算法导论 第三版 第28章 答案 英
- 格式：pdf
- 大小：163.03 KB
- 文档页数：12
下载文档原格式
- 1、下载文档前请自行甄别文档内容的完整性，平台不提供额外的编辑、内容补充、找答案等附加服务。
- 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
- 3、如文档侵犯您的权益，请联系客服反馈,我们会尽快为您处理(人工客服工作时间：9:00-18:30)。
Chapter 28
Michelle Bodnar, Andrew Lohr April 12, 2016
Exercise 28.1-1 We get the solution:
3 3 = 2 14 − 4 · 3 1 −7 − 5 · (14 − 4 · 3) + 6 · 3
Exercise 28.1-2 An LU decomposition of the matrix is 1 0 4 −5 6 8 −6 7 = 2 1 3 2 12 −7 12 Exercise 28.1-3 First, we find the LUP decomposition 1 5 2 0 5 8 of the given matrix 4 3 2 given by 0 4 0 0 1 0
know that L is the identity. Since A only has ones as its nonzero entries, and LU = U . U must also only have ones as its nonzero entries. So, we have U is the identity. This means that P A = I , which means that P = A−1 . This completes showing that the given decomposition is unique. Exercise 28.1-6 The zero matrix always has an LU decomposition by taking L to be any unit lower-triangular matrix and U to be the zero matrix, which is upper triangular. Exercise 28.1-7 For LU decomposition, it is indeed necessary. If we didn’t run the last run of the outermost for loop, unn would be left its initial value of zero instead of being set equal to ann . This can clearly produce incorrect results, because the LU decomposition of any non-singular matrix must have both L and U having positive determinant. However, if unn = 0, the determinant of U will be zero by problem D.2-2. For LUP-decomposition, the iteration of the outermost for loop that occurs with k = n will not change the final answer. Since π would have to be a permutation on a single element, it cannot be non-trivial. and the for loop on line 16 will not run at all. Exercise 28.2-1 Showing that being able to multiply matrices in time M (n) implies being able to square matrices in time M (n) is trivial because squaring a matrix is just multiplying it by itself. The more tricky direction is showing that being able to square matrices in time S (n) implies being able to multiply matrices in time O(S (n)). As we do this, we apply the same regularity condition that S (2n) ∈ O(S (n)). Suppose that we are trying to multiply the matrices, A and B , that is, find AB . Then, define the matrix I A C= 0 B Then, we can find C 2 in time S (2n) ∈ O(S (n)). Since C2 = I 0 A + AB B
−5 4 0
6 −5 . 4
we bring the 5 to the top, and then divide the first column by 5, and use Schur complements to change the rest of the matrix to get 5 8 2 .4 −3.2 2.2 .2 3.4 3.6 Then, we swap the third and second rows, and apply the Schur complement to get 5 8 2 .2 3.4 3.6 11.52 3.2 .4 − 3.4 2.2 + 3.4
1
This gets us the LUP decomposition that 1 0 0 1 0 L = .2 3.2 1 .4 − 3 .4 5 8 2 3.6 U = 0 3.4 11.52 0 0 2.2 + 3.4 0 0 1 P = 1 0 0 0 1 0 Using this, we can get that the solution must be 5 8 2 1 0 0 x1 5 .2 x2 = 12 3.6 1 0 0 3.4 11.52 3.2 9 x3 0 0 2.2 + 3.4 .4 − 3.4 1 Which, by forward substitution, 5 8 2 5 x1 0 3.4 x2 = 3.6 11 .52 35.2 x 0 0 2.2 + 11 7 + 3 3.4 3.4 So, finally by back substitution, 3 − 19 x1 x2 = − 1 19 49 x3 19 Exercise 28.1-4 The LUP decomposition of a diagonal matrix D is D = IDI where I is the identity matrix. Exercise 28.1-5 A LU decomposition of a permutation matrix is letting P be the inverse permutation matrix, and let both L and U be the identity matrix. Now, we need show th来自百度文库t this representation is unique. We know that the permutation matrix A is non-singular. This means that U has nonzero elements all along its diagonal. Now, suppose that there were some nonzero element off of the diagonal in L, which is to say Li,j = 0 for i = j . Then, look at row i in the product LU . This has a nonzero entry both at column j and at column i. Since it has more than one non-zero entry, it cannot be transformed into a permutation matrix by permuting the rows. Similarly, we have that U cannot have any off-diagonal elements. Lastly, since we know that both L and U are diagonal matrices, we 2
Michelle Bodnar, Andrew Lohr April 12, 2016
Exercise 28.1-1 We get the solution:
3 3 = 2 14 − 4 · 3 1 −7 − 5 · (14 − 4 · 3) + 6 · 3
Exercise 28.1-2 An LU decomposition of the matrix is 1 0 4 −5 6 8 −6 7 = 2 1 3 2 12 −7 12 Exercise 28.1-3 First, we find the LUP decomposition 1 5 2 0 5 8 of the given matrix 4 3 2 given by 0 4 0 0 1 0
know that L is the identity. Since A only has ones as its nonzero entries, and LU = U . U must also only have ones as its nonzero entries. So, we have U is the identity. This means that P A = I , which means that P = A−1 . This completes showing that the given decomposition is unique. Exercise 28.1-6 The zero matrix always has an LU decomposition by taking L to be any unit lower-triangular matrix and U to be the zero matrix, which is upper triangular. Exercise 28.1-7 For LU decomposition, it is indeed necessary. If we didn’t run the last run of the outermost for loop, unn would be left its initial value of zero instead of being set equal to ann . This can clearly produce incorrect results, because the LU decomposition of any non-singular matrix must have both L and U having positive determinant. However, if unn = 0, the determinant of U will be zero by problem D.2-2. For LUP-decomposition, the iteration of the outermost for loop that occurs with k = n will not change the final answer. Since π would have to be a permutation on a single element, it cannot be non-trivial. and the for loop on line 16 will not run at all. Exercise 28.2-1 Showing that being able to multiply matrices in time M (n) implies being able to square matrices in time M (n) is trivial because squaring a matrix is just multiplying it by itself. The more tricky direction is showing that being able to square matrices in time S (n) implies being able to multiply matrices in time O(S (n)). As we do this, we apply the same regularity condition that S (2n) ∈ O(S (n)). Suppose that we are trying to multiply the matrices, A and B , that is, find AB . Then, define the matrix I A C= 0 B Then, we can find C 2 in time S (2n) ∈ O(S (n)). Since C2 = I 0 A + AB B
−5 4 0
6 −5 . 4
we bring the 5 to the top, and then divide the first column by 5, and use Schur complements to change the rest of the matrix to get 5 8 2 .4 −3.2 2.2 .2 3.4 3.6 Then, we swap the third and second rows, and apply the Schur complement to get 5 8 2 .2 3.4 3.6 11.52 3.2 .4 − 3.4 2.2 + 3.4
1
This gets us the LUP decomposition that 1 0 0 1 0 L = .2 3.2 1 .4 − 3 .4 5 8 2 3.6 U = 0 3.4 11.52 0 0 2.2 + 3.4 0 0 1 P = 1 0 0 0 1 0 Using this, we can get that the solution must be 5 8 2 1 0 0 x1 5 .2 x2 = 12 3.6 1 0 0 3.4 11.52 3.2 9 x3 0 0 2.2 + 3.4 .4 − 3.4 1 Which, by forward substitution, 5 8 2 5 x1 0 3.4 x2 = 3.6 11 .52 35.2 x 0 0 2.2 + 11 7 + 3 3.4 3.4 So, finally by back substitution, 3 − 19 x1 x2 = − 1 19 49 x3 19 Exercise 28.1-4 The LUP decomposition of a diagonal matrix D is D = IDI where I is the identity matrix. Exercise 28.1-5 A LU decomposition of a permutation matrix is letting P be the inverse permutation matrix, and let both L and U be the identity matrix. Now, we need show th来自百度文库t this representation is unique. We know that the permutation matrix A is non-singular. This means that U has nonzero elements all along its diagonal. Now, suppose that there were some nonzero element off of the diagonal in L, which is to say Li,j = 0 for i = j . Then, look at row i in the product LU . This has a nonzero entry both at column j and at column i. Since it has more than one non-zero entry, it cannot be transformed into a permutation matrix by permuting the rows. Similarly, we have that U cannot have any off-diagonal elements. Lastly, since we know that both L and U are diagonal matrices, we 2