Problem 1. What is the numerical value of the composite trapezoid rule applied to the function f (x) = x−1 using the points 1, 43 and 2. Solution: The basic trapezoid rule is b−a (f (a) + f (b)). (1) T (f, a, b) = 2 To find the numerical value we apply this basic rule to two subintervals [1, 4/3] and [4/3,2]. Then we sum the results: 1 T (f, 1, 4/3) = [1 + 3/4] 6 1 T (f, 4/3, 2) = [3/4 + 1/2] 3 The result then is: Z 2 1 87 1 . x−1 dx ≈ T (f, 1, 4/3) + T (f, 4/3, 2) = [1 + 3/4] + [3/4 + 1/2] = 6 3 24 1 R1 Problem 2 Compute approximate value of 0 (x2 +1)−1 dx by using the composite trapezoid rule with three points. Then compare the actual value of the integral. Next determine the error formula and numerically verify an upper bound on it. Z Solution: As in the previous problem, we have that 1 (x2 + 1)−1 dx ≈ T (f, 0, 1/2) + T (f, 1/2, 1) = 0 The error formula is Z b f (x)dx − T (f, h) = E= a 1 31 1 (1 + 4/5) + (4/5 + 1/2) = . 4 4 40 b − a 2 00 h f (ξ). 12 In our case we have a = 0, b = 1, h = 1/2, f 00 (ξ) = 2 ξ2 − 1 . (ξ 2 + 1)2 A bound on the error then is 2 x −1 1 ≤ 1 . max 2 |E| ≤ 48 x∈[0,1] (x2 + 1)2 24 Problem 3.(Continuation) Having computed R(1, 0) from the previous problem, compute R(2, 0) by using the formula: n−1 2X 1 f (a + (2k − 1)h), R(n, 0) = R(n − 1, 0) + h 2 k=1 where h = 2 −n (b−a). Solution: Just apply the formula (h = 1/4, because n = 2) and use the result from the previous problem. 31 1 + (f (1/4) + f (3/4)) 80 4 31 1 + (16/17 + 16/25) = 0.782794117647059 80 4 R(2, 0) = = 1 2 composite trapezoid rule is used to compute the integral R 2 Problem 4. If the −2 sin xdx with h = 10 give a realistic bound on the error −1 Solution: This is straightforward application of the formula for the error: Z b b − a 2 00 h f (ξ). f (x)dx − T (f, h) = (2) E= 12 a The bound of the second derivative is 1. Therefore the final result will be: 10−4 3 2 h ×1= |E| ≤ 12 4 Problem 5. By the Romberg algorithm, approximate Z 2 4 dx 1 + x2 0 by evaluating R(1, 1). 4 24 1 = 4.8 = R(0, 0) = (2 − 0) 4 + 2 1 + 22 5 According to (1) page 211, Solution: 1 a+b 1 2 1 4 2 1 R(0, 0) + (b − a)f ( ) = × 5 + (2) = 5 = 4.4 2 2 2 2 4 2 1+1 2 Using the extrapolation formula (2) page 211, we have R(1, 0) = 1 4 2 1 2 6 4 R(1, 0) − R(0, 0) = × 5 − × 5 = 15 = 4.267 3 3 3 2 3 4 4 Problem 6. Show that the system of equations R(1, 1) = x1 + 4x2 + αx3 = 6 2x1 − x2 + 2αx3 αx1 + 3x2 + x3 = = 3 5, possesses a unique solution when α = 0, no solution when α = −1, and infinitely many solutions when α = 1. Also investigate the homogeneous case when the right-hand side is replaced by 00 s. Solution: After one step of Naive Gauss elimination we obtain the following equivalent system: x1 + 4x2 + αx3 0x1 − 9x2 + 0x3 = = 6 −9 0x1 + (3 − 4α)x2 + (1 − α2 )x3 = 5 − 6α From the second equation we have x2 = 1 and putting x2 = 1 in the first and the third equation we obtain x1 + αx3 = 2 (1 − α )x3 = 2(1 − α) 2 Now, for α = 0 : x1 = 2, x2 = 1, x3 = 2. 3 for α = −1 : no solutions (0x3 = 4). for α = 1 : infinite number of solutions (0x3 = 0). The homogeneous case: From the second equations we get x2 = 0 and putting x2 = 0 in the first and the third equation we obtain x1 + αx3 (1 − α2 )x3 = = 0 0 for α = 0 : x1 = x2 = x3 = 0 for α = −1 :infinite number of solutions (0x3 = 0 for α = 1 : infinite number of solutions (0x3 = 0). Problem 7. A Gaussian quadrature rule for the interval [−1, 1] can be used on an interval [a, b] by applying a suitable linear transformation. Approximate: Z 2 2 e−x dx 0 using the two-point Gaussian rule on [−1, 1] (i.e. n = 1) as follows: Z 1 1 1 f (x)dx ≈ f (− √ ) + f ( √ ) 3 3 −1 Solution: This is a straightforward application of the Gaussian rule after changing the variable from [0, 2] to [−1, 1]. The change of variable is as follows: x = t + 1, note that dt = dx. Then the integral takes the form: Z 1 Z 2 2 2 1 1 e−x dx = e−(t+1) dt ≈ exp(−(1 − √ )2 ) + exp(−(1 + √ )2 ). (3) 3 3 0 −1 Problem 8. Construct a rule of the form Z 1 1 1 f (x)dx ≈ αf (− ) + βf (0) + γf ( ), 2 2 −1 that is exact for all polynomials of degree 2. Solution: We need only to make the rule exact for the basic polynomials 1, x and x2 . Then it will be exact for all polynomials of degree ≤ 2 because every such polynomial is represented as a sum of basic polynomials multiplied by constants (i.e. every polynomial of degree ≤ 2 is a linear combination of the basic polynomials). This means that for these three fixed particular polynomials (1, x and x2 ) we need to determine α β and γ, such that ≈ becomes = in these three cases. This gives us the following system of equations: Z 1 1dx = α+β+γ xdx − −1 Z 1 Z = −1 1 x2 dx = −1 γ α + 0β + 2 2 γ α + 0β + 4 4 Computing the integrals on the left hand side gives the following system of equations for α, β and γ: 4 α+β+γ = 2 α γ = 0 − + 2 2 2 α γ + = . 4 4 3 Solving this system gives: α = γ = 4/3, β = −2/3. Problem 9 How many storage locations are needed for a system of n linear equations if the coefficient matrix has banded structure in which aij = 0 for |i−j| ≥ k + 1? Solution: For k = 1 the matrix is a tridiagonal matrix. For k=2 the matrix is a pentadiagonal matrix. For a general k a matrix with the above property is a banded matrix with k super diagonals and k subdiagonals. Thus, the number of storage locations is: n + 2(n − 1) + 2(n − 2) + · · · 2(n − k) = n + 2kn − 2(1 + 2 + · · · + k) = k(k + 1) = n + 2kn − k(k + 1). 2 Problem 10 Using naive Gaussian elimination, factor the matrix A in the form A = LU , where L is a unit lower triangular matrix and U is an upper triangular matrix. (a) 3 0 3 1 0 0 3 0 3 1 0 , U = 0 −1 3 A = 0 −1 3 Answer: L = 0 1 3 0 1/3 −3 1 0 0 8 n + 2kn − 2 (b) 1 0 A= 3 0 0 1/3 0 1 0 0 1 1 3 −1 Answer: L = 3 −3 −3 0 6 2 4 −6 0 2 0 0 1 −1/4 0 1 0 0 , U = 0 0 1 0 0 1/3 0 1 3 −1 0 8 3 0 0 −13/4