shift that can be used in the Power Method to compute 1 when the eigenvalues of A satisfy: 1 = 2 > |3 | . . . |n |. 16. Explain why Googles PageRank algorithm is a huge eigenvalue problem.