math 221 linear algebra
play

Math 221: LINEAR ALGEBRA 8-6. Singular Value Decomposition Le Chen 1 - PowerPoint PPT Presentation

Math 221: LINEAR ALGEBRA 8-6. Singular Value Decomposition Le Chen 1 Emory University, 2020 Fall (last updated on 08/27/2020) Creative Commons License (CC BY-NC-SA) 1 Slides are adapted from those by Karen Seyffarth from University of


  1. Math 221: LINEAR ALGEBRA §8-6. Singular Value Decomposition Le Chen 1 Emory University, 2020 Fall (last updated on 08/27/2020) Creative Commons License (CC BY-NC-SA) 1 Slides are adapted from those by Karen Seyffarth from University of Calgary.

  2. Singular Value Decomposition (SVD) Definition Let A be an m × n matrix. The singular values of A are the square roots of the nonzero eigenvalues of A T A.

  3. Singular Value Decomposition (SVD) Definition Let A be an m × n matrix. The singular values of A are the square roots of the nonzero eigenvalues of A T A. Singular Value Decomposition (SVD) can be thought of as a generalization of orthogonal diagonalization of a symmetric matrix to an arbitrary m × n matrix.

  4. Singular Value Decomposition (SVD) Definition Let A be an m × n matrix. The singular values of A are the square roots of the nonzero eigenvalues of A T A. Singular Value Decomposition (SVD) can be thought of as a generalization of orthogonal diagonalization of a symmetric matrix to an arbitrary m × n matrix. Given an m × n matrix A, we will see how to express A as a product A = U Σ V T where

  5. Singular Value Decomposition (SVD) Definition Let A be an m × n matrix. The singular values of A are the square roots of the nonzero eigenvalues of A T A. Singular Value Decomposition (SVD) can be thought of as a generalization of orthogonal diagonalization of a symmetric matrix to an arbitrary m × n matrix. Given an m × n matrix A, we will see how to express A as a product A = U Σ V T where ◮ U is an m × m orthogonal matrix whose columns are eigenvectors of AA T .

  6. Singular Value Decomposition (SVD) Definition Let A be an m × n matrix. The singular values of A are the square roots of the nonzero eigenvalues of A T A. Singular Value Decomposition (SVD) can be thought of as a generalization of orthogonal diagonalization of a symmetric matrix to an arbitrary m × n matrix. Given an m × n matrix A, we will see how to express A as a product A = U Σ V T where ◮ U is an m × m orthogonal matrix whose columns are eigenvectors of AA T . ◮ V is an n × n orthogonal matrix whose columns are eigenvectors of A T A.

  7. Singular Value Decomposition (SVD) Definition Let A be an m × n matrix. The singular values of A are the square roots of the nonzero eigenvalues of A T A. Singular Value Decomposition (SVD) can be thought of as a generalization of orthogonal diagonalization of a symmetric matrix to an arbitrary m × n matrix. Given an m × n matrix A, we will see how to express A as a product A = U Σ V T where ◮ U is an m × m orthogonal matrix whose columns are eigenvectors of AA T . ◮ V is an n × n orthogonal matrix whose columns are eigenvectors of A T A. ◮ Σ is an m × n matrix whose only nonzero values lie on its main diagonal, and are the square roots of the eigenvalues of both AA T and A T A.

  8. Singular Value Decomposition (SVD) Definition Let A be an m × n matrix. The singular values of A are the square roots of the nonzero eigenvalues of A T A. Singular Value Decomposition (SVD) can be thought of as a generalization of orthogonal diagonalization of a symmetric matrix to an arbitrary m × n matrix. Given an m × n matrix A, we will see how to express A as a product A = U Σ V T where ◮ U is an m × m orthogonal matrix whose columns are eigenvectors of AA T . ◮ V is an n × n orthogonal matrix whose columns are eigenvectors of A T A. ◮ Σ is an m × n matrix whose only nonzero values lie on its main diagonal, and are the square roots of the eigenvalues of both AA T and A T A. Remark Although we haven’t proved it, A T A and AA T have the same nonzero eigenvalues.

  9. Example � 1 � − 1 3 Let A = . Then 3 1 1 � 1 � 11 �   1 3 − 1 3 5 � AA T =  = − 1 1 .  3 1 1 5 11 3 1 � 1     1 3 10 2 6 � − 1 3 A T A =  . − 1 1 = 2 2 − 2    3 1 1 3 1 6 − 2 10

  10. Example � 1 � − 1 3 Let A = . Then 3 1 1 � 1 � 11 �   1 3 − 1 3 5 � AA T =  = − 1 1 .  3 1 1 5 11 3 1 � 1     1 3 10 2 6 � − 1 3 A T A =  . − 1 1 = 2 2 − 2    3 1 1 3 1 6 − 2 10

  11. Example (continued) Since AA T is 2 × 2 while A T A is 3 × 3 , and AA T and A T A have the same nonzero eigenvalues, compute c AA T ( x ) (because it’s easier to compute than c A T A ( x ) ). � � x − 11 − 5 det ( xI − AA T ) = � � c AA T ( x ) = � � − 5 x − 11 � � ( x − 11) 2 − 25 = x 2 − 22 x + 121 − 25 = x 2 − 22 x + 96 = = ( x − 16)( x − 6) . Therefore, the eigenvalues of AA T are λ 1 = 16 and λ 2 = 6 .

  12. so . Since the eigenvalues of . : solve so . : solve need only normalize them. are distinct, the corresponding eigenvectors are orthogonal, and we , fjnd eigenvectors for from largest to smallest). eigenvalues (and corresponding singular values) in nonincreasing order (i.e., Example (continued) The eigenvalues of A T A are λ 1 = 16 , λ 2 = 6 , and λ 3 = 0 , and the singular √ √ values of A are σ 1 = 16 = 4 and σ 2 = 6 . By convention, we list the

  13. so need only normalize them. . : solve so . : solve eigenvalues (and corresponding singular values) in nonincreasing order (i.e., from largest to smallest). Example (continued) The eigenvalues of A T A are λ 1 = 16 , λ 2 = 6 , and λ 3 = 0 , and the singular √ √ values of A are σ 1 = 16 = 4 and σ 2 = 6 . By convention, we list the To find the matrix V , fjnd eigenvectors for A T A . Since the eigenvalues of AA T are distinct, the corresponding eigenvectors are orthogonal, and we

  14. so eigenvalues (and corresponding singular values) in nonincreasing order (i.e., . : solve need only normalize them. from largest to smallest). Example (continued) The eigenvalues of A T A are λ 1 = 16 , λ 2 = 6 , and λ 3 = 0 , and the singular √ √ values of A are σ 1 = 16 = 4 and σ 2 = 6 . By convention, we list the To find the matrix V , fjnd eigenvectors for A T A . Since the eigenvalues of AA T are distinct, the corresponding eigenvectors are orthogonal, and we λ 1 = 16 : solve (16 I − A T A ) � y 1 = � 0 .         t 6 − 2 − 6 0 1 0 − 1 0 1  , so �  , t ∈ R .  →  = t − 2 14 2 0 0 1 0 0 y 1 = 0 0     t − 6 2 6 0 0 0 0 0 1

  15. need only normalize them. eigenvalues (and corresponding singular values) in nonincreasing order (i.e., from largest to smallest). Example (continued) The eigenvalues of A T A are λ 1 = 16 , λ 2 = 6 , and λ 3 = 0 , and the singular √ √ values of A are σ 1 = 16 = 4 and σ 2 = 6 . By convention, we list the To find the matrix V , fjnd eigenvectors for A T A . Since the eigenvalues of AA T are distinct, the corresponding eigenvectors are orthogonal, and we λ 1 = 16 : solve (16 I − A T A ) � y 1 = � 0 .         t 6 − 2 − 6 0 1 0 − 1 0 1  , so �  , t ∈ R .  →  = t − 2 14 2 0 0 1 0 0 y 1 = 0 0     t − 6 2 6 0 0 0 0 0 1 λ 2 = 6 : solve (6 I − A T A ) � y 2 = � 0 .         − 4 − 2 − 6 0 1 0 1 0 − s − 1  , so �  = s  , s ∈ R . − 2 4 2 0  → 0 1 1 0 y 2 = − s − 1     − 6 2 − 4 0 0 0 0 0 s 1

  16. . Let to fjnd , and , and we use Also, Then Example (continued) λ 3 = 0 : solve ( − A T A ) � y 3 = � 0 .         − 10 − 2 − 6 0 1 0 1 0 − r − 1  , so �  , r ∈ R .  →  = r − 2 − 2 2 0 0 1 − 2 0 y 3 = 2 r 2     − 6 2 − 10 0 0 0 0 0 r 1

  17. . Let to fjnd , and , and we use Also, Then Example (continued) λ 3 = 0 : solve ( − A T A ) � y 3 = � 0 .         − 10 − 2 − 6 0 1 0 1 0 − r − 1  , so �  , r ∈ R .  →  = r − 2 − 2 2 0 0 1 − 2 0 y 3 = 2 r 2     − 6 2 − 10 0 0 0 0 0 r 1       1 − 1 − 1 1 1 1  ,�  ,�  . v 1 = � 0 v 2 = − 1 v 3 = 2 √  √  √  2 3 6 1 1 1 √ √   3 2 − 1 − 1 √  . V = 0 2 2 √  − 6 √ √ 3 2 1

  18. Also, Then Let Example (continued) λ 3 = 0 : solve ( − A T A ) � y 3 = � 0 .         − 10 − 2 − 6 0 1 0 1 0 − r − 1  , so �  , r ∈ R .  →  = r − 2 − 2 2 0 0 1 − 2 0 y 3 = 2 r 2     − 6 2 − 10 0 0 0 0 0 r 1       1 − 1 − 1 1 1 1  ,�  ,�  . v 1 = � 0 v 2 = − 1 v 3 = 2 √  √  √  2 3 6 1 1 1 √ √   3 2 − 1 − 1 √  . V = 0 2 2 √  − 6 √ √ 3 2 1 � 4 � 0 0 Σ = , √ 0 6 0 and we use A , V T , and Σ to fjnd U .

  19. and Then we have . Thus, and which implies that Example (continued) Since V is orthogonal and A = U Σ V T , it follows that AV = U Σ . Let � � � � V = � v 1 � v 2 � v 3 , and let U = � u 1 � u 2 , where � u 1 and � u 2 are the two columns of U .

Recommend


More recommend