Math 317: Linear Algebra Homework 9 Solutions The following problems are for additional practice and are not to be turned in: (All problems come from Linear Algebra: A Geometric Approach, 2nd Edition by ShifrinAdams.) Exercises: Section 4.2: 1–6, 8, 9 Turn in the following problems. 1. Section 4.2, Problem 2d We apply the Gram-Schmidt procedure to obtain an orthronormal basis {q1 , q2 , q3 } or the subspace spanned by the vectors: v1 = (−1, 2, 0, 2), v2 = (2, −4, 1, −4), v3 = (−1, 3, 1, 1). To begin we let, −1 2 0 2 −1 −1/3 1 2 v1 = 2/3 . = =p q1 = kv1 k 3 0 0 (−1)2 + 22 + 02 + 22 2 2/3 q2 is then given by (noting that kq1 k2 = 1 and so there is no denominator): 2 −1/3 0 −4 2/3 0 − (−6) 1 0 1 0 0 −4 2/3 0 v2 − (q1 · v2 ) q1 = √ = = q2 = 1 . 2 2 + 02 + 1 2 + 02 kv2 − (q1 · v2 ) q1 k −1/3 0 −4 2/3 0 − (−6) 1 0 −4 2/3 q3 is given by 1 Math 317: Linear Algebra Homework 9 Solutions v3 − (q1 · v3 ) q1 − (q2 · v3 ) q2 kv3 − (q1 · v3 ) q1 − (q2 · v3 ) q2 k −1 −1/3 0 3 2/3 0 − (3) 1 0 − (1) 1 1 2/3 0 = −1 −1/3 0 3 2/3 0 − (3) 1 0 − (1) 1 1 2/3 0 0 1 0 0 1/2 −1 . = p = 02 + 12 + 02 + (−1)2 0 −1/2 q3 = 2. Section 4.2, Problem 6 (Hint: Find a matrix A so that V = R(A). Finding a basis for the orthogonal complement of V should now be straightforward.) (a) Let V = span (v1 = (1, 0, 1, 1), v2 = (0, 1, 0, 1)). To find an orthogonal basis for V , {w1 , w2 } we apply the Gram-Schmidt procedure letting, w1 1 0 = v1 = 1 1 w2 0 1 −1/3 1 1 0 1 (w1 · v2 ) w1 = = v2 − 0 − 3 1 = −1/3 2 kw1 k 1 1 2/3 1 0 1 1 (b) To find a basis for V , we let V = R(A) so that A = . −1/3 1 −1/3 2/3 Then V ⊥ = (R(A))⊥ = N (A). So to find a basis for V ⊥ is equivalent to finding a basis for N (A). To this extent, we solve Ax = 0 to obtain ⊥ x1 + x3 + x4 = 0 −(1/3)x1 + x2 − (1/3)x3 + (2/3)x4 = 0 =⇒ −x1 + 3x2 − x3 + 2x4 = 0 2 Math 317: Linear Algebra Homework 9 Solutions Since the rank of V is 2, there are two free variables (x3 , x4 ) and so we obtain x1 = −x3 − x4 and 3x2 = x1 + x3 − 2x4 = (−x3 − x4 )+ x3− 2x4 =−3x4 =⇒ −1 −1 0 −1 x2 = −x4 . Thus a basis for N (A) is given by y1 = 1 , y2 = 0 . 0 1 (c) To find an v ∈ V and a w ∈ V ⊥ such that x = v + w for any x ∈ R3 , we recall that x = projV (x) + projV ⊥ (x) where projV (x) ∈ V and projV ⊥ (x) ∈ V ⊥ . From (a), we have an orthogonal basis for V and so projV (x) = projw1 (x) + projw2 (x), where {w1 , w2 } is an orthogonal basis for V . So, we have v = projV (x) = projw1 (x) + projw2 (x) w2 ·x w1 ·x = kw 2 w1 + kw k2 w2 1k 2 1 1 ,x2 ,x3 ,x4 ) 0 = (1,0,1,1)·(x 12 +02 +12 +12 1 + 1 −1/3 (−1/3,1,−1/3,2/3)·(x1 ,x2 ,x3 ,x4 ) 1 2 2 2 2 (−1/3) +1 +(−1/3) +(2/3) −1/3 2/3 (6/15)x1 − (3/15)x2 + (6/15)x3 + (3/15)x4 (−3/15)x1 + (9/15)x2 − (3/15)x3 + (6/15)x4 = (6/15)x1 − (3/15)x2 + (6/15)x3 + (3/15)x4 (3/15)x1 + (6/15)x2 + (3/15)x3 + (9/15)x4 w = projV ⊥ (x) = x − v where v is given above. 3. Section 4.2, Problem 10 Proof : Suppose that v1 , . . . , vk ∈ Rn forms an orthronormal set and that kxk2 = (x · v1 )2 + . . . + (x · vk )2 . We prove that k = n and hence the set forms an orthronormal basis for Rn . Since v1 , . . . , vk ∈ Rn forms an orthronormal set, it is linearly independent. Thus we know k cannot be greater than n, since we have proven (in a previous 3 Math 317: Linear Algebra Homework 9 Solutions homework assignment) that any set in Rn containing more than n vectors is linearly dependent. If k < n, then V 6= Rn and so since Rn = V + V ⊥ (by the fundamental theorem of linear algebra) there is a nonzero vector x ∈ V ⊥ . Recalling that x ∈ V ⊥ =⇒ x · vi = 0 for each vi ∈ V , we have that kxk2 = (x · v1 )2 + . . . + (x · vk )2 = 0 =⇒ kxk2 = 0 =⇒ x = 0, which contradicts the assumption that x was a nonzero vector. Thus k = n and since we have a linearly independent set of n vectors in Rn , we know that this set will form an (orthronormal) basis for Rn . 4. Section 4.2, Problem 11 (a) Proof : To prove that A−1 = AT , we verify that AT A = I where I denotes the identity matrix. To this extent, we note that [AT A]ij = aTi aj where ai denotes the ith column of A and aTi is the ith row of AT . Noting that since the columns of A form an orthronormal set, we have that aTi aj = ai · aj = 0 when i 6= j and aTi aj = ai · aj = 1 if i = j. So [AT A]ii = 1 and [AT A]ij = 0 when i 6= j and so AT A = I =⇒ A−1 = AT . (b) Using part (a) above, we note that aTi aj = ai · aj = 0 when i 6= j and aTi aj = ai · aj = kai k2 since the columns of A form an orthogonal set. So AT A is the diagonal matrix whose ith diagonal term is given by kai k2 . That is AT A = diag {ka1 k2 , . . . , kan k2 }. Thus, if we multiply both sides by diag {1/ka1 k2 , . . . , 1/kan k2 }, we obtain diag {1/ka1 k2 , . . . , 1/kan k2 } AT A = diag {1/ka1 k2 , . . . , 1/kan k2 } diag {ka1 k2 , . . . , kan k2 } = I, so A−1 = diag {1/ka1 k2 , . . . , 1/kan k2 } AT . 1 1 5. Let A = 1 0 0 −1 1 1 2 1 and b = . 1 1 −3 1 1 1 (a) Determine the QR factorization of A. (b) Use the QR factors from part (a) to determine the least squares solution to Ax = b. To find the QR factorization for A, we begin by finding an othronormal basis for the columns of A. To this extent, we apply the Gram-Schmidt procedure to the columns of A by first letting 4 Math 317: Linear Algebra Homework 9 Solutions q1 = √ 1 1 1 0 12 + 12 + 12 + 02 . q2 is given by √ √ √ √ (0, 2, 1, 1) − 3(1/ 3, 1/ 3, 1/ 3, 0) a2 − (q1 · a2 ) q1 √ √ √ √ = q2 = ka2 − (q1 · a2 ) q1 k k(0, 2, 1, 1) − 3(1/ 3, 1/ 3, 1/ 3, 0)k √ −1/√ 3 1/ 3 = 0 . √ 1/ 3 Finally we have that q3 is given by a3 − (q1 · a3 ) q1 − (q2 · a3 ) q2 ka3 − (q1 · a3 ) q1 − (q2 · a3 ) q2 k √ √ √ √ √ √ √ √ (−1, 1, −3, 1) − (− 3)(1/ 3, 1/ 3, 1/ 3, 0) + 3(−1/ 3, 1/ 3, 0, 1/ 3) √ √ √ √ √ √ √ √ = k(−1, 1, −3, 1) − (− 3)(1/ 3, 1/ 3, 1/ 3, 0) + 3(−1/ 3, 1/ 3, 0, 1/ 3)k √ 1/√6 1/ 6 √ = −2/ 6 0 √ √ √ √ √ √ 1/√3 −1/√ 3 1/√6 3 3 − 1/ 3 1/ 3 √3 √ 1/ 6 √ and R = 0 So, Q = √ 3 √3 . 1/ 3 0√ −2/ 6 0 0 6 0 1/ 3 0 6. (a) Explain what happens when the Gram-Schmidt process is applied to an orthonormal set of vectors. Nothing! The resulting orthonormal set is the same as the original. (b) Explain what happens when the Gram-Schmidt process is applied to a linearly dependent set of vectors. It breaks down at the first vector such that xk ∈ span {x1 , . . . , xk−1 } because if xk ∈ span {x1 , . . . , xk−1 } = span {q1 , . . . , qk−1 } , 5 Math 317: Linear Algebra Homework 9 Solutions then we have that xk = 0 k0k Pk−1 i=1 (qi · xk )qi and so qk = Pk−1 xk − i=1 (qi ·xk )qi Pk−1 kxk − i=1 (qi ·xk )qi k = which is not defined. 6