bounds on the solution of sylvester equation and
play

Bounds on the solution of Sylvester equation and something else - PowerPoint PPT Presentation

Aug 31, 2022 •36 likes •1.5k views

Bounds on the solution of Sylvester equation and something else Ninoslav Truhar Department of Mathematics, University of Osijek ntruhar@mathos.hr 24.04.2009 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 1 / 22 Coauthors:

  1. Estimating the size of the solution AX − XB = GF ∗ the main tool; Consider: factorized version of ADI iterations: 1 solve ( A − β i I ) X i +1 / 2 = X i ( B − β i I ) + GF ∗ for X i +1 / 2 ; 2 solve X i +1 ( B − α i I ) = ( A − α i I ) X i +1 / 2 − GF ∗ for X i +1 , where { α i } � = { β i } are given two sets of parameters. It holods X i +1 − X = ( A − α i I )( A − β i I ) − 1 ( X i − X )( B − β i I )( B − α i I ) − 1 ,     i i � ( A − α j I )( A − β j I ) − 1  ( X 0 − X ) � ( B − β j I )( B − α j I ) − 1  , =   j =0 j =0 In addition, if { α i } n j =1 = σ ( A ) or { β j } m j =1 = σ ( B ), N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 6 / 22

  2. Estimating the size of the solution AX − XB = GF ∗ the main tool; Consider: factorized version of ADI iterations: 1 solve ( A − β i I ) X i +1 / 2 = X i ( B − β i I ) + GF ∗ for X i +1 / 2 ; 2 solve X i +1 ( B − α i I ) = ( A − α i I ) X i +1 / 2 − GF ∗ for X i +1 , where { α i } � = { β i } are given two sets of parameters. It holods X i +1 − X = ( A − α i I )( A − β i I ) − 1 ( X i − X )( B − β i I )( B − α i I ) − 1 ,     i i � ( A − α j I )( A − β j I ) − 1  ( X 0 − X ) � ( B − β j I )( B − α j I ) − 1  , =   j =0 j =0 In addition, if { α i } n j =1 = σ ( A ) or { β j } m j =1 = σ ( B ), then X min { m , n } = X , N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 6 / 22

  3. Estimating the size of the solution AX − XB = GF ∗ the main tool; Consider: factorized version of ADI iterations: 1 solve ( A − β i I ) X i +1 / 2 = X i ( B − β i I ) + GF ∗ for X i +1 / 2 ; 2 solve X i +1 ( B − α i I ) = ( A − α i I ) X i +1 / 2 − GF ∗ for X i +1 , where { α i } � = { β i } are given two sets of parameters. It holods X i +1 − X = ( A − α i I )( A − β i I ) − 1 ( X i − X )( B − β i I )( B − α i I ) − 1 ,     i i � ( A − α j I )( A − β j I ) − 1  ( X 0 − X ) � ( B − β j I )( B − α j I ) − 1  , =   j =0 j =0 In addition, if { α i } n j =1 = σ ( A ) or { β j } m j =1 = σ ( B ), then X min { m , n } = X , the Hamilton-Cayley theorem, p ( A ) ≡ 0 for p ( λ ) def = det( λ I − A ) and q ( B ) ≡ 0 for q ( λ ) def = det( λ I − B ). N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 6 / 22

  4. Low Rank ADI (LRADI) Method for Sylvester Equations N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 7 / 22

  5. Low Rank ADI (LRADI) Method for Sylvester Equations For A ∈ R n × n , B ∈ R m × m G ∈ R m × r and F ∈ R n × r , N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 7 / 22

  6. Low Rank ADI (LRADI) Method for Sylvester Equations For A ∈ R n × n , B ∈ R m × m G ∈ R m × r and F ∈ R n × r , AX − XB = GF ∗ . consider Sylvester equation: N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 7 / 22

  7. Low Rank ADI (LRADI) Method for Sylvester Equations For A ∈ R n × n , B ∈ R m × m G ∈ R m × r and F ∈ R n × r , AX − XB = GF ∗ . consider Sylvester equation: Input : (a) A , B , G , and F ; (b) ADI shifts { β 1 , β 2 , . . . } , { α 1 , α 2 , . . . } ; (c) k , the number of ADI steps; N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 7 / 22

  8. Low Rank ADI (LRADI) Method for Sylvester Equations For A ∈ R n × n , B ∈ R m × m G ∈ R m × r and F ∈ R n × r , AX − XB = GF ∗ . consider Sylvester equation: Input : (a) A , B , G , and F ; (b) ADI shifts { β 1 , β 2 , . . . } , { α 1 , α 2 , . . . } ; (c) k , the number of ADI steps; Output: Z ( m × kr ) , D ( kr × kr ) , and Y ( n × kr ) such that ZDY ∗ ≈ X Z 1 = ( A − β 1 I ) − 1 G ; ( Y ∗ ) 1 = F ∗ ( B − α 1 I ) − 1 ; 1. 2. for i = 1 , 2 , . . . , k do Z i = Z i − 1 + ( β i +1 − α i )( A − β i +1 I ) − 1 Z i − 1 ; 3. ( Y ∗ ) i = ( Y ∗ ) i − 1 + ( α i +1 − β i ) ( Y ∗ ) i − 1 ( B − α i +1 I ) − 1 ; 4. 5. end for; 6. D = diag (( β 1 − α 1 ) I r , . . . , ( β k − α k ) I r ). N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 7 / 22

  9. Low Rank ADI (LRADI) Method for Sylvester Equations For A ∈ R n × n , B ∈ R m × m G ∈ R m × r and F ∈ R n × r , AX − XB = GF ∗ . consider Sylvester equation: Input : (a) A , B , G , and F ; (b) ADI shifts { β 1 , β 2 , . . . } , { α 1 , α 2 , . . . } ; (c) k , the number of ADI steps; Output: Z ( m × kr ) , D ( kr × kr ) , and Y ( n × kr ) such that ZDY ∗ ≈ X Z 1 = ( A − β 1 I ) − 1 G ; ( Y ∗ ) 1 = F ∗ ( B − α 1 I ) − 1 ; 1. 2. for i = 1 , 2 , . . . , k do Z i = Z i − 1 + ( β i +1 − α i )( A − β i +1 I ) − 1 Z i − 1 ; 3. ( Y ∗ ) i = ( Y ∗ ) i − 1 + ( α i +1 − β i ) ( Y ∗ ) i − 1 ( B − α i +1 I ) − 1 ; 4. 5. end for; 6. D = diag (( β 1 − α 1 ) I r , . . . , ( β k − α k ) I r ). Recall: if { α i } n j =1 = σ ( A ) or { β j } m j =1 = σ ( B ), N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 7 / 22

  10. Low Rank ADI (LRADI) Method for Sylvester Equations For A ∈ R n × n , B ∈ R m × m G ∈ R m × r and F ∈ R n × r , AX − XB = GF ∗ . consider Sylvester equation: Input : (a) A , B , G , and F ; (b) ADI shifts { β 1 , β 2 , . . . } , { α 1 , α 2 , . . . } ; (c) k , the number of ADI steps; Output: Z ( m × kr ) , D ( kr × kr ) , and Y ( n × kr ) such that ZDY ∗ ≈ X Z 1 = ( A − β 1 I ) − 1 G ; ( Y ∗ ) 1 = F ∗ ( B − α 1 I ) − 1 ; 1. 2. for i = 1 , 2 , . . . , k do Z i = Z i − 1 + ( β i +1 − α i )( A − β i +1 I ) − 1 Z i − 1 ; 3. ( Y ∗ ) i = ( Y ∗ ) i − 1 + ( α i +1 − β i ) ( Y ∗ ) i − 1 ( B − α i +1 I ) − 1 ; 4. 5. end for; 6. D = diag (( β 1 − α 1 ) I r , . . . , ( β k − α k ) I r ). Recall: if { α i } n j =1 = σ ( A ) or { β j } m j =1 = σ ( B ), min { m , n } ( β j − α j ) Z ( j ) Y ( j ) ∗ � then X = X min { m , n } = j =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 7 / 22

  11. Estimating the size of the solution N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 8 / 22

  12. Estimating the size of the solution Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 8 / 22

  13. Estimating the size of the solution Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , min { m , n } m n | τ ( ℓ, j − 1) | � ˆ | σ ( i , j − 1) | · � ˆ g i � f ℓ � � � � � X � ≤ � S �� T − 1 � | β j − α j | , | λ i − β j | | µ ℓ − α j | j =1 i =1 ℓ =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 8 / 22

  14. Estimating the size of the solution Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , min { m , n } m n | τ ( ℓ, j − 1) | � ˆ | σ ( i , j − 1) | · � ˆ g i � f ℓ � � � � � X � ≤ � S �� T − 1 � | β j − α j | , | λ i − β j | | µ ℓ − α j | j =1 i =1 ℓ =1 j − 1 λ i − α s � σ ( i , j − 1) = , σ ( i , 0) = 1 , i = 1 , . . . , m , λ i − β s s =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 8 / 22

  15. Estimating the size of the solution Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , min { m , n } m n | τ ( ℓ, j − 1) | � ˆ | σ ( i , j − 1) | · � ˆ g i � f ℓ � � � � � X � ≤ � S �� T − 1 � | β j − α j | , | λ i − β j | | µ ℓ − α j | j =1 i =1 ℓ =1 j − 1 λ i − α s � σ ( i , j − 1) = , σ ( i , 0) = 1 , i = 1 , . . . , m , λ i − β s s =1 j − 1 µ i − β s � τ ( i , j − 1) = , τ ( i , 0) = 1 , i = 1 , . . . , n . µ i − α s s =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 8 / 22

  16. Estimating the size of the solution Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , min { m , n } m n | τ ( ℓ, j − 1) | � ˆ | σ ( i , j − 1) | · � ˆ g i � f ℓ � � � � � X � ≤ � S �� T − 1 � | β j − α j | , | λ i − β j | | µ ℓ − α j | j =1 i =1 ℓ =1 j − 1 λ i − α s � σ ( i , j − 1) = , σ ( i , 0) = 1 , i = 1 , . . . , m , λ i − β s s =1 j − 1 µ i − β s � τ ( i , j − 1) = , τ ( i , 0) = 1 , i = 1 , . . . , n . µ i − α s s =1 F ∗ = F ∗ T g i is the i th row of ˆ ˆ f i is the i th column of ˆ G = S − 1 G ; ˆ N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 8 / 22

  17. Example N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 9 / 22

  18. Example Let A = S diag (1 , 2 , 3 , 4 , 5) S − 1 where   5 . 04 − 4 . 2 0 . 0013 − 0 . 003 0 . 002 − 3 . 91 − 3 . 67 0 . 0028 0 . 01 0 . 004     S = 0 . 001 0 . 0035 − 1 . 55 − 0 . 11 − 0 . 21     − 0 . 002 0 . 0011 0 . 32 − 1 . 82 0 . 42   − 0 . 004 0 . 0025 0 . 79 1 . 29 − 0 . 89 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 9 / 22

  19. Example Let A = S diag (1 , 2 , 3 , 4 , 5) S − 1 where   5 . 04 − 4 . 2 0 . 0013 − 0 . 003 0 . 002 − 3 . 91 − 3 . 67 0 . 0028 0 . 01 0 . 004     S = 0 . 001 0 . 0035 − 1 . 55 − 0 . 11 − 0 . 21     − 0 . 002 0 . 0011 0 . 32 − 1 . 82 0 . 42   − 0 . 004 0 . 0025 0 . 79 1 . 29 − 0 . 89 Let B = T diag (11 , 12 , 13 , 14 , 5 . 0001) T − 1 where  5 . 0443 − 4 . 1948 0 . 0022 0 . 0062 0 . 0023  − 3 . 9041 − 3 . 6609 0 . 0032 0 . 0105 0 . 0045   T = 0 . 0019 0 . 0101 − 1 . 5425 − 0 . 1093 − 0 . 2076  .     0 . 0005 0 . 0061 0 . 3232 − 1 . 8143 0 . 4213  0 . 0019 0 . 0088 0 . 7907 1 . 2944 − 0 . 8837 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 9 / 22

  20. Example N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 10 / 22

  21. Example � T � T � 2 � 2 . 2 5 0 0 0 4 . 5 0 0 0 G = , F = . 5 1 0 0 0 5 . 1 1 . 1 0 0 0 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 10 / 22

  22. Example � T � T � 2 � 2 . 2 5 0 0 0 4 . 5 0 0 0 G = , F = . 5 1 0 0 0 5 . 1 1 . 1 0 0 0 Consider AX − XB = GF ∗ , N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 10 / 22

  23. Example � T � T � 2 � 2 . 2 5 0 0 0 4 . 5 0 0 0 G = , F = . 5 1 0 0 0 5 . 1 1 . 1 0 0 0 Consider AX − XB = GF ∗ , � X � = 4 . 4884 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 10 / 22

  24. Example � T � T � 2 � 2 . 2 5 0 0 0 4 . 5 0 0 0 G = , F = . 5 1 0 0 0 5 . 1 1 . 1 0 0 0 Consider AX − XB = GF ∗ , � X � = 4 . 4884 � X � ≤ 176 . 5 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 10 / 22

  25. Example � T � T � 2 � 2 . 2 5 0 0 0 4 . 5 0 0 0 G = , F = . 5 1 0 0 0 5 . 1 1 . 1 0 0 0 Consider AX − XB = GF ∗ , � X � = 4 . 4884 � X � ≤ 176 . 5 � − 1 �� vec ( W ) � ≤ 34469 ( I ⊗ A ) − ( B T ⊗ I ) � � X � ≤ � N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 10 / 22

  26. Error Bound for Sylvester equation N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 11 / 22

  27. Error Bound for Sylvester equation Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 11 / 22

  28. Error Bound for Sylvester equation Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , Let X k obtained by LRADI, s.t. X k ≈ X , δ X = X − X k N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 11 / 22

  29. Error Bound for Sylvester equation Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , Let X k obtained by LRADI, s.t. X k ≈ X , δ X = X − X k min { m , n } m n | τ ( ℓ, j − 1) | � ˆ | σ ( i , j − 1) | · � ˆ g i � f ℓ � � � � � δ X � ≤ � S �� T − 1 � | β j − α j | , | λ i − β j | | µ ℓ − α j | j = k +1 i =1 ℓ =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 11 / 22

  30. Error Bound for Sylvester equation Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , Let X k obtained by LRADI, s.t. X k ≈ X , δ X = X − X k min { m , n } m n | τ ( ℓ, j − 1) | � ˆ | σ ( i , j − 1) | · � ˆ g i � f ℓ � � � � � δ X � ≤ � S �� T − 1 � | β j − α j | , | λ i − β j | | µ ℓ − α j | j = k +1 i =1 ℓ =1 j − 1 λ i − α s � σ ( i , j − 1) = , σ ( i , 0) = 1 , i = 1 , . . . , m , λ i − β s s =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 11 / 22

  31. Error Bound for Sylvester equation Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , Let X k obtained by LRADI, s.t. X k ≈ X , δ X = X − X k min { m , n } m n | τ ( ℓ, j − 1) | � ˆ | σ ( i , j − 1) | · � ˆ g i � f ℓ � � � � � δ X � ≤ � S �� T − 1 � | β j − α j | , | λ i − β j | | µ ℓ − α j | j = k +1 i =1 ℓ =1 j − 1 λ i − α s � σ ( i , j − 1) = , σ ( i , 0) = 1 , i = 1 , . . . , m , λ i − β s s =1 j − 1 µ i − β s � τ ( i , j − 1) = , τ ( i , 0) = 1 , i = 1 , . . . , n . µ i − α s s =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 11 / 22

  32. Error Bound for Sylvester equation Consider AX − XB = GF ∗ , where A and B are diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 , Let X k obtained by LRADI, s.t. X k ≈ X , δ X = X − X k min { m , n } m n | τ ( ℓ, j − 1) | � ˆ | σ ( i , j − 1) | · � ˆ g i � f ℓ � � � � � δ X � ≤ � S �� T − 1 � | β j − α j | , | λ i − β j | | µ ℓ − α j | j = k +1 i =1 ℓ =1 j − 1 λ i − α s � σ ( i , j − 1) = , σ ( i , 0) = 1 , i = 1 , . . . , m , λ i − β s s =1 j − 1 µ i − β s � τ ( i , j − 1) = , τ ( i , 0) = 1 , i = 1 , . . . , n . µ i − α s s =1 F ∗ = F ∗ T g i is the i th row of ˆ ˆ f i is the i th column of ˆ G = S − 1 G ; ˆ N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 11 / 22

  33. Error Bound for Lyapunov equation N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 12 / 22

  34. Error Bound for Lyapunov equation Let X k be obtained by LRADI, s.t. X k ≈ X N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 12 / 22

  35. Error Bound for Lyapunov equation Let X k be obtained by LRADI, s.t. X k ≈ X How close is X k to X ? N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 12 / 22

  36. Error Bound for Lyapunov equation Let X k be obtained by LRADI, s.t. X k ≈ X How close is X k to X ? For the case of Lyapunov equation: AX − XA ∗ = GG ∗ , A = S Λ S − 1 diagonalizable N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 12 / 22

  37. Error Bound for Lyapunov equation Let X k be obtained by LRADI, s.t. X k ≈ X How close is X k to X ? For the case of Lyapunov equation: AX − XA ∗ = GG ∗ , A = S Λ S − 1 diagonalizable ( T . and Veseli´ c 2007): m m σ ( j , k ) 2 · � ˆ � X − X k � ≤ � S � 2 � � g k � 2 ( − 2 Re ( α j )) j = k +1 k =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 12 / 22

  38. Error Bound for Lyapunov equation Let X k be obtained by LRADI, s.t. X k ≈ X How close is X k to X ? For the case of Lyapunov equation: AX − XA ∗ = GG ∗ , A = S Λ S − 1 diagonalizable ( T . and Veseli´ c 2007): m m σ ( j , k ) 2 · � ˆ � X − X k � ≤ � S � 2 � � g k � 2 ( − 2 Re ( α j )) j = k +1 k =1 j − 1 1 1 λ k − ¯ α t � σ (1 , k ) = , and σ ( j , k ) = for j > 1 . λ k + α 1 λ k + α j λ k + α t t =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 12 / 22

  39. Error Bound for Lyapunov equation Let X k be obtained by LRADI, s.t. X k ≈ X How close is X k to X ? For the case of Lyapunov equation: AX − XA ∗ = GG ∗ , A = S Λ S − 1 diagonalizable ( T . and Veseli´ c 2007): m m σ ( j , k ) 2 · � ˆ � X − X k � ≤ � S � 2 � � g k � 2 ( − 2 Re ( α j )) j = k +1 k =1 j − 1 1 1 λ k − ¯ α t � σ (1 , k ) = , and σ ( j , k ) = for j > 1 . λ k + α 1 λ k + α j λ k + α t t =1 g k is the k th row of ˆ G = S − 1 G ˆ N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 12 / 22

  40. Perturbation Bound for Sylvester equation N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 13 / 22

  41. Perturbation Bound for Sylvester equation Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( G + δ G )( F + δ F ) ∗ , N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 13 / 22

  42. Perturbation Bound for Sylvester equation Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( G + δ G )( F + δ F ) ∗ , Using N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 13 / 22

  43. Perturbation Bound for Sylvester equation Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( G + δ G )( F + δ F ) ∗ , Using A δ X − δ X B ≈ G δ F ∗ + δ G F ∗ − δ A X + X δ B , N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 13 / 22

  44. Perturbation Bound for Sylvester equation Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( G + δ G )( F + δ F ) ∗ , Using A δ X − δ X B ≈ G δ F ∗ + δ G F ∗ − δ A X + X δ B , one gets: δ X ≈ δ X 1 + δ X 2 + δ X 3 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 13 / 22

  45. Perturbation Bound for Sylvester equation Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( G + δ G )( F + δ F ) ∗ , Using A δ X − δ X B ≈ G δ F ∗ + δ G F ∗ − δ A X + X δ B , one gets: δ X ≈ δ X 1 + δ X 2 + δ X 3 A δ X 1 − δ X 1 B = − δ A X , A δ X 2 − δ X 2 B = − X δ B , ( G δ F ∗ + δ G F ∗ ) . A δ X 3 − δ X 3 B = N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 13 / 22

  46. Perturbation Bound for Sylvester equation N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 14 / 22

  47. Perturbation Bound for Sylvester equation The following perturbation bound holds: N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 14 / 22

  48. Perturbation Bound for Sylvester equation The following perturbation bound holds: for A and B diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 . N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 14 / 22

  49. Perturbation Bound for Sylvester equation The following perturbation bound holds: for A and B diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 . and sufficiently small η = max {� δ A � , � δ B � , � δ G � , � δ F �} , N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 14 / 22

  50. Perturbation Bound for Sylvester equation The following perturbation bound holds: for A and B diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 . and sufficiently small η = max {� δ A � , � δ B � , � δ G � , � δ F �} , � � δ X � κ ( T ) � S �� S − 1 δ A � + κ ( S ) � T − 1 �� δ BT � ≤ � X � + � S �� T − 1 �� S − 1 ( G δ F ∗ + δ GF ∗ ) T � � γ + O ( η 2 ) , � X � N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 14 / 22

  51. Perturbation Bound for Sylvester equation The following perturbation bound holds: for A and B diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 . and sufficiently small η = max {� δ A � , � δ B � , � δ G � , � δ F �} , � � δ X � κ ( T ) � S �� S − 1 δ A � + κ ( S ) � T − 1 �� δ BT � ≤ � X � + � S �� T − 1 �� S − 1 ( G δ F ∗ + δ GF ∗ ) T � � γ + O ( η 2 ) , � X � n 0 | µ j − λ j | τ ( j ) max σ ( j ) � γ = max j =1 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 14 / 22

  52. Perturbation Bound for Sylvester equation The following perturbation bound holds: for A and B diagonalizable; A = S diag ( λ i ) S − 1 , B = T diag ( µ i ) T − 1 . and sufficiently small η = max {� δ A � , � δ B � , � δ G � , � δ F �} , � � δ X � κ ( T ) � S �� S − 1 δ A � + κ ( S ) � T − 1 �� δ BT � ≤ � X � + � S �� T − 1 �� S − 1 ( G δ F ∗ + δ GF ∗ ) T � � γ + O ( η 2 ) , � X � n 0 | µ j − λ j | τ ( j ) max σ ( j ) � γ = max where: j =1 | σ ( i , j − 1) | | τ ( i , j − 1) | σ ( j ) | λ i − µ j | , τ ( j ) max = � ∆ Φ ( j ) � = max max = � ∆ Ψ ( j ) � = max | µ i − λ j | . i i N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 14 / 22

  53. Illustration of new perturbation bound N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 15 / 22

  54. Illustration of new perturbation bound Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( C + δ C ) , N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 15 / 22

  55. Illustration of new perturbation bound Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( C + δ C ) , The standard bound (N. J. Higham, 1996): √ � δ X � F 3Ψ ǫ + O ( ǫ 2 ) , ≤ � X � F N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 15 / 22

  56. Illustration of new perturbation bound Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( C + δ C ) , The standard bound (N. J. Higham, 1996): √ � δ X � F 3Ψ ǫ + O ( ǫ 2 ) , ≤ � X � F where Ψ = � P − 1 [ α ( X T ⊗ I m ) − β ( I n ⊗ X ) − δ I mn ] � / � X � F , P = I n ⊗ A − B T ⊗ I m N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 15 / 22

  57. Illustration of new perturbation bound Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( C + δ C ) , The standard bound (N. J. Higham, 1996): √ � δ X � F 3Ψ ǫ + O ( ǫ 2 ) , ≤ � X � F where Ψ = � P − 1 [ α ( X T ⊗ I m ) − β ( I n ⊗ X ) − δ I mn ] � / � X � F , P = I n ⊗ A − B T ⊗ I m � � � δ A � F , � δ B � F , � δ C � F ǫ = max , α β δ N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 15 / 22

  58. Illustration of new perturbation bound Consider perturbed Sylvester equation ( A + δ A )( X + δ X ) − ( X + δ X )( B + δ B ) = ( C + δ C ) , The standard bound (N. J. Higham, 1996): √ � δ X � F 3Ψ ǫ + O ( ǫ 2 ) , ≤ � X � F where Ψ = � P − 1 [ α ( X T ⊗ I m ) − β ( I n ⊗ X ) − δ I mn ] � / � X � F , P = I n ⊗ A − B T ⊗ I m � � � δ A � F , � δ B � F , � δ C � F ǫ = max , α = � A � F , β = � B � F and δ = � C � F α β δ N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 15 / 22

  59. Illustration of new perturbation bound N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 16 / 22

  60. Illustration of new perturbation bound Let A = S Λ S − 1 with Λ = diag (30 , 40 , 60 , 70 , 90) N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 16 / 22

  61. Illustration of new perturbation bound Let A = S Λ S − 1 with Λ = diag (30 , 40 , 60 , 70 , 90)  0 . 052 0 . 494 0 . 935 0 . 592 0 . 827  0 . 264 0 . 094 0 . 220 0 . 388 0 . 387     S = 0 . 757 0 . 971 0 . 189 0 . 935 0 . 445 .     0 . 435 0 . 424 0 . 280 0 . 826 0 . 009   0 . 506 0 . 270 0 . 674 0 . 051 0 . 280 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 16 / 22

  62. Illustration of new perturbation bound Let A = S Λ S − 1 with Λ = diag (30 , 40 , 60 , 70 , 90)  0 . 052 0 . 494 0 . 935 0 . 592 0 . 827  0 . 264 0 . 094 0 . 220 0 . 388 0 . 387     S = 0 . 757 0 . 971 0 . 189 0 . 935 0 . 445 .     0 . 435 0 . 424 0 . 280 0 . 826 0 . 009   0 . 506 0 . 270 0 . 674 0 . 051 0 . 280 B = T Ω T − 1 with Ω = diag (100 , 200 , 300 , 400 , 450) N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 16 / 22

  63. Illustration of new perturbation bound Let A = S Λ S − 1 with Λ = diag (30 , 40 , 60 , 70 , 90)  0 . 052 0 . 494 0 . 935 0 . 592 0 . 827  0 . 264 0 . 094 0 . 220 0 . 388 0 . 387     S = 0 . 757 0 . 971 0 . 189 0 . 935 0 . 445 .     0 . 435 0 . 424 0 . 280 0 . 826 0 . 009   0 . 506 0 . 270 0 . 674 0 . 051 0 . 280 B = T Ω T − 1 with Ω = diag (100 , 200 , 300 , 400 , 450)  85610 89420 42300 31670 97380  9709 64180 89230 36420 28030     T = 0 . 3910 0 . 2030 0 . 9910 0 . 6150 0 . 1740 .     0 . 3760 0 . 0680 0 . 4000 0 . 5970 0 . 9210   0 . 5150 0 . 8000 0 . 3400 0 . 6850 0 . 2600 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 16 / 22

  64. Illustration of new perturbation bound N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 17 / 22

  65. Illustration of new perturbation bound Further, � − 10 � T − 60 0 0 0 0 G = , F = G . − 60 400 0 0 0 0 N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 17 / 22

  66. Illustration of new perturbation bound Further, � − 10 � T − 60 0 0 0 0 G = , F = G . − 60 400 0 0 0 0 Perturbations are δ B = 10 − 9 · diag 10 − 5 , 10 − 5 , 1 , 1 , 1 � � , δ A = 0 , δ F = δ G = 0 . N. Truhar (Osijek, 2009) Sylvester equation 24.04.2009 17 / 22

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Consistency and efficient solution of the Sylvester equation for -congruence: AX + X B = C

Consistency and efficient solution of the Sylvester equation for -congruence: AX + X B = C

Consistency and efficient solution of the Sylvester equation for -congruence: AX + X B = C Fernando De Tern and Froiln M. Dopico ICMAT and Departamento de Matemticas, Universidad Carlos III de Madrid, Spain Special thanks to Daniel

668 views • 65 slides
Sherman-Morrison-Woodbury formula for Sylvester and T -Sylvester equation Ivana Kuzmanovi

Sherman-Morrison-Woodbury formula for Sylvester and T -Sylvester equation Ivana Kuzmanovi

Sherman-Morrison-Woodbury formula for Sylvester and T -Sylvester equation Ivana Kuzmanovi Department of Mathematics, University of Osijek, Croatia I.Kuzmanovi (University of Osijek, Croatia) 1 / 9 Problem We consider the Sylvester equation

1.61k views • 17 slides
Equations in One Variable Definition 1 (Equation) . An equation is a state- ment that two

Equations in One Variable Definition 1 (Equation) . An equation is a state- ment that two

Equations in One Variable Definition 1 (Equation) . An equation is a state- ment that two algebraic expressions are equal. Definition 2 (Solution) . A solution or root is a value which yields a true statement when it replaces the variable.

1.15k views • 52 slides
Acoustic and electromagnetic transmission problems: wavenumber-explicit bounds and

Acoustic and electromagnetic transmission problems: wavenumber-explicit bounds and

R OMA , 14 M AY 2019 Acoustic and electromagnetic transmission problems: wavenumber-explicit bounds and resonance-free regions Andrea Moiola Joint work with E.A. Spence (Bath) Part I Helmholtz equation Helmholtz equation Acoustic waves in

981 views • 58 slides
Quarkonium Production: From JLab to an EIC Sylvester Joosten sylvester.joosten@temple.edu QCD

Quarkonium Production: From JLab to an EIC Sylvester Joosten sylvester.joosten@temple.edu QCD

This work is supported by DOE grant DE-FG02-94ER4084 Quarkonium Production: From JLab to an EIC Sylvester Joosten sylvester.joosten@temple.edu QCD Evolution 2018 (Santa Fe, NM) Quarkonium in electro- and photo-production l - Strong gluonic

849 views • 67 slides
Problem Formulation Assumption 1 The local best solution for equation (1) provides the best

Problem Formulation Assumption 1 The local best solution for equation (1) provides the best

Problem Formulation Assumption 1 The local best solution for equation (1) provides the best overall solution. Assumption 2 The derived sub-graph forms a spanning tree structure. Algorithm Optimization Vertex # Degree Weight Vertex #

78 views • 5 slides
Solution of the Gravitational Wave Introduction Constrained evolution Tensor Equation Using

Solution of the Gravitational Wave Introduction Constrained evolution Tensor Equation Using

Tensor Wave Equation J er ome Novak Solution of the Gravitational Wave Introduction Constrained evolution Tensor Equation Using Spectral Evolution Equation Numerical Methods Methods Vector Evolution Spherical Harmonics PDEs

746 views • 63 slides
SOLUTION OF LINEAR ALGEBRAIC EQUATIONS I. Hajj 2017 Linear Equation Solution Methods Consider a

SOLUTION OF LINEAR ALGEBRAIC EQUATIONS I. Hajj 2017 Linear Equation Solution Methods Consider a

ECE 552 Numerical Circuit Analysis Chapter Three SOLUTION OF LINEAR ALGEBRAIC EQUATIONS I. Hajj 2017 Linear Equation Solution Methods Consider a set of n linear algebraic equations Ax = b where A is a real or complex n x n nonsingular matrix.

1.18k views • 96 slides
S -matrix solution of the Lippmann-Schwinger equation for regular and singular potentials J. A.

S -matrix solution of the Lippmann-Schwinger equation for regular and singular potentials J. A.

S -matrix solution of the Lippmann-Schwinger equation for regular and singular potentials S -matrix solution of the Lippmann-Schwinger equation for regular and singular potentials J. A. Oller Departamento de F sica Universidad de Murcia 1

738 views • 58 slides
Parallel Solution of the 3-D Laplace Equation Using a Symmetric-Galerkin Boundary Integral

Parallel Solution of the 3-D Laplace Equation Using a Symmetric-Galerkin Boundary Integral

Parallel Solution of the 3-D Laplace Equation Using a Symmetric-Galerkin Boundary Integral Approximation Talisha Haywood Physics major w/ Emphasis in Computational Science Wofford College, Spartanburg, SC 29303 E-mail:

287 views • 14 slides
THE MONGE-AMP` ERE EQUATION: NOVEL FORMS AND NUMERICAL SOLUTION V.A. Zheligovsky (vlad@mitp.ru) ,

THE MONGE-AMP` ERE EQUATION: NOVEL FORMS AND NUMERICAL SOLUTION V.A. Zheligovsky (vlad@mitp.ru) ,

THE MONGE-AMP` ERE EQUATION: NOVEL FORMS AND NUMERICAL SOLUTION V.A. Zheligovsky (vlad@mitp.ru) , O.M. Podvigina International Institute of Earthquake Prediction Theory and Mathematical Geophysics, 84/32 Profsoyuznaya St., Moscow, Russian

721 views • 38 slides
The IPAT Equation The IPAT Equation The IPAT Equation The IPAT

The IPAT Equation The IPAT Equation The IPAT Equation The IPAT

The IPAT Equation The IPAT Equation The IPAT Equation The IPAT Equation References Ehrlich, P.R. and J.P. Holdren, 1972. Critique: One Dimensional Ecology. Bulletin of the Atomic Scientists 28(5): 16, 18-27.

619 views • 18 slides
Graph of f 1 . Since the equation y = f 1 ( x ) is the same as the equation x = f ( y ),

Graph of f 1 . Since the equation y = f 1 ( x ) is the same as the equation x = f ( y ),

Graph of f 1 . Since the equation y = f 1 ( x ) is the same as the equation x = f ( y ), the graphs of both equations are identical. To graph the equation x = f ( y ), we note that this equation results from switching the roles of x and

268 views • 6 slides
Linear Program Duality Frdric Giroire FG Duality 1/24 Motivation Finding bounds on the

Linear Program Duality Frdric Giroire FG Duality 1/24 Motivation Finding bounds on the

Linear Program Duality Frdric Giroire FG Duality 1/24 Motivation Finding bounds on the optimal solution. Provides a measure of the "goodness" of a solution. Provide certificate of optimality. Economic interpretation

675 views • 36 slides
From bounds on optimal growth towards a theory of good-deal hedging Dirk Becherer,

From bounds on optimal growth towards a theory of good-deal hedging Dirk Becherer,

Motivation General It o processes Good deals by BSDEs Problem Solution Bounds on optional growth From bounds on optimal growth towards a theory of good-deal hedging Dirk Becherer, Humboldt-Universit at Tamerza, Tunesia, Oct.2010

373 views • 35 slides
Solving a Nonlinear Equation Problem: Given f ( x ), find a root (solution) of f ( x ) = 0. If

Solving a Nonlinear Equation Problem: Given f ( x ), find a root (solution) of f ( x ) = 0. If

Solving a Nonlinear Equation Problem: Given f ( x ), find a root (solution) of f ( x ) = 0. If f is a polynomial with degree 4 or less, formulas for roots exist. If f is a polynomial with degree larger than 4, no finite formula exists

482 views • 12 slides
On the construction of several multivariate resultant matrices Weikun Sun School of Science

On the construction of several multivariate resultant matrices Weikun Sun School of Science

On the construction of several multivariate resultant matrices Weikun Sun School of Science Tianjin University of Technology and Education Jan. 1st, 2014 Table of Content 1 Backgrounds and Historical Review 2 Sylvester-Macaulay Type 3

833 views • 35 slides
XML in databases. Some XML-related standards (XLink, XPointer, XForms). Patryk Czarnik XML and

XML in databases. Some XML-related standards (XLink, XPointer, XForms). Patryk Czarnik XML and

XML in databases. Some XML-related standards (XLink, XPointer, XForms). Patryk Czarnik XML and Applications 2015/2016 Lecture 12 30.05.2016 XML support in databases categorisation Classic (usually relational) database with XML

542 views • 35 slides
Module 5: XML Modeling & Storage The major aspects of storing XML include Concepts: Data and

Module 5: XML Modeling & Storage The major aspects of storing XML include Concepts: Data and

Module 5: XML Modeling & Storage The major aspects of storing XML include Concepts: Data and Document Centrism Storage Mapping to relational schemas SQL/XML c Munindar P. Singh, CSC 513, Spring 2010 p.153 Modern Information Systems

291 views • 13 slides
Database Access via Programming Languages SQL is a direct query language; as such, it has

Database Access via Programming Languages SQL is a direct query language; as such, it has

Database Access via Programming Languages SQL is a direct query language; as such, it has limitations. Some reasons why access to databases via programming languages is needed : Complex computational processing of the data.

796 views • 9 slides
Structured matrix methods for the computation of multiple roots of univariate polynomials John D.

Structured matrix methods for the computation of multiple roots of univariate polynomials John D.

Structured matrix methods for the computation of multiple roots of univariate polynomials John D. Allan and Joab R. Winkler Department of Computer Science The University of Sheffield, United Kingdom Overview

537 views • 27 slides
The Bellows Theorem (Proof) Giovanni Viglietta JAIST July 5, 2018 The Bellows Theorem

The Bellows Theorem (Proof) Giovanni Viglietta JAIST July 5, 2018 The Bellows Theorem

The Bellows Theorem (Proof) Giovanni Viglietta JAIST July 5, 2018 The Bellows Theorem (Proof) Bellows theorem: statement Theorem (Sabitov, 1996) The volume V of a polyhedron (of any genus) with edge lengths 1 , , e

1.22k views • 93 slides
Tight Enclosure of Matrix Multiplication with Level 3 BLAS K. Ozaki (Shibaura Institute of

Tight Enclosure of Matrix Multiplication with Level 3 BLAS K. Ozaki (Shibaura Institute of

Tight Enclosure of Matrix Multiplication with Level 3 BLAS K. Ozaki (Shibaura Institute of Technology) joint work with T. Ogita (Tokyo Womans Christian University) 8th Small Workshop on Interval Methods the Charles university, Prague, Czech

591 views • 30 slides
On l 2 -Betti numbers and their analogues in positive characteristic Andrei Jaikin-Zapirain

On l 2 -Betti numbers and their analogues in positive characteristic Andrei Jaikin-Zapirain

On l 2 -Betti numbers and their analogues in positive characteristic Andrei Jaikin-Zapirain Birmingham, August 12th, 2017 L 2 -Betti numbers Andrei Jaikin-Zapirain The initial setting G is a finitely generated group. G > G 1 > G 2 > .

1.23k views • 96 slides

More recommend